Add write_input() before transferring files to remote

[jan-janssen]

This pull request addresses the issue raised by @Leimeroth in pyiron/pyiron_atomistics#1471 that the remote submission for LAMMPS jobs failed after the switch to using the calculate() function. While it is not exactly clear why the VASP job works fine, the work around in this branch fixes the issue for LAMMPS.

[Leimeroth]

Ok, so check if working directory is empty leads to a bunch of test failures and some horrible side effects, because I created the job with delete_existing_job=True. However the contents of the working directory are not deleted on time,
so on my local computer no input is written with the check.

[jan-janssen]

Ok, so check if working directory is empty leads to a bunch of test failures and some horrible side effects, because I created the job with delete_existing_job=True. However the contents of the working directory are not deleted on time, so on my local computer no input is written with the check.

The != and == sign were mixed up. You want to write the input files when the directory is empty and do not write them when it is not empty.

[jan-janssen]

@Leimeroth Can you run your tests again, like submitting a job with a custom potential to a remote cluster? Once this works and you are fine with the changes as they are in this pull request, can you approve this pull request? Then I can make sure it is included in the release of the next version. I aim to update both pyiron_base and pyiron_atomistics before the pyiron meeting on Monday.

[Leimeroth]

Now it works with my own potential and the EAM.
However, only the check prevents it from rewriting input, which could cause trouble when trying to rerun jobs.
When I run them again using delete_existing_job=True on my local computer the old files are still present on the remote cluster, so they will be used instead of my updated inputs.

[jan-janssen]

Are you sure about this? I thought the files are written locally and then transferred. So as the job is first deleted locally before the new input files are written the working directory is removed and a new empty working directory is created. Then at the copying stage we do not use the write_input() function so it should still overwrite the files on the cluster.

[Leimeroth]

Are you sure about this? I thought the files are written locally and then transferred. So as the job is first deleted locally before the new input files are written the working directory is removed and a new empty working directory is created. Then at the copying stage we do not use the write_input() function so it should still overwrite the files on the cluster.

You are right, I just tested it. Fine to merge from my side.