Proof of concept for #9603

[kumaraditya303]

Concept for cloning repo treeless hence saving disk space and bandwidth.
#9603 (comment)

@uranusjr

[sbidoul]

TIL git clone --filter=tree:0. That sounds very interesting.

I tested with a project that uses setuptools-scm and it seems to work fine. @RonnyPfannschmidt you may want to give it a try.

There might be gotchas but this is worth exploring. Thanks @kumaraditya303

[kumaraditya303]

@pradyunsg you may want to give this PR a try.

[sbidoul]

Hi @kumaraditya303,

Are you closing this PR out of lack of patience or frustration, or because you discovered a flaw ?
If it's the former, please be patient, and don't hesitate to call for attention... these things tend to quickly go out of sight.

I personally think this is something worth exploring. I had a look again at this and I found this relatively recent article at github.com which gives food for thought.
This article seems to point to some performance limitations with treeless clones and submodules? Perhaps blobless clones would be good compromise?

Another question I have is about minimum git client and server version. It would be interesting to know the minimum requirements to determine if we need to fallback to the full clone in some circumstances?

Also, if we do this we probably want to do it for non editable requirements first and leave editable requirements doing a full clone, as users who use editable VCS URLs probably want to develop in the local tree.

[kumaraditya303]

Hi @sbidoul,
I closed it because of lack of interest among the maintainers, I called @pradyunsg if he wanted to give it a try but no reply :/, and the PR is of Feb and it is nearly three months so I closed it. Moreover @uranusjr was to eager to close the issue #9603 (comment) I even replied him to reopen the issue but he reopen didn't to discuss the issue

[kumaraditya303]

@sbidoul
If the maintainers have any interest in it then you may try the solution in this PR as it is just a single line change, and if they think it is correct then I can reopen PR as I see no flaw in treeless clone

[sbidoul]

I reopened the issue. You must understand that the pip team is small, and the topics to address numerous, so yeah, patience and occasional gentle nudges, and things move eventually :) I personally tested it and it seems to works, but we must be careful, so if you can help researching the questions I mentioned above that would certainly help!

[kumaraditya303]

I am open for discussion on issue if you need some info about the change

[pradyunsg]

@kumaraditya303 thanks for the nudge and sorry for the delay in responding to it here. Could you file a new PR with this change?

Pinging @RonnyPfannschmidt as well, since setuptools-scm is likely the main project that we want to know that this works fine with, and they're the only person I know who works on that. :)

---

As an aside, what @sbidoul said about lots of things pulling for pip's maintainers' attention is very true! For context, I got around to this from my "unread mentions" search in my GitHub notifications bucket (there's 210 unread notifications total, and many more still-actionable-and-read items). 😅

[dolfinus]

Actually, didn't get why a new PR is need for just the same change...

[RonnyPfannschmidt]

Treeless works fine for setuptools_scm

Just commits and tags are needed

[kumaraditya303]

@kumaraditya303 thanks for the nudge and sorry for the delay in responding to it here. Could you file a new PR with this change?

Will create new one !

[kumaraditya303]

Treeless works fine for setuptools_scm

Just commits and tags are needed

Yes treeless clone works perfectly for setuptools-scm as it has the required metadata