pack-objects.c · leroysheep/rag-system

test

bisect jch maint master next seen test todo

c 5,770 lines 156 KB

Raw

1	#define USE_THE_REPOSITORY_VARIABLE
2	#define DISABLE_SIGN_COMPARE_WARNINGS
3
4	#include "builtin.h"
5	#include "environment.h"
6	#include "gettext.h"
7	#include "hex.h"
8	#include "config.h"
9	#include "attr.h"
10	#include "object.h"
11	#include "commit.h"
12	#include "tag.h"
13	#include "delta.h"
14	#include "pack.h"
15	#include "pack-revindex.h"
16	#include "csum-file.h"
17	#include "tree-walk.h"
18	#include "diff.h"
19	#include "revision.h"
20	#include "list-objects.h"
21	#include "list-objects-filter-options.h"
22	#include "pack-objects.h"
23	#include "progress.h"
24	#include "refs.h"
25	#include "thread-utils.h"
26	#include "pack-bitmap.h"
27	#include "delta-islands.h"
28	#include "reachable.h"
29	#include "oid-array.h"
30	#include "strvec.h"
31	#include "strmap.h"
32	#include "list.h"
33	#include "packfile.h"
34	#include "object-file.h"
35	#include "object-file-convert.h"
36	#include "odb.h"
37	#include "odb/streaming.h"
38	#include "replace-object.h"
39	#include "dir.h"
40	#include "midx.h"
41	#include "trace2.h"
42	#include "shallow.h"
43	#include "promisor-remote.h"
44	#include "pack-mtimes.h"
45	#include "parse-options.h"
46	#include "pkt-line.h"
47	#include "blob.h"
48	#include "tree.h"
49	#include "path-walk.h"
50
51	/*
52	* Objects we are going to pack are collected in the `to_pack` structure.
53	* It contains an array (dynamically expanded) of the object data, and a map
54	* that can resolve SHA1s to their position in the array.
55	*/
56	static struct packing_data to_pack;
57
58	static inline struct object_entry *oe_delta(
59	const struct packing_data *pack,
60	const struct object_entry *e)
61	{
62	if (!e->delta_idx)
63	return NULL;
64	if (e->ext_base)
65	return &pack->ext_bases[e->delta_idx - 1];
66	else
67	return &pack->objects[e->delta_idx - 1];
68	}
69
70	static inline size_t oe_delta_size(struct packing_data *pack,
71	const struct object_entry *e)
72	{
73	if (e->delta_size_valid)
74	return e->delta_size_;
75
76	/*
77	* pack->delta_size[] can't be NULL because oe_set_delta_size()
78	* must have been called when a new delta is saved with
79	* oe_set_delta().
80	* If oe_delta() returns NULL (i.e. default state, which means
81	* delta_size_valid is also false), then the caller must never
82	* call oe_delta_size().
83	*/
84	return pack->delta_size[e - pack->objects];
85	}
86
87	size_t oe_get_size_slow(struct packing_data *pack,
88	const struct object_entry *e);
89
90	static inline size_t oe_size(struct packing_data *pack,
91	const struct object_entry *e)
92	{
93	if (e->size_valid)
94	return e->size_;
95
96	return oe_get_size_slow(pack, e);
97	}
98
99	static inline void oe_set_delta(struct packing_data *pack,
100	struct object_entry *e,
101	struct object_entry *delta)
102	{
103	if (delta)
104	e->delta_idx = (delta - pack->objects) + 1;
105	else
106	e->delta_idx = 0;
107	}
108
109	static inline struct object_entry *oe_delta_sibling(
110	const struct packing_data *pack,
111	const struct object_entry *e)
112	{
113	if (e->delta_sibling_idx)
114	return &pack->objects[e->delta_sibling_idx - 1];
115	return NULL;
116	}
117
118	static inline struct object_entry *oe_delta_child(
119	const struct packing_data *pack,
120	const struct object_entry *e)
121	{
122	if (e->delta_child_idx)
123	return &pack->objects[e->delta_child_idx - 1];
124	return NULL;
125	}
126
127	static inline void oe_set_delta_child(struct packing_data *pack,
128	struct object_entry *e,
129	struct object_entry *delta)
130	{
131	if (delta)
132	e->delta_child_idx = (delta - pack->objects) + 1;
133	else
134	e->delta_child_idx = 0;
135	}
136
137	static inline void oe_set_delta_sibling(struct packing_data *pack,
138	struct object_entry *e,
139	struct object_entry *delta)
140	{
141	if (delta)
142	e->delta_sibling_idx = (delta - pack->objects) + 1;
143	else
144	e->delta_sibling_idx = 0;
145	}
146
147	static inline void oe_set_size(struct packing_data *pack,
148	struct object_entry *e,
149	size_t size)
150	{
151	if (size < pack->oe_size_limit) {
152	e->size_ = size;
153	e->size_valid = 1;
154	} else {
155	e->size_valid = 0;
156	if (oe_get_size_slow(pack, e) != size)
157	BUG("'size' is supposed to be the object size!");
158	}
159	}
160
161	static inline void oe_set_delta_size(struct packing_data *pack,
162	struct object_entry *e,
163	size_t size)
164	{
165	if (size < pack->oe_delta_size_limit) {
166	e->delta_size_ = size;
167	e->delta_size_valid = 1;
168	} else {
169	packing_data_lock(pack);
170	if (!pack->delta_size)
171	ALLOC_ARRAY(pack->delta_size, pack->nr_alloc);
172	packing_data_unlock(pack);
173
174	pack->delta_size[e - pack->objects] = size;
175	e->delta_size_valid = 0;
176	}
177	}
178
179	#define IN_PACK(obj) oe_in_pack(&to_pack, obj)
180	#define SIZE(obj) oe_size(&to_pack, obj)
181	#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
182	#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
183	#define DELTA(obj) oe_delta(&to_pack, obj)
184	#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
185	#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
186	#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
187	#define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid)
188	#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
189	#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
190	#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
191
192	static const char *const pack_usage[] = {
193	N_("git pack-objects [-q \| --progress \| --all-progress] [--all-progress-implied]\n"
194	" [--no-reuse-delta] [--delta-base-offset] [--no-ref-delta]\n"
195	" [--non-empty]\n"
196	" [--local] [--incremental] [--window=<n>] [--depth=<n>]\n"
197	" [--revs [--unpacked \| --all]] [--keep-pack=<pack-name>]\n"
198	" [--cruft] [--cruft-expiration=<time>]\n"
199	" [--stdout [--filter=<filter-spec>] \| <base-name>]\n"
200	" [--shallow] [--keep-true-parents] [--[no-]sparse]\n"
201	" [--name-hash-version=<n>] [--path-walk] < <object-list>"),
202	NULL
203	};
204
205	static struct pack_idx_entry **written_list;
206	static uint32_t nr_result, nr_written, nr_seen;
207	static struct bitmap_index *bitmap_git;
208	static uint32_t write_layer;
209
210	static int non_empty;
211	static int reuse_delta = 1, reuse_object = 1;
212	static int keep_unreachable, unpack_unreachable, include_tag;
213	static timestamp_t unpack_unreachable_expiration;
214	static int pack_loose_unreachable;
215	static int cruft;
216	static int shallow = 0;
217	static timestamp_t cruft_expiration;
218	static int local;
219	static int have_non_local_packs;
220	static int incremental;
221	static int ignore_packed_keep_on_disk;
222	static int ignore_packed_keep_in_core;
223	static int ignore_packed_keep_in_core_open;
224	static const char *stdin_packs_refs_snapshot;
225	static int ignore_packed_keep_in_core_has_cruft;
226	static int allow_ofs_delta;
227	static int allow_ref_delta = 1;
228	static struct pack_idx_option pack_idx_opts;
229	static const char *base_name;
230	static int progress = 1;
231	static int window = 10;
232	static unsigned long pack_size_limit;
233	static int depth = 50;
234	static int delta_search_threads;
235	static int pack_to_stdout;
236	static int sparse;
237	static int thin;
238	static int path_walk = -1;
239	static int num_preferred_base;
240	static struct progress *progress_state;
241
242	static struct bitmapped_pack *reuse_packfiles;
243	static size_t reuse_packfiles_nr;
244	static size_t reuse_packfiles_used_nr;
245	static uint32_t reuse_packfile_objects;
246	static struct bitmap *reuse_packfile_bitmap;
247
248	static int use_bitmap_index_default = 1;
249	static int use_bitmap_index = -1;
250	static enum {
251	NO_PACK_REUSE = 0,
252	SINGLE_PACK_REUSE,
253	MULTI_PACK_REUSE,
254	} allow_pack_reuse = SINGLE_PACK_REUSE;
255	static enum {
256	WRITE_BITMAP_FALSE = 0,
257	WRITE_BITMAP_QUIET,
258	WRITE_BITMAP_TRUE,
259	} write_bitmap_index;
260	static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE;
261
262	static int exclude_promisor_objects;
263	static int exclude_promisor_objects_best_effort;
264
265	static int use_delta_islands;
266
267	static size_t delta_cache_size = 0;
268	static size_t max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE;
269	static unsigned long cache_max_small_delta_size = 1000;
270
271	static unsigned long window_memory_limit = 0;
272
273	static struct string_list uri_protocols = STRING_LIST_INIT_NODUP;
274
275	enum missing_action {
276	MA_ERROR = 0, /* fail if any missing objects are encountered */
277	MA_ALLOW_ANY, /* silently allow ALL missing objects */
278	MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
279	};
280	static enum missing_action arg_missing_action;
281	static show_object_fn fn_show_object;
282
283	struct configured_exclusion {
284	struct oidmap_entry e;
285	char *pack_hash_hex;
286	char *uri;
287	};
288	static struct oidmap configured_exclusions;
289
290	static struct oidset excluded_by_config;
291	static int name_hash_version = -1;
292
293	enum stdin_packs_mode {
294	STDIN_PACKS_MODE_NONE,
295	STDIN_PACKS_MODE_STANDARD,
296	STDIN_PACKS_MODE_FOLLOW,
297	STDIN_PACKS_MODE_FOLLOW_REACHABLE,
298	};
299
300	/**
301	* Check whether the name_hash_version chosen by user input is appropriate,
302	* and also validate whether it is compatible with other features.
303	*/
304	static void validate_name_hash_version(void)
305	{
306	if (name_hash_version < 1 \|\| name_hash_version > 2)
307	die(_("invalid --name-hash-version option: %d"), name_hash_version);
308	if (write_bitmap_index && name_hash_version != 1) {
309	warning(_("currently, --write-bitmap-index requires --name-hash-version=1"));
310	name_hash_version = 1;
311	}
312	}
313
314	static inline uint32_t pack_name_hash_fn(const char *name)
315	{
316	static int seen_version = -1;
317
318	if (seen_version < 0)
319	seen_version = name_hash_version;
320	else if (seen_version != name_hash_version)
321	BUG("name hash version changed from %d to %d mid-process",
322	seen_version, name_hash_version);
323
324	switch (name_hash_version) {
325	case 1:
326	return pack_name_hash(name);
327
328	case 2:
329	return pack_name_hash_v2((const unsigned char *)name);
330
331	default:
332	BUG("invalid name-hash version: %d", name_hash_version);
333	}
334	}
335
336	/*
337	* stats
338	*/
339	static uint32_t written, written_delta;
340	static uint32_t reused, reused_delta;
341
342	/*
343	* Indexed commits
344	*/
345	static struct commit **indexed_commits;
346	static unsigned int indexed_commits_nr;
347	static unsigned int indexed_commits_alloc;
348
349	static void index_commit_for_bitmap(struct commit *commit)
350	{
351	if (indexed_commits_nr >= indexed_commits_alloc) {
352	indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
353	REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
354	}
355
356	indexed_commits[indexed_commits_nr++] = commit;
357	}
358
359	static void get_delta(struct object_entry entry)
360	{
361	unsigned long size, base_size;
362	size_t delta_size;
363	void buf, base_buf, *delta_buf;
364	enum object_type type;
365	size_t size_st = 0, base_size_st = 0;
366
367	buf = odb_read_object(the_repository->objects, &entry->idx.oid,
368	&type, &size_st);
369	size = cast_size_t_to_ulong(size_st);
370	if (!buf)
371	die(_("unable to read %s"), oid_to_hex(&entry->idx.oid));
372	base_buf = odb_read_object(the_repository->objects,
373	&DELTA(entry)->idx.oid, &type,
374	&base_size_st);
375	base_size = cast_size_t_to_ulong(base_size_st);
376	if (!base_buf)
377	die("unable to read %s",
378	oid_to_hex(&DELTA(entry)->idx.oid));
379	delta_buf = diff_delta(base_buf, base_size,
380	buf, size, &delta_size, 0);
381	/*
382	* We successfully computed this delta once but dropped it for
383	* memory reasons. Something is very wrong if this time we
384	* recompute and create a different delta.
385	*/
386	if (!delta_buf \|\| delta_size != DELTA_SIZE(entry))
387	BUG("delta size changed");
388	free(buf);
389	free(base_buf);
390	return delta_buf;
391	}
392
393	static unsigned long do_compress(void **pptr, unsigned long size)
394	{
395	git_zstream stream;
396	void in, out;
397	unsigned long maxsize;
398	struct repo_config_values *cfg = repo_config_values(the_repository);
399
400	git_deflate_init(&stream, cfg->pack_compression_level);
401	maxsize = git_deflate_bound(&stream, size);
402
403	in = *pptr;
404	out = xmalloc(maxsize);
405	*pptr = out;
406
407	stream.next_in = in;
408	stream.avail_in = size;
409	stream.next_out = out;
410	stream.avail_out = maxsize;
411	while (git_deflate(&stream, Z_FINISH) == Z_OK)
412	; /* nothing */
413	git_deflate_end(&stream);
414
415	free(in);
416	return stream.total_out;
417	}
418
419	static unsigned long write_large_blob_data(struct odb_read_stream st, struct hashfile f,
420	const struct object_id *oid)
421	{
422	git_zstream stream;
423	unsigned char ibuf[1024 * 16];
424	unsigned char obuf[1024 * 16];
425	unsigned long olen = 0;
426	struct repo_config_values *cfg = repo_config_values(the_repository);
427
428	git_deflate_init(&stream, cfg->pack_compression_level);
429
430	for (;;) {
431	ssize_t readlen;
432	int zret = Z_OK;
433	readlen = odb_read_stream_read(st, ibuf, sizeof(ibuf));
434	if (readlen == -1)
435	die(_("unable to read %s"), oid_to_hex(oid));
436
437	stream.next_in = ibuf;
438	stream.avail_in = readlen;
439	while ((stream.avail_in \|\| readlen == 0) &&
440	(zret == Z_OK \|\| zret == Z_BUF_ERROR)) {
441	stream.next_out = obuf;
442	stream.avail_out = sizeof(obuf);
443	zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
444	hashwrite(f, obuf, stream.next_out - obuf);
445	olen += stream.next_out - obuf;
446	}
447	if (stream.avail_in)
448	die(_("deflate error (%d)"), zret);
449	if (readlen == 0) {
450	if (zret != Z_STREAM_END)
451	die(_("deflate error (%d)"), zret);
452	break;
453	}
454	}
455	git_deflate_end(&stream);
456	return olen;
457	}
458
459	/*
460	* we are going to reuse the existing object data as is. make
461	* sure it is not corrupt.
462	*/
463	static int check_pack_inflate(struct packed_git *p,
464	struct pack_window **w_curs,
465	off_t offset,
466	off_t len,
467	size_t expect)
468	{
469	git_zstream stream;
470	unsigned char fakebuf[4096], *in;
471	int st;
472
473	memset(&stream, 0, sizeof(stream));
474	git_inflate_init(&stream);
475	do {
476	in = use_pack(p, w_curs, offset, &stream.avail_in);
477	stream.next_in = in;
478	stream.next_out = fakebuf;
479	stream.avail_out = sizeof(fakebuf);
480	st = git_inflate(&stream, Z_FINISH);
481	offset += stream.next_in - in;
482	} while (st == Z_OK \|\| st == Z_BUF_ERROR);
483	git_inflate_end(&stream);
484	return (st == Z_STREAM_END &&
485	stream.total_out == expect &&
486	stream.total_in == len) ? 0 : -1;
487	}
488
489	static void copy_pack_data(struct hashfile *f,
490	struct packed_git *p,
491	struct pack_window **w_curs,
492	off_t offset,
493	off_t len)
494	{
495	unsigned char *in;
496	size_t avail;
497
498	while (len) {
499	in = use_pack(p, w_curs, offset, &avail);
500	if (avail > len)
501	avail = (unsigned long)len;
502	hashwrite(f, in, avail);
503	offset += avail;
504	len -= avail;
505	}
506	}
507
508	static inline int oe_size_greater_than(struct packing_data *pack,
509	const struct object_entry *lhs,
510	size_t rhs)
511	{
512	if (lhs->size_valid)
513	return lhs->size_ > rhs;
514	if (rhs < pack->oe_size_limit) /* rhs < 2^x <= lhs ? */
515	return 1;
516	return oe_get_size_slow(pack, lhs) > rhs;
517	}
518
519	/* Return 0 if we will bust the pack-size limit */
520	static unsigned long write_no_reuse_object(struct hashfile f, struct object_entry entry,
521	unsigned long limit, int usable_delta)
522	{
523	unsigned long size, datalen;
524	unsigned char header[MAX_PACK_OBJECT_HEADER],
525	dheader[MAX_PACK_OBJECT_HEADER];
526	unsigned hdrlen;
527	enum object_type type;
528	void *buf;
529	struct odb_read_stream *st = NULL;
530	const unsigned hashsz = the_hash_algo->rawsz;
531
532	if (!usable_delta) {
533	if (oe_type(entry) == OBJ_BLOB &&
534	oe_size_greater_than(&to_pack, entry,
535	repo_settings_get_big_file_threshold(the_repository)) &&
536	(st = odb_read_stream_open(the_repository->objects, &entry->idx.oid,
537	NULL)) != NULL) {
538	buf = NULL;
539	type = st->type;
540	size = st->size;
541	} else {
542	size_t size_st = 0;
543	buf = odb_read_object(the_repository->objects,
544	&entry->idx.oid, &type,
545	&size_st);
546	size = cast_size_t_to_ulong(size_st);
547	if (!buf)
548	die(_("unable to read %s"),
549	oid_to_hex(&entry->idx.oid));
550	}
551	/*
552	* make sure no cached delta data remains from a
553	* previous attempt before a pack split occurred.
554	*/
555	FREE_AND_NULL(entry->delta_data);
556	entry->z_delta_size = 0;
557	} else if (entry->delta_data) {
558	size = DELTA_SIZE(entry);
559	buf = entry->delta_data;
560	entry->delta_data = NULL;
561	type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
562	OBJ_OFS_DELTA : OBJ_REF_DELTA;
563	} else {
564	buf = get_delta(entry);
565	size = DELTA_SIZE(entry);
566	type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
567	OBJ_OFS_DELTA : OBJ_REF_DELTA;
568	}
569
570	if (st) /* large blob case, just assume we don't compress well */
571	datalen = size;
572	else if (entry->z_delta_size)
573	datalen = entry->z_delta_size;
574	else
575	datalen = do_compress(&buf, size);
576
577	/*
578	* The object header is a byte of 'type' followed by zero or
579	* more bytes of length.
580	*/
581	hdrlen = encode_in_pack_object_header(header, sizeof(header),
582	type, size);
583
584	if (type == OBJ_OFS_DELTA) {
585	/*
586	* Deltas with relative base contain an additional
587	* encoding of the relative offset for the delta
588	* base from this object's position in the pack.
589	*/
590	off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
591	unsigned pos = sizeof(dheader) - 1;
592	dheader[pos] = ofs & 127;
593	while (ofs >>= 7)
594	dheader[--pos] = 128 \| (--ofs & 127);
595	if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
596	if (st)
597	odb_read_stream_close(st);
598	free(buf);
599	return 0;
600	}
601	hashwrite(f, header, hdrlen);
602	hashwrite(f, dheader + pos, sizeof(dheader) - pos);
603	hdrlen += sizeof(dheader) - pos;
604	} else if (type == OBJ_REF_DELTA) {
605	/*
606	* Deltas with a base reference contain
607	* additional bytes for the base object ID.
608	*/
609	if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
610	if (st)
611	odb_read_stream_close(st);
612	free(buf);
613	return 0;
614	}
615	hashwrite(f, header, hdrlen);
616	hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
617	hdrlen += hashsz;
618	} else {
619	if (limit && hdrlen + datalen + hashsz >= limit) {
620	if (st)
621	odb_read_stream_close(st);
622	free(buf);
623	return 0;
624	}
625	hashwrite(f, header, hdrlen);
626	}
627	if (st) {
628	datalen = write_large_blob_data(st, f, &entry->idx.oid);
629	odb_read_stream_close(st);
630	} else {
631	hashwrite(f, buf, datalen);
632	free(buf);
633	}
634
635	return hdrlen + datalen;
636	}
637
638	/* Return 0 if we will bust the pack-size limit */
639	static off_t write_reuse_object(struct hashfile f, struct object_entry entry,
640	unsigned long limit, int usable_delta)
641	{
642	struct packed_git *p = IN_PACK(entry);
643	struct pack_window *w_curs = NULL;
644	uint32_t pos;
645	off_t offset, cur;
646	enum object_type type = oe_type(entry);
647	enum object_type in_pack_type;
648	off_t datalen;
649	unsigned char header[MAX_PACK_OBJECT_HEADER],
650	dheader[MAX_PACK_OBJECT_HEADER];
651	unsigned hdrlen;
652	const unsigned hashsz = the_hash_algo->rawsz;
653	size_t entry_size;
654
655	cur = entry->in_pack_offset;
656	in_pack_type = unpack_object_header(p, &w_curs, &cur, &entry_size);
657	if (in_pack_type < 0)
658	die(_("write_reuse_object: unable to parse object header of %s"),
659	oid_to_hex(&entry->idx.oid));
660
661	if (DELTA(entry))
662	type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
663	OBJ_OFS_DELTA : OBJ_REF_DELTA;
664	hdrlen = encode_in_pack_object_header(header, sizeof(header),
665	type, entry_size);
666
667	offset = entry->in_pack_offset;
668	if (offset_to_pack_pos(p, offset, &pos) < 0)
669	die(_("write_reuse_object: could not locate %s, expected at "
670	"offset %"PRIuMAX" in pack %s"),
671	oid_to_hex(&entry->idx.oid), (uintmax_t)offset,
672	p->pack_name);
673	datalen = pack_pos_to_offset(p, pos + 1) - offset;
674	if (!pack_to_stdout && p->index_version > 1 &&
675	check_pack_crc(p, &w_curs, offset, datalen,
676	pack_pos_to_index(p, pos))) {
677	error(_("bad packed object CRC for %s"),
678	oid_to_hex(&entry->idx.oid));
679	unuse_pack(&w_curs);
680	return write_no_reuse_object(f, entry, limit, usable_delta);
681	}
682
683	offset += entry->in_pack_header_size;
684	datalen -= entry->in_pack_header_size;
685
686	if (!pack_to_stdout && p->index_version == 1 &&
687	check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
688	error(_("corrupt packed object for %s"),
689	oid_to_hex(&entry->idx.oid));
690	unuse_pack(&w_curs);
691	return write_no_reuse_object(f, entry, limit, usable_delta);
692	}
693
694	if (type == OBJ_OFS_DELTA) {
695	off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
696	unsigned pos = sizeof(dheader) - 1;
697	dheader[pos] = ofs & 127;
698	while (ofs >>= 7)
699	dheader[--pos] = 128 \| (--ofs & 127);
700	if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
701	unuse_pack(&w_curs);
702	return 0;
703	}
704	hashwrite(f, header, hdrlen);
705	hashwrite(f, dheader + pos, sizeof(dheader) - pos);
706	hdrlen += sizeof(dheader) - pos;
707	reused_delta++;
708	} else if (type == OBJ_REF_DELTA) {
709	if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
710	unuse_pack(&w_curs);
711	return 0;
712	}
713	hashwrite(f, header, hdrlen);
714	hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
715	hdrlen += hashsz;
716	reused_delta++;
717	} else {
718	if (limit && hdrlen + datalen + hashsz >= limit) {
719	unuse_pack(&w_curs);
720	return 0;
721	}
722	hashwrite(f, header, hdrlen);
723	}
724	copy_pack_data(f, p, &w_curs, offset, datalen);
725	unuse_pack(&w_curs);
726	reused++;
727	return hdrlen + datalen;
728	}
729
730	/* Return 0 if we will bust the pack-size limit */
731	static off_t write_object(struct hashfile *f,
732	struct object_entry *entry,
733	off_t write_offset)
734	{
735	unsigned long limit;
736	off_t len;
737	int usable_delta, to_reuse;
738
739	if (!pack_to_stdout)
740	crc32_begin(f);
741
742	/* apply size limit if limited packsize and not first object */
743	if (!pack_size_limit \|\| !nr_written)
744	limit = 0;
745	else if (pack_size_limit <= write_offset)
746	/*
747	* the earlier object did not fit the limit; avoid
748	* mistaking this with unlimited (i.e. limit = 0).
749	*/
750	limit = 1;
751	else
752	limit = pack_size_limit - write_offset;
753
754	if (!DELTA(entry))
755	usable_delta = 0; /* no delta */
756	else if (!pack_size_limit)
757	usable_delta = 1; /* unlimited packfile */
758	else if (DELTA(entry)->idx.offset == (off_t)-1)
759	usable_delta = 0; /* base was written to another pack */
760	else if (DELTA(entry)->idx.offset)
761	usable_delta = 1; /* base already exists in this pack */
762	else
763	usable_delta = 0; /* base could end up in another pack */
764
765	if (!reuse_object)
766	to_reuse = 0; /* explicit */
767	else if (!IN_PACK(entry))
768	to_reuse = 0; /* can't reuse what we don't have */
769	else if (oe_type(entry) == OBJ_REF_DELTA \|\|
770	oe_type(entry) == OBJ_OFS_DELTA)
771	/* check_object() decided it for us ... */
772	to_reuse = usable_delta;
773	/* ... but pack split may override that */
774	else if (oe_type(entry) != entry->in_pack_type)
775	to_reuse = 0; /* pack has delta which is unusable */
776	else if (DELTA(entry))
777	to_reuse = 0; /* we want to pack afresh */
778	else
779	to_reuse = 1; /* we have it in-pack undeltified,
780	* and we do not need to deltify it.
781	*/
782
783	if (!to_reuse)
784	len = write_no_reuse_object(f, entry, limit, usable_delta);
785	else
786	len = write_reuse_object(f, entry, limit, usable_delta);
787	if (!len)
788	return 0;
789
790	if (usable_delta)
791	written_delta++;
792	written++;
793	if (!pack_to_stdout)
794	entry->idx.crc32 = crc32_end(f);
795	return len;
796	}
797
798	enum write_one_status {
799	WRITE_ONE_SKIP = -1, /* already written */
800	WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
801	WRITE_ONE_WRITTEN = 1, /* normal */
802	WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
803	};
804
805	static enum write_one_status write_one(struct hashfile *f,
806	struct object_entry *e,
807	off_t *offset)
808	{
809	off_t size;
810	int recursing;
811
812	/*
813	* we set offset to 1 (which is an impossible value) to mark
814	* the fact that this object is involved in "write its base
815	* first before writing a deltified object" recursion.
816	*/
817	recursing = (e->idx.offset == 1);
818	if (recursing) {
819	warning(_("recursive delta detected for object %s"),
820	oid_to_hex(&e->idx.oid));
821	return WRITE_ONE_RECURSIVE;
822	} else if (e->idx.offset \|\| e->preferred_base) {
823	/* offset is non zero if object is written already. */
824	return WRITE_ONE_SKIP;
825	}
826
827	/* if we are deltified, write out base object first. */
828	if (DELTA(e)) {
829	e->idx.offset = 1; /* now recurse */
830	switch (write_one(f, DELTA(e), offset)) {
831	case WRITE_ONE_RECURSIVE:
832	/* we cannot depend on this one */
833	SET_DELTA(e, NULL);
834	break;
835	default:
836	break;
837	case WRITE_ONE_BREAK:
838	e->idx.offset = recursing;
839	return WRITE_ONE_BREAK;
840	}
841	}
842
843	e->idx.offset = *offset;
844	size = write_object(f, e, *offset);
845	if (!size) {
846	e->idx.offset = recursing;
847	return WRITE_ONE_BREAK;
848	}
849	written_list[nr_written++] = &e->idx;
850
851	/* make sure off_t is sufficiently large not to wrap */
852	if (signed_add_overflows(*offset, size))
853	die(_("pack too large for current definition of off_t"));
854	*offset += size;
855	return WRITE_ONE_WRITTEN;
856	}
857
858	static int mark_tagged(const struct reference ref, void cb_data UNUSED)
859	{
860	struct object_id peeled;
861	struct object_entry *entry = packlist_find(&to_pack, ref->oid);
862
863	if (entry)
864	entry->tagged = 1;
865	if (!reference_get_peeled_oid(the_repository, ref, &peeled)) {
866	entry = packlist_find(&to_pack, &peeled);
867	if (entry)
868	entry->tagged = 1;
869	}
870	return 0;
871	}
872
873	static inline unsigned char oe_layer(struct packing_data *pack,
874	struct object_entry *e)
875	{
876	if (!pack->layer)
877	return 0;
878	return pack->layer[e - pack->objects];
879	}
880
881	static inline void add_to_write_order(struct object_entry **wo,
882	unsigned int *endp,
883	struct object_entry *e)
884	{
885	if (e->filled \|\| oe_layer(&to_pack, e) != write_layer)
886	return;
887	wo[(*endp)++] = e;
888	e->filled = 1;
889	}
890
891	static void add_descendants_to_write_order(struct object_entry **wo,
892	unsigned int *endp,
893	struct object_entry *e)
894	{
895	int add_to_order = 1;
896	while (e) {
897	if (add_to_order) {
898	struct object_entry *s;
899	/* add this node... */
900	add_to_write_order(wo, endp, e);
901	/* all its siblings... */
902	for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
903	add_to_write_order(wo, endp, s);
904	}
905	}
906	/* drop down a level to add left subtree nodes if possible */
907	if (DELTA_CHILD(e)) {
908	add_to_order = 1;
909	e = DELTA_CHILD(e);
910	} else {
911	add_to_order = 0;
912	/* our sibling might have some children, it is next */
913	if (DELTA_SIBLING(e)) {
914	e = DELTA_SIBLING(e);
915	continue;
916	}
917	/* go back to our parent node */
918	e = DELTA(e);
919	while (e && !DELTA_SIBLING(e)) {
920	/* we're on the right side of a subtree, keep
921	* going up until we can go right again */
922	e = DELTA(e);
923	}
924	if (!e) {
925	/* done- we hit our original root node */
926	return;
927	}
928	/* pass it off to sibling at this level */
929	e = DELTA_SIBLING(e);
930	}
931	};
932	}
933
934	static void add_family_to_write_order(struct object_entry **wo,
935	unsigned int *endp,
936	struct object_entry *e)
937	{
938	struct object_entry *root;
939
940	for (root = e; DELTA(root); root = DELTA(root))
941	; /* nothing */
942	add_descendants_to_write_order(wo, endp, root);
943	}
944
945	static void compute_layer_order(struct object_entry *wo, unsigned int wo_end)
946	{
947	unsigned int i, last_untagged;
948	struct object_entry *objects = to_pack.objects;
949
950	for (i = 0; i < to_pack.nr_objects; i++) {
951	if (objects[i].tagged)
952	break;
953	add_to_write_order(wo, wo_end, &objects[i]);
954	}
955	last_untagged = i;
956
957	/*
958	* Then fill all the tagged tips.
959	*/
960	for (; i < to_pack.nr_objects; i++) {
961	if (objects[i].tagged)
962	add_to_write_order(wo, wo_end, &objects[i]);
963	}
964
965	/*
966	* And then all remaining commits and tags.
967	*/
968	for (i = last_untagged; i < to_pack.nr_objects; i++) {
969	if (oe_type(&objects[i]) != OBJ_COMMIT &&
970	oe_type(&objects[i]) != OBJ_TAG)
971	continue;
972	add_to_write_order(wo, wo_end, &objects[i]);
973	}
974
975	/*
976	* And then all the trees.
977	*/
978	for (i = last_untagged; i < to_pack.nr_objects; i++) {
979	if (oe_type(&objects[i]) != OBJ_TREE)
980	continue;
981	add_to_write_order(wo, wo_end, &objects[i]);
982	}
983
984	/*
985	* Finally all the rest in really tight order
986	*/
987	for (i = last_untagged; i < to_pack.nr_objects; i++) {
988	if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer)
989	add_family_to_write_order(wo, wo_end, &objects[i]);
990	}
991	}
992
993	static struct object_entry **compute_write_order(void)
994	{
995	uint32_t max_layers = 1;
996	unsigned int i, wo_end;
997
998	struct object_entry **wo;
999	struct object_entry *objects = to_pack.objects;
1000
1001	for (i = 0; i < to_pack.nr_objects; i++) {
1002	objects[i].tagged = 0;
1003	objects[i].filled = 0;
1004	SET_DELTA_CHILD(&objects[i], NULL);
1005	SET_DELTA_SIBLING(&objects[i], NULL);
1006	}
1007
1008	/*
1009	* Fully connect delta_child/delta_sibling network.
1010	* Make sure delta_sibling is sorted in the original
1011	* recency order.
1012	*/
1013	for (i = to_pack.nr_objects; i > 0;) {
1014	struct object_entry *e = &objects[--i];
1015	if (!DELTA(e))
1016	continue;
1017	/* Mark me as the first child */
1018	e->delta_sibling_idx = DELTA(e)->delta_child_idx;
1019	SET_DELTA_CHILD(DELTA(e), e);
1020	}
1021
1022	/*
1023	* Mark objects that are at the tip of tags.
1024	*/
1025	refs_for_each_tag_ref(get_main_ref_store(the_repository), mark_tagged,
1026	NULL);
1027
1028	if (use_delta_islands) {
1029	max_layers = compute_pack_layers(&to_pack);
1030	free_island_marks();
1031	}
1032
1033	ALLOC_ARRAY(wo, to_pack.nr_objects);
1034	wo_end = 0;
1035
1036	for (; write_layer < max_layers; ++write_layer)
1037	compute_layer_order(wo, &wo_end);
1038
1039	if (wo_end != to_pack.nr_objects)
1040	die(_("ordered %u objects, expected %"PRIu32),
1041	wo_end, to_pack.nr_objects);
1042
1043	return wo;
1044	}
1045
1046
1047	/*
1048	* A reused set of objects. All objects in a chunk have the same
1049	* relative position in the original packfile and the generated
1050	* packfile.
1051	*/
1052
1053	static struct reused_chunk {
1054	/* The offset of the first object of this chunk in the original
1055	* packfile. */
1056	off_t original;
1057	/* The difference for "original" minus the offset of the first object of
1058	* this chunk in the generated packfile. */
1059	off_t difference;
1060	} *reused_chunks;
1061	static int reused_chunks_nr;
1062	static int reused_chunks_alloc;
1063
1064	static void record_reused_object(off_t where, off_t offset)
1065	{
1066	if (reused_chunks_nr && reused_chunks[reused_chunks_nr-1].difference == offset)
1067	return;
1068
1069	ALLOC_GROW(reused_chunks, reused_chunks_nr + 1,
1070	reused_chunks_alloc);
1071	reused_chunks[reused_chunks_nr].original = where;
1072	reused_chunks[reused_chunks_nr].difference = offset;
1073	reused_chunks_nr++;
1074	}
1075
1076	/*
1077	* Binary search to find the chunk that "where" is in. Note
1078	* that we're not looking for an exact match, just the first
1079	* chunk that contains it (which implicitly ends at the start
1080	* of the next chunk.
1081	*/
1082	static off_t find_reused_offset(off_t where)
1083	{
1084	int lo = 0, hi = reused_chunks_nr;
1085	while (lo < hi) {
1086	int mi = lo + ((hi - lo) / 2);
1087	if (where == reused_chunks[mi].original)
1088	return reused_chunks[mi].difference;
1089	if (where < reused_chunks[mi].original)
1090	hi = mi;
1091	else
1092	lo = mi + 1;
1093	}
1094
1095	/*
1096	* The first chunk starts at zero, so we can't have gone below
1097	* there.
1098	*/
1099	assert(lo);
1100	return reused_chunks[lo-1].difference;
1101	}
1102
1103	static void write_reused_pack_one(struct packed_git *reuse_packfile,
1104	size_t pos, struct hashfile *out,
1105	off_t pack_start,
1106	struct pack_window **w_curs)
1107	{
1108	off_t offset, next, cur;
1109	enum object_type type;
1110	size_t size;
1111
1112	offset = pack_pos_to_offset(reuse_packfile, pos);
1113	next = pack_pos_to_offset(reuse_packfile, pos + 1);
1114
1115	record_reused_object(offset,
1116	offset - (hashfile_total(out) - pack_start));
1117
1118	cur = offset;
1119	type = unpack_object_header(reuse_packfile, w_curs, &cur, &size);
1120	assert(type >= 0);
1121
1122	if (type == OBJ_OFS_DELTA) {
1123	off_t base_offset;
1124	off_t fixup;
1125
1126	unsigned char header[MAX_PACK_OBJECT_HEADER];
1127	unsigned len;
1128
1129	base_offset = get_delta_base(reuse_packfile, w_curs, &cur, type, offset);
1130	assert(base_offset != 0);
1131
1132	/* Convert to REF_DELTA if we must... */
1133	if (!allow_ofs_delta) {
1134	uint32_t base_pos;
1135	struct object_id base_oid;
1136
1137	if (offset_to_pack_pos(reuse_packfile, base_offset, &base_pos) < 0)
1138	die(_("expected object at offset %"PRIuMAX" "
1139	"in pack %s"),
1140	(uintmax_t)base_offset,
1141	reuse_packfile->pack_name);
1142
1143	nth_packed_object_id(&base_oid, reuse_packfile,
1144	pack_pos_to_index(reuse_packfile, base_pos));
1145
1146	len = encode_in_pack_object_header(header, sizeof(header),
1147	OBJ_REF_DELTA, size);
1148	hashwrite(out, header, len);
1149	hashwrite(out, base_oid.hash, the_hash_algo->rawsz);
1150	copy_pack_data(out, reuse_packfile, w_curs, cur, next - cur);
1151	return;
1152	}
1153
1154	/* Otherwise see if we need to rewrite the offset... */
1155	fixup = find_reused_offset(offset) -
1156	find_reused_offset(base_offset);
1157	if (fixup) {
1158	unsigned char ofs_header[MAX_PACK_OBJECT_HEADER];
1159	unsigned i, ofs_len;
1160	off_t ofs = offset - base_offset - fixup;
1161
1162	len = encode_in_pack_object_header(header, sizeof(header),
1163	OBJ_OFS_DELTA, size);
1164
1165	i = sizeof(ofs_header) - 1;
1166	ofs_header[i] = ofs & 127;
1167	while (ofs >>= 7)
1168	ofs_header[--i] = 128 \| (--ofs & 127);
1169
1170	ofs_len = sizeof(ofs_header) - i;
1171
1172	hashwrite(out, header, len);
1173	hashwrite(out, ofs_header + sizeof(ofs_header) - ofs_len, ofs_len);
1174	copy_pack_data(out, reuse_packfile, w_curs, cur, next - cur);
1175	return;
1176	}
1177
1178	/* ...otherwise we have no fixup, and can write it verbatim */
1179	}
1180
1181	copy_pack_data(out, reuse_packfile, w_curs, offset, next - offset);
1182	}
1183
1184	static size_t write_reused_pack_verbatim(struct bitmapped_pack *reuse_packfile,
1185	struct hashfile *out,
1186	struct pack_window **w_curs)
1187	{
1188	size_t pos = 0;
1189	size_t end;
1190
1191	if (reuse_packfile->bitmap_pos) {
1192	/*
1193	* We can't reuse whole chunks verbatim out of
1194	* non-preferred packs since we can't guarantee that
1195	* all duplicate objects were resolved in favor of
1196	* that pack.
1197	*
1198	* Even if we have a whole eword_t worth of bits that
1199	* could be reused, there may be objects between the
1200	* objects corresponding to the first and last bit of
1201	* that word which were selected from a different
1202	* pack, causing us to send duplicate or unwanted
1203	* objects.
1204	*
1205	* Handle non-preferred packs from within
1206	* write_reused_pack(), which inspects and reuses
1207	* individual bits.
1208	*/
1209	return reuse_packfile->bitmap_pos / BITS_IN_EWORD;
1210	}
1211
1212	/*
1213	* Only read through the last word whose bits all correspond
1214	* to objects in the given packfile, since we must stop at a
1215	* word boundary.
1216	*
1217	* If there is no whole word to read (i.e. the packfile
1218	* contains fewer than BITS_IN_EWORD objects), then we'll
1219	* inspect bits one-by-one in write_reused_pack().
1220	*/
1221	end = reuse_packfile->bitmap_nr / BITS_IN_EWORD;
1222	if (reuse_packfile_bitmap->word_alloc < end)
1223	BUG("fewer words than expected in reuse_packfile_bitmap");
1224
1225	while (pos < end && reuse_packfile_bitmap->words[pos] == (eword_t)~0)
1226	pos++;
1227
1228	if (pos) {
1229	off_t to_write;
1230
1231	written = (pos * BITS_IN_EWORD);
1232	to_write = pack_pos_to_offset(reuse_packfile->p, written)
1233	- sizeof(struct pack_header);
1234
1235	/* We're recording one chunk, not one object. */
1236	record_reused_object(sizeof(struct pack_header), 0);
1237	hashflush(out);
1238	copy_pack_data(out, reuse_packfile->p, w_curs,
1239	sizeof(struct pack_header), to_write);
1240
1241	display_progress(progress_state, written);
1242	}
1243	return pos;
1244	}
1245
1246	static void write_reused_pack(struct bitmapped_pack *reuse_packfile,
1247	struct hashfile *f)
1248	{
1249	size_t i = reuse_packfile->bitmap_pos / BITS_IN_EWORD;
1250	uint32_t offset;
1251	off_t pack_start = hashfile_total(f) - sizeof(struct pack_header);
1252	struct pack_window *w_curs = NULL;
1253
1254	if (allow_ofs_delta)
1255	i = write_reused_pack_verbatim(reuse_packfile, f, &w_curs);
1256
1257	for (; i < reuse_packfile_bitmap->word_alloc; ++i) {
1258	eword_t word = reuse_packfile_bitmap->words[i];
1259	size_t pos = (i * BITS_IN_EWORD);
1260
1261	for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
1262	uint32_t pack_pos;
1263	if ((word >> offset) == 0)
1264	break;
1265
1266	offset += ewah_bit_ctz64(word >> offset);
1267	if (pos + offset < reuse_packfile->bitmap_pos)
1268	continue;
1269	if (pos + offset >= reuse_packfile->bitmap_pos + reuse_packfile->bitmap_nr)
1270	goto done;
1271
1272	if (reuse_packfile->bitmap_pos) {
1273	/*
1274	* When doing multi-pack reuse on a
1275	* non-preferred pack, translate bit positions
1276	* from the MIDX pseudo-pack order back to their
1277	* pack-relative positions before attempting
1278	* reuse.
1279	*/
1280	struct multi_pack_index *m = reuse_packfile->from_midx;
1281	uint32_t midx_pos;
1282	off_t pack_ofs;
1283
1284	if (!m)
1285	BUG("non-zero bitmap position without MIDX");
1286
1287	midx_pos = pack_pos_to_midx(m, pos + offset);
1288	pack_ofs = nth_midxed_offset(m, midx_pos);
1289
1290	if (offset_to_pack_pos(reuse_packfile->p,
1291	pack_ofs, &pack_pos) < 0)
1292	BUG("could not find expected object at offset %"PRIuMAX" in pack %s",
1293	(uintmax_t)pack_ofs,
1294	pack_basename(reuse_packfile->p));
1295	} else {
1296	/*
1297	* Can use bit positions directly, even for MIDX
1298	* bitmaps. See comment in try_partial_reuse()
1299	* for why.
1300	*/
1301	pack_pos = pos + offset;
1302	}
1303
1304	write_reused_pack_one(reuse_packfile->p, pack_pos, f,
1305	pack_start, &w_curs);
1306	display_progress(progress_state, ++written);
1307	}
1308	}
1309
1310	done:
1311	unuse_pack(&w_curs);
1312	}
1313
1314	static void write_excluded_by_configs(void)
1315	{
1316	struct oidset_iter iter;
1317	const struct object_id *oid;
1318
1319	oidset_iter_init(&excluded_by_config, &iter);
1320	while ((oid = oidset_iter_next(&iter))) {
1321	struct configured_exclusion *ex =
1322	oidmap_get(&configured_exclusions, oid);
1323
1324	if (!ex)
1325	BUG("configured exclusion wasn't configured");
1326	write_in_full(1, ex->pack_hash_hex, strlen(ex->pack_hash_hex));
1327	write_in_full(1, " ", 1);
1328	write_in_full(1, ex->uri, strlen(ex->uri));
1329	write_in_full(1, "\n", 1);
1330	}
1331	}
1332
1333	static const char no_split_warning[] = N_(
1334	"disabling bitmap writing, packs are split due to pack.packSizeLimit"
1335	);
1336
1337	static void write_pack_file(void)
1338	{
1339	uint32_t i = 0, j;
1340	struct hashfile *f;
1341	off_t offset;
1342	uint32_t nr_remaining = nr_result;
1343	time_t last_mtime = 0;
1344	struct object_entry **write_order;
1345
1346	if (progress > pack_to_stdout)
1347	progress_state = start_progress(the_repository,
1348	_("Writing objects"), nr_result);
1349	ALLOC_ARRAY(written_list, to_pack.nr_objects);
1350	write_order = compute_write_order();
1351
1352	do {
1353	unsigned char hash[GIT_MAX_RAWSZ];
1354	char *pack_tmp_name = NULL;
1355
1356	if (pack_to_stdout) {
1357	/*
1358	* This command is most often invoked via
1359	* git-upload-pack(1), which will typically chunk data
1360	* into pktlines. As such, we use the maximum data
1361	* length of them as buffer length.
1362	*
1363	* Note that we need to subtract one though to
1364	* accommodate for the sideband byte.
1365	*/
1366	struct hashfd_options opts = {
1367	.progress = progress_state,
1368	.buffer_len = LARGE_PACKET_DATA_MAX - 1,
1369	};
1370	f = hashfd_ext(the_repository->hash_algo, 1,
1371	"<stdout>", &opts);
1372	} else {
1373	f = create_tmp_packfile(the_repository, &pack_tmp_name);
1374	}
1375
1376	offset = write_pack_header(f, nr_remaining);
1377
1378	if (reuse_packfiles_nr) {
1379	assert(pack_to_stdout);
1380	for (j = 0; j < reuse_packfiles_nr; j++) {
1381	reused_chunks_nr = 0;
1382	write_reused_pack(&reuse_packfiles[j], f);
1383	if (reused_chunks_nr)
1384	reuse_packfiles_used_nr++;
1385	}
1386	offset = hashfile_total(f);
1387	}
1388
1389	nr_written = 0;
1390	for (; i < to_pack.nr_objects; i++) {
1391	struct object_entry *e = write_order[i];
1392	if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
1393	break;
1394	display_progress(progress_state, written);
1395	}
1396
1397	if (pack_to_stdout) {
1398	/*
1399	* We never fsync when writing to stdout since we may
1400	* not be writing to an actual pack file. For instance,
1401	* the upload-pack code passes a pipe here. Calling
1402	* fsync on a pipe results in unnecessary
1403	* synchronization with the reader on some platforms.
1404	*/
1405	finalize_hashfile(f, hash, FSYNC_COMPONENT_NONE,
1406	CSUM_HASH_IN_STREAM \| CSUM_CLOSE);
1407	} else if (nr_written == nr_remaining) {
1408	finalize_hashfile(f, hash, FSYNC_COMPONENT_PACK,
1409	CSUM_HASH_IN_STREAM \| CSUM_FSYNC \| CSUM_CLOSE);
1410	} else {
1411	/*
1412	* If we wrote the wrong number of entries in the
1413	* header, rewrite it like in fast-import.
1414	*/
1415
1416	int fd = finalize_hashfile(f, hash, FSYNC_COMPONENT_PACK, 0);
1417	fixup_pack_header_footer(the_hash_algo, fd, hash,
1418	pack_tmp_name, nr_written,
1419	hash, offset);
1420	close(fd);
1421	if (write_bitmap_index) {
1422	if (write_bitmap_index != WRITE_BITMAP_QUIET)
1423	warning(_(no_split_warning));
1424	write_bitmap_index = 0;
1425	}
1426	}
1427
1428	if (!pack_to_stdout) {
1429	struct stat st;
1430	struct strbuf tmpname = STRBUF_INIT;
1431	struct bitmap_writer bitmap_writer;
1432	char *idx_tmp_name = NULL;
1433
1434	/*
1435	* Packs are runtime accessed in their mtime
1436	* order since newer packs are more likely to contain
1437	* younger objects. So if we are creating multiple
1438	* packs then we should modify the mtime of later ones
1439	* to preserve this property.
1440	*/
1441	if (stat(pack_tmp_name, &st) < 0) {
1442	warning_errno(_("failed to stat %s"), pack_tmp_name);
1443	} else if (!last_mtime) {
1444	last_mtime = st.st_mtime;
1445	} else {
1446	struct utimbuf utb;
1447	utb.actime = st.st_atime;
1448	utb.modtime = --last_mtime;
1449	if (utime(pack_tmp_name, &utb) < 0)
1450	warning_errno(_("failed utime() on %s"), pack_tmp_name);
1451	}
1452
1453	strbuf_addf(&tmpname, "%s-%s.", base_name,
1454	hash_to_hex(hash));
1455
1456	if (write_bitmap_index) {
1457	bitmap_writer_init(&bitmap_writer,
1458	the_repository, &to_pack,
1459	NULL);
1460	bitmap_writer_set_checksum(&bitmap_writer, hash);
1461	bitmap_writer_build_type_index(&bitmap_writer,
1462	written_list);
1463	}
1464
1465	if (cruft)
1466	pack_idx_opts.flags \|= WRITE_MTIMES;
1467
1468	stage_tmp_packfiles(the_repository, &tmpname,
1469	pack_tmp_name, written_list,
1470	nr_written, &to_pack,
1471	&pack_idx_opts, hash,
1472	&idx_tmp_name);
1473
1474	if (write_bitmap_index) {
1475	size_t tmpname_len = tmpname.len;
1476
1477	strbuf_addstr(&tmpname, "bitmap");
1478	stop_progress(&progress_state);
1479
1480	bitmap_writer_show_progress(&bitmap_writer,
1481	progress);
1482	bitmap_writer_select_commits(&bitmap_writer,
1483	indexed_commits,
1484	indexed_commits_nr);
1485	if (bitmap_writer_build(&bitmap_writer) < 0)
1486	die(_("failed to write bitmap index"));
1487	bitmap_writer_finish(&bitmap_writer,
1488	written_list,
1489	tmpname.buf, write_bitmap_options);
1490	bitmap_writer_free(&bitmap_writer);
1491	write_bitmap_index = 0;
1492	strbuf_setlen(&tmpname, tmpname_len);
1493	}
1494
1495	rename_tmp_packfile_idx(the_repository, &tmpname, &idx_tmp_name);
1496
1497	free(idx_tmp_name);
1498	strbuf_release(&tmpname);
1499	free(pack_tmp_name);
1500	puts(hash_to_hex(hash));
1501	}
1502
1503	/* mark written objects as written to previous pack */
1504	for (j = 0; j < nr_written; j++) {
1505	written_list[j]->offset = (off_t)-1;
1506	}
1507	nr_remaining -= nr_written;
1508	} while (nr_remaining && i < to_pack.nr_objects);
1509
1510	free(written_list);
1511	free(write_order);
1512	stop_progress(&progress_state);
1513	if (written != nr_result)
1514	die(_("wrote %"PRIu32" objects while expecting %"PRIu32),
1515	written, nr_result);
1516	trace2_data_intmax("pack-objects", the_repository,
1517	"write_pack_file/wrote", nr_result);
1518	}
1519
1520	static int no_try_delta(const char *path)
1521	{
1522	static struct attr_check *check;
1523
1524	if (!check)
1525	check = attr_check_initl("delta", NULL);
1526	git_check_attr(the_repository->index, path, check);
1527	if (ATTR_FALSE(check->items[0].value))
1528	return 1;
1529	return 0;
1530	}
1531
1532	/*
1533	* When adding an object, check whether we have already added it
1534	* to our packing list. If so, we can skip. However, if we are
1535	* being asked to excludei t, but the previous mention was to include
1536	* it, make sure to adjust its flags and tweak our numbers accordingly.
1537	*
1538	* As an optimization, we pass out the index position where we would have
1539	* found the item, since that saves us from having to look it up again a
1540	* few lines later when we want to add the new entry.
1541	*/
1542	static int have_duplicate_entry(const struct object_id *oid,
1543	int exclude)
1544	{
1545	struct object_entry *entry;
1546
1547	if (reuse_packfile_bitmap &&
1548	bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid))
1549	return 1;
1550
1551	entry = packlist_find(&to_pack, oid);
1552	if (!entry)
1553	return 0;
1554
1555	if (exclude) {
1556	if (!entry->preferred_base)
1557	nr_result--;
1558	entry->preferred_base = 1;
1559	}
1560
1561	return 1;
1562	}
1563
1564	static int want_cruft_object_mtime(struct repository *r,
1565	const struct object_id *oid,
1566	unsigned flags, uint32_t mtime)
1567	{
1568	struct odb_source *source;
1569
1570	for (source = r->objects->sources; source; source = source->next) {
1571	struct odb_source_files *files = odb_source_files_downcast(source);
1572	struct packed_git **cache = packfile_store_get_kept_pack_cache(files->packed, flags);
1573
1574	for (; *cache; cache++) {
1575	struct packed_git p = cache;
1576	off_t ofs;
1577	uint32_t candidate_mtime;
1578
1579	ofs = find_pack_entry_one(oid, p);
1580	if (!ofs)
1581	continue;
1582
1583	/*
1584	* We have a copy of the object 'oid' in a non-cruft
1585	* pack. We can avoid packing an additional copy
1586	* regardless of what the existing copy's mtime is since
1587	* it is outside of a cruft pack.
1588	*/
1589	if (!p->is_cruft)
1590	return 0;
1591
1592	/*
1593	* If we have a copy of the object 'oid' in a cruft
1594	* pack, then either read the cruft pack's mtime for
1595	* that object, or, if that can't be loaded, assume the
1596	* pack's mtime itself.
1597	*/
1598	if (!load_pack_mtimes(p)) {
1599	uint32_t pos;
1600	if (offset_to_pack_pos(p, ofs, &pos) < 0)
1601	continue;
1602	candidate_mtime = nth_packed_mtime(p, pos);
1603	} else {
1604	candidate_mtime = p->mtime;
1605	}
1606
1607	/*
1608	* We have a surviving copy of the object in a cruft
1609	* pack whose mtime is greater than or equal to the one
1610	* we are considering. We can thus avoid packing an
1611	* additional copy of that object.
1612	*/
1613	if (mtime <= candidate_mtime)
1614	return 0;
1615	}
1616	}
1617
1618	return -1;
1619	}
1620
1621	static int want_found_object(const struct object_id *oid, int exclude,
1622	struct packed_git *p, uint32_t mtime)
1623	{
1624	if (exclude)
1625	return 1;
1626	if (incremental)
1627	return 0;
1628
1629	if (!is_pack_valid(p))
1630	return -1;
1631
1632	/*
1633	* When asked to do --local (do not include an object that appears in a
1634	* pack we borrow from elsewhere) or --honor-pack-keep (do not include
1635	* an object that appears in a pack marked with .keep), finding a pack
1636	* that matches the criteria is sufficient for us to decide to omit it.
1637	* However, even if this pack does not satisfy the criteria, we need to
1638	* make sure no copy of this object appears in _any_ pack that makes us
1639	* to omit the object, so we need to check all the packs.
1640	*
1641	* We can however first check whether these options can possibly matter;
1642	* if they do not matter we know we want the object in generated pack.
1643	* Otherwise, we signal "-1" at the end to tell the caller that we do
1644	* not know either way, and it needs to check more packs.
1645	*/
1646
1647	/*
1648	* Objects in packs borrowed from elsewhere are discarded regardless of
1649	* if they appear in other packs that weren't borrowed.
1650	*/
1651	if (local && !p->pack_local)
1652	return 0;
1653
1654	/*
1655	* Then handle .keep first, as we have a fast(er) path there.
1656	*/
1657	if (ignore_packed_keep_on_disk \|\| ignore_packed_keep_in_core \|\|
1658	ignore_packed_keep_in_core_open) {
1659	/*
1660	* Set the flags for the kept-pack cache to be the ones we want
1661	* to ignore.
1662	*
1663	* That is, if we are ignoring objects in on-disk keep packs,
1664	* then we want to search through the on-disk keep and ignore
1665	* the in-core ones.
1666	*/
1667	unsigned flags = 0;
1668	if (ignore_packed_keep_on_disk)
1669	flags \|= KEPT_PACK_ON_DISK;
1670	if (ignore_packed_keep_in_core)
1671	flags \|= KEPT_PACK_IN_CORE;
1672	if (ignore_packed_keep_in_core_open)
1673	flags \|= KEPT_PACK_IN_CORE_OPEN;
1674
1675	/*
1676	* If the object is in a pack that we want to ignore, and we
1677	* don't have any cruft packs that are being retained, we can
1678	* abort quickly.
1679	*/
1680	if (!ignore_packed_keep_in_core_has_cruft) {
1681	if (ignore_packed_keep_on_disk && p->pack_keep)
1682	return 0;
1683	if (ignore_packed_keep_in_core && p->pack_keep_in_core)
1684	return 0;
1685	if (ignore_packed_keep_in_core_open && p->pack_keep_in_core_open)
1686	return 0;
1687	if (has_object_kept_pack(p->repo, oid, flags))
1688	return 0;
1689	} else {
1690	/*
1691	* But if there is at least one cruft pack which
1692	* is being kept, we only want to include the
1693	* provided object if it has a strictly greater
1694	* mtime than any existing cruft copy.
1695	*/
1696	if (!want_cruft_object_mtime(p->repo, oid, flags,
1697	mtime))
1698	return 0;
1699	}
1700	}
1701
1702	/*
1703	* At this point we know definitively that either we don't care about
1704	* keep-packs, or the object is not in one. Keep checking other
1705	* conditions...
1706	*/
1707	if (!local \|\| !have_non_local_packs)
1708	return 1;
1709
1710	/* we don't know yet; keep looking for more packs */
1711	return -1;
1712	}
1713
1714	static int want_object_in_pack_one(struct packed_git *p,
1715	const struct object_id *oid,
1716	int exclude,
1717	struct packed_git **found_pack,
1718	off_t *found_offset,
1719	uint32_t found_mtime)
1720	{
1721	off_t offset;
1722
1723	if (p == *found_pack)
1724	offset = *found_offset;
1725	else
1726	offset = find_pack_entry_one(oid, p);
1727
1728	if (offset) {
1729	if (!*found_pack) {
1730	if (!is_pack_valid(p))
1731	return -1;
1732	*found_offset = offset;
1733	*found_pack = p;
1734	}
1735	return want_found_object(oid, exclude, p, found_mtime);
1736	}
1737	return -1;
1738	}
1739
1740	/*
1741	* Check whether we want the object in the pack (e.g., we do not want
1742	* objects found in non-local stores if the "--local" option was used).
1743	*
1744	* If the caller already knows an existing pack it wants to take the object
1745	* from, that is passed in found_pack and found_offset; otherwise this
1746	* function finds if there is any pack that has the object and returns the pack
1747	* and its offset in these variables.
1748	*/
1749	static int want_object_in_pack_mtime(const struct object_id *oid,
1750	int exclude,
1751	struct packed_git **found_pack,
1752	off_t *found_offset,
1753	uint32_t found_mtime)
1754	{
1755	int want;
1756	struct packfile_list_entry *e;
1757	struct odb_source *source;
1758
1759	if (!exclude && local) {
1760	/*
1761	* Note that we start iterating at `sources->next` so that we
1762	* skip the local object source.
1763	*/
1764	struct odb_source *source = the_repository->objects->sources->next;
1765	for (; source; source = source->next) {
1766	struct odb_source_files *files = odb_source_files_downcast(source);
1767	if (!odb_source_read_object_info(&files->loose->base, oid, NULL, 0))
1768	return 0;
1769	}
1770	}
1771
1772	/*
1773	* If we already know the pack object lives in, start checks from that
1774	* pack - in the usual case when neither --local was given nor .keep files
1775	* are present we will determine the answer right now.
1776	*/
1777	if (*found_pack) {
1778	want = want_found_object(oid, exclude, *found_pack,
1779	found_mtime);
1780	if (want != -1)
1781	return want;
1782
1783	*found_pack = NULL;
1784	*found_offset = 0;
1785	}
1786
1787	odb_prepare_alternates(the_repository->objects);
1788
1789	for (source = the_repository->objects->sources; source; source = source->next) {
1790	struct odb_source_files *files = odb_source_files_downcast(source);
1791	struct multi_pack_index *m = get_multi_pack_index(files->packed);
1792	struct pack_entry e;
1793
1794	if (m && fill_midx_entry(m, oid, &e)) {
1795	want = want_object_in_pack_one(e.p, oid, exclude, found_pack, found_offset, found_mtime);
1796	if (want != -1)
1797	return want;
1798	}
1799	}
1800
1801	for (source = the_repository->objects->sources; source; source = source->next) {
1802	struct odb_source_files *files = odb_source_files_downcast(source);
1803
1804	for (e = files->packed->packs.head; e; e = e->next) {
1805	struct packed_git *p = e->pack;
1806	want = want_object_in_pack_one(p, oid, exclude, found_pack, found_offset, found_mtime);
1807	if (!exclude && want > 0)
1808	packfile_list_prepend(&files->packed->packs, p);
1809	if (want != -1)
1810	return want;
1811	}
1812	}
1813
1814	if (uri_protocols.nr) {
1815	struct configured_exclusion *ex =
1816	oidmap_get(&configured_exclusions, oid);
1817	int i;
1818	const char *p;
1819
1820	if (ex) {
1821	for (i = 0; i < uri_protocols.nr; i++) {
1822	if (skip_prefix(ex->uri,
1823	uri_protocols.items[i].string,
1824	&p) &&
1825	*p == ':') {
1826	oidset_insert(&excluded_by_config, oid);
1827	return 0;
1828	}
1829	}
1830	}
1831	}
1832
1833	return 1;
1834	}
1835
1836	static inline int want_object_in_pack(const struct object_id *oid,
1837	int exclude,
1838	struct packed_git **found_pack,
1839	off_t *found_offset)
1840	{
1841	return want_object_in_pack_mtime(oid, exclude, found_pack, found_offset,
1842	0);
1843	}
1844
1845	static struct object_entry create_object_entry(const struct object_id oid,
1846	enum object_type type,
1847	uint32_t hash,
1848	int exclude,
1849	int no_try_delta,
1850	struct packed_git *found_pack,
1851	off_t found_offset)
1852	{
1853	struct object_entry *entry;
1854
1855	entry = packlist_alloc(&to_pack, oid);
1856	entry->hash = hash;
1857	oe_set_type(entry, type);
1858	if (exclude)
1859	entry->preferred_base = 1;
1860	else
1861	nr_result++;
1862	if (found_pack) {
1863	oe_set_in_pack(&to_pack, entry, found_pack);
1864	entry->in_pack_offset = found_offset;
1865	}
1866
1867	entry->no_try_delta = no_try_delta;
1868
1869	return entry;
1870	}
1871
1872	static const char no_closure_warning[] = N_(
1873	"disabling bitmap writing, as some objects are not being packed"
1874	);
1875
1876	static void add_object_entry(const struct object_id *oid, enum object_type type,
1877	const char *name, int exclude)
1878	{
1879	struct packed_git *found_pack = NULL;
1880	off_t found_offset = 0;
1881
1882	display_progress(progress_state, ++nr_seen);
1883
1884	if (have_duplicate_entry(oid, exclude))
1885	return;
1886
1887	if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1888	/* The pack is missing an object, so it will not have closure */
1889	if (write_bitmap_index) {
1890	if (write_bitmap_index != WRITE_BITMAP_QUIET)
1891	warning(_(no_closure_warning));
1892	write_bitmap_index = 0;
1893	}
1894	return;
1895	}
1896
1897	create_object_entry(oid, type, pack_name_hash_fn(name),
1898	exclude, name && no_try_delta(name),
1899	found_pack, found_offset);
1900	}
1901
1902	static int add_object_entry_from_bitmap(const struct object_id *oid,
1903	enum object_type type,
1904	int flags UNUSED, uint32_t name_hash,
1905	struct packed_git *pack, off_t offset,
1906	void *payload UNUSED)
1907	{
1908	display_progress(progress_state, ++nr_seen);
1909
1910	if (have_duplicate_entry(oid, 0))
1911	return 0;
1912
1913	if (!want_object_in_pack(oid, 0, &pack, &offset))
1914	return 0;
1915
1916	create_object_entry(oid, type, name_hash, 0, 0, pack, offset);
1917	return 0;
1918	}
1919
1920	struct pbase_tree_cache {
1921	struct object_id oid;
1922	int ref;
1923	int temporary;
1924	void *tree_data;
1925	unsigned long tree_size;
1926	};
1927
1928	static struct pbase_tree_cache *(pbase_tree_cache[256]);
1929	static int pbase_tree_cache_ix(const struct object_id *oid)
1930	{
1931	return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1932	}
1933	static int pbase_tree_cache_ix_incr(int ix)
1934	{
1935	return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1936	}
1937
1938	static struct pbase_tree {
1939	struct pbase_tree *next;
1940	/* This is a phony "cache" entry; we are not
1941	* going to evict it or find it through _get()
1942	* mechanism -- this is for the toplevel node that
1943	* would almost always change with any commit.
1944	*/
1945	struct pbase_tree_cache pcache;
1946	} *pbase_tree;
1947
1948	static struct pbase_tree_cache pbase_tree_get(const struct object_id oid)
1949	{
1950	struct pbase_tree_cache ent, nent;
1951	void *data;
1952	unsigned long size;
1953	size_t size_st = 0;
1954	enum object_type type;
1955	int neigh;
1956	int my_ix = pbase_tree_cache_ix(oid);
1957	int available_ix = -1;
1958
1959	/* pbase-tree-cache acts as a limited hashtable.
1960	* your object will be found at your index or within a few
1961	* slots after that slot if it is cached.
1962	*/
1963	for (neigh = 0; neigh < 8; neigh++) {
1964	ent = pbase_tree_cache[my_ix];
1965	if (ent && oideq(&ent->oid, oid)) {
1966	ent->ref++;
1967	return ent;
1968	}
1969	else if (((available_ix < 0) && (!ent \|\| !ent->ref)) \|\|
1970	((0 <= available_ix) &&
1971	(!ent && pbase_tree_cache[available_ix])))
1972	available_ix = my_ix;
1973	if (!ent)
1974	break;
1975	my_ix = pbase_tree_cache_ix_incr(my_ix);
1976	}
1977
1978	/* Did not find one. Either we got a bogus request or
1979	* we need to read and perhaps cache.
1980	*/
1981	data = odb_read_object(the_repository->objects, oid, &type, &size_st);
1982	size = cast_size_t_to_ulong(size_st);
1983	if (!data)
1984	return NULL;
1985	if (type != OBJ_TREE) {
1986	free(data);
1987	return NULL;
1988	}
1989
1990	/* We need to either cache or return a throwaway copy */
1991
1992	if (available_ix < 0)
1993	ent = NULL;
1994	else {
1995	ent = pbase_tree_cache[available_ix];
1996	my_ix = available_ix;
1997	}
1998
1999	if (!ent) {
2000	nent = xmalloc(sizeof(*nent));
2001	nent->temporary = (available_ix < 0);
2002	}
2003	else {
2004	/* evict and reuse */
2005	free(ent->tree_data);
2006	nent = ent;
2007	}
2008	oidcpy(&nent->oid, oid);
2009	nent->tree_data = data;
2010	nent->tree_size = size;
2011	nent->ref = 1;
2012	if (!nent->temporary)
2013	pbase_tree_cache[my_ix] = nent;
2014	return nent;
2015	}
2016
2017	static void pbase_tree_put(struct pbase_tree_cache *cache)
2018	{
2019	if (!cache->temporary) {
2020	cache->ref--;
2021	return;
2022	}
2023	free(cache->tree_data);
2024	free(cache);
2025	}
2026
2027	static size_t name_cmp_len(const char *name)
2028	{
2029	return strcspn(name, "\n/");
2030	}
2031
2032	static void add_pbase_object(struct tree_desc *tree,
2033	const char *name,
2034	size_t cmplen,
2035	const char *fullname)
2036	{
2037	struct name_entry entry;
2038	int cmp;
2039
2040	while (tree_entry(tree,&entry)) {
2041	if (S_ISGITLINK(entry.mode))
2042	continue;
2043	cmp = tree_entry_len(&entry) != cmplen ? 1 :
2044	memcmp(name, entry.path, cmplen);
2045	if (cmp > 0)
2046	continue;
2047	if (cmp < 0)
2048	return;
2049	if (name[cmplen] != '/') {
2050	add_object_entry(&entry.oid,
2051	object_type(entry.mode),
2052	fullname, 1);
2053	return;
2054	}
2055	if (S_ISDIR(entry.mode)) {
2056	struct tree_desc sub;
2057	struct pbase_tree_cache *tree;
2058	const char *down = name+cmplen+1;
2059	size_t downlen = name_cmp_len(down);
2060
2061	tree = pbase_tree_get(&entry.oid);
2062	if (!tree)
2063	return;
2064	init_tree_desc(&sub, &tree->oid,
2065	tree->tree_data, tree->tree_size);
2066
2067	add_pbase_object(&sub, down, downlen, fullname);
2068	pbase_tree_put(tree);
2069	}
2070	}
2071	}
2072
2073	static unsigned *done_pbase_paths;
2074	static int done_pbase_paths_num;
2075	static int done_pbase_paths_alloc;
2076	static int done_pbase_path_pos(unsigned hash)
2077	{
2078	int lo = 0;
2079	int hi = done_pbase_paths_num;
2080	while (lo < hi) {
2081	int mi = lo + (hi - lo) / 2;
2082	if (done_pbase_paths[mi] == hash)
2083	return mi;
2084	if (done_pbase_paths[mi] < hash)
2085	hi = mi;
2086	else
2087	lo = mi + 1;
2088	}
2089	return -lo-1;
2090	}
2091
2092	static int check_pbase_path(unsigned hash)
2093	{
2094	int pos = done_pbase_path_pos(hash);
2095	if (0 <= pos)
2096	return 1;
2097	pos = -pos - 1;
2098	ALLOC_GROW(done_pbase_paths,
2099	done_pbase_paths_num + 1,
2100	done_pbase_paths_alloc);
2101	done_pbase_paths_num++;
2102	if (pos < done_pbase_paths_num)
2103	MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
2104	done_pbase_paths_num - pos - 1);
2105	done_pbase_paths[pos] = hash;
2106	return 0;
2107	}
2108
2109	static void add_preferred_base_object(const char *name)
2110	{
2111	struct pbase_tree *it;
2112	size_t cmplen;
2113	unsigned hash = pack_name_hash_fn(name);
2114
2115	if (!num_preferred_base \|\| check_pbase_path(hash))
2116	return;
2117
2118	cmplen = name_cmp_len(name);
2119	for (it = pbase_tree; it; it = it->next) {
2120	if (cmplen == 0) {
2121	add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
2122	}
2123	else {
2124	struct tree_desc tree;
2125	init_tree_desc(&tree, &it->pcache.oid,
2126	it->pcache.tree_data, it->pcache.tree_size);
2127	add_pbase_object(&tree, name, cmplen, name);
2128	}
2129	}
2130	}
2131
2132	static void add_preferred_base(struct object_id *oid)
2133	{
2134	struct pbase_tree *it;
2135	void *data;
2136	unsigned long size;
2137	size_t size_st = 0;
2138	struct object_id tree_oid;
2139
2140	if (window <= num_preferred_base++)
2141	return;
2142
2143	data = odb_read_object_peeled(the_repository->objects, oid,
2144	OBJ_TREE, &size_st, &tree_oid);
2145	size = cast_size_t_to_ulong(size_st);
2146	if (!data)
2147	return;
2148
2149	for (it = pbase_tree; it; it = it->next) {
2150	if (oideq(&it->pcache.oid, &tree_oid)) {
2151	free(data);
2152	return;
2153	}
2154	}
2155
2156	CALLOC_ARRAY(it, 1);
2157	it->next = pbase_tree;
2158	pbase_tree = it;
2159
2160	oidcpy(&it->pcache.oid, &tree_oid);
2161	it->pcache.tree_data = data;
2162	it->pcache.tree_size = size;
2163	}
2164
2165	static void cleanup_preferred_base(void)
2166	{
2167	struct pbase_tree *it;
2168	unsigned i;
2169
2170	it = pbase_tree;
2171	pbase_tree = NULL;
2172	while (it) {
2173	struct pbase_tree *tmp = it;
2174	it = tmp->next;
2175	free(tmp->pcache.tree_data);
2176	free(tmp);
2177	}
2178
2179	for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
2180	if (!pbase_tree_cache[i])
2181	continue;
2182	free(pbase_tree_cache[i]->tree_data);
2183	FREE_AND_NULL(pbase_tree_cache[i]);
2184	}
2185
2186	FREE_AND_NULL(done_pbase_paths);
2187	done_pbase_paths_num = done_pbase_paths_alloc = 0;
2188	}
2189
2190	/*
2191	* Return 1 iff the object specified by "delta" can be sent
2192	* literally as a delta against the base in "base_sha1". If
2193	* so, then *base_out will point to the entry in our packing
2194	* list, or NULL if we must use the external-base list.
2195	*
2196	* Depth value does not matter - find_deltas() will
2197	* never consider reused delta as the base object to
2198	* deltify other objects against, in order to avoid
2199	* circular deltas.
2200	*/
2201	static int can_reuse_delta(const struct object_id *base_oid,
2202	struct object_entry *delta,
2203	struct object_entry **base_out)
2204	{
2205	struct object_entry *base;
2206
2207	/*
2208	* First see if we're already sending the base (or it's explicitly in
2209	* our "excluded" list).
2210	*/
2211	base = packlist_find(&to_pack, base_oid);
2212	if (base) {
2213	/*
2214	* A preferred base is omitted from the resulting pack, so it
2215	* can only be referenced by object ID.
2216	*/
2217	if (base->preferred_base && !allow_ref_delta)
2218	return 0;
2219
2220	if (!in_same_island(&delta->idx.oid, &base->idx.oid))
2221	return 0;
2222	*base_out = base;
2223	return 1;
2224	}
2225
2226	/*
2227	* Otherwise, reachability bitmaps may tell us if the receiver has it,
2228	* even if it was buried too deep in history to make it into the
2229	* packing list.
2230	*/
2231	if (allow_ref_delta && thin &&
2232	bitmap_has_oid_in_uninteresting(bitmap_git, base_oid)) {
2233	if (use_delta_islands) {
2234	if (!in_same_island(&delta->idx.oid, base_oid))
2235	return 0;
2236	}
2237	*base_out = NULL;
2238	return 1;
2239	}
2240
2241	return 0;
2242	}
2243
2244	static void prefetch_to_pack(uint32_t object_index_start) {
2245	struct oid_array to_fetch = OID_ARRAY_INIT;
2246	uint32_t i;
2247
2248	for (i = object_index_start; i < to_pack.nr_objects; i++) {
2249	struct object_entry *entry = to_pack.objects + i;
2250
2251	if (!odb_read_object_info_extended(the_repository->objects,
2252	&entry->idx.oid,
2253	NULL,
2254	OBJECT_INFO_FOR_PREFETCH))
2255	continue;
2256	oid_array_append(&to_fetch, &entry->idx.oid);
2257	}
2258	promisor_remote_get_direct(the_repository,
2259	to_fetch.oid, to_fetch.nr);
2260	oid_array_clear(&to_fetch);
2261	}
2262
2263	static void check_object(struct object_entry *entry, uint32_t object_index)
2264	{
2265	size_t canonical_size;
2266	enum object_type type;
2267	struct object_info oi = {.typep = &type, .sizep = &canonical_size};
2268
2269	if (IN_PACK(entry)) {
2270	struct packed_git *p = IN_PACK(entry);
2271	struct pack_window *w_curs = NULL;
2272	int have_base = 0;
2273	struct object_id base_ref;
2274	struct object_entry *base_entry;
2275	unsigned long used, used_0;
2276	size_t avail;
2277	off_t ofs;
2278	unsigned char *buf, c;
2279	enum object_type type;
2280	size_t in_pack_size;
2281
2282	buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
2283
2284	/*
2285	* We want in_pack_type even if we do not reuse delta
2286	* since non-delta representations could still be reused.
2287	*/
2288	used = unpack_object_header_buffer(buf, avail,
2289	&type,
2290	&in_pack_size);
2291	if (used == 0)
2292	goto give_up;
2293
2294	if (type < 0)
2295	BUG("invalid type %d", type);
2296	entry->in_pack_type = type;
2297
2298	/*
2299	* Determine if this is a delta and if so whether we can
2300	* reuse it or not. Otherwise let's find out as cheaply as
2301	* possible what the actual type and size for this object is.
2302	*/
2303	switch (entry->in_pack_type) {
2304	default:
2305	/* Not a delta hence we've already got all we need. */
2306	oe_set_type(entry, entry->in_pack_type);
2307	SET_SIZE(entry, in_pack_size);
2308	entry->in_pack_header_size = used;
2309	if (oe_type(entry) < OBJ_COMMIT \|\| oe_type(entry) > OBJ_BLOB)
2310	goto give_up;
2311	unuse_pack(&w_curs);
2312	return;
2313	case OBJ_REF_DELTA:
2314	if (reuse_delta && !entry->preferred_base) {
2315	oidread(&base_ref,
2316	use_pack(p, &w_curs,
2317	entry->in_pack_offset + used,
2318	NULL),
2319	the_repository->hash_algo);
2320	have_base = 1;
2321	}
2322	entry->in_pack_header_size = used + the_hash_algo->rawsz;
2323	break;
2324	case OBJ_OFS_DELTA:
2325	buf = use_pack(p, &w_curs,
2326	entry->in_pack_offset + used, NULL);
2327	used_0 = 0;
2328	c = buf[used_0++];
2329	ofs = c & 127;
2330	while (c & 128) {
2331	ofs += 1;
2332	if (!ofs \|\| MSB(ofs, 7)) {
2333	error(_("delta base offset overflow in pack for %s"),
2334	oid_to_hex(&entry->idx.oid));
2335	goto give_up;
2336	}
2337	c = buf[used_0++];
2338	ofs = (ofs << 7) + (c & 127);
2339	}
2340	ofs = entry->in_pack_offset - ofs;
2341	if (ofs <= 0 \|\| ofs >= entry->in_pack_offset) {
2342	error(_("delta base offset out of bound for %s"),
2343	oid_to_hex(&entry->idx.oid));
2344	goto give_up;
2345	}
2346	if (reuse_delta && !entry->preferred_base) {
2347	uint32_t pos;
2348	if (offset_to_pack_pos(p, ofs, &pos) < 0)
2349	goto give_up;
2350	if (!nth_packed_object_id(&base_ref, p,
2351	pack_pos_to_index(p, pos)))
2352	have_base = 1;
2353	}
2354	entry->in_pack_header_size = used + used_0;
2355	break;
2356	}
2357
2358	if (have_base &&
2359	can_reuse_delta(&base_ref, entry, &base_entry)) {
2360	oe_set_type(entry, entry->in_pack_type);
2361	SET_SIZE(entry, in_pack_size); /* delta size */
2362	SET_DELTA_SIZE(entry, in_pack_size);
2363
2364	if (base_entry) {
2365	SET_DELTA(entry, base_entry);
2366	entry->delta_sibling_idx = base_entry->delta_child_idx;
2367	SET_DELTA_CHILD(base_entry, entry);
2368	} else {
2369	SET_DELTA_EXT(entry, &base_ref);
2370	}
2371
2372	unuse_pack(&w_curs);
2373	return;
2374	}
2375
2376	if (oe_type(entry)) {
2377	off_t delta_pos;
2378
2379	/*
2380	* This must be a delta and we already know what the
2381	* final object type is. Let's extract the actual
2382	* object size from the delta header.
2383	*/
2384	delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
2385	canonical_size = get_size_from_delta(p, &w_curs,
2386	delta_pos);
2387	if (canonical_size == 0)
2388	goto give_up;
2389	SET_SIZE(entry, canonical_size);
2390	unuse_pack(&w_curs);
2391	return;
2392	}
2393
2394	/*
2395	* No choice but to fall back to the recursive delta walk
2396	* with odb_read_object_info() to find about the object type
2397	* at this point...
2398	*/
2399	give_up:
2400	unuse_pack(&w_curs);
2401	}
2402
2403	if (odb_read_object_info_extended(the_repository->objects, &entry->idx.oid, &oi,
2404	OBJECT_INFO_SKIP_FETCH_OBJECT \| OBJECT_INFO_LOOKUP_REPLACE) < 0) {
2405	if (repo_has_promisor_remote(the_repository)) {
2406	prefetch_to_pack(object_index);
2407	if (odb_read_object_info_extended(the_repository->objects, &entry->idx.oid, &oi,
2408	OBJECT_INFO_SKIP_FETCH_OBJECT \| OBJECT_INFO_LOOKUP_REPLACE) < 0)
2409	type = -1;
2410	} else {
2411	type = -1;
2412	}
2413	}
2414	oe_set_type(entry, type);
2415	if (entry->type_valid) {
2416	SET_SIZE(entry, canonical_size);
2417	} else {
2418	/*
2419	* Bad object type is checked in prepare_pack(). This is
2420	* to permit a missing preferred base object to be ignored
2421	* as a preferred base. Doing so can result in a larger
2422	* pack file, but the transfer will still take place.
2423	*/
2424	}
2425	}
2426
2427	static int pack_offset_sort(const void _a, const void _b)
2428	{
2429	const struct object_entry a = (struct object_entry **)_a;
2430	const struct object_entry b = (struct object_entry **)_b;
2431	const struct packed_git *a_in_pack = IN_PACK(a);
2432	const struct packed_git *b_in_pack = IN_PACK(b);
2433
2434	/* avoid filesystem trashing with loose objects */
2435	if (!a_in_pack && !b_in_pack)
2436	return oidcmp(&a->idx.oid, &b->idx.oid);
2437
2438	if (a_in_pack < b_in_pack)
2439	return -1;
2440	if (a_in_pack > b_in_pack)
2441	return 1;
2442	return a->in_pack_offset < b->in_pack_offset ? -1 :
2443	(a->in_pack_offset > b->in_pack_offset);
2444	}
2445
2446	/*
2447	* Drop an on-disk delta we were planning to reuse. Naively, this would
2448	* just involve blanking out the "delta" field, but we have to deal
2449	* with some extra book-keeping:
2450	*
2451	* 1. Removing ourselves from the delta_sibling linked list.
2452	*
2453	* 2. Updating our size/type to the non-delta representation. These were
2454	* either not recorded initially (size) or overwritten with the delta type
2455	* (type) when check_object() decided to reuse the delta.
2456	*
2457	* 3. Resetting our delta depth, as we are now a base object.
2458	*/
2459	static void drop_reused_delta(struct object_entry *entry)
2460	{
2461	unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
2462	struct object_info oi = OBJECT_INFO_INIT;
2463	enum object_type type;
2464	size_t size;
2465
2466	while (*idx) {
2467	struct object_entry oe = &to_pack.objects[idx - 1];
2468
2469	if (oe == entry)
2470	*idx = oe->delta_sibling_idx;
2471	else
2472	idx = &oe->delta_sibling_idx;
2473	}
2474	SET_DELTA(entry, NULL);
2475	entry->depth = 0;
2476
2477	oi.sizep = &size;
2478	oi.typep = &type;
2479	if (packed_object_info(NULL, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
2480	/*
2481	* We failed to get the info from this pack for some reason;
2482	* fall back to odb_read_object_info, which may find another copy.
2483	* And if that fails, the error will be recorded in oe_type(entry)
2484	* and dealt with in prepare_pack().
2485	*/
2486	oe_set_type(entry,
2487	odb_read_object_info(the_repository->objects,
2488	&entry->idx.oid, &size));
2489	} else {
2490	oe_set_type(entry, type);
2491	}
2492	SET_SIZE(entry, size);
2493	}
2494
2495	/*
2496	* Follow the chain of deltas from this entry onward, throwing away any links
2497	* that cause us to hit a cycle (as determined by the DFS state flags in
2498	* the entries).
2499	*
2500	* We also detect too-long reused chains that would violate our --depth
2501	* limit.
2502	*/
2503	static void break_delta_chains(struct object_entry *entry)
2504	{
2505	/*
2506	* The actual depth of each object we will write is stored as an int,
2507	* as it cannot exceed our int "depth" limit. But before we break
2508	* changes based no that limit, we may potentially go as deep as the
2509	* number of objects, which is elsewhere bounded to a uint32_t.
2510	*/
2511	uint32_t total_depth;
2512	struct object_entry cur, next;
2513
2514	for (cur = entry, total_depth = 0;
2515	cur;
2516	cur = DELTA(cur), total_depth++) {
2517	if (cur->dfs_state == DFS_DONE) {
2518	/*
2519	* We've already seen this object and know it isn't
2520	* part of a cycle. We do need to append its depth
2521	* to our count.
2522	*/
2523	total_depth += cur->depth;
2524	break;
2525	}
2526
2527	/*
2528	* We break cycles before looping, so an ACTIVE state (or any
2529	* other cruft which made its way into the state variable)
2530	* is a bug.
2531	*/
2532	if (cur->dfs_state != DFS_NONE)
2533	BUG("confusing delta dfs state in first pass: %d",
2534	cur->dfs_state);
2535
2536	/*
2537	* Now we know this is the first time we've seen the object. If
2538	* it's not a delta, we're done traversing, but we'll mark it
2539	* done to save time on future traversals.
2540	*/
2541	if (!DELTA(cur)) {
2542	cur->dfs_state = DFS_DONE;
2543	break;
2544	}
2545
2546	/*
2547	* Mark ourselves as active and see if the next step causes
2548	* us to cycle to another active object. It's important to do
2549	* this _before_ we loop, because it impacts where we make the
2550	* cut, and thus how our total_depth counter works.
2551	* E.g., We may see a partial loop like:
2552	*
2553	* A -> B -> C -> D -> B
2554	*
2555	* Cutting B->C breaks the cycle. But now the depth of A is
2556	* only 1, and our total_depth counter is at 3. The size of the
2557	* error is always one less than the size of the cycle we
2558	* broke. Commits C and D were "lost" from A's chain.
2559	*
2560	* If we instead cut D->B, then the depth of A is correct at 3.
2561	* We keep all commits in the chain that we examined.
2562	*/
2563	cur->dfs_state = DFS_ACTIVE;
2564	if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
2565	drop_reused_delta(cur);
2566	cur->dfs_state = DFS_DONE;
2567	break;
2568	}
2569	}
2570
2571	/*
2572	* And now that we've gone all the way to the bottom of the chain, we
2573	* need to clear the active flags and set the depth fields as
2574	* appropriate. Unlike the loop above, which can quit when it drops a
2575	* delta, we need to keep going to look for more depth cuts. So we need
2576	* an extra "next" pointer to keep going after we reset cur->delta.
2577	*/
2578	for (cur = entry; cur; cur = next) {
2579	next = DELTA(cur);
2580
2581	/*
2582	* We should have a chain of zero or more ACTIVE states down to
2583	* a final DONE. We can quit after the DONE, because either it
2584	* has no bases, or we've already handled them in a previous
2585	* call.
2586	*/
2587	if (cur->dfs_state == DFS_DONE)
2588	break;
2589	else if (cur->dfs_state != DFS_ACTIVE)
2590	BUG("confusing delta dfs state in second pass: %d",
2591	cur->dfs_state);
2592
2593	/*
2594	* If the total_depth is more than depth, then we need to snip
2595	* the chain into two or more smaller chains that don't exceed
2596	* the maximum depth. Most of the resulting chains will contain
2597	* (depth + 1) entries (i.e., depth deltas plus one base), and
2598	* the last chain (i.e., the one containing entry) will contain
2599	* whatever entries are left over, namely
2600	* (total_depth % (depth + 1)) of them.
2601	*
2602	* Since we are iterating towards decreasing depth, we need to
2603	* decrement total_depth as we go, and we need to write to the
2604	* entry what its final depth will be after all of the
2605	* snipping. Since we're snipping into chains of length (depth
2606	* + 1) entries, the final depth of an entry will be its
2607	* original depth modulo (depth + 1). Any time we encounter an
2608	* entry whose final depth is supposed to be zero, we snip it
2609	* from its delta base, thereby making it so.
2610	*/
2611	cur->depth = (total_depth--) % (depth + 1);
2612	if (!cur->depth)
2613	drop_reused_delta(cur);
2614
2615	cur->dfs_state = DFS_DONE;
2616	}
2617	}
2618
2619	static void get_object_details(void)
2620	{
2621	uint32_t i;
2622	struct object_entry **sorted_by_offset;
2623
2624	if (progress)
2625	progress_state = start_progress(the_repository,
2626	_("Counting objects"),
2627	to_pack.nr_objects);
2628
2629	CALLOC_ARRAY(sorted_by_offset, to_pack.nr_objects);
2630	for (i = 0; i < to_pack.nr_objects; i++)
2631	sorted_by_offset[i] = to_pack.objects + i;
2632	QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
2633
2634	for (i = 0; i < to_pack.nr_objects; i++) {
2635	struct object_entry *entry = sorted_by_offset[i];
2636	check_object(entry, i);
2637	if (entry->type_valid &&
2638	oe_size_greater_than(&to_pack, entry,
2639	repo_settings_get_big_file_threshold(the_repository)))
2640	entry->no_try_delta = 1;
2641	display_progress(progress_state, i + 1);
2642	}
2643	stop_progress(&progress_state);
2644
2645	/*
2646	* This must happen in a second pass, since we rely on the delta
2647	* information for the whole list being completed.
2648	*/
2649	for (i = 0; i < to_pack.nr_objects; i++)
2650	break_delta_chains(&to_pack.objects[i]);
2651
2652	free(sorted_by_offset);
2653	}
2654
2655	/*
2656	* We search for deltas in a list sorted by type, by filename hash, and then
2657	* by size, so that we see progressively smaller and smaller files.
2658	* That's because we prefer deltas to be from the bigger file
2659	* to the smaller -- deletes are potentially cheaper, but perhaps
2660	* more importantly, the bigger file is likely the more recent
2661	* one. The deepest deltas are therefore the oldest objects which are
2662	* less susceptible to be accessed often.
2663	*/
2664	static int type_size_sort(const void _a, const void _b)
2665	{
2666	const struct object_entry a = (struct object_entry **)_a;
2667	const struct object_entry b = (struct object_entry **)_b;
2668	const enum object_type a_type = oe_type(a);
2669	const enum object_type b_type = oe_type(b);
2670	const unsigned long a_size = SIZE(a);
2671	const unsigned long b_size = SIZE(b);
2672
2673	if (a_type > b_type)
2674	return -1;
2675	if (a_type < b_type)
2676	return 1;
2677	if (a->hash > b->hash)
2678	return -1;
2679	if (a->hash < b->hash)
2680	return 1;
2681	if (a->preferred_base > b->preferred_base)
2682	return -1;
2683	if (a->preferred_base < b->preferred_base)
2684	return 1;
2685	if (use_delta_islands) {
2686	const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);
2687	if (island_cmp)
2688	return island_cmp;
2689	}
2690	if (a_size > b_size)
2691	return -1;
2692	if (a_size < b_size)
2693	return 1;
2694	return a < b ? -1 : (a > b); /* newest first */
2695	}
2696
2697	struct unpacked {
2698	struct object_entry *entry;
2699	void *data;
2700	struct delta_index *index;
2701	unsigned depth;
2702	};
2703
2704	static int delta_cacheable(size_t src_size, size_t trg_size,
2705	size_t delta_size)
2706	{
2707	if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
2708	return 0;
2709
2710	if (delta_size < cache_max_small_delta_size)
2711	return 1;
2712
2713	/* cache delta, if objects are large enough compared to delta size */
2714	if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
2715	return 1;
2716
2717	return 0;
2718	}
2719
2720	/* Protect delta_cache_size */
2721	static pthread_mutex_t cache_mutex;
2722	#define cache_lock() pthread_mutex_lock(&cache_mutex)
2723	#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
2724
2725	/*
2726	* Protect object list partitioning (e.g. struct thread_param) and
2727	* progress_state
2728	*/
2729	static pthread_mutex_t progress_mutex;
2730	#define progress_lock() pthread_mutex_lock(&progress_mutex)
2731	#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
2732
2733	/*
2734	* Access to struct object_entry is unprotected since each thread owns
2735	* a portion of the main object list. Just don't access object entries
2736	* ahead in the list because they can be stolen and would need
2737	* progress_mutex for protection.
2738	*/
2739
2740	static inline int oe_size_less_than(struct packing_data *pack,
2741	const struct object_entry *lhs,
2742	size_t rhs)
2743	{
2744	if (lhs->size_valid)
2745	return lhs->size_ < rhs;
2746	if (rhs < pack->oe_size_limit) /* rhs < 2^x <= lhs ? */
2747	return 0;
2748	return oe_get_size_slow(pack, lhs) < rhs;
2749	}
2750
2751	static inline void oe_set_tree_depth(struct packing_data *pack,
2752	struct object_entry *e,
2753	unsigned int tree_depth)
2754	{
2755	if (!pack->tree_depth)
2756	CALLOC_ARRAY(pack->tree_depth, pack->nr_alloc);
2757	pack->tree_depth[e - pack->objects] = tree_depth;
2758	}
2759
2760	static void record_tree_depth(const struct object_id oid, const char name)
2761	{
2762	const char *p;
2763	unsigned depth;
2764	struct object_entry *ent;
2765
2766	/* the empty string is a root tree, which is depth 0 */
2767	depth = *name ? 1 : 0;
2768	for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))
2769	depth++;
2770
2771	ent = packlist_find(&to_pack, oid);
2772	if (ent && depth > oe_tree_depth(&to_pack, ent))
2773	oe_set_tree_depth(&to_pack, ent, depth);
2774	}
2775
2776	/*
2777	* Return the size of the object without doing any delta
2778	* reconstruction (so non-deltas are true object sizes, but deltas
2779	* return the size of the delta data).
2780	*/
2781	size_t oe_get_size_slow(struct packing_data *pack,
2782	const struct object_entry *e)
2783	{
2784	struct packed_git *p;
2785	struct pack_window *w_curs;
2786	unsigned char *buf;
2787	enum object_type type;
2788	unsigned long used;
2789	size_t avail, size;
2790
2791	if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
2792	size_t sz;
2793	packing_data_lock(&to_pack);
2794	if (odb_read_object_info(the_repository->objects,
2795	&e->idx.oid, &sz) < 0)
2796	die(_("unable to get size of %s"),
2797	oid_to_hex(&e->idx.oid));
2798	packing_data_unlock(&to_pack);
2799	return sz;
2800	}
2801
2802	p = oe_in_pack(pack, e);
2803	if (!p)
2804	BUG("when e->type is a delta, it must belong to a pack");
2805
2806	packing_data_lock(&to_pack);
2807	w_curs = NULL;
2808	buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
2809	used = unpack_object_header_buffer(buf, avail, &type, &size);
2810	if (used == 0)
2811	die(_("unable to parse object header of %s"),
2812	oid_to_hex(&e->idx.oid));
2813
2814	unuse_pack(&w_curs);
2815	packing_data_unlock(&to_pack);
2816	return size;
2817	}
2818
2819	static int try_delta(struct unpacked trg, struct unpacked src,
2820	unsigned max_depth, size_t *mem_usage)
2821	{
2822	struct object_entry *trg_entry = trg->entry;
2823	struct object_entry *src_entry = src->entry;
2824	unsigned long trg_size, src_size, sizediff, max_size, sz;
2825	size_t delta_size;
2826	unsigned ref_depth;
2827	enum object_type type;
2828	void *delta_buf;
2829
2830	/* Don't bother doing diffs between different types */
2831	if (oe_type(trg_entry) != oe_type(src_entry))
2832	return -1;
2833
2834	/*
2835	* We do not bother to try a delta that we discarded on an
2836	* earlier try, but only when reusing delta data. Note that
2837	* src_entry that is marked as the preferred_base should always
2838	* be considered, as even if we produce a suboptimal delta against
2839	* it, we will still save the transfer cost, as we already know
2840	* the other side has it and we won't send src_entry at all.
2841	*/
2842	if (reuse_delta && IN_PACK(trg_entry) &&
2843	IN_PACK(trg_entry) == IN_PACK(src_entry) &&
2844	!src_entry->preferred_base &&
2845	trg_entry->in_pack_type != OBJ_REF_DELTA &&
2846	trg_entry->in_pack_type != OBJ_OFS_DELTA)
2847	return 0;
2848
2849	/* Let's not bust the allowed depth. */
2850	if (src->depth >= max_depth)
2851	return 0;
2852
2853	/* Now some size filtering heuristics. */
2854	trg_size = SIZE(trg_entry);
2855	if (!DELTA(trg_entry)) {
2856	max_size = trg_size/2 - the_hash_algo->rawsz;
2857	ref_depth = 1;
2858	} else {
2859	max_size = DELTA_SIZE(trg_entry);
2860	ref_depth = trg->depth;
2861	}
2862	max_size = (uint64_t)max_size * (max_depth - src->depth) /
2863	(max_depth - ref_depth + 1);
2864	if (max_size == 0)
2865	return 0;
2866	src_size = SIZE(src_entry);
2867	sizediff = src_size < trg_size ? trg_size - src_size : 0;
2868	if (sizediff >= max_size)
2869	return 0;
2870	if (trg_size < src_size / 32)
2871	return 0;
2872
2873	if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))
2874	return 0;
2875
2876	/* Load data if not already done */
2877	if (!trg->data) {
2878	size_t sz_st = 0;
2879	packing_data_lock(&to_pack);
2880	trg->data = odb_read_object(the_repository->objects,
2881	&trg_entry->idx.oid, &type,
2882	&sz_st);
2883	sz = cast_size_t_to_ulong(sz_st);
2884	packing_data_unlock(&to_pack);
2885	if (!trg->data)
2886	die(_("object %s cannot be read"),
2887	oid_to_hex(&trg_entry->idx.oid));
2888	if (sz != trg_size)
2889	die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2890	oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,
2891	(uintmax_t)trg_size);
2892	*mem_usage += sz;
2893	}
2894	if (!src->data) {
2895	size_t sz_st = 0;
2896	packing_data_lock(&to_pack);
2897	src->data = odb_read_object(the_repository->objects,
2898	&src_entry->idx.oid, &type,
2899	&sz_st);
2900	sz = cast_size_t_to_ulong(sz_st);
2901	packing_data_unlock(&to_pack);
2902	if (!src->data) {
2903	if (src_entry->preferred_base) {
2904	static int warned = 0;
2905	if (!warned++)
2906	warning(_("object %s cannot be read"),
2907	oid_to_hex(&src_entry->idx.oid));
2908	/*
2909	* Those objects are not included in the
2910	* resulting pack. Be resilient and ignore
2911	* them if they can't be read, in case the
2912	* pack could be created nevertheless.
2913	*/
2914	return 0;
2915	}
2916	die(_("object %s cannot be read"),
2917	oid_to_hex(&src_entry->idx.oid));
2918	}
2919	if (sz != src_size)
2920	die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2921	oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,
2922	(uintmax_t)src_size);
2923	*mem_usage += sz;
2924	}
2925	if (!src->index) {
2926	src->index = create_delta_index(src->data, src_size);
2927	if (!src->index) {
2928	static int warned = 0;
2929	if (!warned++)
2930	warning(_("suboptimal pack - out of memory"));
2931	return 0;
2932	}
2933	*mem_usage += sizeof_delta_index(src->index);
2934	}
2935
2936	delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
2937	if (!delta_buf)
2938	return 0;
2939
2940	if (DELTA(trg_entry)) {
2941	/* Prefer only shallower same-sized deltas. */
2942	if (delta_size == DELTA_SIZE(trg_entry) &&
2943	src->depth + 1 >= trg->depth) {
2944	free(delta_buf);
2945	return 0;
2946	}
2947	}
2948
2949	/*
2950	* Handle memory allocation outside of the cache
2951	* accounting lock. Compiler will optimize the strangeness
2952	* away when NO_PTHREADS is defined.
2953	*/
2954	free(trg_entry->delta_data);
2955	cache_lock();
2956	if (trg_entry->delta_data) {
2957	delta_cache_size -= DELTA_SIZE(trg_entry);
2958	trg_entry->delta_data = NULL;
2959	}
2960	if (delta_cacheable(src_size, trg_size, delta_size)) {
2961	delta_cache_size += delta_size;
2962	cache_unlock();
2963	trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2964	} else {
2965	cache_unlock();
2966	free(delta_buf);
2967	}
2968
2969	SET_DELTA(trg_entry, src_entry);
2970	SET_DELTA_SIZE(trg_entry, delta_size);
2971	trg->depth = src->depth + 1;
2972
2973	return 1;
2974	}
2975
2976	static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
2977	{
2978	struct object_entry *child = DELTA_CHILD(me);
2979	unsigned int m = n;
2980	while (child) {
2981	const unsigned int c = check_delta_limit(child, n + 1);
2982	if (m < c)
2983	m = c;
2984	child = DELTA_SIBLING(child);
2985	}
2986	return m;
2987	}
2988
2989	static size_t free_unpacked(struct unpacked *n)
2990	{
2991	size_t freed_mem = sizeof_delta_index(n->index);
2992	free_delta_index(n->index);
2993	n->index = NULL;
2994	if (n->data) {
2995	freed_mem += SIZE(n->entry);
2996	FREE_AND_NULL(n->data);
2997	}
2998	n->entry = NULL;
2999	n->depth = 0;
3000	return freed_mem;
3001	}
3002
3003	static void find_deltas(struct object_entry *list, unsigned list_size,
3004	int window, int depth, unsigned *processed)
3005	{
3006	uint32_t i, idx = 0, count = 0;
3007	struct unpacked *array;
3008	size_t mem_usage = 0;
3009
3010	CALLOC_ARRAY(array, window);
3011
3012	for (;;) {
3013	struct object_entry *entry;
3014	struct unpacked *n = array + idx;
3015	int j, max_depth, best_base = -1;
3016
3017	progress_lock();
3018	if (!*list_size) {
3019	progress_unlock();
3020	break;
3021	}
3022	entry = *list++;
3023	(*list_size)--;
3024	if (!entry->preferred_base) {
3025	(*processed)++;
3026	display_progress(progress_state, *processed);
3027	}
3028	progress_unlock();
3029
3030	mem_usage -= free_unpacked(n);
3031	n->entry = entry;
3032
3033	while (window_memory_limit &&
3034	mem_usage > window_memory_limit &&
3035	count > 1) {
3036	const uint32_t tail = (idx + window - count) % window;
3037	mem_usage -= free_unpacked(array + tail);
3038	count--;
3039	}
3040
3041	/* We do not compute delta to create objects we are not
3042	* going to pack.
3043	*/
3044	if (entry->preferred_base)
3045	goto next;
3046
3047	/*
3048	* If the current object is at pack edge, take the depth the
3049	* objects that depend on the current object into account
3050	* otherwise they would become too deep.
3051	*/
3052	max_depth = depth;
3053	if (DELTA_CHILD(entry)) {
3054	max_depth -= check_delta_limit(entry, 0);
3055	if (max_depth <= 0)
3056	goto next;
3057	}
3058
3059	j = window;
3060	while (--j > 0) {
3061	int ret;
3062	uint32_t other_idx = idx + j;
3063	struct unpacked *m;
3064	if (other_idx >= window)
3065	other_idx -= window;
3066	m = array + other_idx;
3067	if (!m->entry)
3068	break;
3069	ret = try_delta(n, m, max_depth, &mem_usage);
3070	if (ret < 0)
3071	break;
3072	else if (ret > 0)
3073	best_base = other_idx;
3074	}
3075
3076	/*
3077	* If we decided to cache the delta data, then it is best
3078	* to compress it right away. First because we have to do
3079	* it anyway, and doing it here while we're threaded will
3080	* save a lot of time in the non threaded write phase,
3081	* as well as allow for caching more deltas within
3082	* the same cache size limit.
3083	* ...
3084	* But only if not writing to stdout, since in that case
3085	* the network is most likely throttling writes anyway,
3086	* and therefore it is best to go to the write phase ASAP
3087	* instead, as we can afford spending more time compressing
3088	* between writes at that moment.
3089	*/
3090	if (entry->delta_data && !pack_to_stdout) {
3091	unsigned long size;
3092
3093	size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
3094	if (size < (1U << OE_Z_DELTA_BITS)) {
3095	entry->z_delta_size = size;
3096	cache_lock();
3097	delta_cache_size -= DELTA_SIZE(entry);
3098	delta_cache_size += entry->z_delta_size;
3099	cache_unlock();
3100	} else {
3101	FREE_AND_NULL(entry->delta_data);
3102	entry->z_delta_size = 0;
3103	}
3104	}
3105
3106	/* if we made n a delta, and if n is already at max
3107	* depth, leaving it in the window is pointless. we
3108	* should evict it first.
3109	*/
3110	if (DELTA(entry) && max_depth <= n->depth)
3111	continue;
3112
3113	/*
3114	* Move the best delta base up in the window, after the
3115	* currently deltified object, to keep it longer. It will
3116	* be the first base object to be attempted next.
3117	*/
3118	if (DELTA(entry)) {
3119	struct unpacked swap = array[best_base];
3120	int dist = (window + idx - best_base) % window;
3121	int dst = best_base;
3122	while (dist--) {
3123	int src = (dst + 1) % window;
3124	array[dst] = array[src];
3125	dst = src;
3126	}
3127	array[dst] = swap;
3128	}
3129
3130	next:
3131	idx++;
3132	if (count + 1 < window)
3133	count++;
3134	if (idx >= window)
3135	idx = 0;
3136	}
3137
3138	for (i = 0; i < window; ++i) {
3139	free_delta_index(array[i].index);
3140	free(array[i].data);
3141	}
3142	free(array);
3143	}
3144
3145	/*
3146	* The main object list is split into smaller lists, each is handed to
3147	* one worker.
3148	*
3149	* The main thread waits on the condition that (at least) one of the workers
3150	* has stopped working (which is indicated in the .working member of
3151	* struct thread_params).
3152	*
3153	* When a work thread has completed its work, it sets .working to 0 and
3154	* signals the main thread and waits on the condition that .data_ready
3155	* becomes 1.
3156	*
3157	* The main thread steals half of the work from the worker that has
3158	* most work left to hand it to the idle worker.
3159	*/
3160
3161	struct thread_params {
3162	pthread_t thread;
3163	struct object_entry **list;
3164	struct packing_region *regions;
3165	unsigned list_size;
3166	unsigned remaining;
3167	int window;
3168	int depth;
3169	int working;
3170	int data_ready;
3171	pthread_mutex_t mutex;
3172	pthread_cond_t cond;
3173	unsigned *processed;
3174	};
3175
3176	static pthread_cond_t progress_cond;
3177
3178	/*
3179	* Mutex and conditional variable can't be statically-initialized on Windows.
3180	*/
3181	static void init_threaded_search(void)
3182	{
3183	pthread_mutex_init(&cache_mutex, NULL);
3184	pthread_mutex_init(&progress_mutex, NULL);
3185	pthread_cond_init(&progress_cond, NULL);
3186	}
3187
3188	static void cleanup_threaded_search(void)
3189	{
3190	pthread_cond_destroy(&progress_cond);
3191	pthread_mutex_destroy(&cache_mutex);
3192	pthread_mutex_destroy(&progress_mutex);
3193	}
3194
3195	static void threaded_find_deltas(void arg)
3196	{
3197	struct thread_params *me = arg;
3198
3199	progress_lock();
3200	while (me->remaining) {
3201	progress_unlock();
3202
3203	find_deltas(me->list, &me->remaining,
3204	me->window, me->depth, me->processed);
3205
3206	progress_lock();
3207	me->working = 0;
3208	pthread_cond_signal(&progress_cond);
3209	progress_unlock();
3210
3211	/*
3212	* We must not set ->data_ready before we wait on the
3213	* condition because the main thread may have set it to 1
3214	* before we get here. In order to be sure that new
3215	* work is available if we see 1 in ->data_ready, it
3216	* was initialized to 0 before this thread was spawned
3217	* and we reset it to 0 right away.
3218	*/
3219	pthread_mutex_lock(&me->mutex);
3220	while (!me->data_ready)
3221	pthread_cond_wait(&me->cond, &me->mutex);
3222	me->data_ready = 0;
3223	pthread_mutex_unlock(&me->mutex);
3224
3225	progress_lock();
3226	}
3227	progress_unlock();
3228	/* leave ->working 1 so that this doesn't get more work assigned */
3229	return NULL;
3230	}
3231
3232	static void ll_find_deltas(struct object_entry **list, unsigned list_size,
3233	int window, int depth, unsigned *processed)
3234	{
3235	struct thread_params *p;
3236	int i, ret, active_threads = 0;
3237
3238	init_threaded_search();
3239
3240	if (delta_search_threads <= 1) {
3241	find_deltas(list, &list_size, window, depth, processed);
3242	cleanup_threaded_search();
3243	return;
3244	}
3245	if (progress > pack_to_stdout)
3246	fprintf_ln(stderr, _("Delta compression using up to %d threads"),
3247	delta_search_threads);
3248	CALLOC_ARRAY(p, delta_search_threads);
3249
3250	/* Partition the work amongst work threads. */
3251	for (i = 0; i < delta_search_threads; i++) {
3252	unsigned sub_size = list_size / (delta_search_threads - i);
3253
3254	/* don't use too small segments or no deltas will be found */
3255	if (sub_size < 2*window && i+1 < delta_search_threads)
3256	sub_size = 0;
3257
3258	p[i].window = window;
3259	p[i].depth = depth;
3260	p[i].processed = processed;
3261	p[i].working = 1;
3262	p[i].data_ready = 0;
3263
3264	/* try to split chunks on "path" boundaries */
3265	while (sub_size && sub_size < list_size &&
3266	list[sub_size]->hash &&
3267	list[sub_size]->hash == list[sub_size-1]->hash)
3268	sub_size++;
3269
3270	p[i].list = list;
3271	p[i].list_size = sub_size;
3272	p[i].remaining = sub_size;
3273
3274	list += sub_size;
3275	list_size -= sub_size;
3276	}
3277
3278	/* Start work threads. */
3279	for (i = 0; i < delta_search_threads; i++) {
3280	if (!p[i].list_size)
3281	continue;
3282	pthread_mutex_init(&p[i].mutex, NULL);
3283	pthread_cond_init(&p[i].cond, NULL);
3284	ret = pthread_create(&p[i].thread, NULL,
3285	threaded_find_deltas, &p[i]);
3286	if (ret)
3287	die(_("unable to create thread: %s"), strerror(ret));
3288	active_threads++;
3289	}
3290
3291	/*
3292	* Now let's wait for work completion. Each time a thread is done
3293	* with its work, we steal half of the remaining work from the
3294	* thread with the largest number of unprocessed objects and give
3295	* it to that newly idle thread. This ensure good load balancing
3296	* until the remaining object list segments are simply too short
3297	* to be worth splitting anymore.
3298	*/
3299	while (active_threads) {
3300	struct thread_params *target = NULL;
3301	struct thread_params *victim = NULL;
3302	unsigned sub_size = 0;
3303
3304	progress_lock();
3305	for (;;) {
3306	for (i = 0; !target && i < delta_search_threads; i++)
3307	if (!p[i].working)
3308	target = &p[i];
3309	if (target)
3310	break;
3311	pthread_cond_wait(&progress_cond, &progress_mutex);
3312	}
3313
3314	for (i = 0; i < delta_search_threads; i++)
3315	if (p[i].remaining > 2*window &&
3316	(!victim \|\| victim->remaining < p[i].remaining))
3317	victim = &p[i];
3318	if (victim) {
3319	sub_size = victim->remaining / 2;
3320	list = victim->list + victim->list_size - sub_size;
3321	while (sub_size && list[0]->hash &&
3322	list[0]->hash == list[-1]->hash) {
3323	list++;
3324	sub_size--;
3325	}
3326	if (!sub_size) {
3327	/*
3328	* It is possible for some "paths" to have
3329	* so many objects that no hash boundary
3330	* might be found. Let's just steal the
3331	* exact half in that case.
3332	*/
3333	sub_size = victim->remaining / 2;
3334	list -= sub_size;
3335	}
3336	target->list = list;
3337	victim->list_size -= sub_size;
3338	victim->remaining -= sub_size;
3339	}
3340	target->list_size = sub_size;
3341	target->remaining = sub_size;
3342	target->working = 1;
3343	progress_unlock();
3344
3345	pthread_mutex_lock(&target->mutex);
3346	target->data_ready = 1;
3347	pthread_cond_signal(&target->cond);
3348	pthread_mutex_unlock(&target->mutex);
3349
3350	if (!sub_size) {
3351	pthread_join(target->thread, NULL);
3352	pthread_cond_destroy(&target->cond);
3353	pthread_mutex_destroy(&target->mutex);
3354	active_threads--;
3355	}
3356	}
3357	cleanup_threaded_search();
3358	free(p);
3359	}
3360
3361	static int obj_is_packed(const struct object_id *oid)
3362	{
3363	return packlist_find(&to_pack, oid) \|\|
3364	(reuse_packfile_bitmap &&
3365	bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid));
3366	}
3367
3368	static void add_tag_chain(const struct object_id *oid)
3369	{
3370	struct tag *tag;
3371
3372	/*
3373	* We catch duplicates already in add_object_entry(), but we'd
3374	* prefer to do this extra check to avoid having to parse the
3375	* tag at all if we already know that it's being packed (e.g., if
3376	* it was included via bitmaps, we would not have parsed it
3377	* previously).
3378	*/
3379	if (obj_is_packed(oid))
3380	return;
3381
3382	tag = lookup_tag(the_repository, oid);
3383	while (1) {
3384	if (!tag \|\| parse_tag(the_repository, tag) \|\| !tag->tagged)
3385	die(_("unable to pack objects reachable from tag %s"),
3386	oid_to_hex(oid));
3387
3388	add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
3389
3390	if (tag->tagged->type != OBJ_TAG)
3391	return;
3392
3393	tag = (struct tag *)tag->tagged;
3394	}
3395	}
3396
3397	static int add_ref_tag(const struct reference ref, void cb_data UNUSED)
3398	{
3399	struct object_id peeled;
3400
3401	if (!reference_get_peeled_oid(the_repository, ref, &peeled) &&
3402	obj_is_packed(&peeled))
3403	add_tag_chain(ref->oid);
3404	return 0;
3405	}
3406
3407	static int should_attempt_deltas(struct object_entry *entry)
3408	{
3409	if (DELTA(entry))
3410	/* This happens if we decided to reuse existing
3411	* delta from a pack. "reuse_delta &&" is implied.
3412	*/
3413	return 0;
3414
3415	if (!entry->type_valid \|\|
3416	oe_size_less_than(&to_pack, entry, 50))
3417	return 0;
3418
3419	if (entry->no_try_delta)
3420	return 0;
3421
3422	if (entry->preferred_base && !allow_ref_delta)
3423	return 0;
3424
3425	if (!entry->preferred_base) {
3426	if (oe_type(entry) < 0)
3427	die(_("unable to get type of object %s"),
3428	oid_to_hex(&entry->idx.oid));
3429	} else if (oe_type(entry) < 0) {
3430	/*
3431	* This object is not found, but we
3432	* don't have to include it anyway.
3433	*/
3434	return 0;
3435	}
3436
3437	return 1;
3438	}
3439
3440	static void find_deltas_for_region(struct object_entry *list,
3441	struct packing_region *region,
3442	unsigned int *processed)
3443	{
3444	struct object_entry **delta_list;
3445	unsigned int delta_list_nr = 0;
3446
3447	ALLOC_ARRAY(delta_list, region->nr);
3448	for (size_t i = 0; i < region->nr; i++) {
3449	struct object_entry *entry = list + region->start + i;
3450	if (should_attempt_deltas(entry))
3451	delta_list[delta_list_nr++] = entry;
3452	}
3453
3454	QSORT(delta_list, delta_list_nr, type_size_sort);
3455	find_deltas(delta_list, &delta_list_nr, window, depth, processed);
3456	free(delta_list);
3457	}
3458
3459	static void find_deltas_by_region(struct object_entry *list,
3460	struct packing_region *regions,
3461	size_t start, size_t nr)
3462	{
3463	unsigned int processed = 0;
3464	size_t progress_nr;
3465
3466	if (!nr)
3467	return;
3468
3469	progress_nr = regions[nr - 1].start + regions[nr - 1].nr;
3470
3471	if (progress)
3472	progress_state = start_progress(the_repository,
3473	_("Compressing objects by path"),
3474	progress_nr);
3475
3476	while (nr--)
3477	find_deltas_for_region(list,
3478	&regions[start++],
3479	&processed);
3480
3481	display_progress(progress_state, progress_nr);
3482	stop_progress(&progress_state);
3483	}
3484
3485	static void threaded_find_deltas_by_path(void arg)
3486	{
3487	struct thread_params *me = arg;
3488
3489	progress_lock();
3490	while (me->remaining) {
3491	while (me->remaining) {
3492	progress_unlock();
3493	find_deltas_for_region(to_pack.objects,
3494	me->regions,
3495	me->processed);
3496	progress_lock();
3497	me->remaining--;
3498	me->regions++;
3499	}
3500
3501	me->working = 0;
3502	pthread_cond_signal(&progress_cond);
3503	progress_unlock();
3504
3505	/*
3506	* We must not set ->data_ready before we wait on the
3507	* condition because the main thread may have set it to 1
3508	* before we get here. In order to be sure that new
3509	* work is available if we see 1 in ->data_ready, it
3510	* was initialized to 0 before this thread was spawned
3511	* and we reset it to 0 right away.
3512	*/
3513	pthread_mutex_lock(&me->mutex);
3514	while (!me->data_ready)
3515	pthread_cond_wait(&me->cond, &me->mutex);
3516	me->data_ready = 0;
3517	pthread_mutex_unlock(&me->mutex);
3518
3519	progress_lock();
3520	}
3521	progress_unlock();
3522	/* leave ->working 1 so that this doesn't get more work assigned */
3523	return NULL;
3524	}
3525
3526	static void ll_find_deltas_by_region(struct object_entry *list,
3527	struct packing_region *regions,
3528	uint32_t start, uint32_t nr)
3529	{
3530	struct thread_params *p;
3531	int i, ret, active_threads = 0;
3532	unsigned int processed = 0;
3533	uint32_t progress_nr;
3534	init_threaded_search();
3535
3536	if (!nr)
3537	return;
3538
3539	progress_nr = regions[nr - 1].start + regions[nr - 1].nr;
3540	if (delta_search_threads <= 1) {
3541	find_deltas_by_region(list, regions, start, nr);
3542	cleanup_threaded_search();
3543	return;
3544	}
3545
3546	if (progress > pack_to_stdout)
3547	fprintf_ln(stderr,
3548	Q_("Path-based delta compression using up to %d thread",
3549	"Path-based delta compression using up to %d threads",
3550	delta_search_threads),
3551	delta_search_threads);
3552	CALLOC_ARRAY(p, delta_search_threads);
3553
3554	if (progress)
3555	progress_state = start_progress(the_repository,
3556	_("Compressing objects by path"),
3557	progress_nr);
3558	/* Partition the work amongst work threads. */
3559	for (i = 0; i < delta_search_threads; i++) {
3560	unsigned sub_size = nr / (delta_search_threads - i);
3561
3562	p[i].window = window;
3563	p[i].depth = depth;
3564	p[i].processed = &processed;
3565	p[i].working = 1;
3566	p[i].data_ready = 0;
3567
3568	p[i].regions = regions;
3569	p[i].list_size = sub_size;
3570	p[i].remaining = sub_size;
3571
3572	regions += sub_size;
3573	nr -= sub_size;
3574	}
3575
3576	/* Start work threads. */
3577	for (i = 0; i < delta_search_threads; i++) {
3578	if (!p[i].list_size)
3579	continue;
3580	pthread_mutex_init(&p[i].mutex, NULL);
3581	pthread_cond_init(&p[i].cond, NULL);
3582	ret = pthread_create(&p[i].thread, NULL,
3583	threaded_find_deltas_by_path, &p[i]);
3584	if (ret)
3585	die(_("unable to create thread: %s"), strerror(ret));
3586	active_threads++;
3587	}
3588
3589	/*
3590	* Now let's wait for work completion. Each time a thread is done
3591	* with its work, we steal half of the remaining work from the
3592	* thread with the largest number of unprocessed objects and give
3593	* it to that newly idle thread. This ensure good load balancing
3594	* until the remaining object list segments are simply too short
3595	* to be worth splitting anymore.
3596	*/
3597	while (active_threads) {
3598	struct thread_params *target = NULL;
3599	struct thread_params *victim = NULL;
3600	unsigned sub_size = 0;
3601
3602	progress_lock();
3603	for (;;) {
3604	for (i = 0; !target && i < delta_search_threads; i++)
3605	if (!p[i].working)
3606	target = &p[i];
3607	if (target)
3608	break;
3609	pthread_cond_wait(&progress_cond, &progress_mutex);
3610	}
3611
3612	for (i = 0; i < delta_search_threads; i++)
3613	if (p[i].remaining > 2*window &&
3614	(!victim \|\| victim->remaining < p[i].remaining))
3615	victim = &p[i];
3616	if (victim) {
3617	sub_size = victim->remaining / 2;
3618	target->regions = victim->regions + victim->remaining - sub_size;
3619	victim->list_size -= sub_size;
3620	victim->remaining -= sub_size;
3621	}
3622	target->list_size = sub_size;
3623	target->remaining = sub_size;
3624	target->working = 1;
3625	progress_unlock();
3626
3627	pthread_mutex_lock(&target->mutex);
3628	target->data_ready = 1;
3629	pthread_cond_signal(&target->cond);
3630	pthread_mutex_unlock(&target->mutex);
3631
3632	if (!sub_size) {
3633	pthread_join(target->thread, NULL);
3634	pthread_cond_destroy(&target->cond);
3635	pthread_mutex_destroy(&target->mutex);
3636	active_threads--;
3637	}
3638	}
3639	cleanup_threaded_search();
3640	free(p);
3641
3642	display_progress(progress_state, progress_nr);
3643	stop_progress(&progress_state);
3644	}
3645
3646	static void prepare_pack(int window, int depth)
3647	{
3648	struct object_entry **delta_list;
3649	uint32_t i, nr_deltas;
3650	unsigned n;
3651
3652	if (use_delta_islands)
3653	resolve_tree_islands(the_repository, progress, &to_pack);
3654
3655	get_object_details();
3656
3657	/*
3658	* If we're locally repacking then we need to be doubly careful
3659	* from now on in order to make sure no stealth corruption gets
3660	* propagated to the new pack. Clients receiving streamed packs
3661	* should validate everything they get anyway so no need to incur
3662	* the additional cost here in that case.
3663	*/
3664	if (!pack_to_stdout)
3665	do_check_packed_object_crc = 1;
3666
3667	if (!to_pack.nr_objects \|\| !window \|\| !depth \|\|
3668	(!allow_ref_delta && !allow_ofs_delta))
3669	return;
3670
3671	if (path_walk)
3672	ll_find_deltas_by_region(to_pack.objects, to_pack.regions,
3673	0, to_pack.nr_regions);
3674
3675	ALLOC_ARRAY(delta_list, to_pack.nr_objects);
3676	nr_deltas = n = 0;
3677
3678	for (i = 0; i < to_pack.nr_objects; i++) {
3679	struct object_entry *entry = to_pack.objects + i;
3680
3681	if (!should_attempt_deltas(entry))
3682	continue;
3683
3684	if (!entry->preferred_base)
3685	nr_deltas++;
3686
3687	delta_list[n++] = entry;
3688	}
3689
3690	if (nr_deltas && n > 1) {
3691	unsigned nr_done = 0;
3692
3693	if (progress)
3694	progress_state = start_progress(the_repository,
3695	_("Compressing objects"),
3696	nr_deltas);
3697	QSORT(delta_list, n, type_size_sort);
3698	ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
3699	stop_progress(&progress_state);
3700	if (nr_done != nr_deltas)
3701	die(_("inconsistency with delta count"));
3702	}
3703	free(delta_list);
3704	}
3705
3706	static int git_pack_config(const char k, const char v,
3707	const struct config_context ctx, void cb)
3708	{
3709	if (!strcmp(k, "pack.window")) {
3710	window = git_config_int(k, v, ctx->kvi);
3711	return 0;
3712	}
3713	if (!strcmp(k, "pack.windowmemory")) {
3714	window_memory_limit = git_config_ulong(k, v, ctx->kvi);
3715	return 0;
3716	}
3717	if (!strcmp(k, "pack.depth")) {
3718	depth = git_config_int(k, v, ctx->kvi);
3719	return 0;
3720	}
3721	if (!strcmp(k, "pack.deltacachesize")) {
3722	max_delta_cache_size = git_config_int(k, v, ctx->kvi);
3723	return 0;
3724	}
3725	if (!strcmp(k, "pack.deltacachelimit")) {
3726	cache_max_small_delta_size = git_config_int(k, v, ctx->kvi);
3727	return 0;
3728	}
3729	if (!strcmp(k, "pack.writebitmaphashcache")) {
3730	if (git_config_bool(k, v))
3731	write_bitmap_options \|= BITMAP_OPT_HASH_CACHE;
3732	else
3733	write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
3734	}
3735
3736	if (!strcmp(k, "pack.writebitmaplookuptable")) {
3737	if (git_config_bool(k, v))
3738	write_bitmap_options \|= BITMAP_OPT_LOOKUP_TABLE;
3739	else
3740	write_bitmap_options &= ~BITMAP_OPT_LOOKUP_TABLE;
3741	}
3742
3743	if (!strcmp(k, "pack.usebitmaps")) {
3744	use_bitmap_index_default = git_config_bool(k, v);
3745	return 0;
3746	}
3747	if (!strcmp(k, "pack.allowpackreuse")) {
3748	int res = git_parse_maybe_bool_text(v);
3749	if (res < 0) {
3750	if (!strcasecmp(v, "single"))
3751	allow_pack_reuse = SINGLE_PACK_REUSE;
3752	else if (!strcasecmp(v, "multi"))
3753	allow_pack_reuse = MULTI_PACK_REUSE;
3754	else
3755	die(_("invalid pack.allowPackReuse value: '%s'"), v);
3756	} else if (res) {
3757	allow_pack_reuse = SINGLE_PACK_REUSE;
3758	} else {
3759	allow_pack_reuse = NO_PACK_REUSE;
3760	}
3761	return 0;
3762	}
3763	if (!strcmp(k, "pack.threads")) {
3764	delta_search_threads = git_config_int(k, v, ctx->kvi);
3765	if (delta_search_threads < 0)
3766	die(_("invalid number of threads specified (%d)"),
3767	delta_search_threads);
3768	if (!HAVE_THREADS && delta_search_threads != 1) {
3769	warning(_("no threads support, ignoring %s"), k);
3770	delta_search_threads = 0;
3771	}
3772	return 0;
3773	}
3774	if (!strcmp(k, "pack.indexversion")) {
3775	pack_idx_opts.version = git_config_int(k, v, ctx->kvi);
3776	if (pack_idx_opts.version > 2)
3777	die(_("bad pack.indexVersion=%"PRIu32),
3778	pack_idx_opts.version);
3779	return 0;
3780	}
3781	if (!strcmp(k, "pack.writereverseindex")) {
3782	if (git_config_bool(k, v))
3783	pack_idx_opts.flags \|= WRITE_REV;
3784	else
3785	pack_idx_opts.flags &= ~WRITE_REV;
3786	return 0;
3787	}
3788	if (!strcmp(k, "uploadpack.blobpackfileuri")) {
3789	struct configured_exclusion *ex;
3790	const char oid_end, pack_end;
3791	/*
3792	* Stores the pack hash. This is not a true object ID, but is
3793	* of the same form.
3794	*/
3795	struct object_id pack_hash;
3796
3797	if (!v)
3798	return config_error_nonbool(k);
3799
3800	ex = xmalloc(sizeof(*ex));
3801	if (parse_oid_hex(v, &ex->e.oid, &oid_end) \|\|
3802	*oid_end != ' ' \|\|
3803	parse_oid_hex(oid_end + 1, &pack_hash, &pack_end) \|\|
3804	*pack_end != ' ')
3805	die(_("value of uploadpack.blobpackfileuri must be "
3806	"of the form '<object-hash> <pack-hash> <uri>' (got '%s')"), v);
3807	if (oidmap_get(&configured_exclusions, &ex->e.oid))
3808	die(_("object already configured in another "
3809	"uploadpack.blobpackfileuri (got '%s')"), v);
3810	ex->pack_hash_hex = xcalloc(1, pack_end - oid_end);
3811	memcpy(ex->pack_hash_hex, oid_end + 1, pack_end - oid_end - 1);
3812	ex->uri = xstrdup(pack_end + 1);
3813	oidmap_put(&configured_exclusions, ex);
3814	}
3815	return git_default_config(k, v, ctx, cb);
3816	}
3817
3818	/* Counters for trace2 output when in --stdin-packs mode. */
3819	static int stdin_packs_found_nr;
3820	static int stdin_packs_hints_nr;
3821
3822	static int add_object_entry_from_pack(const struct object_id *oid,
3823	struct packed_git *p,
3824	uint32_t pos,
3825	void *_data)
3826	{
3827	off_t ofs;
3828	struct object_info oi = OBJECT_INFO_INIT;
3829	enum object_type type = OBJ_NONE;
3830
3831	display_progress(progress_state, ++nr_seen);
3832
3833	if (have_duplicate_entry(oid, 0))
3834	return 0;
3835
3836	stdin_packs_found_nr++;
3837
3838	ofs = nth_packed_object_offset(p, pos);
3839
3840	oi.typep = &type;
3841	if (packed_object_info(NULL, p, ofs, &oi) < 0) {
3842	die(_("could not get type of object %s in pack %s"),
3843	oid_to_hex(oid), p->pack_name);
3844	} else if (type == OBJ_COMMIT) {
3845	struct rev_info *revs = _data;
3846	/*
3847	* commits in included packs are used as starting points
3848	* for the subsequent revision walk
3849	*
3850	* Note that we do want to walk through commits that are
3851	* present in excluded-open ('!') packs to pick up any
3852	* objects reachable from them not present in the
3853	* excluded-closed ('^') packs.
3854	*
3855	* However, we'll only add those objects to the packing
3856	* list after checking `want_object_in_pack()` below.
3857	*/
3858	add_pending_oid(revs, NULL, oid, 0);
3859	}
3860
3861	if (!want_object_in_pack(oid, 0, &p, &ofs))
3862	return 0;
3863
3864	create_object_entry(oid, type, 0, 0, 0, p, ofs);
3865
3866	return 0;
3867	}
3868
3869	static void show_object_pack_hint(struct object object, const char name,
3870	void *data)
3871	{
3872	enum stdin_packs_mode mode = (enum stdin_packs_mode )data;
3873	if (mode == STDIN_PACKS_MODE_FOLLOW \|\|
3874	mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE) {
3875	if (object->type == OBJ_BLOB &&
3876	!odb_has_object(the_repository->objects, &object->oid, 0))
3877	return;
3878	add_object_entry(&object->oid, object->type, name, 0);
3879	} else {
3880	struct object_entry *oe = packlist_find(&to_pack, &object->oid);
3881	if (!oe)
3882	return;
3883
3884	/*
3885	* Our 'to_pack' list was constructed by iterating all
3886	* objects packed in included packs, and so doesn't have
3887	* a non-zero hash field that you would typically pick
3888	* up during a reachability traversal.
3889	*
3890	* Make a best-effort attempt to fill in the ->hash and
3891	* ->no_try_delta fields here in order to perhaps
3892	* improve the delta selection process.
3893	*/
3894	oe->hash = pack_name_hash_fn(name);
3895	oe->no_try_delta = name && no_try_delta(name);
3896
3897	stdin_packs_hints_nr++;
3898	}
3899	}
3900
3901	static void show_commit_pack_hint(struct commit commit, void data)
3902	{
3903	enum stdin_packs_mode mode = (enum stdin_packs_mode )data;
3904
3905	if (mode == STDIN_PACKS_MODE_FOLLOW \|\|
3906	mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE) {
3907	show_object_pack_hint((struct object *)commit, "", data);
3908	return;
3909	}
3910
3911	/* nothing to do; commits don't have a namehash */
3912
3913	}
3914
3915	/*
3916	* stdin_pack_info_kind specifies how a pack specified over stdin
3917	* should be treated when pack-objects is invoked with --stdin-packs.
3918	*
3919	* - STDIN_PACK_INCLUDE: objects in any packs with this flag bit set
3920	* should be included in the output pack, unless they appear in an
3921	* excluded pack.
3922	*
3923	* - STDIN_PACK_EXCLUDE_CLOSED: objects in any packs with this flag
3924	* bit set should be excluded from the output pack.
3925	*
3926	* - STDIN_PACK_EXCLUDE_OPEN: objects in any packs with this flag
3927	* bit set should be excluded from the output pack, but are not
3928	* guaranteed to be closed under reachability.
3929	*
3930	* Objects in packs whose 'kind' bits include STDIN_PACK_INCLUDE or
3931	* STDIN_PACK_EXCLUDE_OPEN are used as traversal tips when invoked
3932	* with --stdin-packs=follow.
3933	*/
3934	enum stdin_pack_info_kind {
3935	STDIN_PACK_INCLUDE = (1<<0),
3936	STDIN_PACK_EXCLUDE_CLOSED = (1<<1),
3937	STDIN_PACK_EXCLUDE_OPEN = (1<<2),
3938	};
3939
3940	struct stdin_pack_info {
3941	struct packed_git *p;
3942	enum stdin_pack_info_kind kind;
3943	};
3944
3945	static int pack_mtime_cmp(const void _a, const void _b)
3946	{
3947	struct stdin_pack_info a = ((const struct string_list_item)_a)->util;
3948	struct stdin_pack_info b = ((const struct string_list_item)_b)->util;
3949
3950	/*
3951	* order packs by descending mtime so that objects are laid out
3952	* roughly as newest-to-oldest
3953	*/
3954	if (a->p->mtime < b->p->mtime)
3955	return 1;
3956	else if (b->p->mtime < a->p->mtime)
3957	return -1;
3958	else
3959	return 0;
3960	}
3961
3962	static int stdin_packs_include_check_obj(struct object obj, void data UNUSED)
3963	{
3964	return !has_object_kept_pack(to_pack.repo, &obj->oid,
3965	KEPT_PACK_IN_CORE);
3966	}
3967
3968	static int stdin_packs_include_check(struct commit commit, void data)
3969	{
3970	return stdin_packs_include_check_obj((struct object *)commit, data);
3971	}
3972
3973	/*
3974	* Flag bit set on commits that belong to an included pack during
3975	* '--stdin-packs=follow-reachable'. Used by the pre-walk to
3976	* identify which reachable commits should be tips for the main
3977	* object traversal.
3978	*/
3979	#define IN_INCLUDED_PACK (1u<<11)
3980
3981	static int mark_included_pack_tip(const struct object_id *oid,
3982	struct packed_git *p,
3983	uint32_t pos,
3984	void *data)
3985	{
3986	struct rev_info *main_revs = data;
3987	off_t ofs = nth_packed_object_offset(p, pos);
3988	enum object_type type;
3989	struct object_info oi = OBJECT_INFO_INIT;
3990	struct object *obj;
3991
3992	oi.typep = &type;
3993	if (packed_object_info(NULL, p, ofs, &oi) < 0)
3994	return 0;
3995	if (type != OBJ_COMMIT && type != OBJ_TAG)
3996	return 0;
3997
3998	obj = parse_object(the_repository, oid);
3999	if (!obj)
4000	return 0;
4001
4002	obj->flags \|= IN_INCLUDED_PACK;
4003
4004	if (type == OBJ_TAG && main_revs)
4005	add_pending_object(main_revs, obj, "");
4006	return 0;
4007	}
4008
4009	static int mark_loose_object_tip(const struct object_id *oid,
4010	struct object_info *oi UNUSED,
4011	void *data)
4012	{
4013	struct rev_info *main_revs = data;
4014	struct object *obj;
4015	enum object_type type;
4016
4017	type = odb_read_object_info(the_repository->objects, oid, NULL);
4018	if (type != OBJ_COMMIT && type != OBJ_TAG)
4019	return 0;
4020
4021	obj = parse_object(the_repository, oid);
4022	if (!obj)
4023	return 0;
4024
4025	obj->flags \|= IN_INCLUDED_PACK;
4026
4027	if (type == OBJ_TAG && main_revs)
4028	add_pending_object(main_revs, obj, "");
4029
4030	return 0;
4031	}
4032
4033	static int add_ref_to_pending(const struct reference ref, void cb_data)
4034	{
4035	struct rev_info *revs = cb_data;
4036	struct object *object;
4037
4038	object = parse_object(the_repository, ref->oid);
4039	if (!object)
4040	return 0;
4041
4042	add_pending_object(revs, object, "");
4043	return 0;
4044	}
4045
4046	static void read_refs_snapshot(const char *refs_snapshot,
4047	struct rev_info *revs)
4048	{
4049	struct strbuf buf = STRBUF_INIT;
4050	struct object_id oid;
4051	FILE *f = xfopen(refs_snapshot, "r");
4052
4053	while (strbuf_getline(&buf, f) != EOF) {
4054	struct object *object;
4055	const char *hex = buf.buf;
4056	const char *end = NULL;
4057
4058	if (*hex == '+')
4059	hex++;
4060
4061	if (parse_oid_hex_algop(hex, &oid, &end,
4062	the_repository->hash_algo) < 0)
4063	die(_("could not parse line: %s"), buf.buf);
4064	if (*end)
4065	die(_("malformed line: %s"), buf.buf);
4066
4067	object = parse_object(the_repository, &oid);
4068	if (!object)
4069	continue;
4070
4071	add_pending_object(revs, object, "");
4072	}
4073
4074	fclose(f);
4075	strbuf_release(&buf);
4076	}
4077
4078	static void stdin_packs_add_reachable_pack_entries(struct string_list *keys,
4079	struct rev_info *revs,
4080	int rev_list_unpacked)
4081	{
4082	struct rev_info pre_walk;
4083	struct commit *commit;
4084	struct string_list_item *item;
4085
4086	/*
4087	* Phase 1: mark commits in included packs, then walk from
4088	* ref tips to discover which of them are reachable. The walk
4089	* halts at excluded-closed packs (via no_kept_objects) and
4090	* continues through excluded-open ones.
4091	*
4092	* Also set include_check on the outer revs so that phase 2
4093	* (the main object traversal) halts at closed packs.
4094	*/
4095	revs->include_check = stdin_packs_include_check;
4096	revs->include_check_obj = stdin_packs_include_check_obj;
4097
4098	for_each_string_list_item(item, keys) {
4099	struct stdin_pack_info *info = item->util;
4100	if (info->kind & STDIN_PACK_INCLUDE)
4101	for_each_object_in_pack(info->p,
4102	mark_included_pack_tip,
4103	revs,
4104	ODB_FOR_EACH_OBJECT_PACK_ORDER);
4105	}
4106
4107	if (rev_list_unpacked) {
4108	/*
4109	* With '--stdin-packs=follow-reachable', specifying
4110	* '--unpacked' instructs pack-objects to pack any loose
4111	* objects which are reachable.
4112	*
4113	* Pretend as if all loose objects are in an included
4114	* pack in order to make them eligible for packing.
4115	*/
4116	struct odb_source *source = revs->repo->objects->sources;
4117	for (; source; source = source->next) {
4118	struct odb_source_files *files = odb_source_files_downcast(source);
4119	struct odb_for_each_object_options opts = { 0 };
4120	if (local)
4121	opts.flags \|= ODB_FOR_EACH_OBJECT_LOCAL_ONLY;
4122
4123	odb_source_for_each_object(&files->loose->base, NULL,
4124	mark_loose_object_tip,
4125	revs, &opts);
4126	}
4127	}
4128
4129	repo_init_revisions(the_repository, &pre_walk, NULL);
4130	pre_walk.no_kept_objects = 1;
4131	pre_walk.keep_pack_cache_flags \|= KEPT_PACK_IN_CORE;
4132	pre_walk.ignore_missing_links = 1;
4133
4134	if (stdin_packs_refs_snapshot)
4135	read_refs_snapshot(stdin_packs_refs_snapshot, &pre_walk);
4136	else
4137	refs_for_each_ref(get_main_ref_store(the_repository),
4138	add_ref_to_pending, &pre_walk);
4139
4140	if (prepare_revision_walk(&pre_walk))
4141	die(_("revision walk setup failed"));
4142
4143	/*
4144	* Phase 2 tips: every reachable commit that is in an
4145	* included pack becomes a starting point for the main
4146	* object traversal.
4147	*/
4148	while ((commit = get_revision(&pre_walk)) != NULL) {
4149	if (commit->object.flags & IN_INCLUDED_PACK)
4150	add_pending_oid(revs, NULL,
4151	&commit->object.oid, 0);
4152	}
4153
4154	reset_revision_walk();
4155	release_revisions(&pre_walk);
4156	}
4157
4158	static void stdin_packs_add_all_pack_entries(struct string_list *keys,
4159	struct rev_info *revs)
4160	{
4161	struct string_list_item *item;
4162
4163	for_each_string_list_item(item, keys) {
4164	struct stdin_pack_info *info = item->util;
4165
4166	if (info->kind & STDIN_PACK_EXCLUDE_OPEN) {
4167	/*
4168	* When open-excluded packs ("!") are present, stop
4169	* the parent walk at closed-excluded ("^") packs.
4170	* Objects behind a "^" boundary are guaranteed to
4171	* have closure and should not be rescued.
4172	*/
4173	revs->include_check = stdin_packs_include_check;
4174	revs->include_check_obj = stdin_packs_include_check_obj;
4175	}
4176
4177	if ((info->kind & STDIN_PACK_INCLUDE) \|\|
4178	(info->kind & STDIN_PACK_EXCLUDE_OPEN))
4179	for_each_object_in_pack(info->p,
4180	add_object_entry_from_pack,
4181	revs,
4182	ODB_FOR_EACH_OBJECT_PACK_ORDER);
4183	}
4184	}
4185
4186	static void stdin_packs_add_pack_entries(struct strmap *packs,
4187	struct rev_info *revs,
4188	enum stdin_packs_mode mode,
4189	int rev_list_unpacked)
4190	{
4191	struct string_list keys = STRING_LIST_INIT_NODUP;
4192	struct hashmap_iter iter;
4193	struct strmap_entry *entry;
4194
4195	strmap_for_each_entry(packs, &iter, entry) {
4196	struct stdin_pack_info *info = entry->value;
4197	if (!info->p)
4198	die(_("could not find pack '%s'"), entry->key);
4199
4200	string_list_append(&keys, entry->key)->util = info;
4201	}
4202
4203	/*
4204	* Order packs by ascending mtime; use QSORT directly to access the
4205	* string_list_item's ->util pointer, which string_list_sort() does not
4206	* provide.
4207	*/
4208	QSORT(keys.items, keys.nr, pack_mtime_cmp);
4209
4210	if (mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE)
4211	stdin_packs_add_reachable_pack_entries(&keys, revs,
4212	rev_list_unpacked);
4213	else
4214	stdin_packs_add_all_pack_entries(&keys, revs);
4215
4216	string_list_clear(&keys, 0);
4217	}
4218
4219	static void stdin_packs_read_input(struct rev_info *revs,
4220	enum stdin_packs_mode mode,
4221	int rev_list_unpacked)
4222	{
4223	struct strbuf buf = STRBUF_INIT;
4224	struct strmap packs = STRMAP_INIT;
4225	struct packed_git *p;
4226
4227	while (strbuf_getline(&buf, stdin) != EOF) {
4228	struct stdin_pack_info *info;
4229	enum stdin_pack_info_kind kind = STDIN_PACK_INCLUDE;
4230	const char *key = buf.buf;
4231
4232	if (!*key)
4233	continue;
4234	else if (*key == '^')
4235	kind = STDIN_PACK_EXCLUDE_CLOSED;
4236	else if (*key == '!' &&
4237	(mode == STDIN_PACKS_MODE_FOLLOW \|\|
4238	mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE))
4239	kind = STDIN_PACK_EXCLUDE_OPEN;
4240
4241	if (kind != STDIN_PACK_INCLUDE)
4242	key++;
4243
4244	info = strmap_get(&packs, key);
4245	if (!info) {
4246	CALLOC_ARRAY(info, 1);
4247	strmap_put(&packs, key, info);
4248	}
4249
4250	info->kind \|= kind;
4251
4252	strbuf_reset(&buf);
4253	}
4254
4255	repo_for_each_pack(the_repository, p) {
4256	struct stdin_pack_info *info;
4257
4258	info = strmap_get(&packs, pack_basename(p));
4259	if (!info)
4260	continue;
4261
4262	if (info->kind & STDIN_PACK_INCLUDE) {
4263	if (exclude_promisor_objects && p->pack_promisor)
4264	die(_("packfile %s is a promisor but --exclude-promisor-objects was given"), p->pack_name);
4265
4266	/*
4267	* Arguments we got on stdin may not even be
4268	* packs. First check that to avoid segfaulting
4269	* later on in e.g. pack_mtime_cmp(), excluded
4270	* packs are handled below.
4271	*/
4272	if (!is_pack_valid(p))
4273	die(_("packfile %s cannot be accessed"), p->pack_name);
4274	}
4275
4276	if (info->kind & STDIN_PACK_EXCLUDE_CLOSED) {
4277	/*
4278	* Marking excluded packs as kept in-core so
4279	* that later calls to add_object_entry()
4280	* discards any objects that are also found in
4281	* excluded packs.
4282	*/
4283	p->pack_keep_in_core = 1;
4284	}
4285
4286	if (info->kind & STDIN_PACK_EXCLUDE_OPEN) {
4287	/*
4288	* Marking excluded open packs as kept in-core
4289	* (open) for the same reason as we marked
4290	* exclude closed packs as kept in-core.
4291	*
4292	* Use a separate flag here to ensure we don't
4293	* halt our traversal at these packs, since they
4294	* are not guaranteed to have closure.
4295	*
4296	*/
4297	p->pack_keep_in_core_open = 1;
4298	}
4299
4300	info->p = p;
4301	}
4302
4303	stdin_packs_add_pack_entries(&packs, revs, mode, rev_list_unpacked);
4304
4305	strbuf_release(&buf);
4306	strmap_clear(&packs, 1);
4307	}
4308
4309	static void add_unreachable_loose_objects(struct rev_info *revs);
4310
4311	static void read_stdin_packs(enum stdin_packs_mode mode, int rev_list_unpacked)
4312	{
4313	int prev_fetch_if_missing = fetch_if_missing;
4314	struct rev_info revs;
4315
4316	/*
4317	* The revision walk may hit objects that are promised, only. As the
4318	* walk is best-effort though we don't want to perform backfill fetches
4319	* for them.
4320	*/
4321	fetch_if_missing = 0;
4322
4323	repo_init_revisions(the_repository, &revs, NULL);
4324	/*
4325	* Use a revision walk to fill in the namehash of objects in the include
4326	* packs. To save time, we'll avoid traversing through objects that are
4327	* in excluded packs.
4328	*
4329	* That may cause us to avoid populating all of the namehash fields of
4330	* all included objects, but our goal is best-effort, since this is only
4331	* an optimization during delta selection.
4332	*/
4333	revs.no_kept_objects = 1;
4334	revs.keep_pack_cache_flags \|= KEPT_PACK_IN_CORE;
4335	revs.blob_objects = 1;
4336	revs.tree_objects = 1;
4337	revs.tag_objects = 1;
4338	revs.ignore_missing_links = 1;
4339	revs.exclude_promisor_objects = exclude_promisor_objects;
4340
4341	/* avoids adding objects in excluded packs */
4342	ignore_packed_keep_in_core = 1;
4343	if (mode == STDIN_PACKS_MODE_FOLLOW \|\|
4344	mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE) {
4345	/*
4346	* In '--stdin-packs=follow' mode, additionally ignore
4347	* objects in excluded-open packs to prevent them from
4348	* appearing in the resulting pack.
4349	*/
4350	ignore_packed_keep_in_core_open = 1;
4351	}
4352	stdin_packs_read_input(&revs, mode, rev_list_unpacked);
4353	if (rev_list_unpacked && mode != STDIN_PACKS_MODE_FOLLOW_REACHABLE)
4354	add_unreachable_loose_objects(&revs);
4355
4356	if (prepare_revision_walk(&revs))
4357	die(_("revision walk setup failed"));
4358	traverse_commit_list(&revs,
4359	show_commit_pack_hint,
4360	show_object_pack_hint,
4361	&mode);
4362
4363	release_revisions(&revs);
4364
4365	trace2_data_intmax("pack-objects", the_repository, "stdin_packs_found",
4366	stdin_packs_found_nr);
4367	trace2_data_intmax("pack-objects", the_repository, "stdin_packs_hints",
4368	stdin_packs_hints_nr);
4369
4370	fetch_if_missing = prev_fetch_if_missing;
4371	}
4372
4373	static void add_cruft_object_entry(const struct object_id *oid, enum object_type type,
4374	struct packed_git *pack, off_t offset,
4375	const char *name, uint32_t mtime)
4376	{
4377	struct object_entry *entry;
4378
4379	display_progress(progress_state, ++nr_seen);
4380
4381	entry = packlist_find(&to_pack, oid);
4382	if (entry) {
4383	if (name) {
4384	entry->hash = pack_name_hash_fn(name);
4385	entry->no_try_delta = no_try_delta(name);
4386	}
4387	} else {
4388	if (!want_object_in_pack_mtime(oid, 0, &pack, &offset, mtime))
4389	return;
4390	if (!pack && type == OBJ_BLOB) {
4391	struct odb_source *source = the_repository->objects->sources;
4392	int found = 0;
4393
4394	for (; !found && source; source = source->next) {
4395	struct odb_source_files *files = odb_source_files_downcast(source);
4396	if (!odb_source_read_object_info(&files->loose->base, oid, NULL, 0))
4397	found = 1;
4398	}
4399
4400	/*
4401	* If a traversed tree has a missing blob then we want
4402	* to avoid adding that missing object to our pack.
4403	*
4404	* This only applies to missing blobs, not trees,
4405	* because the traversal needs to parse sub-trees but
4406	* not blobs.
4407	*
4408	* Note we only perform this check when we couldn't
4409	* already find the object in a pack, so we're really
4410	* limited to "ensure non-tip blobs which don't exist in
4411	* packs do exist via loose objects". Confused?
4412	*/
4413	if (!found)
4414	return;
4415	}
4416
4417	entry = create_object_entry(oid, type, pack_name_hash_fn(name),
4418	0, name && no_try_delta(name),
4419	pack, offset);
4420	}
4421
4422	if (mtime > oe_cruft_mtime(&to_pack, entry))
4423	oe_set_cruft_mtime(&to_pack, entry, mtime);
4424	return;
4425	}
4426
4427	static void show_cruft_object(struct object obj, const char name, void *data UNUSED)
4428	{
4429	/*
4430	* if we did not record it earlier, it's at least as old as our
4431	* expiration value. Rather than find it exactly, just use that
4432	* value. This may bump it forward from its real mtime, but it
4433	* will still be "too old" next time we run with the same
4434	* expiration.
4435	*
4436	* if obj does appear in the packing list, this call is a noop (or may
4437	* set the namehash).
4438	*/
4439	add_cruft_object_entry(&obj->oid, obj->type, NULL, 0, name, cruft_expiration);
4440	}
4441
4442	static void show_cruft_commit(struct commit commit, void data)
4443	{
4444	show_cruft_object((struct object*)commit, NULL, data);
4445	}
4446
4447	static int cruft_include_check_obj(struct object obj, void data UNUSED)
4448	{
4449	return !has_object_kept_pack(to_pack.repo, &obj->oid, KEPT_PACK_IN_CORE);
4450	}
4451
4452	static int cruft_include_check(struct commit commit, void data)
4453	{
4454	return cruft_include_check_obj((struct object*)commit, data);
4455	}
4456
4457	static void set_cruft_mtime(const struct object *object,
4458	struct packed_git *pack,
4459	off_t offset, time_t mtime)
4460	{
4461	add_cruft_object_entry(&object->oid, object->type, pack, offset, NULL,
4462	mtime);
4463	}
4464
4465	static void mark_pack_kept_in_core(struct string_list *packs, unsigned keep)
4466	{
4467	struct string_list_item *item = NULL;
4468	for_each_string_list_item(item, packs) {
4469	struct packed_git *p = item->util;
4470	if (!p)
4471	die(_("could not find pack '%s'"), item->string);
4472	if (p->is_cruft && keep)
4473	ignore_packed_keep_in_core_has_cruft = 1;
4474	p->pack_keep_in_core = keep;
4475	}
4476	}
4477
4478	static void add_objects_in_unpacked_packs(void);
4479
4480	static void enumerate_cruft_objects(void)
4481	{
4482	if (progress)
4483	progress_state = start_progress(the_repository,
4484	_("Enumerating cruft objects"), 0);
4485
4486	add_objects_in_unpacked_packs();
4487	add_unreachable_loose_objects(NULL);
4488
4489	stop_progress(&progress_state);
4490	}
4491
4492	static void enumerate_and_traverse_cruft_objects(struct string_list *fresh_packs)
4493	{
4494	struct packed_git *p;
4495	struct rev_info revs;
4496	int ret;
4497
4498	repo_init_revisions(the_repository, &revs, NULL);
4499
4500	revs.tag_objects = 1;
4501	revs.tree_objects = 1;
4502	revs.blob_objects = 1;
4503
4504	revs.include_check = cruft_include_check;
4505	revs.include_check_obj = cruft_include_check_obj;
4506
4507	revs.ignore_missing_links = 1;
4508
4509	if (progress)
4510	progress_state = start_progress(the_repository,
4511	_("Enumerating cruft objects"), 0);
4512	ret = add_unseen_recent_objects_to_traversal(&revs, cruft_expiration,
4513	set_cruft_mtime, 1);
4514	stop_progress(&progress_state);
4515
4516	if (ret)
4517	die(_("unable to add cruft objects"));
4518
4519	/*
4520	* Re-mark only the fresh packs as kept so that objects in
4521	* unknown packs do not halt the reachability traversal early.
4522	*/
4523	repo_for_each_pack(the_repository, p)
4524	p->pack_keep_in_core = 0;
4525	mark_pack_kept_in_core(fresh_packs, 1);
4526
4527	if (prepare_revision_walk(&revs))
4528	die(_("revision walk setup failed"));
4529	if (progress)
4530	progress_state = start_progress(the_repository,
4531	_("Traversing cruft objects"), 0);
4532	nr_seen = 0;
4533	traverse_commit_list(&revs, show_cruft_commit, show_cruft_object, NULL);
4534
4535	stop_progress(&progress_state);
4536	release_revisions(&revs);
4537	}
4538
4539	static void read_cruft_objects(void)
4540	{
4541	struct strbuf buf = STRBUF_INIT;
4542	struct string_list discard_packs = STRING_LIST_INIT_DUP;
4543	struct string_list fresh_packs = STRING_LIST_INIT_DUP;
4544	struct packed_git *p;
4545
4546	ignore_packed_keep_in_core = 1;
4547
4548	while (strbuf_getline(&buf, stdin) != EOF) {
4549	if (!buf.len)
4550	continue;
4551
4552	if (*buf.buf == '-')
4553	string_list_append(&discard_packs, buf.buf + 1);
4554	else
4555	string_list_append(&fresh_packs, buf.buf);
4556	}
4557
4558	string_list_sort(&discard_packs);
4559	string_list_sort(&fresh_packs);
4560
4561	repo_for_each_pack(the_repository, p) {
4562	const char *pack_name = pack_basename(p);
4563	struct string_list_item *item;
4564
4565	item = string_list_lookup(&fresh_packs, pack_name);
4566	if (!item)
4567	item = string_list_lookup(&discard_packs, pack_name);
4568
4569	if (item) {
4570	item->util = p;
4571	} else {
4572	/*
4573	* This pack wasn't mentioned in either the "fresh" or
4574	* "discard" list, so the caller didn't know about it.
4575	*
4576	* Mark it as kept so that its objects are ignored by
4577	* add_unseen_recent_objects_to_traversal(). We'll
4578	* unmark it before starting the traversal so it doesn't
4579	* halt the traversal early.
4580	*/
4581	p->pack_keep_in_core = 1;
4582	}
4583	}
4584
4585	mark_pack_kept_in_core(&fresh_packs, 1);
4586	mark_pack_kept_in_core(&discard_packs, 0);
4587
4588	if (cruft_expiration)
4589	enumerate_and_traverse_cruft_objects(&fresh_packs);
4590	else
4591	enumerate_cruft_objects();
4592
4593	strbuf_release(&buf);
4594	string_list_clear(&discard_packs, 0);
4595	string_list_clear(&fresh_packs, 0);
4596	}
4597
4598	static void read_object_list_from_stdin(void)
4599	{
4600	char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
4601	struct object_id oid;
4602	const char *p;
4603
4604	for (;;) {
4605	if (!fgets(line, sizeof(line), stdin)) {
4606	if (feof(stdin))
4607	break;
4608	if (!ferror(stdin))
4609	BUG("fgets returned NULL, not EOF, not error!");
4610	if (errno != EINTR)
4611	die_errno("fgets");
4612	clearerr(stdin);
4613	continue;
4614	}
4615	if (line[0] == '-') {
4616	if (get_oid_hex(line+1, &oid))
4617	die(_("expected edge object ID, got garbage:\n %s"),
4618	line);
4619	add_preferred_base(&oid);
4620	continue;
4621	}
4622	if (parse_oid_hex(line, &oid, &p))
4623	die(_("expected object ID, got garbage:\n %s"), line);
4624
4625	add_preferred_base_object(p + 1);
4626	add_object_entry(&oid, OBJ_NONE, p + 1, 0);
4627	}
4628	}
4629
4630	static void show_commit(struct commit commit, void data UNUSED)
4631	{
4632	add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
4633
4634	if (write_bitmap_index)
4635	index_commit_for_bitmap(commit);
4636
4637	if (use_delta_islands)
4638	propagate_island_marks(the_repository, commit);
4639	}
4640
4641	static void show_object(struct object obj, const char name,
4642	void *data UNUSED)
4643	{
4644	add_preferred_base_object(name);
4645	add_object_entry(&obj->oid, obj->type, name, 0);
4646
4647	if (use_delta_islands)
4648	record_tree_depth(&obj->oid, name);
4649	}
4650
4651	static void show_object__ma_allow_any(struct object obj, const char name, void *data)
4652	{
4653	assert(arg_missing_action == MA_ALLOW_ANY);
4654
4655	/*
4656	* Quietly ignore ALL missing objects. This avoids problems with
4657	* staging them now and getting an odd error later.
4658	*/
4659	if (!odb_has_object(the_repository->objects, &obj->oid, 0))
4660	return;
4661
4662	show_object(obj, name, data);
4663	}
4664
4665	static void show_object__ma_allow_promisor(struct object obj, const char name, void *data)
4666	{
4667	assert(arg_missing_action == MA_ALLOW_PROMISOR);
4668
4669	/*
4670	* Quietly ignore EXPECTED missing objects. This avoids problems with
4671	* staging them now and getting an odd error later.
4672	*/
4673	if (!odb_has_object(the_repository->objects, &obj->oid, 0) &&
4674	is_promisor_object(to_pack.repo, &obj->oid))
4675	return;
4676
4677	show_object(obj, name, data);
4678	}
4679
4680	static int option_parse_missing_action(const struct option *opt UNUSED,
4681	const char *arg, int unset)
4682	{
4683	assert(arg);
4684	assert(!unset);
4685
4686	if (!strcmp(arg, "error")) {
4687	arg_missing_action = MA_ERROR;
4688	fn_show_object = show_object;
4689	return 0;
4690	}
4691
4692	if (!strcmp(arg, "allow-any")) {
4693	arg_missing_action = MA_ALLOW_ANY;
4694	fetch_if_missing = 0;
4695	fn_show_object = show_object__ma_allow_any;
4696	return 0;
4697	}
4698
4699	if (!strcmp(arg, "allow-promisor")) {
4700	arg_missing_action = MA_ALLOW_PROMISOR;
4701	fetch_if_missing = 0;
4702	fn_show_object = show_object__ma_allow_promisor;
4703	return 0;
4704	}
4705
4706	die(_("invalid value for '%s': '%s'"), "--missing", arg);
4707	return 0;
4708	}
4709
4710	static void show_edge(struct commit *commit)
4711	{
4712	add_preferred_base(&commit->object.oid);
4713	}
4714
4715	static int add_object_in_unpacked_pack(const struct object_id *oid,
4716	struct object_info *oi,
4717	void *data UNUSED)
4718	{
4719	if (cruft) {
4720	add_cruft_object_entry(oid, OBJ_NONE, oi->source_infop->u.packed.pack,
4721	oi->source_infop->u.packed.offset, NULL,
4722	*oi->mtimep);
4723	} else {
4724	add_object_entry(oid, OBJ_NONE, "", 0);
4725	}
4726	return 0;
4727	}
4728
4729	static void add_objects_in_unpacked_packs(void)
4730	{
4731	struct odb_source *source;
4732	time_t mtime;
4733	struct odb_for_each_object_options opts = {
4734	.flags = ODB_FOR_EACH_OBJECT_PACK_ORDER \|
4735	ODB_FOR_EACH_OBJECT_LOCAL_ONLY \|
4736	ODB_FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS \|
4737	ODB_FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS,
4738	};
4739	struct odb_source_info source_info;
4740	struct object_info oi = {
4741	.mtimep = &mtime,
4742	.source_infop = &source_info,
4743	};
4744
4745	odb_prepare_alternates(to_pack.repo->objects);
4746	for (source = to_pack.repo->objects->sources; source; source = source->next) {
4747	struct odb_source_files *files = odb_source_files_downcast(source);
4748
4749	if (!source->local)
4750	continue;
4751
4752	if (odb_source_for_each_object(&files->packed->base, &oi,
4753	add_object_in_unpacked_pack, NULL, &opts))
4754	die(_("cannot open pack index"));
4755	}
4756	}
4757
4758	static int add_loose_object(const struct object_id oid, const char path,
4759	void *data)
4760	{
4761	struct rev_info *revs = data;
4762	enum object_type type = odb_read_object_info(the_repository->objects, oid, NULL);
4763
4764	if (type < 0) {
4765	warning(_("loose object at %s could not be examined"), path);
4766	return 0;
4767	}
4768
4769	if (cruft) {
4770	struct stat st;
4771	if (stat(path, &st) < 0) {
4772	if (errno == ENOENT)
4773	return 0;
4774	return error_errno("unable to stat %s", oid_to_hex(oid));
4775	}
4776
4777	add_cruft_object_entry(oid, type, NULL, 0, NULL,
4778	st.st_mtime);
4779	} else {
4780	add_object_entry(oid, type, "", 0);
4781	}
4782
4783	if (revs && type == OBJ_COMMIT)
4784	add_pending_oid(revs, NULL, oid, 0);
4785
4786	return 0;
4787	}
4788
4789	/*
4790	* We actually don't even have to worry about reachability here.
4791	* add_object_entry will weed out duplicates, so we just add every
4792	* loose object we find.
4793	*/
4794	static void add_unreachable_loose_objects(struct rev_info *revs)
4795	{
4796	for_each_loose_file_in_source(the_repository->objects->sources,
4797	add_loose_object, NULL, NULL, revs);
4798	}
4799
4800	static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
4801	{
4802	static struct packed_git *last_found = NULL;
4803	struct packed_git *p;
4804
4805	if (last_found && find_pack_entry_one(oid, last_found))
4806	return 1;
4807
4808	repo_for_each_pack(the_repository, p) {
4809	/*
4810	* We have already checked `last_found`, so there is no need to
4811	* re-check here.
4812	*/
4813	if (p == last_found)
4814	continue;
4815
4816	if ((!p->pack_local \|\| p->pack_keep \|\| p->pack_keep_in_core) &&
4817	find_pack_entry_one(oid, p)) {
4818	last_found = p;
4819	return 1;
4820	}
4821	}
4822
4823	return 0;
4824	}
4825
4826	/*
4827	* Store a list of sha1s that are should not be discarded
4828	* because they are either written too recently, or are
4829	* reachable from another object that was.
4830	*
4831	* This is filled by get_object_list.
4832	*/
4833	static struct oid_array recent_objects;
4834
4835	static int loosened_object_can_be_discarded(const struct object_id *oid,
4836	timestamp_t mtime)
4837	{
4838	if (!unpack_unreachable_expiration)
4839	return 0;
4840	if (mtime > unpack_unreachable_expiration)
4841	return 0;
4842	if (oid_array_lookup(&recent_objects, oid) >= 0)
4843	return 0;
4844	return 1;
4845	}
4846
4847	static int force_object_loose(struct odb_source *source,
4848	const struct object_id *oid,
4849	const time_t *mtime)
4850	{
4851	struct odb_source_files *files = odb_source_files_downcast(source);
4852	const struct git_hash_algo *compat = source->odb->repo->compat_hash_algo;
4853	struct object_info oi = OBJECT_INFO_INIT;
4854	struct object_id compat_oid, *compat_oid_p = NULL;
4855	enum object_type type;
4856	void *buf = NULL;
4857	size_t len;
4858	int ret;
4859
4860	for (struct odb_source *s = source->odb->sources; s; s = s->next) {
4861	struct odb_source_files *files = odb_source_files_downcast(s);
4862	if (!odb_source_read_object_info(&files->loose->base, oid, NULL, 0))
4863	return 0;
4864	}
4865
4866	oi.typep = &type;
4867	oi.sizep = &len;
4868	oi.contentp = &buf;
4869	if (odb_read_object_info_extended(source->odb, oid, &oi, 0)) {
4870	ret = error(_("cannot read object for %s"), oid_to_hex(oid));
4871	goto out;
4872	}
4873
4874	if (compat) {
4875	if (repo_oid_to_algop(source->odb->repo, oid, compat, &compat_oid)) {
4876	ret = error(_("cannot map object %s to %s"),
4877	oid_to_hex(oid), compat->name);
4878	goto out;
4879	}
4880
4881	compat_oid_p = &compat_oid;
4882	}
4883
4884	ret = odb_source_write_object(&files->loose->base, buf, len, type, oid,
4885	compat_oid_p, mtime, 0);
4886
4887	out:
4888	free(buf);
4889	return ret;
4890	}
4891
4892	static void loosen_unused_packed_objects(void)
4893	{
4894	struct packed_git *p;
4895	uint32_t i;
4896	uint32_t loosened_objects_nr = 0;
4897	struct object_id oid;
4898
4899	repo_for_each_pack(the_repository, p) {
4900	if (!p->pack_local \|\| p->pack_keep \|\| p->pack_keep_in_core)
4901	continue;
4902
4903	if (open_pack_index(p))
4904	die(_("cannot open pack index"));
4905
4906	for (i = 0; i < p->num_objects; i++) {
4907	nth_packed_object_id(&oid, p, i);
4908	if (!packlist_find(&to_pack, &oid) &&
4909	!has_sha1_pack_kept_or_nonlocal(&oid) &&
4910	!loosened_object_can_be_discarded(&oid, p->mtime)) {
4911	if (force_object_loose(the_repository->objects->sources,
4912	&oid, &p->mtime))
4913	die(_("unable to force loose object"));
4914	loosened_objects_nr++;
4915	}
4916	}
4917	}
4918
4919	trace2_data_intmax("pack-objects", the_repository,
4920	"loosen_unused_packed_objects/loosened", loosened_objects_nr);
4921	}
4922
4923	/*
4924	* This tracks any options which pack-reuse code expects to be on, or which a
4925	* reader of the pack might not understand, and which would therefore prevent
4926	* blind reuse of what we have on disk.
4927	*/
4928	static int pack_options_allow_reuse(void)
4929	{
4930	return allow_pack_reuse != NO_PACK_REUSE &&
4931	pack_to_stdout &&
4932	!ignore_packed_keep_on_disk &&
4933	!ignore_packed_keep_in_core &&
4934	(!local \|\| !have_non_local_packs) &&
4935	!incremental && (allow_ref_delta \|\| allow_ofs_delta);
4936	}
4937
4938	static int get_object_list_from_bitmap(struct rev_info *revs)
4939	{
4940	if (!(bitmap_git = prepare_bitmap_walk(revs, 0)))
4941	return -1;
4942
4943	/*
4944	* For now, force the name-hash version to be 1 since that
4945	* is the version implied by the bitmap format. Later, the
4946	* format can include this version explicitly in its format,
4947	* allowing readers to know the version that was used during
4948	* the bitmap write.
4949	*/
4950	name_hash_version = 1;
4951
4952	if (pack_options_allow_reuse())
4953	reuse_partial_packfile_from_bitmap(bitmap_git,
4954	&reuse_packfiles,
4955	&reuse_packfiles_nr,
4956	&reuse_packfile_bitmap,
4957	allow_pack_reuse == MULTI_PACK_REUSE,
4958	allow_ref_delta);
4959
4960	if (reuse_packfiles) {
4961	reuse_packfile_objects = bitmap_popcount(reuse_packfile_bitmap);
4962	if (!reuse_packfile_objects)
4963	BUG("expected non-empty reuse bitmap");
4964
4965	nr_result += reuse_packfile_objects;
4966	nr_seen += reuse_packfile_objects;
4967	display_progress(progress_state, nr_seen);
4968	}
4969
4970	traverse_bitmap_commit_list(bitmap_git, revs,
4971	&add_object_entry_from_bitmap);
4972	return 0;
4973	}
4974
4975	static void record_recent_object(struct object *obj,
4976	const char *name UNUSED,
4977	void *data UNUSED)
4978	{
4979	oid_array_append(&recent_objects, &obj->oid);
4980	}
4981
4982	static void record_recent_commit(struct commit commit, void data UNUSED)
4983	{
4984	oid_array_append(&recent_objects, &commit->object.oid);
4985	}
4986
4987	static int mark_bitmap_preferred_tip(const struct reference ref, void data UNUSED)
4988	{
4989	const struct object_id *maybe_peeled = ref->oid;
4990	struct object_id peeled;
4991	struct object *object;
4992
4993	if (!reference_get_peeled_oid(the_repository, ref, &peeled))
4994	maybe_peeled = &peeled;
4995
4996	object = parse_object_or_die(the_repository, maybe_peeled, ref->name);
4997	if (object->type == OBJ_COMMIT)
4998	object->flags \|= NEEDS_BITMAP;
4999
5000	return 0;

Showing first 5,000 of 5,770 lines. View raw