Re: [PATCH v13 6/6] remote: resolve @{push} through remote.<name>.fetchRefmap
- From
D. Ben Knoble <ben.knoble@gmail.com>
- Date
- Nov 11, 2026, 18:27 UTC
- Message-ID
- <CALnO6CB+AP=osF8mjkGGHWZz8YX322UoSjE6DCfYZkZ0rXxVcg@mail.gmail.com>
- In-Reply-To
- <25eec8d9676d4adf82570616ca88a2b38d20afa4.1793357550.git.gitgitgadget@gmail.com>
Hi Harald,
On Fri, Oct 30, 2026 at 6:52 AM Harald Nordgren via GitGitGadget <gitgitgadget@gmail.com> wrote:
Show 7 quoted lines
>
> From: Harald Nordgren <haraldnordgren@gmail.com>
>
> A branch pushed to a remote set up with "git remote add
> --limited-fetch" has no remote.<name>.fetch entry to map it
> through, so @{push} cannot tell where it landed and fails with
> "has no local tracking branch".I don't have the rest of the series paged in right now (esp. patches 4 & 5 which deal with --limited-fetch), so forgive me if this comes from ignorance.
Wouldn't the same problem happen if we applied the first 3 patches, then setup a repository's remote with a fetchRefmap and limited (or none) default fetch? If so, that almost makes this seem like a bugfix for an earlier patch, and perhaps it belongs together? It also would then be better justified in those terms, I think.
Such a reorganization would maybe let the early part of the series be mostly about fetchRefmap (which I'm keen on) and let the end be about --limited-fetch, which seems nice but I don't have a strong motivation to review right now (personally---it's just limited time to spend!).
Show 21 quoted lines
> Give tracking_for_push_dest() the same fallback to
> remote.<name>.fetchRefmap that get_ref_map() uses on the fetch
> side, so a --limited-fetch remote supports @{push} resolution
> the same way any other remote does.
>
> Signed-off-by: Harald Nordgren <haraldnordgren@gmail.com>
> ---
> remote.c | 2 ++
> t/t1514-rev-parse-push.sh | 19 +++++++++++++++++++
> 2 files changed, 21 insertions(+)
>
> diff --git a/remote.c b/remote.c
> index 15cd96aa8e..ecd104766c 100644
> --- a/remote.c
> +++ b/remote.c
> @@ -2018,6 +2018,8 @@ static char *tracking_for_push_dest(struct repository *repo,
>
> remote = repo_remote_for_push_tracking(repo, remote);
> ret = apply_refspecs(&remote->fetch, refname);
> + if (!ret)
> + ret = apply_refspecs(&remote->refmap, refname);Ok, if we don't find the information in the fetch spec, we fallback to the refmap. That also means the former overrides the latter in case both are configured.
Does that match what a fetch does if both are configured and overlap? (I'm not sure, as I can't recall what we decided the fetching behavior is!)
Show 30 quoted lines
> if (!ret)
> return error_buf(err,
> _("push destination '%s' on remote '%s' has no local tracking branch"),
> diff --git a/t/t1514-rev-parse-push.sh b/t/t1514-rev-parse-push.sh
> index 5a4f16867a..bcf00107f5 100755
> --- a/t/t1514-rev-parse-push.sh
> +++ b/t/t1514-rev-parse-push.sh
> @@ -80,4 +80,23 @@ test_expect_success 'resolving @{push} fails with a detached HEAD' '
> test_must_fail git rev-parse @{push}
> '
>
> +test_expect_success '@{push} resolves after pushing to a --limited-fetch remote' '
> + test_when_finished "rm -fr upstream fork downstream" &&
> + git init -b main upstream &&
> + test_commit -C upstream base &&
> + git init --bare fork &&
> + git clone --sparse upstream downstream &&
> + (
> + cd downstream &&
> + git remote add --limited-fetch fork ../fork &&
> + git config push.default current &&
> + git config remote.pushDefault fork &&
> + git checkout -b topic main &&
> + test_commit on-topic &&
> + git push &&
> + resolve topic@{push} refs/remotes/fork/topic &&
> + test_cmp_rev topic topic@{push}
> + )
> +'
> +If you take my suggestion to make this change with the early parts of the series, the test would probably need to be in terms of fetchRefmap instead of --limited-fetch, as well. That might be one reason to consider leaving it at the end, but… really the "problem" is the interaction of the earlier feature more than --limited-fetch, right?
HTH,
-- D. Ben Knoble