Closed jgfouca closed 8 years ago
@jgfouca : How can you recreate this issue? Does this issue occur only on the ANL workstations?
@jayeshkrishna : That's the only place I've seen it. I think @ndkeen was seeing it elsewhere. Maybe he can comment real quick to say what machine that was.
@ndkeen was seeing it when building standalone MPAS with PIO2 I think. This is different than the config (ACME+PIO1) we have right now and might be a different problem.
I am wondering if this is related to the netcdf install (whether the nf_fill* are available via the netcdf mod?) on the ANL workstations. @sarich : Were you able to verify that the netcdf installations on the ANL workstations worked correctly?
@jgfouca : How are you building the case (Can you provide the commands so that I can reproduce it?)?
@jayeshkrishna : sure.
create_test ERS.f09_g16_g.MPASLI_ONLY
... should do it. On the ANL workstations.
thanks, I will give it a try and let you know the results
@jayeshkrishna , thanks! I should mention, it looks like my branch to get ANL workstations working hasn't been merged yet: https://github.com/ACME-Climate/ACME/pull/909
You'll have to work off that branch until it gets merged.
Thanks, did the other tests build correctly (Is this the only one that fails?)?
All the tests that use MPAS compsets failed to build.
@jgfouca : I just tried building the test on compute001 and saw that env_mach_specific was empty. Is that expected?
@jayeshkrishna : classic git mistake on my part. I did a commit -a but I had untracked files. Please pull and try again.
ok, thanks. Now I am able to reproduce the problem mentioned above.
I was trying to build stand-alone mpas on edison/cori. As part of that, I used a PIO2 build and I saw those compiler errors (that you referred to). They were easy to fix and it builds (on edison/cori with intel). Without PIO2, we had some other issues, but the PIO being used was an old build from a few years ago.
After adding the paths to the pnetcdf library the case is building for me on compute001. Please try out the patch and incorporate it into your branch if it works for you,
See the patch here - https://gist.github.com/jayeshkrishna/1ad746aa105acce21b590ededa239e49
@jayeshkrishna So, this (proposed) fix doesn't need any modifications to the MPAS framework? Does this also fix the issues you previously had related to PIO2, or have you tried that?
No, this change only affects the machine files. This fix does not handle the issues that we previously saw with ACME+PIO2 (PIO2 no longer exposes NF_FILL* - PIO1 exposed, probably not intended, NF_FILL* in pnetcdf via the pio modules). For fixing that issue I suggest the following,
@jayeshkrishna I don't know that adding the PIO2 fill values to PIO1 would be appropriate, as that will break backwards compatibility with previous versions of PIO for MPAS only users outside of ACME.
It might be better to just include the correct NetCDF include file / module, since that should be a more robust solution.
At that point though, it's really just a matter of choosing the correct file to include / use. Does include 'netcdf.inc'
fix the issues for you? (I suspect use netcdf
doesn't).
Including netcdf.inc in MPAS should also work (if you are using "use netcdf" you might have to use NF90_FILL*).
@jayeshkrishna , your patch fixed the problem, thanks
From slack: