Re: [PATCH v2] checkout: print blank line after autostash conflict advice
- From
Junio C Hamano <gitster@pobox.com>
- Date
- Aug 31, 2026, 17:19 UTC
- Message-ID
- <xmqq4igaxl5t.fsf@gitster.g>
- In-Reply-To
- <pull.2364.v2.git.git.1788177601572.gitgitgadget@gmail.com>
"Harald Nordgren via GitGitGadget" <gitgitgadget@gmail.com> writes:
Show 5 quoted lines
> To learn whether the stash apply resulted in conflicts, make the > autostash apply functions return 1 when the stash could not be applied > due to conflicts (and was stored back), while success and error remain > 0 and -1 respectively. Checkout can then use the return value to decide > whether to print the blank line.
Show 40 quoted lines
> @@ -1237,12 +1238,12 @@ static int switch_branches(const struct checkout_opts *opts,
> git_config_push_parameter(cfg.buf);
> strbuf_release(&cfg);
> }
> - apply_autostash_ref(the_repository,
> - "CHECKOUT_AUTOSTASH_HEAD",
> - new_branch_info->name,
> - "local",
> - stash_label_base,
> - autostash_msg.buf);
> + autostash_res = apply_autostash_ref(the_repository,
> + "CHECKOUT_AUTOSTASH_HEAD",
> + new_branch_info->name,
> + "local",
> + stash_label_base,
> + autostash_msg.buf);
> }
> if (ret) {
> branch_info_release(&old_branch_info);
> @@ -1255,6 +1256,8 @@ static int switch_branches(const struct checkout_opts *opts,
> if (!opts->quiet && !old_branch_info.path && old_branch_info.commit && new_branch_info->commit != old_branch_info.commit)
> orphaned_commit_warning(old_branch_info.commit, new_branch_info->commit);
>
> + if (autostash_res == 1 && !opts->quiet)
> + fputc('\n', stderr);
> update_refs_for_switch(opts, &old_branch_info, new_branch_info);
>
> if (created_autostash) {
> diff --git a/sequencer.c b/sequencer.c
> index 65afd100d9..5ed9ae86c4 100644
> --- a/sequencer.c
> +++ b/sequencer.c
> @@ -4815,7 +4815,8 @@ static int apply_save_autostash_oid(const char *stash_oid, int attempt_apply,
> if (label_base)
> strvec_pushf(&child.args, "--label-base=%s", label_base);
> strvec_push(&child.args, stash_oid);
> - ret = run_command(&child);
> + if (run_command(&child))
> + ret = 1;
> }This does not look like the right way to have the function return 1 if the objective is to do so only when the spawned "git stash apply <oid>" process fails due to conflicts.
The reasons for a non-zero return value from run_command() range from failing to find the command to run, to the command crashing, to the command voluntarily exiting with a fixed exit status to signal how exactly it exited to the spawning process.
Skimming bultin/stash.c::do_apply_stash(), I find this bit:
clean = merge_ort_nonrecursive(&o, head, merge, merge_base);
/* * 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. */ ret = clean >= 0 ? !clean : clean;
If 'clean' is 0 (i.e., unclean merge), we assign 1 (== !clean) to ret. If 'clean' is 1 or more (presumably a clean merge? Check what merge_ort_nonrecursive() does for details---I didn't), return 0. If merge-ort gave an error (i.e., negative), use it as-is. And this 'ret' is eventually returned to the caller apply_stash() , which returns that value as-is.
The value is given back to builtin/stash.c::cmd_stash() and the more detailed exit value unfortunately discarded ehre:
if (fn) return !!fn(argc, argv, prefix, repo);
but if one wants to implement
> To learn whether the stash apply resulted in conflicts, make the
> autostash apply functions return 1 when the stash could not be applied
> due to conflicts (and was stored back),...correctly, I think this patch needs a bit more work.
(1) Find the values returned by foo_stash() implementations of all
'git stash foo' subcommands. Do they follow the 0 == success,
negative == failure, positive == something else convention?(2) Update the code above to something like
if (fn) {
int ret = fn(argc, argv, prefix, repo);if (ret < 0) return 1; /* as before */ if (!ret) return 0; /* as before */ return ret + 1; }
or with whatever necessary tweak to allow the positive "I have
something to tell the spawning process" case from the bog
standard and boring 'return error("message");' cases. (3) document what each of 'git stash foo' signals with its exit
status.For expediency, it may be OK to assume any and all failures from "git stash apply <oid>" come from a conflicted stash application in your first version. If that is what your reviewer recommended, I would agree. But let's help users and future developers (who do not necessarily have to be you) by leaving a note that this code is not doing what it claims to do and needs more work in the code.
Thanks.