Don't use binary for gitattributes for patch files about conda-smithy HOT 3 OPEN

conda-forge/emacs-feedstock#80 (comment) :

We might be able to do either

*.path -text
*.diff -text

or

*.path binary diff
*.diff binary diff

. IDK which one would work/be preferred.

Looking at https://github.com/git/git/blob/v2.43.0/attr.c#L662 it seems we have binary defined as

[attr]binary -diff -merge -text

. So,

*.path -merge -text
*.diff -merge -text

would probably suffice to address the reported issue while still retaining the desired behavior with Git not break source line ending depending on the users' configurations.

from conda-smithy.

Maybe we can try one of these changes in a PR?

IIRC the line ending issue was tricky to get right depending on whether the user was on UNIX or Windows also depending on whether the patch/source code was written on UNIX or Windows. So we probably want to test all 4 cases to make sure that things behave as intended

from conda-smithy.

Maybe we can try one of these changes in a PR?

Yes, it would be nice if someone could try it out.

IIRC the line ending issue was tricky to get right depending on whether the user was on UNIX or Windows also depending on whether the patch/source code was written on UNIX or Windows.

Yes, depending on the users' Git settings and platforms they use things can break easily.
A good example to test things out for would be python-feedstock since we do have patches for files with different line endings in there.

So we probably want to test all 4 cases to make sure that things behave as intended

No, we'd only need to test the

*.patch -merge -text
*.diff -merge -text

case.
binary is just an alias for -diff -merge -text.
So, the minimal change to get diff to display contents of *.patch/*.diff without risking further issues is just to remove the -diff part of binary.
See https://git-scm.com/docs/gitattributes/2.43.0 for details (search for -diff/-text).

from conda-smithy.