EDIT: all my problems were fixed by touch src/sage/__init__.py which has the immediate effect of isolating the current sage package from the system (as in: the presence of this __init__.py file "closes" the sage package and nothing else will be taken from the system site-packages).

I'm leaving the original comment below in case someone has a similar issue. The problem below was caused by sage/modules/__init__.py existing in the system site-packages, but not existing in the local tree, so the sage.modules package (and all its modules) get imported from site-packages instead of locally. And of course sage.modules.with_basis.all doesn't exist in system site-packages (but it is expected to exist in the local tree)...

You are just making my life miserable:

ModuleNotFoundError: No module named 'sage.modules.with_basis.all'

I did nuke all the empty __init__.py files in my system site-packages. Then my system sagemath is broken because I'm missing some all.py files. Also, some non-empty __init__.py files have to be removed as here. Producing a patch from #35322, #35366 and this to apply to 9.8 is a pain because of the huge amount of churn. Also, patching 9.8 is a moving pain since what has to be done changes depending on which of these three I want to test...

One thing I don't understand is: why on earth are the empty all.py files needed? Aren't you repeating python mistakes? It feels crazy that some doctest failures disappear just by adding an empty __init__.py or an empty all.py.

I would like to argue that the behaviour with or without these empty files should be identical. That's the point of PEP 420: that directories without __init__.py are packages just as much as those with __init__.py files. Requiring an all.py file goes against this.

In any case, as I requested before, it would be better if there is JUST ONE single point of "removing empty files that may break havoc" and then don't do it again.

Even more: could we please have

(1) one PR that introduces all the changes needed to support removing empty __init__.py files, but leaves all the empty __init__.py files there? The goal is that I get something where I can safely do

find  .../sage/ -name "__init__.py" -size 0 -delete

without having any type of trouble and without having to add random all.py files.

(2) a separate PR with only removal of empty __init__.py files ?

Ideally (1) is merged in one release, and (2) is merged in a later release.

After the move to gh, I was able to contribute a lot more efficiently to sage (both code and reviews). But the last 2 weeks, since this happened with the merge of #35100, I've been stuck unable to do anything at all. I'm getting tempted of pruning all my sage git trees for 6 months to wait until this is over.

Kindly rephrase for clarity

Rebased away from #35322. Ready for review.

Rebased on top of #35418

Merge conflict with #35358 (see 63c3ab6)

FWIW, after fixing the trivial merge conflict, all tests pass for me on void linux.

Merged & resolved

LGTM

Thanks!