Adding a gitignore after the fact is difficult to clean up

[peacememories]

Description

When adding files to the gitignore after jj has recognized them as new files, it is rather difficult to remove them from the automatically created commit again.

I ended up using git to manually create a new commit and work from there, but ideally, jj would recognize when a file added in the current commit has also been added to the gitignore.

Steps to Reproduce the Problem

1. create a new repository somewhere with git init
2. jj init --git-repo=.
3. touch testfile
4. jj st
5. echo "testfile" > .gitignore
6. jj st

Expected Behavior

Ideally, the testfile would not be part of the created commit.

Actual Behavior

testfile as well as .gitignore are marked as "added" in the commit.

Specifications

• Platform: NixOS 23.11
• Version: jj 0.13.0

[yuja]

5. `echo "testfile" > .gitignore`
6. `jj st`

After the step 5, you'll need to run jj untrack testfile.

Perhaps, jj status can show hint for the added files matching .gitignore.

[peacememories]

Ah thanks, I guess I missed untrack somehow. Adding a hint would be a good idea I think. I know some repos make use of the "committed but ignored file" behavior of git, but I think for most people it is unexpected.

[NoahTheDuke]

This has tripped me up before, and while jj untrack testfile works, it feel unintuitive compared to git's immediate respect of .gitignore.

[joyously]

It seems just the opposite of #3204.
Does jj have its own ignore file also?

[martinvonz]

We may want to consider the parent of the working copy when snapshotting the working copy. We currently only consider the previous snapshot of the working copy itself.

Does jj have its own ignore file also?

No. Maybe we will one day, or maybe not.

[joyously]

Is there special handling for undo when the ignore file is involved?

[martinvonz]

Is there special handling for undo when the ignore file is involved?

No

[PhilipMetzger]

This has tripped me up before, and while jj untrack testfile works, it feel unintuitive compared to git's immediate respect of .gitignore.

It makes sense though as it inverses the git add pattern which we cover with auto-tracking. But a hint for jj st is definitely a good idea.

[farnoy]

Encountered this as a fresh git convert. What compounded my confusion is that the documentation is outdated: https://github.com/martinvonz/jj/blob/6498a0cf8a5053e8a55fb503800e11b00953ecdc/docs/working-copy.md?plain=1#L60-L61 and contradicts an existing test case https://github.com/martinvonz/jj/blob/8e4d1af98eccf1ed8a79a7ddf9500467aa7684f5/cli/tests/test_gitignores.rs#L41-L47

[martinvonz]

What compounded my confusion is that the documentation is outdated:

Thanks for letting us know! I'll send a PR to fix it.

[dbarnett]

+1 would love a hint to use untrack. Would it be better for status to include a hint (a) every time you run it while a file in the repo is committed + ignored or (b) once only after jj detects a new (git)ignore edit disagrees with the set of committed files?

[dbarnett]

An interesting special case of this: when switching the working copy to before/after .gitignore changes (see #4349). While it's not "wrong" for jj to leave the untracked file around and then snapshot them as new/deleted, that's especially frustrating and silent in jj vs. git (like after doing jj edit xyz; jj new main I can later discover junk files dirtying up xyz). I wonder if jj could explicitly detect those ignore changes as it updates the working copy and warn about them, if nothing else...

[martinvonz]

+1 would love a hint to use untrack. Would it be better for status to include a hint (a) every time you run it while a file in the repo is committed + ignored or (b) once only after jj detects a new (git)ignore edit disagrees with the set of committed files?

I think every time you have a added file that's also ignored. I think we would have to implement that in the snapshotting code, so the working copy is the only thing that knows about the .gitignore files. I think @hooper was thinking of doing some refactoring of the gitignore code for a jj track command for tracking otherwise ignored files. I don't know if that would impact this feature.

[martinvonz]

Btw, note that both listing untracked files (after #4338) and listing ignored-but-added files requires walking the working copy, which means these features won't work well with background snapshotting. I don't know what to do about that.