"Invalid script entry point" on Nuitka 0.5.29 installation

The latest Nuitka 0.5.29 cannot be installed from pip:

$ pip install nuitka
Collecting nuitka
  Downloading Nuitka-0.5.29.tar.gz (2.0MB)
    100% |████████████████████████████████| 2.0MB 660kB/s
Building wheels for collected packages: nuitka
  Running setup.py bdist_wheel for nuitka ... done
  Stored in directory: /home/frol/.cache/pip/wheels/46/56/4d/65ba15f23b85dbbb1fa09b2dd6c704b403cc641ec313812b53
Successfully built nuitka
Installing collected packages: nuitka
Invalid script entry point: <ExportEntry nuitka3 = nuitka.__main__:None []> for req: nuitka - A callable suffix is required. Cf https://packaging.python.org/en/latest/distributing.html#console-scripts for more information.

The error is the same for Python 2 and Python 3 environments (only nuitka3 gets nuitka in Python 2 environment).

• Linux, Windows, Mac OS are all affected
• Python 2.7.14 and 3.6.4
• pip 9.0.1 and 9.0.3

I am one of the maintainers of Nuitka package for Conda-Forge (Anaconda) and it fails on the CI: https://github.com/conda-forge/nuitka-feedstock/pull/12 (follow the CI links at the bottom of the page)

[kayhayen]

I created the check step for PyPI uploads, and the archive as created by “sdist” fails. The one you take from GH has hard coded runners for use in checkout, they are going to work, but sort of expect to be inside source checkout layout, not generally true. It might not run.

This reminds me, to check if my factory instructions will still work, as they do something similar.

However, the thing is, pip has its own code to create runners during installation, according to setup.py entry points, throwing away what is there, and inhibiting the setuptools code that I hook to make them work. So, I can get around that issue of having the complaint, but the runners didn’t work, as they cannot be run again with “-S” disabling site parameter, and then find the “nuitka” package again.

This issue is worked around by restoring the sys.path in the runners from an environment variable, then importing nuitka, and it works. However, pip and its wheel code are running beyond my control, and will not be patchable.

So what I did instead is to go back to what the original pull request did and make nuitka.main call itself instead of the runner. It used to give wrong error messages, about “main: some error message”, but I added code that preserves the sys.argv[0], and restores it now, and therefore calling the runner again has become unnecessary.

It now merely is a performance optimization to patch the runners, so they are leaner and do not import stuff for a second, but pip does that itself too, so it ought to be good.

I will put it through my CI, but in principle it will be nicer as it’s one less module to load on restart. And with that, my PyPI upload job now passes, and will fail if in the future it breaks again as severe as this.

[frol]

Conda-Forge CI passed: https://github.com/conda-forge/nuitka-feedstock/pull/13