-
Notifications
You must be signed in to change notification settings - Fork 40
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
The ESMF build system strikes back: failure to build on Narwhal #517
Comments
From the linker errors, it looks like it is not seeing the Fortran MPI routines:
|
@jedwards4b @mathomp4 Here I have another ESMF build error, this time on Narwhal. If you have time, please see the description above. Everything is the same (OS, Compiler, MPI, ...) and it built just fine last time round. The difference is probably the move from But I have no idea how this would change something, since the I got it to build by making the following change:
I am hesitant creating a PR for our spack fork (or the authoritative spack repo), because
Any ideas? |
It's not at all clear to me why you would need that, but I'll try it and let you know if I run into any problems. |
Thanks @jedwards4b . Fortunately, it allowed me to finish the build on Narwhal making this change. |
I'd give advice but I've never even had access to a Cray! I was going to say "No one uses It might be worth talking with ESMF (pinging @theurich) and having them add a |
|
@climbfuji one thing I am wondering is the usage of When this worked fine before, was this also with direct usage of |
Update/somewhat related. I am also getting linker errors for esmf@8.4.1 on macOS now that I am building static libraries (related to nlohmann-json):
|
See #545 (comment) for the solution for macOS (but not yet for Narwhal). |
@climbfuji - in order to move forward with the Narwhal issue... I had a few questions wrt whether the Cray compiler wrappers are used or not. Having complete ESMF build logs under Spack for previous successful build vs now failing build would be great. |
Unfortunately I don't have those, because I made that manual change (add |
Don't worry about failing one. I am actually more interested in the previous build that worked before you had to add the extra And again, I am mostly interested in verifying the suspicion that previously, when it was working, the |
log.install.intel-2021.4.0-narwhal-esmf-ok.gz Please see here for the last successful install w/o adding |
@climbfuji - the issue on Narwhal is a bit interesting because really all that changed here is the Spack part. You are still using the same compiler and ESMF version. |
Unfortunately I cannot test on Narwhal this week. Having v8.3.0b09 there is not critical, let's check next week if v8.4.0b03 works. |
Okay, that's fine. I will move forward with the v8.4.2 release either way, since EMC (i.e. Jun) is really pushing for it. |
Yes please. Definitely solves the macOS problem! |
@theurich my understanding is that your team has taken over the spack build of ESMF. Is that correct? If not, when will that happen? |
@edwardhartnett : esmf in spack develop represents a collaboration which includes Gerhard as well as Dom, myself, Ufuk and Alexander Richart. |
@jedwards4b in a recent meeting the ESMF team volunteered to take on this task and to ensure that spack can always build the latest ESMF releases. This is intended to reduce the resources that NOAA has to spend supporting the home-rolled ESMF build, which has been a considerable drain. We have only a small team to manage the installs of over 100 packages, and ESMF has, for a long time, been the tall pole in the tent, requiring a great deal of effort from the NOAA team - far more than any other single package. It is expected that @AlexanderRichert-NOAA will no longer need to devote time to maintaining the spack build, and @Hang-Lei-NOAA will no longer have to spend so much time debugging and fixing ESMF build problems on NOAA platforms. Is the latest ESMF release supported by spack? If not, when will it be? |
The 8.4.2 release was added to spack develop earlier today. |
@edwardhartnett - The ESMF team is taking an active role in maintaining the ESMF package under Spack. Of course it is a collaborative effort. |
@edwardhartnett I think this has worked very well over the last weeks or months. For sure @AlexanderRichert-NOAA and I have to spend some time on helping to debug and test on systems where the ESMF developers have no access, but apart from that @theurich and his colleagues have taken over the maintenance tasks for ESMF(, and @mathomp4 for MAPL). |
FWIW, I had to apply the same hack (add |
But fortunately we have a way forward with esmf@8.4.2! |
Describe the bug
ESMF no longer builds on Narwhal. Same versions as in
spack-stack-1.2.0
:esmf@8.3.0b09
, same compiler, same MPI library. Error is as follows. It's apparently missing the-lmpifort
linker flag, and I don't know why:To see the difference between the spack recipes, checkout our spack fork and run:
The main spack build options seem to be all correct (and same as for the successful build in spack-stack-1.2.0):
My guess is that the change to use
build_environment
introduced the error.To Reproduce
Try to build ESMF on Narwhal.
Expected behavior
For once, only for once, no ESMF build failures when we roll out a new release.
System:
Narwhal (Cray) with Intel-2021.4.0, cray-mpich, Cray compiler wrappers.
Additional context
Add any other context about the problem here.
The text was updated successfully, but these errors were encountered: