[Python-Dev] second opinion for proposed fix for bug #754449

[Brett C.]

Guido van Rossum wrote:

<SNIP>
>>Basically the exception stems from a call to threading.currentThread in 
>>the _Condition class (which is just Condition) which, according to the 
>>comment along with the call, is just for its side-effect.  Well, I 
>>looked and I can't find the need for the side-effect.  I took out the 
>>code and ran the regression tests and nothing failed because of it.
>>
>>Since the code says the call is for the explicit side-effect I just 
>>wanted another pair of eyes to quickly scan the code in threading.py and 
>>either agree with me or tell me I am blind and point out where the 
>>side-effect is needed for Condition instances.  The offending 
>>currentThread calls can be found on lines 196 and 238 (can also just 
>>search for "side-effect" to find them).
> 
> 
> This is my code.  It was long ago.  I think the intended side effect
> is creating a dummy thread *IF* the calling thread was not created by
> the threading module but by thread.start_new_thread().  The current
> thread in turn is only relevant when the Condition's lock is an RLock.
> But the only thing that is ever done with that is comparing it with
> the outcome of another call to currentThread().

Right.  And RLock makes the call itself so it seems RLock on its own 
will trigger the side-effect when it needs it.

>  So I *THINK* that the
> side effect isn't particularly needed.
> 

OK.  That's good enough for me.

> (But the threading module references other globals, so I'm not sure
> that ripping out the currentThread() calls is really enough, except in
> this particular app's case.)
> 

Yeah.  To force this problem (couldn't trigger it) I set everything to 
None in threading and then tried this.  It triggered the right error, 
but the function registered with atexit also threw a fit.  But that is a 
forced error since atexit will do its thing before teardown, right?

Regardless, I will try to go through the code and see if there are any 
other points during the shutdown the reference a global and see if there 
is a way to solve it cleanly.  Otherwise I guess people just need to 
shutdown there threading instances properly.

-Brett