[PyKDE] pyqt/mysqld thread deadlock

[Phil Thompson]

On Wednesday 01 January 2003 7:55 pm, Kaleb Pederson wrote:
> On Wednesday 01 January 2003 11:40 am, you wrote:
> > On Wednesday 01 January 2003 5:58 pm, Kaleb Pederson wrote:
> > > -----BEGIN PGP SIGNED MESSAGE-----
> > > Hash: SHA1
>
> <snip>
>
> > > It turns out, however, that the thread locks in the middle of my run()
> > > call and never completes. I'm cross-posting this message as I'm not
> > > familiar with Python extension modules and I don't know if PyQt could
> > > have caused this, it is a Python bug (or feature) or if there is a
> > > problem in the mysqldb module code.
>
> <snip>
>
> > > So does anyone have any idea what's wrong or how to fix it? Please
> > > email me as well if you aren't posting to the PyQt/Python list (kibab
> > > at icehouse dot net).
> >
> > To quote the Python source the lock is "to prevent parallel imports of
> > the same module in different threads to return with a partially loaded
> > module". What other imports (or the same import in other threads) might
> > be happening at the same time?
>
> At this point in my application there is nothing else "NEW" going on.  That
> is, I'm in the middle of the event loop and this occurs in response to a
> button press.  The thread, of course, returns control to my application
> which then waits for more user input.  I can think of no reason anything
> else would be trying to import ListType, TupleType, or exc_info. 
> Furthermore, this is the first thread I spawn (although the behavior is
> similar for all subsequent threads) thus, I would hope there would be no
> other blocked threads.

It would be any import, not just those specific things.

> Phil, wouldn't the block cease once the "other" thread had successfully
> imported the module thereby unblocking my thread?  Never through the rest
> of my application life does that thread unblock.

Should do - but something obviously isn't right. If there was no other import 
going on then the first call to PyThread_acquire_lock() in lock_import() in 
import.c would succeed. It's not so it's blocking in the second call.

You could try instrumenting lock_import(), or replace the mysqldb calls with 
code that still does the imports but nothing else.

Phil

Phil