During collective hdf5 reads, if the distributed axis is smaller than the number of processes, the I/O can hang on the processes with an empty slice. Possibly related to #165, but it could also be a separate issue.
I seem to remember running into (and resolving) this issue in the past, so it's possible that there was a regression at some point.
Needs further testing.
During collective
hdf5reads, if the distributed axis is smaller than the number of processes, the I/O can hang on the processes with an empty slice. Possibly related to #165, but it could also be a separate issue.I seem to remember running into (and resolving) this issue in the past, so it's possible that there was a regression at some point.
Needs further testing.