kernel-doc: strip attributes even if they have an argument
authorPaolo Bonzini <pbonzini@redhat.com>
Mon, 2 Jan 2017 15:22:24 +0000 (16:22 +0100)
committerJonathan Corbet <corbet@lwn.net>
Wed, 4 Jan 2017 22:07:42 +0000 (15:07 -0700)
An inline function can have an attribute, as in include/linux/log2.h,
and kernel-doc handles this already for simple cases.  However,
some attributes have arguments (e.g. the "target" attribute).
Handle those too.

Furthermore, attributes could be at the beginning of a function
declaration, before the return type.  To correctly handle this case,
you need to strip spaces after the attributes; otherwise, dump_function
is left confused.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
scripts/kernel-doc

index c1ea91c2e497ce5abc0f73241a61044094d66723..99b4847f9bb225dde733a0ecbb78c1370db9ccb1 100755 (executable)
@@ -2506,7 +2506,13 @@ sub dump_function($$) {
     $prototype =~ s/__must_check +//;
     $prototype =~ s/__weak +//;
     my $define = $prototype =~ s/^#\s*define\s+//; #ak added
-    $prototype =~ s/__attribute__\s*\(\([a-z,]*\)\)//;
+    $prototype =~ s/__attribute__\s*\(\(
+            (?:
+                 [\w\s]++          # attribute name
+                 (?:\([^)]*+\))?   # attribute arguments
+                 \s*+,?            # optional comma at the end
+            )+
+          \)\)\s+//x;
 
     # Yes, this truly is vile.  We are looking for:
     # 1. Return type (may be nothing if we're looking at a macro)