> but if you make a blocking call such as in the following program,
> you'll get a hang

Yeah, and for that we could use the same approach as for the maps.

But, I've just implemented the accept callback approach, which should be superior. Maps/Apply fails instantly as soon as a worker process crashes, but the pool remains fully functional. Patch  multiprocessing-trunk@82502-termination-trackjobs.patch added.

There seems to be some race conditions left, because some of the tests breaks from time to time. Maybe you can pinpoint it before me.