Problems with latest VM

[Lee Salzman]

The problem is caused by our conservative GC which scans the C stack to
ensure anything pointing to it is not GC'd. VC is bungling up references
to objects on the stack with optimizations somehow. What is needed is to
find the specific object that is being inappropriately GC'd despite
being referenced on the C stack and to guard it with rootStackPush.

Lee

On Sun, Aug 08, 2004 at 02:39:11PM +0200, Jaco van der Merwe wrote:
> Hi all,
> 
> I've just downloaded the latest VM and slate sources. I compiled the VM with
> Visual C++. I made both a debug and a release build. The debug build VM
> seems to work (at least it loads the image without problems). However, the
> release build fails fairly early while loading the image with an unhandled
> signal message, after which the VM exits.
> 
> The fact that the debug and release builds behave differently still
> indicates to me that the VM code accesses memory somewhere that is
> uninitialised, or not intended to be accessed. Therefore, although the debug
> build seems to run, it may be hiding the problem or just postponing it to
> many instructions later. I'm not sure how to proceed with debugging the
> unhandled signal case mostly because I don't understand the underlying
> mechanisms used in the vm. I'll appreciate it if someone can explain it to
> me. Then, maybe I'll be able to debug this case.
> 
> Also, when compiling the sources the compiler generates the following
> warning which seems suspicious to me:
> 
>    vm.c(3021) : warning C4047: '=' : 'ObjectPointer *' differs in levels of
> indirection from 'ObjectPointer (*)[]'
> 
> The offending line of code is as follows:
> 
>    vars = &lexicalContext->variables;
> 
> where vars and lexicalContext are defined as follows:
> 
>    ObjectPointer * vars;
>    struct LexicalContext * lexicalContext;
> 
> and LexicalContext is as follows:
> 
>    struct LexicalContext {
>       struct ObjectHeader header;
>       struct Map * map;
>       ObjectPointer traits;
>       ObjectPointer unwindBlock;
>       ObjectPointer variables[0];
>    };
> 
> This does not seem correct to me. Shouldn't the offending line of code read
> as follows?
> 
>    vars = lexicalContext->variables;
> 
> Also, the following suspicious warning still remains. This has previously
> been reported by me and other people.
> 
>    file.c(136) : warning C4146: unary minus operator applied to unsigned
> type, result still unsigned
> 
> The offending line of code is:
> 
>    PSObjectHeap_adjustAllOopsBy_ (CurrentMemory, - (unsigned)
> CurrentMemory -> memory);
> 
> where the function signature is:
> 
>    unsigned long int PSObjectHeap_adjustAllOopsBy_(struct ObjectHeap * h,
> unsigned long int shiftAmountInBytes)
> 
> Why give an explicit negative argument to a formal argument that is
> unsigned?
> 
> Regards
> 
> Jaco van der Merwe
>