3.10-6 shaky?

 
Hey!

I am prepping a Gjallar demo for later today and had to back down to
3.10-4 due to segfaults etc. Not fully certified yet, but have anyone
else seen 3.10-6 go boom more than "usual"? :)

regards, Göran

Hi Göran,

2009/8/19 Göran Krampe <[hidden email]>:
> I am prepping a Gjallar demo for later today and had to back down to 3.10-4
> due to segfaults etc. Not fully certified yet, but have anyone else seen
> 3.10-6 go boom more than "usual"? :)

you mean the Linux VM, right?

I have observed a crash when running all the tests. I could not track
it down to a particular test because when I ran the tests in small
batches the crash did not occur.

Note that this time my hard disk is not full. ;-)

Best,

Michael

Is this a closure VM?

On 19-Aug-09, at 12:59 AM, Michael Haupt wrote:

--
=
=
=
========================================================================
John M. McIntosh <[hidden email]>   Twitter:  
squeaker68882
Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
=
=
=
========================================================================

 
Hi John,

On Wed, Aug 19, 2009 at 10:01 AM, John M
McIntosh<[hidden email]> wrote:
> Is this a closure VM?

it seems it is. (Eliot's factorial example produces a sensible
result.) I don't have the time to look at the code right now, though.

Best,

Michael

 
There is a known problem where unbalanced primitive calls, where the  
logic pop/push variables in an unbalanced manner
as it passes variables to primitive calls will cause the VM to crash.

This was identified in eToys where it would make calls to check  
sandboxing permissions.
Normally those calls aren't made in the squeak vm, but perhaps others  
exist.

What happens is the unbalanced pop leaves garbage on the stack where  
in the past that slot was not used, but now it's
used for the block closure logic, so when the method returns it  
attempts to use that slot data, and BOOM.

The crash happens on the return from a method call where the method is  
invoking a primitive. The trick then is to find
which method is being called.


Of course this all presumes that is the issue, and not some other  
problem, since I've not built a VM from the latest
code base for a few months now and there has been a few changes which  
I've not eyeballed.


On 19-Aug-09, at 1:08 AM, Michael Haupt wrote:

--
=
=
=
========================================================================
John M. McIntosh <[hidden email]>   Twitter:  
squeaker68882
Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
=
=
=
========================================================================

 
Hi!

John M McIntosh wrote:
> There is a known problem where unbalanced primitive calls, where the
> logic pop/push variables in an unbalanced manner
> as it passes variables to primitive calls will cause the VM to crash.

Ok, I will take a closer look in a day or two and check the logs etc.

regards, Göran

I bet this is because of UUID plugin, not because closures :)

2009/8/19 Göran Krampe <[hidden email]>:


--
Best regards,
Igor Stasenko AKA sig.