Bug 5518 – Consistent crash using liveconnect/javascript @ www.binarybet.com

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Andrew Wright]

Our site uses a heavy mixture of frames, javascript, liveconnect to an applet....you name it. It basically 
works but is quite flakey - Safari (or the nightly builds) crash just during navigation around the site. 
The crash reports are all very similiar, down in KJS::Bindings::dispatchJNICall.

This happens across all versions of Safari - I'm testing with fully patched 10.4.2. I've gotten as far as 
debugging ToT source in xcode but my obj-c skills are somewhat lacking :-/

Relevant portion of crashdump attached. Happy to build/update/test/provide more info if it will help.



Date/Time:      2005-10-27 09:47:26.366 +0100
OS Version:     10.4.2 (Build 8C46)
Report Version: 3

Command: Safari
Path:    /Applications/Safari.app/Contents/MacOS/Safari
Parent:  WindowServer [87]

Version:        2.0.1 (412.5)
Build Version:  7
Project Name:   WebBrowser
Source Version: 4120500

PID:    23018
Thread: 0

Exception:  EXC_BAD_ACCESS (0x0001)
Codes:      KERN_INVALID_ADDRESS (0x0001) at 0x162a096c

Thread 0 Crashed:
0   <<00000000>> 	0xfffeff30 objc_msgSend_rtp + 48
1   com.apple.JavaScriptCore  	0x95a71f14 KJS::Bindings::dispatchJNICall(void const*, _jobject*, bool, 
JNIType, _jmethodID*, jvalue*, jvalue&, char const*, KJS::Value&) + 80
2   com.apple.JavaScriptCore  	0x95a68840 KJS::Bindings::JavaInstance::invokeMethod(KJS::ExecState*, 
KJS::Bindings::MethodList const&, KJS::List const&) + 460
3   com.apple.JavaScriptCore  	0x95a691ac KJS::RuntimeMethodImp::call(KJS::ExecState*, KJS::Object&, 
KJS::List const&) + 268
4   com.apple.JavaScriptCore  	0x95a34d70 KJS::Object::call(KJS::ExecState*, KJS::Object&, KJS::List 
const&) + 172
5   com.apple.JavaScriptCore  	0x95a33670 KJS::FunctionCallNode::evaluate(KJS::ExecState*) + 932
6   com.apple.JavaScriptCore  	0x95a37d7c KJS::ArgumentListNode::evaluateList(KJS::ExecState*) + 80
7   com.apple.JavaScriptCore  	0x95a347d0 KJS::ArgumentsNode::evaluateList(KJS::ExecState*) + 44
8   com.apple.JavaScriptCore  	0x95a33370 KJS::FunctionCallNode::evaluate(KJS::ExecState*) + 164
9   com.apple.JavaScriptCore  	0x95a332b4 KJS::AssignExprNode::evaluate(KJS::ExecState*) + 40
10  com.apple.JavaScriptCore  	0x95a331c4 KJS::VarDeclNode::evaluate(KJS::ExecState*) + 124
11  com.apple.JavaScriptCore  	0x95a330f4 KJS::VarDeclListNode::evaluate(KJS::ExecState*) + 68
12  com.apple.JavaScriptCore  	0x95a32f50 KJS::VarStatementNode::execute(KJS::ExecState*) + 128
13  com.apple.JavaScriptCore  	0x95a55ba4 KJS::StatListNode::execute(KJS::ExecState*) + 52
14  com.apple.JavaScriptCore  	0x95a55b24 KJS::CaseClauseNode::evalStatements(KJS::ExecState*) + 52
15  com.apple.JavaScriptCore  	0x95a55598 KJS::CaseBlockNode::evalBlock(KJS::ExecState*, KJS::Value 
const&) + 548
16  com.apple.JavaScriptCore  	0x95a55274 KJS::SwitchNode::execute(KJS::ExecState*) + 304
17  com.apple.JavaScriptCore  	0x95a32dc8 KJS::SourceElementsNode::execute(KJS::ExecState*) + 500
18  com.apple.JavaScriptCore  	0x95a32b64 KJS::BlockNode::execute(KJS::ExecState*) + 136
19  com.apple.JavaScriptCore  	0x95a3f260 KJS::DeclaredFunctionImp::execute(KJS::ExecState*) + 44
20  com.apple.JavaScriptCore  	0x95a3eb60 KJS::FunctionImp::call(KJS::ExecState*, KJS::Object&, KJS::List 
const&) + 428
21  com.apple.JavaScriptCore  	0x95a69cf4 KJS::Bindings::JSObject::call(_jstring*, _jobjectArray*) const + 
496
22  com.apple.JavaScriptCore  	0x95a69968 KJS::Bindings::JSObject::invoke
(KJS::Bindings::JSObjectCallContext*) + 316
23  com.apple.JavaScriptCore  	0x95a6afb4 KJS_JSObject_JSObjectCall + 48
24  com.apple.JavaPluginCocoa 	0x059b7688 nsToJavaString + 3740
25  com.apple.JavaPluginCocoa 	0x059b6cf8 nsToJavaString + 1292
26  com.apple.Foundation      	0x92890760 __NSFireMainThreadPerform + 276
27  com.apple.CoreFoundation  	0x9077c108 __CFRunLoopPerformPerform + 104
28  com.apple.CoreFoundation  	0x9074bc8c __CFRunLoopDoSources0 + 384
29  com.apple.CoreFoundation  	0x9074b1bc __CFRunLoopRun + 452
30  com.apple.CoreFoundation  	0x9074ac3c CFRunLoopRunSpecific + 268
31  com.apple.HIToolbox       	0x93129ac0 RunCurrentEventLoopInMode + 264
32  com.apple.HIToolbox       	0x931d66e0 HIMenuView::SimulateHitSelf(short, unsigned long, short*) 
+ 200
33  com.apple.HIToolbox       	0x931d65d0 HIStandardMenuView::SimulateHitSelf(short, unsigned long, 
short*) + 172
34  com.apple.HIToolbox       	0x9314b630 HIView::EventHandler(OpaqueEventHandlerCallRef*, 
OpaqueEventRef*, void*) + 1932
35  com.apple.HIToolbox       	0x931288d4 DispatchEventToHandlers(EventTargetRec*, 
OpaqueEventRef*, HandlerCallRec*) + 692
36  com.apple.HIToolbox       	0x9312802c SendEventToEventTargetInternal(OpaqueEventRef*, 
OpaqueEventTargetRef*, HandlerCallRec*) + 372
37  com.apple.HIToolbox       	0x93127ea8 SendEventToEventTargetWithOptions + 40
38  com.apple.HIToolbox       	0x931d64c8 SendControlSimulateHit(HIView*, short, unsigned long, 
short*) + 172
39  com.apple.HIToolbox       	0x932ae8b8 HIView::SimulateClickSelf(short, unsigned long, short*) + 
36
40  com.apple.HIToolbox       	0x9314b58c HIView::EventHandler(OpaqueEventHandlerCallRef*, 
OpaqueEventRef*, void*) + 1768
41  com.apple.HIToolbox       	0x931288d4 DispatchEventToHandlers(EventTargetRec*, 
OpaqueEventRef*, HandlerCallRec*) + 692
42  com.apple.HIToolbox       	0x9312802c SendEventToEventTargetInternal(OpaqueEventRef*, 
OpaqueEventTargetRef*, HandlerCallRec*) + 372
43  com.apple.HIToolbox       	0x93127ea8 SendEventToEventTargetWithOptions + 40
44  com.apple.HIToolbox       	0x931d63b8 HIView::SimulateClick(short, unsigned long, short*) + 192
45  com.apple.HIToolbox       	0x931d62dc HIViewSimulateClick + 60
46  com.apple.HIToolbox       	0x931d619c FlashFeedback(MenuSelectData*) + 96
47  com.apple.HIToolbox       	0x931d6100 SelectItemAndRestoreAllMenuBits(MenuSelectData&) + 436
48  com.apple.HIToolbox       	0x931c0dec TrackMenuCommon(MenuSelectData&, unsigned char*) + 
828
49  com.apple.HIToolbox       	0x931f9428 PopUpMenuSelectCore(MenuData*, Point, double, Point, 
GDevice**, Rect const*, unsigned short, unsigned long, Rect const*, Rect const*, __CFString const*, 
OpaqueMenuRef**, unsigned short*) + 312
50  com.apple.HIToolbox       	0x931f92a4 _HandlePopUpMenuSelection5 + 364
51  com.apple.AppKit          	0x93755ce0 _NSPopUpCarbonMenu2 + 2268
52  com.apple.AppKit          	0x937553f4 _NSPopUpCarbonMenu1 + 44
53  com.apple.AppKit          	0x937553b0 -[NSCarbonMenuImpl 
popUpMenu:atLocation:width:forView:withSelectedItem:withFont:] + 224
54  com.apple.AppKit          	0x93755050 -[NSPopUpButtonCell 
trackMouse:inRect:ofView:untilMouseUp:] + 1184
55  com.apple.WebCore         	0x95ca5348 -[KWQPopUpButtonCell 
trackMouse:inRect:ofView:untilMouseUp:] + 148
56  com.apple.AppKit          	0x936ec164 -[NSControl mouseDown:] + 536
57  com.apple.WebCore         	0x95ca5578 -[KWQPopUpButton mouseDown:] + 72
58  com.apple.WebCore         	0x95c495c8 KWQKHTMLPart::passWidgetMouseDownEventToWidget
(QWidget*) + 644
59  com.apple.WebCore         	0x95c1dfb8 KWQKHTMLPart::passWidgetMouseDownEventToWidget
(khtml::MouseEvent*) + 260
60  com.apple.WebCore         	0x95c1ddec KWQKHTMLPart::khtmlMousePressEvent
(khtml::MousePressEvent*) + 168
61  com.apple.WebCore         	0x95dc6c28 KParts::Part::event(QEvent*) + 28
62  com.apple.WebCore         	0x95c1d508 KHTMLView::viewportMousePressEvent(QMouseEvent*) + 
1008
63  com.apple.WebCore         	0x95c1cfd0 KWQKHTMLPart::mouseDown(NSEvent*) + 424
64  com.apple.WebKit          	0x9595c444 -[WebHTMLView mouseDown:] + 200
65  com.apple.WebCore         	0x95c495c8 KWQKHTMLPart::passWidgetMouseDownEventToWidget
(QWidget*) + 644
66  com.apple.WebCore         	0x95bfd10c KWQKHTMLPart::passSubframeEventToSubframe
(DOM::NodeImpl::MouseEvent&) + 208
67  com.apple.WebCore         	0x95c1d394 KHTMLView::viewportMousePressEvent(QMouseEvent*) + 
636
68  com.apple.WebCore         	0x95c1cfd0 KWQKHTMLPart::mouseDown(NSEvent*) + 424
69  com.apple.WebKit          	0x9595c444 -[WebHTMLView mouseDown:] + 200
70  com.apple.WebCore         	0x95c495c8 KWQKHTMLPart::passWidgetMouseDownEventToWidget
(QWidget*) + 644
71  com.apple.WebCore         	0x95bfd10c KWQKHTMLPart::passSubframeEventToSubframe
(DOM::NodeImpl::MouseEvent&) + 208
72  com.apple.WebCore         	0x95c1d394 KHTMLView::viewportMousePressEvent(QMouseEvent*) + 
636
73  com.apple.WebCore         	0x95c1cfd0 KWQKHTMLPart::mouseDown(NSEvent*) + 424
74  com.apple.WebKit          	0x9595c444 -[WebHTMLView mouseDown:] + 200
75  com.apple.WebCore         	0x95c495c8 KWQKHTMLPart::passWidgetMouseDownEventToWidget
(QWidget*) + 644
76  com.apple.WebCore         	0x95bfd10c KWQKHTMLPart::passSubframeEventToSubframe
(DOM::NodeImpl::MouseEvent&) + 208
77  com.apple.WebCore         	0x95c1d394 KHTMLView::viewportMousePressEvent(QMouseEvent*) + 
636
78  com.apple.WebCore         	0x95c1cfd0 KWQKHTMLPart::mouseDown(NSEvent*) + 424
79  com.apple.WebKit          	0x9595c444 -[WebHTMLView mouseDown:] + 200
80  com.apple.WebCore         	0x95c495c8 KWQKHTMLPart::passWidgetMouseDownEventToWidget
(QWidget*) + 644
81  com.apple.WebCore         	0x95bfd10c KWQKHTMLPart::passSubframeEventToSubframe
(DOM::NodeImpl::MouseEvent&) + 208
82  com.apple.WebCore         	0x95c1d394 KHTMLView::viewportMousePressEvent(QMouseEvent*) + 
636
83  com.apple.WebCore         	0x95c1cfd0 KWQKHTMLPart::mouseDown(NSEvent*) + 424
84  com.apple.WebKit          	0x9595c444 -[WebHTMLView mouseDown:] + 200
85  com.apple.AppKit          	0x9368d9c8 -[NSWindow sendEvent:] + 4616
86  com.apple.Safari          	0x0001d2d8 0x1000 + 115416
87  com.apple.AppKit          	0x93636bfc -[NSApplication sendEvent:] + 4172
88  com.apple.Safari          	0x0001a2b8 0x1000 + 103096
89  com.apple.AppKit          	0x9362e090 -[NSApplication run] + 508
90  com.apple.AppKit          	0x9371e8bc NSApplicationMain + 452
91  com.apple.Safari          	0x000021e8 0x1000 + 4584
92  com.apple.Safari          	0x00056e28 0x1000 + 351784

[Alexey Proskuryakov]

(In reply to comment #0)
> Relevant portion of crashdump attached. Happy to build/update/test/provide more info if it will help.

Please provide some way to reproduce the problem (even if inconsistently). Attaching (as a file) a complete 
crash log with all loaded binary images could also be useful, especially since this is JNI-related.

[Andrew Wright]

The most reproducible crasher involves selecting choices from 3 drop-down menus. The first two drop 
downs affect the next one along the row, selecting a row in the last menu calls an applet, which 
requests updated info from a server. Once this info is received, individual table cells are being updated 
with the new values (basically a stock price streaming/ticker solution).

It is particularly bad if you change a menu selection without waiting for the entire table to load & begin 
updating - seems like cutting things off during an update causes issues.

I can provide a login to our test system if needed but would rather not post it here - am happy to email 
someone direct though.

BTW: are there any preferences for where to test/take crashdumps - latest public Safari, or a nightly 
build?

[Andrew Wright]

Created attachment 4548 [details]
Full dump from latest public Safari

[Andrew Wright]

Some more info - the crash is consistently on this line:

    if ([view respondsToSelector:@selector
(webPlugInCallJava:isStatic:returnType:method:arguments:callingURL:exceptionDescription:)]) {

of dispatchJNICall in jni_objc.mm, and strangely also appears to be when calling lastIndexOf (although 
that may just be a side-effect of our site).

Adding a 'if (!view) return false' line to the above method made no difference.

Some sample output from the debug log, with a bit of extra debug. There are many successful calls, 
then the crash. There didn't seem to be any correlation between
whether a global ref had been used before or not (before having a method called on it), and a crash - 
happened with both new & previously gc'ed references.

/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:353 -- JObjectWrapper:  
new global ref 0xce2e224 for 0xce369fc
/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:353 -- JObjectWrapper:  
new global ref 0xce2e220 for 0xce36d48
/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:127 -- invokeMethod:  call 
lastIndexOf (Ljava/lang/String;)I on 0xce2e224
/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:130 -- invokeMethod:  orig 
str: [18:42:55:496]
/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:143 -- invokeMethod:  arg
[0] = [:]
/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:273 -- invokeMethod:  int 
result: [8]

...

/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:353 -- JObjectWrapper:  
new global ref 0xce30e94 for 0xce35cf0
/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:353 -- JObjectWrapper:  
new global ref 0xce30e90 for 0xce37a78
/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:127 -- invokeMethod:  call 
lastIndexOf (Ljava/lang/String;)I on 0xce30e94
/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:130 -- invokeMethod:  orig 
str: [13:52:02:531]
/Users/andrew/Code/WebKit/JavaScriptCore/bindings/jni/jni_instance.cpp:143 -- invokeMethod:  arg
[0] = [:]
[Switching to process 2313 local thread 0xf03]
Program received signal:  "EXC_BAD_ACCESS".

Suspect this means the problem is elsewhere :-/

[Andrew Wright]

Created attachment 4562 [details]
JVM crash at same time of Safari crash

[Andrew Wright]

Hadn't realised the JVM was crashing - attached sample log. Happens with 1.4.2 and 5.0 plugins. Possibly 
browser/plugin/jvm interacting badly?

[Andrew Wright]

JVM crash logged as radar 4325722

[Andrew Wright]

Symptoms become markedly worse using the Java5 VM - either update 3 publicly released, or the current 
DP from dev connection.

Unfortunately our site goes from 'bascially working' to 'badly broken' as a result. This is of particular 
concern as the next update will make Java 5 the default VM for users.

There is also a much easier way to reproduce the crash. Go to www.binarybet.com, and click on one of the 
'highlights' for events on that page. This will open a page which uses the applet in question. Prices should 
load into the applet - I only see the first price load (a single row) then updates will either stop or a crash 
will occur. If it doesn't crash immediately, navigating the popup menus will trigger one.

[Andrew Wright]

Created attachment 5060 [details]
JVM 5 dump

Running JVM Update 4 DP1

[Andrew Wright]

Created attachment 5061 [details]
Safari dump to go with Attachment 5060 [details]

[Geoffrey Garen]

Thanks for the great bug report(s). I'm on the case.

[Geoffrey Garen]

Andrew, 

Once again, thanks for the great bug report.

The basic problem here is that the Java plugin continues to execute code after Safari has told it to stop. 
Some of this code accesses plugin data structures that Safari/WebKit has destroyed.

I'm preparing a patch that should fix this in a future release. In the meantime, you can try to work 
around it. I see two options. 

(1) Reload the page. Currently, your drop-down menus do a form submission and load the data you 
want into an iframe. If you could load the whole page instead, (still using the form submission to tell 
the server what data you wanted on the new page), the new page would ignore the old plugin's errant 
messages.

(2) Remove Java calls to JavaScript; use JavaScript calls to Java instead. If I'm reading things right, your 
Java plugin periodically invokes javaScript methods like 'handleAppletTick' in order to update data. 
Instead, you could use a JavaScript method like setInterval, along with getElementById, to periodically 
call into Java and get the new data. (I'm not sure if this is practical on your end or not.)

Hope that helps.

[Geoffrey Garen]

Created attachment 5348 [details]
Fix

Never has a programmer worked so hard to produce so little code.

No layout test because the bug is difficult to reproduce.

[Maciej Stachowiak]

Comment on attachment 5348 [details]
Fix

r=me, I will grudingly accept the lack of test.

[Maciej Stachowiak]

Comment on attachment 5348 [details]
Fix

r=me, I will grudingly accept the lack of test.

[Geoffrey Garen]

Landed.