ImportError when using PEP 420 namespaces

[zacchiro]

Steps to reproduce

1. git clone http://forge.softwareheritage.org/source/swh-core.git
2. export PYTHONPATH=$(pwd)/swh-core
3. pylint3 swh.core

Current behavior

ImportError: Unable to find module for /tmp/swh-core/swh/core/api.py in /tmp,

(or equivalent)

Expected behavior

Should be able to walk the module and pylint it.
Note that both "import swh.core" and manually pylinting the specific module that pylint claims is not loadable (pylint3 swh.core.api) work just fine.

I've given more details about this issue on the code-quality at https://mail.python.org/pipermail/code-quality/2017-September/000937.html (no answer yet at the time of filing this).

I've seen there are a number of PEP 420 related issues, but as far as I can tell they are either supposed to be already fixed in the version of pylint I'm using, or the fail in different ways (e.g., with a proper "cannot load module" message instead of a runaway ImportError exception like in my case).

pylint --version output

pylint3 1.7.2,
astroid 1.5.3
Python 3.5.4 (default, Aug 12 2017, 14:08:14)
[GCC 7.1.0]

[PCManticore]

Thanks for the report! We do have a couple of issues related to PEP 420, I'll have to take a look to see exactly why this is not working as intended. As you mentioned, linting with pylint swh/core works fine, but it should work with dot syntax as well.

[PCManticore]

In the meantime, I'm going to assign this for the next release so we could fix all these PEP 420 problems before releasing 1.8.

[arcivanov]

I'm surprised this hasn't been fixed yet. Pylint simply does not work with implicit namespaces at all. Case in point: pybuilder/pybuilder#818 (comment)

[DanielNoord]

I have tried to reproduce this on main but was able to lint the package. Are we sure this is still not fixed?

There are many issues with namespace packages but I wonder if this one has been fixed already.

[Pierre-Sassoulas]

I created a label for namespace-package so we can track and close them. There's probably a lot of near-duplicates.

[jacobtylerwalls]

Fixed in pylint-dev/astroid#1576 -- I'll try to get to a regression test tomorrow.