Re: [PATCH v3] status: reword message for a missing upstream branch
- From
Junio C Hamano <gitster@pobox.com>
- Date
- Nov 19, 2026, 18:05 UTC
- Message-ID
- <xmqqjym8y9vy.fsf@gitster.g>
- In-Reply-To
- <20261119055726.774485-1-gitter.spiros@gmail.com>
Elia Pinto <gitter.spiros@gmail.com> writes:
Show 48 quoted lines
> diff --git a/remote.c b/remote.c
> index 49ee2ea318..7e26088f65 100644
> --- a/remote.c
> +++ b/remote.c
> @@ -2449,6 +2449,47 @@ static void format_branch_comparison(struct strbuf *sb,
> }
> }
>
> +/*
> + * Would a plain "git push" create the remote-tracking ref "upstream"?
> + *
> + * Only if "push_ref", what @{push} resolves to, is the upstream. That
> + * rules out push.default=nothing, push refspecs that do not include
> + * the branch, and pushes that go elsewhere, e.g. a mirror remote when
> + * the branch tracks a branch of a different name: @{push} does not
> + * resolve or names another ref.
> + *
> + * Even then, "matching" only updates branches that already exist on
> + * the remote, unless push refspecs or mirroring override it, as in
> + * branch_get_push_1().
> + */
> +static bool push_creates(struct branch *branch, const char *push_ref,
> + const char *upstream)
> +{
> + struct remote *remote;
> +
> + if (!push_ref || strcmp(push_ref, upstream))
> + return false;
> + remote = pushremote_get(pushremote_for_branch(branch, NULL));
> + if (remote && (remote->push.nr || remote->mirror))
> + return true;
> + return repo_config_values(the_repository)->push_default !=
> + PUSH_DEFAULT_MATCHING;
> +}
> +
> +static void format_missing_upstream(struct strbuf *sb, struct branch *branch,
> + const char *name, const char *push_ref,
> + const char *upstream)
> +{
> + strbuf_addf(sb, _("Your branch is set up to track '%s', "
> + "which does not exist.\n"), name);
> + if (!advice_enabled(ADVICE_STATUS_HINTS))
> + return;
> + if (push_creates(branch, push_ref, upstream))
> + strbuf_addstr(sb, _(" (use \"git push\" to create it)\n"));
> + strbuf_addstr(sb, _(" (use \"git branch --unset-upstream\" "
> + "to stop tracking it)\n"));
> +}OK. This kills two birds with one stone. An early return based on ADVICE_STATUS_HINTS omits the call to push_creates() that is only relevant when giving the advance, and the overly long lines in the caller due to a deeply nested block are now much more readable.
Show 18 quoted lines
> /*
> * Return true when there is anything to report, otherwise false.
> */
> @@ -2510,12 +2551,8 @@ int format_tracking_info(struct branch *branch, struct strbuf *sb,
>
> if (cmp < 0) {
> if (is_upstream) {
> - strbuf_addf(sb,
> - _("Your branch is based on '%s', but the upstream is gone.\n"),
> - short_ref);
> - if (advice_enabled(ADVICE_STATUS_HINTS))
> - strbuf_addstr(sb,
> - _(" (use \"git branch --unset-upstream\" to fixup)\n"));
> + format_missing_upstream(sb, branch, short_ref,
> + push_ref, full_ref);
> reported = 1;
> }
> free(full_ref);When the push.default configuration is set to 'upstream', and the branch.<name>.pushRemote (or remote.pushDefault) configuration is set to a remote that is different from branch.<name>.remote, because "git push" would detect misconfiguration and immediately abort like
fatal: You are pushing to remote '<remote>', which is not the
upstream of your current branch '<name>', without telling me
what to push to update which remote branch.Yet, we suggest '(use "git push" to create it)' with the advice, it seems (reproduction added at the end).
branch_get_push_1() returns the upstream for PUSH_DEFAULT_UPSTREAM without checking whether the push remote matches the upstream remote, and that is what comes back from branch_get_push() and stored in push_ref.
push_creates() checks !remote->push.nr and !remote->mirror and push_default != PUSH_DEFAULT_MATCHING to decide if "git push" would create the upstream we want, which says "yes" unfortunately in this situation.
But in builtin/push.c, setup_default_push_refspecs() has a guard "if (!same_remote)" in PUSH_DEFAULT_UPSTREAM before allowing the branch to be used as the destination. That is what makes the advice message incorrect in this case.
I am not sure what the right fix should be.
Intuitively, the correct change may be to update branch_get_push_1() to pay attention to the remote being the same (in other words, we are not in a triangular workflow), as that would match what "git push" does. Such a change, however, may have fallouts and break other callers of branch_get_push() that have been assuming the current behaviour.
Adding a band-aid guard to push_creates() to specifically detect this particular case would be safer, but it is far less satisfactory and makes me feel dirty, playing whack-a-mole with the code.
Perhaps branch_get_push_simple(), which is a relatively recent addition, might point us in a good direction?
I dunno.
t/t6040-tracking-info.sh | 8 ++++++++ 1 file changed, 8 insertions(+)
diff --git c/t/t6040-tracking-info.sh w/t/t6040-tracking-info.sh index 1f0583b391..4752559280 100755 --- c/t/t6040-tracking-info.sh +++ w/t/t6040-tracking-info.sh @@ -221,6 +221,14 @@ test_expect_success 'status (missing upstream, pushRemote is another remote)' ' test_cmp expect-no-push actual ' +test_expect_success 'status (missing upstream, push.default=upstream with pushRemote)' ' + test_must_fail git -C empty-clone -c push.default=upstream \ + -c branch.main.pushRemote=fork push && + git -C empty-clone -c push.default=upstream \ + -c branch.main.pushRemote=fork status >actual && + test_cmp expect-no-push actual +' + test_expect_success 'status (missing upstream, advice.statusHints=false)' ' cat >expect <<-EOF && On branch main