pack-objects: support reuse with `--no-ref-delta`
The previous commit disables delta- and bitmap-reuse entirely whenever pack-objects is given '--no-ref-delta' for the sake of simplicity. This is overly pessimistic. When '--delta-base-offset' is also given, delta reuse can remain enabled. A reused delta whose base is written earlier in the output can be encoded as an `OFS_DELTA`, even when its source copy was encoded as a `REF_DELTA`. Preferred bases and external thin-pack bases are different: neither appears in the output, so deltas against either still require encoding the object as a `REF_DELTA`, and thus cannot be reused. Without '--delta-base-offset', delta reuse remains disabled, since no delta representation remains. Bitmap reuse follows a different path, since selected entries may be copied without passing through the code which chooses a delta representation. When given '--no-ref-delta', we must inspect candidate objects individually, and leave `REF_DELTA` entries to the normal object path outside of pack-reuse. We must likewise avoid the special-case for reusing either the single or preferred pack corresponding to the bitmap by whole `eword_t`'s at a time. Signed-off-by: Taylor Blau <ttaylorr@openai.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Taylor Blau committed
Jul 12, 2026 at 18:12 UTC
aca978d38668f74766e14cd75b4461ae356e47bb
5 files changed
+71
-16
builtin/pack-objects.c
+13
-4
@@ -2207,6 +2207,13 @@ static int can_reuse_delta(const struct object_id *base_oid,
2207
*/
2208
base = packlist_find(&to_pack, base_oid);
2209
if (base) {
2210
+ /*
2211
+ * A preferred base is omitted from the resulting pack, so it
2212
+ * can only be referenced by object ID.
2213
+ */
2214
+ if (base->preferred_base && !allow_ref_delta)
2215
+ return 0;
2216
+
2217
if (!in_same_island(&delta->idx.oid, &base->idx.oid))
2218
return 0;
2219
*base_out = base;
@@ -2218,7 +2225,8 @@ static int can_reuse_delta(const struct object_id *base_oid,
2225
* even if it was buried too deep in history to make it into the
2226
* packing list.
2227
*/
2221
- if (thin && bitmap_has_oid_in_uninteresting(bitmap_git, base_oid)) {
2228
+ if (allow_ref_delta && thin &&
2229
+ bitmap_has_oid_in_uninteresting(bitmap_git, base_oid)) {
2230
if (use_delta_islands) {
2231
if (!in_same_island(&delta->idx.oid, base_oid))
2232
return 0;
@@ -4668,7 +4676,7 @@ static int pack_options_allow_reuse(void)
4676
!ignore_packed_keep_on_disk &&
4677
!ignore_packed_keep_in_core &&
4678
(!local || !have_non_local_packs) &&
4671
- !incremental && allow_ref_delta;
4679
+ !incremental && (allow_ref_delta || allow_ofs_delta);
4680
}
4681
4682
static int get_object_list_from_bitmap(struct rev_info *revs)
@@ -4690,7 +4698,8 @@ static int get_object_list_from_bitmap(struct rev_info *revs)
4698
&reuse_packfiles,
4699
&reuse_packfiles_nr,
4700
&reuse_packfile_bitmap,
4693
- allow_pack_reuse == MULTI_PACK_REUSE);
4701
+ allow_pack_reuse == MULTI_PACK_REUSE,
4702
+ allow_ref_delta);
4703
4704
if (reuse_packfiles) {
4705
reuse_packfile_objects = bitmap_popcount(reuse_packfile_bitmap);
@@ -5317,7 +5326,7 @@ int cmd_pack_objects(int argc,
5326
if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)
5327
use_internal_rev_list = 1;
5328
5320
- if (!reuse_object || !allow_ref_delta)
5329
+ if (!reuse_object || (!allow_ref_delta && !allow_ofs_delta))
5330
reuse_delta = 0;
5331
if (cfg->pack_compression_level == -1)
5332
cfg->pack_compression_level = Z_DEFAULT_COMPRESSION;
pack-bitmap.c
+20
-10
@@ -2267,7 +2267,8 @@ static int try_partial_reuse(struct bitmap_index *bitmap_git,
2267
uint32_t pack_pos,
2268
off_t offset,
2269
struct bitmap *reuse,
2270
- struct pack_window **w_curs)
2270
+ struct pack_window **w_curs,
2271
+ int allow_ref_delta)
2272
{
2273
off_t delta_obj_offset;
2274
enum object_type type;
@@ -2286,6 +2287,9 @@ static int try_partial_reuse(struct bitmap_index *bitmap_git,
2287
uint32_t base_pos;
2288
uint32_t base_bitmap_pos;
2289
2290
+ if (type == OBJ_REF_DELTA && !allow_ref_delta)
2291
+ return 0;
2292
+
2293
/*
2294
* Find the position of the base object so we can look it up
2295
* in our bitmaps. If we can't come up with an offset, or if
@@ -2358,20 +2362,19 @@ static int try_partial_reuse(struct bitmap_index *bitmap_git,
2362
2363
static void reuse_partial_packfile_from_bitmap_1(struct bitmap_index *bitmap_git,
2364
struct bitmapped_pack *pack,
2361
- struct bitmap *reuse)
2365
+ struct bitmap *reuse,
2366
+ int allow_ref_delta)
2367
{
2368
struct bitmap *result = bitmap_git->result;
2369
struct pack_window *w_curs = NULL;
2370
size_t pos = pack->bitmap_pos / BITS_IN_EWORD;
2371
2367
- if (!pack->bitmap_pos) {
2372
+ if (allow_ref_delta && !pack->bitmap_pos) {
2373
/*
2374
* If we're processing the first (in the case of a MIDX, the
2375
* preferred pack) or the only (in the case of single-pack
2371
- * bitmaps) pack, then we can reuse whole words at a time.
2372
- *
2373
- * This is because we know that any deltas in this range *must*
2374
- * have their bases chosen from the same pack, since:
2376
+ * bitmaps) pack, then any delta in this range must have its
2377
+ * base chosen from the same pack:
2378
*
2379
* - In the single pack case, there is no other pack to choose
2380
* them from.
@@ -2380,6 +2383,10 @@ static void reuse_partial_packfile_from_bitmap_1(struct bitmap_index *bitmap_git
2383
* all ties are broken in favor of that pack (i.e. the one
2384
* we're currently processing). So any duplicate bases will be
2385
* resolved in favor of the pack we're processing.
2386
+ *
2387
+ * When REF_DELTAs are allowed, we can therefore reuse whole
2388
+ * words at a time without inspecting object headers. Otherwise,
2389
+ * inspect each object below to avoid reusing a REF_DELTA entry.
2390
*/
2391
while (pos < result->word_alloc &&
2392
pos < pack->bitmap_nr / BITS_IN_EWORD &&
@@ -2429,7 +2436,8 @@ static void reuse_partial_packfile_from_bitmap_1(struct bitmap_index *bitmap_git
2436
}
2437
2438
if (try_partial_reuse(bitmap_git, pack, bit_pos,
2432
- pack_pos, ofs, reuse, &w_curs) < 0) {
2439
+ pack_pos, ofs, reuse, &w_curs,
2440
+ allow_ref_delta) < 0) {
2441
/*
2442
* try_partial_reuse indicated we couldn't reuse
2443
* any bits, so there is no point in trying more
@@ -2464,7 +2472,8 @@ void reuse_partial_packfile_from_bitmap(struct bitmap_index *bitmap_git,
2472
struct bitmapped_pack **packs_out,
2473
size_t *packs_nr_out,
2474
struct bitmap **reuse_out,
2467
- int multi_pack_reuse)
2475
+ int multi_pack_reuse,
2476
+ int allow_ref_delta)
2477
{
2478
struct repository *r = bitmap_repo(bitmap_git);
2479
struct bitmapped_pack *packs = NULL;
@@ -2559,7 +2568,8 @@ void reuse_partial_packfile_from_bitmap(struct bitmap_index *bitmap_git,
2568
reuse = bitmap_word_alloc(word_alloc);
2569
2570
for (i = 0; i < packs_nr; i++)
2562
- reuse_partial_packfile_from_bitmap_1(bitmap_git, &packs[i], reuse);
2571
+ reuse_partial_packfile_from_bitmap_1(bitmap_git, &packs[i], reuse,
2572
+ allow_ref_delta);
2573
2574
if (bitmap_is_empty(reuse)) {
2575
free(packs);
pack-bitmap.h
+2
-1
@@ -116,7 +116,8 @@ void reuse_partial_packfile_from_bitmap(struct bitmap_index *bitmap_git,
116
struct bitmapped_pack **packs_out,
117
size_t *packs_nr_out,
118
struct bitmap **reuse_out,
119
- int multi_pack_reuse);
119
+ int multi_pack_reuse,
120
+ int allow_ref_delta);
121
int rebuild_existing_bitmaps(struct bitmap_index *, struct packing_data *mapping,
122
kh_oid_map_t *reused_bitmaps, int show_progress);
123
void free_bitmap_index(struct bitmap_index *);
t/t5300-pack-object.sh
+20
-1
@@ -229,6 +229,20 @@ test_expect_success 'pack without REF_DELTA with OFS_DELTA' '
229
test_grep ! " REF_DELTA " deltas
230
'
231
232
+test_expect_success 'pack without REF_DELTA reuses deltas as OFS_DELTA' '
233
+ # Install the REF_DELTA pack above and disable delta search, so any
234
+ # output delta must be a reused REF_DELTA rewritten as OFS_DELTA.
235
+ test_when_finished "rm -f .git/objects/pack/pack-$packname_2.*" &&
236
+ git index-pack --stdin <test-2-${packname_2}.pack >/dev/null &&
237
+
238
+ git pack-objects --window=0 --delta-base-offset \
239
+ --no-ref-delta --stdout <obj-list >reused.pack &&
240
+ git index-pack -o reused.idx reused.pack &&
241
+ test-tool pack-deltas --list-deltas reused.idx >deltas &&
242
+ test_grep " OFS_DELTA " deltas &&
243
+ test_grep ! " REF_DELTA " deltas
244
+'
245
+
246
test_expect_success 'pack without REF_DELTA skips excluded delta bases' '
247
test_when_finished "git read-tree $tree" &&
248
@@ -253,7 +267,12 @@ test_expect_success 'pack without REF_DELTA skips excluded delta bases' '
267
test_grep ! " OFS_DELTA " deltas &&
268
test_grep " REF_DELTA " deltas &&
269
256
- git pack-objects --thin --stdout --revs \
270
+ # Store the REF_DELTA entries above and disable delta search below,
271
+ # so any output delta would have to reuse an excluded-base
272
+ # REF_DELTA.
273
+ git index-pack --stdin <thin-fixed.pack >/dev/null &&
274
+
275
+ git pack-objects --thin --window=0 --stdout --revs \
276
--delta-base-offset --no-ref-delta \
277
<thin-revs >no-ref-thin.pack &&
278
git index-pack --fix-thin --stdin no-ref-thin-fixed.pack \
t/t5332-multi-pack-reuse.sh
+16
@@ -111,6 +111,22 @@ test_expect_success 'reuse all objects from all packs' '
111
test_pack_objects_reused_all 9 3
112
'
113
114
+test_expect_success '--no-ref-delta reuses REF_DELTA-free bitmapped packs' '
115
+ # Whole-word reuse is unavailable under --no-ref-delta, so reusing
116
+ # every object below exercises the per-object bitmap path.
117
+ : >trace2.txt &&
118
+ GIT_TRACE2_EVENT="$PWD/trace2.txt" \
119
+ git pack-objects --stdout --revs --all --delta-base-offset \
120
+ --no-ref-delta >got.pack &&
121
+
122
+ test_pack_reused 9 <trace2.txt &&
123
+ test_packs_reused 3 <trace2.txt &&
124
+
125
+ git index-pack --strict -o got.idx got.pack &&
126
+ test-tool pack-deltas --list-deltas got.idx >deltas &&
127
+ test_grep ! " REF_DELTA " deltas
128
+'
129
+
130
test_expect_success 'reuse objects from first pack with middle gap' '
131
for i in D E F
132
do