kernel
/
git
mirror of https://git.kernel.org/pub/scm/git/git.git
1
0
Fork 0
Code Releases Activity

mktree: do not use OBJECT_INFO_QUICK when checking objects 

mktree_line() checks each referenced object's type with
odb_read_object_info_extended() under OBJECT_INFO_QUICK.  QUICK skips the
reprepare-and-retry that reloads the on-disk pack set, so a resident
"git mktree --batch" reader reports an object that a concurrent repack
just relocated into a new pack as missing, and rejects the entry.

QUICK entered this lookup in 817b0f6027 (mktree: do not check type of
remote objects, 2022-06-21) only to avoid lazily fetching promisor
objects; OBJECT_INFO_SKIP_FETCH_OBJECT already provides that.  Drop
OBJECT_INFO_QUICK and keep OBJECT_INFO_SKIP_FETCH_OBJECT, so mktree still
avoids a promisor fetch but recovers an object that was merely repacked.

Add a regression test driving a resident mktree --batch reader across a
concurrent repack that retires a pack.

Assisted-by: Claude Opus 4.8 & GPT-5.6 Sol
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
seen
Elijah Newren 2026-08-29 07:00:30 +00:00 committed by Junio C Hamano
parent 6272f3bd17
commit 22eef58fba
2 changed files with 48 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
builtin/mktree.c
View File

@ -125,7 +125,6 @@ static void mktree_line(struct repository *repo, char *buf, int nul_term_line, i
oi.typep = &obj_type;
if (odb_read_object_info_extended(repo->objects, &oid, &oi,
OBJECT_INFO_LOOKUP_REPLACE |
OBJECT_INFO_QUICK |
OBJECT_INFO_SKIP_FETCH_OBJECT) < 0)
obj_type = -1;


48
t/t1010-mktree.sh
View File

@ -69,4 +69,52 @@ test_expect_success 'mktree refuses to read ls-tree -r output (2)' '
test_must_fail git mktree <all.withsub
'

test_expect_success PIPE 'mktree --batch survives a concurrent repack retiring a pack' '
test_when_finished "rm -fr race" &&
git init race &&
(
cd race &&
test_commit seed &&
a=$(echo A | git hash-object -w --stdin) &&
b=$(echo B | git hash-object -w --stdin) &&
echo "$a" | git pack-objects .git/objects/pack/pack >pack-a &&
echo "$b" | git pack-objects .git/objects/pack/pack >pack-b &&

# Drop the loose copies so the blobs resolve only through the
# packs the multi-pack-index names.
git prune-packed &&
git multi-pack-index write &&
printf "100644 blob %s\ta\n" "$a" >tree-a &&
printf "100644 blob %s\tb\n" "$b" >tree-b &&

victim=".git/objects/pack/pack-$(cat pack-b)" &&
mkfifo in out &&

# mktree --batch stays resident, so its pack view predates the
# repack below; feed it one tree at a time over a fifo. The
# subshell exit closes the fifos, letting mktree see EOF and quit.
(git mktree --batch <in >out 2>err &) &&
exec 9>in &&
exec 8<out &&

# The first tree makes the reader cache its (soon stale) view.
cat tree-a >&9 && echo >&9 && read tree_a <&8 &&

# Mimic a concurrent repack: a replacement pack holds every
# object, and the pack for b loses its .idx (its .pack lingers),
# matching the order in which unlink_pack_path() removes files.
git cat-file --batch-all-objects --batch-check="%(objectname)" >oids &&
git pack-objects .git/objects/pack/pack <oids >/dev/null &&
rm -f "$victim.idx" &&

# Resolving b used to fail, as its QUICK lookup accepted the
# miss; without QUICK the reader repreps and finds b in the
# replacement pack.
cat tree-b >&9 && echo >&9 && read tree_b <&8 &&
exec 9>&- &&

test -n "$tree_b"
)
'

test_done