pack-objects.c · matheusbsbs/matheus-bagliano

seen

bisect jch maint master next seen test todo

c 5,768 lines 157 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	reuse_packfile->bitmap_nr != reuse_packfile->p->num_objects) {
1193	/*
1194	* We can't reuse whole chunks verbatim from non-preferred
1195	* packs or packs with entries missing from the bitmap because
1196	* bitmap and pack positions may differ.
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 these 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	reuse_packfile->bitmap_nr != reuse_packfile->p->num_objects) {
1274	/*
1275	* Translate MIDX bitmap positions which do not
1276	* correspond directly to physical pack positions.
1277	*/
1278	struct multi_pack_index *m = reuse_packfile->from_midx;
1279	uint32_t midx_pos;
1280	off_t pack_ofs;
1281
1282	if (!m)
1283	BUG("cannot translate bitmap position without MIDX");
1284
1285	midx_pos = pack_pos_to_midx(m, pos + offset);
1286	pack_ofs = nth_midxed_offset(m, midx_pos);
1287
1288	if (offset_to_pack_pos(reuse_packfile->p,
1289	pack_ofs, &pack_pos) < 0)
1290	BUG("could not find expected object at offset %"PRIuMAX" in pack %s",
1291	(uintmax_t)pack_ofs,
1292	pack_basename(reuse_packfile->p));
1293	} else {
1294	/*
1295	* Can use bit positions directly, even for MIDX
1296	* bitmaps. See comment in try_partial_reuse()
1297	* for why.
1298	*/
1299	pack_pos = pos + offset;
1300	}
1301
1302	write_reused_pack_one(reuse_packfile->p, pack_pos, f,
1303	pack_start, &w_curs);
1304	display_progress(progress_state, ++written);
1305	}
1306	}
1307
1308	done:
1309	unuse_pack(&w_curs);
1310	}
1311
1312	static void write_excluded_by_configs(void)
1313	{
1314	struct oidset_iter iter;
1315	const struct object_id *oid;
1316
1317	oidset_iter_init(&excluded_by_config, &iter);
1318	while ((oid = oidset_iter_next(&iter))) {
1319	struct configured_exclusion *ex =
1320	oidmap_get(&configured_exclusions, oid);
1321
1322	if (!ex)
1323	BUG("configured exclusion wasn't configured");
1324	write_in_full(1, ex->pack_hash_hex, strlen(ex->pack_hash_hex));
1325	write_in_full(1, " ", 1);
1326	write_in_full(1, ex->uri, strlen(ex->uri));
1327	write_in_full(1, "\n", 1);
1328	}
1329	}
1330
1331	static const char no_split_warning[] = N_(
1332	"disabling bitmap writing, packs are split due to pack.packSizeLimit"
1333	);
1334
1335	static void write_pack_file(void)
1336	{
1337	uint32_t i = 0, j;
1338	struct hashfile *f;
1339	off_t offset;
1340	uint32_t nr_remaining = nr_result;
1341	time_t last_mtime = 0;
1342	struct object_entry **write_order;
1343
1344	if (progress > pack_to_stdout)
1345	progress_state = start_progress(the_repository,
1346	_("Writing objects"), nr_result);
1347	ALLOC_ARRAY(written_list, to_pack.nr_objects);
1348	write_order = compute_write_order();
1349
1350	do {
1351	unsigned char hash[GIT_MAX_RAWSZ];
1352	char *pack_tmp_name = NULL;
1353
1354	if (pack_to_stdout) {
1355	/*
1356	* This command is most often invoked via
1357	* git-upload-pack(1), which will typically chunk data
1358	* into pktlines. As such, we use the maximum data
1359	* length of them as buffer length.
1360	*
1361	* Note that we need to subtract one though to
1362	* accommodate for the sideband byte.
1363	*/
1364	struct hashfd_options opts = {
1365	.progress = progress_state,
1366	.buffer_len = LARGE_PACKET_DATA_MAX - 1,
1367	};
1368	f = hashfd_ext(the_repository->hash_algo, 1,
1369	"<stdout>", &opts);
1370	} else {
1371	f = create_tmp_packfile(the_repository, &pack_tmp_name);
1372	}
1373
1374	offset = write_pack_header(f, nr_remaining);
1375
1376	if (reuse_packfiles_nr) {
1377	assert(pack_to_stdout);
1378	for (j = 0; j < reuse_packfiles_nr; j++) {
1379	reused_chunks_nr = 0;
1380	write_reused_pack(&reuse_packfiles[j], f);
1381	if (reused_chunks_nr)
1382	reuse_packfiles_used_nr++;
1383	}
1384	offset = hashfile_total(f);
1385	}
1386
1387	nr_written = 0;
1388	for (; i < to_pack.nr_objects; i++) {
1389	struct object_entry *e = write_order[i];
1390	if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
1391	break;
1392	display_progress(progress_state, written);
1393	}
1394
1395	if (pack_to_stdout) {
1396	/*
1397	* We never fsync when writing to stdout since we may
1398	* not be writing to an actual pack file. For instance,
1399	* the upload-pack code passes a pipe here. Calling
1400	* fsync on a pipe results in unnecessary
1401	* synchronization with the reader on some platforms.
1402	*/
1403	finalize_hashfile(f, hash, FSYNC_COMPONENT_NONE,
1404	CSUM_HASH_IN_STREAM \| CSUM_CLOSE);
1405	} else if (nr_written == nr_remaining) {
1406	finalize_hashfile(f, hash, FSYNC_COMPONENT_PACK,
1407	CSUM_HASH_IN_STREAM \| CSUM_FSYNC \| CSUM_CLOSE);
1408	} else {
1409	/*
1410	* If we wrote the wrong number of entries in the
1411	* header, rewrite it like in fast-import.
1412	*/
1413
1414	int fd = finalize_hashfile(f, hash, FSYNC_COMPONENT_PACK, 0);
1415	fixup_pack_header_footer(the_hash_algo, fd, hash,
1416	pack_tmp_name, nr_written,
1417	hash, offset);
1418	close(fd);
1419	if (write_bitmap_index) {
1420	if (write_bitmap_index != WRITE_BITMAP_QUIET)
1421	warning(_(no_split_warning));
1422	write_bitmap_index = 0;
1423	}
1424	}
1425
1426	if (!pack_to_stdout) {
1427	struct stat st;
1428	struct strbuf tmpname = STRBUF_INIT;
1429	struct bitmap_writer bitmap_writer;
1430	char *idx_tmp_name = NULL;
1431
1432	/*
1433	* Packs are runtime accessed in their mtime
1434	* order since newer packs are more likely to contain
1435	* younger objects. So if we are creating multiple
1436	* packs then we should modify the mtime of later ones
1437	* to preserve this property.
1438	*/
1439	if (stat(pack_tmp_name, &st) < 0) {
1440	warning_errno(_("failed to stat %s"), pack_tmp_name);
1441	} else if (!last_mtime) {
1442	last_mtime = st.st_mtime;
1443	} else {
1444	struct utimbuf utb;
1445	utb.actime = st.st_atime;
1446	utb.modtime = --last_mtime;
1447	if (utime(pack_tmp_name, &utb) < 0)
1448	warning_errno(_("failed utime() on %s"), pack_tmp_name);
1449	}
1450
1451	strbuf_addf(&tmpname, "%s-%s.", base_name,
1452	hash_to_hex(hash));
1453
1454	if (write_bitmap_index) {
1455	bitmap_writer_init(&bitmap_writer,
1456	the_repository, &to_pack,
1457	NULL);
1458	bitmap_writer_set_checksum(&bitmap_writer, hash);
1459	bitmap_writer_build_type_index(&bitmap_writer,
1460	written_list);
1461	}
1462
1463	if (cruft)
1464	pack_idx_opts.flags \|= WRITE_MTIMES;
1465
1466	stage_tmp_packfiles(the_repository, &tmpname,
1467	pack_tmp_name, written_list,
1468	nr_written, &to_pack,
1469	&pack_idx_opts, hash,
1470	&idx_tmp_name);
1471
1472	if (write_bitmap_index) {
1473	size_t tmpname_len = tmpname.len;
1474
1475	strbuf_addstr(&tmpname, "bitmap");
1476	stop_progress(&progress_state);
1477
1478	bitmap_writer_show_progress(&bitmap_writer,
1479	progress);
1480	bitmap_writer_select_commits(&bitmap_writer,
1481	indexed_commits,
1482	indexed_commits_nr);
1483	if (bitmap_writer_build(&bitmap_writer) < 0)
1484	die(_("failed to write bitmap index"));
1485	bitmap_writer_finish(&bitmap_writer,
1486	written_list,
1487	tmpname.buf, write_bitmap_options);
1488	bitmap_writer_free(&bitmap_writer);
1489	write_bitmap_index = 0;
1490	strbuf_setlen(&tmpname, tmpname_len);
1491	}
1492
1493	rename_tmp_packfile_idx(the_repository, &tmpname, &idx_tmp_name);
1494
1495	free(idx_tmp_name);
1496	strbuf_release(&tmpname);
1497	free(pack_tmp_name);
1498	puts(hash_to_hex(hash));
1499	}
1500
1501	/* mark written objects as written to previous pack */
1502	for (j = 0; j < nr_written; j++) {
1503	written_list[j]->offset = (off_t)-1;
1504	}
1505	nr_remaining -= nr_written;
1506	} while (nr_remaining && i < to_pack.nr_objects);
1507
1508	free(written_list);
1509	free(write_order);
1510	stop_progress(&progress_state);
1511	if (written != nr_result)
1512	die(_("wrote %"PRIu32" objects while expecting %"PRIu32),
1513	written, nr_result);
1514	trace2_data_intmax("pack-objects", the_repository,
1515	"write_pack_file/wrote", nr_result);
1516	}
1517
1518	static int no_try_delta(const char *path)
1519	{
1520	static struct attr_check *check;
1521
1522	if (!check)
1523	check = attr_check_initl("delta", NULL);
1524	git_check_attr(the_repository->index, path, check);
1525	if (ATTR_FALSE(check->items[0].value))
1526	return 1;
1527	return 0;
1528	}
1529
1530	/*
1531	* When adding an object, check whether we have already added it
1532	* to our packing list. If so, we can skip. However, if we are
1533	* being asked to excludei t, but the previous mention was to include
1534	* it, make sure to adjust its flags and tweak our numbers accordingly.
1535	*
1536	* As an optimization, we pass out the index position where we would have
1537	* found the item, since that saves us from having to look it up again a
1538	* few lines later when we want to add the new entry.
1539	*/
1540	static int have_duplicate_entry(const struct object_id *oid,
1541	int exclude)
1542	{
1543	struct object_entry *entry;
1544
1545	if (reuse_packfile_bitmap &&
1546	bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid))
1547	return 1;
1548
1549	entry = packlist_find(&to_pack, oid);
1550	if (!entry)
1551	return 0;
1552
1553	if (exclude) {
1554	if (!entry->preferred_base)
1555	nr_result--;
1556	entry->preferred_base = 1;
1557	}
1558
1559	return 1;
1560	}
1561
1562	static int want_cruft_object_mtime(struct repository *r,
1563	const struct object_id *oid,
1564	unsigned flags, uint32_t mtime)
1565	{
1566	struct odb_source *source;
1567
1568	for (source = r->objects->sources; source; source = source->next) {
1569	struct odb_source_files *files = odb_source_files_downcast(source);
1570	struct packed_git **cache = packfile_store_get_kept_pack_cache(files->packed, flags);
1571
1572	for (; *cache; cache++) {
1573	struct packed_git p = cache;
1574	off_t ofs;
1575	uint32_t candidate_mtime;
1576
1577	ofs = find_pack_entry_one(oid, p);
1578	if (!ofs)
1579	continue;
1580
1581	/*
1582	* We have a copy of the object 'oid' in a non-cruft
1583	* pack. We can avoid packing an additional copy
1584	* regardless of what the existing copy's mtime is since
1585	* it is outside of a cruft pack.
1586	*/
1587	if (!p->is_cruft)
1588	return 0;
1589
1590	/*
1591	* If we have a copy of the object 'oid' in a cruft
1592	* pack, then either read the cruft pack's mtime for
1593	* that object, or, if that can't be loaded, assume the
1594	* pack's mtime itself.
1595	*/
1596	if (!load_pack_mtimes(p)) {
1597	uint32_t pos;
1598	if (offset_to_pack_pos(p, ofs, &pos) < 0)
1599	continue;
1600	candidate_mtime = nth_packed_mtime(p, pos);
1601	} else {
1602	candidate_mtime = p->mtime;
1603	}
1604
1605	/*
1606	* We have a surviving copy of the object in a cruft
1607	* pack whose mtime is greater than or equal to the one
1608	* we are considering. We can thus avoid packing an
1609	* additional copy of that object.
1610	*/
1611	if (mtime <= candidate_mtime)
1612	return 0;
1613	}
1614	}
1615
1616	return -1;
1617	}
1618
1619	static int want_found_object(const struct object_id *oid, int exclude,
1620	struct packed_git *p, uint32_t mtime)
1621	{
1622	if (exclude)
1623	return 1;
1624	if (incremental)
1625	return 0;
1626
1627	if (!is_pack_valid(p))
1628	return -1;
1629
1630	/*
1631	* When asked to do --local (do not include an object that appears in a
1632	* pack we borrow from elsewhere) or --honor-pack-keep (do not include
1633	* an object that appears in a pack marked with .keep), finding a pack
1634	* that matches the criteria is sufficient for us to decide to omit it.
1635	* However, even if this pack does not satisfy the criteria, we need to
1636	* make sure no copy of this object appears in _any_ pack that makes us
1637	* to omit the object, so we need to check all the packs.
1638	*
1639	* We can however first check whether these options can possibly matter;
1640	* if they do not matter we know we want the object in generated pack.
1641	* Otherwise, we signal "-1" at the end to tell the caller that we do
1642	* not know either way, and it needs to check more packs.
1643	*/
1644
1645	/*
1646	* Objects in packs borrowed from elsewhere are discarded regardless of
1647	* if they appear in other packs that weren't borrowed.
1648	*/
1649	if (local && !p->pack_local)
1650	return 0;
1651
1652	/*
1653	* Then handle .keep first, as we have a fast(er) path there.
1654	*/
1655	if (ignore_packed_keep_on_disk \|\| ignore_packed_keep_in_core \|\|
1656	ignore_packed_keep_in_core_open) {
1657	/*
1658	* Set the flags for the kept-pack cache to be the ones we want
1659	* to ignore.
1660	*
1661	* That is, if we are ignoring objects in on-disk keep packs,
1662	* then we want to search through the on-disk keep and ignore
1663	* the in-core ones.
1664	*/
1665	unsigned flags = 0;
1666	if (ignore_packed_keep_on_disk)
1667	flags \|= KEPT_PACK_ON_DISK;
1668	if (ignore_packed_keep_in_core)
1669	flags \|= KEPT_PACK_IN_CORE;
1670	if (ignore_packed_keep_in_core_open)
1671	flags \|= KEPT_PACK_IN_CORE_OPEN;
1672
1673	/*
1674	* If the object is in a pack that we want to ignore, and we
1675	* don't have any cruft packs that are being retained, we can
1676	* abort quickly.
1677	*/
1678	if (!ignore_packed_keep_in_core_has_cruft) {
1679	if (ignore_packed_keep_on_disk && p->pack_keep)
1680	return 0;
1681	if (ignore_packed_keep_in_core && p->pack_keep_in_core)
1682	return 0;
1683	if (ignore_packed_keep_in_core_open && p->pack_keep_in_core_open)
1684	return 0;
1685	if (has_object_kept_pack(p->repo, oid, flags))
1686	return 0;
1687	} else {
1688	/*
1689	* But if there is at least one cruft pack which
1690	* is being kept, we only want to include the
1691	* provided object if it has a strictly greater
1692	* mtime than any existing cruft copy.
1693	*/
1694	if (!want_cruft_object_mtime(p->repo, oid, flags,
1695	mtime))
1696	return 0;
1697	}
1698	}
1699
1700	/*
1701	* At this point we know definitively that either we don't care about
1702	* keep-packs, or the object is not in one. Keep checking other
1703	* conditions...
1704	*/
1705	if (!local \|\| !have_non_local_packs)
1706	return 1;
1707
1708	/* we don't know yet; keep looking for more packs */
1709	return -1;
1710	}
1711
1712	static int want_object_in_pack_one(struct packed_git *p,
1713	const struct object_id *oid,
1714	int exclude,
1715	struct packed_git **found_pack,
1716	off_t *found_offset,
1717	uint32_t found_mtime)
1718	{
1719	off_t offset;
1720
1721	if (p == *found_pack)
1722	offset = *found_offset;
1723	else
1724	offset = find_pack_entry_one(oid, p);
1725
1726	if (offset) {
1727	if (!*found_pack) {
1728	if (!is_pack_valid(p))
1729	return -1;
1730	*found_offset = offset;
1731	*found_pack = p;
1732	}
1733	return want_found_object(oid, exclude, p, found_mtime);
1734	}
1735	return -1;
1736	}
1737
1738	/*
1739	* Check whether we want the object in the pack (e.g., we do not want
1740	* objects found in non-local stores if the "--local" option was used).
1741	*
1742	* If the caller already knows an existing pack it wants to take the object
1743	* from, that is passed in found_pack and found_offset; otherwise this
1744	* function finds if there is any pack that has the object and returns the pack
1745	* and its offset in these variables.
1746	*/
1747	static int want_object_in_pack_mtime(const struct object_id *oid,
1748	int exclude,
1749	struct packed_git **found_pack,
1750	off_t *found_offset,
1751	uint32_t found_mtime)
1752	{
1753	int want;
1754	struct packfile_list_entry *e;
1755	struct odb_source *source;
1756
1757	if (!exclude && local) {
1758	/*
1759	* Note that we start iterating at `sources->next` so that we
1760	* skip the local object source.
1761	*/
1762	struct odb_source *source = the_repository->objects->sources->next;
1763	for (; source; source = source->next) {
1764	struct odb_source_files *files = odb_source_files_downcast(source);
1765	if (!odb_source_read_object_info(&files->loose->base, oid, NULL, 0))
1766	return 0;
1767	}
1768	}
1769
1770	/*
1771	* If we already know the pack object lives in, start checks from that
1772	* pack - in the usual case when neither --local was given nor .keep files
1773	* are present we will determine the answer right now.
1774	*/
1775	if (*found_pack) {
1776	want = want_found_object(oid, exclude, *found_pack,
1777	found_mtime);
1778	if (want != -1)
1779	return want;
1780
1781	*found_pack = NULL;
1782	*found_offset = 0;
1783	}
1784
1785	odb_prepare_alternates(the_repository->objects);
1786
1787	for (source = the_repository->objects->sources; source; source = source->next) {
1788	struct odb_source_files *files = odb_source_files_downcast(source);
1789	struct multi_pack_index *m = get_multi_pack_index(files->packed);
1790	struct pack_entry e;
1791
1792	if (m && fill_midx_entry(m, oid, &e)) {
1793	want = want_object_in_pack_one(e.p, oid, exclude, found_pack, found_offset, found_mtime);
1794	if (want != -1)
1795	return want;
1796	}
1797	}
1798
1799	for (source = the_repository->objects->sources; source; source = source->next) {
1800	struct odb_source_files *files = odb_source_files_downcast(source);
1801
1802	for (e = files->packed->packs.head; e; e = e->next) {
1803	struct packed_git *p = e->pack;
1804	want = want_object_in_pack_one(p, oid, exclude, found_pack, found_offset, found_mtime);
1805	if (!exclude && want > 0)
1806	packfile_list_prepend(&files->packed->packs, p);
1807	if (want != -1)
1808	return want;
1809	}
1810	}
1811
1812	if (uri_protocols.nr) {
1813	struct configured_exclusion *ex =
1814	oidmap_get(&configured_exclusions, oid);
1815	int i;
1816	const char *p;
1817
1818	if (ex) {
1819	for (i = 0; i < uri_protocols.nr; i++) {
1820	if (skip_prefix(ex->uri,
1821	uri_protocols.items[i].string,
1822	&p) &&
1823	*p == ':') {
1824	oidset_insert(&excluded_by_config, oid);
1825	return 0;
1826	}
1827	}
1828	}
1829	}
1830
1831	return 1;
1832	}
1833
1834	static inline int want_object_in_pack(const struct object_id *oid,
1835	int exclude,
1836	struct packed_git **found_pack,
1837	off_t *found_offset)
1838	{
1839	return want_object_in_pack_mtime(oid, exclude, found_pack, found_offset,
1840	0);
1841	}
1842
1843	static struct object_entry create_object_entry(const struct object_id oid,
1844	enum object_type type,
1845	uint32_t hash,
1846	int exclude,
1847	int no_try_delta,
1848	struct packed_git *found_pack,
1849	off_t found_offset)
1850	{
1851	struct object_entry *entry;
1852
1853	entry = packlist_alloc(&to_pack, oid);
1854	entry->hash = hash;
1855	oe_set_type(entry, type);
1856	if (exclude)
1857	entry->preferred_base = 1;
1858	else
1859	nr_result++;
1860	if (found_pack) {
1861	oe_set_in_pack(&to_pack, entry, found_pack);
1862	entry->in_pack_offset = found_offset;
1863	}
1864
1865	entry->no_try_delta = no_try_delta;
1866
1867	return entry;
1868	}
1869
1870	static const char no_closure_warning[] = N_(
1871	"disabling bitmap writing, as some objects are not being packed"
1872	);
1873
1874	static void add_object_entry(const struct object_id *oid, enum object_type type,
1875	const char *name, int exclude)
1876	{
1877	struct packed_git *found_pack = NULL;
1878	off_t found_offset = 0;
1879
1880	display_progress(progress_state, ++nr_seen);
1881
1882	if (have_duplicate_entry(oid, exclude))
1883	return;
1884
1885	if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1886	/* The pack is missing an object, so it will not have closure */
1887	if (write_bitmap_index) {
1888	if (write_bitmap_index != WRITE_BITMAP_QUIET)
1889	warning(_(no_closure_warning));
1890	write_bitmap_index = 0;
1891	}
1892	return;
1893	}
1894
1895	create_object_entry(oid, type, pack_name_hash_fn(name),
1896	exclude, name && no_try_delta(name),
1897	found_pack, found_offset);
1898	}
1899
1900	static int add_object_entry_from_bitmap(const struct object_id *oid,
1901	enum object_type type,
1902	int flags UNUSED, uint32_t name_hash,
1903	struct packed_git *pack, off_t offset,
1904	void *payload UNUSED)
1905	{
1906	display_progress(progress_state, ++nr_seen);
1907
1908	if (have_duplicate_entry(oid, 0))
1909	return 0;
1910
1911	if (!want_object_in_pack(oid, 0, &pack, &offset))
1912	return 0;
1913
1914	create_object_entry(oid, type, name_hash, 0, 0, pack, offset);
1915	return 0;
1916	}
1917
1918	struct pbase_tree_cache {
1919	struct object_id oid;
1920	int ref;
1921	int temporary;
1922	void *tree_data;
1923	unsigned long tree_size;
1924	};
1925
1926	static struct pbase_tree_cache *(pbase_tree_cache[256]);
1927	static int pbase_tree_cache_ix(const struct object_id *oid)
1928	{
1929	return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1930	}
1931	static int pbase_tree_cache_ix_incr(int ix)
1932	{
1933	return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1934	}
1935
1936	static struct pbase_tree {
1937	struct pbase_tree *next;
1938	/* This is a phony "cache" entry; we are not
1939	* going to evict it or find it through _get()
1940	* mechanism -- this is for the toplevel node that
1941	* would almost always change with any commit.
1942	*/
1943	struct pbase_tree_cache pcache;
1944	} *pbase_tree;
1945
1946	static struct pbase_tree_cache pbase_tree_get(const struct object_id oid)
1947	{
1948	struct pbase_tree_cache ent, nent;
1949	void *data;
1950	unsigned long size;
1951	size_t size_st = 0;
1952	enum object_type type;
1953	int neigh;
1954	int my_ix = pbase_tree_cache_ix(oid);
1955	int available_ix = -1;
1956
1957	/* pbase-tree-cache acts as a limited hashtable.
1958	* your object will be found at your index or within a few
1959	* slots after that slot if it is cached.
1960	*/
1961	for (neigh = 0; neigh < 8; neigh++) {
1962	ent = pbase_tree_cache[my_ix];
1963	if (ent && oideq(&ent->oid, oid)) {
1964	ent->ref++;
1965	return ent;
1966	}
1967	else if (((available_ix < 0) && (!ent \|\| !ent->ref)) \|\|
1968	((0 <= available_ix) &&
1969	(!ent && pbase_tree_cache[available_ix])))
1970	available_ix = my_ix;
1971	if (!ent)
1972	break;
1973	my_ix = pbase_tree_cache_ix_incr(my_ix);
1974	}
1975
1976	/* Did not find one. Either we got a bogus request or
1977	* we need to read and perhaps cache.
1978	*/
1979	data = odb_read_object(the_repository->objects, oid, &type, &size_st);
1980	size = cast_size_t_to_ulong(size_st);
1981	if (!data)
1982	return NULL;
1983	if (type != OBJ_TREE) {
1984	free(data);
1985	return NULL;
1986	}
1987
1988	/* We need to either cache or return a throwaway copy */
1989
1990	if (available_ix < 0)
1991	ent = NULL;
1992	else {
1993	ent = pbase_tree_cache[available_ix];
1994	my_ix = available_ix;
1995	}
1996
1997	if (!ent) {
1998	nent = xmalloc(sizeof(*nent));
1999	nent->temporary = (available_ix < 0);
2000	}
2001	else {
2002	/* evict and reuse */
2003	free(ent->tree_data);
2004	nent = ent;
2005	}
2006	oidcpy(&nent->oid, oid);
2007	nent->tree_data = data;
2008	nent->tree_size = size;
2009	nent->ref = 1;
2010	if (!nent->temporary)
2011	pbase_tree_cache[my_ix] = nent;
2012	return nent;
2013	}
2014
2015	static void pbase_tree_put(struct pbase_tree_cache *cache)
2016	{
2017	if (!cache->temporary) {
2018	cache->ref--;
2019	return;
2020	}
2021	free(cache->tree_data);
2022	free(cache);
2023	}
2024
2025	static size_t name_cmp_len(const char *name)
2026	{
2027	return strcspn(name, "\n/");
2028	}
2029
2030	static void add_pbase_object(struct tree_desc *tree,
2031	const char *name,
2032	size_t cmplen,
2033	const char *fullname)
2034	{
2035	struct name_entry entry;
2036	int cmp;
2037
2038	while (tree_entry(tree,&entry)) {
2039	if (S_ISGITLINK(entry.mode))
2040	continue;
2041	cmp = tree_entry_len(&entry) != cmplen ? 1 :
2042	memcmp(name, entry.path, cmplen);
2043	if (cmp > 0)
2044	continue;
2045	if (cmp < 0)
2046	return;
2047	if (name[cmplen] != '/') {
2048	add_object_entry(&entry.oid,
2049	object_type(entry.mode),
2050	fullname, 1);
2051	return;
2052	}
2053	if (S_ISDIR(entry.mode)) {
2054	struct tree_desc sub;
2055	struct pbase_tree_cache *tree;
2056	const char *down = name+cmplen+1;
2057	size_t downlen = name_cmp_len(down);
2058
2059	tree = pbase_tree_get(&entry.oid);
2060	if (!tree)
2061	return;
2062	init_tree_desc(&sub, &tree->oid,
2063	tree->tree_data, tree->tree_size);
2064
2065	add_pbase_object(&sub, down, downlen, fullname);
2066	pbase_tree_put(tree);
2067	}
2068	}
2069	}
2070
2071	static unsigned *done_pbase_paths;
2072	static int done_pbase_paths_num;
2073	static int done_pbase_paths_alloc;
2074	static int done_pbase_path_pos(unsigned hash)
2075	{
2076	int lo = 0;
2077	int hi = done_pbase_paths_num;
2078	while (lo < hi) {
2079	int mi = lo + (hi - lo) / 2;
2080	if (done_pbase_paths[mi] == hash)
2081	return mi;
2082	if (done_pbase_paths[mi] < hash)
2083	hi = mi;
2084	else
2085	lo = mi + 1;
2086	}
2087	return -lo-1;
2088	}
2089
2090	static int check_pbase_path(unsigned hash)
2091	{
2092	int pos = done_pbase_path_pos(hash);
2093	if (0 <= pos)
2094	return 1;
2095	pos = -pos - 1;
2096	ALLOC_GROW(done_pbase_paths,
2097	done_pbase_paths_num + 1,
2098	done_pbase_paths_alloc);
2099	done_pbase_paths_num++;
2100	if (pos < done_pbase_paths_num)
2101	MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
2102	done_pbase_paths_num - pos - 1);
2103	done_pbase_paths[pos] = hash;
2104	return 0;
2105	}
2106
2107	static void add_preferred_base_object(const char *name)
2108	{
2109	struct pbase_tree *it;
2110	size_t cmplen;
2111	unsigned hash = pack_name_hash_fn(name);
2112
2113	if (!num_preferred_base \|\| check_pbase_path(hash))
2114	return;
2115
2116	cmplen = name_cmp_len(name);
2117	for (it = pbase_tree; it; it = it->next) {
2118	if (cmplen == 0) {
2119	add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
2120	}
2121	else {
2122	struct tree_desc tree;
2123	init_tree_desc(&tree, &it->pcache.oid,
2124	it->pcache.tree_data, it->pcache.tree_size);
2125	add_pbase_object(&tree, name, cmplen, name);
2126	}
2127	}
2128	}
2129
2130	static void add_preferred_base(struct object_id *oid)
2131	{
2132	struct pbase_tree *it;
2133	void *data;
2134	unsigned long size;
2135	size_t size_st = 0;
2136	struct object_id tree_oid;
2137
2138	if (window <= num_preferred_base++)
2139	return;
2140
2141	data = odb_read_object_peeled(the_repository->objects, oid,
2142	OBJ_TREE, &size_st, &tree_oid);
2143	size = cast_size_t_to_ulong(size_st);
2144	if (!data)
2145	return;
2146
2147	for (it = pbase_tree; it; it = it->next) {
2148	if (oideq(&it->pcache.oid, &tree_oid)) {
2149	free(data);
2150	return;
2151	}
2152	}
2153
2154	CALLOC_ARRAY(it, 1);
2155	it->next = pbase_tree;
2156	pbase_tree = it;
2157
2158	oidcpy(&it->pcache.oid, &tree_oid);
2159	it->pcache.tree_data = data;
2160	it->pcache.tree_size = size;
2161	}
2162
2163	static void cleanup_preferred_base(void)
2164	{
2165	struct pbase_tree *it;
2166	unsigned i;
2167
2168	it = pbase_tree;
2169	pbase_tree = NULL;
2170	while (it) {
2171	struct pbase_tree *tmp = it;
2172	it = tmp->next;
2173	free(tmp->pcache.tree_data);
2174	free(tmp);
2175	}
2176
2177	for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
2178	if (!pbase_tree_cache[i])
2179	continue;
2180	free(pbase_tree_cache[i]->tree_data);
2181	FREE_AND_NULL(pbase_tree_cache[i]);
2182	}
2183
2184	FREE_AND_NULL(done_pbase_paths);
2185	done_pbase_paths_num = done_pbase_paths_alloc = 0;
2186	}
2187
2188	/*
2189	* Return 1 iff the object specified by "delta" can be sent
2190	* literally as a delta against the base in "base_sha1". If
2191	* so, then *base_out will point to the entry in our packing
2192	* list, or NULL if we must use the external-base list.
2193	*
2194	* Depth value does not matter - find_deltas() will
2195	* never consider reused delta as the base object to
2196	* deltify other objects against, in order to avoid
2197	* circular deltas.
2198	*/
2199	static int can_reuse_delta(const struct object_id *base_oid,
2200	struct object_entry *delta,
2201	struct object_entry **base_out)
2202	{
2203	struct object_entry *base;
2204
2205	/*
2206	* First see if we're already sending the base (or it's explicitly in
2207	* our "excluded" list).
2208	*/
2209	base = packlist_find(&to_pack, base_oid);
2210	if (base) {
2211	/*
2212	* A preferred base is omitted from the resulting pack, so it
2213	* can only be referenced by object ID.
2214	*/
2215	if (base->preferred_base && !allow_ref_delta)
2216	return 0;
2217
2218	if (!in_same_island(&delta->idx.oid, &base->idx.oid))
2219	return 0;
2220	*base_out = base;
2221	return 1;
2222	}
2223
2224	/*
2225	* Otherwise, reachability bitmaps may tell us if the receiver has it,
2226	* even if it was buried too deep in history to make it into the
2227	* packing list.
2228	*/
2229	if (allow_ref_delta && thin &&
2230	bitmap_has_oid_in_uninteresting(bitmap_git, base_oid)) {
2231	if (use_delta_islands) {
2232	if (!in_same_island(&delta->idx.oid, base_oid))
2233	return 0;
2234	}
2235	*base_out = NULL;
2236	return 1;
2237	}
2238
2239	return 0;
2240	}
2241
2242	static void prefetch_to_pack(uint32_t object_index_start) {
2243	struct oid_array to_fetch = OID_ARRAY_INIT;
2244	uint32_t i;
2245
2246	for (i = object_index_start; i < to_pack.nr_objects; i++) {
2247	struct object_entry *entry = to_pack.objects + i;
2248
2249	if (!odb_read_object_info_extended(the_repository->objects,
2250	&entry->idx.oid,
2251	NULL,
2252	OBJECT_INFO_FOR_PREFETCH))
2253	continue;
2254	oid_array_append(&to_fetch, &entry->idx.oid);
2255	}
2256	promisor_remote_get_direct(the_repository,
2257	to_fetch.oid, to_fetch.nr);
2258	oid_array_clear(&to_fetch);
2259	}
2260
2261	static void check_object(struct object_entry *entry, uint32_t object_index)
2262	{
2263	size_t canonical_size;
2264	enum object_type type;
2265	struct object_info oi = {.typep = &type, .sizep = &canonical_size};
2266
2267	if (IN_PACK(entry)) {
2268	struct packed_git *p = IN_PACK(entry);
2269	struct pack_window *w_curs = NULL;
2270	int have_base = 0;
2271	struct object_id base_ref;
2272	struct object_entry *base_entry;
2273	unsigned long used, used_0;
2274	size_t avail;
2275	off_t ofs;
2276	unsigned char *buf, c;
2277	enum object_type type;
2278	size_t in_pack_size;
2279
2280	buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
2281
2282	/*
2283	* We want in_pack_type even if we do not reuse delta
2284	* since non-delta representations could still be reused.
2285	*/
2286	used = unpack_object_header_buffer(buf, avail,
2287	&type,
2288	&in_pack_size);
2289	if (used == 0)
2290	goto give_up;
2291
2292	if (type < 0)
2293	BUG("invalid type %d", type);
2294	entry->in_pack_type = type;
2295
2296	/*
2297	* Determine if this is a delta and if so whether we can
2298	* reuse it or not. Otherwise let's find out as cheaply as
2299	* possible what the actual type and size for this object is.
2300	*/
2301	switch (entry->in_pack_type) {
2302	default:
2303	/* Not a delta hence we've already got all we need. */
2304	oe_set_type(entry, entry->in_pack_type);
2305	SET_SIZE(entry, in_pack_size);
2306	entry->in_pack_header_size = used;
2307	if (oe_type(entry) < OBJ_COMMIT \|\| oe_type(entry) > OBJ_BLOB)
2308	goto give_up;
2309	unuse_pack(&w_curs);
2310	return;
2311	case OBJ_REF_DELTA:
2312	if (reuse_delta && !entry->preferred_base) {
2313	oidread(&base_ref,
2314	use_pack(p, &w_curs,
2315	entry->in_pack_offset + used,
2316	NULL),
2317	the_repository->hash_algo);
2318	have_base = 1;
2319	}
2320	entry->in_pack_header_size = used + the_hash_algo->rawsz;
2321	break;
2322	case OBJ_OFS_DELTA:
2323	buf = use_pack(p, &w_curs,
2324	entry->in_pack_offset + used, NULL);
2325	used_0 = 0;
2326	c = buf[used_0++];
2327	ofs = c & 127;
2328	while (c & 128) {
2329	ofs += 1;
2330	if (!ofs \|\| MSB(ofs, 7)) {
2331	error(_("delta base offset overflow in pack for %s"),
2332	oid_to_hex(&entry->idx.oid));
2333	goto give_up;
2334	}
2335	c = buf[used_0++];
2336	ofs = (ofs << 7) + (c & 127);
2337	}
2338	ofs = entry->in_pack_offset - ofs;
2339	if (ofs <= 0 \|\| ofs >= entry->in_pack_offset) {
2340	error(_("delta base offset out of bound for %s"),
2341	oid_to_hex(&entry->idx.oid));
2342	goto give_up;
2343	}
2344	if (reuse_delta && !entry->preferred_base) {
2345	uint32_t pos;
2346	if (offset_to_pack_pos(p, ofs, &pos) < 0)
2347	goto give_up;
2348	if (!nth_packed_object_id(&base_ref, p,
2349	pack_pos_to_index(p, pos)))
2350	have_base = 1;
2351	}
2352	entry->in_pack_header_size = used + used_0;
2353	break;
2354	}
2355
2356	if (have_base &&
2357	can_reuse_delta(&base_ref, entry, &base_entry)) {
2358	oe_set_type(entry, entry->in_pack_type);
2359	SET_SIZE(entry, in_pack_size); /* delta size */
2360	SET_DELTA_SIZE(entry, in_pack_size);
2361
2362	if (base_entry) {
2363	SET_DELTA(entry, base_entry);
2364	entry->delta_sibling_idx = base_entry->delta_child_idx;
2365	SET_DELTA_CHILD(base_entry, entry);
2366	} else {
2367	SET_DELTA_EXT(entry, &base_ref);
2368	}
2369
2370	unuse_pack(&w_curs);
2371	return;
2372	}
2373
2374	if (oe_type(entry)) {
2375	off_t delta_pos;
2376
2377	/*
2378	* This must be a delta and we already know what the
2379	* final object type is. Let's extract the actual
2380	* object size from the delta header.
2381	*/
2382	delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
2383	canonical_size = get_size_from_delta(p, &w_curs,
2384	delta_pos);
2385	if (canonical_size == 0)
2386	goto give_up;
2387	SET_SIZE(entry, canonical_size);
2388	unuse_pack(&w_curs);
2389	return;
2390	}
2391
2392	/*
2393	* No choice but to fall back to the recursive delta walk
2394	* with odb_read_object_info() to find about the object type
2395	* at this point...
2396	*/
2397	give_up:
2398	unuse_pack(&w_curs);
2399	}
2400
2401	if (odb_read_object_info_extended(the_repository->objects, &entry->idx.oid, &oi,
2402	OBJECT_INFO_SKIP_FETCH_OBJECT \| OBJECT_INFO_LOOKUP_REPLACE) < 0) {
2403	if (repo_has_promisor_remote(the_repository)) {
2404	prefetch_to_pack(object_index);
2405	if (odb_read_object_info_extended(the_repository->objects, &entry->idx.oid, &oi,
2406	OBJECT_INFO_SKIP_FETCH_OBJECT \| OBJECT_INFO_LOOKUP_REPLACE) < 0)
2407	type = -1;
2408	} else {
2409	type = -1;
2410	}
2411	}
2412	oe_set_type(entry, type);
2413	if (entry->type_valid) {
2414	SET_SIZE(entry, canonical_size);
2415	} else {
2416	/*
2417	* Bad object type is checked in prepare_pack(). This is
2418	* to permit a missing preferred base object to be ignored
2419	* as a preferred base. Doing so can result in a larger
2420	* pack file, but the transfer will still take place.
2421	*/
2422	}
2423	}
2424
2425	static int pack_offset_sort(const void _a, const void _b)
2426	{
2427	const struct object_entry a = (struct object_entry **)_a;
2428	const struct object_entry b = (struct object_entry **)_b;
2429	const struct packed_git *a_in_pack = IN_PACK(a);
2430	const struct packed_git *b_in_pack = IN_PACK(b);
2431
2432	/* avoid filesystem trashing with loose objects */
2433	if (!a_in_pack && !b_in_pack)
2434	return oidcmp(&a->idx.oid, &b->idx.oid);
2435
2436	if (a_in_pack < b_in_pack)
2437	return -1;
2438	if (a_in_pack > b_in_pack)
2439	return 1;
2440	return a->in_pack_offset < b->in_pack_offset ? -1 :
2441	(a->in_pack_offset > b->in_pack_offset);
2442	}
2443
2444	/*
2445	* Drop an on-disk delta we were planning to reuse. Naively, this would
2446	* just involve blanking out the "delta" field, but we have to deal
2447	* with some extra book-keeping:
2448	*
2449	* 1. Removing ourselves from the delta_sibling linked list.
2450	*
2451	* 2. Updating our size/type to the non-delta representation. These were
2452	* either not recorded initially (size) or overwritten with the delta type
2453	* (type) when check_object() decided to reuse the delta.
2454	*
2455	* 3. Resetting our delta depth, as we are now a base object.
2456	*/
2457	static void drop_reused_delta(struct object_entry *entry)
2458	{
2459	unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
2460	struct object_info oi = OBJECT_INFO_INIT;
2461	enum object_type type;
2462	size_t size;
2463
2464	while (*idx) {
2465	struct object_entry oe = &to_pack.objects[idx - 1];
2466
2467	if (oe == entry)
2468	*idx = oe->delta_sibling_idx;
2469	else
2470	idx = &oe->delta_sibling_idx;
2471	}
2472	SET_DELTA(entry, NULL);
2473	entry->depth = 0;
2474
2475	oi.sizep = &size;
2476	oi.typep = &type;
2477	if (packed_object_info(NULL, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
2478	/*
2479	* We failed to get the info from this pack for some reason;
2480	* fall back to odb_read_object_info, which may find another copy.
2481	* And if that fails, the error will be recorded in oe_type(entry)
2482	* and dealt with in prepare_pack().
2483	*/
2484	oe_set_type(entry,
2485	odb_read_object_info(the_repository->objects,
2486	&entry->idx.oid, &size));
2487	} else {
2488	oe_set_type(entry, type);
2489	}
2490	SET_SIZE(entry, size);
2491	}
2492
2493	/*
2494	* Follow the chain of deltas from this entry onward, throwing away any links
2495	* that cause us to hit a cycle (as determined by the DFS state flags in
2496	* the entries).
2497	*
2498	* We also detect too-long reused chains that would violate our --depth
2499	* limit.
2500	*/
2501	static void break_delta_chains(struct object_entry *entry)
2502	{
2503	/*
2504	* The actual depth of each object we will write is stored as an int,
2505	* as it cannot exceed our int "depth" limit. But before we break
2506	* changes based no that limit, we may potentially go as deep as the
2507	* number of objects, which is elsewhere bounded to a uint32_t.
2508	*/
2509	uint32_t total_depth;
2510	struct object_entry cur, next;
2511
2512	for (cur = entry, total_depth = 0;
2513	cur;
2514	cur = DELTA(cur), total_depth++) {
2515	if (cur->dfs_state == DFS_DONE) {
2516	/*
2517	* We've already seen this object and know it isn't
2518	* part of a cycle. We do need to append its depth
2519	* to our count.
2520	*/
2521	total_depth += cur->depth;
2522	break;
2523	}
2524
2525	/*
2526	* We break cycles before looping, so an ACTIVE state (or any
2527	* other cruft which made its way into the state variable)
2528	* is a bug.
2529	*/
2530	if (cur->dfs_state != DFS_NONE)
2531	BUG("confusing delta dfs state in first pass: %d",
2532	cur->dfs_state);
2533
2534	/*
2535	* Now we know this is the first time we've seen the object. If
2536	* it's not a delta, we're done traversing, but we'll mark it
2537	* done to save time on future traversals.
2538	*/
2539	if (!DELTA(cur)) {
2540	cur->dfs_state = DFS_DONE;
2541	break;
2542	}
2543
2544	/*
2545	* Mark ourselves as active and see if the next step causes
2546	* us to cycle to another active object. It's important to do
2547	* this _before_ we loop, because it impacts where we make the
2548	* cut, and thus how our total_depth counter works.
2549	* E.g., We may see a partial loop like:
2550	*
2551	* A -> B -> C -> D -> B
2552	*
2553	* Cutting B->C breaks the cycle. But now the depth of A is
2554	* only 1, and our total_depth counter is at 3. The size of the
2555	* error is always one less than the size of the cycle we
2556	* broke. Commits C and D were "lost" from A's chain.
2557	*
2558	* If we instead cut D->B, then the depth of A is correct at 3.
2559	* We keep all commits in the chain that we examined.
2560	*/
2561	cur->dfs_state = DFS_ACTIVE;
2562	if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
2563	drop_reused_delta(cur);
2564	cur->dfs_state = DFS_DONE;
2565	break;
2566	}
2567	}
2568
2569	/*
2570	* And now that we've gone all the way to the bottom of the chain, we
2571	* need to clear the active flags and set the depth fields as
2572	* appropriate. Unlike the loop above, which can quit when it drops a
2573	* delta, we need to keep going to look for more depth cuts. So we need
2574	* an extra "next" pointer to keep going after we reset cur->delta.
2575	*/
2576	for (cur = entry; cur; cur = next) {
2577	next = DELTA(cur);
2578
2579	/*
2580	* We should have a chain of zero or more ACTIVE states down to
2581	* a final DONE. We can quit after the DONE, because either it
2582	* has no bases, or we've already handled them in a previous
2583	* call.
2584	*/
2585	if (cur->dfs_state == DFS_DONE)
2586	break;
2587	else if (cur->dfs_state != DFS_ACTIVE)
2588	BUG("confusing delta dfs state in second pass: %d",
2589	cur->dfs_state);
2590
2591	/*
2592	* If the total_depth is more than depth, then we need to snip
2593	* the chain into two or more smaller chains that don't exceed
2594	* the maximum depth. Most of the resulting chains will contain
2595	* (depth + 1) entries (i.e., depth deltas plus one base), and
2596	* the last chain (i.e., the one containing entry) will contain
2597	* whatever entries are left over, namely
2598	* (total_depth % (depth + 1)) of them.
2599	*
2600	* Since we are iterating towards decreasing depth, we need to
2601	* decrement total_depth as we go, and we need to write to the
2602	* entry what its final depth will be after all of the
2603	* snipping. Since we're snipping into chains of length (depth
2604	* + 1) entries, the final depth of an entry will be its
2605	* original depth modulo (depth + 1). Any time we encounter an
2606	* entry whose final depth is supposed to be zero, we snip it
2607	* from its delta base, thereby making it so.
2608	*/
2609	cur->depth = (total_depth--) % (depth + 1);
2610	if (!cur->depth)
2611	drop_reused_delta(cur);
2612
2613	cur->dfs_state = DFS_DONE;
2614	}
2615	}
2616
2617	static void get_object_details(void)
2618	{
2619	uint32_t i;
2620	struct object_entry **sorted_by_offset;
2621
2622	if (progress)
2623	progress_state = start_progress(the_repository,
2624	_("Counting objects"),
2625	to_pack.nr_objects);
2626
2627	CALLOC_ARRAY(sorted_by_offset, to_pack.nr_objects);
2628	for (i = 0; i < to_pack.nr_objects; i++)
2629	sorted_by_offset[i] = to_pack.objects + i;
2630	QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
2631
2632	for (i = 0; i < to_pack.nr_objects; i++) {
2633	struct object_entry *entry = sorted_by_offset[i];
2634	check_object(entry, i);
2635	if (entry->type_valid &&
2636	oe_size_greater_than(&to_pack, entry,
2637	repo_settings_get_big_file_threshold(the_repository)))
2638	entry->no_try_delta = 1;
2639	display_progress(progress_state, i + 1);
2640	}
2641	stop_progress(&progress_state);
2642
2643	/*
2644	* This must happen in a second pass, since we rely on the delta
2645	* information for the whole list being completed.
2646	*/
2647	for (i = 0; i < to_pack.nr_objects; i++)
2648	break_delta_chains(&to_pack.objects[i]);
2649
2650	free(sorted_by_offset);
2651	}
2652
2653	/*
2654	* We search for deltas in a list sorted by type, by filename hash, and then
2655	* by size, so that we see progressively smaller and smaller files.
2656	* That's because we prefer deltas to be from the bigger file
2657	* to the smaller -- deletes are potentially cheaper, but perhaps
2658	* more importantly, the bigger file is likely the more recent
2659	* one. The deepest deltas are therefore the oldest objects which are
2660	* less susceptible to be accessed often.
2661	*/
2662	static int type_size_sort(const void _a, const void _b)
2663	{
2664	const struct object_entry a = (struct object_entry **)_a;
2665	const struct object_entry b = (struct object_entry **)_b;
2666	const enum object_type a_type = oe_type(a);
2667	const enum object_type b_type = oe_type(b);
2668	const unsigned long a_size = SIZE(a);
2669	const unsigned long b_size = SIZE(b);
2670
2671	if (a_type > b_type)
2672	return -1;
2673	if (a_type < b_type)
2674	return 1;
2675	if (a->hash > b->hash)
2676	return -1;
2677	if (a->hash < b->hash)
2678	return 1;
2679	if (a->preferred_base > b->preferred_base)
2680	return -1;
2681	if (a->preferred_base < b->preferred_base)
2682	return 1;
2683	if (use_delta_islands) {
2684	const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);
2685	if (island_cmp)
2686	return island_cmp;
2687	}
2688	if (a_size > b_size)
2689	return -1;
2690	if (a_size < b_size)
2691	return 1;
2692	return a < b ? -1 : (a > b); /* newest first */
2693	}
2694
2695	struct unpacked {
2696	struct object_entry *entry;
2697	void *data;
2698	struct delta_index *index;
2699	unsigned depth;
2700	};
2701
2702	static int delta_cacheable(size_t src_size, size_t trg_size,
2703	size_t delta_size)
2704	{
2705	if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
2706	return 0;
2707
2708	if (delta_size < cache_max_small_delta_size)
2709	return 1;
2710
2711	/* cache delta, if objects are large enough compared to delta size */
2712	if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
2713	return 1;
2714
2715	return 0;
2716	}
2717
2718	/* Protect delta_cache_size */
2719	static pthread_mutex_t cache_mutex;
2720	#define cache_lock() pthread_mutex_lock(&cache_mutex)
2721	#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
2722
2723	/*
2724	* Protect object list partitioning (e.g. struct thread_param) and
2725	* progress_state
2726	*/
2727	static pthread_mutex_t progress_mutex;
2728	#define progress_lock() pthread_mutex_lock(&progress_mutex)
2729	#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
2730
2731	/*
2732	* Access to struct object_entry is unprotected since each thread owns
2733	* a portion of the main object list. Just don't access object entries
2734	* ahead in the list because they can be stolen and would need
2735	* progress_mutex for protection.
2736	*/
2737
2738	static inline int oe_size_less_than(struct packing_data *pack,
2739	const struct object_entry *lhs,
2740	size_t rhs)
2741	{
2742	if (lhs->size_valid)
2743	return lhs->size_ < rhs;
2744	if (rhs < pack->oe_size_limit) /* rhs < 2^x <= lhs ? */
2745	return 0;
2746	return oe_get_size_slow(pack, lhs) < rhs;
2747	}
2748
2749	static inline void oe_set_tree_depth(struct packing_data *pack,
2750	struct object_entry *e,
2751	unsigned int tree_depth)
2752	{
2753	if (!pack->tree_depth)
2754	CALLOC_ARRAY(pack->tree_depth, pack->nr_alloc);
2755	pack->tree_depth[e - pack->objects] = tree_depth;
2756	}
2757
2758	static void record_tree_depth(const struct object_id oid, const char name)
2759	{
2760	const char *p;
2761	unsigned depth;
2762	struct object_entry *ent;
2763
2764	/* the empty string is a root tree, which is depth 0 */
2765	depth = *name ? 1 : 0;
2766	for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))
2767	depth++;
2768
2769	ent = packlist_find(&to_pack, oid);
2770	if (ent && depth > oe_tree_depth(&to_pack, ent))
2771	oe_set_tree_depth(&to_pack, ent, depth);
2772	}
2773
2774	/*
2775	* Return the size of the object without doing any delta
2776	* reconstruction (so non-deltas are true object sizes, but deltas
2777	* return the size of the delta data).
2778	*/
2779	size_t oe_get_size_slow(struct packing_data *pack,
2780	const struct object_entry *e)
2781	{
2782	struct packed_git *p;
2783	struct pack_window *w_curs;
2784	unsigned char *buf;
2785	enum object_type type;
2786	unsigned long used;
2787	size_t avail, size;
2788
2789	if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
2790	size_t sz;
2791	packing_data_lock(&to_pack);
2792	if (odb_read_object_info(the_repository->objects,
2793	&e->idx.oid, &sz) < 0)
2794	die(_("unable to get size of %s"),
2795	oid_to_hex(&e->idx.oid));
2796	packing_data_unlock(&to_pack);
2797	return sz;
2798	}
2799
2800	p = oe_in_pack(pack, e);
2801	if (!p)
2802	BUG("when e->type is a delta, it must belong to a pack");
2803
2804	packing_data_lock(&to_pack);
2805	w_curs = NULL;
2806	buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
2807	used = unpack_object_header_buffer(buf, avail, &type, &size);
2808	if (used == 0)
2809	die(_("unable to parse object header of %s"),
2810	oid_to_hex(&e->idx.oid));
2811
2812	unuse_pack(&w_curs);
2813	packing_data_unlock(&to_pack);
2814	return size;
2815	}
2816
2817	static int try_delta(struct unpacked trg, struct unpacked src,
2818	unsigned max_depth, size_t *mem_usage)
2819	{
2820	struct object_entry *trg_entry = trg->entry;
2821	struct object_entry *src_entry = src->entry;
2822	unsigned long trg_size, src_size, sizediff, max_size, sz;
2823	size_t delta_size;
2824	unsigned ref_depth;
2825	enum object_type type;
2826	void *delta_buf;
2827
2828	/* Don't bother doing diffs between different types */
2829	if (oe_type(trg_entry) != oe_type(src_entry))
2830	return -1;
2831
2832	/*
2833	* We do not bother to try a delta that we discarded on an
2834	* earlier try, but only when reusing delta data. Note that
2835	* src_entry that is marked as the preferred_base should always
2836	* be considered, as even if we produce a suboptimal delta against
2837	* it, we will still save the transfer cost, as we already know
2838	* the other side has it and we won't send src_entry at all.
2839	*/
2840	if (reuse_delta && IN_PACK(trg_entry) &&
2841	IN_PACK(trg_entry) == IN_PACK(src_entry) &&
2842	!src_entry->preferred_base &&
2843	trg_entry->in_pack_type != OBJ_REF_DELTA &&
2844	trg_entry->in_pack_type != OBJ_OFS_DELTA)
2845	return 0;
2846
2847	/* Let's not bust the allowed depth. */
2848	if (src->depth >= max_depth)
2849	return 0;
2850
2851	/* Now some size filtering heuristics. */
2852	trg_size = SIZE(trg_entry);
2853	if (!DELTA(trg_entry)) {
2854	max_size = trg_size/2 - the_hash_algo->rawsz;
2855	ref_depth = 1;
2856	} else {
2857	max_size = DELTA_SIZE(trg_entry);
2858	ref_depth = trg->depth;
2859	}
2860	max_size = (uint64_t)max_size * (max_depth - src->depth) /
2861	(max_depth - ref_depth + 1);
2862	if (max_size == 0)
2863	return 0;
2864	src_size = SIZE(src_entry);
2865	sizediff = src_size < trg_size ? trg_size - src_size : 0;
2866	if (sizediff >= max_size)
2867	return 0;
2868	if (trg_size < src_size / 32)
2869	return 0;
2870
2871	if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))
2872	return 0;
2873
2874	/* Load data if not already done */
2875	if (!trg->data) {
2876	size_t sz_st = 0;
2877	packing_data_lock(&to_pack);
2878	trg->data = odb_read_object(the_repository->objects,
2879	&trg_entry->idx.oid, &type,
2880	&sz_st);
2881	sz = cast_size_t_to_ulong(sz_st);
2882	packing_data_unlock(&to_pack);
2883	if (!trg->data)
2884	die(_("object %s cannot be read"),
2885	oid_to_hex(&trg_entry->idx.oid));
2886	if (sz != trg_size)
2887	die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2888	oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,
2889	(uintmax_t)trg_size);
2890	*mem_usage += sz;
2891	}
2892	if (!src->data) {
2893	size_t sz_st = 0;
2894	packing_data_lock(&to_pack);
2895	src->data = odb_read_object(the_repository->objects,
2896	&src_entry->idx.oid, &type,
2897	&sz_st);
2898	sz = cast_size_t_to_ulong(sz_st);
2899	packing_data_unlock(&to_pack);
2900	if (!src->data) {
2901	if (src_entry->preferred_base) {
2902	static int warned = 0;
2903	if (!warned++)
2904	warning(_("object %s cannot be read"),
2905	oid_to_hex(&src_entry->idx.oid));
2906	/*
2907	* Those objects are not included in the
2908	* resulting pack. Be resilient and ignore
2909	* them if they can't be read, in case the
2910	* pack could be created nevertheless.
2911	*/
2912	return 0;
2913	}
2914	die(_("object %s cannot be read"),
2915	oid_to_hex(&src_entry->idx.oid));
2916	}
2917	if (sz != src_size)
2918	die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2919	oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,
2920	(uintmax_t)src_size);
2921	*mem_usage += sz;
2922	}
2923	if (!src->index) {
2924	src->index = create_delta_index(src->data, src_size);
2925	if (!src->index) {
2926	static int warned = 0;
2927	if (!warned++)
2928	warning(_("suboptimal pack - out of memory"));
2929	return 0;
2930	}
2931	*mem_usage += sizeof_delta_index(src->index);
2932	}
2933
2934	delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
2935	if (!delta_buf)
2936	return 0;
2937
2938	if (DELTA(trg_entry)) {
2939	/* Prefer only shallower same-sized deltas. */
2940	if (delta_size == DELTA_SIZE(trg_entry) &&
2941	src->depth + 1 >= trg->depth) {
2942	free(delta_buf);
2943	return 0;
2944	}
2945	}
2946
2947	/*
2948	* Handle memory allocation outside of the cache
2949	* accounting lock. Compiler will optimize the strangeness
2950	* away when NO_PTHREADS is defined.
2951	*/
2952	free(trg_entry->delta_data);
2953	cache_lock();
2954	if (trg_entry->delta_data) {
2955	delta_cache_size -= DELTA_SIZE(trg_entry);
2956	trg_entry->delta_data = NULL;
2957	}
2958	if (delta_cacheable(src_size, trg_size, delta_size)) {
2959	delta_cache_size += delta_size;
2960	cache_unlock();
2961	trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2962	} else {
2963	cache_unlock();
2964	free(delta_buf);
2965	}
2966
2967	SET_DELTA(trg_entry, src_entry);
2968	SET_DELTA_SIZE(trg_entry, delta_size);
2969	trg->depth = src->depth + 1;
2970
2971	return 1;
2972	}
2973
2974	static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
2975	{
2976	struct object_entry *child = DELTA_CHILD(me);
2977	unsigned int m = n;
2978	while (child) {
2979	const unsigned int c = check_delta_limit(child, n + 1);
2980	if (m < c)
2981	m = c;
2982	child = DELTA_SIBLING(child);
2983	}
2984	return m;
2985	}
2986
2987	static size_t free_unpacked(struct unpacked *n)
2988	{
2989	size_t freed_mem = sizeof_delta_index(n->index);
2990	free_delta_index(n->index);
2991	n->index = NULL;
2992	if (n->data) {
2993	freed_mem += SIZE(n->entry);
2994	FREE_AND_NULL(n->data);
2995	}
2996	n->entry = NULL;
2997	n->depth = 0;
2998	return freed_mem;
2999	}
3000
3001	static void find_deltas(struct object_entry *list, unsigned list_size,
3002	int window, int depth, unsigned *processed)
3003	{
3004	uint32_t i, idx = 0, count = 0;
3005	struct unpacked *array;
3006	size_t mem_usage = 0;
3007
3008	CALLOC_ARRAY(array, window);
3009
3010	for (;;) {
3011	struct object_entry *entry;
3012	struct unpacked *n = array + idx;
3013	int j, max_depth, best_base = -1;
3014
3015	progress_lock();
3016	if (!*list_size) {
3017	progress_unlock();
3018	break;
3019	}
3020	entry = *list++;
3021	(*list_size)--;
3022	if (!entry->preferred_base) {
3023	(*processed)++;
3024	display_progress(progress_state, *processed);
3025	}
3026	progress_unlock();
3027
3028	mem_usage -= free_unpacked(n);
3029	n->entry = entry;
3030
3031	while (window_memory_limit &&
3032	mem_usage > window_memory_limit &&
3033	count > 1) {
3034	const uint32_t tail = (idx + window - count) % window;
3035	mem_usage -= free_unpacked(array + tail);
3036	count--;
3037	}
3038
3039	/* We do not compute delta to create objects we are not
3040	* going to pack.
3041	*/
3042	if (entry->preferred_base)
3043	goto next;
3044
3045	/*
3046	* If the current object is at pack edge, take the depth the
3047	* objects that depend on the current object into account
3048	* otherwise they would become too deep.
3049	*/
3050	max_depth = depth;
3051	if (DELTA_CHILD(entry)) {
3052	max_depth -= check_delta_limit(entry, 0);
3053	if (max_depth <= 0)
3054	goto next;
3055	}
3056
3057	j = window;
3058	while (--j > 0) {
3059	int ret;
3060	uint32_t other_idx = idx + j;
3061	struct unpacked *m;
3062	if (other_idx >= window)
3063	other_idx -= window;
3064	m = array + other_idx;
3065	if (!m->entry)
3066	break;
3067	ret = try_delta(n, m, max_depth, &mem_usage);
3068	if (ret < 0)
3069	break;
3070	else if (ret > 0)
3071	best_base = other_idx;
3072	}
3073
3074	/*
3075	* If we decided to cache the delta data, then it is best
3076	* to compress it right away. First because we have to do
3077	* it anyway, and doing it here while we're threaded will
3078	* save a lot of time in the non threaded write phase,
3079	* as well as allow for caching more deltas within
3080	* the same cache size limit.
3081	* ...
3082	* But only if not writing to stdout, since in that case
3083	* the network is most likely throttling writes anyway,
3084	* and therefore it is best to go to the write phase ASAP
3085	* instead, as we can afford spending more time compressing
3086	* between writes at that moment.
3087	*/
3088	if (entry->delta_data && !pack_to_stdout) {
3089	unsigned long size;
3090
3091	size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
3092	if (size < (1U << OE_Z_DELTA_BITS)) {
3093	entry->z_delta_size = size;
3094	cache_lock();
3095	delta_cache_size -= DELTA_SIZE(entry);
3096	delta_cache_size += entry->z_delta_size;
3097	cache_unlock();
3098	} else {
3099	FREE_AND_NULL(entry->delta_data);
3100	entry->z_delta_size = 0;
3101	}
3102	}
3103
3104	/* if we made n a delta, and if n is already at max
3105	* depth, leaving it in the window is pointless. we
3106	* should evict it first.
3107	*/
3108	if (DELTA(entry) && max_depth <= n->depth)
3109	continue;
3110
3111	/*
3112	* Move the best delta base up in the window, after the
3113	* currently deltified object, to keep it longer. It will
3114	* be the first base object to be attempted next.
3115	*/
3116	if (DELTA(entry)) {
3117	struct unpacked swap = array[best_base];
3118	int dist = (window + idx - best_base) % window;
3119	int dst = best_base;
3120	while (dist--) {
3121	int src = (dst + 1) % window;
3122	array[dst] = array[src];
3123	dst = src;
3124	}
3125	array[dst] = swap;
3126	}
3127
3128	next:
3129	idx++;
3130	if (count + 1 < window)
3131	count++;
3132	if (idx >= window)
3133	idx = 0;
3134	}
3135
3136	for (i = 0; i < window; ++i) {
3137	free_delta_index(array[i].index);
3138	free(array[i].data);
3139	}
3140	free(array);
3141	}
3142
3143	/*
3144	* The main object list is split into smaller lists, each is handed to
3145	* one worker.
3146	*
3147	* The main thread waits on the condition that (at least) one of the workers
3148	* has stopped working (which is indicated in the .working member of
3149	* struct thread_params).
3150	*
3151	* When a work thread has completed its work, it sets .working to 0 and
3152	* signals the main thread and waits on the condition that .data_ready
3153	* becomes 1.
3154	*
3155	* The main thread steals half of the work from the worker that has
3156	* most work left to hand it to the idle worker.
3157	*/
3158
3159	struct thread_params {
3160	pthread_t thread;
3161	struct object_entry **list;
3162	struct packing_region *regions;
3163	unsigned list_size;
3164	unsigned remaining;
3165	int window;
3166	int depth;
3167	int working;
3168	int data_ready;
3169	pthread_mutex_t mutex;
3170	pthread_cond_t cond;
3171	unsigned *processed;
3172	};
3173
3174	static pthread_cond_t progress_cond;
3175
3176	/*
3177	* Mutex and conditional variable can't be statically-initialized on Windows.
3178	*/
3179	static void init_threaded_search(void)
3180	{
3181	pthread_mutex_init(&cache_mutex, NULL);
3182	pthread_mutex_init(&progress_mutex, NULL);
3183	pthread_cond_init(&progress_cond, NULL);
3184	}
3185
3186	static void cleanup_threaded_search(void)
3187	{
3188	pthread_cond_destroy(&progress_cond);
3189	pthread_mutex_destroy(&cache_mutex);
3190	pthread_mutex_destroy(&progress_mutex);
3191	}
3192
3193	static void threaded_find_deltas(void arg)
3194	{
3195	struct thread_params *me = arg;
3196
3197	progress_lock();
3198	while (me->remaining) {
3199	progress_unlock();
3200
3201	find_deltas(me->list, &me->remaining,
3202	me->window, me->depth, me->processed);
3203
3204	progress_lock();
3205	me->working = 0;
3206	pthread_cond_signal(&progress_cond);
3207	progress_unlock();
3208
3209	/*
3210	* We must not set ->data_ready before we wait on the
3211	* condition because the main thread may have set it to 1
3212	* before we get here. In order to be sure that new
3213	* work is available if we see 1 in ->data_ready, it
3214	* was initialized to 0 before this thread was spawned
3215	* and we reset it to 0 right away.
3216	*/
3217	pthread_mutex_lock(&me->mutex);
3218	while (!me->data_ready)
3219	pthread_cond_wait(&me->cond, &me->mutex);
3220	me->data_ready = 0;
3221	pthread_mutex_unlock(&me->mutex);
3222
3223	progress_lock();
3224	}
3225	progress_unlock();
3226	/* leave ->working 1 so that this doesn't get more work assigned */
3227	return NULL;
3228	}
3229
3230	static void ll_find_deltas(struct object_entry **list, unsigned list_size,
3231	int window, int depth, unsigned *processed)
3232	{
3233	struct thread_params *p;
3234	int i, ret, active_threads = 0;
3235
3236	init_threaded_search();
3237
3238	if (delta_search_threads <= 1) {
3239	find_deltas(list, &list_size, window, depth, processed);
3240	cleanup_threaded_search();
3241	return;
3242	}
3243	if (progress > pack_to_stdout)
3244	fprintf_ln(stderr, _("Delta compression using up to %d threads"),
3245	delta_search_threads);
3246	CALLOC_ARRAY(p, delta_search_threads);
3247
3248	/* Partition the work amongst work threads. */
3249	for (i = 0; i < delta_search_threads; i++) {
3250	unsigned sub_size = list_size / (delta_search_threads - i);
3251
3252	/* don't use too small segments or no deltas will be found */
3253	if (sub_size < 2*window && i+1 < delta_search_threads)
3254	sub_size = 0;
3255
3256	p[i].window = window;
3257	p[i].depth = depth;
3258	p[i].processed = processed;
3259	p[i].working = 1;
3260	p[i].data_ready = 0;
3261
3262	/* try to split chunks on "path" boundaries */
3263	while (sub_size && sub_size < list_size &&
3264	list[sub_size]->hash &&
3265	list[sub_size]->hash == list[sub_size-1]->hash)
3266	sub_size++;
3267
3268	p[i].list = list;
3269	p[i].list_size = sub_size;
3270	p[i].remaining = sub_size;
3271
3272	list += sub_size;
3273	list_size -= sub_size;
3274	}
3275
3276	/* Start work threads. */
3277	for (i = 0; i < delta_search_threads; i++) {
3278	if (!p[i].list_size)
3279	continue;
3280	pthread_mutex_init(&p[i].mutex, NULL);
3281	pthread_cond_init(&p[i].cond, NULL);
3282	ret = pthread_create(&p[i].thread, NULL,
3283	threaded_find_deltas, &p[i]);
3284	if (ret)
3285	die(_("unable to create thread: %s"), strerror(ret));
3286	active_threads++;
3287	}
3288
3289	/*
3290	* Now let's wait for work completion. Each time a thread is done
3291	* with its work, we steal half of the remaining work from the
3292	* thread with the largest number of unprocessed objects and give
3293	* it to that newly idle thread. This ensure good load balancing
3294	* until the remaining object list segments are simply too short
3295	* to be worth splitting anymore.
3296	*/
3297	while (active_threads) {
3298	struct thread_params *target = NULL;
3299	struct thread_params *victim = NULL;
3300	unsigned sub_size = 0;
3301
3302	progress_lock();
3303	for (;;) {
3304	for (i = 0; !target && i < delta_search_threads; i++)
3305	if (!p[i].working)
3306	target = &p[i];
3307	if (target)
3308	break;
3309	pthread_cond_wait(&progress_cond, &progress_mutex);
3310	}
3311
3312	for (i = 0; i < delta_search_threads; i++)
3313	if (p[i].remaining > 2*window &&
3314	(!victim \|\| victim->remaining < p[i].remaining))
3315	victim = &p[i];
3316	if (victim) {
3317	sub_size = victim->remaining / 2;
3318	list = victim->list + victim->list_size - sub_size;
3319	while (sub_size && list[0]->hash &&
3320	list[0]->hash == list[-1]->hash) {
3321	list++;
3322	sub_size--;
3323	}
3324	if (!sub_size) {
3325	/*
3326	* It is possible for some "paths" to have
3327	* so many objects that no hash boundary
3328	* might be found. Let's just steal the
3329	* exact half in that case.
3330	*/
3331	sub_size = victim->remaining / 2;
3332	list -= sub_size;
3333	}
3334	target->list = list;
3335	victim->list_size -= sub_size;
3336	victim->remaining -= sub_size;
3337	}
3338	target->list_size = sub_size;
3339	target->remaining = sub_size;
3340	target->working = 1;
3341	progress_unlock();
3342
3343	pthread_mutex_lock(&target->mutex);
3344	target->data_ready = 1;
3345	pthread_cond_signal(&target->cond);
3346	pthread_mutex_unlock(&target->mutex);
3347
3348	if (!sub_size) {
3349	pthread_join(target->thread, NULL);
3350	pthread_cond_destroy(&target->cond);
3351	pthread_mutex_destroy(&target->mutex);
3352	active_threads--;
3353	}
3354	}
3355	cleanup_threaded_search();
3356	free(p);
3357	}
3358
3359	static int obj_is_packed(const struct object_id *oid)
3360	{
3361	return packlist_find(&to_pack, oid) \|\|
3362	(reuse_packfile_bitmap &&
3363	bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid));
3364	}
3365
3366	static void add_tag_chain(const struct object_id *oid)
3367	{
3368	struct tag *tag;
3369
3370	/*
3371	* We catch duplicates already in add_object_entry(), but we'd
3372	* prefer to do this extra check to avoid having to parse the
3373	* tag at all if we already know that it's being packed (e.g., if
3374	* it was included via bitmaps, we would not have parsed it
3375	* previously).
3376	*/
3377	if (obj_is_packed(oid))
3378	return;
3379
3380	tag = lookup_tag(the_repository, oid);
3381	while (1) {
3382	if (!tag \|\| parse_tag(the_repository, tag) \|\| !tag->tagged)
3383	die(_("unable to pack objects reachable from tag %s"),
3384	oid_to_hex(oid));
3385
3386	add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
3387
3388	if (tag->tagged->type != OBJ_TAG)
3389	return;
3390
3391	tag = (struct tag *)tag->tagged;
3392	}
3393	}
3394
3395	static int add_ref_tag(const struct reference ref, void cb_data UNUSED)
3396	{
3397	struct object_id peeled;
3398
3399	if (!reference_get_peeled_oid(the_repository, ref, &peeled) &&
3400	obj_is_packed(&peeled))
3401	add_tag_chain(ref->oid);
3402	return 0;
3403	}
3404
3405	static int should_attempt_deltas(struct object_entry *entry)
3406	{
3407	if (DELTA(entry))
3408	/* This happens if we decided to reuse existing
3409	* delta from a pack. "reuse_delta &&" is implied.
3410	*/
3411	return 0;
3412
3413	if (!entry->type_valid \|\|
3414	oe_size_less_than(&to_pack, entry, 50))
3415	return 0;
3416
3417	if (entry->no_try_delta)
3418	return 0;
3419
3420	if (entry->preferred_base && !allow_ref_delta)
3421	return 0;
3422
3423	if (!entry->preferred_base) {
3424	if (oe_type(entry) < 0)
3425	die(_("unable to get type of object %s"),
3426	oid_to_hex(&entry->idx.oid));
3427	} else if (oe_type(entry) < 0) {
3428	/*
3429	* This object is not found, but we
3430	* don't have to include it anyway.
3431	*/
3432	return 0;
3433	}
3434
3435	return 1;
3436	}
3437
3438	static void find_deltas_for_region(struct object_entry *list,
3439	struct packing_region *region,
3440	unsigned int *processed)
3441	{
3442	struct object_entry **delta_list;
3443	unsigned int delta_list_nr = 0;
3444
3445	ALLOC_ARRAY(delta_list, region->nr);
3446	for (size_t i = 0; i < region->nr; i++) {
3447	struct object_entry *entry = list + region->start + i;
3448	if (should_attempt_deltas(entry))
3449	delta_list[delta_list_nr++] = entry;
3450	}
3451
3452	QSORT(delta_list, delta_list_nr, type_size_sort);
3453	find_deltas(delta_list, &delta_list_nr, window, depth, processed);
3454	free(delta_list);
3455	}
3456
3457	static void find_deltas_by_region(struct object_entry *list,
3458	struct packing_region *regions,
3459	size_t start, size_t nr)
3460	{
3461	unsigned int processed = 0;
3462	size_t progress_nr;
3463
3464	if (!nr)
3465	return;
3466
3467	progress_nr = regions[nr - 1].start + regions[nr - 1].nr;
3468
3469	if (progress)
3470	progress_state = start_progress(the_repository,
3471	_("Compressing objects by path"),
3472	progress_nr);
3473
3474	while (nr--)
3475	find_deltas_for_region(list,
3476	&regions[start++],
3477	&processed);
3478
3479	display_progress(progress_state, progress_nr);
3480	stop_progress(&progress_state);
3481	}
3482
3483	static void threaded_find_deltas_by_path(void arg)
3484	{
3485	struct thread_params *me = arg;
3486
3487	progress_lock();
3488	while (me->remaining) {
3489	while (me->remaining) {
3490	progress_unlock();
3491	find_deltas_for_region(to_pack.objects,
3492	me->regions,
3493	me->processed);
3494	progress_lock();
3495	me->remaining--;
3496	me->regions++;
3497	}
3498
3499	me->working = 0;
3500	pthread_cond_signal(&progress_cond);
3501	progress_unlock();
3502
3503	/*
3504	* We must not set ->data_ready before we wait on the
3505	* condition because the main thread may have set it to 1
3506	* before we get here. In order to be sure that new
3507	* work is available if we see 1 in ->data_ready, it
3508	* was initialized to 0 before this thread was spawned
3509	* and we reset it to 0 right away.
3510	*/
3511	pthread_mutex_lock(&me->mutex);
3512	while (!me->data_ready)
3513	pthread_cond_wait(&me->cond, &me->mutex);
3514	me->data_ready = 0;
3515	pthread_mutex_unlock(&me->mutex);
3516
3517	progress_lock();
3518	}
3519	progress_unlock();
3520	/* leave ->working 1 so that this doesn't get more work assigned */
3521	return NULL;
3522	}
3523
3524	static void ll_find_deltas_by_region(struct object_entry *list,
3525	struct packing_region *regions,
3526	uint32_t start, uint32_t nr)
3527	{
3528	struct thread_params *p;
3529	int i, ret, active_threads = 0;
3530	unsigned int processed = 0;
3531	uint32_t progress_nr;
3532	init_threaded_search();
3533
3534	if (!nr)
3535	return;
3536
3537	progress_nr = regions[nr - 1].start + regions[nr - 1].nr;
3538	if (delta_search_threads <= 1) {
3539	find_deltas_by_region(list, regions, start, nr);
3540	cleanup_threaded_search();
3541	return;
3542	}
3543
3544	if (progress > pack_to_stdout)
3545	fprintf_ln(stderr,
3546	Q_("Path-based delta compression using up to %d thread",
3547	"Path-based delta compression using up to %d threads",
3548	delta_search_threads),
3549	delta_search_threads);
3550	CALLOC_ARRAY(p, delta_search_threads);
3551
3552	if (progress)
3553	progress_state = start_progress(the_repository,
3554	_("Compressing objects by path"),
3555	progress_nr);
3556	/* Partition the work amongst work threads. */
3557	for (i = 0; i < delta_search_threads; i++) {
3558	unsigned sub_size = nr / (delta_search_threads - i);
3559
3560	p[i].window = window;
3561	p[i].depth = depth;
3562	p[i].processed = &processed;
3563	p[i].working = 1;
3564	p[i].data_ready = 0;
3565
3566	p[i].regions = regions;
3567	p[i].list_size = sub_size;
3568	p[i].remaining = sub_size;
3569
3570	regions += sub_size;
3571	nr -= sub_size;
3572	}
3573
3574	/* Start work threads. */
3575	for (i = 0; i < delta_search_threads; i++) {
3576	if (!p[i].list_size)
3577	continue;
3578	pthread_mutex_init(&p[i].mutex, NULL);
3579	pthread_cond_init(&p[i].cond, NULL);
3580	ret = pthread_create(&p[i].thread, NULL,
3581	threaded_find_deltas_by_path, &p[i]);
3582	if (ret)
3583	die(_("unable to create thread: %s"), strerror(ret));
3584	active_threads++;
3585	}
3586
3587	/*
3588	* Now let's wait for work completion. Each time a thread is done
3589	* with its work, we steal half of the remaining work from the
3590	* thread with the largest number of unprocessed objects and give
3591	* it to that newly idle thread. This ensure good load balancing
3592	* until the remaining object list segments are simply too short
3593	* to be worth splitting anymore.
3594	*/
3595	while (active_threads) {
3596	struct thread_params *target = NULL;
3597	struct thread_params *victim = NULL;
3598	unsigned sub_size = 0;
3599
3600	progress_lock();
3601	for (;;) {
3602	for (i = 0; !target && i < delta_search_threads; i++)
3603	if (!p[i].working)
3604	target = &p[i];
3605	if (target)
3606	break;
3607	pthread_cond_wait(&progress_cond, &progress_mutex);
3608	}
3609
3610	for (i = 0; i < delta_search_threads; i++)
3611	if (p[i].remaining > 2*window &&
3612	(!victim \|\| victim->remaining < p[i].remaining))
3613	victim = &p[i];
3614	if (victim) {
3615	sub_size = victim->remaining / 2;
3616	target->regions = victim->regions + victim->remaining - sub_size;
3617	victim->list_size -= sub_size;
3618	victim->remaining -= sub_size;
3619	}
3620	target->list_size = sub_size;
3621	target->remaining = sub_size;
3622	target->working = 1;
3623	progress_unlock();
3624
3625	pthread_mutex_lock(&target->mutex);
3626	target->data_ready = 1;
3627	pthread_cond_signal(&target->cond);
3628	pthread_mutex_unlock(&target->mutex);
3629
3630	if (!sub_size) {
3631	pthread_join(target->thread, NULL);
3632	pthread_cond_destroy(&target->cond);
3633	pthread_mutex_destroy(&target->mutex);
3634	active_threads--;
3635	}
3636	}
3637	cleanup_threaded_search();
3638	free(p);
3639
3640	display_progress(progress_state, progress_nr);
3641	stop_progress(&progress_state);
3642	}
3643
3644	static void prepare_pack(int window, int depth)
3645	{
3646	struct object_entry **delta_list;
3647	uint32_t i, nr_deltas;
3648	unsigned n;
3649
3650	if (use_delta_islands)
3651	resolve_tree_islands(the_repository, progress, &to_pack);
3652
3653	get_object_details();
3654
3655	/*
3656	* If we're locally repacking then we need to be doubly careful
3657	* from now on in order to make sure no stealth corruption gets
3658	* propagated to the new pack. Clients receiving streamed packs
3659	* should validate everything they get anyway so no need to incur
3660	* the additional cost here in that case.
3661	*/
3662	if (!pack_to_stdout)
3663	do_check_packed_object_crc = 1;
3664
3665	if (!to_pack.nr_objects \|\| !window \|\| !depth \|\|
3666	(!allow_ref_delta && !allow_ofs_delta))
3667	return;
3668
3669	if (path_walk)
3670	ll_find_deltas_by_region(to_pack.objects, to_pack.regions,
3671	0, to_pack.nr_regions);
3672
3673	ALLOC_ARRAY(delta_list, to_pack.nr_objects);
3674	nr_deltas = n = 0;
3675
3676	for (i = 0; i < to_pack.nr_objects; i++) {
3677	struct object_entry *entry = to_pack.objects + i;
3678
3679	if (!should_attempt_deltas(entry))
3680	continue;
3681
3682	if (!entry->preferred_base)
3683	nr_deltas++;
3684
3685	delta_list[n++] = entry;
3686	}
3687
3688	if (nr_deltas && n > 1) {
3689	unsigned nr_done = 0;
3690
3691	if (progress)
3692	progress_state = start_progress(the_repository,
3693	_("Compressing objects"),
3694	nr_deltas);
3695	QSORT(delta_list, n, type_size_sort);
3696	ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
3697	stop_progress(&progress_state);
3698	if (nr_done != nr_deltas)
3699	die(_("inconsistency with delta count"));
3700	}
3701	free(delta_list);
3702	}
3703
3704	static int git_pack_config(const char k, const char v,
3705	const struct config_context ctx, void cb)
3706	{
3707	if (!strcmp(k, "pack.window")) {
3708	window = git_config_int(k, v, ctx->kvi);
3709	return 0;
3710	}
3711	if (!strcmp(k, "pack.windowmemory")) {
3712	window_memory_limit = git_config_ulong(k, v, ctx->kvi);
3713	return 0;
3714	}
3715	if (!strcmp(k, "pack.depth")) {
3716	depth = git_config_int(k, v, ctx->kvi);
3717	return 0;
3718	}
3719	if (!strcmp(k, "pack.deltacachesize")) {
3720	max_delta_cache_size = git_config_int(k, v, ctx->kvi);
3721	return 0;
3722	}
3723	if (!strcmp(k, "pack.deltacachelimit")) {
3724	cache_max_small_delta_size = git_config_int(k, v, ctx->kvi);
3725	return 0;
3726	}
3727	if (!strcmp(k, "pack.writebitmaphashcache")) {
3728	if (git_config_bool(k, v))
3729	write_bitmap_options \|= BITMAP_OPT_HASH_CACHE;
3730	else
3731	write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
3732	}
3733
3734	if (!strcmp(k, "pack.writebitmaplookuptable")) {
3735	if (git_config_bool(k, v))
3736	write_bitmap_options \|= BITMAP_OPT_LOOKUP_TABLE;
3737	else
3738	write_bitmap_options &= ~BITMAP_OPT_LOOKUP_TABLE;
3739	}
3740
3741	if (!strcmp(k, "pack.usebitmaps")) {
3742	use_bitmap_index_default = git_config_bool(k, v);
3743	return 0;
3744	}
3745	if (!strcmp(k, "pack.allowpackreuse")) {
3746	int res = git_parse_maybe_bool_text(v);
3747	if (res < 0) {
3748	if (!strcasecmp(v, "single"))
3749	allow_pack_reuse = SINGLE_PACK_REUSE;
3750	else if (!strcasecmp(v, "multi"))
3751	allow_pack_reuse = MULTI_PACK_REUSE;
3752	else
3753	die(_("invalid pack.allowPackReuse value: '%s'"), v);
3754	} else if (res) {
3755	allow_pack_reuse = SINGLE_PACK_REUSE;
3756	} else {
3757	allow_pack_reuse = NO_PACK_REUSE;
3758	}
3759	return 0;
3760	}
3761	if (!strcmp(k, "pack.threads")) {
3762	delta_search_threads = git_config_int(k, v, ctx->kvi);
3763	if (delta_search_threads < 0)
3764	die(_("invalid number of threads specified (%d)"),
3765	delta_search_threads);
3766	if (!HAVE_THREADS && delta_search_threads != 1) {
3767	warning(_("no threads support, ignoring %s"), k);
3768	delta_search_threads = 0;
3769	}
3770	return 0;
3771	}
3772	if (!strcmp(k, "pack.indexversion")) {
3773	pack_idx_opts.version = git_config_int(k, v, ctx->kvi);
3774	if (pack_idx_opts.version > 2)
3775	die(_("bad pack.indexVersion=%"PRIu32),
3776	pack_idx_opts.version);
3777	return 0;
3778	}
3779	if (!strcmp(k, "pack.writereverseindex")) {
3780	if (git_config_bool(k, v))
3781	pack_idx_opts.flags \|= WRITE_REV;
3782	else
3783	pack_idx_opts.flags &= ~WRITE_REV;
3784	return 0;
3785	}
3786	if (!strcmp(k, "uploadpack.blobpackfileuri")) {
3787	struct configured_exclusion *ex;
3788	const char oid_end, pack_end;
3789	/*
3790	* Stores the pack hash. This is not a true object ID, but is
3791	* of the same form.
3792	*/
3793	struct object_id pack_hash;
3794
3795	if (!v)
3796	return config_error_nonbool(k);
3797
3798	ex = xmalloc(sizeof(*ex));
3799	if (parse_oid_hex(v, &ex->e.oid, &oid_end) \|\|
3800	*oid_end != ' ' \|\|
3801	parse_oid_hex(oid_end + 1, &pack_hash, &pack_end) \|\|
3802	*pack_end != ' ')
3803	die(_("value of uploadpack.blobpackfileuri must be "
3804	"of the form '<object-hash> <pack-hash> <uri>' (got '%s')"), v);
3805	if (oidmap_get(&configured_exclusions, &ex->e.oid))
3806	die(_("object already configured in another "
3807	"uploadpack.blobpackfileuri (got '%s')"), v);
3808	ex->pack_hash_hex = xcalloc(1, pack_end - oid_end);
3809	memcpy(ex->pack_hash_hex, oid_end + 1, pack_end - oid_end - 1);
3810	ex->uri = xstrdup(pack_end + 1);
3811	oidmap_put(&configured_exclusions, ex);
3812	}
3813	return git_default_config(k, v, ctx, cb);
3814	}
3815
3816	/* Counters for trace2 output when in --stdin-packs mode. */
3817	static int stdin_packs_found_nr;
3818	static int stdin_packs_hints_nr;
3819
3820	static int add_object_entry_from_pack(const struct object_id *oid,
3821	struct packed_git *p,
3822	uint32_t pos,
3823	void *_data)
3824	{
3825	off_t ofs;
3826	struct object_info oi = OBJECT_INFO_INIT;
3827	enum object_type type = OBJ_NONE;
3828
3829	display_progress(progress_state, ++nr_seen);
3830
3831	if (have_duplicate_entry(oid, 0))
3832	return 0;
3833
3834	stdin_packs_found_nr++;
3835
3836	ofs = nth_packed_object_offset(p, pos);
3837
3838	oi.typep = &type;
3839	if (packed_object_info(NULL, p, ofs, &oi) < 0) {
3840	die(_("could not get type of object %s in pack %s"),
3841	oid_to_hex(oid), p->pack_name);
3842	} else if (type == OBJ_COMMIT) {
3843	struct rev_info *revs = _data;
3844	/*
3845	* commits in included packs are used as starting points
3846	* for the subsequent revision walk
3847	*
3848	* Note that we do want to walk through commits that are
3849	* present in excluded-open ('!') packs to pick up any
3850	* objects reachable from them not present in the
3851	* excluded-closed ('^') packs.
3852	*
3853	* However, we'll only add those objects to the packing
3854	* list after checking `want_object_in_pack()` below.
3855	*/
3856	add_pending_oid(revs, NULL, oid, 0);
3857	}
3858
3859	if (!want_object_in_pack(oid, 0, &p, &ofs))
3860	return 0;
3861
3862	create_object_entry(oid, type, 0, 0, 0, p, ofs);
3863
3864	return 0;
3865	}
3866
3867	static void show_object_pack_hint(struct object object, const char name,
3868	void *data)
3869	{
3870	enum stdin_packs_mode mode = (enum stdin_packs_mode )data;
3871	if (mode == STDIN_PACKS_MODE_FOLLOW \|\|
3872	mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE) {
3873	if (object->type == OBJ_BLOB &&
3874	!odb_has_object(the_repository->objects, &object->oid, 0))
3875	return;
3876	add_object_entry(&object->oid, object->type, name, 0);
3877	} else {
3878	struct object_entry *oe = packlist_find(&to_pack, &object->oid);
3879	if (!oe)
3880	return;
3881
3882	/*
3883	* Our 'to_pack' list was constructed by iterating all
3884	* objects packed in included packs, and so doesn't have
3885	* a non-zero hash field that you would typically pick
3886	* up during a reachability traversal.
3887	*
3888	* Make a best-effort attempt to fill in the ->hash and
3889	* ->no_try_delta fields here in order to perhaps
3890	* improve the delta selection process.
3891	*/
3892	oe->hash = pack_name_hash_fn(name);
3893	oe->no_try_delta = name && no_try_delta(name);
3894
3895	stdin_packs_hints_nr++;
3896	}
3897	}
3898
3899	static void show_commit_pack_hint(struct commit commit, void data)
3900	{
3901	enum stdin_packs_mode mode = (enum stdin_packs_mode )data;
3902
3903	if (mode == STDIN_PACKS_MODE_FOLLOW \|\|
3904	mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE) {
3905	show_object_pack_hint((struct object *)commit, "", data);
3906	return;
3907	}
3908
3909	/* nothing to do; commits don't have a namehash */
3910
3911	}
3912
3913	/*
3914	* stdin_pack_info_kind specifies how a pack specified over stdin
3915	* should be treated when pack-objects is invoked with --stdin-packs.
3916	*
3917	* - STDIN_PACK_INCLUDE: objects in any packs with this flag bit set
3918	* should be included in the output pack, unless they appear in an
3919	* excluded pack.
3920	*
3921	* - STDIN_PACK_EXCLUDE_CLOSED: objects in any packs with this flag
3922	* bit set should be excluded from the output pack.
3923	*
3924	* - STDIN_PACK_EXCLUDE_OPEN: objects in any packs with this flag
3925	* bit set should be excluded from the output pack, but are not
3926	* guaranteed to be closed under reachability.
3927	*
3928	* Objects in packs whose 'kind' bits include STDIN_PACK_INCLUDE or
3929	* STDIN_PACK_EXCLUDE_OPEN are used as traversal tips when invoked
3930	* with --stdin-packs=follow.
3931	*/
3932	enum stdin_pack_info_kind {
3933	STDIN_PACK_INCLUDE = (1<<0),
3934	STDIN_PACK_EXCLUDE_CLOSED = (1<<1),
3935	STDIN_PACK_EXCLUDE_OPEN = (1<<2),
3936	};
3937
3938	struct stdin_pack_info {
3939	struct packed_git *p;
3940	enum stdin_pack_info_kind kind;
3941	};
3942
3943	static int pack_mtime_cmp(const void _a, const void _b)
3944	{
3945	struct stdin_pack_info a = ((const struct string_list_item)_a)->util;
3946	struct stdin_pack_info b = ((const struct string_list_item)_b)->util;
3947
3948	/*
3949	* order packs by descending mtime so that objects are laid out
3950	* roughly as newest-to-oldest
3951	*/
3952	if (a->p->mtime < b->p->mtime)
3953	return 1;
3954	else if (b->p->mtime < a->p->mtime)
3955	return -1;
3956	else
3957	return 0;
3958	}
3959
3960	static int stdin_packs_include_check_obj(struct object obj, void data UNUSED)
3961	{
3962	return !has_object_kept_pack(to_pack.repo, &obj->oid,
3963	KEPT_PACK_IN_CORE);
3964	}
3965
3966	static int stdin_packs_include_check(struct commit commit, void data)
3967	{
3968	return stdin_packs_include_check_obj((struct object *)commit, data);
3969	}
3970
3971	/*
3972	* Flag bit set on commits that belong to an included pack during
3973	* '--stdin-packs=follow-reachable'. Used by the pre-walk to
3974	* identify which reachable commits should be tips for the main
3975	* object traversal.
3976	*/
3977	#define IN_INCLUDED_PACK (1u<<11)
3978
3979	static int mark_included_pack_tip(const struct object_id *oid,
3980	struct packed_git *p,
3981	uint32_t pos,
3982	void *data)
3983	{
3984	struct rev_info *main_revs = data;
3985	off_t ofs = nth_packed_object_offset(p, pos);
3986	enum object_type type;
3987	struct object_info oi = OBJECT_INFO_INIT;
3988	struct object *obj;
3989
3990	oi.typep = &type;
3991	if (packed_object_info(NULL, p, ofs, &oi) < 0)
3992	return 0;
3993	if (type != OBJ_COMMIT && type != OBJ_TAG)
3994	return 0;
3995
3996	obj = parse_object(the_repository, oid);
3997	if (!obj)
3998	return 0;
3999
4000	obj->flags \|= IN_INCLUDED_PACK;
4001
4002	if (type == OBJ_TAG && main_revs)
4003	add_pending_object(main_revs, obj, "");
4004	return 0;
4005	}
4006
4007	static int mark_loose_object_tip(const struct object_id *oid,
4008	struct object_info *oi UNUSED,
4009	void *data)
4010	{
4011	struct rev_info *main_revs = data;
4012	struct object *obj;
4013	enum object_type type;
4014
4015	type = odb_read_object_info(the_repository->objects, oid, NULL);
4016	if (type != OBJ_COMMIT && type != OBJ_TAG)
4017	return 0;
4018
4019	obj = parse_object(the_repository, oid);
4020	if (!obj)
4021	return 0;
4022
4023	obj->flags \|= IN_INCLUDED_PACK;
4024
4025	if (type == OBJ_TAG && main_revs)
4026	add_pending_object(main_revs, obj, "");
4027
4028	return 0;
4029	}
4030
4031	static int add_ref_to_pending(const struct reference ref, void cb_data)
4032	{
4033	struct rev_info *revs = cb_data;
4034	struct object *object;
4035
4036	object = parse_object(the_repository, ref->oid);
4037	if (!object)
4038	return 0;
4039
4040	add_pending_object(revs, object, "");
4041	return 0;
4042	}
4043
4044	static void read_refs_snapshot(const char *refs_snapshot,
4045	struct rev_info *revs)
4046	{
4047	struct strbuf buf = STRBUF_INIT;
4048	struct object_id oid;
4049	FILE *f = xfopen(refs_snapshot, "r");
4050
4051	while (strbuf_getline(&buf, f) != EOF) {
4052	struct object *object;
4053	const char *hex = buf.buf;
4054	const char *end = NULL;
4055
4056	if (*hex == '+')
4057	hex++;
4058
4059	if (parse_oid_hex_algop(hex, &oid, &end,
4060	the_repository->hash_algo) < 0)
4061	die(_("could not parse line: %s"), buf.buf);
4062	if (*end)
4063	die(_("malformed line: %s"), buf.buf);
4064
4065	object = parse_object(the_repository, &oid);
4066	if (!object)
4067	continue;
4068
4069	add_pending_object(revs, object, "");
4070	}
4071
4072	fclose(f);
4073	strbuf_release(&buf);
4074	}
4075
4076	static void stdin_packs_add_reachable_pack_entries(struct string_list *keys,
4077	struct rev_info *revs,
4078	int rev_list_unpacked)
4079	{
4080	struct rev_info pre_walk;
4081	struct commit *commit;
4082	struct string_list_item *item;
4083
4084	/*
4085	* Phase 1: mark commits in included packs, then walk from
4086	* ref tips to discover which of them are reachable. The walk
4087	* halts at excluded-closed packs (via no_kept_objects) and
4088	* continues through excluded-open ones.
4089	*
4090	* Also set include_check on the outer revs so that phase 2
4091	* (the main object traversal) halts at closed packs.
4092	*/
4093	revs->include_check = stdin_packs_include_check;
4094	revs->include_check_obj = stdin_packs_include_check_obj;
4095
4096	for_each_string_list_item(item, keys) {
4097	struct stdin_pack_info *info = item->util;
4098	if (info->kind & STDIN_PACK_INCLUDE)
4099	for_each_object_in_pack(info->p,
4100	mark_included_pack_tip,
4101	revs,
4102	ODB_FOR_EACH_OBJECT_PACK_ORDER);
4103	}
4104
4105	if (rev_list_unpacked) {
4106	/*
4107	* With '--stdin-packs=follow-reachable', specifying
4108	* '--unpacked' instructs pack-objects to pack any loose
4109	* objects which are reachable.
4110	*
4111	* Pretend as if all loose objects are in an included
4112	* pack in order to make them eligible for packing.
4113	*/
4114	struct odb_source *source = revs->repo->objects->sources;
4115	for (; source; source = source->next) {
4116	struct odb_source_files *files = odb_source_files_downcast(source);
4117	struct odb_for_each_object_options opts = { 0 };
4118	if (local)
4119	opts.flags \|= ODB_FOR_EACH_OBJECT_LOCAL_ONLY;
4120
4121	odb_source_for_each_object(&files->loose->base, NULL,
4122	mark_loose_object_tip,
4123	revs, &opts);
4124	}
4125	}
4126
4127	repo_init_revisions(the_repository, &pre_walk, NULL);
4128	pre_walk.no_kept_objects = 1;
4129	pre_walk.keep_pack_cache_flags \|= KEPT_PACK_IN_CORE;
4130	pre_walk.ignore_missing_links = 1;
4131
4132	if (stdin_packs_refs_snapshot)
4133	read_refs_snapshot(stdin_packs_refs_snapshot, &pre_walk);
4134	else
4135	refs_for_each_ref(get_main_ref_store(the_repository),
4136	add_ref_to_pending, &pre_walk);
4137
4138	if (prepare_revision_walk(&pre_walk))
4139	die(_("revision walk setup failed"));
4140
4141	/*
4142	* Phase 2 tips: every reachable commit that is in an
4143	* included pack becomes a starting point for the main
4144	* object traversal.
4145	*/
4146	while ((commit = get_revision(&pre_walk)) != NULL) {
4147	if (commit->object.flags & IN_INCLUDED_PACK)
4148	add_pending_oid(revs, NULL,
4149	&commit->object.oid, 0);
4150	}
4151
4152	reset_revision_walk();
4153	release_revisions(&pre_walk);
4154	}
4155
4156	static void stdin_packs_add_all_pack_entries(struct string_list *keys,
4157	struct rev_info *revs)
4158	{
4159	struct string_list_item *item;
4160
4161	for_each_string_list_item(item, keys) {
4162	struct stdin_pack_info *info = item->util;
4163
4164	if (info->kind & STDIN_PACK_EXCLUDE_OPEN) {
4165	/*
4166	* When open-excluded packs ("!") are present, stop
4167	* the parent walk at closed-excluded ("^") packs.
4168	* Objects behind a "^" boundary are guaranteed to
4169	* have closure and should not be rescued.
4170	*/
4171	revs->include_check = stdin_packs_include_check;
4172	revs->include_check_obj = stdin_packs_include_check_obj;
4173	}
4174
4175	if ((info->kind & STDIN_PACK_INCLUDE) \|\|
4176	(info->kind & STDIN_PACK_EXCLUDE_OPEN))
4177	for_each_object_in_pack(info->p,
4178	add_object_entry_from_pack,
4179	revs,
4180	ODB_FOR_EACH_OBJECT_PACK_ORDER);
4181	}
4182	}
4183
4184	static void stdin_packs_add_pack_entries(struct strmap *packs,
4185	struct rev_info *revs,
4186	enum stdin_packs_mode mode,
4187	int rev_list_unpacked)
4188	{
4189	struct string_list keys = STRING_LIST_INIT_NODUP;
4190	struct hashmap_iter iter;
4191	struct strmap_entry *entry;
4192
4193	strmap_for_each_entry(packs, &iter, entry) {
4194	struct stdin_pack_info *info = entry->value;
4195	if (!info->p)
4196	die(_("could not find pack '%s'"), entry->key);
4197
4198	string_list_append(&keys, entry->key)->util = info;
4199	}
4200
4201	/*
4202	* Order packs by ascending mtime; use QSORT directly to access the
4203	* string_list_item's ->util pointer, which string_list_sort() does not
4204	* provide.
4205	*/
4206	QSORT(keys.items, keys.nr, pack_mtime_cmp);
4207
4208	if (mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE)
4209	stdin_packs_add_reachable_pack_entries(&keys, revs,
4210	rev_list_unpacked);
4211	else
4212	stdin_packs_add_all_pack_entries(&keys, revs);
4213
4214	string_list_clear(&keys, 0);
4215	}
4216
4217	static void stdin_packs_read_input(struct rev_info *revs,
4218	enum stdin_packs_mode mode,
4219	int rev_list_unpacked)
4220	{
4221	struct strbuf buf = STRBUF_INIT;
4222	struct strmap packs = STRMAP_INIT;
4223	struct packed_git *p;
4224
4225	while (strbuf_getline(&buf, stdin) != EOF) {
4226	struct stdin_pack_info *info;
4227	enum stdin_pack_info_kind kind = STDIN_PACK_INCLUDE;
4228	const char *key = buf.buf;
4229
4230	if (!*key)
4231	continue;
4232	else if (*key == '^')
4233	kind = STDIN_PACK_EXCLUDE_CLOSED;
4234	else if (*key == '!' &&
4235	(mode == STDIN_PACKS_MODE_FOLLOW \|\|
4236	mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE))
4237	kind = STDIN_PACK_EXCLUDE_OPEN;
4238
4239	if (kind != STDIN_PACK_INCLUDE)
4240	key++;
4241
4242	info = strmap_get(&packs, key);
4243	if (!info) {
4244	CALLOC_ARRAY(info, 1);
4245	strmap_put(&packs, key, info);
4246	}
4247
4248	info->kind \|= kind;
4249
4250	strbuf_reset(&buf);
4251	}
4252
4253	repo_for_each_pack(the_repository, p) {
4254	struct stdin_pack_info *info;
4255
4256	info = strmap_get(&packs, pack_basename(p));
4257	if (!info)
4258	continue;
4259
4260	if (info->kind & STDIN_PACK_INCLUDE) {
4261	if (exclude_promisor_objects && p->pack_promisor)
4262	die(_("packfile %s is a promisor but --exclude-promisor-objects was given"), p->pack_name);
4263
4264	/*
4265	* Arguments we got on stdin may not even be
4266	* packs. First check that to avoid segfaulting
4267	* later on in e.g. pack_mtime_cmp(), excluded
4268	* packs are handled below.
4269	*/
4270	if (!is_pack_valid(p))
4271	die(_("packfile %s cannot be accessed"), p->pack_name);
4272	}
4273
4274	if (info->kind & STDIN_PACK_EXCLUDE_CLOSED) {
4275	/*
4276	* Marking excluded packs as kept in-core so
4277	* that later calls to add_object_entry()
4278	* discards any objects that are also found in
4279	* excluded packs.
4280	*/
4281	p->pack_keep_in_core = 1;
4282	}
4283
4284	if (info->kind & STDIN_PACK_EXCLUDE_OPEN) {
4285	/*
4286	* Marking excluded open packs as kept in-core
4287	* (open) for the same reason as we marked
4288	* exclude closed packs as kept in-core.
4289	*
4290	* Use a separate flag here to ensure we don't
4291	* halt our traversal at these packs, since they
4292	* are not guaranteed to have closure.
4293	*
4294	*/
4295	p->pack_keep_in_core_open = 1;
4296	}
4297
4298	info->p = p;
4299	}
4300
4301	stdin_packs_add_pack_entries(&packs, revs, mode, rev_list_unpacked);
4302
4303	strbuf_release(&buf);
4304	strmap_clear(&packs, 1);
4305	}
4306
4307	static void add_unreachable_loose_objects(struct rev_info *revs);
4308
4309	static void read_stdin_packs(enum stdin_packs_mode mode, int rev_list_unpacked)
4310	{
4311	int prev_fetch_if_missing = fetch_if_missing;
4312	struct rev_info revs;
4313
4314	/*
4315	* The revision walk may hit objects that are promised, only. As the
4316	* walk is best-effort though we don't want to perform backfill fetches
4317	* for them.
4318	*/
4319	fetch_if_missing = 0;
4320
4321	repo_init_revisions(the_repository, &revs, NULL);
4322	/*
4323	* Use a revision walk to fill in the namehash of objects in the include
4324	* packs. To save time, we'll avoid traversing through objects that are
4325	* in excluded packs.
4326	*
4327	* That may cause us to avoid populating all of the namehash fields of
4328	* all included objects, but our goal is best-effort, since this is only
4329	* an optimization during delta selection.
4330	*/
4331	revs.no_kept_objects = 1;
4332	revs.keep_pack_cache_flags \|= KEPT_PACK_IN_CORE;
4333	revs.blob_objects = 1;
4334	revs.tree_objects = 1;
4335	revs.tag_objects = 1;
4336	revs.ignore_missing_links = 1;
4337	revs.exclude_promisor_objects = exclude_promisor_objects;
4338
4339	/* avoids adding objects in excluded packs */
4340	ignore_packed_keep_in_core = 1;
4341	if (mode == STDIN_PACKS_MODE_FOLLOW \|\|
4342	mode == STDIN_PACKS_MODE_FOLLOW_REACHABLE) {
4343	/*
4344	* In '--stdin-packs=follow' mode, additionally ignore
4345	* objects in excluded-open packs to prevent them from
4346	* appearing in the resulting pack.
4347	*/
4348	ignore_packed_keep_in_core_open = 1;
4349	}
4350	stdin_packs_read_input(&revs, mode, rev_list_unpacked);
4351	if (rev_list_unpacked && mode != STDIN_PACKS_MODE_FOLLOW_REACHABLE)
4352	add_unreachable_loose_objects(&revs);
4353
4354	if (prepare_revision_walk(&revs))
4355	die(_("revision walk setup failed"));
4356	traverse_commit_list(&revs,
4357	show_commit_pack_hint,
4358	show_object_pack_hint,
4359	&mode);
4360
4361	release_revisions(&revs);
4362
4363	trace2_data_intmax("pack-objects", the_repository, "stdin_packs_found",
4364	stdin_packs_found_nr);
4365	trace2_data_intmax("pack-objects", the_repository, "stdin_packs_hints",
4366	stdin_packs_hints_nr);
4367
4368	fetch_if_missing = prev_fetch_if_missing;
4369	}
4370
4371	static void add_cruft_object_entry(const struct object_id *oid, enum object_type type,
4372	struct packed_git *pack, off_t offset,
4373	const char *name, uint32_t mtime)
4374	{
4375	struct object_entry *entry;
4376
4377	display_progress(progress_state, ++nr_seen);
4378
4379	entry = packlist_find(&to_pack, oid);
4380	if (entry) {
4381	if (name) {
4382	entry->hash = pack_name_hash_fn(name);
4383	entry->no_try_delta = no_try_delta(name);
4384	}
4385	} else {
4386	if (!want_object_in_pack_mtime(oid, 0, &pack, &offset, mtime))
4387	return;
4388	if (!pack && type == OBJ_BLOB) {
4389	struct odb_source *source = the_repository->objects->sources;
4390	int found = 0;
4391
4392	for (; !found && source; source = source->next) {
4393	struct odb_source_files *files = odb_source_files_downcast(source);
4394	if (!odb_source_read_object_info(&files->loose->base, oid, NULL, 0))
4395	found = 1;
4396	}
4397
4398	/*
4399	* If a traversed tree has a missing blob then we want
4400	* to avoid adding that missing object to our pack.
4401	*
4402	* This only applies to missing blobs, not trees,
4403	* because the traversal needs to parse sub-trees but
4404	* not blobs.
4405	*
4406	* Note we only perform this check when we couldn't
4407	* already find the object in a pack, so we're really
4408	* limited to "ensure non-tip blobs which don't exist in
4409	* packs do exist via loose objects". Confused?
4410	*/
4411	if (!found)
4412	return;
4413	}
4414
4415	entry = create_object_entry(oid, type, pack_name_hash_fn(name),
4416	0, name && no_try_delta(name),
4417	pack, offset);
4418	}
4419
4420	if (mtime > oe_cruft_mtime(&to_pack, entry))
4421	oe_set_cruft_mtime(&to_pack, entry, mtime);
4422	return;
4423	}
4424
4425	static void show_cruft_object(struct object obj, const char name, void *data UNUSED)
4426	{
4427	/*
4428	* if we did not record it earlier, it's at least as old as our
4429	* expiration value. Rather than find it exactly, just use that
4430	* value. This may bump it forward from its real mtime, but it
4431	* will still be "too old" next time we run with the same
4432	* expiration.
4433	*
4434	* if obj does appear in the packing list, this call is a noop (or may
4435	* set the namehash).
4436	*/
4437	add_cruft_object_entry(&obj->oid, obj->type, NULL, 0, name, cruft_expiration);
4438	}
4439
4440	static void show_cruft_commit(struct commit commit, void data)
4441	{
4442	show_cruft_object((struct object*)commit, NULL, data);
4443	}
4444
4445	static int cruft_include_check_obj(struct object obj, void data UNUSED)
4446	{
4447	return !has_object_kept_pack(to_pack.repo, &obj->oid, KEPT_PACK_IN_CORE);
4448	}
4449
4450	static int cruft_include_check(struct commit commit, void data)
4451	{
4452	return cruft_include_check_obj((struct object*)commit, data);
4453	}
4454
4455	static void set_cruft_mtime(const struct object *object,
4456	struct packed_git *pack,
4457	off_t offset, time_t mtime)
4458	{
4459	add_cruft_object_entry(&object->oid, object->type, pack, offset, NULL,
4460	mtime);
4461	}
4462
4463	static void mark_pack_kept_in_core(struct string_list *packs, unsigned keep)
4464	{
4465	struct string_list_item *item = NULL;
4466	for_each_string_list_item(item, packs) {
4467	struct packed_git *p = item->util;
4468	if (!p)
4469	die(_("could not find pack '%s'"), item->string);
4470	if (p->is_cruft && keep)
4471	ignore_packed_keep_in_core_has_cruft = 1;
4472	p->pack_keep_in_core = keep;
4473	}
4474	}
4475
4476	static void add_objects_in_unpacked_packs(void);
4477
4478	static void enumerate_cruft_objects(void)
4479	{
4480	if (progress)
4481	progress_state = start_progress(the_repository,
4482	_("Enumerating cruft objects"), 0);
4483
4484	add_objects_in_unpacked_packs();
4485	add_unreachable_loose_objects(NULL);
4486
4487	stop_progress(&progress_state);
4488	}
4489
4490	static void enumerate_and_traverse_cruft_objects(struct string_list *fresh_packs)
4491	{
4492	struct packed_git *p;
4493	struct rev_info revs;
4494	int ret;
4495
4496	repo_init_revisions(the_repository, &revs, NULL);
4497
4498	revs.tag_objects = 1;
4499	revs.tree_objects = 1;
4500	revs.blob_objects = 1;
4501
4502	revs.include_check = cruft_include_check;
4503	revs.include_check_obj = cruft_include_check_obj;
4504
4505	revs.ignore_missing_links = 1;
4506
4507	if (progress)
4508	progress_state = start_progress(the_repository,
4509	_("Enumerating cruft objects"), 0);
4510	ret = add_unseen_recent_objects_to_traversal(&revs, cruft_expiration,
4511	set_cruft_mtime, 1);
4512	stop_progress(&progress_state);
4513
4514	if (ret)
4515	die(_("unable to add cruft objects"));
4516
4517	/*
4518	* Re-mark only the fresh packs as kept so that objects in
4519	* unknown packs do not halt the reachability traversal early.
4520	*/
4521	repo_for_each_pack(the_repository, p)
4522	p->pack_keep_in_core = 0;
4523	mark_pack_kept_in_core(fresh_packs, 1);
4524
4525	if (prepare_revision_walk(&revs))
4526	die(_("revision walk setup failed"));
4527	if (progress)
4528	progress_state = start_progress(the_repository,
4529	_("Traversing cruft objects"), 0);
4530	nr_seen = 0;
4531	traverse_commit_list(&revs, show_cruft_commit, show_cruft_object, NULL);
4532
4533	stop_progress(&progress_state);
4534	release_revisions(&revs);
4535	}
4536
4537	static void read_cruft_objects(void)
4538	{
4539	struct strbuf buf = STRBUF_INIT;
4540	struct string_list discard_packs = STRING_LIST_INIT_DUP;
4541	struct string_list fresh_packs = STRING_LIST_INIT_DUP;
4542	struct packed_git *p;
4543
4544	ignore_packed_keep_in_core = 1;
4545
4546	while (strbuf_getline(&buf, stdin) != EOF) {
4547	if (!buf.len)
4548	continue;
4549
4550	if (*buf.buf == '-')
4551	string_list_append(&discard_packs, buf.buf + 1);
4552	else
4553	string_list_append(&fresh_packs, buf.buf);
4554	}
4555
4556	string_list_sort(&discard_packs);
4557	string_list_sort(&fresh_packs);
4558
4559	repo_for_each_pack(the_repository, p) {
4560	const char *pack_name = pack_basename(p);
4561	struct string_list_item *item;
4562
4563	item = string_list_lookup(&fresh_packs, pack_name);
4564	if (!item)
4565	item = string_list_lookup(&discard_packs, pack_name);
4566
4567	if (item) {
4568	item->util = p;
4569	} else {
4570	/*
4571	* This pack wasn't mentioned in either the "fresh" or
4572	* "discard" list, so the caller didn't know about it.
4573	*
4574	* Mark it as kept so that its objects are ignored by
4575	* add_unseen_recent_objects_to_traversal(). We'll
4576	* unmark it before starting the traversal so it doesn't
4577	* halt the traversal early.
4578	*/
4579	p->pack_keep_in_core = 1;
4580	}
4581	}
4582
4583	mark_pack_kept_in_core(&fresh_packs, 1);
4584	mark_pack_kept_in_core(&discard_packs, 0);
4585
4586	if (cruft_expiration)
4587	enumerate_and_traverse_cruft_objects(&fresh_packs);
4588	else
4589	enumerate_cruft_objects();
4590
4591	strbuf_release(&buf);
4592	string_list_clear(&discard_packs, 0);
4593	string_list_clear(&fresh_packs, 0);
4594	}
4595
4596	static void read_object_list_from_stdin(void)
4597	{
4598	char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
4599	struct object_id oid;
4600	const char *p;
4601
4602	for (;;) {
4603	if (!fgets(line, sizeof(line), stdin)) {
4604	if (feof(stdin))
4605	break;
4606	if (!ferror(stdin))
4607	BUG("fgets returned NULL, not EOF, not error!");
4608	if (errno != EINTR)
4609	die_errno("fgets");
4610	clearerr(stdin);
4611	continue;
4612	}
4613	if (line[0] == '-') {
4614	if (get_oid_hex(line+1, &oid))
4615	die(_("expected edge object ID, got garbage:\n %s"),
4616	line);
4617	add_preferred_base(&oid);
4618	continue;
4619	}
4620	if (parse_oid_hex(line, &oid, &p))
4621	die(_("expected object ID, got garbage:\n %s"), line);
4622
4623	add_preferred_base_object(p + 1);
4624	add_object_entry(&oid, OBJ_NONE, p + 1, 0);
4625	}
4626	}
4627
4628	static void show_commit(struct commit commit, void data UNUSED)
4629	{
4630	add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
4631
4632	if (write_bitmap_index)
4633	index_commit_for_bitmap(commit);
4634
4635	if (use_delta_islands)
4636	propagate_island_marks(the_repository, commit);
4637	}
4638
4639	static void show_object(struct object obj, const char name,
4640	void *data UNUSED)
4641	{
4642	add_preferred_base_object(name);
4643	add_object_entry(&obj->oid, obj->type, name, 0);
4644
4645	if (use_delta_islands)
4646	record_tree_depth(&obj->oid, name);
4647	}
4648
4649	static void show_object__ma_allow_any(struct object obj, const char name, void *data)
4650	{
4651	assert(arg_missing_action == MA_ALLOW_ANY);
4652
4653	/*
4654	* Quietly ignore ALL missing objects. This avoids problems with
4655	* staging them now and getting an odd error later.
4656	*/
4657	if (!odb_has_object(the_repository->objects, &obj->oid, 0))
4658	return;
4659
4660	show_object(obj, name, data);
4661	}
4662
4663	static void show_object__ma_allow_promisor(struct object obj, const char name, void *data)
4664	{
4665	assert(arg_missing_action == MA_ALLOW_PROMISOR);
4666
4667	/*
4668	* Quietly ignore EXPECTED missing objects. This avoids problems with
4669	* staging them now and getting an odd error later.
4670	*/
4671	if (!odb_has_object(the_repository->objects, &obj->oid, 0) &&
4672	is_promisor_object(to_pack.repo, &obj->oid))
4673	return;
4674
4675	show_object(obj, name, data);
4676	}
4677
4678	static int option_parse_missing_action(const struct option *opt UNUSED,
4679	const char *arg, int unset)
4680	{
4681	assert(arg);
4682	assert(!unset);
4683
4684	if (!strcmp(arg, "error")) {
4685	arg_missing_action = MA_ERROR;
4686	fn_show_object = show_object;
4687	return 0;
4688	}
4689
4690	if (!strcmp(arg, "allow-any")) {
4691	arg_missing_action = MA_ALLOW_ANY;
4692	fetch_if_missing = 0;
4693	fn_show_object = show_object__ma_allow_any;
4694	return 0;
4695	}
4696
4697	if (!strcmp(arg, "allow-promisor")) {
4698	arg_missing_action = MA_ALLOW_PROMISOR;
4699	fetch_if_missing = 0;
4700	fn_show_object = show_object__ma_allow_promisor;
4701	return 0;
4702	}
4703
4704	die(_("invalid value for '%s': '%s'"), "--missing", arg);
4705	return 0;
4706	}
4707
4708	static void show_edge(struct commit *commit)
4709	{
4710	add_preferred_base(&commit->object.oid);
4711	}
4712
4713	static int add_object_in_unpacked_pack(const struct object_id *oid,
4714	struct object_info *oi,
4715	void *data UNUSED)
4716	{
4717	if (cruft) {
4718	add_cruft_object_entry(oid, OBJ_NONE, oi->source_infop->u.packed.pack,
4719	oi->source_infop->u.packed.offset, NULL,
4720	*oi->mtimep);
4721	} else {
4722	add_object_entry(oid, OBJ_NONE, "", 0);
4723	}
4724	return 0;
4725	}
4726
4727	static void add_objects_in_unpacked_packs(void)
4728	{
4729	struct odb_source *source;
4730	time_t mtime;
4731	struct odb_for_each_object_options opts = {
4732	.flags = ODB_FOR_EACH_OBJECT_PACK_ORDER \|
4733	ODB_FOR_EACH_OBJECT_LOCAL_ONLY \|
4734	ODB_FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS \|
4735	ODB_FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS,
4736	};
4737	struct odb_source_info source_info;
4738	struct object_info oi = {
4739	.mtimep = &mtime,
4740	.source_infop = &source_info,
4741	};
4742
4743	odb_prepare_alternates(to_pack.repo->objects);
4744	for (source = to_pack.repo->objects->sources; source; source = source->next) {
4745	struct odb_source_files *files = odb_source_files_downcast(source);
4746
4747	if (!source->local)
4748	continue;
4749
4750	if (odb_source_for_each_object(&files->packed->base, &oi,
4751	add_object_in_unpacked_pack, NULL, &opts))
4752	die(_("cannot open pack index"));
4753	}
4754	}
4755
4756	static int add_loose_object(const struct object_id oid, const char path,
4757	void *data)
4758	{
4759	struct rev_info *revs = data;
4760	enum object_type type = odb_read_object_info(the_repository->objects, oid, NULL);
4761
4762	if (type < 0) {
4763	warning(_("loose object at %s could not be examined"), path);
4764	return 0;
4765	}
4766
4767	if (cruft) {
4768	struct stat st;
4769	if (stat(path, &st) < 0) {
4770	if (errno == ENOENT)
4771	return 0;
4772	return error_errno("unable to stat %s", oid_to_hex(oid));
4773	}
4774
4775	add_cruft_object_entry(oid, type, NULL, 0, NULL,
4776	st.st_mtime);
4777	} else {
4778	add_object_entry(oid, type, "", 0);
4779	}
4780
4781	if (revs && type == OBJ_COMMIT)
4782	add_pending_oid(revs, NULL, oid, 0);
4783
4784	return 0;
4785	}
4786
4787	/*
4788	* We actually don't even have to worry about reachability here.
4789	* add_object_entry will weed out duplicates, so we just add every
4790	* loose object we find.
4791	*/
4792	static void add_unreachable_loose_objects(struct rev_info *revs)
4793	{
4794	for_each_loose_file_in_source(the_repository->objects->sources,
4795	add_loose_object, NULL, NULL, revs);
4796	}
4797
4798	static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
4799	{
4800	static struct packed_git *last_found = NULL;
4801	struct packed_git *p;
4802
4803	if (last_found && find_pack_entry_one(oid, last_found))
4804	return 1;
4805
4806	repo_for_each_pack(the_repository, p) {
4807	/*
4808	* We have already checked `last_found`, so there is no need to
4809	* re-check here.
4810	*/
4811	if (p == last_found)
4812	continue;
4813
4814	if ((!p->pack_local \|\| p->pack_keep \|\| p->pack_keep_in_core) &&
4815	find_pack_entry_one(oid, p)) {
4816	last_found = p;
4817	return 1;
4818	}
4819	}
4820
4821	return 0;
4822	}
4823
4824	/*
4825	* Store a list of sha1s that are should not be discarded
4826	* because they are either written too recently, or are
4827	* reachable from another object that was.
4828	*
4829	* This is filled by get_object_list.
4830	*/
4831	static struct oid_array recent_objects;
4832
4833	static int loosened_object_can_be_discarded(const struct object_id *oid,
4834	timestamp_t mtime)
4835	{
4836	if (!unpack_unreachable_expiration)
4837	return 0;
4838	if (mtime > unpack_unreachable_expiration)
4839	return 0;
4840	if (oid_array_lookup(&recent_objects, oid) >= 0)
4841	return 0;
4842	return 1;
4843	}
4844
4845	static int force_object_loose(struct odb_source *source,
4846	const struct object_id *oid,
4847	const time_t *mtime)
4848	{
4849	struct odb_source_files *files = odb_source_files_downcast(source);
4850	const struct git_hash_algo *compat = source->odb->repo->compat_hash_algo;
4851	struct object_info oi = OBJECT_INFO_INIT;
4852	struct object_id compat_oid, *compat_oid_p = NULL;
4853	enum object_type type;
4854	void *buf = NULL;
4855	size_t len;
4856	int ret;
4857
4858	for (struct odb_source *s = source->odb->sources; s; s = s->next) {
4859	struct odb_source_files *files = odb_source_files_downcast(s);
4860	if (!odb_source_read_object_info(&files->loose->base, oid, NULL, 0))
4861	return 0;
4862	}
4863
4864	oi.typep = &type;
4865	oi.sizep = &len;
4866	oi.contentp = &buf;
4867	if (odb_read_object_info_extended(source->odb, oid, &oi, 0)) {
4868	ret = error(_("cannot read object for %s"), oid_to_hex(oid));
4869	goto out;
4870	}
4871
4872	if (compat) {
4873	if (repo_oid_to_algop(source->odb->repo, oid, compat, &compat_oid)) {
4874	ret = error(_("cannot map object %s to %s"),
4875	oid_to_hex(oid), compat->name);
4876	goto out;
4877	}
4878
4879	compat_oid_p = &compat_oid;
4880	}
4881
4882	ret = odb_source_write_object(&files->loose->base, buf, len, type, oid,
4883	compat_oid_p, mtime, 0);
4884
4885	out:
4886	free(buf);
4887	return ret;
4888	}
4889
4890	static void loosen_unused_packed_objects(void)
4891	{
4892	struct packed_git *p;
4893	uint32_t i;
4894	uint32_t loosened_objects_nr = 0;
4895	struct object_id oid;
4896
4897	repo_for_each_pack(the_repository, p) {
4898	if (!p->pack_local \|\| p->pack_keep \|\| p->pack_keep_in_core)
4899	continue;
4900
4901	if (open_pack_index(p))
4902	die(_("cannot open pack index"));
4903
4904	for (i = 0; i < p->num_objects; i++) {
4905	nth_packed_object_id(&oid, p, i);
4906	if (!packlist_find(&to_pack, &oid) &&
4907	!has_sha1_pack_kept_or_nonlocal(&oid) &&
4908	!loosened_object_can_be_discarded(&oid, p->mtime)) {
4909	if (force_object_loose(the_repository->objects->sources,
4910	&oid, &p->mtime))
4911	die(_("unable to force loose object"));
4912	loosened_objects_nr++;
4913	}
4914	}
4915	}
4916
4917	trace2_data_intmax("pack-objects", the_repository,
4918	"loosen_unused_packed_objects/loosened", loosened_objects_nr);
4919	}
4920
4921	/*
4922	* This tracks any options which pack-reuse code expects to be on, or which a
4923	* reader of the pack might not understand, and which would therefore prevent
4924	* blind reuse of what we have on disk.
4925	*/
4926	static int pack_options_allow_reuse(void)
4927	{
4928	return allow_pack_reuse != NO_PACK_REUSE &&
4929	pack_to_stdout &&
4930	!ignore_packed_keep_on_disk &&
4931	!ignore_packed_keep_in_core &&
4932	(!local \|\| !have_non_local_packs) &&
4933	!incremental && (allow_ref_delta \|\| allow_ofs_delta);
4934	}
4935
4936	static int get_object_list_from_bitmap(struct rev_info *revs)
4937	{
4938	if (!(bitmap_git = prepare_bitmap_walk(revs, 0)))
4939	return -1;
4940
4941	/*
4942	* For now, force the name-hash version to be 1 since that
4943	* is the version implied by the bitmap format. Later, the
4944	* format can include this version explicitly in its format,
4945	* allowing readers to know the version that was used during
4946	* the bitmap write.
4947	*/
4948	name_hash_version = 1;
4949
4950	if (pack_options_allow_reuse())
4951	reuse_partial_packfile_from_bitmap(bitmap_git,
4952	&reuse_packfiles,
4953	&reuse_packfiles_nr,
4954	&reuse_packfile_bitmap,
4955	allow_pack_reuse == MULTI_PACK_REUSE,
4956	allow_ref_delta);
4957
4958	if (reuse_packfiles) {
4959	reuse_packfile_objects = bitmap_popcount(reuse_packfile_bitmap);
4960	if (!reuse_packfile_objects)
4961	BUG("expected non-empty reuse bitmap");
4962
4963	nr_result += reuse_packfile_objects;
4964	nr_seen += reuse_packfile_objects;
4965	display_progress(progress_state, nr_seen);
4966	}
4967
4968	traverse_bitmap_commit_list(bitmap_git, revs,
4969	&add_object_entry_from_bitmap);
4970	return 0;
4971	}
4972
4973	static void record_recent_object(struct object *obj,
4974	const char *name UNUSED,
4975	void *data UNUSED)
4976	{
4977	oid_array_append(&recent_objects, &obj->oid);
4978	}
4979
4980	static void record_recent_commit(struct commit commit, void data UNUSED)
4981	{
4982	oid_array_append(&recent_objects, &commit->object.oid);
4983	}
4984
4985	static int mark_bitmap_preferred_tip(const struct reference ref, void data UNUSED)
4986	{
4987	const struct object_id *maybe_peeled = ref->oid;
4988	struct object_id peeled;
4989	struct object *object;
4990
4991	if (!reference_get_peeled_oid(the_repository, ref, &peeled))
4992	maybe_peeled = &peeled;
4993
4994	object = parse_object_or_die(the_repository, maybe_peeled, ref->name);
4995	if (object->type == OBJ_COMMIT)
4996	object->flags \|= NEEDS_BITMAP;
4997
4998	return 0;
4999	}
5000

Showing first 5,000 of 5,768 lines. View raw