[Webkit-unassigned] [Bug 90098] [GTK] LLint build fails with -g -02

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=90098





--- Comment #2 from Filip Pizlo <fpizlo at apple.com>  2012-07-03 08:46:05 PST ---
(In reply to comment #0)
> Created an attachment (id=149791)
 --> (https://bugs.webkit.org/attachment.cgi?id=149791&action=review) [details]
> fix llint build wwith -02 -g 
> 
> /usr/bin/ruby ./Source/JavaScriptCore/offlineasm/asm.rb ./Source/JavaScriptCore/llint/LowLevelInterpreter.asm Programs/LLIntOffsetsExtractor DerivedSources/JavaScriptCore/LLIntAssembly.h
>  offsetsAndConfigurationIndex in Source/JavaScriptCore/offlineasm/offsets.rb raise an unhandled exception during build if -02 -g flags are used (ie set CFLAGS to "-O0 -g" and not set webkit debug configure switch which ends up with "-00 -g - 02" => "-g -O2").
> the exception is raised at:
> raise if result.map{|v| v[1]}.uniq.size < result.map{|v| v[1]}.size
> 
> It turns out that the way extractorTable is defined inside the LLIntOffsetsExtractor::dummy which itself is defined as a class member inside the class declaration leads to it behing defined in the assembly twice.

In that case, the correct solution is to simply remove the assertion and harden the code against duplicates (i.e. if it finds a duplicate then return the first one, or the last one, or the middle one, or whichever one you like).

On the other hand, your fix will break fat binary builds, which is a show-stopper for us.  A fat binary build produces a binary with two or more different builds in it; in that case the offsets extractor needs to find all of the extractor tables.  Your patch prevents finding all extractor tables because all but the first will have a corrupted magic number.

> That is the class member with -O2 is inlined (which is common for class members at least with g++). The static local variable extractorTable is thus duplicated for each translation unit. And ends up in the usual symbols  and the debug_info symbols. Ie twice. Then offsetsAndConfigurationIndex in Source/JavaScriptCore/offlineasm/offsets.rb  finds extractorTable magic numbers twice in the binary object and the result.map{|v| v[1]}.uniq.size < result.map{|v| v[1]}.size becomes 1 < 2 and the exception is raised.
> 
> Even if done by mistake (I intended to build -O0 -g) -g -02 is quite common for distribution that provide -dbg packages. So this issue might be major instead of normal.
> 
> Running:
> $ /usr/bin/ruby ./Source/JavaScriptCore/offlineasm/generate_offset_extractor.rb ./Source/JavaScriptCore/llint/LowLevelInterpreter.asm DerivedSources/JavaScriptCore/LLIntDesiredOffsets.h
> $ g++ -DHAVE_CONFIG_H -I.  -Wall -W -Wcast-align -Wchar-subscripts -Wreturn-type -Wformat -Wformat-security -Wno-format-y2k -Wundef -Wmissing-format-attribute -Wpointer-arith -Wwrite-strings -Wno-unused-parameter -Wno-parentheses -fno-exceptions -DENABLE_GLIB_SUPPORT=1 -DBUILDING_CAIRO__=1 -DBUILDING_GTK__=1 -DBUILDING_SOUP__=1 -DWTF_CHANGES -DBUILDING_WEBKIT2__=1 -DXP_UNIX -DMOZ_X11  -DWTF_USE_ICU_UNICODE=1   -DWTF_USE_GSTREAMER=1 -DWTF_USE_ACCELERATED_COMPOSITING=1 -DENABLE_3D_RENDERING=1  -DWTF_USE_TEXTURE_MAPPER=1 -DWTF_USE_TEXTURE_MAPPER_GL=1  -DWTF_USE_GLX=1 -DWTF_USE_OPENGL=1  -DNDEBUG    -I./Source -I./Source/JavaScriptCore -I./Source/JavaScriptCore/API -I./Source/JavaScriptCore/assembler -I./Source/JavaScriptCore/bytecode -I./Source/JavaScriptCore/bytecompiler -I./Source/JavaScriptCore/dfg -I./Source/JavaScriptCore/heap -I./Source/JavaScriptCore/debugger -I./Source/JavaScriptCore/ForwardingHeaders -I./Source/JavaScriptCore/interpreter -I./Source/JavaScriptCore/jit 
 -I./Source/JavaScriptCore/jit -I./Source/JavaScriptCore/llint -I./Source/JavaScriptCore/parser -I./Source/JavaScriptCore/profiler -I./Source/JavaScriptCore/runtime -I./Source/JavaScriptCore/tools -I./Source/JavaScriptCore/yarr -I./DerivedSources/JavaScriptCore -I./Source/WTF   -O0 -g -Wno-c++0x-compat -O2 -MT Source/JavaScriptCore/llint/Programs_LLIntOffsetsExtractor-LLIntOffsetsExtractor.o -MD -MP -MF Source/JavaScriptCore/llint/.deps/Programs_LLIntOffsetsExtractor-LLIntOffsetsExtractor.Tpo -c -o Source/JavaScriptCore/llint/Programs_LLIntOffsetsExtractor-LLIntOffsetsExtractor.o `test -f 'Source/JavaScriptCore/llint/LLIntOffsetsExtractor.cpp' || echo './'`Source/JavaScriptCore/llint/LLIntOffsetsExtractor.cpp
> $ /usr/bin/ruby ./Source/JavaScriptCore/offlineasm/asm.rb ./Source/JavaScriptCore/llint/LowLevelInterpreter.asm Programs/LLIntOffsetsExtractor DerivedSources/JavaScriptCore/LLIntAssembly.h
> from an existing build tree of webkit should give the above error.
> 
> 
> The attached patch fixes this.
> 
> 
> This is tested against git mirror master 4c742312f5b69a166256a28390aedc4c0bd45ac1, ie :
> 
> commit 4c742312f5b69a166256a28390aedc4c0bd45ac1
> Author: commit-queue at webkit.org <commit-queue at webkit.org@268f45cc-cd09-0410-ab3c-d52691b4dbfc>
> Date:   Thu Jun 21 02:03:44 2012 +0000
> 
>     Unreviewed, rolling out r120889.
>     http://trac.webkit.org/changeset/120889
>     https://bugs.webkit.org/show_bug.cgi?id=89630
> 
>     [Chromium] webkit_unit_tests didDrawNotCalledOnHiddenLayer
>     start failing (Requested by ukai on #webkit).
> 
>     Patch by Sheriff Bot <webkit.review.bot at gmail.com> on 2012-06-20
> 
>     Source/WebCore:
> 
>     * platform/graphics/chromium/cc/CCLayerTreeHostImpl.cpp:
>     (WebCore::CCLayerTreeHostImpl::calculateRenderSurfaceLayerList):
> 
>     Source/WebKit/chromium:
> 
>     * tests/CCLayerTreeHostImplTest.cpp:
> 
>     git-svn-id: http://svn.webkit.org/repository/webkit/trunk@120899 268f45cc-cd09-0410-ab3c-d5

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.