Strange problem when using re module with threads

[Gabriel Genellina]

En Thu, 01 Nov 2007 17:01:36 -0300, Wei Lee Woon <wlwoon at gmail.com>  
escribió:

> I've been getting a rather strange problem with the following  
> multithreaded
> code (reduced to the minimum which still results in the problem):
>
> import threading
> import re
>
> class hey(threading.Thread):
>     def run(self):
>         print re.compile("\d+").search("hey95you").group();
>
> thlist=[]
> for tech in range(2):
>     thlist.append(hey())
>     thlist[-1].start()
>
> for th in thlist:
>     th.join()
>
> After saving this to a file (say "test.py"), if I try to run this from  
> the
> console using "python test.py", it seems to work fine, but when i try to  
> run
> it from the python interactive shell using "import test", it freezes up  
> (if
> i don't issue the join() it is fine, though). Any ideas why this is so?

It's the "import lock" in action.
When an import is being made, no other thread can import anything, they  
block waiting for the lock to be released. re.compile internally tries to  
import sre_parse, and blocks. Then both threads are waiting for the import  
lock, and the main thread is waiting in the join() call [inside the  
"import test" call] - a deadlock.
Note that you get the same effect if you replace the re.compile(...) with  
a simple "import os"

So, don't do blocking actions (like join()) at import time.

-- 
Gabriel Genellina