[PATCH 0/3] contrib/subtree: process out-of-prefix subtrees

[PATCH 0/3] contrib/subtree: process out-of-prefix subtrees

[Colin Stagner]

* cs/subtree-remove-optimization:
  Remove an optimization that can exclude too much
  history during a "subtree split."

Remove an optimization introduced in 98ba49ccc2 (subtree: fix split
processing with multiple subtrees present, 2023-12-01). The
optimization incorrectly excludes commits from the split history
that must be included.

The above commit, and my later attempt to fix it in 83f9dad7d6
(contrib/subtree: fix split with squashed subtrees, 2025-09-09),
both introduce known regressions.

Improve test coverage for `copy_or_skip`, which will help detect
future regressions.

Colin Stagner (3):
  contrib/subtree: capture additional test-cases
  contrib/subtree: test history depth
  contrib/subtree: process out-of-prefix subtrees

 contrib/subtree/git-subtree.sh     |  50 +-------
 contrib/subtree/t/t7900-subtree.sh | 198 +++++++++++++++++++++++++++--
 2 files changed, 190 insertions(+), 58 deletions(-)


base-commit: 852829b3dd2fe4e7c7fc4d8badde644cf1b66c74
-- 
2.43.0

[PATCH 1/3] contrib/subtree: capture additional test-cases

[Colin Stagner]

Patch series e7b07376e5 (Merge branch 'rs/subtree-fixes',
2018-10-26) corrects several defects in `git subtree split`.
The defects affect `split --rejoin` and merge commit processing.

There is no test coverage for this, and e7b07376e5 did not
introduce any.

Convert the minimum working example [1] from the original patch
submission [2] into test cases.

[1]: https://gist.github.com/FoxFireX/1b794384612b7fd5e7cd157cff96269e

[2]: <20180928183540.48968-1-roger.strain@swri.org>

Signed-off-by: Colin Stagner <ask+git@howdoi.land>
---
 contrib/subtree/t/t7900-subtree.sh | 110 +++++++++++++++++++++++++++++
 1 file changed, 110 insertions(+)

diff --git a/contrib/subtree/t/t7900-subtree.sh b/contrib/subtree/t/t7900-subtree.sh
index e7040718f2..3ee2f95d86 100755
--- a/contrib/subtree/t/t7900-subtree.sh
+++ b/contrib/subtree/t/t7900-subtree.sh
@@ -1575,6 +1575,116 @@ test_expect_success 'push split to subproj' '
 	)
 '
 
+# --ignore-joins must ignore mainline content outside of the
+# subtree. This test verifies that the logic in
+# `find_existing_splits()` correctly handles a `git subtree add`
+# In this test, the split history must not contain a commit titled
+#
+#     Add 'sub/' from commit ...
+#
+# see: dd21d43b58 (subtree: make --ignore-joins pay
+#      attention to adds, 2018-09-28)
+test_expect_success 'split --ignore-joins respects subtree add' '
+	subtree_test_create_repo "$test_count" &&
+	(
+		cd "$test_count" &&
+		test_commit main_must_not_be_in_subtree &&
+		test_create_subtree_add . mksubtree sub sub1 &&
+		test_commit sub/sub2 &&
+		test_commit main_must_not_be_in_subtree2 &&
+		git subtree split --prefix sub -b first_split --rejoin &&
+		test_commit sub/sub3 &&
+		no_ignore_joins="$(git subtree split --prefix sub -b no_ignore_joins)" &&
+		ignore_joins="$(git subtree split --prefix sub --ignore-joins -b ignore_joins)" &&
+		git checkout ignore_joins &&
+		test_path_is_file sub1.t &&
+		test_path_is_file sub2.t &&
+		test_path_is_file sub3.t &&
+		! test_path_is_file main_must_not_be_in_subtree.t &&
+		! test_path_is_file main_must_not_be_in_subtree2.t &&
+		test -z "$(git log -1 --grep "Add '''sub/''' from commit" ignore_joins)" &&
+		test "$no_ignore_joins" = "$ignore_joins" &&
+		test "$(git rev-list --count ignore_joins)" -eq 3;
+	)
+'
+
+# split excludes commits reachable from any previous --rejoin.
+# These ignored commits can still be the basis for new work
+# after the --rejoin. These commits must be processed, even
+# if they are excluded. Otherwise, the split history will be
+# incorrect.
+#
+# here, the merge
+#
+#     git merge --no-ff new_work_based_on_prejoin
+#
+# doesn't contain any subtree changes and so should not end
+# up in the split history. this subtree should be flat,
+# with no merges.
+#
+# see: 315a84f9aa (subtree: use commits before rejoins for
+#      splits, 2018-09-28)
+test_expect_success 'split links out-of-tree pre --rejoin commits with post --rejoin commits' '
+	subtree_test_create_repo "$test_count" &&
+	(
+		cd "$test_count" &&
+		test_commit main_must_not_be_in_subtree &&
+		mkdir sub &&
+		test_commit sub/sub1 &&
+		test_commit sub/sub2 &&
+		git subtree split --prefix sub --rejoin &&
+		test "$(git rev-list --count HEAD)" -eq 6 &&
+		git checkout sub/sub1 &&
+		git checkout -b new_work_based_on_prejoin &&
+		test_commit main_must_not_be_in_subtree2 &&
+		git checkout main &&
+		git merge --no-ff new_work_based_on_prejoin &&
+		test_commit sub/sub3 &&
+		git subtree split -d --prefix sub -b second_split &&
+		git checkout second_split &&
+		test_path_is_file sub1.t &&
+		test_path_is_file sub2.t &&
+		test_path_is_file sub3.t &&
+		! test_path_is_file main_must_not_be_in_subtree.t &&
+		! test_path_is_file main_must_not_be_in_subtree2.t &&
+		test "$(git rev-list --count --merges second_split)" -eq 0 &&
+		test "$(git rev-list --count second_split)" -eq 3;
+	)
+'
+
+# split must keep merge commits with unrelated histories, even
+# if both parents are treesame. When deciding whether or not
+# to eliminate a parent, copy_or_skip compares the merge-base
+# of each parent.
+#
+# in the split_of_merges branch:
+#
+#   * expect 4 commits
+#   * HEAD~ must be a merge
+#
+# see: 68f8ff8151 (subtree: improve decision on merges kept
+#      in split, 2018-09-28)
+test_expect_success 'split preserves merges with unrelated history' '
+	subtree_test_create_repo "$test_count" &&
+	(
+		cd "$test_count" &&
+		test_commit main_must_not_be_in_subtree &&
+		mkdir sub &&
+		test_commit sub/sub1 &&
+		git checkout --orphan new_history &&
+		git checkout sub/sub1 -- . &&
+		git add . &&
+		git commit -m "treesame history but not a merge-base" &&
+		git checkout main &&
+		git merge --allow-unrelated-histories --no-ff new_history &&
+		test "$(git rev-parse "HEAD^1^{tree}")" = "$(git rev-parse "HEAD^2^{tree}")" &&
+		test_commit sub/sub2 &&
+		git subtree split -d --prefix sub -b split_of_merges &&
+		test "$(git rev-list --count split_of_merges)" -eq 4 &&
+		test -n "$(git rev-list --merges HEAD~)";
+	)
+'
+
 #
 # This test covers 2 cases in subtree split copy_or_skip code
 # 1) Merges where one parent is a superset of the changes of the other
-- 
2.43.0

[PATCH 2/3] contrib/subtree: test history depth

[Colin Stagner]

Add history depth checks to some of the subtree unit tests.

These checks were previously introduced as part of 28a7e27cff
(contrib/subtree: detect rewritten subtree commits, 2026-01-09),
which has since been reverted.

Signed-off-by: Colin Stagner <ask+git@howdoi.land>
---
 contrib/subtree/t/t7900-subtree.sh | 22 ++++++++++++++++------
 1 file changed, 16 insertions(+), 6 deletions(-)

diff --git a/contrib/subtree/t/t7900-subtree.sh b/contrib/subtree/t/t7900-subtree.sh
index 3ee2f95d86..dad8dea63a 100755
--- a/contrib/subtree/t/t7900-subtree.sh
+++ b/contrib/subtree/t/t7900-subtree.sh
@@ -411,8 +411,9 @@ test_expect_success 'split sub dir/ with --rejoin' '
 		git fetch ./"sub proj" HEAD &&
 		git subtree merge --prefix="sub dir" FETCH_HEAD &&
 		split_hash=$(git subtree split --prefix="sub dir" --annotate="*") &&
-		git subtree split --prefix="sub dir" --annotate="*" --rejoin &&
-		test "$(last_commit_subject)" = "Split '\''sub dir/'\'' into commit '\''$split_hash'\''"
+		git subtree split --prefix="sub dir" --annotate="*" -b spl --rejoin &&
+		test "$(last_commit_subject)" = "Split '\''sub dir/'\'' into commit '\''$split_hash'\''" &&
+		test "$(git rev-list --count spl)" -eq 5
 	)
 '
 
@@ -442,18 +443,25 @@ test_expect_success 'split with multiple subtrees' '
 	git -C "$test_count" subtree add --prefix=subADir FETCH_HEAD &&
 	git -C "$test_count" fetch ./subB HEAD &&
 	git -C "$test_count" subtree add --prefix=subBDir FETCH_HEAD &&
+	test "$(git -C "$test_count" rev-list --count main)" -eq 7 &&
 	test_create_commit "$test_count" subADir/main-subA1 &&
 	test_create_commit "$test_count" subBDir/main-subB1 &&
 	git -C "$test_count" subtree split --prefix=subADir \
-		--squash --rejoin -m "Sub A Split 1" &&
+		--squash --rejoin -m "Sub A Split 1" -b a1 &&
+	test "$(git -C "$test_count" rev-list --count main..a1)" -eq 1 &&
 	git -C "$test_count" subtree split --prefix=subBDir \
-		--squash --rejoin -m "Sub B Split 1" &&
+		--squash --rejoin -m "Sub B Split 1" -b b1 &&
+	test "$(git -C "$test_count" rev-list --count main..b1)" -eq 1 &&
 	test_create_commit "$test_count" subADir/main-subA2 &&
 	test_create_commit "$test_count" subBDir/main-subB2 &&
 	git -C "$test_count" subtree split --prefix=subADir \
-		--squash --rejoin -m "Sub A Split 2" &&
+		--squash --rejoin -m "Sub A Split 2" -b a2 &&
+	test "$(git -C "$test_count" rev-list --count main..a2)" -eq 2 &&
+	test "$(git -C "$test_count" rev-list --count a1..a2)" -eq 1 &&
 	test "$(git -C "$test_count" subtree split --prefix=subBDir \
-		--squash --rejoin -d -m "Sub B Split 1" 2>&1 | grep -w "\[1\]")" = ""
+		--squash --rejoin -d -m "Sub B Split 1" -b b2 2>&1 | grep -w "\[1\]")" = "" &&
+	test "$(git -C "$test_count" rev-list --count main..b2)" -eq 2 &&
+	test "$(git -C "$test_count" rev-list --count b1..b2)" -eq 1
 '
 
 # When subtree split-ing a directory that has other subtree
@@ -477,6 +485,7 @@ do
 			test_path_is_file subA/file1.t &&
 			test_path_is_file subA/subB/file2.t &&
 			git subtree split --prefix=subA --branch=bsplit &&
+			test "$(git rev-list --count bsplit)" -eq 2 &&
 			git checkout bsplit &&
 			test_path_is_file file1.t &&
 			test_path_is_file subB/file2.t &&
@@ -489,6 +498,7 @@ do
 				--prefix=subA/subB mksubtree &&
 			test_path_is_file subA/subB/file3.t &&
 			git subtree split --prefix=subA --branch=bsplit &&
+			test "$(git rev-list --count bsplit)" -eq 3 &&
 			git checkout bsplit &&
 			test_path_is_file file1.t &&
 			test_path_is_file subB/file2.t &&
-- 
2.43.0

[PATCH 3/3] contrib/subtree: process out-of-prefix subtrees

[Colin Stagner]

`should_ignore_subtree_split_commit` detects subtrees which are
outside of the current path --prefix and ignores them. This can
speed up splits of repositories that have many subtrees.

Since its inception [1], every iteration of this logic [2], [3]
incorrectly excludes commits. This alters the split history. The
split history and its commit hashes are API contract, so this is
not permissible.

While a commit from a different subtree may look like it doesn't
contribute anything to a split, sometimes it does. Merge commits
are a particular hot spot. For these, the pruning logic in
`copy_or_skip` performs:

1. a check for "treesame" parents
2. two different common ancestry checks

These checks operate on the **split history**, not the input
history. The split history omits commits that do not affect the
--prefix. This can significantly alter the ancestry of a merge.
In order to determine if `copy_or_skip` will skip a merge, it
is likely necessary to compute all the split history... which
is what `should_ignore_subtree_split_commit` tries to avoid.

To make this logic API-preserving, we could gate it behind a
new CLI argument. The present implementation is actually a
speed penalty in many cases, however, so this is not done here.

Remove the `should_ignore_subtree_split_commit` logic. This
fixes the regression reported in [4].

[1]: 98ba49ccc2 (subtree: fix split processing with multiple subtrees present, 2023-12-01)

[2]: 83f9dad7d6 (contrib/subtree: fix split with squashed subtrees, 2025-09-09)

[3]: 28a7e27cff (contrib/subtree: detect rewritten subtree commits, 2026-01-09)

[4]: <20251230170719.845029-1-george@mail.dietrich.pub>

Reported-by: George <george@mail.dietrich.pub>

Reported-by: Christian Heusel <christian@heusel.eu>

Signed-off-by: Colin Stagner <ask+git@howdoi.land>
---
 contrib/subtree/git-subtree.sh     | 50 +---------------------
 contrib/subtree/t/t7900-subtree.sh | 68 ++++++++++++++++++++++++++++--
 2 files changed, 65 insertions(+), 53 deletions(-)

diff --git a/contrib/subtree/git-subtree.sh b/contrib/subtree/git-subtree.sh
index 17106d1a72..ba9fb2ee5d 100755
--- a/contrib/subtree/git-subtree.sh
+++ b/contrib/subtree/git-subtree.sh
@@ -785,42 +785,6 @@ ensure_valid_ref_format () {
 		die "fatal: '$1' does not look like a ref"
 }
 
-# Usage: should_ignore_subtree_split_commit REV
-#
-# Check if REV is a commit from another subtree and should be
-# ignored from processing for splits
-should_ignore_subtree_split_commit () {
-	assert test $# = 1
-
-	git show \
-		--no-patch \
-		--no-show-signature \
-		--format='%(trailers:key=git-subtree-dir,key=git-subtree-mainline)' \
-		"$1" |
-	(
-	have_mainline=
-	subtree_dir=
-
-	while read -r trailer val
-	do
-		case "$trailer" in
-		git-subtree-dir:)
-			subtree_dir="${val%/}" ;;
-		git-subtree-mainline:)
-			have_mainline=y ;;
-		esac
-	done
-
-	if test -n "${subtree_dir}" &&
-		test -z "${have_mainline}" &&
-		test "${subtree_dir}" != "$arg_prefix"
-	then
-		return 0
-	fi
-	return 1
-	)
-}
-
 # Usage: process_split_commit REV PARENTS
 process_split_commit () {
 	assert test $# = 2
@@ -1006,19 +970,7 @@ cmd_split () {
 	eval "$grl" |
 	while read rev parents
 	do
-		if should_ignore_subtree_split_commit "$rev"
-		then
-			continue
-		fi
-		parsedparents=''
-		for parent in $parents
-		do
-			if ! should_ignore_subtree_split_commit "$parent"
-			then
-				parsedparents="$parsedparents$parent "
-			fi
-		done
-		process_split_commit "$rev" "$parsedparents"
+		process_split_commit "$rev" "$parents"
 	done || exit $?
 
 	latest_new=$(cache_get latest_new) || exit $?
diff --git a/contrib/subtree/t/t7900-subtree.sh b/contrib/subtree/t/t7900-subtree.sh
index dad8dea63a..05a774ad47 100755
--- a/contrib/subtree/t/t7900-subtree.sh
+++ b/contrib/subtree/t/t7900-subtree.sh
@@ -428,8 +428,7 @@ test_expect_success 'split sub dir/ with --rejoin' '
 # 	- Perform 'split' on subtree B
 # 	- Create new commits with changes to subtree A and B
 # 	- Perform split on subtree A
-# 	- Check that the commits in subtree B are not processed
-#			as part of the subtree A split
+# 	- Check for expected history
 test_expect_success 'split with multiple subtrees' '
 	subtree_test_create_repo "$test_count" &&
 	subtree_test_create_repo "$test_count/subA" &&
@@ -458,8 +457,8 @@ test_expect_success 'split with multiple subtrees' '
 		--squash --rejoin -m "Sub A Split 2" -b a2 &&
 	test "$(git -C "$test_count" rev-list --count main..a2)" -eq 2 &&
 	test "$(git -C "$test_count" rev-list --count a1..a2)" -eq 1 &&
-	test "$(git -C "$test_count" subtree split --prefix=subBDir \
-		--squash --rejoin -d -m "Sub B Split 1" -b b2 2>&1 | grep -w "\[1\]")" = "" &&
+	git -C "$test_count" subtree split --prefix=subBDir \
+		--squash --rejoin -d -m "Sub B Split 1" -b b2 &&
 	test "$(git -C "$test_count" rev-list --count main..b2)" -eq 2 &&
 	test "$(git -C "$test_count" rev-list --count b1..b2)" -eq 1
 '
@@ -507,6 +506,67 @@ do
 	'
 done
 
+# Usually,
+#
+#    git subtree merge -P subA --squash f00...
+#
+# makes two commits, in this order:
+#
+# 1. Squashed 'subA/' content from commit f00...
+# 2. Merge commit (1) as 'subA'
+#
+# Commit 1 updates the subtree but does *not* rewrite paths.
+# Commit 2 rewrites all trees to start with `subA/`
+#
+# Commit 1 either has no parents or depends only on other
+# "Squashed 'subA/' content" commits.
+#
+# For merge without --squash, subtree produces just one commit:
+# a merge commit with git-subtree trailers.
+#
+# In either case, if the user rebases these commits, they will
+# still have the git-subtree-* trailers… but will NOT have
+# the layout described above.
+#
+# Test that subsequent `git subtree split` are not confused by this.
+test_expect_success 'split with rebased subtree commit' '
+	subtree_test_create_repo "$test_count" &&
+	(
+		cd "$test_count" &&
+		test_commit file0 &&
+		test_create_subtree_add \
+			. mksubtree subA file1 --squash &&
+		test_path_is_file subA/file1.t &&
+		mkdir subB &&
+		test_commit subB/bfile &&
+		git commit --amend -F - <<'EOF' &&
+Squashed '\''subB/'\'' content from commit '\''badf00da911bbe895347b4b236f5461d55dc9877'\''
+
+Simulate a cherry-picked or rebased subtree commit.
+
+git-subtree-dir: subB
+git-subtree-split: badf00da911bbe895347b4b236f5461d55dc9877
+EOF
+		test_commit subA/file2 &&
+		test_commit subB/bfile2 &&
+		git commit --amend -F - <<'EOF' &&
+Split '\''subB/'\'' into commit '\''badf00da911bbe895347b4b236f5461d55dc9877'\''
+
+Simulate a cherry-picked or rebased subtree commit.
+
+git-subtree-dir: subB
+git-subtree-mainline: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+git-subtree-split: badf00da911bbe895347b4b236f5461d55dc9877
+EOF
+		git subtree split --prefix=subA --branch=bsplit &&
+		git checkout bsplit &&
+		test_path_is_file file1.t &&
+		test_path_is_file file2.t &&
+		test "$(last_commit_subject)" = "subA/file2" &&
+		test "$(git rev-list --count bsplit)" -eq 2
+	)
+'
+
 test_expect_success 'split sub dir/ with --rejoin from scratch' '
 	subtree_test_create_repo "$test_count" &&
 	test_create_commit "$test_count" main1 &&
-- 
2.43.0

Re: [PATCH 3/3] contrib/subtree: process out-of-prefix subtrees

[D. Ben Knoble]

On Mon, Feb 16, 2026 at 4:16 PM Colin Stagner <ask+git@howdoi.land> wrote:
>
> `should_ignore_subtree_split_commit` detects subtrees which are
> outside of the current path --prefix and ignores them. This can
> speed up splits of repositories that have many subtrees.
>
> Since its inception [1], every iteration of this logic [2], [3]
> incorrectly excludes commits. This alters the split history. The
> split history and its commit hashes are API contract, so this is
> not permissible.
>
> While a commit from a different subtree may look like it doesn't
> contribute anything to a split, sometimes it does. Merge commits
> are a particular hot spot. For these, the pruning logic in
> `copy_or_skip` performs:
>
> 1. a check for "treesame" parents
> 2. two different common ancestry checks
>
> These checks operate on the **split history**, not the input
> history. The split history omits commits that do not affect the
> --prefix. This can significantly alter the ancestry of a merge.
> In order to determine if `copy_or_skip` will skip a merge, it
> is likely necessary to compute all the split history... which
> is what `should_ignore_subtree_split_commit` tries to avoid.
>
> To make this logic API-preserving, we could gate it behind a
> new CLI argument. The present implementation is actually a
> speed penalty in many cases, however, so this is not done here.
>
> Remove the `should_ignore_subtree_split_commit` logic. This
> fixes the regression reported in [4].
>
> [1]: 98ba49ccc2 (subtree: fix split processing with multiple subtrees present, 2023-12-01)
>
> [2]: 83f9dad7d6 (contrib/subtree: fix split with squashed subtrees, 2025-09-09)
>
> [3]: 28a7e27cff (contrib/subtree: detect rewritten subtree commits, 2026-01-09)
>
> [4]: <20251230170719.845029-1-george@mail.dietrich.pub>
>
> Reported-by: George <george@mail.dietrich.pub>
>
> Reported-by: Christian Heusel <christian@heusel.eu>
>
> Signed-off-by: Colin Stagner <ask+git@howdoi.land>

nit: "git interpret-trailers --parse" doesn't understand
line-separated trailers, so only finds the SoB line. See
git-interpret-trailers(1).

(A useful test locally was something like

    b4 am https://lore.kernel.org/git/20260215201906.889951-4-ask+git@howdoi.land/
    git mailsplit -op
20260215_ask_git_contrib_subtree_process_out_of_prefix_subtrees.mbx
    git interpret-trailers --parse p/0003

which output only

    Signed-off-by: Colin Stagner <ask+git@howdoi.land>

which I don't think you intended.)

-- 
D. Ben Knoble

Re: [PATCH 3/3] contrib/subtree: process out-of-prefix subtrees

[Colin Stagner]

On 2/16/26 15:33, D. Ben Knoble wrote:
> On Mon, Feb 16, 2026 at 4:16 PM Colin Stagner <ask+git@howdoi.land> wrote:
>>
>> Reported-by: George <george@mail.dietrich.pub>
>>
>> Reported-by: Christian Heusel <christian@heusel.eu>
>>
>> Signed-off-by: Colin Stagner <ask+git@howdoi.land>
> 
> nit: "git interpret-trailers --parse" doesn't understand
> line-separated trailers

Will fix in v2.

And I think I'll give b4 a shot, while I'm at it.

[PATCH v2 0/3] contrib/subtree: process out-of-prefix subtrees

[Colin Stagner]

* cs/subtree-remove-optimization:
  Remove an optimization that can exclude too much
  history during a "subtree split."

Remove an optimization introduced in 98ba49ccc2 (subtree: fix split
processing with multiple subtrees present, 2023-12-01). The
optimization incorrectly excludes commits from the split history
that must be included.

The above commit, and my later attempt to fix it in 83f9dad7d6
(contrib/subtree: fix split with squashed subtrees, 2025-09-09),
both introduce known regressions.

Improve test coverage for `copy_or_skip`, which will help detect
future regressions.

Signed-off-by: Colin Stagner <ask+git@howdoi.land>
---
Changes in v2:
- remove incorrect newlines from commit message trailers
- patch diffs are unchanged

---
Colin Stagner (3):
      contrib/subtree: capture additional test-cases
      contrib/subtree: test history depth
      contrib/subtree: process out-of-prefix subtrees

 contrib/subtree/git-subtree.sh     |  50 +---------
 contrib/subtree/t/t7900-subtree.sh | 198 +++++++++++++++++++++++++++++++++++--
 2 files changed, 190 insertions(+), 58 deletions(-)
---
base-commit: 852829b3dd2fe4e7c7fc4d8badde644cf1b66c74
change-id: 20260217-cs-subtree-remove-optimization-aca442f748eb

Best regards,
--  
Colin Stagner <ask+git@howdoi.land>

[PATCH v2 1/3] contrib/subtree: capture additional test-cases

[Colin Stagner]

Patch series e7b07376e5 (Merge branch 'rs/subtree-fixes',
2018-10-26) corrects several defects in `git subtree split`.
The defects affect `split --rejoin` and merge commit processing.

There is no test coverage for this, and e7b07376e5 did not
introduce any.

Convert the minimum working example [1] from the original patch
submission [2] into test cases.

[1]: https://gist.github.com/FoxFireX/1b794384612b7fd5e7cd157cff96269e

[2]: <20180928183540.48968-1-roger.strain@swri.org>

Signed-off-by: Colin Stagner <ask+git@howdoi.land>
---
 contrib/subtree/t/t7900-subtree.sh | 110 +++++++++++++++++++++++++++++++++++++
 1 file changed, 110 insertions(+)

diff --git a/contrib/subtree/t/t7900-subtree.sh b/contrib/subtree/t/t7900-subtree.sh
index e7040718f2..3ee2f95d86 100755
--- a/contrib/subtree/t/t7900-subtree.sh
+++ b/contrib/subtree/t/t7900-subtree.sh
@@ -1575,6 +1575,116 @@ test_expect_success 'push split to subproj' '
 	)
 '
 
+# --ignore-joins must ignore mainline content outside of the
+# subtree. This test verifies that the logic in
+# `find_existing_splits()` correctly handles a `git subtree add`
+# In this test, the split history must not contain a commit titled
+#
+#     Add 'sub/' from commit ...
+#
+# see: dd21d43b58 (subtree: make --ignore-joins pay
+#      attention to adds, 2018-09-28)
+test_expect_success 'split --ignore-joins respects subtree add' '
+	subtree_test_create_repo "$test_count" &&
+	(
+		cd "$test_count" &&
+		test_commit main_must_not_be_in_subtree &&
+		test_create_subtree_add . mksubtree sub sub1 &&
+		test_commit sub/sub2 &&
+		test_commit main_must_not_be_in_subtree2 &&
+		git subtree split --prefix sub -b first_split --rejoin &&
+		test_commit sub/sub3 &&
+		no_ignore_joins="$(git subtree split --prefix sub -b no_ignore_joins)" &&
+		ignore_joins="$(git subtree split --prefix sub --ignore-joins -b ignore_joins)" &&
+		git checkout ignore_joins &&
+		test_path_is_file sub1.t &&
+		test_path_is_file sub2.t &&
+		test_path_is_file sub3.t &&
+		! test_path_is_file main_must_not_be_in_subtree.t &&
+		! test_path_is_file main_must_not_be_in_subtree2.t &&
+		test -z "$(git log -1 --grep "Add '''sub/''' from commit" ignore_joins)" &&
+		test "$no_ignore_joins" = "$ignore_joins" &&
+		test "$(git rev-list --count ignore_joins)" -eq 3;
+	)
+'
+
+# split excludes commits reachable from any previous --rejoin.
+# These ignored commits can still be the basis for new work
+# after the --rejoin. These commits must be processed, even
+# if they are excluded. Otherwise, the split history will be
+# incorrect.
+#
+# here, the merge
+#
+#     git merge --no-ff new_work_based_on_prejoin
+#
+# doesn't contain any subtree changes and so should not end
+# up in the split history. this subtree should be flat,
+# with no merges.
+#
+# see: 315a84f9aa (subtree: use commits before rejoins for
+#      splits, 2018-09-28)
+test_expect_success 'split links out-of-tree pre --rejoin commits with post --rejoin commits' '
+	subtree_test_create_repo "$test_count" &&
+	(
+		cd "$test_count" &&
+		test_commit main_must_not_be_in_subtree &&
+		mkdir sub &&
+		test_commit sub/sub1 &&
+		test_commit sub/sub2 &&
+		git subtree split --prefix sub --rejoin &&
+		test "$(git rev-list --count HEAD)" -eq 6 &&
+		git checkout sub/sub1 &&
+		git checkout -b new_work_based_on_prejoin &&
+		test_commit main_must_not_be_in_subtree2 &&
+		git checkout main &&
+		git merge --no-ff new_work_based_on_prejoin &&
+		test_commit sub/sub3 &&
+		git subtree split -d --prefix sub -b second_split &&
+		git checkout second_split &&
+		test_path_is_file sub1.t &&
+		test_path_is_file sub2.t &&
+		test_path_is_file sub3.t &&
+		! test_path_is_file main_must_not_be_in_subtree.t &&
+		! test_path_is_file main_must_not_be_in_subtree2.t &&
+		test "$(git rev-list --count --merges second_split)" -eq 0 &&
+		test "$(git rev-list --count second_split)" -eq 3;
+	)
+'
+
+# split must keep merge commits with unrelated histories, even
+# if both parents are treesame. When deciding whether or not
+# to eliminate a parent, copy_or_skip compares the merge-base
+# of each parent.
+#
+# in the split_of_merges branch:
+#
+#   * expect 4 commits
+#   * HEAD~ must be a merge
+#
+# see: 68f8ff8151 (subtree: improve decision on merges kept
+#      in split, 2018-09-28)
+test_expect_success 'split preserves merges with unrelated history' '
+	subtree_test_create_repo "$test_count" &&
+	(
+		cd "$test_count" &&
+		test_commit main_must_not_be_in_subtree &&
+		mkdir sub &&
+		test_commit sub/sub1 &&
+		git checkout --orphan new_history &&
+		git checkout sub/sub1 -- . &&
+		git add . &&
+		git commit -m "treesame history but not a merge-base" &&
+		git checkout main &&
+		git merge --allow-unrelated-histories --no-ff new_history &&
+		test "$(git rev-parse "HEAD^1^{tree}")" = "$(git rev-parse "HEAD^2^{tree}")" &&
+		test_commit sub/sub2 &&
+		git subtree split -d --prefix sub -b split_of_merges &&
+		test "$(git rev-list --count split_of_merges)" -eq 4 &&
+		test -n "$(git rev-list --merges HEAD~)";
+	)
+'
+
 #
 # This test covers 2 cases in subtree split copy_or_skip code
 # 1) Merges where one parent is a superset of the changes of the other

-- 
2.43.0

[PATCH v2 2/3] contrib/subtree: test history depth

[Colin Stagner]

Add history depth checks to some of the subtree unit tests.

These checks were previously introduced as part of 28a7e27cff
(contrib/subtree: detect rewritten subtree commits, 2026-01-09),
which has since been reverted.

Signed-off-by: Colin Stagner <ask+git@howdoi.land>
---
 contrib/subtree/t/t7900-subtree.sh | 22 ++++++++++++++++------
 1 file changed, 16 insertions(+), 6 deletions(-)

diff --git a/contrib/subtree/t/t7900-subtree.sh b/contrib/subtree/t/t7900-subtree.sh
index 3ee2f95d86..dad8dea63a 100755
--- a/contrib/subtree/t/t7900-subtree.sh
+++ b/contrib/subtree/t/t7900-subtree.sh
@@ -411,8 +411,9 @@ test_expect_success 'split sub dir/ with --rejoin' '
 		git fetch ./"sub proj" HEAD &&
 		git subtree merge --prefix="sub dir" FETCH_HEAD &&
 		split_hash=$(git subtree split --prefix="sub dir" --annotate="*") &&
-		git subtree split --prefix="sub dir" --annotate="*" --rejoin &&
-		test "$(last_commit_subject)" = "Split '\''sub dir/'\'' into commit '\''$split_hash'\''"
+		git subtree split --prefix="sub dir" --annotate="*" -b spl --rejoin &&
+		test "$(last_commit_subject)" = "Split '\''sub dir/'\'' into commit '\''$split_hash'\''" &&
+		test "$(git rev-list --count spl)" -eq 5
 	)
 '
 
@@ -442,18 +443,25 @@ test_expect_success 'split with multiple subtrees' '
 	git -C "$test_count" subtree add --prefix=subADir FETCH_HEAD &&
 	git -C "$test_count" fetch ./subB HEAD &&
 	git -C "$test_count" subtree add --prefix=subBDir FETCH_HEAD &&
+	test "$(git -C "$test_count" rev-list --count main)" -eq 7 &&
 	test_create_commit "$test_count" subADir/main-subA1 &&
 	test_create_commit "$test_count" subBDir/main-subB1 &&
 	git -C "$test_count" subtree split --prefix=subADir \
-		--squash --rejoin -m "Sub A Split 1" &&
+		--squash --rejoin -m "Sub A Split 1" -b a1 &&
+	test "$(git -C "$test_count" rev-list --count main..a1)" -eq 1 &&
 	git -C "$test_count" subtree split --prefix=subBDir \
-		--squash --rejoin -m "Sub B Split 1" &&
+		--squash --rejoin -m "Sub B Split 1" -b b1 &&
+	test "$(git -C "$test_count" rev-list --count main..b1)" -eq 1 &&
 	test_create_commit "$test_count" subADir/main-subA2 &&
 	test_create_commit "$test_count" subBDir/main-subB2 &&
 	git -C "$test_count" subtree split --prefix=subADir \
-		--squash --rejoin -m "Sub A Split 2" &&
+		--squash --rejoin -m "Sub A Split 2" -b a2 &&
+	test "$(git -C "$test_count" rev-list --count main..a2)" -eq 2 &&
+	test "$(git -C "$test_count" rev-list --count a1..a2)" -eq 1 &&
 	test "$(git -C "$test_count" subtree split --prefix=subBDir \
-		--squash --rejoin -d -m "Sub B Split 1" 2>&1 | grep -w "\[1\]")" = ""
+		--squash --rejoin -d -m "Sub B Split 1" -b b2 2>&1 | grep -w "\[1\]")" = "" &&
+	test "$(git -C "$test_count" rev-list --count main..b2)" -eq 2 &&
+	test "$(git -C "$test_count" rev-list --count b1..b2)" -eq 1
 '
 
 # When subtree split-ing a directory that has other subtree
@@ -477,6 +485,7 @@ do
 			test_path_is_file subA/file1.t &&
 			test_path_is_file subA/subB/file2.t &&
 			git subtree split --prefix=subA --branch=bsplit &&
+			test "$(git rev-list --count bsplit)" -eq 2 &&
 			git checkout bsplit &&
 			test_path_is_file file1.t &&
 			test_path_is_file subB/file2.t &&
@@ -489,6 +498,7 @@ do
 				--prefix=subA/subB mksubtree &&
 			test_path_is_file subA/subB/file3.t &&
 			git subtree split --prefix=subA --branch=bsplit &&
+			test "$(git rev-list --count bsplit)" -eq 3 &&
 			git checkout bsplit &&
 			test_path_is_file file1.t &&
 			test_path_is_file subB/file2.t &&

-- 
2.43.0

[PATCH v2 3/3] contrib/subtree: process out-of-prefix subtrees

[Colin Stagner]

`should_ignore_subtree_split_commit` detects subtrees which are
outside of the current path --prefix and ignores them. This can
speed up splits of repositories that have many subtrees.

Since its inception [1], every iteration of this logic [2], [3]
incorrectly excludes commits. This alters the split history. The
split history and its commit hashes are API contract, so this is
not permissible.

While a commit from a different subtree may look like it doesn't
contribute anything to a split, sometimes it does. Merge commits
are a particular hot spot. For these, the pruning logic in
`copy_or_skip` performs:

1. a check for "treesame" parents
2. two different common ancestry checks

These checks operate on the **split history**, not the input
history. The split history omits commits that do not affect the
--prefix. This can significantly alter the ancestry of a merge.
In order to determine if `copy_or_skip` will skip a merge, it
is likely necessary to compute all the split history... which
is what `should_ignore_subtree_split_commit` tries to avoid.

To make this logic API-preserving, we could gate it behind a
new CLI argument. The present implementation is actually a
speed penalty in many cases, however, so this is not done here.

Remove the `should_ignore_subtree_split_commit` logic. This
fixes the regression reported in [4].

[1]: 98ba49ccc2 (subtree: fix split processing with multiple subtrees present, 2023-12-01)

[2]: 83f9dad7d6 (contrib/subtree: fix split with squashed subtrees, 2025-09-09)

[3]: 28a7e27cff (contrib/subtree: detect rewritten subtree commits, 2026-01-09)

[4]: <20251230170719.845029-1-george@mail.dietrich.pub>

Reported-by: George <george@mail.dietrich.pub>
Reported-by: Christian Heusel <christian@heusel.eu>
Signed-off-by: Colin Stagner <ask+git@howdoi.land>
---
 contrib/subtree/git-subtree.sh     | 50 +---------------------------
 contrib/subtree/t/t7900-subtree.sh | 68 +++++++++++++++++++++++++++++++++++---
 2 files changed, 65 insertions(+), 53 deletions(-)

diff --git a/contrib/subtree/git-subtree.sh b/contrib/subtree/git-subtree.sh
index 17106d1a72..ba9fb2ee5d 100755
--- a/contrib/subtree/git-subtree.sh
+++ b/contrib/subtree/git-subtree.sh
@@ -785,42 +785,6 @@ ensure_valid_ref_format () {
 		die "fatal: '$1' does not look like a ref"
 }
 
-# Usage: should_ignore_subtree_split_commit REV
-#
-# Check if REV is a commit from another subtree and should be
-# ignored from processing for splits
-should_ignore_subtree_split_commit () {
-	assert test $# = 1
-
-	git show \
-		--no-patch \
-		--no-show-signature \
-		--format='%(trailers:key=git-subtree-dir,key=git-subtree-mainline)' \
-		"$1" |
-	(
-	have_mainline=
-	subtree_dir=
-
-	while read -r trailer val
-	do
-		case "$trailer" in
-		git-subtree-dir:)
-			subtree_dir="${val%/}" ;;
-		git-subtree-mainline:)
-			have_mainline=y ;;
-		esac
-	done
-
-	if test -n "${subtree_dir}" &&
-		test -z "${have_mainline}" &&
-		test "${subtree_dir}" != "$arg_prefix"
-	then
-		return 0
-	fi
-	return 1
-	)
-}
-
 # Usage: process_split_commit REV PARENTS
 process_split_commit () {
 	assert test $# = 2
@@ -1006,19 +970,7 @@ cmd_split () {
 	eval "$grl" |
 	while read rev parents
 	do
-		if should_ignore_subtree_split_commit "$rev"
-		then
-			continue
-		fi
-		parsedparents=''
-		for parent in $parents
-		do
-			if ! should_ignore_subtree_split_commit "$parent"
-			then
-				parsedparents="$parsedparents$parent "
-			fi
-		done
-		process_split_commit "$rev" "$parsedparents"
+		process_split_commit "$rev" "$parents"
 	done || exit $?
 
 	latest_new=$(cache_get latest_new) || exit $?
diff --git a/contrib/subtree/t/t7900-subtree.sh b/contrib/subtree/t/t7900-subtree.sh
index dad8dea63a..05a774ad47 100755
--- a/contrib/subtree/t/t7900-subtree.sh
+++ b/contrib/subtree/t/t7900-subtree.sh
@@ -428,8 +428,7 @@ test_expect_success 'split sub dir/ with --rejoin' '
 # 	- Perform 'split' on subtree B
 # 	- Create new commits with changes to subtree A and B
 # 	- Perform split on subtree A
-# 	- Check that the commits in subtree B are not processed
-#			as part of the subtree A split
+# 	- Check for expected history
 test_expect_success 'split with multiple subtrees' '
 	subtree_test_create_repo "$test_count" &&
 	subtree_test_create_repo "$test_count/subA" &&
@@ -458,8 +457,8 @@ test_expect_success 'split with multiple subtrees' '
 		--squash --rejoin -m "Sub A Split 2" -b a2 &&
 	test "$(git -C "$test_count" rev-list --count main..a2)" -eq 2 &&
 	test "$(git -C "$test_count" rev-list --count a1..a2)" -eq 1 &&
-	test "$(git -C "$test_count" subtree split --prefix=subBDir \
-		--squash --rejoin -d -m "Sub B Split 1" -b b2 2>&1 | grep -w "\[1\]")" = "" &&
+	git -C "$test_count" subtree split --prefix=subBDir \
+		--squash --rejoin -d -m "Sub B Split 1" -b b2 &&
 	test "$(git -C "$test_count" rev-list --count main..b2)" -eq 2 &&
 	test "$(git -C "$test_count" rev-list --count b1..b2)" -eq 1
 '
@@ -507,6 +506,67 @@ do
 	'
 done
 
+# Usually,
+#
+#    git subtree merge -P subA --squash f00...
+#
+# makes two commits, in this order:
+#
+# 1. Squashed 'subA/' content from commit f00...
+# 2. Merge commit (1) as 'subA'
+#
+# Commit 1 updates the subtree but does *not* rewrite paths.
+# Commit 2 rewrites all trees to start with `subA/`
+#
+# Commit 1 either has no parents or depends only on other
+# "Squashed 'subA/' content" commits.
+#
+# For merge without --squash, subtree produces just one commit:
+# a merge commit with git-subtree trailers.
+#
+# In either case, if the user rebases these commits, they will
+# still have the git-subtree-* trailers… but will NOT have
+# the layout described above.
+#
+# Test that subsequent `git subtree split` are not confused by this.
+test_expect_success 'split with rebased subtree commit' '
+	subtree_test_create_repo "$test_count" &&
+	(
+		cd "$test_count" &&
+		test_commit file0 &&
+		test_create_subtree_add \
+			. mksubtree subA file1 --squash &&
+		test_path_is_file subA/file1.t &&
+		mkdir subB &&
+		test_commit subB/bfile &&
+		git commit --amend -F - <<'EOF' &&
+Squashed '\''subB/'\'' content from commit '\''badf00da911bbe895347b4b236f5461d55dc9877'\''
+
+Simulate a cherry-picked or rebased subtree commit.
+
+git-subtree-dir: subB
+git-subtree-split: badf00da911bbe895347b4b236f5461d55dc9877
+EOF
+		test_commit subA/file2 &&
+		test_commit subB/bfile2 &&
+		git commit --amend -F - <<'EOF' &&
+Split '\''subB/'\'' into commit '\''badf00da911bbe895347b4b236f5461d55dc9877'\''
+
+Simulate a cherry-picked or rebased subtree commit.
+
+git-subtree-dir: subB
+git-subtree-mainline: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+git-subtree-split: badf00da911bbe895347b4b236f5461d55dc9877
+EOF
+		git subtree split --prefix=subA --branch=bsplit &&
+		git checkout bsplit &&
+		test_path_is_file file1.t &&
+		test_path_is_file file2.t &&
+		test "$(last_commit_subject)" = "subA/file2" &&
+		test "$(git rev-list --count bsplit)" -eq 2
+	)
+'
+
 test_expect_success 'split sub dir/ with --rejoin from scratch' '
 	subtree_test_create_repo "$test_count" &&
 	test_create_commit "$test_count" main1 &&

-- 
2.43.0