Bug 127909 – [GTK] Disable optimizations for JSC that turned out malignant after jsCStack branch merge

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Zan Dobersek]

[GTK] Disable optimizations for JSC that turned out malignant after jsCStack branch merge

[Zan Dobersek]

Created attachment 222668 [details]
Patch

[Csaba Osztrogonác]

-fno-omit-frame-pointer is reasonable, but I'm a little bit sceptic about
-fno-tree-dce . Of course it can be good as a workaround, but it would be
great to find the root of this problem. Could you possibly add a detailed 
comment with bug URL to the GNUmakefile.am too?

I'm not so familiar with autotools files, so I'd like 
to let the final review for the GTK maintainers.

[Carlos Garcia Campos]

(In reply to comment #2)
> -fno-omit-frame-pointer is reasonable, but I'm a little bit sceptic about
> -fno-tree-dce . Of course it can be good as a workaround, but it would be
> great to find the root of this problem. Could you possibly add a detailed 
> comment with bug URL to the GNUmakefile.am too?
> 
> I'm not so familiar with autotools files, so I'd like 
> to let the final review for the GTK maintainers.

I agree, let's land this just as a workaround, but we should re-enable the optimizations at some point. Maybe we could file bug reports for that.

[Alberto Garcia]

I think that should go to CFLAGS (compiler flags), not CPPFLAGS
(preprocessor flags).

--- Source/JavaScriptCore/GNUmakefile.am	(revision 163081)
+++ Source/JavaScriptCore/GNUmakefile.am	(working copy)
@@ -70,6 +70,8 @@ javascriptcore_cppflags += \
 	-I$(srcdir)/Source/JavaScriptCore/yarr \
 	-I$(top_builddir)/DerivedSources/JavaScriptCore \
 	-I$(srcdir)/Source/WTF \
+	-fno-omit-frame-pointer \
+	-fno-tree-dce \
 	$(LLVM_CFLAGS)
 
 javascriptcore_cflags += \

[Alberto Garcia]

(In reply to comment #4)
> I think that should go to CFLAGS (compiler flags), not CPPFLAGS
> (preprocessor flags).

CXXFLAGS actually, or are we using CPPFLAGS to make sure that they're
included in both cases? :)

[Zan Dobersek]

(In reply to comment #5)
> (In reply to comment #4)
> > I think that should go to CFLAGS (compiler flags), not CPPFLAGS
> > (preprocessor flags).
> 
> CXXFLAGS actually, or are we using CPPFLAGS to make sure that they're
> included in both cases? :)

Yes, we're trying to cover both cases.

[Alberto Garcia]

My clang installation (3.4) doesn't seem to like this flag:

clang: error: unknown argument: '-fno-tree-dce'

[Geoffrey Garen]

> -fno-omit-frame-pointer is reasonable, but I'm a little bit sceptic about
> -fno-tree-dce . 

Me too.

[Zan Dobersek]

(In reply to comment #7)
> My clang installation (3.4) doesn't seem to like this flag:
> 
> clang: error: unknown argument: '-fno-tree-dce'

Already addressed in bug #127911.

[Zan Dobersek]

(In reply to comment #8)
> > -fno-omit-frame-pointer is reasonable, but I'm a little bit sceptic about
> > -fno-tree-dce . 
> 
> Me too.

The optimization is only problematic with GCC 4.8 but not with GCC 4.7, so this might be a bug introduced in the compiler.

[Zan Dobersek]

Landed in r163083.
http://trac.webkit.org/changeset/163083