Fix 930 pyreverse regression

[DudeNr33]

Steps

• For new features or bug fixes, add a ChangeLog entry describing what your PR does.
• Write a good description on what the PR does.

Description

PR #857 introduced a regression for modutils.is_relative().
Calling this function failed if the inspected import statement was inside a module of a package at least two levels deep (e.g. astroid.interpreter._import.spec, see the added unit tests).

Inspecting the old implementation of is_relative which relied on imp and the implementation of imp.find_module, this only checked of the given modname (or better to say the first part after splitting it on ".") was a module or package inside the same package as the from_file itself.

The new implementation seemed to do a lot more, and I could not really understand what exactly.

Anyways, importlib.machinery.PathFinder().find_spec seems to be a better replacement for imp.find_module, makes the implementation simpler, passes all the tests and pyreverse is working again.

Type of Changes

	Type
✓	🐛 Bug fix

Related Issue

Closes #930
Closes pylint-dev/pylint#4186

[DudeNr33]

@Pierre-Sassoulas or @cdce8p could you please approve the workflow run?
I fixed the pre-commit checks that were failing on the previous run.

[hippo91]

@DudeNr33 thanks for this clear, concise and tested MR!
Maybe @cdce8p could also give his opinion on this.
Before merging it to master i would also like to have @pkolbus (author of #857) opinion on the use of PathFinder class.

[hippo91]

From what i can read in the PathFinder doc, it seems to do the job. However i would like to have @pkolbus opinion on this. Maybe he could tell us if did not use the PathFInder for a specific reason.
Instead of split you can use partition. Performance is a bit better as it doesn't split multiple times the string but just one. Moreover, as the PathFinder doc say, there is no need to instantiate this class as all its methods are @classmethod.
We thus could directly write:

return importlib.machinery.PathFinder.find_spec(modname.partition(".")[0], [from_file])

[pkolbus]

This part of #857 wasn’t my work actually, it came from @degustaf in #686. I didn’t run into any errors with it, so didn’t evaluate alternatives. @degustaf, can you provide insight here?

[cdce8p]

It should probably be:

    return bool(
        importlib.machinery.PathFinder.find_spec(
            modname.split(".", maxsplit=1)[0], [from_file]
        )
    )

[hippo91]

This part of #857 wasn’t my work actually, it came from @degustaf in #686. I didn’t run into any errors with it, so didn’t evaluate alternatives. @degustaf, can you provide insight here?

Thanks @pkolbus !

[cdce8p]

I noticed that astroid/modutils.py seems to be missing some imports

import importlib
import importlib.machinery

[cdce8p]

It should probably be:

    return bool(
        importlib.machinery.PathFinder.find_spec(
            modname.split(".", maxsplit=1)[0], [from_file]
        )
    )

[DudeNr33]

I noticed that astroid/modutils.py seems to be missing some imports

import importlib
import importlib.machinery

Added the imports and removed the instantiation of the PathFinder class.
Interesting that it already worked without the imports - and while PyCharm actually complained about the imports, VS Code did not.

[cdce8p]

Suggested change

	importlib.machinery.PathFinder.find_spec(modname.partition(".")[0], [from_file])
	importlib.machinery.PathFinder.find_spec(modname.split(".", maxsplit=1)[0], [from_file])

We started to use split with maxsplit=1 instead for pylint. Might be an idea to use here as well.

[DudeNr33]

Changed it, thanks for the suggestion.