Re: [Pyobjc-dev] threading

[Bob I. <bo...@re...>]

On Mar 30, 2005, at 2:34 PM, Ronald Oussoren wrote:

>
> On 30-mrt-05, at 21:02, Bob Ippolito wrote:
>> The *real* problem is that you're calling NSLog with str objects, not 
>> unicode objects, and it looks like it may be triggering some kind of 
>> GIL issue in PyObjC's type conversion stuff, which I'll look into 
>> later this week but...  NEVER USE STR OBJECTS WHEN CROSSING THE 
>> BRIDGE.
> <mumbles a bit>
>
> Be happy that he does use strings, this uncovered a bug in the bridge 
> ;-)
>
> The reason you shouldn't use str objects is internationalisation: your 
> strings are converted to unicode when the cross the bridge. This may 
> work for you now, but could suddenly fail when someone passes text 
> with accents into your program.

Further more, because Python uses str to represent text *and* data, it 
makes life hard on PyObjC as Foundation has different classes to 
represent each.  It is currently impossible to use a str type to 
represent data across the bridge, and it's not a good idea to use them 
to represent text.  So, you shouldn't use them at all whenever humanly 
possible in the context of a PyObjC application.

> According to gdb ThreadTest.py blocks in thread 2 because NSLog wants 
> to grab the GIL and in thread 1 because it also wants to fetch the GIL 
> somewhere in a function named 'inittime'. That's very odd, the 
> function name suggests the init function of the time extension, and 
> that shouldn't be called at this time.
>
> Debugging threading related bugs can be lots of fun ;-).
>
> I'm currently building python-2.4.1 with debugging hooks turned on, 
> that should make debugging slightly easier.

Except for the fact that it will work using 2.4.1 ;)

-bob