Bug#946986: python3-h5py: h5py does not play nicely with subprocess and mpirun

[Drew Parsons]

On 2020-01-22 20:19, Rafael Laboissière wrote:
> * Drew Parsons <dparsons at debian.org> [2020-01-15 22:25]:
> 
>> This is one very strange bug.  Why would the mere fact that h5py is 
>> loaded affect what subprocess does?
> 
> Yes, this is completely mysterious.  However, the bug is real.
> 
> It is probably an upstream problem.  I would love to forward the
> present report upstream, but I have no idea to where send it.
> 
> I am inclined to raise the severity of this bug report from
> “important” to “serious”, because it is blocking Bug#938925, which has
> severity “serious” and is preventing xmds2 to enter testing.


h5py have done some work on their mpi support (cf. 
https://github.com/h5py/h5py/issues/1364).  They've created pytest-mpi 
to help with tests. It's currently in the debian NEW queue.

Thibaut's idea in bug#944769 to ignore mpi if there's only one process 
will be relatively straightforward to implement I think (I've had other 
tasks to finish before I get onto that). I guess that will fix (or at 
least workaround) this problem.

Might be worth bringing to the attention of the subprocess developers. I 
figure they'll either be interested to know about the problem or will 
have a simple explanation.

Drew