[BUG] replay: segmentation fault when mistyping target to --onto

    $ ./bin-wrappers/git replay --onto="$doesntexist" "$commit"'^!'
    Segmentation fault (core dumped)

I did a bisect starting on current `seen` at a0bdfe7b (Merge branch
'bc/sha1-256-interop-02' into seen, 2025-12-11). That found the “first
bad commit” 15cd4ef1 (replay: make atomic ref updates the default
behavior, 2025-11-06) (which is on `master`). I started with v2.52.0 as
the first known “good”, which I manually checked.

Same segmentation fault on `next` at 674ac2bd (Merge branch
'kh/doc-send-email-paragraph-fix' into next, 2025-12-10).

The following is basically the bisect script except I changed it to make
sense outside my own repo.

#!/bin/sh

make || exit 125

# Mistyped `seen` for example
doesntexist=boh1eixe
# Current commit for topic kh/doc-pre-commit-fix
commit=8cbbdc92f77a20014d9c425c8b9e4af46e492204

./bin-wrappers/git replay --onto="$doesntexist" "$commit"'^!'

if test $? = 139
then
    exit 1
else
    # Presumably regular failure:
    #     fatal: Replaying down to root commit is not supported yet!
    exit 0
fi

cmd_replay() aborts if the pointer "onto" is NULL after argument
parsing, e.g. when specifying a non-existing commit with --onto.
15cd4ef1f4 (replay: make atomic ref updates the default behavior,
2025-11-06) added code that dereferences this pointer before the check.
Switch their places to avoid a segmentation fault.

Reported-by: Kristoffer Haugsbakk <redacted>
Signed-off-by: René Scharfe <redacted>
---
 builtin/replay.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/builtin/replay.c b/builtin/replay.c
index 507b909df7..64ad2f0f04 100644
--- a/builtin/replay.c
+++ b/builtin/replay.c

@@ -454,6 +454,9 @@ int cmd_replay(int argc,
 	determine_replay_mode(repo, &revs.cmdline, onto_name, &advance_name,
 			      &onto, &update_refs);
 
+	if (!onto) /* FIXME: Should handle replaying down to root commit */
+		die("Replaying down to root commit is not supported yet!");
+
 	/* Build reflog message */
 	if (advance_name_opt)
 		strbuf_addf(&reflog_msg, "replay --advance %s", advance_name_opt);

@@ -472,9 +475,6 @@ int cmd_replay(int argc,
 		}
 	}
 
-	if (!onto) /* FIXME: Should handle replaying down to root commit */
-		die("Replaying down to root commit is not supported yet!");
-
 	if (prepare_revision_walk(&revs) < 0) {
 		ret = error(_("error preparing revisions"));
 		goto cleanup;

-- 
2.52.0

On 11/12/2025 17:56, René Scharfe wrote:

cmd_replay() aborts if the pointer "onto" is NULL after argument
parsing, e.g. when specifying a non-existing commit with --onto.
15cd4ef1f4 (replay: make atomic ref updates the default behavior,
2025-11-06) added code that dereferences this pointer before the check.
Switch their places to avoid a segmentation fault.

This fixes the regression nicely. There is a preexisting bug that we 
treat an invalid --onto argument the same as a missing argument but that 
can be fixed separately.

Thanks

Phillip

On Mon, Dec 15, 2025, at 11:10, Phillip Wood wrote:

On 11/12/2025 17:56, René Scharfe wrote:

quoted

cmd_replay() aborts if the pointer "onto" is NULL after argument
parsing, e.g. when specifying a non-existing commit with --onto.
15cd4ef1f4 (replay: make atomic ref updates the default behavior,
2025-11-06) added code that dereferences this pointer before the check.
Switch their places to avoid a segmentation fault.

This fixes the regression nicely. There is a preexisting bug that we
treat an invalid --onto argument the same as a missing argument but that
can be fixed separately.

I have a commit cooking (locally) which makes the command die when it
cannot find commit-ish for `--onto` or `--advance` (whitespace mangled
diff):

diff --git a/builtin/replay.c b/builtin/replay.c
index 507b909df7d..72d62aa34a6 100644
--- a/builtin/replay.c
+++ b/builtin/replay.c
@@ -39,7 +39,7 @@ static struct commit *peel_committish(struct repository *repo, const char *name)
 	struct object_id oid;

 	if (repo_get_oid(repo, name, &oid))
-		return NULL;
+		die(_("'%s' is not a valid commit-ish"), name);
 	obj = parse_object(repo, &oid);
 	return (struct commit *)repo_peel_to_type(repo, name, 0, obj,
 						  OBJ_COMMIT);

Instead of dieing like this:

    Replaying down to root commit is not supported yet!

I hope that doesn’t cause any leak issues when I test it later.