sympy test suite fails on FileNotFoundError because path to temporary directory has symlinks

[robogast]

I was getting many errors in sympy's test suite because of failed shutil cp calls, such as:

__ sympy/codegen/tests/test_algorithms.py:test_newtons_method_function__ccode __
Traceback (most recent call last):
  File "/gpfs/nvme1/1/robertsc/EasyBuild/ebbuildpath/sympy/1.11.1/foss-2022a/sympy-1.11.1/sympy/utilities/_compilation/compilation.py", line 516, in compile_link_import_py_ext
    mod = import_module_from_file(os.path.join(build_dir, extname), sources)
  File "/gpfs/nvme1/1/robertsc/EasyBuild/ebbuildpath/sympy/1.11.1/foss-2022a/sympy-1.11.1/sympy/utilities/_compilation/util.py", line 249, in import_module_from_file
    raise ImportError("Failed to import: '%s'" % filename)
ImportError: Failed to import: '/home/robertsc/eb-workdir/ebtmpdir/eb-omfzk4u0/tmpct97j9ge/_newton'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/robertsc/eb-workdir/easybuild/generic/software/Python/3.10.4-GCCcore-11.3.0/lib/python3.10/shutil.py", line 813, in move
    os.rename(src, real_dst)
FileNotFoundError: [Errno 2] No such file or directory: '/home/robertsc/eb-workdir/ebtmpdir/eb-omfzk4u0/tmpct97j9ge/_newton.c' -> './_newton.c'

Fixed by getting the abspath with readlink -f /home/robertsc/eb-workdir, which got me /gpfs/nvme1/1/robertsc/EasyBuild, and after setting eb's tmpdir to this abspath / ebtmpdir, the test suite passed.

[casparvl]

I had the same problem in #17467

The old patch here would fix that by calling os.path.realpath() in some places in the code. I did not manage to recreate a patch for the version used in #17467 however

Maybe the easier solution from a maintenance perspective is just to have the EasyBlock do what @robogast did above: take the current EasyBuild tmpdir in an early stage, call a realpath on it, and overwrite the EasyBuild tmpdir with that realpath. I don't really see any downsides to that approach, but I might be missing something...?

[boegel]

@casparvl I think that approach makes sense. Are you up for opening a pull request for this?