Abinit dfpt

[emarazzi]

Hi,
this is the PR for the latest implementation of dfpt wf in atomate2 wrt the materialsproject main branch.
We align everything to the discussion we had last week.

To run the dfpt workflow you will need the abipy version you can find in

https://github.com/emarazzi/abipy/tree/phonons_atomate2

there is a lot of differences in the tests part but I think we can discuss them when we will deal with the tests in a later stage

Future things to discuss:

• do we want to keep the output['dirs'] in the generate_perts as a list of dirs or a dict of lists of dirs?
• do we want to add some classmethod to perform some standard "modifications" of the phonon workflow predefined for the user, e.g. computing LOTO splitting?

[gpetretto]

Thanks @emarazzi and @VicTrqt for all the work.
I have added several comments, but hopefully these should not require large changes in the logic and in the code.

I have a few general comments:

• I think that there are sometimes some commented parts of old version of the code. If something is outdated or not relevant for the current choices it would be better to remove them.

• according to github, there are 206 .gz files that will be added with this PR, for a single test. I just wanted to be sure if this is really the bare minimum that it is required, since they are quite a lot. If all the jobs and files are needed, than it is fine like this. Just one thing though. I think there is a problem in this old part of the code:

  atomate2/src/atomate2/cli/dev.py

  Lines 370 to 373 in aeee107

  	empty_file = dest_dirdata / ".empty"
  	empty_file.write_text(
  	"Empty file for git to be able to have an empty directory"
  	)

  
  Here the empty file is always created. As a result:
  • the .empty file is zipped. I don't know if it is better this, but I am not sure it was intended. It can also stay this way, but would be better to decide.
  • there are a lot of .empty.gz in folders that are not empty.

  So I am wondering:

  • Is it really needed to have those empty folders in the references?
  • If yes (or at least we prefer to keep it this way for any reason), can you modify the abinit_test_data function, so that it only creates the file if the directory is actually empty? (Either by creating it only if the directory is really empty or by removing it when a file is copied there).

[gpetretto]

Why is abinit needed? To test stuff like k/q points generation?

In general I am not sure if it is a good idea to have this dependence, since the version on conda-forge is 10.0.3. It may be that in the future this will give raise to incompatibilities between the abinit version availabe and the python, that may require a later or older version.

I think we can leave it like this at the moment and monitor how things will evolve, but if this will be cause for tests failing they may ask us to disable them.

[VicTrqt]

If I remember correctly, it is because it is needed for the tests to generate the proper perturbations

[VicTrqt]

I left a comment to indicate why we need abinit

[gpetretto]

Fine with me, maybe let's mention this when the real PR is opened.

[gpetretto]

just a note to remember to switch to a released abipy version before opening the PR