jj op undo/restore no longer restore git-tracking branches by default

[ilyagr]

This is attempt number ~3-4 to fix #922 and is heavily based on #1541 (two more attempts happened as two versions of #1487). There is a long discussion in that PR, the plan for this PR specifically is discussed towards the end.

The first 2 commits are somewhat unrelated and could be moved to a separate PR. They add tests pertaining to the question of whether we should import/export remote-tracking branches by default in normal repos (see discussion in #1739). It is unfortunate that changes to import/export & git-tracking branches affect push and fetch. However, I thought it might be interesting which of these tests are and are not changed by this PR.

See also #1741 for what tests change if we have this behavior in non-colocated repos as well (making this equivalent to #1541).

This may still be a bit unpolished, and I may have gone a bit overboard with the tests. Oh well.

I also discovered some minor unexpected behaviors and some mysteries about the test behavior still remain, mostly having to do with fetch/push behavior. We've been discussing what to do about these in #1739, I think it's a separate issue. I think they are acceptable (mostly changing from one weird behavior to another); I'll look at them again and suggestions are welcome.

Checklist

If applicable:

• I have updated CHANGELOG.md
• (maybe in followup PR) I have updated the documentation (README.md, docs/, demos/)
• (not planned) I have updated the config schema (src/config-schema.json)
• I have added tests to cover my changes

[ilyagr]

@yuja @martinvonz I'm not sure you noticed this PR is ready for review. It's not in perfect shape, but I think it'd benefit from other people's eyes.

Let me know if it would benefit from explaining something in more detail, either in commit messages, tests, or comments.

As I said in the description, the first two commits add tests that become a separate puzzle and (I would say) demonstrate new bugs. I could separate them in a separate PR.

[ilyagr]

It turns out test_git_push_undo_colocated (one of the puzzling tests from the first two commits) is also affected by #1771. The test change in the merge of the two PRs is:

diff --git a/tests/test_undo.rs b/tests/test_undo.rs
index eda4b2f1..506111fe 100644
--- a/tests/test_undo.rs
+++ b/tests/test_undo.rs
@@ -220,8 +220,12 @@ fn test_git_push_undo_colocated() {
     // was undone, but keeps track of the latest local version. TODO: Would this
     // be improved if `jj git export` exported remote-tracking branches?
     insta::assert_snapshot!(get_branch_output(&test_env, &repo_path), @r###"
-    main: 0a3e99f08a48 CC
-      @origin (ahead by 1 commits, behind by 1 commits): 0cffb6146141 AA
+    main (conflicted):
+      - 0cffb6146141 AA
+      + 0a3e99f08a48 CC
+      + 8c05de152218 BB
+      @git (behind by 1 commits): 0a3e99f08a48 CC
+      @origin (behind by 1 commits): 8c05de152218 BB
     "###);
 }

Anyway, we should probably focus on the last four commits.

[yuja]

The change LGTM, many thanks!

Just repeating my comment: I don't think --what should be a user-facing option, but being experimental seems fine.

[ilyagr]

Thank you very much, Yuya!

🎉