From: Phillip Wood Date: Thu, 27 Aug 2026 15:19:29 GMT Subject: Re: [PATCH v2 0/3] commit: refuse to amend during conflict resolution Message-ID: <6a04525c-d9e7-415a-845f-f11ed1c0f0d5@gmail.com> In-Reply-To: Hi Elijah On 27/08/2026 02:02, Elijah Newren via GitGitGadget wrote: > Both git commit --amend and a partial commit (git commit ) are > foot-guns while the user is in the middle of an operation that resolves > conflicts on top of HEAD: recording a conflict resolution is about capturing > the state of the whole tree as a new commit, not about rewriting HEAD or > committing a subset of paths. > > Historically we only rejected these during a merge or a cherry-pick or when > resolving an empty pick during a rebase. The same hazard exists for am, > revert, and rebase conflict stops, none of which were covered. This series > extends the refusal to all of them. > > The three patches: > > 1. reword the two pre-existing "empty commit" rebase messages, which were > misleadingly generic > 2. refuse git commit --amend during these additional operations > 3. refuse partial commits during the same operations. Thanks for working on this, it is a useful improvement to our UI. I've left a couple of comments but this all looks pretty sound to me. Thanks Phillip > Elijah Newren (3): > commit: reword the empty-commit rebase errors > commit: refuse to amend during conflict resolution > commit: refuse partial commits during conflict resolution > > builtin/commit.c | 51 +++++++++---- > sequencer.c | 65 +++++++++++++++++ > sequencer.h | 24 ++++++ > t/t3404-rebase-interactive.sh | 125 +++++++++++++++++++++++++++++++- > t/t3507-cherry-pick-conflict.sh | 22 ++++++ > t/t4151-am-abort.sh | 22 ++++++ > 6 files changed, 293 insertions(+), 16 deletions(-) > > > base-commit: 2c3adbb2c475981e340c79fdc5e7f4f9b5d9054e > Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-git-2389%2Fnewren%2Frefuse-amend-during-conflicts-v2 > Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-git-2389/newren/refuse-amend-during-conflicts-v2 > Pull-Request: https://github.com/git/git/pull/2389 > > Range-diff vs v1: > > -: ---------- > 1: 65c48ed3cb commit: reword the empty-commit rebase errors > 1: a3d6b059c6 ! 2: 4a1461e527 commit: refuse to amend during conflict resolution > @@ Commit message > Signed-off-by: Elijah Newren > > ## builtin/commit.c ## > -@@ > - #include "path.h" > - #include "preload-index.h" > - #include "read-cache.h" > -+#include "refs.h" > - #include "repository.h" > - #include "string-list.h" > - #include "rerere.h" > @@ builtin/commit.c: static int parse_and_validate_options(int argc, const char *argv[], > - else if (whence == FROM_REBASE_PICK) > - die(_("You are in the middle of a rebase -- cannot amend.")); > + use_editor = 0; > + > + /* Sanity check options */ > +- if (amend && !current_head) > +- die(_("You have nothing to amend.")); > +- if (amend && whence != FROM_COMMIT) { > +- if (whence == FROM_MERGE) > ++ if (amend) { > ++ if (!current_head) > ++ die(_("You have nothing to amend.")); > ++ /* > ++ * Refuse to amend in the middle of any operation that is > ++ * meant to record its result as a new commit on top of HEAD > ++ * rather than by rewriting HEAD. > ++ */ > ++ switch (sequencer_ongoing_operation(s->repo, whence)) { > ++ case ONGOING_NONE: > ++ break; > ++ case ONGOING_MERGE: > + die(_("You are in the middle of a merge -- cannot amend.")); > +- else if (is_from_cherry_pick(whence)) > ++ case ONGOING_CHERRY_PICK: > + die(_("You are in the middle of a cherry-pick -- cannot amend.")); > +- else if (whence == FROM_REBASE_PICK) > ++ case ONGOING_REBASE_EMPTY: > + die(_("You are resolving a commit that became empty -- cannot amend.")); > ++ case ONGOING_REVERT: > ++ die(_("You are in the middle of a revert -- cannot amend.")); > ++ case ONGOING_AM: > ++ die(_("You are in the middle of an am session -- cannot amend.")); > ++ case ONGOING_REBASE_CONFLICT: > ++ die(_("You are resolving conflicts during a rebase -- cannot amend.")); > ++ } > } > -+ if (amend && whence == FROM_COMMIT) { > -+ char *applying, *apply_dir, *stopped_sha, *amend_marker; > -+ int in_am, conflicted_stop; > + if (fixup_message && squash_message) > + die(_("options '%s' and '%s' cannot be used together"), "--squash", "--fixup"); > + > + ## sequencer.c ## > +@@ sequencer.c: int sequencer_determine_whence(struct repository *r, enum commit_whence *whence) > + return 0; > + } > + > ++enum ongoing_operation sequencer_ongoing_operation(struct repository *r, > ++ enum commit_whence whence) > ++{ > ++ char *path; > ++ int found; > + > -+ /* Check middle of revert */ > -+ if (refs_ref_exists(get_main_ref_store(the_repository), > -+ "REVERT_HEAD")) > -+ die(_("You are in the middle of a revert -- cannot amend.")); > ++ /* > ++ * The merge, cherry-pick, and (empty) rebase-pick stops are already > ++ * distinguished by 'whence'. > ++ */ > ++ switch (whence) { > ++ case FROM_MERGE: > ++ return ONGOING_MERGE; > ++ case FROM_CHERRY_PICK_SINGLE: > ++ case FROM_CHERRY_PICK_MULTI: > ++ return ONGOING_CHERRY_PICK; > ++ case FROM_REBASE_PICK: > ++ return ONGOING_REBASE_EMPTY; > ++ case FROM_COMMIT: > ++ break; > ++ } > + > -+ /* Check middle of `am` */ > -+ applying = repo_git_path(the_repository, > -+ "rebase-apply/applying"); > -+ in_am = file_exists(applying); > ++ /* > ++ * 'whence' is FROM_COMMIT, but we may still be in the middle of an > ++ * operation that records its result on top of HEAD; detect those > ++ * from their on-disk state. > ++ */ > + > -+ free(applying); > -+ if (in_am) > -+ die(_("You are in the middle of an am session -- cannot amend.")); > ++ /* In the middle of a revert? */ > ++ if (refs_ref_exists(get_main_ref_store(r), "REVERT_HEAD")) > ++ return ONGOING_REVERT; > + > -+ /* Check middle of rebase specifically stopped for conflicts */ > -+ apply_dir = repo_git_path(the_repository, > -+ "rebase-apply"); > -+ stopped_sha = repo_git_path(the_repository, > -+ "rebase-merge/stopped-sha"); > -+ amend_marker = repo_git_path(the_repository, > -+ "rebase-merge/amend"); > -+ /* > -+ * The apply backend only ever stops for conflicts; the > -+ * merge backend writes stopped-sha but omits `amend`, > -+ * which it writes only at a clean edit/reword stop. > -+ */ > -+ conflicted_stop = > -+ file_exists(apply_dir) || > -+ (file_exists(stopped_sha) && !file_exists(amend_marker)); > ++ /* In the middle of an `am`? */ > ++ path = repo_git_path(r, "rebase-apply/applying"); > ++ found = file_exists(path); > ++ free(path); > ++ if (found) > ++ return ONGOING_AM; > ++ > ++ /* > ++ * In the middle of a rebase that stopped for conflict resolution? > ++ * The apply backend only ever stops for conflicts, so the presence > ++ * of its state directory is enough. The merge backend writes > ++ * stopped-sha whenever it hands control back to the user, but omits > ++ * `amend` unless it stopped with HEAD already pointing at the commit > ++ * to be amended (a clean edit/reword stop); its absence therefore > ++ * marks a conflicted stop. > ++ */ > ++ path = repo_git_path(r, "rebase-apply"); > ++ found = file_exists(path); > ++ free(path); > ++ if (!found) { > ++ char *stopped_sha = repo_git_path(r, "rebase-merge/stopped-sha"); > ++ char *amend_marker = repo_git_path(r, "rebase-merge/amend"); > + > -+ free(apply_dir); > ++ found = file_exists(stopped_sha) && !file_exists(amend_marker); > + free(stopped_sha); > + free(amend_marker); > -+ if (conflicted_stop) > -+ die(_("You are resolving conflicts during a rebase -- cannot amend.")); > + } > - if (fixup_message && squash_message) > - die(_("options '%s' and '%s' cannot be used together"), "--squash", "--fixup"); > - die_for_incompatible_opt4(!!use_message, "-C", > ++ if (found) > ++ return ONGOING_REBASE_CONFLICT; > ++ > ++ return ONGOING_NONE; > ++} > ++ > + int sequencer_get_update_refs_state(const char *wt_dir, > + struct string_list *refs) > + { > + > + ## sequencer.h ## > +@@ sequencer.h: int sequencer_get_last_command(struct repository* r, > + enum replay_action *action); > + int sequencer_determine_whence(struct repository *r, enum commit_whence *whence); > + > ++/* > ++ * An in-progress operation that records its result (often a conflict > ++ * resolution) as a new commit on top of HEAD, during which amending > ++ * HEAD via "git commit --amend" is almost always a mistake. > ++ */ > ++enum ongoing_operation { > ++ ONGOING_NONE = 0, > ++ ONGOING_MERGE, > ++ ONGOING_CHERRY_PICK, > ++ ONGOING_REBASE_EMPTY, > ++ ONGOING_REVERT, > ++ ONGOING_AM, > ++ ONGOING_REBASE_CONFLICT > ++}; > ++ > ++/* > ++ * Return which in-progress operation, if any, is underway; see enum > ++ * ongoing_operation. 'whence' is the origin already computed for the > ++ * pending commit. > ++ */ > ++enum ongoing_operation sequencer_ongoing_operation(struct repository *r, > ++ enum commit_whence whence); > ++ > + /** > + * Append the set of ref-OID pairs that are currently stored for the 'git > + * rebase --update-refs' feature if such a rebase is currently happening. > > ## t/t3404-rebase-interactive.sh ## > @@ t/t3404-rebase-interactive.sh: test_expect_success 'correct error message for commit --amend after empty pick' > - test_grep "middle of a rebase -- cannot amend." err > + test_grep "resolving a commit that became empty -- cannot amend." err > ' > > +test_expect_success 'commit --amend is refused at a rebase conflict stop' ' > -: ---------- > 3: e0be8cdf63 commit: refuse partial commits during conflict resolution >