Merge branch 'master' of github.com:nafmo/git-l10n-sv
[alt-git.git] / builtin / fast-import.c
blob20406f677542c0c1e34b623fc3f4b2b349700525
1 #include "builtin.h"
2 #include "cache.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "lockfile.h"
6 #include "object.h"
7 #include "blob.h"
8 #include "tree.h"
9 #include "commit.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "refs.h"
13 #include "csum-file.h"
14 #include "quote.h"
15 #include "dir.h"
16 #include "run-command.h"
17 #include "packfile.h"
18 #include "object-store.h"
19 #include "mem-pool.h"
20 #include "commit-reach.h"
21 #include "khash.h"
23 #define PACK_ID_BITS 16
24 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
25 #define DEPTH_BITS 13
26 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
29 * We abuse the setuid bit on directories to mean "do not delta".
31 #define NO_DELTA S_ISUID
34 * The amount of additional space required in order to write an object into the
35 * current pack. This is the hash lengths at the end of the pack, plus the
36 * length of one object ID.
38 #define PACK_SIZE_THRESHOLD (the_hash_algo->rawsz * 3)
40 struct object_entry {
41 struct pack_idx_entry idx;
42 struct hashmap_entry ent;
43 uint32_t type : TYPE_BITS,
44 pack_id : PACK_ID_BITS,
45 depth : DEPTH_BITS;
48 static int object_entry_hashcmp(const void *map_data,
49 const struct hashmap_entry *eptr,
50 const struct hashmap_entry *entry_or_key,
51 const void *keydata)
53 const struct object_id *oid = keydata;
54 const struct object_entry *e1, *e2;
56 e1 = container_of(eptr, const struct object_entry, ent);
57 if (oid)
58 return oidcmp(&e1->idx.oid, oid);
60 e2 = container_of(entry_or_key, const struct object_entry, ent);
61 return oidcmp(&e1->idx.oid, &e2->idx.oid);
64 struct object_entry_pool {
65 struct object_entry_pool *next_pool;
66 struct object_entry *next_free;
67 struct object_entry *end;
68 struct object_entry entries[FLEX_ARRAY]; /* more */
71 struct mark_set {
72 union {
73 struct object_id *oids[1024];
74 struct object_entry *marked[1024];
75 struct mark_set *sets[1024];
76 } data;
77 unsigned int shift;
80 struct last_object {
81 struct strbuf data;
82 off_t offset;
83 unsigned int depth;
84 unsigned no_swap : 1;
87 struct atom_str {
88 struct atom_str *next_atom;
89 unsigned short str_len;
90 char str_dat[FLEX_ARRAY]; /* more */
93 struct tree_content;
94 struct tree_entry {
95 struct tree_content *tree;
96 struct atom_str *name;
97 struct tree_entry_ms {
98 uint16_t mode;
99 struct object_id oid;
100 } versions[2];
103 struct tree_content {
104 unsigned int entry_capacity; /* must match avail_tree_content */
105 unsigned int entry_count;
106 unsigned int delta_depth;
107 struct tree_entry *entries[FLEX_ARRAY]; /* more */
110 struct avail_tree_content {
111 unsigned int entry_capacity; /* must match tree_content */
112 struct avail_tree_content *next_avail;
115 struct branch {
116 struct branch *table_next_branch;
117 struct branch *active_next_branch;
118 const char *name;
119 struct tree_entry branch_tree;
120 uintmax_t last_commit;
121 uintmax_t num_notes;
122 unsigned active : 1;
123 unsigned delete : 1;
124 unsigned pack_id : PACK_ID_BITS;
125 struct object_id oid;
128 struct tag {
129 struct tag *next_tag;
130 const char *name;
131 unsigned int pack_id;
132 struct object_id oid;
135 struct hash_list {
136 struct hash_list *next;
137 struct object_id oid;
140 typedef enum {
141 WHENSPEC_RAW = 1,
142 WHENSPEC_RAW_PERMISSIVE,
143 WHENSPEC_RFC2822,
144 WHENSPEC_NOW
145 } whenspec_type;
147 struct recent_command {
148 struct recent_command *prev;
149 struct recent_command *next;
150 char *buf;
153 typedef void (*mark_set_inserter_t)(struct mark_set **s, struct object_id *oid, uintmax_t mark);
154 typedef void (*each_mark_fn_t)(uintmax_t mark, void *obj, void *cbp);
156 /* Configured limits on output */
157 static unsigned long max_depth = 50;
158 static off_t max_packsize;
159 static int unpack_limit = 100;
160 static int force_update;
162 /* Stats and misc. counters */
163 static uintmax_t alloc_count;
164 static uintmax_t marks_set_count;
165 static uintmax_t object_count_by_type[1 << TYPE_BITS];
166 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
167 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
168 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
169 static unsigned long object_count;
170 static unsigned long branch_count;
171 static unsigned long branch_load_count;
172 static int failure;
173 static FILE *pack_edges;
174 static unsigned int show_stats = 1;
175 static int global_argc;
176 static const char **global_argv;
178 /* Memory pools */
179 static struct mem_pool fi_mem_pool = {NULL, 2*1024*1024 -
180 sizeof(struct mp_block), 0 };
182 /* Atom management */
183 static unsigned int atom_table_sz = 4451;
184 static unsigned int atom_cnt;
185 static struct atom_str **atom_table;
187 /* The .pack file being generated */
188 static struct pack_idx_option pack_idx_opts;
189 static unsigned int pack_id;
190 static struct hashfile *pack_file;
191 static struct packed_git *pack_data;
192 static struct packed_git **all_packs;
193 static off_t pack_size;
195 /* Table of objects we've written. */
196 static unsigned int object_entry_alloc = 5000;
197 static struct object_entry_pool *blocks;
198 static struct hashmap object_table;
199 static struct mark_set *marks;
200 static const char *export_marks_file;
201 static const char *import_marks_file;
202 static int import_marks_file_from_stream;
203 static int import_marks_file_ignore_missing;
204 static int import_marks_file_done;
205 static int relative_marks_paths;
207 /* Our last blob */
208 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
210 /* Tree management */
211 static unsigned int tree_entry_alloc = 1000;
212 static void *avail_tree_entry;
213 static unsigned int avail_tree_table_sz = 100;
214 static struct avail_tree_content **avail_tree_table;
215 static size_t tree_entry_allocd;
216 static struct strbuf old_tree = STRBUF_INIT;
217 static struct strbuf new_tree = STRBUF_INIT;
219 /* Branch data */
220 static unsigned long max_active_branches = 5;
221 static unsigned long cur_active_branches;
222 static unsigned long branch_table_sz = 1039;
223 static struct branch **branch_table;
224 static struct branch *active_branches;
226 /* Tag data */
227 static struct tag *first_tag;
228 static struct tag *last_tag;
230 /* Input stream parsing */
231 static whenspec_type whenspec = WHENSPEC_RAW;
232 static struct strbuf command_buf = STRBUF_INIT;
233 static int unread_command_buf;
234 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
235 static struct recent_command *cmd_tail = &cmd_hist;
236 static struct recent_command *rc_free;
237 static unsigned int cmd_save = 100;
238 static uintmax_t next_mark;
239 static struct strbuf new_data = STRBUF_INIT;
240 static int seen_data_command;
241 static int require_explicit_termination;
242 static int allow_unsafe_features;
244 /* Signal handling */
245 static volatile sig_atomic_t checkpoint_requested;
247 /* Submodule marks */
248 static struct string_list sub_marks_from = STRING_LIST_INIT_DUP;
249 static struct string_list sub_marks_to = STRING_LIST_INIT_DUP;
250 static kh_oid_map_t *sub_oid_map;
252 /* Where to write output of cat-blob commands */
253 static int cat_blob_fd = STDOUT_FILENO;
255 static void parse_argv(void);
256 static void parse_get_mark(const char *p);
257 static void parse_cat_blob(const char *p);
258 static void parse_ls(const char *p, struct branch *b);
260 static void for_each_mark(struct mark_set *m, uintmax_t base, each_mark_fn_t callback, void *p)
262 uintmax_t k;
263 if (m->shift) {
264 for (k = 0; k < 1024; k++) {
265 if (m->data.sets[k])
266 for_each_mark(m->data.sets[k], base + (k << m->shift), callback, p);
268 } else {
269 for (k = 0; k < 1024; k++) {
270 if (m->data.marked[k])
271 callback(base + k, m->data.marked[k], p);
276 static void dump_marks_fn(uintmax_t mark, void *object, void *cbp) {
277 struct object_entry *e = object;
278 FILE *f = cbp;
280 fprintf(f, ":%" PRIuMAX " %s\n", mark, oid_to_hex(&e->idx.oid));
283 static void write_branch_report(FILE *rpt, struct branch *b)
285 fprintf(rpt, "%s:\n", b->name);
287 fprintf(rpt, " status :");
288 if (b->active)
289 fputs(" active", rpt);
290 if (b->branch_tree.tree)
291 fputs(" loaded", rpt);
292 if (is_null_oid(&b->branch_tree.versions[1].oid))
293 fputs(" dirty", rpt);
294 fputc('\n', rpt);
296 fprintf(rpt, " tip commit : %s\n", oid_to_hex(&b->oid));
297 fprintf(rpt, " old tree : %s\n",
298 oid_to_hex(&b->branch_tree.versions[0].oid));
299 fprintf(rpt, " cur tree : %s\n",
300 oid_to_hex(&b->branch_tree.versions[1].oid));
301 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
303 fputs(" last pack : ", rpt);
304 if (b->pack_id < MAX_PACK_ID)
305 fprintf(rpt, "%u", b->pack_id);
306 fputc('\n', rpt);
308 fputc('\n', rpt);
311 static void write_crash_report(const char *err)
313 char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
314 FILE *rpt = fopen(loc, "w");
315 struct branch *b;
316 unsigned long lu;
317 struct recent_command *rc;
319 if (!rpt) {
320 error_errno("can't write crash report %s", loc);
321 free(loc);
322 return;
325 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
327 fprintf(rpt, "fast-import crash report:\n");
328 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
329 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
330 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
331 fputc('\n', rpt);
333 fputs("fatal: ", rpt);
334 fputs(err, rpt);
335 fputc('\n', rpt);
337 fputc('\n', rpt);
338 fputs("Most Recent Commands Before Crash\n", rpt);
339 fputs("---------------------------------\n", rpt);
340 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
341 if (rc->next == &cmd_hist)
342 fputs("* ", rpt);
343 else
344 fputs(" ", rpt);
345 fputs(rc->buf, rpt);
346 fputc('\n', rpt);
349 fputc('\n', rpt);
350 fputs("Active Branch LRU\n", rpt);
351 fputs("-----------------\n", rpt);
352 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
353 cur_active_branches,
354 max_active_branches);
355 fputc('\n', rpt);
356 fputs(" pos clock name\n", rpt);
357 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
358 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
359 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
360 ++lu, b->last_commit, b->name);
362 fputc('\n', rpt);
363 fputs("Inactive Branches\n", rpt);
364 fputs("-----------------\n", rpt);
365 for (lu = 0; lu < branch_table_sz; lu++) {
366 for (b = branch_table[lu]; b; b = b->table_next_branch)
367 write_branch_report(rpt, b);
370 if (first_tag) {
371 struct tag *tg;
372 fputc('\n', rpt);
373 fputs("Annotated Tags\n", rpt);
374 fputs("--------------\n", rpt);
375 for (tg = first_tag; tg; tg = tg->next_tag) {
376 fputs(oid_to_hex(&tg->oid), rpt);
377 fputc(' ', rpt);
378 fputs(tg->name, rpt);
379 fputc('\n', rpt);
383 fputc('\n', rpt);
384 fputs("Marks\n", rpt);
385 fputs("-----\n", rpt);
386 if (export_marks_file)
387 fprintf(rpt, " exported to %s\n", export_marks_file);
388 else
389 for_each_mark(marks, 0, dump_marks_fn, rpt);
391 fputc('\n', rpt);
392 fputs("-------------------\n", rpt);
393 fputs("END OF CRASH REPORT\n", rpt);
394 fclose(rpt);
395 free(loc);
398 static void end_packfile(void);
399 static void unkeep_all_packs(void);
400 static void dump_marks(void);
402 static NORETURN void die_nicely(const char *err, va_list params)
404 static int zombie;
405 char message[2 * PATH_MAX];
407 vsnprintf(message, sizeof(message), err, params);
408 fputs("fatal: ", stderr);
409 fputs(message, stderr);
410 fputc('\n', stderr);
412 if (!zombie) {
413 zombie = 1;
414 write_crash_report(message);
415 end_packfile();
416 unkeep_all_packs();
417 dump_marks();
419 exit(128);
422 #ifndef SIGUSR1 /* Windows, for example */
424 static void set_checkpoint_signal(void)
428 #else
430 static void checkpoint_signal(int signo)
432 checkpoint_requested = 1;
435 static void set_checkpoint_signal(void)
437 struct sigaction sa;
439 memset(&sa, 0, sizeof(sa));
440 sa.sa_handler = checkpoint_signal;
441 sigemptyset(&sa.sa_mask);
442 sa.sa_flags = SA_RESTART;
443 sigaction(SIGUSR1, &sa, NULL);
446 #endif
448 static void alloc_objects(unsigned int cnt)
450 struct object_entry_pool *b;
452 b = xmalloc(sizeof(struct object_entry_pool)
453 + cnt * sizeof(struct object_entry));
454 b->next_pool = blocks;
455 b->next_free = b->entries;
456 b->end = b->entries + cnt;
457 blocks = b;
458 alloc_count += cnt;
461 static struct object_entry *new_object(struct object_id *oid)
463 struct object_entry *e;
465 if (blocks->next_free == blocks->end)
466 alloc_objects(object_entry_alloc);
468 e = blocks->next_free++;
469 oidcpy(&e->idx.oid, oid);
470 return e;
473 static struct object_entry *find_object(struct object_id *oid)
475 return hashmap_get_entry_from_hash(&object_table, oidhash(oid), oid,
476 struct object_entry, ent);
479 static struct object_entry *insert_object(struct object_id *oid)
481 struct object_entry *e;
482 unsigned int hash = oidhash(oid);
484 e = hashmap_get_entry_from_hash(&object_table, hash, oid,
485 struct object_entry, ent);
486 if (!e) {
487 e = new_object(oid);
488 e->idx.offset = 0;
489 hashmap_entry_init(&e->ent, hash);
490 hashmap_add(&object_table, &e->ent);
493 return e;
496 static void invalidate_pack_id(unsigned int id)
498 unsigned long lu;
499 struct tag *t;
500 struct hashmap_iter iter;
501 struct object_entry *e;
503 hashmap_for_each_entry(&object_table, &iter, e, ent) {
504 if (e->pack_id == id)
505 e->pack_id = MAX_PACK_ID;
508 for (lu = 0; lu < branch_table_sz; lu++) {
509 struct branch *b;
511 for (b = branch_table[lu]; b; b = b->table_next_branch)
512 if (b->pack_id == id)
513 b->pack_id = MAX_PACK_ID;
516 for (t = first_tag; t; t = t->next_tag)
517 if (t->pack_id == id)
518 t->pack_id = MAX_PACK_ID;
521 static unsigned int hc_str(const char *s, size_t len)
523 unsigned int r = 0;
524 while (len-- > 0)
525 r = r * 31 + *s++;
526 return r;
529 static void insert_mark(struct mark_set **top, uintmax_t idnum, struct object_entry *oe)
531 struct mark_set *s = *top;
533 while ((idnum >> s->shift) >= 1024) {
534 s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
535 s->shift = (*top)->shift + 10;
536 s->data.sets[0] = *top;
537 *top = s;
539 while (s->shift) {
540 uintmax_t i = idnum >> s->shift;
541 idnum -= i << s->shift;
542 if (!s->data.sets[i]) {
543 s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
544 s->data.sets[i]->shift = s->shift - 10;
546 s = s->data.sets[i];
548 if (!s->data.marked[idnum])
549 marks_set_count++;
550 s->data.marked[idnum] = oe;
553 static void *find_mark(struct mark_set *s, uintmax_t idnum)
555 uintmax_t orig_idnum = idnum;
556 struct object_entry *oe = NULL;
557 if ((idnum >> s->shift) < 1024) {
558 while (s && s->shift) {
559 uintmax_t i = idnum >> s->shift;
560 idnum -= i << s->shift;
561 s = s->data.sets[i];
563 if (s)
564 oe = s->data.marked[idnum];
566 if (!oe)
567 die("mark :%" PRIuMAX " not declared", orig_idnum);
568 return oe;
571 static struct atom_str *to_atom(const char *s, unsigned short len)
573 unsigned int hc = hc_str(s, len) % atom_table_sz;
574 struct atom_str *c;
576 for (c = atom_table[hc]; c; c = c->next_atom)
577 if (c->str_len == len && !strncmp(s, c->str_dat, len))
578 return c;
580 c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
581 c->str_len = len;
582 memcpy(c->str_dat, s, len);
583 c->str_dat[len] = 0;
584 c->next_atom = atom_table[hc];
585 atom_table[hc] = c;
586 atom_cnt++;
587 return c;
590 static struct branch *lookup_branch(const char *name)
592 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
593 struct branch *b;
595 for (b = branch_table[hc]; b; b = b->table_next_branch)
596 if (!strcmp(name, b->name))
597 return b;
598 return NULL;
601 static struct branch *new_branch(const char *name)
603 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
604 struct branch *b = lookup_branch(name);
606 if (b)
607 die("Invalid attempt to create duplicate branch: %s", name);
608 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
609 die("Branch name doesn't conform to GIT standards: %s", name);
611 b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
612 b->name = mem_pool_strdup(&fi_mem_pool, name);
613 b->table_next_branch = branch_table[hc];
614 b->branch_tree.versions[0].mode = S_IFDIR;
615 b->branch_tree.versions[1].mode = S_IFDIR;
616 b->num_notes = 0;
617 b->active = 0;
618 b->pack_id = MAX_PACK_ID;
619 branch_table[hc] = b;
620 branch_count++;
621 return b;
624 static unsigned int hc_entries(unsigned int cnt)
626 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
627 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
630 static struct tree_content *new_tree_content(unsigned int cnt)
632 struct avail_tree_content *f, *l = NULL;
633 struct tree_content *t;
634 unsigned int hc = hc_entries(cnt);
636 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
637 if (f->entry_capacity >= cnt)
638 break;
640 if (f) {
641 if (l)
642 l->next_avail = f->next_avail;
643 else
644 avail_tree_table[hc] = f->next_avail;
645 } else {
646 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
647 f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
648 f->entry_capacity = cnt;
651 t = (struct tree_content*)f;
652 t->entry_count = 0;
653 t->delta_depth = 0;
654 return t;
657 static void release_tree_entry(struct tree_entry *e);
658 static void release_tree_content(struct tree_content *t)
660 struct avail_tree_content *f = (struct avail_tree_content*)t;
661 unsigned int hc = hc_entries(f->entry_capacity);
662 f->next_avail = avail_tree_table[hc];
663 avail_tree_table[hc] = f;
666 static void release_tree_content_recursive(struct tree_content *t)
668 unsigned int i;
669 for (i = 0; i < t->entry_count; i++)
670 release_tree_entry(t->entries[i]);
671 release_tree_content(t);
674 static struct tree_content *grow_tree_content(
675 struct tree_content *t,
676 int amt)
678 struct tree_content *r = new_tree_content(t->entry_count + amt);
679 r->entry_count = t->entry_count;
680 r->delta_depth = t->delta_depth;
681 COPY_ARRAY(r->entries, t->entries, t->entry_count);
682 release_tree_content(t);
683 return r;
686 static struct tree_entry *new_tree_entry(void)
688 struct tree_entry *e;
690 if (!avail_tree_entry) {
691 unsigned int n = tree_entry_alloc;
692 tree_entry_allocd += n * sizeof(struct tree_entry);
693 ALLOC_ARRAY(e, n);
694 avail_tree_entry = e;
695 while (n-- > 1) {
696 *((void**)e) = e + 1;
697 e++;
699 *((void**)e) = NULL;
702 e = avail_tree_entry;
703 avail_tree_entry = *((void**)e);
704 return e;
707 static void release_tree_entry(struct tree_entry *e)
709 if (e->tree)
710 release_tree_content_recursive(e->tree);
711 *((void**)e) = avail_tree_entry;
712 avail_tree_entry = e;
715 static struct tree_content *dup_tree_content(struct tree_content *s)
717 struct tree_content *d;
718 struct tree_entry *a, *b;
719 unsigned int i;
721 if (!s)
722 return NULL;
723 d = new_tree_content(s->entry_count);
724 for (i = 0; i < s->entry_count; i++) {
725 a = s->entries[i];
726 b = new_tree_entry();
727 memcpy(b, a, sizeof(*a));
728 if (a->tree && is_null_oid(&b->versions[1].oid))
729 b->tree = dup_tree_content(a->tree);
730 else
731 b->tree = NULL;
732 d->entries[i] = b;
734 d->entry_count = s->entry_count;
735 d->delta_depth = s->delta_depth;
737 return d;
740 static void start_packfile(void)
742 struct strbuf tmp_file = STRBUF_INIT;
743 struct packed_git *p;
744 int pack_fd;
746 pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
747 FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
748 strbuf_release(&tmp_file);
750 p->pack_fd = pack_fd;
751 p->do_not_close = 1;
752 pack_file = hashfd(pack_fd, p->pack_name);
754 pack_data = p;
755 pack_size = write_pack_header(pack_file, 0);
756 object_count = 0;
758 REALLOC_ARRAY(all_packs, pack_id + 1);
759 all_packs[pack_id] = p;
762 static const char *create_index(void)
764 const char *tmpfile;
765 struct pack_idx_entry **idx, **c, **last;
766 struct object_entry *e;
767 struct object_entry_pool *o;
769 /* Build the table of object IDs. */
770 ALLOC_ARRAY(idx, object_count);
771 c = idx;
772 for (o = blocks; o; o = o->next_pool)
773 for (e = o->next_free; e-- != o->entries;)
774 if (pack_id == e->pack_id)
775 *c++ = &e->idx;
776 last = idx + object_count;
777 if (c != last)
778 die("internal consistency error creating the index");
780 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts,
781 pack_data->hash);
782 free(idx);
783 return tmpfile;
786 static char *keep_pack(const char *curr_index_name)
788 static const char *keep_msg = "fast-import";
789 struct strbuf name = STRBUF_INIT;
790 int keep_fd;
792 odb_pack_name(&name, pack_data->hash, "keep");
793 keep_fd = odb_pack_keep(name.buf);
794 if (keep_fd < 0)
795 die_errno("cannot create keep file");
796 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
797 if (close(keep_fd))
798 die_errno("failed to write keep file");
800 odb_pack_name(&name, pack_data->hash, "pack");
801 if (finalize_object_file(pack_data->pack_name, name.buf))
802 die("cannot store pack file");
804 odb_pack_name(&name, pack_data->hash, "idx");
805 if (finalize_object_file(curr_index_name, name.buf))
806 die("cannot store index file");
807 free((void *)curr_index_name);
808 return strbuf_detach(&name, NULL);
811 static void unkeep_all_packs(void)
813 struct strbuf name = STRBUF_INIT;
814 int k;
816 for (k = 0; k < pack_id; k++) {
817 struct packed_git *p = all_packs[k];
818 odb_pack_name(&name, p->hash, "keep");
819 unlink_or_warn(name.buf);
821 strbuf_release(&name);
824 static int loosen_small_pack(const struct packed_git *p)
826 struct child_process unpack = CHILD_PROCESS_INIT;
828 if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
829 die_errno("Failed seeking to start of '%s'", p->pack_name);
831 unpack.in = p->pack_fd;
832 unpack.git_cmd = 1;
833 unpack.stdout_to_stderr = 1;
834 strvec_push(&unpack.args, "unpack-objects");
835 if (!show_stats)
836 strvec_push(&unpack.args, "-q");
838 return run_command(&unpack);
841 static void end_packfile(void)
843 static int running;
845 if (running || !pack_data)
846 return;
848 running = 1;
849 clear_delta_base_cache();
850 if (object_count) {
851 struct packed_git *new_p;
852 struct object_id cur_pack_oid;
853 char *idx_name;
854 int i;
855 struct branch *b;
856 struct tag *t;
858 close_pack_windows(pack_data);
859 finalize_hashfile(pack_file, cur_pack_oid.hash, 0);
860 fixup_pack_header_footer(pack_data->pack_fd, pack_data->hash,
861 pack_data->pack_name, object_count,
862 cur_pack_oid.hash, pack_size);
864 if (object_count <= unpack_limit) {
865 if (!loosen_small_pack(pack_data)) {
866 invalidate_pack_id(pack_id);
867 goto discard_pack;
871 close(pack_data->pack_fd);
872 idx_name = keep_pack(create_index());
874 /* Register the packfile with core git's machinery. */
875 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
876 if (!new_p)
877 die("core git rejected index %s", idx_name);
878 all_packs[pack_id] = new_p;
879 install_packed_git(the_repository, new_p);
880 free(idx_name);
882 /* Print the boundary */
883 if (pack_edges) {
884 fprintf(pack_edges, "%s:", new_p->pack_name);
885 for (i = 0; i < branch_table_sz; i++) {
886 for (b = branch_table[i]; b; b = b->table_next_branch) {
887 if (b->pack_id == pack_id)
888 fprintf(pack_edges, " %s",
889 oid_to_hex(&b->oid));
892 for (t = first_tag; t; t = t->next_tag) {
893 if (t->pack_id == pack_id)
894 fprintf(pack_edges, " %s",
895 oid_to_hex(&t->oid));
897 fputc('\n', pack_edges);
898 fflush(pack_edges);
901 pack_id++;
903 else {
904 discard_pack:
905 close(pack_data->pack_fd);
906 unlink_or_warn(pack_data->pack_name);
908 FREE_AND_NULL(pack_data);
909 running = 0;
911 /* We can't carry a delta across packfiles. */
912 strbuf_release(&last_blob.data);
913 last_blob.offset = 0;
914 last_blob.depth = 0;
917 static void cycle_packfile(void)
919 end_packfile();
920 start_packfile();
923 static int store_object(
924 enum object_type type,
925 struct strbuf *dat,
926 struct last_object *last,
927 struct object_id *oidout,
928 uintmax_t mark)
930 void *out, *delta;
931 struct object_entry *e;
932 unsigned char hdr[96];
933 struct object_id oid;
934 unsigned long hdrlen, deltalen;
935 git_hash_ctx c;
936 git_zstream s;
938 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
939 type_name(type), (unsigned long)dat->len) + 1;
940 the_hash_algo->init_fn(&c);
941 the_hash_algo->update_fn(&c, hdr, hdrlen);
942 the_hash_algo->update_fn(&c, dat->buf, dat->len);
943 the_hash_algo->final_oid_fn(&oid, &c);
944 if (oidout)
945 oidcpy(oidout, &oid);
947 e = insert_object(&oid);
948 if (mark)
949 insert_mark(&marks, mark, e);
950 if (e->idx.offset) {
951 duplicate_count_by_type[type]++;
952 return 1;
953 } else if (find_sha1_pack(oid.hash,
954 get_all_packs(the_repository))) {
955 e->type = type;
956 e->pack_id = MAX_PACK_ID;
957 e->idx.offset = 1; /* just not zero! */
958 duplicate_count_by_type[type]++;
959 return 1;
962 if (last && last->data.len && last->data.buf && last->depth < max_depth
963 && dat->len > the_hash_algo->rawsz) {
965 delta_count_attempts_by_type[type]++;
966 delta = diff_delta(last->data.buf, last->data.len,
967 dat->buf, dat->len,
968 &deltalen, dat->len - the_hash_algo->rawsz);
969 } else
970 delta = NULL;
972 git_deflate_init(&s, pack_compression_level);
973 if (delta) {
974 s.next_in = delta;
975 s.avail_in = deltalen;
976 } else {
977 s.next_in = (void *)dat->buf;
978 s.avail_in = dat->len;
980 s.avail_out = git_deflate_bound(&s, s.avail_in);
981 s.next_out = out = xmalloc(s.avail_out);
982 while (git_deflate(&s, Z_FINISH) == Z_OK)
983 ; /* nothing */
984 git_deflate_end(&s);
986 /* Determine if we should auto-checkpoint. */
987 if ((max_packsize
988 && (pack_size + PACK_SIZE_THRESHOLD + s.total_out) > max_packsize)
989 || (pack_size + PACK_SIZE_THRESHOLD + s.total_out) < pack_size) {
991 /* This new object needs to *not* have the current pack_id. */
992 e->pack_id = pack_id + 1;
993 cycle_packfile();
995 /* We cannot carry a delta into the new pack. */
996 if (delta) {
997 FREE_AND_NULL(delta);
999 git_deflate_init(&s, pack_compression_level);
1000 s.next_in = (void *)dat->buf;
1001 s.avail_in = dat->len;
1002 s.avail_out = git_deflate_bound(&s, s.avail_in);
1003 s.next_out = out = xrealloc(out, s.avail_out);
1004 while (git_deflate(&s, Z_FINISH) == Z_OK)
1005 ; /* nothing */
1006 git_deflate_end(&s);
1010 e->type = type;
1011 e->pack_id = pack_id;
1012 e->idx.offset = pack_size;
1013 object_count++;
1014 object_count_by_type[type]++;
1016 crc32_begin(pack_file);
1018 if (delta) {
1019 off_t ofs = e->idx.offset - last->offset;
1020 unsigned pos = sizeof(hdr) - 1;
1022 delta_count_by_type[type]++;
1023 e->depth = last->depth + 1;
1025 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1026 OBJ_OFS_DELTA, deltalen);
1027 hashwrite(pack_file, hdr, hdrlen);
1028 pack_size += hdrlen;
1030 hdr[pos] = ofs & 127;
1031 while (ofs >>= 7)
1032 hdr[--pos] = 128 | (--ofs & 127);
1033 hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1034 pack_size += sizeof(hdr) - pos;
1035 } else {
1036 e->depth = 0;
1037 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1038 type, dat->len);
1039 hashwrite(pack_file, hdr, hdrlen);
1040 pack_size += hdrlen;
1043 hashwrite(pack_file, out, s.total_out);
1044 pack_size += s.total_out;
1046 e->idx.crc32 = crc32_end(pack_file);
1048 free(out);
1049 free(delta);
1050 if (last) {
1051 if (last->no_swap) {
1052 last->data = *dat;
1053 } else {
1054 strbuf_swap(&last->data, dat);
1056 last->offset = e->idx.offset;
1057 last->depth = e->depth;
1059 return 0;
1062 static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1064 if (hashfile_truncate(pack_file, checkpoint))
1065 die_errno("cannot truncate pack to skip duplicate");
1066 pack_size = checkpoint->offset;
1069 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1071 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1072 unsigned char *in_buf = xmalloc(in_sz);
1073 unsigned char *out_buf = xmalloc(out_sz);
1074 struct object_entry *e;
1075 struct object_id oid;
1076 unsigned long hdrlen;
1077 off_t offset;
1078 git_hash_ctx c;
1079 git_zstream s;
1080 struct hashfile_checkpoint checkpoint;
1081 int status = Z_OK;
1083 /* Determine if we should auto-checkpoint. */
1084 if ((max_packsize
1085 && (pack_size + PACK_SIZE_THRESHOLD + len) > max_packsize)
1086 || (pack_size + PACK_SIZE_THRESHOLD + len) < pack_size)
1087 cycle_packfile();
1089 hashfile_checkpoint(pack_file, &checkpoint);
1090 offset = checkpoint.offset;
1092 hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1094 the_hash_algo->init_fn(&c);
1095 the_hash_algo->update_fn(&c, out_buf, hdrlen);
1097 crc32_begin(pack_file);
1099 git_deflate_init(&s, pack_compression_level);
1101 hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1103 s.next_out = out_buf + hdrlen;
1104 s.avail_out = out_sz - hdrlen;
1106 while (status != Z_STREAM_END) {
1107 if (0 < len && !s.avail_in) {
1108 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1109 size_t n = fread(in_buf, 1, cnt, stdin);
1110 if (!n && feof(stdin))
1111 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1113 the_hash_algo->update_fn(&c, in_buf, n);
1114 s.next_in = in_buf;
1115 s.avail_in = n;
1116 len -= n;
1119 status = git_deflate(&s, len ? 0 : Z_FINISH);
1121 if (!s.avail_out || status == Z_STREAM_END) {
1122 size_t n = s.next_out - out_buf;
1123 hashwrite(pack_file, out_buf, n);
1124 pack_size += n;
1125 s.next_out = out_buf;
1126 s.avail_out = out_sz;
1129 switch (status) {
1130 case Z_OK:
1131 case Z_BUF_ERROR:
1132 case Z_STREAM_END:
1133 continue;
1134 default:
1135 die("unexpected deflate failure: %d", status);
1138 git_deflate_end(&s);
1139 the_hash_algo->final_oid_fn(&oid, &c);
1141 if (oidout)
1142 oidcpy(oidout, &oid);
1144 e = insert_object(&oid);
1146 if (mark)
1147 insert_mark(&marks, mark, e);
1149 if (e->idx.offset) {
1150 duplicate_count_by_type[OBJ_BLOB]++;
1151 truncate_pack(&checkpoint);
1153 } else if (find_sha1_pack(oid.hash,
1154 get_all_packs(the_repository))) {
1155 e->type = OBJ_BLOB;
1156 e->pack_id = MAX_PACK_ID;
1157 e->idx.offset = 1; /* just not zero! */
1158 duplicate_count_by_type[OBJ_BLOB]++;
1159 truncate_pack(&checkpoint);
1161 } else {
1162 e->depth = 0;
1163 e->type = OBJ_BLOB;
1164 e->pack_id = pack_id;
1165 e->idx.offset = offset;
1166 e->idx.crc32 = crc32_end(pack_file);
1167 object_count++;
1168 object_count_by_type[OBJ_BLOB]++;
1171 free(in_buf);
1172 free(out_buf);
1175 /* All calls must be guarded by find_object() or find_mark() to
1176 * ensure the 'struct object_entry' passed was written by this
1177 * process instance. We unpack the entry by the offset, avoiding
1178 * the need for the corresponding .idx file. This unpacking rule
1179 * works because we only use OBJ_REF_DELTA within the packfiles
1180 * created by fast-import.
1182 * oe must not be NULL. Such an oe usually comes from giving
1183 * an unknown SHA-1 to find_object() or an undefined mark to
1184 * find_mark(). Callers must test for this condition and use
1185 * the standard read_sha1_file() when it happens.
1187 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1188 * find_mark(), where the mark was reloaded from an existing marks
1189 * file and is referencing an object that this fast-import process
1190 * instance did not write out to a packfile. Callers must test for
1191 * this condition and use read_sha1_file() instead.
1193 static void *gfi_unpack_entry(
1194 struct object_entry *oe,
1195 unsigned long *sizep)
1197 enum object_type type;
1198 struct packed_git *p = all_packs[oe->pack_id];
1199 if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1200 /* The object is stored in the packfile we are writing to
1201 * and we have modified it since the last time we scanned
1202 * back to read a previously written object. If an old
1203 * window covered [p->pack_size, p->pack_size + rawsz) its
1204 * data is stale and is not valid. Closing all windows
1205 * and updating the packfile length ensures we can read
1206 * the newly written data.
1208 close_pack_windows(p);
1209 hashflush(pack_file);
1211 /* We have to offer rawsz bytes additional on the end of
1212 * the packfile as the core unpacker code assumes the
1213 * footer is present at the file end and must promise
1214 * at least rawsz bytes within any window it maps. But
1215 * we don't actually create the footer here.
1217 p->pack_size = pack_size + the_hash_algo->rawsz;
1219 return unpack_entry(the_repository, p, oe->idx.offset, &type, sizep);
1222 static const char *get_mode(const char *str, uint16_t *modep)
1224 unsigned char c;
1225 uint16_t mode = 0;
1227 while ((c = *str++) != ' ') {
1228 if (c < '0' || c > '7')
1229 return NULL;
1230 mode = (mode << 3) + (c - '0');
1232 *modep = mode;
1233 return str;
1236 static void load_tree(struct tree_entry *root)
1238 struct object_id *oid = &root->versions[1].oid;
1239 struct object_entry *myoe;
1240 struct tree_content *t;
1241 unsigned long size;
1242 char *buf;
1243 const char *c;
1245 root->tree = t = new_tree_content(8);
1246 if (is_null_oid(oid))
1247 return;
1249 myoe = find_object(oid);
1250 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1251 if (myoe->type != OBJ_TREE)
1252 die("Not a tree: %s", oid_to_hex(oid));
1253 t->delta_depth = myoe->depth;
1254 buf = gfi_unpack_entry(myoe, &size);
1255 if (!buf)
1256 die("Can't load tree %s", oid_to_hex(oid));
1257 } else {
1258 enum object_type type;
1259 buf = read_object_file(oid, &type, &size);
1260 if (!buf || type != OBJ_TREE)
1261 die("Can't load tree %s", oid_to_hex(oid));
1264 c = buf;
1265 while (c != (buf + size)) {
1266 struct tree_entry *e = new_tree_entry();
1268 if (t->entry_count == t->entry_capacity)
1269 root->tree = t = grow_tree_content(t, t->entry_count);
1270 t->entries[t->entry_count++] = e;
1272 e->tree = NULL;
1273 c = get_mode(c, &e->versions[1].mode);
1274 if (!c)
1275 die("Corrupt mode in %s", oid_to_hex(oid));
1276 e->versions[0].mode = e->versions[1].mode;
1277 e->name = to_atom(c, strlen(c));
1278 c += e->name->str_len + 1;
1279 oidread(&e->versions[0].oid, (unsigned char *)c);
1280 oidread(&e->versions[1].oid, (unsigned char *)c);
1281 c += the_hash_algo->rawsz;
1283 free(buf);
1286 static int tecmp0 (const void *_a, const void *_b)
1288 struct tree_entry *a = *((struct tree_entry**)_a);
1289 struct tree_entry *b = *((struct tree_entry**)_b);
1290 return base_name_compare(
1291 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1292 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1295 static int tecmp1 (const void *_a, const void *_b)
1297 struct tree_entry *a = *((struct tree_entry**)_a);
1298 struct tree_entry *b = *((struct tree_entry**)_b);
1299 return base_name_compare(
1300 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1301 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1304 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1306 size_t maxlen = 0;
1307 unsigned int i;
1309 if (!v)
1310 QSORT(t->entries, t->entry_count, tecmp0);
1311 else
1312 QSORT(t->entries, t->entry_count, tecmp1);
1314 for (i = 0; i < t->entry_count; i++) {
1315 if (t->entries[i]->versions[v].mode)
1316 maxlen += t->entries[i]->name->str_len + 34;
1319 strbuf_reset(b);
1320 strbuf_grow(b, maxlen);
1321 for (i = 0; i < t->entry_count; i++) {
1322 struct tree_entry *e = t->entries[i];
1323 if (!e->versions[v].mode)
1324 continue;
1325 strbuf_addf(b, "%o %s%c",
1326 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1327 e->name->str_dat, '\0');
1328 strbuf_add(b, e->versions[v].oid.hash, the_hash_algo->rawsz);
1332 static void store_tree(struct tree_entry *root)
1334 struct tree_content *t;
1335 unsigned int i, j, del;
1336 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1337 struct object_entry *le = NULL;
1339 if (!is_null_oid(&root->versions[1].oid))
1340 return;
1342 if (!root->tree)
1343 load_tree(root);
1344 t = root->tree;
1346 for (i = 0; i < t->entry_count; i++) {
1347 if (t->entries[i]->tree)
1348 store_tree(t->entries[i]);
1351 if (!(root->versions[0].mode & NO_DELTA))
1352 le = find_object(&root->versions[0].oid);
1353 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1354 mktree(t, 0, &old_tree);
1355 lo.data = old_tree;
1356 lo.offset = le->idx.offset;
1357 lo.depth = t->delta_depth;
1360 mktree(t, 1, &new_tree);
1361 store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1363 t->delta_depth = lo.depth;
1364 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1365 struct tree_entry *e = t->entries[i];
1366 if (e->versions[1].mode) {
1367 e->versions[0].mode = e->versions[1].mode;
1368 oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1369 t->entries[j++] = e;
1370 } else {
1371 release_tree_entry(e);
1372 del++;
1375 t->entry_count -= del;
1378 static void tree_content_replace(
1379 struct tree_entry *root,
1380 const struct object_id *oid,
1381 const uint16_t mode,
1382 struct tree_content *newtree)
1384 if (!S_ISDIR(mode))
1385 die("Root cannot be a non-directory");
1386 oidclr(&root->versions[0].oid);
1387 oidcpy(&root->versions[1].oid, oid);
1388 if (root->tree)
1389 release_tree_content_recursive(root->tree);
1390 root->tree = newtree;
1393 static int tree_content_set(
1394 struct tree_entry *root,
1395 const char *p,
1396 const struct object_id *oid,
1397 const uint16_t mode,
1398 struct tree_content *subtree)
1400 struct tree_content *t;
1401 const char *slash1;
1402 unsigned int i, n;
1403 struct tree_entry *e;
1405 slash1 = strchrnul(p, '/');
1406 n = slash1 - p;
1407 if (!n)
1408 die("Empty path component found in input");
1409 if (!*slash1 && !S_ISDIR(mode) && subtree)
1410 die("Non-directories cannot have subtrees");
1412 if (!root->tree)
1413 load_tree(root);
1414 t = root->tree;
1415 for (i = 0; i < t->entry_count; i++) {
1416 e = t->entries[i];
1417 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1418 if (!*slash1) {
1419 if (!S_ISDIR(mode)
1420 && e->versions[1].mode == mode
1421 && oideq(&e->versions[1].oid, oid))
1422 return 0;
1423 e->versions[1].mode = mode;
1424 oidcpy(&e->versions[1].oid, oid);
1425 if (e->tree)
1426 release_tree_content_recursive(e->tree);
1427 e->tree = subtree;
1430 * We need to leave e->versions[0].sha1 alone
1431 * to avoid modifying the preimage tree used
1432 * when writing out the parent directory.
1433 * But after replacing the subdir with a
1434 * completely different one, it's not a good
1435 * delta base any more, and besides, we've
1436 * thrown away the tree entries needed to
1437 * make a delta against it.
1439 * So let's just explicitly disable deltas
1440 * for the subtree.
1442 if (S_ISDIR(e->versions[0].mode))
1443 e->versions[0].mode |= NO_DELTA;
1445 oidclr(&root->versions[1].oid);
1446 return 1;
1448 if (!S_ISDIR(e->versions[1].mode)) {
1449 e->tree = new_tree_content(8);
1450 e->versions[1].mode = S_IFDIR;
1452 if (!e->tree)
1453 load_tree(e);
1454 if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1455 oidclr(&root->versions[1].oid);
1456 return 1;
1458 return 0;
1462 if (t->entry_count == t->entry_capacity)
1463 root->tree = t = grow_tree_content(t, t->entry_count);
1464 e = new_tree_entry();
1465 e->name = to_atom(p, n);
1466 e->versions[0].mode = 0;
1467 oidclr(&e->versions[0].oid);
1468 t->entries[t->entry_count++] = e;
1469 if (*slash1) {
1470 e->tree = new_tree_content(8);
1471 e->versions[1].mode = S_IFDIR;
1472 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1473 } else {
1474 e->tree = subtree;
1475 e->versions[1].mode = mode;
1476 oidcpy(&e->versions[1].oid, oid);
1478 oidclr(&root->versions[1].oid);
1479 return 1;
1482 static int tree_content_remove(
1483 struct tree_entry *root,
1484 const char *p,
1485 struct tree_entry *backup_leaf,
1486 int allow_root)
1488 struct tree_content *t;
1489 const char *slash1;
1490 unsigned int i, n;
1491 struct tree_entry *e;
1493 slash1 = strchrnul(p, '/');
1494 n = slash1 - p;
1496 if (!root->tree)
1497 load_tree(root);
1499 if (!*p && allow_root) {
1500 e = root;
1501 goto del_entry;
1504 t = root->tree;
1505 for (i = 0; i < t->entry_count; i++) {
1506 e = t->entries[i];
1507 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1508 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1510 * If p names a file in some subdirectory, and a
1511 * file or symlink matching the name of the
1512 * parent directory of p exists, then p cannot
1513 * exist and need not be deleted.
1515 return 1;
1516 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1517 goto del_entry;
1518 if (!e->tree)
1519 load_tree(e);
1520 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1521 for (n = 0; n < e->tree->entry_count; n++) {
1522 if (e->tree->entries[n]->versions[1].mode) {
1523 oidclr(&root->versions[1].oid);
1524 return 1;
1527 backup_leaf = NULL;
1528 goto del_entry;
1530 return 0;
1533 return 0;
1535 del_entry:
1536 if (backup_leaf)
1537 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1538 else if (e->tree)
1539 release_tree_content_recursive(e->tree);
1540 e->tree = NULL;
1541 e->versions[1].mode = 0;
1542 oidclr(&e->versions[1].oid);
1543 oidclr(&root->versions[1].oid);
1544 return 1;
1547 static int tree_content_get(
1548 struct tree_entry *root,
1549 const char *p,
1550 struct tree_entry *leaf,
1551 int allow_root)
1553 struct tree_content *t;
1554 const char *slash1;
1555 unsigned int i, n;
1556 struct tree_entry *e;
1558 slash1 = strchrnul(p, '/');
1559 n = slash1 - p;
1560 if (!n && !allow_root)
1561 die("Empty path component found in input");
1563 if (!root->tree)
1564 load_tree(root);
1566 if (!n) {
1567 e = root;
1568 goto found_entry;
1571 t = root->tree;
1572 for (i = 0; i < t->entry_count; i++) {
1573 e = t->entries[i];
1574 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1575 if (!*slash1)
1576 goto found_entry;
1577 if (!S_ISDIR(e->versions[1].mode))
1578 return 0;
1579 if (!e->tree)
1580 load_tree(e);
1581 return tree_content_get(e, slash1 + 1, leaf, 0);
1584 return 0;
1586 found_entry:
1587 memcpy(leaf, e, sizeof(*leaf));
1588 if (e->tree && is_null_oid(&e->versions[1].oid))
1589 leaf->tree = dup_tree_content(e->tree);
1590 else
1591 leaf->tree = NULL;
1592 return 1;
1595 static int update_branch(struct branch *b)
1597 static const char *msg = "fast-import";
1598 struct ref_transaction *transaction;
1599 struct object_id old_oid;
1600 struct strbuf err = STRBUF_INIT;
1602 if (is_null_oid(&b->oid)) {
1603 if (b->delete)
1604 delete_ref(NULL, b->name, NULL, 0);
1605 return 0;
1607 if (read_ref(b->name, &old_oid))
1608 oidclr(&old_oid);
1609 if (!force_update && !is_null_oid(&old_oid)) {
1610 struct commit *old_cmit, *new_cmit;
1612 old_cmit = lookup_commit_reference_gently(the_repository,
1613 &old_oid, 0);
1614 new_cmit = lookup_commit_reference_gently(the_repository,
1615 &b->oid, 0);
1616 if (!old_cmit || !new_cmit)
1617 return error("Branch %s is missing commits.", b->name);
1619 if (!in_merge_bases(old_cmit, new_cmit)) {
1620 warning("Not updating %s"
1621 " (new tip %s does not contain %s)",
1622 b->name, oid_to_hex(&b->oid),
1623 oid_to_hex(&old_oid));
1624 return -1;
1627 transaction = ref_transaction_begin(&err);
1628 if (!transaction ||
1629 ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1630 0, msg, &err) ||
1631 ref_transaction_commit(transaction, &err)) {
1632 ref_transaction_free(transaction);
1633 error("%s", err.buf);
1634 strbuf_release(&err);
1635 return -1;
1637 ref_transaction_free(transaction);
1638 strbuf_release(&err);
1639 return 0;
1642 static void dump_branches(void)
1644 unsigned int i;
1645 struct branch *b;
1647 for (i = 0; i < branch_table_sz; i++) {
1648 for (b = branch_table[i]; b; b = b->table_next_branch)
1649 failure |= update_branch(b);
1653 static void dump_tags(void)
1655 static const char *msg = "fast-import";
1656 struct tag *t;
1657 struct strbuf ref_name = STRBUF_INIT;
1658 struct strbuf err = STRBUF_INIT;
1659 struct ref_transaction *transaction;
1661 transaction = ref_transaction_begin(&err);
1662 if (!transaction) {
1663 failure |= error("%s", err.buf);
1664 goto cleanup;
1666 for (t = first_tag; t; t = t->next_tag) {
1667 strbuf_reset(&ref_name);
1668 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1670 if (ref_transaction_update(transaction, ref_name.buf,
1671 &t->oid, NULL, 0, msg, &err)) {
1672 failure |= error("%s", err.buf);
1673 goto cleanup;
1676 if (ref_transaction_commit(transaction, &err))
1677 failure |= error("%s", err.buf);
1679 cleanup:
1680 ref_transaction_free(transaction);
1681 strbuf_release(&ref_name);
1682 strbuf_release(&err);
1685 static void dump_marks(void)
1687 struct lock_file mark_lock = LOCK_INIT;
1688 FILE *f;
1690 if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1691 return;
1693 if (safe_create_leading_directories_const(export_marks_file)) {
1694 failure |= error_errno("unable to create leading directories of %s",
1695 export_marks_file);
1696 return;
1699 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1700 failure |= error_errno("Unable to write marks file %s",
1701 export_marks_file);
1702 return;
1705 f = fdopen_lock_file(&mark_lock, "w");
1706 if (!f) {
1707 int saved_errno = errno;
1708 rollback_lock_file(&mark_lock);
1709 failure |= error("Unable to write marks file %s: %s",
1710 export_marks_file, strerror(saved_errno));
1711 return;
1714 for_each_mark(marks, 0, dump_marks_fn, f);
1715 if (commit_lock_file(&mark_lock)) {
1716 failure |= error_errno("Unable to write file %s",
1717 export_marks_file);
1718 return;
1722 static void insert_object_entry(struct mark_set **s, struct object_id *oid, uintmax_t mark)
1724 struct object_entry *e;
1725 e = find_object(oid);
1726 if (!e) {
1727 enum object_type type = oid_object_info(the_repository,
1728 oid, NULL);
1729 if (type < 0)
1730 die("object not found: %s", oid_to_hex(oid));
1731 e = insert_object(oid);
1732 e->type = type;
1733 e->pack_id = MAX_PACK_ID;
1734 e->idx.offset = 1; /* just not zero! */
1736 insert_mark(s, mark, e);
1739 static void insert_oid_entry(struct mark_set **s, struct object_id *oid, uintmax_t mark)
1741 insert_mark(s, mark, xmemdupz(oid, sizeof(*oid)));
1744 static void read_mark_file(struct mark_set **s, FILE *f, mark_set_inserter_t inserter)
1746 char line[512];
1747 while (fgets(line, sizeof(line), f)) {
1748 uintmax_t mark;
1749 char *end;
1750 struct object_id oid;
1752 /* Ensure SHA-1 objects are padded with zeros. */
1753 memset(oid.hash, 0, sizeof(oid.hash));
1755 end = strchr(line, '\n');
1756 if (line[0] != ':' || !end)
1757 die("corrupt mark line: %s", line);
1758 *end = 0;
1759 mark = strtoumax(line + 1, &end, 10);
1760 if (!mark || end == line + 1
1761 || *end != ' '
1762 || get_oid_hex_any(end + 1, &oid) == GIT_HASH_UNKNOWN)
1763 die("corrupt mark line: %s", line);
1764 inserter(s, &oid, mark);
1768 static void read_marks(void)
1770 FILE *f = fopen(import_marks_file, "r");
1771 if (f)
1773 else if (import_marks_file_ignore_missing && errno == ENOENT)
1774 goto done; /* Marks file does not exist */
1775 else
1776 die_errno("cannot read '%s'", import_marks_file);
1777 read_mark_file(&marks, f, insert_object_entry);
1778 fclose(f);
1779 done:
1780 import_marks_file_done = 1;
1784 static int read_next_command(void)
1786 static int stdin_eof = 0;
1788 if (stdin_eof) {
1789 unread_command_buf = 0;
1790 return EOF;
1793 for (;;) {
1794 if (unread_command_buf) {
1795 unread_command_buf = 0;
1796 } else {
1797 struct recent_command *rc;
1799 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1800 if (stdin_eof)
1801 return EOF;
1803 if (!seen_data_command
1804 && !starts_with(command_buf.buf, "feature ")
1805 && !starts_with(command_buf.buf, "option ")) {
1806 parse_argv();
1809 rc = rc_free;
1810 if (rc)
1811 rc_free = rc->next;
1812 else {
1813 rc = cmd_hist.next;
1814 cmd_hist.next = rc->next;
1815 cmd_hist.next->prev = &cmd_hist;
1816 free(rc->buf);
1819 rc->buf = xstrdup(command_buf.buf);
1820 rc->prev = cmd_tail;
1821 rc->next = cmd_hist.prev;
1822 rc->prev->next = rc;
1823 cmd_tail = rc;
1825 if (command_buf.buf[0] == '#')
1826 continue;
1827 return 0;
1831 static void skip_optional_lf(void)
1833 int term_char = fgetc(stdin);
1834 if (term_char != '\n' && term_char != EOF)
1835 ungetc(term_char, stdin);
1838 static void parse_mark(void)
1840 const char *v;
1841 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1842 next_mark = strtoumax(v, NULL, 10);
1843 read_next_command();
1845 else
1846 next_mark = 0;
1849 static void parse_original_identifier(void)
1851 const char *v;
1852 if (skip_prefix(command_buf.buf, "original-oid ", &v))
1853 read_next_command();
1856 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1858 const char *data;
1859 strbuf_reset(sb);
1861 if (!skip_prefix(command_buf.buf, "data ", &data))
1862 die("Expected 'data n' command, found: %s", command_buf.buf);
1864 if (skip_prefix(data, "<<", &data)) {
1865 char *term = xstrdup(data);
1866 size_t term_len = command_buf.len - (data - command_buf.buf);
1868 for (;;) {
1869 if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1870 die("EOF in data (terminator '%s' not found)", term);
1871 if (term_len == command_buf.len
1872 && !strcmp(term, command_buf.buf))
1873 break;
1874 strbuf_addbuf(sb, &command_buf);
1875 strbuf_addch(sb, '\n');
1877 free(term);
1879 else {
1880 uintmax_t len = strtoumax(data, NULL, 10);
1881 size_t n = 0, length = (size_t)len;
1883 if (limit && limit < len) {
1884 *len_res = len;
1885 return 0;
1887 if (length < len)
1888 die("data is too large to use in this context");
1890 while (n < length) {
1891 size_t s = strbuf_fread(sb, length - n, stdin);
1892 if (!s && feof(stdin))
1893 die("EOF in data (%lu bytes remaining)",
1894 (unsigned long)(length - n));
1895 n += s;
1899 skip_optional_lf();
1900 return 1;
1903 static int validate_raw_date(const char *src, struct strbuf *result, int strict)
1905 const char *orig_src = src;
1906 char *endp;
1907 unsigned long num;
1909 errno = 0;
1911 num = strtoul(src, &endp, 10);
1913 * NEEDSWORK: perhaps check for reasonable values? For example, we
1914 * could error on values representing times more than a
1915 * day in the future.
1917 if (errno || endp == src || *endp != ' ')
1918 return -1;
1920 src = endp + 1;
1921 if (*src != '-' && *src != '+')
1922 return -1;
1924 num = strtoul(src + 1, &endp, 10);
1926 * NEEDSWORK: check for brokenness other than num > 1400, such as
1927 * (num % 100) >= 60, or ((num % 100) % 15) != 0 ?
1929 if (errno || endp == src + 1 || *endp || /* did not parse */
1930 (strict && (1400 < num)) /* parsed a broken timezone */
1932 return -1;
1934 strbuf_addstr(result, orig_src);
1935 return 0;
1938 static char *parse_ident(const char *buf)
1940 const char *ltgt;
1941 size_t name_len;
1942 struct strbuf ident = STRBUF_INIT;
1944 /* ensure there is a space delimiter even if there is no name */
1945 if (*buf == '<')
1946 --buf;
1948 ltgt = buf + strcspn(buf, "<>");
1949 if (*ltgt != '<')
1950 die("Missing < in ident string: %s", buf);
1951 if (ltgt != buf && ltgt[-1] != ' ')
1952 die("Missing space before < in ident string: %s", buf);
1953 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
1954 if (*ltgt != '>')
1955 die("Missing > in ident string: %s", buf);
1956 ltgt++;
1957 if (*ltgt != ' ')
1958 die("Missing space after > in ident string: %s", buf);
1959 ltgt++;
1960 name_len = ltgt - buf;
1961 strbuf_add(&ident, buf, name_len);
1963 switch (whenspec) {
1964 case WHENSPEC_RAW:
1965 if (validate_raw_date(ltgt, &ident, 1) < 0)
1966 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1967 break;
1968 case WHENSPEC_RAW_PERMISSIVE:
1969 if (validate_raw_date(ltgt, &ident, 0) < 0)
1970 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1971 break;
1972 case WHENSPEC_RFC2822:
1973 if (parse_date(ltgt, &ident) < 0)
1974 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
1975 break;
1976 case WHENSPEC_NOW:
1977 if (strcmp("now", ltgt))
1978 die("Date in ident must be 'now': %s", buf);
1979 datestamp(&ident);
1980 break;
1983 return strbuf_detach(&ident, NULL);
1986 static void parse_and_store_blob(
1987 struct last_object *last,
1988 struct object_id *oidout,
1989 uintmax_t mark)
1991 static struct strbuf buf = STRBUF_INIT;
1992 uintmax_t len;
1994 if (parse_data(&buf, big_file_threshold, &len))
1995 store_object(OBJ_BLOB, &buf, last, oidout, mark);
1996 else {
1997 if (last) {
1998 strbuf_release(&last->data);
1999 last->offset = 0;
2000 last->depth = 0;
2002 stream_blob(len, oidout, mark);
2003 skip_optional_lf();
2007 static void parse_new_blob(void)
2009 read_next_command();
2010 parse_mark();
2011 parse_original_identifier();
2012 parse_and_store_blob(&last_blob, NULL, next_mark);
2015 static void unload_one_branch(void)
2017 while (cur_active_branches
2018 && cur_active_branches >= max_active_branches) {
2019 uintmax_t min_commit = ULONG_MAX;
2020 struct branch *e, *l = NULL, *p = NULL;
2022 for (e = active_branches; e; e = e->active_next_branch) {
2023 if (e->last_commit < min_commit) {
2024 p = l;
2025 min_commit = e->last_commit;
2027 l = e;
2030 if (p) {
2031 e = p->active_next_branch;
2032 p->active_next_branch = e->active_next_branch;
2033 } else {
2034 e = active_branches;
2035 active_branches = e->active_next_branch;
2037 e->active = 0;
2038 e->active_next_branch = NULL;
2039 if (e->branch_tree.tree) {
2040 release_tree_content_recursive(e->branch_tree.tree);
2041 e->branch_tree.tree = NULL;
2043 cur_active_branches--;
2047 static void load_branch(struct branch *b)
2049 load_tree(&b->branch_tree);
2050 if (!b->active) {
2051 b->active = 1;
2052 b->active_next_branch = active_branches;
2053 active_branches = b;
2054 cur_active_branches++;
2055 branch_load_count++;
2059 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2061 unsigned char fanout = 0;
2062 while ((num_notes >>= 8))
2063 fanout++;
2064 return fanout;
2067 static void construct_path_with_fanout(const char *hex_sha1,
2068 unsigned char fanout, char *path)
2070 unsigned int i = 0, j = 0;
2071 if (fanout >= the_hash_algo->rawsz)
2072 die("Too large fanout (%u)", fanout);
2073 while (fanout) {
2074 path[i++] = hex_sha1[j++];
2075 path[i++] = hex_sha1[j++];
2076 path[i++] = '/';
2077 fanout--;
2079 memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2080 path[i + the_hash_algo->hexsz - j] = '\0';
2083 static uintmax_t do_change_note_fanout(
2084 struct tree_entry *orig_root, struct tree_entry *root,
2085 char *hex_oid, unsigned int hex_oid_len,
2086 char *fullpath, unsigned int fullpath_len,
2087 unsigned char fanout)
2089 struct tree_content *t;
2090 struct tree_entry *e, leaf;
2091 unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2092 uintmax_t num_notes = 0;
2093 struct object_id oid;
2094 /* hex oid + '/' between each pair of hex digits + NUL */
2095 char realpath[GIT_MAX_HEXSZ + ((GIT_MAX_HEXSZ / 2) - 1) + 1];
2096 const unsigned hexsz = the_hash_algo->hexsz;
2098 if (!root->tree)
2099 load_tree(root);
2100 t = root->tree;
2102 for (i = 0; t && i < t->entry_count; i++) {
2103 e = t->entries[i];
2104 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2105 tmp_fullpath_len = fullpath_len;
2108 * We're interested in EITHER existing note entries (entries
2109 * with exactly 40 hex chars in path, not including directory
2110 * separators), OR directory entries that may contain note
2111 * entries (with < 40 hex chars in path).
2112 * Also, each path component in a note entry must be a multiple
2113 * of 2 chars.
2115 if (!e->versions[1].mode ||
2116 tmp_hex_oid_len > hexsz ||
2117 e->name->str_len % 2)
2118 continue;
2120 /* This _may_ be a note entry, or a subdir containing notes */
2121 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2122 e->name->str_len);
2123 if (tmp_fullpath_len)
2124 fullpath[tmp_fullpath_len++] = '/';
2125 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2126 e->name->str_len);
2127 tmp_fullpath_len += e->name->str_len;
2128 fullpath[tmp_fullpath_len] = '\0';
2130 if (tmp_hex_oid_len == hexsz && !get_oid_hex(hex_oid, &oid)) {
2131 /* This is a note entry */
2132 if (fanout == 0xff) {
2133 /* Counting mode, no rename */
2134 num_notes++;
2135 continue;
2137 construct_path_with_fanout(hex_oid, fanout, realpath);
2138 if (!strcmp(fullpath, realpath)) {
2139 /* Note entry is in correct location */
2140 num_notes++;
2141 continue;
2144 /* Rename fullpath to realpath */
2145 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2146 die("Failed to remove path %s", fullpath);
2147 tree_content_set(orig_root, realpath,
2148 &leaf.versions[1].oid,
2149 leaf.versions[1].mode,
2150 leaf.tree);
2151 } else if (S_ISDIR(e->versions[1].mode)) {
2152 /* This is a subdir that may contain note entries */
2153 num_notes += do_change_note_fanout(orig_root, e,
2154 hex_oid, tmp_hex_oid_len,
2155 fullpath, tmp_fullpath_len, fanout);
2158 /* The above may have reallocated the current tree_content */
2159 t = root->tree;
2161 return num_notes;
2164 static uintmax_t change_note_fanout(struct tree_entry *root,
2165 unsigned char fanout)
2168 * The size of path is due to one slash between every two hex digits,
2169 * plus the terminating NUL. Note that there is no slash at the end, so
2170 * the number of slashes is one less than half the number of hex
2171 * characters.
2173 char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2174 return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2177 static int parse_mapped_oid_hex(const char *hex, struct object_id *oid, const char **end)
2179 int algo;
2180 khiter_t it;
2182 /* Make SHA-1 object IDs have all-zero padding. */
2183 memset(oid->hash, 0, sizeof(oid->hash));
2185 algo = parse_oid_hex_any(hex, oid, end);
2186 if (algo == GIT_HASH_UNKNOWN)
2187 return -1;
2189 it = kh_get_oid_map(sub_oid_map, *oid);
2190 /* No such object? */
2191 if (it == kh_end(sub_oid_map)) {
2192 /* If we're using the same algorithm, pass it through. */
2193 if (hash_algos[algo].format_id == the_hash_algo->format_id)
2194 return 0;
2195 return -1;
2197 oidcpy(oid, kh_value(sub_oid_map, it));
2198 return 0;
2202 * Given a pointer into a string, parse a mark reference:
2204 * idnum ::= ':' bigint;
2206 * Return the first character after the value in *endptr.
2208 * Complain if the following character is not what is expected,
2209 * either a space or end of the string.
2211 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2213 uintmax_t mark;
2215 assert(*p == ':');
2216 p++;
2217 mark = strtoumax(p, endptr, 10);
2218 if (*endptr == p)
2219 die("No value after ':' in mark: %s", command_buf.buf);
2220 return mark;
2224 * Parse the mark reference, and complain if this is not the end of
2225 * the string.
2227 static uintmax_t parse_mark_ref_eol(const char *p)
2229 char *end;
2230 uintmax_t mark;
2232 mark = parse_mark_ref(p, &end);
2233 if (*end != '\0')
2234 die("Garbage after mark: %s", command_buf.buf);
2235 return mark;
2239 * Parse the mark reference, demanding a trailing space. Return a
2240 * pointer to the space.
2242 static uintmax_t parse_mark_ref_space(const char **p)
2244 uintmax_t mark;
2245 char *end;
2247 mark = parse_mark_ref(*p, &end);
2248 if (*end++ != ' ')
2249 die("Missing space after mark: %s", command_buf.buf);
2250 *p = end;
2251 return mark;
2254 static void file_change_m(const char *p, struct branch *b)
2256 static struct strbuf uq = STRBUF_INIT;
2257 const char *endp;
2258 struct object_entry *oe;
2259 struct object_id oid;
2260 uint16_t mode, inline_data = 0;
2262 p = get_mode(p, &mode);
2263 if (!p)
2264 die("Corrupt mode: %s", command_buf.buf);
2265 switch (mode) {
2266 case 0644:
2267 case 0755:
2268 mode |= S_IFREG;
2269 case S_IFREG | 0644:
2270 case S_IFREG | 0755:
2271 case S_IFLNK:
2272 case S_IFDIR:
2273 case S_IFGITLINK:
2274 /* ok */
2275 break;
2276 default:
2277 die("Corrupt mode: %s", command_buf.buf);
2280 if (*p == ':') {
2281 oe = find_mark(marks, parse_mark_ref_space(&p));
2282 oidcpy(&oid, &oe->idx.oid);
2283 } else if (skip_prefix(p, "inline ", &p)) {
2284 inline_data = 1;
2285 oe = NULL; /* not used with inline_data, but makes gcc happy */
2286 } else {
2287 if (parse_mapped_oid_hex(p, &oid, &p))
2288 die("Invalid dataref: %s", command_buf.buf);
2289 oe = find_object(&oid);
2290 if (*p++ != ' ')
2291 die("Missing space after SHA1: %s", command_buf.buf);
2294 strbuf_reset(&uq);
2295 if (!unquote_c_style(&uq, p, &endp)) {
2296 if (*endp)
2297 die("Garbage after path in: %s", command_buf.buf);
2298 p = uq.buf;
2301 /* Git does not track empty, non-toplevel directories. */
2302 if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2303 tree_content_remove(&b->branch_tree, p, NULL, 0);
2304 return;
2307 if (S_ISGITLINK(mode)) {
2308 if (inline_data)
2309 die("Git links cannot be specified 'inline': %s",
2310 command_buf.buf);
2311 else if (oe) {
2312 if (oe->type != OBJ_COMMIT)
2313 die("Not a commit (actually a %s): %s",
2314 type_name(oe->type), command_buf.buf);
2317 * Accept the sha1 without checking; it expected to be in
2318 * another repository.
2320 } else if (inline_data) {
2321 if (S_ISDIR(mode))
2322 die("Directories cannot be specified 'inline': %s",
2323 command_buf.buf);
2324 if (p != uq.buf) {
2325 strbuf_addstr(&uq, p);
2326 p = uq.buf;
2328 while (read_next_command() != EOF) {
2329 const char *v;
2330 if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2331 parse_cat_blob(v);
2332 else {
2333 parse_and_store_blob(&last_blob, &oid, 0);
2334 break;
2337 } else {
2338 enum object_type expected = S_ISDIR(mode) ?
2339 OBJ_TREE: OBJ_BLOB;
2340 enum object_type type = oe ? oe->type :
2341 oid_object_info(the_repository, &oid,
2342 NULL);
2343 if (type < 0)
2344 die("%s not found: %s",
2345 S_ISDIR(mode) ? "Tree" : "Blob",
2346 command_buf.buf);
2347 if (type != expected)
2348 die("Not a %s (actually a %s): %s",
2349 type_name(expected), type_name(type),
2350 command_buf.buf);
2353 if (!*p) {
2354 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2355 return;
2357 tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2360 static void file_change_d(const char *p, struct branch *b)
2362 static struct strbuf uq = STRBUF_INIT;
2363 const char *endp;
2365 strbuf_reset(&uq);
2366 if (!unquote_c_style(&uq, p, &endp)) {
2367 if (*endp)
2368 die("Garbage after path in: %s", command_buf.buf);
2369 p = uq.buf;
2371 tree_content_remove(&b->branch_tree, p, NULL, 1);
2374 static void file_change_cr(const char *s, struct branch *b, int rename)
2376 const char *d;
2377 static struct strbuf s_uq = STRBUF_INIT;
2378 static struct strbuf d_uq = STRBUF_INIT;
2379 const char *endp;
2380 struct tree_entry leaf;
2382 strbuf_reset(&s_uq);
2383 if (!unquote_c_style(&s_uq, s, &endp)) {
2384 if (*endp != ' ')
2385 die("Missing space after source: %s", command_buf.buf);
2386 } else {
2387 endp = strchr(s, ' ');
2388 if (!endp)
2389 die("Missing space after source: %s", command_buf.buf);
2390 strbuf_add(&s_uq, s, endp - s);
2392 s = s_uq.buf;
2394 endp++;
2395 if (!*endp)
2396 die("Missing dest: %s", command_buf.buf);
2398 d = endp;
2399 strbuf_reset(&d_uq);
2400 if (!unquote_c_style(&d_uq, d, &endp)) {
2401 if (*endp)
2402 die("Garbage after dest in: %s", command_buf.buf);
2403 d = d_uq.buf;
2406 memset(&leaf, 0, sizeof(leaf));
2407 if (rename)
2408 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2409 else
2410 tree_content_get(&b->branch_tree, s, &leaf, 1);
2411 if (!leaf.versions[1].mode)
2412 die("Path %s not in branch", s);
2413 if (!*d) { /* C "path/to/subdir" "" */
2414 tree_content_replace(&b->branch_tree,
2415 &leaf.versions[1].oid,
2416 leaf.versions[1].mode,
2417 leaf.tree);
2418 return;
2420 tree_content_set(&b->branch_tree, d,
2421 &leaf.versions[1].oid,
2422 leaf.versions[1].mode,
2423 leaf.tree);
2426 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2428 static struct strbuf uq = STRBUF_INIT;
2429 struct object_entry *oe;
2430 struct branch *s;
2431 struct object_id oid, commit_oid;
2432 char path[GIT_MAX_RAWSZ * 3];
2433 uint16_t inline_data = 0;
2434 unsigned char new_fanout;
2437 * When loading a branch, we don't traverse its tree to count the real
2438 * number of notes (too expensive to do this for all non-note refs).
2439 * This means that recently loaded notes refs might incorrectly have
2440 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2442 * Fix this by traversing the tree and counting the number of notes
2443 * when b->num_notes == 0. If the notes tree is truly empty, the
2444 * calculation should not take long.
2446 if (b->num_notes == 0 && *old_fanout == 0) {
2447 /* Invoke change_note_fanout() in "counting mode". */
2448 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2449 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2452 /* Now parse the notemodify command. */
2453 /* <dataref> or 'inline' */
2454 if (*p == ':') {
2455 oe = find_mark(marks, parse_mark_ref_space(&p));
2456 oidcpy(&oid, &oe->idx.oid);
2457 } else if (skip_prefix(p, "inline ", &p)) {
2458 inline_data = 1;
2459 oe = NULL; /* not used with inline_data, but makes gcc happy */
2460 } else {
2461 if (parse_mapped_oid_hex(p, &oid, &p))
2462 die("Invalid dataref: %s", command_buf.buf);
2463 oe = find_object(&oid);
2464 if (*p++ != ' ')
2465 die("Missing space after SHA1: %s", command_buf.buf);
2468 /* <commit-ish> */
2469 s = lookup_branch(p);
2470 if (s) {
2471 if (is_null_oid(&s->oid))
2472 die("Can't add a note on empty branch.");
2473 oidcpy(&commit_oid, &s->oid);
2474 } else if (*p == ':') {
2475 uintmax_t commit_mark = parse_mark_ref_eol(p);
2476 struct object_entry *commit_oe = find_mark(marks, commit_mark);
2477 if (commit_oe->type != OBJ_COMMIT)
2478 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2479 oidcpy(&commit_oid, &commit_oe->idx.oid);
2480 } else if (!get_oid(p, &commit_oid)) {
2481 unsigned long size;
2482 char *buf = read_object_with_reference(the_repository,
2483 &commit_oid,
2484 commit_type, &size,
2485 &commit_oid);
2486 if (!buf || size < the_hash_algo->hexsz + 6)
2487 die("Not a valid commit: %s", p);
2488 free(buf);
2489 } else
2490 die("Invalid ref name or SHA1 expression: %s", p);
2492 if (inline_data) {
2493 if (p != uq.buf) {
2494 strbuf_addstr(&uq, p);
2495 p = uq.buf;
2497 read_next_command();
2498 parse_and_store_blob(&last_blob, &oid, 0);
2499 } else if (oe) {
2500 if (oe->type != OBJ_BLOB)
2501 die("Not a blob (actually a %s): %s",
2502 type_name(oe->type), command_buf.buf);
2503 } else if (!is_null_oid(&oid)) {
2504 enum object_type type = oid_object_info(the_repository, &oid,
2505 NULL);
2506 if (type < 0)
2507 die("Blob not found: %s", command_buf.buf);
2508 if (type != OBJ_BLOB)
2509 die("Not a blob (actually a %s): %s",
2510 type_name(type), command_buf.buf);
2513 construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2514 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2515 b->num_notes--;
2517 if (is_null_oid(&oid))
2518 return; /* nothing to insert */
2520 b->num_notes++;
2521 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2522 construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2523 tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2526 static void file_change_deleteall(struct branch *b)
2528 release_tree_content_recursive(b->branch_tree.tree);
2529 oidclr(&b->branch_tree.versions[0].oid);
2530 oidclr(&b->branch_tree.versions[1].oid);
2531 load_tree(&b->branch_tree);
2532 b->num_notes = 0;
2535 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2537 if (!buf || size < the_hash_algo->hexsz + 6)
2538 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2539 if (memcmp("tree ", buf, 5)
2540 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2541 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2542 oidcpy(&b->branch_tree.versions[0].oid,
2543 &b->branch_tree.versions[1].oid);
2546 static void parse_from_existing(struct branch *b)
2548 if (is_null_oid(&b->oid)) {
2549 oidclr(&b->branch_tree.versions[0].oid);
2550 oidclr(&b->branch_tree.versions[1].oid);
2551 } else {
2552 unsigned long size;
2553 char *buf;
2555 buf = read_object_with_reference(the_repository,
2556 &b->oid, commit_type, &size,
2557 &b->oid);
2558 parse_from_commit(b, buf, size);
2559 free(buf);
2563 static int parse_objectish(struct branch *b, const char *objectish)
2565 struct branch *s;
2566 struct object_id oid;
2568 oidcpy(&oid, &b->branch_tree.versions[1].oid);
2570 s = lookup_branch(objectish);
2571 if (b == s)
2572 die("Can't create a branch from itself: %s", b->name);
2573 else if (s) {
2574 struct object_id *t = &s->branch_tree.versions[1].oid;
2575 oidcpy(&b->oid, &s->oid);
2576 oidcpy(&b->branch_tree.versions[0].oid, t);
2577 oidcpy(&b->branch_tree.versions[1].oid, t);
2578 } else if (*objectish == ':') {
2579 uintmax_t idnum = parse_mark_ref_eol(objectish);
2580 struct object_entry *oe = find_mark(marks, idnum);
2581 if (oe->type != OBJ_COMMIT)
2582 die("Mark :%" PRIuMAX " not a commit", idnum);
2583 if (!oideq(&b->oid, &oe->idx.oid)) {
2584 oidcpy(&b->oid, &oe->idx.oid);
2585 if (oe->pack_id != MAX_PACK_ID) {
2586 unsigned long size;
2587 char *buf = gfi_unpack_entry(oe, &size);
2588 parse_from_commit(b, buf, size);
2589 free(buf);
2590 } else
2591 parse_from_existing(b);
2593 } else if (!get_oid(objectish, &b->oid)) {
2594 parse_from_existing(b);
2595 if (is_null_oid(&b->oid))
2596 b->delete = 1;
2598 else
2599 die("Invalid ref name or SHA1 expression: %s", objectish);
2601 if (b->branch_tree.tree && !oideq(&oid, &b->branch_tree.versions[1].oid)) {
2602 release_tree_content_recursive(b->branch_tree.tree);
2603 b->branch_tree.tree = NULL;
2606 read_next_command();
2607 return 1;
2610 static int parse_from(struct branch *b)
2612 const char *from;
2614 if (!skip_prefix(command_buf.buf, "from ", &from))
2615 return 0;
2617 return parse_objectish(b, from);
2620 static int parse_objectish_with_prefix(struct branch *b, const char *prefix)
2622 const char *base;
2624 if (!skip_prefix(command_buf.buf, prefix, &base))
2625 return 0;
2627 return parse_objectish(b, base);
2630 static struct hash_list *parse_merge(unsigned int *count)
2632 struct hash_list *list = NULL, **tail = &list, *n;
2633 const char *from;
2634 struct branch *s;
2636 *count = 0;
2637 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2638 n = xmalloc(sizeof(*n));
2639 s = lookup_branch(from);
2640 if (s)
2641 oidcpy(&n->oid, &s->oid);
2642 else if (*from == ':') {
2643 uintmax_t idnum = parse_mark_ref_eol(from);
2644 struct object_entry *oe = find_mark(marks, idnum);
2645 if (oe->type != OBJ_COMMIT)
2646 die("Mark :%" PRIuMAX " not a commit", idnum);
2647 oidcpy(&n->oid, &oe->idx.oid);
2648 } else if (!get_oid(from, &n->oid)) {
2649 unsigned long size;
2650 char *buf = read_object_with_reference(the_repository,
2651 &n->oid,
2652 commit_type,
2653 &size, &n->oid);
2654 if (!buf || size < the_hash_algo->hexsz + 6)
2655 die("Not a valid commit: %s", from);
2656 free(buf);
2657 } else
2658 die("Invalid ref name or SHA1 expression: %s", from);
2660 n->next = NULL;
2661 *tail = n;
2662 tail = &n->next;
2664 (*count)++;
2665 read_next_command();
2667 return list;
2670 static void parse_new_commit(const char *arg)
2672 static struct strbuf msg = STRBUF_INIT;
2673 struct branch *b;
2674 char *author = NULL;
2675 char *committer = NULL;
2676 char *encoding = NULL;
2677 struct hash_list *merge_list = NULL;
2678 unsigned int merge_count;
2679 unsigned char prev_fanout, new_fanout;
2680 const char *v;
2682 b = lookup_branch(arg);
2683 if (!b)
2684 b = new_branch(arg);
2686 read_next_command();
2687 parse_mark();
2688 parse_original_identifier();
2689 if (skip_prefix(command_buf.buf, "author ", &v)) {
2690 author = parse_ident(v);
2691 read_next_command();
2693 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2694 committer = parse_ident(v);
2695 read_next_command();
2697 if (!committer)
2698 die("Expected committer but didn't get one");
2699 if (skip_prefix(command_buf.buf, "encoding ", &v)) {
2700 encoding = xstrdup(v);
2701 read_next_command();
2703 parse_data(&msg, 0, NULL);
2704 read_next_command();
2705 parse_from(b);
2706 merge_list = parse_merge(&merge_count);
2708 /* ensure the branch is active/loaded */
2709 if (!b->branch_tree.tree || !max_active_branches) {
2710 unload_one_branch();
2711 load_branch(b);
2714 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2716 /* file_change* */
2717 while (command_buf.len > 0) {
2718 if (skip_prefix(command_buf.buf, "M ", &v))
2719 file_change_m(v, b);
2720 else if (skip_prefix(command_buf.buf, "D ", &v))
2721 file_change_d(v, b);
2722 else if (skip_prefix(command_buf.buf, "R ", &v))
2723 file_change_cr(v, b, 1);
2724 else if (skip_prefix(command_buf.buf, "C ", &v))
2725 file_change_cr(v, b, 0);
2726 else if (skip_prefix(command_buf.buf, "N ", &v))
2727 note_change_n(v, b, &prev_fanout);
2728 else if (!strcmp("deleteall", command_buf.buf))
2729 file_change_deleteall(b);
2730 else if (skip_prefix(command_buf.buf, "ls ", &v))
2731 parse_ls(v, b);
2732 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2733 parse_cat_blob(v);
2734 else {
2735 unread_command_buf = 1;
2736 break;
2738 if (read_next_command() == EOF)
2739 break;
2742 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2743 if (new_fanout != prev_fanout)
2744 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2746 /* build the tree and the commit */
2747 store_tree(&b->branch_tree);
2748 oidcpy(&b->branch_tree.versions[0].oid,
2749 &b->branch_tree.versions[1].oid);
2751 strbuf_reset(&new_data);
2752 strbuf_addf(&new_data, "tree %s\n",
2753 oid_to_hex(&b->branch_tree.versions[1].oid));
2754 if (!is_null_oid(&b->oid))
2755 strbuf_addf(&new_data, "parent %s\n",
2756 oid_to_hex(&b->oid));
2757 while (merge_list) {
2758 struct hash_list *next = merge_list->next;
2759 strbuf_addf(&new_data, "parent %s\n",
2760 oid_to_hex(&merge_list->oid));
2761 free(merge_list);
2762 merge_list = next;
2764 strbuf_addf(&new_data,
2765 "author %s\n"
2766 "committer %s\n",
2767 author ? author : committer, committer);
2768 if (encoding)
2769 strbuf_addf(&new_data,
2770 "encoding %s\n",
2771 encoding);
2772 strbuf_addch(&new_data, '\n');
2773 strbuf_addbuf(&new_data, &msg);
2774 free(author);
2775 free(committer);
2776 free(encoding);
2778 if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2779 b->pack_id = pack_id;
2780 b->last_commit = object_count_by_type[OBJ_COMMIT];
2783 static void parse_new_tag(const char *arg)
2785 static struct strbuf msg = STRBUF_INIT;
2786 const char *from;
2787 char *tagger;
2788 struct branch *s;
2789 struct tag *t;
2790 uintmax_t from_mark = 0;
2791 struct object_id oid;
2792 enum object_type type;
2793 const char *v;
2795 t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2796 memset(t, 0, sizeof(struct tag));
2797 t->name = mem_pool_strdup(&fi_mem_pool, arg);
2798 if (last_tag)
2799 last_tag->next_tag = t;
2800 else
2801 first_tag = t;
2802 last_tag = t;
2803 read_next_command();
2804 parse_mark();
2806 /* from ... */
2807 if (!skip_prefix(command_buf.buf, "from ", &from))
2808 die("Expected from command, got %s", command_buf.buf);
2809 s = lookup_branch(from);
2810 if (s) {
2811 if (is_null_oid(&s->oid))
2812 die("Can't tag an empty branch.");
2813 oidcpy(&oid, &s->oid);
2814 type = OBJ_COMMIT;
2815 } else if (*from == ':') {
2816 struct object_entry *oe;
2817 from_mark = parse_mark_ref_eol(from);
2818 oe = find_mark(marks, from_mark);
2819 type = oe->type;
2820 oidcpy(&oid, &oe->idx.oid);
2821 } else if (!get_oid(from, &oid)) {
2822 struct object_entry *oe = find_object(&oid);
2823 if (!oe) {
2824 type = oid_object_info(the_repository, &oid, NULL);
2825 if (type < 0)
2826 die("Not a valid object: %s", from);
2827 } else
2828 type = oe->type;
2829 } else
2830 die("Invalid ref name or SHA1 expression: %s", from);
2831 read_next_command();
2833 /* original-oid ... */
2834 parse_original_identifier();
2836 /* tagger ... */
2837 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2838 tagger = parse_ident(v);
2839 read_next_command();
2840 } else
2841 tagger = NULL;
2843 /* tag payload/message */
2844 parse_data(&msg, 0, NULL);
2846 /* build the tag object */
2847 strbuf_reset(&new_data);
2849 strbuf_addf(&new_data,
2850 "object %s\n"
2851 "type %s\n"
2852 "tag %s\n",
2853 oid_to_hex(&oid), type_name(type), t->name);
2854 if (tagger)
2855 strbuf_addf(&new_data,
2856 "tagger %s\n", tagger);
2857 strbuf_addch(&new_data, '\n');
2858 strbuf_addbuf(&new_data, &msg);
2859 free(tagger);
2861 if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, next_mark))
2862 t->pack_id = MAX_PACK_ID;
2863 else
2864 t->pack_id = pack_id;
2867 static void parse_reset_branch(const char *arg)
2869 struct branch *b;
2870 const char *tag_name;
2872 b = lookup_branch(arg);
2873 if (b) {
2874 oidclr(&b->oid);
2875 oidclr(&b->branch_tree.versions[0].oid);
2876 oidclr(&b->branch_tree.versions[1].oid);
2877 if (b->branch_tree.tree) {
2878 release_tree_content_recursive(b->branch_tree.tree);
2879 b->branch_tree.tree = NULL;
2882 else
2883 b = new_branch(arg);
2884 read_next_command();
2885 parse_from(b);
2886 if (b->delete && skip_prefix(b->name, "refs/tags/", &tag_name)) {
2888 * Elsewhere, we call dump_branches() before dump_tags(),
2889 * and dump_branches() will handle ref deletions first, so
2890 * in order to make sure the deletion actually takes effect,
2891 * we need to remove the tag from our list of tags to update.
2893 * NEEDSWORK: replace list of tags with hashmap for faster
2894 * deletion?
2896 struct tag *t, *prev = NULL;
2897 for (t = first_tag; t; t = t->next_tag) {
2898 if (!strcmp(t->name, tag_name))
2899 break;
2900 prev = t;
2902 if (t) {
2903 if (prev)
2904 prev->next_tag = t->next_tag;
2905 else
2906 first_tag = t->next_tag;
2907 if (!t->next_tag)
2908 last_tag = prev;
2909 /* There is no mem_pool_free(t) function to call. */
2912 if (command_buf.len > 0)
2913 unread_command_buf = 1;
2916 static void cat_blob_write(const char *buf, unsigned long size)
2918 if (write_in_full(cat_blob_fd, buf, size) < 0)
2919 die_errno("Write to frontend failed");
2922 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2924 struct strbuf line = STRBUF_INIT;
2925 unsigned long size;
2926 enum object_type type = 0;
2927 char *buf;
2929 if (!oe || oe->pack_id == MAX_PACK_ID) {
2930 buf = read_object_file(oid, &type, &size);
2931 } else {
2932 type = oe->type;
2933 buf = gfi_unpack_entry(oe, &size);
2937 * Output based on batch_one_object() from cat-file.c.
2939 if (type <= 0) {
2940 strbuf_reset(&line);
2941 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2942 cat_blob_write(line.buf, line.len);
2943 strbuf_release(&line);
2944 free(buf);
2945 return;
2947 if (!buf)
2948 die("Can't read object %s", oid_to_hex(oid));
2949 if (type != OBJ_BLOB)
2950 die("Object %s is a %s but a blob was expected.",
2951 oid_to_hex(oid), type_name(type));
2952 strbuf_reset(&line);
2953 strbuf_addf(&line, "%s %s %"PRIuMAX"\n", oid_to_hex(oid),
2954 type_name(type), (uintmax_t)size);
2955 cat_blob_write(line.buf, line.len);
2956 strbuf_release(&line);
2957 cat_blob_write(buf, size);
2958 cat_blob_write("\n", 1);
2959 if (oe && oe->pack_id == pack_id) {
2960 last_blob.offset = oe->idx.offset;
2961 strbuf_attach(&last_blob.data, buf, size, size);
2962 last_blob.depth = oe->depth;
2963 } else
2964 free(buf);
2967 static void parse_get_mark(const char *p)
2969 struct object_entry *oe;
2970 char output[GIT_MAX_HEXSZ + 2];
2972 /* get-mark SP <object> LF */
2973 if (*p != ':')
2974 die("Not a mark: %s", p);
2976 oe = find_mark(marks, parse_mark_ref_eol(p));
2977 if (!oe)
2978 die("Unknown mark: %s", command_buf.buf);
2980 xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2981 cat_blob_write(output, the_hash_algo->hexsz + 1);
2984 static void parse_cat_blob(const char *p)
2986 struct object_entry *oe;
2987 struct object_id oid;
2989 /* cat-blob SP <object> LF */
2990 if (*p == ':') {
2991 oe = find_mark(marks, parse_mark_ref_eol(p));
2992 if (!oe)
2993 die("Unknown mark: %s", command_buf.buf);
2994 oidcpy(&oid, &oe->idx.oid);
2995 } else {
2996 if (parse_mapped_oid_hex(p, &oid, &p))
2997 die("Invalid dataref: %s", command_buf.buf);
2998 if (*p)
2999 die("Garbage after SHA1: %s", command_buf.buf);
3000 oe = find_object(&oid);
3003 cat_blob(oe, &oid);
3006 static struct object_entry *dereference(struct object_entry *oe,
3007 struct object_id *oid)
3009 unsigned long size;
3010 char *buf = NULL;
3011 const unsigned hexsz = the_hash_algo->hexsz;
3013 if (!oe) {
3014 enum object_type type = oid_object_info(the_repository, oid,
3015 NULL);
3016 if (type < 0)
3017 die("object not found: %s", oid_to_hex(oid));
3018 /* cache it! */
3019 oe = insert_object(oid);
3020 oe->type = type;
3021 oe->pack_id = MAX_PACK_ID;
3022 oe->idx.offset = 1;
3024 switch (oe->type) {
3025 case OBJ_TREE: /* easy case. */
3026 return oe;
3027 case OBJ_COMMIT:
3028 case OBJ_TAG:
3029 break;
3030 default:
3031 die("Not a tree-ish: %s", command_buf.buf);
3034 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3035 buf = gfi_unpack_entry(oe, &size);
3036 } else {
3037 enum object_type unused;
3038 buf = read_object_file(oid, &unused, &size);
3040 if (!buf)
3041 die("Can't load object %s", oid_to_hex(oid));
3043 /* Peel one layer. */
3044 switch (oe->type) {
3045 case OBJ_TAG:
3046 if (size < hexsz + strlen("object ") ||
3047 get_oid_hex(buf + strlen("object "), oid))
3048 die("Invalid SHA1 in tag: %s", command_buf.buf);
3049 break;
3050 case OBJ_COMMIT:
3051 if (size < hexsz + strlen("tree ") ||
3052 get_oid_hex(buf + strlen("tree "), oid))
3053 die("Invalid SHA1 in commit: %s", command_buf.buf);
3056 free(buf);
3057 return find_object(oid);
3060 static void insert_mapped_mark(uintmax_t mark, void *object, void *cbp)
3062 struct object_id *fromoid = object;
3063 struct object_id *tooid = find_mark(cbp, mark);
3064 int ret;
3065 khiter_t it;
3067 it = kh_put_oid_map(sub_oid_map, *fromoid, &ret);
3068 /* We've already seen this object. */
3069 if (ret == 0)
3070 return;
3071 kh_value(sub_oid_map, it) = tooid;
3074 static void build_mark_map_one(struct mark_set *from, struct mark_set *to)
3076 for_each_mark(from, 0, insert_mapped_mark, to);
3079 static void build_mark_map(struct string_list *from, struct string_list *to)
3081 struct string_list_item *fromp, *top;
3083 sub_oid_map = kh_init_oid_map();
3085 for_each_string_list_item(fromp, from) {
3086 top = string_list_lookup(to, fromp->string);
3087 if (!fromp->util) {
3088 die(_("Missing from marks for submodule '%s'"), fromp->string);
3089 } else if (!top || !top->util) {
3090 die(_("Missing to marks for submodule '%s'"), fromp->string);
3092 build_mark_map_one(fromp->util, top->util);
3096 static struct object_entry *parse_treeish_dataref(const char **p)
3098 struct object_id oid;
3099 struct object_entry *e;
3101 if (**p == ':') { /* <mark> */
3102 e = find_mark(marks, parse_mark_ref_space(p));
3103 if (!e)
3104 die("Unknown mark: %s", command_buf.buf);
3105 oidcpy(&oid, &e->idx.oid);
3106 } else { /* <sha1> */
3107 if (parse_mapped_oid_hex(*p, &oid, p))
3108 die("Invalid dataref: %s", command_buf.buf);
3109 e = find_object(&oid);
3110 if (*(*p)++ != ' ')
3111 die("Missing space after tree-ish: %s", command_buf.buf);
3114 while (!e || e->type != OBJ_TREE)
3115 e = dereference(e, &oid);
3116 return e;
3119 static void print_ls(int mode, const unsigned char *hash, const char *path)
3121 static struct strbuf line = STRBUF_INIT;
3123 /* See show_tree(). */
3124 const char *type =
3125 S_ISGITLINK(mode) ? commit_type :
3126 S_ISDIR(mode) ? tree_type :
3127 blob_type;
3129 if (!mode) {
3130 /* missing SP path LF */
3131 strbuf_reset(&line);
3132 strbuf_addstr(&line, "missing ");
3133 quote_c_style(path, &line, NULL, 0);
3134 strbuf_addch(&line, '\n');
3135 } else {
3136 /* mode SP type SP object_name TAB path LF */
3137 strbuf_reset(&line);
3138 strbuf_addf(&line, "%06o %s %s\t",
3139 mode & ~NO_DELTA, type, hash_to_hex(hash));
3140 quote_c_style(path, &line, NULL, 0);
3141 strbuf_addch(&line, '\n');
3143 cat_blob_write(line.buf, line.len);
3146 static void parse_ls(const char *p, struct branch *b)
3148 struct tree_entry *root = NULL;
3149 struct tree_entry leaf = {NULL};
3151 /* ls SP (<tree-ish> SP)? <path> */
3152 if (*p == '"') {
3153 if (!b)
3154 die("Not in a commit: %s", command_buf.buf);
3155 root = &b->branch_tree;
3156 } else {
3157 struct object_entry *e = parse_treeish_dataref(&p);
3158 root = new_tree_entry();
3159 oidcpy(&root->versions[1].oid, &e->idx.oid);
3160 if (!is_null_oid(&root->versions[1].oid))
3161 root->versions[1].mode = S_IFDIR;
3162 load_tree(root);
3164 if (*p == '"') {
3165 static struct strbuf uq = STRBUF_INIT;
3166 const char *endp;
3167 strbuf_reset(&uq);
3168 if (unquote_c_style(&uq, p, &endp))
3169 die("Invalid path: %s", command_buf.buf);
3170 if (*endp)
3171 die("Garbage after path in: %s", command_buf.buf);
3172 p = uq.buf;
3174 tree_content_get(root, p, &leaf, 1);
3176 * A directory in preparation would have a sha1 of zero
3177 * until it is saved. Save, for simplicity.
3179 if (S_ISDIR(leaf.versions[1].mode))
3180 store_tree(&leaf);
3182 print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3183 if (leaf.tree)
3184 release_tree_content_recursive(leaf.tree);
3185 if (!b || root != &b->branch_tree)
3186 release_tree_entry(root);
3189 static void checkpoint(void)
3191 checkpoint_requested = 0;
3192 if (object_count) {
3193 cycle_packfile();
3195 dump_branches();
3196 dump_tags();
3197 dump_marks();
3200 static void parse_checkpoint(void)
3202 checkpoint_requested = 1;
3203 skip_optional_lf();
3206 static void parse_progress(void)
3208 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3209 fputc('\n', stdout);
3210 fflush(stdout);
3211 skip_optional_lf();
3214 static void parse_alias(void)
3216 struct object_entry *e;
3217 struct branch b;
3219 skip_optional_lf();
3220 read_next_command();
3222 /* mark ... */
3223 parse_mark();
3224 if (!next_mark)
3225 die(_("Expected 'mark' command, got %s"), command_buf.buf);
3227 /* to ... */
3228 memset(&b, 0, sizeof(b));
3229 if (!parse_objectish_with_prefix(&b, "to "))
3230 die(_("Expected 'to' command, got %s"), command_buf.buf);
3231 e = find_object(&b.oid);
3232 assert(e);
3233 insert_mark(&marks, next_mark, e);
3236 static char* make_fast_import_path(const char *path)
3238 if (!relative_marks_paths || is_absolute_path(path))
3239 return xstrdup(path);
3240 return git_pathdup("info/fast-import/%s", path);
3243 static void option_import_marks(const char *marks,
3244 int from_stream, int ignore_missing)
3246 if (import_marks_file) {
3247 if (from_stream)
3248 die("Only one import-marks command allowed per stream");
3250 /* read previous mark file */
3251 if(!import_marks_file_from_stream)
3252 read_marks();
3255 import_marks_file = make_fast_import_path(marks);
3256 import_marks_file_from_stream = from_stream;
3257 import_marks_file_ignore_missing = ignore_missing;
3260 static void option_date_format(const char *fmt)
3262 if (!strcmp(fmt, "raw"))
3263 whenspec = WHENSPEC_RAW;
3264 else if (!strcmp(fmt, "raw-permissive"))
3265 whenspec = WHENSPEC_RAW_PERMISSIVE;
3266 else if (!strcmp(fmt, "rfc2822"))
3267 whenspec = WHENSPEC_RFC2822;
3268 else if (!strcmp(fmt, "now"))
3269 whenspec = WHENSPEC_NOW;
3270 else
3271 die("unknown --date-format argument %s", fmt);
3274 static unsigned long ulong_arg(const char *option, const char *arg)
3276 char *endptr;
3277 unsigned long rv = strtoul(arg, &endptr, 0);
3278 if (strchr(arg, '-') || endptr == arg || *endptr)
3279 die("%s: argument must be a non-negative integer", option);
3280 return rv;
3283 static void option_depth(const char *depth)
3285 max_depth = ulong_arg("--depth", depth);
3286 if (max_depth > MAX_DEPTH)
3287 die("--depth cannot exceed %u", MAX_DEPTH);
3290 static void option_active_branches(const char *branches)
3292 max_active_branches = ulong_arg("--active-branches", branches);
3295 static void option_export_marks(const char *marks)
3297 export_marks_file = make_fast_import_path(marks);
3300 static void option_cat_blob_fd(const char *fd)
3302 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3303 if (n > (unsigned long) INT_MAX)
3304 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3305 cat_blob_fd = (int) n;
3308 static void option_export_pack_edges(const char *edges)
3310 if (pack_edges)
3311 fclose(pack_edges);
3312 pack_edges = xfopen(edges, "a");
3315 static void option_rewrite_submodules(const char *arg, struct string_list *list)
3317 struct mark_set *ms;
3318 FILE *fp;
3319 char *s = xstrdup(arg);
3320 char *f = strchr(s, ':');
3321 if (!f)
3322 die(_("Expected format name:filename for submodule rewrite option"));
3323 *f = '\0';
3324 f++;
3325 CALLOC_ARRAY(ms, 1);
3327 fp = fopen(f, "r");
3328 if (!fp)
3329 die_errno("cannot read '%s'", f);
3330 read_mark_file(&ms, fp, insert_oid_entry);
3331 fclose(fp);
3333 string_list_insert(list, s)->util = ms;
3336 static int parse_one_option(const char *option)
3338 if (skip_prefix(option, "max-pack-size=", &option)) {
3339 unsigned long v;
3340 if (!git_parse_ulong(option, &v))
3341 return 0;
3342 if (v < 8192) {
3343 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3344 v *= 1024 * 1024;
3345 } else if (v < 1024 * 1024) {
3346 warning("minimum max-pack-size is 1 MiB");
3347 v = 1024 * 1024;
3349 max_packsize = v;
3350 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3351 unsigned long v;
3352 if (!git_parse_ulong(option, &v))
3353 return 0;
3354 big_file_threshold = v;
3355 } else if (skip_prefix(option, "depth=", &option)) {
3356 option_depth(option);
3357 } else if (skip_prefix(option, "active-branches=", &option)) {
3358 option_active_branches(option);
3359 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3360 option_export_pack_edges(option);
3361 } else if (!strcmp(option, "quiet")) {
3362 show_stats = 0;
3363 } else if (!strcmp(option, "stats")) {
3364 show_stats = 1;
3365 } else if (!strcmp(option, "allow-unsafe-features")) {
3366 ; /* already handled during early option parsing */
3367 } else {
3368 return 0;
3371 return 1;
3374 static void check_unsafe_feature(const char *feature, int from_stream)
3376 if (from_stream && !allow_unsafe_features)
3377 die(_("feature '%s' forbidden in input without --allow-unsafe-features"),
3378 feature);
3381 static int parse_one_feature(const char *feature, int from_stream)
3383 const char *arg;
3385 if (skip_prefix(feature, "date-format=", &arg)) {
3386 option_date_format(arg);
3387 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3388 check_unsafe_feature("import-marks", from_stream);
3389 option_import_marks(arg, from_stream, 0);
3390 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3391 check_unsafe_feature("import-marks-if-exists", from_stream);
3392 option_import_marks(arg, from_stream, 1);
3393 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3394 check_unsafe_feature(feature, from_stream);
3395 option_export_marks(arg);
3396 } else if (!strcmp(feature, "alias")) {
3397 ; /* Don't die - this feature is supported */
3398 } else if (skip_prefix(feature, "rewrite-submodules-to=", &arg)) {
3399 option_rewrite_submodules(arg, &sub_marks_to);
3400 } else if (skip_prefix(feature, "rewrite-submodules-from=", &arg)) {
3401 option_rewrite_submodules(arg, &sub_marks_from);
3402 } else if (!strcmp(feature, "get-mark")) {
3403 ; /* Don't die - this feature is supported */
3404 } else if (!strcmp(feature, "cat-blob")) {
3405 ; /* Don't die - this feature is supported */
3406 } else if (!strcmp(feature, "relative-marks")) {
3407 relative_marks_paths = 1;
3408 } else if (!strcmp(feature, "no-relative-marks")) {
3409 relative_marks_paths = 0;
3410 } else if (!strcmp(feature, "done")) {
3411 require_explicit_termination = 1;
3412 } else if (!strcmp(feature, "force")) {
3413 force_update = 1;
3414 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3415 ; /* do nothing; we have the feature */
3416 } else {
3417 return 0;
3420 return 1;
3423 static void parse_feature(const char *feature)
3425 if (seen_data_command)
3426 die("Got feature command '%s' after data command", feature);
3428 if (parse_one_feature(feature, 1))
3429 return;
3431 die("This version of fast-import does not support feature %s.", feature);
3434 static void parse_option(const char *option)
3436 if (seen_data_command)
3437 die("Got option command '%s' after data command", option);
3439 if (parse_one_option(option))
3440 return;
3442 die("This version of fast-import does not support option: %s", option);
3445 static void git_pack_config(void)
3447 int indexversion_value;
3448 int limit;
3449 unsigned long packsizelimit_value;
3451 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3452 if (max_depth > MAX_DEPTH)
3453 max_depth = MAX_DEPTH;
3455 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3456 pack_idx_opts.version = indexversion_value;
3457 if (pack_idx_opts.version > 2)
3458 git_die_config("pack.indexversion",
3459 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3461 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3462 max_packsize = packsizelimit_value;
3464 if (!git_config_get_int("fastimport.unpacklimit", &limit))
3465 unpack_limit = limit;
3466 else if (!git_config_get_int("transfer.unpacklimit", &limit))
3467 unpack_limit = limit;
3469 git_config(git_default_config, NULL);
3472 static const char fast_import_usage[] =
3473 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3475 static void parse_argv(void)
3477 unsigned int i;
3479 for (i = 1; i < global_argc; i++) {
3480 const char *a = global_argv[i];
3482 if (*a != '-' || !strcmp(a, "--"))
3483 break;
3485 if (!skip_prefix(a, "--", &a))
3486 die("unknown option %s", a);
3488 if (parse_one_option(a))
3489 continue;
3491 if (parse_one_feature(a, 0))
3492 continue;
3494 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3495 option_cat_blob_fd(a);
3496 continue;
3499 die("unknown option --%s", a);
3501 if (i != global_argc)
3502 usage(fast_import_usage);
3504 seen_data_command = 1;
3505 if (import_marks_file)
3506 read_marks();
3507 build_mark_map(&sub_marks_from, &sub_marks_to);
3510 int cmd_fast_import(int argc, const char **argv, const char *prefix)
3512 unsigned int i;
3514 if (argc == 2 && !strcmp(argv[1], "-h"))
3515 usage(fast_import_usage);
3517 reset_pack_idx_option(&pack_idx_opts);
3518 git_pack_config();
3520 alloc_objects(object_entry_alloc);
3521 strbuf_init(&command_buf, 0);
3522 CALLOC_ARRAY(atom_table, atom_table_sz);
3523 CALLOC_ARRAY(branch_table, branch_table_sz);
3524 CALLOC_ARRAY(avail_tree_table, avail_tree_table_sz);
3525 marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3527 hashmap_init(&object_table, object_entry_hashcmp, NULL, 0);
3530 * We don't parse most options until after we've seen the set of
3531 * "feature" lines at the start of the stream (which allows the command
3532 * line to override stream data). But we must do an early parse of any
3533 * command-line options that impact how we interpret the feature lines.
3535 for (i = 1; i < argc; i++) {
3536 const char *arg = argv[i];
3537 if (*arg != '-' || !strcmp(arg, "--"))
3538 break;
3539 if (!strcmp(arg, "--allow-unsafe-features"))
3540 allow_unsafe_features = 1;
3543 global_argc = argc;
3544 global_argv = argv;
3546 rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3547 for (i = 0; i < (cmd_save - 1); i++)
3548 rc_free[i].next = &rc_free[i + 1];
3549 rc_free[cmd_save - 1].next = NULL;
3551 start_packfile();
3552 set_die_routine(die_nicely);
3553 set_checkpoint_signal();
3554 while (read_next_command() != EOF) {
3555 const char *v;
3556 if (!strcmp("blob", command_buf.buf))
3557 parse_new_blob();
3558 else if (skip_prefix(command_buf.buf, "commit ", &v))
3559 parse_new_commit(v);
3560 else if (skip_prefix(command_buf.buf, "tag ", &v))
3561 parse_new_tag(v);
3562 else if (skip_prefix(command_buf.buf, "reset ", &v))
3563 parse_reset_branch(v);
3564 else if (skip_prefix(command_buf.buf, "ls ", &v))
3565 parse_ls(v, NULL);
3566 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
3567 parse_cat_blob(v);
3568 else if (skip_prefix(command_buf.buf, "get-mark ", &v))
3569 parse_get_mark(v);
3570 else if (!strcmp("checkpoint", command_buf.buf))
3571 parse_checkpoint();
3572 else if (!strcmp("done", command_buf.buf))
3573 break;
3574 else if (!strcmp("alias", command_buf.buf))
3575 parse_alias();
3576 else if (starts_with(command_buf.buf, "progress "))
3577 parse_progress();
3578 else if (skip_prefix(command_buf.buf, "feature ", &v))
3579 parse_feature(v);
3580 else if (skip_prefix(command_buf.buf, "option git ", &v))
3581 parse_option(v);
3582 else if (starts_with(command_buf.buf, "option "))
3583 /* ignore non-git options*/;
3584 else
3585 die("Unsupported command: %s", command_buf.buf);
3587 if (checkpoint_requested)
3588 checkpoint();
3591 /* argv hasn't been parsed yet, do so */
3592 if (!seen_data_command)
3593 parse_argv();
3595 if (require_explicit_termination && feof(stdin))
3596 die("stream ends early");
3598 end_packfile();
3600 dump_branches();
3601 dump_tags();
3602 unkeep_all_packs();
3603 dump_marks();
3605 if (pack_edges)
3606 fclose(pack_edges);
3608 if (show_stats) {
3609 uintmax_t total_count = 0, duplicate_count = 0;
3610 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3611 total_count += object_count_by_type[i];
3612 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3613 duplicate_count += duplicate_count_by_type[i];
3615 fprintf(stderr, "%s statistics:\n", argv[0]);
3616 fprintf(stderr, "---------------------------------------------------------------------\n");
3617 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3618 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3619 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]);
3620 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]);
3621 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]);
3622 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]);
3623 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3624 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3625 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3626 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3627 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3628 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3629 fprintf(stderr, "---------------------------------------------------------------------\n");
3630 pack_report();
3631 fprintf(stderr, "---------------------------------------------------------------------\n");
3632 fprintf(stderr, "\n");
3635 return failure ? 1 : 0;