Is this a closure VM?

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

Hi Göran,

2009/8/19 Göran Krampe goran@krampe.se:

...

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

-- = = = ======================================================================== John M. McIntosh johnmci@smalltalkconsulting.com Twitter: squeaker68882 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com = = = ========================================================================

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:

Hi John,

On Wed, Aug 19, 2009 at 10:01 AM, John M McIntoshjohnmci@smalltalkconsulting.com 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

-- = = = ======================================================================== John M. McIntosh johnmci@smalltalkconsulting.com Twitter: squeaker68882 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com = = = ========================================================================

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

2009/8/19 Göran Krampe goran@krampe.se:

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