threads / patch / 66553

patchbranch: let inherit track a remote-tracking start point

Subject: [PATCH] branch: let inherit track a remote-tracking start point

## tl;dr

Today inherit leaves a branch created from origin/topic with no upstream. Junio Hamano asks whether the rule should be broader. Read the story.

replies: 7people: 4as markdown or json

Harald Nordgren via GitGitGadget· Oct 16, 2026, 07:02 UTC · lore
From: Harald Nordgren <haraldnordgren@gmail.com>

With branch.autoSetupMerge=inherit, "git switch topic" creates a local topic branch from origin/topic with no upstream, and warns that no remote is set. "git switch -c topic origin/topic" and an explicit "--track=inherit" do the same.

A remote-tracking branch has no upstream of its own to copy. Make inherit use it as the upstream directly, as the default "true" does, so that inherit sets up tracking in every case the default does.

Signed-off-by: Harald Nordgren <haraldnordgren@gmail.com>
---
    branch: let inherit track a remote-tracking start point
    
    With branch.autoSetupMerge=inherit, make git switch -c new_branch
    upstream/master work without a warning, and sets the remote branch
    upstream/master as the upstream.
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-git-2450%2FHaraldNordgren%2Fhn%2Fbranch-inherit-remote-tracking-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-git-2450/HaraldNordgren/hn/branch-inherit-remote-tracking-v1
Pull-Request: https://github.com/git/git/pull/2450
 Documentation/config/branch.adoc |  2 +-
 Documentation/git-branch.adoc    |  4 +++-
 branch.c                         |  3 +++
 t/t2060-switch.sh                | 17 +++++++++++++++++
 4 files changed, 24 insertions(+), 2 deletions(-)
Show changes to 4 files +24 −2

Documentation/config/branch.adoc, Documentation/git-branch.adoc, branch.c, t/t2060-switch.sh

diff --git a/Documentation/config/branch.adoc b/Documentation/config/branch.adoc
index c2acd89c99..6d9fbe5353 100644
--- a/Documentation/config/branch.adoc
+++ b/Documentation/config/branch.adoc
@@ -10,7 +10,7 @@
 `always`;; automatic setup is done when the starting point is either a
 	local branch or remote-tracking branch
 `inherit`;; if the starting point has a tracking configuration, it is copied to the new
-	branch
+	branch. A remote-tracking starting point is tracked directly, as with `true`.
 `simple`;; automatic setup is done only when the starting point
 	is a remote-tracking branch and the new branch has the same name as the
 	remote branch.
diff --git a/Documentation/git-branch.adoc b/Documentation/git-branch.adoc
index bfdf459329..238d99fd27 100644
--- a/Documentation/git-branch.adoc
+++ b/Documentation/git-branch.adoc
@@ -280,7 +280,9 @@ remove it anyway.
 The exact upstream branch is chosen depending on the optional argument:
 `-t`, `--track`, or `--track=direct` means to use the start-point branch
 itself as the upstream; `--track=inherit` means to copy the upstream
-configuration of the start-point branch.
+configuration of the start-point branch. A remote-tracking start-point
+has no upstream of its own, so `--track=inherit` uses it as the upstream
+directly.
 +
 The `branch.autoSetupMerge` configuration variable specifies how `git switch`,
 `git checkout` and `git branch` should behave when neither `--track` nor
diff --git a/branch.c b/branch.c
index 22f4f46b96..2c23179775 100644
--- a/branch.c
+++ b/branch.c
@@ -263,6 +263,9 @@ static void setup_tracking(const char *new_ref, const char *orig_ref,
 	if (!track)
 		BUG("asked to set up tracking, but tracking is disallowed");
 
+	if (track == BRANCH_TRACK_INHERIT && !starts_with(orig_ref, "refs/heads/"))
+		track = BRANCH_TRACK_REMOTE;
+
 	memset(&tracking, 0, sizeof(tracking));
 	tracking.spec.dst = (char *)orig_ref;
 	tracking.srcs = &tracking_srcs;
diff --git a/t/t2060-switch.sh b/t/t2060-switch.sh
index cec2fe6876..da9d01ab84 100755
--- a/t/t2060-switch.sh
+++ b/t/t2060-switch.sh
@@ -146,6 +146,23 @@ test_expect_success 'tracking info copied with autoSetupMerge=inherit' '
 	test_cmp_config "" --default "" branch.main2.merge
 '
 
+test_expect_success 'autoSetupMerge=inherit tracks a remote-tracking start point' '
+	git clone . downstream &&
+	(
+		cd downstream &&
+		git config branch.autoSetupMerge inherit &&
+		git switch -c new_branch origin/main &&
+		git status >../actual &&
+		cat >expect <<-EOF &&
+		On branch new_branch
+		Your branch is up to date with ${SQ}origin/main${SQ}.
+
+		nothing to commit, working tree clean
+		EOF
+		test_cmp expect ../actual
+	)
+'
+
 test_expect_success 'switch back when temporarily detached and checked out elsewhere ' '
 	test_when_finished "
 		git worktree remove wt1 ||:

base-commit: f57c9a73884400380bddd172acc88b569834f909
-- 
gitgitgadget
Junio C Hamano· Oct 16, 2026, 16:55 UTC · re: Harald Nordgren via GitGitGadget · lore

Re: [PATCH] branch: let inherit track a remote-tracking start point

"Harald Nordgren via GitGitGadget" <gitgitgadget@gmail.com> writes:
>  `inherit`;; if the starting point has a tracking configuration, it is copied to the new
> -	branch
> +	branch. A remote-tracking starting point is tracked directly, as with `true`.

That is "A remote-tracking branch given as a starting point", not "A remote-tracking starting point". Don't abbreviate and drop "branch".

If a starting point is not a remote-tracking branch, shouldn't the same thing happen, by the way? IOW, shouldn't the updated text and behaviour be more like

	`inherit`;; if the starting point has a tracking
	configuration, it is copied to the new branch.  Otherwise,
	the new branch is set to track the starting point itself.

I am not sure if it is the users of `inherit` want that forking a new branch from something that they know is *not* tracking anything suddenly starts to track the starting point (whether it is a remote-tracking branch or something else).

Looking at the originating commit d3115660b4 (branch: add flags and config to inherit tracking, 2021-12-20), it only talks about what they want to do when the starting point does track something, and forgets to say what they want when it does not track anything (the author CC'ed in case he retains the memory of the motivation from 5 years ago), but if you ask me _now_, not doing anything, i.e., the behaviour the old commit happened to have implemented without explanation does sound like the right thing to do.

What we should do if this change of behaviour turns out to be undesirable?

Perhaps `inherit` should gain a synonym `inherit-or-ignore` for those who prefer the current behaviour, and we want two extra behaviours, `inherit-or-fail` that complains and dies if the specified starting point does not have anything for the new branch to inherit, and `inherit-or-direct` that uses the fallback definition, or something?

Harald Nordgren· Oct 17, 2026, 09:54 UTC · re: Junio C Hamano · lore

Re: [PATCH] branch: let inherit track a remote-tracking start point

On Fri, Oct 16, 2026 at 6:55 PM Junio C Hamano <gitster@pobox.com> wrote:
Show 9 quoted lines
>
> "Harald Nordgren via GitGitGadget" <gitgitgadget@gmail.com> writes:
>
> >  `inherit`;; if the starting point has a tracking configuration, it is copied to the new
> > -     branch
> > +     branch. A remote-tracking starting point is tracked directly, as with `true`.
>
> That is "A remote-tracking branch given as a starting point", not "A
> remote-tracking starting point".  Don't abbreviate and drop "branch".
Thanks!
> I am not sure if it is the users of `inherit` want that forking a
> new branch from something that they know is *not* tracking anything
> suddenly starts to track the starting point (whether it is a
> remote-tracking branch or something else).
For me it's simple, only tracking when the starting point is a remote branch.
Show 18 quoted lines
> Looking at the originating commit d3115660b4 (branch: add flags and
> config to inherit tracking, 2021-12-20), it only talks about what
> they want to do when the starting point does track something, and
> forgets to say what they want when it does not track anything (the
> author CC'ed in case he retains the memory of the motivation from 5
> years ago), but if you ask me _now_, not doing anything, i.e., the
> behaviour the old commit happened to have implemented without
> explanation does sound like the right thing to do.
>
> What we should do if this change of behaviour turns out to be
> undesirable?
>
> Perhaps `inherit` should gain a synonym `inherit-or-ignore` for
> those who prefer the current behaviour, and we want two extra
> behaviours, `inherit-or-fail` that complains and dies if the
> specified starting point does not have anything for the new branch
> to inherit, and `inherit-or-direct` that uses the fallback
> definition, or something?

Seems quite overkill for the relatively small problem this is solving. But it would be interesting to hear the original author's POV.

Harald
Ben Knoble· Oct 19, 2026, 14:25 UTC · re: Harald Nordgren · lore

Re: [PATCH] branch: let inherit track a remote-tracking start point

Show 22 quoted lines
> Le 17 oct. 2026 à 05:55, Harald Nordgren <haraldnordgren@gmail.com> a écrit :
> 
> On Fri, Oct 16, 2026 at 6:55 PM Junio C Hamano <gitster@pobox.com> wrote:
>> Looking at the originating commit d3115660b4 (branch: add flags and
>> config to inherit tracking, 2021-12-20), it only talks about what
>> they want to do when the starting point does track something, and
>> forgets to say what they want when it does not track anything (the
>> author CC'ed in case he retains the memory of the motivation from 5
>> years ago), but if you ask me _now_, not doing anything, i.e., the
>> behaviour the old commit happened to have implemented without
>> explanation does sound like the right thing to do.
>> What we should do if this change of behaviour turns out to be
>> undesirable?
>> Perhaps `inherit` should gain a synonym `inherit-or-ignore` for
>> those who prefer the current behaviour, and we want two extra
>> behaviours, `inherit-or-fail` that complains and dies if the
>> specified starting point does not have anything for the new branch
>> to inherit, and `inherit-or-direct` that uses the fallback
>> definition, or something?
> 
> Seems quite overkill for the relatively small problem this is solving.
> But it would be interesting to hear the original author's POV.

Overkill, possibly, but easier to expand in a compatible way over time (other than piling new modifier --options on top, which makes the CLI confusing IMO).

Junio C Hamano· Oct 19, 2026, 20:06 UTC · re: Ben Knoble · lore

Re: [PATCH] branch: let inherit track a remote-tracking start point

Ben Knoble <ben.knoble@gmail.com> writes:
> Overkill, possibly, but easier to expand in a compatible way over time
> (other than piling new modifier --options on top, which makes the CLI confusing IMO).

Yes, exactly. A myopic initial design can easily paint us into a UI corner from which we cannot easily escape without awkward options. That is why I care.

Thanks.
Phillip Wood· Oct 18, 2026, 09:59 UTC · re: Junio C Hamano · lore

Re: [PATCH] branch: let inherit track a remote-tracking start point

On 16/10/2026 17:55, Junio C Hamano wrote:
Show 16 quoted lines
> "Harald Nordgren via GitGitGadget" <gitgitgadget@gmail.com> writes:
> >>   `inherit`;; if the starting point has a tracking configuration, it is copied to the new
>> -	branch
>> +	branch. A remote-tracking starting point is tracked directly, as with `true`.
> > That is "A remote-tracking branch given as a starting point", not "A
> remote-tracking starting point".  Don't abbreviate and drop "branch".
> > If a starting point is not a remote-tracking branch, shouldn't the
> same thing happen, by the way?  IOW, shouldn't the updated text and
> behaviour be more like
> > 	`inherit`;; if the starting point has a tracking
> 	configuration, it is copied to the new branch.  Otherwise,
> 	the new branch is set to track the starting point itself.
> > I am not sure if it is the users of `inherit` want that forking a
> new branch from something that they know is *not* tracking anything
> suddenly starts to track the starting point (whether it is a
> remote-tracking branch or something else).
For me it depends on what I'm forking from. If I create a branch on my desktop, push it to my server and then create a local branch from that on my laptop, I normally want the new branch to track something like origin/master rather than the starting point. However, if I create a new branch from origin/master then I do want to track the starting point. So it is not clear to me that the behavior proposed here is better than the current behavior, but equally I don't think it is any worse. Having some way of saying "set up tracking when the starting point matches one of these patterns" might be useful, but I find the tracking options to "git switch -c" so confusing I normally just run "git branch --set-upstream-to" afterwards to set the upstream I want.
Thanks
Phillip
Show 16 quoted lines
> Looking at the originating commit d3115660b4 (branch: add flags and
> config to inherit tracking, 2021-12-20), it only talks about what
> they want to do when the starting point does track something, and
> forgets to say what they want when it does not track anything (the
> author CC'ed in case he retains the memory of the motivation from 5
> years ago), but if you ask me _now_, not doing anything, i.e., the
> behaviour the old commit happened to have implemented without
> explanation does sound like the right thing to do.
> > What we should do if this change of behaviour turns out to be
> undesirable?
> > Perhaps `inherit` should gain a synonym `inherit-or-ignore` for
> those who prefer the current behaviour, and we want two extra
> behaviours, `inherit-or-fail` that complains and dies if the
> specified starting point does not have anything for the new branch
> to inherit, and `inherit-or-direct` that uses the fallback
> definition, or something?
Harald Nordgren· Oct 18, 2026, 15:19 UTC · re: Phillip Wood · lore

Re: [PATCH] branch: let inherit track a remote-tracking start point

Show 11 quoted lines
> For me it depends on what I'm forking from. If I create a branch on my
> desktop, push it to my server and then create a local branch from that
> on my laptop, I normally want the new branch to track something like
> origin/master rather than the starting point. However, if I create a new
> branch from origin/master then I do want to track the starting point. So
> it is not clear to me that the behavior proposed here is better than the
> current behavior, but equally I don't think it is any worse. Having some
> way of saying "set up tracking when the starting point matches one of
> these patterns" might be useful, but I find the tracking options to "git
> switch -c" so confusing I normally just run "git branch
> --set-upstream-to" afterwards to set the upstream I want.

I don't actually use 'git switch -c', I just used that in the example because it's the *modern* way to do things. In reality I always run 'git checkout -b'. I agree about `set-upstream-to`, there is probably no single terminal command that I run more than

    git branch --set-upstream-to=origin

Your example got me thinking. Maybe we should go the other way? If we can inherit from the branch we are coming from, should we do that as first prio? Because of the `--set-upstream-to=origin` usage, we will likely already be tracking the right one. So

    git checkout -b first_branch
    git branch --set-upstream-to=origin/master
    git checkout -b new_branch origin/another_branch

could give us a 'new_branch' that tracks what 'first_branch' tracked, i.e. 'origin/master'. Maybe as a fallback, we could have the new behavior from my v1, so if 'first_branch' has no tracking branch, then

    git checkout -b first_branch
    git checkout -b new_branch origin/another_branch

falls back to 'new_branch ' tracking 'origin/another_branch' instead? Or not, maybe this fallback is just bad?

Harald
Ben Knoble· Oct 19, 2026, 14:26 UTC · re: Harald Nordgren · lore

Re: [PATCH] branch: let inherit track a remote-tracking start point

Show 22 quoted lines
> Le 18 oct. 2026 à 11:20, Harald Nordgren <haraldnordgren@gmail.com> a écrit :
> 
> 
>> 
>> For me it depends on what I'm forking from. If I create a branch on my
>> desktop, push it to my server and then create a local branch from that
>> on my laptop, I normally want the new branch to track something like
>> origin/master rather than the starting point. However, if I create a new
>> branch from origin/master then I do want to track the starting point. So
>> it is not clear to me that the behavior proposed here is better than the
>> current behavior, but equally I don't think it is any worse. Having some
>> way of saying "set up tracking when the starting point matches one of
>> these patterns" might be useful, but I find the tracking options to "git
>> switch -c" so confusing I normally just run "git branch
>> --set-upstream-to" afterwards to set the upstream I want.
> 
> I don't actually use 'git switch -c', I just used that in the example
> because it's the *modern* way to do things. In reality I always run
> 'git checkout -b'. I agree about `set-upstream-to`, there is probably
> no single terminal command that I run more than
> 
>    git branch --set-upstream-to=origin
Curious. I typically do
    git switch -c topic origin

and am happy. (When creating a branch that *doesn’t* start at origin but still should track it, I use « branch -u » as shown, too.)

← back to recent threads