On 2020-04-06 20:23, Gilles Filippini wrote:
Drew Parsons a écrit le 06/04/2020 à 05:08 :
Do we want to use this as a solution? Or would you prefer an
environment
variable that h5py can check to allow mpi invocation on a serial
process?
I let this decision up to you.
I think it's reasonable to give h5py an environment variable to work
with, since the same problem might occur for other applications
(including user apps). It's conceivable some users might want their hdf5
(h5py) apps to build against mpi, even if also used in serial mode.
Whatever you choose it deserve a bit fat
note in README.Debian.
Certainly.
Note that this means bitshuffle as built now is expressly tied in with
hdf5-mpi and h5py-mpi (this seems intentional by debian/rules and
debian/control, though the Build-Depends must be updated to
python3-h5py-mpi). It's a separate question whether it's desirable to
also support a hdf5-serial build of bitshuffle. Likewise we need to
think about what we want to happen when bitshuffle is invoked in a
serial process.
I'll let that to the bitshuffle maintainer.
More puzzles for Thorsten to spend the weekends on :)
I'll propose a patch to fix
the current FTBFS, sticking on the mpi flavour to be conservative vs
bitshuffle's previous builds.
Thanks for your help.
Drew
