Using git LFS x setuptools-scm x cibuildwheel - project always dirty?

[jemand771]

I have this rather big project (that I sadly can't share; will create a reproducer if necessary+possible). It has a bunch of .tar.gz files checked in using git LFS.

When I build this project using cibuildwheel, the resulting version is always dirty, i.e. contains .dYYYYMMDD. Running a quick before-build = "git diff" shows me that all LFS-managed files (and a submodule that contains LFS-managed files itself) are considered "modified". However, the files' contents are fine, their hashes inside /project inside the container match my host filesystem. git only thinks the files changed, I assume because git-lfs (the command) isn't available inside the container.

As a test, I tracked a text file using git LFS and sure enough, the (text) diff for it inside the container is something like

@@ -1,3 +1,11 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a1ed0c3e8cf7c447b2c6744f68f9bfe7efe65deaf05a22dd92f1135e1f94b67b
-size 516
+hello
+world
+123

now, how would I go about this?

• I don't want to run custom builder images that have git-lfs (maintaining/distributing is a pain)
• Installing git-lfs on the fly in before-build would technically work but is wasteful (and manylinux/musllinux use different package managers)
• "just don't use lfs" - thanks, sadly not my call
• is this a problem for https://github.com/pypa/setuptools-scm ? not really, I mean it's doing everything it's supposed to.
• do I ask over at https://github.com/pypa/manylinux to bake git-lfs into the default image? That feels rude and overkill for such a nieche use case

I'm a bit lost as to what I should do now. None of my ideas/options above sound all that great :/

[mayeut]

git-lfs will be available in manylinux images, see pypa/manylinux#1857