Missing files in resulting package

[lokik]

When I clone ase2sprkkr package

git clone ase2sprkkr
cd ase2sprkkr
git checkout 047223aedc0e06ba678ae93c8368572a13ab38fe

and install it using
pip install --no-deps .
using scikit-build-core version 0.12.1, there are many missing files in the resulting installed package (in site_packages), e.g.
the whole subpackages

ase2sprkkr.ase
ase2sprkkr.common.grammar_types
ase2sprkkr.bindings.nomad

or e.g. half of the examples from
ase2sprkkr.examples

If I pin the build backend to 0.11.5, the package is installed correctly.
There is no apparent reason, why some of the directories are excluded and some of them not.

[henryiii]

Is there a repo I can look at? If not, do you at least have your .gitignore and any relevant tool.scikit-build settings?

I'm assuming that it's the new inclusion mode. sdist.inclusion-mode = "classic" would fix it, as would setting minimum-version="0.11" (or any value less than 0.12), but I'd like to know why these are missing with the new mode.

[henryiii]

Ahh, I think I found it. It's this line:

https://github.com/ase2sprkkr/ase2sprkkr/blob/4e524b9cf3d7b67c8ac3b42037e2376d7f881554/pyproject.toml#L57

It's really expensive to recurse into every single subdirectory, so the modern logic doesn't go into an excluded subdirectory unless something inside it is explicitly included by a full path.

If you do want to use this pattern, you need the "classic" mode. Setting a minimum-version will protect you from changes in the future.

I wonder if the "manual" mode should be reversed, actually, so it includes then excludes? I think what you are trying to do with **/*, and it's what I'd hope manual mode would be able to do. I'll investigate. (GH is having issues at the moment)

[lokik]

Hi Henryii,
thanks for the reply and investigation. The repo is available here
https://github.com/ase2sprkkr/ase2sprkkr

I just want:

1. first exclude all.
2. then include files of given types.
   just not to need to list each particular file to exclude.

If the modern logic does not descend into excluded directories, it explain, why it should not work. However, it does not explain the actual behavior. E.g. in the examples:
https://github.com/ase2sprkkr/ase2sprkkr/tree/master/src/ase2sprkkr/examples
in the 01_... and 02_.... example directory, there are exactly the same filenames. However, the first one is (by 0.12.1) included, the second one is not. However, In the pyproject toml
https://github.com/ase2sprkkr/ase2sprkkr/blob/master/pyproject.toml
(nor .gitignore), there is no special handling of these directories (or files in them). So they should behave just the same. - and they do not.
So I think, that the actual behavior is not deterministic, which seems wrong to me.

[henryiii]

I did verify that tool.scikit-build.sdist.inclusion-mode = "classic" does restore the old behavior (as does setting a tool.scikit-build.minimum-version below "0.12").

By the way, you don't have to hard-code logging.level, you can pass -Clogging.level=DEBUG when building, like uv build --sdist -Clogging.level=DEBUG, so you don't always have to see the long printout, just when you need it.

I've added #1250 which should help us understand which rule is being hit for inclusion or exclusion.