Hi Tim,

On Tue, May 12, 2015 at 7:59 AM, Tim Olson <tim.olson.mail at gmail.com> wrote:

>
> On May 12, 2015, at 3:10 AM, marcel.taeumel <Marcel.Taeumel at hpi.de> wrote:
>
> > I think that's only a symptom. Why else is the bug related to different
> VM
> > versions?
>
> I ran into this and various variants of it a little while ago, when I
> tried to update my somewhat outdated image using the latest Cog VM.  When I
> ran using an older interpreter VM, the build went fine.
>
> In tracking down one of the MNU failures I was seeing with the Cog build,
> I saw this:
>
> SmallInteger(Object) >> doesNotUnderstand: #isEmptyOrNil
> PluggableListMorphOfMany(PluggableListMorph) >> hasFilter
>
> where:
>
> hasFilter
>         ^ lastKeystrokes isEmptyOrNil not
>
> the debugger in the context of hasFilter says that lastKeystrokes is an
> empty ByteString, but when I click on self for the SmallInteger DNU, it
> says that the receiver is a SmallInteger 0.
>
> Looking at the object context of PluggableListMorphOfMany, I saw that the
> only SmallInteger 0 value was the instance variable “lastKeystrokeTime”
> just before “lastKeystrokes”.
>
> So it looks to me like there may be a (rare) instance variable to register
> mapping bug in the latest Cog VMs.
>

Yes, very nearly.  In fact it was a bug in new register allocation code for
compiling #==.  It is fixed in the r3343 VMs.  Sorry for the inconvenience.
 http://www.mirandabanda.org/files/Cog/VM/VM.r3343


>         — tim
>
>
>


-- 
best,
Eliot
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20150515/8cb36de4/attachment.htm