[Debichem-devel] Bug#1086272: gromacs: FTBFS with OpenMPI 5 on systems with less than 4 CPUs

[Santiago Vila]

retitle 1086272 gromacs: FTBFS with OpenMPI 5 on systems with less than 4 CPUs
severity 1086272 important
thanks

Hi. Thanks for finding a fix for this so quickly.

Nicholas Breen wrote:
> Looks like OpenMPI 5 has changed the options to allow oversubscription -
> but this should only affect running tests on single core/single thread
> hosts.

Well, just to clarify: I included a build log made in a single-cpu system
because sequential logs are usually easier to read and debug, but I also tried
on a system with 2 CPUs and it also failed.

In fact, the build log contains this:

28: There are not enough slots available in the system to satisfy the 4
28: slots that were requested by the application:

So any system with less than 4 CPUs will be also affected.

(I understand that you don't want this bug to be RC, but please do not
downgrade below important, this is the kind of bug which I usually try
to fix in stable (on the teams I am a member of), and for that the
RMs require the bug to be >= important).

While we are at it: Could you please fix a similar bug in the "elpa"
package? (both in trixie/sid, which fails in the same way as this
one because of OpenMPI 5, and in bookworm, for which the previous
fix in trixie did not reach propose-updates yet).

Thanks.