[3.0.1] pyproject.toml/setuptools: automatic package discovery does not work on debian

[picca]

In order to build guidata I need to add this in pyproject.toml

+[tool.setuptools.packages.find]
+include = ["guidata*"]
+

If not it finds multiple directory guidata and debian and understand that this is a multiple package project.
I think that it will be the same when you will re-add the doc directory.

[PierreRaybaut]

Thanks for your feedback fred!

I think that Python package management has never been worse...
Using setuptools and the new official way (pyproject.toml) is a nightmare.
Unfortunately, using setuptools is my only option since I need to use the same solution with all my projects and some of them (e.g. guiqwt) have C++ extensions.

[picca]

pyfai switch to mesonpy like numpy, and next scipy. I think that at the end this will be the officiel way to build most of c/c++ extensions... here the link https://github.com/silx-kit/pyFAI

I think that Python package management has never been worse...

I prefer the new pyproject whuich is more declarativ. the problem with setup.py was thaht so many peoples where customizing it and it was a nightmare for packager ;)

Using setuptools and the new official way (pyproject.toml) is a nightmare. Unfortunately, using setuptools is my only option since I need to use the same solution with all my projects and some of them (e.g. guiqwt) have C++ extensions.

One problématic area is the package data which are not installed in the package directory. at least in haskell there is a dedicated method common to all package in order to find the data once installed... Cheers Just for info when will be release datalab ? and are you aware of this https://datalab.sspcloud.fr/home

[PierreRaybaut]

Again, our minds are somehow connected :).

I did some tests with mesonpy and it seems that it could be the right long-term solution indeed.
But, for now, as I have a solution that works, I will stick with setuptools for a while. I think I will get back to this in a few months or so. So, definitely, mesonpy should be my final choice.

[PierreRaybaut]

Regarding DataLab, I'm not sure: hopefully in September? It depends on a financial support that we are currently working on.

There are multiple projects called DataLab out there: that is quite a generic term actually. So we chose DataLab in spite of this situation and we do not pretend that this will be the DataLab, but just a DataLab among others.