From: Phillip Wood Date: Mon, 02 Nov 2026 14:07:33 GMT Subject: Re: [PATCH 1/2] worktree remove: refuse a worktree in the middle of a rebase or bisect Message-ID: In-Reply-To: <20261030142137.54750-2-me@jaspermayone.com> Hi Jasper On 30/10/2026 14:21, Jasper Mayone wrote: > "git worktree remove" refuses a worktree with modified or untracked "refuses to move" would be clearer > files unless --force is given, but it removes a worktree that is in > the middle of a rebase, a "git am" session, a bisect or a cherry-pick > or revert sequence when its files happen to be clean. The state of > that operation lives in the worktree's own $GIT_DIR, so it is lost > with the worktree. A rebase stopped at an "edit" or "break" line, an > "exec" that failed, or a cherry-pick sequence after a conflict was > resolved and committed is clean in this sense and still has work in > progress. Cherry picking or reverting a single commit that contains no new changes also looks clean, but there is still a cherry-pick operation in progress as CHERRY_PICK_HEAD exists. One could argue that, because the user doesn't really lose any work, we don't lose much by deleting such a worktree but we're still deleting a worktree that contains an in-progress operation. Fixing that case simply requires us to see if CHERRY_PICK_HEAD, or REVERT_HEAD exist. > Treat such a worktree as unclean: refuse to remove it unless --force > is given, and name the operation in the message. > > A worktree whose directory is already gone is still removed without > --force, as before: the check is part of the check for a clean > worktree, which is only done when the directory exists. Sensible > Reported-by: D. Ben Knoble > Helped-by: Phillip Wood I'm not sure what I've done anything to help with this patch > diff --git a/Documentation/git-worktree.adoc b/Documentation/git-worktree.adoc > index e6e77252ab..da4c38f909 100644 > --- a/Documentation/git-worktree.adoc > +++ b/Documentation/git-worktree.adoc > @@ -142,9 +142,10 @@ the new working tree location. > `remove`:: > > Remove a worktree. Only clean worktrees (no untracked files and no > -modification in tracked files) can be removed. Unclean worktrees or ones > -with submodules can be removed with `--force`. The main worktree cannot be > -removed. > +modification in tracked files) can be removed. A worktree in the middle > +of a rebase, a `git am` session, a bisect or a cherry-pick or revert "bisect, cherry-pick or revert" > +sequence is not clean either, "either" doesn't make sense, because you've not described the unclean case yet. This line needs folding as well. > because removing it loses that state. Unclean worktrees or ones with submodules > +can be removed with `--force`. The main worktree cannot be removed. A list of the conditions that require --force might make this section clearer. > `repair [...]`:: > > @@ -189,7 +190,9 @@ twice. If the destination is already assigned to some other worktree but is > missing (for instance, if __ was deleted manually), then `--force` > allows the move to proceed; use `--force` twice if the destination is locked. > + > -`remove` refuses to remove an unclean worktree unless `--force` is used. > +`remove` refuses to remove an unclean worktree, or one in the middle of a > +rebase, `git am` session, bisect, or cherry-pick or revert sequence, too many "or"s as above > +unless `--force` is used. > To remove a locked worktree, specify `--force` twice. > > `-b `:: > diff --git a/builtin/worktree.c b/builtin/worktree.c > index 77ecd0f71f..ed08de1f43 100644 > --- a/builtin/worktree.c > +++ b/builtin/worktree.c > @@ -26,6 +26,7 @@ > #include "submodule.h" > #include "utf8.h" > #include "worktree.h" > +#include "wt-status.h" > #include "quote.h" > > #define BUILTIN_WORKTREE_ADD_USAGE \ > @@ -1408,6 +1409,32 @@ static void check_clean_worktree(struct worktree *wt, > original_path, ret); > } > > +/* > + * A rebase, "git am", bisect or cherry-pick/revert sequence keeps its > + * state in the worktree's own $GIT_DIR, and removing the worktree > + * throws that away even when "git status" says the files are clean. > + */ > +static void check_no_operation_in_progress(struct worktree *wt, > + const char *original_path) > +{ > + struct wt_status_state state = { 0 }; > + > + if (wt_status_check_rebase(wt, &state)) { > + if (state.am_in_progress) > + die(_("'%s' is in the middle of an am session, use --force to delete it"), > + original_path); > + die(_("'%s' is in the middle of a rebase, use --force to delete it"), > + original_path); > + } > + if (wt_status_check_bisect(wt, &state)) > + die(_("'%s' is in the middle of a bisect, use --force to delete it"), > + original_path); > + wt_status_state_free_buffers(&state); > + if (is_directory(worktree_git_path(wt, "sequencer"))) > + die(_("'%s' is in the middle of a cherry-pick or revert, use --force to delete it"), > + original_path); > +} This looks correct, though I wish we had a standard function that returned an enum describing which operation was in progress, rather than having to open code this sort of thing. > static int delete_git_work_tree(struct worktree *wt) > { > struct strbuf sb = STRBUF_INIT; > @@ -1461,8 +1488,10 @@ static int remove_worktree(int ac, const char **av, const char *prefix, > strbuf_release(&errmsg); > > if (file_exists(wt->path)) { > - if (!force) > + if (!force) { > + check_no_operation_in_progress(wt, av[0]); This triggers the new check when the worktree directory exists - good. > check_clean_worktree(wt, av[0]); > + } > > ret |= delete_git_work_tree(wt); > } > diff --git a/t/t2403-worktree-move.sh b/t/t2403-worktree-move.sh > index 69768c1207..4eb56a8911 100755 > --- a/t/t2403-worktree-move.sh > +++ b/t/t2403-worktree-move.sh > @@ -271,4 +271,63 @@ test_expect_success 'move worktree with relative path to absolute path' ' > test_cmp expect .git/worktrees/absolute/gitdir > ' > > +test_expect_success 'remove worktree in the middle of a rebase' ' > + test_when_finished "rm -rf mid-rebase; git worktree prune" && > + git worktree add --detach mid-rebase init && > + test_commit -C mid-rebase rebase-me && > + test_must_fail git -C mid-rebase rebase -f -x false init && > + test_must_fail git worktree remove mid-rebase 2>err && > + test_grep "in the middle of a rebase" err && > + test_path_is_dir mid-rebase && > + git worktree remove --force mid-rebase && > + test_path_is_missing mid-rebase > +' > + > +test_expect_success 'remove worktree in the middle of an am session' ' > + test_when_finished "rm -rf mid-am patch; git worktree prune" && > + git worktree add --detach mid-am init && > + test_commit -C mid-am am-me && > + git -C mid-am format-patch -1 --stdout >patch && > + git -C mid-am reset --hard init && > + echo conflict >mid-am/am-me.t && > + git -C mid-am add am-me.t && > + git -C mid-am commit -m conflict && > + test_must_fail git -C mid-am am ../patch && > + git -C mid-am reset --hard && If we're running a lot of consecutive commands in a worktree, it is clearer to use a subshell ( cd mid-am && ... ) && I wonder how hard it would be to extend the "setup" test to create a branch that has conflicts when it is applied with am, cherry-pick, or rebase, so that we don't have to set up the conflicts in each test. Thanks Phillip > + test_must_fail git worktree remove mid-am 2>err && > + test_grep "in the middle of an am session" err && > + git worktree remove --force mid-am > +' > + > +test_expect_success 'remove worktree in the middle of a bisect' ' > + test_when_finished "rm -rf mid-bisect; git worktree prune" && > + git worktree add --detach mid-bisect init && > + test_commit -C mid-bisect bisect-one && > + test_commit -C mid-bisect bisect-two && > + git -C mid-bisect bisect start HEAD init && > + test_must_fail git worktree remove mid-bisect 2>err && > + test_grep "in the middle of a bisect" err && > + git worktree remove --force mid-bisect > +' > + > +test_expect_success 'remove worktree in the middle of a cherry-pick sequence' ' > + test_when_finished "rm -rf mid-pick; git worktree prune" && > + git worktree add --detach mid-pick init && > + git -C mid-pick switch -c pick-src && > + test_commit -C mid-pick pick-one && > + test_commit -C mid-pick pick-two && > + git -C mid-pick switch --detach init && > + echo conflict >mid-pick/pick-one.t && > + git -C mid-pick add pick-one.t && > + git -C mid-pick commit -m conflict && > + test_must_fail git -C mid-pick cherry-pick pick-src~1 pick-src && > + git -C mid-pick checkout --theirs pick-one.t && > + git -C mid-pick add pick-one.t && > + git -C mid-pick commit --no-edit && > + test_path_is_dir .git/worktrees/mid-pick/sequencer && > + test_must_fail git worktree remove mid-pick 2>err && > + test_grep "in the middle of a cherry-pick or revert" err && > + git worktree remove --force mid-pick > +' > + > test_done