search

MolSSI / QCEngine

Quantum chemistry program executor and IO standardizer (QCSchema).
https://molssi.github.io/QCEngine/
BSD 3-Clause "New" or "Revised" License
163 stars 79 forks source link

Some tests fail #324

Open yurivict opened 2 years ago

yurivict commented 2 years ago

Describe the bug See log.

Version: 0.20.1 OS: FreeBSD 13 (installed from the package)

loriab commented 2 years ago

That's peculiar -- it looks like you're getting a memory segfault whenever nwchem invoked. You're running 7.0.2 but nwchem 7.0.0 from Ubuntu was fine in our testing https://github.com/MolSSI/QCEngine/runs/3840808437#step:4:211 . When did you last have this running correctly, v0.20.0 or v0.19.0 (or is this the first packaging?)? There was a change in 0.20 in how mem was passed to nwc, but I confirmed the new way to work with v5 and v7.

yurivict commented 2 years ago

When did you last have this running correctly, [...]

The FreeBSD port didn't run tests before. I just added the ability to run tests and discovered this problem.

WardLT commented 2 years ago

The memory segfault is surprising and I'm puzzled by it.

I can point out another problem is that the NWChem is not compiled with Python support. That's currently required by QCEngine to compute gradients with NWChem. You can see that issue in how many of the tests crash with

python_input is not in this build of NWChem

in the STDOUT.

yurivict commented 2 years ago

I can point out another problem is that the NWChem is not compiled with Python support.

I didn't realize that NWChem has Python support. The Arch Linux port doesn't enable python support either.

Do you know how to enable python support in NWChem?

WardLT commented 2 years ago

It should be pretty straightforward. I don't have my Makefiles on hand (filesystems at ALCF are down for maintence), but you should be able to set the PYTHONVERSION environment variable to the major.minor version of Python in your environment.

For further details: https://nwchemgit.github.io/Compiling-NWChem.html#setting-python-environment-variables

yurivict commented 2 years ago

MPI was broken on that machine. I fixed MPI, added the PYTHON option to nwchem, and updated the test log in the URL in my OP.

loriab commented 2 years ago

could your geometric be very old? all those tests are failing. they don't have any exactly new releases, but I could believe that anything less than the latest wasn't fully compatible with qca. https://github.com/leeping/geomeTRIC/tags

yurivict commented 2 years ago

geomeTRIC is of the latest version 0.9.7.2 released in 2019. Maybe qcengine expects the current revision? They didn't have any releases after 2019.

Also log contains NWChem SEGVs. They probably should be reported to nwchem but it's unclear how to reproduce them.

yurivict commented 2 years ago

0.21.0 test failures py38-qcengine-0.21.0-test.log