[BUG] licence file included despite explicitly excluded via MANIFEST.in

[calestyo]

setuptools version

68.0.0

Python version

3.11.4

OS

Debian sid

Additional environment information

Also using python3-setuptools-scm 7.1.0

Description

I have a project using pyproject.toml which includes a file LICENCE and a symlink COPYING to that.

It seems that setuptools cannot include symlinks (🙁) but instead resolves the, so the very same file is included twice, making the sdist and wheel needlessly bigger.

One way of solving this, would be supporting symlinks ;-) ... or perhaps not including the file if it's a symlink to an already included file (an perhaps restricting this to the recognised licences files?).

Anyway, my issue here is, that I've set up a MANIFEST.in with:

exclude COPYING

yet it still includes the file, which I think it shouldn't do, or should it?

Only if I set:

[tool.setuptools]
license-files = ['LICEN[CS]E*', 'NOTICE*', 'AUTHORS*']

in pyproject.toml it really excludes the file (or probably rather not even includes it in the first place, regardless of my MANIFEST.in.

Expected behavior

Files explicitly excluded, should be excluded. :-)

How to Reproduce

1. Create a simple empty project with LICENCE and COPYING as a symlink to it, as well a a default pyproject.toml
2. Also MANIFEST.in as described above
3. Create a sdist
4. It will have the COPYING despite the exclusion.

Output

not applicable

[abravalheri]

Hi @calestyo, thank you very much for opening the issue.

Setuptools is a project that depends on voluntary work from the community, with the maintainer team (also volunteers) working as facilitators/catalysts.

Users are encouraged to discuss possible implementations and propose PRs based on these proposal's feedback if they find a problem that is pressing for their workflow or would like to request a feature that is meaningful to them. Otherwise it is a bit difficult to guarantee everything gets fixed/implemented given the limitations we have in terms of volunteers' time.

I understand that not everyone can/would like to provide contributions, and if that is your case (and this issue is urgent for you) please consider implementing the workaround you discussed in a more permanent basis, until a member of the community volunteers to tackle this issue.

[calestyo]

From what I saw I'd rather say it may go a bit too deep into setuptool's internals for me to provide a PR for that - I rather still struggle with even using setuptools ;-)

So I guess for now I just use the workaround, and maybe someone may pick it up sooner or later.

[jpakkane]

I have a different symptom of the same issue (probably).

In Meson we have a large directory of test files. These include symlinks that must be preserved as is, because they are used to test our code that handles symlinks.

The package is created with Setuptools by having the entire directory in MANIFEST.in as graft test?cases. Setuptools expands symlinks so the resulting tar file contains duplicates of the pointed file instead of the symlinks. It is arguable whether individually listead files in the manifest should be deduplicated or not but graft dirname really should store the entire directory as is, especially since all the symlinks point to other things inside the directory.

Thanks,