VM Crash on Mac with Input Method ([Vm-dev] Re: [squeak-dev] RFB)

[Tobias Pape]

On 17.09.2016, at 00:25, John McIntosh <johnmci at smalltalkconsulting.com> wrote:

> >compiled with the toolchain for XCode 7.2.
> 
> So how are you doing the compile? 

I presume:

git clone https://github.com/OpenSmalltalk/opensmalltalk-vm.git
cd opensmalltalk-vm/build.macos32x86/squeak.cog.spur/
./mvm

?

> 
> 
> On Fri, Sep 16, 2016 at 3:13 PM, Yoshiki Ohshima <Yoshiki.Ohshima at acm.org> wrote:
> 
> Great!
> 
> The crash happens with the 32-bit VM shipped in the 5.1 all in one.  I
> can reproduce it the latest code of opensmalltalk-vm on github, and
> compiled with the toolchain for XCode 7.2.1.
> 
> 
> 
> On Fri, Sep 16, 2016 at 3:04 PM, John McIntosh
> <johnmci at smalltalkconsulting.com> wrote:
> >
> > Ok, we'll I"m not sure which version of the source code is being compiled here, or how this is being done (Xcode project, cmd line script?)
> >
> > First If the VM is compiled for 32bit it uses NON-ARC memory management, if it is compiled for 64bit then it should be ARC aware.
> >
> > The returnAndRemoveOldest is from the ancient Apple Queue.m/h which is ill named and returns for ARC a retain count of one, which then requires the RELEASEOBJ.
> >
> >
> > - (void) ioGetNextEvent: (sqInputEvent *) evt {
> >
> > ioProcessEvents();
> >
> > id event = [eventQueue returnAndRemoveOldest];
> >
> > if (event) {
> >
> > @autoreleasepool {
> >
> >             [self processAsOldEventOrComplexEvent: event placeIn: evt];
> >
> >         }
> >
> >         RELEASEOBJ(event);
> >
> > }
> >
> > }
> >
> >
> > The use of the autoreleasepool here is for the iOS platform where a more complex procedure occurs to return data from the touch data. For OSX it just drops the bytes (length sizeof(sqInputEvent))into the given sqInputEvent buffer for consumption by Smalltalk.
> >
> >
> > In general it just takes a possible object off the queue and pulls the data bytes into a C structure for consumption. Under non-ARC the object is retained within the Queue.m/h and has to be later released.  I think I should take one of the more standard GCD queue examples and replace the older logic in the near future.
> >
> >
> > - (void ) processAsOldEventOrComplexEvent: (id) event placeIn: (sqInputEvent *) evt {
> >
> > if ([event[0] intValue] == 1) {
> >
> > [(NSData *)event[1] getBytes: evt length: sizeof(sqInputEvent)];
> >
> > if (evt->type == EventTypeKeyboard) {
> >
> > // NSLog(@"keyboard pc %i cc %i uc %i m %i",((sqKeyboardEvent *)evt)->pressCode,((sqKeyboardEvent *) evt)->charCode,((sqKeyboardEvent *) evt)->utf32Code,((sqKeyboardEvent *) evt)->modifiers);
> >
> > }
> >
> > return;
> >
> > }
> >
> > }
> >
> >
> > As for the crash. it appears when we are doing the self interpretedKeyEvents
> >
> > -(void)keyDown:(NSEvent*)theEvent {
> >
> > keyBoardStrokeDetails *aKeyBoardStrokeDetails = AUTORELEASEOBJ([[keyBoardStrokeDetails alloc] init]);
> >
> > aKeyBoardStrokeDetails.keyCode = [theEvent keyCode];
> >
> > aKeyBoardStrokeDetails.modifierFlags = [theEvent modifierFlags];
> >
> > NSArray *down = @[theEvent];
> >
> > @synchronized(self) {
> >
> > lastSeenKeyBoardStrokeDetails = aKeyBoardStrokeDetails;
> >
> > [self interpretKeyEvents: down];
> >
> > self.lastSeenKeyBoardStrokeDetails = NULL;
> >
> > }
> >
> > }
> >
> >
> >
> > which triggers a failure via...
> >
> >>> 6   org.squeak.Squeak5.1.32.All-in-One 0x00130512 getCrashDumpFilenameInto + 82
> >>> 7   com.apple.CoreFoundation       0x98c2ddaf _CFAutoreleasePoolPop + 47
> >
> >
> > On saturday I"ll see if I can recreate it and see what might be causing it, mind nothing comes to mind at the moment.
> >
> >
> >
> > On Fri, Sep 16, 2016 at 10:16 AM, Tobias Pape <Das.Linux at gmx.de> wrote:
> >>
> >>
> >>
> >> On 16.09.2016, at 19:09, Yoshiki Ohshima <Yoshiki.Ohshima at acm.org> wrote:
> >>
> >> > On Fri, Sep 16, 2016 at 6:22 AM, Yoshiki Ohshima
> >> > <Yoshiki.Ohshima at acm.org> wrote:
> >> >> Would it be a feasible approach to try to use XCode (clang) Analyze?
> >> >
> >> > I got 9 errors from clang-checker.  One of those that looks most
> >> > suspicious is this one attached:
> >> >
> >> > Apparently just commenting the RELEASEOBJ line is not the right
> >> > thing... But does anybody has some ideas?
> >>
> >> Somewhere a RETAINOBJECT might be missing?
> >>
> >> Best regards
> >>         -Tobias
> >>
> >> >
> >> > --
> >> > -- Yoshiki
> >> > <event.png>
> >>
> >
> >
> >
> > --
> > ===========================================================================
> > John M. McIntosh. Corporate Smalltalk Consulting Ltd https://www.linkedin.com/in/smalltalk
> > ===========================================================================
> >
> 
> 
> 
> --
> -- Yoshiki
> 
> 
> 
> -- 
> ===========================================================================
> John M. McIntosh. Corporate Smalltalk Consulting Ltd https://www.linkedin.com/in/smalltalk
> ===========================================================================