On Wed, Jan 11, 2017 at 09:03:53PM -0500, David T. Lewis wrote:
> On Wed, Jan 11, 2017 at 05:10:24PM +0100, Bert Freudenberg wrote:
> > On Tue, Jan 10, 2017 at 3:13 PM, David T. Lewis <lewis at mail.msen.com> wrote:
> > 
> > >
> > > Thanks Bert. It appears to be something related to MethodContext. Since it
> > > is VM related, I initially guessed that it might be something related to
> > > the method cache, but I was able to disable that in the VM and the symptoms
> > > did not change. I'm thinking now that it might be a failure in primitive 61
> > > when trying to copy a MethodContext, this based on the debugger stack
> > > (although I cannot actually display anything useful in the debugger).
> > >
> > > I'll report back if I end up figuring it out, meanwhile any and all
> > > historical
> > > clues are appreciated :-)
> > >
> > 
> > Ah, slowly all the hacks I had to put in are coming back to me ;)
> > 
> > Some early images fill the context stack before advancing its stack
> > pointer. I have a flag to allow that, it's pretty certainly used in
> > primitive 61. Normally the VM does not allow access beyond the SP because
> > there is garbage there (stack pops do not nil out the context slot):
> > 
> > https://github.com/bertfreudenberg/SqueakJS/search?q=allowAccessBeyondSP
> > 
> > But since the regular VM does not allow it, no image (except the really old
> > ones) ever does it, so I just leave the flag enabled whenever "oldPrims" is
> > in effect ;) Would be better if we could come up with a better way to
> > identify these images.
> > 
> 
> Brilliant, thanks Bert.
> 
> I see several places where we check "fmt = 3 and: [self isContextHeader: hdr]",
> including in #stObject:at:put: which is probably the immediate cause of the
> failure that I was seeing. And #stSizeOf: does this:
> 
> 	(fmt = 3 and: [self isContextHeader: hdr])
> 		ifTrue: [stSize := self fetchStackPointerOf: array]
> 		ifFalse: [stSize := totalLength - fixedFields].
> 
> I'll try playing with it this weekend and see if I can get it working for the
> older images.
> 
> But I have to ask ... how in the world did you figure this out?
> 
> :-)
> 

Indeed, disabling the stack pointer check in #stObject:at:put: results in
a VM that now works with the old Squeak1.13u.image.

Dave