merge: simplify parsing of "-n" option
[git.git] / builtin / fast-import.c
blob4dbb10aff3dacf58b0ecb5daf648cefb16d41f4b
1 #include "builtin.h"
2 #include "abspath.h"
3 #include "environment.h"
4 #include "gettext.h"
5 #include "hex.h"
6 #include "repository.h"
7 #include "config.h"
8 #include "lockfile.h"
9 #include "object.h"
10 #include "blob.h"
11 #include "tree.h"
12 #include "commit.h"
13 #include "delta.h"
14 #include "pack.h"
15 #include "path.h"
16 #include "refs.h"
17 #include "csum-file.h"
18 #include "quote.h"
19 #include "dir.h"
20 #include "run-command.h"
21 #include "packfile.h"
22 #include "object-file.h"
23 #include "object-name.h"
24 #include "object-store-ll.h"
25 #include "mem-pool.h"
26 #include "commit-reach.h"
27 #include "khash.h"
28 #include "date.h"
30 #define PACK_ID_BITS 16
31 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
32 #define DEPTH_BITS 13
33 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
36 * We abuse the setuid bit on directories to mean "do not delta".
38 #define NO_DELTA S_ISUID
41 * The amount of additional space required in order to write an object into the
42 * current pack. This is the hash lengths at the end of the pack, plus the
43 * length of one object ID.
45 #define PACK_SIZE_THRESHOLD (the_hash_algo->rawsz * 3)
47 struct object_entry {
48 struct pack_idx_entry idx;
49 struct hashmap_entry ent;
50 uint32_t type : TYPE_BITS,
51 pack_id : PACK_ID_BITS,
52 depth : DEPTH_BITS;
55 static int object_entry_hashcmp(const void *map_data UNUSED,
56 const struct hashmap_entry *eptr,
57 const struct hashmap_entry *entry_or_key,
58 const void *keydata)
60 const struct object_id *oid = keydata;
61 const struct object_entry *e1, *e2;
63 e1 = container_of(eptr, const struct object_entry, ent);
64 if (oid)
65 return oidcmp(&e1->idx.oid, oid);
67 e2 = container_of(entry_or_key, const struct object_entry, ent);
68 return oidcmp(&e1->idx.oid, &e2->idx.oid);
71 struct object_entry_pool {
72 struct object_entry_pool *next_pool;
73 struct object_entry *next_free;
74 struct object_entry *end;
75 struct object_entry entries[FLEX_ARRAY]; /* more */
78 struct mark_set {
79 union {
80 struct object_id *oids[1024];
81 struct object_entry *marked[1024];
82 struct mark_set *sets[1024];
83 } data;
84 unsigned int shift;
87 struct last_object {
88 struct strbuf data;
89 off_t offset;
90 unsigned int depth;
91 unsigned no_swap : 1;
94 struct atom_str {
95 struct atom_str *next_atom;
96 unsigned short str_len;
97 char str_dat[FLEX_ARRAY]; /* more */
100 struct tree_content;
101 struct tree_entry {
102 struct tree_content *tree;
103 struct atom_str *name;
104 struct tree_entry_ms {
105 uint16_t mode;
106 struct object_id oid;
107 } versions[2];
110 struct tree_content {
111 unsigned int entry_capacity; /* must match avail_tree_content */
112 unsigned int entry_count;
113 unsigned int delta_depth;
114 struct tree_entry *entries[FLEX_ARRAY]; /* more */
117 struct avail_tree_content {
118 unsigned int entry_capacity; /* must match tree_content */
119 struct avail_tree_content *next_avail;
122 struct branch {
123 struct branch *table_next_branch;
124 struct branch *active_next_branch;
125 const char *name;
126 struct tree_entry branch_tree;
127 uintmax_t last_commit;
128 uintmax_t num_notes;
129 unsigned active : 1;
130 unsigned delete : 1;
131 unsigned pack_id : PACK_ID_BITS;
132 struct object_id oid;
135 struct tag {
136 struct tag *next_tag;
137 const char *name;
138 unsigned int pack_id;
139 struct object_id oid;
142 struct hash_list {
143 struct hash_list *next;
144 struct object_id oid;
147 typedef enum {
148 WHENSPEC_RAW = 1,
149 WHENSPEC_RAW_PERMISSIVE,
150 WHENSPEC_RFC2822,
151 WHENSPEC_NOW
152 } whenspec_type;
154 struct recent_command {
155 struct recent_command *prev;
156 struct recent_command *next;
157 char *buf;
160 typedef void (*mark_set_inserter_t)(struct mark_set **s, struct object_id *oid, uintmax_t mark);
161 typedef void (*each_mark_fn_t)(uintmax_t mark, void *obj, void *cbp);
163 /* Configured limits on output */
164 static unsigned long max_depth = 50;
165 static off_t max_packsize;
166 static int unpack_limit = 100;
167 static int force_update;
169 /* Stats and misc. counters */
170 static uintmax_t alloc_count;
171 static uintmax_t marks_set_count;
172 static uintmax_t object_count_by_type[1 << TYPE_BITS];
173 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
174 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
175 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
176 static unsigned long object_count;
177 static unsigned long branch_count;
178 static unsigned long branch_load_count;
179 static int failure;
180 static FILE *pack_edges;
181 static unsigned int show_stats = 1;
182 static int global_argc;
183 static const char **global_argv;
184 static const char *global_prefix;
186 /* Memory pools */
187 static struct mem_pool fi_mem_pool = {
188 .block_alloc = 2*1024*1024 - sizeof(struct mp_block),
191 /* Atom management */
192 static unsigned int atom_table_sz = 4451;
193 static unsigned int atom_cnt;
194 static struct atom_str **atom_table;
196 /* The .pack file being generated */
197 static struct pack_idx_option pack_idx_opts;
198 static unsigned int pack_id;
199 static struct hashfile *pack_file;
200 static struct packed_git *pack_data;
201 static struct packed_git **all_packs;
202 static off_t pack_size;
204 /* Table of objects we've written. */
205 static unsigned int object_entry_alloc = 5000;
206 static struct object_entry_pool *blocks;
207 static struct hashmap object_table;
208 static struct mark_set *marks;
209 static const char *export_marks_file;
210 static const char *import_marks_file;
211 static int import_marks_file_from_stream;
212 static int import_marks_file_ignore_missing;
213 static int import_marks_file_done;
214 static int relative_marks_paths;
216 /* Our last blob */
217 static struct last_object last_blob = {
218 .data = STRBUF_INIT,
221 /* Tree management */
222 static unsigned int tree_entry_alloc = 1000;
223 static void *avail_tree_entry;
224 static unsigned int avail_tree_table_sz = 100;
225 static struct avail_tree_content **avail_tree_table;
226 static size_t tree_entry_allocd;
227 static struct strbuf old_tree = STRBUF_INIT;
228 static struct strbuf new_tree = STRBUF_INIT;
230 /* Branch data */
231 static unsigned long max_active_branches = 5;
232 static unsigned long cur_active_branches;
233 static unsigned long branch_table_sz = 1039;
234 static struct branch **branch_table;
235 static struct branch *active_branches;
237 /* Tag data */
238 static struct tag *first_tag;
239 static struct tag *last_tag;
241 /* Input stream parsing */
242 static whenspec_type whenspec = WHENSPEC_RAW;
243 static struct strbuf command_buf = STRBUF_INIT;
244 static int unread_command_buf;
245 static struct recent_command cmd_hist = {
246 .prev = &cmd_hist,
247 .next = &cmd_hist,
249 static struct recent_command *cmd_tail = &cmd_hist;
250 static struct recent_command *rc_free;
251 static unsigned int cmd_save = 100;
252 static uintmax_t next_mark;
253 static struct strbuf new_data = STRBUF_INIT;
254 static int seen_data_command;
255 static int require_explicit_termination;
256 static int allow_unsafe_features;
258 /* Signal handling */
259 static volatile sig_atomic_t checkpoint_requested;
261 /* Submodule marks */
262 static struct string_list sub_marks_from = STRING_LIST_INIT_DUP;
263 static struct string_list sub_marks_to = STRING_LIST_INIT_DUP;
264 static kh_oid_map_t *sub_oid_map;
266 /* Where to write output of cat-blob commands */
267 static int cat_blob_fd = STDOUT_FILENO;
269 static void parse_argv(void);
270 static void parse_get_mark(const char *p);
271 static void parse_cat_blob(const char *p);
272 static void parse_ls(const char *p, struct branch *b);
274 static void for_each_mark(struct mark_set *m, uintmax_t base, each_mark_fn_t callback, void *p)
276 uintmax_t k;
277 if (m->shift) {
278 for (k = 0; k < 1024; k++) {
279 if (m->data.sets[k])
280 for_each_mark(m->data.sets[k], base + (k << m->shift), callback, p);
282 } else {
283 for (k = 0; k < 1024; k++) {
284 if (m->data.marked[k])
285 callback(base + k, m->data.marked[k], p);
290 static void dump_marks_fn(uintmax_t mark, void *object, void *cbp) {
291 struct object_entry *e = object;
292 FILE *f = cbp;
294 fprintf(f, ":%" PRIuMAX " %s\n", mark, oid_to_hex(&e->idx.oid));
297 static void write_branch_report(FILE *rpt, struct branch *b)
299 fprintf(rpt, "%s:\n", b->name);
301 fprintf(rpt, " status :");
302 if (b->active)
303 fputs(" active", rpt);
304 if (b->branch_tree.tree)
305 fputs(" loaded", rpt);
306 if (is_null_oid(&b->branch_tree.versions[1].oid))
307 fputs(" dirty", rpt);
308 fputc('\n', rpt);
310 fprintf(rpt, " tip commit : %s\n", oid_to_hex(&b->oid));
311 fprintf(rpt, " old tree : %s\n",
312 oid_to_hex(&b->branch_tree.versions[0].oid));
313 fprintf(rpt, " cur tree : %s\n",
314 oid_to_hex(&b->branch_tree.versions[1].oid));
315 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
317 fputs(" last pack : ", rpt);
318 if (b->pack_id < MAX_PACK_ID)
319 fprintf(rpt, "%u", b->pack_id);
320 fputc('\n', rpt);
322 fputc('\n', rpt);
325 static void write_crash_report(const char *err)
327 char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
328 FILE *rpt = fopen(loc, "w");
329 struct branch *b;
330 unsigned long lu;
331 struct recent_command *rc;
333 if (!rpt) {
334 error_errno("can't write crash report %s", loc);
335 free(loc);
336 return;
339 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
341 fprintf(rpt, "fast-import crash report:\n");
342 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
343 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
344 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
345 fputc('\n', rpt);
347 fputs("fatal: ", rpt);
348 fputs(err, rpt);
349 fputc('\n', rpt);
351 fputc('\n', rpt);
352 fputs("Most Recent Commands Before Crash\n", rpt);
353 fputs("---------------------------------\n", rpt);
354 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
355 if (rc->next == &cmd_hist)
356 fputs("* ", rpt);
357 else
358 fputs(" ", rpt);
359 fputs(rc->buf, rpt);
360 fputc('\n', rpt);
363 fputc('\n', rpt);
364 fputs("Active Branch LRU\n", rpt);
365 fputs("-----------------\n", rpt);
366 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
367 cur_active_branches,
368 max_active_branches);
369 fputc('\n', rpt);
370 fputs(" pos clock name\n", rpt);
371 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
372 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
373 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
374 ++lu, b->last_commit, b->name);
376 fputc('\n', rpt);
377 fputs("Inactive Branches\n", rpt);
378 fputs("-----------------\n", rpt);
379 for (lu = 0; lu < branch_table_sz; lu++) {
380 for (b = branch_table[lu]; b; b = b->table_next_branch)
381 write_branch_report(rpt, b);
384 if (first_tag) {
385 struct tag *tg;
386 fputc('\n', rpt);
387 fputs("Annotated Tags\n", rpt);
388 fputs("--------------\n", rpt);
389 for (tg = first_tag; tg; tg = tg->next_tag) {
390 fputs(oid_to_hex(&tg->oid), rpt);
391 fputc(' ', rpt);
392 fputs(tg->name, rpt);
393 fputc('\n', rpt);
397 fputc('\n', rpt);
398 fputs("Marks\n", rpt);
399 fputs("-----\n", rpt);
400 if (export_marks_file)
401 fprintf(rpt, " exported to %s\n", export_marks_file);
402 else
403 for_each_mark(marks, 0, dump_marks_fn, rpt);
405 fputc('\n', rpt);
406 fputs("-------------------\n", rpt);
407 fputs("END OF CRASH REPORT\n", rpt);
408 fclose(rpt);
409 free(loc);
412 static void end_packfile(void);
413 static void unkeep_all_packs(void);
414 static void dump_marks(void);
416 static NORETURN void die_nicely(const char *err, va_list params)
418 va_list cp;
419 static int zombie;
420 report_fn die_message_fn = get_die_message_routine();
422 va_copy(cp, params);
423 die_message_fn(err, params);
425 if (!zombie) {
426 char message[2 * PATH_MAX];
428 zombie = 1;
429 vsnprintf(message, sizeof(message), err, cp);
430 write_crash_report(message);
431 end_packfile();
432 unkeep_all_packs();
433 dump_marks();
435 exit(128);
438 #ifndef SIGUSR1 /* Windows, for example */
440 static void set_checkpoint_signal(void)
444 #else
446 static void checkpoint_signal(int signo UNUSED)
448 checkpoint_requested = 1;
451 static void set_checkpoint_signal(void)
453 struct sigaction sa;
455 memset(&sa, 0, sizeof(sa));
456 sa.sa_handler = checkpoint_signal;
457 sigemptyset(&sa.sa_mask);
458 sa.sa_flags = SA_RESTART;
459 sigaction(SIGUSR1, &sa, NULL);
462 #endif
464 static void alloc_objects(unsigned int cnt)
466 struct object_entry_pool *b;
468 b = xmalloc(sizeof(struct object_entry_pool)
469 + cnt * sizeof(struct object_entry));
470 b->next_pool = blocks;
471 b->next_free = b->entries;
472 b->end = b->entries + cnt;
473 blocks = b;
474 alloc_count += cnt;
477 static struct object_entry *new_object(struct object_id *oid)
479 struct object_entry *e;
481 if (blocks->next_free == blocks->end)
482 alloc_objects(object_entry_alloc);
484 e = blocks->next_free++;
485 oidcpy(&e->idx.oid, oid);
486 return e;
489 static struct object_entry *find_object(struct object_id *oid)
491 return hashmap_get_entry_from_hash(&object_table, oidhash(oid), oid,
492 struct object_entry, ent);
495 static struct object_entry *insert_object(struct object_id *oid)
497 struct object_entry *e;
498 unsigned int hash = oidhash(oid);
500 e = hashmap_get_entry_from_hash(&object_table, hash, oid,
501 struct object_entry, ent);
502 if (!e) {
503 e = new_object(oid);
504 e->idx.offset = 0;
505 hashmap_entry_init(&e->ent, hash);
506 hashmap_add(&object_table, &e->ent);
509 return e;
512 static void invalidate_pack_id(unsigned int id)
514 unsigned long lu;
515 struct tag *t;
516 struct hashmap_iter iter;
517 struct object_entry *e;
519 hashmap_for_each_entry(&object_table, &iter, e, ent) {
520 if (e->pack_id == id)
521 e->pack_id = MAX_PACK_ID;
524 for (lu = 0; lu < branch_table_sz; lu++) {
525 struct branch *b;
527 for (b = branch_table[lu]; b; b = b->table_next_branch)
528 if (b->pack_id == id)
529 b->pack_id = MAX_PACK_ID;
532 for (t = first_tag; t; t = t->next_tag)
533 if (t->pack_id == id)
534 t->pack_id = MAX_PACK_ID;
537 static unsigned int hc_str(const char *s, size_t len)
539 unsigned int r = 0;
540 while (len-- > 0)
541 r = r * 31 + *s++;
542 return r;
545 static void insert_mark(struct mark_set **top, uintmax_t idnum, struct object_entry *oe)
547 struct mark_set *s = *top;
549 while ((idnum >> s->shift) >= 1024) {
550 s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
551 s->shift = (*top)->shift + 10;
552 s->data.sets[0] = *top;
553 *top = s;
555 while (s->shift) {
556 uintmax_t i = idnum >> s->shift;
557 idnum -= i << s->shift;
558 if (!s->data.sets[i]) {
559 s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
560 s->data.sets[i]->shift = s->shift - 10;
562 s = s->data.sets[i];
564 if (!s->data.marked[idnum])
565 marks_set_count++;
566 s->data.marked[idnum] = oe;
569 static void *find_mark(struct mark_set *s, uintmax_t idnum)
571 uintmax_t orig_idnum = idnum;
572 struct object_entry *oe = NULL;
573 if ((idnum >> s->shift) < 1024) {
574 while (s && s->shift) {
575 uintmax_t i = idnum >> s->shift;
576 idnum -= i << s->shift;
577 s = s->data.sets[i];
579 if (s)
580 oe = s->data.marked[idnum];
582 if (!oe)
583 die("mark :%" PRIuMAX " not declared", orig_idnum);
584 return oe;
587 static struct atom_str *to_atom(const char *s, unsigned short len)
589 unsigned int hc = hc_str(s, len) % atom_table_sz;
590 struct atom_str *c;
592 for (c = atom_table[hc]; c; c = c->next_atom)
593 if (c->str_len == len && !strncmp(s, c->str_dat, len))
594 return c;
596 c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
597 c->str_len = len;
598 memcpy(c->str_dat, s, len);
599 c->str_dat[len] = 0;
600 c->next_atom = atom_table[hc];
601 atom_table[hc] = c;
602 atom_cnt++;
603 return c;
606 static struct branch *lookup_branch(const char *name)
608 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
609 struct branch *b;
611 for (b = branch_table[hc]; b; b = b->table_next_branch)
612 if (!strcmp(name, b->name))
613 return b;
614 return NULL;
617 static struct branch *new_branch(const char *name)
619 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
620 struct branch *b = lookup_branch(name);
622 if (b)
623 die("Invalid attempt to create duplicate branch: %s", name);
624 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
625 die("Branch name doesn't conform to GIT standards: %s", name);
627 b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
628 b->name = mem_pool_strdup(&fi_mem_pool, name);
629 b->table_next_branch = branch_table[hc];
630 b->branch_tree.versions[0].mode = S_IFDIR;
631 b->branch_tree.versions[1].mode = S_IFDIR;
632 b->num_notes = 0;
633 b->active = 0;
634 b->pack_id = MAX_PACK_ID;
635 branch_table[hc] = b;
636 branch_count++;
637 return b;
640 static unsigned int hc_entries(unsigned int cnt)
642 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
643 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
646 static struct tree_content *new_tree_content(unsigned int cnt)
648 struct avail_tree_content *f, *l = NULL;
649 struct tree_content *t;
650 unsigned int hc = hc_entries(cnt);
652 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
653 if (f->entry_capacity >= cnt)
654 break;
656 if (f) {
657 if (l)
658 l->next_avail = f->next_avail;
659 else
660 avail_tree_table[hc] = f->next_avail;
661 } else {
662 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
663 f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
664 f->entry_capacity = cnt;
667 t = (struct tree_content*)f;
668 t->entry_count = 0;
669 t->delta_depth = 0;
670 return t;
673 static void release_tree_entry(struct tree_entry *e);
674 static void release_tree_content(struct tree_content *t)
676 struct avail_tree_content *f = (struct avail_tree_content*)t;
677 unsigned int hc = hc_entries(f->entry_capacity);
678 f->next_avail = avail_tree_table[hc];
679 avail_tree_table[hc] = f;
682 static void release_tree_content_recursive(struct tree_content *t)
684 unsigned int i;
685 for (i = 0; i < t->entry_count; i++)
686 release_tree_entry(t->entries[i]);
687 release_tree_content(t);
690 static struct tree_content *grow_tree_content(
691 struct tree_content *t,
692 int amt)
694 struct tree_content *r = new_tree_content(t->entry_count + amt);
695 r->entry_count = t->entry_count;
696 r->delta_depth = t->delta_depth;
697 COPY_ARRAY(r->entries, t->entries, t->entry_count);
698 release_tree_content(t);
699 return r;
702 static struct tree_entry *new_tree_entry(void)
704 struct tree_entry *e;
706 if (!avail_tree_entry) {
707 unsigned int n = tree_entry_alloc;
708 tree_entry_allocd += n * sizeof(struct tree_entry);
709 ALLOC_ARRAY(e, n);
710 avail_tree_entry = e;
711 while (n-- > 1) {
712 *((void**)e) = e + 1;
713 e++;
715 *((void**)e) = NULL;
718 e = avail_tree_entry;
719 avail_tree_entry = *((void**)e);
720 return e;
723 static void release_tree_entry(struct tree_entry *e)
725 if (e->tree)
726 release_tree_content_recursive(e->tree);
727 *((void**)e) = avail_tree_entry;
728 avail_tree_entry = e;
731 static struct tree_content *dup_tree_content(struct tree_content *s)
733 struct tree_content *d;
734 struct tree_entry *a, *b;
735 unsigned int i;
737 if (!s)
738 return NULL;
739 d = new_tree_content(s->entry_count);
740 for (i = 0; i < s->entry_count; i++) {
741 a = s->entries[i];
742 b = new_tree_entry();
743 memcpy(b, a, sizeof(*a));
744 if (a->tree && is_null_oid(&b->versions[1].oid))
745 b->tree = dup_tree_content(a->tree);
746 else
747 b->tree = NULL;
748 d->entries[i] = b;
750 d->entry_count = s->entry_count;
751 d->delta_depth = s->delta_depth;
753 return d;
756 static void start_packfile(void)
758 struct strbuf tmp_file = STRBUF_INIT;
759 struct packed_git *p;
760 int pack_fd;
762 pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
763 FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
764 strbuf_release(&tmp_file);
766 p->pack_fd = pack_fd;
767 p->do_not_close = 1;
768 pack_file = hashfd(pack_fd, p->pack_name);
770 pack_data = p;
771 pack_size = write_pack_header(pack_file, 0);
772 object_count = 0;
774 REALLOC_ARRAY(all_packs, pack_id + 1);
775 all_packs[pack_id] = p;
778 static const char *create_index(void)
780 const char *tmpfile;
781 struct pack_idx_entry **idx, **c, **last;
782 struct object_entry *e;
783 struct object_entry_pool *o;
785 /* Build the table of object IDs. */
786 ALLOC_ARRAY(idx, object_count);
787 c = idx;
788 for (o = blocks; o; o = o->next_pool)
789 for (e = o->next_free; e-- != o->entries;)
790 if (pack_id == e->pack_id)
791 *c++ = &e->idx;
792 last = idx + object_count;
793 if (c != last)
794 die("internal consistency error creating the index");
796 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts,
797 pack_data->hash);
798 free(idx);
799 return tmpfile;
802 static char *keep_pack(const char *curr_index_name)
804 static const char *keep_msg = "fast-import";
805 struct strbuf name = STRBUF_INIT;
806 int keep_fd;
808 odb_pack_name(&name, pack_data->hash, "keep");
809 keep_fd = odb_pack_keep(name.buf);
810 if (keep_fd < 0)
811 die_errno("cannot create keep file");
812 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
813 if (close(keep_fd))
814 die_errno("failed to write keep file");
816 odb_pack_name(&name, pack_data->hash, "pack");
817 if (finalize_object_file(pack_data->pack_name, name.buf))
818 die("cannot store pack file");
820 odb_pack_name(&name, pack_data->hash, "idx");
821 if (finalize_object_file(curr_index_name, name.buf))
822 die("cannot store index file");
823 free((void *)curr_index_name);
824 return strbuf_detach(&name, NULL);
827 static void unkeep_all_packs(void)
829 struct strbuf name = STRBUF_INIT;
830 int k;
832 for (k = 0; k < pack_id; k++) {
833 struct packed_git *p = all_packs[k];
834 odb_pack_name(&name, p->hash, "keep");
835 unlink_or_warn(name.buf);
837 strbuf_release(&name);
840 static int loosen_small_pack(const struct packed_git *p)
842 struct child_process unpack = CHILD_PROCESS_INIT;
844 if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
845 die_errno("Failed seeking to start of '%s'", p->pack_name);
847 unpack.in = p->pack_fd;
848 unpack.git_cmd = 1;
849 unpack.stdout_to_stderr = 1;
850 strvec_push(&unpack.args, "unpack-objects");
851 if (!show_stats)
852 strvec_push(&unpack.args, "-q");
854 return run_command(&unpack);
857 static void end_packfile(void)
859 static int running;
861 if (running || !pack_data)
862 return;
864 running = 1;
865 clear_delta_base_cache();
866 if (object_count) {
867 struct packed_git *new_p;
868 struct object_id cur_pack_oid;
869 char *idx_name;
870 int i;
871 struct branch *b;
872 struct tag *t;
874 close_pack_windows(pack_data);
875 finalize_hashfile(pack_file, cur_pack_oid.hash, FSYNC_COMPONENT_PACK, 0);
876 fixup_pack_header_footer(pack_data->pack_fd, pack_data->hash,
877 pack_data->pack_name, object_count,
878 cur_pack_oid.hash, pack_size);
880 if (object_count <= unpack_limit) {
881 if (!loosen_small_pack(pack_data)) {
882 invalidate_pack_id(pack_id);
883 goto discard_pack;
887 close(pack_data->pack_fd);
888 idx_name = keep_pack(create_index());
890 /* Register the packfile with core git's machinery. */
891 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
892 if (!new_p)
893 die("core git rejected index %s", idx_name);
894 all_packs[pack_id] = new_p;
895 install_packed_git(the_repository, new_p);
896 free(idx_name);
898 /* Print the boundary */
899 if (pack_edges) {
900 fprintf(pack_edges, "%s:", new_p->pack_name);
901 for (i = 0; i < branch_table_sz; i++) {
902 for (b = branch_table[i]; b; b = b->table_next_branch) {
903 if (b->pack_id == pack_id)
904 fprintf(pack_edges, " %s",
905 oid_to_hex(&b->oid));
908 for (t = first_tag; t; t = t->next_tag) {
909 if (t->pack_id == pack_id)
910 fprintf(pack_edges, " %s",
911 oid_to_hex(&t->oid));
913 fputc('\n', pack_edges);
914 fflush(pack_edges);
917 pack_id++;
919 else {
920 discard_pack:
921 close(pack_data->pack_fd);
922 unlink_or_warn(pack_data->pack_name);
924 FREE_AND_NULL(pack_data);
925 running = 0;
927 /* We can't carry a delta across packfiles. */
928 strbuf_release(&last_blob.data);
929 last_blob.offset = 0;
930 last_blob.depth = 0;
933 static void cycle_packfile(void)
935 end_packfile();
936 start_packfile();
939 static int store_object(
940 enum object_type type,
941 struct strbuf *dat,
942 struct last_object *last,
943 struct object_id *oidout,
944 uintmax_t mark)
946 void *out, *delta;
947 struct object_entry *e;
948 unsigned char hdr[96];
949 struct object_id oid;
950 unsigned long hdrlen, deltalen;
951 git_hash_ctx c;
952 git_zstream s;
954 hdrlen = format_object_header((char *)hdr, sizeof(hdr), type,
955 dat->len);
956 the_hash_algo->init_fn(&c);
957 the_hash_algo->update_fn(&c, hdr, hdrlen);
958 the_hash_algo->update_fn(&c, dat->buf, dat->len);
959 the_hash_algo->final_oid_fn(&oid, &c);
960 if (oidout)
961 oidcpy(oidout, &oid);
963 e = insert_object(&oid);
964 if (mark)
965 insert_mark(&marks, mark, e);
966 if (e->idx.offset) {
967 duplicate_count_by_type[type]++;
968 return 1;
969 } else if (find_sha1_pack(oid.hash,
970 get_all_packs(the_repository))) {
971 e->type = type;
972 e->pack_id = MAX_PACK_ID;
973 e->idx.offset = 1; /* just not zero! */
974 duplicate_count_by_type[type]++;
975 return 1;
978 if (last && last->data.len && last->data.buf && last->depth < max_depth
979 && dat->len > the_hash_algo->rawsz) {
981 delta_count_attempts_by_type[type]++;
982 delta = diff_delta(last->data.buf, last->data.len,
983 dat->buf, dat->len,
984 &deltalen, dat->len - the_hash_algo->rawsz);
985 } else
986 delta = NULL;
988 git_deflate_init(&s, pack_compression_level);
989 if (delta) {
990 s.next_in = delta;
991 s.avail_in = deltalen;
992 } else {
993 s.next_in = (void *)dat->buf;
994 s.avail_in = dat->len;
996 s.avail_out = git_deflate_bound(&s, s.avail_in);
997 s.next_out = out = xmalloc(s.avail_out);
998 while (git_deflate(&s, Z_FINISH) == Z_OK)
999 ; /* nothing */
1000 git_deflate_end(&s);
1002 /* Determine if we should auto-checkpoint. */
1003 if ((max_packsize
1004 && (pack_size + PACK_SIZE_THRESHOLD + s.total_out) > max_packsize)
1005 || (pack_size + PACK_SIZE_THRESHOLD + s.total_out) < pack_size) {
1007 /* This new object needs to *not* have the current pack_id. */
1008 e->pack_id = pack_id + 1;
1009 cycle_packfile();
1011 /* We cannot carry a delta into the new pack. */
1012 if (delta) {
1013 FREE_AND_NULL(delta);
1015 git_deflate_init(&s, pack_compression_level);
1016 s.next_in = (void *)dat->buf;
1017 s.avail_in = dat->len;
1018 s.avail_out = git_deflate_bound(&s, s.avail_in);
1019 s.next_out = out = xrealloc(out, s.avail_out);
1020 while (git_deflate(&s, Z_FINISH) == Z_OK)
1021 ; /* nothing */
1022 git_deflate_end(&s);
1026 e->type = type;
1027 e->pack_id = pack_id;
1028 e->idx.offset = pack_size;
1029 object_count++;
1030 object_count_by_type[type]++;
1032 crc32_begin(pack_file);
1034 if (delta) {
1035 off_t ofs = e->idx.offset - last->offset;
1036 unsigned pos = sizeof(hdr) - 1;
1038 delta_count_by_type[type]++;
1039 e->depth = last->depth + 1;
1041 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1042 OBJ_OFS_DELTA, deltalen);
1043 hashwrite(pack_file, hdr, hdrlen);
1044 pack_size += hdrlen;
1046 hdr[pos] = ofs & 127;
1047 while (ofs >>= 7)
1048 hdr[--pos] = 128 | (--ofs & 127);
1049 hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1050 pack_size += sizeof(hdr) - pos;
1051 } else {
1052 e->depth = 0;
1053 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1054 type, dat->len);
1055 hashwrite(pack_file, hdr, hdrlen);
1056 pack_size += hdrlen;
1059 hashwrite(pack_file, out, s.total_out);
1060 pack_size += s.total_out;
1062 e->idx.crc32 = crc32_end(pack_file);
1064 free(out);
1065 free(delta);
1066 if (last) {
1067 if (last->no_swap) {
1068 last->data = *dat;
1069 } else {
1070 strbuf_swap(&last->data, dat);
1072 last->offset = e->idx.offset;
1073 last->depth = e->depth;
1075 return 0;
1078 static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1080 if (hashfile_truncate(pack_file, checkpoint))
1081 die_errno("cannot truncate pack to skip duplicate");
1082 pack_size = checkpoint->offset;
1085 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1087 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1088 unsigned char *in_buf = xmalloc(in_sz);
1089 unsigned char *out_buf = xmalloc(out_sz);
1090 struct object_entry *e;
1091 struct object_id oid;
1092 unsigned long hdrlen;
1093 off_t offset;
1094 git_hash_ctx c;
1095 git_zstream s;
1096 struct hashfile_checkpoint checkpoint;
1097 int status = Z_OK;
1099 /* Determine if we should auto-checkpoint. */
1100 if ((max_packsize
1101 && (pack_size + PACK_SIZE_THRESHOLD + len) > max_packsize)
1102 || (pack_size + PACK_SIZE_THRESHOLD + len) < pack_size)
1103 cycle_packfile();
1105 hashfile_checkpoint(pack_file, &checkpoint);
1106 offset = checkpoint.offset;
1108 hdrlen = format_object_header((char *)out_buf, out_sz, OBJ_BLOB, len);
1110 the_hash_algo->init_fn(&c);
1111 the_hash_algo->update_fn(&c, out_buf, hdrlen);
1113 crc32_begin(pack_file);
1115 git_deflate_init(&s, pack_compression_level);
1117 hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1119 s.next_out = out_buf + hdrlen;
1120 s.avail_out = out_sz - hdrlen;
1122 while (status != Z_STREAM_END) {
1123 if (0 < len && !s.avail_in) {
1124 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1125 size_t n = fread(in_buf, 1, cnt, stdin);
1126 if (!n && feof(stdin))
1127 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1129 the_hash_algo->update_fn(&c, in_buf, n);
1130 s.next_in = in_buf;
1131 s.avail_in = n;
1132 len -= n;
1135 status = git_deflate(&s, len ? 0 : Z_FINISH);
1137 if (!s.avail_out || status == Z_STREAM_END) {
1138 size_t n = s.next_out - out_buf;
1139 hashwrite(pack_file, out_buf, n);
1140 pack_size += n;
1141 s.next_out = out_buf;
1142 s.avail_out = out_sz;
1145 switch (status) {
1146 case Z_OK:
1147 case Z_BUF_ERROR:
1148 case Z_STREAM_END:
1149 continue;
1150 default:
1151 die("unexpected deflate failure: %d", status);
1154 git_deflate_end(&s);
1155 the_hash_algo->final_oid_fn(&oid, &c);
1157 if (oidout)
1158 oidcpy(oidout, &oid);
1160 e = insert_object(&oid);
1162 if (mark)
1163 insert_mark(&marks, mark, e);
1165 if (e->idx.offset) {
1166 duplicate_count_by_type[OBJ_BLOB]++;
1167 truncate_pack(&checkpoint);
1169 } else if (find_sha1_pack(oid.hash,
1170 get_all_packs(the_repository))) {
1171 e->type = OBJ_BLOB;
1172 e->pack_id = MAX_PACK_ID;
1173 e->idx.offset = 1; /* just not zero! */
1174 duplicate_count_by_type[OBJ_BLOB]++;
1175 truncate_pack(&checkpoint);
1177 } else {
1178 e->depth = 0;
1179 e->type = OBJ_BLOB;
1180 e->pack_id = pack_id;
1181 e->idx.offset = offset;
1182 e->idx.crc32 = crc32_end(pack_file);
1183 object_count++;
1184 object_count_by_type[OBJ_BLOB]++;
1187 free(in_buf);
1188 free(out_buf);
1191 /* All calls must be guarded by find_object() or find_mark() to
1192 * ensure the 'struct object_entry' passed was written by this
1193 * process instance. We unpack the entry by the offset, avoiding
1194 * the need for the corresponding .idx file. This unpacking rule
1195 * works because we only use OBJ_REF_DELTA within the packfiles
1196 * created by fast-import.
1198 * oe must not be NULL. Such an oe usually comes from giving
1199 * an unknown SHA-1 to find_object() or an undefined mark to
1200 * find_mark(). Callers must test for this condition and use
1201 * the standard read_sha1_file() when it happens.
1203 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1204 * find_mark(), where the mark was reloaded from an existing marks
1205 * file and is referencing an object that this fast-import process
1206 * instance did not write out to a packfile. Callers must test for
1207 * this condition and use read_sha1_file() instead.
1209 static void *gfi_unpack_entry(
1210 struct object_entry *oe,
1211 unsigned long *sizep)
1213 enum object_type type;
1214 struct packed_git *p = all_packs[oe->pack_id];
1215 if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1216 /* The object is stored in the packfile we are writing to
1217 * and we have modified it since the last time we scanned
1218 * back to read a previously written object. If an old
1219 * window covered [p->pack_size, p->pack_size + rawsz) its
1220 * data is stale and is not valid. Closing all windows
1221 * and updating the packfile length ensures we can read
1222 * the newly written data.
1224 close_pack_windows(p);
1225 hashflush(pack_file);
1227 /* We have to offer rawsz bytes additional on the end of
1228 * the packfile as the core unpacker code assumes the
1229 * footer is present at the file end and must promise
1230 * at least rawsz bytes within any window it maps. But
1231 * we don't actually create the footer here.
1233 p->pack_size = pack_size + the_hash_algo->rawsz;
1235 return unpack_entry(the_repository, p, oe->idx.offset, &type, sizep);
1238 static const char *get_mode(const char *str, uint16_t *modep)
1240 unsigned char c;
1241 uint16_t mode = 0;
1243 while ((c = *str++) != ' ') {
1244 if (c < '0' || c > '7')
1245 return NULL;
1246 mode = (mode << 3) + (c - '0');
1248 *modep = mode;
1249 return str;
1252 static void load_tree(struct tree_entry *root)
1254 struct object_id *oid = &root->versions[1].oid;
1255 struct object_entry *myoe;
1256 struct tree_content *t;
1257 unsigned long size;
1258 char *buf;
1259 const char *c;
1261 root->tree = t = new_tree_content(8);
1262 if (is_null_oid(oid))
1263 return;
1265 myoe = find_object(oid);
1266 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1267 if (myoe->type != OBJ_TREE)
1268 die("Not a tree: %s", oid_to_hex(oid));
1269 t->delta_depth = myoe->depth;
1270 buf = gfi_unpack_entry(myoe, &size);
1271 if (!buf)
1272 die("Can't load tree %s", oid_to_hex(oid));
1273 } else {
1274 enum object_type type;
1275 buf = repo_read_object_file(the_repository, oid, &type, &size);
1276 if (!buf || type != OBJ_TREE)
1277 die("Can't load tree %s", oid_to_hex(oid));
1280 c = buf;
1281 while (c != (buf + size)) {
1282 struct tree_entry *e = new_tree_entry();
1284 if (t->entry_count == t->entry_capacity)
1285 root->tree = t = grow_tree_content(t, t->entry_count);
1286 t->entries[t->entry_count++] = e;
1288 e->tree = NULL;
1289 c = get_mode(c, &e->versions[1].mode);
1290 if (!c)
1291 die("Corrupt mode in %s", oid_to_hex(oid));
1292 e->versions[0].mode = e->versions[1].mode;
1293 e->name = to_atom(c, strlen(c));
1294 c += e->name->str_len + 1;
1295 oidread(&e->versions[0].oid, (unsigned char *)c);
1296 oidread(&e->versions[1].oid, (unsigned char *)c);
1297 c += the_hash_algo->rawsz;
1299 free(buf);
1302 static int tecmp0 (const void *_a, const void *_b)
1304 struct tree_entry *a = *((struct tree_entry**)_a);
1305 struct tree_entry *b = *((struct tree_entry**)_b);
1306 return base_name_compare(
1307 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1308 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1311 static int tecmp1 (const void *_a, const void *_b)
1313 struct tree_entry *a = *((struct tree_entry**)_a);
1314 struct tree_entry *b = *((struct tree_entry**)_b);
1315 return base_name_compare(
1316 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1317 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1320 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1322 size_t maxlen = 0;
1323 unsigned int i;
1325 if (!v)
1326 QSORT(t->entries, t->entry_count, tecmp0);
1327 else
1328 QSORT(t->entries, t->entry_count, tecmp1);
1330 for (i = 0; i < t->entry_count; i++) {
1331 if (t->entries[i]->versions[v].mode)
1332 maxlen += t->entries[i]->name->str_len + 34;
1335 strbuf_reset(b);
1336 strbuf_grow(b, maxlen);
1337 for (i = 0; i < t->entry_count; i++) {
1338 struct tree_entry *e = t->entries[i];
1339 if (!e->versions[v].mode)
1340 continue;
1341 strbuf_addf(b, "%o %s%c",
1342 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1343 e->name->str_dat, '\0');
1344 strbuf_add(b, e->versions[v].oid.hash, the_hash_algo->rawsz);
1348 static void store_tree(struct tree_entry *root)
1350 struct tree_content *t;
1351 unsigned int i, j, del;
1352 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1353 struct object_entry *le = NULL;
1355 if (!is_null_oid(&root->versions[1].oid))
1356 return;
1358 if (!root->tree)
1359 load_tree(root);
1360 t = root->tree;
1362 for (i = 0; i < t->entry_count; i++) {
1363 if (t->entries[i]->tree)
1364 store_tree(t->entries[i]);
1367 if (!(root->versions[0].mode & NO_DELTA))
1368 le = find_object(&root->versions[0].oid);
1369 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1370 mktree(t, 0, &old_tree);
1371 lo.data = old_tree;
1372 lo.offset = le->idx.offset;
1373 lo.depth = t->delta_depth;
1376 mktree(t, 1, &new_tree);
1377 store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1379 t->delta_depth = lo.depth;
1380 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1381 struct tree_entry *e = t->entries[i];
1382 if (e->versions[1].mode) {
1383 e->versions[0].mode = e->versions[1].mode;
1384 oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1385 t->entries[j++] = e;
1386 } else {
1387 release_tree_entry(e);
1388 del++;
1391 t->entry_count -= del;
1394 static void tree_content_replace(
1395 struct tree_entry *root,
1396 const struct object_id *oid,
1397 const uint16_t mode,
1398 struct tree_content *newtree)
1400 if (!S_ISDIR(mode))
1401 die("Root cannot be a non-directory");
1402 oidclr(&root->versions[0].oid);
1403 oidcpy(&root->versions[1].oid, oid);
1404 if (root->tree)
1405 release_tree_content_recursive(root->tree);
1406 root->tree = newtree;
1409 static int tree_content_set(
1410 struct tree_entry *root,
1411 const char *p,
1412 const struct object_id *oid,
1413 const uint16_t mode,
1414 struct tree_content *subtree)
1416 struct tree_content *t;
1417 const char *slash1;
1418 unsigned int i, n;
1419 struct tree_entry *e;
1421 slash1 = strchrnul(p, '/');
1422 n = slash1 - p;
1423 if (!n)
1424 die("Empty path component found in input");
1425 if (!*slash1 && !S_ISDIR(mode) && subtree)
1426 die("Non-directories cannot have subtrees");
1428 if (!root->tree)
1429 load_tree(root);
1430 t = root->tree;
1431 for (i = 0; i < t->entry_count; i++) {
1432 e = t->entries[i];
1433 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1434 if (!*slash1) {
1435 if (!S_ISDIR(mode)
1436 && e->versions[1].mode == mode
1437 && oideq(&e->versions[1].oid, oid))
1438 return 0;
1439 e->versions[1].mode = mode;
1440 oidcpy(&e->versions[1].oid, oid);
1441 if (e->tree)
1442 release_tree_content_recursive(e->tree);
1443 e->tree = subtree;
1446 * We need to leave e->versions[0].sha1 alone
1447 * to avoid modifying the preimage tree used
1448 * when writing out the parent directory.
1449 * But after replacing the subdir with a
1450 * completely different one, it's not a good
1451 * delta base any more, and besides, we've
1452 * thrown away the tree entries needed to
1453 * make a delta against it.
1455 * So let's just explicitly disable deltas
1456 * for the subtree.
1458 if (S_ISDIR(e->versions[0].mode))
1459 e->versions[0].mode |= NO_DELTA;
1461 oidclr(&root->versions[1].oid);
1462 return 1;
1464 if (!S_ISDIR(e->versions[1].mode)) {
1465 e->tree = new_tree_content(8);
1466 e->versions[1].mode = S_IFDIR;
1468 if (!e->tree)
1469 load_tree(e);
1470 if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1471 oidclr(&root->versions[1].oid);
1472 return 1;
1474 return 0;
1478 if (t->entry_count == t->entry_capacity)
1479 root->tree = t = grow_tree_content(t, t->entry_count);
1480 e = new_tree_entry();
1481 e->name = to_atom(p, n);
1482 e->versions[0].mode = 0;
1483 oidclr(&e->versions[0].oid);
1484 t->entries[t->entry_count++] = e;
1485 if (*slash1) {
1486 e->tree = new_tree_content(8);
1487 e->versions[1].mode = S_IFDIR;
1488 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1489 } else {
1490 e->tree = subtree;
1491 e->versions[1].mode = mode;
1492 oidcpy(&e->versions[1].oid, oid);
1494 oidclr(&root->versions[1].oid);
1495 return 1;
1498 static int tree_content_remove(
1499 struct tree_entry *root,
1500 const char *p,
1501 struct tree_entry *backup_leaf,
1502 int allow_root)
1504 struct tree_content *t;
1505 const char *slash1;
1506 unsigned int i, n;
1507 struct tree_entry *e;
1509 slash1 = strchrnul(p, '/');
1510 n = slash1 - p;
1512 if (!root->tree)
1513 load_tree(root);
1515 if (!*p && allow_root) {
1516 e = root;
1517 goto del_entry;
1520 t = root->tree;
1521 for (i = 0; i < t->entry_count; i++) {
1522 e = t->entries[i];
1523 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1524 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1526 * If p names a file in some subdirectory, and a
1527 * file or symlink matching the name of the
1528 * parent directory of p exists, then p cannot
1529 * exist and need not be deleted.
1531 return 1;
1532 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1533 goto del_entry;
1534 if (!e->tree)
1535 load_tree(e);
1536 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1537 for (n = 0; n < e->tree->entry_count; n++) {
1538 if (e->tree->entries[n]->versions[1].mode) {
1539 oidclr(&root->versions[1].oid);
1540 return 1;
1543 backup_leaf = NULL;
1544 goto del_entry;
1546 return 0;
1549 return 0;
1551 del_entry:
1552 if (backup_leaf)
1553 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1554 else if (e->tree)
1555 release_tree_content_recursive(e->tree);
1556 e->tree = NULL;
1557 e->versions[1].mode = 0;
1558 oidclr(&e->versions[1].oid);
1559 oidclr(&root->versions[1].oid);
1560 return 1;
1563 static int tree_content_get(
1564 struct tree_entry *root,
1565 const char *p,
1566 struct tree_entry *leaf,
1567 int allow_root)
1569 struct tree_content *t;
1570 const char *slash1;
1571 unsigned int i, n;
1572 struct tree_entry *e;
1574 slash1 = strchrnul(p, '/');
1575 n = slash1 - p;
1576 if (!n && !allow_root)
1577 die("Empty path component found in input");
1579 if (!root->tree)
1580 load_tree(root);
1582 if (!n) {
1583 e = root;
1584 goto found_entry;
1587 t = root->tree;
1588 for (i = 0; i < t->entry_count; i++) {
1589 e = t->entries[i];
1590 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1591 if (!*slash1)
1592 goto found_entry;
1593 if (!S_ISDIR(e->versions[1].mode))
1594 return 0;
1595 if (!e->tree)
1596 load_tree(e);
1597 return tree_content_get(e, slash1 + 1, leaf, 0);
1600 return 0;
1602 found_entry:
1603 memcpy(leaf, e, sizeof(*leaf));
1604 if (e->tree && is_null_oid(&e->versions[1].oid))
1605 leaf->tree = dup_tree_content(e->tree);
1606 else
1607 leaf->tree = NULL;
1608 return 1;
1611 static int update_branch(struct branch *b)
1613 static const char *msg = "fast-import";
1614 struct ref_transaction *transaction;
1615 struct object_id old_oid;
1616 struct strbuf err = STRBUF_INIT;
1618 if (is_null_oid(&b->oid)) {
1619 if (b->delete)
1620 delete_ref(NULL, b->name, NULL, 0);
1621 return 0;
1623 if (read_ref(b->name, &old_oid))
1624 oidclr(&old_oid);
1625 if (!force_update && !is_null_oid(&old_oid)) {
1626 struct commit *old_cmit, *new_cmit;
1628 old_cmit = lookup_commit_reference_gently(the_repository,
1629 &old_oid, 0);
1630 new_cmit = lookup_commit_reference_gently(the_repository,
1631 &b->oid, 0);
1632 if (!old_cmit || !new_cmit)
1633 return error("Branch %s is missing commits.", b->name);
1635 if (!repo_in_merge_bases(the_repository, old_cmit, new_cmit)) {
1636 warning("Not updating %s"
1637 " (new tip %s does not contain %s)",
1638 b->name, oid_to_hex(&b->oid),
1639 oid_to_hex(&old_oid));
1640 return -1;
1643 transaction = ref_transaction_begin(&err);
1644 if (!transaction ||
1645 ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1646 0, msg, &err) ||
1647 ref_transaction_commit(transaction, &err)) {
1648 ref_transaction_free(transaction);
1649 error("%s", err.buf);
1650 strbuf_release(&err);
1651 return -1;
1653 ref_transaction_free(transaction);
1654 strbuf_release(&err);
1655 return 0;
1658 static void dump_branches(void)
1660 unsigned int i;
1661 struct branch *b;
1663 for (i = 0; i < branch_table_sz; i++) {
1664 for (b = branch_table[i]; b; b = b->table_next_branch)
1665 failure |= update_branch(b);
1669 static void dump_tags(void)
1671 static const char *msg = "fast-import";
1672 struct tag *t;
1673 struct strbuf ref_name = STRBUF_INIT;
1674 struct strbuf err = STRBUF_INIT;
1675 struct ref_transaction *transaction;
1677 transaction = ref_transaction_begin(&err);
1678 if (!transaction) {
1679 failure |= error("%s", err.buf);
1680 goto cleanup;
1682 for (t = first_tag; t; t = t->next_tag) {
1683 strbuf_reset(&ref_name);
1684 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1686 if (ref_transaction_update(transaction, ref_name.buf,
1687 &t->oid, NULL, 0, msg, &err)) {
1688 failure |= error("%s", err.buf);
1689 goto cleanup;
1692 if (ref_transaction_commit(transaction, &err))
1693 failure |= error("%s", err.buf);
1695 cleanup:
1696 ref_transaction_free(transaction);
1697 strbuf_release(&ref_name);
1698 strbuf_release(&err);
1701 static void dump_marks(void)
1703 struct lock_file mark_lock = LOCK_INIT;
1704 FILE *f;
1706 if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1707 return;
1709 if (safe_create_leading_directories_const(export_marks_file)) {
1710 failure |= error_errno("unable to create leading directories of %s",
1711 export_marks_file);
1712 return;
1715 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1716 failure |= error_errno("Unable to write marks file %s",
1717 export_marks_file);
1718 return;
1721 f = fdopen_lock_file(&mark_lock, "w");
1722 if (!f) {
1723 int saved_errno = errno;
1724 rollback_lock_file(&mark_lock);
1725 failure |= error("Unable to write marks file %s: %s",
1726 export_marks_file, strerror(saved_errno));
1727 return;
1730 for_each_mark(marks, 0, dump_marks_fn, f);
1731 if (commit_lock_file(&mark_lock)) {
1732 failure |= error_errno("Unable to write file %s",
1733 export_marks_file);
1734 return;
1738 static void insert_object_entry(struct mark_set **s, struct object_id *oid, uintmax_t mark)
1740 struct object_entry *e;
1741 e = find_object(oid);
1742 if (!e) {
1743 enum object_type type = oid_object_info(the_repository,
1744 oid, NULL);
1745 if (type < 0)
1746 die("object not found: %s", oid_to_hex(oid));
1747 e = insert_object(oid);
1748 e->type = type;
1749 e->pack_id = MAX_PACK_ID;
1750 e->idx.offset = 1; /* just not zero! */
1752 insert_mark(s, mark, e);
1755 static void insert_oid_entry(struct mark_set **s, struct object_id *oid, uintmax_t mark)
1757 insert_mark(s, mark, xmemdupz(oid, sizeof(*oid)));
1760 static void read_mark_file(struct mark_set **s, FILE *f, mark_set_inserter_t inserter)
1762 char line[512];
1763 while (fgets(line, sizeof(line), f)) {
1764 uintmax_t mark;
1765 char *end;
1766 struct object_id oid;
1768 /* Ensure SHA-1 objects are padded with zeros. */
1769 memset(oid.hash, 0, sizeof(oid.hash));
1771 end = strchr(line, '\n');
1772 if (line[0] != ':' || !end)
1773 die("corrupt mark line: %s", line);
1774 *end = 0;
1775 mark = strtoumax(line + 1, &end, 10);
1776 if (!mark || end == line + 1
1777 || *end != ' '
1778 || get_oid_hex_any(end + 1, &oid) == GIT_HASH_UNKNOWN)
1779 die("corrupt mark line: %s", line);
1780 inserter(s, &oid, mark);
1784 static void read_marks(void)
1786 FILE *f = fopen(import_marks_file, "r");
1787 if (f)
1789 else if (import_marks_file_ignore_missing && errno == ENOENT)
1790 goto done; /* Marks file does not exist */
1791 else
1792 die_errno("cannot read '%s'", import_marks_file);
1793 read_mark_file(&marks, f, insert_object_entry);
1794 fclose(f);
1795 done:
1796 import_marks_file_done = 1;
1800 static int read_next_command(void)
1802 static int stdin_eof = 0;
1804 if (stdin_eof) {
1805 unread_command_buf = 0;
1806 return EOF;
1809 for (;;) {
1810 if (unread_command_buf) {
1811 unread_command_buf = 0;
1812 } else {
1813 struct recent_command *rc;
1815 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1816 if (stdin_eof)
1817 return EOF;
1819 if (!seen_data_command
1820 && !starts_with(command_buf.buf, "feature ")
1821 && !starts_with(command_buf.buf, "option ")) {
1822 parse_argv();
1825 rc = rc_free;
1826 if (rc)
1827 rc_free = rc->next;
1828 else {
1829 rc = cmd_hist.next;
1830 cmd_hist.next = rc->next;
1831 cmd_hist.next->prev = &cmd_hist;
1832 free(rc->buf);
1835 rc->buf = xstrdup(command_buf.buf);
1836 rc->prev = cmd_tail;
1837 rc->next = cmd_hist.prev;
1838 rc->prev->next = rc;
1839 cmd_tail = rc;
1841 if (command_buf.buf[0] == '#')
1842 continue;
1843 return 0;
1847 static void skip_optional_lf(void)
1849 int term_char = fgetc(stdin);
1850 if (term_char != '\n' && term_char != EOF)
1851 ungetc(term_char, stdin);
1854 static void parse_mark(void)
1856 const char *v;
1857 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1858 next_mark = strtoumax(v, NULL, 10);
1859 read_next_command();
1861 else
1862 next_mark = 0;
1865 static void parse_original_identifier(void)
1867 const char *v;
1868 if (skip_prefix(command_buf.buf, "original-oid ", &v))
1869 read_next_command();
1872 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1874 const char *data;
1875 strbuf_reset(sb);
1877 if (!skip_prefix(command_buf.buf, "data ", &data))
1878 die("Expected 'data n' command, found: %s", command_buf.buf);
1880 if (skip_prefix(data, "<<", &data)) {
1881 char *term = xstrdup(data);
1882 size_t term_len = command_buf.len - (data - command_buf.buf);
1884 for (;;) {
1885 if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1886 die("EOF in data (terminator '%s' not found)", term);
1887 if (term_len == command_buf.len
1888 && !strcmp(term, command_buf.buf))
1889 break;
1890 strbuf_addbuf(sb, &command_buf);
1891 strbuf_addch(sb, '\n');
1893 free(term);
1895 else {
1896 uintmax_t len = strtoumax(data, NULL, 10);
1897 size_t n = 0, length = (size_t)len;
1899 if (limit && limit < len) {
1900 *len_res = len;
1901 return 0;
1903 if (length < len)
1904 die("data is too large to use in this context");
1906 while (n < length) {
1907 size_t s = strbuf_fread(sb, length - n, stdin);
1908 if (!s && feof(stdin))
1909 die("EOF in data (%lu bytes remaining)",
1910 (unsigned long)(length - n));
1911 n += s;
1915 skip_optional_lf();
1916 return 1;
1919 static int validate_raw_date(const char *src, struct strbuf *result, int strict)
1921 const char *orig_src = src;
1922 char *endp;
1923 unsigned long num;
1925 errno = 0;
1927 num = strtoul(src, &endp, 10);
1929 * NEEDSWORK: perhaps check for reasonable values? For example, we
1930 * could error on values representing times more than a
1931 * day in the future.
1933 if (errno || endp == src || *endp != ' ')
1934 return -1;
1936 src = endp + 1;
1937 if (*src != '-' && *src != '+')
1938 return -1;
1940 num = strtoul(src + 1, &endp, 10);
1942 * NEEDSWORK: check for brokenness other than num > 1400, such as
1943 * (num % 100) >= 60, or ((num % 100) % 15) != 0 ?
1945 if (errno || endp == src + 1 || *endp || /* did not parse */
1946 (strict && (1400 < num)) /* parsed a broken timezone */
1948 return -1;
1950 strbuf_addstr(result, orig_src);
1951 return 0;
1954 static char *parse_ident(const char *buf)
1956 const char *ltgt;
1957 size_t name_len;
1958 struct strbuf ident = STRBUF_INIT;
1960 /* ensure there is a space delimiter even if there is no name */
1961 if (*buf == '<')
1962 --buf;
1964 ltgt = buf + strcspn(buf, "<>");
1965 if (*ltgt != '<')
1966 die("Missing < in ident string: %s", buf);
1967 if (ltgt != buf && ltgt[-1] != ' ')
1968 die("Missing space before < in ident string: %s", buf);
1969 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
1970 if (*ltgt != '>')
1971 die("Missing > in ident string: %s", buf);
1972 ltgt++;
1973 if (*ltgt != ' ')
1974 die("Missing space after > in ident string: %s", buf);
1975 ltgt++;
1976 name_len = ltgt - buf;
1977 strbuf_add(&ident, buf, name_len);
1979 switch (whenspec) {
1980 case WHENSPEC_RAW:
1981 if (validate_raw_date(ltgt, &ident, 1) < 0)
1982 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1983 break;
1984 case WHENSPEC_RAW_PERMISSIVE:
1985 if (validate_raw_date(ltgt, &ident, 0) < 0)
1986 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1987 break;
1988 case WHENSPEC_RFC2822:
1989 if (parse_date(ltgt, &ident) < 0)
1990 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
1991 break;
1992 case WHENSPEC_NOW:
1993 if (strcmp("now", ltgt))
1994 die("Date in ident must be 'now': %s", buf);
1995 datestamp(&ident);
1996 break;
1999 return strbuf_detach(&ident, NULL);
2002 static void parse_and_store_blob(
2003 struct last_object *last,
2004 struct object_id *oidout,
2005 uintmax_t mark)
2007 static struct strbuf buf = STRBUF_INIT;
2008 uintmax_t len;
2010 if (parse_data(&buf, big_file_threshold, &len))
2011 store_object(OBJ_BLOB, &buf, last, oidout, mark);
2012 else {
2013 if (last) {
2014 strbuf_release(&last->data);
2015 last->offset = 0;
2016 last->depth = 0;
2018 stream_blob(len, oidout, mark);
2019 skip_optional_lf();
2023 static void parse_new_blob(void)
2025 read_next_command();
2026 parse_mark();
2027 parse_original_identifier();
2028 parse_and_store_blob(&last_blob, NULL, next_mark);
2031 static void unload_one_branch(void)
2033 while (cur_active_branches
2034 && cur_active_branches >= max_active_branches) {
2035 uintmax_t min_commit = ULONG_MAX;
2036 struct branch *e, *l = NULL, *p = NULL;
2038 for (e = active_branches; e; e = e->active_next_branch) {
2039 if (e->last_commit < min_commit) {
2040 p = l;
2041 min_commit = e->last_commit;
2043 l = e;
2046 if (p) {
2047 e = p->active_next_branch;
2048 p->active_next_branch = e->active_next_branch;
2049 } else {
2050 e = active_branches;
2051 active_branches = e->active_next_branch;
2053 e->active = 0;
2054 e->active_next_branch = NULL;
2055 if (e->branch_tree.tree) {
2056 release_tree_content_recursive(e->branch_tree.tree);
2057 e->branch_tree.tree = NULL;
2059 cur_active_branches--;
2063 static void load_branch(struct branch *b)
2065 load_tree(&b->branch_tree);
2066 if (!b->active) {
2067 b->active = 1;
2068 b->active_next_branch = active_branches;
2069 active_branches = b;
2070 cur_active_branches++;
2071 branch_load_count++;
2075 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2077 unsigned char fanout = 0;
2078 while ((num_notes >>= 8))
2079 fanout++;
2080 return fanout;
2083 static void construct_path_with_fanout(const char *hex_sha1,
2084 unsigned char fanout, char *path)
2086 unsigned int i = 0, j = 0;
2087 if (fanout >= the_hash_algo->rawsz)
2088 die("Too large fanout (%u)", fanout);
2089 while (fanout) {
2090 path[i++] = hex_sha1[j++];
2091 path[i++] = hex_sha1[j++];
2092 path[i++] = '/';
2093 fanout--;
2095 memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2096 path[i + the_hash_algo->hexsz - j] = '\0';
2099 static uintmax_t do_change_note_fanout(
2100 struct tree_entry *orig_root, struct tree_entry *root,
2101 char *hex_oid, unsigned int hex_oid_len,
2102 char *fullpath, unsigned int fullpath_len,
2103 unsigned char fanout)
2105 struct tree_content *t;
2106 struct tree_entry *e, leaf;
2107 unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2108 uintmax_t num_notes = 0;
2109 struct object_id oid;
2110 /* hex oid + '/' between each pair of hex digits + NUL */
2111 char realpath[GIT_MAX_HEXSZ + ((GIT_MAX_HEXSZ / 2) - 1) + 1];
2112 const unsigned hexsz = the_hash_algo->hexsz;
2114 if (!root->tree)
2115 load_tree(root);
2116 t = root->tree;
2118 for (i = 0; t && i < t->entry_count; i++) {
2119 e = t->entries[i];
2120 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2121 tmp_fullpath_len = fullpath_len;
2124 * We're interested in EITHER existing note entries (entries
2125 * with exactly 40 hex chars in path, not including directory
2126 * separators), OR directory entries that may contain note
2127 * entries (with < 40 hex chars in path).
2128 * Also, each path component in a note entry must be a multiple
2129 * of 2 chars.
2131 if (!e->versions[1].mode ||
2132 tmp_hex_oid_len > hexsz ||
2133 e->name->str_len % 2)
2134 continue;
2136 /* This _may_ be a note entry, or a subdir containing notes */
2137 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2138 e->name->str_len);
2139 if (tmp_fullpath_len)
2140 fullpath[tmp_fullpath_len++] = '/';
2141 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2142 e->name->str_len);
2143 tmp_fullpath_len += e->name->str_len;
2144 fullpath[tmp_fullpath_len] = '\0';
2146 if (tmp_hex_oid_len == hexsz && !get_oid_hex(hex_oid, &oid)) {
2147 /* This is a note entry */
2148 if (fanout == 0xff) {
2149 /* Counting mode, no rename */
2150 num_notes++;
2151 continue;
2153 construct_path_with_fanout(hex_oid, fanout, realpath);
2154 if (!strcmp(fullpath, realpath)) {
2155 /* Note entry is in correct location */
2156 num_notes++;
2157 continue;
2160 /* Rename fullpath to realpath */
2161 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2162 die("Failed to remove path %s", fullpath);
2163 tree_content_set(orig_root, realpath,
2164 &leaf.versions[1].oid,
2165 leaf.versions[1].mode,
2166 leaf.tree);
2167 } else if (S_ISDIR(e->versions[1].mode)) {
2168 /* This is a subdir that may contain note entries */
2169 num_notes += do_change_note_fanout(orig_root, e,
2170 hex_oid, tmp_hex_oid_len,
2171 fullpath, tmp_fullpath_len, fanout);
2174 /* The above may have reallocated the current tree_content */
2175 t = root->tree;
2177 return num_notes;
2180 static uintmax_t change_note_fanout(struct tree_entry *root,
2181 unsigned char fanout)
2184 * The size of path is due to one slash between every two hex digits,
2185 * plus the terminating NUL. Note that there is no slash at the end, so
2186 * the number of slashes is one less than half the number of hex
2187 * characters.
2189 char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2190 return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2193 static int parse_mapped_oid_hex(const char *hex, struct object_id *oid, const char **end)
2195 int algo;
2196 khiter_t it;
2198 /* Make SHA-1 object IDs have all-zero padding. */
2199 memset(oid->hash, 0, sizeof(oid->hash));
2201 algo = parse_oid_hex_any(hex, oid, end);
2202 if (algo == GIT_HASH_UNKNOWN)
2203 return -1;
2205 it = kh_get_oid_map(sub_oid_map, *oid);
2206 /* No such object? */
2207 if (it == kh_end(sub_oid_map)) {
2208 /* If we're using the same algorithm, pass it through. */
2209 if (hash_algos[algo].format_id == the_hash_algo->format_id)
2210 return 0;
2211 return -1;
2213 oidcpy(oid, kh_value(sub_oid_map, it));
2214 return 0;
2218 * Given a pointer into a string, parse a mark reference:
2220 * idnum ::= ':' bigint;
2222 * Return the first character after the value in *endptr.
2224 * Complain if the following character is not what is expected,
2225 * either a space or end of the string.
2227 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2229 uintmax_t mark;
2231 assert(*p == ':');
2232 p++;
2233 mark = strtoumax(p, endptr, 10);
2234 if (*endptr == p)
2235 die("No value after ':' in mark: %s", command_buf.buf);
2236 return mark;
2240 * Parse the mark reference, and complain if this is not the end of
2241 * the string.
2243 static uintmax_t parse_mark_ref_eol(const char *p)
2245 char *end;
2246 uintmax_t mark;
2248 mark = parse_mark_ref(p, &end);
2249 if (*end != '\0')
2250 die("Garbage after mark: %s", command_buf.buf);
2251 return mark;
2255 * Parse the mark reference, demanding a trailing space. Return a
2256 * pointer to the space.
2258 static uintmax_t parse_mark_ref_space(const char **p)
2260 uintmax_t mark;
2261 char *end;
2263 mark = parse_mark_ref(*p, &end);
2264 if (*end++ != ' ')
2265 die("Missing space after mark: %s", command_buf.buf);
2266 *p = end;
2267 return mark;
2270 static void file_change_m(const char *p, struct branch *b)
2272 static struct strbuf uq = STRBUF_INIT;
2273 const char *endp;
2274 struct object_entry *oe;
2275 struct object_id oid;
2276 uint16_t mode, inline_data = 0;
2278 p = get_mode(p, &mode);
2279 if (!p)
2280 die("Corrupt mode: %s", command_buf.buf);
2281 switch (mode) {
2282 case 0644:
2283 case 0755:
2284 mode |= S_IFREG;
2285 case S_IFREG | 0644:
2286 case S_IFREG | 0755:
2287 case S_IFLNK:
2288 case S_IFDIR:
2289 case S_IFGITLINK:
2290 /* ok */
2291 break;
2292 default:
2293 die("Corrupt mode: %s", command_buf.buf);
2296 if (*p == ':') {
2297 oe = find_mark(marks, parse_mark_ref_space(&p));
2298 oidcpy(&oid, &oe->idx.oid);
2299 } else if (skip_prefix(p, "inline ", &p)) {
2300 inline_data = 1;
2301 oe = NULL; /* not used with inline_data, but makes gcc happy */
2302 } else {
2303 if (parse_mapped_oid_hex(p, &oid, &p))
2304 die("Invalid dataref: %s", command_buf.buf);
2305 oe = find_object(&oid);
2306 if (*p++ != ' ')
2307 die("Missing space after SHA1: %s", command_buf.buf);
2310 strbuf_reset(&uq);
2311 if (!unquote_c_style(&uq, p, &endp)) {
2312 if (*endp)
2313 die("Garbage after path in: %s", command_buf.buf);
2314 p = uq.buf;
2317 /* Git does not track empty, non-toplevel directories. */
2318 if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2319 tree_content_remove(&b->branch_tree, p, NULL, 0);
2320 return;
2323 if (S_ISGITLINK(mode)) {
2324 if (inline_data)
2325 die("Git links cannot be specified 'inline': %s",
2326 command_buf.buf);
2327 else if (oe) {
2328 if (oe->type != OBJ_COMMIT)
2329 die("Not a commit (actually a %s): %s",
2330 type_name(oe->type), command_buf.buf);
2333 * Accept the sha1 without checking; it expected to be in
2334 * another repository.
2336 } else if (inline_data) {
2337 if (S_ISDIR(mode))
2338 die("Directories cannot be specified 'inline': %s",
2339 command_buf.buf);
2340 if (p != uq.buf) {
2341 strbuf_addstr(&uq, p);
2342 p = uq.buf;
2344 while (read_next_command() != EOF) {
2345 const char *v;
2346 if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2347 parse_cat_blob(v);
2348 else {
2349 parse_and_store_blob(&last_blob, &oid, 0);
2350 break;
2353 } else {
2354 enum object_type expected = S_ISDIR(mode) ?
2355 OBJ_TREE: OBJ_BLOB;
2356 enum object_type type = oe ? oe->type :
2357 oid_object_info(the_repository, &oid,
2358 NULL);
2359 if (type < 0)
2360 die("%s not found: %s",
2361 S_ISDIR(mode) ? "Tree" : "Blob",
2362 command_buf.buf);
2363 if (type != expected)
2364 die("Not a %s (actually a %s): %s",
2365 type_name(expected), type_name(type),
2366 command_buf.buf);
2369 if (!*p) {
2370 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2371 return;
2373 tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2376 static void file_change_d(const char *p, struct branch *b)
2378 static struct strbuf uq = STRBUF_INIT;
2379 const char *endp;
2381 strbuf_reset(&uq);
2382 if (!unquote_c_style(&uq, p, &endp)) {
2383 if (*endp)
2384 die("Garbage after path in: %s", command_buf.buf);
2385 p = uq.buf;
2387 tree_content_remove(&b->branch_tree, p, NULL, 1);
2390 static void file_change_cr(const char *s, struct branch *b, int rename)
2392 const char *d;
2393 static struct strbuf s_uq = STRBUF_INIT;
2394 static struct strbuf d_uq = STRBUF_INIT;
2395 const char *endp;
2396 struct tree_entry leaf;
2398 strbuf_reset(&s_uq);
2399 if (!unquote_c_style(&s_uq, s, &endp)) {
2400 if (*endp != ' ')
2401 die("Missing space after source: %s", command_buf.buf);
2402 } else {
2403 endp = strchr(s, ' ');
2404 if (!endp)
2405 die("Missing space after source: %s", command_buf.buf);
2406 strbuf_add(&s_uq, s, endp - s);
2408 s = s_uq.buf;
2410 endp++;
2411 if (!*endp)
2412 die("Missing dest: %s", command_buf.buf);
2414 d = endp;
2415 strbuf_reset(&d_uq);
2416 if (!unquote_c_style(&d_uq, d, &endp)) {
2417 if (*endp)
2418 die("Garbage after dest in: %s", command_buf.buf);
2419 d = d_uq.buf;
2422 memset(&leaf, 0, sizeof(leaf));
2423 if (rename)
2424 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2425 else
2426 tree_content_get(&b->branch_tree, s, &leaf, 1);
2427 if (!leaf.versions[1].mode)
2428 die("Path %s not in branch", s);
2429 if (!*d) { /* C "path/to/subdir" "" */
2430 tree_content_replace(&b->branch_tree,
2431 &leaf.versions[1].oid,
2432 leaf.versions[1].mode,
2433 leaf.tree);
2434 return;
2436 tree_content_set(&b->branch_tree, d,
2437 &leaf.versions[1].oid,
2438 leaf.versions[1].mode,
2439 leaf.tree);
2442 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2444 static struct strbuf uq = STRBUF_INIT;
2445 struct object_entry *oe;
2446 struct branch *s;
2447 struct object_id oid, commit_oid;
2448 char path[GIT_MAX_RAWSZ * 3];
2449 uint16_t inline_data = 0;
2450 unsigned char new_fanout;
2453 * When loading a branch, we don't traverse its tree to count the real
2454 * number of notes (too expensive to do this for all non-note refs).
2455 * This means that recently loaded notes refs might incorrectly have
2456 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2458 * Fix this by traversing the tree and counting the number of notes
2459 * when b->num_notes == 0. If the notes tree is truly empty, the
2460 * calculation should not take long.
2462 if (b->num_notes == 0 && *old_fanout == 0) {
2463 /* Invoke change_note_fanout() in "counting mode". */
2464 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2465 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2468 /* Now parse the notemodify command. */
2469 /* <dataref> or 'inline' */
2470 if (*p == ':') {
2471 oe = find_mark(marks, parse_mark_ref_space(&p));
2472 oidcpy(&oid, &oe->idx.oid);
2473 } else if (skip_prefix(p, "inline ", &p)) {
2474 inline_data = 1;
2475 oe = NULL; /* not used with inline_data, but makes gcc happy */
2476 } else {
2477 if (parse_mapped_oid_hex(p, &oid, &p))
2478 die("Invalid dataref: %s", command_buf.buf);
2479 oe = find_object(&oid);
2480 if (*p++ != ' ')
2481 die("Missing space after SHA1: %s", command_buf.buf);
2484 /* <commit-ish> */
2485 s = lookup_branch(p);
2486 if (s) {
2487 if (is_null_oid(&s->oid))
2488 die("Can't add a note on empty branch.");
2489 oidcpy(&commit_oid, &s->oid);
2490 } else if (*p == ':') {
2491 uintmax_t commit_mark = parse_mark_ref_eol(p);
2492 struct object_entry *commit_oe = find_mark(marks, commit_mark);
2493 if (commit_oe->type != OBJ_COMMIT)
2494 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2495 oidcpy(&commit_oid, &commit_oe->idx.oid);
2496 } else if (!repo_get_oid(the_repository, p, &commit_oid)) {
2497 unsigned long size;
2498 char *buf = read_object_with_reference(the_repository,
2499 &commit_oid,
2500 OBJ_COMMIT, &size,
2501 &commit_oid);
2502 if (!buf || size < the_hash_algo->hexsz + 6)
2503 die("Not a valid commit: %s", p);
2504 free(buf);
2505 } else
2506 die("Invalid ref name or SHA1 expression: %s", p);
2508 if (inline_data) {
2509 if (p != uq.buf) {
2510 strbuf_addstr(&uq, p);
2511 p = uq.buf;
2513 read_next_command();
2514 parse_and_store_blob(&last_blob, &oid, 0);
2515 } else if (oe) {
2516 if (oe->type != OBJ_BLOB)
2517 die("Not a blob (actually a %s): %s",
2518 type_name(oe->type), command_buf.buf);
2519 } else if (!is_null_oid(&oid)) {
2520 enum object_type type = oid_object_info(the_repository, &oid,
2521 NULL);
2522 if (type < 0)
2523 die("Blob not found: %s", command_buf.buf);
2524 if (type != OBJ_BLOB)
2525 die("Not a blob (actually a %s): %s",
2526 type_name(type), command_buf.buf);
2529 construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2530 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2531 b->num_notes--;
2533 if (is_null_oid(&oid))
2534 return; /* nothing to insert */
2536 b->num_notes++;
2537 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2538 construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2539 tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2542 static void file_change_deleteall(struct branch *b)
2544 release_tree_content_recursive(b->branch_tree.tree);
2545 oidclr(&b->branch_tree.versions[0].oid);
2546 oidclr(&b->branch_tree.versions[1].oid);
2547 load_tree(&b->branch_tree);
2548 b->num_notes = 0;
2551 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2553 if (!buf || size < the_hash_algo->hexsz + 6)
2554 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2555 if (memcmp("tree ", buf, 5)
2556 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2557 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2558 oidcpy(&b->branch_tree.versions[0].oid,
2559 &b->branch_tree.versions[1].oid);
2562 static void parse_from_existing(struct branch *b)
2564 if (is_null_oid(&b->oid)) {
2565 oidclr(&b->branch_tree.versions[0].oid);
2566 oidclr(&b->branch_tree.versions[1].oid);
2567 } else {
2568 unsigned long size;
2569 char *buf;
2571 buf = read_object_with_reference(the_repository,
2572 &b->oid, OBJ_COMMIT, &size,
2573 &b->oid);
2574 parse_from_commit(b, buf, size);
2575 free(buf);
2579 static int parse_objectish(struct branch *b, const char *objectish)
2581 struct branch *s;
2582 struct object_id oid;
2584 oidcpy(&oid, &b->branch_tree.versions[1].oid);
2586 s = lookup_branch(objectish);
2587 if (b == s)
2588 die("Can't create a branch from itself: %s", b->name);
2589 else if (s) {
2590 struct object_id *t = &s->branch_tree.versions[1].oid;
2591 oidcpy(&b->oid, &s->oid);
2592 oidcpy(&b->branch_tree.versions[0].oid, t);
2593 oidcpy(&b->branch_tree.versions[1].oid, t);
2594 } else if (*objectish == ':') {
2595 uintmax_t idnum = parse_mark_ref_eol(objectish);
2596 struct object_entry *oe = find_mark(marks, idnum);
2597 if (oe->type != OBJ_COMMIT)
2598 die("Mark :%" PRIuMAX " not a commit", idnum);
2599 if (!oideq(&b->oid, &oe->idx.oid)) {
2600 oidcpy(&b->oid, &oe->idx.oid);
2601 if (oe->pack_id != MAX_PACK_ID) {
2602 unsigned long size;
2603 char *buf = gfi_unpack_entry(oe, &size);
2604 parse_from_commit(b, buf, size);
2605 free(buf);
2606 } else
2607 parse_from_existing(b);
2609 } else if (!repo_get_oid(the_repository, objectish, &b->oid)) {
2610 parse_from_existing(b);
2611 if (is_null_oid(&b->oid))
2612 b->delete = 1;
2614 else
2615 die("Invalid ref name or SHA1 expression: %s", objectish);
2617 if (b->branch_tree.tree && !oideq(&oid, &b->branch_tree.versions[1].oid)) {
2618 release_tree_content_recursive(b->branch_tree.tree);
2619 b->branch_tree.tree = NULL;
2622 read_next_command();
2623 return 1;
2626 static int parse_from(struct branch *b)
2628 const char *from;
2630 if (!skip_prefix(command_buf.buf, "from ", &from))
2631 return 0;
2633 return parse_objectish(b, from);
2636 static int parse_objectish_with_prefix(struct branch *b, const char *prefix)
2638 const char *base;
2640 if (!skip_prefix(command_buf.buf, prefix, &base))
2641 return 0;
2643 return parse_objectish(b, base);
2646 static struct hash_list *parse_merge(unsigned int *count)
2648 struct hash_list *list = NULL, **tail = &list, *n;
2649 const char *from;
2650 struct branch *s;
2652 *count = 0;
2653 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2654 n = xmalloc(sizeof(*n));
2655 s = lookup_branch(from);
2656 if (s)
2657 oidcpy(&n->oid, &s->oid);
2658 else if (*from == ':') {
2659 uintmax_t idnum = parse_mark_ref_eol(from);
2660 struct object_entry *oe = find_mark(marks, idnum);
2661 if (oe->type != OBJ_COMMIT)
2662 die("Mark :%" PRIuMAX " not a commit", idnum);
2663 oidcpy(&n->oid, &oe->idx.oid);
2664 } else if (!repo_get_oid(the_repository, from, &n->oid)) {
2665 unsigned long size;
2666 char *buf = read_object_with_reference(the_repository,
2667 &n->oid,
2668 OBJ_COMMIT,
2669 &size, &n->oid);
2670 if (!buf || size < the_hash_algo->hexsz + 6)
2671 die("Not a valid commit: %s", from);
2672 free(buf);
2673 } else
2674 die("Invalid ref name or SHA1 expression: %s", from);
2676 n->next = NULL;
2677 *tail = n;
2678 tail = &n->next;
2680 (*count)++;
2681 read_next_command();
2683 return list;
2686 static void parse_new_commit(const char *arg)
2688 static struct strbuf msg = STRBUF_INIT;
2689 struct branch *b;
2690 char *author = NULL;
2691 char *committer = NULL;
2692 char *encoding = NULL;
2693 struct hash_list *merge_list = NULL;
2694 unsigned int merge_count;
2695 unsigned char prev_fanout, new_fanout;
2696 const char *v;
2698 b = lookup_branch(arg);
2699 if (!b)
2700 b = new_branch(arg);
2702 read_next_command();
2703 parse_mark();
2704 parse_original_identifier();
2705 if (skip_prefix(command_buf.buf, "author ", &v)) {
2706 author = parse_ident(v);
2707 read_next_command();
2709 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2710 committer = parse_ident(v);
2711 read_next_command();
2713 if (!committer)
2714 die("Expected committer but didn't get one");
2715 if (skip_prefix(command_buf.buf, "encoding ", &v)) {
2716 encoding = xstrdup(v);
2717 read_next_command();
2719 parse_data(&msg, 0, NULL);
2720 read_next_command();
2721 parse_from(b);
2722 merge_list = parse_merge(&merge_count);
2724 /* ensure the branch is active/loaded */
2725 if (!b->branch_tree.tree || !max_active_branches) {
2726 unload_one_branch();
2727 load_branch(b);
2730 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2732 /* file_change* */
2733 while (command_buf.len > 0) {
2734 if (skip_prefix(command_buf.buf, "M ", &v))
2735 file_change_m(v, b);
2736 else if (skip_prefix(command_buf.buf, "D ", &v))
2737 file_change_d(v, b);
2738 else if (skip_prefix(command_buf.buf, "R ", &v))
2739 file_change_cr(v, b, 1);
2740 else if (skip_prefix(command_buf.buf, "C ", &v))
2741 file_change_cr(v, b, 0);
2742 else if (skip_prefix(command_buf.buf, "N ", &v))
2743 note_change_n(v, b, &prev_fanout);
2744 else if (!strcmp("deleteall", command_buf.buf))
2745 file_change_deleteall(b);
2746 else if (skip_prefix(command_buf.buf, "ls ", &v))
2747 parse_ls(v, b);
2748 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2749 parse_cat_blob(v);
2750 else {
2751 unread_command_buf = 1;
2752 break;
2754 if (read_next_command() == EOF)
2755 break;
2758 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2759 if (new_fanout != prev_fanout)
2760 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2762 /* build the tree and the commit */
2763 store_tree(&b->branch_tree);
2764 oidcpy(&b->branch_tree.versions[0].oid,
2765 &b->branch_tree.versions[1].oid);
2767 strbuf_reset(&new_data);
2768 strbuf_addf(&new_data, "tree %s\n",
2769 oid_to_hex(&b->branch_tree.versions[1].oid));
2770 if (!is_null_oid(&b->oid))
2771 strbuf_addf(&new_data, "parent %s\n",
2772 oid_to_hex(&b->oid));
2773 while (merge_list) {
2774 struct hash_list *next = merge_list->next;
2775 strbuf_addf(&new_data, "parent %s\n",
2776 oid_to_hex(&merge_list->oid));
2777 free(merge_list);
2778 merge_list = next;
2780 strbuf_addf(&new_data,
2781 "author %s\n"
2782 "committer %s\n",
2783 author ? author : committer, committer);
2784 if (encoding)
2785 strbuf_addf(&new_data,
2786 "encoding %s\n",
2787 encoding);
2788 strbuf_addch(&new_data, '\n');
2789 strbuf_addbuf(&new_data, &msg);
2790 free(author);
2791 free(committer);
2792 free(encoding);
2794 if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2795 b->pack_id = pack_id;
2796 b->last_commit = object_count_by_type[OBJ_COMMIT];
2799 static void parse_new_tag(const char *arg)
2801 static struct strbuf msg = STRBUF_INIT;
2802 const char *from;
2803 char *tagger;
2804 struct branch *s;
2805 struct tag *t;
2806 uintmax_t from_mark = 0;
2807 struct object_id oid;
2808 enum object_type type;
2809 const char *v;
2811 t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2812 memset(t, 0, sizeof(struct tag));
2813 t->name = mem_pool_strdup(&fi_mem_pool, arg);
2814 if (last_tag)
2815 last_tag->next_tag = t;
2816 else
2817 first_tag = t;
2818 last_tag = t;
2819 read_next_command();
2820 parse_mark();
2822 /* from ... */
2823 if (!skip_prefix(command_buf.buf, "from ", &from))
2824 die("Expected from command, got %s", command_buf.buf);
2825 s = lookup_branch(from);
2826 if (s) {
2827 if (is_null_oid(&s->oid))
2828 die("Can't tag an empty branch.");
2829 oidcpy(&oid, &s->oid);
2830 type = OBJ_COMMIT;
2831 } else if (*from == ':') {
2832 struct object_entry *oe;
2833 from_mark = parse_mark_ref_eol(from);
2834 oe = find_mark(marks, from_mark);
2835 type = oe->type;
2836 oidcpy(&oid, &oe->idx.oid);
2837 } else if (!repo_get_oid(the_repository, from, &oid)) {
2838 struct object_entry *oe = find_object(&oid);
2839 if (!oe) {
2840 type = oid_object_info(the_repository, &oid, NULL);
2841 if (type < 0)
2842 die("Not a valid object: %s", from);
2843 } else
2844 type = oe->type;
2845 } else
2846 die("Invalid ref name or SHA1 expression: %s", from);
2847 read_next_command();
2849 /* original-oid ... */
2850 parse_original_identifier();
2852 /* tagger ... */
2853 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2854 tagger = parse_ident(v);
2855 read_next_command();
2856 } else
2857 tagger = NULL;
2859 /* tag payload/message */
2860 parse_data(&msg, 0, NULL);
2862 /* build the tag object */
2863 strbuf_reset(&new_data);
2865 strbuf_addf(&new_data,
2866 "object %s\n"
2867 "type %s\n"
2868 "tag %s\n",
2869 oid_to_hex(&oid), type_name(type), t->name);
2870 if (tagger)
2871 strbuf_addf(&new_data,
2872 "tagger %s\n", tagger);
2873 strbuf_addch(&new_data, '\n');
2874 strbuf_addbuf(&new_data, &msg);
2875 free(tagger);
2877 if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, next_mark))
2878 t->pack_id = MAX_PACK_ID;
2879 else
2880 t->pack_id = pack_id;
2883 static void parse_reset_branch(const char *arg)
2885 struct branch *b;
2886 const char *tag_name;
2888 b = lookup_branch(arg);
2889 if (b) {
2890 oidclr(&b->oid);
2891 oidclr(&b->branch_tree.versions[0].oid);
2892 oidclr(&b->branch_tree.versions[1].oid);
2893 if (b->branch_tree.tree) {
2894 release_tree_content_recursive(b->branch_tree.tree);
2895 b->branch_tree.tree = NULL;
2898 else
2899 b = new_branch(arg);
2900 read_next_command();
2901 parse_from(b);
2902 if (b->delete && skip_prefix(b->name, "refs/tags/", &tag_name)) {
2904 * Elsewhere, we call dump_branches() before dump_tags(),
2905 * and dump_branches() will handle ref deletions first, so
2906 * in order to make sure the deletion actually takes effect,
2907 * we need to remove the tag from our list of tags to update.
2909 * NEEDSWORK: replace list of tags with hashmap for faster
2910 * deletion?
2912 struct tag *t, *prev = NULL;
2913 for (t = first_tag; t; t = t->next_tag) {
2914 if (!strcmp(t->name, tag_name))
2915 break;
2916 prev = t;
2918 if (t) {
2919 if (prev)
2920 prev->next_tag = t->next_tag;
2921 else
2922 first_tag = t->next_tag;
2923 if (!t->next_tag)
2924 last_tag = prev;
2925 /* There is no mem_pool_free(t) function to call. */
2928 if (command_buf.len > 0)
2929 unread_command_buf = 1;
2932 static void cat_blob_write(const char *buf, unsigned long size)
2934 if (write_in_full(cat_blob_fd, buf, size) < 0)
2935 die_errno("Write to frontend failed");
2938 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2940 struct strbuf line = STRBUF_INIT;
2941 unsigned long size;
2942 enum object_type type = 0;
2943 char *buf;
2945 if (!oe || oe->pack_id == MAX_PACK_ID) {
2946 buf = repo_read_object_file(the_repository, oid, &type, &size);
2947 } else {
2948 type = oe->type;
2949 buf = gfi_unpack_entry(oe, &size);
2953 * Output based on batch_one_object() from cat-file.c.
2955 if (type <= 0) {
2956 strbuf_reset(&line);
2957 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2958 cat_blob_write(line.buf, line.len);
2959 strbuf_release(&line);
2960 free(buf);
2961 return;
2963 if (!buf)
2964 die("Can't read object %s", oid_to_hex(oid));
2965 if (type != OBJ_BLOB)
2966 die("Object %s is a %s but a blob was expected.",
2967 oid_to_hex(oid), type_name(type));
2968 strbuf_reset(&line);
2969 strbuf_addf(&line, "%s %s %"PRIuMAX"\n", oid_to_hex(oid),
2970 type_name(type), (uintmax_t)size);
2971 cat_blob_write(line.buf, line.len);
2972 strbuf_release(&line);
2973 cat_blob_write(buf, size);
2974 cat_blob_write("\n", 1);
2975 if (oe && oe->pack_id == pack_id) {
2976 last_blob.offset = oe->idx.offset;
2977 strbuf_attach(&last_blob.data, buf, size, size);
2978 last_blob.depth = oe->depth;
2979 } else
2980 free(buf);
2983 static void parse_get_mark(const char *p)
2985 struct object_entry *oe;
2986 char output[GIT_MAX_HEXSZ + 2];
2988 /* get-mark SP <object> LF */
2989 if (*p != ':')
2990 die("Not a mark: %s", p);
2992 oe = find_mark(marks, parse_mark_ref_eol(p));
2993 if (!oe)
2994 die("Unknown mark: %s", command_buf.buf);
2996 xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2997 cat_blob_write(output, the_hash_algo->hexsz + 1);
3000 static void parse_cat_blob(const char *p)
3002 struct object_entry *oe;
3003 struct object_id oid;
3005 /* cat-blob SP <object> LF */
3006 if (*p == ':') {
3007 oe = find_mark(marks, parse_mark_ref_eol(p));
3008 if (!oe)
3009 die("Unknown mark: %s", command_buf.buf);
3010 oidcpy(&oid, &oe->idx.oid);
3011 } else {
3012 if (parse_mapped_oid_hex(p, &oid, &p))
3013 die("Invalid dataref: %s", command_buf.buf);
3014 if (*p)
3015 die("Garbage after SHA1: %s", command_buf.buf);
3016 oe = find_object(&oid);
3019 cat_blob(oe, &oid);
3022 static struct object_entry *dereference(struct object_entry *oe,
3023 struct object_id *oid)
3025 unsigned long size;
3026 char *buf = NULL;
3027 const unsigned hexsz = the_hash_algo->hexsz;
3029 if (!oe) {
3030 enum object_type type = oid_object_info(the_repository, oid,
3031 NULL);
3032 if (type < 0)
3033 die("object not found: %s", oid_to_hex(oid));
3034 /* cache it! */
3035 oe = insert_object(oid);
3036 oe->type = type;
3037 oe->pack_id = MAX_PACK_ID;
3038 oe->idx.offset = 1;
3040 switch (oe->type) {
3041 case OBJ_TREE: /* easy case. */
3042 return oe;
3043 case OBJ_COMMIT:
3044 case OBJ_TAG:
3045 break;
3046 default:
3047 die("Not a tree-ish: %s", command_buf.buf);
3050 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3051 buf = gfi_unpack_entry(oe, &size);
3052 } else {
3053 enum object_type unused;
3054 buf = repo_read_object_file(the_repository, oid, &unused,
3055 &size);
3057 if (!buf)
3058 die("Can't load object %s", oid_to_hex(oid));
3060 /* Peel one layer. */
3061 switch (oe->type) {
3062 case OBJ_TAG:
3063 if (size < hexsz + strlen("object ") ||
3064 get_oid_hex(buf + strlen("object "), oid))
3065 die("Invalid SHA1 in tag: %s", command_buf.buf);
3066 break;
3067 case OBJ_COMMIT:
3068 if (size < hexsz + strlen("tree ") ||
3069 get_oid_hex(buf + strlen("tree "), oid))
3070 die("Invalid SHA1 in commit: %s", command_buf.buf);
3073 free(buf);
3074 return find_object(oid);
3077 static void insert_mapped_mark(uintmax_t mark, void *object, void *cbp)
3079 struct object_id *fromoid = object;
3080 struct object_id *tooid = find_mark(cbp, mark);
3081 int ret;
3082 khiter_t it;
3084 it = kh_put_oid_map(sub_oid_map, *fromoid, &ret);
3085 /* We've already seen this object. */
3086 if (ret == 0)
3087 return;
3088 kh_value(sub_oid_map, it) = tooid;
3091 static void build_mark_map_one(struct mark_set *from, struct mark_set *to)
3093 for_each_mark(from, 0, insert_mapped_mark, to);
3096 static void build_mark_map(struct string_list *from, struct string_list *to)
3098 struct string_list_item *fromp, *top;
3100 sub_oid_map = kh_init_oid_map();
3102 for_each_string_list_item(fromp, from) {
3103 top = string_list_lookup(to, fromp->string);
3104 if (!fromp->util) {
3105 die(_("Missing from marks for submodule '%s'"), fromp->string);
3106 } else if (!top || !top->util) {
3107 die(_("Missing to marks for submodule '%s'"), fromp->string);
3109 build_mark_map_one(fromp->util, top->util);
3113 static struct object_entry *parse_treeish_dataref(const char **p)
3115 struct object_id oid;
3116 struct object_entry *e;
3118 if (**p == ':') { /* <mark> */
3119 e = find_mark(marks, parse_mark_ref_space(p));
3120 if (!e)
3121 die("Unknown mark: %s", command_buf.buf);
3122 oidcpy(&oid, &e->idx.oid);
3123 } else { /* <sha1> */
3124 if (parse_mapped_oid_hex(*p, &oid, p))
3125 die("Invalid dataref: %s", command_buf.buf);
3126 e = find_object(&oid);
3127 if (*(*p)++ != ' ')
3128 die("Missing space after tree-ish: %s", command_buf.buf);
3131 while (!e || e->type != OBJ_TREE)
3132 e = dereference(e, &oid);
3133 return e;
3136 static void print_ls(int mode, const unsigned char *hash, const char *path)
3138 static struct strbuf line = STRBUF_INIT;
3140 /* See show_tree(). */
3141 const char *type =
3142 S_ISGITLINK(mode) ? commit_type :
3143 S_ISDIR(mode) ? tree_type :
3144 blob_type;
3146 if (!mode) {
3147 /* missing SP path LF */
3148 strbuf_reset(&line);
3149 strbuf_addstr(&line, "missing ");
3150 quote_c_style(path, &line, NULL, 0);
3151 strbuf_addch(&line, '\n');
3152 } else {
3153 /* mode SP type SP object_name TAB path LF */
3154 strbuf_reset(&line);
3155 strbuf_addf(&line, "%06o %s %s\t",
3156 mode & ~NO_DELTA, type, hash_to_hex(hash));
3157 quote_c_style(path, &line, NULL, 0);
3158 strbuf_addch(&line, '\n');
3160 cat_blob_write(line.buf, line.len);
3163 static void parse_ls(const char *p, struct branch *b)
3165 struct tree_entry *root = NULL;
3166 struct tree_entry leaf = {NULL};
3168 /* ls SP (<tree-ish> SP)? <path> */
3169 if (*p == '"') {
3170 if (!b)
3171 die("Not in a commit: %s", command_buf.buf);
3172 root = &b->branch_tree;
3173 } else {
3174 struct object_entry *e = parse_treeish_dataref(&p);
3175 root = new_tree_entry();
3176 oidcpy(&root->versions[1].oid, &e->idx.oid);
3177 if (!is_null_oid(&root->versions[1].oid))
3178 root->versions[1].mode = S_IFDIR;
3179 load_tree(root);
3181 if (*p == '"') {
3182 static struct strbuf uq = STRBUF_INIT;
3183 const char *endp;
3184 strbuf_reset(&uq);
3185 if (unquote_c_style(&uq, p, &endp))
3186 die("Invalid path: %s", command_buf.buf);
3187 if (*endp)
3188 die("Garbage after path in: %s", command_buf.buf);
3189 p = uq.buf;
3191 tree_content_get(root, p, &leaf, 1);
3193 * A directory in preparation would have a sha1 of zero
3194 * until it is saved. Save, for simplicity.
3196 if (S_ISDIR(leaf.versions[1].mode))
3197 store_tree(&leaf);
3199 print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3200 if (leaf.tree)
3201 release_tree_content_recursive(leaf.tree);
3202 if (!b || root != &b->branch_tree)
3203 release_tree_entry(root);
3206 static void checkpoint(void)
3208 checkpoint_requested = 0;
3209 if (object_count) {
3210 cycle_packfile();
3212 dump_branches();
3213 dump_tags();
3214 dump_marks();
3217 static void parse_checkpoint(void)
3219 checkpoint_requested = 1;
3220 skip_optional_lf();
3223 static void parse_progress(void)
3225 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3226 fputc('\n', stdout);
3227 fflush(stdout);
3228 skip_optional_lf();
3231 static void parse_alias(void)
3233 struct object_entry *e;
3234 struct branch b;
3236 skip_optional_lf();
3237 read_next_command();
3239 /* mark ... */
3240 parse_mark();
3241 if (!next_mark)
3242 die(_("Expected 'mark' command, got %s"), command_buf.buf);
3244 /* to ... */
3245 memset(&b, 0, sizeof(b));
3246 if (!parse_objectish_with_prefix(&b, "to "))
3247 die(_("Expected 'to' command, got %s"), command_buf.buf);
3248 e = find_object(&b.oid);
3249 assert(e);
3250 insert_mark(&marks, next_mark, e);
3253 static char* make_fast_import_path(const char *path)
3255 if (!relative_marks_paths || is_absolute_path(path))
3256 return prefix_filename(global_prefix, path);
3257 return git_pathdup("info/fast-import/%s", path);
3260 static void option_import_marks(const char *marks,
3261 int from_stream, int ignore_missing)
3263 if (import_marks_file) {
3264 if (from_stream)
3265 die("Only one import-marks command allowed per stream");
3267 /* read previous mark file */
3268 if(!import_marks_file_from_stream)
3269 read_marks();
3272 import_marks_file = make_fast_import_path(marks);
3273 import_marks_file_from_stream = from_stream;
3274 import_marks_file_ignore_missing = ignore_missing;
3277 static void option_date_format(const char *fmt)
3279 if (!strcmp(fmt, "raw"))
3280 whenspec = WHENSPEC_RAW;
3281 else if (!strcmp(fmt, "raw-permissive"))
3282 whenspec = WHENSPEC_RAW_PERMISSIVE;
3283 else if (!strcmp(fmt, "rfc2822"))
3284 whenspec = WHENSPEC_RFC2822;
3285 else if (!strcmp(fmt, "now"))
3286 whenspec = WHENSPEC_NOW;
3287 else
3288 die("unknown --date-format argument %s", fmt);
3291 static unsigned long ulong_arg(const char *option, const char *arg)
3293 char *endptr;
3294 unsigned long rv = strtoul(arg, &endptr, 0);
3295 if (strchr(arg, '-') || endptr == arg || *endptr)
3296 die("%s: argument must be a non-negative integer", option);
3297 return rv;
3300 static void option_depth(const char *depth)
3302 max_depth = ulong_arg("--depth", depth);
3303 if (max_depth > MAX_DEPTH)
3304 die("--depth cannot exceed %u", MAX_DEPTH);
3307 static void option_active_branches(const char *branches)
3309 max_active_branches = ulong_arg("--active-branches", branches);
3312 static void option_export_marks(const char *marks)
3314 export_marks_file = make_fast_import_path(marks);
3317 static void option_cat_blob_fd(const char *fd)
3319 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3320 if (n > (unsigned long) INT_MAX)
3321 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3322 cat_blob_fd = (int) n;
3325 static void option_export_pack_edges(const char *edges)
3327 char *fn = prefix_filename(global_prefix, edges);
3328 if (pack_edges)
3329 fclose(pack_edges);
3330 pack_edges = xfopen(fn, "a");
3331 free(fn);
3334 static void option_rewrite_submodules(const char *arg, struct string_list *list)
3336 struct mark_set *ms;
3337 FILE *fp;
3338 char *s = xstrdup(arg);
3339 char *f = strchr(s, ':');
3340 if (!f)
3341 die(_("Expected format name:filename for submodule rewrite option"));
3342 *f = '\0';
3343 f++;
3344 CALLOC_ARRAY(ms, 1);
3346 f = prefix_filename(global_prefix, f);
3347 fp = fopen(f, "r");
3348 if (!fp)
3349 die_errno("cannot read '%s'", f);
3350 read_mark_file(&ms, fp, insert_oid_entry);
3351 fclose(fp);
3352 free(f);
3354 string_list_insert(list, s)->util = ms;
3357 static int parse_one_option(const char *option)
3359 if (skip_prefix(option, "max-pack-size=", &option)) {
3360 unsigned long v;
3361 if (!git_parse_ulong(option, &v))
3362 return 0;
3363 if (v < 8192) {
3364 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3365 v *= 1024 * 1024;
3366 } else if (v < 1024 * 1024) {
3367 warning("minimum max-pack-size is 1 MiB");
3368 v = 1024 * 1024;
3370 max_packsize = v;
3371 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3372 unsigned long v;
3373 if (!git_parse_ulong(option, &v))
3374 return 0;
3375 big_file_threshold = v;
3376 } else if (skip_prefix(option, "depth=", &option)) {
3377 option_depth(option);
3378 } else if (skip_prefix(option, "active-branches=", &option)) {
3379 option_active_branches(option);
3380 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3381 option_export_pack_edges(option);
3382 } else if (!strcmp(option, "quiet")) {
3383 show_stats = 0;
3384 } else if (!strcmp(option, "stats")) {
3385 show_stats = 1;
3386 } else if (!strcmp(option, "allow-unsafe-features")) {
3387 ; /* already handled during early option parsing */
3388 } else {
3389 return 0;
3392 return 1;
3395 static void check_unsafe_feature(const char *feature, int from_stream)
3397 if (from_stream && !allow_unsafe_features)
3398 die(_("feature '%s' forbidden in input without --allow-unsafe-features"),
3399 feature);
3402 static int parse_one_feature(const char *feature, int from_stream)
3404 const char *arg;
3406 if (skip_prefix(feature, "date-format=", &arg)) {
3407 option_date_format(arg);
3408 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3409 check_unsafe_feature("import-marks", from_stream);
3410 option_import_marks(arg, from_stream, 0);
3411 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3412 check_unsafe_feature("import-marks-if-exists", from_stream);
3413 option_import_marks(arg, from_stream, 1);
3414 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3415 check_unsafe_feature(feature, from_stream);
3416 option_export_marks(arg);
3417 } else if (!strcmp(feature, "alias")) {
3418 ; /* Don't die - this feature is supported */
3419 } else if (skip_prefix(feature, "rewrite-submodules-to=", &arg)) {
3420 option_rewrite_submodules(arg, &sub_marks_to);
3421 } else if (skip_prefix(feature, "rewrite-submodules-from=", &arg)) {
3422 option_rewrite_submodules(arg, &sub_marks_from);
3423 } else if (!strcmp(feature, "get-mark")) {
3424 ; /* Don't die - this feature is supported */
3425 } else if (!strcmp(feature, "cat-blob")) {
3426 ; /* Don't die - this feature is supported */
3427 } else if (!strcmp(feature, "relative-marks")) {
3428 relative_marks_paths = 1;
3429 } else if (!strcmp(feature, "no-relative-marks")) {
3430 relative_marks_paths = 0;
3431 } else if (!strcmp(feature, "done")) {
3432 require_explicit_termination = 1;
3433 } else if (!strcmp(feature, "force")) {
3434 force_update = 1;
3435 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3436 ; /* do nothing; we have the feature */
3437 } else {
3438 return 0;
3441 return 1;
3444 static void parse_feature(const char *feature)
3446 if (seen_data_command)
3447 die("Got feature command '%s' after data command", feature);
3449 if (parse_one_feature(feature, 1))
3450 return;
3452 die("This version of fast-import does not support feature %s.", feature);
3455 static void parse_option(const char *option)
3457 if (seen_data_command)
3458 die("Got option command '%s' after data command", option);
3460 if (parse_one_option(option))
3461 return;
3463 die("This version of fast-import does not support option: %s", option);
3466 static void git_pack_config(void)
3468 int indexversion_value;
3469 int limit;
3470 unsigned long packsizelimit_value;
3472 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3473 if (max_depth > MAX_DEPTH)
3474 max_depth = MAX_DEPTH;
3476 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3477 pack_idx_opts.version = indexversion_value;
3478 if (pack_idx_opts.version > 2)
3479 git_die_config("pack.indexversion",
3480 "bad pack.indexVersion=%"PRIu32, pack_idx_opts.version);
3482 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3483 max_packsize = packsizelimit_value;
3485 if (!git_config_get_int("fastimport.unpacklimit", &limit))
3486 unpack_limit = limit;
3487 else if (!git_config_get_int("transfer.unpacklimit", &limit))
3488 unpack_limit = limit;
3490 git_config(git_default_config, NULL);
3493 static const char fast_import_usage[] =
3494 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3496 static void parse_argv(void)
3498 unsigned int i;
3500 for (i = 1; i < global_argc; i++) {
3501 const char *a = global_argv[i];
3503 if (*a != '-' || !strcmp(a, "--"))
3504 break;
3506 if (!skip_prefix(a, "--", &a))
3507 die("unknown option %s", a);
3509 if (parse_one_option(a))
3510 continue;
3512 if (parse_one_feature(a, 0))
3513 continue;
3515 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3516 option_cat_blob_fd(a);
3517 continue;
3520 die("unknown option --%s", a);
3522 if (i != global_argc)
3523 usage(fast_import_usage);
3525 seen_data_command = 1;
3526 if (import_marks_file)
3527 read_marks();
3528 build_mark_map(&sub_marks_from, &sub_marks_to);
3531 int cmd_fast_import(int argc, const char **argv, const char *prefix)
3533 unsigned int i;
3535 if (argc == 2 && !strcmp(argv[1], "-h"))
3536 usage(fast_import_usage);
3538 reset_pack_idx_option(&pack_idx_opts);
3539 git_pack_config();
3541 alloc_objects(object_entry_alloc);
3542 strbuf_init(&command_buf, 0);
3543 CALLOC_ARRAY(atom_table, atom_table_sz);
3544 CALLOC_ARRAY(branch_table, branch_table_sz);
3545 CALLOC_ARRAY(avail_tree_table, avail_tree_table_sz);
3546 marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3548 hashmap_init(&object_table, object_entry_hashcmp, NULL, 0);
3551 * We don't parse most options until after we've seen the set of
3552 * "feature" lines at the start of the stream (which allows the command
3553 * line to override stream data). But we must do an early parse of any
3554 * command-line options that impact how we interpret the feature lines.
3556 for (i = 1; i < argc; i++) {
3557 const char *arg = argv[i];
3558 if (*arg != '-' || !strcmp(arg, "--"))
3559 break;
3560 if (!strcmp(arg, "--allow-unsafe-features"))
3561 allow_unsafe_features = 1;
3564 global_argc = argc;
3565 global_argv = argv;
3566 global_prefix = prefix;
3568 rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3569 for (i = 0; i < (cmd_save - 1); i++)
3570 rc_free[i].next = &rc_free[i + 1];
3571 rc_free[cmd_save - 1].next = NULL;
3573 start_packfile();
3574 set_die_routine(die_nicely);
3575 set_checkpoint_signal();
3576 while (read_next_command() != EOF) {
3577 const char *v;
3578 if (!strcmp("blob", command_buf.buf))
3579 parse_new_blob();
3580 else if (skip_prefix(command_buf.buf, "commit ", &v))
3581 parse_new_commit(v);
3582 else if (skip_prefix(command_buf.buf, "tag ", &v))
3583 parse_new_tag(v);
3584 else if (skip_prefix(command_buf.buf, "reset ", &v))
3585 parse_reset_branch(v);
3586 else if (skip_prefix(command_buf.buf, "ls ", &v))
3587 parse_ls(v, NULL);
3588 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
3589 parse_cat_blob(v);
3590 else if (skip_prefix(command_buf.buf, "get-mark ", &v))
3591 parse_get_mark(v);
3592 else if (!strcmp("checkpoint", command_buf.buf))
3593 parse_checkpoint();
3594 else if (!strcmp("done", command_buf.buf))
3595 break;
3596 else if (!strcmp("alias", command_buf.buf))
3597 parse_alias();
3598 else if (starts_with(command_buf.buf, "progress "))
3599 parse_progress();
3600 else if (skip_prefix(command_buf.buf, "feature ", &v))
3601 parse_feature(v);
3602 else if (skip_prefix(command_buf.buf, "option git ", &v))
3603 parse_option(v);
3604 else if (starts_with(command_buf.buf, "option "))
3605 /* ignore non-git options*/;
3606 else
3607 die("Unsupported command: %s", command_buf.buf);
3609 if (checkpoint_requested)
3610 checkpoint();
3613 /* argv hasn't been parsed yet, do so */
3614 if (!seen_data_command)
3615 parse_argv();
3617 if (require_explicit_termination && feof(stdin))
3618 die("stream ends early");
3620 end_packfile();
3622 dump_branches();
3623 dump_tags();
3624 unkeep_all_packs();
3625 dump_marks();
3627 if (pack_edges)
3628 fclose(pack_edges);
3630 if (show_stats) {
3631 uintmax_t total_count = 0, duplicate_count = 0;
3632 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3633 total_count += object_count_by_type[i];
3634 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3635 duplicate_count += duplicate_count_by_type[i];
3637 fprintf(stderr, "%s statistics:\n", argv[0]);
3638 fprintf(stderr, "---------------------------------------------------------------------\n");
3639 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3640 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3641 fprintf(stderr, " blobs : %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB], delta_count_attempts_by_type[OBJ_BLOB]);
3642 fprintf(stderr, " trees : %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE], delta_count_attempts_by_type[OBJ_TREE]);
3643 fprintf(stderr, " commits: %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT], delta_count_attempts_by_type[OBJ_COMMIT]);
3644 fprintf(stderr, " tags : %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG], delta_count_attempts_by_type[OBJ_TAG]);
3645 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3646 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3647 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3648 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3649 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3650 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3651 fprintf(stderr, "---------------------------------------------------------------------\n");
3652 pack_report();
3653 fprintf(stderr, "---------------------------------------------------------------------\n");
3654 fprintf(stderr, "\n");
3657 return failure ? 1 : 0;