Re: [Pyobjc-dev] crasher demo app

[Ronald O. <ous...@ci...>]

On 26-mrt-04, at 14:55, Bob Ippolito wrote:

>
> On Mar 26, 2004, at 1:54 AM, Ronald Oussoren wrote:
>
>>
>> On 26-mrt-04, at 0:03, Bob Ippolito wrote:
>>
>>>
>>> On Mar 25, 2004, at 4:23 PM, Ronald Oussoren wrote:
>>>
>>>>
>>>> On 25-mrt-04, at 22:08, Burris T. Ewell wrote:
>>>>
>>>>> First off, I appreciate all of the help here...
>>>>>
>>>>> On Mar 25, 2004, at 12:19 PM, Ronald Oussoren wrote:
>>>>>> If I remove those calls the application runs longer, e.g. it 
>>>>>> didn't crash yet :-)
>>>>>
>>>>> I don't think that's it...  I tried that, holding a reference to 
>>>>> the cell, and also retaining the application delegate that holds 
>>>>> references to the controller objects.  Still, inside the cell, the 
>>>>> controller attribute disappears inside drawInteriorWithFrame.  
>>>>> Here is a traceback from the full application, which is pretty 
>>>>> much the same as the demonstration.  This shouldn't be possible 
>>>>> since the cell sets self.controller inside of init() and nowhere 
>>>>> else is the attribute removed.
>>>>
>>>> Hmm, doesn't NSTableView use -copy? That might explain why 
>>>> self.controller disappears.
>>>>
>>>> There's definitly something fishy going on,
>>>
>>> There's fishy things with threads in PyObjC in general.. try this, 
>>> for example (I just ran into this bug today):
>>>
>>> http://undefined.org/~bob/PyURLProtocol_broken.tgz
>>>
>>> My guess is that it doesn't release the GIL when doing the call, so 
>>> when it gets the callback it can't acquire the GIL and deadlocks.
>>
>> Yup, we don't release the GIL when calling into Objective-C, and we 
>> get a callback on another thread here.
>>
>> This won't be completely solved before the 1.1 release, I'd prefer to 
>> get that out soon (we should have released that long ago). Doing a 
>> 90% solution should be possible by releasing the GIL when calling 
>> ffi_call (libffi_support.m).
>>
>> That won't solve the problems with mcfoo.
>
> I spent a few minutes playing with that, but it gets complicated 
> really fast and about 24 of the tests start crashing or deadlocking ;)

I noticed the crashing/deadlocking part :( and I wonder if these have 
anything to do with the problems we're seeing with mcfoo, to be honest 
I'm afraid they are related.

Just releaseing the GIL around calls to ffi_call is not too 
complicated, it adds less then 20 lines of code to libffi_support.m.  
More fine-grained releasing of the GIL will be a lot of work.

Ronald
>
> -bob
>
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by: IBM Linux Tutorials
> Free Linux tutorial presented by Daniel Robbins, President and CEO of
> GenToo technologies. Learn everything from fundamentals to system
> administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
> _______________________________________________
> Pyobjc-dev mailing list
> Pyo...@li...
> https://lists.sourceforge.net/lists/listinfo/pyobjc-dev
>
>
--
X|support bv            http://www.xsupport.nl/
T:  +31 610271479       F:  +31 204416173