threading test locks up python on windows

[sgillies@…]

But passes on Linux.  In the windows case, mapserver was build with /MD and 
USE_THREAD.  Here is the test (tests/cases/threadtest.py 1.6):

import os, sys
import unittest
 
# the testing module helps us import the pre-installed mapscript
from testing import mapscript, TESTMAPFILE
 
def make_map(i):
    print "making map in thread %d" % (i)
    print mapscript.MS_ON
    po = mapscript.pointObj(1,1)
    print po
 
class MultipleThreadsTestCase(unittest.TestCase):
 
    def testLayerAdditionMultiThreads(self):
        """mapscripting in multiple threads"""
 
        import threading

        workers = []

        for i in range(10):
            thread = threading.Thread(target=make_map, args=(i,))
            workers.append(thread)
            thread.start()
 
        for thread in workers:
            print "waiting ... " + str(thread)
            thread.join()
            print str(thread) + " done."
        
# -----------------------------------------------------------------------------
if __name__ == '__main__':
    unittest.main()

the make_map() function is run by the 10 "worker" threads.  Curious that even
an operation as simple as init-ing a pointObj hangs.  Note that printing the
module variable mapscript.MS_ON works just fine by itself.  I also substituted
code from a non-mapscript module into make_map() and had no problems at all.

[fwarmerdam]

I'll take this one for now.  

[fwarmerdam]

As Sean noted, windows multithreading support was completely borked, and has
been since the beginning of time.  The initialization function was creating
the mutexes as "creator owned" and so no other thread was ever able to acquire
them.   It worked in single threaded mode, since multi-acquisition of a mutex
by the same thread is permitted on win32.  

The fix is comitted in the 4.5 stream.  If we are ever going to have a 4.4.3
the fix should be migrated back.   Just change the mutex creation from this:

    for( ; mutexes_initialized < TLOCK_STATIC_MAX; mutexes_initialized++ )
        mutex_locks[mutexes_initialized] = CreateMutex( NULL, TRUE, NULL );

to this:

    for( ; mutexes_initialized < TLOCK_STATIC_MAX; mutexes_initialized++ )
        mutex_locks[mutexes_initialized] = CreateMutex( NULL, FALSE, NULL );

in mapthread.c

I'm not closing till someone verifies the fix, and we discuss whether it 
needs to be backported.

[sgillies@…]

Sweet.  I'll test and have a report this morning.

[sgillies@…]

Frank, thread initialization seems to be fixed.  I committed 2 new tests 
(threadtest.py 1.7).  One triggers exceptions in all worker threads, the other
triggers an exception in one thread and makes valid requests in subsequent 
threads as a simple-minded effort to check that exceptions are contained to one
thread.  All the tests pass on windows and linux with USE_THREAD.

Do we need a 4.4.3 release?  I don't know.  While the number of multi-threading
users is growing, it's much smaller than the traditional CGI users.  Cutting a
4.4.3 release that most users could ignore might cause some confudion.  It's
more important that those of us distributing 4.4.x or 4.5 windows mapscript 
binaries make the fix and update our binaries.

[fwarmerdam]

Closing ... I don't think there will be a 4.4.3 now that we are into the 4.6
betas.