[Python-Dev] Who can make test_fork1 fail?
Sat, 19 Aug 2000 01:59:12 -0400
> Since nobody has made real progress on figuring out why test_fork1
> fails on some systems, would somebody who is able to make it fail
> please just try this patch & see what happens?
[Charles G Waldman]
> Or try this program (based on Neil's example), which will fail almost
> immediately unless you apply my patch:
Not "or", please, "both". Without understanding the problem in detail, we
have no idea how many bugs are lurking here. For example, Python allocates
at least two locks besides "the global lock", and "doing something" about
the latter alone may not help with all the failing test cases. Note too
that the pthread_atfork docs were discussed earlier, and neither Guido nor I
were able to dream up a scenario that accounted for the details of most
failures people *saw*: we both stumbled into another (and the same) failing
scenario, but it didn't match the stacktraces people posted (which showed
deadlocks/hangs in the *parent* thread; but at a fork, only the state of the
locks in the child "can" get screwed up). The patch you posted should plug
the "deadlock in the child" scenario we did understand, but that scenario
didn't appear to be relevant in most cases.
The more info the better, let's just be careful to test *everything* that
failed before writing this off.