From: Phillip Wood Date: Thu, 03 Sep 2026 13:57:43 GMT Subject: Re: [PATCH v4 1/2] stash: reserve exit status 1 for conflicts Message-ID: <002b1324-0f7f-45ec-9f67-624a41801ff7@gmail.com> In-Reply-To: On 02/09/2026 20:51, Junio C Hamano wrote: > "Harald Nordgren via GitGitGadget" writes: > >> From: Harald Nordgren >> >> "git stash apply", "pop" and "branch" exit with status 1 both when >> applying the stash entry resulted in conflicts and when they fail for >> other reasons, so callers cannot tell the two apart. >> >> Follow the convention of "git merge-tree" and the merge strategies, >> which exit with status 1 to indicate conflicts and with a different >> non-zero status for errors: those subcommands now exit with status 1 >> only when applying the stash entry resulted in conflicts, in which >> case the stash entry is left in place, and exit with status 128, the >> status die() uses, when they fail for other reasons. Document the >> exit statuses. >> >> cmd_stash() used to collapse the return values of the subcommand >> implementations to a boolean. It now maps negative values, which >> signal a failure, to 128 and passes everything else through as-is. >> The only implementations that return a positive value are "apply", >> "pop" and "branch", which return the value of do_apply_stash(): >> "apply" returns it directly, and "pop" and "branch" drop the stash >> entry, via do_drop_stash(), which always returns 0, only when the >> application succeeded. The positive value is always 1, as >> do_apply_stash() only returns a positive value when the three-way >> merge was unclean. >> >> Make the convention explicit by introducing enum stash_apply_result >> with the values STASH_APPLY_CLEAN, STASH_APPLY_CONFLICT and >> STASH_APPLY_ERROR, and use it for the in-process autostash helpers, >> too. They spawn "git stash apply" and can now tell conflicts apart >> from other failures, e.g. a crash or death by signal of the child, >> which map to exit statuses above 1. Since we know the stash entry >> was saved, tell users so in the error message instead of leaving them >> wondering what happened to their stashed changes. >> >> Signed-off-by: Harald Nordgren >> --- > > The above is on the overly verbose side. The first two paragraphs > give enough discussion and the remainder mostly repeats with small > details sprinkled in, which can probably be shortened to 1/4 of the > amount of text, but it is OK. I think the analysis in the middle of the third paragraph is useful to make it clear that the return paths have been audited correctly. I agree the rest could be condensed or cut. Thanks Phillip > >> diff --git a/Documentation/git-stash.adoc b/Documentation/git-stash.adoc >> index 50bb89f483..fc6a9a008c 100644 >> --- a/Documentation/git-stash.adoc >> +++ b/Documentation/git-stash.adoc >> @@ -426,6 +426,15 @@ include::includes/cmd-config-section-all.adoc[] >> :git-stash: 1 >> include::config/stash.adoc[] >> >> +EXIT STATUS >> +----------- >> + >> +The `git stash` subcommands exit with status 0 on success. The >> +subcommands that apply a stash entry, i.e. `apply`, `pop` and `branch`, >> +exit with status 1 when applying the stash entry resulted in conflicts, >> +in which case the stash entry is left in place, and with a non-zero >> +status other than 1 when they fail for other reasons. >> + > > Great. > >> +static enum stash_apply_result do_apply_stash(const char *prefix, >> + struct stash_info *info, >> + int index, int quiet, >> + const char *label_ours, >> + const char *label_theirs, >> + const char *label_base) >> { >> int clean, ret; >> int has_index = index; >> @@ -717,8 +720,8 @@ static int do_apply_stash(const char *prefix, struct stash_info *info, >> >> /* >> * If 'clean' >= 0, reverse the value for 'ret' so 'ret' is 0 when the >> - * merge was clean, and nonzero if the merge was unclean or encountered >> - * an error. >> + * merge was clean, and 1 if the merge was unclean or a negative value >> + * if it encountered an error. >> */ >> ret = clean >= 0 ? !clean : clean; > > OK. > >> + if (fn) { >> + ret = fn(argc, argv, prefix, repo); >> + >> + /* >> + * The subcommand implementations return 0 on success, a >> + * negative value on failure, and STASH_APPLY_CONFLICT >> + * when applying a stash entry resulted in conflicts. >> + * Map failures to 128, the status die() uses, so that >> + * exit status 1 unambiguously indicates conflicts. >> + */ >> + if (ret < 0) >> + return 128; >> + return ret; >> + } else if (!argc) >> return !!push_stash_unassumed(0, NULL, prefix, repo); > > Style. Once one of "if", "else if" and "else" cascade gains > {braches}, others should do so as well. > >> +static enum stash_apply_result apply_save_autostash_oid(const char *stash_oid, >> + int attempt_apply, >> + const char *label_ours, >> + const char *label_theirs, >> + const char *label_base, >> + const char *stash_msg) >> { >> struct child_process child = CHILD_PROCESS_INIT; >> - int ret = 0; >> + enum stash_apply_result ret = STASH_APPLY_CLEAN; >> >> if (attempt_apply) { >> child.git_cmd = 1; >> @@ -4816,9 +4819,11 @@ static int apply_save_autostash_oid(const char *stash_oid, int attempt_apply, >> strvec_pushf(&child.args, "--label-base=%s", label_base); >> strvec_push(&child.args, stash_oid); >> ret = run_command(&child); >> + if (ret > 1) >> + ret = STASH_APPLY_ERROR; > > This kind of code that assigns any random "int" that is returned by > run_command() to "enum ret" that has much narrower valid value range > and then makes corrections annoys me a bit. > > One way to do this cleanly might be to make a small helper function > do_stash_apply(), and use it like so: > > if (attempt_apply) > ret = do_stash_apply(stash_oid, label_ours, label_theirs, > label_base); > > The implementation of do_stash_apply() would be like what you have > in "if (attempt_apply) {...}" block, perhaps like: > > static enum stash_apply_result do_stash_apply(const char *stash_oid, > const char *label_ours, > const char *label_theirs, > const char *label_base) > { > struct child_process child = CHILD_PROCESS_INIT; > > child.git_cmd = 1; > ... > strvec_push(&child.args, stash_oid); > switch (run_command(&child)) { > case 0: return STASH_APPLY_CLEAN; > case 1: return STASH_APPLY_CONFLICT; > default: return STASH_APPLY_ERROR; > } > } > >> - if (attempt_apply && !ret) >> + if (attempt_apply && ret == STASH_APPLY_CLEAN) >> fprintf(stderr, _("Applied autostash.\n")); >> else { >> struct child_process store = CHILD_PROCESS_INIT; > > Good, and the rest of this function is good.