pip install git+... from a branch/tag could be faster

[fandreuz]

What's the problem this feature will solve?

pip install git+...@foo-branch clones the whole repository and then checks out foo-branch. This process takes some time if the repository has a long history or many branches.

Describe the solution you'd like

I'd like to leverage git clone parameters to make the whole thing faster when a specific branch is requested. In particular, I'd use `the following:

• --branch foo-branch
• --depth 1

Alternative Solutions

The current solution works fine, but is slow in some specific situations.

Additional context

https://git-scm.com/docs/git-clone

Code of Conduct

• I agree to follow the PSF Code of Conduct.

[fandreuz]

I can take this if this improvement is approved, I've spotted where the changes should go

[notatallshaw]

Thanks for reporting, but check this comment: #12949 (comment)

we actually tried this before and had to revert due to packages relying on SCM metadata to build

Also, other previous discussions:

• git clone only the last/required commit (--depth 1) #344
• git checkout could/should be done with --depth=1 #2432
• Shallow git clones #4938

[ichard26]

It's also not true that pip always clones the entire repository. Since pip 21.3, pip performs blobless clones where possible (aka git >= 2.17) to speed up clones without breaking projects which rely on git history in their build process.

pip/src/pip/_internal/vcs/git.py

Lines 274 to 280 in 4204359

	# Git added support for partial clone in 2.17
	# https://git-scm.com/docs/partial-clone
	# Speeds up cloning by functioning without a complete copy of repository
	self.run_command(
	make_command(
	"clone",
	"--filter=blob:none",

Perhaps the --single-branch flag could be added when a specific branch is required, but that'd require additional investigation. It's not immediately clear how pip would know how to determine whether a specified revision is a branch (and not a tag or commit SHA).

[notatallshaw]

pip performs blobless clones where possible (aka git >= 2.17) to speed up clones without breaking projects which rely on git history in their build process

And even that optimization broke some people's build process: #10652

[fandreuz]

Thanks @notatallshaw, indeed I did not consider those kinds of builds. I guess some flag to enable "unsafe" optimizations is out of the question, right?

Perhaps the --single-branch flag could be added when a specific branch is required, but that'd require additional investigation.

We could query the repository for the existence of a branch with the given name. However, this would not solve the which compelled me to open the issue, which is more about the history than the number of branches.

[notatallshaw]

I guess some flag to enable "unsafe" optimizations is out of the question, right?

I don't think it's "out of the question", but I do think it's a question of complexity vs. benefit, if it's not the default behavior of pip it doesn't add significant benefit, so the complexity should be very low.

I'll reopen this ticket with it being the idea of adding a "non-default" optimization, but I don't know how much traction it will get with pip maintainers.