Bug 106521

Summary: [WinCE][V8]document.write(i) broken on 3rd loop on WebKit's WinCE port integrated with V8 JIT engine
Product: WebKit Reporter: Chen Zhixiang <chenzx>
Component: New BugsAssignee: Nobody <webkit-unassigned>
Status: RESOLVED WONTFIX    
Severity: Blocker CC: paroga
Priority: P5    
Version: 528+ (Nightly build)   
Hardware: Other   
OS: Other   
Attachments:
Description Flags
V8 dissemable log for test case none

Chen Zhixiang
Reported 2013-01-09 19:37:38 PST
Created attachment 182049 [details] V8 dissemable log for test case I'm doing WebKit's WinCE(arm) port, using VS2005, WinCE 6.0 and port v8 engine into it However, i wrote a simple html test file: <body> <script> for(var =0; i<5; i++) document.write(i); </script> </body> It should output 01234, but when I testes it on WinCE device simulator, It only output 01 and then crashed. That's to say, it can successfully run the first 2 loop, but fails at 3rd loop. The crash breakpoint is: v8\include\v8.h: Local<Object> Arguments::Holder() const { return Local<Object>(reinterpret_cast<Object*>( > &implicit_args_[kHolderIndex])); } Source\WebCore\bindings\v8\custom\V8HTMLDocumentCustom.cpp: v8::Handle<v8::Value> V8HTMLDocument::writelnCallback(const v8::Arguments& args) { INC_STATS("DOM.HTMLDocument.writeln()"); > HTMLDocument* htmlDocument = V8HTMLDocument::toNative(args.Holder()); htmlDocument->writeln(writeHelperGetString(args), activeDOMWindow(BindingState::instance())->document()); return v8::Undefined(); } I can not provide the full call stack, this 2 nearest stack trace is only deferred from crash log. Can't figure out Why, for i cannot debug on WinCE simulator. Attachment is V8 dissemble log for this test html. The key point is: 1、Using loop, for/while, or even recursion like this: function test(i){ if(i==4) return; document.write(i); test(++i); } test(0); 2、WinCE simulator run will be broken at exactly the 3rd loop run! 3、I'm using the V8 trunk code I hesitated if it's V8 doesn't support old WinCE 6.0 port, since VS2005 arm compiler generated code may be in-consistent with V8-arm's JIT engine? If I Enable USE_SIMULATOR, that is to say, Running WebKit-WinCE-launcher-with-V8 on VS2005 WinCE 6.0 Simulator, but the V8 JIT-ed code is not directly run by simulator, but interpreted by the v8's arm-simulator, which is C++ code running on WinCE simulator, This settings compiled app runs OK, nothing wrong. So, where does the difference stands between the 2 situation?
Attachments
V8 dissemable log for test case (56.47 KB, text/plain)
2013-01-09 19:37 PST, Chen Zhixiang 		no flags
Details
View All Add attachment proposed patch, testcase, etc.
Patrick R. Gansterer
Comment 1 2013-01-28 10:01:46 PST
Do you know if it is a V8 or WebKit problem? Does the same code work in the v8 shell?
Chen Zhixiang
Comment 2 2013-01-28 18:22:22 PST
We now doubt it's the `OEMCacheRangeFlush` problem. JIT engine always accquires this API, but we are now using `FlushInstructionCache` from winbase.h. Don't know if WinCE 6.0 emulator implements the same behavior in `OEMCacheRangeFlush` and `FlushInstructionCache`. V8 Shell has no such problem, for it can't use `document.write`. If it is a webkit problem, why does the first 2 loop `document.write` success? I've used 3 kinds of loop: for、while、tail-recursive-call, all the same result. We are using a customed PB build for which we can't link to `OEMCacheRangeFlush` API, so I hesitate this may be the reason of problem.
Chen Zhixiang
Comment 3 2013-03-20 02:00:16 PDT
QtWebKit guys recently seems to having solved this problem, By disable inline cache fast path? https://codereview.qt-project.org/#change,45152
Patrick R. Gansterer
Comment 4 2013-04-06 08:24:15 PDT
Closing since V8 won't be supported with WebKit any more.
Note You need to log in before you can comment on or make changes to this bug.