From 9452e7cb702b2461f8fb6bf270666806aede6141 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Tim=20D=C3=BCsterhus?= Date: Wed, 5 Jun 2013 18:21:09 +0200 Subject: [PATCH] Remove modes that are unlikely to be used --- .../js/3rdParty/codemirror/mode/apl/apl.js | 160 ---- .../3rdParty/codemirror/mode/apl/index.html | 61 -- .../codemirror/mode/asterisk/asterisk.js | 183 ---- .../codemirror/mode/asterisk/index.html | 142 --- .../codemirror/mode/clojure/clojure.js | 224 ----- .../codemirror/mode/clojure/index.html | 76 -- .../3rdParty/codemirror/mode/cobol/cobol.js | 240 ------ .../3rdParty/codemirror/mode/cobol/index.html | 195 ----- .../codemirror/mode/commonlisp/commonlisp.js | 105 --- .../codemirror/mode/commonlisp/index.html | 165 ---- .../files/js/3rdParty/codemirror/mode/d/d.js | 205 ----- .../js/3rdParty/codemirror/mode/d/index.html | 262 ------ .../js/3rdParty/codemirror/mode/diff/diff.js | 32 - .../3rdParty/codemirror/mode/diff/index.html | 105 --- .../js/3rdParty/codemirror/mode/ecl/ecl.js | 192 ----- .../3rdParty/codemirror/mode/ecl/index.html | 39 - .../3rdParty/codemirror/mode/erlang/erlang.js | 464 ---------- .../codemirror/mode/erlang/index.html | 64 -- .../js/3rdParty/codemirror/mode/gas/gas.js | 330 ------- .../3rdParty/codemirror/mode/gas/index.html | 57 -- .../js/3rdParty/codemirror/mode/gfm/gfm.js | 96 --- .../3rdParty/codemirror/mode/gfm/index.html | 74 -- .../js/3rdParty/codemirror/mode/gfm/test.js | 112 --- .../js/3rdParty/codemirror/mode/go/go.js | 168 ---- .../js/3rdParty/codemirror/mode/go/index.html | 74 -- .../3rdParty/codemirror/mode/groovy/groovy.js | 210 ----- .../codemirror/mode/groovy/index.html | 73 -- .../js/3rdParty/codemirror/mode/haml/haml.js | 153 ---- .../3rdParty/codemirror/mode/haml/index.html | 67 -- .../js/3rdParty/codemirror/mode/haml/test.js | 94 -- .../codemirror/mode/haskell/haskell.js | 246 ------ .../codemirror/mode/haskell/index.html | 62 -- .../js/3rdParty/codemirror/mode/haxe/haxe.js | 429 --------- .../3rdParty/codemirror/mode/haxe/index.html | 90 -- .../js/3rdParty/codemirror/mode/http/http.js | 98 --- .../3rdParty/codemirror/mode/http/index.html | 32 - .../codemirror/mode/jinja2/index.html | 38 - .../3rdParty/codemirror/mode/jinja2/jinja2.js | 42 - .../codemirror/mode/livescript/LICENSE | 23 - .../codemirror/mode/livescript/index.html | 446 ---------- .../codemirror/mode/livescript/livescript.js | 267 ------ .../codemirror/mode/livescript/livescript.ls | 266 ------ .../3rdParty/codemirror/mode/lua/index.html | 74 -- .../js/3rdParty/codemirror/mode/lua/lua.js | 144 ---- .../codemirror/mode/markdown/index.html | 344 -------- .../codemirror/mode/markdown/markdown.js | 526 ----------- .../3rdParty/codemirror/mode/markdown/test.js | 636 -------------- .../files/js/3rdParty/codemirror/mode/meta.js | 77 -- .../3rdParty/codemirror/mode/mirc/index.html | 149 ---- .../js/3rdParty/codemirror/mode/mirc/mirc.js | 177 ---- .../codemirror/mode/ntriples/index.html | 33 - .../codemirror/mode/ntriples/ntriples.js | 170 ---- .../3rdParty/codemirror/mode/ocaml/index.html | 131 --- .../3rdParty/codemirror/mode/ocaml/ocaml.js | 116 --- .../3rdParty/codemirror/mode/pascal/LICENSE | 7 - .../codemirror/mode/pascal/index.html | 48 -- .../3rdParty/codemirror/mode/pascal/pascal.js | 94 -- .../js/3rdParty/codemirror/mode/perl/LICENSE | 19 - .../3rdParty/codemirror/mode/perl/index.html | 62 -- .../js/3rdParty/codemirror/mode/perl/perl.js | 816 ------------------ .../3rdParty/codemirror/mode/pig/index.html | 42 - .../js/3rdParty/codemirror/mode/pig/pig.js | 171 ---- .../codemirror/mode/properties/index.html | 41 - .../codemirror/mode/properties/properties.js | 63 -- .../codemirror/mode/python/LICENSE.txt | 21 - .../codemirror/mode/python/index.html | 135 --- .../3rdParty/codemirror/mode/python/python.js | 341 -------- .../js/3rdParty/codemirror/mode/q/index.html | 131 --- .../files/js/3rdParty/codemirror/mode/q/q.js | 124 --- .../js/3rdParty/codemirror/mode/r/LICENSE | 24 - .../js/3rdParty/codemirror/mode/r/index.html | 74 -- .../files/js/3rdParty/codemirror/mode/r/r.js | 141 --- .../codemirror/mode/rpm/changes/changes.js | 19 - .../codemirror/mode/rpm/changes/index.html | 53 -- .../codemirror/mode/rpm/spec/index.html | 99 --- .../codemirror/mode/rpm/spec/spec.css | 5 - .../3rdParty/codemirror/mode/rpm/spec/spec.js | 66 -- .../3rdParty/codemirror/mode/rst/LICENSE.txt | 21 - .../3rdParty/codemirror/mode/rst/index.html | 524 ----------- .../js/3rdParty/codemirror/mode/rst/rst.js | 550 ------------ .../js/3rdParty/codemirror/mode/ruby/LICENSE | 24 - .../3rdParty/codemirror/mode/ruby/index.html | 173 ---- .../js/3rdParty/codemirror/mode/ruby/ruby.js | 196 ----- .../3rdParty/codemirror/mode/rust/index.html | 48 -- .../js/3rdParty/codemirror/mode/rust/rust.js | 435 ---------- .../3rdParty/codemirror/mode/sass/index.html | 54 -- .../js/3rdParty/codemirror/mode/sass/sass.js | 330 ------- .../codemirror/mode/scheme/index.html | 65 -- .../3rdParty/codemirror/mode/scheme/scheme.js | 232 ----- .../js/3rdParty/codemirror/mode/sieve/LICENSE | 19 - .../3rdParty/codemirror/mode/sieve/index.html | 81 -- .../3rdParty/codemirror/mode/sieve/sieve.js | 183 ---- .../codemirror/mode/smalltalk/index.html | 57 -- .../codemirror/mode/smalltalk/smalltalk.js | 141 --- .../codemirror/mode/sparql/index.html | 42 - .../3rdParty/codemirror/mode/sparql/sparql.js | 143 --- .../3rdParty/codemirror/mode/stex/index.html | 98 --- .../js/3rdParty/codemirror/mode/stex/stex.js | 246 ------ .../js/3rdParty/codemirror/mode/stex/test.js | 120 --- .../3rdParty/codemirror/mode/tcl/index.html | 129 --- .../js/3rdParty/codemirror/mode/tcl/tcl.js | 131 --- .../codemirror/mode/tiddlywiki/index.html | 142 --- .../codemirror/mode/tiddlywiki/tiddlywiki.css | 14 - .../codemirror/mode/tiddlywiki/tiddlywiki.js | 353 -------- .../3rdParty/codemirror/mode/tiki/index.html | 81 -- .../js/3rdParty/codemirror/mode/tiki/tiki.css | 26 - .../js/3rdParty/codemirror/mode/tiki/tiki.js | 308 ------- .../codemirror/mode/turtle/index.html | 39 - .../3rdParty/codemirror/mode/turtle/turtle.js | 145 ---- .../3rdParty/codemirror/mode/vb/LICENSE.txt | 21 - .../js/3rdParty/codemirror/mode/vb/index.html | 88 -- .../js/3rdParty/codemirror/mode/vb/vb.js | 259 ------ .../codemirror/mode/vbscript/index.html | 42 - .../codemirror/mode/vbscript/vbscript.js | 26 - .../codemirror/mode/velocity/index.html | 103 --- .../codemirror/mode/velocity/velocity.js | 144 ---- .../codemirror/mode/verilog/index.html | 121 --- .../codemirror/mode/verilog/verilog.js | 182 ---- .../3rdParty/codemirror/mode/xquery/LICENSE | 20 - .../codemirror/mode/xquery/index.html | 221 ----- .../3rdParty/codemirror/mode/xquery/test.js | 64 -- .../3rdParty/codemirror/mode/xquery/xquery.js | 450 ---------- .../3rdParty/codemirror/mode/yaml/index.html | 68 -- .../js/3rdParty/codemirror/mode/yaml/yaml.js | 97 --- .../3rdParty/codemirror/mode/z80/index.html | 39 - .../js/3rdParty/codemirror/mode/z80/z80.js | 85 -- 126 files changed, 18814 deletions(-) delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/apl/apl.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/apl/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/asterisk/asterisk.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/asterisk/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/clojure/clojure.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/clojure/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/cobol/cobol.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/cobol/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/commonlisp/commonlisp.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/commonlisp/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/d/d.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/d/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/diff/diff.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/diff/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/ecl/ecl.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/ecl/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/erlang/erlang.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/erlang/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/gas/gas.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/gas/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/gfm.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/test.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/go/go.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/go/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/groovy/groovy.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/groovy/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/haml.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/test.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/haskell/haskell.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/haskell/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/haxe/haxe.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/haxe/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/http/http.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/http/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/jinja2/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/jinja2/jinja2.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/LICENSE delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/livescript.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/livescript.ls delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/lua/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/lua/lua.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/markdown.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/test.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/meta.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/mirc/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/mirc/mirc.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/ntriples/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/ntriples/ntriples.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/ocaml/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/ocaml/ocaml.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/LICENSE delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/pascal.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/LICENSE delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/perl.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/pig/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/pig/pig.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/properties/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/properties/properties.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/python/LICENSE.txt delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/python/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/python/python.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/q/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/q/q.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/r/LICENSE delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/r/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/r/r.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/changes/changes.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/changes/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/spec.css delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/spec.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/LICENSE.txt delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/rst.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/LICENSE delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/ruby.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rust/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/rust/rust.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/sass/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/sass/sass.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/scheme/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/scheme/scheme.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/LICENSE delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/sieve.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/smalltalk/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/smalltalk/smalltalk.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/sparql/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/sparql/sparql.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/stex.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/test.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/tcl/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/tcl/tcl.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/tiddlywiki.css delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/tiddlywiki.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/tiki.css delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/tiki.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/turtle/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/turtle/turtle.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/LICENSE.txt delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/vb.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/vbscript/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/vbscript/vbscript.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/velocity/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/velocity/velocity.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/verilog/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/verilog/verilog.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/LICENSE delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/test.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/xquery.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/yaml/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/yaml/yaml.js delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/z80/index.html delete mode 100644 wcfsetup/install/files/js/3rdParty/codemirror/mode/z80/z80.js diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/apl/apl.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/apl/apl.js deleted file mode 100644 index 5c23af85da..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/apl/apl.js +++ /dev/null @@ -1,160 +0,0 @@ -CodeMirror.defineMode("apl", function() { - var builtInOps = { - ".": "innerProduct", - "\\": "scan", - "/": "reduce", - "⌿": "reduce1Axis", - "⍀": "scan1Axis", - "¨": "each", - "⍣": "power" - }; - var builtInFuncs = { - "+": ["conjugate", "add"], - "−": ["negate", "subtract"], - "×": ["signOf", "multiply"], - "÷": ["reciprocal", "divide"], - "⌈": ["ceiling", "greaterOf"], - "⌊": ["floor", "lesserOf"], - "∣": ["absolute", "residue"], - "⍳": ["indexGenerate", "indexOf"], - "?": ["roll", "deal"], - "⋆": ["exponentiate", "toThePowerOf"], - "⍟": ["naturalLog", "logToTheBase"], - "○": ["piTimes", "circularFuncs"], - "!": ["factorial", "binomial"], - "⌹": ["matrixInverse", "matrixDivide"], - "<": [null, "lessThan"], - "≤": [null, "lessThanOrEqual"], - "=": [null, "equals"], - ">": [null, "greaterThan"], - "≥": [null, "greaterThanOrEqual"], - "≠": [null, "notEqual"], - "≡": ["depth", "match"], - "≢": [null, "notMatch"], - "∈": ["enlist", "membership"], - "⍷": [null, "find"], - "∪": ["unique", "union"], - "∩": [null, "intersection"], - "∼": ["not", "without"], - "∨": [null, "or"], - "∧": [null, "and"], - "⍱": [null, "nor"], - "⍲": [null, "nand"], - "⍴": ["shapeOf", "reshape"], - ",": ["ravel", "catenate"], - "⍪": [null, "firstAxisCatenate"], - "⌽": ["reverse", "rotate"], - "⊖": ["axis1Reverse", "axis1Rotate"], - "⍉": ["transpose", null], - "↑": ["first", "take"], - "↓": [null, "drop"], - "⊂": ["enclose", "partitionWithAxis"], - "⊃": ["diclose", "pick"], - "⌷": [null, "index"], - "⍋": ["gradeUp", null], - "⍒": ["gradeDown", null], - "⊤": ["encode", null], - "⊥": ["decode", null], - "⍕": ["format", "formatByExample"], - "⍎": ["execute", null], - "⊣": ["stop", "left"], - "⊢": ["pass", "right"] - }; - - var isOperator = /[\.\/⌿⍀¨⍣]/; - var isNiladic = /⍬/; - var isFunction = /[\+−×÷⌈⌊∣⍳\?⋆⍟○!⌹<≤=>≥≠≡≢∈⍷∪∩∼∨∧⍱⍲⍴,⍪⌽⊖⍉↑↓⊂⊃⌷⍋⍒⊤⊥⍕⍎⊣⊢]/; - var isArrow = /←/; - var isComment = /[⍝#].*$/; - - var stringEater = function(type) { - var prev; - prev = false; - return function(c) { - prev = c; - if (c === type) { - return prev === "\\"; - } - return true; - }; - }; - return { - startState: function() { - return { - prev: false, - func: false, - op: false, - string: false, - escape: false - }; - }, - token: function(stream, state) { - var ch, funcName, word; - if (stream.eatSpace()) { - return null; - } - ch = stream.next(); - if (ch === '"' || ch === "'") { - stream.eatWhile(stringEater(ch)); - stream.next(); - state.prev = true; - return "string"; - } - if (/[\[{\(]/.test(ch)) { - state.prev = false; - return null; - } - if (/[\]}\)]/.test(ch)) { - state.prev = true; - return null; - } - if (isNiladic.test(ch)) { - state.prev = false; - return "niladic"; - } - if (/[¯\d]/.test(ch)) { - if (state.func) { - state.func = false; - state.prev = false; - } else { - state.prev = true; - } - stream.eatWhile(/[\w\.]/); - return "number"; - } - if (isOperator.test(ch)) { - return "operator apl-" + builtInOps[ch]; - } - if (isArrow.test(ch)) { - return "apl-arrow"; - } - if (isFunction.test(ch)) { - funcName = "apl-"; - if (builtInFuncs[ch] != null) { - if (state.prev) { - funcName += builtInFuncs[ch][1]; - } else { - funcName += builtInFuncs[ch][0]; - } - } - state.func = true; - state.prev = false; - return "function " + funcName; - } - if (isComment.test(ch)) { - stream.skipToEnd(); - return "comment"; - } - if (ch === "∘" && stream.peek() === ".") { - stream.next(); - return "function jot-dot"; - } - stream.eatWhile(/[\w\$_]/); - word = stream.current(); - state.prev = true; - return "keyword"; - } - }; -}); - -CodeMirror.defineMIME("text/apl", "apl"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/apl/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/apl/index.html deleted file mode 100644 index 119ff17f19..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/apl/index.html +++ /dev/null @@ -1,61 +0,0 @@ - - - - - CodeMirror: APL mode - - - - - - - - -

CodeMirror: APL mode

- -
- - - -

Simple mode that tries to handle APL as well as it can.

-

It attempts to label functions/operators based upon - monadic/dyadic usage (but this is far from fully fleshed out). - This means there are meaningful classnames so hover states can - have popups etc.

- -

MIME types defined: text/apl (APL code)

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/asterisk/asterisk.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/asterisk/asterisk.js deleted file mode 100644 index 60b689d1d5..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/asterisk/asterisk.js +++ /dev/null @@ -1,183 +0,0 @@ -/* - * ===================================================================================== - * - * Filename: mode/asterisk/asterisk.js - * - * Description: CodeMirror mode for Asterisk dialplan - * - * Created: 05/17/2012 09:20:25 PM - * Revision: none - * - * Author: Stas Kobzar (stas@modulis.ca), - * Company: Modulis.ca Inc. - * - * ===================================================================================== - */ - -CodeMirror.defineMode("asterisk", function() { - var atoms = ["exten", "same", "include","ignorepat","switch"], - dpcmd = ["#include","#exec"], - apps = [ - "addqueuemember","adsiprog","aelsub","agentlogin","agentmonitoroutgoing","agi", - "alarmreceiver","amd","answer","authenticate","background","backgrounddetect", - "bridge","busy","callcompletioncancel","callcompletionrequest","celgenuserevent", - "changemonitor","chanisavail","channelredirect","chanspy","clearhash","confbridge", - "congestion","continuewhile","controlplayback","dahdiacceptr2call","dahdibarge", - "dahdiras","dahdiscan","dahdisendcallreroutingfacility","dahdisendkeypadfacility", - "datetime","dbdel","dbdeltree","deadagi","dial","dictate","directory","disa", - "dumpchan","eagi","echo","endwhile","exec","execif","execiftime","exitwhile","extenspy", - "externalivr","festival","flash","followme","forkcdr","getcpeid","gosub","gosubif", - "goto","gotoif","gotoiftime","hangup","iax2provision","ices","importvar","incomplete", - "ivrdemo","jabberjoin","jabberleave","jabbersend","jabbersendgroup","jabberstatus", - "jack","log","macro","macroexclusive","macroexit","macroif","mailboxexists","meetme", - "meetmeadmin","meetmechanneladmin","meetmecount","milliwatt","minivmaccmess","minivmdelete", - "minivmgreet","minivmmwi","minivmnotify","minivmrecord","mixmonitor","monitor","morsecode", - "mp3player","mset","musiconhold","nbscat","nocdr","noop","odbc","odbc","odbcfinish", - "originate","ospauth","ospfinish","osplookup","ospnext","page","park","parkandannounce", - "parkedcall","pausemonitor","pausequeuemember","pickup","pickupchan","playback","playtones", - "privacymanager","proceeding","progress","queue","queuelog","raiseexception","read","readexten", - "readfile","receivefax","receivefax","receivefax","record","removequeuemember", - "resetcdr","retrydial","return","ringing","sayalpha","saycountedadj","saycountednoun", - "saycountpl","saydigits","saynumber","sayphonetic","sayunixtime","senddtmf","sendfax", - "sendfax","sendfax","sendimage","sendtext","sendurl","set","setamaflags", - "setcallerpres","setmusiconhold","sipaddheader","sipdtmfmode","sipremoveheader","skel", - "slastation","slatrunk","sms","softhangup","speechactivategrammar","speechbackground", - "speechcreate","speechdeactivategrammar","speechdestroy","speechloadgrammar","speechprocessingsound", - "speechstart","speechunloadgrammar","stackpop","startmusiconhold","stopmixmonitor","stopmonitor", - "stopmusiconhold","stopplaytones","system","testclient","testserver","transfer","tryexec", - "trysystem","unpausemonitor","unpausequeuemember","userevent","verbose","vmauthenticate", - "vmsayname","voicemail","voicemailmain","wait","waitexten","waitfornoise","waitforring", - "waitforsilence","waitmusiconhold","waituntil","while","zapateller" - ]; - - function basicToken(stream,state){ - var cur = ''; - var ch = ''; - ch = stream.next(); - // comment - if(ch == ";") { - stream.skipToEnd(); - return "comment"; - } - // context - if(ch == '[') { - stream.skipTo(']'); - stream.eat(']'); - return "header"; - } - // string - if(ch == '"') { - stream.skipTo('"'); - return "string"; - } - if(ch == "'") { - stream.skipTo("'"); - return "string-2"; - } - // dialplan commands - if(ch == '#') { - stream.eatWhile(/\w/); - cur = stream.current(); - if(dpcmd.indexOf(cur) !== -1) { - stream.skipToEnd(); - return "strong"; - } - } - // application args - if(ch == '$'){ - var ch1 = stream.peek(); - if(ch1 == '{'){ - stream.skipTo('}'); - stream.eat('}'); - return "variable-3"; - } - } - // extension - stream.eatWhile(/\w/); - cur = stream.current(); - if(atoms.indexOf(cur) !== -1) { - state.extenStart = true; - switch(cur) { - case 'same': state.extenSame = true; break; - case 'include': - case 'switch': - case 'ignorepat': - state.extenInclude = true;break; - default:break; - } - return "atom"; - } - } - - return { - startState: function() { - return { - extenStart: false, - extenSame: false, - extenInclude: false, - extenExten: false, - extenPriority: false, - extenApplication: false - }; - }, - token: function(stream, state) { - - var cur = ''; - var ch = ''; - if(stream.eatSpace()) return null; - // extension started - if(state.extenStart){ - stream.eatWhile(/[^\s]/); - cur = stream.current(); - if(/^=>?$/.test(cur)){ - state.extenExten = true; - state.extenStart = false; - return "strong"; - } else { - state.extenStart = false; - stream.skipToEnd(); - return "error"; - } - } else if(state.extenExten) { - // set exten and priority - state.extenExten = false; - state.extenPriority = true; - stream.eatWhile(/[^,]/); - if(state.extenInclude) { - stream.skipToEnd(); - state.extenPriority = false; - state.extenInclude = false; - } - if(state.extenSame) { - state.extenPriority = false; - state.extenSame = false; - state.extenApplication = true; - } - return "tag"; - } else if(state.extenPriority) { - state.extenPriority = false; - state.extenApplication = true; - ch = stream.next(); // get comma - if(state.extenSame) return null; - stream.eatWhile(/[^,]/); - return "number"; - } else if(state.extenApplication) { - stream.eatWhile(/,/); - cur = stream.current(); - if(cur === ',') return null; - stream.eatWhile(/\w/); - cur = stream.current().toLowerCase(); - state.extenApplication = false; - if(apps.indexOf(cur) !== -1){ - return "def strong"; - } - } else{ - return basicToken(stream,state); - } - - return null; - } - }; -}); - -CodeMirror.defineMIME("text/x-asterisk", "asterisk"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/asterisk/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/asterisk/index.html deleted file mode 100644 index 0a796a0119..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/asterisk/index.html +++ /dev/null @@ -1,142 +0,0 @@ - - - - - CodeMirror: Asterisk dialplan mode - - - - - - - -

CodeMirror: Asterisk dialplan mode

-
- - -

MIME types defined: text/x-asterisk.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/clojure/clojure.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/clojure/clojure.js deleted file mode 100644 index ee22a12fed..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/clojure/clojure.js +++ /dev/null @@ -1,224 +0,0 @@ -/** - * Author: Hans Engel - * Branched from CodeMirror's Scheme mode (by Koh Zi Han, based on implementation by Koh Zi Chun) - */ -CodeMirror.defineMode("clojure", function () { - var BUILTIN = "builtin", COMMENT = "comment", STRING = "string", CHARACTER = "string-2", - ATOM = "atom", NUMBER = "number", BRACKET = "bracket", KEYWORD = "keyword"; - var INDENT_WORD_SKIP = 2; - - function makeKeywords(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - - var atoms = makeKeywords("true false nil"); - - var keywords = makeKeywords( - "defn defn- def def- defonce defmulti defmethod defmacro defstruct deftype defprotocol defrecord defproject deftest slice defalias defhinted defmacro- defn-memo defnk defnk defonce- defunbound defunbound- defvar defvar- let letfn do case cond condp for loop recur when when-not when-let when-first if if-let if-not . .. -> ->> doto and or dosync doseq dotimes dorun doall load import unimport ns in-ns refer try catch finally throw with-open with-local-vars binding gen-class gen-and-load-class gen-and-save-class handler-case handle"); - - var builtins = makeKeywords( - "* *' *1 *2 *3 *agent* *allow-unresolved-vars* *assert* *clojure-version* *command-line-args* *compile-files* *compile-path* *compiler-options* *data-readers* *e *err* *file* *flush-on-newline* *fn-loader* *in* *math-context* *ns* *out* *print-dup* *print-length* *print-level* *print-meta* *print-readably* *read-eval* *source-path* *unchecked-math* *use-context-classloader* *verbose-defrecords* *warn-on-reflection* + +' - -' -> ->> ->ArrayChunk ->Vec ->VecNode ->VecSeq -cache-protocol-fn -reset-methods .. / < <= = == > >= EMPTY-NODE accessor aclone add-classpath add-watch agent agent-error agent-errors aget alength alias all-ns alter alter-meta! alter-var-root amap ancestors and apply areduce array-map aset aset-boolean aset-byte aset-char aset-double aset-float aset-int aset-long aset-short assert assoc assoc! assoc-in associative? atom await await-for await1 bases bean bigdec bigint biginteger binding bit-and bit-and-not bit-clear bit-flip bit-not bit-or bit-set bit-shift-left bit-shift-right bit-test bit-xor boolean boolean-array booleans bound-fn bound-fn* bound? butlast byte byte-array bytes case cast char char-array char-escape-string char-name-string char? chars chunk chunk-append chunk-buffer chunk-cons chunk-first chunk-next chunk-rest chunked-seq? class class? clear-agent-errors clojure-version coll? comment commute comp comparator compare compare-and-set! compile complement concat cond condp conj conj! cons constantly construct-proxy contains? count counted? create-ns create-struct cycle dec dec' decimal? declare default-data-readers definline definterface defmacro defmethod defmulti defn defn- defonce defprotocol defrecord defstruct deftype delay delay? deliver denominator deref derive descendants destructure disj disj! dissoc dissoc! distinct distinct? doall dorun doseq dosync dotimes doto double double-array doubles drop drop-last drop-while empty empty? ensure enumeration-seq error-handler error-mode eval even? every-pred every? ex-data ex-info extend extend-protocol extend-type extenders extends? false? ffirst file-seq filter filterv find find-keyword find-ns find-protocol-impl find-protocol-method find-var first flatten float float-array float? floats flush fn fn? fnext fnil for force format frequencies future future-call future-cancel future-cancelled? future-done? future? gen-class gen-interface gensym get get-in get-method get-proxy-class get-thread-bindings get-validator group-by hash hash-combine hash-map hash-set identical? identity if-let if-not ifn? import in-ns inc inc' init-proxy instance? int int-array integer? interleave intern interpose into into-array ints io! isa? iterate iterator-seq juxt keep keep-indexed key keys keyword keyword? last lazy-cat lazy-seq let letfn line-seq list list* list? load load-file load-reader load-string loaded-libs locking long long-array longs loop macroexpand macroexpand-1 make-array make-hierarchy map map-indexed map? mapcat mapv max max-key memfn memoize merge merge-with meta method-sig methods min min-key mod munge name namespace namespace-munge neg? newline next nfirst nil? nnext not not-any? not-empty not-every? not= ns ns-aliases ns-imports ns-interns ns-map ns-name ns-publics ns-refers ns-resolve ns-unalias ns-unmap nth nthnext nthrest num number? numerator object-array odd? or parents partial partition partition-all partition-by pcalls peek persistent! pmap pop pop! pop-thread-bindings pos? pr pr-str prefer-method prefers primitives-classnames print print-ctor print-dup print-method print-simple print-str printf println println-str prn prn-str promise proxy proxy-call-with-super proxy-mappings proxy-name proxy-super push-thread-bindings pvalues quot rand rand-int rand-nth range ratio? rational? rationalize re-find re-groups re-matcher re-matches re-pattern re-seq read read-line read-string realized? reduce reduce-kv reductions ref ref-history-count ref-max-history ref-min-history ref-set refer refer-clojure reify release-pending-sends rem remove remove-all-methods remove-method remove-ns remove-watch repeat repeatedly replace replicate require reset! reset-meta! resolve rest restart-agent resultset-seq reverse reversible? rseq rsubseq satisfies? second select-keys send send-off seq seq? seque sequence sequential? set set-error-handler! set-error-mode! set-validator! set? short short-array shorts shuffle shutdown-agents slurp some some-fn sort sort-by sorted-map sorted-map-by sorted-set sorted-set-by sorted? special-symbol? spit split-at split-with str string? struct struct-map subs subseq subvec supers swap! symbol symbol? sync take take-last take-nth take-while test the-ns thread-bound? time to-array to-array-2d trampoline transient tree-seq true? type unchecked-add unchecked-add-int unchecked-byte unchecked-char unchecked-dec unchecked-dec-int unchecked-divide-int unchecked-double unchecked-float unchecked-inc unchecked-inc-int unchecked-int unchecked-long unchecked-multiply unchecked-multiply-int unchecked-negate unchecked-negate-int unchecked-remainder-int unchecked-short unchecked-subtract unchecked-subtract-int underive unquote unquote-splicing update-in update-proxy use val vals var-get var-set var? vary-meta vec vector vector-of vector? when when-first when-let when-not while with-bindings with-bindings* with-in-str with-loading-context with-local-vars with-meta with-open with-out-str with-precision with-redefs with-redefs-fn xml-seq zero? zipmap *default-data-reader-fn* as-> cond-> cond->> reduced reduced? send-via set-agent-send-executor! set-agent-send-off-executor! some-> some->>"); - - var indentKeys = makeKeywords( - // Built-ins - "ns fn def defn defmethod bound-fn if if-not case condp when while when-not when-first do future comment doto locking proxy with-open with-precision reify deftype defrecord defprotocol extend extend-protocol extend-type try catch " + - - // Binding forms - "let letfn binding loop for doseq dotimes when-let if-let " + - - // Data structures - "defstruct struct-map assoc " + - - // clojure.test - "testing deftest " + - - // contrib - "handler-case handle dotrace deftrace"); - - var tests = { - digit: /\d/, - digit_or_colon: /[\d:]/, - hex: /[0-9a-f]/i, - sign: /[+-]/, - exponent: /e/i, - keyword_char: /[^\s\(\[\;\)\]]/, - symbol: /[\w*+!\-\._?:\/]/ - }; - - function stateStack(indent, type, prev) { // represents a state stack object - this.indent = indent; - this.type = type; - this.prev = prev; - } - - function pushStack(state, indent, type) { - state.indentStack = new stateStack(indent, type, state.indentStack); - } - - function popStack(state) { - state.indentStack = state.indentStack.prev; - } - - function isNumber(ch, stream){ - // hex - if ( ch === '0' && stream.eat(/x/i) ) { - stream.eatWhile(tests.hex); - return true; - } - - // leading sign - if ( ( ch == '+' || ch == '-' ) && ( tests.digit.test(stream.peek()) ) ) { - stream.eat(tests.sign); - ch = stream.next(); - } - - if ( tests.digit.test(ch) ) { - stream.eat(ch); - stream.eatWhile(tests.digit); - - if ( '.' == stream.peek() ) { - stream.eat('.'); - stream.eatWhile(tests.digit); - } - - if ( stream.eat(tests.exponent) ) { - stream.eat(tests.sign); - stream.eatWhile(tests.digit); - } - - return true; - } - - return false; - } - - // Eat character that starts after backslash \ - function eatCharacter(stream) { - var first = stream.next(); - // Read special literals: backspace, newline, space, return. - // Just read all lowercase letters. - if (first.match(/[a-z]/) && stream.match(/[a-z]+/, true)) { - return; - } - // Read unicode character: \u1000 \uA0a1 - if (first === "u") { - stream.match(/[0-9a-z]{4}/i, true); - } - } - - return { - startState: function () { - return { - indentStack: null, - indentation: 0, - mode: false - }; - }, - - token: function (stream, state) { - if (state.indentStack == null && stream.sol()) { - // update indentation, but only if indentStack is empty - state.indentation = stream.indentation(); - } - - // skip spaces - if (stream.eatSpace()) { - return null; - } - var returnType = null; - - switch(state.mode){ - case "string": // multi-line string parsing mode - var next, escaped = false; - while ((next = stream.next()) != null) { - if (next == "\"" && !escaped) { - - state.mode = false; - break; - } - escaped = !escaped && next == "\\"; - } - returnType = STRING; // continue on in string mode - break; - default: // default parsing mode - var ch = stream.next(); - - if (ch == "\"") { - state.mode = "string"; - returnType = STRING; - } else if (ch == "\\") { - eatCharacter(stream); - returnType = CHARACTER; - } else if (ch == "'" && !( tests.digit_or_colon.test(stream.peek()) )) { - returnType = ATOM; - } else if (ch == ";") { // comment - stream.skipToEnd(); // rest of the line is a comment - returnType = COMMENT; - } else if (isNumber(ch,stream)){ - returnType = NUMBER; - } else if (ch == "(" || ch == "[" || ch == "{" ) { - var keyWord = '', indentTemp = stream.column(), letter; - /** - Either - (indent-word .. - (non-indent-word .. - (;something else, bracket, etc. - */ - - if (ch == "(") while ((letter = stream.eat(tests.keyword_char)) != null) { - keyWord += letter; - } - - if (keyWord.length > 0 && (indentKeys.propertyIsEnumerable(keyWord) || - /^(?:def|with)/.test(keyWord))) { // indent-word - pushStack(state, indentTemp + INDENT_WORD_SKIP, ch); - } else { // non-indent word - // we continue eating the spaces - stream.eatSpace(); - if (stream.eol() || stream.peek() == ";") { - // nothing significant after - // we restart indentation 1 space after - pushStack(state, indentTemp + 1, ch); - } else { - pushStack(state, indentTemp + stream.current().length, ch); // else we match - } - } - stream.backUp(stream.current().length - 1); // undo all the eating - - returnType = BRACKET; - } else if (ch == ")" || ch == "]" || ch == "}") { - returnType = BRACKET; - if (state.indentStack != null && state.indentStack.type == (ch == ")" ? "(" : (ch == "]" ? "[" :"{"))) { - popStack(state); - } - } else if ( ch == ":" ) { - stream.eatWhile(tests.symbol); - return ATOM; - } else { - stream.eatWhile(tests.symbol); - - if (keywords && keywords.propertyIsEnumerable(stream.current())) { - returnType = KEYWORD; - } else if (builtins && builtins.propertyIsEnumerable(stream.current())) { - returnType = BUILTIN; - } else if (atoms && atoms.propertyIsEnumerable(stream.current())) { - returnType = ATOM; - } else returnType = null; - } - } - - return returnType; - }, - - indent: function (state) { - if (state.indentStack == null) return state.indentation; - return state.indentStack.indent; - }, - - lineComment: ";;" - }; -}); - -CodeMirror.defineMIME("text/x-clojure", "clojure"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/clojure/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/clojure/index.html deleted file mode 100644 index bfe6fc955a..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/clojure/index.html +++ /dev/null @@ -1,76 +0,0 @@ - - - - - CodeMirror: Clojure mode - - - - - - - -

CodeMirror: Clojure mode

-
- - -

MIME types defined: text/x-clojure.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/cobol/cobol.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/cobol/cobol.js deleted file mode 100644 index d92491dde8..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/cobol/cobol.js +++ /dev/null @@ -1,240 +0,0 @@ -/** - * Author: Gautam Mehta - * Branched from CodeMirror's Scheme mode - */ -CodeMirror.defineMode("cobol", function () { - var BUILTIN = "builtin", COMMENT = "comment", STRING = "string", - ATOM = "atom", NUMBER = "number", KEYWORD = "keyword", MODTAG = "header", - COBOLLINENUM = "def", PERIOD = "link"; - function makeKeywords(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - var atoms = makeKeywords("TRUE FALSE ZEROES ZEROS ZERO SPACES SPACE LOW-VALUE LOW-VALUES "); - var keywords = makeKeywords( - "ACCEPT ACCESS ACQUIRE ADD ADDRESS " + - "ADVANCING AFTER ALIAS ALL ALPHABET " + - "ALPHABETIC ALPHABETIC-LOWER ALPHABETIC-UPPER ALPHANUMERIC ALPHANUMERIC-EDITED " + - "ALSO ALTER ALTERNATE AND ANY " + - "ARE AREA AREAS ARITHMETIC ASCENDING " + - "ASSIGN AT ATTRIBUTE AUTHOR AUTO " + - "AUTO-SKIP AUTOMATIC B-AND B-EXOR B-LESS " + - "B-NOT B-OR BACKGROUND-COLOR BACKGROUND-COLOUR BEEP " + - "BEFORE BELL BINARY BIT BITS " + - "BLANK BLINK BLOCK BOOLEAN BOTTOM " + - "BY CALL CANCEL CD CF " + - "CH CHARACTER CHARACTERS CLASS CLOCK-UNITS " + - "CLOSE COBOL CODE CODE-SET COL " + - "COLLATING COLUMN COMMA COMMIT COMMITMENT " + - "COMMON COMMUNICATION COMP COMP-0 COMP-1 " + - "COMP-2 COMP-3 COMP-4 COMP-5 COMP-6 " + - "COMP-7 COMP-8 COMP-9 COMPUTATIONAL COMPUTATIONAL-0 " + - "COMPUTATIONAL-1 COMPUTATIONAL-2 COMPUTATIONAL-3 COMPUTATIONAL-4 COMPUTATIONAL-5 " + - "COMPUTATIONAL-6 COMPUTATIONAL-7 COMPUTATIONAL-8 COMPUTATIONAL-9 COMPUTE " + - "CONFIGURATION CONNECT CONSOLE CONTAINED CONTAINS " + - "CONTENT CONTINUE CONTROL CONTROL-AREA CONTROLS " + - "CONVERTING COPY CORR CORRESPONDING COUNT " + - "CRT CRT-UNDER CURRENCY CURRENT CURSOR " + - "DATA DATE DATE-COMPILED DATE-WRITTEN DAY " + - "DAY-OF-WEEK DB DB-ACCESS-CONTROL-KEY DB-DATA-NAME DB-EXCEPTION " + - "DB-FORMAT-NAME DB-RECORD-NAME DB-SET-NAME DB-STATUS DBCS " + - "DBCS-EDITED DE DEBUG-CONTENTS DEBUG-ITEM DEBUG-LINE " + - "DEBUG-NAME DEBUG-SUB-1 DEBUG-SUB-2 DEBUG-SUB-3 DEBUGGING " + - "DECIMAL-POINT DECLARATIVES DEFAULT DELETE DELIMITED " + - "DELIMITER DEPENDING DESCENDING DESCRIBED DESTINATION " + - "DETAIL DISABLE DISCONNECT DISPLAY DISPLAY-1 " + - "DISPLAY-2 DISPLAY-3 DISPLAY-4 DISPLAY-5 DISPLAY-6 " + - "DISPLAY-7 DISPLAY-8 DISPLAY-9 DIVIDE DIVISION " + - "DOWN DROP DUPLICATE DUPLICATES DYNAMIC " + - "EBCDIC EGI EJECT ELSE EMI " + - "EMPTY EMPTY-CHECK ENABLE END END. END-ACCEPT END-ACCEPT. " + - "END-ADD END-CALL END-COMPUTE END-DELETE END-DISPLAY " + - "END-DIVIDE END-EVALUATE END-IF END-INVOKE END-MULTIPLY " + - "END-OF-PAGE END-PERFORM END-READ END-RECEIVE END-RETURN " + - "END-REWRITE END-SEARCH END-START END-STRING END-SUBTRACT " + - "END-UNSTRING END-WRITE END-XML ENTER ENTRY " + - "ENVIRONMENT EOP EQUAL EQUALS ERASE " + - "ERROR ESI EVALUATE EVERY EXCEEDS " + - "EXCEPTION EXCLUSIVE EXIT EXTEND EXTERNAL " + - "EXTERNALLY-DESCRIBED-KEY FD FETCH FILE FILE-CONTROL " + - "FILE-STREAM FILES FILLER FINAL FIND " + - "FINISH FIRST FOOTING FOR FOREGROUND-COLOR " + - "FOREGROUND-COLOUR FORMAT FREE FROM FULL " + - "FUNCTION GENERATE GET GIVING GLOBAL " + - "GO GOBACK GREATER GROUP HEADING " + - "HIGH-VALUE HIGH-VALUES HIGHLIGHT I-O I-O-CONTROL " + - "ID IDENTIFICATION IF IN INDEX " + - "INDEX-1 INDEX-2 INDEX-3 INDEX-4 INDEX-5 " + - "INDEX-6 INDEX-7 INDEX-8 INDEX-9 INDEXED " + - "INDIC INDICATE INDICATOR INDICATORS INITIAL " + - "INITIALIZE INITIATE INPUT INPUT-OUTPUT INSPECT " + - "INSTALLATION INTO INVALID INVOKE IS " + - "JUST JUSTIFIED KANJI KEEP KEY " + - "LABEL LAST LD LEADING LEFT " + - "LEFT-JUSTIFY LENGTH LENGTH-CHECK LESS LIBRARY " + - "LIKE LIMIT LIMITS LINAGE LINAGE-COUNTER " + - "LINE LINE-COUNTER LINES LINKAGE LOCAL-STORAGE " + - "LOCALE LOCALLY LOCK " + - "MEMBER MEMORY MERGE MESSAGE METACLASS " + - "MODE MODIFIED MODIFY MODULES MOVE " + - "MULTIPLE MULTIPLY NATIONAL NATIVE NEGATIVE " + - "NEXT NO NO-ECHO NONE NOT " + - "NULL NULL-KEY-MAP NULL-MAP NULLS NUMBER " + - "NUMERIC NUMERIC-EDITED OBJECT OBJECT-COMPUTER OCCURS " + - "OF OFF OMITTED ON ONLY " + - "OPEN OPTIONAL OR ORDER ORGANIZATION " + - "OTHER OUTPUT OVERFLOW OWNER PACKED-DECIMAL " + - "PADDING PAGE PAGE-COUNTER PARSE PERFORM " + - "PF PH PIC PICTURE PLUS " + - "POINTER POSITION POSITIVE PREFIX PRESENT " + - "PRINTING PRIOR PROCEDURE PROCEDURE-POINTER PROCEDURES " + - "PROCEED PROCESS PROCESSING PROGRAM PROGRAM-ID " + - "PROMPT PROTECTED PURGE QUEUE QUOTE " + - "QUOTES RANDOM RD READ READY " + - "REALM RECEIVE RECONNECT RECORD RECORD-NAME " + - "RECORDS RECURSIVE REDEFINES REEL REFERENCE " + - "REFERENCE-MONITOR REFERENCES RELATION RELATIVE RELEASE " + - "REMAINDER REMOVAL RENAMES REPEATED REPLACE " + - "REPLACING REPORT REPORTING REPORTS REPOSITORY " + - "REQUIRED RERUN RESERVE RESET RETAINING " + - "RETRIEVAL RETURN RETURN-CODE RETURNING REVERSE-VIDEO " + - "REVERSED REWIND REWRITE RF RH " + - "RIGHT RIGHT-JUSTIFY ROLLBACK ROLLING ROUNDED " + - "RUN SAME SCREEN SD SEARCH " + - "SECTION SECURE SECURITY SEGMENT SEGMENT-LIMIT " + - "SELECT SEND SENTENCE SEPARATE SEQUENCE " + - "SEQUENTIAL SET SHARED SIGN SIZE " + - "SKIP1 SKIP2 SKIP3 SORT SORT-MERGE " + - "SORT-RETURN SOURCE SOURCE-COMPUTER SPACE-FILL " + - "SPECIAL-NAMES STANDARD STANDARD-1 STANDARD-2 " + - "START STARTING STATUS STOP STORE " + - "STRING SUB-QUEUE-1 SUB-QUEUE-2 SUB-QUEUE-3 SUB-SCHEMA " + - "SUBFILE SUBSTITUTE SUBTRACT SUM SUPPRESS " + - "SYMBOLIC SYNC SYNCHRONIZED SYSIN SYSOUT " + - "TABLE TALLYING TAPE TENANT TERMINAL " + - "TERMINATE TEST TEXT THAN THEN " + - "THROUGH THRU TIME TIMES TITLE " + - "TO TOP TRAILING TRAILING-SIGN TRANSACTION " + - "TYPE TYPEDEF UNDERLINE UNEQUAL UNIT " + - "UNSTRING UNTIL UP UPDATE UPON " + - "USAGE USAGE-MODE USE USING VALID " + - "VALIDATE VALUE VALUES VARYING VLR " + - "WAIT WHEN WHEN-COMPILED WITH WITHIN " + - "WORDS WORKING-STORAGE WRITE XML XML-CODE " + - "XML-EVENT XML-NTEXT XML-TEXT ZERO ZERO-FILL " ); - - var builtins = makeKeywords("- * ** / + < <= = > >= "); - var tests = { - digit: /\d/, - digit_or_colon: /[\d:]/, - hex: /[0-9a-f]/i, - sign: /[+-]/, - exponent: /e/i, - keyword_char: /[^\s\(\[\;\)\]]/, - symbol: /[\w*+\-]/ - }; - function isNumber(ch, stream){ - // hex - if ( ch === '0' && stream.eat(/x/i) ) { - stream.eatWhile(tests.hex); - return true; - } - // leading sign - if ( ( ch == '+' || ch == '-' ) && ( tests.digit.test(stream.peek()) ) ) { - stream.eat(tests.sign); - ch = stream.next(); - } - if ( tests.digit.test(ch) ) { - stream.eat(ch); - stream.eatWhile(tests.digit); - if ( '.' == stream.peek()) { - stream.eat('.'); - stream.eatWhile(tests.digit); - } - if ( stream.eat(tests.exponent) ) { - stream.eat(tests.sign); - stream.eatWhile(tests.digit); - } - return true; - } - return false; - } - return { - startState: function () { - return { - indentStack: null, - indentation: 0, - mode: false - }; - }, - token: function (stream, state) { - if (state.indentStack == null && stream.sol()) { - // update indentation, but only if indentStack is empty - state.indentation = 6 ; //stream.indentation(); - } - // skip spaces - if (stream.eatSpace()) { - return null; - } - var returnType = null; - switch(state.mode){ - case "string": // multi-line string parsing mode - var next = false; - while ((next = stream.next()) != null) { - if (next == "\"" || next == "\'") { - state.mode = false; - break; - } - } - returnType = STRING; // continue on in string mode - break; - default: // default parsing mode - var ch = stream.next(); - var col = stream.column(); - if (col >= 0 && col <= 5) { - returnType = COBOLLINENUM; - } else if (col >= 72 && col <= 79) { - stream.skipToEnd(); - returnType = MODTAG; - } else if (ch == "*" && col == 6) { // comment - stream.skipToEnd(); // rest of the line is a comment - returnType = COMMENT; - } else if (ch == "\"" || ch == "\'") { - state.mode = "string"; - returnType = STRING; - } else if (ch == "'" && !( tests.digit_or_colon.test(stream.peek()) )) { - returnType = ATOM; - } else if (ch == ".") { - returnType = PERIOD; - } else if (isNumber(ch,stream)){ - returnType = NUMBER; - } else { - if (stream.current().match(tests.symbol)) { - while (col < 71) { - if (stream.eat(tests.symbol) === undefined) { - break; - } else { - col++; - } - } - } - if (keywords && keywords.propertyIsEnumerable(stream.current().toUpperCase())) { - returnType = KEYWORD; - } else if (builtins && builtins.propertyIsEnumerable(stream.current().toUpperCase())) { - returnType = BUILTIN; - } else if (atoms && atoms.propertyIsEnumerable(stream.current().toUpperCase())) { - returnType = ATOM; - } else returnType = null; - } - } - return returnType; - }, - indent: function (state) { - if (state.indentStack == null) return state.indentation; - return state.indentStack.indent; - } - }; -}); - -CodeMirror.defineMIME("text/x-cobol", "cobol"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/cobol/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/cobol/index.html deleted file mode 100644 index 71cc2fa151..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/cobol/index.html +++ /dev/null @@ -1,195 +0,0 @@ - - - - - CodeMirror: COBOL mode - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

Select Theme Select Font Size - - - - -

- - - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/commonlisp/commonlisp.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/commonlisp/commonlisp.js deleted file mode 100644 index 8fa08c8ac2..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/commonlisp/commonlisp.js +++ /dev/null @@ -1,105 +0,0 @@ -CodeMirror.defineMode("commonlisp", function (config) { - var assumeBody = /^with|^def|^do|^prog|case$|^cond$|bind$|when$|unless$/; - var numLiteral = /^(?:[+\-]?(?:\d+|\d*\.\d+)(?:[efd][+\-]?\d+)?|[+\-]?\d+(?:\/[+\-]?\d+)?|#b[+\-]?[01]+|#o[+\-]?[0-7]+|#x[+\-]?[\da-f]+)/; - var symbol = /[^\s'`,@()\[\]";]/; - var type; - - function readSym(stream) { - var ch; - while (ch = stream.next()) { - if (ch == "\\") stream.next(); - else if (!symbol.test(ch)) { stream.backUp(1); break; } - } - return stream.current(); - } - - function base(stream, state) { - if (stream.eatSpace()) {type = "ws"; return null;} - if (stream.match(numLiteral)) return "number"; - var ch = stream.next(); - if (ch == "\\") ch = stream.next(); - - if (ch == '"') return (state.tokenize = inString)(stream, state); - else if (ch == "(") { type = "open"; return "bracket"; } - else if (ch == ")" || ch == "]") { type = "close"; return "bracket"; } - else if (ch == ";") { stream.skipToEnd(); type = "ws"; return "comment"; } - else if (/['`,@]/.test(ch)) return null; - else if (ch == "|") { - if (stream.skipTo("|")) { stream.next(); return "symbol"; } - else { stream.skipToEnd(); return "error"; } - } else if (ch == "#") { - var ch = stream.next(); - if (ch == "[") { type = "open"; return "bracket"; } - else if (/[+\-=\.']/.test(ch)) return null; - else if (/\d/.test(ch) && stream.match(/^\d*#/)) return null; - else if (ch == "|") return (state.tokenize = inComment)(stream, state); - else if (ch == ":") { readSym(stream); return "meta"; } - else return "error"; - } else { - var name = readSym(stream); - if (name == ".") return null; - type = "symbol"; - if (name == "nil" || name == "t") return "atom"; - if (name.charAt(0) == ":") return "keyword"; - if (name.charAt(0) == "&") return "variable-2"; - return "variable"; - } - } - - function inString(stream, state) { - var escaped = false, next; - while (next = stream.next()) { - if (next == '"' && !escaped) { state.tokenize = base; break; } - escaped = !escaped && next == "\\"; - } - return "string"; - } - - function inComment(stream, state) { - var next, last; - while (next = stream.next()) { - if (next == "#" && last == "|") { state.tokenize = base; break; } - last = next; - } - type = "ws"; - return "comment"; - } - - return { - startState: function () { - return {ctx: {prev: null, start: 0, indentTo: 0}, tokenize: base}; - }, - - token: function (stream, state) { - if (stream.sol() && typeof state.ctx.indentTo != "number") - state.ctx.indentTo = state.ctx.start + 1; - - type = null; - var style = state.tokenize(stream, state); - if (type != "ws") { - if (state.ctx.indentTo == null) { - if (type == "symbol" && assumeBody.test(stream.current())) - state.ctx.indentTo = state.ctx.start + config.indentUnit; - else - state.ctx.indentTo = "next"; - } else if (state.ctx.indentTo == "next") { - state.ctx.indentTo = stream.column(); - } - } - if (type == "open") state.ctx = {prev: state.ctx, start: stream.column(), indentTo: null}; - else if (type == "close") state.ctx = state.ctx.prev || state.ctx; - return style; - }, - - indent: function (state, _textAfter) { - var i = state.ctx.indentTo; - return typeof i == "number" ? i : state.ctx.start + 1; - }, - - lineComment: ";;", - blockCommentStart: "#|", - blockCommentEnd: "|#" - }; -}); - -CodeMirror.defineMIME("text/x-common-lisp", "commonlisp"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/commonlisp/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/commonlisp/index.html deleted file mode 100644 index f9766a844c..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/commonlisp/index.html +++ /dev/null @@ -1,165 +0,0 @@ - - - - - CodeMirror: Common Lisp mode - - - - - - - -

CodeMirror: Common Lisp mode

-
- - -

MIME types defined: text/x-common-lisp.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/d/d.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/d/d.js deleted file mode 100644 index ab345f1a0d..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/d/d.js +++ /dev/null @@ -1,205 +0,0 @@ -CodeMirror.defineMode("d", function(config, parserConfig) { - var indentUnit = config.indentUnit, - statementIndentUnit = parserConfig.statementIndentUnit || indentUnit, - keywords = parserConfig.keywords || {}, - builtin = parserConfig.builtin || {}, - blockKeywords = parserConfig.blockKeywords || {}, - atoms = parserConfig.atoms || {}, - hooks = parserConfig.hooks || {}, - multiLineStrings = parserConfig.multiLineStrings; - var isOperatorChar = /[+\-*&%=<>!?|\/]/; - - var curPunc; - - function tokenBase(stream, state) { - var ch = stream.next(); - if (hooks[ch]) { - var result = hooks[ch](stream, state); - if (result !== false) return result; - } - if (ch == '"' || ch == "'" || ch == "`") { - state.tokenize = tokenString(ch); - return state.tokenize(stream, state); - } - if (/[\[\]{}\(\),;\:\.]/.test(ch)) { - curPunc = ch; - return null; - } - if (/\d/.test(ch)) { - stream.eatWhile(/[\w\.]/); - return "number"; - } - if (ch == "/") { - if (stream.eat("+")) { - state.tokenize = tokenComment; - return tokenNestedComment(stream, state); - } - if (stream.eat("*")) { - state.tokenize = tokenComment; - return tokenComment(stream, state); - } - if (stream.eat("/")) { - stream.skipToEnd(); - return "comment"; - } - } - if (isOperatorChar.test(ch)) { - stream.eatWhile(isOperatorChar); - return "operator"; - } - stream.eatWhile(/[\w\$_]/); - var cur = stream.current(); - if (keywords.propertyIsEnumerable(cur)) { - if (blockKeywords.propertyIsEnumerable(cur)) curPunc = "newstatement"; - return "keyword"; - } - if (builtin.propertyIsEnumerable(cur)) { - if (blockKeywords.propertyIsEnumerable(cur)) curPunc = "newstatement"; - return "builtin"; - } - if (atoms.propertyIsEnumerable(cur)) return "atom"; - return "variable"; - } - - function tokenString(quote) { - return function(stream, state) { - var escaped = false, next, end = false; - while ((next = stream.next()) != null) { - if (next == quote && !escaped) {end = true; break;} - escaped = !escaped && next == "\\"; - } - if (end || !(escaped || multiLineStrings)) - state.tokenize = null; - return "string"; - }; - } - - function tokenComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == "/" && maybeEnd) { - state.tokenize = null; - break; - } - maybeEnd = (ch == "*"); - } - return "comment"; - } - - function tokenNestedComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == "/" && maybeEnd) { - state.tokenize = null; - break; - } - maybeEnd = (ch == "+"); - } - return "comment"; - } - - function Context(indented, column, type, align, prev) { - this.indented = indented; - this.column = column; - this.type = type; - this.align = align; - this.prev = prev; - } - function pushContext(state, col, type) { - var indent = state.indented; - if (state.context && state.context.type == "statement") - indent = state.context.indented; - return state.context = new Context(indent, col, type, null, state.context); - } - function popContext(state) { - var t = state.context.type; - if (t == ")" || t == "]" || t == "}") - state.indented = state.context.indented; - return state.context = state.context.prev; - } - - // Interface - - return { - startState: function(basecolumn) { - return { - tokenize: null, - context: new Context((basecolumn || 0) - indentUnit, 0, "top", false), - indented: 0, - startOfLine: true - }; - }, - - token: function(stream, state) { - var ctx = state.context; - if (stream.sol()) { - if (ctx.align == null) ctx.align = false; - state.indented = stream.indentation(); - state.startOfLine = true; - } - if (stream.eatSpace()) return null; - curPunc = null; - var style = (state.tokenize || tokenBase)(stream, state); - if (style == "comment" || style == "meta") return style; - if (ctx.align == null) ctx.align = true; - - if ((curPunc == ";" || curPunc == ":" || curPunc == ",") && ctx.type == "statement") popContext(state); - else if (curPunc == "{") pushContext(state, stream.column(), "}"); - else if (curPunc == "[") pushContext(state, stream.column(), "]"); - else if (curPunc == "(") pushContext(state, stream.column(), ")"); - else if (curPunc == "}") { - while (ctx.type == "statement") ctx = popContext(state); - if (ctx.type == "}") ctx = popContext(state); - while (ctx.type == "statement") ctx = popContext(state); - } - else if (curPunc == ctx.type) popContext(state); - else if (((ctx.type == "}" || ctx.type == "top") && curPunc != ';') || (ctx.type == "statement" && curPunc == "newstatement")) - pushContext(state, stream.column(), "statement"); - state.startOfLine = false; - return style; - }, - - indent: function(state, textAfter) { - if (state.tokenize != tokenBase && state.tokenize != null) return CodeMirror.Pass; - var ctx = state.context, firstChar = textAfter && textAfter.charAt(0); - if (ctx.type == "statement" && firstChar == "}") ctx = ctx.prev; - var closing = firstChar == ctx.type; - if (ctx.type == "statement") return ctx.indented + (firstChar == "{" ? 0 : statementIndentUnit); - else if (ctx.align) return ctx.column + (closing ? 0 : 1); - else return ctx.indented + (closing ? 0 : indentUnit); - }, - - electricChars: "{}" - }; -}); - -(function() { - function words(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - - var blockKeywords = "body catch class do else enum for foreach foreach_reverse if in interface mixin " + - "out scope struct switch try union unittest version while with"; - - CodeMirror.defineMIME("text/x-d", { - name: "d", - keywords: words("abstract alias align asm assert auto break case cast cdouble cent cfloat const continue " + - "debug default delegate delete deprecated export extern final finally function goto immutable " + - "import inout invariant is lazy macro module new nothrow override package pragma private " + - "protected public pure ref return shared short static super synchronized template this " + - "throw typedef typeid typeof volatile __FILE__ __LINE__ __gshared __traits __vector __parameters " + - blockKeywords), - blockKeywords: words(blockKeywords), - builtin: words("bool byte char creal dchar double float idouble ifloat int ireal long real short ubyte " + - "ucent uint ulong ushort wchar wstring void size_t sizediff_t"), - atoms: words("exit failure success true false null"), - hooks: { - "@": function(stream, _state) { - stream.eatWhile(/[\w\$_]/); - return "meta"; - } - } - }); -}()); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/d/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/d/index.html deleted file mode 100644 index 13332727ad..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/d/index.html +++ /dev/null @@ -1,262 +0,0 @@ - - - - - CodeMirror: D mode - - - - - - - - -

CodeMirror: D mode

- -
- - - -

Simple mode that handle D-Syntax (DLang Homepage).

- -

MIME types defined: text/x-d - .

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/diff/diff.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/diff/diff.js deleted file mode 100644 index 9a0d90ea55..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/diff/diff.js +++ /dev/null @@ -1,32 +0,0 @@ -CodeMirror.defineMode("diff", function() { - - var TOKEN_NAMES = { - '+': 'positive', - '-': 'negative', - '@': 'meta' - }; - - return { - token: function(stream) { - var tw_pos = stream.string.search(/[\t ]+?$/); - - if (!stream.sol() || tw_pos === 0) { - stream.skipToEnd(); - return ("error " + ( - TOKEN_NAMES[stream.string.charAt(0)] || '')).replace(/ $/, ''); - } - - var token_name = TOKEN_NAMES[stream.peek()] || stream.skipToEnd(); - - if (tw_pos === -1) { - stream.skipToEnd(); - } else { - stream.pos = tw_pos; - } - - return token_name; - } - }; -}); - -CodeMirror.defineMIME("text/x-diff", "diff"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/diff/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/diff/index.html deleted file mode 100644 index 556025204d..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/diff/index.html +++ /dev/null @@ -1,105 +0,0 @@ - - - - - CodeMirror: Diff mode - - - - - - - -

CodeMirror: Diff mode

-
- - -

MIME types defined: text/x-diff.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ecl/ecl.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/ecl/ecl.js deleted file mode 100644 index 7601b189a0..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ecl/ecl.js +++ /dev/null @@ -1,192 +0,0 @@ -CodeMirror.defineMode("ecl", function(config) { - - function words(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - - function metaHook(stream, state) { - if (!state.startOfLine) return false; - stream.skipToEnd(); - return "meta"; - } - - var indentUnit = config.indentUnit; - var keyword = words("abs acos allnodes ascii asin asstring atan atan2 ave case choose choosen choosesets clustersize combine correlation cos cosh count covariance cron dataset dedup define denormalize distribute distributed distribution ebcdic enth error evaluate event eventextra eventname exists exp failcode failmessage fetch fromunicode getisvalid global graph group hash hash32 hash64 hashcrc hashmd5 having if index intformat isvalid iterate join keyunicode length library limit ln local log loop map matched matchlength matchposition matchtext matchunicode max merge mergejoin min nolocal nonempty normalize parse pipe power preload process project pull random range rank ranked realformat recordof regexfind regexreplace regroup rejected rollup round roundup row rowdiff sample set sin sinh sizeof soapcall sort sorted sqrt stepped stored sum table tan tanh thisnode topn tounicode transfer trim truncate typeof ungroup unicodeorder variance which workunit xmldecode xmlencode xmltext xmlunicode"); - var variable = words("apply assert build buildindex evaluate fail keydiff keypatch loadxml nothor notify output parallel sequential soapcall wait"); - var variable_2 = words("__compressed__ all and any as atmost before beginc++ best between case const counter csv descend encrypt end endc++ endmacro except exclusive expire export extend false few first flat from full function group header heading hole ifblock import in interface joined keep keyed last left limit load local locale lookup macro many maxcount maxlength min skew module named nocase noroot noscan nosort not of only opt or outer overwrite packed partition penalty physicallength pipe quote record relationship repeat return right scan self separator service shared skew skip sql store terminator thor threshold token transform trim true type unicodeorder unsorted validate virtual whole wild within xml xpath"); - var variable_3 = words("ascii big_endian boolean data decimal ebcdic integer pattern qstring real record rule set of string token udecimal unicode unsigned varstring varunicode"); - var builtin = words("checkpoint deprecated failcode failmessage failure global independent onwarning persist priority recovery stored success wait when"); - var blockKeywords = words("catch class do else finally for if switch try while"); - var atoms = words("true false null"); - var hooks = {"#": metaHook}; - var multiLineStrings; - var isOperatorChar = /[+\-*&%=<>!?|\/]/; - - var curPunc; - - function tokenBase(stream, state) { - var ch = stream.next(); - if (hooks[ch]) { - var result = hooks[ch](stream, state); - if (result !== false) return result; - } - if (ch == '"' || ch == "'") { - state.tokenize = tokenString(ch); - return state.tokenize(stream, state); - } - if (/[\[\]{}\(\),;\:\.]/.test(ch)) { - curPunc = ch; - return null; - } - if (/\d/.test(ch)) { - stream.eatWhile(/[\w\.]/); - return "number"; - } - if (ch == "/") { - if (stream.eat("*")) { - state.tokenize = tokenComment; - return tokenComment(stream, state); - } - if (stream.eat("/")) { - stream.skipToEnd(); - return "comment"; - } - } - if (isOperatorChar.test(ch)) { - stream.eatWhile(isOperatorChar); - return "operator"; - } - stream.eatWhile(/[\w\$_]/); - var cur = stream.current().toLowerCase(); - if (keyword.propertyIsEnumerable(cur)) { - if (blockKeywords.propertyIsEnumerable(cur)) curPunc = "newstatement"; - return "keyword"; - } else if (variable.propertyIsEnumerable(cur)) { - if (blockKeywords.propertyIsEnumerable(cur)) curPunc = "newstatement"; - return "variable"; - } else if (variable_2.propertyIsEnumerable(cur)) { - if (blockKeywords.propertyIsEnumerable(cur)) curPunc = "newstatement"; - return "variable-2"; - } else if (variable_3.propertyIsEnumerable(cur)) { - if (blockKeywords.propertyIsEnumerable(cur)) curPunc = "newstatement"; - return "variable-3"; - } else if (builtin.propertyIsEnumerable(cur)) { - if (blockKeywords.propertyIsEnumerable(cur)) curPunc = "newstatement"; - return "builtin"; - } else { //Data types are of from KEYWORD## - var i = cur.length - 1; - while(i >= 0 && (!isNaN(cur[i]) || cur[i] == '_')) - --i; - - if (i > 0) { - var cur2 = cur.substr(0, i + 1); - if (variable_3.propertyIsEnumerable(cur2)) { - if (blockKeywords.propertyIsEnumerable(cur2)) curPunc = "newstatement"; - return "variable-3"; - } - } - } - if (atoms.propertyIsEnumerable(cur)) return "atom"; - return null; - } - - function tokenString(quote) { - return function(stream, state) { - var escaped = false, next, end = false; - while ((next = stream.next()) != null) { - if (next == quote && !escaped) {end = true; break;} - escaped = !escaped && next == "\\"; - } - if (end || !(escaped || multiLineStrings)) - state.tokenize = tokenBase; - return "string"; - }; - } - - function tokenComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == "/" && maybeEnd) { - state.tokenize = tokenBase; - break; - } - maybeEnd = (ch == "*"); - } - return "comment"; - } - - function Context(indented, column, type, align, prev) { - this.indented = indented; - this.column = column; - this.type = type; - this.align = align; - this.prev = prev; - } - function pushContext(state, col, type) { - return state.context = new Context(state.indented, col, type, null, state.context); - } - function popContext(state) { - var t = state.context.type; - if (t == ")" || t == "]" || t == "}") - state.indented = state.context.indented; - return state.context = state.context.prev; - } - - // Interface - - return { - startState: function(basecolumn) { - return { - tokenize: null, - context: new Context((basecolumn || 0) - indentUnit, 0, "top", false), - indented: 0, - startOfLine: true - }; - }, - - token: function(stream, state) { - var ctx = state.context; - if (stream.sol()) { - if (ctx.align == null) ctx.align = false; - state.indented = stream.indentation(); - state.startOfLine = true; - } - if (stream.eatSpace()) return null; - curPunc = null; - var style = (state.tokenize || tokenBase)(stream, state); - if (style == "comment" || style == "meta") return style; - if (ctx.align == null) ctx.align = true; - - if ((curPunc == ";" || curPunc == ":") && ctx.type == "statement") popContext(state); - else if (curPunc == "{") pushContext(state, stream.column(), "}"); - else if (curPunc == "[") pushContext(state, stream.column(), "]"); - else if (curPunc == "(") pushContext(state, stream.column(), ")"); - else if (curPunc == "}") { - while (ctx.type == "statement") ctx = popContext(state); - if (ctx.type == "}") ctx = popContext(state); - while (ctx.type == "statement") ctx = popContext(state); - } - else if (curPunc == ctx.type) popContext(state); - else if (ctx.type == "}" || ctx.type == "top" || (ctx.type == "statement" && curPunc == "newstatement")) - pushContext(state, stream.column(), "statement"); - state.startOfLine = false; - return style; - }, - - indent: function(state, textAfter) { - if (state.tokenize != tokenBase && state.tokenize != null) return 0; - var ctx = state.context, firstChar = textAfter && textAfter.charAt(0); - if (ctx.type == "statement" && firstChar == "}") ctx = ctx.prev; - var closing = firstChar == ctx.type; - if (ctx.type == "statement") return ctx.indented + (firstChar == "{" ? 0 : indentUnit); - else if (ctx.align) return ctx.column + (closing ? 0 : 1); - else return ctx.indented + (closing ? 0 : indentUnit); - }, - - electricChars: "{}" - }; -}); - -CodeMirror.defineMIME("text/x-ecl", "ecl"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ecl/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/ecl/index.html deleted file mode 100644 index 0ba88c3995..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ecl/index.html +++ /dev/null @@ -1,39 +0,0 @@ - - - - CodeMirror: ECL mode - - - - - - - -

CodeMirror: ECL mode

-
- - -

Based on CodeMirror's clike mode. For more information see HPCC Systems web site.

-

MIME types defined: text/x-ecl.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/erlang/erlang.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/erlang/erlang.js deleted file mode 100644 index 79e0434d1f..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/erlang/erlang.js +++ /dev/null @@ -1,464 +0,0 @@ -// block; "begin", "case", "fun", "if", "receive", "try": closed by "end" -// block internal; "after", "catch", "of" -// guard; "when", closed by "->" -// "->" opens a clause, closed by ";" or "." -// "<<" opens a binary, closed by ">>" -// "," appears in arglists, lists, tuples and terminates lines of code -// "." resets indentation to 0 -// obsolete; "cond", "let", "query" - -CodeMirror.defineMIME("text/x-erlang", "erlang"); - -CodeMirror.defineMode("erlang", function(cmCfg) { - - function rval(state,stream,type) { - // distinguish between "." as terminator and record field operator - if (type == "record") { - state.context = "record"; - }else{ - state.context = false; - } - - // remember last significant bit on last line for indenting - if (type != "whitespace" && type != "comment") { - state.lastToken = stream.current(); - } - // erlang -> CodeMirror tag - switch (type) { - case "atom": return "atom"; - case "attribute": return "attribute"; - case "builtin": return "builtin"; - case "comment": return "comment"; - case "fun": return "meta"; - case "function": return "tag"; - case "guard": return "property"; - case "keyword": return "keyword"; - case "macro": return "variable-2"; - case "number": return "number"; - case "operator": return "operator"; - case "record": return "bracket"; - case "string": return "string"; - case "type": return "def"; - case "variable": return "variable"; - case "error": return "error"; - case "separator": return null; - case "open_paren": return null; - case "close_paren": return null; - default: return null; - } - } - - var typeWords = [ - "-type", "-spec", "-export_type", "-opaque"]; - - var keywordWords = [ - "after","begin","catch","case","cond","end","fun","if", - "let","of","query","receive","try","when"]; - - var separatorWords = [ - "->",";",":",".",","]; - - var operatorWords = [ - "and","andalso","band","bnot","bor","bsl","bsr","bxor", - "div","not","or","orelse","rem","xor"]; - - var symbolWords = [ - "+","-","*","/",">",">=","<","=<","=:=","==","=/=","/=","||","<-"]; - - var openParenWords = [ - "<<","(","[","{"]; - - var closeParenWords = [ - "}","]",")",">>"]; - - var guardWords = [ - "is_atom","is_binary","is_bitstring","is_boolean","is_float", - "is_function","is_integer","is_list","is_number","is_pid", - "is_port","is_record","is_reference","is_tuple", - "atom","binary","bitstring","boolean","function","integer","list", - "number","pid","port","record","reference","tuple"]; - - var bifWords = [ - "abs","adler32","adler32_combine","alive","apply","atom_to_binary", - "atom_to_list","binary_to_atom","binary_to_existing_atom", - "binary_to_list","binary_to_term","bit_size","bitstring_to_list", - "byte_size","check_process_code","contact_binary","crc32", - "crc32_combine","date","decode_packet","delete_module", - "disconnect_node","element","erase","exit","float","float_to_list", - "garbage_collect","get","get_keys","group_leader","halt","hd", - "integer_to_list","internal_bif","iolist_size","iolist_to_binary", - "is_alive","is_atom","is_binary","is_bitstring","is_boolean", - "is_float","is_function","is_integer","is_list","is_number","is_pid", - "is_port","is_process_alive","is_record","is_reference","is_tuple", - "length","link","list_to_atom","list_to_binary","list_to_bitstring", - "list_to_existing_atom","list_to_float","list_to_integer", - "list_to_pid","list_to_tuple","load_module","make_ref","module_loaded", - "monitor_node","node","node_link","node_unlink","nodes","notalive", - "now","open_port","pid_to_list","port_close","port_command", - "port_connect","port_control","pre_loaded","process_flag", - "process_info","processes","purge_module","put","register", - "registered","round","self","setelement","size","spawn","spawn_link", - "spawn_monitor","spawn_opt","split_binary","statistics", - "term_to_binary","time","throw","tl","trunc","tuple_size", - "tuple_to_list","unlink","unregister","whereis"]; - - // ignored for indenting purposes - var ignoreWords = [ - ",", ":", "catch", "after", "of", "cond", "let", "query"]; - - - var smallRE = /[a-z_]/; - var largeRE = /[A-Z_]/; - var digitRE = /[0-9]/; - var octitRE = /[0-7]/; - var anumRE = /[a-z_A-Z0-9]/; - var symbolRE = /[\+\-\*\/<>=\|:]/; - var openParenRE = /[<\(\[\{]/; - var closeParenRE = /[>\)\]\}]/; - var sepRE = /[\->\.,:;]/; - - function isMember(element,list) { - return (-1 < list.indexOf(element)); - } - - function isPrev(stream,string) { - var start = stream.start; - var len = string.length; - if (len <= start) { - var word = stream.string.slice(start-len,start); - return word == string; - }else{ - return false; - } - } - - function tokenize(stream, state) { - if (stream.eatSpace()) { - return rval(state,stream,"whitespace"); - } - - // attributes and type specs - if ((peekToken(state).token == "" || peekToken(state).token == ".") && - stream.peek() == '-') { - stream.next(); - if (stream.eat(smallRE) && stream.eatWhile(anumRE)) { - if (isMember(stream.current(),typeWords)) { - return rval(state,stream,"type"); - }else{ - return rval(state,stream,"attribute"); - } - } - stream.backUp(1); - } - - var ch = stream.next(); - - // comment - if (ch == '%') { - stream.skipToEnd(); - return rval(state,stream,"comment"); - } - - // macro - if (ch == '?') { - stream.eatWhile(anumRE); - return rval(state,stream,"macro"); - } - - // record - if ( ch == "#") { - stream.eatWhile(anumRE); - return rval(state,stream,"record"); - } - - // char - if ( ch == "$") { - if (stream.next() == "\\") { - if (!stream.eatWhile(octitRE)) { - stream.next(); - } - } - return rval(state,stream,"string"); - } - - // quoted atom - if (ch == '\'') { - if (singleQuote(stream)) { - return rval(state,stream,"atom"); - }else{ - return rval(state,stream,"error"); - } - } - - // string - if (ch == '"') { - if (doubleQuote(stream)) { - return rval(state,stream,"string"); - }else{ - return rval(state,stream,"error"); - } - } - - // variable - if (largeRE.test(ch)) { - stream.eatWhile(anumRE); - return rval(state,stream,"variable"); - } - - // atom/keyword/BIF/function - if (smallRE.test(ch)) { - stream.eatWhile(anumRE); - - if (stream.peek() == "/") { - stream.next(); - if (stream.eatWhile(digitRE)) { - return rval(state,stream,"fun"); // f/0 style fun - }else{ - stream.backUp(1); - return rval(state,stream,"atom"); - } - } - - var w = stream.current(); - - if (isMember(w,keywordWords)) { - pushToken(state,stream); - return rval(state,stream,"keyword"); - } - if (stream.peek() == "(") { - // 'put' and 'erlang:put' are bifs, 'foo:put' is not - if (isMember(w,bifWords) && - (!isPrev(stream,":") || isPrev(stream,"erlang:"))) { - return rval(state,stream,"builtin"); - }else{ - return rval(state,stream,"function"); - } - } - if (isMember(w,guardWords)) { - return rval(state,stream,"guard"); - } - if (isMember(w,operatorWords)) { - return rval(state,stream,"operator"); - } - if (stream.peek() == ":") { - if (w == "erlang") { - return rval(state,stream,"builtin"); - } else { - return rval(state,stream,"function"); - } - } - return rval(state,stream,"atom"); - } - - // number - if (digitRE.test(ch)) { - stream.eatWhile(digitRE); - if (stream.eat('#')) { - stream.eatWhile(digitRE); // 16#10 style integer - } else { - if (stream.eat('.')) { // float - stream.eatWhile(digitRE); - } - if (stream.eat(/[eE]/)) { - stream.eat(/[-+]/); // float with exponent - stream.eatWhile(digitRE); - } - } - return rval(state,stream,"number"); // normal integer - } - - // open parens - if (nongreedy(stream,openParenRE,openParenWords)) { - pushToken(state,stream); - return rval(state,stream,"open_paren"); - } - - // close parens - if (nongreedy(stream,closeParenRE,closeParenWords)) { - pushToken(state,stream); - return rval(state,stream,"close_paren"); - } - - // separators - if (greedy(stream,sepRE,separatorWords)) { - // distinguish between "." as terminator and record field operator - if (state.context == false) { - pushToken(state,stream); - } - return rval(state,stream,"separator"); - } - - // operators - if (greedy(stream,symbolRE,symbolWords)) { - return rval(state,stream,"operator"); - } - - return rval(state,stream,null); - } - - function nongreedy(stream,re,words) { - if (stream.current().length == 1 && re.test(stream.current())) { - stream.backUp(1); - while (re.test(stream.peek())) { - stream.next(); - if (isMember(stream.current(),words)) { - return true; - } - } - stream.backUp(stream.current().length-1); - } - return false; - } - - function greedy(stream,re,words) { - if (stream.current().length == 1 && re.test(stream.current())) { - while (re.test(stream.peek())) { - stream.next(); - } - while (0 < stream.current().length) { - if (isMember(stream.current(),words)) { - return true; - }else{ - stream.backUp(1); - } - } - stream.next(); - } - return false; - } - - function doubleQuote(stream) { - return quote(stream, '"', '\\'); - } - - function singleQuote(stream) { - return quote(stream,'\'','\\'); - } - - function quote(stream,quoteChar,escapeChar) { - while (!stream.eol()) { - var ch = stream.next(); - if (ch == quoteChar) { - return true; - }else if (ch == escapeChar) { - stream.next(); - } - } - return false; - } - - function Token(stream) { - this.token = stream ? stream.current() : ""; - this.column = stream ? stream.column() : 0; - this.indent = stream ? stream.indentation() : 0; - } - - function myIndent(state,textAfter) { - var indent = cmCfg.indentUnit; - var outdentWords = ["after","catch"]; - var token = (peekToken(state)).token; - var wordAfter = takewhile(textAfter,/[^a-z]/); - - if (isMember(token,openParenWords)) { - return (peekToken(state)).column+token.length; - }else if (token == "." || token == ""){ - return 0; - }else if (token == "->") { - if (wordAfter == "end") { - return peekToken(state,2).column; - }else if (peekToken(state,2).token == "fun") { - return peekToken(state,2).column+indent; - }else{ - return (peekToken(state)).indent+indent; - } - }else if (isMember(wordAfter,outdentWords)) { - return (peekToken(state)).indent; - }else{ - return (peekToken(state)).column+indent; - } - } - - function takewhile(str,re) { - var m = str.match(re); - return m ? str.slice(0,m.index) : str; - } - - function popToken(state) { - return state.tokenStack.pop(); - } - - function peekToken(state,depth) { - var len = state.tokenStack.length; - var dep = (depth ? depth : 1); - if (len < dep) { - return new Token; - }else{ - return state.tokenStack[len-dep]; - } - } - - function pushToken(state,stream) { - var token = stream.current(); - var prev_token = peekToken(state).token; - if (isMember(token,ignoreWords)) { - return false; - }else if (drop_both(prev_token,token)) { - popToken(state); - return false; - }else if (drop_first(prev_token,token)) { - popToken(state); - return pushToken(state,stream); - }else{ - state.tokenStack.push(new Token(stream)); - return true; - } - } - - function drop_first(open, close) { - switch (open+" "+close) { - case "when ->": return true; - case "-> end": return true; - case "-> .": return true; - case ". .": return true; - default: return false; - } - } - - function drop_both(open, close) { - switch (open+" "+close) { - case "( )": return true; - case "[ ]": return true; - case "{ }": return true; - case "<< >>": return true; - case "begin end": return true; - case "case end": return true; - case "fun end": return true; - case "if end": return true; - case "receive end": return true; - case "try end": return true; - case "-> ;": return true; - default: return false; - } - } - - return { - startState: - function() { - return {tokenStack: [], - context: false, - lastToken: null}; - }, - - token: - function(stream, state) { - return tokenize(stream, state); - }, - - indent: - function(state, textAfter) { - return myIndent(state,textAfter); - }, - - lineComment: "%" - }; -}); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/erlang/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/erlang/index.html deleted file mode 100644 index fd21521c88..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/erlang/index.html +++ /dev/null @@ -1,64 +0,0 @@ - - - - - CodeMirror: Erlang mode - - - - - - - - - -

CodeMirror: Erlang mode

- -
- - - -

MIME types defined: text/x-erlang.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gas/gas.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/gas/gas.js deleted file mode 100644 index a6e6892908..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gas/gas.js +++ /dev/null @@ -1,330 +0,0 @@ -CodeMirror.defineMode("gas", function(_config, parserConfig) { - 'use strict'; - - // If an architecture is specified, its initialization function may - // populate this array with custom parsing functions which will be - // tried in the event that the standard functions do not find a match. - var custom = []; - - // The symbol used to start a line comment changes based on the target - // architecture. - // If no architecture is pased in "parserConfig" then only multiline - // comments will have syntax support. - var lineCommentStartSymbol = ""; - - // These directives are architecture independent. - // Machine specific directives should go in their respective - // architecture initialization function. - // Reference: - // http://sourceware.org/binutils/docs/as/Pseudo-Ops.html#Pseudo-Ops - var directives = { - ".abort" : "builtin", - ".align" : "builtin", - ".altmacro" : "builtin", - ".ascii" : "builtin", - ".asciz" : "builtin", - ".balign" : "builtin", - ".balignw" : "builtin", - ".balignl" : "builtin", - ".bundle_align_mode" : "builtin", - ".bundle_lock" : "builtin", - ".bundle_unlock" : "builtin", - ".byte" : "builtin", - ".cfi_startproc" : "builtin", - ".comm" : "builtin", - ".data" : "builtin", - ".def" : "builtin", - ".desc" : "builtin", - ".dim" : "builtin", - ".double" : "builtin", - ".eject" : "builtin", - ".else" : "builtin", - ".elseif" : "builtin", - ".end" : "builtin", - ".endef" : "builtin", - ".endfunc" : "builtin", - ".endif" : "builtin", - ".equ" : "builtin", - ".equiv" : "builtin", - ".eqv" : "builtin", - ".err" : "builtin", - ".error" : "builtin", - ".exitm" : "builtin", - ".extern" : "builtin", - ".fail" : "builtin", - ".file" : "builtin", - ".fill" : "builtin", - ".float" : "builtin", - ".func" : "builtin", - ".global" : "builtin", - ".gnu_attribute" : "builtin", - ".hidden" : "builtin", - ".hword" : "builtin", - ".ident" : "builtin", - ".if" : "builtin", - ".incbin" : "builtin", - ".include" : "builtin", - ".int" : "builtin", - ".internal" : "builtin", - ".irp" : "builtin", - ".irpc" : "builtin", - ".lcomm" : "builtin", - ".lflags" : "builtin", - ".line" : "builtin", - ".linkonce" : "builtin", - ".list" : "builtin", - ".ln" : "builtin", - ".loc" : "builtin", - ".loc_mark_labels" : "builtin", - ".local" : "builtin", - ".long" : "builtin", - ".macro" : "builtin", - ".mri" : "builtin", - ".noaltmacro" : "builtin", - ".nolist" : "builtin", - ".octa" : "builtin", - ".offset" : "builtin", - ".org" : "builtin", - ".p2align" : "builtin", - ".popsection" : "builtin", - ".previous" : "builtin", - ".print" : "builtin", - ".protected" : "builtin", - ".psize" : "builtin", - ".purgem" : "builtin", - ".pushsection" : "builtin", - ".quad" : "builtin", - ".reloc" : "builtin", - ".rept" : "builtin", - ".sbttl" : "builtin", - ".scl" : "builtin", - ".section" : "builtin", - ".set" : "builtin", - ".short" : "builtin", - ".single" : "builtin", - ".size" : "builtin", - ".skip" : "builtin", - ".sleb128" : "builtin", - ".space" : "builtin", - ".stab" : "builtin", - ".string" : "builtin", - ".struct" : "builtin", - ".subsection" : "builtin", - ".symver" : "builtin", - ".tag" : "builtin", - ".text" : "builtin", - ".title" : "builtin", - ".type" : "builtin", - ".uleb128" : "builtin", - ".val" : "builtin", - ".version" : "builtin", - ".vtable_entry" : "builtin", - ".vtable_inherit" : "builtin", - ".warning" : "builtin", - ".weak" : "builtin", - ".weakref" : "builtin", - ".word" : "builtin" - }; - - var registers = {}; - - function x86(_parserConfig) { - lineCommentStartSymbol = "#"; - - registers.ax = "variable"; - registers.eax = "variable-2"; - registers.rax = "variable-3"; - - registers.bx = "variable"; - registers.ebx = "variable-2"; - registers.rbx = "variable-3"; - - registers.cx = "variable"; - registers.ecx = "variable-2"; - registers.rcx = "variable-3"; - - registers.dx = "variable"; - registers.edx = "variable-2"; - registers.rdx = "variable-3"; - - registers.si = "variable"; - registers.esi = "variable-2"; - registers.rsi = "variable-3"; - - registers.di = "variable"; - registers.edi = "variable-2"; - registers.rdi = "variable-3"; - - registers.sp = "variable"; - registers.esp = "variable-2"; - registers.rsp = "variable-3"; - - registers.bp = "variable"; - registers.ebp = "variable-2"; - registers.rbp = "variable-3"; - - registers.ip = "variable"; - registers.eip = "variable-2"; - registers.rip = "variable-3"; - - registers.cs = "keyword"; - registers.ds = "keyword"; - registers.ss = "keyword"; - registers.es = "keyword"; - registers.fs = "keyword"; - registers.gs = "keyword"; - } - - function armv6(_parserConfig) { - // Reference: - // http://infocenter.arm.com/help/topic/com.arm.doc.qrc0001l/QRC0001_UAL.pdf - // http://infocenter.arm.com/help/topic/com.arm.doc.ddi0301h/DDI0301H_arm1176jzfs_r0p7_trm.pdf - lineCommentStartSymbol = "@"; - directives.syntax = "builtin"; - - registers.r0 = "variable"; - registers.r1 = "variable"; - registers.r2 = "variable"; - registers.r3 = "variable"; - registers.r4 = "variable"; - registers.r5 = "variable"; - registers.r6 = "variable"; - registers.r7 = "variable"; - registers.r8 = "variable"; - registers.r9 = "variable"; - registers.r10 = "variable"; - registers.r11 = "variable"; - registers.r12 = "variable"; - - registers.sp = "variable-2"; - registers.lr = "variable-2"; - registers.pc = "variable-2"; - registers.r13 = registers.sp; - registers.r14 = registers.lr; - registers.r15 = registers.pc; - - custom.push(function(ch, stream) { - if (ch === '#') { - stream.eatWhile(/\w/); - return "number"; - } - }); - } - - var arch = parserConfig.architecture.toLowerCase(); - if (arch === "x86") { - x86(parserConfig); - } else if (arch === "arm" || arch === "armv6") { - armv6(parserConfig); - } - - function nextUntilUnescaped(stream, end) { - var escaped = false, next; - while ((next = stream.next()) != null) { - if (next === end && !escaped) { - return false; - } - escaped = !escaped && next === "\\"; - } - return escaped; - } - - function clikeComment(stream, state) { - var maybeEnd = false, ch; - while ((ch = stream.next()) != null) { - if (ch === "/" && maybeEnd) { - state.tokenize = null; - break; - } - maybeEnd = (ch === "*"); - } - return "comment"; - } - - return { - startState: function() { - return { - tokenize: null - }; - }, - - token: function(stream, state) { - if (state.tokenize) { - return state.tokenize(stream, state); - } - - if (stream.eatSpace()) { - return null; - } - - var style, cur, ch = stream.next(); - - if (ch === "/") { - if (stream.eat("*")) { - state.tokenize = clikeComment; - return clikeComment(stream, state); - } - } - - if (ch === lineCommentStartSymbol) { - stream.skipToEnd(); - return "comment"; - } - - if (ch === '"') { - nextUntilUnescaped(stream, '"'); - return "string"; - } - - if (ch === '.') { - stream.eatWhile(/\w/); - cur = stream.current().toLowerCase(); - style = directives[cur]; - return style || null; - } - - if (ch === '=') { - stream.eatWhile(/\w/); - return "tag"; - } - - if (ch === '{') { - return "braket"; - } - - if (ch === '}') { - return "braket"; - } - - if (/\d/.test(ch)) { - if (ch === "0" && stream.eat("x")) { - stream.eatWhile(/[0-9a-fA-F]/); - return "number"; - } - stream.eatWhile(/\d/); - return "number"; - } - - if (/\w/.test(ch)) { - stream.eatWhile(/\w/); - if (stream.eat(":")) { - return 'tag'; - } - cur = stream.current().toLowerCase(); - style = registers[cur]; - return style || null; - } - - for (var i = 0; i < custom.length; i++) { - style = custom[i](ch, stream, state); - if (style) { - return style; - } - } - }, - - lineComment: lineCommentStartSymbol, - blockCommentStart: "/*", - blockCommentEnd: "*/" - }; -}); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gas/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/gas/index.html deleted file mode 100644 index 7684bc18df..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gas/index.html +++ /dev/null @@ -1,57 +0,0 @@ - - - - - CodeMirror: Gas mode - - - - - - - -

CodeMirror: Gas mode

- -
- -
- - - -

Handles AT&T assembler syntax (more specifically this handles - the GNU Assembler (gas) syntax.) - It takes a single optional configuration parameter: - architecture, which can be one of "ARM", - "ARMv6" or "x86". - Including the parameter adds syntax for the registers and special - directives for the supplied architecture. - -

MIME types defined: text/x-gas

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/gfm.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/gfm.js deleted file mode 100644 index 1179b53dc6..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/gfm.js +++ /dev/null @@ -1,96 +0,0 @@ -CodeMirror.defineMode("gfm", function(config) { - var codeDepth = 0; - function blankLine(state) { - state.code = false; - return null; - } - var gfmOverlay = { - startState: function() { - return { - code: false, - codeBlock: false, - ateSpace: false - }; - }, - copyState: function(s) { - return { - code: s.code, - codeBlock: s.codeBlock, - ateSpace: s.ateSpace - }; - }, - token: function(stream, state) { - // Hack to prevent formatting override inside code blocks (block and inline) - if (state.codeBlock) { - if (stream.match(/^```/)) { - state.codeBlock = false; - return null; - } - stream.skipToEnd(); - return null; - } - if (stream.sol()) { - state.code = false; - } - if (stream.sol() && stream.match(/^```/)) { - stream.skipToEnd(); - state.codeBlock = true; - return null; - } - // If this block is changed, it may need to be updated in Markdown mode - if (stream.peek() === '`') { - stream.next(); - var before = stream.pos; - stream.eatWhile('`'); - var difference = 1 + stream.pos - before; - if (!state.code) { - codeDepth = difference; - state.code = true; - } else { - if (difference === codeDepth) { // Must be exact - state.code = false; - } - } - return null; - } else if (state.code) { - stream.next(); - return null; - } - // Check if space. If so, links can be formatted later on - if (stream.eatSpace()) { - state.ateSpace = true; - return null; - } - if (stream.sol() || state.ateSpace) { - state.ateSpace = false; - if(stream.match(/^(?:[a-zA-Z0-9\-_]+\/)?(?:[a-zA-Z0-9\-_]+@)?(?:[a-f0-9]{7,40}\b)/)) { - // User/Project@SHA - // User@SHA - // SHA - return "link"; - } else if (stream.match(/^(?:[a-zA-Z0-9\-_]+\/)?(?:[a-zA-Z0-9\-_]+)?#[0-9]+\b/)) { - // User/Project#Num - // User#Num - // #Num - return "link"; - } - } - if (stream.match(/^((?:[a-z][\w-]+:(?:\/{1,3}|[a-z0-9%])|www\d{0,3}[.]|[a-z0-9.\-]+[.][a-z]{2,4}\/)(?:[^\s()<>]+|\([^\s()<>]*\))+(?:\([^\s()<>]*\)|[^\s`!()\[\]{};:'".,<>?«»“”‘’]))/i)) { - // URLs - // Taken from http://daringfireball.net/2010/07/improved_regex_for_matching_urls - // And then (issue #1160) simplified to make it not crash the Chrome Regexp engine - return "link"; - } - stream.next(); - return null; - }, - blankLine: blankLine - }; - CodeMirror.defineMIME("gfmBase", { - name: "markdown", - underscoresBreakWords: false, - taskLists: true, - fencedCodeBlocks: true - }); - return CodeMirror.overlayMode(CodeMirror.getMode(config, "gfmBase"), gfmOverlay); -}, "markdown"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/index.html deleted file mode 100644 index 826a96d2de..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/index.html +++ /dev/null @@ -1,74 +0,0 @@ - - - - - CodeMirror: GFM mode - - - - - - - - - - - - - - - - - -

CodeMirror: GFM mode

- -
- - - -

Optionally depends on other modes for properly highlighted code blocks.

- -

Parsing/Highlighting Tests: normal, verbose.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/test.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/test.js deleted file mode 100644 index 3ccaec5010..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/gfm/test.js +++ /dev/null @@ -1,112 +0,0 @@ -(function() { - var mode = CodeMirror.getMode({tabSize: 4}, "gfm"); - function MT(name) { test.mode(name, mode, Array.prototype.slice.call(arguments, 1)); } - - MT("emInWordAsterisk", - "foo[em *bar*]hello"); - - MT("emInWordUnderscore", - "foo_bar_hello"); - - MT("emStrongUnderscore", - "[strong __][em&strong _foo__][em _] bar"); - - MT("fencedCodeBlocks", - "[comment ```]", - "[comment foo]", - "", - "[comment ```]", - "bar"); - - MT("fencedCodeBlockModeSwitching", - "[comment ```javascript]", - "[variable foo]", - "", - "[comment ```]", - "bar"); - - MT("taskListAsterisk", - "[variable-2 * []] foo]", // Invalid; must have space or x between [] - "[variable-2 * [ ]]bar]", // Invalid; must have space after ] - "[variable-2 * [x]]hello]", // Invalid; must have space after ] - "[variable-2 * ][meta [ ]]][variable-2 [world]]]", // Valid; tests reference style links - " [variable-3 * ][property [x]]][variable-3 foo]"); // Valid; can be nested - - MT("taskListPlus", - "[variable-2 + []] foo]", // Invalid; must have space or x between [] - "[variable-2 + [ ]]bar]", // Invalid; must have space after ] - "[variable-2 + [x]]hello]", // Invalid; must have space after ] - "[variable-2 + ][meta [ ]]][variable-2 [world]]]", // Valid; tests reference style links - " [variable-3 + ][property [x]]][variable-3 foo]"); // Valid; can be nested - - MT("taskListDash", - "[variable-2 - []] foo]", // Invalid; must have space or x between [] - "[variable-2 - [ ]]bar]", // Invalid; must have space after ] - "[variable-2 - [x]]hello]", // Invalid; must have space after ] - "[variable-2 - ][meta [ ]]][variable-2 [world]]]", // Valid; tests reference style links - " [variable-3 - ][property [x]]][variable-3 foo]"); // Valid; can be nested - - MT("taskListNumber", - "[variable-2 1. []] foo]", // Invalid; must have space or x between [] - "[variable-2 2. [ ]]bar]", // Invalid; must have space after ] - "[variable-2 3. [x]]hello]", // Invalid; must have space after ] - "[variable-2 4. ][meta [ ]]][variable-2 [world]]]", // Valid; tests reference style links - " [variable-3 1. ][property [x]]][variable-3 foo]"); // Valid; can be nested - - MT("SHA", - "foo [link be6a8cc1c1ecfe9489fb51e4869af15a13fc2cd2] bar"); - - MT("shortSHA", - "foo [link be6a8cc] bar"); - - MT("tooShortSHA", - "foo be6a8c bar"); - - MT("longSHA", - "foo be6a8cc1c1ecfe9489fb51e4869af15a13fc2cd22 bar"); - - MT("badSHA", - "foo be6a8cc1c1ecfe9489fb51e4869af15a13fc2cg2 bar"); - - MT("userSHA", - "foo [link bar@be6a8cc1c1ecfe9489fb51e4869af15a13fc2cd2] hello"); - - MT("userProjectSHA", - "foo [link bar/hello@be6a8cc1c1ecfe9489fb51e4869af15a13fc2cd2] world"); - - MT("num", - "foo [link #1] bar"); - - MT("badNum", - "foo #1bar hello"); - - MT("userNum", - "foo [link bar#1] hello"); - - MT("userProjectNum", - "foo [link bar/hello#1] world"); - - MT("vanillaLink", - "foo [link http://www.example.com/] bar"); - - MT("vanillaLinkPunctuation", - "foo [link http://www.example.com/]. bar"); - - MT("vanillaLinkExtension", - "foo [link http://www.example.com/index.html] bar"); - - MT("notALink", - "[comment ```css]", - "[tag foo] {[property color][operator :][keyword black];}", - "[comment ```][link http://www.example.com/]"); - - MT("notALink", - "[comment ``foo `bar` http://www.example.com/``] hello"); - - MT("notALink", - "[comment `foo]", - "[link http://www.example.com/]", - "[comment `foo]", - "", - "[link http://www.example.com/]"); -})(); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/go/go.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/go/go.js deleted file mode 100644 index 6a458a6fe9..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/go/go.js +++ /dev/null @@ -1,168 +0,0 @@ -CodeMirror.defineMode("go", function(config) { - var indentUnit = config.indentUnit; - - var keywords = { - "break":true, "case":true, "chan":true, "const":true, "continue":true, - "default":true, "defer":true, "else":true, "fallthrough":true, "for":true, - "func":true, "go":true, "goto":true, "if":true, "import":true, - "interface":true, "map":true, "package":true, "range":true, "return":true, - "select":true, "struct":true, "switch":true, "type":true, "var":true, - "bool":true, "byte":true, "complex64":true, "complex128":true, - "float32":true, "float64":true, "int8":true, "int16":true, "int32":true, - "int64":true, "string":true, "uint8":true, "uint16":true, "uint32":true, - "uint64":true, "int":true, "uint":true, "uintptr":true - }; - - var atoms = { - "true":true, "false":true, "iota":true, "nil":true, "append":true, - "cap":true, "close":true, "complex":true, "copy":true, "imag":true, - "len":true, "make":true, "new":true, "panic":true, "print":true, - "println":true, "real":true, "recover":true - }; - - var isOperatorChar = /[+\-*&^%:=<>!|\/]/; - - var curPunc; - - function tokenBase(stream, state) { - var ch = stream.next(); - if (ch == '"' || ch == "'" || ch == "`") { - state.tokenize = tokenString(ch); - return state.tokenize(stream, state); - } - if (/[\d\.]/.test(ch)) { - if (ch == ".") { - stream.match(/^[0-9]+([eE][\-+]?[0-9]+)?/); - } else if (ch == "0") { - stream.match(/^[xX][0-9a-fA-F]+/) || stream.match(/^0[0-7]+/); - } else { - stream.match(/^[0-9]*\.?[0-9]*([eE][\-+]?[0-9]+)?/); - } - return "number"; - } - if (/[\[\]{}\(\),;\:\.]/.test(ch)) { - curPunc = ch; - return null; - } - if (ch == "/") { - if (stream.eat("*")) { - state.tokenize = tokenComment; - return tokenComment(stream, state); - } - if (stream.eat("/")) { - stream.skipToEnd(); - return "comment"; - } - } - if (isOperatorChar.test(ch)) { - stream.eatWhile(isOperatorChar); - return "operator"; - } - stream.eatWhile(/[\w\$_]/); - var cur = stream.current(); - if (keywords.propertyIsEnumerable(cur)) { - if (cur == "case" || cur == "default") curPunc = "case"; - return "keyword"; - } - if (atoms.propertyIsEnumerable(cur)) return "atom"; - return "variable"; - } - - function tokenString(quote) { - return function(stream, state) { - var escaped = false, next, end = false; - while ((next = stream.next()) != null) { - if (next == quote && !escaped) {end = true; break;} - escaped = !escaped && next == "\\"; - } - if (end || !(escaped || quote == "`")) - state.tokenize = tokenBase; - return "string"; - }; - } - - function tokenComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == "/" && maybeEnd) { - state.tokenize = tokenBase; - break; - } - maybeEnd = (ch == "*"); - } - return "comment"; - } - - function Context(indented, column, type, align, prev) { - this.indented = indented; - this.column = column; - this.type = type; - this.align = align; - this.prev = prev; - } - function pushContext(state, col, type) { - return state.context = new Context(state.indented, col, type, null, state.context); - } - function popContext(state) { - var t = state.context.type; - if (t == ")" || t == "]" || t == "}") - state.indented = state.context.indented; - return state.context = state.context.prev; - } - - // Interface - - return { - startState: function(basecolumn) { - return { - tokenize: null, - context: new Context((basecolumn || 0) - indentUnit, 0, "top", false), - indented: 0, - startOfLine: true - }; - }, - - token: function(stream, state) { - var ctx = state.context; - if (stream.sol()) { - if (ctx.align == null) ctx.align = false; - state.indented = stream.indentation(); - state.startOfLine = true; - if (ctx.type == "case") ctx.type = "}"; - } - if (stream.eatSpace()) return null; - curPunc = null; - var style = (state.tokenize || tokenBase)(stream, state); - if (style == "comment") return style; - if (ctx.align == null) ctx.align = true; - - if (curPunc == "{") pushContext(state, stream.column(), "}"); - else if (curPunc == "[") pushContext(state, stream.column(), "]"); - else if (curPunc == "(") pushContext(state, stream.column(), ")"); - else if (curPunc == "case") ctx.type = "case"; - else if (curPunc == "}" && ctx.type == "}") ctx = popContext(state); - else if (curPunc == ctx.type) popContext(state); - state.startOfLine = false; - return style; - }, - - indent: function(state, textAfter) { - if (state.tokenize != tokenBase && state.tokenize != null) return 0; - var ctx = state.context, firstChar = textAfter && textAfter.charAt(0); - if (ctx.type == "case" && /^(?:case|default)\b/.test(textAfter)) { - state.context.type = "}"; - return ctx.indented; - } - var closing = firstChar == ctx.type; - if (ctx.align) return ctx.column + (closing ? 0 : 1); - else return ctx.indented + (closing ? 0 : indentUnit); - }, - - electricChars: "{}:", - blockCommentStart: "/*", - blockCommentEnd: "*/", - lineComment: "//" - }; -}); - -CodeMirror.defineMIME("text/x-go", "go"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/go/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/go/index.html deleted file mode 100644 index 8a6aafca29..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/go/index.html +++ /dev/null @@ -1,74 +0,0 @@ - - - - - CodeMirror: Go mode - - - - - - - - - -

CodeMirror: Go mode

- -
- - - -

MIME type: text/x-go

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/groovy/groovy.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/groovy/groovy.js deleted file mode 100644 index 92b948192e..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/groovy/groovy.js +++ /dev/null @@ -1,210 +0,0 @@ -CodeMirror.defineMode("groovy", function(config) { - function words(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - var keywords = words( - "abstract as assert boolean break byte case catch char class const continue def default " + - "do double else enum extends final finally float for goto if implements import in " + - "instanceof int interface long native new package private protected public return " + - "short static strictfp super switch synchronized threadsafe throw throws transient " + - "try void volatile while"); - var blockKeywords = words("catch class do else finally for if switch try while enum interface def"); - var atoms = words("null true false this"); - - var curPunc; - function tokenBase(stream, state) { - var ch = stream.next(); - if (ch == '"' || ch == "'") { - return startString(ch, stream, state); - } - if (/[\[\]{}\(\),;\:\.]/.test(ch)) { - curPunc = ch; - return null; - } - if (/\d/.test(ch)) { - stream.eatWhile(/[\w\.]/); - if (stream.eat(/eE/)) { stream.eat(/\+\-/); stream.eatWhile(/\d/); } - return "number"; - } - if (ch == "/") { - if (stream.eat("*")) { - state.tokenize.push(tokenComment); - return tokenComment(stream, state); - } - if (stream.eat("/")) { - stream.skipToEnd(); - return "comment"; - } - if (expectExpression(state.lastToken)) { - return startString(ch, stream, state); - } - } - if (ch == "-" && stream.eat(">")) { - curPunc = "->"; - return null; - } - if (/[+\-*&%=<>!?|\/~]/.test(ch)) { - stream.eatWhile(/[+\-*&%=<>|~]/); - return "operator"; - } - stream.eatWhile(/[\w\$_]/); - if (ch == "@") { stream.eatWhile(/[\w\$_\.]/); return "meta"; } - if (state.lastToken == ".") return "property"; - if (stream.eat(":")) { curPunc = "proplabel"; return "property"; } - var cur = stream.current(); - if (atoms.propertyIsEnumerable(cur)) { return "atom"; } - if (keywords.propertyIsEnumerable(cur)) { - if (blockKeywords.propertyIsEnumerable(cur)) curPunc = "newstatement"; - return "keyword"; - } - return "variable"; - } - tokenBase.isBase = true; - - function startString(quote, stream, state) { - var tripleQuoted = false; - if (quote != "/" && stream.eat(quote)) { - if (stream.eat(quote)) tripleQuoted = true; - else return "string"; - } - function t(stream, state) { - var escaped = false, next, end = !tripleQuoted; - while ((next = stream.next()) != null) { - if (next == quote && !escaped) { - if (!tripleQuoted) { break; } - if (stream.match(quote + quote)) { end = true; break; } - } - if (quote == '"' && next == "$" && !escaped && stream.eat("{")) { - state.tokenize.push(tokenBaseUntilBrace()); - return "string"; - } - escaped = !escaped && next == "\\"; - } - if (end) state.tokenize.pop(); - return "string"; - } - state.tokenize.push(t); - return t(stream, state); - } - - function tokenBaseUntilBrace() { - var depth = 1; - function t(stream, state) { - if (stream.peek() == "}") { - depth--; - if (depth == 0) { - state.tokenize.pop(); - return state.tokenize[state.tokenize.length-1](stream, state); - } - } else if (stream.peek() == "{") { - depth++; - } - return tokenBase(stream, state); - } - t.isBase = true; - return t; - } - - function tokenComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == "/" && maybeEnd) { - state.tokenize.pop(); - break; - } - maybeEnd = (ch == "*"); - } - return "comment"; - } - - function expectExpression(last) { - return !last || last == "operator" || last == "->" || /[\.\[\{\(,;:]/.test(last) || - last == "newstatement" || last == "keyword" || last == "proplabel"; - } - - function Context(indented, column, type, align, prev) { - this.indented = indented; - this.column = column; - this.type = type; - this.align = align; - this.prev = prev; - } - function pushContext(state, col, type) { - return state.context = new Context(state.indented, col, type, null, state.context); - } - function popContext(state) { - var t = state.context.type; - if (t == ")" || t == "]" || t == "}") - state.indented = state.context.indented; - return state.context = state.context.prev; - } - - // Interface - - return { - startState: function(basecolumn) { - return { - tokenize: [tokenBase], - context: new Context((basecolumn || 0) - config.indentUnit, 0, "top", false), - indented: 0, - startOfLine: true, - lastToken: null - }; - }, - - token: function(stream, state) { - var ctx = state.context; - if (stream.sol()) { - if (ctx.align == null) ctx.align = false; - state.indented = stream.indentation(); - state.startOfLine = true; - // Automatic semicolon insertion - if (ctx.type == "statement" && !expectExpression(state.lastToken)) { - popContext(state); ctx = state.context; - } - } - if (stream.eatSpace()) return null; - curPunc = null; - var style = state.tokenize[state.tokenize.length-1](stream, state); - if (style == "comment") return style; - if (ctx.align == null) ctx.align = true; - - if ((curPunc == ";" || curPunc == ":") && ctx.type == "statement") popContext(state); - // Handle indentation for {x -> \n ... } - else if (curPunc == "->" && ctx.type == "statement" && ctx.prev.type == "}") { - popContext(state); - state.context.align = false; - } - else if (curPunc == "{") pushContext(state, stream.column(), "}"); - else if (curPunc == "[") pushContext(state, stream.column(), "]"); - else if (curPunc == "(") pushContext(state, stream.column(), ")"); - else if (curPunc == "}") { - while (ctx.type == "statement") ctx = popContext(state); - if (ctx.type == "}") ctx = popContext(state); - while (ctx.type == "statement") ctx = popContext(state); - } - else if (curPunc == ctx.type) popContext(state); - else if (ctx.type == "}" || ctx.type == "top" || (ctx.type == "statement" && curPunc == "newstatement")) - pushContext(state, stream.column(), "statement"); - state.startOfLine = false; - state.lastToken = curPunc || style; - return style; - }, - - indent: function(state, textAfter) { - if (!state.tokenize[state.tokenize.length-1].isBase) return 0; - var firstChar = textAfter && textAfter.charAt(0), ctx = state.context; - if (ctx.type == "statement" && !expectExpression(state.lastToken)) ctx = ctx.prev; - var closing = firstChar == ctx.type; - if (ctx.type == "statement") return ctx.indented + (firstChar == "{" ? 0 : config.indentUnit); - else if (ctx.align) return ctx.column + (closing ? 0 : 1); - else return ctx.indented + (closing ? 0 : config.indentUnit); - }, - - electricChars: "{}" - }; -}); - -CodeMirror.defineMIME("text/x-groovy", "groovy"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/groovy/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/groovy/index.html deleted file mode 100644 index 3d39595774..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/groovy/index.html +++ /dev/null @@ -1,73 +0,0 @@ - - - - - CodeMirror: Groovy mode - - - - - - - - -

CodeMirror: Groovy mode

- -
- - - -

MIME types defined: text/x-groovy

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/haml.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/haml.js deleted file mode 100644 index 793308f6fb..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/haml.js +++ /dev/null @@ -1,153 +0,0 @@ -(function() { - "use strict"; - - // full haml mode. This handled embeded ruby and html fragments too - CodeMirror.defineMode("haml", function(config) { - var htmlMode = CodeMirror.getMode(config, {name: "htmlmixed"}); - var rubyMode = CodeMirror.getMode(config, "ruby"); - - function rubyInQuote(endQuote) { - return function(stream, state) { - var ch = stream.peek(); - if (ch == endQuote && state.rubyState.tokenize.length == 1) { - // step out of ruby context as it seems to complete processing all the braces - stream.next(); - state.tokenize = html; - return "closeAttributeTag"; - } else { - return ruby(stream, state); - } - }; - } - - function ruby(stream, state) { - if (stream.match("-#")) { - stream.skipToEnd(); - return "comment"; - } - return rubyMode.token(stream, state.rubyState); - } - - function html(stream, state) { - var ch = stream.peek(); - - // handle haml declarations. All declarations that cant be handled here - // will be passed to html mode - if (state.previousToken.style == "comment" ) { - if (state.indented > state.previousToken.indented) { - stream.skipToEnd(); - return "commentLine"; - } - } - - if (state.startOfLine) { - if (ch == "!" && stream.match("!!")) { - stream.skipToEnd(); - return "tag"; - } else if (stream.match(/^%[\w:#\.]+=/)) { - state.tokenize = ruby; - return "hamlTag"; - } else if (stream.match(/^%[\w:]+/)) { - return "hamlTag"; - } else if (ch == "/" ) { - stream.skipToEnd(); - return "comment"; - } - } - - if (state.startOfLine || state.previousToken.style == "hamlTag") { - if ( ch == "#" || ch == ".") { - stream.match(/[\w-#\.]*/); - return "hamlAttribute"; - } - } - - // donot handle --> as valid ruby, make it HTML close comment instead - if (state.startOfLine && !stream.match("-->", false) && (ch == "=" || ch == "-" )) { - state.tokenize = ruby; - return null; - } - - if (state.previousToken.style == "hamlTag" || - state.previousToken.style == "closeAttributeTag" || - state.previousToken.style == "hamlAttribute") { - if (ch == "(") { - state.tokenize = rubyInQuote(")"); - return null; - } else if (ch == "{") { - state.tokenize = rubyInQuote("}"); - return null; - } - } - - return htmlMode.token(stream, state.htmlState); - } - - return { - // default to html mode - startState: function() { - var htmlState = htmlMode.startState(); - var rubyState = rubyMode.startState(); - return { - htmlState: htmlState, - rubyState: rubyState, - indented: 0, - previousToken: { style: null, indented: 0}, - tokenize: html - }; - }, - - copyState: function(state) { - return { - htmlState : CodeMirror.copyState(htmlMode, state.htmlState), - rubyState: CodeMirror.copyState(rubyMode, state.rubyState), - indented: state.indented, - previousToken: state.previousToken, - tokenize: state.tokenize - }; - }, - - token: function(stream, state) { - if (stream.sol()) { - state.indented = stream.indentation(); - state.startOfLine = true; - } - if (stream.eatSpace()) return null; - var style = state.tokenize(stream, state); - state.startOfLine = false; - // dont record comment line as we only want to measure comment line with - // the opening comment block - if (style && style != "commentLine") { - state.previousToken = { style: style, indented: state.indented }; - } - // if current state is ruby and the previous token is not `,` reset the - // tokenize to html - if (stream.eol() && state.tokenize == ruby) { - stream.backUp(1); - var ch = stream.peek(); - stream.next(); - if (ch && ch != ",") { - state.tokenize = html; - } - } - // reprocess some of the specific style tag when finish setting previousToken - if (style == "hamlTag") { - style = "tag"; - } else if (style == "commentLine") { - style = "comment"; - } else if (style == "hamlAttribute") { - style = "attribute"; - } else if (style == "closeAttributeTag") { - style = null; - } - return style; - }, - - indent: function(state) { - return state.indented; - } - }; - }, "htmlmixed", "ruby"); - - CodeMirror.defineMIME("text/x-haml", "haml"); -})(); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/index.html deleted file mode 100644 index 7da378fc80..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/index.html +++ /dev/null @@ -1,67 +0,0 @@ - - - - - CodeMirror: HAML mode - - - - - - - - - - - -

CodeMirror: HAML mode

-
- - -

MIME types defined: text/x-haml.

- -

Parsing/Highlighting Tests: normal, verbose.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/test.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/test.js deleted file mode 100644 index b7178d40fb..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haml/test.js +++ /dev/null @@ -1,94 +0,0 @@ -(function() { - var mode = CodeMirror.getMode({tabSize: 4}, "haml"); - function MT(name) { test.mode(name, mode, Array.prototype.slice.call(arguments, 1)); } - - // Requires at least one media query - MT("elementName", - "[tag %h1] Hey There"); - - MT("oneElementPerLine", - "[tag %h1] Hey There %h2"); - - MT("idSelector", - "[tag %h1][attribute #test] Hey There"); - - MT("classSelector", - "[tag %h1][attribute .hello] Hey There"); - - MT("docType", - "[tag !!! XML]"); - - MT("comment", - "[comment / Hello WORLD]"); - - MT("notComment", - "[tag %h1] This is not a / comment "); - - MT("attributes", - "[tag %a]([variable title][operator =][string \"test\"]){[atom :title] [operator =>] [string \"test\"]}"); - - MT("htmlCode", - "[tag

]Title[tag

]"); - - MT("rubyBlock", - "[operator =][variable-2 @item]"); - - MT("selectorRubyBlock", - "[tag %a.selector=] [variable-2 @item]"); - - MT("nestedRubyBlock", - "[tag %a]", - " [operator =][variable puts] [string \"test\"]"); - - MT("multilinePlaintext", - "[tag %p]", - " Hello,", - " World"); - - MT("multilineRuby", - "[tag %p]", - " [comment -# this is a comment]", - " [comment and this is a comment too]", - " Date/Time", - " [operator -] [variable now] [operator =] [tag DateTime][operator .][variable now]", - " [tag %strong=] [variable now]", - " [operator -] [keyword if] [variable now] [operator >] [tag DateTime][operator .][variable parse]([string \"December 31, 2006\"])", - " [operator =][string \"Happy\"]", - " [operator =][string \"Belated\"]", - " [operator =][string \"Birthday\"]"); - - MT("multilineComment", - "[comment /]", - " [comment Multiline]", - " [comment Comment]"); - - MT("hamlComment", - "[comment -# this is a comment]"); - - MT("multilineHamlComment", - "[comment -# this is a comment]", - " [comment and this is a comment too]"); - - MT("multilineHTMLComment", - "[comment ]"); - - MT("hamlAfterRubyTag", - "[attribute .block]", - " [tag %strong=] [variable now]", - " [attribute .test]", - " [operator =][variable now]", - " [attribute .right]"); - - MT("stretchedRuby", - "[operator =] [variable puts] [string \"Hello\"],", - " [string \"World\"]"); - - MT("interpolationInHashAttribute", - //"[tag %div]{[atom :id] [operator =>] [string \"#{][variable test][string }_#{][variable ting][string }\"]} test"); - "[tag %div]{[atom :id] [operator =>] [string \"#{][variable test][string }_#{][variable ting][string }\"]} test"); - - MT("interpolationInHTMLAttribute", - "[tag %div]([variable title][operator =][string \"#{][variable test][string }_#{][variable ting]()[string }\"]) Test"); -})(); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haskell/haskell.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/haskell/haskell.js deleted file mode 100644 index b18d5ced1a..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haskell/haskell.js +++ /dev/null @@ -1,246 +0,0 @@ -CodeMirror.defineMode("haskell", function() { - - function switchState(source, setState, f) { - setState(f); - return f(source, setState); - } - - // These should all be Unicode extended, as per the Haskell 2010 report - var smallRE = /[a-z_]/; - var largeRE = /[A-Z]/; - var digitRE = /[0-9]/; - var hexitRE = /[0-9A-Fa-f]/; - var octitRE = /[0-7]/; - var idRE = /[a-z_A-Z0-9']/; - var symbolRE = /[-!#$%&*+.\/<=>?@\\^|~:]/; - var specialRE = /[(),;[\]`{}]/; - var whiteCharRE = /[ \t\v\f]/; // newlines are handled in tokenizer - - function normal(source, setState) { - if (source.eatWhile(whiteCharRE)) { - return null; - } - - var ch = source.next(); - if (specialRE.test(ch)) { - if (ch == '{' && source.eat('-')) { - var t = "comment"; - if (source.eat('#')) { - t = "meta"; - } - return switchState(source, setState, ncomment(t, 1)); - } - return null; - } - - if (ch == '\'') { - if (source.eat('\\')) { - source.next(); // should handle other escapes here - } - else { - source.next(); - } - if (source.eat('\'')) { - return "string"; - } - return "error"; - } - - if (ch == '"') { - return switchState(source, setState, stringLiteral); - } - - if (largeRE.test(ch)) { - source.eatWhile(idRE); - if (source.eat('.')) { - return "qualifier"; - } - return "variable-2"; - } - - if (smallRE.test(ch)) { - source.eatWhile(idRE); - return "variable"; - } - - if (digitRE.test(ch)) { - if (ch == '0') { - if (source.eat(/[xX]/)) { - source.eatWhile(hexitRE); // should require at least 1 - return "integer"; - } - if (source.eat(/[oO]/)) { - source.eatWhile(octitRE); // should require at least 1 - return "number"; - } - } - source.eatWhile(digitRE); - var t = "number"; - if (source.eat('.')) { - t = "number"; - source.eatWhile(digitRE); // should require at least 1 - } - if (source.eat(/[eE]/)) { - t = "number"; - source.eat(/[-+]/); - source.eatWhile(digitRE); // should require at least 1 - } - return t; - } - - if (symbolRE.test(ch)) { - if (ch == '-' && source.eat(/-/)) { - source.eatWhile(/-/); - if (!source.eat(symbolRE)) { - source.skipToEnd(); - return "comment"; - } - } - var t = "variable"; - if (ch == ':') { - t = "variable-2"; - } - source.eatWhile(symbolRE); - return t; - } - - return "error"; - } - - function ncomment(type, nest) { - if (nest == 0) { - return normal; - } - return function(source, setState) { - var currNest = nest; - while (!source.eol()) { - var ch = source.next(); - if (ch == '{' && source.eat('-')) { - ++currNest; - } - else if (ch == '-' && source.eat('}')) { - --currNest; - if (currNest == 0) { - setState(normal); - return type; - } - } - } - setState(ncomment(type, currNest)); - return type; - }; - } - - function stringLiteral(source, setState) { - while (!source.eol()) { - var ch = source.next(); - if (ch == '"') { - setState(normal); - return "string"; - } - if (ch == '\\') { - if (source.eol() || source.eat(whiteCharRE)) { - setState(stringGap); - return "string"; - } - if (source.eat('&')) { - } - else { - source.next(); // should handle other escapes here - } - } - } - setState(normal); - return "error"; - } - - function stringGap(source, setState) { - if (source.eat('\\')) { - return switchState(source, setState, stringLiteral); - } - source.next(); - setState(normal); - return "error"; - } - - - var wellKnownWords = (function() { - var wkw = {}; - function setType(t) { - return function () { - for (var i = 0; i < arguments.length; i++) - wkw[arguments[i]] = t; - }; - } - - setType("keyword")( - "case", "class", "data", "default", "deriving", "do", "else", "foreign", - "if", "import", "in", "infix", "infixl", "infixr", "instance", "let", - "module", "newtype", "of", "then", "type", "where", "_"); - - setType("keyword")( - "\.\.", ":", "::", "=", "\\", "\"", "<-", "->", "@", "~", "=>"); - - setType("builtin")( - "!!", "$!", "$", "&&", "+", "++", "-", ".", "/", "/=", "<", "<=", "=<<", - "==", ">", ">=", ">>", ">>=", "^", "^^", "||", "*", "**"); - - setType("builtin")( - "Bool", "Bounded", "Char", "Double", "EQ", "Either", "Enum", "Eq", - "False", "FilePath", "Float", "Floating", "Fractional", "Functor", "GT", - "IO", "IOError", "Int", "Integer", "Integral", "Just", "LT", "Left", - "Maybe", "Monad", "Nothing", "Num", "Ord", "Ordering", "Rational", "Read", - "ReadS", "Real", "RealFloat", "RealFrac", "Right", "Show", "ShowS", - "String", "True"); - - setType("builtin")( - "abs", "acos", "acosh", "all", "and", "any", "appendFile", "asTypeOf", - "asin", "asinh", "atan", "atan2", "atanh", "break", "catch", "ceiling", - "compare", "concat", "concatMap", "const", "cos", "cosh", "curry", - "cycle", "decodeFloat", "div", "divMod", "drop", "dropWhile", "either", - "elem", "encodeFloat", "enumFrom", "enumFromThen", "enumFromThenTo", - "enumFromTo", "error", "even", "exp", "exponent", "fail", "filter", - "flip", "floatDigits", "floatRadix", "floatRange", "floor", "fmap", - "foldl", "foldl1", "foldr", "foldr1", "fromEnum", "fromInteger", - "fromIntegral", "fromRational", "fst", "gcd", "getChar", "getContents", - "getLine", "head", "id", "init", "interact", "ioError", "isDenormalized", - "isIEEE", "isInfinite", "isNaN", "isNegativeZero", "iterate", "last", - "lcm", "length", "lex", "lines", "log", "logBase", "lookup", "map", - "mapM", "mapM_", "max", "maxBound", "maximum", "maybe", "min", "minBound", - "minimum", "mod", "negate", "not", "notElem", "null", "odd", "or", - "otherwise", "pi", "pred", "print", "product", "properFraction", - "putChar", "putStr", "putStrLn", "quot", "quotRem", "read", "readFile", - "readIO", "readList", "readLn", "readParen", "reads", "readsPrec", - "realToFrac", "recip", "rem", "repeat", "replicate", "return", "reverse", - "round", "scaleFloat", "scanl", "scanl1", "scanr", "scanr1", "seq", - "sequence", "sequence_", "show", "showChar", "showList", "showParen", - "showString", "shows", "showsPrec", "significand", "signum", "sin", - "sinh", "snd", "span", "splitAt", "sqrt", "subtract", "succ", "sum", - "tail", "take", "takeWhile", "tan", "tanh", "toEnum", "toInteger", - "toRational", "truncate", "uncurry", "undefined", "unlines", "until", - "unwords", "unzip", "unzip3", "userError", "words", "writeFile", "zip", - "zip3", "zipWith", "zipWith3"); - - return wkw; - })(); - - - - return { - startState: function () { return { f: normal }; }, - copyState: function (s) { return { f: s.f }; }, - - token: function(stream, state) { - var t = state.f(stream, function(s) { state.f = s; }); - var w = stream.current(); - return (w in wellKnownWords) ? wellKnownWords[w] : t; - }, - - blockCommentStart: "{-", - blockCommentEnd: "-}", - lineComment: "--" - }; - -}); - -CodeMirror.defineMIME("text/x-haskell", "haskell"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haskell/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/haskell/index.html deleted file mode 100644 index 56307b8a95..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haskell/index.html +++ /dev/null @@ -1,62 +0,0 @@ - - - - - CodeMirror: Haskell mode - - - - - - - - - -

CodeMirror: Haskell mode

- -
- - - -

MIME types defined: text/x-haskell.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haxe/haxe.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/haxe/haxe.js deleted file mode 100644 index 28f9b00ec3..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haxe/haxe.js +++ /dev/null @@ -1,429 +0,0 @@ -CodeMirror.defineMode("haxe", function(config, parserConfig) { - var indentUnit = config.indentUnit; - - // Tokenizer - - var keywords = function(){ - function kw(type) {return {type: type, style: "keyword"};} - var A = kw("keyword a"), B = kw("keyword b"), C = kw("keyword c"); - var operator = kw("operator"), atom = {type: "atom", style: "atom"}, attribute = {type:"attribute", style: "attribute"}; - var type = kw("typedef"); - return { - "if": A, "while": A, "else": B, "do": B, "try": B, - "return": C, "break": C, "continue": C, "new": C, "throw": C, - "var": kw("var"), "inline":attribute, "static": attribute, "using":kw("import"), - "public": attribute, "private": attribute, "cast": kw("cast"), "import": kw("import"), "macro": kw("macro"), - "function": kw("function"), "catch": kw("catch"), "untyped": kw("untyped"), "callback": kw("cb"), - "for": kw("for"), "switch": kw("switch"), "case": kw("case"), "default": kw("default"), - "in": operator, "never": kw("property_access"), "trace":kw("trace"), - "class": type, "enum":type, "interface":type, "typedef":type, "extends":type, "implements":type, "dynamic":type, - "true": atom, "false": atom, "null": atom - }; - }(); - - var isOperatorChar = /[+\-*&%=<>!?|]/; - - function chain(stream, state, f) { - state.tokenize = f; - return f(stream, state); - } - - function nextUntilUnescaped(stream, end) { - var escaped = false, next; - while ((next = stream.next()) != null) { - if (next == end && !escaped) - return false; - escaped = !escaped && next == "\\"; - } - return escaped; - } - - // Used as scratch variables to communicate multiple values without - // consing up tons of objects. - var type, content; - function ret(tp, style, cont) { - type = tp; content = cont; - return style; - } - - function haxeTokenBase(stream, state) { - var ch = stream.next(); - if (ch == '"' || ch == "'") - return chain(stream, state, haxeTokenString(ch)); - else if (/[\[\]{}\(\),;\:\.]/.test(ch)) - return ret(ch); - else if (ch == "0" && stream.eat(/x/i)) { - stream.eatWhile(/[\da-f]/i); - return ret("number", "number"); - } - else if (/\d/.test(ch) || ch == "-" && stream.eat(/\d/)) { - stream.match(/^\d*(?:\.\d*)?(?:[eE][+\-]?\d+)?/); - return ret("number", "number"); - } - else if (state.reAllowed && (ch == "~" && stream.eat(/\//))) { - nextUntilUnescaped(stream, "/"); - stream.eatWhile(/[gimsu]/); - return ret("regexp", "string-2"); - } - else if (ch == "/") { - if (stream.eat("*")) { - return chain(stream, state, haxeTokenComment); - } - else if (stream.eat("/")) { - stream.skipToEnd(); - return ret("comment", "comment"); - } - else { - stream.eatWhile(isOperatorChar); - return ret("operator", null, stream.current()); - } - } - else if (ch == "#") { - stream.skipToEnd(); - return ret("conditional", "meta"); - } - else if (ch == "@") { - stream.eat(/:/); - stream.eatWhile(/[\w_]/); - return ret ("metadata", "meta"); - } - else if (isOperatorChar.test(ch)) { - stream.eatWhile(isOperatorChar); - return ret("operator", null, stream.current()); - } - else { - var word; - if(/[A-Z]/.test(ch)) - { - stream.eatWhile(/[\w_<>]/); - word = stream.current(); - return ret("type", "variable-3", word); - } - else - { - stream.eatWhile(/[\w_]/); - var word = stream.current(), known = keywords.propertyIsEnumerable(word) && keywords[word]; - return (known && state.kwAllowed) ? ret(known.type, known.style, word) : - ret("variable", "variable", word); - } - } - } - - function haxeTokenString(quote) { - return function(stream, state) { - if (!nextUntilUnescaped(stream, quote)) - state.tokenize = haxeTokenBase; - return ret("string", "string"); - }; - } - - function haxeTokenComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == "/" && maybeEnd) { - state.tokenize = haxeTokenBase; - break; - } - maybeEnd = (ch == "*"); - } - return ret("comment", "comment"); - } - - // Parser - - var atomicTypes = {"atom": true, "number": true, "variable": true, "string": true, "regexp": true}; - - function HaxeLexical(indented, column, type, align, prev, info) { - this.indented = indented; - this.column = column; - this.type = type; - this.prev = prev; - this.info = info; - if (align != null) this.align = align; - } - - function inScope(state, varname) { - for (var v = state.localVars; v; v = v.next) - if (v.name == varname) return true; - } - - function parseHaxe(state, style, type, content, stream) { - var cc = state.cc; - // Communicate our context to the combinators. - // (Less wasteful than consing up a hundred closures on every call.) - cx.state = state; cx.stream = stream; cx.marked = null, cx.cc = cc; - - if (!state.lexical.hasOwnProperty("align")) - state.lexical.align = true; - - while(true) { - var combinator = cc.length ? cc.pop() : statement; - if (combinator(type, content)) { - while(cc.length && cc[cc.length - 1].lex) - cc.pop()(); - if (cx.marked) return cx.marked; - if (type == "variable" && inScope(state, content)) return "variable-2"; - if (type == "variable" && imported(state, content)) return "variable-3"; - return style; - } - } - } - - function imported(state, typename) - { - if (/[a-z]/.test(typename.charAt(0))) - return false; - var len = state.importedtypes.length; - for (var i = 0; i= 0; i--) cx.cc.push(arguments[i]); - } - function cont() { - pass.apply(null, arguments); - return true; - } - function register(varname) { - var state = cx.state; - if (state.context) { - cx.marked = "def"; - for (var v = state.localVars; v; v = v.next) - if (v.name == varname) return; - state.localVars = {name: varname, next: state.localVars}; - } - } - - // Combinators - - var defaultVars = {name: "this", next: null}; - function pushcontext() { - if (!cx.state.context) cx.state.localVars = defaultVars; - cx.state.context = {prev: cx.state.context, vars: cx.state.localVars}; - } - function popcontext() { - cx.state.localVars = cx.state.context.vars; - cx.state.context = cx.state.context.prev; - } - function pushlex(type, info) { - var result = function() { - var state = cx.state; - state.lexical = new HaxeLexical(state.indented, cx.stream.column(), type, null, state.lexical, info); - }; - result.lex = true; - return result; - } - function poplex() { - var state = cx.state; - if (state.lexical.prev) { - if (state.lexical.type == ")") - state.indented = state.lexical.indented; - state.lexical = state.lexical.prev; - } - } - poplex.lex = true; - - function expect(wanted) { - return function(type) { - if (type == wanted) return cont(); - else if (wanted == ";") return pass(); - else return cont(arguments.callee); - }; - } - - function statement(type) { - if (type == "@") return cont(metadef); - if (type == "var") return cont(pushlex("vardef"), vardef1, expect(";"), poplex); - if (type == "keyword a") return cont(pushlex("form"), expression, statement, poplex); - if (type == "keyword b") return cont(pushlex("form"), statement, poplex); - if (type == "{") return cont(pushlex("}"), pushcontext, block, poplex, popcontext); - if (type == ";") return cont(); - if (type == "attribute") return cont(maybeattribute); - if (type == "function") return cont(functiondef); - if (type == "for") return cont(pushlex("form"), expect("("), pushlex(")"), forspec1, expect(")"), - poplex, statement, poplex); - if (type == "variable") return cont(pushlex("stat"), maybelabel); - if (type == "switch") return cont(pushlex("form"), expression, pushlex("}", "switch"), expect("{"), - block, poplex, poplex); - if (type == "case") return cont(expression, expect(":")); - if (type == "default") return cont(expect(":")); - if (type == "catch") return cont(pushlex("form"), pushcontext, expect("("), funarg, expect(")"), - statement, poplex, popcontext); - if (type == "import") return cont(importdef, expect(";")); - if (type == "typedef") return cont(typedef); - return pass(pushlex("stat"), expression, expect(";"), poplex); - } - function expression(type) { - if (atomicTypes.hasOwnProperty(type)) return cont(maybeoperator); - if (type == "function") return cont(functiondef); - if (type == "keyword c") return cont(maybeexpression); - if (type == "(") return cont(pushlex(")"), maybeexpression, expect(")"), poplex, maybeoperator); - if (type == "operator") return cont(expression); - if (type == "[") return cont(pushlex("]"), commasep(expression, "]"), poplex, maybeoperator); - if (type == "{") return cont(pushlex("}"), commasep(objprop, "}"), poplex, maybeoperator); - return cont(); - } - function maybeexpression(type) { - if (type.match(/[;\}\)\],]/)) return pass(); - return pass(expression); - } - - function maybeoperator(type, value) { - if (type == "operator" && /\+\+|--/.test(value)) return cont(maybeoperator); - if (type == "operator" || type == ":") return cont(expression); - if (type == ";") return; - if (type == "(") return cont(pushlex(")"), commasep(expression, ")"), poplex, maybeoperator); - if (type == ".") return cont(property, maybeoperator); - if (type == "[") return cont(pushlex("]"), expression, expect("]"), poplex, maybeoperator); - } - - function maybeattribute(type) { - if (type == "attribute") return cont(maybeattribute); - if (type == "function") return cont(functiondef); - if (type == "var") return cont(vardef1); - } - - function metadef(type) { - if(type == ":") return cont(metadef); - if(type == "variable") return cont(metadef); - if(type == "(") return cont(pushlex(")"), comasep(metaargs, ")"), poplex, statement); - } - function metaargs(type) { - if(type == "variable") return cont(); - } - - function importdef (type, value) { - if(type == "variable" && /[A-Z]/.test(value.charAt(0))) { registerimport(value); return cont(); } - else if(type == "variable" || type == "property" || type == ".") return cont(importdef); - } - - function typedef (type, value) - { - if(type == "variable" && /[A-Z]/.test(value.charAt(0))) { registerimport(value); return cont(); } - } - - function maybelabel(type) { - if (type == ":") return cont(poplex, statement); - return pass(maybeoperator, expect(";"), poplex); - } - function property(type) { - if (type == "variable") {cx.marked = "property"; return cont();} - } - function objprop(type) { - if (type == "variable") cx.marked = "property"; - if (atomicTypes.hasOwnProperty(type)) return cont(expect(":"), expression); - } - function commasep(what, end) { - function proceed(type) { - if (type == ",") return cont(what, proceed); - if (type == end) return cont(); - return cont(expect(end)); - } - return function(type) { - if (type == end) return cont(); - else return pass(what, proceed); - }; - } - function block(type) { - if (type == "}") return cont(); - return pass(statement, block); - } - function vardef1(type, value) { - if (type == "variable"){register(value); return cont(typeuse, vardef2);} - return cont(); - } - function vardef2(type, value) { - if (value == "=") return cont(expression, vardef2); - if (type == ",") return cont(vardef1); - } - function forspec1(type, value) { - if (type == "variable") { - register(value); - } - return cont(pushlex(")"), pushcontext, forin, expression, poplex, statement, popcontext); - } - function forin(_type, value) { - if (value == "in") return cont(); - } - function functiondef(type, value) { - if (type == "variable") {register(value); return cont(functiondef);} - if (value == "new") return cont(functiondef); - if (type == "(") return cont(pushlex(")"), pushcontext, commasep(funarg, ")"), poplex, typeuse, statement, popcontext); - } - function typeuse(type) { - if(type == ":") return cont(typestring); - } - function typestring(type) { - if(type == "type") return cont(); - if(type == "variable") return cont(); - if(type == "{") return cont(pushlex("}"), commasep(typeprop, "}"), poplex); - } - function typeprop(type) { - if(type == "variable") return cont(typeuse); - } - function funarg(type, value) { - if (type == "variable") {register(value); return cont(typeuse);} - } - - // Interface - - return { - startState: function(basecolumn) { - var defaulttypes = ["Int", "Float", "String", "Void", "Std", "Bool", "Dynamic", "Array"]; - return { - tokenize: haxeTokenBase, - reAllowed: true, - kwAllowed: true, - cc: [], - lexical: new HaxeLexical((basecolumn || 0) - indentUnit, 0, "block", false), - localVars: parserConfig.localVars, - importedtypes: defaulttypes, - context: parserConfig.localVars && {vars: parserConfig.localVars}, - indented: 0 - }; - }, - - token: function(stream, state) { - if (stream.sol()) { - if (!state.lexical.hasOwnProperty("align")) - state.lexical.align = false; - state.indented = stream.indentation(); - } - if (stream.eatSpace()) return null; - var style = state.tokenize(stream, state); - if (type == "comment") return style; - state.reAllowed = !!(type == "operator" || type == "keyword c" || type.match(/^[\[{}\(,;:]$/)); - state.kwAllowed = type != '.'; - return parseHaxe(state, style, type, content, stream); - }, - - indent: function(state, textAfter) { - if (state.tokenize != haxeTokenBase) return 0; - var firstChar = textAfter && textAfter.charAt(0), lexical = state.lexical; - if (lexical.type == "stat" && firstChar == "}") lexical = lexical.prev; - var type = lexical.type, closing = firstChar == type; - if (type == "vardef") return lexical.indented + 4; - else if (type == "form" && firstChar == "{") return lexical.indented; - else if (type == "stat" || type == "form") return lexical.indented + indentUnit; - else if (lexical.info == "switch" && !closing) - return lexical.indented + (/^(?:case|default)\b/.test(textAfter) ? indentUnit : 2 * indentUnit); - else if (lexical.align) return lexical.column + (closing ? 0 : 1); - else return lexical.indented + (closing ? 0 : indentUnit); - }, - - electricChars: "{}" - }; -}); - -CodeMirror.defineMIME("text/x-haxe", "haxe"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haxe/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/haxe/index.html deleted file mode 100644 index 1125741ad5..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/haxe/index.html +++ /dev/null @@ -1,90 +0,0 @@ - - - - - CodeMirror: Haxe mode - - - - - - - -

CodeMirror: Haxe mode

- -
- - - -

MIME types defined: text/x-haxe.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/http/http.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/http/http.js deleted file mode 100644 index 5a51636027..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/http/http.js +++ /dev/null @@ -1,98 +0,0 @@ -CodeMirror.defineMode("http", function() { - function failFirstLine(stream, state) { - stream.skipToEnd(); - state.cur = header; - return "error"; - } - - function start(stream, state) { - if (stream.match(/^HTTP\/\d\.\d/)) { - state.cur = responseStatusCode; - return "keyword"; - } else if (stream.match(/^[A-Z]+/) && /[ \t]/.test(stream.peek())) { - state.cur = requestPath; - return "keyword"; - } else { - return failFirstLine(stream, state); - } - } - - function responseStatusCode(stream, state) { - var code = stream.match(/^\d+/); - if (!code) return failFirstLine(stream, state); - - state.cur = responseStatusText; - var status = Number(code[0]); - if (status >= 100 && status < 200) { - return "positive informational"; - } else if (status >= 200 && status < 300) { - return "positive success"; - } else if (status >= 300 && status < 400) { - return "positive redirect"; - } else if (status >= 400 && status < 500) { - return "negative client-error"; - } else if (status >= 500 && status < 600) { - return "negative server-error"; - } else { - return "error"; - } - } - - function responseStatusText(stream, state) { - stream.skipToEnd(); - state.cur = header; - return null; - } - - function requestPath(stream, state) { - stream.eatWhile(/\S/); - state.cur = requestProtocol; - return "string-2"; - } - - function requestProtocol(stream, state) { - if (stream.match(/^HTTP\/\d\.\d$/)) { - state.cur = header; - return "keyword"; - } else { - return failFirstLine(stream, state); - } - } - - function header(stream) { - if (stream.sol() && !stream.eat(/[ \t]/)) { - if (stream.match(/^.*?:/)) { - return "atom"; - } else { - stream.skipToEnd(); - return "error"; - } - } else { - stream.skipToEnd(); - return "string"; - } - } - - function body(stream) { - stream.skipToEnd(); - return null; - } - - return { - token: function(stream, state) { - var cur = state.cur; - if (cur != header && cur != body && stream.eatSpace()) return null; - return cur(stream, state); - }, - - blankLine: function(state) { - state.cur = body; - }, - - startState: function() { - return {cur: start}; - } - }; -}); - -CodeMirror.defineMIME("message/http", "http"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/http/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/http/index.html deleted file mode 100644 index 124eb84f95..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/http/index.html +++ /dev/null @@ -1,32 +0,0 @@ - - - - - CodeMirror: HTTP mode - - - - - - - -

CodeMirror: HTTP mode

- -
- - - -

MIME types defined: message/http.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/jinja2/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/jinja2/index.html deleted file mode 100644 index 7cd1da2335..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/jinja2/index.html +++ /dev/null @@ -1,38 +0,0 @@ - - - - - CodeMirror: Jinja2 mode - - - - - - - -

CodeMirror: Jinja2 mode

-
- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/jinja2/jinja2.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/jinja2/jinja2.js deleted file mode 100644 index 16b06c48ef..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/jinja2/jinja2.js +++ /dev/null @@ -1,42 +0,0 @@ -CodeMirror.defineMode("jinja2", function() { - var keywords = ["block", "endblock", "for", "endfor", "in", "true", "false", - "loop", "none", "self", "super", "if", "as", "not", "and", - "else", "import", "with", "without", "context"]; - keywords = new RegExp("^((" + keywords.join(")|(") + "))\\b"); - - function tokenBase (stream, state) { - var ch = stream.next(); - if (ch == "{") { - if (ch = stream.eat(/\{|%|#/)) { - stream.eat("-"); - state.tokenize = inTag(ch); - return "tag"; - } - } - } - function inTag (close) { - if (close == "{") { - close = "}"; - } - return function (stream, state) { - var ch = stream.next(); - if ((ch == close || (ch == "-" && stream.eat(close))) - && stream.eat("}")) { - state.tokenize = tokenBase; - return "tag"; - } - if (stream.match(keywords)) { - return "keyword"; - } - return close == "#" ? "comment" : "string"; - }; - } - return { - startState: function () { - return {tokenize: tokenBase}; - }, - token: function (stream, state) { - return state.tokenize(stream, state); - } - }; -}); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/LICENSE b/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/LICENSE deleted file mode 100644 index a675c40233..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/LICENSE +++ /dev/null @@ -1,23 +0,0 @@ -The MIT License - -Copyright (c) 2013 Kenneth Bentley -Modified from the CoffeeScript CodeMirror mode, Copyright (c) 2011 Jeff Pickhardt -Modified from the Python CodeMirror mode, Copyright (c) 2010 Timothy Farrell - -Permission is hereby granted, free of charge, to any person obtaining a copy -of this software and associated documentation files (the "Software"), to deal -in the Software without restriction, including without limitation the rights -to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -copies of the Software, and to permit persons to whom the Software is -furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in -all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -THE SOFTWARE. diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/index.html deleted file mode 100644 index 3054e35b0f..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/index.html +++ /dev/null @@ -1,446 +0,0 @@ - - - - CodeMirror: LiveScript mode - - - - - - - - -

CodeMirror: LiveScript mode

-
- - -

MIME types defined: text/x-livescript.

- -

The LiveScript mode was written by Kenneth Bentley (license).

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/livescript.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/livescript.js deleted file mode 100644 index c000324b8c..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/livescript.js +++ /dev/null @@ -1,267 +0,0 @@ -/** - * Link to the project's GitHub page: - * https://github.com/duralog/CodeMirror - */ -(function() { - CodeMirror.defineMode('livescript', function(){ - var tokenBase, external; - tokenBase = function(stream, state){ - var next_rule, nr, i$, len$, r, m; - if (next_rule = state.next || 'start') { - state.next = state.next; - if (Array.isArray(nr = Rules[next_rule])) { - for (i$ = 0, len$ = nr.length; i$ < len$; ++i$) { - r = nr[i$]; - if (r.regex && (m = stream.match(r.regex))) { - state.next = r.next; - return r.token; - } - } - stream.next(); - return 'error'; - } - if (stream.match(r = Rules[next_rule])) { - if (r.regex && stream.match(r.regex)) { - state.next = r.next; - return r.token; - } else { - stream.next(); - return 'error'; - } - } - } - stream.next(); - return 'error'; - }; - external = { - startState: function(){ - return { - next: 'start', - lastToken: null - }; - }, - token: function(stream, state){ - var style; - style = tokenBase(stream, state); - state.lastToken = { - style: style, - indent: stream.indentation(), - content: stream.current() - }; - return style.replace(/\./g, ' '); - }, - indent: function(state){ - var indentation; - indentation = state.lastToken.indent; - if (state.lastToken.content.match(indenter)) { - indentation += 2; - } - return indentation; - } - }; - return external; - }); - - var identifier = '(?![\\d\\s])[$\\w\\xAA-\\uFFDC](?:(?!\\s)[$\\w\\xAA-\\uFFDC]|-[A-Za-z])*'; - var indenter = RegExp('(?:[({[=:]|[-~]>|\\b(?:e(?:lse|xport)|d(?:o|efault)|t(?:ry|hen)|finally|import(?:\\s*all)?|const|var|let|new|catch(?:\\s*' + identifier + ')?))\\s*$'); - var keywordend = '(?![$\\w]|-[A-Za-z]|\\s*:(?![:=]))'; - var stringfill = { - token: 'string', - regex: '.+' - }; - var Rules = { - start: [ - { - token: 'comment.doc', - regex: '/\\*', - next: 'comment' - }, { - token: 'comment', - regex: '#.*' - }, { - token: 'keyword', - regex: '(?:t(?:h(?:is|row|en)|ry|ypeof!?)|c(?:on(?:tinue|st)|a(?:se|tch)|lass)|i(?:n(?:stanceof)?|mp(?:ort(?:\\s+all)?|lements)|[fs])|d(?:e(?:fault|lete|bugger)|o)|f(?:or(?:\\s+own)?|inally|unction)|s(?:uper|witch)|e(?:lse|x(?:tends|port)|val)|a(?:nd|rguments)|n(?:ew|ot)|un(?:less|til)|w(?:hile|ith)|o[fr]|return|break|let|var|loop)' + keywordend - }, { - token: 'constant.language', - regex: '(?:true|false|yes|no|on|off|null|void|undefined)' + keywordend - }, { - token: 'invalid.illegal', - regex: '(?:p(?:ackage|r(?:ivate|otected)|ublic)|i(?:mplements|nterface)|enum|static|yield)' + keywordend - }, { - token: 'language.support.class', - regex: '(?:R(?:e(?:gExp|ferenceError)|angeError)|S(?:tring|yntaxError)|E(?:rror|valError)|Array|Boolean|Date|Function|Number|Object|TypeError|URIError)' + keywordend - }, { - token: 'language.support.function', - regex: '(?:is(?:NaN|Finite)|parse(?:Int|Float)|Math|JSON|(?:en|de)codeURI(?:Component)?)' + keywordend - }, { - token: 'variable.language', - regex: '(?:t(?:hat|il|o)|f(?:rom|allthrough)|it|by|e)' + keywordend - }, { - token: 'identifier', - regex: identifier + '\\s*:(?![:=])' - }, { - token: 'variable', - regex: identifier - }, { - token: 'keyword.operator', - regex: '(?:\\.{3}|\\s+\\?)' - }, { - token: 'keyword.variable', - regex: '(?:@+|::|\\.\\.)', - next: 'key' - }, { - token: 'keyword.operator', - regex: '\\.\\s*', - next: 'key' - }, { - token: 'string', - regex: '\\\\\\S[^\\s,;)}\\]]*' - }, { - token: 'string.doc', - regex: '\'\'\'', - next: 'qdoc' - }, { - token: 'string.doc', - regex: '"""', - next: 'qqdoc' - }, { - token: 'string', - regex: '\'', - next: 'qstring' - }, { - token: 'string', - regex: '"', - next: 'qqstring' - }, { - token: 'string', - regex: '`', - next: 'js' - }, { - token: 'string', - regex: '<\\[', - next: 'words' - }, { - token: 'string.regex', - regex: '//', - next: 'heregex' - }, { - token: 'string.regex', - regex: '\\/(?:[^[\\/\\n\\\\]*(?:(?:\\\\.|\\[[^\\]\\n\\\\]*(?:\\\\.[^\\]\\n\\\\]*)*\\])[^[\\/\\n\\\\]*)*)\\/[gimy$]{0,4}', - next: 'key' - }, { - token: 'constant.numeric', - regex: '(?:0x[\\da-fA-F][\\da-fA-F_]*|(?:[2-9]|[12]\\d|3[0-6])r[\\da-zA-Z][\\da-zA-Z_]*|(?:\\d[\\d_]*(?:\\.\\d[\\d_]*)?|\\.\\d[\\d_]*)(?:e[+-]?\\d[\\d_]*)?[\\w$]*)' - }, { - token: 'lparen', - regex: '[({[]' - }, { - token: 'rparen', - regex: '[)}\\]]', - next: 'key' - }, { - token: 'keyword.operator', - regex: '\\S+' - }, { - token: 'text', - regex: '\\s+' - } - ], - heregex: [ - { - token: 'string.regex', - regex: '.*?//[gimy$?]{0,4}', - next: 'start' - }, { - token: 'string.regex', - regex: '\\s*#{' - }, { - token: 'comment.regex', - regex: '\\s+(?:#.*)?' - }, { - token: 'string.regex', - regex: '\\S+' - } - ], - key: [ - { - token: 'keyword.operator', - regex: '[.?@!]+' - }, { - token: 'identifier', - regex: identifier, - next: 'start' - }, { - token: 'text', - regex: '.', - next: 'start' - } - ], - comment: [ - { - token: 'comment.doc', - regex: '.*?\\*/', - next: 'start' - }, { - token: 'comment.doc', - regex: '.+' - } - ], - qdoc: [ - { - token: 'string', - regex: ".*?'''", - next: 'key' - }, stringfill - ], - qqdoc: [ - { - token: 'string', - regex: '.*?"""', - next: 'key' - }, stringfill - ], - qstring: [ - { - token: 'string', - regex: '[^\\\\\']*(?:\\\\.[^\\\\\']*)*\'', - next: 'key' - }, stringfill - ], - qqstring: [ - { - token: 'string', - regex: '[^\\\\"]*(?:\\\\.[^\\\\"]*)*"', - next: 'key' - }, stringfill - ], - js: [ - { - token: 'string', - regex: '[^\\\\`]*(?:\\\\.[^\\\\`]*)*`', - next: 'key' - }, stringfill - ], - words: [ - { - token: 'string', - regex: '.*?\\]>', - next: 'key' - }, stringfill - ] - }; - for (var idx in Rules) { - var r = Rules[idx]; - if (Array.isArray(r)) { - for (var i = 0, len = r.length; i < len; ++i) { - var rr = r[i]; - if (rr.regex) { - Rules[idx][i].regex = new RegExp('^' + rr.regex); - } - } - } else if (r.regex) { - Rules[idx].regex = new RegExp('^' + r.regex); - } - } -})(); - -CodeMirror.defineMIME('text/x-livescript', 'livescript'); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/livescript.ls b/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/livescript.ls deleted file mode 100644 index 065242312a..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/livescript/livescript.ls +++ /dev/null @@ -1,266 +0,0 @@ -/** - * Link to the project's GitHub page: - * https://github.com/duralog/CodeMirror - */ -CodeMirror.defineMode 'livescript', (conf) -> - tokenBase = (stream, state) -> - #indent = - if next_rule = state.next or \start - state.next = state.next - if Array.isArray nr = Rules[next_rule] - for r in nr - if r.regex and m = stream.match r.regex - state.next = r.next - return r.token - stream.next! - return \error - if stream.match r = Rules[next_rule] - if r.regex and stream.match r.regex - state.next = r.next - return r.token - else - stream.next! - return \error - stream.next! - return 'error' - external = { - startState: (basecolumn) -> - { - next: \start - lastToken: null - } - token: (stream, state) -> - style = tokenBase stream, state #tokenLexer stream, state - state.lastToken = { - style: style - indent: stream.indentation! - content: stream.current! - } - style.replace /\./g, ' ' - indent: (state, textAfter) -> - # XXX this won't work with backcalls - indentation = state.lastToken.indent - if state.lastToken.content.match indenter then indentation += 2 - return indentation - } - external - -### Highlight Rules -# taken from mode-ls.ls - -indenter = // (? - : [({[=:] - | [-~]> - | \b (?: e(?:lse|xport) | d(?:o|efault) | t(?:ry|hen) | finally | - import (?:\s* all)? | const | var | - let | new | catch (?:\s* #identifier)? ) - ) \s* $ // - -identifier = /(?![\d\s])[$\w\xAA-\uFFDC](?:(?!\s)[$\w\xAA-\uFFDC]|-[A-Za-z])*/$ -keywordend = /(?![$\w]|-[A-Za-z]|\s*:(?![:=]))/$ -stringfill = token: \string, regex: '.+' - -Rules = - start: - * token: \comment.doc - regex: '/\\*' - next : \comment - - * token: \comment - regex: '#.*' - - * token: \keyword - regex: //(? - :t(?:h(?:is|row|en)|ry|ypeof!?) - |c(?:on(?:tinue|st)|a(?:se|tch)|lass) - |i(?:n(?:stanceof)?|mp(?:ort(?:\s+all)?|lements)|[fs]) - |d(?:e(?:fault|lete|bugger)|o) - |f(?:or(?:\s+own)?|inally|unction) - |s(?:uper|witch) - |e(?:lse|x(?:tends|port)|val) - |a(?:nd|rguments) - |n(?:ew|ot) - |un(?:less|til) - |w(?:hile|ith) - |o[fr]|return|break|let|var|loop - )//$ + keywordend - - * token: \constant.language - regex: '(?:true|false|yes|no|on|off|null|void|undefined)' + keywordend - - * token: \invalid.illegal - regex: '(? - :p(?:ackage|r(?:ivate|otected)|ublic) - |i(?:mplements|nterface) - |enum|static|yield - )' + keywordend - - * token: \language.support.class - regex: '(? - :R(?:e(?:gExp|ferenceError)|angeError) - |S(?:tring|yntaxError) - |E(?:rror|valError) - |Array|Boolean|Date|Function|Number|Object|TypeError|URIError - )' + keywordend - - * token: \language.support.function - regex: '(? - :is(?:NaN|Finite) - |parse(?:Int|Float) - |Math|JSON - |(?:en|de)codeURI(?:Component)? - )' + keywordend - - * token: \variable.language - regex: '(?:t(?:hat|il|o)|f(?:rom|allthrough)|it|by|e)' + keywordend - - * token: \identifier - regex: identifier + /\s*:(?![:=])/$ - - * token: \variable - regex: identifier - - * token: \keyword.operator - regex: /(?:\.{3}|\s+\?)/$ - - * token: \keyword.variable - regex: /(?:@+|::|\.\.)/$ - next : \key - - * token: \keyword.operator - regex: /\.\s*/$ - next : \key - - * token: \string - regex: /\\\S[^\s,;)}\]]*/$ - - * token: \string.doc - regex: \''' - next : \qdoc - - * token: \string.doc - regex: \""" - next : \qqdoc - - * token: \string - regex: \' - next : \qstring - - * token: \string - regex: \" - next : \qqstring - - * token: \string - regex: \` - next : \js - - * token: \string - regex: '<\\[' - next : \words - - * token: \string.regex - regex: \// - next : \heregex - - * token: \string.regex - regex: // - /(?: [^ [ / \n \\ ]* - (?: (?: \\. - | \[ [^\]\n\\]* (?:\\.[^\]\n\\]*)* \] - ) [^ [ / \n \\ ]* - )* - )/ [gimy$]{0,4} - //$ - next : \key - - * token: \constant.numeric - regex: '(?:0x[\\da-fA-F][\\da-fA-F_]* - |(?:[2-9]|[12]\\d|3[0-6])r[\\da-zA-Z][\\da-zA-Z_]* - |(?:\\d[\\d_]*(?:\\.\\d[\\d_]*)?|\\.\\d[\\d_]*) - (?:e[+-]?\\d[\\d_]*)?[\\w$]*)' - - * token: \lparen - regex: '[({[]' - - * token: \rparen - regex: '[)}\\]]' - next : \key - - * token: \keyword.operator - regex: \\\S+ - - * token: \text - regex: \\\s+ - - heregex: - * token: \string.regex - regex: '.*?//[gimy$?]{0,4}' - next : \start - * token: \string.regex - regex: '\\s*#{' - * token: \comment.regex - regex: '\\s+(?:#.*)?' - * token: \string.regex - regex: '\\S+' - - key: - * token: \keyword.operator - regex: '[.?@!]+' - * token: \identifier - regex: identifier - next : \start - * token: \text - regex: '.' - next : \start - - comment: - * token: \comment.doc - regex: '.*?\\*/' - next : \start - * token: \comment.doc - regex: '.+' - - qdoc: - token: \string - regex: ".*?'''" - next : \key - stringfill - - qqdoc: - token: \string - regex: '.*?"""' - next : \key - stringfill - - qstring: - token: \string - regex: /[^\\']*(?:\\.[^\\']*)*'/$ - next : \key - stringfill - - qqstring: - token: \string - regex: /[^\\"]*(?:\\.[^\\"]*)*"/$ - next : \key - stringfill - - js: - token: \string - regex: /[^\\`]*(?:\\.[^\\`]*)*`/$ - next : \key - stringfill - - words: - token: \string - regex: '.*?\\]>' - next : \key - stringfill - -# for optimization, precompile the regexps -for idx, r of Rules - if Array.isArray r - for rr, i in r - if rr.regex then Rules[idx][i].regex = new RegExp '^'+rr.regex - else if r.regex then Rules[idx].regex = new RegExp '^'+r.regex - -CodeMirror.defineMIME 'text/x-livescript', 'livescript' diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/lua/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/lua/index.html deleted file mode 100644 index a0a42d91c4..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/lua/index.html +++ /dev/null @@ -1,74 +0,0 @@ - - - - - CodeMirror: Lua mode - - - - - - - - - -

CodeMirror: Lua mode

-
- - -

Loosely based on Franciszek - Wawrzak's CodeMirror - 1 mode. One configuration parameter is - supported, specials, to which you can provide an - array of strings to have those identifiers highlighted with - the lua-special style.

-

MIME types defined: text/x-lua.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/lua/lua.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/lua/lua.js deleted file mode 100644 index b8deaa2575..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/lua/lua.js +++ /dev/null @@ -1,144 +0,0 @@ -// LUA mode. Ported to CodeMirror 2 from Franciszek Wawrzak's -// CodeMirror 1 mode. -// highlights keywords, strings, comments (no leveling supported! ("[==[")), tokens, basic indenting - -CodeMirror.defineMode("lua", function(config, parserConfig) { - var indentUnit = config.indentUnit; - - function prefixRE(words) { - return new RegExp("^(?:" + words.join("|") + ")", "i"); - } - function wordRE(words) { - return new RegExp("^(?:" + words.join("|") + ")$", "i"); - } - var specials = wordRE(parserConfig.specials || []); - - // long list of standard functions from lua manual - var builtins = wordRE([ - "_G","_VERSION","assert","collectgarbage","dofile","error","getfenv","getmetatable","ipairs","load", - "loadfile","loadstring","module","next","pairs","pcall","print","rawequal","rawget","rawset","require", - "select","setfenv","setmetatable","tonumber","tostring","type","unpack","xpcall", - - "coroutine.create","coroutine.resume","coroutine.running","coroutine.status","coroutine.wrap","coroutine.yield", - - "debug.debug","debug.getfenv","debug.gethook","debug.getinfo","debug.getlocal","debug.getmetatable", - "debug.getregistry","debug.getupvalue","debug.setfenv","debug.sethook","debug.setlocal","debug.setmetatable", - "debug.setupvalue","debug.traceback", - - "close","flush","lines","read","seek","setvbuf","write", - - "io.close","io.flush","io.input","io.lines","io.open","io.output","io.popen","io.read","io.stderr","io.stdin", - "io.stdout","io.tmpfile","io.type","io.write", - - "math.abs","math.acos","math.asin","math.atan","math.atan2","math.ceil","math.cos","math.cosh","math.deg", - "math.exp","math.floor","math.fmod","math.frexp","math.huge","math.ldexp","math.log","math.log10","math.max", - "math.min","math.modf","math.pi","math.pow","math.rad","math.random","math.randomseed","math.sin","math.sinh", - "math.sqrt","math.tan","math.tanh", - - "os.clock","os.date","os.difftime","os.execute","os.exit","os.getenv","os.remove","os.rename","os.setlocale", - "os.time","os.tmpname", - - "package.cpath","package.loaded","package.loaders","package.loadlib","package.path","package.preload", - "package.seeall", - - "string.byte","string.char","string.dump","string.find","string.format","string.gmatch","string.gsub", - "string.len","string.lower","string.match","string.rep","string.reverse","string.sub","string.upper", - - "table.concat","table.insert","table.maxn","table.remove","table.sort" - ]); - var keywords = wordRE(["and","break","elseif","false","nil","not","or","return", - "true","function", "end", "if", "then", "else", "do", - "while", "repeat", "until", "for", "in", "local" ]); - - var indentTokens = wordRE(["function", "if","repeat","do", "\\(", "{"]); - var dedentTokens = wordRE(["end", "until", "\\)", "}"]); - var dedentPartial = prefixRE(["end", "until", "\\)", "}", "else", "elseif"]); - - function readBracket(stream) { - var level = 0; - while (stream.eat("=")) ++level; - stream.eat("["); - return level; - } - - function normal(stream, state) { - var ch = stream.next(); - if (ch == "-" && stream.eat("-")) { - if (stream.eat("[") && stream.eat("[")) - return (state.cur = bracketed(readBracket(stream), "comment"))(stream, state); - stream.skipToEnd(); - return "comment"; - } - if (ch == "\"" || ch == "'") - return (state.cur = string(ch))(stream, state); - if (ch == "[" && /[\[=]/.test(stream.peek())) - return (state.cur = bracketed(readBracket(stream), "string"))(stream, state); - if (/\d/.test(ch)) { - stream.eatWhile(/[\w.%]/); - return "number"; - } - if (/[\w_]/.test(ch)) { - stream.eatWhile(/[\w\\\-_.]/); - return "variable"; - } - return null; - } - - function bracketed(level, style) { - return function(stream, state) { - var curlev = null, ch; - while ((ch = stream.next()) != null) { - if (curlev == null) {if (ch == "]") curlev = 0;} - else if (ch == "=") ++curlev; - else if (ch == "]" && curlev == level) { state.cur = normal; break; } - else curlev = null; - } - return style; - }; - } - - function string(quote) { - return function(stream, state) { - var escaped = false, ch; - while ((ch = stream.next()) != null) { - if (ch == quote && !escaped) break; - escaped = !escaped && ch == "\\"; - } - if (!escaped) state.cur = normal; - return "string"; - }; - } - - return { - startState: function(basecol) { - return {basecol: basecol || 0, indentDepth: 0, cur: normal}; - }, - - token: function(stream, state) { - if (stream.eatSpace()) return null; - var style = state.cur(stream, state); - var word = stream.current(); - if (style == "variable") { - if (keywords.test(word)) style = "keyword"; - else if (builtins.test(word)) style = "builtin"; - else if (specials.test(word)) style = "variable-2"; - } - if ((style != "comment") && (style != "string")){ - if (indentTokens.test(word)) ++state.indentDepth; - else if (dedentTokens.test(word)) --state.indentDepth; - } - return style; - }, - - indent: function(state, textAfter) { - var closing = dedentPartial.test(textAfter); - return state.basecol + indentUnit * (state.indentDepth - (closing ? 1 : 0)); - }, - - lineComment: "--", - blockCommentStart: "--[[", - blockCommentEnd: "]]" - }; -}); - -CodeMirror.defineMIME("text/x-lua", "lua"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/index.html deleted file mode 100644 index 6f97b10e73..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/index.html +++ /dev/null @@ -1,344 +0,0 @@ - - - - - CodeMirror: Markdown mode - - - - - - - - - -

CodeMirror: Markdown mode

- - -
- - - -

Optionally depends on the XML mode for properly highlighted inline XML blocks.

- -

MIME types defined: text/x-markdown.

- -

Parsing/Highlighting Tests: normal, verbose.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/markdown.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/markdown.js deleted file mode 100644 index d93d55590e..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/markdown.js +++ /dev/null @@ -1,526 +0,0 @@ -CodeMirror.defineMode("markdown", function(cmCfg, modeCfg) { - - var htmlFound = CodeMirror.modes.hasOwnProperty("xml"); - var htmlMode = CodeMirror.getMode(cmCfg, htmlFound ? {name: "xml", htmlMode: true} : "text/plain"); - var aliases = { - html: "htmlmixed", - js: "javascript", - json: "application/json", - c: "text/x-csrc", - "c++": "text/x-c++src", - java: "text/x-java", - csharp: "text/x-csharp", - "c#": "text/x-csharp", - scala: "text/x-scala" - }; - - var getMode = (function () { - var i, modes = {}, mimes = {}, mime; - - var list = []; - for (var m in CodeMirror.modes) - if (CodeMirror.modes.propertyIsEnumerable(m)) list.push(m); - for (i = 0; i < list.length; i++) { - modes[list[i]] = list[i]; - } - var mimesList = []; - for (var m in CodeMirror.mimeModes) - if (CodeMirror.mimeModes.propertyIsEnumerable(m)) - mimesList.push({mime: m, mode: CodeMirror.mimeModes[m]}); - for (i = 0; i < mimesList.length; i++) { - mime = mimesList[i].mime; - mimes[mime] = mimesList[i].mime; - } - - for (var a in aliases) { - if (aliases[a] in modes || aliases[a] in mimes) - modes[a] = aliases[a]; - } - - return function (lang) { - return modes[lang] ? CodeMirror.getMode(cmCfg, modes[lang]) : null; - }; - }()); - - // Should underscores in words open/close em/strong? - if (modeCfg.underscoresBreakWords === undefined) - modeCfg.underscoresBreakWords = true; - - // Turn on fenced code blocks? ("```" to start/end) - if (modeCfg.fencedCodeBlocks === undefined) modeCfg.fencedCodeBlocks = false; - - // Turn on task lists? ("- [ ] " and "- [x] ") - if (modeCfg.taskLists === undefined) modeCfg.taskLists = false; - - var codeDepth = 0; - - var header = 'header' - , code = 'comment' - , quote1 = 'atom' - , quote2 = 'number' - , list1 = 'variable-2' - , list2 = 'variable-3' - , list3 = 'keyword' - , hr = 'hr' - , image = 'tag' - , linkinline = 'link' - , linkemail = 'link' - , linktext = 'link' - , linkhref = 'string' - , em = 'em' - , strong = 'strong'; - - var hrRE = /^([*\-=_])(?:\s*\1){2,}\s*$/ - , ulRE = /^[*\-+]\s+/ - , olRE = /^[0-9]+\.\s+/ - , taskListRE = /^\[(x| )\](?=\s)/ // Must follow ulRE or olRE - , headerRE = /^(?:\={1,}|-{1,})$/ - , textRE = /^[^!\[\]*_\\<>` "'(]+/; - - function switchInline(stream, state, f) { - state.f = state.inline = f; - return f(stream, state); - } - - function switchBlock(stream, state, f) { - state.f = state.block = f; - return f(stream, state); - } - - - // Blocks - - function blankLine(state) { - // Reset linkTitle state - state.linkTitle = false; - // Reset EM state - state.em = false; - // Reset STRONG state - state.strong = false; - // Reset state.quote - state.quote = 0; - if (!htmlFound && state.f == htmlBlock) { - state.f = inlineNormal; - state.block = blockNormal; - } - // Mark this line as blank - state.thisLineHasContent = false; - return null; - } - - function blockNormal(stream, state) { - - var prevLineIsList = (state.list !== false); - if (state.list !== false && state.indentationDiff >= 0) { // Continued list - if (state.indentationDiff < 4) { // Only adjust indentation if *not* a code block - state.indentation -= state.indentationDiff; - } - state.list = null; - } else if (state.list !== false && state.indentation > 0) { - state.list = null; - state.listDepth = Math.floor(state.indentation / 4); - } else if (state.list !== false) { // No longer a list - state.list = false; - state.listDepth = 0; - } - - if (state.indentationDiff >= 4) { - state.indentation -= 4; - stream.skipToEnd(); - return code; - } else if (stream.eatSpace()) { - return null; - } else if (stream.peek() === '#' || (state.prevLineHasContent && stream.match(headerRE)) ) { - state.header = true; - } else if (stream.eat('>')) { - state.indentation++; - state.quote = 1; - stream.eatSpace(); - while (stream.eat('>')) { - stream.eatSpace(); - state.quote++; - } - } else if (stream.peek() === '[') { - return switchInline(stream, state, footnoteLink); - } else if (stream.match(hrRE, true)) { - return hr; - } else if ((!state.prevLineHasContent || prevLineIsList) && (stream.match(ulRE, true) || stream.match(olRE, true))) { - state.indentation += 4; - state.list = true; - state.listDepth++; - if (modeCfg.taskLists && stream.match(taskListRE, false)) { - state.taskList = true; - } - } else if (modeCfg.fencedCodeBlocks && stream.match(/^```([\w+#]*)/, true)) { - // try switching mode - state.localMode = getMode(RegExp.$1); - if (state.localMode) state.localState = state.localMode.startState(); - switchBlock(stream, state, local); - return code; - } - - return switchInline(stream, state, state.inline); - } - - function htmlBlock(stream, state) { - var style = htmlMode.token(stream, state.htmlState); - if (htmlFound && style === 'tag' && state.htmlState.type !== 'openTag' && !state.htmlState.context) { - state.f = inlineNormal; - state.block = blockNormal; - } - if (state.md_inside && stream.current().indexOf(">")!=-1) { - state.f = inlineNormal; - state.block = blockNormal; - state.htmlState.context = undefined; - } - return style; - } - - function local(stream, state) { - if (stream.sol() && stream.match(/^```/, true)) { - state.localMode = state.localState = null; - state.f = inlineNormal; - state.block = blockNormal; - return code; - } else if (state.localMode) { - return state.localMode.token(stream, state.localState); - } else { - stream.skipToEnd(); - return code; - } - } - - // Inline - function getType(state) { - var styles = []; - - if (state.taskOpen) { return "meta"; } - if (state.taskClosed) { return "property"; } - - if (state.strong) { styles.push(strong); } - if (state.em) { styles.push(em); } - - if (state.linkText) { styles.push(linktext); } - - if (state.code) { styles.push(code); } - - if (state.header) { styles.push(header); } - if (state.quote) { styles.push(state.quote % 2 ? quote1 : quote2); } - if (state.list !== false) { - var listMod = (state.listDepth - 1) % 3; - if (!listMod) { - styles.push(list1); - } else if (listMod === 1) { - styles.push(list2); - } else { - styles.push(list3); - } - } - - return styles.length ? styles.join(' ') : null; - } - - function handleText(stream, state) { - if (stream.match(textRE, true)) { - return getType(state); - } - return undefined; - } - - function inlineNormal(stream, state) { - var style = state.text(stream, state); - if (typeof style !== 'undefined') - return style; - - if (state.list) { // List marker (*, +, -, 1., etc) - state.list = null; - return getType(state); - } - - if (state.taskList) { - var taskOpen = stream.match(taskListRE, true)[1] !== "x"; - if (taskOpen) state.taskOpen = true; - else state.taskClosed = true; - state.taskList = false; - return getType(state); - } - - state.taskOpen = false; - state.taskClosed = false; - - var ch = stream.next(); - - if (ch === '\\') { - stream.next(); - return getType(state); - } - - // Matches link titles present on next line - if (state.linkTitle) { - state.linkTitle = false; - var matchCh = ch; - if (ch === '(') { - matchCh = ')'; - } - matchCh = (matchCh+'').replace(/([.?*+^$[\]\\(){}|-])/g, "\\$1"); - var regex = '^\\s*(?:[^' + matchCh + '\\\\]+|\\\\\\\\|\\\\.)' + matchCh; - if (stream.match(new RegExp(regex), true)) { - return linkhref; - } - } - - // If this block is changed, it may need to be updated in GFM mode - if (ch === '`') { - var t = getType(state); - var before = stream.pos; - stream.eatWhile('`'); - var difference = 1 + stream.pos - before; - if (!state.code) { - codeDepth = difference; - state.code = true; - return getType(state); - } else { - if (difference === codeDepth) { // Must be exact - state.code = false; - return t; - } - return getType(state); - } - } else if (state.code) { - return getType(state); - } - - if (ch === '!' && stream.match(/\[[^\]]*\] ?(?:\(|\[)/, false)) { - stream.match(/\[[^\]]*\]/); - state.inline = state.f = linkHref; - return image; - } - - if (ch === '[' && stream.match(/.*\](\(| ?\[)/, false)) { - state.linkText = true; - return getType(state); - } - - if (ch === ']' && state.linkText) { - var type = getType(state); - state.linkText = false; - state.inline = state.f = linkHref; - return type; - } - - if (ch === '<' && stream.match(/^(https?|ftps?):\/\/(?:[^\\>]|\\.)+>/, true)) { - return switchInline(stream, state, inlineElement(linkinline, '>')); - } - - if (ch === '<' && stream.match(/^[^> \\]+@(?:[^\\>]|\\.)+>/, true)) { - return switchInline(stream, state, inlineElement(linkemail, '>')); - } - - if (ch === '<' && stream.match(/^\w/, false)) { - if (stream.string.indexOf(">")!=-1) { - var atts = stream.string.substring(1,stream.string.indexOf(">")); - if (/markdown\s*=\s*('|"){0,1}1('|"){0,1}/.test(atts)) { - state.md_inside = true; - } - } - stream.backUp(1); - return switchBlock(stream, state, htmlBlock); - } - - if (ch === '<' && stream.match(/^\/\w*?>/)) { - state.md_inside = false; - return "tag"; - } - - var ignoreUnderscore = false; - if (!modeCfg.underscoresBreakWords) { - if (ch === '_' && stream.peek() !== '_' && stream.match(/(\w)/, false)) { - var prevPos = stream.pos - 2; - if (prevPos >= 0) { - var prevCh = stream.string.charAt(prevPos); - if (prevCh !== '_' && prevCh.match(/(\w)/, false)) { - ignoreUnderscore = true; - } - } - } - } - var t = getType(state); - if (ch === '*' || (ch === '_' && !ignoreUnderscore)) { - if (state.strong === ch && stream.eat(ch)) { // Remove STRONG - state.strong = false; - return t; - } else if (!state.strong && stream.eat(ch)) { // Add STRONG - state.strong = ch; - return getType(state); - } else if (state.em === ch) { // Remove EM - state.em = false; - return t; - } else if (!state.em) { // Add EM - state.em = ch; - return getType(state); - } - } else if (ch === ' ') { - if (stream.eat('*') || stream.eat('_')) { // Probably surrounded by spaces - if (stream.peek() === ' ') { // Surrounded by spaces, ignore - return getType(state); - } else { // Not surrounded by spaces, back up pointer - stream.backUp(1); - } - } - } - - return getType(state); - } - - function linkHref(stream, state) { - // Check if space, and return NULL if so (to avoid marking the space) - if(stream.eatSpace()){ - return null; - } - var ch = stream.next(); - if (ch === '(' || ch === '[') { - return switchInline(stream, state, inlineElement(linkhref, ch === '(' ? ')' : ']')); - } - return 'error'; - } - - function footnoteLink(stream, state) { - if (stream.match(/^[^\]]*\]:/, true)) { - state.f = footnoteUrl; - return linktext; - } - return switchInline(stream, state, inlineNormal); - } - - function footnoteUrl(stream, state) { - // Check if space, and return NULL if so (to avoid marking the space) - if(stream.eatSpace()){ - return null; - } - // Match URL - stream.match(/^[^\s]+/, true); - // Check for link title - if (stream.peek() === undefined) { // End of line, set flag to check next line - state.linkTitle = true; - } else { // More content on line, check if link title - stream.match(/^(?:\s+(?:"(?:[^"\\]|\\\\|\\.)+"|'(?:[^'\\]|\\\\|\\.)+'|\((?:[^)\\]|\\\\|\\.)+\)))?/, true); - } - state.f = state.inline = inlineNormal; - return linkhref; - } - - var savedInlineRE = []; - function inlineRE(endChar) { - if (!savedInlineRE[endChar]) { - // Escape endChar for RegExp (taken from http://stackoverflow.com/a/494122/526741) - endChar = (endChar+'').replace(/([.?*+^$[\]\\(){}|-])/g, "\\$1"); - // Match any non-endChar, escaped character, as well as the closing - // endChar. - savedInlineRE[endChar] = new RegExp('^(?:[^\\\\]|\\\\.)*?(' + endChar + ')'); - } - return savedInlineRE[endChar]; - } - - function inlineElement(type, endChar, next) { - next = next || inlineNormal; - return function(stream, state) { - stream.match(inlineRE(endChar)); - state.inline = state.f = next; - return type; - }; - } - - return { - startState: function() { - return { - f: blockNormal, - - prevLineHasContent: false, - thisLineHasContent: false, - - block: blockNormal, - htmlState: CodeMirror.startState(htmlMode), - indentation: 0, - - inline: inlineNormal, - text: handleText, - - linkText: false, - linkTitle: false, - em: false, - strong: false, - header: false, - taskList: false, - list: false, - listDepth: 0, - quote: 0 - }; - }, - - copyState: function(s) { - return { - f: s.f, - - prevLineHasContent: s.prevLineHasContent, - thisLineHasContent: s.thisLineHasContent, - - block: s.block, - htmlState: CodeMirror.copyState(htmlMode, s.htmlState), - indentation: s.indentation, - - localMode: s.localMode, - localState: s.localMode ? CodeMirror.copyState(s.localMode, s.localState) : null, - - inline: s.inline, - text: s.text, - linkTitle: s.linkTitle, - em: s.em, - strong: s.strong, - header: s.header, - taskList: s.taskList, - list: s.list, - listDepth: s.listDepth, - quote: s.quote, - md_inside: s.md_inside - }; - }, - - token: function(stream, state) { - if (stream.sol()) { - if (stream.match(/^\s*$/, true)) { - state.prevLineHasContent = false; - return blankLine(state); - } else { - state.prevLineHasContent = state.thisLineHasContent; - state.thisLineHasContent = true; - } - - // Reset state.header - state.header = false; - - // Reset state.taskList - state.taskList = false; - - // Reset state.code - state.code = false; - - state.f = state.block; - var indentation = stream.match(/^\s*/, true)[0].replace(/\t/g, ' ').length; - var difference = Math.floor((indentation - state.indentation) / 4) * 4; - if (difference > 4) difference = 4; - var adjustedIndentation = state.indentation + difference; - state.indentationDiff = adjustedIndentation - state.indentation; - state.indentation = adjustedIndentation; - if (indentation > 0) return null; - } - return state.f(stream, state); - }, - - blankLine: blankLine, - - getType: getType - }; - -}, "xml"); - -CodeMirror.defineMIME("text/x-markdown", "markdown"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/test.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/test.js deleted file mode 100644 index c451412fee..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/markdown/test.js +++ /dev/null @@ -1,636 +0,0 @@ -(function() { - var mode = CodeMirror.getMode({tabSize: 4}, "markdown"); - function MT(name) { test.mode(name, mode, Array.prototype.slice.call(arguments, 1)); } - - MT("plainText", - "foo"); - - // Code blocks using 4 spaces (regardless of CodeMirror.tabSize value) - MT("codeBlocksUsing4Spaces", - " [comment foo]"); - - // Code blocks using 4 spaces with internal indentation - MT("codeBlocksUsing4SpacesIndentation", - " [comment bar]", - " [comment hello]", - " [comment world]", - " [comment foo]", - "bar"); - - // Code blocks using 4 spaces with internal indentation - MT("codeBlocksUsing4SpacesIndentation", - " foo", - " [comment bar]", - " [comment hello]", - " [comment world]"); - - // Code blocks using 1 tab (regardless of CodeMirror.indentWithTabs value) - MT("codeBlocksUsing1Tab", - "\t[comment foo]"); - - // Inline code using backticks - MT("inlineCodeUsingBackticks", - "foo [comment `bar`]"); - - // Block code using single backtick (shouldn't work) - MT("blockCodeSingleBacktick", - "[comment `]", - "foo", - "[comment `]"); - - // Unclosed backticks - // Instead of simply marking as CODE, it would be nice to have an - // incomplete flag for CODE, that is styled slightly different. - MT("unclosedBackticks", - "foo [comment `bar]"); - - // Per documentation: "To include a literal backtick character within a - // code span, you can use multiple backticks as the opening and closing - // delimiters" - MT("doubleBackticks", - "[comment ``foo ` bar``]"); - - // Tests based on Dingus - // http://daringfireball.net/projects/markdown/dingus - // - // Multiple backticks within an inline code block - MT("consecutiveBackticks", - "[comment `foo```bar`]"); - - // Multiple backticks within an inline code block with a second code block - MT("consecutiveBackticks", - "[comment `foo```bar`] hello [comment `world`]"); - - // Unclosed with several different groups of backticks - MT("unclosedBackticks", - "[comment ``foo ``` bar` hello]"); - - // Closed with several different groups of backticks - MT("closedBackticks", - "[comment ``foo ``` bar` hello``] world"); - - // atx headers - // http://daringfireball.net/projects/markdown/syntax#header - - MT("atxH1", - "[header # foo]"); - - MT("atxH2", - "[header ## foo]"); - - MT("atxH3", - "[header ### foo]"); - - MT("atxH4", - "[header #### foo]"); - - MT("atxH5", - "[header ##### foo]"); - - MT("atxH6", - "[header ###### foo]"); - - // H6 - 7x '#' should still be H6, per Dingus - // http://daringfireball.net/projects/markdown/dingus - MT("atxH6NotH7", - "[header ####### foo]"); - - // Setext headers - H1, H2 - // Per documentation, "Any number of underlining =’s or -’s will work." - // http://daringfireball.net/projects/markdown/syntax#header - // Ideally, the text would be marked as `header` as well, but this is - // not really feasible at the moment. So, instead, we're testing against - // what works today, to avoid any regressions. - // - // Check if single underlining = works - MT("setextH1", - "foo", - "[header =]"); - - // Check if 3+ ='s work - MT("setextH1", - "foo", - "[header ===]"); - - // Check if single underlining - works - MT("setextH2", - "foo", - "[header -]"); - - // Check if 3+ -'s work - MT("setextH2", - "foo", - "[header ---]"); - - // Single-line blockquote with trailing space - MT("blockquoteSpace", - "[atom > foo]"); - - // Single-line blockquote - MT("blockquoteNoSpace", - "[atom >foo]"); - - // No blank line before blockquote - MT("blockquoteNoBlankLine", - "foo", - "[atom > bar]"); - - // Nested blockquote - MT("blockquoteSpace", - "[atom > foo]", - "[number > > foo]", - "[atom > > > foo]"); - - // Single-line blockquote followed by normal paragraph - MT("blockquoteThenParagraph", - "[atom >foo]", - "", - "bar"); - - // Multi-line blockquote (lazy mode) - MT("multiBlockquoteLazy", - "[atom >foo]", - "[atom bar]"); - - // Multi-line blockquote followed by normal paragraph (lazy mode) - MT("multiBlockquoteLazyThenParagraph", - "[atom >foo]", - "[atom bar]", - "", - "hello"); - - // Multi-line blockquote (non-lazy mode) - MT("multiBlockquote", - "[atom >foo]", - "[atom >bar]"); - - // Multi-line blockquote followed by normal paragraph (non-lazy mode) - MT("multiBlockquoteThenParagraph", - "[atom >foo]", - "[atom >bar]", - "", - "hello"); - - // Check list types - - MT("listAsterisk", - "foo", - "bar", - "", - "[variable-2 * foo]", - "[variable-2 * bar]"); - - MT("listPlus", - "foo", - "bar", - "", - "[variable-2 + foo]", - "[variable-2 + bar]"); - - MT("listDash", - "foo", - "bar", - "", - "[variable-2 - foo]", - "[variable-2 - bar]"); - - MT("listNumber", - "foo", - "bar", - "", - "[variable-2 1. foo]", - "[variable-2 2. bar]"); - - // Lists require a preceding blank line (per Dingus) - MT("listBogus", - "foo", - "1. bar", - "2. hello"); - - // Formatting in lists (*) - MT("listAsteriskFormatting", - "[variable-2 * ][variable-2&em *foo*][variable-2 bar]", - "[variable-2 * ][variable-2&strong **foo**][variable-2 bar]", - "[variable-2 * ][variable-2&strong **][variable-2&em&strong *foo**][variable-2&em *][variable-2 bar]", - "[variable-2 * ][variable-2&comment `foo`][variable-2 bar]"); - - // Formatting in lists (+) - MT("listPlusFormatting", - "[variable-2 + ][variable-2&em *foo*][variable-2 bar]", - "[variable-2 + ][variable-2&strong **foo**][variable-2 bar]", - "[variable-2 + ][variable-2&strong **][variable-2&em&strong *foo**][variable-2&em *][variable-2 bar]", - "[variable-2 + ][variable-2&comment `foo`][variable-2 bar]"); - - // Formatting in lists (-) - MT("listDashFormatting", - "[variable-2 - ][variable-2&em *foo*][variable-2 bar]", - "[variable-2 - ][variable-2&strong **foo**][variable-2 bar]", - "[variable-2 - ][variable-2&strong **][variable-2&em&strong *foo**][variable-2&em *][variable-2 bar]", - "[variable-2 - ][variable-2&comment `foo`][variable-2 bar]"); - - // Formatting in lists (1.) - MT("listNumberFormatting", - "[variable-2 1. ][variable-2&em *foo*][variable-2 bar]", - "[variable-2 2. ][variable-2&strong **foo**][variable-2 bar]", - "[variable-2 3. ][variable-2&strong **][variable-2&em&strong *foo**][variable-2&em *][variable-2 bar]", - "[variable-2 4. ][variable-2&comment `foo`][variable-2 bar]"); - - // Paragraph lists - MT("listParagraph", - "[variable-2 * foo]", - "", - "[variable-2 * bar]"); - - // Multi-paragraph lists - // - // 4 spaces - MT("listMultiParagraph", - "[variable-2 * foo]", - "", - "[variable-2 * bar]", - "", - " [variable-2 hello]"); - - // 4 spaces, extra blank lines (should still be list, per Dingus) - MT("listMultiParagraphExtra", - "[variable-2 * foo]", - "", - "[variable-2 * bar]", - "", - "", - " [variable-2 hello]"); - - // 4 spaces, plus 1 space (should still be list, per Dingus) - MT("listMultiParagraphExtraSpace", - "[variable-2 * foo]", - "", - "[variable-2 * bar]", - "", - " [variable-2 hello]", - "", - " [variable-2 world]"); - - // 1 tab - MT("listTab", - "[variable-2 * foo]", - "", - "[variable-2 * bar]", - "", - "\t[variable-2 hello]"); - - // No indent - MT("listNoIndent", - "[variable-2 * foo]", - "", - "[variable-2 * bar]", - "", - "hello"); - - // Blockquote - MT("blockquote", - "[variable-2 * foo]", - "", - "[variable-2 * bar]", - "", - " [variable-2&atom > hello]"); - - // Code block - MT("blockquoteCode", - "[variable-2 * foo]", - "", - "[variable-2 * bar]", - "", - " [comment > hello]", - "", - " [variable-2 world]"); - - // Code block followed by text - MT("blockquoteCodeText", - "[variable-2 * foo]", - "", - " [variable-2 bar]", - "", - " [comment hello]", - "", - " [variable-2 world]"); - - // Nested list - - MT("listAsteriskNested", - "[variable-2 * foo]", - "", - " [variable-3 * bar]"); - - MT("listPlusNested", - "[variable-2 + foo]", - "", - " [variable-3 + bar]"); - - MT("listDashNested", - "[variable-2 - foo]", - "", - " [variable-3 - bar]"); - - MT("listNumberNested", - "[variable-2 1. foo]", - "", - " [variable-3 2. bar]"); - - MT("listMixed", - "[variable-2 * foo]", - "", - " [variable-3 + bar]", - "", - " [keyword - hello]", - "", - " [variable-2 1. world]"); - - MT("listBlockquote", - "[variable-2 * foo]", - "", - " [variable-3 + bar]", - "", - " [atom&variable-3 > hello]"); - - MT("listCode", - "[variable-2 * foo]", - "", - " [variable-3 + bar]", - "", - " [comment hello]"); - - // Code with internal indentation - MT("listCodeIndentation", - "[variable-2 * foo]", - "", - " [comment bar]", - " [comment hello]", - " [comment world]", - " [comment foo]", - " [variable-2 bar]"); - - // List nesting edge cases - MT("listNested", - "[variable-2 * foo]", - "", - " [variable-3 * bar]", - "", - " [variable-2 hello]" - ); - MT("listNested", - "[variable-2 * foo]", - "", - " [variable-3 * bar]", - "", - " [variable-3 * foo]" - ); - - // Code followed by text - MT("listCodeText", - "[variable-2 * foo]", - "", - " [comment bar]", - "", - "hello"); - - // Following tests directly from official Markdown documentation - // http://daringfireball.net/projects/markdown/syntax#hr - - MT("hrSpace", - "[hr * * *]"); - - MT("hr", - "[hr ***]"); - - MT("hrLong", - "[hr *****]"); - - MT("hrSpaceDash", - "[hr - - -]"); - - MT("hrDashLong", - "[hr ---------------------------------------]"); - - // Inline link with title - MT("linkTitle", - "[link [[foo]]][string (http://example.com/ \"bar\")] hello"); - - // Inline link without title - MT("linkNoTitle", - "[link [[foo]]][string (http://example.com/)] bar"); - - // Inline link with image - MT("linkImage", - "[link [[][tag ![[foo]]][string (http://example.com/)][link ]]][string (http://example.com/)] bar"); - - // Inline link with Em - MT("linkEm", - "[link [[][link&em *foo*][link ]]][string (http://example.com/)] bar"); - - // Inline link with Strong - MT("linkStrong", - "[link [[][link&strong **foo**][link ]]][string (http://example.com/)] bar"); - - // Inline link with EmStrong - MT("linkEmStrong", - "[link [[][link&strong **][link&em&strong *foo**][link&em *][link ]]][string (http://example.com/)] bar"); - - // Image with title - MT("imageTitle", - "[tag ![[foo]]][string (http://example.com/ \"bar\")] hello"); - - // Image without title - MT("imageNoTitle", - "[tag ![[foo]]][string (http://example.com/)] bar"); - - // Image with asterisks - MT("imageAsterisks", - "[tag ![[*foo*]]][string (http://example.com/)] bar"); - - // Not a link. Should be normal text due to square brackets being used - // regularly in text, especially in quoted material, and no space is allowed - // between square brackets and parentheses (per Dingus). - MT("notALink", - "[[foo]] (bar)"); - - // Reference-style links - MT("linkReference", - "[link [[foo]]][string [[bar]]] hello"); - - // Reference-style links with Em - MT("linkReferenceEm", - "[link [[][link&em *foo*][link ]]][string [[bar]]] hello"); - - // Reference-style links with Strong - MT("linkReferenceStrong", - "[link [[][link&strong **foo**][link ]]][string [[bar]]] hello"); - - // Reference-style links with EmStrong - MT("linkReferenceEmStrong", - "[link [[][link&strong **][link&em&strong *foo**][link&em *][link ]]][string [[bar]]] hello"); - - // Reference-style links with optional space separator (per docuentation) - // "You can optionally use a space to separate the sets of brackets" - MT("linkReferenceSpace", - "[link [[foo]]] [string [[bar]]] hello"); - - // Should only allow a single space ("...use *a* space...") - MT("linkReferenceDoubleSpace", - "[[foo]] [[bar]] hello"); - - // Reference-style links with implicit link name - MT("linkImplicit", - "[link [[foo]]][string [[]]] hello"); - - // @todo It would be nice if, at some point, the document was actually - // checked to see if the referenced link exists - - // Link label, for reference-style links (taken from documentation) - - MT("labelNoTitle", - "[link [[foo]]:] [string http://example.com/]"); - - MT("labelIndented", - " [link [[foo]]:] [string http://example.com/]"); - - MT("labelSpaceTitle", - "[link [[foo bar]]:] [string http://example.com/ \"hello\"]"); - - MT("labelDoubleTitle", - "[link [[foo bar]]:] [string http://example.com/ \"hello\"] \"world\""); - - MT("labelTitleDoubleQuotes", - "[link [[foo]]:] [string http://example.com/ \"bar\"]"); - - MT("labelTitleSingleQuotes", - "[link [[foo]]:] [string http://example.com/ 'bar']"); - - MT("labelTitleParenthese", - "[link [[foo]]:] [string http://example.com/ (bar)]"); - - MT("labelTitleInvalid", - "[link [[foo]]:] [string http://example.com/] bar"); - - MT("labelLinkAngleBrackets", - "[link [[foo]]:] [string \"bar\"]"); - - MT("labelTitleNextDoubleQuotes", - "[link [[foo]]:] [string http://example.com/]", - "[string \"bar\"] hello"); - - MT("labelTitleNextSingleQuotes", - "[link [[foo]]:] [string http://example.com/]", - "[string 'bar'] hello"); - - MT("labelTitleNextParenthese", - "[link [[foo]]:] [string http://example.com/]", - "[string (bar)] hello"); - - MT("labelTitleNextMixed", - "[link [[foo]]:] [string http://example.com/]", - "(bar\" hello"); - - MT("linkWeb", - "[link ] foo"); - - MT("linkEmail", - "[link ] foo"); - - MT("emAsterisk", - "[em *foo*] bar"); - - MT("emUnderscore", - "[em _foo_] bar"); - - MT("emInWordAsterisk", - "foo[em *bar*]hello"); - - MT("emInWordUnderscore", - "foo[em _bar_]hello"); - - // Per documentation: "...surround an * or _ with spaces, it’ll be - // treated as a literal asterisk or underscore." - - MT("emEscapedBySpaceIn", - "foo [em _bar _ hello_] world"); - - MT("emEscapedBySpaceOut", - "foo _ bar[em _hello_]world"); - - // Unclosed emphasis characters - // Instead of simply marking as EM / STRONG, it would be nice to have an - // incomplete flag for EM and STRONG, that is styled slightly different. - MT("emIncompleteAsterisk", - "foo [em *bar]"); - - MT("emIncompleteUnderscore", - "foo [em _bar]"); - - MT("strongAsterisk", - "[strong **foo**] bar"); - - MT("strongUnderscore", - "[strong __foo__] bar"); - - MT("emStrongAsterisk", - "[em *foo][em&strong **bar*][strong hello**] world"); - - MT("emStrongUnderscore", - "[em _foo][em&strong __bar_][strong hello__] world"); - - // "...same character must be used to open and close an emphasis span."" - MT("emStrongMixed", - "[em _foo][em&strong **bar*hello__ world]"); - - MT("emStrongMixed", - "[em *foo][em&strong __bar_hello** world]"); - - // These characters should be escaped: - // \ backslash - // ` backtick - // * asterisk - // _ underscore - // {} curly braces - // [] square brackets - // () parentheses - // # hash mark - // + plus sign - // - minus sign (hyphen) - // . dot - // ! exclamation mark - - MT("escapeBacktick", - "foo \\`bar\\`"); - - MT("doubleEscapeBacktick", - "foo \\\\[comment `bar\\\\`]"); - - MT("escapeAsterisk", - "foo \\*bar\\*"); - - MT("doubleEscapeAsterisk", - "foo \\\\[em *bar\\\\*]"); - - MT("escapeUnderscore", - "foo \\_bar\\_"); - - MT("doubleEscapeUnderscore", - "foo \\\\[em _bar\\\\_]"); - - MT("escapeHash", - "\\# foo"); - - MT("doubleEscapeHash", - "\\\\# foo"); - - - // Tests to make sure GFM-specific things aren't getting through - - MT("taskList", - "[variable-2 * [ ]] bar]"); - - MT("fencedCodeBlocks", - "[comment ```]", - "foo", - "[comment ```]"); -})(); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/meta.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/meta.js deleted file mode 100644 index e8cfc8fb63..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/meta.js +++ /dev/null @@ -1,77 +0,0 @@ -CodeMirror.modeInfo = [ - {name: 'APL', mime: 'text/apl', mode: 'apl'}, - {name: 'Asterisk', mime: 'text/x-asterisk', mode: 'asterisk'}, - {name: 'C', mime: 'text/x-csrc', mode: 'clike'}, - {name: 'C++', mime: 'text/x-c++src', mode: 'clike'}, - {name: 'Cobol', mime: 'text/x-cobol', mode: 'cobol'}, - {name: 'Java', mime: 'text/x-java', mode: 'clike'}, - {name: 'C#', mime: 'text/x-csharp', mode: 'clike'}, - {name: 'Scala', mime: 'text/x-scala', mode: 'clike'}, - {name: 'Clojure', mime: 'text/x-clojure', mode: 'clojure'}, - {name: 'CoffeeScript', mime: 'text/x-coffeescript', mode: 'coffeescript'}, - {name: 'Common Lisp', mime: 'text/x-common-lisp', mode: 'commonlisp'}, - {name: 'CSS', mime: 'text/css', mode: 'css'}, - {name: 'D', mime: 'text/x-d', mode: 'd'}, - {name: 'diff', mime: 'text/x-diff', mode: 'diff'}, - {name: 'ECL', mime: 'text/x-ecl', mode: 'ecl'}, - {name: 'Erlang', mime: 'text/x-erlang', mode: 'erlang'}, - {name: 'Gas', mime: 'text/x-gas', mode: 'gas'}, - {name: 'GitHub Flavored Markdown', mode: 'gfm'}, - {name: 'GO', mime: 'text/x-go', mode: 'go'}, - {name: 'Groovy', mime: 'text/x-groovy', mode: 'groovy'}, - {name: 'Haskell', mime: 'text/x-haskell', mode: 'haskell'}, - {name: 'Haxe', mime: 'text/x-haxe', mode: 'haxe'}, - {name: 'ASP.NET', mime: 'application/x-aspx', mode: 'htmlembedded'}, - {name: 'Embedded Javascript', mime: 'application/x-ejs', mode: 'htmlembedded'}, - {name: 'JavaServer Pages', mime: 'application/x-jsp', mode: 'htmlembedded'}, - {name: 'HTML', mime: 'text/html', mode: 'htmlmixed'}, - {name: 'HTTP', mime: 'message/http', mode: 'http'}, - {name: 'JavaScript', mime: 'text/javascript', mode: 'javascript'}, - {name: 'JSON', mime: 'application/x-json', mode: 'javascript'}, - {name: 'JSON', mime: 'application/json', mode: 'javascript'}, - {name: 'TypeScript', mime: 'application/typescript', mode: 'javascript'}, - {name: 'Jinja2', mime: 'jinja2', mode: 'jinja2'}, - {name: 'LESS', mime: 'text/x-less', mode: 'less'}, - {name: 'LiveScript', mime: 'text/x-livescript', mode: 'livescript'}, - {name: 'Lua', mime: 'text/x-lua', mode: 'lua'}, - {name: 'Markdown (GitHub-flavour)', mime: 'text/x-markdown', mode: 'markdown'}, - {name: 'mIRC', mime: 'text/mirc', mode: 'mirc'}, - {name: 'NTriples', mime: 'text/n-triples', mode: 'ntriples'}, - {name: 'OCaml', mime: 'text/x-ocaml', mode: 'ocaml'}, - {name: 'Pascal', mime: 'text/x-pascal', mode: 'pascal'}, - {name: 'Perl', mime: 'text/x-perl', mode: 'perl'}, - {name: 'PHP', mime: 'text/x-php', mode: 'php'}, - {name: 'PHP(HTML)', mime: 'application/x-httpd-php', mode: 'php'}, - {name: 'Pig', mime: 'text/x-pig', mode: 'pig'}, - {name: 'Plain Text', mime: 'text/plain', mode: 'null'}, - {name: 'Properties files', mime: 'text/x-properties', mode: 'clike'}, - {name: 'Python', mime: 'text/x-python', mode: 'python'}, - {name: 'R', mime: 'text/x-rsrc', mode: 'r'}, - {name: 'reStructuredText', mime: 'text/x-rst', mode: 'rst'}, - {name: 'Ruby', mime: 'text/x-ruby', mode: 'ruby'}, - {name: 'Rust', mime: 'text/x-rustsrc', mode: 'rust'}, - {name: 'Sass', mime: 'text/x-sass', mode: 'sass'}, - {name: 'Scheme', mime: 'text/x-scheme', mode: 'scheme'}, - {name: 'SCSS', mime: 'text/x-scss', mode: 'css'}, - {name: 'Shell', mime: 'text/x-sh', mode: 'shell'}, - {name: 'Sieve', mime: 'application/sieve', mode: 'sieve'}, - {name: 'Smalltalk', mime: 'text/x-stsrc', mode: 'smalltalk'}, - {name: 'Smarty', mime: 'text/x-smarty', mode: 'smarty'}, - {name: 'SPARQL', mime: 'application/x-sparql-query', mode: 'sparql'}, - {name: 'SQL', mime: 'text/x-sql', mode: 'sql'}, - {name: 'MariaDB', mime: 'text/x-mariadb', mode: 'sql'}, - {name: 'sTeX', mime: 'text/x-stex', mode: 'stex'}, - {name: 'LaTeX', mime: 'text/x-latex', mode: 'stex'}, - {name: 'Tcl', mime: 'text/x-tcl', mode: 'tcl'}, - {name: 'TiddlyWiki ', mime: 'text/x-tiddlywiki', mode: 'tiddlywiki'}, - {name: 'Tiki wiki', mime: 'text/tiki', mode: 'tiki'}, - {name: 'VB.NET', mime: 'text/x-vb', mode: 'vb'}, - {name: 'VBScript', mime: 'text/vbscript', mode: 'vbscript'}, - {name: 'Velocity', mime: 'text/velocity', mode: 'velocity'}, - {name: 'Verilog', mime: 'text/x-verilog', mode: 'verilog'}, - {name: 'XML', mime: 'application/xml', mode: 'xml'}, - {name: 'HTML', mime: 'text/html', mode: 'xml'}, - {name: 'XQuery', mime: 'application/xquery', mode: 'xquery'}, - {name: 'YAML', mime: 'text/x-yaml', mode: 'yaml'}, - {name: 'Z80', mime: 'text/x-z80', mode: 'z80'} -]; diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/mirc/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/mirc/index.html deleted file mode 100644 index 0ff5ec9a20..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/mirc/index.html +++ /dev/null @@ -1,149 +0,0 @@ - - - - - CodeMirror: mIRC mode - - - - - - - - -

CodeMirror: mIRC mode

-
- - -

MIME types defined: text/mirc.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/mirc/mirc.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/mirc/mirc.js deleted file mode 100644 index fc88bc56fb..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/mirc/mirc.js +++ /dev/null @@ -1,177 +0,0 @@ -//mIRC mode by Ford_Lawnmower :: Based on Velocity mode by Steve O'Hara -CodeMirror.defineMIME("text/mirc", "mirc"); -CodeMirror.defineMode("mirc", function() { - function parseWords(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - var specials = parseWords("$! $$ $& $? $+ $abook $abs $active $activecid " + - "$activewid $address $addtok $agent $agentname $agentstat $agentver " + - "$alias $and $anick $ansi2mirc $aop $appactive $appstate $asc $asctime " + - "$asin $atan $avoice $away $awaymsg $awaytime $banmask $base $bfind " + - "$binoff $biton $bnick $bvar $bytes $calc $cb $cd $ceil $chan $chanmodes " + - "$chantypes $chat $chr $cid $clevel $click $cmdbox $cmdline $cnick $color " + - "$com $comcall $comchan $comerr $compact $compress $comval $cos $count " + - "$cr $crc $creq $crlf $ctime $ctimer $ctrlenter $date $day $daylight " + - "$dbuh $dbuw $dccignore $dccport $dde $ddename $debug $decode $decompress " + - "$deltok $devent $dialog $did $didreg $didtok $didwm $disk $dlevel $dll " + - "$dllcall $dname $dns $duration $ebeeps $editbox $emailaddr $encode $error " + - "$eval $event $exist $feof $ferr $fgetc $file $filename $filtered $finddir " + - "$finddirn $findfile $findfilen $findtok $fline $floor $fopen $fread $fserve " + - "$fulladdress $fulldate $fullname $fullscreen $get $getdir $getdot $gettok $gmt " + - "$group $halted $hash $height $hfind $hget $highlight $hnick $hotline " + - "$hotlinepos $ial $ialchan $ibl $idle $iel $ifmatch $ignore $iif $iil " + - "$inelipse $ini $inmidi $inpaste $inpoly $input $inrect $inroundrect " + - "$insong $instok $int $inwave $ip $isalias $isbit $isdde $isdir $isfile " + - "$isid $islower $istok $isupper $keychar $keyrpt $keyval $knick $lactive " + - "$lactivecid $lactivewid $left $len $level $lf $line $lines $link $lock " + - "$lock $locked $log $logstamp $logstampfmt $longfn $longip $lower $ltimer " + - "$maddress $mask $matchkey $matchtok $md5 $me $menu $menubar $menucontext " + - "$menutype $mid $middir $mircdir $mircexe $mircini $mklogfn $mnick $mode " + - "$modefirst $modelast $modespl $mouse $msfile $network $newnick $nick $nofile " + - "$nopath $noqt $not $notags $notify $null $numeric $numok $oline $onpoly " + - "$opnick $or $ord $os $passivedcc $pic $play $pnick $port $portable $portfree " + - "$pos $prefix $prop $protect $puttok $qt $query $rand $r $rawmsg $read $readomo " + - "$readn $regex $regml $regsub $regsubex $remove $remtok $replace $replacex " + - "$reptok $result $rgb $right $round $scid $scon $script $scriptdir $scriptline " + - "$sdir $send $server $serverip $sfile $sha1 $shortfn $show $signal $sin " + - "$site $sline $snick $snicks $snotify $sock $sockbr $sockerr $sockname " + - "$sorttok $sound $sqrt $ssl $sreq $sslready $status $strip $str $stripped " + - "$syle $submenu $switchbar $tan $target $ticks $time $timer $timestamp " + - "$timestampfmt $timezone $tip $titlebar $toolbar $treebar $trust $ulevel " + - "$ulist $upper $uptime $url $usermode $v1 $v2 $var $vcmd $vcmdstat $vcmdver " + - "$version $vnick $vol $wid $width $wildsite $wildtok $window $wrap $xor"); - var keywords = parseWords("abook ajinvite alias aline ame amsg anick aop auser autojoin avoice " + - "away background ban bcopy beep bread break breplace bset btrunc bunset bwrite " + - "channel clear clearall cline clipboard close cnick color comclose comopen " + - "comreg continue copy creq ctcpreply ctcps dcc dccserver dde ddeserver " + - "debug dec describe dialog did didtok disable disconnect dlevel dline dll " + - "dns dqwindow drawcopy drawdot drawfill drawline drawpic drawrect drawreplace " + - "drawrot drawsave drawscroll drawtext ebeeps echo editbox emailaddr enable " + - "events exit fclose filter findtext finger firewall flash flist flood flush " + - "flushini font fopen fseek fsend fserve fullname fwrite ghide gload gmove " + - "gopts goto gplay gpoint gqreq groups gshow gsize gstop gtalk gunload hadd " + - "halt haltdef hdec hdel help hfree hinc hload hmake hop hsave ial ialclear " + - "ialmark identd if ignore iline inc invite iuser join kick linesep links list " + - "load loadbuf localinfo log mdi me menubar mkdir mnick mode msg nick noop notice " + - "notify omsg onotice part partall pdcc perform play playctrl pop protect pvoice " + - "qme qmsg query queryn quit raw reload remini remote remove rename renwin " + - "reseterror resetidle return rlevel rline rmdir run ruser save savebuf saveini " + - "say scid scon server set showmirc signam sline sockaccept sockclose socklist " + - "socklisten sockmark sockopen sockpause sockread sockrename sockudp sockwrite " + - "sound speak splay sreq strip switchbar timer timestamp titlebar tnick tokenize " + - "toolbar topic tray treebar ulist unload unset unsetall updatenl url uwho " + - "var vcadd vcmd vcrem vol while whois window winhelp write writeint if isalnum " + - "isalpha isaop isavoice isban ischan ishop isignore isin isincs isletter islower " + - "isnotify isnum ison isop isprotect isreg isupper isvoice iswm iswmcs " + - "elseif else goto menu nicklist status title icon size option text edit " + - "button check radio box scroll list combo link tab item"); - var functions = parseWords("if elseif else and not or eq ne in ni for foreach while switch"); - var isOperatorChar = /[+\-*&%=<>!?^\/\|]/; - function chain(stream, state, f) { - state.tokenize = f; - return f(stream, state); - } - function tokenBase(stream, state) { - var beforeParams = state.beforeParams; - state.beforeParams = false; - var ch = stream.next(); - if (/[\[\]{}\(\),\.]/.test(ch)) { - if (ch == "(" && beforeParams) state.inParams = true; - else if (ch == ")") state.inParams = false; - return null; - } - else if (/\d/.test(ch)) { - stream.eatWhile(/[\w\.]/); - return "number"; - } - else if (ch == "\\") { - stream.eat("\\"); - stream.eat(/./); - return "number"; - } - else if (ch == "/" && stream.eat("*")) { - return chain(stream, state, tokenComment); - } - else if (ch == ";" && stream.match(/ *\( *\(/)) { - return chain(stream, state, tokenUnparsed); - } - else if (ch == ";" && !state.inParams) { - stream.skipToEnd(); - return "comment"; - } - else if (ch == '"') { - stream.eat(/"/); - return "keyword"; - } - else if (ch == "$") { - stream.eatWhile(/[$_a-z0-9A-Z\.:]/); - if (specials && specials.propertyIsEnumerable(stream.current().toLowerCase())) { - return "keyword"; - } - else { - state.beforeParams = true; - return "builtin"; - } - } - else if (ch == "%") { - stream.eatWhile(/[^,^\s^\(^\)]/); - state.beforeParams = true; - return "string"; - } - else if (isOperatorChar.test(ch)) { - stream.eatWhile(isOperatorChar); - return "operator"; - } - else { - stream.eatWhile(/[\w\$_{}]/); - var word = stream.current().toLowerCase(); - if (keywords && keywords.propertyIsEnumerable(word)) - return "keyword"; - if (functions && functions.propertyIsEnumerable(word)) { - state.beforeParams = true; - return "keyword"; - } - return null; - } - } - function tokenComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == "/" && maybeEnd) { - state.tokenize = tokenBase; - break; - } - maybeEnd = (ch == "*"); - } - return "comment"; - } - function tokenUnparsed(stream, state) { - var maybeEnd = 0, ch; - while (ch = stream.next()) { - if (ch == ";" && maybeEnd == 2) { - state.tokenize = tokenBase; - break; - } - if (ch == ")") - maybeEnd++; - else if (ch != " ") - maybeEnd = 0; - } - return "meta"; - } - return { - startState: function() { - return { - tokenize: tokenBase, - beforeParams: false, - inParams: false - }; - }, - token: function(stream, state) { - if (stream.eatSpace()) return null; - return state.tokenize(stream, state); - } - }; -}); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ntriples/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/ntriples/index.html deleted file mode 100644 index 052a53d86a..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ntriples/index.html +++ /dev/null @@ -1,33 +0,0 @@ - - - - - CodeMirror: NTriples mode - - - - - - - -

CodeMirror: NTriples mode

-
- -
- - -

MIME types defined: text/n-triples.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ntriples/ntriples.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/ntriples/ntriples.js deleted file mode 100644 index ed0cee34b0..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ntriples/ntriples.js +++ /dev/null @@ -1,170 +0,0 @@ -/********************************************************** -* This script provides syntax highlighting support for -* the Ntriples format. -* Ntriples format specification: -* http://www.w3.org/TR/rdf-testcases/#ntriples -***********************************************************/ - -/* - The following expression defines the defined ASF grammar transitions. - - pre_subject -> - { - ( writing_subject_uri | writing_bnode_uri ) - -> pre_predicate - -> writing_predicate_uri - -> pre_object - -> writing_object_uri | writing_object_bnode | - ( - writing_object_literal - -> writing_literal_lang | writing_literal_type - ) - -> post_object - -> BEGIN - } otherwise { - -> ERROR - } -*/ -CodeMirror.defineMode("ntriples", function() { - - var Location = { - PRE_SUBJECT : 0, - WRITING_SUB_URI : 1, - WRITING_BNODE_URI : 2, - PRE_PRED : 3, - WRITING_PRED_URI : 4, - PRE_OBJ : 5, - WRITING_OBJ_URI : 6, - WRITING_OBJ_BNODE : 7, - WRITING_OBJ_LITERAL : 8, - WRITING_LIT_LANG : 9, - WRITING_LIT_TYPE : 10, - POST_OBJ : 11, - ERROR : 12 - }; - function transitState(currState, c) { - var currLocation = currState.location; - var ret; - - // Opening. - if (currLocation == Location.PRE_SUBJECT && c == '<') ret = Location.WRITING_SUB_URI; - else if(currLocation == Location.PRE_SUBJECT && c == '_') ret = Location.WRITING_BNODE_URI; - else if(currLocation == Location.PRE_PRED && c == '<') ret = Location.WRITING_PRED_URI; - else if(currLocation == Location.PRE_OBJ && c == '<') ret = Location.WRITING_OBJ_URI; - else if(currLocation == Location.PRE_OBJ && c == '_') ret = Location.WRITING_OBJ_BNODE; - else if(currLocation == Location.PRE_OBJ && c == '"') ret = Location.WRITING_OBJ_LITERAL; - - // Closing. - else if(currLocation == Location.WRITING_SUB_URI && c == '>') ret = Location.PRE_PRED; - else if(currLocation == Location.WRITING_BNODE_URI && c == ' ') ret = Location.PRE_PRED; - else if(currLocation == Location.WRITING_PRED_URI && c == '>') ret = Location.PRE_OBJ; - else if(currLocation == Location.WRITING_OBJ_URI && c == '>') ret = Location.POST_OBJ; - else if(currLocation == Location.WRITING_OBJ_BNODE && c == ' ') ret = Location.POST_OBJ; - else if(currLocation == Location.WRITING_OBJ_LITERAL && c == '"') ret = Location.POST_OBJ; - else if(currLocation == Location.WRITING_LIT_LANG && c == ' ') ret = Location.POST_OBJ; - else if(currLocation == Location.WRITING_LIT_TYPE && c == '>') ret = Location.POST_OBJ; - - // Closing typed and language literal. - else if(currLocation == Location.WRITING_OBJ_LITERAL && c == '@') ret = Location.WRITING_LIT_LANG; - else if(currLocation == Location.WRITING_OBJ_LITERAL && c == '^') ret = Location.WRITING_LIT_TYPE; - - // Spaces. - else if( c == ' ' && - ( - currLocation == Location.PRE_SUBJECT || - currLocation == Location.PRE_PRED || - currLocation == Location.PRE_OBJ || - currLocation == Location.POST_OBJ - ) - ) ret = currLocation; - - // Reset. - else if(currLocation == Location.POST_OBJ && c == '.') ret = Location.PRE_SUBJECT; - - // Error - else ret = Location.ERROR; - - currState.location=ret; - } - - return { - startState: function() { - return { - location : Location.PRE_SUBJECT, - uris : [], - anchors : [], - bnodes : [], - langs : [], - types : [] - }; - }, - token: function(stream, state) { - var ch = stream.next(); - if(ch == '<') { - transitState(state, ch); - var parsedURI = ''; - stream.eatWhile( function(c) { if( c != '#' && c != '>' ) { parsedURI += c; return true; } return false;} ); - state.uris.push(parsedURI); - if( stream.match('#', false) ) return 'variable'; - stream.next(); - transitState(state, '>'); - return 'variable'; - } - if(ch == '#') { - var parsedAnchor = ''; - stream.eatWhile(function(c) { if(c != '>' && c != ' ') { parsedAnchor+= c; return true; } return false;}); - state.anchors.push(parsedAnchor); - return 'variable-2'; - } - if(ch == '>') { - transitState(state, '>'); - return 'variable'; - } - if(ch == '_') { - transitState(state, ch); - var parsedBNode = ''; - stream.eatWhile(function(c) { if( c != ' ' ) { parsedBNode += c; return true; } return false;}); - state.bnodes.push(parsedBNode); - stream.next(); - transitState(state, ' '); - return 'builtin'; - } - if(ch == '"') { - transitState(state, ch); - stream.eatWhile( function(c) { return c != '"'; } ); - stream.next(); - if( stream.peek() != '@' && stream.peek() != '^' ) { - transitState(state, '"'); - } - return 'string'; - } - if( ch == '@' ) { - transitState(state, '@'); - var parsedLang = ''; - stream.eatWhile(function(c) { if( c != ' ' ) { parsedLang += c; return true; } return false;}); - state.langs.push(parsedLang); - stream.next(); - transitState(state, ' '); - return 'string-2'; - } - if( ch == '^' ) { - stream.next(); - transitState(state, '^'); - var parsedType = ''; - stream.eatWhile(function(c) { if( c != '>' ) { parsedType += c; return true; } return false;} ); - state.types.push(parsedType); - stream.next(); - transitState(state, '>'); - return 'variable'; - } - if( ch == ' ' ) { - transitState(state, ch); - } - if( ch == '.' ) { - transitState(state, ch); - } - } - }; -}); - -CodeMirror.defineMIME("text/n-triples", "ntriples"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ocaml/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/ocaml/index.html deleted file mode 100644 index c10a84fade..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ocaml/index.html +++ /dev/null @@ -1,131 +0,0 @@ - - -CodeMirror: OCaml mode - - - - - - - - - - -

CodeMirror: OCaml mode

- - - - - -

MIME types defined: text/x-ocaml.

diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ocaml/ocaml.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/ocaml/ocaml.js deleted file mode 100644 index 32cbc0b7ba..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ocaml/ocaml.js +++ /dev/null @@ -1,116 +0,0 @@ -CodeMirror.defineMode('ocaml', function() { - - var words = { - 'true': 'atom', - 'false': 'atom', - 'let': 'keyword', - 'rec': 'keyword', - 'in': 'keyword', - 'of': 'keyword', - 'and': 'keyword', - 'succ': 'keyword', - 'if': 'keyword', - 'then': 'keyword', - 'else': 'keyword', - 'for': 'keyword', - 'to': 'keyword', - 'while': 'keyword', - 'do': 'keyword', - 'done': 'keyword', - 'fun': 'keyword', - 'function': 'keyword', - 'val': 'keyword', - 'type': 'keyword', - 'mutable': 'keyword', - 'match': 'keyword', - 'with': 'keyword', - 'try': 'keyword', - 'raise': 'keyword', - 'begin': 'keyword', - 'end': 'keyword', - 'open': 'builtin', - 'trace': 'builtin', - 'ignore': 'builtin', - 'exit': 'builtin', - 'print_string': 'builtin', - 'print_endline': 'builtin' - }; - - function tokenBase(stream, state) { - var ch = stream.next(); - - if (ch === '"') { - state.tokenize = tokenString; - return state.tokenize(stream, state); - } - if (ch === '(') { - if (stream.eat('*')) { - state.commentLevel++; - state.tokenize = tokenComment; - return state.tokenize(stream, state); - } - } - if (ch === '~') { - stream.eatWhile(/\w/); - return 'variable-2'; - } - if (ch === '`') { - stream.eatWhile(/\w/); - return 'quote'; - } - if (/\d/.test(ch)) { - stream.eatWhile(/[\d]/); - if (stream.eat('.')) { - stream.eatWhile(/[\d]/); - } - return 'number'; - } - if ( /[+\-*&%=<>!?|]/.test(ch)) { - return 'operator'; - } - stream.eatWhile(/\w/); - var cur = stream.current(); - return words[cur] || 'variable'; - } - - function tokenString(stream, state) { - var next, end = false, escaped = false; - while ((next = stream.next()) != null) { - if (next === '"' && !escaped) { - end = true; - break; - } - escaped = !escaped && next === '\\'; - } - if (end && !escaped) { - state.tokenize = tokenBase; - } - return 'string'; - }; - - function tokenComment(stream, state) { - var prev, next; - while(state.commentLevel > 0 && (next = stream.next()) != null) { - if (prev === '(' && next === '*') state.commentLevel++; - if (prev === '*' && next === ')') state.commentLevel--; - prev = next; - } - if (state.commentLevel <= 0) { - state.tokenize = tokenBase; - } - return 'comment'; - } - - return { - startState: function() {return {tokenize: tokenBase, commentLevel: 0};}, - token: function(stream, state) { - if (stream.eatSpace()) return null; - return state.tokenize(stream, state); - }, - - blockCommentStart: "(*", - blockCommentEnd: "*)" - }; -}); - -CodeMirror.defineMIME('text/x-ocaml', 'ocaml'); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/LICENSE b/wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/LICENSE deleted file mode 100644 index 8e3747e748..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/LICENSE +++ /dev/null @@ -1,7 +0,0 @@ -Copyright (c) 2011 souceLair - -Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/index.html deleted file mode 100644 index b3016afb10..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/index.html +++ /dev/null @@ -1,48 +0,0 @@ - - - - - CodeMirror: Pascal mode - - - - - - - -

CodeMirror: Pascal mode

- -
- - - -

MIME types defined: text/x-pascal.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/pascal.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/pascal.js deleted file mode 100644 index 09d9b0617b..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pascal/pascal.js +++ /dev/null @@ -1,94 +0,0 @@ -CodeMirror.defineMode("pascal", function() { - function words(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - var keywords = words("and array begin case const div do downto else end file for forward integer " + - "boolean char function goto if in label mod nil not of or packed procedure " + - "program record repeat set string then to type until var while with"); - var atoms = {"null": true}; - - var isOperatorChar = /[+\-*&%=<>!?|\/]/; - - function tokenBase(stream, state) { - var ch = stream.next(); - if (ch == "#" && state.startOfLine) { - stream.skipToEnd(); - return "meta"; - } - if (ch == '"' || ch == "'") { - state.tokenize = tokenString(ch); - return state.tokenize(stream, state); - } - if (ch == "(" && stream.eat("*")) { - state.tokenize = tokenComment; - return tokenComment(stream, state); - } - if (/[\[\]{}\(\),;\:\.]/.test(ch)) { - return null; - } - if (/\d/.test(ch)) { - stream.eatWhile(/[\w\.]/); - return "number"; - } - if (ch == "/") { - if (stream.eat("/")) { - stream.skipToEnd(); - return "comment"; - } - } - if (isOperatorChar.test(ch)) { - stream.eatWhile(isOperatorChar); - return "operator"; - } - stream.eatWhile(/[\w\$_]/); - var cur = stream.current(); - if (keywords.propertyIsEnumerable(cur)) return "keyword"; - if (atoms.propertyIsEnumerable(cur)) return "atom"; - return "variable"; - } - - function tokenString(quote) { - return function(stream, state) { - var escaped = false, next, end = false; - while ((next = stream.next()) != null) { - if (next == quote && !escaped) {end = true; break;} - escaped = !escaped && next == "\\"; - } - if (end || !escaped) state.tokenize = null; - return "string"; - }; - } - - function tokenComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == ")" && maybeEnd) { - state.tokenize = null; - break; - } - maybeEnd = (ch == "*"); - } - return "comment"; - } - - // Interface - - return { - startState: function() { - return {tokenize: null}; - }, - - token: function(stream, state) { - if (stream.eatSpace()) return null; - var style = (state.tokenize || tokenBase)(stream, state); - if (style == "comment" || style == "meta") return style; - return style; - }, - - electricChars: "{}" - }; -}); - -CodeMirror.defineMIME("text/x-pascal", "pascal"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/LICENSE b/wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/LICENSE deleted file mode 100644 index 96f4115af8..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/LICENSE +++ /dev/null @@ -1,19 +0,0 @@ -Copyright (C) 2011 by Sabaca under the MIT license. - -Permission is hereby granted, free of charge, to any person obtaining a copy -of this software and associated documentation files (the "Software"), to deal -in the Software without restriction, including without limitation the rights -to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -copies of the Software, and to permit persons to whom the Software is -furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in -all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -THE SOFTWARE. diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/index.html deleted file mode 100644 index 13c7af64d3..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/index.html +++ /dev/null @@ -1,62 +0,0 @@ - - - - - CodeMirror: Perl mode - - - - - - - -

CodeMirror: Perl mode

- -
- - - -

MIME types defined: text/x-perl.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/perl.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/perl.js deleted file mode 100644 index 5954b1a61c..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/perl/perl.js +++ /dev/null @@ -1,816 +0,0 @@ -// CodeMirror2 mode/perl/perl.js (text/x-perl) beta 0.10 (2011-11-08) -// This is a part of CodeMirror from https://github.com/sabaca/CodeMirror_mode_perl (mail@sabaca.com) -CodeMirror.defineMode("perl",function(){ - // http://perldoc.perl.org - var PERL={ // null - magic touch - // 1 - keyword - // 2 - def - // 3 - atom - // 4 - operator - // 5 - variable-2 (predefined) - // [x,y] - x=1,2,3; y=must be defined if x{...} - // PERL operators - '->' : 4, - '++' : 4, - '--' : 4, - '**' : 4, - // ! ~ \ and unary + and - - '=~' : 4, - '!~' : 4, - '*' : 4, - '/' : 4, - '%' : 4, - 'x' : 4, - '+' : 4, - '-' : 4, - '.' : 4, - '<<' : 4, - '>>' : 4, - // named unary operators - '<' : 4, - '>' : 4, - '<=' : 4, - '>=' : 4, - 'lt' : 4, - 'gt' : 4, - 'le' : 4, - 'ge' : 4, - '==' : 4, - '!=' : 4, - '<=>' : 4, - 'eq' : 4, - 'ne' : 4, - 'cmp' : 4, - '~~' : 4, - '&' : 4, - '|' : 4, - '^' : 4, - '&&' : 4, - '||' : 4, - '//' : 4, - '..' : 4, - '...' : 4, - '?' : 4, - ':' : 4, - '=' : 4, - '+=' : 4, - '-=' : 4, - '*=' : 4, // etc. ??? - ',' : 4, - '=>' : 4, - '::' : 4, - // list operators (rightward) - 'not' : 4, - 'and' : 4, - 'or' : 4, - 'xor' : 4, - // PERL predefined variables (I know, what this is a paranoid idea, but may be needed for people, who learn PERL, and for me as well, ...and may be for you?;) - 'BEGIN' : [5,1], - 'END' : [5,1], - 'PRINT' : [5,1], - 'PRINTF' : [5,1], - 'GETC' : [5,1], - 'READ' : [5,1], - 'READLINE' : [5,1], - 'DESTROY' : [5,1], - 'TIE' : [5,1], - 'TIEHANDLE' : [5,1], - 'UNTIE' : [5,1], - 'STDIN' : 5, - 'STDIN_TOP' : 5, - 'STDOUT' : 5, - 'STDOUT_TOP' : 5, - 'STDERR' : 5, - 'STDERR_TOP' : 5, - '$ARG' : 5, - '$_' : 5, - '@ARG' : 5, - '@_' : 5, - '$LIST_SEPARATOR' : 5, - '$"' : 5, - '$PROCESS_ID' : 5, - '$PID' : 5, - '$$' : 5, - '$REAL_GROUP_ID' : 5, - '$GID' : 5, - '$(' : 5, - '$EFFECTIVE_GROUP_ID' : 5, - '$EGID' : 5, - '$)' : 5, - '$PROGRAM_NAME' : 5, - '$0' : 5, - '$SUBSCRIPT_SEPARATOR' : 5, - '$SUBSEP' : 5, - '$;' : 5, - '$REAL_USER_ID' : 5, - '$UID' : 5, - '$<' : 5, - '$EFFECTIVE_USER_ID' : 5, - '$EUID' : 5, - '$>' : 5, - '$a' : 5, - '$b' : 5, - '$COMPILING' : 5, - '$^C' : 5, - '$DEBUGGING' : 5, - '$^D' : 5, - '${^ENCODING}' : 5, - '$ENV' : 5, - '%ENV' : 5, - '$SYSTEM_FD_MAX' : 5, - '$^F' : 5, - '@F' : 5, - '${^GLOBAL_PHASE}' : 5, - '$^H' : 5, - '%^H' : 5, - '@INC' : 5, - '%INC' : 5, - '$INPLACE_EDIT' : 5, - '$^I' : 5, - '$^M' : 5, - '$OSNAME' : 5, - '$^O' : 5, - '${^OPEN}' : 5, - '$PERLDB' : 5, - '$^P' : 5, - '$SIG' : 5, - '%SIG' : 5, - '$BASETIME' : 5, - '$^T' : 5, - '${^TAINT}' : 5, - '${^UNICODE}' : 5, - '${^UTF8CACHE}' : 5, - '${^UTF8LOCALE}' : 5, - '$PERL_VERSION' : 5, - '$^V' : 5, - '${^WIN32_SLOPPY_STAT}' : 5, - '$EXECUTABLE_NAME' : 5, - '$^X' : 5, - '$1' : 5, // - regexp $1, $2... - '$MATCH' : 5, - '$&' : 5, - '${^MATCH}' : 5, - '$PREMATCH' : 5, - '$`' : 5, - '${^PREMATCH}' : 5, - '$POSTMATCH' : 5, - "$'" : 5, - '${^POSTMATCH}' : 5, - '$LAST_PAREN_MATCH' : 5, - '$+' : 5, - '$LAST_SUBMATCH_RESULT' : 5, - '$^N' : 5, - '@LAST_MATCH_END' : 5, - '@+' : 5, - '%LAST_PAREN_MATCH' : 5, - '%+' : 5, - '@LAST_MATCH_START' : 5, - '@-' : 5, - '%LAST_MATCH_START' : 5, - '%-' : 5, - '$LAST_REGEXP_CODE_RESULT' : 5, - '$^R' : 5, - '${^RE_DEBUG_FLAGS}' : 5, - '${^RE_TRIE_MAXBUF}' : 5, - '$ARGV' : 5, - '@ARGV' : 5, - 'ARGV' : 5, - 'ARGVOUT' : 5, - '$OUTPUT_FIELD_SEPARATOR' : 5, - '$OFS' : 5, - '$,' : 5, - '$INPUT_LINE_NUMBER' : 5, - '$NR' : 5, - '$.' : 5, - '$INPUT_RECORD_SEPARATOR' : 5, - '$RS' : 5, - '$/' : 5, - '$OUTPUT_RECORD_SEPARATOR' : 5, - '$ORS' : 5, - '$\\' : 5, - '$OUTPUT_AUTOFLUSH' : 5, - '$|' : 5, - '$ACCUMULATOR' : 5, - '$^A' : 5, - '$FORMAT_FORMFEED' : 5, - '$^L' : 5, - '$FORMAT_PAGE_NUMBER' : 5, - '$%' : 5, - '$FORMAT_LINES_LEFT' : 5, - '$-' : 5, - '$FORMAT_LINE_BREAK_CHARACTERS' : 5, - '$:' : 5, - '$FORMAT_LINES_PER_PAGE' : 5, - '$=' : 5, - '$FORMAT_TOP_NAME' : 5, - '$^' : 5, - '$FORMAT_NAME' : 5, - '$~' : 5, - '${^CHILD_ERROR_NATIVE}' : 5, - '$EXTENDED_OS_ERROR' : 5, - '$^E' : 5, - '$EXCEPTIONS_BEING_CAUGHT' : 5, - '$^S' : 5, - '$WARNING' : 5, - '$^W' : 5, - '${^WARNING_BITS}' : 5, - '$OS_ERROR' : 5, - '$ERRNO' : 5, - '$!' : 5, - '%OS_ERROR' : 5, - '%ERRNO' : 5, - '%!' : 5, - '$CHILD_ERROR' : 5, - '$?' : 5, - '$EVAL_ERROR' : 5, - '$@' : 5, - '$OFMT' : 5, - '$#' : 5, - '$*' : 5, - '$ARRAY_BASE' : 5, - '$[' : 5, - '$OLD_PERL_VERSION' : 5, - '$]' : 5, - // PERL blocks - 'if' :[1,1], - elsif :[1,1], - 'else' :[1,1], - 'while' :[1,1], - unless :[1,1], - 'for' :[1,1], - foreach :[1,1], - // PERL functions - 'abs' :1, // - absolute value function - accept :1, // - accept an incoming socket connect - alarm :1, // - schedule a SIGALRM - 'atan2' :1, // - arctangent of Y/X in the range -PI to PI - bind :1, // - binds an address to a socket - binmode :1, // - prepare binary files for I/O - bless :1, // - create an object - bootstrap :1, // - 'break' :1, // - break out of a "given" block - caller :1, // - get context of the current subroutine call - chdir :1, // - change your current working directory - chmod :1, // - changes the permissions on a list of files - chomp :1, // - remove a trailing record separator from a string - chop :1, // - remove the last character from a string - chown :1, // - change the owership on a list of files - chr :1, // - get character this number represents - chroot :1, // - make directory new root for path lookups - close :1, // - close file (or pipe or socket) handle - closedir :1, // - close directory handle - connect :1, // - connect to a remote socket - 'continue' :[1,1], // - optional trailing block in a while or foreach - 'cos' :1, // - cosine function - crypt :1, // - one-way passwd-style encryption - dbmclose :1, // - breaks binding on a tied dbm file - dbmopen :1, // - create binding on a tied dbm file - 'default' :1, // - defined :1, // - test whether a value, variable, or function is defined - 'delete' :1, // - deletes a value from a hash - die :1, // - raise an exception or bail out - 'do' :1, // - turn a BLOCK into a TERM - dump :1, // - create an immediate core dump - each :1, // - retrieve the next key/value pair from a hash - endgrent :1, // - be done using group file - endhostent :1, // - be done using hosts file - endnetent :1, // - be done using networks file - endprotoent :1, // - be done using protocols file - endpwent :1, // - be done using passwd file - endservent :1, // - be done using services file - eof :1, // - test a filehandle for its end - 'eval' :1, // - catch exceptions or compile and run code - 'exec' :1, // - abandon this program to run another - exists :1, // - test whether a hash key is present - exit :1, // - terminate this program - 'exp' :1, // - raise I to a power - fcntl :1, // - file control system call - fileno :1, // - return file descriptor from filehandle - flock :1, // - lock an entire file with an advisory lock - fork :1, // - create a new process just like this one - format :1, // - declare a picture format with use by the write() function - formline :1, // - internal function used for formats - getc :1, // - get the next character from the filehandle - getgrent :1, // - get next group record - getgrgid :1, // - get group record given group user ID - getgrnam :1, // - get group record given group name - gethostbyaddr :1, // - get host record given its address - gethostbyname :1, // - get host record given name - gethostent :1, // - get next hosts record - getlogin :1, // - return who logged in at this tty - getnetbyaddr :1, // - get network record given its address - getnetbyname :1, // - get networks record given name - getnetent :1, // - get next networks record - getpeername :1, // - find the other end of a socket connection - getpgrp :1, // - get process group - getppid :1, // - get parent process ID - getpriority :1, // - get current nice value - getprotobyname :1, // - get protocol record given name - getprotobynumber :1, // - get protocol record numeric protocol - getprotoent :1, // - get next protocols record - getpwent :1, // - get next passwd record - getpwnam :1, // - get passwd record given user login name - getpwuid :1, // - get passwd record given user ID - getservbyname :1, // - get services record given its name - getservbyport :1, // - get services record given numeric port - getservent :1, // - get next services record - getsockname :1, // - retrieve the sockaddr for a given socket - getsockopt :1, // - get socket options on a given socket - given :1, // - glob :1, // - expand filenames using wildcards - gmtime :1, // - convert UNIX time into record or string using Greenwich time - 'goto' :1, // - create spaghetti code - grep :1, // - locate elements in a list test true against a given criterion - hex :1, // - convert a string to a hexadecimal number - 'import' :1, // - patch a module's namespace into your own - index :1, // - find a substring within a string - 'int' :1, // - get the integer portion of a number - ioctl :1, // - system-dependent device control system call - 'join' :1, // - join a list into a string using a separator - keys :1, // - retrieve list of indices from a hash - kill :1, // - send a signal to a process or process group - last :1, // - exit a block prematurely - lc :1, // - return lower-case version of a string - lcfirst :1, // - return a string with just the next letter in lower case - length :1, // - return the number of bytes in a string - 'link' :1, // - create a hard link in the filesytem - listen :1, // - register your socket as a server - local : 2, // - create a temporary value for a global variable (dynamic scoping) - localtime :1, // - convert UNIX time into record or string using local time - lock :1, // - get a thread lock on a variable, subroutine, or method - 'log' :1, // - retrieve the natural logarithm for a number - lstat :1, // - stat a symbolic link - m :null, // - match a string with a regular expression pattern - map :1, // - apply a change to a list to get back a new list with the changes - mkdir :1, // - create a directory - msgctl :1, // - SysV IPC message control operations - msgget :1, // - get SysV IPC message queue - msgrcv :1, // - receive a SysV IPC message from a message queue - msgsnd :1, // - send a SysV IPC message to a message queue - my : 2, // - declare and assign a local variable (lexical scoping) - 'new' :1, // - next :1, // - iterate a block prematurely - no :1, // - unimport some module symbols or semantics at compile time - oct :1, // - convert a string to an octal number - open :1, // - open a file, pipe, or descriptor - opendir :1, // - open a directory - ord :1, // - find a character's numeric representation - our : 2, // - declare and assign a package variable (lexical scoping) - pack :1, // - convert a list into a binary representation - 'package' :1, // - declare a separate global namespace - pipe :1, // - open a pair of connected filehandles - pop :1, // - remove the last element from an array and return it - pos :1, // - find or set the offset for the last/next m//g search - print :1, // - output a list to a filehandle - printf :1, // - output a formatted list to a filehandle - prototype :1, // - get the prototype (if any) of a subroutine - push :1, // - append one or more elements to an array - q :null, // - singly quote a string - qq :null, // - doubly quote a string - qr :null, // - Compile pattern - quotemeta :null, // - quote regular expression magic characters - qw :null, // - quote a list of words - qx :null, // - backquote quote a string - rand :1, // - retrieve the next pseudorandom number - read :1, // - fixed-length buffered input from a filehandle - readdir :1, // - get a directory from a directory handle - readline :1, // - fetch a record from a file - readlink :1, // - determine where a symbolic link is pointing - readpipe :1, // - execute a system command and collect standard output - recv :1, // - receive a message over a Socket - redo :1, // - start this loop iteration over again - ref :1, // - find out the type of thing being referenced - rename :1, // - change a filename - require :1, // - load in external functions from a library at runtime - reset :1, // - clear all variables of a given name - 'return' :1, // - get out of a function early - reverse :1, // - flip a string or a list - rewinddir :1, // - reset directory handle - rindex :1, // - right-to-left substring search - rmdir :1, // - remove a directory - s :null, // - replace a pattern with a string - say :1, // - print with newline - scalar :1, // - force a scalar context - seek :1, // - reposition file pointer for random-access I/O - seekdir :1, // - reposition directory pointer - select :1, // - reset default output or do I/O multiplexing - semctl :1, // - SysV semaphore control operations - semget :1, // - get set of SysV semaphores - semop :1, // - SysV semaphore operations - send :1, // - send a message over a socket - setgrent :1, // - prepare group file for use - sethostent :1, // - prepare hosts file for use - setnetent :1, // - prepare networks file for use - setpgrp :1, // - set the process group of a process - setpriority :1, // - set a process's nice value - setprotoent :1, // - prepare protocols file for use - setpwent :1, // - prepare passwd file for use - setservent :1, // - prepare services file for use - setsockopt :1, // - set some socket options - shift :1, // - remove the first element of an array, and return it - shmctl :1, // - SysV shared memory operations - shmget :1, // - get SysV shared memory segment identifier - shmread :1, // - read SysV shared memory - shmwrite :1, // - write SysV shared memory - shutdown :1, // - close down just half of a socket connection - 'sin' :1, // - return the sine of a number - sleep :1, // - block for some number of seconds - socket :1, // - create a socket - socketpair :1, // - create a pair of sockets - 'sort' :1, // - sort a list of values - splice :1, // - add or remove elements anywhere in an array - 'split' :1, // - split up a string using a regexp delimiter - sprintf :1, // - formatted print into a string - 'sqrt' :1, // - square root function - srand :1, // - seed the random number generator - stat :1, // - get a file's status information - state :1, // - declare and assign a state variable (persistent lexical scoping) - study :1, // - optimize input data for repeated searches - 'sub' :1, // - declare a subroutine, possibly anonymously - 'substr' :1, // - get or alter a portion of a stirng - symlink :1, // - create a symbolic link to a file - syscall :1, // - execute an arbitrary system call - sysopen :1, // - open a file, pipe, or descriptor - sysread :1, // - fixed-length unbuffered input from a filehandle - sysseek :1, // - position I/O pointer on handle used with sysread and syswrite - system :1, // - run a separate program - syswrite :1, // - fixed-length unbuffered output to a filehandle - tell :1, // - get current seekpointer on a filehandle - telldir :1, // - get current seekpointer on a directory handle - tie :1, // - bind a variable to an object class - tied :1, // - get a reference to the object underlying a tied variable - time :1, // - return number of seconds since 1970 - times :1, // - return elapsed time for self and child processes - tr :null, // - transliterate a string - truncate :1, // - shorten a file - uc :1, // - return upper-case version of a string - ucfirst :1, // - return a string with just the next letter in upper case - umask :1, // - set file creation mode mask - undef :1, // - remove a variable or function definition - unlink :1, // - remove one link to a file - unpack :1, // - convert binary structure into normal perl variables - unshift :1, // - prepend more elements to the beginning of a list - untie :1, // - break a tie binding to a variable - use :1, // - load in a module at compile time - utime :1, // - set a file's last access and modify times - values :1, // - return a list of the values in a hash - vec :1, // - test or set particular bits in a string - wait :1, // - wait for any child process to die - waitpid :1, // - wait for a particular child process to die - wantarray :1, // - get void vs scalar vs list context of current subroutine call - warn :1, // - print debugging info - when :1, // - write :1, // - print a picture record - y :null}; // - transliterate a string - - var RXstyle="string-2"; - var RXmodifiers=/[goseximacplud]/; // NOTE: "m", "s", "y" and "tr" need to correct real modifiers for each regexp type - - function tokenChain(stream,state,chain,style,tail){ // NOTE: chain.length > 2 is not working now (it's for s[...][...]geos;) - state.chain=null; // 12 3tail - state.style=null; - state.tail=null; - state.tokenize=function(stream,state){ - var e=false,c,i=0; - while(c=stream.next()){ - if(c===chain[i]&&!e){ - if(chain[++i]!==undefined){ - state.chain=chain[i]; - state.style=style; - state.tail=tail;} - else if(tail) - stream.eatWhile(tail); - state.tokenize=tokenPerl; - return style;} - e=!e&&c=="\\";} - return style;}; - return state.tokenize(stream,state);} - - function tokenSOMETHING(stream,state,string){ - state.tokenize=function(stream,state){ - if(stream.string==string) - state.tokenize=tokenPerl; - stream.skipToEnd(); - return "string";}; - return state.tokenize(stream,state);} - - function tokenPerl(stream,state){ - if(stream.eatSpace()) - return null; - if(state.chain) - return tokenChain(stream,state,state.chain,state.style,state.tail); - if(stream.match(/^\-?[\d\.]/,false)) - if(stream.match(/^(\-?(\d*\.\d+(e[+-]?\d+)?|\d+\.\d*)|0x[\da-fA-F]+|0b[01]+|\d+(e[+-]?\d+)?)/)) - return 'number'; - if(stream.match(/^<<(?=\w)/)){ // NOTE: <"],RXstyle,RXmodifiers);} - if(/[\^'"!~\/]/.test(c)){ - stream.eatSuffix(1); - return tokenChain(stream,state,[stream.eat(c)],RXstyle,RXmodifiers);}} - else if(c=="q"){ - c=stream.look(1); - if(c=="("){ - stream.eatSuffix(2); - return tokenChain(stream,state,[")"],"string");} - if(c=="["){ - stream.eatSuffix(2); - return tokenChain(stream,state,["]"],"string");} - if(c=="{"){ -stream.eatSuffix(2); - return tokenChain(stream,state,["}"],"string");} - if(c=="<"){ - stream.eatSuffix(2); - return tokenChain(stream,state,[">"],"string");} - if(/[\^'"!~\/]/.test(c)){ - stream.eatSuffix(1); - return tokenChain(stream,state,[stream.eat(c)],"string");}} - else if(c=="w"){ - c=stream.look(1); - if(c=="("){ - stream.eatSuffix(2); - return tokenChain(stream,state,[")"],"bracket");} - if(c=="["){ - stream.eatSuffix(2); - return tokenChain(stream,state,["]"],"bracket");} - if(c=="{"){ - stream.eatSuffix(2); - return tokenChain(stream,state,["}"],"bracket");} - if(c=="<"){ - stream.eatSuffix(2); - return tokenChain(stream,state,[">"],"bracket");} - if(/[\^'"!~\/]/.test(c)){ - stream.eatSuffix(1); - return tokenChain(stream,state,[stream.eat(c)],"bracket");}} - else if(c=="r"){ - c=stream.look(1); - if(c=="("){ - stream.eatSuffix(2); - return tokenChain(stream,state,[")"],RXstyle,RXmodifiers);} - if(c=="["){ - stream.eatSuffix(2); - return tokenChain(stream,state,["]"],RXstyle,RXmodifiers);} - if(c=="{"){ - stream.eatSuffix(2); - return tokenChain(stream,state,["}"],RXstyle,RXmodifiers);} - if(c=="<"){ - stream.eatSuffix(2); - return tokenChain(stream,state,[">"],RXstyle,RXmodifiers);} - if(/[\^'"!~\/]/.test(c)){ - stream.eatSuffix(1); - return tokenChain(stream,state,[stream.eat(c)],RXstyle,RXmodifiers);}} - else if(/[\^'"!~\/(\[{<]/.test(c)){ - if(c=="("){ - stream.eatSuffix(1); - return tokenChain(stream,state,[")"],"string");} - if(c=="["){ - stream.eatSuffix(1); - return tokenChain(stream,state,["]"],"string");} - if(c=="{"){ - stream.eatSuffix(1); - return tokenChain(stream,state,["}"],"string");} - if(c=="<"){ - stream.eatSuffix(1); - return tokenChain(stream,state,[">"],"string");} - if(/[\^'"!~\/]/.test(c)){ - return tokenChain(stream,state,[stream.eat(c)],"string");}}}} - if(ch=="m"){ - var c=stream.look(-2); - if(!(c&&/\w/.test(c))){ - c=stream.eat(/[(\[{<\^'"!~\/]/); - if(c){ - if(/[\^'"!~\/]/.test(c)){ - return tokenChain(stream,state,[c],RXstyle,RXmodifiers);} - if(c=="("){ - return tokenChain(stream,state,[")"],RXstyle,RXmodifiers);} - if(c=="["){ - return tokenChain(stream,state,["]"],RXstyle,RXmodifiers);} - if(c=="{"){ - return tokenChain(stream,state,["}"],RXstyle,RXmodifiers);} - if(c=="<"){ - return tokenChain(stream,state,[">"],RXstyle,RXmodifiers);}}}} - if(ch=="s"){ - var c=/[\/>\]})\w]/.test(stream.look(-2)); - if(!c){ - c=stream.eat(/[(\[{<\^'"!~\/]/); - if(c){ - if(c=="[") - return tokenChain(stream,state,["]","]"],RXstyle,RXmodifiers); - if(c=="{") - return tokenChain(stream,state,["}","}"],RXstyle,RXmodifiers); - if(c=="<") - return tokenChain(stream,state,[">",">"],RXstyle,RXmodifiers); - if(c=="(") - return tokenChain(stream,state,[")",")"],RXstyle,RXmodifiers); - return tokenChain(stream,state,[c,c],RXstyle,RXmodifiers);}}} - if(ch=="y"){ - var c=/[\/>\]})\w]/.test(stream.look(-2)); - if(!c){ - c=stream.eat(/[(\[{<\^'"!~\/]/); - if(c){ - if(c=="[") - return tokenChain(stream,state,["]","]"],RXstyle,RXmodifiers); - if(c=="{") - return tokenChain(stream,state,["}","}"],RXstyle,RXmodifiers); - if(c=="<") - return tokenChain(stream,state,[">",">"],RXstyle,RXmodifiers); - if(c=="(") - return tokenChain(stream,state,[")",")"],RXstyle,RXmodifiers); - return tokenChain(stream,state,[c,c],RXstyle,RXmodifiers);}}} - if(ch=="t"){ - var c=/[\/>\]})\w]/.test(stream.look(-2)); - if(!c){ - c=stream.eat("r");if(c){ - c=stream.eat(/[(\[{<\^'"!~\/]/); - if(c){ - if(c=="[") - return tokenChain(stream,state,["]","]"],RXstyle,RXmodifiers); - if(c=="{") - return tokenChain(stream,state,["}","}"],RXstyle,RXmodifiers); - if(c=="<") - return tokenChain(stream,state,[">",">"],RXstyle,RXmodifiers); - if(c=="(") - return tokenChain(stream,state,[")",")"],RXstyle,RXmodifiers); - return tokenChain(stream,state,[c,c],RXstyle,RXmodifiers);}}}} - if(ch=="`"){ - return tokenChain(stream,state,[ch],"variable-2");} - if(ch=="/"){ - if(!/~\s*$/.test(stream.prefix())) - return "operator"; - else - return tokenChain(stream,state,[ch],RXstyle,RXmodifiers);} - if(ch=="$"){ - var p=stream.pos; - if(stream.eatWhile(/\d/)||stream.eat("{")&&stream.eatWhile(/\d/)&&stream.eat("}")) - return "variable-2"; - else - stream.pos=p;} - if(/[$@%]/.test(ch)){ - var p=stream.pos; - if(stream.eat("^")&&stream.eat(/[A-Z]/)||!/[@$%&]/.test(stream.look(-2))&&stream.eat(/[=|\\\-#?@;:&`~\^!\[\]*'"$+.,\/<>()]/)){ - var c=stream.current(); - if(PERL[c]) - return "variable-2";} - stream.pos=p;} - if(/[$@%&]/.test(ch)){ - if(stream.eatWhile(/[\w$\[\]]/)||stream.eat("{")&&stream.eatWhile(/[\w$\[\]]/)&&stream.eat("}")){ - var c=stream.current(); - if(PERL[c]) - return "variable-2"; - else - return "variable";}} - if(ch=="#"){ - if(stream.look(-2)!="$"){ - stream.skipToEnd(); - return "comment";}} - if(/[:+\-\^*$&%@=<>!?|\/~\.]/.test(ch)){ - var p=stream.pos; - stream.eatWhile(/[:+\-\^*$&%@=<>!?|\/~\.]/); - if(PERL[stream.current()]) - return "operator"; - else - stream.pos=p;} - if(ch=="_"){ - if(stream.pos==1){ - if(stream.suffix(6)=="_END__"){ - return tokenChain(stream,state,['\0'],"comment");} - else if(stream.suffix(7)=="_DATA__"){ - return tokenChain(stream,state,['\0'],"variable-2");} - else if(stream.suffix(7)=="_C__"){ - return tokenChain(stream,state,['\0'],"string");}}} - if(/\w/.test(ch)){ - var p=stream.pos; - if(stream.look(-2)=="{"&&(stream.look(0)=="}"||stream.eatWhile(/\w/)&&stream.look(0)=="}")) - return "string"; - else - stream.pos=p;} - if(/[A-Z]/.test(ch)){ - var l=stream.look(-2); - var p=stream.pos; - stream.eatWhile(/[A-Z_]/); - if(/[\da-z]/.test(stream.look(0))){ - stream.pos=p;} - else{ - var c=PERL[stream.current()]; - if(!c) - return "meta"; - if(c[1]) - c=c[0]; - if(l!=":"){ - if(c==1) - return "keyword"; - else if(c==2) - return "def"; - else if(c==3) - return "atom"; - else if(c==4) - return "operator"; - else if(c==5) - return "variable-2"; - else - return "meta";} - else - return "meta";}} - if(/[a-zA-Z_]/.test(ch)){ - var l=stream.look(-2); - stream.eatWhile(/\w/); - var c=PERL[stream.current()]; - if(!c) - return "meta"; - if(c[1]) - c=c[0]; - if(l!=":"){ - if(c==1) - return "keyword"; - else if(c==2) - return "def"; - else if(c==3) - return "atom"; - else if(c==4) - return "operator"; - else if(c==5) - return "variable-2"; - else - return "meta";} - else - return "meta";} - return null;} - - return{ - startState:function(){ - return{ - tokenize:tokenPerl, - chain:null, - style:null, - tail:null};}, - token:function(stream,state){ - return (state.tokenize||tokenPerl)(stream,state);}, - electricChars:"{}"};}); - -CodeMirror.defineMIME("text/x-perl", "perl"); - -// it's like "peek", but need for look-ahead or look-behind if index < 0 -CodeMirror.StringStream.prototype.look=function(c){ - return this.string.charAt(this.pos+(c||0));}; - -// return a part of prefix of current stream from current position -CodeMirror.StringStream.prototype.prefix=function(c){ - if(c){ - var x=this.pos-c; - return this.string.substr((x>=0?x:0),c);} - else{ - return this.string.substr(0,this.pos-1);}}; - -// return a part of suffix of current stream from current position -CodeMirror.StringStream.prototype.suffix=function(c){ - var y=this.string.length; - var x=y-this.pos+1; - return this.string.substr(this.pos,(c&&c=(y=this.string.length-1)) - this.pos=y; - else - this.pos=x;}; diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pig/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/pig/index.html deleted file mode 100644 index 1b0c60267f..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pig/index.html +++ /dev/null @@ -1,42 +0,0 @@ - - - - - CodeMirror: Pig Latin mode - - - - - - - -

CodeMirror: Pig Latin mode

- -
- - - -

- Simple mode that handles Pig Latin language. -

- -

MIME type defined: text/x-pig - (PIG code) - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pig/pig.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/pig/pig.js deleted file mode 100644 index c2f611a1a0..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/pig/pig.js +++ /dev/null @@ -1,171 +0,0 @@ -/* - * Pig Latin Mode for CodeMirror 2 - * @author Prasanth Jayachandran - * @link https://github.com/prasanthj/pig-codemirror-2 - * This implementation is adapted from PL/SQL mode in CodeMirror 2. - */ -CodeMirror.defineMode("pig", function(_config, parserConfig) { - var keywords = parserConfig.keywords, - builtins = parserConfig.builtins, - types = parserConfig.types, - multiLineStrings = parserConfig.multiLineStrings; - - var isOperatorChar = /[*+\-%<>=&?:\/!|]/; - - function chain(stream, state, f) { - state.tokenize = f; - return f(stream, state); - } - - var type; - function ret(tp, style) { - type = tp; - return style; - } - - function tokenComment(stream, state) { - var isEnd = false; - var ch; - while(ch = stream.next()) { - if(ch == "/" && isEnd) { - state.tokenize = tokenBase; - break; - } - isEnd = (ch == "*"); - } - return ret("comment", "comment"); - } - - function tokenString(quote) { - return function(stream, state) { - var escaped = false, next, end = false; - while((next = stream.next()) != null) { - if (next == quote && !escaped) { - end = true; break; - } - escaped = !escaped && next == "\\"; - } - if (end || !(escaped || multiLineStrings)) - state.tokenize = tokenBase; - return ret("string", "error"); - }; - } - - function tokenBase(stream, state) { - var ch = stream.next(); - - // is a start of string? - if (ch == '"' || ch == "'") - return chain(stream, state, tokenString(ch)); - // is it one of the special chars - else if(/[\[\]{}\(\),;\.]/.test(ch)) - return ret(ch); - // is it a number? - else if(/\d/.test(ch)) { - stream.eatWhile(/[\w\.]/); - return ret("number", "number"); - } - // multi line comment or operator - else if (ch == "/") { - if (stream.eat("*")) { - return chain(stream, state, tokenComment); - } - else { - stream.eatWhile(isOperatorChar); - return ret("operator", "operator"); - } - } - // single line comment or operator - else if (ch=="-") { - if(stream.eat("-")){ - stream.skipToEnd(); - return ret("comment", "comment"); - } - else { - stream.eatWhile(isOperatorChar); - return ret("operator", "operator"); - } - } - // is it an operator - else if (isOperatorChar.test(ch)) { - stream.eatWhile(isOperatorChar); - return ret("operator", "operator"); - } - else { - // get the while word - stream.eatWhile(/[\w\$_]/); - // is it one of the listed keywords? - if (keywords && keywords.propertyIsEnumerable(stream.current().toUpperCase())) { - if (stream.eat(")") || stream.eat(".")) { - //keywords can be used as variables like flatten(group), group.$0 etc.. - } - else { - return ("keyword", "keyword"); - } - } - // is it one of the builtin functions? - if (builtins && builtins.propertyIsEnumerable(stream.current().toUpperCase())) - { - return ("keyword", "variable-2"); - } - // is it one of the listed types? - if (types && types.propertyIsEnumerable(stream.current().toUpperCase())) - return ("keyword", "variable-3"); - // default is a 'variable' - return ret("variable", "pig-word"); - } - } - - // Interface - return { - startState: function() { - return { - tokenize: tokenBase, - startOfLine: true - }; - }, - - token: function(stream, state) { - if(stream.eatSpace()) return null; - var style = state.tokenize(stream, state); - return style; - } - }; -}); - -(function() { - function keywords(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - - // builtin funcs taken from trunk revision 1303237 - var pBuiltins = "ABS ACOS ARITY ASIN ATAN AVG BAGSIZE BINSTORAGE BLOOM BUILDBLOOM CBRT CEIL " - + "CONCAT COR COS COSH COUNT COUNT_STAR COV CONSTANTSIZE CUBEDIMENSIONS DIFF DISTINCT DOUBLEABS " - + "DOUBLEAVG DOUBLEBASE DOUBLEMAX DOUBLEMIN DOUBLEROUND DOUBLESUM EXP FLOOR FLOATABS FLOATAVG " - + "FLOATMAX FLOATMIN FLOATROUND FLOATSUM GENERICINVOKER INDEXOF INTABS INTAVG INTMAX INTMIN " - + "INTSUM INVOKEFORDOUBLE INVOKEFORFLOAT INVOKEFORINT INVOKEFORLONG INVOKEFORSTRING INVOKER " - + "ISEMPTY JSONLOADER JSONMETADATA JSONSTORAGE LAST_INDEX_OF LCFIRST LOG LOG10 LOWER LONGABS " - + "LONGAVG LONGMAX LONGMIN LONGSUM MAX MIN MAPSIZE MONITOREDUDF NONDETERMINISTIC OUTPUTSCHEMA " - + "PIGSTORAGE PIGSTREAMING RANDOM REGEX_EXTRACT REGEX_EXTRACT_ALL REPLACE ROUND SIN SINH SIZE " - + "SQRT STRSPLIT SUBSTRING SUM STRINGCONCAT STRINGMAX STRINGMIN STRINGSIZE TAN TANH TOBAG " - + "TOKENIZE TOMAP TOP TOTUPLE TRIM TEXTLOADER TUPLESIZE UCFIRST UPPER UTF8STORAGECONVERTER "; - - // taken from QueryLexer.g - var pKeywords = "VOID IMPORT RETURNS DEFINE LOAD FILTER FOREACH ORDER CUBE DISTINCT COGROUP " - + "JOIN CROSS UNION SPLIT INTO IF OTHERWISE ALL AS BY USING INNER OUTER ONSCHEMA PARALLEL " - + "PARTITION GROUP AND OR NOT GENERATE FLATTEN ASC DESC IS STREAM THROUGH STORE MAPREDUCE " - + "SHIP CACHE INPUT OUTPUT STDERROR STDIN STDOUT LIMIT SAMPLE LEFT RIGHT FULL EQ GT LT GTE LTE " - + "NEQ MATCHES TRUE FALSE "; - - // data types - var pTypes = "BOOLEAN INT LONG FLOAT DOUBLE CHARARRAY BYTEARRAY BAG TUPLE MAP "; - - CodeMirror.defineMIME("text/x-pig", { - name: "pig", - builtins: keywords(pBuiltins), - keywords: keywords(pKeywords), - types: keywords(pTypes) - }); -}()); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/properties/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/properties/index.html deleted file mode 100644 index e21e02abd2..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/properties/index.html +++ /dev/null @@ -1,41 +0,0 @@ - - - - - CodeMirror: Properties files mode - - - - - - - -

CodeMirror: Properties files mode

-
- - -

MIME types defined: text/x-properties, - text/x-ini.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/properties/properties.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/properties/properties.js deleted file mode 100644 index d3a13c765d..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/properties/properties.js +++ /dev/null @@ -1,63 +0,0 @@ -CodeMirror.defineMode("properties", function() { - return { - token: function(stream, state) { - var sol = stream.sol() || state.afterSection; - var eol = stream.eol(); - - state.afterSection = false; - - if (sol) { - if (state.nextMultiline) { - state.inMultiline = true; - state.nextMultiline = false; - } else { - state.position = "def"; - } - } - - if (eol && ! state.nextMultiline) { - state.inMultiline = false; - state.position = "def"; - } - - if (sol) { - while(stream.eatSpace()); - } - - var ch = stream.next(); - - if (sol && (ch === "#" || ch === "!" || ch === ";")) { - state.position = "comment"; - stream.skipToEnd(); - return "comment"; - } else if (sol && ch === "[") { - state.afterSection = true; - stream.skipTo("]"); stream.eat("]"); - return "header"; - } else if (ch === "=" || ch === ":") { - state.position = "quote"; - return null; - } else if (ch === "\\" && state.position === "quote") { - if (stream.next() !== "u") { // u = Unicode sequence \u1234 - // Multiline value - state.nextMultiline = true; - } - } - - return state.position; - }, - - startState: function() { - return { - position : "def", // Current position, "def", "quote" or "comment" - nextMultiline : false, // Is the next line multiline value - inMultiline : false, // Is the current line a multiline value - afterSection : false // Did we just open a section - }; - } - - }; -}); - -CodeMirror.defineMIME("text/x-properties", "properties"); -CodeMirror.defineMIME("text/x-ini", "properties"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/python/LICENSE.txt b/wcfsetup/install/files/js/3rdParty/codemirror/mode/python/LICENSE.txt deleted file mode 100644 index 918866b42a..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/python/LICENSE.txt +++ /dev/null @@ -1,21 +0,0 @@ -The MIT License - -Copyright (c) 2010 Timothy Farrell - -Permission is hereby granted, free of charge, to any person obtaining a copy -of this software and associated documentation files (the "Software"), to deal -in the Software without restriction, including without limitation the rights -to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -copies of the Software, and to permit persons to whom the Software is -furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in -all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -THE SOFTWARE. \ No newline at end of file diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/python/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/python/index.html deleted file mode 100644 index 4244c6fc5a..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/python/index.html +++ /dev/null @@ -1,135 +0,0 @@ - - - - - CodeMirror: Python mode - - - - - - - - -

CodeMirror: Python mode

- -
- -

Configuration Options:

-
    -
  • version - 2/3 - The version of Python to recognize. Default is 2.
  • -
  • singleLineStringErrors - true/false - If you have a single-line string that is not terminated at the end of the line, this will show subsequent lines as errors if true, otherwise it will consider the newline as the end of the string. Default is false.
  • -
-

Advanced Configuration Options:

-

Usefull for superset of python syntax like Enthought enaml, IPython magics and questionmark help

-
    -
  • singleOperators - RegEx - Regular Expression for single operator matching, default :
    ^[\\+\\-\\*/%&|\\^~<>!]
  • -
  • singleDelimiters - RegEx - Regular Expression for single delimiter matching, default :
    ^[\\(\\)\\[\\]\\{\\}@,:`=;\\.]
  • -
  • doubleOperators - RegEx - Regular Expression for double operators matching, default :
    ^((==)|(!=)|(<=)|(>=)|(<>)|(<<)|(>>)|(//)|(\\*\\*))
  • -
  • doubleDelimiters - RegEx - Regular Expressoin for double delimiters matching, default :
    ^((\\+=)|(\\-=)|(\\*=)|(%=)|(/=)|(&=)|(\\|=)|(\\^=))
  • -
  • tripleDelimiters - RegEx - Regular Expression for triple delimiters matching, default :
    ^((//=)|(>>=)|(<<=)|(\\*\\*=))
  • -
  • identifiers - RegEx - Regular Expression for identifier, default :
    ^[_A-Za-z][_A-Za-z0-9]*
  • -
- - -

MIME types defined: text/x-python.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/python/python.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/python/python.js deleted file mode 100644 index b623972b88..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/python/python.js +++ /dev/null @@ -1,341 +0,0 @@ -CodeMirror.defineMode("python", function(conf, parserConf) { - var ERRORCLASS = 'error'; - - function wordRegexp(words) { - return new RegExp("^((" + words.join(")|(") + "))\\b"); - } - - var singleOperators = parserConf.singleOperators || new RegExp("^[\\+\\-\\*/%&|\\^~<>!]"); - var singleDelimiters = parserConf.singleDelimiters || new RegExp('^[\\(\\)\\[\\]\\{\\}@,:`=;\\.]'); - var doubleOperators = parserConf.doubleOperators || new RegExp("^((==)|(!=)|(<=)|(>=)|(<>)|(<<)|(>>)|(//)|(\\*\\*))"); - var doubleDelimiters = parserConf.doubleDelimiters || new RegExp("^((\\+=)|(\\-=)|(\\*=)|(%=)|(/=)|(&=)|(\\|=)|(\\^=))"); - var tripleDelimiters = parserConf.tripleDelimiters || new RegExp("^((//=)|(>>=)|(<<=)|(\\*\\*=))"); - var identifiers = parserConf.identifiers|| new RegExp("^[_A-Za-z][_A-Za-z0-9]*"); - - var wordOperators = wordRegexp(['and', 'or', 'not', 'is', 'in']); - var commonkeywords = ['as', 'assert', 'break', 'class', 'continue', - 'def', 'del', 'elif', 'else', 'except', 'finally', - 'for', 'from', 'global', 'if', 'import', - 'lambda', 'pass', 'raise', 'return', - 'try', 'while', 'with', 'yield']; - var commonBuiltins = ['abs', 'all', 'any', 'bin', 'bool', 'bytearray', 'callable', 'chr', - 'classmethod', 'compile', 'complex', 'delattr', 'dict', 'dir', 'divmod', - 'enumerate', 'eval', 'filter', 'float', 'format', 'frozenset', - 'getattr', 'globals', 'hasattr', 'hash', 'help', 'hex', 'id', - 'input', 'int', 'isinstance', 'issubclass', 'iter', 'len', - 'list', 'locals', 'map', 'max', 'memoryview', 'min', 'next', - 'object', 'oct', 'open', 'ord', 'pow', 'property', 'range', - 'repr', 'reversed', 'round', 'set', 'setattr', 'slice', - 'sorted', 'staticmethod', 'str', 'sum', 'super', 'tuple', - 'type', 'vars', 'zip', '__import__', 'NotImplemented', - 'Ellipsis', '__debug__']; - var py2 = {'builtins': ['apply', 'basestring', 'buffer', 'cmp', 'coerce', 'execfile', - 'file', 'intern', 'long', 'raw_input', 'reduce', 'reload', - 'unichr', 'unicode', 'xrange', 'False', 'True', 'None'], - 'keywords': ['exec', 'print']}; - var py3 = {'builtins': ['ascii', 'bytes', 'exec', 'print'], - 'keywords': ['nonlocal', 'False', 'True', 'None']}; - - if (!!parserConf.version && parseInt(parserConf.version, 10) === 3) { - commonkeywords = commonkeywords.concat(py3.keywords); - commonBuiltins = commonBuiltins.concat(py3.builtins); - var stringPrefixes = new RegExp("^(([rb]|(br))?('{3}|\"{3}|['\"]))", "i"); - } else { - commonkeywords = commonkeywords.concat(py2.keywords); - commonBuiltins = commonBuiltins.concat(py2.builtins); - var stringPrefixes = new RegExp("^(([rub]|(ur)|(br))?('{3}|\"{3}|['\"]))", "i"); - } - var keywords = wordRegexp(commonkeywords); - var builtins = wordRegexp(commonBuiltins); - - var indentInfo = null; - - // tokenizers - function tokenBase(stream, state) { - // Handle scope changes - if (stream.sol()) { - var scopeOffset = state.scopes[0].offset; - if (stream.eatSpace()) { - var lineOffset = stream.indentation(); - if (lineOffset > scopeOffset) { - indentInfo = 'indent'; - } else if (lineOffset < scopeOffset) { - indentInfo = 'dedent'; - } - return null; - } else { - if (scopeOffset > 0) { - dedent(stream, state); - } - } - } - if (stream.eatSpace()) { - return null; - } - - var ch = stream.peek(); - - // Handle Comments - if (ch === '#') { - stream.skipToEnd(); - return 'comment'; - } - - // Handle Number Literals - if (stream.match(/^[0-9\.]/, false)) { - var floatLiteral = false; - // Floats - if (stream.match(/^\d*\.\d+(e[\+\-]?\d+)?/i)) { floatLiteral = true; } - if (stream.match(/^\d+\.\d*/)) { floatLiteral = true; } - if (stream.match(/^\.\d+/)) { floatLiteral = true; } - if (floatLiteral) { - // Float literals may be "imaginary" - stream.eat(/J/i); - return 'number'; - } - // Integers - var intLiteral = false; - // Hex - if (stream.match(/^0x[0-9a-f]+/i)) { intLiteral = true; } - // Binary - if (stream.match(/^0b[01]+/i)) { intLiteral = true; } - // Octal - if (stream.match(/^0o[0-7]+/i)) { intLiteral = true; } - // Decimal - if (stream.match(/^[1-9]\d*(e[\+\-]?\d+)?/)) { - // Decimal literals may be "imaginary" - stream.eat(/J/i); - // TODO - Can you have imaginary longs? - intLiteral = true; - } - // Zero by itself with no other piece of number. - if (stream.match(/^0(?![\dx])/i)) { intLiteral = true; } - if (intLiteral) { - // Integer literals may be "long" - stream.eat(/L/i); - return 'number'; - } - } - - // Handle Strings - if (stream.match(stringPrefixes)) { - state.tokenize = tokenStringFactory(stream.current()); - return state.tokenize(stream, state); - } - - // Handle operators and Delimiters - if (stream.match(tripleDelimiters) || stream.match(doubleDelimiters)) { - return null; - } - if (stream.match(doubleOperators) - || stream.match(singleOperators) - || stream.match(wordOperators)) { - return 'operator'; - } - if (stream.match(singleDelimiters)) { - return null; - } - - if (stream.match(keywords)) { - return 'keyword'; - } - - if (stream.match(builtins)) { - return 'builtin'; - } - - if (stream.match(identifiers)) { - return 'variable'; - } - - // Handle non-detected items - stream.next(); - return ERRORCLASS; - } - - function tokenStringFactory(delimiter) { - while ('rub'.indexOf(delimiter.charAt(0).toLowerCase()) >= 0) { - delimiter = delimiter.substr(1); - } - var singleline = delimiter.length == 1; - var OUTCLASS = 'string'; - - function tokenString(stream, state) { - while (!stream.eol()) { - stream.eatWhile(/[^'"\\]/); - if (stream.eat('\\')) { - stream.next(); - if (singleline && stream.eol()) { - return OUTCLASS; - } - } else if (stream.match(delimiter)) { - state.tokenize = tokenBase; - return OUTCLASS; - } else { - stream.eat(/['"]/); - } - } - if (singleline) { - if (parserConf.singleLineStringErrors) { - return ERRORCLASS; - } else { - state.tokenize = tokenBase; - } - } - return OUTCLASS; - } - tokenString.isString = true; - return tokenString; - } - - function indent(stream, state, type) { - type = type || 'py'; - var indentUnit = 0; - if (type === 'py') { - if (state.scopes[0].type !== 'py') { - state.scopes[0].offset = stream.indentation(); - return; - } - for (var i = 0; i < state.scopes.length; ++i) { - if (state.scopes[i].type === 'py') { - indentUnit = state.scopes[i].offset + conf.indentUnit; - break; - } - } - } else { - indentUnit = stream.column() + stream.current().length; - } - state.scopes.unshift({ - offset: indentUnit, - type: type - }); - } - - function dedent(stream, state, type) { - type = type || 'py'; - if (state.scopes.length == 1) return; - if (state.scopes[0].type === 'py') { - var _indent = stream.indentation(); - var _indent_index = -1; - for (var i = 0; i < state.scopes.length; ++i) { - if (_indent === state.scopes[i].offset) { - _indent_index = i; - break; - } - } - if (_indent_index === -1) { - return true; - } - while (state.scopes[0].offset !== _indent) { - state.scopes.shift(); - } - return false; - } else { - if (type === 'py') { - state.scopes[0].offset = stream.indentation(); - return false; - } else { - if (state.scopes[0].type != type) { - return true; - } - state.scopes.shift(); - return false; - } - } - } - - function tokenLexer(stream, state) { - indentInfo = null; - var style = state.tokenize(stream, state); - var current = stream.current(); - - // Handle '.' connected identifiers - if (current === '.') { - style = stream.match(identifiers, false) ? null : ERRORCLASS; - if (style === null && state.lastToken === 'meta') { - // Apply 'meta' style to '.' connected identifiers when - // appropriate. - style = 'meta'; - } - return style; - } - - // Handle decorators - if (current === '@') { - return stream.match(identifiers, false) ? 'meta' : ERRORCLASS; - } - - if ((style === 'variable' || style === 'builtin') - && state.lastToken === 'meta') { - style = 'meta'; - } - - // Handle scope changes. - if (current === 'pass' || current === 'return') { - state.dedent += 1; - } - if (current === 'lambda') state.lambda = true; - if ((current === ':' && !state.lambda && state.scopes[0].type == 'py') - || indentInfo === 'indent') { - indent(stream, state); - } - var delimiter_index = '[({'.indexOf(current); - if (delimiter_index !== -1) { - indent(stream, state, '])}'.slice(delimiter_index, delimiter_index+1)); - } - if (indentInfo === 'dedent') { - if (dedent(stream, state)) { - return ERRORCLASS; - } - } - delimiter_index = '])}'.indexOf(current); - if (delimiter_index !== -1) { - if (dedent(stream, state, current)) { - return ERRORCLASS; - } - } - if (state.dedent > 0 && stream.eol() && state.scopes[0].type == 'py') { - if (state.scopes.length > 1) state.scopes.shift(); - state.dedent -= 1; - } - - return style; - } - - var external = { - startState: function(basecolumn) { - return { - tokenize: tokenBase, - scopes: [{offset:basecolumn || 0, type:'py'}], - lastToken: null, - lambda: false, - dedent: 0 - }; - }, - - token: function(stream, state) { - var style = tokenLexer(stream, state); - - state.lastToken = style; - - if (stream.eol() && stream.lambda) { - state.lambda = false; - } - - return style; - }, - - indent: function(state) { - if (state.tokenize != tokenBase) { - return state.tokenize.isString ? CodeMirror.Pass : 0; - } - - return state.scopes[0].offset; - }, - - lineComment: "#" - }; - return external; -}); - -CodeMirror.defineMIME("text/x-python", "python"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/q/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/q/index.html deleted file mode 100644 index 303ec1d3ad..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/q/index.html +++ /dev/null @@ -1,131 +0,0 @@ - - - - - CodeMirror: Q mode - - - - - - - - -

CodeMirror: Q mode

- -
- - - -

MIME type defined: text/x-q.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/q/q.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/q/q.js deleted file mode 100644 index 56017e30ad..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/q/q.js +++ /dev/null @@ -1,124 +0,0 @@ -CodeMirror.defineMode("q",function(config){ - var indentUnit=config.indentUnit, - curPunc, - keywords=buildRE(["abs","acos","aj","aj0","all","and","any","asc","asin","asof","atan","attr","avg","avgs","bin","by","ceiling","cols","cor","cos","count","cov","cross","csv","cut","delete","deltas","desc","dev","differ","distinct","div","do","each","ej","enlist","eval","except","exec","exit","exp","fby","fills","first","fkeys","flip","floor","from","get","getenv","group","gtime","hclose","hcount","hdel","hopen","hsym","iasc","idesc","if","ij","in","insert","inter","inv","key","keys","last","like","list","lj","load","log","lower","lsq","ltime","ltrim","mavg","max","maxs","mcount","md5","mdev","med","meta","min","mins","mmax","mmin","mmu","mod","msum","neg","next","not","null","or","over","parse","peach","pj","plist","prd","prds","prev","prior","rand","rank","ratios","raze","read0","read1","reciprocal","reverse","rload","rotate","rsave","rtrim","save","scan","select","set","setenv","show","signum","sin","sqrt","ss","ssr","string","sublist","sum","sums","sv","system","tables","tan","til","trim","txf","type","uj","ungroup","union","update","upper","upsert","value","var","view","views","vs","wavg","where","where","while","within","wj","wj1","wsum","xasc","xbar","xcol","xcols","xdesc","xexp","xgroup","xkey","xlog","xprev","xrank"]), - E=/[|/&^!+:\\\-*%$=~#;@><,?_\'\"\[\(\]\)\s{}]/; - function buildRE(w){return new RegExp("^("+w.join("|")+")$");} - function tokenBase(stream,state){ - var sol=stream.sol(),c=stream.next(); - curPunc=null; - if(sol) - if(c=="/") - return(state.tokenize=tokenLineComment)(stream,state); - else if(c=="\\"){ - if(stream.eol()||/\s/.test(stream.peek())) - return stream.skipToEnd(),/^\\\s*$/.test(stream.current())?(state.tokenize=tokenCommentToEOF)(stream, state):state.tokenize=tokenBase,"comment"; - else - return state.tokenize=tokenBase,"builtin"; - } - if(/\s/.test(c)) - return stream.peek()=="/"?(stream.skipToEnd(),"comment"):"whitespace"; - if(c=='"') - return(state.tokenize=tokenString)(stream,state); - if(c=='`') - return stream.eatWhile(/[A-Z|a-z|\d|_|:|\/|\.]/),"symbol"; - if(("."==c&&/\d/.test(stream.peek()))||/\d/.test(c)){ - var t=null; - stream.backUp(1); - if(stream.match(/^\d{4}\.\d{2}(m|\.\d{2}([D|T](\d{2}(:\d{2}(:\d{2}(\.\d{1,9})?)?)?)?)?)/) - || stream.match(/^\d+D(\d{2}(:\d{2}(:\d{2}(\.\d{1,9})?)?)?)/) - || stream.match(/^\d{2}:\d{2}(:\d{2}(\.\d{1,9})?)?/) - || stream.match(/^\d+[ptuv]{1}/)) - t="temporal"; - else if(stream.match(/^0[NwW]{1}/) - || stream.match(/^0x[\d|a-f|A-F]*/) - || stream.match(/^[0|1]+[b]{1}/) - || stream.match(/^\d+[chijn]{1}/) - || stream.match(/-?\d*(\.\d*)?(e[+\-]?\d+)?(e|f)?/)) - t="number"; - return(t&&(!(c=stream.peek())||E.test(c)))?t:(stream.next(),"error"); - } - if(/[A-Z|a-z]|\./.test(c)) - return stream.eatWhile(/[A-Z|a-z|\.|_|\d]/),keywords.test(stream.current())?"keyword":"variable"; - if(/[|/&^!+:\\\-*%$=~#;@><\.,?_\']/.test(c)) - return null; - if(/[{}\(\[\]\)]/.test(c)) - return null; - return"error"; - } - function tokenLineComment(stream,state){ - return stream.skipToEnd(),/\/\s*$/.test(stream.current())?(state.tokenize=tokenBlockComment)(stream,state):(state.tokenize=tokenBase),"comment"; - } - function tokenBlockComment(stream,state){ - var f=stream.sol()&&stream.peek()=="\\"; - stream.skipToEnd(); - if(f&&/^\\\s*$/.test(stream.current())) - state.tokenize=tokenBase; - return"comment"; - } - function tokenCommentToEOF(stream){return stream.skipToEnd(),"comment";} - function tokenString(stream,state){ - var escaped=false,next,end=false; - while((next=stream.next())){ - if(next=="\""&&!escaped){end=true;break;} - escaped=!escaped&&next=="\\"; - } - if(end)state.tokenize=tokenBase; - return"string"; - } - function pushContext(state,type,col){state.context={prev:state.context,indent:state.indent,col:col,type:type};} - function popContext(state){state.indent=state.context.indent;state.context=state.context.prev;} - return{ - startState:function(){ - return{tokenize:tokenBase, - context:null, - indent:0, - col:0}; - }, - token:function(stream,state){ - if(stream.sol()){ - if(state.context&&state.context.align==null) - state.context.align=false; - state.indent=stream.indentation(); - } - //if (stream.eatSpace()) return null; - var style=state.tokenize(stream,state); - if(style!="comment"&&state.context&&state.context.align==null&&state.context.type!="pattern"){ - state.context.align=true; - } - if(curPunc=="(")pushContext(state,")",stream.column()); - else if(curPunc=="[")pushContext(state,"]",stream.column()); - else if(curPunc=="{")pushContext(state,"}",stream.column()); - else if(/[\]\}\)]/.test(curPunc)){ - while(state.context&&state.context.type=="pattern")popContext(state); - if(state.context&&curPunc==state.context.type)popContext(state); - } - else if(curPunc=="."&&state.context&&state.context.type=="pattern")popContext(state); - else if(/atom|string|variable/.test(style)&&state.context){ - if(/[\}\]]/.test(state.context.type)) - pushContext(state,"pattern",stream.column()); - else if(state.context.type=="pattern"&&!state.context.align){ - state.context.align=true; - state.context.col=stream.column(); - } - } - return style; - }, - indent:function(state,textAfter){ - var firstChar=textAfter&&textAfter.charAt(0); - var context=state.context; - if(/[\]\}]/.test(firstChar)) - while (context&&context.type=="pattern")context=context.prev; - var closing=context&&firstChar==context.type; - if(!context) - return 0; - else if(context.type=="pattern") - return context.col; - else if(context.align) - return context.col+(closing?0:1); - else - return context.indent+(closing?0:indentUnit); - } - }; -}); -CodeMirror.defineMIME("text/x-q","q"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/r/LICENSE b/wcfsetup/install/files/js/3rdParty/codemirror/mode/r/LICENSE deleted file mode 100644 index 2510ae16cf..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/r/LICENSE +++ /dev/null @@ -1,24 +0,0 @@ -Copyright (c) 2011, Ubalo, Inc. -All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are met: - * Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - * Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - * Neither the name of the Ubalo, Inc nor the names of its - contributors may be used to endorse or promote products derived - from this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND -ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED -WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE -DISCLAIMED. IN NO EVENT SHALL UBALO, INC BE LIABLE FOR ANY -DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES -(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; -LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND -ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS -SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/r/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/r/index.html deleted file mode 100644 index 12819553e9..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/r/index.html +++ /dev/null @@ -1,74 +0,0 @@ - - - - - CodeMirror: R mode - - - - - - - -

CodeMirror: R mode

-
- - -

MIME types defined: text/x-rsrc.

- -

Development of the CodeMirror R mode was kindly sponsored - by Ubalo, who hold - the license.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/r/r.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/r/r.js deleted file mode 100644 index 6410efbb22..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/r/r.js +++ /dev/null @@ -1,141 +0,0 @@ -CodeMirror.defineMode("r", function(config) { - function wordObj(str) { - var words = str.split(" "), res = {}; - for (var i = 0; i < words.length; ++i) res[words[i]] = true; - return res; - } - var atoms = wordObj("NULL NA Inf NaN NA_integer_ NA_real_ NA_complex_ NA_character_"); - var builtins = wordObj("list quote bquote eval return call parse deparse"); - var keywords = wordObj("if else repeat while function for in next break"); - var blockkeywords = wordObj("if else repeat while function for"); - var opChars = /[+\-*\/^<>=!&|~$:]/; - var curPunc; - - function tokenBase(stream, state) { - curPunc = null; - var ch = stream.next(); - if (ch == "#") { - stream.skipToEnd(); - return "comment"; - } else if (ch == "0" && stream.eat("x")) { - stream.eatWhile(/[\da-f]/i); - return "number"; - } else if (ch == "." && stream.eat(/\d/)) { - stream.match(/\d*(?:e[+\-]?\d+)?/); - return "number"; - } else if (/\d/.test(ch)) { - stream.match(/\d*(?:\.\d+)?(?:e[+\-]\d+)?L?/); - return "number"; - } else if (ch == "'" || ch == '"') { - state.tokenize = tokenString(ch); - return "string"; - } else if (ch == "." && stream.match(/.[.\d]+/)) { - return "keyword"; - } else if (/[\w\.]/.test(ch) && ch != "_") { - stream.eatWhile(/[\w\.]/); - var word = stream.current(); - if (atoms.propertyIsEnumerable(word)) return "atom"; - if (keywords.propertyIsEnumerable(word)) { - if (blockkeywords.propertyIsEnumerable(word)) curPunc = "block"; - return "keyword"; - } - if (builtins.propertyIsEnumerable(word)) return "builtin"; - return "variable"; - } else if (ch == "%") { - if (stream.skipTo("%")) stream.next(); - return "variable-2"; - } else if (ch == "<" && stream.eat("-")) { - return "arrow"; - } else if (ch == "=" && state.ctx.argList) { - return "arg-is"; - } else if (opChars.test(ch)) { - if (ch == "$") return "dollar"; - stream.eatWhile(opChars); - return "operator"; - } else if (/[\(\){}\[\];]/.test(ch)) { - curPunc = ch; - if (ch == ";") return "semi"; - return null; - } else { - return null; - } - } - - function tokenString(quote) { - return function(stream, state) { - if (stream.eat("\\")) { - var ch = stream.next(); - if (ch == "x") stream.match(/^[a-f0-9]{2}/i); - else if ((ch == "u" || ch == "U") && stream.eat("{") && stream.skipTo("}")) stream.next(); - else if (ch == "u") stream.match(/^[a-f0-9]{4}/i); - else if (ch == "U") stream.match(/^[a-f0-9]{8}/i); - else if (/[0-7]/.test(ch)) stream.match(/^[0-7]{1,2}/); - return "string-2"; - } else { - var next; - while ((next = stream.next()) != null) { - if (next == quote) { state.tokenize = tokenBase; break; } - if (next == "\\") { stream.backUp(1); break; } - } - return "string"; - } - }; - } - - function push(state, type, stream) { - state.ctx = {type: type, - indent: state.indent, - align: null, - column: stream.column(), - prev: state.ctx}; - } - function pop(state) { - state.indent = state.ctx.indent; - state.ctx = state.ctx.prev; - } - - return { - startState: function() { - return {tokenize: tokenBase, - ctx: {type: "top", - indent: -config.indentUnit, - align: false}, - indent: 0, - afterIdent: false}; - }, - - token: function(stream, state) { - if (stream.sol()) { - if (state.ctx.align == null) state.ctx.align = false; - state.indent = stream.indentation(); - } - if (stream.eatSpace()) return null; - var style = state.tokenize(stream, state); - if (style != "comment" && state.ctx.align == null) state.ctx.align = true; - - var ctype = state.ctx.type; - if ((curPunc == ";" || curPunc == "{" || curPunc == "}") && ctype == "block") pop(state); - if (curPunc == "{") push(state, "}", stream); - else if (curPunc == "(") { - push(state, ")", stream); - if (state.afterIdent) state.ctx.argList = true; - } - else if (curPunc == "[") push(state, "]", stream); - else if (curPunc == "block") push(state, "block", stream); - else if (curPunc == ctype) pop(state); - state.afterIdent = style == "variable" || style == "keyword"; - return style; - }, - - indent: function(state, textAfter) { - if (state.tokenize != tokenBase) return 0; - var firstChar = textAfter && textAfter.charAt(0), ctx = state.ctx, - closing = firstChar == ctx.type; - if (ctx.type == "block") return ctx.indent + (firstChar == "{" ? 0 : config.indentUnit); - else if (ctx.align) return ctx.column + (closing ? 0 : 1); - else return ctx.indent + (closing ? 0 : config.indentUnit); - } - }; -}); - -CodeMirror.defineMIME("text/x-rsrc", "r"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/changes/changes.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/changes/changes.js deleted file mode 100644 index 14a08d9700..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/changes/changes.js +++ /dev/null @@ -1,19 +0,0 @@ -CodeMirror.defineMode("changes", function() { - var headerSeperator = /^-+$/; - var headerLine = /^(Mon|Tue|Wed|Thu|Fri|Sat|Sun) (Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|Dec) ?\d{1,2} \d{2}:\d{2}(:\d{2})? [A-Z]{3,4} \d{4} - /; - var simpleEmail = /^[\w+.-]+@[\w.-]+/; - - return { - token: function(stream) { - if (stream.sol()) { - if (stream.match(headerSeperator)) { return 'tag'; } - if (stream.match(headerLine)) { return 'tag'; } - } - if (stream.match(simpleEmail)) { return 'string'; } - stream.next(); - return null; - } - }; -}); - -CodeMirror.defineMIME("text/x-rpm-changes", "changes"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/changes/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/changes/index.html deleted file mode 100644 index e0e2d8778b..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/changes/index.html +++ /dev/null @@ -1,53 +0,0 @@ - - - - - CodeMirror: RPM changes mode - - - - - - - -

CodeMirror: RPM changes mode

- -
- - -

MIME types defined: text/x-rpm-changes.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/index.html deleted file mode 100644 index 8be98b63ec..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/index.html +++ /dev/null @@ -1,99 +0,0 @@ - - - - - CodeMirror: RPM spec mode - - - - - - - - -

CodeMirror: RPM spec mode

- -
- - -

MIME types defined: text/x-rpm-spec.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/spec.css b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/spec.css deleted file mode 100644 index d0a5d430ca..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/spec.css +++ /dev/null @@ -1,5 +0,0 @@ -.cm-s-default span.cm-preamble {color: #b26818; font-weight: bold;} -.cm-s-default span.cm-macro {color: #b218b2;} -.cm-s-default span.cm-section {color: green; font-weight: bold;} -.cm-s-default span.cm-script {color: red;} -.cm-s-default span.cm-issue {color: yellow;} diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/spec.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/spec.js deleted file mode 100644 index 9f339c21b1..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rpm/spec/spec.js +++ /dev/null @@ -1,66 +0,0 @@ -// Quick and dirty spec file highlighting - -CodeMirror.defineMode("spec", function() { - var arch = /^(i386|i586|i686|x86_64|ppc64|ppc|ia64|s390x|s390|sparc64|sparcv9|sparc|noarch|alphaev6|alpha|hppa|mipsel)/; - - var preamble = /^(Name|Version|Release|License|Summary|Url|Group|Source|BuildArch|BuildRequires|BuildRoot|AutoReqProv|Provides|Requires(\(\w+\))?|Obsoletes|Conflicts|Recommends|Source\d*|Patch\d*|ExclusiveArch|NoSource|Supplements):/; - var section = /^%(debug_package|package|description|prep|build|install|files|clean|changelog|preun|postun|pre|post|triggerin|triggerun|pretrans|posttrans|verifyscript|check|triggerpostun|triggerprein|trigger)/; - var control_flow_complex = /^%(ifnarch|ifarch|if)/; // rpm control flow macros - var control_flow_simple = /^%(else|endif)/; // rpm control flow macros - var operators = /^(\!|\?|\<\=|\<|\>\=|\>|\=\=|\&\&|\|\|)/; // operators in control flow macros - - return { - startState: function () { - return { - controlFlow: false, - macroParameters: false, - section: false - }; - }, - token: function (stream, state) { - var ch = stream.peek(); - if (ch == "#") { stream.skipToEnd(); return "comment"; } - - if (stream.sol()) { - if (stream.match(preamble)) { return "preamble"; } - if (stream.match(section)) { return "section"; } - } - - if (stream.match(/^\$\w+/)) { return "def"; } // Variables like '$RPM_BUILD_ROOT' - if (stream.match(/^\$\{\w+\}/)) { return "def"; } // Variables like '${RPM_BUILD_ROOT}' - - if (stream.match(control_flow_simple)) { return "keyword"; } - if (stream.match(control_flow_complex)) { - state.controlFlow = true; - return "keyword"; - } - if (state.controlFlow) { - if (stream.match(operators)) { return "operator"; } - if (stream.match(/^(\d+)/)) { return "number"; } - if (stream.eol()) { state.controlFlow = false; } - } - - if (stream.match(arch)) { return "number"; } - - // Macros like '%make_install' or '%attr(0775,root,root)' - if (stream.match(/^%[\w]+/)) { - if (stream.match(/^\(/)) { state.macroParameters = true; } - return "macro"; - } - if (state.macroParameters) { - if (stream.match(/^\d+/)) { return "number";} - if (stream.match(/^\)/)) { - state.macroParameters = false; - return "macro"; - } - } - if (stream.match(/^%\{\??[\w \-]+\}/)) { return "macro"; } // Macros like '%{defined fedora}' - - //TODO: Include bash script sub-parser (CodeMirror supports that) - stream.next(); - return null; - } - }; -}); - -CodeMirror.defineMIME("text/x-rpm-spec", "spec"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/LICENSE.txt b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/LICENSE.txt deleted file mode 100644 index 39484fabb9..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/LICENSE.txt +++ /dev/null @@ -1,21 +0,0 @@ -The MIT License - -Copyright (c) 2013 Hasan Karahan - -Permission is hereby granted, free of charge, to any person obtaining a copy -of this software and associated documentation files (the "Software"), to deal -in the Software without restriction, including without limitation the rights -to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -copies of the Software, and to permit persons to whom the Software is -furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in -all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -THE SOFTWARE. \ No newline at end of file diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/index.html deleted file mode 100644 index b3ab64b80e..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/index.html +++ /dev/null @@ -1,524 +0,0 @@ - - - - - CodeMirror: reStructuredText mode - - - - - - - -

CodeMirror: reStructuredText mode

- -
- - -

- The python mode will be used for highlighting blocks - containing Python/IPython terminal sessions: blocks starting with - >>> (for Python) or In [num]: (for - IPython). - - Further, the stex mode will be used for highlighting - blocks containing LaTex code. -

- -

MIME types defined: text/x-rst.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/rst.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/rst.js deleted file mode 100644 index 8de9d75415..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rst/rst.js +++ /dev/null @@ -1,550 +0,0 @@ -CodeMirror.defineMode('rst-base', function (config) { - - /////////////////////////////////////////////////////////////////////////// - /////////////////////////////////////////////////////////////////////////// - - function format(string) { - var args = Array.prototype.slice.call(arguments, 1); - return string.replace(/{(\d+)}/g, function (match, n) { - return typeof args[n] != 'undefined' ? args[n] : match; - }); - } - - function AssertException(message) { - this.message = message; - } - - AssertException.prototype.toString = function () { - return 'AssertException: ' + this.message; - }; - - function assert(expression, message) { - if (!expression) throw new AssertException(message); - return expression; - } - - /////////////////////////////////////////////////////////////////////////// - /////////////////////////////////////////////////////////////////////////// - - var mode_python = CodeMirror.getMode(config, 'python'); - var mode_stex = CodeMirror.getMode(config, 'stex'); - - /////////////////////////////////////////////////////////////////////////// - /////////////////////////////////////////////////////////////////////////// - - var SEPA = "\\s+"; - var TAIL = "(?:\\s*|\\W|$)", - rx_TAIL = new RegExp(format('^{0}', TAIL)); - - var NAME = "(?:[^\\W\\d_](?:[\\w\\+\\.\\-:]*[^\\W_])?)", - rx_NAME = new RegExp(format('^{0}', NAME)); - var NAME_WWS = "(?:[^\\W\\d_](?:[\\w\\s\\+\\.\\-:]*[^\\W_])?)"; - var REF_NAME = format('(?:{0}|`{1}`)', NAME, NAME_WWS); - - var TEXT1 = "(?:[^\\s\\|](?:[^\\|]*[^\\s\\|])?)"; - var TEXT2 = "(?:[^\\`]+)", - rx_TEXT2 = new RegExp(format('^{0}', TEXT2)); - - var rx_section = new RegExp( - "^([!'#$%&\"()*+,-./:;<=>?@\\[\\\\\\]^_`{|}~])\\1{3,}\\s*$"); - var rx_explicit = new RegExp( - format('^\\.\\.{0}', SEPA)); - var rx_link = new RegExp( - format('^_{0}:{1}|^__:{1}', REF_NAME, TAIL)); - var rx_directive = new RegExp( - format('^{0}::{1}', REF_NAME, TAIL)); - var rx_substitution = new RegExp( - format('^\\|{0}\\|{1}{2}::{3}', TEXT1, SEPA, REF_NAME, TAIL)); - var rx_footnote = new RegExp( - format('^\\[(?:\\d+|#{0}?|\\*)]{1}', REF_NAME, TAIL)); - var rx_citation = new RegExp( - format('^\\[{0}\\]{1}', REF_NAME, TAIL)); - - var rx_substitution_ref = new RegExp( - format('^\\|{0}\\|', TEXT1)); - var rx_footnote_ref = new RegExp( - format('^\\[(?:\\d+|#{0}?|\\*)]_', REF_NAME)); - var rx_citation_ref = new RegExp( - format('^\\[{0}\\]_', REF_NAME)); - var rx_link_ref1 = new RegExp( - format('^{0}__?', REF_NAME)); - var rx_link_ref2 = new RegExp( - format('^`{0}`_', TEXT2)); - - var rx_role_pre = new RegExp( - format('^:{0}:`{1}`{2}', NAME, TEXT2, TAIL)); - var rx_role_suf = new RegExp( - format('^`{1}`:{0}:{2}', NAME, TEXT2, TAIL)); - var rx_role = new RegExp( - format('^:{0}:{1}', NAME, TAIL)); - - var rx_directive_name = new RegExp(format('^{0}', REF_NAME)); - var rx_directive_tail = new RegExp(format('^::{0}', TAIL)); - var rx_substitution_text = new RegExp(format('^\\|{0}\\|', TEXT1)); - var rx_substitution_sepa = new RegExp(format('^{0}', SEPA)); - var rx_substitution_name = new RegExp(format('^{0}', REF_NAME)); - var rx_substitution_tail = new RegExp(format('^::{0}', TAIL)); - var rx_link_head = new RegExp("^_"); - var rx_link_name = new RegExp(format('^{0}|_', REF_NAME)); - var rx_link_tail = new RegExp(format('^:{0}', TAIL)); - - var rx_verbatim = new RegExp('^::\\s*$'); - var rx_examples = new RegExp('^\\s+(?:>>>|In \\[\\d+\\]:)\\s'); - - /////////////////////////////////////////////////////////////////////////// - /////////////////////////////////////////////////////////////////////////// - - function to_normal(stream, state) { - var token = null; - - if (stream.sol() && stream.match(rx_examples, false)) { - change(state, to_mode, { - mode: mode_python, local: mode_python.startState() - }); - } else if (stream.sol() && stream.match(rx_explicit)) { - change(state, to_explicit); - token = 'meta'; - } else if (stream.sol() && stream.match(rx_section)) { - change(state, to_normal); - token = 'header'; - } else if (phase(state) == rx_role_pre || - stream.match(rx_role_pre, false)) { - - switch (stage(state)) { - case 0: - change(state, to_normal, context(rx_role_pre, 1)); - assert(stream.match(/^:/)); - token = 'meta'; - break; - case 1: - change(state, to_normal, context(rx_role_pre, 2)); - assert(stream.match(rx_NAME)); - token = 'keyword'; - - if (stream.current().match(/^(?:math|latex)/)) { - state.tmp = { - mode: mode_stex, local: mode_stex.startState() - }; - } - break; - case 2: - change(state, to_normal, context(rx_role_pre, 3)); - assert(stream.match(/^:`/)); - token = 'meta'; - break; - case 3: - if (state.tmp) { - if (stream.peek() == '`') { - change(state, to_normal, context(rx_role_pre, 4)); - state.tmp = undefined; - break; - } - - token = state.tmp.mode.token(stream, state.tmp.local); - break; - } - - change(state, to_normal, context(rx_role_pre, 4)); - assert(stream.match(rx_TEXT2)); - token = 'string'; - break; - case 4: - change(state, to_normal, context(rx_role_pre, 5)); - assert(stream.match(/^`/)); - token = 'meta'; - break; - case 5: - change(state, to_normal, context(rx_role_pre, 6)); - assert(stream.match(rx_TAIL)); - break; - default: - change(state, to_normal); - assert(stream.current() == ''); - } - } else if (phase(state) == rx_role_suf || - stream.match(rx_role_suf, false)) { - - switch (stage(state)) { - case 0: - change(state, to_normal, context(rx_role_suf, 1)); - assert(stream.match(/^`/)); - token = 'meta'; - break; - case 1: - change(state, to_normal, context(rx_role_suf, 2)); - assert(stream.match(rx_TEXT2)); - token = 'string'; - break; - case 2: - change(state, to_normal, context(rx_role_suf, 3)); - assert(stream.match(/^`:/)); - token = 'meta'; - break; - case 3: - change(state, to_normal, context(rx_role_suf, 4)); - assert(stream.match(rx_NAME)); - token = 'keyword'; - break; - case 4: - change(state, to_normal, context(rx_role_suf, 5)); - assert(stream.match(/^:/)); - token = 'meta'; - break; - case 5: - change(state, to_normal, context(rx_role_suf, 6)); - assert(stream.match(rx_TAIL)); - break; - default: - change(state, to_normal); - assert(stream.current() == ''); - } - } else if (phase(state) == rx_role || stream.match(rx_role, false)) { - - switch (stage(state)) { - case 0: - change(state, to_normal, context(rx_role, 1)); - assert(stream.match(/^:/)); - token = 'meta'; - break; - case 1: - change(state, to_normal, context(rx_role, 2)); - assert(stream.match(rx_NAME)); - token = 'keyword'; - break; - case 2: - change(state, to_normal, context(rx_role, 3)); - assert(stream.match(/^:/)); - token = 'meta'; - break; - case 3: - change(state, to_normal, context(rx_role, 4)); - assert(stream.match(rx_TAIL)); - break; - default: - change(state, to_normal); - assert(stream.current() == ''); - } - } else if (phase(state) == rx_substitution_ref || - stream.match(rx_substitution_ref, false)) { - - switch (stage(state)) { - case 0: - change(state, to_normal, context(rx_substitution_ref, 1)); - assert(stream.match(rx_substitution_text)); - token = 'variable-2'; - break; - case 1: - change(state, to_normal, context(rx_substitution_ref, 2)); - if (stream.match(/^_?_?/)) token = 'link'; - break; - default: - change(state, to_normal); - assert(stream.current() == ''); - } - } else if (stream.match(rx_footnote_ref)) { - change(state, to_normal); - token = 'quote'; - } else if (stream.match(rx_citation_ref)) { - change(state, to_normal); - token = 'quote'; - } else if (stream.match(rx_link_ref1)) { - change(state, to_normal); - if (!stream.peek() || stream.peek().match(/^\W$/)) { - token = 'link'; - } - } else if (phase(state) == rx_link_ref2 || - stream.match(rx_link_ref2, false)) { - - switch (stage(state)) { - case 0: - if (!stream.peek() || stream.peek().match(/^\W$/)) { - change(state, to_normal, context(rx_link_ref2, 1)); - } else { - stream.match(rx_link_ref2); - } - break; - case 1: - change(state, to_normal, context(rx_link_ref2, 2)); - assert(stream.match(/^`/)); - token = 'link'; - break; - case 2: - change(state, to_normal, context(rx_link_ref2, 3)); - assert(stream.match(rx_TEXT2)); - break; - case 3: - change(state, to_normal, context(rx_link_ref2, 4)); - assert(stream.match(/^`_/)); - token = 'link'; - break; - default: - change(state, to_normal); - assert(stream.current() == ''); - } - } else if (stream.match(rx_verbatim)) { - change(state, to_verbatim); - } - - else { - if (stream.next()) change(state, to_normal); - } - - return token; - } - - /////////////////////////////////////////////////////////////////////////// - /////////////////////////////////////////////////////////////////////////// - - function to_explicit(stream, state) { - var token = null; - - if (phase(state) == rx_substitution || - stream.match(rx_substitution, false)) { - - switch (stage(state)) { - case 0: - change(state, to_explicit, context(rx_substitution, 1)); - assert(stream.match(rx_substitution_text)); - token = 'variable-2'; - break; - case 1: - change(state, to_explicit, context(rx_substitution, 2)); - assert(stream.match(rx_substitution_sepa)); - break; - case 2: - change(state, to_explicit, context(rx_substitution, 3)); - assert(stream.match(rx_substitution_name)); - token = 'keyword'; - break; - case 3: - change(state, to_explicit, context(rx_substitution, 4)); - assert(stream.match(rx_substitution_tail)); - token = 'meta'; - break; - default: - change(state, to_normal); - assert(stream.current() == ''); - } - } else if (phase(state) == rx_directive || - stream.match(rx_directive, false)) { - - switch (stage(state)) { - case 0: - change(state, to_explicit, context(rx_directive, 1)); - assert(stream.match(rx_directive_name)); - token = 'keyword'; - - if (stream.current().match(/^(?:math|latex)/)) - state.tmp_stex = true; - else if (stream.current().match(/^python/)) - state.tmp_py = true; - break; - case 1: - change(state, to_explicit, context(rx_directive, 2)); - assert(stream.match(rx_directive_tail)); - token = 'meta'; - break; - default: - if (stream.match(/^latex\s*$/) || state.tmp_stex) { - state.tmp_stex = undefined; - change(state, to_mode, { - mode: mode_stex, local: mode_stex.startState() - }); - } else if (stream.match(/^python\s*$/) || state.tmp_py) { - state.tmp_py = undefined; - change(state, to_mode, { - mode: mode_python, local: mode_python.startState() - }); - } - - else { - change(state, to_normal); - assert(stream.current() == ''); - } - } - } else if (phase(state) == rx_link || stream.match(rx_link, false)) { - - switch (stage(state)) { - case 0: - change(state, to_explicit, context(rx_link, 1)); - assert(stream.match(rx_link_head)); - assert(stream.match(rx_link_name)); - token = 'link'; - break; - case 1: - change(state, to_explicit, context(rx_link, 2)); - assert(stream.match(rx_link_tail)); - token = 'meta'; - break; - default: - change(state, to_normal); - assert(stream.current() == ''); - } - } else if (stream.match(rx_footnote)) { - change(state, to_normal); - token = 'quote'; - } else if (stream.match(rx_citation)) { - change(state, to_normal); - token = 'quote'; - } - - else { - stream.eatSpace(); - if (stream.eol()) { - change(state, to_normal); - } else { - stream.skipToEnd(); - change(state, to_comment); - token = 'comment'; - } - } - - return token; - } - - /////////////////////////////////////////////////////////////////////////// - /////////////////////////////////////////////////////////////////////////// - - function to_comment(stream, state) { - return as_block(stream, state, 'comment'); - } - - function to_verbatim(stream, state) { - return as_block(stream, state, 'meta'); - } - - function as_block(stream, state, token) { - if (stream.eol() || stream.eatSpace()) { - stream.skipToEnd(); - return token; - } else { - change(state, to_normal); - return null; - } - } - - /////////////////////////////////////////////////////////////////////////// - /////////////////////////////////////////////////////////////////////////// - - function to_mode(stream, state) { - - if (state.ctx.mode && state.ctx.local) { - - if (stream.sol()) { - if (!stream.eatSpace()) change(state, to_normal); - return null; - } - - try { - return state.ctx.mode.token(stream, state.ctx.local); - } catch (ex) { - change(state, to_normal); - return null; - } - } - - change(state, to_normal); - return null; - } - - /////////////////////////////////////////////////////////////////////////// - /////////////////////////////////////////////////////////////////////////// - - function context(phase, stage, mode, local) { - return {phase: phase, stage: stage, mode: mode, local: local}; - } - - function change(state, tok, ctx) { - state.tok = tok; - state.ctx = ctx || {}; - } - - function stage(state) { - return state.ctx.stage || 0; - } - - function phase(state) { - return state.ctx.phase; - } - - /////////////////////////////////////////////////////////////////////////// - /////////////////////////////////////////////////////////////////////////// - - return { - startState: function () { - return {tok: to_normal, ctx: context(undefined, 0)}; - }, - - copyState: function (state) { - return {tok: state.tok, ctx: state.ctx}; - }, - - innerMode: function (state) { - return {state: state.ctx.local, mode: state.ctx.mode}; - }, - - token: function (stream, state) { - return state.tok(stream, state); - } - }; -}, 'python', 'stex'); - -/////////////////////////////////////////////////////////////////////////////// -/////////////////////////////////////////////////////////////////////////////// - -CodeMirror.defineMode('rst', function (config, options) { - - var rx_uri_protocol = "[Hh][Tt][Tt][Pp][Ss]?://"; - var rx_uri_domain = "(?:[\\d\\w.-]+)\\.(?:\\w{2,6})"; - var rx_uri_path = "(?:/[\\d\\w\\#\\%\\&\\-\\.\\,\\/\\:\\=\\?\\~]+)*"; - var rx_uri = new RegExp("^" + - rx_uri_protocol + rx_uri_domain + rx_uri_path - ); - - var rx_strong = /^\*\*[^\*\s](?:[^\*]*[^\*\s])?\*\*(\s+|$)/; - var rx_emphasis = /^[^\*]\*[^\*\s](?:[^\*]*[^\*\s])?\*(\s+|$)/; - var rx_literal = /^``[^`\s](?:[^`]*[^`\s])``(\s+|$)/; - - var rx_number = /^(?:[\d]+(?:[\.,]\d+)*)/; - var rx_positive = /^(?:\s\+[\d]+(?:[\.,]\d+)*)/; - var rx_negative = /^(?:\s\-[\d]+(?:[\.,]\d+)*)/; - - var overlay = { - token: function (stream) { - - if (stream.match(rx_uri)) return 'link'; - if (stream.match(rx_strong)) return 'strong'; - if (stream.match(rx_emphasis)) return 'em'; - if (stream.match(rx_literal)) return 'string-2'; - if (stream.match(rx_number)) return 'number'; - if (stream.match(rx_positive)) return 'positive'; - if (stream.match(rx_negative)) return 'negative'; - - while (stream.next() != null) { - if (stream.match(rx_uri, false)) break; - if (stream.match(rx_strong, false)) break; - if (stream.match(rx_emphasis, false)) break; - if (stream.match(rx_literal, false)) break; - if (stream.match(rx_number, false)) break; - if (stream.match(rx_positive, false)) break; - if (stream.match(rx_negative, false)) break; - } - - return null; - } - }; - - var mode = CodeMirror.getMode( - config, options.backdrop || 'rst-base' - ); - - return CodeMirror.overlayMode(mode, overlay, true); // combine -}, 'python', 'stex'); - -/////////////////////////////////////////////////////////////////////////////// -/////////////////////////////////////////////////////////////////////////////// - -CodeMirror.defineMIME('text/x-rst', 'rst'); - -/////////////////////////////////////////////////////////////////////////////// -/////////////////////////////////////////////////////////////////////////////// diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/LICENSE b/wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/LICENSE deleted file mode 100644 index ac09fc4035..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/LICENSE +++ /dev/null @@ -1,24 +0,0 @@ -Copyright (c) 2011, Ubalo, Inc. -All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are met: - * Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - * Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - * Neither the name of the Ubalo, Inc. nor the names of its - contributors may be used to endorse or promote products derived - from this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND -ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED -WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE -DISCLAIMED. IN NO EVENT SHALL UBALO, INC BE LIABLE FOR ANY -DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES -(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; -LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND -ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS -SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/index.html deleted file mode 100644 index 64cfe5ef34..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/index.html +++ /dev/null @@ -1,173 +0,0 @@ - - - - - CodeMirror: Ruby mode - - - - - - - - -

CodeMirror: Ruby mode

-
- - -

MIME types defined: text/x-ruby.

- -

Development of the CodeMirror Ruby mode was kindly sponsored - by Ubalo, who hold - the license.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/ruby.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/ruby.js deleted file mode 100644 index 883244b1c7..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/ruby/ruby.js +++ /dev/null @@ -1,196 +0,0 @@ -CodeMirror.defineMode("ruby", function(config) { - function wordObj(words) { - var o = {}; - for (var i = 0, e = words.length; i < e; ++i) o[words[i]] = true; - return o; - } - var keywords = wordObj([ - "alias", "and", "BEGIN", "begin", "break", "case", "class", "def", "defined?", "do", "else", - "elsif", "END", "end", "ensure", "false", "for", "if", "in", "module", "next", "not", "or", - "redo", "rescue", "retry", "return", "self", "super", "then", "true", "undef", "unless", - "until", "when", "while", "yield", "nil", "raise", "throw", "catch", "fail", "loop", "callcc", - "caller", "lambda", "proc", "public", "protected", "private", "require", "load", - "require_relative", "extend", "autoload" - ]); - var indentWords = wordObj(["def", "class", "case", "for", "while", "do", "module", "then", - "catch", "loop", "proc", "begin"]); - var dedentWords = wordObj(["end", "until"]); - var matching = {"[": "]", "{": "}", "(": ")"}; - var curPunc; - - function chain(newtok, stream, state) { - state.tokenize.push(newtok); - return newtok(stream, state); - } - - function tokenBase(stream, state) { - curPunc = null; - if (stream.sol() && stream.match("=begin") && stream.eol()) { - state.tokenize.push(readBlockComment); - return "comment"; - } - if (stream.eatSpace()) return null; - var ch = stream.next(), m; - if (ch == "`" || ch == "'" || ch == '"' || - (ch == "/" && !stream.eol() && stream.peek() != " ")) { - return chain(readQuoted(ch, "string", ch == '"' || ch == "`"), stream, state); - } else if (ch == "%") { - var style, embed = false; - if (stream.eat("s")) style = "atom"; - else if (stream.eat(/[WQ]/)) { style = "string"; embed = true; } - else if (stream.eat(/[wxqr]/)) style = "string"; - var delim = stream.eat(/[^\w\s]/); - if (!delim) return "operator"; - if (matching.propertyIsEnumerable(delim)) delim = matching[delim]; - return chain(readQuoted(delim, style, embed, true), stream, state); - } else if (ch == "#") { - stream.skipToEnd(); - return "comment"; - } else if (ch == "<" && (m = stream.match(/^<-?[\`\"\']?([a-zA-Z_?]\w*)[\`\"\']?(?:;|$)/))) { - return chain(readHereDoc(m[1]), stream, state); - } else if (ch == "0") { - if (stream.eat("x")) stream.eatWhile(/[\da-fA-F]/); - else if (stream.eat("b")) stream.eatWhile(/[01]/); - else stream.eatWhile(/[0-7]/); - return "number"; - } else if (/\d/.test(ch)) { - stream.match(/^[\d_]*(?:\.[\d_]+)?(?:[eE][+\-]?[\d_]+)?/); - return "number"; - } else if (ch == "?") { - while (stream.match(/^\\[CM]-/)) {} - if (stream.eat("\\")) stream.eatWhile(/\w/); - else stream.next(); - return "string"; - } else if (ch == ":") { - if (stream.eat("'")) return chain(readQuoted("'", "atom", false), stream, state); - if (stream.eat('"')) return chain(readQuoted('"', "atom", true), stream, state); - stream.eatWhile(/[\w\?]/); - return "atom"; - } else if (ch == "@") { - stream.eat("@"); - stream.eatWhile(/[\w\?]/); - return "variable-2"; - } else if (ch == "$") { - stream.next(); - stream.eatWhile(/[\w\?]/); - return "variable-3"; - } else if (/\w/.test(ch)) { - stream.eatWhile(/[\w\?]/); - if (stream.eat(":")) return "atom"; - return "ident"; - } else if (ch == "|" && (state.varList || state.lastTok == "{" || state.lastTok == "do")) { - curPunc = "|"; - return null; - } else if (/[\(\)\[\]{}\\;]/.test(ch)) { - curPunc = ch; - return null; - } else if (ch == "-" && stream.eat(">")) { - return "arrow"; - } else if (/[=+\-\/*:\.^%<>~|]/.test(ch)) { - stream.eatWhile(/[=+\-\/*:\.^%<>~|]/); - return "operator"; - } else { - return null; - } - } - - function tokenBaseUntilBrace() { - var depth = 1; - return function(stream, state) { - if (stream.peek() == "}") { - depth--; - if (depth == 0) { - state.tokenize.pop(); - return state.tokenize[state.tokenize.length-1](stream, state); - } - } else if (stream.peek() == "{") { - depth++; - } - return tokenBase(stream, state); - }; - } - function readQuoted(quote, style, embed, unescaped) { - return function(stream, state) { - var escaped = false, ch; - while ((ch = stream.next()) != null) { - if (ch == quote && (unescaped || !escaped)) { - state.tokenize.pop(); - break; - } - if (embed && ch == "#" && !escaped && stream.eat("{")) { - state.tokenize.push(tokenBaseUntilBrace(arguments.callee)); - break; - } - escaped = !escaped && ch == "\\"; - } - return style; - }; - } - function readHereDoc(phrase) { - return function(stream, state) { - if (stream.match(phrase)) state.tokenize.pop(); - else stream.skipToEnd(); - return "string"; - }; - } - function readBlockComment(stream, state) { - if (stream.sol() && stream.match("=end") && stream.eol()) - state.tokenize.pop(); - stream.skipToEnd(); - return "comment"; - } - - return { - startState: function() { - return {tokenize: [tokenBase], - indented: 0, - context: {type: "top", indented: -config.indentUnit}, - continuedLine: false, - lastTok: null, - varList: false}; - }, - - token: function(stream, state) { - if (stream.sol()) state.indented = stream.indentation(); - var style = state.tokenize[state.tokenize.length-1](stream, state), kwtype; - if (style == "ident") { - var word = stream.current(); - style = keywords.propertyIsEnumerable(stream.current()) ? "keyword" - : /^[A-Z]/.test(word) ? "tag" - : (state.lastTok == "def" || state.lastTok == "class" || state.varList) ? "def" - : "variable"; - if (indentWords.propertyIsEnumerable(word)) kwtype = "indent"; - else if (dedentWords.propertyIsEnumerable(word)) kwtype = "dedent"; - else if ((word == "if" || word == "unless") && stream.column() == stream.indentation()) - kwtype = "indent"; - } - if (curPunc || (style && style != "comment")) state.lastTok = word || curPunc || style; - if (curPunc == "|") state.varList = !state.varList; - - if (kwtype == "indent" || /[\(\[\{]/.test(curPunc)) - state.context = {prev: state.context, type: curPunc || style, indented: state.indented}; - else if ((kwtype == "dedent" || /[\)\]\}]/.test(curPunc)) && state.context.prev) - state.context = state.context.prev; - - if (stream.eol()) - state.continuedLine = (curPunc == "\\" || style == "operator"); - return style; - }, - - indent: function(state, textAfter) { - if (state.tokenize[state.tokenize.length-1] != tokenBase) return 0; - var firstChar = textAfter && textAfter.charAt(0); - var ct = state.context; - var closing = ct.type == matching[firstChar] || - ct.type == "keyword" && /^(?:end|until|else|elsif|when|rescue)\b/.test(textAfter); - return ct.indented + (closing ? 0 : config.indentUnit) + - (state.continuedLine ? config.indentUnit : 0); - }, - - electricChars: "}de", // enD and rescuE - lineComment: "#" - }; -}); - -CodeMirror.defineMIME("text/x-ruby", "ruby"); - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rust/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rust/index.html deleted file mode 100644 index a6d47fe84d..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rust/index.html +++ /dev/null @@ -1,48 +0,0 @@ - - - - - CodeMirror: Rust mode - - - - - - - -

CodeMirror: Rust mode

- -
- - - -

MIME types defined: text/x-rustsrc.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rust/rust.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/rust/rust.js deleted file mode 100644 index 7bee489b47..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/rust/rust.js +++ /dev/null @@ -1,435 +0,0 @@ -CodeMirror.defineMode("rust", function() { - var indentUnit = 4, altIndentUnit = 2; - var valKeywords = { - "if": "if-style", "while": "if-style", "else": "else-style", - "do": "else-style", "ret": "else-style", "fail": "else-style", - "break": "atom", "cont": "atom", "const": "let", "resource": "fn", - "let": "let", "fn": "fn", "for": "for", "alt": "alt", "iface": "iface", - "impl": "impl", "type": "type", "enum": "enum", "mod": "mod", - "as": "op", "true": "atom", "false": "atom", "assert": "op", "check": "op", - "claim": "op", "native": "ignore", "unsafe": "ignore", "import": "else-style", - "export": "else-style", "copy": "op", "log": "op", "log_err": "op", - "use": "op", "bind": "op", "self": "atom" - }; - var typeKeywords = function() { - var keywords = {"fn": "fn", "block": "fn", "obj": "obj"}; - var atoms = "bool uint int i8 i16 i32 i64 u8 u16 u32 u64 float f32 f64 str char".split(" "); - for (var i = 0, e = atoms.length; i < e; ++i) keywords[atoms[i]] = "atom"; - return keywords; - }(); - var operatorChar = /[+\-*&%=<>!?|\.@]/; - - // Tokenizer - - // Used as scratch variable to communicate multiple values without - // consing up tons of objects. - var tcat, content; - function r(tc, style) { - tcat = tc; - return style; - } - - function tokenBase(stream, state) { - var ch = stream.next(); - if (ch == '"') { - state.tokenize = tokenString; - return state.tokenize(stream, state); - } - if (ch == "'") { - tcat = "atom"; - if (stream.eat("\\")) { - if (stream.skipTo("'")) { stream.next(); return "string"; } - else { return "error"; } - } else { - stream.next(); - return stream.eat("'") ? "string" : "error"; - } - } - if (ch == "/") { - if (stream.eat("/")) { stream.skipToEnd(); return "comment"; } - if (stream.eat("*")) { - state.tokenize = tokenComment(1); - return state.tokenize(stream, state); - } - } - if (ch == "#") { - if (stream.eat("[")) { tcat = "open-attr"; return null; } - stream.eatWhile(/\w/); - return r("macro", "meta"); - } - if (ch == ":" && stream.match(":<")) { - return r("op", null); - } - if (ch.match(/\d/) || (ch == "." && stream.eat(/\d/))) { - var flp = false; - if (!stream.match(/^x[\da-f]+/i) && !stream.match(/^b[01]+/)) { - stream.eatWhile(/\d/); - if (stream.eat(".")) { flp = true; stream.eatWhile(/\d/); } - if (stream.match(/^e[+\-]?\d+/i)) { flp = true; } - } - if (flp) stream.match(/^f(?:32|64)/); - else stream.match(/^[ui](?:8|16|32|64)/); - return r("atom", "number"); - } - if (ch.match(/[()\[\]{}:;,]/)) return r(ch, null); - if (ch == "-" && stream.eat(">")) return r("->", null); - if (ch.match(operatorChar)) { - stream.eatWhile(operatorChar); - return r("op", null); - } - stream.eatWhile(/\w/); - content = stream.current(); - if (stream.match(/^::\w/)) { - stream.backUp(1); - return r("prefix", "variable-2"); - } - if (state.keywords.propertyIsEnumerable(content)) - return r(state.keywords[content], content.match(/true|false/) ? "atom" : "keyword"); - return r("name", "variable"); - } - - function tokenString(stream, state) { - var ch, escaped = false; - while (ch = stream.next()) { - if (ch == '"' && !escaped) { - state.tokenize = tokenBase; - return r("atom", "string"); - } - escaped = !escaped && ch == "\\"; - } - // Hack to not confuse the parser when a string is split in - // pieces. - return r("op", "string"); - } - - function tokenComment(depth) { - return function(stream, state) { - var lastCh = null, ch; - while (ch = stream.next()) { - if (ch == "/" && lastCh == "*") { - if (depth == 1) { - state.tokenize = tokenBase; - break; - } else { - state.tokenize = tokenComment(depth - 1); - return state.tokenize(stream, state); - } - } - if (ch == "*" && lastCh == "/") { - state.tokenize = tokenComment(depth + 1); - return state.tokenize(stream, state); - } - lastCh = ch; - } - return "comment"; - }; - } - - // Parser - - var cx = {state: null, stream: null, marked: null, cc: null}; - function pass() { - for (var i = arguments.length - 1; i >= 0; i--) cx.cc.push(arguments[i]); - } - function cont() { - pass.apply(null, arguments); - return true; - } - - function pushlex(type, info) { - var result = function() { - var state = cx.state; - state.lexical = {indented: state.indented, column: cx.stream.column(), - type: type, prev: state.lexical, info: info}; - }; - result.lex = true; - return result; - } - function poplex() { - var state = cx.state; - if (state.lexical.prev) { - if (state.lexical.type == ")") - state.indented = state.lexical.indented; - state.lexical = state.lexical.prev; - } - } - function typecx() { cx.state.keywords = typeKeywords; } - function valcx() { cx.state.keywords = valKeywords; } - poplex.lex = typecx.lex = valcx.lex = true; - - function commasep(comb, end) { - function more(type) { - if (type == ",") return cont(comb, more); - if (type == end) return cont(); - return cont(more); - } - return function(type) { - if (type == end) return cont(); - return pass(comb, more); - }; - } - - function stat_of(comb, tag) { - return cont(pushlex("stat", tag), comb, poplex, block); - } - function block(type) { - if (type == "}") return cont(); - if (type == "let") return stat_of(letdef1, "let"); - if (type == "fn") return stat_of(fndef); - if (type == "type") return cont(pushlex("stat"), tydef, endstatement, poplex, block); - if (type == "enum") return stat_of(enumdef); - if (type == "mod") return stat_of(mod); - if (type == "iface") return stat_of(iface); - if (type == "impl") return stat_of(impl); - if (type == "open-attr") return cont(pushlex("]"), commasep(expression, "]"), poplex); - if (type == "ignore" || type.match(/[\]\);,]/)) return cont(block); - return pass(pushlex("stat"), expression, poplex, endstatement, block); - } - function endstatement(type) { - if (type == ";") return cont(); - return pass(); - } - function expression(type) { - if (type == "atom" || type == "name") return cont(maybeop); - if (type == "{") return cont(pushlex("}"), exprbrace, poplex); - if (type.match(/[\[\(]/)) return matchBrackets(type, expression); - if (type.match(/[\]\)\};,]/)) return pass(); - if (type == "if-style") return cont(expression, expression); - if (type == "else-style" || type == "op") return cont(expression); - if (type == "for") return cont(pattern, maybetype, inop, expression, expression); - if (type == "alt") return cont(expression, altbody); - if (type == "fn") return cont(fndef); - if (type == "macro") return cont(macro); - return cont(); - } - function maybeop(type) { - if (content == ".") return cont(maybeprop); - if (content == "::<"){return cont(typarams, maybeop);} - if (type == "op" || content == ":") return cont(expression); - if (type == "(" || type == "[") return matchBrackets(type, expression); - return pass(); - } - function maybeprop() { - if (content.match(/^\w+$/)) {cx.marked = "variable"; return cont(maybeop);} - return pass(expression); - } - function exprbrace(type) { - if (type == "op") { - if (content == "|") return cont(blockvars, poplex, pushlex("}", "block"), block); - if (content == "||") return cont(poplex, pushlex("}", "block"), block); - } - if (content == "mutable" || (content.match(/^\w+$/) && cx.stream.peek() == ":" - && !cx.stream.match("::", false))) - return pass(record_of(expression)); - return pass(block); - } - function record_of(comb) { - function ro(type) { - if (content == "mutable" || content == "with") {cx.marked = "keyword"; return cont(ro);} - if (content.match(/^\w*$/)) {cx.marked = "variable"; return cont(ro);} - if (type == ":") return cont(comb, ro); - if (type == "}") return cont(); - return cont(ro); - } - return ro; - } - function blockvars(type) { - if (type == "name") {cx.marked = "def"; return cont(blockvars);} - if (type == "op" && content == "|") return cont(); - return cont(blockvars); - } - - function letdef1(type) { - if (type.match(/[\]\)\};]/)) return cont(); - if (content == "=") return cont(expression, letdef2); - if (type == ",") return cont(letdef1); - return pass(pattern, maybetype, letdef1); - } - function letdef2(type) { - if (type.match(/[\]\)\};,]/)) return pass(letdef1); - else return pass(expression, letdef2); - } - function maybetype(type) { - if (type == ":") return cont(typecx, rtype, valcx); - return pass(); - } - function inop(type) { - if (type == "name" && content == "in") {cx.marked = "keyword"; return cont();} - return pass(); - } - function fndef(type) { - if (content == "@" || content == "~") {cx.marked = "keyword"; return cont(fndef);} - if (type == "name") {cx.marked = "def"; return cont(fndef);} - if (content == "<") return cont(typarams, fndef); - if (type == "{") return pass(expression); - if (type == "(") return cont(pushlex(")"), commasep(argdef, ")"), poplex, fndef); - if (type == "->") return cont(typecx, rtype, valcx, fndef); - if (type == ";") return cont(); - return cont(fndef); - } - function tydef(type) { - if (type == "name") {cx.marked = "def"; return cont(tydef);} - if (content == "<") return cont(typarams, tydef); - if (content == "=") return cont(typecx, rtype, valcx); - return cont(tydef); - } - function enumdef(type) { - if (type == "name") {cx.marked = "def"; return cont(enumdef);} - if (content == "<") return cont(typarams, enumdef); - if (content == "=") return cont(typecx, rtype, valcx, endstatement); - if (type == "{") return cont(pushlex("}"), typecx, enumblock, valcx, poplex); - return cont(enumdef); - } - function enumblock(type) { - if (type == "}") return cont(); - if (type == "(") return cont(pushlex(")"), commasep(rtype, ")"), poplex, enumblock); - if (content.match(/^\w+$/)) cx.marked = "def"; - return cont(enumblock); - } - function mod(type) { - if (type == "name") {cx.marked = "def"; return cont(mod);} - if (type == "{") return cont(pushlex("}"), block, poplex); - return pass(); - } - function iface(type) { - if (type == "name") {cx.marked = "def"; return cont(iface);} - if (content == "<") return cont(typarams, iface); - if (type == "{") return cont(pushlex("}"), block, poplex); - return pass(); - } - function impl(type) { - if (content == "<") return cont(typarams, impl); - if (content == "of" || content == "for") {cx.marked = "keyword"; return cont(rtype, impl);} - if (type == "name") {cx.marked = "def"; return cont(impl);} - if (type == "{") return cont(pushlex("}"), block, poplex); - return pass(); - } - function typarams() { - if (content == ">") return cont(); - if (content == ",") return cont(typarams); - if (content == ":") return cont(rtype, typarams); - return pass(rtype, typarams); - } - function argdef(type) { - if (type == "name") {cx.marked = "def"; return cont(argdef);} - if (type == ":") return cont(typecx, rtype, valcx); - return pass(); - } - function rtype(type) { - if (type == "name") {cx.marked = "variable-3"; return cont(rtypemaybeparam); } - if (content == "mutable") {cx.marked = "keyword"; return cont(rtype);} - if (type == "atom") return cont(rtypemaybeparam); - if (type == "op" || type == "obj") return cont(rtype); - if (type == "fn") return cont(fntype); - if (type == "{") return cont(pushlex("{"), record_of(rtype), poplex); - return matchBrackets(type, rtype); - } - function rtypemaybeparam() { - if (content == "<") return cont(typarams); - return pass(); - } - function fntype(type) { - if (type == "(") return cont(pushlex("("), commasep(rtype, ")"), poplex, fntype); - if (type == "->") return cont(rtype); - return pass(); - } - function pattern(type) { - if (type == "name") {cx.marked = "def"; return cont(patternmaybeop);} - if (type == "atom") return cont(patternmaybeop); - if (type == "op") return cont(pattern); - if (type.match(/[\]\)\};,]/)) return pass(); - return matchBrackets(type, pattern); - } - function patternmaybeop(type) { - if (type == "op" && content == ".") return cont(); - if (content == "to") {cx.marked = "keyword"; return cont(pattern);} - else return pass(); - } - function altbody(type) { - if (type == "{") return cont(pushlex("}", "alt"), altblock1, poplex); - return pass(); - } - function altblock1(type) { - if (type == "}") return cont(); - if (type == "|") return cont(altblock1); - if (content == "when") {cx.marked = "keyword"; return cont(expression, altblock2);} - if (type.match(/[\]\);,]/)) return cont(altblock1); - return pass(pattern, altblock2); - } - function altblock2(type) { - if (type == "{") return cont(pushlex("}", "alt"), block, poplex, altblock1); - else return pass(altblock1); - } - - function macro(type) { - if (type.match(/[\[\(\{]/)) return matchBrackets(type, expression); - return pass(); - } - function matchBrackets(type, comb) { - if (type == "[") return cont(pushlex("]"), commasep(comb, "]"), poplex); - if (type == "(") return cont(pushlex(")"), commasep(comb, ")"), poplex); - if (type == "{") return cont(pushlex("}"), commasep(comb, "}"), poplex); - return cont(); - } - - function parse(state, stream, style) { - var cc = state.cc; - // Communicate our context to the combinators. - // (Less wasteful than consing up a hundred closures on every call.) - cx.state = state; cx.stream = stream; cx.marked = null, cx.cc = cc; - - while (true) { - var combinator = cc.length ? cc.pop() : block; - if (combinator(tcat)) { - while(cc.length && cc[cc.length - 1].lex) - cc.pop()(); - return cx.marked || style; - } - } - } - - return { - startState: function() { - return { - tokenize: tokenBase, - cc: [], - lexical: {indented: -indentUnit, column: 0, type: "top", align: false}, - keywords: valKeywords, - indented: 0 - }; - }, - - token: function(stream, state) { - if (stream.sol()) { - if (!state.lexical.hasOwnProperty("align")) - state.lexical.align = false; - state.indented = stream.indentation(); - } - if (stream.eatSpace()) return null; - tcat = content = null; - var style = state.tokenize(stream, state); - if (style == "comment") return style; - if (!state.lexical.hasOwnProperty("align")) - state.lexical.align = true; - if (tcat == "prefix") return style; - if (!content) content = stream.current(); - return parse(state, stream, style); - }, - - indent: function(state, textAfter) { - if (state.tokenize != tokenBase) return 0; - var firstChar = textAfter && textAfter.charAt(0), lexical = state.lexical, - type = lexical.type, closing = firstChar == type; - if (type == "stat") return lexical.indented + indentUnit; - if (lexical.align) return lexical.column + (closing ? 0 : 1); - return lexical.indented + (closing ? 0 : (lexical.info == "alt" ? altIndentUnit : indentUnit)); - }, - - electricChars: "{}", - blockCommentStart: "/*", - blockCommentEnd: "*/", - lineComment: "//" - }; -}); - -CodeMirror.defineMIME("text/x-rustsrc", "rust"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sass/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/sass/index.html deleted file mode 100644 index 3af7bff9ed..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sass/index.html +++ /dev/null @@ -1,54 +0,0 @@ - - - - - CodeMirror: Sass mode - - - - - - - - -

CodeMirror: Sass mode

-
- - -

MIME types defined: text/x-sass.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sass/sass.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/sass/sass.js deleted file mode 100644 index 9c9a0dae07..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sass/sass.js +++ /dev/null @@ -1,330 +0,0 @@ -CodeMirror.defineMode("sass", function(config) { - var tokenRegexp = function(words){ - return new RegExp("^" + words.join("|")); - }; - - var keywords = ["true", "false", "null", "auto"]; - var keywordsRegexp = new RegExp("^" + keywords.join("|")); - - var operators = ["\\(", "\\)", "=", ">", "<", "==", ">=", "<=", "\\+", "-", "\\!=", "/", "\\*", "%", "and", "or", "not"]; - var opRegexp = tokenRegexp(operators); - - var pseudoElementsRegexp = /^::?[\w\-]+/; - - var urlTokens = function(stream, state){ - var ch = stream.peek(); - - if (ch === ")"){ - stream.next(); - state.tokenizer = tokenBase; - return "operator"; - }else if (ch === "("){ - stream.next(); - stream.eatSpace(); - - return "operator"; - }else if (ch === "'" || ch === '"'){ - state.tokenizer = buildStringTokenizer(stream.next()); - return "string"; - }else{ - state.tokenizer = buildStringTokenizer(")", false); - return "string"; - } - }; - var multilineComment = function(stream, state) { - if (stream.skipTo("*/")){ - stream.next(); - stream.next(); - state.tokenizer = tokenBase; - }else { - stream.next(); - } - - return "comment"; - }; - - var buildStringTokenizer = function(quote, greedy){ - if(greedy == null){ greedy = true; } - - function stringTokenizer(stream, state){ - var nextChar = stream.next(); - var peekChar = stream.peek(); - var previousChar = stream.string.charAt(stream.pos-2); - - var endingString = ((nextChar !== "\\" && peekChar === quote) || (nextChar === quote && previousChar !== "\\")); - - /* - console.log("previousChar: " + previousChar); - console.log("nextChar: " + nextChar); - console.log("peekChar: " + peekChar); - console.log("ending: " + endingString); - */ - - if (endingString){ - if (nextChar !== quote && greedy) { stream.next(); } - state.tokenizer = tokenBase; - return "string"; - }else if (nextChar === "#" && peekChar === "{"){ - state.tokenizer = buildInterpolationTokenizer(stringTokenizer); - stream.next(); - return "operator"; - }else { - return "string"; - } - } - - return stringTokenizer; - }; - - var buildInterpolationTokenizer = function(currentTokenizer){ - return function(stream, state){ - if (stream.peek() === "}"){ - stream.next(); - state.tokenizer = currentTokenizer; - return "operator"; - }else{ - return tokenBase(stream, state); - } - }; - }; - - var indent = function(state){ - if (state.indentCount == 0){ - state.indentCount++; - var lastScopeOffset = state.scopes[0].offset; - var currentOffset = lastScopeOffset + config.indentUnit; - state.scopes.unshift({ offset:currentOffset }); - } - }; - - var dedent = function(state){ - if (state.scopes.length == 1) { return; } - - state.scopes.shift(); - }; - - var tokenBase = function(stream, state) { - var ch = stream.peek(); - - // Single line Comment - if (stream.match('//')) { - stream.skipToEnd(); - return "comment"; - } - - // Multiline Comment - if (stream.match('/*')){ - state.tokenizer = multilineComment; - return state.tokenizer(stream, state); - } - - // Interpolation - if (stream.match('#{')){ - state.tokenizer = buildInterpolationTokenizer(tokenBase); - return "operator"; - } - - if (ch === "."){ - stream.next(); - - // Match class selectors - if (stream.match(/^[\w-]+/)){ - indent(state); - return "atom"; - }else if (stream.peek() === "#"){ - indent(state); - return "atom"; - }else{ - return "operator"; - } - } - - if (ch === "#"){ - stream.next(); - - // Hex numbers - if (stream.match(/[0-9a-fA-F]{6}|[0-9a-fA-F]{3}/)){ - return "number"; - } - - // ID selectors - if (stream.match(/^[\w-]+/)){ - indent(state); - return "atom"; - } - - if (stream.peek() === "#"){ - indent(state); - return "atom"; - } - } - - // Numbers - if (stream.match(/^-?[0-9\.]+/)){ - return "number"; - } - - // Units - if (stream.match(/^(px|em|in)\b/)){ - return "unit"; - } - - if (stream.match(keywordsRegexp)){ - return "keyword"; - } - - if (stream.match(/^url/) && stream.peek() === "("){ - state.tokenizer = urlTokens; - return "atom"; - } - - // Variables - if (ch === "$"){ - stream.next(); - stream.eatWhile(/[\w-]/); - - if (stream.peek() === ":"){ - stream.next(); - return "variable-2"; - }else{ - return "variable-3"; - } - } - - if (ch === "!"){ - stream.next(); - - if (stream.match(/^[\w]+/)){ - return "keyword"; - } - - return "operator"; - } - - if (ch === "="){ - stream.next(); - - // Match shortcut mixin definition - if (stream.match(/^[\w-]+/)){ - indent(state); - return "meta"; - }else { - return "operator"; - } - } - - if (ch === "+"){ - stream.next(); - - // Match shortcut mixin definition - if (stream.match(/^[\w-]+/)){ - return "variable-3"; - }else { - return "operator"; - } - } - - // Indent Directives - if (stream.match(/^@(else if|if|media|else|for|each|while|mixin|function)/)){ - indent(state); - return "meta"; - } - - // Other Directives - if (ch === "@"){ - stream.next(); - stream.eatWhile(/[\w-]/); - return "meta"; - } - - // Strings - if (ch === '"' || ch === "'"){ - stream.next(); - state.tokenizer = buildStringTokenizer(ch); - return "string"; - } - - // Pseudo element selectors - if (ch == ':' && stream.match(pseudoElementsRegexp)){ - return "keyword"; - } - - // atoms - if (stream.eatWhile(/[\w-&]/)){ - // matches a property definition - if (stream.peek() === ":" && !stream.match(pseudoElementsRegexp, false)) - return "property"; - else - return "atom"; - } - - if (stream.match(opRegexp)){ - return "operator"; - } - - // If we haven't returned by now, we move 1 character - // and return an error - stream.next(); - return null; - }; - - var tokenLexer = function(stream, state) { - if (stream.sol()){ - state.indentCount = 0; - } - var style = state.tokenizer(stream, state); - var current = stream.current(); - - if (current === "@return"){ - dedent(state); - } - - if (style === "atom"){ - indent(state); - } - - if (style !== null){ - var startOfToken = stream.pos - current.length; - var withCurrentIndent = startOfToken + (config.indentUnit * state.indentCount); - - var newScopes = []; - - for (var i = 0; i < state.scopes.length; i++){ - var scope = state.scopes[i]; - - if (scope.offset <= withCurrentIndent){ - newScopes.push(scope); - } - } - - state.scopes = newScopes; - } - - - return style; - }; - - return { - startState: function() { - return { - tokenizer: tokenBase, - scopes: [{offset: 0, type: 'sass'}], - definedVars: [], - definedMixins: [] - }; - }, - token: function(stream, state) { - var style = tokenLexer(stream, state); - - state.lastToken = { style: style, content: stream.current() }; - - return style; - }, - - indent: function(state) { - return state.scopes[0].offset; - } - }; -}); - -CodeMirror.defineMIME("text/x-sass", "sass"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/scheme/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/scheme/index.html deleted file mode 100644 index 5936a0241f..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/scheme/index.html +++ /dev/null @@ -1,65 +0,0 @@ - - - - - CodeMirror: Scheme mode - - - - - - - -

CodeMirror: Scheme mode

-
- - -

MIME types defined: text/x-scheme.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/scheme/scheme.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/scheme/scheme.js deleted file mode 100644 index c5990ae92c..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/scheme/scheme.js +++ /dev/null @@ -1,232 +0,0 @@ -/** - * Author: Koh Zi Han, based on implementation by Koh Zi Chun - */ -CodeMirror.defineMode("scheme", function () { - var BUILTIN = "builtin", COMMENT = "comment", STRING = "string", - ATOM = "atom", NUMBER = "number", BRACKET = "bracket"; - var INDENT_WORD_SKIP = 2; - - function makeKeywords(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - - var keywords = makeKeywords("λ case-lambda call/cc class define-class exit-handler field import inherit init-field interface let*-values let-values let/ec mixin opt-lambda override protect provide public rename require require-for-syntax syntax syntax-case syntax-error unit/sig unless when with-syntax and begin call-with-current-continuation call-with-input-file call-with-output-file case cond define define-syntax delay do dynamic-wind else for-each if lambda let let* let-syntax letrec letrec-syntax map or syntax-rules abs acos angle append apply asin assoc assq assv atan boolean? caar cadr call-with-input-file call-with-output-file call-with-values car cdddar cddddr cdr ceiling char->integer char-alphabetic? char-ci<=? char-ci=? char-ci>? char-downcase char-lower-case? char-numeric? char-ready? char-upcase char-upper-case? char-whitespace? char<=? char=? char>? char? close-input-port close-output-port complex? cons cos current-input-port current-output-port denominator display eof-object? eq? equal? eqv? eval even? exact->inexact exact? exp expt #f floor force gcd imag-part inexact->exact inexact? input-port? integer->char integer? interaction-environment lcm length list list->string list->vector list-ref list-tail list? load log magnitude make-polar make-rectangular make-string make-vector max member memq memv min modulo negative? newline not null-environment null? number->string number? numerator odd? open-input-file open-output-file output-port? pair? peek-char port? positive? procedure? quasiquote quote quotient rational? rationalize read read-char real-part real? remainder reverse round scheme-report-environment set! set-car! set-cdr! sin sqrt string string->list string->number string->symbol string-append string-ci<=? string-ci=? string-ci>? string-copy string-fill! string-length string-ref string-set! string<=? string=? string>? string? substring symbol->string symbol? #t tan transcript-off transcript-on truncate values vector vector->list vector-fill! vector-length vector-ref vector-set! with-input-from-file with-output-to-file write write-char zero?"); - var indentKeys = makeKeywords("define let letrec let* lambda"); - - function stateStack(indent, type, prev) { // represents a state stack object - this.indent = indent; - this.type = type; - this.prev = prev; - } - - function pushStack(state, indent, type) { - state.indentStack = new stateStack(indent, type, state.indentStack); - } - - function popStack(state) { - state.indentStack = state.indentStack.prev; - } - - var binaryMatcher = new RegExp(/^(?:[-+]i|[-+][01]+#*(?:\/[01]+#*)?i|[-+]?[01]+#*(?:\/[01]+#*)?@[-+]?[01]+#*(?:\/[01]+#*)?|[-+]?[01]+#*(?:\/[01]+#*)?[-+](?:[01]+#*(?:\/[01]+#*)?)?i|[-+]?[01]+#*(?:\/[01]+#*)?)(?=[()\s;"]|$)/i); - var octalMatcher = new RegExp(/^(?:[-+]i|[-+][0-7]+#*(?:\/[0-7]+#*)?i|[-+]?[0-7]+#*(?:\/[0-7]+#*)?@[-+]?[0-7]+#*(?:\/[0-7]+#*)?|[-+]?[0-7]+#*(?:\/[0-7]+#*)?[-+](?:[0-7]+#*(?:\/[0-7]+#*)?)?i|[-+]?[0-7]+#*(?:\/[0-7]+#*)?)(?=[()\s;"]|$)/i); - var hexMatcher = new RegExp(/^(?:[-+]i|[-+][\da-f]+#*(?:\/[\da-f]+#*)?i|[-+]?[\da-f]+#*(?:\/[\da-f]+#*)?@[-+]?[\da-f]+#*(?:\/[\da-f]+#*)?|[-+]?[\da-f]+#*(?:\/[\da-f]+#*)?[-+](?:[\da-f]+#*(?:\/[\da-f]+#*)?)?i|[-+]?[\da-f]+#*(?:\/[\da-f]+#*)?)(?=[()\s;"]|$)/i); - var decimalMatcher = new RegExp(/^(?:[-+]i|[-+](?:(?:(?:\d+#+\.?#*|\d+\.\d*#*|\.\d+#*|\d+)(?:[esfdl][-+]?\d+)?)|\d+#*\/\d+#*)i|[-+]?(?:(?:(?:\d+#+\.?#*|\d+\.\d*#*|\.\d+#*|\d+)(?:[esfdl][-+]?\d+)?)|\d+#*\/\d+#*)@[-+]?(?:(?:(?:\d+#+\.?#*|\d+\.\d*#*|\.\d+#*|\d+)(?:[esfdl][-+]?\d+)?)|\d+#*\/\d+#*)|[-+]?(?:(?:(?:\d+#+\.?#*|\d+\.\d*#*|\.\d+#*|\d+)(?:[esfdl][-+]?\d+)?)|\d+#*\/\d+#*)[-+](?:(?:(?:\d+#+\.?#*|\d+\.\d*#*|\.\d+#*|\d+)(?:[esfdl][-+]?\d+)?)|\d+#*\/\d+#*)?i|(?:(?:(?:\d+#+\.?#*|\d+\.\d*#*|\.\d+#*|\d+)(?:[esfdl][-+]?\d+)?)|\d+#*\/\d+#*))(?=[()\s;"]|$)/i); - - function isBinaryNumber (stream) { - return stream.match(binaryMatcher); - } - - function isOctalNumber (stream) { - return stream.match(octalMatcher); - } - - function isDecimalNumber (stream, backup) { - if (backup === true) { - stream.backUp(1); - } - return stream.match(decimalMatcher); - } - - function isHexNumber (stream) { - return stream.match(hexMatcher); - } - - return { - startState: function () { - return { - indentStack: null, - indentation: 0, - mode: false, - sExprComment: false - }; - }, - - token: function (stream, state) { - if (state.indentStack == null && stream.sol()) { - // update indentation, but only if indentStack is empty - state.indentation = stream.indentation(); - } - - // skip spaces - if (stream.eatSpace()) { - return null; - } - var returnType = null; - - switch(state.mode){ - case "string": // multi-line string parsing mode - var next, escaped = false; - while ((next = stream.next()) != null) { - if (next == "\"" && !escaped) { - - state.mode = false; - break; - } - escaped = !escaped && next == "\\"; - } - returnType = STRING; // continue on in scheme-string mode - break; - case "comment": // comment parsing mode - var next, maybeEnd = false; - while ((next = stream.next()) != null) { - if (next == "#" && maybeEnd) { - - state.mode = false; - break; - } - maybeEnd = (next == "|"); - } - returnType = COMMENT; - break; - case "s-expr-comment": // s-expr commenting mode - state.mode = false; - if(stream.peek() == "(" || stream.peek() == "["){ - // actually start scheme s-expr commenting mode - state.sExprComment = 0; - }else{ - // if not we just comment the entire of the next token - stream.eatWhile(/[^/s]/); // eat non spaces - returnType = COMMENT; - break; - } - default: // default parsing mode - var ch = stream.next(); - - if (ch == "\"") { - state.mode = "string"; - returnType = STRING; - - } else if (ch == "'") { - returnType = ATOM; - } else if (ch == '#') { - if (stream.eat("|")) { // Multi-line comment - state.mode = "comment"; // toggle to comment mode - returnType = COMMENT; - } else if (stream.eat(/[tf]/i)) { // #t/#f (atom) - returnType = ATOM; - } else if (stream.eat(';')) { // S-Expr comment - state.mode = "s-expr-comment"; - returnType = COMMENT; - } else { - var numTest = null, hasExactness = false, hasRadix = true; - if (stream.eat(/[ei]/i)) { - hasExactness = true; - } else { - stream.backUp(1); // must be radix specifier - } - if (stream.match(/^#b/i)) { - numTest = isBinaryNumber; - } else if (stream.match(/^#o/i)) { - numTest = isOctalNumber; - } else if (stream.match(/^#x/i)) { - numTest = isHexNumber; - } else if (stream.match(/^#d/i)) { - numTest = isDecimalNumber; - } else if (stream.match(/^[-+0-9.]/, false)) { - hasRadix = false; - numTest = isDecimalNumber; - // re-consume the intial # if all matches failed - } else if (!hasExactness) { - stream.eat('#'); - } - if (numTest != null) { - if (hasRadix && !hasExactness) { - // consume optional exactness after radix - stream.match(/^#[ei]/i); - } - if (numTest(stream)) - returnType = NUMBER; - } - } - } else if (/^[-+0-9.]/.test(ch) && isDecimalNumber(stream, true)) { // match non-prefixed number, must be decimal - returnType = NUMBER; - } else if (ch == ";") { // comment - stream.skipToEnd(); // rest of the line is a comment - returnType = COMMENT; - } else if (ch == "(" || ch == "[") { - var keyWord = ''; var indentTemp = stream.column(), letter; - /** - Either - (indent-word .. - (non-indent-word .. - (;something else, bracket, etc. - */ - - while ((letter = stream.eat(/[^\s\(\[\;\)\]]/)) != null) { - keyWord += letter; - } - - if (keyWord.length > 0 && indentKeys.propertyIsEnumerable(keyWord)) { // indent-word - - pushStack(state, indentTemp + INDENT_WORD_SKIP, ch); - } else { // non-indent word - // we continue eating the spaces - stream.eatSpace(); - if (stream.eol() || stream.peek() == ";") { - // nothing significant after - // we restart indentation 1 space after - pushStack(state, indentTemp + 1, ch); - } else { - pushStack(state, indentTemp + stream.current().length, ch); // else we match - } - } - stream.backUp(stream.current().length - 1); // undo all the eating - - if(typeof state.sExprComment == "number") state.sExprComment++; - - returnType = BRACKET; - } else if (ch == ")" || ch == "]") { - returnType = BRACKET; - if (state.indentStack != null && state.indentStack.type == (ch == ")" ? "(" : "[")) { - popStack(state); - - if(typeof state.sExprComment == "number"){ - if(--state.sExprComment == 0){ - returnType = COMMENT; // final closing bracket - state.sExprComment = false; // turn off s-expr commenting mode - } - } - } - } else { - stream.eatWhile(/[\w\$_\-!$%&*+\.\/:<=>?@\^~]/); - - if (keywords && keywords.propertyIsEnumerable(stream.current())) { - returnType = BUILTIN; - } else returnType = "variable"; - } - } - return (typeof state.sExprComment == "number") ? COMMENT : returnType; - }, - - indent: function (state) { - if (state.indentStack == null) return state.indentation; - return state.indentStack.indent; - }, - - lineComment: ";;" - }; -}); - -CodeMirror.defineMIME("text/x-scheme", "scheme"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/LICENSE b/wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/LICENSE deleted file mode 100644 index 8a74612cba..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/LICENSE +++ /dev/null @@ -1,19 +0,0 @@ -Copyright (C) 2012 Thomas Schmid - -Permission is hereby granted, free of charge, to any person obtaining a copy -of this software and associated documentation files (the "Software"), to deal -in the Software without restriction, including without limitation the rights -to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -copies of the Software, and to permit persons to whom the Software is -furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in -all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -THE SOFTWARE. diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/index.html deleted file mode 100644 index 8b549815ce..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/index.html +++ /dev/null @@ -1,81 +0,0 @@ - - - - - CodeMirror: Sieve (RFC5228) mode - - - - - - - -

CodeMirror: Sieve (RFC5228) mode

-
- - -

MIME types defined: application/sieve.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/sieve.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/sieve.js deleted file mode 100644 index 8ca2a4cb82..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sieve/sieve.js +++ /dev/null @@ -1,183 +0,0 @@ -/* - * See LICENSE in this directory for the license under which this code - * is released. - */ - -CodeMirror.defineMode("sieve", function(config) { - function words(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - - var keywords = words("if elsif else stop require"); - var atoms = words("true false not"); - var indentUnit = config.indentUnit; - - function tokenBase(stream, state) { - - var ch = stream.next(); - if (ch == "/" && stream.eat("*")) { - state.tokenize = tokenCComment; - return tokenCComment(stream, state); - } - - if (ch === '#') { - stream.skipToEnd(); - return "comment"; - } - - if (ch == "\"") { - state.tokenize = tokenString(ch); - return state.tokenize(stream, state); - } - - if (ch == "(") { - state._indent.push("("); - // add virtual angel wings so that editor behaves... - // ...more sane incase of broken brackets - state._indent.push("{"); - return null; - } - - if (ch === "{") { - state._indent.push("{"); - return null; - } - - if (ch == ")") { - state._indent.pop(); - state._indent.pop(); - } - - if (ch === "}") { - state._indent.pop(); - return null; - } - - if (ch == ",") - return null; - - if (ch == ";") - return null; - - - if (/[{}\(\),;]/.test(ch)) - return null; - - // 1*DIGIT "K" / "M" / "G" - if (/\d/.test(ch)) { - stream.eatWhile(/[\d]/); - stream.eat(/[KkMmGg]/); - return "number"; - } - - // ":" (ALPHA / "_") *(ALPHA / DIGIT / "_") - if (ch == ":") { - stream.eatWhile(/[a-zA-Z_]/); - stream.eatWhile(/[a-zA-Z0-9_]/); - - return "operator"; - } - - stream.eatWhile(/\w/); - var cur = stream.current(); - - // "text:" *(SP / HTAB) (hash-comment / CRLF) - // *(multiline-literal / multiline-dotstart) - // "." CRLF - if ((cur == "text") && stream.eat(":")) - { - state.tokenize = tokenMultiLineString; - return "string"; - } - - if (keywords.propertyIsEnumerable(cur)) - return "keyword"; - - if (atoms.propertyIsEnumerable(cur)) - return "atom"; - - return null; - } - - function tokenMultiLineString(stream, state) - { - state._multiLineString = true; - // the first line is special it may contain a comment - if (!stream.sol()) { - stream.eatSpace(); - - if (stream.peek() == "#") { - stream.skipToEnd(); - return "comment"; - } - - stream.skipToEnd(); - return "string"; - } - - if ((stream.next() == ".") && (stream.eol())) - { - state._multiLineString = false; - state.tokenize = tokenBase; - } - - return "string"; - } - - function tokenCComment(stream, state) { - var maybeEnd = false, ch; - while ((ch = stream.next()) != null) { - if (maybeEnd && ch == "/") { - state.tokenize = tokenBase; - break; - } - maybeEnd = (ch == "*"); - } - return "comment"; - } - - function tokenString(quote) { - return function(stream, state) { - var escaped = false, ch; - while ((ch = stream.next()) != null) { - if (ch == quote && !escaped) - break; - escaped = !escaped && ch == "\\"; - } - if (!escaped) state.tokenize = tokenBase; - return "string"; - }; - } - - return { - startState: function(base) { - return {tokenize: tokenBase, - baseIndent: base || 0, - _indent: []}; - }, - - token: function(stream, state) { - if (stream.eatSpace()) - return null; - - return (state.tokenize || tokenBase)(stream, state);; - }, - - indent: function(state, _textAfter) { - var length = state._indent.length; - if (_textAfter && (_textAfter[0] == "}")) - length--; - - if (length <0) - length = 0; - - return length * indentUnit; - }, - - electricChars: "}" - }; -}); - -CodeMirror.defineMIME("application/sieve", "sieve"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/smalltalk/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/smalltalk/index.html deleted file mode 100644 index b7aebdb7fb..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/smalltalk/index.html +++ /dev/null @@ -1,57 +0,0 @@ - - - - - CodeMirror: Smalltalk mode - - - - - - - - -

CodeMirror: Smalltalk mode

- -
- - - -

Simple Smalltalk mode.

- -

MIME types defined: text/x-stsrc.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/smalltalk/smalltalk.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/smalltalk/smalltalk.js deleted file mode 100644 index eb32417514..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/smalltalk/smalltalk.js +++ /dev/null @@ -1,141 +0,0 @@ -CodeMirror.defineMode('smalltalk', function(config) { - - var specialChars = /[+\-\/\\*~<>=@%|&?!.,:;^]/; - var keywords = /true|false|nil|self|super|thisContext/; - - var Context = function(tokenizer, parent) { - this.next = tokenizer; - this.parent = parent; - }; - - var Token = function(name, context, eos) { - this.name = name; - this.context = context; - this.eos = eos; - }; - - var State = function() { - this.context = new Context(next, null); - this.expectVariable = true; - this.indentation = 0; - this.userIndentationDelta = 0; - }; - - State.prototype.userIndent = function(indentation) { - this.userIndentationDelta = indentation > 0 ? (indentation / config.indentUnit - this.indentation) : 0; - }; - - var next = function(stream, context, state) { - var token = new Token(null, context, false); - var aChar = stream.next(); - - if (aChar === '"') { - token = nextComment(stream, new Context(nextComment, context)); - - } else if (aChar === '\'') { - token = nextString(stream, new Context(nextString, context)); - - } else if (aChar === '#') { - stream.eatWhile(/[^ .\[\]()]/); - token.name = 'string-2'; - - } else if (aChar === '$') { - if (stream.next() === '<') { - stream.eatWhile(/[^ >]/); - stream.next(); - } - token.name = 'string-2'; - - } else if (aChar === '|' && state.expectVariable) { - token.context = new Context(nextTemporaries, context); - - } else if (/[\[\]{}()]/.test(aChar)) { - token.name = 'bracket'; - token.eos = /[\[{(]/.test(aChar); - - if (aChar === '[') { - state.indentation++; - } else if (aChar === ']') { - state.indentation = Math.max(0, state.indentation - 1); - } - - } else if (specialChars.test(aChar)) { - stream.eatWhile(specialChars); - token.name = 'operator'; - token.eos = aChar !== ';'; // ; cascaded message expression - - } else if (/\d/.test(aChar)) { - stream.eatWhile(/[\w\d]/); - token.name = 'number'; - - } else if (/[\w_]/.test(aChar)) { - stream.eatWhile(/[\w\d_]/); - token.name = state.expectVariable ? (keywords.test(stream.current()) ? 'keyword' : 'variable') : null; - - } else { - token.eos = state.expectVariable; - } - - return token; - }; - - var nextComment = function(stream, context) { - stream.eatWhile(/[^"]/); - return new Token('comment', stream.eat('"') ? context.parent : context, true); - }; - - var nextString = function(stream, context) { - stream.eatWhile(/[^']/); - return new Token('string', stream.eat('\'') ? context.parent : context, false); - }; - - var nextTemporaries = function(stream, context) { - var token = new Token(null, context, false); - var aChar = stream.next(); - - if (aChar === '|') { - token.context = context.parent; - token.eos = true; - - } else { - stream.eatWhile(/[^|]/); - token.name = 'variable'; - } - - return token; - }; - - return { - startState: function() { - return new State; - }, - - token: function(stream, state) { - state.userIndent(stream.indentation()); - - if (stream.eatSpace()) { - return null; - } - - var token = state.context.next(stream, state.context, state); - state.context = token.context; - state.expectVariable = token.eos; - - return token.name; - }, - - blankLine: function(state) { - state.userIndent(0); - }, - - indent: function(state, textAfter) { - var i = state.context.next === next && textAfter && textAfter.charAt(0) === ']' ? -1 : state.userIndentationDelta; - return (state.indentation + i) * config.indentUnit; - }, - - electricChars: ']' - }; - -}); - -CodeMirror.defineMIME('text/x-stsrc', {name: 'smalltalk'}); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sparql/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/sparql/index.html deleted file mode 100644 index e7433871c7..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sparql/index.html +++ /dev/null @@ -1,42 +0,0 @@ - - - - - CodeMirror: SPARQL mode - - - - - - - - -

CodeMirror: SPARQL mode

-
- - -

MIME types defined: application/x-sparql-query.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sparql/sparql.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/sparql/sparql.js deleted file mode 100644 index f7237698d3..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/sparql/sparql.js +++ /dev/null @@ -1,143 +0,0 @@ -CodeMirror.defineMode("sparql", function(config) { - var indentUnit = config.indentUnit; - var curPunc; - - function wordRegexp(words) { - return new RegExp("^(?:" + words.join("|") + ")$", "i"); - } - var ops = wordRegexp(["str", "lang", "langmatches", "datatype", "bound", "sameterm", "isiri", "isuri", - "isblank", "isliteral", "union", "a"]); - var keywords = wordRegexp(["base", "prefix", "select", "distinct", "reduced", "construct", "describe", - "ask", "from", "named", "where", "order", "limit", "offset", "filter", "optional", - "graph", "by", "asc", "desc"]); - var operatorChars = /[*+\-<>=&|]/; - - function tokenBase(stream, state) { - var ch = stream.next(); - curPunc = null; - if (ch == "$" || ch == "?") { - stream.match(/^[\w\d]*/); - return "variable-2"; - } - else if (ch == "<" && !stream.match(/^[\s\u00a0=]/, false)) { - stream.match(/^[^\s\u00a0>]*>?/); - return "atom"; - } - else if (ch == "\"" || ch == "'") { - state.tokenize = tokenLiteral(ch); - return state.tokenize(stream, state); - } - else if (/[{}\(\),\.;\[\]]/.test(ch)) { - curPunc = ch; - return null; - } - else if (ch == "#") { - stream.skipToEnd(); - return "comment"; - } - else if (operatorChars.test(ch)) { - stream.eatWhile(operatorChars); - return null; - } - else if (ch == ":") { - stream.eatWhile(/[\w\d\._\-]/); - return "atom"; - } - else { - stream.eatWhile(/[_\w\d]/); - if (stream.eat(":")) { - stream.eatWhile(/[\w\d_\-]/); - return "atom"; - } - var word = stream.current(); - if (ops.test(word)) - return null; - else if (keywords.test(word)) - return "keyword"; - else - return "variable"; - } - } - - function tokenLiteral(quote) { - return function(stream, state) { - var escaped = false, ch; - while ((ch = stream.next()) != null) { - if (ch == quote && !escaped) { - state.tokenize = tokenBase; - break; - } - escaped = !escaped && ch == "\\"; - } - return "string"; - }; - } - - function pushContext(state, type, col) { - state.context = {prev: state.context, indent: state.indent, col: col, type: type}; - } - function popContext(state) { - state.indent = state.context.indent; - state.context = state.context.prev; - } - - return { - startState: function() { - return {tokenize: tokenBase, - context: null, - indent: 0, - col: 0}; - }, - - token: function(stream, state) { - if (stream.sol()) { - if (state.context && state.context.align == null) state.context.align = false; - state.indent = stream.indentation(); - } - if (stream.eatSpace()) return null; - var style = state.tokenize(stream, state); - - if (style != "comment" && state.context && state.context.align == null && state.context.type != "pattern") { - state.context.align = true; - } - - if (curPunc == "(") pushContext(state, ")", stream.column()); - else if (curPunc == "[") pushContext(state, "]", stream.column()); - else if (curPunc == "{") pushContext(state, "}", stream.column()); - else if (/[\]\}\)]/.test(curPunc)) { - while (state.context && state.context.type == "pattern") popContext(state); - if (state.context && curPunc == state.context.type) popContext(state); - } - else if (curPunc == "." && state.context && state.context.type == "pattern") popContext(state); - else if (/atom|string|variable/.test(style) && state.context) { - if (/[\}\]]/.test(state.context.type)) - pushContext(state, "pattern", stream.column()); - else if (state.context.type == "pattern" && !state.context.align) { - state.context.align = true; - state.context.col = stream.column(); - } - } - - return style; - }, - - indent: function(state, textAfter) { - var firstChar = textAfter && textAfter.charAt(0); - var context = state.context; - if (/[\]\}]/.test(firstChar)) - while (context && context.type == "pattern") context = context.prev; - - var closing = context && firstChar == context.type; - if (!context) - return 0; - else if (context.type == "pattern") - return context.col; - else if (context.align) - return context.col + (closing ? 0 : 1); - else - return context.indent + (closing ? 0 : indentUnit); - } - }; -}); - -CodeMirror.defineMIME("application/x-sparql-query", "sparql"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/index.html deleted file mode 100644 index 2dafe69816..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/index.html +++ /dev/null @@ -1,98 +0,0 @@ - - - - - CodeMirror: sTeX mode - - - - - - - -

CodeMirror: sTeX mode

-
- - -

MIME types defined: text/x-stex.

- -

Parsing/Highlighting Tests: normal, verbose.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/stex.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/stex.js deleted file mode 100644 index ca04c24f22..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/stex.js +++ /dev/null @@ -1,246 +0,0 @@ -/* - * Author: Constantin Jucovschi (c.jucovschi@jacobs-university.de) - * Licence: MIT - */ - -CodeMirror.defineMode("stex", function() { - "use strict"; - - function pushCommand(state, command) { - state.cmdState.push(command); - } - - function peekCommand(state) { - if (state.cmdState.length > 0) { - return state.cmdState[state.cmdState.length - 1]; - } else { - return null; - } - } - - function popCommand(state) { - var plug = state.cmdState.pop(); - if (plug) { - plug.closeBracket(); - } - } - - // returns the non-default plugin closest to the end of the list - function getMostPowerful(state) { - var context = state.cmdState; - for (var i = context.length - 1; i >= 0; i--) { - var plug = context[i]; - if (plug.name == "DEFAULT") { - continue; - } - return plug; - } - return { styleIdentifier: function() { return null; } }; - } - - function addPluginPattern(pluginName, cmdStyle, styles) { - return function () { - this.name = pluginName; - this.bracketNo = 0; - this.style = cmdStyle; - this.styles = styles; - this.argument = null; // \begin and \end have arguments that follow. These are stored in the plugin - - this.styleIdentifier = function() { - return this.styles[this.bracketNo - 1] || null; - }; - this.openBracket = function() { - this.bracketNo++; - return "bracket"; - }; - this.closeBracket = function() {}; - }; - } - - var plugins = {}; - - plugins["importmodule"] = addPluginPattern("importmodule", "tag", ["string", "builtin"]); - plugins["documentclass"] = addPluginPattern("documentclass", "tag", ["", "atom"]); - plugins["usepackage"] = addPluginPattern("usepackage", "tag", ["atom"]); - plugins["begin"] = addPluginPattern("begin", "tag", ["atom"]); - plugins["end"] = addPluginPattern("end", "tag", ["atom"]); - - plugins["DEFAULT"] = function () { - this.name = "DEFAULT"; - this.style = "tag"; - - this.styleIdentifier = this.openBracket = this.closeBracket = function() {}; - }; - - function setState(state, f) { - state.f = f; - } - - // called when in a normal (no environment) context - function normal(source, state) { - var plug; - // Do we look like '\command' ? If so, attempt to apply the plugin 'command' - if (source.match(/^\\[a-zA-Z@]+/)) { - var cmdName = source.current().slice(1); - plug = plugins[cmdName] || plugins["DEFAULT"]; - plug = new plug(); - pushCommand(state, plug); - setState(state, beginParams); - return plug.style; - } - - // escape characters - if (source.match(/^\\[$&%#{}_]/)) { - return "tag"; - } - - // white space control characters - if (source.match(/^\\[,;!\/\\]/)) { - return "tag"; - } - - // find if we're starting various math modes - if (source.match("\\[")) { - setState(state, function(source, state){ return inMathMode(source, state, "\\]"); }); - return "keyword"; - } - if (source.match("$$")) { - setState(state, function(source, state){ return inMathMode(source, state, "$$"); }); - return "keyword"; - } - if (source.match("$")) { - setState(state, function(source, state){ return inMathMode(source, state, "$"); }); - return "keyword"; - } - - var ch = source.next(); - if (ch == "%") { - // special case: % at end of its own line; stay in same state - if (!source.eol()) { - setState(state, inCComment); - } - return "comment"; - } - else if (ch == '}' || ch == ']') { - plug = peekCommand(state); - if (plug) { - plug.closeBracket(ch); - setState(state, beginParams); - } else { - return "error"; - } - return "bracket"; - } else if (ch == '{' || ch == '[') { - plug = plugins["DEFAULT"]; - plug = new plug(); - pushCommand(state, plug); - return "bracket"; - } - else if (/\d/.test(ch)) { - source.eatWhile(/[\w.%]/); - return "atom"; - } - else { - source.eatWhile(/[\w\-_]/); - plug = getMostPowerful(state); - if (plug.name == 'begin') { - plug.argument = source.current(); - } - return plug.styleIdentifier(); - } - } - - function inCComment(source, state) { - source.skipToEnd(); - setState(state, normal); - return "comment"; - } - - function inMathMode(source, state, endModeSeq) { - if (source.eatSpace()) { - return null; - } - if (source.match(endModeSeq)) { - setState(state, normal); - return "keyword"; - } - if (source.match(/^\\[a-zA-Z@]+/)) { - return "tag"; - } - if (source.match(/^[a-zA-Z]+/)) { - return "variable-2"; - } - // escape characters - if (source.match(/^\\[$&%#{}_]/)) { - return "tag"; - } - // white space control characters - if (source.match(/^\\[,;!\/]/)) { - return "tag"; - } - // special math-mode characters - if (source.match(/^[\^_&]/)) { - return "tag"; - } - // non-special characters - if (source.match(/^[+\-<>|=,\/@!*:;'"`~#?]/)) { - return null; - } - if (source.match(/^(\d+\.\d*|\d*\.\d+|\d+)/)) { - return "number"; - } - var ch = source.next(); - if (ch == "{" || ch == "}" || ch == "[" || ch == "]" || ch == "(" || ch == ")") { - return "bracket"; - } - - // eat comments here, because inCComment returns us to normal state! - if (ch == "%") { - if (!source.eol()) { - source.skipToEnd(); - } - return "comment"; - } - return "error"; - } - - function beginParams(source, state) { - var ch = source.peek(), lastPlug; - if (ch == '{' || ch == '[') { - lastPlug = peekCommand(state); - lastPlug.openBracket(ch); - source.eat(ch); - setState(state, normal); - return "bracket"; - } - if (/[ \t\r]/.test(ch)) { - source.eat(ch); - return null; - } - setState(state, normal); - popCommand(state); - - return normal(source, state); - } - - return { - startState: function() { - return { - cmdState: [], - f: normal - }; - }, - copyState: function(s) { - return { - cmdState: s.cmdState.slice(), - f: s.f - }; - }, - token: function(stream, state) { - return state.f(stream, state); - } - }; -}); - -CodeMirror.defineMIME("text/x-stex", "stex"); -CodeMirror.defineMIME("text/x-latex", "stex"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/test.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/test.js deleted file mode 100644 index ab629e81ea..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/stex/test.js +++ /dev/null @@ -1,120 +0,0 @@ -(function() { - var mode = CodeMirror.getMode({tabSize: 4}, "stex"); - function MT(name) { test.mode(name, mode, Array.prototype.slice.call(arguments, 1)); } - - MT("word", - "foo"); - - MT("twoWords", - "foo bar"); - - MT("beginEndDocument", - "[tag \\begin][bracket {][atom document][bracket }]", - "[tag \\end][bracket {][atom document][bracket }]"); - - MT("beginEndEquation", - "[tag \\begin][bracket {][atom equation][bracket }]", - " E=mc^2", - "[tag \\end][bracket {][atom equation][bracket }]"); - - MT("beginModule", - "[tag \\begin][bracket {][atom module][bracket }[[]]]"); - - MT("beginModuleId", - "[tag \\begin][bracket {][atom module][bracket }[[]id=bbt-size[bracket ]]]"); - - MT("importModule", - "[tag \\importmodule][bracket [[][string b-b-t][bracket ]]{][builtin b-b-t][bracket }]"); - - MT("importModulePath", - "[tag \\importmodule][bracket [[][tag \\KWARCslides][bracket {][string dmath/en/cardinality][bracket }]]{][builtin card][bracket }]"); - - MT("psForPDF", - "[tag \\PSforPDF][bracket [[][atom 1][bracket ]]{]#1[bracket }]"); - - MT("comment", - "[comment % foo]"); - - MT("tagComment", - "[tag \\item][comment % bar]"); - - MT("commentTag", - " [comment % \\item]"); - - MT("commentLineBreak", - "[comment %]", - "foo"); - - MT("tagErrorCurly", - "[tag \\begin][error }][bracket {]"); - - MT("tagErrorSquare", - "[tag \\item][error ]]][bracket {]"); - - MT("commentCurly", - "[comment % }]"); - - MT("tagHash", - "the [tag \\#] key"); - - MT("tagNumber", - "a [tag \\$][atom 5] stetson"); - - MT("tagPercent", - "[atom 100][tag \\%] beef"); - - MT("tagAmpersand", - "L [tag \\&] N"); - - MT("tagUnderscore", - "foo[tag \\_]bar"); - - MT("tagBracketOpen", - "[tag \\emph][bracket {][tag \\{][bracket }]"); - - MT("tagBracketClose", - "[tag \\emph][bracket {][tag \\}][bracket }]"); - - MT("tagLetterNumber", - "section [tag \\S][atom 1]"); - - MT("textTagNumber", - "para [tag \\P][atom 2]"); - - MT("thinspace", - "x[tag \\,]y"); - - MT("thickspace", - "x[tag \\;]y"); - - MT("negativeThinspace", - "x[tag \\!]y"); - - MT("periodNotSentence", - "J.\\ L.\\ is"); - - MT("periodSentence", - "X[tag \\@]. The"); - - MT("italicCorrection", - "[bracket {][tag \\em] If[tag \\/][bracket }] I"); - - MT("tagBracket", - "[tag \\newcommand][bracket {][tag \\pop][bracket }]"); - - MT("inlineMathTagFollowedByNumber", - "[keyword $][tag \\pi][number 2][keyword $]"); - - MT("inlineMath", - "[keyword $][number 3][variable-2 x][tag ^][number 2.45]-[tag \\sqrt][bracket {][tag \\$\\alpha][bracket }] = [number 2][keyword $] other text"); - - MT("displayMath", - "More [keyword $$]\t[variable-2 S][tag ^][variable-2 n][tag \\sum] [variable-2 i][keyword $$] other text"); - - MT("mathWithComment", - "[keyword $][variable-2 x] [comment % $]", - "[variable-2 y][keyword $] other text"); - - MT("lineBreakArgument", - "[tag \\\\][bracket [[][atom 1cm][bracket ]]]"); -})(); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tcl/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/tcl/index.html deleted file mode 100644 index e2e42a04be..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tcl/index.html +++ /dev/null @@ -1,129 +0,0 @@ - - - - - CodeMirror: Tcl mode - - - - - - - -

CodeMirror: Tcl mode

-
- - -

MIME types defined: text/x-tcl.

- - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tcl/tcl.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/tcl/tcl.js deleted file mode 100644 index ed2c697217..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tcl/tcl.js +++ /dev/null @@ -1,131 +0,0 @@ -//tcl mode by Ford_Lawnmower :: Based on Velocity mode by Steve O'Hara -CodeMirror.defineMode("tcl", function() { - function parseWords(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - var keywords = parseWords("Tcl safe after append array auto_execok auto_import auto_load " + - "auto_mkindex auto_mkindex_old auto_qualify auto_reset bgerror " + - "binary break catch cd close concat continue dde eof encoding error " + - "eval exec exit expr fblocked fconfigure fcopy file fileevent filename " + - "filename flush for foreach format gets glob global history http if " + - "incr info interp join lappend lindex linsert list llength load lrange " + - "lreplace lsearch lset lsort memory msgcat namespace open package parray " + - "pid pkg::create pkg_mkIndex proc puts pwd re_syntax read regex regexp " + - "registry regsub rename resource return scan seek set socket source split " + - "string subst switch tcl_endOfWord tcl_findLibrary tcl_startOfNextWord " + - "tcl_wordBreakAfter tcl_startOfPreviousWord tcl_wordBreakBefore tcltest " + - "tclvars tell time trace unknown unset update uplevel upvar variable " + - "vwait"); - var functions = parseWords("if elseif else and not or eq ne in ni for foreach while switch"); - var isOperatorChar = /[+\-*&%=<>!?^\/\|]/; - function chain(stream, state, f) { - state.tokenize = f; - return f(stream, state); - } - function tokenBase(stream, state) { - var beforeParams = state.beforeParams; - state.beforeParams = false; - var ch = stream.next(); - if ((ch == '"' || ch == "'") && state.inParams) - return chain(stream, state, tokenString(ch)); - else if (/[\[\]{}\(\),;\.]/.test(ch)) { - if (ch == "(" && beforeParams) state.inParams = true; - else if (ch == ")") state.inParams = false; - return null; - } - else if (/\d/.test(ch)) { - stream.eatWhile(/[\w\.]/); - return "number"; - } - else if (ch == "#" && stream.eat("*")) { - return chain(stream, state, tokenComment); - } - else if (ch == "#" && stream.match(/ *\[ *\[/)) { - return chain(stream, state, tokenUnparsed); - } - else if (ch == "#" && stream.eat("#")) { - stream.skipToEnd(); - return "comment"; - } - else if (ch == '"') { - stream.skipTo(/"/); - return "comment"; - } - else if (ch == "$") { - stream.eatWhile(/[$_a-z0-9A-Z\.{:]/); - stream.eatWhile(/}/); - state.beforeParams = true; - return "builtin"; - } - else if (isOperatorChar.test(ch)) { - stream.eatWhile(isOperatorChar); - return "comment"; - } - else { - stream.eatWhile(/[\w\$_{}]/); - var word = stream.current().toLowerCase(); - if (keywords && keywords.propertyIsEnumerable(word)) - return "keyword"; - if (functions && functions.propertyIsEnumerable(word)) { - state.beforeParams = true; - return "keyword"; - } - return null; - } - } - function tokenString(quote) { - return function(stream, state) { - var escaped = false, next, end = false; - while ((next = stream.next()) != null) { - if (next == quote && !escaped) { - end = true; - break; - } - escaped = !escaped && next == "\\"; - } - if (end) state.tokenize = tokenBase; - return "string"; - }; - } - function tokenComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == "#" && maybeEnd) { - state.tokenize = tokenBase; - break; - } - maybeEnd = (ch == "*"); - } - return "comment"; - } - function tokenUnparsed(stream, state) { - var maybeEnd = 0, ch; - while (ch = stream.next()) { - if (ch == "#" && maybeEnd == 2) { - state.tokenize = tokenBase; - break; - } - if (ch == "]") - maybeEnd++; - else if (ch != " ") - maybeEnd = 0; - } - return "meta"; - } - return { - startState: function() { - return { - tokenize: tokenBase, - beforeParams: false, - inParams: false - }; - }, - token: function(stream, state) { - if (stream.eatSpace()) return null; - return state.tokenize(stream, state); - } - }; -}); -CodeMirror.defineMIME("text/x-tcl", "tcl"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/index.html deleted file mode 100644 index 848f33a592..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/index.html +++ /dev/null @@ -1,142 +0,0 @@ - - - - - CodeMirror: TiddlyWiki mode - - - - - - - - - -

CodeMirror: TiddlyWiki mode

- -
- - - -

TiddlyWiki mode supports a single configuration.

- -

MIME types defined: text/x-tiddlywiki.

- - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/tiddlywiki.css b/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/tiddlywiki.css deleted file mode 100644 index 9a69b639f8..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/tiddlywiki.css +++ /dev/null @@ -1,14 +0,0 @@ -span.cm-underlined { - text-decoration: underline; -} -span.cm-strikethrough { - text-decoration: line-through; -} -span.cm-brace { - color: #170; - font-weight: bold; -} -span.cm-table { - color: blue; - font-weight: bold; -} diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/tiddlywiki.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/tiddlywiki.js deleted file mode 100644 index 24a24786bc..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiddlywiki/tiddlywiki.js +++ /dev/null @@ -1,353 +0,0 @@ -/*** - |''Name''|tiddlywiki.js| - |''Description''|Enables TiddlyWikiy syntax highlighting using CodeMirror| - |''Author''|PMario| - |''Version''|0.1.7| - |''Status''|''stable''| - |''Source''|[[GitHub|https://github.com/pmario/CodeMirror2/blob/tw-syntax/mode/tiddlywiki]]| - |''Documentation''|http://codemirror.tiddlyspace.com/| - |''License''|[[MIT License|http://www.opensource.org/licenses/mit-license.php]]| - |''CoreVersion''|2.5.0| - |''Requires''|codemirror.js| - |''Keywords''|syntax highlighting color code mirror codemirror| - ! Info - CoreVersion parameter is needed for TiddlyWiki only! -***/ -//{{{ -CodeMirror.defineMode("tiddlywiki", function () { - // Tokenizer - var textwords = {}; - - var keywords = function () { - function kw(type) { - return { type: type, style: "macro"}; - } - return { - "allTags": kw('allTags'), "closeAll": kw('closeAll'), "list": kw('list'), - "newJournal": kw('newJournal'), "newTiddler": kw('newTiddler'), - "permaview": kw('permaview'), "saveChanges": kw('saveChanges'), - "search": kw('search'), "slider": kw('slider'), "tabs": kw('tabs'), - "tag": kw('tag'), "tagging": kw('tagging'), "tags": kw('tags'), - "tiddler": kw('tiddler'), "timeline": kw('timeline'), - "today": kw('today'), "version": kw('version'), "option": kw('option'), - - "with": kw('with'), - "filter": kw('filter') - }; - }(); - - var isSpaceName = /[\w_\-]/i, - reHR = /^\-\-\-\-+$/, //
- reWikiCommentStart = /^\/\*\*\*$/, // /*** - reWikiCommentStop = /^\*\*\*\/$/, // ***/ - reBlockQuote = /^<<<$/, - - reJsCodeStart = /^\/\/\{\{\{$/, // //{{{ js block start - reJsCodeStop = /^\/\/\}\}\}$/, // //}}} js stop - reXmlCodeStart = /^$/, // xml block start - reXmlCodeStop = /^$/, // xml stop - - reCodeBlockStart = /^\{\{\{$/, // {{{ TW text div block start - reCodeBlockStop = /^\}\}\}$/, // }}} TW text stop - - reUntilCodeStop = /.*?\}\}\}/; - - function chain(stream, state, f) { - state.tokenize = f; - return f(stream, state); - } - - // Used as scratch variables to communicate multiple values without - // consing up tons of objects. - var type, content; - - function ret(tp, style, cont) { - type = tp; - content = cont; - return style; - } - - function jsTokenBase(stream, state) { - var sol = stream.sol(), ch; - - state.block = false; // indicates the start of a code block. - - ch = stream.peek(); // don't eat, to make matching simpler - - // check start of blocks - if (sol && /[<\/\*{}\-]/.test(ch)) { - if (stream.match(reCodeBlockStart)) { - state.block = true; - return chain(stream, state, twTokenCode); - } - if (stream.match(reBlockQuote)) { - return ret('quote', 'quote'); - } - if (stream.match(reWikiCommentStart) || stream.match(reWikiCommentStop)) { - return ret('code', 'comment'); - } - if (stream.match(reJsCodeStart) || stream.match(reJsCodeStop) || stream.match(reXmlCodeStart) || stream.match(reXmlCodeStop)) { - return ret('code', 'comment'); - } - if (stream.match(reHR)) { - return ret('hr', 'hr'); - } - } // sol - ch = stream.next(); - - if (sol && /[\/\*!#;:>|]/.test(ch)) { - if (ch == "!") { // tw header - stream.skipToEnd(); - return ret("header", "header"); - } - if (ch == "*") { // tw list - stream.eatWhile('*'); - return ret("list", "comment"); - } - if (ch == "#") { // tw numbered list - stream.eatWhile('#'); - return ret("list", "comment"); - } - if (ch == ";") { // definition list, term - stream.eatWhile(';'); - return ret("list", "comment"); - } - if (ch == ":") { // definition list, description - stream.eatWhile(':'); - return ret("list", "comment"); - } - if (ch == ">") { // single line quote - stream.eatWhile(">"); - return ret("quote", "quote"); - } - if (ch == '|') { - return ret('table', 'header'); - } - } - - if (ch == '{' && stream.match(/\{\{/)) { - return chain(stream, state, twTokenCode); - } - - // rudimentary html:// file:// link matching. TW knows much more ... - if (/[hf]/i.test(ch)) { - if (/[ti]/i.test(stream.peek()) && stream.match(/\b(ttps?|tp|ile):\/\/[\-A-Z0-9+&@#\/%?=~_|$!:,.;]*[A-Z0-9+&@#\/%=~_|$]/i)) { - return ret("link", "link"); - } - } - // just a little string indicator, don't want to have the whole string covered - if (ch == '"') { - return ret('string', 'string'); - } - if (ch == '~') { // _no_ CamelCase indicator should be bold - return ret('text', 'brace'); - } - if (/[\[\]]/.test(ch)) { // check for [[..]] - if (stream.peek() == ch) { - stream.next(); - return ret('brace', 'brace'); - } - } - if (ch == "@") { // check for space link. TODO fix @@...@@ highlighting - stream.eatWhile(isSpaceName); - return ret("link", "link"); - } - if (/\d/.test(ch)) { // numbers - stream.eatWhile(/\d/); - return ret("number", "number"); - } - if (ch == "/") { // tw invisible comment - if (stream.eat("%")) { - return chain(stream, state, twTokenComment); - } - else if (stream.eat("/")) { // - return chain(stream, state, twTokenEm); - } - } - if (ch == "_") { // tw underline - if (stream.eat("_")) { - return chain(stream, state, twTokenUnderline); - } - } - // strikethrough and mdash handling - if (ch == "-") { - if (stream.eat("-")) { - // if strikethrough looks ugly, change CSS. - if (stream.peek() != ' ') - return chain(stream, state, twTokenStrike); - // mdash - if (stream.peek() == ' ') - return ret('text', 'brace'); - } - } - if (ch == "'") { // tw bold - if (stream.eat("'")) { - return chain(stream, state, twTokenStrong); - } - } - if (ch == "<") { // tw macro - if (stream.eat("<")) { - return chain(stream, state, twTokenMacro); - } - } - else { - return ret(ch); - } - - // core macro handling - stream.eatWhile(/[\w\$_]/); - var word = stream.current(), - known = textwords.propertyIsEnumerable(word) && textwords[word]; - - return known ? ret(known.type, known.style, word) : ret("text", null, word); - - } // jsTokenBase() - - // tw invisible comment - function twTokenComment(stream, state) { - var maybeEnd = false, - ch; - while (ch = stream.next()) { - if (ch == "/" && maybeEnd) { - state.tokenize = jsTokenBase; - break; - } - maybeEnd = (ch == "%"); - } - return ret("comment", "comment"); - } - - // tw strong / bold - function twTokenStrong(stream, state) { - var maybeEnd = false, - ch; - while (ch = stream.next()) { - if (ch == "'" && maybeEnd) { - state.tokenize = jsTokenBase; - break; - } - maybeEnd = (ch == "'"); - } - return ret("text", "strong"); - } - - // tw code - function twTokenCode(stream, state) { - var ch, sb = state.block; - - if (sb && stream.current()) { - return ret("code", "comment"); - } - - if (!sb && stream.match(reUntilCodeStop)) { - state.tokenize = jsTokenBase; - return ret("code", "comment"); - } - - if (sb && stream.sol() && stream.match(reCodeBlockStop)) { - state.tokenize = jsTokenBase; - return ret("code", "comment"); - } - - ch = stream.next(); - return (sb) ? ret("code", "comment") : ret("code", "comment"); - } - - // tw em / italic - function twTokenEm(stream, state) { - var maybeEnd = false, - ch; - while (ch = stream.next()) { - if (ch == "/" && maybeEnd) { - state.tokenize = jsTokenBase; - break; - } - maybeEnd = (ch == "/"); - } - return ret("text", "em"); - } - - // tw underlined text - function twTokenUnderline(stream, state) { - var maybeEnd = false, - ch; - while (ch = stream.next()) { - if (ch == "_" && maybeEnd) { - state.tokenize = jsTokenBase; - break; - } - maybeEnd = (ch == "_"); - } - return ret("text", "underlined"); - } - - // tw strike through text looks ugly - // change CSS if needed - function twTokenStrike(stream, state) { - var maybeEnd = false, ch; - - while (ch = stream.next()) { - if (ch == "-" && maybeEnd) { - state.tokenize = jsTokenBase; - break; - } - maybeEnd = (ch == "-"); - } - return ret("text", "strikethrough"); - } - - // macro - function twTokenMacro(stream, state) { - var ch, word, known; - - if (stream.current() == '<<') { - return ret('brace', 'macro'); - } - - ch = stream.next(); - if (!ch) { - state.tokenize = jsTokenBase; - return ret(ch); - } - if (ch == ">") { - if (stream.peek() == '>') { - stream.next(); - state.tokenize = jsTokenBase; - return ret("brace", "macro"); - } - } - - stream.eatWhile(/[\w\$_]/); - word = stream.current(); - known = keywords.propertyIsEnumerable(word) && keywords[word]; - - if (known) { - return ret(known.type, known.style, word); - } - else { - return ret("macro", null, word); - } - } - - // Interface - return { - startState: function () { - return { - tokenize: jsTokenBase, - indented: 0, - level: 0 - }; - }, - - token: function (stream, state) { - if (stream.eatSpace()) return null; - var style = state.tokenize(stream, state); - return style; - }, - - electricChars: "" - }; -}); - -CodeMirror.defineMIME("text/x-tiddlywiki", "tiddlywiki"); -//}}} diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/index.html deleted file mode 100644 index 7b85a44a7e..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/index.html +++ /dev/null @@ -1,81 +0,0 @@ - - - - CodeMirror: Tiki wiki mode - - - - - - - - -

CodeMirror: Tiki wiki mode

- -
- - - - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/tiki.css b/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/tiki.css deleted file mode 100644 index 0dbc3ea0e0..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/tiki.css +++ /dev/null @@ -1,26 +0,0 @@ -.cm-tw-syntaxerror { - color: #FFF; - background-color: #900; -} - -.cm-tw-deleted { - text-decoration: line-through; -} - -.cm-tw-header5 { - font-weight: bold; -} -.cm-tw-listitem:first-child { /*Added first child to fix duplicate padding when highlighting*/ - padding-left: 10px; -} - -.cm-tw-box { - border-top-width: 0px ! important; - border-style: solid; - border-width: 1px; - border-color: inherit; -} - -.cm-tw-underline { - text-decoration: underline; -} \ No newline at end of file diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/tiki.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/tiki.js deleted file mode 100644 index e789163dc7..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/tiki/tiki.js +++ /dev/null @@ -1,308 +0,0 @@ -CodeMirror.defineMode('tiki', function(config) { - function inBlock(style, terminator, returnTokenizer) { - return function(stream, state) { - while (!stream.eol()) { - if (stream.match(terminator)) { - state.tokenize = inText; - break; - } - stream.next(); - } - - if (returnTokenizer) state.tokenize = returnTokenizer; - - return style; - }; - } - - function inLine(style) { - return function(stream, state) { - while(!stream.eol()) { - stream.next(); - } - state.tokenize = inText; - return style; - }; - } - - function inText(stream, state) { - function chain(parser) { - state.tokenize = parser; - return parser(stream, state); - } - - var sol = stream.sol(); - var ch = stream.next(); - - //non start of line - switch (ch) { //switch is generally much faster than if, so it is used here - case "{": //plugin - stream.eat("/"); - stream.eatSpace(); - var tagName = ""; - var c; - while ((c = stream.eat(/[^\s\u00a0=\"\'\/?(}]/))) tagName += c; - state.tokenize = inPlugin; - return "tag"; - break; - case "_": //bold - if (stream.eat("_")) { - return chain(inBlock("strong", "__", inText)); - } - break; - case "'": //italics - if (stream.eat("'")) { - // Italic text - return chain(inBlock("em", "''", inText)); - } - break; - case "(":// Wiki Link - if (stream.eat("(")) { - return chain(inBlock("variable-2", "))", inText)); - } - break; - case "[":// Weblink - return chain(inBlock("variable-3", "]", inText)); - break; - case "|": //table - if (stream.eat("|")) { - return chain(inBlock("comment", "||")); - } - break; - case "-": - if (stream.eat("=")) {//titleBar - return chain(inBlock("header string", "=-", inText)); - } else if (stream.eat("-")) {//deleted - return chain(inBlock("error tw-deleted", "--", inText)); - } - break; - case "=": //underline - if (stream.match("==")) { - return chain(inBlock("tw-underline", "===", inText)); - } - break; - case ":": - if (stream.eat(":")) { - return chain(inBlock("comment", "::")); - } - break; - case "^": //box - return chain(inBlock("tw-box", "^")); - break; - case "~": //np - if (stream.match("np~")) { - return chain(inBlock("meta", "~/np~")); - } - break; - } - - //start of line types - if (sol) { - switch (ch) { - case "!": //header at start of line - if (stream.match('!!!!!')) { - return chain(inLine("header string")); - } else if (stream.match('!!!!')) { - return chain(inLine("header string")); - } else if (stream.match('!!!')) { - return chain(inLine("header string")); - } else if (stream.match('!!')) { - return chain(inLine("header string")); - } else { - return chain(inLine("header string")); - } - break; - case "*": //unordered list line item, or
  • at start of line - case "#": //ordered list line item, or
  • at start of line - case "+": //ordered list line item, or
  • at start of line - return chain(inLine("tw-listitem bracket")); - break; - } - } - - //stream.eatWhile(/[&{]/); was eating up plugins, turned off to act less like html and more like tiki - return null; - } - - var indentUnit = config.indentUnit; - - // Return variables for tokenizers - var pluginName, type; - function inPlugin(stream, state) { - var ch = stream.next(); - var peek = stream.peek(); - - if (ch == "}") { - state.tokenize = inText; - //type = ch == ")" ? "endPlugin" : "selfclosePlugin"; inPlugin - return "tag"; - } else if (ch == "(" || ch == ")") { - return "bracket"; - } else if (ch == "=") { - type = "equals"; - - if (peek == ">") { - ch = stream.next(); - peek = stream.peek(); - } - - //here we detect values directly after equal character with no quotes - if (!/[\'\"]/.test(peek)) { - state.tokenize = inAttributeNoQuote(); - } - //end detect values - - return "operator"; - } else if (/[\'\"]/.test(ch)) { - state.tokenize = inAttribute(ch); - return state.tokenize(stream, state); - } else { - stream.eatWhile(/[^\s\u00a0=\"\'\/?]/); - return "keyword"; - } - } - - function inAttribute(quote) { - return function(stream, state) { - while (!stream.eol()) { - if (stream.next() == quote) { - state.tokenize = inPlugin; - break; - } - } - return "string"; - }; - } - - function inAttributeNoQuote() { - return function(stream, state) { - while (!stream.eol()) { - var ch = stream.next(); - var peek = stream.peek(); - if (ch == " " || ch == "," || /[ )}]/.test(peek)) { - state.tokenize = inPlugin; - break; - } - } - return "string"; -}; - } - -var curState, setStyle; -function pass() { - for (var i = arguments.length - 1; i >= 0; i--) curState.cc.push(arguments[i]); -} - -function cont() { - pass.apply(null, arguments); - return true; -} - -function pushContext(pluginName, startOfLine) { - var noIndent = curState.context && curState.context.noIndent; - curState.context = { - prev: curState.context, - pluginName: pluginName, - indent: curState.indented, - startOfLine: startOfLine, - noIndent: noIndent - }; -} - -function popContext() { - if (curState.context) curState.context = curState.context.prev; -} - -function element(type) { - if (type == "openPlugin") {curState.pluginName = pluginName; return cont(attributes, endplugin(curState.startOfLine));} - else if (type == "closePlugin") { - var err = false; - if (curState.context) { - err = curState.context.pluginName != pluginName; - popContext(); - } else { - err = true; - } - if (err) setStyle = "error"; - return cont(endcloseplugin(err)); - } - else if (type == "string") { - if (!curState.context || curState.context.name != "!cdata") pushContext("!cdata"); - if (curState.tokenize == inText) popContext(); - return cont(); - } - else return cont(); -} - -function endplugin(startOfLine) { - return function(type) { - if ( - type == "selfclosePlugin" || - type == "endPlugin" - ) - return cont(); - if (type == "endPlugin") {pushContext(curState.pluginName, startOfLine); return cont();} - return cont(); - }; -} - -function endcloseplugin(err) { - return function(type) { - if (err) setStyle = "error"; - if (type == "endPlugin") return cont(); - return pass(); - }; -} - -function attributes(type) { - if (type == "keyword") {setStyle = "attribute"; return cont(attributes);} - if (type == "equals") return cont(attvalue, attributes); - return pass(); -} -function attvalue(type) { - if (type == "keyword") {setStyle = "string"; return cont();} - if (type == "string") return cont(attvaluemaybe); - return pass(); -} -function attvaluemaybe(type) { - if (type == "string") return cont(attvaluemaybe); - else return pass(); -} -return { - startState: function() { - return {tokenize: inText, cc: [], indented: 0, startOfLine: true, pluginName: null, context: null}; - }, - token: function(stream, state) { - if (stream.sol()) { - state.startOfLine = true; - state.indented = stream.indentation(); - } - if (stream.eatSpace()) return null; - - setStyle = type = pluginName = null; - var style = state.tokenize(stream, state); - if ((style || type) && style != "comment") { - curState = state; - while (true) { - var comb = state.cc.pop() || element; - if (comb(type || style)) break; - } - } - state.startOfLine = false; - return setStyle || style; - }, - indent: function(state, textAfter) { - var context = state.context; - if (context && context.noIndent) return 0; - if (context && /^{\//.test(textAfter)) - context = context.prev; - while (context && !context.startOfLine) - context = context.prev; - if (context) return context.indent + indentUnit; - else return 0; - }, - electricChars: "/" - }; -}); - -CodeMirror.defineMIME("text/tiki", "tiki"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/turtle/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/turtle/index.html deleted file mode 100644 index 5e56e5755c..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/turtle/index.html +++ /dev/null @@ -1,39 +0,0 @@ - - - - - CodeMirror: Turtle mode - - - - - - - -

    CodeMirror: Turtle mode

    -
    - - -

    MIME types defined: text/turtle.

    - - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/turtle/turtle.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/turtle/turtle.js deleted file mode 100644 index e118bfbced..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/turtle/turtle.js +++ /dev/null @@ -1,145 +0,0 @@ -CodeMirror.defineMode("turtle", function(config) { - var indentUnit = config.indentUnit; - var curPunc; - - function wordRegexp(words) { - return new RegExp("^(?:" + words.join("|") + ")$", "i"); - } - var ops = wordRegexp([]); - var keywords = wordRegexp(["@prefix", "@base", "a"]); - var operatorChars = /[*+\-<>=&|]/; - - function tokenBase(stream, state) { - var ch = stream.next(); - curPunc = null; - if (ch == "<" && !stream.match(/^[\s\u00a0=]/, false)) { - stream.match(/^[^\s\u00a0>]*>?/); - return "atom"; - } - else if (ch == "\"" || ch == "'") { - state.tokenize = tokenLiteral(ch); - return state.tokenize(stream, state); - } - else if (/[{}\(\),\.;\[\]]/.test(ch)) { - curPunc = ch; - return null; - } - else if (ch == "#") { - stream.skipToEnd(); - return "comment"; - } - else if (operatorChars.test(ch)) { - stream.eatWhile(operatorChars); - return null; - } - else if (ch == ":") { - return "operator"; - } else { - stream.eatWhile(/[_\w\d]/); - if(stream.peek() == ":") { - return "variable-3"; - } else { - var word = stream.current(); - - if(keywords.test(word)) { - return "meta"; - } - - if(ch >= "A" && ch <= "Z") { - return "comment"; - } else { - return "keyword"; - } - } - var word = stream.current(); - if (ops.test(word)) - return null; - else if (keywords.test(word)) - return "meta"; - else - return "variable"; - } - } - - function tokenLiteral(quote) { - return function(stream, state) { - var escaped = false, ch; - while ((ch = stream.next()) != null) { - if (ch == quote && !escaped) { - state.tokenize = tokenBase; - break; - } - escaped = !escaped && ch == "\\"; - } - return "string"; - }; - } - - function pushContext(state, type, col) { - state.context = {prev: state.context, indent: state.indent, col: col, type: type}; - } - function popContext(state) { - state.indent = state.context.indent; - state.context = state.context.prev; - } - - return { - startState: function() { - return {tokenize: tokenBase, - context: null, - indent: 0, - col: 0}; - }, - - token: function(stream, state) { - if (stream.sol()) { - if (state.context && state.context.align == null) state.context.align = false; - state.indent = stream.indentation(); - } - if (stream.eatSpace()) return null; - var style = state.tokenize(stream, state); - - if (style != "comment" && state.context && state.context.align == null && state.context.type != "pattern") { - state.context.align = true; - } - - if (curPunc == "(") pushContext(state, ")", stream.column()); - else if (curPunc == "[") pushContext(state, "]", stream.column()); - else if (curPunc == "{") pushContext(state, "}", stream.column()); - else if (/[\]\}\)]/.test(curPunc)) { - while (state.context && state.context.type == "pattern") popContext(state); - if (state.context && curPunc == state.context.type) popContext(state); - } - else if (curPunc == "." && state.context && state.context.type == "pattern") popContext(state); - else if (/atom|string|variable/.test(style) && state.context) { - if (/[\}\]]/.test(state.context.type)) - pushContext(state, "pattern", stream.column()); - else if (state.context.type == "pattern" && !state.context.align) { - state.context.align = true; - state.context.col = stream.column(); - } - } - - return style; - }, - - indent: function(state, textAfter) { - var firstChar = textAfter && textAfter.charAt(0); - var context = state.context; - if (/[\]\}]/.test(firstChar)) - while (context && context.type == "pattern") context = context.prev; - - var closing = context && firstChar == context.type; - if (!context) - return 0; - else if (context.type == "pattern") - return context.col; - else if (context.align) - return context.col + (closing ? 0 : 1); - else - return context.indent + (closing ? 0 : indentUnit); - } - }; -}); - -CodeMirror.defineMIME("text/turtle", "turtle"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/LICENSE.txt b/wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/LICENSE.txt deleted file mode 100644 index 60839703a9..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/LICENSE.txt +++ /dev/null @@ -1,21 +0,0 @@ -The MIT License - -Copyright (c) 2012 Codility Limited, 107 Cheapside, London EC2V 6DN, UK - -Permission is hereby granted, free of charge, to any person obtaining a copy -of this software and associated documentation files (the "Software"), to deal -in the Software without restriction, including without limitation the rights -to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -copies of the Software, and to permit persons to whom the Software is -furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in -all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -THE SOFTWARE. diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/index.html deleted file mode 100644 index 74dd5e8161..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/index.html +++ /dev/null @@ -1,88 +0,0 @@ - - - - CodeMirror: VB.NET mode - - - - - - - - - -

    CodeMirror: VB.NET mode

    - - - -
    - -
    -
    
    -  

    MIME type defined: text/x-vb.

    - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/vb.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/vb.js deleted file mode 100644 index 27b2271951..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vb/vb.js +++ /dev/null @@ -1,259 +0,0 @@ -CodeMirror.defineMode("vb", function(conf, parserConf) { - var ERRORCLASS = 'error'; - - function wordRegexp(words) { - return new RegExp("^((" + words.join(")|(") + "))\\b", "i"); - } - - var singleOperators = new RegExp("^[\\+\\-\\*/%&\\\\|\\^~<>!]"); - var singleDelimiters = new RegExp('^[\\(\\)\\[\\]\\{\\}@,:`=;\\.]'); - var doubleOperators = new RegExp("^((==)|(<>)|(<=)|(>=)|(<>)|(<<)|(>>)|(//)|(\\*\\*))"); - var doubleDelimiters = new RegExp("^((\\+=)|(\\-=)|(\\*=)|(%=)|(/=)|(&=)|(\\|=)|(\\^=))"); - var tripleDelimiters = new RegExp("^((//=)|(>>=)|(<<=)|(\\*\\*=))"); - var identifiers = new RegExp("^[_A-Za-z][_A-Za-z0-9]*"); - - var openingKeywords = ['class','module', 'sub','enum','select','while','if','function', 'get','set','property', 'try']; - var middleKeywords = ['else','elseif','case', 'catch']; - var endKeywords = ['next','loop']; - - var wordOperators = wordRegexp(['and', 'or', 'not', 'xor', 'in']); - var commonkeywords = ['as', 'dim', 'break', 'continue','optional', 'then', 'until', - 'goto', 'byval','byref','new','handles','property', 'return', - 'const','private', 'protected', 'friend', 'public', 'shared', 'static', 'true','false']; - var commontypes = ['integer','string','double','decimal','boolean','short','char', 'float','single']; - - var keywords = wordRegexp(commonkeywords); - var types = wordRegexp(commontypes); - var stringPrefixes = '"'; - - var opening = wordRegexp(openingKeywords); - var middle = wordRegexp(middleKeywords); - var closing = wordRegexp(endKeywords); - var doubleClosing = wordRegexp(['end']); - var doOpening = wordRegexp(['do']); - - var indentInfo = null; - - - - - function indent(_stream, state) { - state.currentIndent++; - } - - function dedent(_stream, state) { - state.currentIndent--; - } - // tokenizers - function tokenBase(stream, state) { - if (stream.eatSpace()) { - return null; - } - - var ch = stream.peek(); - - // Handle Comments - if (ch === "'") { - stream.skipToEnd(); - return 'comment'; - } - - - // Handle Number Literals - if (stream.match(/^((&H)|(&O))?[0-9\.a-f]/i, false)) { - var floatLiteral = false; - // Floats - if (stream.match(/^\d*\.\d+F?/i)) { floatLiteral = true; } - else if (stream.match(/^\d+\.\d*F?/)) { floatLiteral = true; } - else if (stream.match(/^\.\d+F?/)) { floatLiteral = true; } - - if (floatLiteral) { - // Float literals may be "imaginary" - stream.eat(/J/i); - return 'number'; - } - // Integers - var intLiteral = false; - // Hex - if (stream.match(/^&H[0-9a-f]+/i)) { intLiteral = true; } - // Octal - else if (stream.match(/^&O[0-7]+/i)) { intLiteral = true; } - // Decimal - else if (stream.match(/^[1-9]\d*F?/)) { - // Decimal literals may be "imaginary" - stream.eat(/J/i); - // TODO - Can you have imaginary longs? - intLiteral = true; - } - // Zero by itself with no other piece of number. - else if (stream.match(/^0(?![\dx])/i)) { intLiteral = true; } - if (intLiteral) { - // Integer literals may be "long" - stream.eat(/L/i); - return 'number'; - } - } - - // Handle Strings - if (stream.match(stringPrefixes)) { - state.tokenize = tokenStringFactory(stream.current()); - return state.tokenize(stream, state); - } - - // Handle operators and Delimiters - if (stream.match(tripleDelimiters) || stream.match(doubleDelimiters)) { - return null; - } - if (stream.match(doubleOperators) - || stream.match(singleOperators) - || stream.match(wordOperators)) { - return 'operator'; - } - if (stream.match(singleDelimiters)) { - return null; - } - if (stream.match(doOpening)) { - indent(stream,state); - state.doInCurrentLine = true; - return 'keyword'; - } - if (stream.match(opening)) { - if (! state.doInCurrentLine) - indent(stream,state); - else - state.doInCurrentLine = false; - return 'keyword'; - } - if (stream.match(middle)) { - return 'keyword'; - } - - if (stream.match(doubleClosing)) { - dedent(stream,state); - dedent(stream,state); - return 'keyword'; - } - if (stream.match(closing)) { - dedent(stream,state); - return 'keyword'; - } - - if (stream.match(types)) { - return 'keyword'; - } - - if (stream.match(keywords)) { - return 'keyword'; - } - - if (stream.match(identifiers)) { - return 'variable'; - } - - // Handle non-detected items - stream.next(); - return ERRORCLASS; - } - - function tokenStringFactory(delimiter) { - var singleline = delimiter.length == 1; - var OUTCLASS = 'string'; - - return function(stream, state) { - while (!stream.eol()) { - stream.eatWhile(/[^'"]/); - if (stream.match(delimiter)) { - state.tokenize = tokenBase; - return OUTCLASS; - } else { - stream.eat(/['"]/); - } - } - if (singleline) { - if (parserConf.singleLineStringErrors) { - return ERRORCLASS; - } else { - state.tokenize = tokenBase; - } - } - return OUTCLASS; - }; - } - - - function tokenLexer(stream, state) { - var style = state.tokenize(stream, state); - var current = stream.current(); - - // Handle '.' connected identifiers - if (current === '.') { - style = state.tokenize(stream, state); - current = stream.current(); - if (style === 'variable') { - return 'variable'; - } else { - return ERRORCLASS; - } - } - - - var delimiter_index = '[({'.indexOf(current); - if (delimiter_index !== -1) { - indent(stream, state ); - } - if (indentInfo === 'dedent') { - if (dedent(stream, state)) { - return ERRORCLASS; - } - } - delimiter_index = '])}'.indexOf(current); - if (delimiter_index !== -1) { - if (dedent(stream, state)) { - return ERRORCLASS; - } - } - - return style; - } - - var external = { - electricChars:"dDpPtTfFeE ", - startState: function() { - return { - tokenize: tokenBase, - lastToken: null, - currentIndent: 0, - nextLineIndent: 0, - doInCurrentLine: false - - - }; - }, - - token: function(stream, state) { - if (stream.sol()) { - state.currentIndent += state.nextLineIndent; - state.nextLineIndent = 0; - state.doInCurrentLine = 0; - } - var style = tokenLexer(stream, state); - - state.lastToken = {style:style, content: stream.current()}; - - - - return style; - }, - - indent: function(state, textAfter) { - var trueText = textAfter.replace(/^\s+|\s+$/g, '') ; - if (trueText.match(closing) || trueText.match(doubleClosing) || trueText.match(middle)) return conf.indentUnit*(state.currentIndent-1); - if(state.currentIndent < 0) return 0; - return state.currentIndent * conf.indentUnit; - } - - }; - return external; -}); - -CodeMirror.defineMIME("text/x-vb", "vb"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vbscript/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/vbscript/index.html deleted file mode 100644 index 8c86f9ef94..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vbscript/index.html +++ /dev/null @@ -1,42 +0,0 @@ - - - - - CodeMirror: VBScript mode - - - - - - - -

    CodeMirror: VBScript mode

    - -
    - - - -

    MIME types defined: text/vbscript.

    - - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vbscript/vbscript.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/vbscript/vbscript.js deleted file mode 100644 index 3e1eedc7db..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/vbscript/vbscript.js +++ /dev/null @@ -1,26 +0,0 @@ -CodeMirror.defineMode("vbscript", function() { - var regexVBScriptKeyword = /^(?:Call|Case|CDate|Clear|CInt|CLng|Const|CStr|Description|Dim|Do|Each|Else|ElseIf|End|Err|Error|Exit|False|For|Function|If|LCase|Loop|LTrim|Next|Nothing|Now|Number|On|Preserve|Quit|ReDim|Resume|RTrim|Select|Set|Sub|Then|To|Trim|True|UBound|UCase|Until|VbCr|VbCrLf|VbLf|VbTab)$/im; - - return { - token: function(stream) { - if (stream.eatSpace()) return null; - var ch = stream.next(); - if (ch == "'") { - stream.skipToEnd(); - return "comment"; - } - if (ch == '"') { - stream.skipTo('"'); - return "string"; - } - - if (/\w/.test(ch)) { - stream.eatWhile(/\w/); - if (regexVBScriptKeyword.test(stream.current())) return "keyword"; - } - return null; - } - }; -}); - -CodeMirror.defineMIME("text/vbscript", "vbscript"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/velocity/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/velocity/index.html deleted file mode 100644 index fb59cb590d..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/velocity/index.html +++ /dev/null @@ -1,103 +0,0 @@ - - - - - CodeMirror: Velocity mode - - - - - - - - -

    CodeMirror: Velocity mode

    -
    - - -

    MIME types defined: text/velocity.

    - - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/velocity/velocity.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/velocity/velocity.js deleted file mode 100644 index 43a97ba676..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/velocity/velocity.js +++ /dev/null @@ -1,144 +0,0 @@ -CodeMirror.defineMode("velocity", function() { - function parseWords(str) { - var obj = {}, words = str.split(" "); - for (var i = 0; i < words.length; ++i) obj[words[i]] = true; - return obj; - } - - var keywords = parseWords("#end #else #break #stop #[[ #]] " + - "#{end} #{else} #{break} #{stop}"); - var functions = parseWords("#if #elseif #foreach #set #include #parse #macro #define #evaluate " + - "#{if} #{elseif} #{foreach} #{set} #{include} #{parse} #{macro} #{define} #{evaluate}"); - var specials = parseWords("$foreach.count $foreach.hasNext $foreach.first $foreach.last $foreach.topmost $foreach.parent $velocityCount"); - var isOperatorChar = /[+\-*&%=<>!?:\/|]/; - - function chain(stream, state, f) { - state.tokenize = f; - return f(stream, state); - } - function tokenBase(stream, state) { - var beforeParams = state.beforeParams; - state.beforeParams = false; - var ch = stream.next(); - // start of string? - if ((ch == '"' || ch == "'") && state.inParams) - return chain(stream, state, tokenString(ch)); - // is it one of the special signs []{}().,;? Seperator? - else if (/[\[\]{}\(\),;\.]/.test(ch)) { - if (ch == "(" && beforeParams) state.inParams = true; - else if (ch == ")") state.inParams = false; - return null; - } - // start of a number value? - else if (/\d/.test(ch)) { - stream.eatWhile(/[\w\.]/); - return "number"; - } - // multi line comment? - else if (ch == "#" && stream.eat("*")) { - return chain(stream, state, tokenComment); - } - // unparsed content? - else if (ch == "#" && stream.match(/ *\[ *\[/)) { - return chain(stream, state, tokenUnparsed); - } - // single line comment? - else if (ch == "#" && stream.eat("#")) { - stream.skipToEnd(); - return "comment"; - } - // variable? - else if (ch == "$") { - stream.eatWhile(/[\w\d\$_\.{}]/); - // is it one of the specials? - if (specials && specials.propertyIsEnumerable(stream.current().toLowerCase())) { - return "keyword"; - } - else { - state.beforeParams = true; - return "builtin"; - } - } - // is it a operator? - else if (isOperatorChar.test(ch)) { - stream.eatWhile(isOperatorChar); - return "operator"; - } - else { - // get the whole word - stream.eatWhile(/[\w\$_{}]/); - var word = stream.current().toLowerCase(); - // is it one of the listed keywords? - if (keywords && keywords.propertyIsEnumerable(word)) - return "keyword"; - // is it one of the listed functions? - if (functions && functions.propertyIsEnumerable(word) || - stream.current().match(/^#[a-z0-9_]+ *$/i) && stream.peek()=="(") { - state.beforeParams = true; - return "keyword"; - } - // default: just a "word" - return null; - } - } - - function tokenString(quote) { - return function(stream, state) { - var escaped = false, next, end = false; - while ((next = stream.next()) != null) { - if (next == quote && !escaped) { - end = true; - break; - } - escaped = !escaped && next == "\\"; - } - if (end) state.tokenize = tokenBase; - return "string"; - }; - } - - function tokenComment(stream, state) { - var maybeEnd = false, ch; - while (ch = stream.next()) { - if (ch == "#" && maybeEnd) { - state.tokenize = tokenBase; - break; - } - maybeEnd = (ch == "*"); - } - return "comment"; - } - - function tokenUnparsed(stream, state) { - var maybeEnd = 0, ch; - while (ch = stream.next()) { - if (ch == "#" && maybeEnd == 2) { - state.tokenize = tokenBase; - break; - } - if (ch == "]") - maybeEnd++; - else if (ch != " ") - maybeEnd = 0; - } - return "meta"; - } - // Interface - - return { - startState: function() { - return { - tokenize: tokenBase, - beforeParams: false, - inParams: false - }; - }, - - token: function(stream, state) { - if (stream.eatSpace()) return null; - return state.tokenize(stream, state); - } - }; -}); - -CodeMirror.defineMIME("text/velocity", "velocity"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/verilog/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/verilog/index.html deleted file mode 100644 index f7c88c6231..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/verilog/index.html +++ /dev/null @@ -1,121 +0,0 @@ - - - - - CodeMirror: Verilog mode - - - - - - - -

    CodeMirror: Verilog mode

    - -
    - - - -

    Simple mode that tries to handle Verilog-like languages as well as it - can. Takes one configuration parameters: keywords, an - object whose property names are the keywords in the language.

    - -

    MIME types defined: text/x-verilog (Verilog code).

    - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/verilog/verilog.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/verilog/verilog.js deleted file mode 100644 index 708de23f49..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/verilog/verilog.js +++ /dev/null @@ -1,182 +0,0 @@ -CodeMirror.defineMode("verilog", function(config, parserConfig) { - var indentUnit = config.indentUnit, - keywords = parserConfig.keywords || {}, - blockKeywords = parserConfig.blockKeywords || {}, - atoms = parserConfig.atoms || {}, - hooks = parserConfig.hooks || {}, - multiLineStrings = parserConfig.multiLineStrings; - var isOperatorChar = /[&|~> - -Permission is hereby granted, free of charge, to any person obtaining a copy -of this software and associated documentation files (the "Software"), to deal -in the Software without restriction, including without limitation the rights -to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -copies of the Software, and to permit persons to whom the Software is -furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in -all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -THE SOFTWARE. \ No newline at end of file diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/index.html deleted file mode 100644 index 27acb89786..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/index.html +++ /dev/null @@ -1,221 +0,0 @@ - - - - - - CodeMirror: XQuery mode - - - - - - - - -

    CodeMirror: XQuery mode

    - -
    - -
    - - - -

    MIME types defined: application/xquery.

    - -

    Development of the CodeMirror XQuery mode was sponsored by - MarkLogic and developed by - Mike Brevoort. -

    - - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/test.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/test.js deleted file mode 100644 index 41719dd169..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/test.js +++ /dev/null @@ -1,64 +0,0 @@ -// Don't take these too seriously -- the expected results appear to be -// based on the results of actual runs without any serious manual -// verification. If a change you made causes them to fail, the test is -// as likely to wrong as the code. - -(function() { - var mode = CodeMirror.getMode({tabSize: 4}, "xquery"); - function MT(name) { test.mode(name, mode, Array.prototype.slice.call(arguments, 1)); } - - MT("eviltest", - "[keyword xquery] [keyword version] [variable "1][keyword .][atom 0][keyword -][variable ml"][def&variable ;] [comment (: this is : a \"comment\" :)]", - " [keyword let] [variable $let] [keyword :=] [variable <x] [variable attr][keyword =][variable "value">"test"<func>][def&variable ;function]() [variable $var] {[keyword function]()} {[variable $var]}[variable <][keyword /][variable func><][keyword /][variable x>]", - " [keyword let] [variable $joe][keyword :=][atom 1]", - " [keyword return] [keyword element] [variable element] {", - " [keyword attribute] [variable attribute] { [atom 1] },", - " [keyword element] [variable test] { [variable 'a'] }, [keyword attribute] [variable foo] { [variable "bar"] },", - " [def&variable fn:doc]()[[ [variable foo][keyword /][variable @bar] [keyword eq] [variable $let] ]],", - " [keyword //][variable x] } [comment (: a more 'evil' test :)]", - " [comment (: Modified Blakeley example (: with nested comment :) ... :)]", - " [keyword declare] [keyword private] [keyword function] [def&variable local:declare]() {()}[variable ;]", - " [keyword declare] [keyword private] [keyword function] [def&variable local:private]() {()}[variable ;]", - " [keyword declare] [keyword private] [keyword function] [def&variable local:function]() {()}[variable ;]", - " [keyword declare] [keyword private] [keyword function] [def&variable local:local]() {()}[variable ;]", - " [keyword let] [variable $let] [keyword :=] [variable <let>let] [variable $let] [keyword :=] [variable "let"<][keyword /let][variable >]", - " [keyword return] [keyword element] [variable element] {", - " [keyword attribute] [variable attribute] { [keyword try] { [def&variable xdmp:version]() } [keyword catch]([variable $e]) { [def&variable xdmp:log]([variable $e]) } },", - " [keyword attribute] [variable fn:doc] { [variable "bar"] [variable castable] [keyword as] [atom xs:string] },", - " [keyword element] [variable text] { [keyword text] { [variable "text"] } },", - " [def&variable fn:doc]()[[ [qualifier child::][variable eq][keyword /]([variable @bar] [keyword |] [qualifier attribute::][variable attribute]) [keyword eq] [variable $let] ]],", - " [keyword //][variable fn:doc]", - " }"); - - MT("testEmptySequenceKeyword", - "[string \"foo\"] [keyword instance] [keyword of] [keyword empty-sequence]()"); - - MT("testMultiAttr", - "[tag

    ][variable hello] [variable world][tag

    ]"); - - MT("test namespaced variable", - "[keyword declare] [keyword namespace] [variable e] [keyword =] [string \"http://example.com/ANamespace\"][variable ;declare] [keyword variable] [variable $e:exampleComThisVarIsNotRecognized] [keyword as] [keyword element]([keyword *]) [variable external;]"); - - MT("test EQName variable", - "[keyword declare] [keyword variable] [variable $\"http://www.example.com/ns/my\":var] [keyword :=] [atom 12][variable ;]", - "[tag ]{[variable $\"http://www.example.com/ns/my\":var]}[tag ]"); - - MT("test EQName function", - "[keyword declare] [keyword function] [def&variable \"http://www.example.com/ns/my\":fn] ([variable $a] [keyword as] [atom xs:integer]) [keyword as] [atom xs:integer] {", - " [variable $a] [keyword +] [atom 2]", - "}[variable ;]", - "[tag ]{[def&variable \"http://www.example.com/ns/my\":fn]([atom 12])}[tag ]"); - - MT("test EQName function with single quotes", - "[keyword declare] [keyword function] [def&variable 'http://www.example.com/ns/my':fn] ([variable $a] [keyword as] [atom xs:integer]) [keyword as] [atom xs:integer] {", - " [variable $a] [keyword +] [atom 2]", - "}[variable ;]", - "[tag ]{[def&variable 'http://www.example.com/ns/my':fn]([atom 12])}[tag ]"); - - MT("testProcessingInstructions", - "[def&variable data]([comment&meta ]) [keyword instance] [keyword of] [atom xs:string]"); - - MT("testQuoteEscapeDouble", - "[keyword let] [variable $rootfolder] [keyword :=] [string \"c:\\builds\\winnt\\HEAD\\qa\\scripts\\\"]", - "[keyword let] [variable $keysfolder] [keyword :=] [def&variable concat]([variable $rootfolder], [string \"keys\\\"])"); -})(); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/xquery.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/xquery.js deleted file mode 100644 index 95decc11b7..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/xquery/xquery.js +++ /dev/null @@ -1,450 +0,0 @@ -/* -Copyright (C) 2011 by MarkLogic Corporation -Author: Mike Brevoort - -Permission is hereby granted, free of charge, to any person obtaining a copy -of this software and associated documentation files (the "Software"), to deal -in the Software without restriction, including without limitation the rights -to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -copies of the Software, and to permit persons to whom the Software is -furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in -all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -THE SOFTWARE. -*/ -CodeMirror.defineMode("xquery", function() { - - // The keywords object is set to the result of this self executing - // function. Each keyword is a property of the keywords object whose - // value is {type: atype, style: astyle} - var keywords = function(){ - // conveinence functions used to build keywords object - function kw(type) {return {type: type, style: "keyword"};} - var A = kw("keyword a") - , B = kw("keyword b") - , C = kw("keyword c") - , operator = kw("operator") - , atom = {type: "atom", style: "atom"} - , punctuation = {type: "punctuation", style: null} - , qualifier = {type: "axis_specifier", style: "qualifier"}; - - // kwObj is what is return from this function at the end - var kwObj = { - 'if': A, 'switch': A, 'while': A, 'for': A, - 'else': B, 'then': B, 'try': B, 'finally': B, 'catch': B, - 'element': C, 'attribute': C, 'let': C, 'implements': C, 'import': C, 'module': C, 'namespace': C, - 'return': C, 'super': C, 'this': C, 'throws': C, 'where': C, 'private': C, - ',': punctuation, - 'null': atom, 'fn:false()': atom, 'fn:true()': atom - }; - - // a list of 'basic' keywords. For each add a property to kwObj with the value of - // {type: basic[i], style: "keyword"} e.g. 'after' --> {type: "after", style: "keyword"} - var basic = ['after','ancestor','ancestor-or-self','and','as','ascending','assert','attribute','before', - 'by','case','cast','child','comment','declare','default','define','descendant','descendant-or-self', - 'descending','document','document-node','element','else','eq','every','except','external','following', - 'following-sibling','follows','for','function','if','import','in','instance','intersect','item', - 'let','module','namespace','node','node','of','only','or','order','parent','precedes','preceding', - 'preceding-sibling','processing-instruction','ref','return','returns','satisfies','schema','schema-element', - 'self','some','sortby','stable','text','then','to','treat','typeswitch','union','variable','version','where', - 'xquery', 'empty-sequence']; - for(var i=0, l=basic.length; i < l; i++) { kwObj[basic[i]] = kw(basic[i]);}; - - // a list of types. For each add a property to kwObj with the value of - // {type: "atom", style: "atom"} - var types = ['xs:string', 'xs:float', 'xs:decimal', 'xs:double', 'xs:integer', 'xs:boolean', 'xs:date', 'xs:dateTime', - 'xs:time', 'xs:duration', 'xs:dayTimeDuration', 'xs:time', 'xs:yearMonthDuration', 'numeric', 'xs:hexBinary', - 'xs:base64Binary', 'xs:anyURI', 'xs:QName', 'xs:byte','xs:boolean','xs:anyURI','xf:yearMonthDuration']; - for(var i=0, l=types.length; i < l; i++) { kwObj[types[i]] = atom;}; - - // each operator will add a property to kwObj with value of {type: "operator", style: "keyword"} - var operators = ['eq', 'ne', 'lt', 'le', 'gt', 'ge', ':=', '=', '>', '>=', '<', '<=', '.', '|', '?', 'and', 'or', 'div', 'idiv', 'mod', '*', '/', '+', '-']; - for(var i=0, l=operators.length; i < l; i++) { kwObj[operators[i]] = operator;}; - - // each axis_specifiers will add a property to kwObj with value of {type: "axis_specifier", style: "qualifier"} - var axis_specifiers = ["self::", "attribute::", "child::", "descendant::", "descendant-or-self::", "parent::", - "ancestor::", "ancestor-or-self::", "following::", "preceding::", "following-sibling::", "preceding-sibling::"]; - for(var i=0, l=axis_specifiers.length; i < l; i++) { kwObj[axis_specifiers[i]] = qualifier; }; - - return kwObj; - }(); - - // Used as scratch variables to communicate multiple values without - // consing up tons of objects. - var type, content; - - function ret(tp, style, cont) { - type = tp; content = cont; - return style; - } - - function chain(stream, state, f) { - state.tokenize = f; - return f(stream, state); - } - - // the primary mode tokenizer - function tokenBase(stream, state) { - var ch = stream.next(), - mightBeFunction = false, - isEQName = isEQNameAhead(stream); - - // an XML tag (if not in some sub, chained tokenizer) - if (ch == "<") { - if(stream.match("!--", true)) - return chain(stream, state, tokenXMLComment); - - if(stream.match("![CDATA", false)) { - state.tokenize = tokenCDATA; - return ret("tag", "tag"); - } - - if(stream.match("?", false)) { - return chain(stream, state, tokenPreProcessing); - } - - var isclose = stream.eat("/"); - stream.eatSpace(); - var tagName = "", c; - while ((c = stream.eat(/[^\s\u00a0=<>\"\'\/?]/))) tagName += c; - - return chain(stream, state, tokenTag(tagName, isclose)); - } - // start code block - else if(ch == "{") { - pushStateStack(state,{ type: "codeblock"}); - return ret("", null); - } - // end code block - else if(ch == "}") { - popStateStack(state); - return ret("", null); - } - // if we're in an XML block - else if(isInXmlBlock(state)) { - if(ch == ">") - return ret("tag", "tag"); - else if(ch == "/" && stream.eat(">")) { - popStateStack(state); - return ret("tag", "tag"); - } - else - return ret("word", "variable"); - } - // if a number - else if (/\d/.test(ch)) { - stream.match(/^\d*(?:\.\d*)?(?:E[+\-]?\d+)?/); - return ret("number", "atom"); - } - // comment start - else if (ch === "(" && stream.eat(":")) { - pushStateStack(state, { type: "comment"}); - return chain(stream, state, tokenComment); - } - // quoted string - else if ( !isEQName && (ch === '"' || ch === "'")) - return chain(stream, state, tokenString(ch)); - // variable - else if(ch === "$") { - return chain(stream, state, tokenVariable); - } - // assignment - else if(ch ===":" && stream.eat("=")) { - return ret("operator", "keyword"); - } - // open paren - else if(ch === "(") { - pushStateStack(state, { type: "paren"}); - return ret("", null); - } - // close paren - else if(ch === ")") { - popStateStack(state); - return ret("", null); - } - // open paren - else if(ch === "[") { - pushStateStack(state, { type: "bracket"}); - return ret("", null); - } - // close paren - else if(ch === "]") { - popStateStack(state); - return ret("", null); - } - else { - var known = keywords.propertyIsEnumerable(ch) && keywords[ch]; - - // if there's a EQName ahead, consume the rest of the string portion, it's likely a function - if(isEQName && ch === '\"') while(stream.next() !== '"'){} - if(isEQName && ch === '\'') while(stream.next() !== '\''){} - - // gobble up a word if the character is not known - if(!known) stream.eatWhile(/[\w\$_-]/); - - // gobble a colon in the case that is a lib func type call fn:doc - var foundColon = stream.eat(":"); - - // if there's not a second colon, gobble another word. Otherwise, it's probably an axis specifier - // which should get matched as a keyword - if(!stream.eat(":") && foundColon) { - stream.eatWhile(/[\w\$_-]/); - } - // if the next non whitespace character is an open paren, this is probably a function (if not a keyword of other sort) - if(stream.match(/^[ \t]*\(/, false)) { - mightBeFunction = true; - } - // is the word a keyword? - var word = stream.current(); - known = keywords.propertyIsEnumerable(word) && keywords[word]; - - // if we think it's a function call but not yet known, - // set style to variable for now for lack of something better - if(mightBeFunction && !known) known = {type: "function_call", style: "variable def"}; - - // if the previous word was element, attribute, axis specifier, this word should be the name of that - if(isInXmlConstructor(state)) { - popStateStack(state); - return ret("word", "variable", word); - } - // as previously checked, if the word is element,attribute, axis specifier, call it an "xmlconstructor" and - // push the stack so we know to look for it on the next word - if(word == "element" || word == "attribute" || known.type == "axis_specifier") pushStateStack(state, {type: "xmlconstructor"}); - - // if the word is known, return the details of that else just call this a generic 'word' - return known ? ret(known.type, known.style, word) : - ret("word", "variable", word); - } - } - - // handle comments, including nested - function tokenComment(stream, state) { - var maybeEnd = false, maybeNested = false, nestedCount = 0, ch; - while (ch = stream.next()) { - if (ch == ")" && maybeEnd) { - if(nestedCount > 0) - nestedCount--; - else { - popStateStack(state); - break; - } - } - else if(ch == ":" && maybeNested) { - nestedCount++; - } - maybeEnd = (ch == ":"); - maybeNested = (ch == "("); - } - - return ret("comment", "comment"); - } - - // tokenizer for string literals - // optionally pass a tokenizer function to set state.tokenize back to when finished - function tokenString(quote, f) { - return function(stream, state) { - var ch; - - if(isInString(state) && stream.current() == quote) { - popStateStack(state); - if(f) state.tokenize = f; - return ret("string", "string"); - } - - pushStateStack(state, { type: "string", name: quote, tokenize: tokenString(quote, f) }); - - // if we're in a string and in an XML block, allow an embedded code block - if(stream.match("{", false) && isInXmlAttributeBlock(state)) { - state.tokenize = tokenBase; - return ret("string", "string"); - } - - - while (ch = stream.next()) { - if (ch == quote) { - popStateStack(state); - if(f) state.tokenize = f; - break; - } - else { - // if we're in a string and in an XML block, allow an embedded code block in an attribute - if(stream.match("{", false) && isInXmlAttributeBlock(state)) { - state.tokenize = tokenBase; - return ret("string", "string"); - } - - } - } - - return ret("string", "string"); - }; - } - - // tokenizer for variables - function tokenVariable(stream, state) { - var isVariableChar = /[\w\$_-]/; - - // a variable may start with a quoted EQName so if the next character is quote, consume to the next quote - if(stream.eat("\"")) { - while(stream.next() !== '\"'){}; - stream.eat(":"); - } else { - stream.eatWhile(isVariableChar); - if(!stream.match(":=", false)) stream.eat(":"); - } - stream.eatWhile(isVariableChar); - state.tokenize = tokenBase; - return ret("variable", "variable"); - } - - // tokenizer for XML tags - function tokenTag(name, isclose) { - return function(stream, state) { - stream.eatSpace(); - if(isclose && stream.eat(">")) { - popStateStack(state); - state.tokenize = tokenBase; - return ret("tag", "tag"); - } - // self closing tag without attributes? - if(!stream.eat("/")) - pushStateStack(state, { type: "tag", name: name, tokenize: tokenBase}); - if(!stream.eat(">")) { - state.tokenize = tokenAttribute; - return ret("tag", "tag"); - } - else { - state.tokenize = tokenBase; - } - return ret("tag", "tag"); - }; - } - - // tokenizer for XML attributes - function tokenAttribute(stream, state) { - var ch = stream.next(); - - if(ch == "/" && stream.eat(">")) { - if(isInXmlAttributeBlock(state)) popStateStack(state); - if(isInXmlBlock(state)) popStateStack(state); - return ret("tag", "tag"); - } - if(ch == ">") { - if(isInXmlAttributeBlock(state)) popStateStack(state); - return ret("tag", "tag"); - } - if(ch == "=") - return ret("", null); - // quoted string - if (ch == '"' || ch == "'") - return chain(stream, state, tokenString(ch, tokenAttribute)); - - if(!isInXmlAttributeBlock(state)) - pushStateStack(state, { type: "attribute", name: name, tokenize: tokenAttribute}); - - stream.eat(/[a-zA-Z_:]/); - stream.eatWhile(/[-a-zA-Z0-9_:.]/); - stream.eatSpace(); - - // the case where the attribute has not value and the tag was closed - if(stream.match(">", false) || stream.match("/", false)) { - popStateStack(state); - state.tokenize = tokenBase; - } - - return ret("attribute", "attribute"); - } - - // handle comments, including nested - function tokenXMLComment(stream, state) { - var ch; - while (ch = stream.next()) { - if (ch == "-" && stream.match("->", true)) { - state.tokenize = tokenBase; - return ret("comment", "comment"); - } - } - } - - - // handle CDATA - function tokenCDATA(stream, state) { - var ch; - while (ch = stream.next()) { - if (ch == "]" && stream.match("]", true)) { - state.tokenize = tokenBase; - return ret("comment", "comment"); - } - } - } - - // handle preprocessing instructions - function tokenPreProcessing(stream, state) { - var ch; - while (ch = stream.next()) { - if (ch == "?" && stream.match(">", true)) { - state.tokenize = tokenBase; - return ret("comment", "comment meta"); - } - } - } - - - // functions to test the current context of the state - function isInXmlBlock(state) { return isIn(state, "tag"); } - function isInXmlAttributeBlock(state) { return isIn(state, "attribute"); } - function isInXmlConstructor(state) { return isIn(state, "xmlconstructor"); } - function isInString(state) { return isIn(state, "string"); } - - function isEQNameAhead(stream) { - // assume we've already eaten a quote (") - if(stream.current() === '"') - return stream.match(/^[^\"]+\"\:/, false); - else if(stream.current() === '\'') - return stream.match(/^[^\"]+\'\:/, false); - else - return false; - } - - function isIn(state, type) { - return (state.stack.length && state.stack[state.stack.length - 1].type == type); - } - - function pushStateStack(state, newState) { - state.stack.push(newState); - } - - function popStateStack(state) { - state.stack.pop(); - var reinstateTokenize = state.stack.length && state.stack[state.stack.length-1].tokenize; - state.tokenize = reinstateTokenize || tokenBase; - } - - // the interface for the mode API - return { - startState: function() { - return { - tokenize: tokenBase, - cc: [], - stack: [] - }; - }, - - token: function(stream, state) { - if (stream.eatSpace()) return null; - var style = state.tokenize(stream, state); - return style; - } - }; - -}); - -CodeMirror.defineMIME("application/xquery", "xquery"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/yaml/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/yaml/index.html deleted file mode 100644 index 65e1ea73fb..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/yaml/index.html +++ /dev/null @@ -1,68 +0,0 @@ - - - - - CodeMirror: YAML mode - - - - - - - -

    CodeMirror: YAML mode

    -
    - - -

    MIME types defined: text/x-yaml.

    - - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/yaml/yaml.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/yaml/yaml.js deleted file mode 100644 index 7a095b31db..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/yaml/yaml.js +++ /dev/null @@ -1,97 +0,0 @@ -CodeMirror.defineMode("yaml", function() { - - var cons = ['true', 'false', 'on', 'off', 'yes', 'no']; - var keywordRegex = new RegExp("\\b(("+cons.join(")|(")+"))$", 'i'); - - return { - token: function(stream, state) { - var ch = stream.peek(); - var esc = state.escaped; - state.escaped = false; - /* comments */ - if (ch == "#" && (stream.pos == 0 || /\s/.test(stream.string.charAt(stream.pos - 1)))) { - stream.skipToEnd(); return "comment"; - } - if (state.literal && stream.indentation() > state.keyCol) { - stream.skipToEnd(); return "string"; - } else if (state.literal) { state.literal = false; } - if (stream.sol()) { - state.keyCol = 0; - state.pair = false; - state.pairStart = false; - /* document start */ - if(stream.match(/---/)) { return "def"; } - /* document end */ - if (stream.match(/\.\.\./)) { return "def"; } - /* array list item */ - if (stream.match(/\s*-\s+/)) { return 'meta'; } - } - /* pairs (associative arrays) -> key */ - if (!state.pair && stream.match(/^\s*([a-z0-9\._-])+(?=\s*:)/i)) { - state.pair = true; - state.keyCol = stream.indentation(); - return "atom"; - } - if (state.pair && stream.match(/^:\s*/)) { state.pairStart = true; return 'meta'; } - - /* inline pairs/lists */ - if (stream.match(/^(\{|\}|\[|\])/)) { - if (ch == '{') - state.inlinePairs++; - else if (ch == '}') - state.inlinePairs--; - else if (ch == '[') - state.inlineList++; - else - state.inlineList--; - return 'meta'; - } - - /* list seperator */ - if (state.inlineList > 0 && !esc && ch == ',') { - stream.next(); - return 'meta'; - } - /* pairs seperator */ - if (state.inlinePairs > 0 && !esc && ch == ',') { - state.keyCol = 0; - state.pair = false; - state.pairStart = false; - stream.next(); - return 'meta'; - } - - /* start of value of a pair */ - if (state.pairStart) { - /* block literals */ - if (stream.match(/^\s*(\||\>)\s*/)) { state.literal = true; return 'meta'; }; - /* references */ - if (stream.match(/^\s*(\&|\*)[a-z0-9\._-]+\b/i)) { return 'variable-2'; } - /* numbers */ - if (state.inlinePairs == 0 && stream.match(/^\s*-?[0-9\.\,]+\s?$/)) { return 'number'; } - if (state.inlinePairs > 0 && stream.match(/^\s*-?[0-9\.\,]+\s?(?=(,|}))/)) { return 'number'; } - /* keywords */ - if (stream.match(keywordRegex)) { return 'keyword'; } - } - - /* nothing found, continue */ - state.pairStart = false; - state.escaped = (ch == '\\'); - stream.next(); - return null; - }, - startState: function() { - return { - pair: false, - pairStart: false, - keyCol: 0, - inlinePairs: 0, - inlineList: 0, - literal: false, - escaped: false - }; - } - }; -}); - -CodeMirror.defineMIME("text/x-yaml", "yaml"); diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/z80/index.html b/wcfsetup/install/files/js/3rdParty/codemirror/mode/z80/index.html deleted file mode 100644 index 133c870e32..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/z80/index.html +++ /dev/null @@ -1,39 +0,0 @@ - - - - - CodeMirror: Z80 assembly mode - - - - - - - -

    CodeMirror: Z80 assembly mode

    - -
    - - - -

    MIME type defined: text/x-z80.

    - - diff --git a/wcfsetup/install/files/js/3rdParty/codemirror/mode/z80/z80.js b/wcfsetup/install/files/js/3rdParty/codemirror/mode/z80/z80.js deleted file mode 100644 index ff43d32b52..0000000000 --- a/wcfsetup/install/files/js/3rdParty/codemirror/mode/z80/z80.js +++ /dev/null @@ -1,85 +0,0 @@ -CodeMirror.defineMode('z80', function() { - var keywords1 = /^(exx?|(ld|cp|in)([di]r?)?|pop|push|ad[cd]|cpl|daa|dec|inc|neg|sbc|sub|and|bit|[cs]cf|x?or|res|set|r[lr]c?a?|r[lr]d|s[lr]a|srl|djnz|nop|rst|[de]i|halt|im|ot[di]r|out[di]?)\b/i; - var keywords2 = /^(call|j[pr]|ret[in]?)\b/i; - var keywords3 = /^b_?(call|jump)\b/i; - var variables1 = /^(af?|bc?|c|de?|e|hl?|l|i[xy]?|r|sp)\b/i; - var variables2 = /^(n?[zc]|p[oe]?|m)\b/i; - var errors = /^([hl][xy]|i[xy][hl]|slia|sll)\b/i; - var numbers = /^([\da-f]+h|[0-7]+o|[01]+b|\d+)\b/i; - - return { - startState: function() { - return {context: 0}; - }, - token: function(stream, state) { - if (!stream.column()) - state.context = 0; - - if (stream.eatSpace()) - return null; - - var w; - - if (stream.eatWhile(/\w/)) { - w = stream.current(); - - if (stream.indentation()) { - if (state.context == 1 && variables1.test(w)) - return 'variable-2'; - - if (state.context == 2 && variables2.test(w)) - return 'variable-3'; - - if (keywords1.test(w)) { - state.context = 1; - return 'keyword'; - } else if (keywords2.test(w)) { - state.context = 2; - return 'keyword'; - } else if (keywords3.test(w)) { - state.context = 3; - return 'keyword'; - } - - if (errors.test(w)) - return 'error'; - } else if (numbers.test(w)) { - return 'number'; - } else { - return null; - } - } else if (stream.eat(';')) { - stream.skipToEnd(); - return 'comment'; - } else if (stream.eat('"')) { - while (w = stream.next()) { - if (w == '"') - break; - - if (w == '\\') - stream.next(); - } - return 'string'; - } else if (stream.eat('\'')) { - if (stream.match(/\\?.'/)) - return 'number'; - } else if (stream.eat('.') || stream.sol() && stream.eat('#')) { - state.context = 4; - - if (stream.eatWhile(/\w/)) - return 'def'; - } else if (stream.eat('$')) { - if (stream.eatWhile(/[\da-f]/i)) - return 'number'; - } else if (stream.eat('%')) { - if (stream.eatWhile(/[01]/)) - return 'number'; - } else { - stream.next(); - } - return null; - } - }; -}); - -CodeMirror.defineMIME("text/x-z80", "z80"); -- 2.20.1