I'm thinking about trying to sneak in openmpi 3.1.3 before the f30 branch. My COPR rebuilds have been looking pretty good lately so I hope it will go fairly smoothly. Does anyone have any objections? I'm planning on starting tomorrow if not.
- Orion
On Wednesday, 13 February 2019 at 06:13, Orion Poplawski wrote:
I'm thinking about trying to sneak in openmpi 3.1.3 before the f30 branch. My COPR rebuilds have been looking pretty good lately so I hope it will go fairly smoothly. Does anyone have any objections? I'm planning on starting tomorrow if not.
Why not 4.0.0? It was released in September 2018 and it's supposedly ABI compatible with 3.x: https://www.mail-archive.com/announce@lists.open-mpi.org/msg00119.html
It's a big jump from the currently packaged 2.1.6. Any idea why we are so behind upstream? 3.0.0 was released in September 2017, 3.1.0 in May 2018. 3.0.0 should have been in F27 or F28 and 3.1.0 in F29 latest.
Also, they say 2.1.6 is the last release in 2.1.x series, so I'd say an update to a supported version is a must.
Regards, Dominik
On 2/13/19 3:56 AM, Dominik 'Rathann' Mierzejewski wrote:
On Wednesday, 13 February 2019 at 06:13, Orion Poplawski wrote:
I'm thinking about trying to sneak in openmpi 3.1.3 before the f30 branch. My COPR rebuilds have been looking pretty good lately so I hope it will go fairly smoothly. Does anyone have any objections? I'm planning on starting tomorrow if not.
Why not 4.0.0? It was released in September 2018 and it's supposedly ABI compatible with 3.x: https://www.mail-archive.com/announce@lists.open-mpi.org/msg00119.html
Because it's still relatively new. I've hit enough problems with major bumps in openmpi to be a little wary. And I don't have the COPR testing record for it. We are doing this last minute so I'm trying to be a little more careful.
It's a big jump from the currently packaged 2.1.6. Any idea why we are so behind upstream? 3.0.0 was released in September 2017, 3.1.0 in May 2018. 3.0.0 should have been in F27 or F28 and 3.1.0 in F29 latest.
Because no one else has stepped up to help maintain it and I'm stretched too thin.
Also, they say 2.1.6 is the last release in 2.1.x series, so I'd say an update to a supported version is a must.
I concur as well.
On 2/13/19 5:05 PM, Orion Poplawski wrote:
On 2/13/19 3:56 AM, Dominik 'Rathann' Mierzejewski wrote:
On Wednesday, 13 February 2019 at 06:13, Orion Poplawski wrote:
I'm thinking about trying to sneak in openmpi 3.1.3 before the f30 branch. My COPR rebuilds have been looking pretty good lately so I hope it will go fairly smoothly. Does anyone have any objections? I'm planning on starting tomorrow if not.
Why not 4.0.0? It was released in September 2018 and it's supposedly ABI compatible with 3.x: https://www.mail-archive.com/announce@lists.open-mpi.org/msg00119.html
Because it's still relatively new. I've hit enough problems with major bumps in openmpi to be a little wary. And I don't have the COPR testing record for it. We are doing this last minute so I'm trying to be a little more careful.
It's a big jump from the currently packaged 2.1.6. Any idea why we are so behind upstream? 3.0.0 was released in September 2017, 3.1.0 in May 2018. 3.0.0 should have been in F27 or F28 and 3.1.0 in F29 latest.
Because no one else has stepped up to help maintain it and I'm stretched too thin.
Also, they say 2.1.6 is the last release in 2.1.x series, so I'd say an update to a supported version is a must.
I concur as well.
Can help review if needed.
Benson Muite benson_muite@emailplus.org writes:
Because no one else has stepped up to help maintain it and I'm stretched too thin.
Also, they say 2.1.6 is the last release in 2.1.x series, so I'd say an update to a supported version is a must.
I concur as well.
Can help review if needed.
openmpi 3 is already packaged (in epel7), and epel7 MPI packages should be rebuilt to include it, which may help somewhat if there's any particular problem with 3. I don't think I've needed to do anything more than add an extra build to openmpi and mpich for the packages I've tackled so far, i.e. there's no obvious problem with v3, at least for building.
Presumably anyone doing openmpi breaking changes needs to be able to do rebuilds of dependencies, as Orion can.
[Re-sending from a subscribed address, apologies for any duplicates.]
Dominik 'Rathann' Mierzejewski dominik@greysector.net writes:
Why not 4.0.0? It was released in September 2018 and it's supposedly ABI compatible with 3.x: https://www.mail-archive.com/announce@lists.open-mpi.org/msg00119.html
I guess you haven't tried it :-/. It may be ABI compatible, but it's not API compatible (at least by default); that's apart from the usual sort of breakage you expect from a major version. I've wasted too much time on it, and given up locally for now.
I wish there was an MPI one could be happy with, providing the flexibility of OMPI without the maintenance issues.
On Thursday, 14 February 2019 at 11:11, Dave Love wrote:
[Re-sending from a subscribed address, apologies for any duplicates.]
Dominik 'Rathann' Mierzejewski dominik@greysector.net writes:
Why not 4.0.0? It was released in September 2018 and it's supposedly ABI compatible with 3.x: https://www.mail-archive.com/announce@lists.open-mpi.org/msg00119.html
I guess you haven't tried it :-/.
Indeed, I haven't.
It may be ABI compatible, but it's not API compatible (at least by default); that's apart from the usual sort of breakage you expect from a major version. I've wasted too much time on it, and given up locally for now.
Understood. I guess the way forward is to update to latest 3.1.x now and start working on 4.0.0 after F30 is branched in COPR. I might be able to lend a hand or two. ;)
I wish there was an MPI one could be happy with, providing the flexibility of OMPI without the maintenance issues.
Seconded. For example, there are subtle differences between MPICH and OpenMPI regarding MPI_OFFSET_KIND int type on 32bit arches and neither upstream considered changing to be consistent with the other: https://github.com/open-mpi/ompi/issues/4846 https://github.com/pmodels/mpich/issues/3002
I'm sure there are other pitfalls as well.
Regards, Dominik
On 2/12/19 10:13 PM, Orion Poplawski wrote:
I'm thinking about trying to sneak in openmpi 3.1.3 before the f30 branch. My COPR rebuilds have been looking pretty good lately so I hope it will go fairly smoothly. Does anyone have any objections? I'm planning on starting tomorrow if not.
- Orion
Building now. Will start dependent builds when it is complete.
On Tue, Feb 12, 2019 22:13:59 -0700, Orion Poplawski wrote:
I'm thinking about trying to sneak in openmpi 3.1.3 before the f30 branch. My COPR rebuilds have been looking pretty good lately so I hope it will go fairly smoothly. Does anyone have any objections? I'm planning on starting tomorrow if not.
No objections from me. Please do let me know if I can help.
scitech@lists.fedoraproject.org
-
Ankur Sinha -
Benson Muite -
Dave Love -
Dominik 'Rathann' Mierzejewski -
Orion Poplawski