Hooray for Chris and Nicolas!

frank

On 17 October 2013 01:45, Chris Muller <asqueaker at gmail.com> wrote:
> On Wed, Oct 16, 2013 at 4:07 PM, Nicolas Cellier
> <nicolas.cellier.aka.nice at gmail.com> wrote:
>> To reproduce, just load System-cmm.602 from a MC browser and it quits quite
>> fast...
>> From assert cog vm, here is the end of the report:
>>
>> **IncrementalGC**
>> **FullGC**
>>
>> stack page bytes 2048 available headroom 1252 minimum unused headroom 724
>>
>>     (sweep failed to find exact end of memory)
>> Abort trap
>>
>> and the beginning is:
>>
>> sweep failed to find exact end of memory
>>
>> Squeak VM version: 4.0 4.0.2778 Mac OS X built on Aug  8 2013 07:43:35
>> Compiler: 4.2.1 (Apple Inc. build 5666) (dot 3) [Assert VM]
>> Built from: CoInterpreter * VMMaker.oscog-nice.336 uuid:
>> 409e5084-5ffa-466c-a844-2473662c1ebf Sep  8 2013
>> With: StackToRegisterMappingCogit * VMMaker.oscog-nice.336 uuid:
>> 409e5084-5ffa-466c-a844-2473662c1ebf Sep  8 2013
>> Revision: VM: r2778 http://squeakvm.org/svn/squeak/branches/Cog
>> Plugins: r2545 http://squeakvm.org/svn/squeak/trunk/platforms/Cross/plugins
>>
>>
>> C stack backtrace:
>> 0   Squeak                              0x0004deec reportStackState + 147
>> 1   Squeak                              0x0004e26b error + 31
>> 2   Squeak                              0x000b43ec fullGC + 897
>> 3   Squeak                              0x000b4cd5 sufficientSpaceAfterGC +
>> 64
>> 4   Squeak                              0x000b6094 primitiveNewWithArg + 147
>> 5   ???                                 0x11acc789 0x0 + 296535945
>> 6   Squeak                              0x000c9795 interpret + 32627
>> 7   Squeak                              0x000429dd EventLoopEventHandler +
>> 28
>> 8   HIToolbox                           0x911eac2f
>> _ZL23DispatchEventToHandlersP14EventTargetRecP14OpaqueEventRefP14HandlerCallRec
>> + 1567
>> 9   HIToolbox                           0x911e9ef6
>> _ZL30SendEventToEventTargetInternalP14OpaqueEventRefP20OpaqueEventTargetRefP14HandlerCallRec
>> + 411
>> 10  HIToolbox                           0x911e9d55
>> SendEventToEventTargetWithOptions + 58
>> 11  HIToolbox                           0x9121ea24
>> _ZL29ToolboxEventDispatcherHandlerP25OpaqueEventHandlerCallRefP14OpaqueEventRefPv
>> + 3006
>> 12  HIToolbox                           0x911eb080
>> _ZL23DispatchEventToHandlersP14EventTargetRecP14OpaqueEventRefP14HandlerCallRec
>> + 2672
>> 13  HIToolbox                           0x911e9ef6
>> _ZL30SendEventToEventTargetInternalP14OpaqueEventRefP20OpaqueEventTargetRefP14HandlerCallRec
>> + 411
>> 14  HIToolbox                           0x9120c7f3 SendEventToEventTarget +
>> 52
>> 15  HIToolbox                           0x91395c17 ToolboxEventDispatcher +
>> 86
>> 16  HIToolbox                           0x91395d4f RunApplicationEventLoop +
>> 243
>> 17  Squeak                              0x00040e61
>> RunApplicationEventLoopWithSqueak + 185
>> 18  Squeak                              0x0004de37 main + 1079
>> 19  Squeak                              0x000027be start + 54
>>
>>
>> Smalltalk stack dump:
>> 0xbffeb058 M Array(SequenceableCollection)>select: 0x13fe1d04: a(n) Array
>> 0xbffeb074 M MCFileRepositoryInspector>versionNamesForSelectedPackage
>> 0x135d28b0: a(n) MCFileRepositoryInspector
>> 0xbffeb094 M MCFileRepositoryInspector(MCRepositoryInspector)>versionList
>> 0x135d28b0: a(n) MCFileRepositoryInspector
>>
>> Squeak stack has no importance, I just clicked somewhere in a MC trunk repo
>> inspector...
>>
>> It sounds like memory corruption...
>>
>> What it interesting is that loading this mcz also blow a 4.10.10 interpreter
>> VM...
>>
>> I just fail to see what could cause such a violent ... Ah WAIT WAIT WAIT:
>>
>> MCPackageLoader>>basicLoad invoke RecentMessages default suspendWhile: [ ]
>> suspendWhile is using the last instance variable... which is being shifted
>> inside the suspendWhile: []...
>> So when we're back from the block, it's not long before die...
>
> Ahh!!  THANK YOU and compliments for an awesome diagnosis!  I was
> leisurely digging for a couple of hours today, I was stuck thinking my
> changes were benign.  My image updated fine after I committed it.
>
> I guess I found a tricky land mine!
>
>> At least i'm happy it's not related to CharacterScanner stuff.
>
> My bad.  :)  I'll post a fix.
>