On Mon, Jan 17, 2011 at 4:21 AM, Torsten Bergmann <astares@gmx.de> wrote:
Looks like a Cog issue, I was able to reproduce it:

When I download "PharoCore-1.2.zip" from
https://pharo-ic.lille.inria.fr/hudson/job/Pharo%20Core%201.2/126/

(which is a Pharo 1.2 Core #12303 image) and I directly start the image
�with the latest Win32 Cog VM from Eliot then I'm not able
�to go through update #12304 via the update mechanism
�as I already wrote in [1]. I get this error:

� => ClassBuilder(Object)>>doesNotUnderstand: #reshapeClass:toSuper:

Thanks,�Torsten!

Yes, the StackToRegisterMappingCogit gets confused by the initial assignment of nil in

oldClass subclasses do: [:oldSubclass | | newSubclass |
copyOfOldTraitComposition := copyOfOldClassTraitComposition := nil.

assuming the nil is a temp initialization rather than merely an operand, and hence gets confused about the stack layout (is off by one) and ends up sending�reshapeClass:toSuper: to the first temp (which is nil) instead of the receiver. �The fix is probably to duplicate the implementation of�BlockLocalTempCounter in the block scanning part of the code generator and hence to accurately count the number of temps in a block. �This will have to wait for the weekend :/

By the way it looks to me that making�copyOfOldTraitComposition & copyOfOldClassTraitComposition local temps in each of the do: blocks won't change the semantics and will remove the need to initialize them explicitly.

�

When I use the standard VM I'm able to update correctly
to the latest version pharo core 1.2 #12307

Maybe Eliot is able to reproduce and fix it. Until then I switch
to standard VM.

Thx
Torsten


[1] http://lists.gforge.inria.fr/pipermail/pharo-project/2011-January/039747.html
--
Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir
belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de