5 #include "parse-options.h"
6 #include "run-command.h"
9 #include "string-list.h"
13 #include "prune-packed.h"
14 #include "object-store.h"
15 #include "promisor-remote.h"
19 static int delta_base_offset
= 1;
20 static int pack_kept_objects
= -1;
21 static int write_bitmaps
= -1;
22 static int use_delta_islands
;
23 static char *packdir
, *packtmp_name
, *packtmp
;
25 static const char *const git_repack_usage
[] = {
26 N_("git repack [<options>]"),
30 static const char incremental_bitmap_conflict_error
[] = N_(
31 "Incremental repacks are incompatible with bitmap indexes. Use\n"
32 "--no-write-bitmap-index or disable the pack.writebitmaps configuration."
36 static int repack_config(const char *var
, const char *value
, void *cb
)
38 if (!strcmp(var
, "repack.usedeltabaseoffset")) {
39 delta_base_offset
= git_config_bool(var
, value
);
42 if (!strcmp(var
, "repack.packkeptobjects")) {
43 pack_kept_objects
= git_config_bool(var
, value
);
46 if (!strcmp(var
, "repack.writebitmaps") ||
47 !strcmp(var
, "pack.writebitmaps")) {
48 write_bitmaps
= git_config_bool(var
, value
);
51 if (!strcmp(var
, "repack.usedeltaislands")) {
52 use_delta_islands
= git_config_bool(var
, value
);
55 return git_default_config(var
, value
, cb
);
59 * Remove temporary $GIT_OBJECT_DIRECTORY/pack/.tmp-$$-pack-* files.
61 static void remove_temporary_files(void)
63 struct strbuf buf
= STRBUF_INIT
;
64 size_t dirlen
, prefixlen
;
68 dir
= opendir(packdir
);
72 /* Point at the slash at the end of ".../objects/pack/" */
73 dirlen
= strlen(packdir
) + 1;
74 strbuf_addstr(&buf
, packtmp
);
75 /* Hold the length of ".tmp-%d-pack-" */
76 prefixlen
= buf
.len
- dirlen
;
78 while ((e
= readdir(dir
))) {
79 if (strncmp(e
->d_name
, buf
.buf
+ dirlen
, prefixlen
))
81 strbuf_setlen(&buf
, dirlen
);
82 strbuf_addstr(&buf
, e
->d_name
);
89 static void remove_pack_on_signal(int signo
)
91 remove_temporary_files();
97 * Adds all packs hex strings to the fname list, which do not
98 * have a corresponding .keep file. These packs are not to
99 * be kept if we are going to pack everything into one file.
101 static void get_non_kept_pack_filenames(struct string_list
*fname_list
,
102 const struct string_list
*extra_keep
)
108 if (!(dir
= opendir(packdir
)))
111 while ((e
= readdir(dir
)) != NULL
) {
115 for (i
= 0; i
< extra_keep
->nr
; i
++)
116 if (!fspathcmp(e
->d_name
, extra_keep
->items
[i
].string
))
118 if (extra_keep
->nr
> 0 && i
< extra_keep
->nr
)
121 if (!strip_suffix(e
->d_name
, ".pack", &len
))
124 fname
= xmemdupz(e
->d_name
, len
);
126 if (!file_exists(mkpath("%s/%s.keep", packdir
, fname
)))
127 string_list_append_nodup(fname_list
, fname
);
134 static void remove_redundant_pack(const char *dir_name
, const char *base_name
)
136 struct strbuf buf
= STRBUF_INIT
;
137 struct multi_pack_index
*m
= get_local_multi_pack_index(the_repository
);
138 strbuf_addf(&buf
, "%s.pack", base_name
);
139 if (m
&& midx_contains_pack(m
, buf
.buf
))
140 clear_midx_file(the_repository
);
141 strbuf_insertf(&buf
, 0, "%s/", dir_name
);
142 unlink_pack_path(buf
.buf
, 1);
143 strbuf_release(&buf
);
146 struct pack_objects_args
{
148 const char *window_memory
;
151 const char *max_pack_size
;
158 static void prepare_pack_objects(struct child_process
*cmd
,
159 const struct pack_objects_args
*args
)
161 strvec_push(&cmd
->args
, "pack-objects");
163 strvec_pushf(&cmd
->args
, "--window=%s", args
->window
);
164 if (args
->window_memory
)
165 strvec_pushf(&cmd
->args
, "--window-memory=%s", args
->window_memory
);
167 strvec_pushf(&cmd
->args
, "--depth=%s", args
->depth
);
169 strvec_pushf(&cmd
->args
, "--threads=%s", args
->threads
);
170 if (args
->max_pack_size
)
171 strvec_pushf(&cmd
->args
, "--max-pack-size=%s", args
->max_pack_size
);
172 if (args
->no_reuse_delta
)
173 strvec_pushf(&cmd
->args
, "--no-reuse-delta");
174 if (args
->no_reuse_object
)
175 strvec_pushf(&cmd
->args
, "--no-reuse-object");
177 strvec_push(&cmd
->args
, "--local");
179 strvec_push(&cmd
->args
, "--quiet");
180 if (delta_base_offset
)
181 strvec_push(&cmd
->args
, "--delta-base-offset");
182 strvec_push(&cmd
->args
, packtmp
);
188 * Write oid to the given struct child_process's stdin, starting it first if
191 static int write_oid(const struct object_id
*oid
, struct packed_git
*pack
,
192 uint32_t pos
, void *data
)
194 struct child_process
*cmd
= data
;
197 if (start_command(cmd
))
198 die(_("could not start pack-objects to repack promisor objects"));
201 xwrite(cmd
->in
, oid_to_hex(oid
), the_hash_algo
->hexsz
);
202 xwrite(cmd
->in
, "\n", 1);
217 static unsigned populate_pack_exts(char *name
)
220 struct strbuf path
= STRBUF_INIT
;
224 for (i
= 0; i
< ARRAY_SIZE(exts
); i
++) {
226 strbuf_addf(&path
, "%s-%s%s", packtmp
, name
, exts
[i
].name
);
228 if (stat(path
.buf
, &statbuf
))
234 strbuf_release(&path
);
238 static void repack_promisor_objects(const struct pack_objects_args
*args
,
239 struct string_list
*names
)
241 struct child_process cmd
= CHILD_PROCESS_INIT
;
243 struct strbuf line
= STRBUF_INIT
;
245 prepare_pack_objects(&cmd
, args
);
249 * NEEDSWORK: Giving pack-objects only the OIDs without any ordering
250 * hints may result in suboptimal deltas in the resulting pack. See if
251 * the OIDs can be sent with fake paths such that pack-objects can use a
252 * {type -> existing pack order} ordering when computing deltas instead
253 * of a {type -> size} ordering, which may produce better deltas.
255 for_each_packed_object(write_oid
, &cmd
,
256 FOR_EACH_OBJECT_PROMISOR_ONLY
);
259 /* No packed objects; cmd was never started */
264 out
= xfdopen(cmd
.out
, "r");
265 while (strbuf_getline_lf(&line
, out
) != EOF
) {
266 struct string_list_item
*item
;
269 if (line
.len
!= the_hash_algo
->hexsz
)
270 die(_("repack: Expecting full hex object ID lines only from pack-objects."));
271 item
= string_list_append(names
, line
.buf
);
274 * pack-objects creates the .pack and .idx files, but not the
275 * .promisor file. Create the .promisor file, which is empty.
277 * NEEDSWORK: fetch-pack sometimes generates non-empty
278 * .promisor files containing the ref names and associated
279 * hashes at the point of generation of the corresponding
280 * packfile, but this would not preserve their contents. Maybe
281 * concatenate the contents of all .promisor files instead of
282 * just creating a new empty file.
284 promisor_name
= mkpathdup("%s-%s.promisor", packtmp
,
286 write_promisor_file(promisor_name
, NULL
, 0);
288 item
->util
= (void *)(uintptr_t)populate_pack_exts(item
->string
);
293 if (finish_command(&cmd
))
294 die(_("could not finish pack-objects to repack promisor objects"));
297 #define ALL_INTO_ONE 1
298 #define LOOSEN_UNREACHABLE 2
300 struct pack_geometry
{
301 struct packed_git
**pack
;
302 uint32_t pack_nr
, pack_alloc
;
306 static uint32_t geometry_pack_weight(struct packed_git
*p
)
308 if (open_pack_index(p
))
309 die(_("cannot open index for %s"), p
->pack_name
);
310 return p
->num_objects
;
313 static int geometry_cmp(const void *va
, const void *vb
)
315 uint32_t aw
= geometry_pack_weight(*(struct packed_git
**)va
),
316 bw
= geometry_pack_weight(*(struct packed_git
**)vb
);
325 static void init_pack_geometry(struct pack_geometry
**geometry_p
)
327 struct packed_git
*p
;
328 struct pack_geometry
*geometry
;
330 *geometry_p
= xcalloc(1, sizeof(struct pack_geometry
));
331 geometry
= *geometry_p
;
333 for (p
= get_all_packs(the_repository
); p
; p
= p
->next
) {
334 if (!pack_kept_objects
&& p
->pack_keep
)
337 ALLOC_GROW(geometry
->pack
,
338 geometry
->pack_nr
+ 1,
339 geometry
->pack_alloc
);
341 geometry
->pack
[geometry
->pack_nr
] = p
;
345 QSORT(geometry
->pack
, geometry
->pack_nr
, geometry_cmp
);
348 static void split_pack_geometry(struct pack_geometry
*geometry
, int factor
)
352 off_t total_size
= 0;
354 if (!geometry
->pack_nr
) {
355 geometry
->split
= geometry
->pack_nr
;
360 * First, count the number of packs (in descending order of size) which
361 * already form a geometric progression.
363 for (i
= geometry
->pack_nr
- 1; i
> 0; i
--) {
364 struct packed_git
*ours
= geometry
->pack
[i
];
365 struct packed_git
*prev
= geometry
->pack
[i
- 1];
367 if (unsigned_mult_overflows(factor
, geometry_pack_weight(prev
)))
368 die(_("pack %s too large to consider in geometric "
372 if (geometry_pack_weight(ours
) < factor
* geometry_pack_weight(prev
))
380 * Move the split one to the right, since the top element in the
381 * last-compared pair can't be in the progression. Only do this
382 * when we split in the middle of the array (otherwise if we got
383 * to the end, then the split is in the right place).
389 * Then, anything to the left of 'split' must be in a new pack. But,
390 * creating that new pack may cause packs in the heavy half to no longer
391 * form a geometric progression.
393 * Compute an expected size of the new pack, and then determine how many
394 * packs in the heavy half need to be joined into it (if any) to restore
395 * the geometric progression.
397 for (i
= 0; i
< split
; i
++) {
398 struct packed_git
*p
= geometry
->pack
[i
];
400 if (unsigned_add_overflows(total_size
, geometry_pack_weight(p
)))
401 die(_("pack %s too large to roll up"), p
->pack_name
);
402 total_size
+= geometry_pack_weight(p
);
404 for (i
= split
; i
< geometry
->pack_nr
; i
++) {
405 struct packed_git
*ours
= geometry
->pack
[i
];
407 if (unsigned_mult_overflows(factor
, total_size
))
408 die(_("pack %s too large to roll up"), ours
->pack_name
);
410 if (geometry_pack_weight(ours
) < factor
* total_size
) {
411 if (unsigned_add_overflows(total_size
,
412 geometry_pack_weight(ours
)))
413 die(_("pack %s too large to roll up"),
417 total_size
+= geometry_pack_weight(ours
);
422 geometry
->split
= split
;
425 static void clear_pack_geometry(struct pack_geometry
*geometry
)
430 free(geometry
->pack
);
431 geometry
->pack_nr
= 0;
432 geometry
->pack_alloc
= 0;
436 int cmd_repack(int argc
, const char **argv
, const char *prefix
)
438 struct child_process cmd
= CHILD_PROCESS_INIT
;
439 struct string_list_item
*item
;
440 struct string_list names
= STRING_LIST_INIT_DUP
;
441 struct string_list rollback
= STRING_LIST_INIT_NODUP
;
442 struct string_list existing_packs
= STRING_LIST_INIT_DUP
;
443 struct pack_geometry
*geometry
= NULL
;
444 struct strbuf line
= STRBUF_INIT
;
448 /* variables to be filled by option parsing */
449 int pack_everything
= 0;
450 int delete_redundant
= 0;
451 const char *unpack_unreachable
= NULL
;
452 int keep_unreachable
= 0;
453 struct string_list keep_pack_list
= STRING_LIST_INIT_NODUP
;
454 int no_update_server_info
= 0;
455 struct pack_objects_args po_args
= {NULL
};
456 int geometric_factor
= 0;
458 struct option builtin_repack_options
[] = {
459 OPT_BIT('a', NULL
, &pack_everything
,
460 N_("pack everything in a single pack"), ALL_INTO_ONE
),
461 OPT_BIT('A', NULL
, &pack_everything
,
462 N_("same as -a, and turn unreachable objects loose"),
463 LOOSEN_UNREACHABLE
| ALL_INTO_ONE
),
464 OPT_BOOL('d', NULL
, &delete_redundant
,
465 N_("remove redundant packs, and run git-prune-packed")),
466 OPT_BOOL('f', NULL
, &po_args
.no_reuse_delta
,
467 N_("pass --no-reuse-delta to git-pack-objects")),
468 OPT_BOOL('F', NULL
, &po_args
.no_reuse_object
,
469 N_("pass --no-reuse-object to git-pack-objects")),
470 OPT_BOOL('n', NULL
, &no_update_server_info
,
471 N_("do not run git-update-server-info")),
472 OPT__QUIET(&po_args
.quiet
, N_("be quiet")),
473 OPT_BOOL('l', "local", &po_args
.local
,
474 N_("pass --local to git-pack-objects")),
475 OPT_BOOL('b', "write-bitmap-index", &write_bitmaps
,
476 N_("write bitmap index")),
477 OPT_BOOL('i', "delta-islands", &use_delta_islands
,
478 N_("pass --delta-islands to git-pack-objects")),
479 OPT_STRING(0, "unpack-unreachable", &unpack_unreachable
, N_("approxidate"),
480 N_("with -A, do not loosen objects older than this")),
481 OPT_BOOL('k', "keep-unreachable", &keep_unreachable
,
482 N_("with -a, repack unreachable objects")),
483 OPT_STRING(0, "window", &po_args
.window
, N_("n"),
484 N_("size of the window used for delta compression")),
485 OPT_STRING(0, "window-memory", &po_args
.window_memory
, N_("bytes"),
486 N_("same as the above, but limit memory size instead of entries count")),
487 OPT_STRING(0, "depth", &po_args
.depth
, N_("n"),
488 N_("limits the maximum delta depth")),
489 OPT_STRING(0, "threads", &po_args
.threads
, N_("n"),
490 N_("limits the maximum number of threads")),
491 OPT_STRING(0, "max-pack-size", &po_args
.max_pack_size
, N_("bytes"),
492 N_("maximum size of each packfile")),
493 OPT_BOOL(0, "pack-kept-objects", &pack_kept_objects
,
494 N_("repack objects in packs marked with .keep")),
495 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list
, N_("name"),
496 N_("do not repack this pack")),
497 OPT_INTEGER('g', "geometric", &geometric_factor
,
498 N_("find a geometric progression with factor <N>")),
502 git_config(repack_config
, NULL
);
504 argc
= parse_options(argc
, argv
, prefix
, builtin_repack_options
,
505 git_repack_usage
, 0);
507 if (delete_redundant
&& repository_format_precious_objects
)
508 die(_("cannot delete packs in a precious-objects repo"));
510 if (keep_unreachable
&&
511 (unpack_unreachable
|| (pack_everything
& LOOSEN_UNREACHABLE
)))
512 die(_("--keep-unreachable and -A are incompatible"));
514 if (write_bitmaps
< 0) {
515 if (!(pack_everything
& ALL_INTO_ONE
) ||
516 !is_bare_repository())
519 if (pack_kept_objects
< 0)
520 pack_kept_objects
= write_bitmaps
> 0;
522 if (write_bitmaps
&& !(pack_everything
& ALL_INTO_ONE
))
523 die(_(incremental_bitmap_conflict_error
));
525 if (geometric_factor
) {
527 die(_("--geometric is incompatible with -A, -a"));
528 init_pack_geometry(&geometry
);
529 split_pack_geometry(geometry
, geometric_factor
);
532 packdir
= mkpathdup("%s/pack", get_object_directory());
533 packtmp_name
= xstrfmt(".tmp-%d-pack", (int)getpid());
534 packtmp
= mkpathdup("%s/%s", packdir
, packtmp_name
);
536 sigchain_push_common(remove_pack_on_signal
);
538 prepare_pack_objects(&cmd
, &po_args
);
540 strvec_push(&cmd
.args
, "--keep-true-parents");
541 if (!pack_kept_objects
)
542 strvec_push(&cmd
.args
, "--honor-pack-keep");
543 for (i
= 0; i
< keep_pack_list
.nr
; i
++)
544 strvec_pushf(&cmd
.args
, "--keep-pack=%s",
545 keep_pack_list
.items
[i
].string
);
546 strvec_push(&cmd
.args
, "--non-empty");
549 * We need to grab all reachable objects, including those that
550 * are reachable from reflogs and the index.
552 * When repacking into a geometric progression of packs,
553 * however, we ask 'git pack-objects --stdin-packs', and it is
554 * not about packing objects based on reachability but about
555 * repacking all the objects in specified packs and loose ones
556 * (indeed, --stdin-packs is incompatible with these options).
558 strvec_push(&cmd
.args
, "--all");
559 strvec_push(&cmd
.args
, "--reflog");
560 strvec_push(&cmd
.args
, "--indexed-objects");
562 if (has_promisor_remote())
563 strvec_push(&cmd
.args
, "--exclude-promisor-objects");
564 if (write_bitmaps
> 0)
565 strvec_push(&cmd
.args
, "--write-bitmap-index");
566 else if (write_bitmaps
< 0)
567 strvec_push(&cmd
.args
, "--write-bitmap-index-quiet");
568 if (use_delta_islands
)
569 strvec_push(&cmd
.args
, "--delta-islands");
571 if (pack_everything
& ALL_INTO_ONE
) {
572 get_non_kept_pack_filenames(&existing_packs
, &keep_pack_list
);
574 repack_promisor_objects(&po_args
, &names
);
576 if (existing_packs
.nr
&& delete_redundant
) {
577 for_each_string_list_item(item
, &names
) {
578 strvec_pushf(&cmd
.args
, "--keep-pack=%s-%s.pack",
579 packtmp_name
, item
->string
);
581 if (unpack_unreachable
) {
582 strvec_pushf(&cmd
.args
,
583 "--unpack-unreachable=%s",
585 strvec_push(&cmd
.env_array
, "GIT_REF_PARANOIA=1");
586 } else if (pack_everything
& LOOSEN_UNREACHABLE
) {
587 strvec_push(&cmd
.args
,
588 "--unpack-unreachable");
589 } else if (keep_unreachable
) {
590 strvec_push(&cmd
.args
, "--keep-unreachable");
591 strvec_push(&cmd
.args
, "--pack-loose-unreachable");
593 strvec_push(&cmd
.env_array
, "GIT_REF_PARANOIA=1");
596 } else if (geometry
) {
597 strvec_push(&cmd
.args
, "--stdin-packs");
598 strvec_push(&cmd
.args
, "--unpacked");
600 strvec_push(&cmd
.args
, "--unpacked");
601 strvec_push(&cmd
.args
, "--incremental");
609 ret
= start_command(&cmd
);
614 FILE *in
= xfdopen(cmd
.in
, "w");
616 * The resulting pack should contain all objects in packs that
617 * are going to be rolled up, but exclude objects in packs which
618 * are being left alone.
620 for (i
= 0; i
< geometry
->split
; i
++)
621 fprintf(in
, "%s\n", pack_basename(geometry
->pack
[i
]));
622 for (i
= geometry
->split
; i
< geometry
->pack_nr
; i
++)
623 fprintf(in
, "^%s\n", pack_basename(geometry
->pack
[i
]));
627 out
= xfdopen(cmd
.out
, "r");
628 while (strbuf_getline_lf(&line
, out
) != EOF
) {
629 if (line
.len
!= the_hash_algo
->hexsz
)
630 die(_("repack: Expecting full hex object ID lines only from pack-objects."));
631 string_list_append(&names
, line
.buf
);
634 ret
= finish_command(&cmd
);
638 if (!names
.nr
&& !po_args
.quiet
)
639 printf_ln(_("Nothing new to pack."));
641 for_each_string_list_item(item
, &names
) {
642 item
->util
= (void *)(uintptr_t)populate_pack_exts(item
->string
);
645 close_object_store(the_repository
->objects
);
648 * Ok we have prepared all new packfiles.
650 for_each_string_list_item(item
, &names
) {
651 for (ext
= 0; ext
< ARRAY_SIZE(exts
); ext
++) {
652 char *fname
, *fname_old
;
654 fname
= mkpathdup("%s/pack-%s%s",
655 packdir
, item
->string
, exts
[ext
].name
);
656 fname_old
= mkpathdup("%s-%s%s",
657 packtmp
, item
->string
, exts
[ext
].name
);
659 if (((uintptr_t)item
->util
) & (1 << ext
)) {
660 struct stat statbuffer
;
661 if (!stat(fname_old
, &statbuffer
)) {
662 statbuffer
.st_mode
&= ~(S_IWUSR
| S_IWGRP
| S_IWOTH
);
663 chmod(fname_old
, statbuffer
.st_mode
);
666 if (rename(fname_old
, fname
))
667 die_errno(_("renaming '%s' failed"), fname_old
);
668 } else if (!exts
[ext
].optional
)
669 die(_("missing required file: %s"), fname_old
);
670 else if (unlink(fname
) < 0 && errno
!= ENOENT
)
671 die_errno(_("could not unlink: %s"), fname
);
677 /* End of pack replacement. */
679 reprepare_packed_git(the_repository
);
681 if (delete_redundant
) {
682 const int hexsz
= the_hash_algo
->hexsz
;
684 string_list_sort(&names
);
685 for_each_string_list_item(item
, &existing_packs
) {
687 size_t len
= strlen(item
->string
);
690 sha1
= item
->string
+ len
- hexsz
;
691 if (!string_list_has_string(&names
, sha1
))
692 remove_redundant_pack(packdir
, item
->string
);
696 struct strbuf buf
= STRBUF_INIT
;
699 for (i
= 0; i
< geometry
->split
; i
++) {
700 struct packed_git
*p
= geometry
->pack
[i
];
701 if (string_list_has_string(&names
,
702 hash_to_hex(p
->hash
)))
706 strbuf_addstr(&buf
, pack_basename(p
));
707 strbuf_strip_suffix(&buf
, ".pack");
709 remove_redundant_pack(packdir
, buf
.buf
);
711 strbuf_release(&buf
);
713 if (!po_args
.quiet
&& isatty(2))
714 opts
|= PRUNE_PACKED_VERBOSE
;
715 prune_packed_objects(opts
);
717 if (!keep_unreachable
&&
718 (!(pack_everything
& LOOSEN_UNREACHABLE
) ||
719 unpack_unreachable
) &&
720 is_repository_shallow(the_repository
))
721 prune_shallow(PRUNE_QUICK
);
724 if (!no_update_server_info
)
725 update_server_info(0);
726 remove_temporary_files();
728 if (git_env_bool(GIT_TEST_MULTI_PACK_INDEX
, 0))
729 write_midx_file(get_object_directory(), NULL
, 0);
731 string_list_clear(&names
, 0);
732 string_list_clear(&rollback
, 0);
733 string_list_clear(&existing_packs
, 0);
734 clear_pack_geometry(geometry
);
735 strbuf_release(&line
);