Bug 21057 – Crash in RegisterID::deref() running fast/canvas/canvas-putImageData.html

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Alexey Proskuryakov]

Reliably crashes under GuardMalloc:

run-webkit-tests -g fast/canvas/canvas-putImageData.html

Thread 0 Crashed:
0   com.apple.JavaScriptCore      	0x004569c8 JSC::RegisterID::deref() + 16 (RegisterID.h:91)
1   com.apple.JavaScriptCore      	0x00459ad9 WTF::RefPtr<JSC::RegisterID>::~RefPtr() + 31
2   com.apple.JavaScriptCore      	0x00490b8e JSC::CodeGenerator::~CodeGenerator() + 194
3   com.apple.JavaScriptCore      	0x00446098 JSC::ProgramNode::generateCode(JSC::ScopeChainNode*) + 430 (nodes.cpp:1851)
4   com.apple.JavaScriptCore      	0x004ca159 JSC::ProgramNode::byteCode(JSC::ScopeChainNode*) + 43 (nodes.h:2205)
5   com.apple.JavaScriptCore      	0x004b7a52 JSC::Machine::execute(JSC::ProgramNode*, JSC::ExecState*, JSC::ScopeChainNode*, JSC::JSObject*, JSC::JSValue**) + 62 (Machine.cpp:885)
6   com.apple.JavaScriptCore      	0x00442927 JSC::Interpreter::evaluate(JSC::ExecState*, JSC::ScopeChain&, JSC::UString const&, int, WTF::PassRefPtr<JSC::SourceProvider>, JSC::JSValue*) + 427 (interpreter.cpp:83)

[Cameron Zwarich (cpst)]

This is likely caused by r36821, since that is where the only RefPtr<RegisterID> instance variable for CodeGenerator was defined:

http://trac.webkit.org/changeset/36821

[Geoffrey Garen]

Ah, looks like the RefPtr destructor runs after the SegmentedVector destructor. Should be easy to fix.

[Geoffrey Garen]

Created attachment 23750 [details]
patch

[Geoffrey Garen]

I haven't been able to verify this patch with the original test case, because DRT with GuardMalloc crashes in LaunchServices on my machine. However, I did verify that this patch fixes the destructor order.

[Cameron Zwarich (cpst)]

Comment on attachment 23750 [details]
patch

r=me

[Geoffrey Garen]

Committed revision 36853.