Fix wait() in @parallel

[Jeroen Demeyer]

The new doctesting framework sometimes gets:

sage -t --long devel/sage/sage/homology/simplicial_complex.py
**********************************************************************
File "devel/sage/sage/homology/simplicial_complex.py", line 2188, in sage.homology.simplicial_complex.SimplicialComplex.is_cohen_macaulay
Failed example:
    S.is_cohen_macaulay(ncpus=3)
Expected:
    False
Got:
    Exception OSError: (10, 'No child processes') in <generator object __call__ at 0x646f1e0> ignored
    False
**********************************************************************

Fix this.

[David Roe]

Looks good. You didn't add a doctest: is there one that consistently fails in the new framework?

[Jeroen Demeyer]

Replying to roed:

is there one that consistently fails in the new framework?

To be honest, I cannot say for sure why the problem occurs.

I'm fairly certain the problem is the big gap between

pid = os.wait()[0]

and

del workers[pid]

If some exception would occur in between, then workers[pid] would still be in the list of workers, even though we saw it exit. However, I cannot really explain where the exception could come from (can yield raise exceptions?)

[David Roe]

How consistently is this error occurring in the new framework? Every time? Do the fixes here make the problem disappear? If so, I'm willing to give this a positive review, since there is a doctest failing that will no longer fail.

[Jeroen Demeyer]

Replying to roed:

How consistently is this error occurring in the new framework? Every time?

Not every time, but it's still reasonably common that I have seen it multiple times.

[Jeroen Demeyer]

Added doctest which always shows the OSError without this patch.

[David Roe]

You should fix the unpickling after breaking it. Add

sage: from sage.structure.sage_object import unpickle_override
sage: del unpickle_override[('sage.rings.polynomial.polynomial_rational_flint', 'Polynomial_rational_flint')]

after the end of the TESTS block. Otherwise it looks good.

[Jeroen Demeyer]

Good point :-)

[David Roe]

Great.

[Volker Braun]

Possibly related: #15585