[Vm-dev] [Cog] primitiveLoadImageSegment fails

[Eliot Miranda]

Hi Bert,

    (when you're back from Scotland, birthplace of my grandmother)

I've been looking at the failing ImageSegmentTest (actually vm-crashing
ImageSegmentTest) and I've made it work in the simulator.  I now want to
check whether the reading or the writing is at fault.  If in the real VM I
can load the segment written by the simulator I'll know its VM writing
that's at fault.  But I can't see how to crate an image segment to load a
specific image segment file.  Do you have a snippet handy?

On Wed, Apr 17, 2013 at 9:58 AM, Bert Freudenberg <bert at freudenbergs.de>wrote:

>
> Hi Eliot,
>
> thanks for looking into this.
>
> On 15.04.2013, at 18:02, Eliot Miranda <eliot.miranda at gmail.com> wrote:
>
> Hi Bert,
>
>     the version-compatibility check is failing.  In
> NewObjectMemory>>loadImageSegmentFrom:outPointers: there's the check
>         (self readableFormat: (data bitAnd: 16rFFFF "low 2 bytes"))
> ifFalse:
> and this is failing.  The version info in the segment is 6502, which is
> pre-Cog, BlockContext.  Cog expects either 6504 (closures, but big-endian
> order floats) or 6505 (closures, platform-order floats).  So one might hope
> that just munging the project file so that the version is 6504 will work,
> alas no.
>
> Adding e.g.
>
> ((segment at: 1) bitAnd: 16rFFFF) = 6502 ifTrue:
> [(UIManager default confirm: 'Image Segment is version 6502; expecting
> either 6504 or 6505.  OK to force it to 6504?') ifFalse:
>  [^self error: 'bad version'].
>  segment at: 1 put: ((segment at: 1) bitAnd: 16rffff0000) + 6504].
>  arrayOfRoots := self loadSegmentFrom: segment outPointers: outPointers.
>
> to comeUpFullyOnReload: fails because some objects ned up having a null
> class
>
>
> Really? How could that be? If it's just the BlockContexts and
> MethodContexts, the primitive could just replace them with nil.
>
> and this will just crash the image.  Alas, a) the image code doesn't
> return primitiveFailed if the load primitive does fail, and b) the failure
> code in the VM doesn't fail properly.  If invalid objects are created
> during an attempted load the failure code doesn't nuke them.
>
>
> Could this be done on the image side, or will it crash before?
>
> So there's more work to do here, a) to figure out how to convert a project
> file so that it's compatible with cog images, and b) to make the image
> segment loading primitive more robust when it fails.
>
>
> I really don't want to have to convert project files, but load them as
> they are.
>
> - Bert -
>
>
>
> On Thu, Apr 11, 2013 at 3:52 PM, Bert Freudenberg <bert at freudenbergs.de>wrote:
>
>>
>> That's primitive 99. With an updated Squeak trunk image I can load this
>> project fine in the interpreter:
>>
>>         http://etoys.squeak.org/svn/trunk/Etoys/Home.007.pr
>>
>> On Cog the primitive fails. Any idea what might be wrong?
>>
>> - Bert -
>>
>>
>
>
> --
> best,
> Eliot
>
>
>
>
>


-- 
best,
Eliot
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20130719/1e8d15b5/attachment.htm