[PATCH] replay: support replaying down from root commit

[PATCH] replay: support replaying down from root commit

[Toon Claes]

git-replay(1) doesn't allow replaying commits all the way down to the
root commit. Fix that.

Signed-off-by: Toon Claes <toon@iotcl.com>
---
These changes might conflict Siddharth's series[1] to add '--revert' to
git-replay(1), although resolving that should be trivial.

[1]: https://lore.kernel.org/git/20260313054035.26605-1-siddharthasthana31@gmail.com/
---
 replay.c                 | 18 ++++++++++--------
 t/t3650-replay-basics.sh | 10 +++++++---
 2 files changed, 17 insertions(+), 11 deletions(-)

diff --git a/replay.c b/replay.c
index a63f6714c4..63ff56552e 100644
--- a/replay.c
+++ b/replay.c
@@ -225,12 +225,18 @@ static struct commit *pick_regular_commit(struct repository *repo,
 	struct commit *base, *replayed_base;
 	struct tree *pickme_tree, *base_tree, *replayed_base_tree;
 
-	base = pickme->parents->item;
-	replayed_base = mapped_commit(replayed_commits, base, onto);
+	if (pickme->parents) {
+		base = pickme->parents->item;
+		replayed_base = mapped_commit(replayed_commits, base, onto);
+		base_tree = repo_get_commit_tree(repo, base);
+	} else {
+		base = NULL;
+		replayed_base = onto;
+		base_tree = lookup_tree(repo, repo->hash_algo->empty_tree);
+	}
 
 	replayed_base_tree = repo_get_commit_tree(repo, replayed_base);
 	pickme_tree = repo_get_commit_tree(repo, pickme);
-	base_tree = repo_get_commit_tree(repo, base);
 
 	merge_opt->branch1 = short_commit_name(repo, replayed_base);
 	merge_opt->branch2 = short_commit_name(repo, pickme);
@@ -293,8 +299,6 @@ int replay_revisions(struct rev_info *revs,
 	set_up_replay_mode(revs->repo, &revs->cmdline, opts->onto,
 			   &detached_head, &advance, &onto, &update_refs);
 
-	/* FIXME: Should allow replaying commits with the first as a root commit */
-
 	if (prepare_revision_walk(revs) < 0) {
 		ret = error(_("error preparing revisions"));
 		goto out;
@@ -309,9 +313,7 @@ int replay_revisions(struct rev_info *revs,
 		khint_t pos;
 		int hr;
 
-		if (!commit->parents)
-			die(_("replaying down from root commit is not supported yet!"));
-		if (commit->parents->next)
+		if (commit->parents && commit->parents->next)
 			die(_("replaying merge commits is not supported yet!"));
 
 		last_commit = pick_regular_commit(revs->repo, commit, replayed_commits,
diff --git a/t/t3650-replay-basics.sh b/t/t3650-replay-basics.sh
index a03f8f9293..9c55b62757 100755
--- a/t/t3650-replay-basics.sh
+++ b/t/t3650-replay-basics.sh
@@ -81,9 +81,13 @@ test_expect_success 'option --onto or --advance is mandatory' '
 	test_cmp expect actual
 '
 
-test_expect_success 'no base or negative ref gives no-replaying down to root error' '
-	echo "fatal: replaying down from root commit is not supported yet!" >expect &&
-	test_must_fail git replay --onto=topic1 topic2 2>actual &&
+test_expect_success 'replay down to root onto another branch' '
+	git replay --ref-action=print --onto main topic2 >result &&
+
+	test_line_count = 1 result &&
+
+	git log --format=%s $(cut -f 3 -d " " result) >actual &&
+	test_write_lines E D C M L B A >expect &&
 	test_cmp expect actual
 '
 

---
base-commit: ca1db8a0f7dc0dbea892e99f5b37c5fe5861be71
change-id: 20260317-toon-replay-down-to-root-d412048f1741

Re: [PATCH] replay: support replaying down from root commit

[Junio C Hamano]

Toon Claes <toon@iotcl.com> writes:

> git-replay(1) doesn't allow replaying commits all the way down to the
> root commit. Fix that.

OK.

> -	base = pickme->parents->item;
> -	replayed_base = mapped_commit(replayed_commits, base, onto);
> +	if (pickme->parents) {
> +		base = pickme->parents->item;
> +		replayed_base = mapped_commit(replayed_commits, base, onto);
> +		base_tree = repo_get_commit_tree(repo, base);

So, if we are replaying a commit with parent(s), we do the same as
before (base_tree used to be computed a bit later).  But ...

> +	} else {
> +		base = NULL;
> +		replayed_base = onto;
> +		base_tree = lookup_tree(repo, repo->hash_algo->empty_tree);
> +	}

... if we are replaying the root commit, there is no base (in
contrast to "the first parent of the original commit" used in the
other branch of this if-else construct).  We use an empty tree for
the base_tree, which is the natural thing to use to replay for a
root commit, of course.

I am not sure why replayed_base is computed differently, though?  Is
it because mapped_commit() would not work when base==NULL?

I have to wonder if the handling of that case should also be
encapsulated inside mapped_commit() helper, just like the helper
knows to "fallback" when the commit is not yet mapped, but that is
minor.  After all, if we drive that line of thought to the extreme,
we would end up making repo_get_commit_tree() to return an empty
tree object for base==NULL, too, which may be logical but it is
probably too much.

Re: [PATCH] replay: support replaying down from root commit

[Toon Claes]

Junio C Hamano <gitster@pobox.com> writes:

> Toon Claes <toon@iotcl.com> writes:
>
>> git-replay(1) doesn't allow replaying commits all the way down to the
>> root commit. Fix that.
>
> OK.
>
>> -	base = pickme->parents->item;
>> -	replayed_base = mapped_commit(replayed_commits, base, onto);
>> +	if (pickme->parents) {
>> +		base = pickme->parents->item;
>> +		replayed_base = mapped_commit(replayed_commits, base, onto);
>> +		base_tree = repo_get_commit_tree(repo, base);
>
> So, if we are replaying a commit with parent(s), we do the same as
> before (base_tree used to be computed a bit later).  But ...
>
>> +	} else {
>> +		base = NULL;
>> +		replayed_base = onto;
>> +		base_tree = lookup_tree(repo, repo->hash_algo->empty_tree);
>> +	}
>
> ... if we are replaying the root commit, there is no base (in
> contrast to "the first parent of the original commit" used in the
> other branch of this if-else construct).  We use an empty tree for
> the base_tree, which is the natural thing to use to replay for a
> root commit, of course.
>
> I am not sure why replayed_base is computed differently, though?  Is
> it because mapped_commit() would not work when base==NULL?

That's correct, mapped_commit() dereferences commit->object.oid.

> I have to wonder if the handling of that case should also be
> encapsulated inside mapped_commit() helper, just like the helper
> knows to "fallback" when the commit is not yet mapped

I'm fine either way, so I'll do that in v2.

> After all, if we drive that line of thought to the extreme, we would
> end up making repo_get_commit_tree() to return an empty tree object
> for base==NULL, too, which may be logical but it is probably too much.

Well, it's a bit annoying `struct commit::parents` is a `struct
commit_list`, so we have to first check that pointer before we can
dereference `item`, so that guard do we need anyway. So I agree it's too
much.

One other thing, in v2 I'm also fixing the merge ancestor label. For
root commits it will say "empty tree" instead of "parent of <commit>",
so conflict messages aren't misleading.

-- 
Cheers,
Toon

[PATCH v2] replay: support replaying down from root commit

[Toon Claes]

git-replay(1) doesn't allow replaying commits all the way down to the
root commit. Fix that.

Signed-off-by: Toon Claes <toon@iotcl.com>
---
These changes might conflict Siddharth's series[1] to add '--revert' to
git-replay(1), although resolving that should be trivial.

[1]: https://lore.kernel.org/git/20260313054035.26605-1-siddharthasthana31@gmail.com/
---
Changes in v2:
- Add NULL pointer check for `commit` in mapped_commit().
- Change ancestor message to "empty tree" when replaying root commit.
- Link to v1: https://patch.msgid.link/20260317-toon-replay-down-to-root-v1-1-cb5c249e15fd@iotcl.com
---
 replay.c                 | 27 +++++++++++++++++----------
 t/t3650-replay-basics.sh | 10 +++++++---
 2 files changed, 24 insertions(+), 13 deletions(-)

diff --git a/replay.c b/replay.c
index a63f6714c4..92f2279156 100644
--- a/replay.c
+++ b/replay.c
@@ -209,7 +209,10 @@ static struct commit *mapped_commit(kh_oid_map_t *replayed_commits,
 				    struct commit *commit,
 				    struct commit *fallback)
 {
-	khint_t pos = kh_get_oid_map(replayed_commits, commit->object.oid);
+	khint_t pos;
+	if (!commit)
+		return fallback;
+	pos = kh_get_oid_map(replayed_commits, commit->object.oid);
 	if (pos == kh_end(replayed_commits))
 		return fallback;
 	return kh_value(replayed_commits, pos);
@@ -225,16 +228,24 @@ static struct commit *pick_regular_commit(struct repository *repo,
 	struct commit *base, *replayed_base;
 	struct tree *pickme_tree, *base_tree, *replayed_base_tree;
 
-	base = pickme->parents->item;
-	replayed_base = mapped_commit(replayed_commits, base, onto);
+	if (pickme->parents) {
+		base = pickme->parents->item;
+		base_tree = repo_get_commit_tree(repo, base);
+	} else {
+		base = NULL;
+		base_tree = lookup_tree(repo, repo->hash_algo->empty_tree);
+	}
 
+	replayed_base = mapped_commit(replayed_commits, base, onto);
 	replayed_base_tree = repo_get_commit_tree(repo, replayed_base);
 	pickme_tree = repo_get_commit_tree(repo, pickme);
-	base_tree = repo_get_commit_tree(repo, base);
 
 	merge_opt->branch1 = short_commit_name(repo, replayed_base);
 	merge_opt->branch2 = short_commit_name(repo, pickme);
-	merge_opt->ancestor = xstrfmt("parent of %s", merge_opt->branch2);
+	if (pickme->parents)
+		merge_opt->ancestor = xstrfmt("parent of %s", merge_opt->branch2);
+	else
+		merge_opt->ancestor = xstrdup("empty tree");
 
 	merge_incore_nonrecursive(merge_opt,
 				  base_tree,
@@ -293,8 +304,6 @@ int replay_revisions(struct rev_info *revs,
 	set_up_replay_mode(revs->repo, &revs->cmdline, opts->onto,
 			   &detached_head, &advance, &onto, &update_refs);
 
-	/* FIXME: Should allow replaying commits with the first as a root commit */
-
 	if (prepare_revision_walk(revs) < 0) {
 		ret = error(_("error preparing revisions"));
 		goto out;
@@ -309,9 +318,7 @@ int replay_revisions(struct rev_info *revs,
 		khint_t pos;
 		int hr;
 
-		if (!commit->parents)
-			die(_("replaying down from root commit is not supported yet!"));
-		if (commit->parents->next)
+		if (commit->parents && commit->parents->next)
 			die(_("replaying merge commits is not supported yet!"));
 
 		last_commit = pick_regular_commit(revs->repo, commit, replayed_commits,
diff --git a/t/t3650-replay-basics.sh b/t/t3650-replay-basics.sh
index a03f8f9293..9c55b62757 100755
--- a/t/t3650-replay-basics.sh
+++ b/t/t3650-replay-basics.sh
@@ -81,9 +81,13 @@ test_expect_success 'option --onto or --advance is mandatory' '
 	test_cmp expect actual
 '
 
-test_expect_success 'no base or negative ref gives no-replaying down to root error' '
-	echo "fatal: replaying down from root commit is not supported yet!" >expect &&
-	test_must_fail git replay --onto=topic1 topic2 2>actual &&
+test_expect_success 'replay down to root onto another branch' '
+	git replay --ref-action=print --onto main topic2 >result &&
+
+	test_line_count = 1 result &&
+
+	git log --format=%s $(cut -f 3 -d " " result) >actual &&
+	test_write_lines E D C M L B A >expect &&
 	test_cmp expect actual
 '
 

---
base-commit: ca1db8a0f7dc0dbea892e99f5b37c5fe5861be71
change-id: 20260317-toon-replay-down-to-root-d412048f1741

Re: [PATCH v2] replay: support replaying down from root commit

[Junio C Hamano]

Toon Claes <toon@iotcl.com> writes:

> git-replay(1) doesn't allow replaying commits all the way down to the
> root commit. Fix that.
>
> Signed-off-by: Toon Claes <toon@iotcl.com>
> ---
> These changes might conflict Siddharth's series[1] to add '--revert' to
> git-replay(1), although resolving that should be trivial.

True.  This round looks great to me.  Will queue.

Shall we mark the topic for 'next' now?

Thanks.

Re: [PATCH v2] replay: support replaying down from root commit

[Christian Couder]

On Tue, Mar 24, 2026 at 8:56 PM Junio C Hamano <gitster@pobox.com> wrote:
>
> Toon Claes <toon@iotcl.com> writes:
>
> > git-replay(1) doesn't allow replaying commits all the way down to the
> > root commit. Fix that.
> >
> > Signed-off-by: Toon Claes <toon@iotcl.com>
> > ---
> > These changes might conflict Siddharth's series[1] to add '--revert' to
> > git-replay(1), although resolving that should be trivial.
>
> True.  This round looks great to me.  Will queue.
>
> Shall we mark the topic for 'next' now?

The patch looks good to me, but I wonder if the docs should be updated
somehow, especially to try to avoid confusion in case users don't
properly specify a range.

For example, before this, `git replay --onto main topic` would fail,
but emit "fatal: replaying down from root commit is not supported
yet!". This would likely help users understand that they might need to
properly specify a range like "main..topic" instead of 'topic".

Now it would likely fail without any error message.

Maybe something like the following could help:

--- a/Documentation/git-replay.adoc
+++ b/Documentation/git-replay.adoc
@@ -23,6 +23,10 @@ instead get update commands that can be piped to
`git update-ref --stdin`

 THIS COMMAND IS EXPERIMENTAL. THE BEHAVIOR MAY CHANGE.

+Note that `git replay --onto main topic` replays the topic branch starting
+from the root commit, not from main. What you might want instead is
+`git replay --onto main main..topic`.
+
 OPTIONS
 -------

?

And yeah currently `git replay` is a plumbing command that most
regular users shouldn't likely use, but I think Elijah's goal was to
eventually make it user friendly enough for advanced users with
stacked branches.

Re: [PATCH v2] replay: support replaying down from root commit

[Ben Knoble]

> Le 25 mars 2026 à 06:05, Christian Couder <christian.couder@gmail.com> a écrit :
> 
> On Tue, Mar 24, 2026 at 8:56 PM Junio C Hamano <gitster@pobox.com> wrote:
>> 
>> Toon Claes <toon@iotcl.com> writes:
>> 
>>> git-replay(1) doesn't allow replaying commits all the way down to the
>>> root commit. Fix that.
>>> 
>>> Signed-off-by: Toon Claes <toon@iotcl.com>
>>> ---
>>> These changes might conflict Siddharth's series[1] to add '--revert' to
>>> git-replay(1), although resolving that should be trivial.
>> 
>> True.  This round looks great to me.  Will queue.
>> 
>> Shall we mark the topic for 'next' now?
> 
> The patch looks good to me, but I wonder if the docs should be updated
> somehow, especially to try to avoid confusion in case users don't
> properly specify a range.
> 
> For example, before this, `git replay --onto main topic` would fail,
> but emit "fatal: replaying down from root commit is not supported
> yet!". This would likely help users understand that they might need to
> properly specify a range like "main..topic" instead of 'topic".
> 
> Now it would likely fail without any error message.

Having used replay in a large monorepo where I juggle many branches (so rebasing another in-flight topic without otherwise interrupting my work is valuable), I’ve made this mistake a few times. Some way of handling it more gracefully would be appreciated: perhaps the root case is rare enough to warrant an option or confirmation prompt (when attached interactively)?

> 
> Maybe something like the following could help:
> 
> --- a/Documentation/git-replay.adoc
> +++ b/Documentation/git-replay.adoc
> @@ -23,6 +23,10 @@ instead get update commands that can be piped to
> `git update-ref --stdin`
> 
> THIS COMMAND IS EXPERIMENTAL. THE BEHAVIOR MAY CHANGE.
> 
> +Note that `git replay --onto main topic` replays the topic branch starting
> +from the root commit, not from main. What you might want instead is
> +`git replay --onto main main..topic`.
> +
> OPTIONS
> -------
> 
> ?
> 
> And yeah currently `git replay` is a plumbing command that most
> regular users shouldn't likely use, but I think Elijah's goal was to
> eventually make it user friendly enough for advanced users with
> stacked branches.
> 

Re: [PATCH v2] replay: support replaying down from root commit

[Toon Claes]

Christian Couder <christian.couder@gmail.com> writes:

> On Tue, Mar 24, 2026 at 8:56 PM Junio C Hamano <gitster@pobox.com> wrote:
>>
>> Toon Claes <toon@iotcl.com> writes:
>>
>> > git-replay(1) doesn't allow replaying commits all the way down to the
>> > root commit. Fix that.
>> >
>> > Signed-off-by: Toon Claes <toon@iotcl.com>
>> > ---
>> > These changes might conflict Siddharth's series[1] to add '--revert' to
>> > git-replay(1), although resolving that should be trivial.
>>
>> True.  This round looks great to me.  Will queue.
>>
>> Shall we mark the topic for 'next' now?
>
> The patch looks good to me, but I wonder if the docs should be updated
> somehow, especially to try to avoid confusion in case users don't
> properly specify a range.
>
> For example, before this, `git replay --onto main topic` would fail,
> but emit "fatal: replaying down from root commit is not supported
> yet!".

I'm fixing that in a separate series.

> This would likely help users understand that they might need to
> properly specify a range like "main..topic" instead of 'topic".
>
> Now it would likely fail without any error message.
>
> Maybe something like the following could help:
>
> --- a/Documentation/git-replay.adoc
> +++ b/Documentation/git-replay.adoc
> @@ -23,6 +23,10 @@ instead get update commands that can be piped to
> `git update-ref --stdin`
>
>  THIS COMMAND IS EXPERIMENTAL. THE BEHAVIOR MAY CHANGE.
>
> +Note that `git replay --onto main topic` replays the topic branch starting
> +from the root commit, not from main. What you might want instead is
> +`git replay --onto main main..topic`.
> +

Definitely would help, not sure it needs to be part of this series.

-- 
Cheers,
Toon

Re: Make git-replay(1) warn if revision-range isn't a range (was: Re: [PATCH v2] replay: support replaying down from root commit)

[Toon Claes]

Ben Knoble <ben.knoble@gmail.com> writes:

>> The patch looks good to me, but I wonder if the docs should be updated
>> somehow, especially to try to avoid confusion in case users don't
>> properly specify a range.
>> 
>> For example, before this, `git replay --onto main topic` would fail,
>> but emit "fatal: replaying down from root commit is not supported
>> yet!". This would likely help users understand that they might need to
>> properly specify a range like "main..topic" instead of 'topic".
>> 
>> Now it would likely fail without any error message.
>
> Having used replay in a large monorepo where I juggle many branches
> (so rebasing another in-flight topic without otherwise interrupting my
> work is valuable), I’ve made this mistake a few times. Some way of
> handling it more gracefully would be appreciated: perhaps the root
> case is rare enough to warrant an option or confirmation prompt (when
> attached interactively)?

Yeah, there is definetily room for improvement here. But I consider that
outside the scope of this series and more like #leftoverbits.

-- 
Cheers,
Toon

Re: [PATCH v2] replay: support replaying down from root commit

[Toon Claes]

Toon Claes <toon@iotcl.com> writes:

> I'm fixing that in a separate series.

LOL, whoops I'm mixing up my own series. Sorry about that.

-- 
Cheers,
Toon

Re: [PATCH v2] replay: support replaying down from root commit

[Junio C Hamano]

Christian Couder <christian.couder@gmail.com> writes:

> For example, before this, `git replay --onto main topic` would fail,
> but emit "fatal: replaying down from root commit is not supported
> yet!". This would likely help users understand that they might need to
> properly specify a range like "main..topic" instead of 'topic".
>
> Now it would likely fail without any error message.

The fact that I do not quite understand the suggested change is a
very strong sign that people would be helped by a bit more
documentation ;-).

Depending on what is in "topic" and what "main" has, wouldn't it be
possible that the history leads to "topic" replay cleanly on top of
"main"?  And if there are problems (e.g., the replayed history may
want to add a file where "main"'s history already has contents with
a different ancestry sitting there, or the path may be taken by a
directory), wouldn't "git replay" fail loudly explaining what got
conflicted, instead of failing silently?

> Maybe something like the following could help:
>
> --- a/Documentation/git-replay.adoc
> +++ b/Documentation/git-replay.adoc
> @@ -23,6 +23,10 @@ instead get update commands that can be piped to
> `git update-ref --stdin`
>
>  THIS COMMAND IS EXPERIMENTAL. THE BEHAVIOR MAY CHANGE.
>
> +Note that `git replay --onto main topic` replays the topic branch starting
> +from the root commit, not from main. What you might want instead is
> +`git replay --onto main main..topic`.

To be helpful, "might want" needs to be accompanied by "if you want
to do X", I think.  In this case, that X probably is "recreate what
was done on 'topic' since it forked from 'main' on top of 'main'".

Thanks.

Re: [PATCH v2] replay: support replaying down from root commit

[Junio C Hamano]

Toon Claes <toon@iotcl.com> writes:

>> Maybe something like the following could help:
>>
>> --- a/Documentation/git-replay.adoc
>> +++ b/Documentation/git-replay.adoc
>> @@ -23,6 +23,10 @@ instead get update commands that can be piped to
>> `git update-ref --stdin`
>>
>>  THIS COMMAND IS EXPERIMENTAL. THE BEHAVIOR MAY CHANGE.
>>
>> +Note that `git replay --onto main topic` replays the topic branch starting
>> +from the root commit, not from main. What you might want instead is
>> +`git replay --onto main main..topic`.
>> +
>
> Definitely would help, not sure it needs to be part of this series.

Where else should the patch to add such a note to the documentation
go, though?  Without this patch, we do not is because the command
will not take such a command line.  With this patch that adds the
"now we allow replay to take a single tip commit and replay the
history leading to the tip all the way down to root" feature, the
note may become relevant.

So to me, it looks like it is either we will never add such a note
because it is irrelevant and everybody should know the consequence
of passing "topic", not "main..topic", or we will have to add such a
note as part of the series (if the note would help the readers).

Even though I am on the fence about the need for this specific note
in the documentation, it does not make sense to me to say "this will
help but we are not doing so here".

My comment on "might" in "What you might" in the thread still
applies, by the way.

Thanks.