git: Add support for repository excludes file

[errmayank]

Closes #4824

Release Notes:

• Added support for Git repository excludes file .git/info/exclude

[errmayank]

@cole-miller ready for review :)

[cole-miller]

Thank you! This looks really good, I just have one minor comment.

[cole-miller]

Thanks! I tried running this locally and it looks like changes to .git/info/exclude do not propagate to the Zed ignore statuses. Rereading the code, it seems like repo_exclude_by_work_directory_abs_path probably needs to track needs_update like ignores_by_parent_abs_path already does. It looks like the test was meant to cover this, but right now it asserts that .env.example is still tracked after adding it to .git/info/exclude, which is the opposite of the behavior we want.

[errmayank]

Thanks! I tried running this locally and it looks like changes to .git/info/exclude do not propagate to the Zed ignore statuses. Rereading the code, it seems like repo_exclude_by_work_directory_abs_path probably needs to track needs_update like ignores_by_parent_abs_path already does.

I see, from my manual tests it seemed to work. Can you tell me how you debug the ignore statuses in Zed? I'll check how ignores_by_parent_abs_path does it. Thanks!

It looks like the test was meant to cover this, but right now it asserts that .env.example is still tracked after adding it to .git/info/exclude, which is the opposite of the behavior we want.

I think the test could be a bit clearer, but .env.example is asserted as tracked coz .gitignore still has !.env.example and local .gitignore has higher priority.

[cole-miller]

I tested by cloning the https://github.com/agentclientprotocol/agent-client-protocol repo, adding README.md to .git/info/exclude, and opening Zed---this causes README.md to have a faded appearance in the project panel (depending on the theme):

Then I did echo '' >.git/info/exclude from the terminal. After that, README.md still shows as faded in the project panel, but if I restart Zed it appears normal. So .git/info/exclude is being applied correctly at startup, but when it changes it seems like we're not recomputing the ignore statuses for the repo.

I think the test could be a bit clearer, but .env.example is asserted as tracked coz .gitignore still has !.env.example and local .gitignore has higher priority.

My mistake, thanks for pointing that out--seems like the test is correct then, I'm sort of surprised it does since it seems like the code for recomputing ignore statuses when .git/info/exclude changes isn't there right now. Maybe something else is causing us to recompute ignore statuses in the test?

[errmayank]

I realized that I only tested this for untracked files before and weirdly even that didn't seem to work properly, my bad.

I've added updates and it seems to work fine now, tested it on agent-client-protocol repository.

Here's the before/after:

screen-recording-repo-exclude.mov

Maybe something else is causing us to recompute ignore statuses in the test?

Hmm, I'm not sure about that either.

[esthertrapadoux]

Hey @errmayank just following up to let you know the team went pretty deep in more git feature work and we got behind on PR reviews. We're ramping back up and appreciate the patience! 👍

[errmayank]

@esthertrapadoux no worries at all, thanks for letting me know! :)

[cole-miller]

Thanks @errmayank!

[errmayank]

I'm not sure why editor_tests::test_outdent_after_input_for_bash is failing consistently on Linux. Seems totally unrelated.

For @cameron1024 as well it failed locally during our pairing session.

[cole-miller]

Sorry about that flaky test failure! I'll try to figure out what's going on with it and get this merged.

[errmayank]

Thanks!