🐛 [BUG] NotADirectoryError when attempting to run git

[fxcoudert]

Describe the bug
A clear and concise description of what the bug is.

To Reproduce

After compiling nequip 0.5.5 from source, when trying to use it we see this error:

$ nequip-train config.yaml
Traceback (most recent call last):
  File "/ccc/work/cont003/gen7069/couderfx/nequip/bin/nequip-train", line 33, in <module>
    sys.exit(load_entry_point('nequip==0.5.5', 'console_scripts', 'nequip-train')())
  File "/ccc/work/cont003/gen7069/couderfx/nequip/lib/python3.8/site-packages/nequip-0.5.5-py3.8.egg/nequip/scripts/train.py", line 72, in main
  File "/ccc/work/cont003/gen7069/couderfx/nequip/lib/python3.8/site-packages/nequip-0.5.5-py3.8.egg/nequip/scripts/train.py", line 120, in fresh_start
  File "/ccc/work/cont003/gen7069/couderfx/nequip/lib/python3.8/site-packages/nequip-0.5.5-py3.8.egg/nequip/utils/versions.py", line 48, in check_code_version
  File "/ccc/work/cont003/gen7069/couderfx/nequip/lib/python3.8/site-packages/nequip-0.5.5-py3.8.egg/nequip/utils/versions.py", line 42, in get_current_code_versions
  File "/ccc/work/cont003/gen7069/couderfx/nequip/lib/python3.8/site-packages/nequip-0.5.5-py3.8.egg/nequip/utils/versions.py", line 42, in <dictcomp>
  File "/ccc/work/cont003/gen7069/couderfx/nequip/lib/python3.8/site-packages/nequip-0.5.5-py3.8.egg/nequip/utils/git.py", line 13, in get_commit
  File "/ccc/products2/python3-3.8.10/Rhel_8__x86_64/gcc--8.3.0__openmpi--4.0.1/cuda/lib/python3.8/subprocess.py", line 493, in run
    with Popen(*popenargs, **kwargs) as process:
  File "/ccc/products2/python3-3.8.10/Rhel_8__x86_64/gcc--8.3.0__openmpi--4.0.1/cuda/lib/python3.8/subprocess.py", line 858, in __init__
    self._execute_child(args, executable, preexec_fn, close_fds,
  File "/ccc/products2/python3-3.8.10/Rhel_8__x86_64/gcc--8.3.0__openmpi--4.0.1/cuda/lib/python3.8/subprocess.py", line 1704, in _execute_child
    raise child_exception_type(errno_num, err_msg, err_filename)
NotADirectoryError: [Errno 20] Not a directory: '/ccc/work/cont003/gen7069/couderfx/nequip/lib/python3.8/site-packages/nequip-0.5.5-py3.8.egg/nequip/..'

This does not appear to depend on the details of the config.yaml file, any example from nequip itself will reproduce the issue.

The error is that get_commit calls git with a cwd set by appending /.. to $VENV/lib/python3.8/site-packages/nequip-0.5.5-py3.8.egg/nequip (where VENV is the virtualenv where nequip is installed). But that fails, because nequip-0.5.5-py3.8.egg is not actually a directory, it's a ZIP file, so the OS does not recognise $VENV/lib/python3.8/site-packages/nequip-0.5.5-py3.8.egg/nequip/.. as a valid directory (and rightly so).

Environment (please complete the following information):

• OS: CentOS Linux release 8.4.2105
• Python 3.8.10
• python environment (commands are given for python interpreter):
  • nequip version: 0.5.5
  • e3nn version: 0.5.0
  • pytorch version: 1.10.2

Additional context

nequip was installed in the virtualenv from source by running: python setup.py install

[fxcoudert]

Ugly work-around: expand the nequip-0.5.5-py3.8.egg ZIP file into a directory of the same name at the same path.

[Linux-cpp-lisp]

Hi @fxcoudert ,

Thanks for finding this issue, I think I should have resolved it in d090f32, but let me know. I've never seen this before since for some reason on my systems it never zips the .eggs as far as I can tell.