fix(git): convert SCP-like submodule URLs to ssh:// format

[cibil47]

What does this PR try to resolve?

PR #16246 introduced a regression where submodules using SCP-like URLs fail because child_remote_url.into_url() requires WHATWG-parsable URLs.

This commit fixes the issue by detecting SCP-like URLs in absolute_submodule_url() and converting them to the equivalent ssh:// format.

How to test and review this PR?

1. Add a dependency to Cargo.toml and configure the dependency repository to pull its submodules using the following .gitmodules entry:

[submodule "common/path/apis"]
	path = third_party/apis
	url = git@github.com:common/apis.git

2. cargo fetch should succeed

[rustbot]

r? @weihanglo

rustbot has assigned @weihanglo.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: @ehuss, @epage, @weihanglo
• @ehuss, @epage, @weihanglo expanded to ehuss, epage, weihanglo
• Random selection from ehuss, epage, weihanglo

[weihanglo]

Thanks for taking a stab. This is indeed a regression.

Since Cargo doesn't support SCP-like git dependency URL in Cargo.toml, this case was overlooked. However, libgit2 supports SCP-like URL, so we should follow if it is in .gitmodules not in Cargo.toml. I think it is fine we scope this for supporting submodule only. For general SCP-like URL support, we track in #1851.

[cibil47]

@weihanglo One task failed because the CI runner was unable to verify the SSH host key. I could disable host key checking inside the test case using the GIT_SSH_COMMAND environment variable, but I am not sure if that is the right way to fix this. What is your suggestion?

[weihanglo]

@weihanglo One task failed because the CI runner was unable to verify the SSH host key. I could disable host key checking inside the test case using the GIT_SSH_COMMAND environment variable, but I am not sure if that is the right way to fix this. What is your suggestion?

Sure, we could try it. The best case is that we don't need to diverge our snapshot based on different platforms. The other alternative here would be we only run this test on certain platform that is guaranteed to pass.

[cibil47]

Sure, we could try it. The best case is that we don't need to diverge our snapshot based on different platforms. The other alternative here would be we only run this test on certain platform that is guaranteed to pass.

I updated the commit, and the test case now passes on all platforms

[weihanglo]

Thanks for working with us!

One thing to note that the path semantic is a bit different between the two forms

• In SCP-like git@host:path/repo.git, path is relative to the user's home directory (~git/path/repo.git)
• In SSH ssh:// ssh://git@host/path/repo.git, path is absolute from root (/path/repo.git)

For like GitHub and GitLab, this doesn't really matter as they intercept the path and resolve it. But for a self-hosted git server this might be an issue after the conversion.

That said, this conversion is as same as the suggestion in IntoUrl::into_url, and also this is a regression already a hard error, so this change is still better than nothing.

We may probably want a way to track the original Git submodule URL and use that during fetching, if possible.

View changes since this review