Is pywin32 actually used?

[mattip]

setup.py has a line to install pywin32 which is a huge package of c-extensions that does not work with PyPy. Is it actually required for jupyter_server?

[mattip]

Answering my own question: it seems it is used in secure_write, which calls win32_restrict_file_to_user https://github.com/jupyter/jupyter_server/blob/d1cb381195cacb82de530e57f0f8bebed07bf518/jupyter_server/utils.py#L226, which imports and uses win32api and win32security. This code is duplicated in jupyter_core. The jupyter_core implementation seems to be the code used according to a user @SleepyRoy who reported the issue in PyPy.

[kevin-bates]

Correct - there was a bit of a migration exercise a while ago and this last step was not performed on this repo. jupyter_core should be used for this and setup.py should be updated to not include this dependency. There are a number of functions in utils.py that should get pulled from jupyter_core as well (e.g.. is_hidden()) and their usages updating accordingly - nice catch.

Would you be interested in making these changes?

[bollwyvl]

Can concur that this would help downstream packaging, and that we should strive to keep this package's dependencies as cross-platform as possible for as long as possible. Perhaps:

• on the 1.x line
  • provide shim imports in utils for everything that was moved from jupyter_core
  • add a deprecation warning
• drop entirely on 2.x

cc @maartenbreddels

[stonebig]

version 2 seems far far away , could the jupyter-core simplification be anticipated a bit ?
... it still leaves the problem in one place, though :https://github.com/jupyter/jupyter_core/blob/master/jupyter_core/paths.py#L387.L389

[kevin-bates]

I think that all that needs to happen is the dependency gets dropped from setup.cfg. The dependent methods were moved to jupyter_core via #382 but I failed to follow up on the removal of pywin32. I'll submit a PR for this.