[Vm-dev] [urgent] Spur sometimes produces unloadable, image segments

[Max Leske]

Hi Dave, hi Eliot,

Thank you for responding so quickly!


> Hi Max,
>
> I'm not sure if it would help, but if there is a code snippet that
> reproduces the failure, I can try running it on Squeak to see if it
> gives any clues.

The images have lots of code in them and there are a couple of Pharo specifics in there, so I'm pretty sure we'd run into issues trying that. But thanks for the offer.

>
> Dave

> On Sat, May 30, 2020 at 07:00:29PM +0200, Max Leske wrote:
>
>>
>> Hi everyone,
>>
>> I desperately need some help with image segments in spur (Pharo). After having tested close to 1000 image segments that were written with a spur VM I have ended up with a couple that can't be loaded. I can reproducibly generate them from our old segements (Squeak 4 VM).
>>
>> Without full confidence in the image segment code I can't upgrade and couldn't upgrade those few instances anyway. I would be very thankful if someone had the time to look into this quickly.
>>
>> I will try to debug it myself but it will probably take me a couple of hours (if not days) until I have succeeded in building a VM that I can debug.
>>
>>
>> Cheers,
>> Max



> Hi Max,
>
>> On May 30, 2020, at 10:00 AM, Max Leske <maxleske at gmail.com> wrote:
>>
>> Hi everyone,
>>
>> I desperately need some help with image segments in spur (Pharo). After having tested close to 1000 image segments that were written with a spur VM I have ended up with a couple that can't be loaded. I can reproducibly generate them from our old segements (Squeak 4 VM).
>>
>> Without full confidence in the image segment code I can't upgrade and couldn't upgrade those few instances anyway. I would be very thankful if someone had the time to look into this quickly.
>>
>> I will try to debug it myself but it will probably take me a couple of hours (if not days) until I have succeeded in building a VM that I can debug.
>
> The only way this can be debugged realistically is to be able to simulate the reproducible case in the VMMaker simulator.  Is there any possibility of this working?  How big is the image in the reproducible case?  What facilities does it use (I’m guessing external imaging which as yet can’t be simulated).  If this can be made to occur headless and is hundreds of megabytes or less not gigabytes then we have a good chance of reproducing it in the simulator.
>

I don't know how big the image would be because we never save it. When we start the image, however, it's currently 54 MB.
I currently don't know the exact size but it's very likely somwhere between 100 and 200 MB.
The images run headless already when this happens.

I'm not sure what you mean by external imaging and facilities. Nothing fancy, at least not in the snapshot.

The only thing that correlates the failing instances is that their segments are bigger in comparison to thos of other instance, although not by too much.


Let me know what you need (or what I can do).

Thanks Eliot!

>>
>>
>> Cheers,
>> Max
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20200530/6da41cb1/attachment.sig>