Merge branch 'cb/test-use-ere-for-alternation'
[git.git] / fast-import.c
blobc98970274c4056178b0e9a37e3ba38b323707b50
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_RFC2822,
143 WHENSPEC_NOW
144 } whenspec_type;
146 struct recent_command {
147 struct recent_command *prev;
148 struct recent_command *next;
149 char *buf;
152 typedef void (*mark_set_inserter_t)(struct mark_set *s, struct object_id *oid, uintmax_t mark);
153 typedef void (*each_mark_fn_t)(uintmax_t mark, void *obj, void *cbp);
155 /* Configured limits on output */
156 static unsigned long max_depth = 50;
157 static off_t max_packsize;
158 static int unpack_limit = 100;
159 static int force_update;
161 /* Stats and misc. counters */
162 static uintmax_t alloc_count;
163 static uintmax_t marks_set_count;
164 static uintmax_t object_count_by_type[1 << TYPE_BITS];
165 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
166 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
167 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
168 static unsigned long object_count;
169 static unsigned long branch_count;
170 static unsigned long branch_load_count;
171 static int failure;
172 static FILE *pack_edges;
173 static unsigned int show_stats = 1;
174 static int global_argc;
175 static const char **global_argv;
177 /* Memory pools */
178 static struct mem_pool fi_mem_pool = {NULL, 2*1024*1024 -
179 sizeof(struct mp_block), 0 };
181 /* Atom management */
182 static unsigned int atom_table_sz = 4451;
183 static unsigned int atom_cnt;
184 static struct atom_str **atom_table;
186 /* The .pack file being generated */
187 static struct pack_idx_option pack_idx_opts;
188 static unsigned int pack_id;
189 static struct hashfile *pack_file;
190 static struct packed_git *pack_data;
191 static struct packed_git **all_packs;
192 static off_t pack_size;
194 /* Table of objects we've written. */
195 static unsigned int object_entry_alloc = 5000;
196 static struct object_entry_pool *blocks;
197 static struct hashmap object_table;
198 static struct mark_set *marks;
199 static const char *export_marks_file;
200 static const char *import_marks_file;
201 static int import_marks_file_from_stream;
202 static int import_marks_file_ignore_missing;
203 static int import_marks_file_done;
204 static int relative_marks_paths;
206 /* Our last blob */
207 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
209 /* Tree management */
210 static unsigned int tree_entry_alloc = 1000;
211 static void *avail_tree_entry;
212 static unsigned int avail_tree_table_sz = 100;
213 static struct avail_tree_content **avail_tree_table;
214 static size_t tree_entry_allocd;
215 static struct strbuf old_tree = STRBUF_INIT;
216 static struct strbuf new_tree = STRBUF_INIT;
218 /* Branch data */
219 static unsigned long max_active_branches = 5;
220 static unsigned long cur_active_branches;
221 static unsigned long branch_table_sz = 1039;
222 static struct branch **branch_table;
223 static struct branch *active_branches;
225 /* Tag data */
226 static struct tag *first_tag;
227 static struct tag *last_tag;
229 /* Input stream parsing */
230 static whenspec_type whenspec = WHENSPEC_RAW;
231 static struct strbuf command_buf = STRBUF_INIT;
232 static int unread_command_buf;
233 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
234 static struct recent_command *cmd_tail = &cmd_hist;
235 static struct recent_command *rc_free;
236 static unsigned int cmd_save = 100;
237 static uintmax_t next_mark;
238 static struct strbuf new_data = STRBUF_INIT;
239 static int seen_data_command;
240 static int require_explicit_termination;
241 static int allow_unsafe_features;
243 /* Signal handling */
244 static volatile sig_atomic_t checkpoint_requested;
246 /* Submodule marks */
247 static struct string_list sub_marks_from = STRING_LIST_INIT_DUP;
248 static struct string_list sub_marks_to = STRING_LIST_INIT_DUP;
249 static kh_oid_map_t *sub_oid_map;
251 /* Where to write output of cat-blob commands */
252 static int cat_blob_fd = STDOUT_FILENO;
254 static void parse_argv(void);
255 static void parse_get_mark(const char *p);
256 static void parse_cat_blob(const char *p);
257 static void parse_ls(const char *p, struct branch *b);
259 static void for_each_mark(struct mark_set *m, uintmax_t base, each_mark_fn_t callback, void *p)
261 uintmax_t k;
262 if (m->shift) {
263 for (k = 0; k < 1024; k++) {
264 if (m->data.sets[k])
265 for_each_mark(m->data.sets[k], base + (k << m->shift), callback, p);
267 } else {
268 for (k = 0; k < 1024; k++) {
269 if (m->data.marked[k])
270 callback(base + k, m->data.marked[k], p);
275 static void dump_marks_fn(uintmax_t mark, void *object, void *cbp) {
276 struct object_entry *e = object;
277 FILE *f = cbp;
279 fprintf(f, ":%" PRIuMAX " %s\n", mark, oid_to_hex(&e->idx.oid));
282 static void write_branch_report(FILE *rpt, struct branch *b)
284 fprintf(rpt, "%s:\n", b->name);
286 fprintf(rpt, " status :");
287 if (b->active)
288 fputs(" active", rpt);
289 if (b->branch_tree.tree)
290 fputs(" loaded", rpt);
291 if (is_null_oid(&b->branch_tree.versions[1].oid))
292 fputs(" dirty", rpt);
293 fputc('\n', rpt);
295 fprintf(rpt, " tip commit : %s\n", oid_to_hex(&b->oid));
296 fprintf(rpt, " old tree : %s\n",
297 oid_to_hex(&b->branch_tree.versions[0].oid));
298 fprintf(rpt, " cur tree : %s\n",
299 oid_to_hex(&b->branch_tree.versions[1].oid));
300 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
302 fputs(" last pack : ", rpt);
303 if (b->pack_id < MAX_PACK_ID)
304 fprintf(rpt, "%u", b->pack_id);
305 fputc('\n', rpt);
307 fputc('\n', rpt);
310 static void write_crash_report(const char *err)
312 char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
313 FILE *rpt = fopen(loc, "w");
314 struct branch *b;
315 unsigned long lu;
316 struct recent_command *rc;
318 if (!rpt) {
319 error_errno("can't write crash report %s", loc);
320 free(loc);
321 return;
324 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
326 fprintf(rpt, "fast-import crash report:\n");
327 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
328 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
329 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
330 fputc('\n', rpt);
332 fputs("fatal: ", rpt);
333 fputs(err, rpt);
334 fputc('\n', rpt);
336 fputc('\n', rpt);
337 fputs("Most Recent Commands Before Crash\n", rpt);
338 fputs("---------------------------------\n", rpt);
339 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
340 if (rc->next == &cmd_hist)
341 fputs("* ", rpt);
342 else
343 fputs(" ", rpt);
344 fputs(rc->buf, rpt);
345 fputc('\n', rpt);
348 fputc('\n', rpt);
349 fputs("Active Branch LRU\n", rpt);
350 fputs("-----------------\n", rpt);
351 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
352 cur_active_branches,
353 max_active_branches);
354 fputc('\n', rpt);
355 fputs(" pos clock name\n", rpt);
356 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
357 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
358 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
359 ++lu, b->last_commit, b->name);
361 fputc('\n', rpt);
362 fputs("Inactive Branches\n", rpt);
363 fputs("-----------------\n", rpt);
364 for (lu = 0; lu < branch_table_sz; lu++) {
365 for (b = branch_table[lu]; b; b = b->table_next_branch)
366 write_branch_report(rpt, b);
369 if (first_tag) {
370 struct tag *tg;
371 fputc('\n', rpt);
372 fputs("Annotated Tags\n", rpt);
373 fputs("--------------\n", rpt);
374 for (tg = first_tag; tg; tg = tg->next_tag) {
375 fputs(oid_to_hex(&tg->oid), rpt);
376 fputc(' ', rpt);
377 fputs(tg->name, rpt);
378 fputc('\n', rpt);
382 fputc('\n', rpt);
383 fputs("Marks\n", rpt);
384 fputs("-----\n", rpt);
385 if (export_marks_file)
386 fprintf(rpt, " exported to %s\n", export_marks_file);
387 else
388 for_each_mark(marks, 0, dump_marks_fn, rpt);
390 fputc('\n', rpt);
391 fputs("-------------------\n", rpt);
392 fputs("END OF CRASH REPORT\n", rpt);
393 fclose(rpt);
394 free(loc);
397 static void end_packfile(void);
398 static void unkeep_all_packs(void);
399 static void dump_marks(void);
401 static NORETURN void die_nicely(const char *err, va_list params)
403 static int zombie;
404 char message[2 * PATH_MAX];
406 vsnprintf(message, sizeof(message), err, params);
407 fputs("fatal: ", stderr);
408 fputs(message, stderr);
409 fputc('\n', stderr);
411 if (!zombie) {
412 zombie = 1;
413 write_crash_report(message);
414 end_packfile();
415 unkeep_all_packs();
416 dump_marks();
418 exit(128);
421 #ifndef SIGUSR1 /* Windows, for example */
423 static void set_checkpoint_signal(void)
427 #else
429 static void checkpoint_signal(int signo)
431 checkpoint_requested = 1;
434 static void set_checkpoint_signal(void)
436 struct sigaction sa;
438 memset(&sa, 0, sizeof(sa));
439 sa.sa_handler = checkpoint_signal;
440 sigemptyset(&sa.sa_mask);
441 sa.sa_flags = SA_RESTART;
442 sigaction(SIGUSR1, &sa, NULL);
445 #endif
447 static void alloc_objects(unsigned int cnt)
449 struct object_entry_pool *b;
451 b = xmalloc(sizeof(struct object_entry_pool)
452 + cnt * sizeof(struct object_entry));
453 b->next_pool = blocks;
454 b->next_free = b->entries;
455 b->end = b->entries + cnt;
456 blocks = b;
457 alloc_count += cnt;
460 static struct object_entry *new_object(struct object_id *oid)
462 struct object_entry *e;
464 if (blocks->next_free == blocks->end)
465 alloc_objects(object_entry_alloc);
467 e = blocks->next_free++;
468 oidcpy(&e->idx.oid, oid);
469 return e;
472 static struct object_entry *find_object(struct object_id *oid)
474 return hashmap_get_entry_from_hash(&object_table, oidhash(oid), oid,
475 struct object_entry, ent);
478 static struct object_entry *insert_object(struct object_id *oid)
480 struct object_entry *e;
481 unsigned int hash = oidhash(oid);
483 e = hashmap_get_entry_from_hash(&object_table, hash, oid,
484 struct object_entry, ent);
485 if (!e) {
486 e = new_object(oid);
487 e->idx.offset = 0;
488 hashmap_entry_init(&e->ent, hash);
489 hashmap_add(&object_table, &e->ent);
492 return e;
495 static void invalidate_pack_id(unsigned int id)
497 unsigned long lu;
498 struct tag *t;
499 struct hashmap_iter iter;
500 struct object_entry *e;
502 hashmap_for_each_entry(&object_table, &iter, e, ent) {
503 if (e->pack_id == id)
504 e->pack_id = MAX_PACK_ID;
507 for (lu = 0; lu < branch_table_sz; lu++) {
508 struct branch *b;
510 for (b = branch_table[lu]; b; b = b->table_next_branch)
511 if (b->pack_id == id)
512 b->pack_id = MAX_PACK_ID;
515 for (t = first_tag; t; t = t->next_tag)
516 if (t->pack_id == id)
517 t->pack_id = MAX_PACK_ID;
520 static unsigned int hc_str(const char *s, size_t len)
522 unsigned int r = 0;
523 while (len-- > 0)
524 r = r * 31 + *s++;
525 return r;
528 static char *pool_strdup(const char *s)
530 size_t len = strlen(s) + 1;
531 char *r = mem_pool_alloc(&fi_mem_pool, len);
532 memcpy(r, s, len);
533 return r;
536 static void insert_mark(struct mark_set *s, uintmax_t idnum, struct object_entry *oe)
538 while ((idnum >> s->shift) >= 1024) {
539 s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
540 s->shift = marks->shift + 10;
541 s->data.sets[0] = marks;
542 marks = s;
544 while (s->shift) {
545 uintmax_t i = idnum >> s->shift;
546 idnum -= i << s->shift;
547 if (!s->data.sets[i]) {
548 s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
549 s->data.sets[i]->shift = s->shift - 10;
551 s = s->data.sets[i];
553 if (!s->data.marked[idnum])
554 marks_set_count++;
555 s->data.marked[idnum] = oe;
558 static void *find_mark(struct mark_set *s, uintmax_t idnum)
560 uintmax_t orig_idnum = idnum;
561 struct object_entry *oe = NULL;
562 if ((idnum >> s->shift) < 1024) {
563 while (s && s->shift) {
564 uintmax_t i = idnum >> s->shift;
565 idnum -= i << s->shift;
566 s = s->data.sets[i];
568 if (s)
569 oe = s->data.marked[idnum];
571 if (!oe)
572 die("mark :%" PRIuMAX " not declared", orig_idnum);
573 return oe;
576 static struct atom_str *to_atom(const char *s, unsigned short len)
578 unsigned int hc = hc_str(s, len) % atom_table_sz;
579 struct atom_str *c;
581 for (c = atom_table[hc]; c; c = c->next_atom)
582 if (c->str_len == len && !strncmp(s, c->str_dat, len))
583 return c;
585 c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
586 c->str_len = len;
587 memcpy(c->str_dat, s, len);
588 c->str_dat[len] = 0;
589 c->next_atom = atom_table[hc];
590 atom_table[hc] = c;
591 atom_cnt++;
592 return c;
595 static struct branch *lookup_branch(const char *name)
597 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
598 struct branch *b;
600 for (b = branch_table[hc]; b; b = b->table_next_branch)
601 if (!strcmp(name, b->name))
602 return b;
603 return NULL;
606 static struct branch *new_branch(const char *name)
608 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
609 struct branch *b = lookup_branch(name);
611 if (b)
612 die("Invalid attempt to create duplicate branch: %s", name);
613 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
614 die("Branch name doesn't conform to GIT standards: %s", name);
616 b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
617 b->name = pool_strdup(name);
618 b->table_next_branch = branch_table[hc];
619 b->branch_tree.versions[0].mode = S_IFDIR;
620 b->branch_tree.versions[1].mode = S_IFDIR;
621 b->num_notes = 0;
622 b->active = 0;
623 b->pack_id = MAX_PACK_ID;
624 branch_table[hc] = b;
625 branch_count++;
626 return b;
629 static unsigned int hc_entries(unsigned int cnt)
631 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
632 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
635 static struct tree_content *new_tree_content(unsigned int cnt)
637 struct avail_tree_content *f, *l = NULL;
638 struct tree_content *t;
639 unsigned int hc = hc_entries(cnt);
641 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
642 if (f->entry_capacity >= cnt)
643 break;
645 if (f) {
646 if (l)
647 l->next_avail = f->next_avail;
648 else
649 avail_tree_table[hc] = f->next_avail;
650 } else {
651 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
652 f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
653 f->entry_capacity = cnt;
656 t = (struct tree_content*)f;
657 t->entry_count = 0;
658 t->delta_depth = 0;
659 return t;
662 static void release_tree_entry(struct tree_entry *e);
663 static void release_tree_content(struct tree_content *t)
665 struct avail_tree_content *f = (struct avail_tree_content*)t;
666 unsigned int hc = hc_entries(f->entry_capacity);
667 f->next_avail = avail_tree_table[hc];
668 avail_tree_table[hc] = f;
671 static void release_tree_content_recursive(struct tree_content *t)
673 unsigned int i;
674 for (i = 0; i < t->entry_count; i++)
675 release_tree_entry(t->entries[i]);
676 release_tree_content(t);
679 static struct tree_content *grow_tree_content(
680 struct tree_content *t,
681 int amt)
683 struct tree_content *r = new_tree_content(t->entry_count + amt);
684 r->entry_count = t->entry_count;
685 r->delta_depth = t->delta_depth;
686 COPY_ARRAY(r->entries, t->entries, t->entry_count);
687 release_tree_content(t);
688 return r;
691 static struct tree_entry *new_tree_entry(void)
693 struct tree_entry *e;
695 if (!avail_tree_entry) {
696 unsigned int n = tree_entry_alloc;
697 tree_entry_allocd += n * sizeof(struct tree_entry);
698 ALLOC_ARRAY(e, n);
699 avail_tree_entry = e;
700 while (n-- > 1) {
701 *((void**)e) = e + 1;
702 e++;
704 *((void**)e) = NULL;
707 e = avail_tree_entry;
708 avail_tree_entry = *((void**)e);
709 return e;
712 static void release_tree_entry(struct tree_entry *e)
714 if (e->tree)
715 release_tree_content_recursive(e->tree);
716 *((void**)e) = avail_tree_entry;
717 avail_tree_entry = e;
720 static struct tree_content *dup_tree_content(struct tree_content *s)
722 struct tree_content *d;
723 struct tree_entry *a, *b;
724 unsigned int i;
726 if (!s)
727 return NULL;
728 d = new_tree_content(s->entry_count);
729 for (i = 0; i < s->entry_count; i++) {
730 a = s->entries[i];
731 b = new_tree_entry();
732 memcpy(b, a, sizeof(*a));
733 if (a->tree && is_null_oid(&b->versions[1].oid))
734 b->tree = dup_tree_content(a->tree);
735 else
736 b->tree = NULL;
737 d->entries[i] = b;
739 d->entry_count = s->entry_count;
740 d->delta_depth = s->delta_depth;
742 return d;
745 static void start_packfile(void)
747 struct strbuf tmp_file = STRBUF_INIT;
748 struct packed_git *p;
749 struct pack_header hdr;
750 int pack_fd;
752 pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
753 FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
754 strbuf_release(&tmp_file);
756 p->pack_fd = pack_fd;
757 p->do_not_close = 1;
758 pack_file = hashfd(pack_fd, p->pack_name);
760 hdr.hdr_signature = htonl(PACK_SIGNATURE);
761 hdr.hdr_version = htonl(2);
762 hdr.hdr_entries = 0;
763 hashwrite(pack_file, &hdr, sizeof(hdr));
765 pack_data = p;
766 pack_size = sizeof(hdr);
767 object_count = 0;
769 REALLOC_ARRAY(all_packs, pack_id + 1);
770 all_packs[pack_id] = p;
773 static const char *create_index(void)
775 const char *tmpfile;
776 struct pack_idx_entry **idx, **c, **last;
777 struct object_entry *e;
778 struct object_entry_pool *o;
780 /* Build the table of object IDs. */
781 ALLOC_ARRAY(idx, object_count);
782 c = idx;
783 for (o = blocks; o; o = o->next_pool)
784 for (e = o->next_free; e-- != o->entries;)
785 if (pack_id == e->pack_id)
786 *c++ = &e->idx;
787 last = idx + object_count;
788 if (c != last)
789 die("internal consistency error creating the index");
791 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts,
792 pack_data->hash);
793 free(idx);
794 return tmpfile;
797 static char *keep_pack(const char *curr_index_name)
799 static const char *keep_msg = "fast-import";
800 struct strbuf name = STRBUF_INIT;
801 int keep_fd;
803 odb_pack_name(&name, pack_data->hash, "keep");
804 keep_fd = odb_pack_keep(name.buf);
805 if (keep_fd < 0)
806 die_errno("cannot create keep file");
807 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
808 if (close(keep_fd))
809 die_errno("failed to write keep file");
811 odb_pack_name(&name, pack_data->hash, "pack");
812 if (finalize_object_file(pack_data->pack_name, name.buf))
813 die("cannot store pack file");
815 odb_pack_name(&name, pack_data->hash, "idx");
816 if (finalize_object_file(curr_index_name, name.buf))
817 die("cannot store index file");
818 free((void *)curr_index_name);
819 return strbuf_detach(&name, NULL);
822 static void unkeep_all_packs(void)
824 struct strbuf name = STRBUF_INIT;
825 int k;
827 for (k = 0; k < pack_id; k++) {
828 struct packed_git *p = all_packs[k];
829 odb_pack_name(&name, p->hash, "keep");
830 unlink_or_warn(name.buf);
832 strbuf_release(&name);
835 static int loosen_small_pack(const struct packed_git *p)
837 struct child_process unpack = CHILD_PROCESS_INIT;
839 if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
840 die_errno("Failed seeking to start of '%s'", p->pack_name);
842 unpack.in = p->pack_fd;
843 unpack.git_cmd = 1;
844 unpack.stdout_to_stderr = 1;
845 argv_array_push(&unpack.args, "unpack-objects");
846 if (!show_stats)
847 argv_array_push(&unpack.args, "-q");
849 return run_command(&unpack);
852 static void end_packfile(void)
854 static int running;
856 if (running || !pack_data)
857 return;
859 running = 1;
860 clear_delta_base_cache();
861 if (object_count) {
862 struct packed_git *new_p;
863 struct object_id cur_pack_oid;
864 char *idx_name;
865 int i;
866 struct branch *b;
867 struct tag *t;
869 close_pack_windows(pack_data);
870 finalize_hashfile(pack_file, cur_pack_oid.hash, 0);
871 fixup_pack_header_footer(pack_data->pack_fd, pack_data->hash,
872 pack_data->pack_name, object_count,
873 cur_pack_oid.hash, pack_size);
875 if (object_count <= unpack_limit) {
876 if (!loosen_small_pack(pack_data)) {
877 invalidate_pack_id(pack_id);
878 goto discard_pack;
882 close(pack_data->pack_fd);
883 idx_name = keep_pack(create_index());
885 /* Register the packfile with core git's machinery. */
886 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
887 if (!new_p)
888 die("core git rejected index %s", idx_name);
889 all_packs[pack_id] = new_p;
890 install_packed_git(the_repository, new_p);
891 free(idx_name);
893 /* Print the boundary */
894 if (pack_edges) {
895 fprintf(pack_edges, "%s:", new_p->pack_name);
896 for (i = 0; i < branch_table_sz; i++) {
897 for (b = branch_table[i]; b; b = b->table_next_branch) {
898 if (b->pack_id == pack_id)
899 fprintf(pack_edges, " %s",
900 oid_to_hex(&b->oid));
903 for (t = first_tag; t; t = t->next_tag) {
904 if (t->pack_id == pack_id)
905 fprintf(pack_edges, " %s",
906 oid_to_hex(&t->oid));
908 fputc('\n', pack_edges);
909 fflush(pack_edges);
912 pack_id++;
914 else {
915 discard_pack:
916 close(pack_data->pack_fd);
917 unlink_or_warn(pack_data->pack_name);
919 FREE_AND_NULL(pack_data);
920 running = 0;
922 /* We can't carry a delta across packfiles. */
923 strbuf_release(&last_blob.data);
924 last_blob.offset = 0;
925 last_blob.depth = 0;
928 static void cycle_packfile(void)
930 end_packfile();
931 start_packfile();
934 static int store_object(
935 enum object_type type,
936 struct strbuf *dat,
937 struct last_object *last,
938 struct object_id *oidout,
939 uintmax_t mark)
941 void *out, *delta;
942 struct object_entry *e;
943 unsigned char hdr[96];
944 struct object_id oid;
945 unsigned long hdrlen, deltalen;
946 git_hash_ctx c;
947 git_zstream s;
949 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
950 type_name(type), (unsigned long)dat->len) + 1;
951 the_hash_algo->init_fn(&c);
952 the_hash_algo->update_fn(&c, hdr, hdrlen);
953 the_hash_algo->update_fn(&c, dat->buf, dat->len);
954 the_hash_algo->final_fn(oid.hash, &c);
955 if (oidout)
956 oidcpy(oidout, &oid);
958 e = insert_object(&oid);
959 if (mark)
960 insert_mark(marks, mark, e);
961 if (e->idx.offset) {
962 duplicate_count_by_type[type]++;
963 return 1;
964 } else if (find_sha1_pack(oid.hash,
965 get_all_packs(the_repository))) {
966 e->type = type;
967 e->pack_id = MAX_PACK_ID;
968 e->idx.offset = 1; /* just not zero! */
969 duplicate_count_by_type[type]++;
970 return 1;
973 if (last && last->data.len && last->data.buf && last->depth < max_depth
974 && dat->len > the_hash_algo->rawsz) {
976 delta_count_attempts_by_type[type]++;
977 delta = diff_delta(last->data.buf, last->data.len,
978 dat->buf, dat->len,
979 &deltalen, dat->len - the_hash_algo->rawsz);
980 } else
981 delta = NULL;
983 git_deflate_init(&s, pack_compression_level);
984 if (delta) {
985 s.next_in = delta;
986 s.avail_in = deltalen;
987 } else {
988 s.next_in = (void *)dat->buf;
989 s.avail_in = dat->len;
991 s.avail_out = git_deflate_bound(&s, s.avail_in);
992 s.next_out = out = xmalloc(s.avail_out);
993 while (git_deflate(&s, Z_FINISH) == Z_OK)
994 ; /* nothing */
995 git_deflate_end(&s);
997 /* Determine if we should auto-checkpoint. */
998 if ((max_packsize
999 && (pack_size + PACK_SIZE_THRESHOLD + s.total_out) > max_packsize)
1000 || (pack_size + PACK_SIZE_THRESHOLD + s.total_out) < pack_size) {
1002 /* This new object needs to *not* have the current pack_id. */
1003 e->pack_id = pack_id + 1;
1004 cycle_packfile();
1006 /* We cannot carry a delta into the new pack. */
1007 if (delta) {
1008 FREE_AND_NULL(delta);
1010 git_deflate_init(&s, pack_compression_level);
1011 s.next_in = (void *)dat->buf;
1012 s.avail_in = dat->len;
1013 s.avail_out = git_deflate_bound(&s, s.avail_in);
1014 s.next_out = out = xrealloc(out, s.avail_out);
1015 while (git_deflate(&s, Z_FINISH) == Z_OK)
1016 ; /* nothing */
1017 git_deflate_end(&s);
1021 e->type = type;
1022 e->pack_id = pack_id;
1023 e->idx.offset = pack_size;
1024 object_count++;
1025 object_count_by_type[type]++;
1027 crc32_begin(pack_file);
1029 if (delta) {
1030 off_t ofs = e->idx.offset - last->offset;
1031 unsigned pos = sizeof(hdr) - 1;
1033 delta_count_by_type[type]++;
1034 e->depth = last->depth + 1;
1036 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1037 OBJ_OFS_DELTA, deltalen);
1038 hashwrite(pack_file, hdr, hdrlen);
1039 pack_size += hdrlen;
1041 hdr[pos] = ofs & 127;
1042 while (ofs >>= 7)
1043 hdr[--pos] = 128 | (--ofs & 127);
1044 hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1045 pack_size += sizeof(hdr) - pos;
1046 } else {
1047 e->depth = 0;
1048 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1049 type, dat->len);
1050 hashwrite(pack_file, hdr, hdrlen);
1051 pack_size += hdrlen;
1054 hashwrite(pack_file, out, s.total_out);
1055 pack_size += s.total_out;
1057 e->idx.crc32 = crc32_end(pack_file);
1059 free(out);
1060 free(delta);
1061 if (last) {
1062 if (last->no_swap) {
1063 last->data = *dat;
1064 } else {
1065 strbuf_swap(&last->data, dat);
1067 last->offset = e->idx.offset;
1068 last->depth = e->depth;
1070 return 0;
1073 static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1075 if (hashfile_truncate(pack_file, checkpoint))
1076 die_errno("cannot truncate pack to skip duplicate");
1077 pack_size = checkpoint->offset;
1080 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1082 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1083 unsigned char *in_buf = xmalloc(in_sz);
1084 unsigned char *out_buf = xmalloc(out_sz);
1085 struct object_entry *e;
1086 struct object_id oid;
1087 unsigned long hdrlen;
1088 off_t offset;
1089 git_hash_ctx c;
1090 git_zstream s;
1091 struct hashfile_checkpoint checkpoint;
1092 int status = Z_OK;
1094 /* Determine if we should auto-checkpoint. */
1095 if ((max_packsize
1096 && (pack_size + PACK_SIZE_THRESHOLD + len) > max_packsize)
1097 || (pack_size + PACK_SIZE_THRESHOLD + len) < pack_size)
1098 cycle_packfile();
1100 hashfile_checkpoint(pack_file, &checkpoint);
1101 offset = checkpoint.offset;
1103 hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1105 the_hash_algo->init_fn(&c);
1106 the_hash_algo->update_fn(&c, out_buf, hdrlen);
1108 crc32_begin(pack_file);
1110 git_deflate_init(&s, pack_compression_level);
1112 hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1114 s.next_out = out_buf + hdrlen;
1115 s.avail_out = out_sz - hdrlen;
1117 while (status != Z_STREAM_END) {
1118 if (0 < len && !s.avail_in) {
1119 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1120 size_t n = fread(in_buf, 1, cnt, stdin);
1121 if (!n && feof(stdin))
1122 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1124 the_hash_algo->update_fn(&c, in_buf, n);
1125 s.next_in = in_buf;
1126 s.avail_in = n;
1127 len -= n;
1130 status = git_deflate(&s, len ? 0 : Z_FINISH);
1132 if (!s.avail_out || status == Z_STREAM_END) {
1133 size_t n = s.next_out - out_buf;
1134 hashwrite(pack_file, out_buf, n);
1135 pack_size += n;
1136 s.next_out = out_buf;
1137 s.avail_out = out_sz;
1140 switch (status) {
1141 case Z_OK:
1142 case Z_BUF_ERROR:
1143 case Z_STREAM_END:
1144 continue;
1145 default:
1146 die("unexpected deflate failure: %d", status);
1149 git_deflate_end(&s);
1150 the_hash_algo->final_fn(oid.hash, &c);
1152 if (oidout)
1153 oidcpy(oidout, &oid);
1155 e = insert_object(&oid);
1157 if (mark)
1158 insert_mark(marks, mark, e);
1160 if (e->idx.offset) {
1161 duplicate_count_by_type[OBJ_BLOB]++;
1162 truncate_pack(&checkpoint);
1164 } else if (find_sha1_pack(oid.hash,
1165 get_all_packs(the_repository))) {
1166 e->type = OBJ_BLOB;
1167 e->pack_id = MAX_PACK_ID;
1168 e->idx.offset = 1; /* just not zero! */
1169 duplicate_count_by_type[OBJ_BLOB]++;
1170 truncate_pack(&checkpoint);
1172 } else {
1173 e->depth = 0;
1174 e->type = OBJ_BLOB;
1175 e->pack_id = pack_id;
1176 e->idx.offset = offset;
1177 e->idx.crc32 = crc32_end(pack_file);
1178 object_count++;
1179 object_count_by_type[OBJ_BLOB]++;
1182 free(in_buf);
1183 free(out_buf);
1186 /* All calls must be guarded by find_object() or find_mark() to
1187 * ensure the 'struct object_entry' passed was written by this
1188 * process instance. We unpack the entry by the offset, avoiding
1189 * the need for the corresponding .idx file. This unpacking rule
1190 * works because we only use OBJ_REF_DELTA within the packfiles
1191 * created by fast-import.
1193 * oe must not be NULL. Such an oe usually comes from giving
1194 * an unknown SHA-1 to find_object() or an undefined mark to
1195 * find_mark(). Callers must test for this condition and use
1196 * the standard read_sha1_file() when it happens.
1198 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1199 * find_mark(), where the mark was reloaded from an existing marks
1200 * file and is referencing an object that this fast-import process
1201 * instance did not write out to a packfile. Callers must test for
1202 * this condition and use read_sha1_file() instead.
1204 static void *gfi_unpack_entry(
1205 struct object_entry *oe,
1206 unsigned long *sizep)
1208 enum object_type type;
1209 struct packed_git *p = all_packs[oe->pack_id];
1210 if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1211 /* The object is stored in the packfile we are writing to
1212 * and we have modified it since the last time we scanned
1213 * back to read a previously written object. If an old
1214 * window covered [p->pack_size, p->pack_size + rawsz) its
1215 * data is stale and is not valid. Closing all windows
1216 * and updating the packfile length ensures we can read
1217 * the newly written data.
1219 close_pack_windows(p);
1220 hashflush(pack_file);
1222 /* We have to offer rawsz bytes additional on the end of
1223 * the packfile as the core unpacker code assumes the
1224 * footer is present at the file end and must promise
1225 * at least rawsz bytes within any window it maps. But
1226 * we don't actually create the footer here.
1228 p->pack_size = pack_size + the_hash_algo->rawsz;
1230 return unpack_entry(the_repository, p, oe->idx.offset, &type, sizep);
1233 static const char *get_mode(const char *str, uint16_t *modep)
1235 unsigned char c;
1236 uint16_t mode = 0;
1238 while ((c = *str++) != ' ') {
1239 if (c < '0' || c > '7')
1240 return NULL;
1241 mode = (mode << 3) + (c - '0');
1243 *modep = mode;
1244 return str;
1247 static void load_tree(struct tree_entry *root)
1249 struct object_id *oid = &root->versions[1].oid;
1250 struct object_entry *myoe;
1251 struct tree_content *t;
1252 unsigned long size;
1253 char *buf;
1254 const char *c;
1256 root->tree = t = new_tree_content(8);
1257 if (is_null_oid(oid))
1258 return;
1260 myoe = find_object(oid);
1261 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1262 if (myoe->type != OBJ_TREE)
1263 die("Not a tree: %s", oid_to_hex(oid));
1264 t->delta_depth = myoe->depth;
1265 buf = gfi_unpack_entry(myoe, &size);
1266 if (!buf)
1267 die("Can't load tree %s", oid_to_hex(oid));
1268 } else {
1269 enum object_type type;
1270 buf = read_object_file(oid, &type, &size);
1271 if (!buf || type != OBJ_TREE)
1272 die("Can't load tree %s", oid_to_hex(oid));
1275 c = buf;
1276 while (c != (buf + size)) {
1277 struct tree_entry *e = new_tree_entry();
1279 if (t->entry_count == t->entry_capacity)
1280 root->tree = t = grow_tree_content(t, t->entry_count);
1281 t->entries[t->entry_count++] = e;
1283 e->tree = NULL;
1284 c = get_mode(c, &e->versions[1].mode);
1285 if (!c)
1286 die("Corrupt mode in %s", oid_to_hex(oid));
1287 e->versions[0].mode = e->versions[1].mode;
1288 e->name = to_atom(c, strlen(c));
1289 c += e->name->str_len + 1;
1290 hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1291 hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1292 c += the_hash_algo->rawsz;
1294 free(buf);
1297 static int tecmp0 (const void *_a, const void *_b)
1299 struct tree_entry *a = *((struct tree_entry**)_a);
1300 struct tree_entry *b = *((struct tree_entry**)_b);
1301 return base_name_compare(
1302 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1303 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1306 static int tecmp1 (const void *_a, const void *_b)
1308 struct tree_entry *a = *((struct tree_entry**)_a);
1309 struct tree_entry *b = *((struct tree_entry**)_b);
1310 return base_name_compare(
1311 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1312 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1315 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1317 size_t maxlen = 0;
1318 unsigned int i;
1320 if (!v)
1321 QSORT(t->entries, t->entry_count, tecmp0);
1322 else
1323 QSORT(t->entries, t->entry_count, tecmp1);
1325 for (i = 0; i < t->entry_count; i++) {
1326 if (t->entries[i]->versions[v].mode)
1327 maxlen += t->entries[i]->name->str_len + 34;
1330 strbuf_reset(b);
1331 strbuf_grow(b, maxlen);
1332 for (i = 0; i < t->entry_count; i++) {
1333 struct tree_entry *e = t->entries[i];
1334 if (!e->versions[v].mode)
1335 continue;
1336 strbuf_addf(b, "%o %s%c",
1337 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1338 e->name->str_dat, '\0');
1339 strbuf_add(b, e->versions[v].oid.hash, the_hash_algo->rawsz);
1343 static void store_tree(struct tree_entry *root)
1345 struct tree_content *t;
1346 unsigned int i, j, del;
1347 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1348 struct object_entry *le = NULL;
1350 if (!is_null_oid(&root->versions[1].oid))
1351 return;
1353 if (!root->tree)
1354 load_tree(root);
1355 t = root->tree;
1357 for (i = 0; i < t->entry_count; i++) {
1358 if (t->entries[i]->tree)
1359 store_tree(t->entries[i]);
1362 if (!(root->versions[0].mode & NO_DELTA))
1363 le = find_object(&root->versions[0].oid);
1364 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1365 mktree(t, 0, &old_tree);
1366 lo.data = old_tree;
1367 lo.offset = le->idx.offset;
1368 lo.depth = t->delta_depth;
1371 mktree(t, 1, &new_tree);
1372 store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1374 t->delta_depth = lo.depth;
1375 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1376 struct tree_entry *e = t->entries[i];
1377 if (e->versions[1].mode) {
1378 e->versions[0].mode = e->versions[1].mode;
1379 oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1380 t->entries[j++] = e;
1381 } else {
1382 release_tree_entry(e);
1383 del++;
1386 t->entry_count -= del;
1389 static void tree_content_replace(
1390 struct tree_entry *root,
1391 const struct object_id *oid,
1392 const uint16_t mode,
1393 struct tree_content *newtree)
1395 if (!S_ISDIR(mode))
1396 die("Root cannot be a non-directory");
1397 oidclr(&root->versions[0].oid);
1398 oidcpy(&root->versions[1].oid, oid);
1399 if (root->tree)
1400 release_tree_content_recursive(root->tree);
1401 root->tree = newtree;
1404 static int tree_content_set(
1405 struct tree_entry *root,
1406 const char *p,
1407 const struct object_id *oid,
1408 const uint16_t mode,
1409 struct tree_content *subtree)
1411 struct tree_content *t;
1412 const char *slash1;
1413 unsigned int i, n;
1414 struct tree_entry *e;
1416 slash1 = strchrnul(p, '/');
1417 n = slash1 - p;
1418 if (!n)
1419 die("Empty path component found in input");
1420 if (!*slash1 && !S_ISDIR(mode) && subtree)
1421 die("Non-directories cannot have subtrees");
1423 if (!root->tree)
1424 load_tree(root);
1425 t = root->tree;
1426 for (i = 0; i < t->entry_count; i++) {
1427 e = t->entries[i];
1428 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1429 if (!*slash1) {
1430 if (!S_ISDIR(mode)
1431 && e->versions[1].mode == mode
1432 && oideq(&e->versions[1].oid, oid))
1433 return 0;
1434 e->versions[1].mode = mode;
1435 oidcpy(&e->versions[1].oid, oid);
1436 if (e->tree)
1437 release_tree_content_recursive(e->tree);
1438 e->tree = subtree;
1441 * We need to leave e->versions[0].sha1 alone
1442 * to avoid modifying the preimage tree used
1443 * when writing out the parent directory.
1444 * But after replacing the subdir with a
1445 * completely different one, it's not a good
1446 * delta base any more, and besides, we've
1447 * thrown away the tree entries needed to
1448 * make a delta against it.
1450 * So let's just explicitly disable deltas
1451 * for the subtree.
1453 if (S_ISDIR(e->versions[0].mode))
1454 e->versions[0].mode |= NO_DELTA;
1456 oidclr(&root->versions[1].oid);
1457 return 1;
1459 if (!S_ISDIR(e->versions[1].mode)) {
1460 e->tree = new_tree_content(8);
1461 e->versions[1].mode = S_IFDIR;
1463 if (!e->tree)
1464 load_tree(e);
1465 if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1466 oidclr(&root->versions[1].oid);
1467 return 1;
1469 return 0;
1473 if (t->entry_count == t->entry_capacity)
1474 root->tree = t = grow_tree_content(t, t->entry_count);
1475 e = new_tree_entry();
1476 e->name = to_atom(p, n);
1477 e->versions[0].mode = 0;
1478 oidclr(&e->versions[0].oid);
1479 t->entries[t->entry_count++] = e;
1480 if (*slash1) {
1481 e->tree = new_tree_content(8);
1482 e->versions[1].mode = S_IFDIR;
1483 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1484 } else {
1485 e->tree = subtree;
1486 e->versions[1].mode = mode;
1487 oidcpy(&e->versions[1].oid, oid);
1489 oidclr(&root->versions[1].oid);
1490 return 1;
1493 static int tree_content_remove(
1494 struct tree_entry *root,
1495 const char *p,
1496 struct tree_entry *backup_leaf,
1497 int allow_root)
1499 struct tree_content *t;
1500 const char *slash1;
1501 unsigned int i, n;
1502 struct tree_entry *e;
1504 slash1 = strchrnul(p, '/');
1505 n = slash1 - p;
1507 if (!root->tree)
1508 load_tree(root);
1510 if (!*p && allow_root) {
1511 e = root;
1512 goto del_entry;
1515 t = root->tree;
1516 for (i = 0; i < t->entry_count; i++) {
1517 e = t->entries[i];
1518 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1519 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1521 * If p names a file in some subdirectory, and a
1522 * file or symlink matching the name of the
1523 * parent directory of p exists, then p cannot
1524 * exist and need not be deleted.
1526 return 1;
1527 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1528 goto del_entry;
1529 if (!e->tree)
1530 load_tree(e);
1531 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1532 for (n = 0; n < e->tree->entry_count; n++) {
1533 if (e->tree->entries[n]->versions[1].mode) {
1534 oidclr(&root->versions[1].oid);
1535 return 1;
1538 backup_leaf = NULL;
1539 goto del_entry;
1541 return 0;
1544 return 0;
1546 del_entry:
1547 if (backup_leaf)
1548 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1549 else if (e->tree)
1550 release_tree_content_recursive(e->tree);
1551 e->tree = NULL;
1552 e->versions[1].mode = 0;
1553 oidclr(&e->versions[1].oid);
1554 oidclr(&root->versions[1].oid);
1555 return 1;
1558 static int tree_content_get(
1559 struct tree_entry *root,
1560 const char *p,
1561 struct tree_entry *leaf,
1562 int allow_root)
1564 struct tree_content *t;
1565 const char *slash1;
1566 unsigned int i, n;
1567 struct tree_entry *e;
1569 slash1 = strchrnul(p, '/');
1570 n = slash1 - p;
1571 if (!n && !allow_root)
1572 die("Empty path component found in input");
1574 if (!root->tree)
1575 load_tree(root);
1577 if (!n) {
1578 e = root;
1579 goto found_entry;
1582 t = root->tree;
1583 for (i = 0; i < t->entry_count; i++) {
1584 e = t->entries[i];
1585 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1586 if (!*slash1)
1587 goto found_entry;
1588 if (!S_ISDIR(e->versions[1].mode))
1589 return 0;
1590 if (!e->tree)
1591 load_tree(e);
1592 return tree_content_get(e, slash1 + 1, leaf, 0);
1595 return 0;
1597 found_entry:
1598 memcpy(leaf, e, sizeof(*leaf));
1599 if (e->tree && is_null_oid(&e->versions[1].oid))
1600 leaf->tree = dup_tree_content(e->tree);
1601 else
1602 leaf->tree = NULL;
1603 return 1;
1606 static int update_branch(struct branch *b)
1608 static const char *msg = "fast-import";
1609 struct ref_transaction *transaction;
1610 struct object_id old_oid;
1611 struct strbuf err = STRBUF_INIT;
1613 if (is_null_oid(&b->oid)) {
1614 if (b->delete)
1615 delete_ref(NULL, b->name, NULL, 0);
1616 return 0;
1618 if (read_ref(b->name, &old_oid))
1619 oidclr(&old_oid);
1620 if (!force_update && !is_null_oid(&old_oid)) {
1621 struct commit *old_cmit, *new_cmit;
1623 old_cmit = lookup_commit_reference_gently(the_repository,
1624 &old_oid, 0);
1625 new_cmit = lookup_commit_reference_gently(the_repository,
1626 &b->oid, 0);
1627 if (!old_cmit || !new_cmit)
1628 return error("Branch %s is missing commits.", b->name);
1630 if (!in_merge_bases(old_cmit, new_cmit)) {
1631 warning("Not updating %s"
1632 " (new tip %s does not contain %s)",
1633 b->name, oid_to_hex(&b->oid),
1634 oid_to_hex(&old_oid));
1635 return -1;
1638 transaction = ref_transaction_begin(&err);
1639 if (!transaction ||
1640 ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1641 0, msg, &err) ||
1642 ref_transaction_commit(transaction, &err)) {
1643 ref_transaction_free(transaction);
1644 error("%s", err.buf);
1645 strbuf_release(&err);
1646 return -1;
1648 ref_transaction_free(transaction);
1649 strbuf_release(&err);
1650 return 0;
1653 static void dump_branches(void)
1655 unsigned int i;
1656 struct branch *b;
1658 for (i = 0; i < branch_table_sz; i++) {
1659 for (b = branch_table[i]; b; b = b->table_next_branch)
1660 failure |= update_branch(b);
1664 static void dump_tags(void)
1666 static const char *msg = "fast-import";
1667 struct tag *t;
1668 struct strbuf ref_name = STRBUF_INIT;
1669 struct strbuf err = STRBUF_INIT;
1670 struct ref_transaction *transaction;
1672 transaction = ref_transaction_begin(&err);
1673 if (!transaction) {
1674 failure |= error("%s", err.buf);
1675 goto cleanup;
1677 for (t = first_tag; t; t = t->next_tag) {
1678 strbuf_reset(&ref_name);
1679 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1681 if (ref_transaction_update(transaction, ref_name.buf,
1682 &t->oid, NULL, 0, msg, &err)) {
1683 failure |= error("%s", err.buf);
1684 goto cleanup;
1687 if (ref_transaction_commit(transaction, &err))
1688 failure |= error("%s", err.buf);
1690 cleanup:
1691 ref_transaction_free(transaction);
1692 strbuf_release(&ref_name);
1693 strbuf_release(&err);
1696 static void dump_marks(void)
1698 struct lock_file mark_lock = LOCK_INIT;
1699 FILE *f;
1701 if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1702 return;
1704 if (safe_create_leading_directories_const(export_marks_file)) {
1705 failure |= error_errno("unable to create leading directories of %s",
1706 export_marks_file);
1707 return;
1710 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1711 failure |= error_errno("Unable to write marks file %s",
1712 export_marks_file);
1713 return;
1716 f = fdopen_lock_file(&mark_lock, "w");
1717 if (!f) {
1718 int saved_errno = errno;
1719 rollback_lock_file(&mark_lock);
1720 failure |= error("Unable to write marks file %s: %s",
1721 export_marks_file, strerror(saved_errno));
1722 return;
1725 for_each_mark(marks, 0, dump_marks_fn, f);
1726 if (commit_lock_file(&mark_lock)) {
1727 failure |= error_errno("Unable to write file %s",
1728 export_marks_file);
1729 return;
1733 static void insert_object_entry(struct mark_set *s, struct object_id *oid, uintmax_t mark)
1735 struct object_entry *e;
1736 e = find_object(oid);
1737 if (!e) {
1738 enum object_type type = oid_object_info(the_repository,
1739 oid, NULL);
1740 if (type < 0)
1741 die("object not found: %s", oid_to_hex(oid));
1742 e = insert_object(oid);
1743 e->type = type;
1744 e->pack_id = MAX_PACK_ID;
1745 e->idx.offset = 1; /* just not zero! */
1747 insert_mark(s, mark, e);
1750 static void insert_oid_entry(struct mark_set *s, struct object_id *oid, uintmax_t mark)
1752 insert_mark(s, mark, xmemdupz(oid, sizeof(*oid)));
1755 static void read_mark_file(struct mark_set *s, FILE *f, mark_set_inserter_t inserter)
1757 char line[512];
1758 while (fgets(line, sizeof(line), f)) {
1759 uintmax_t mark;
1760 char *end;
1761 struct object_id oid;
1763 /* Ensure SHA-1 objects are padded with zeros. */
1764 memset(oid.hash, 0, sizeof(oid.hash));
1766 end = strchr(line, '\n');
1767 if (line[0] != ':' || !end)
1768 die("corrupt mark line: %s", line);
1769 *end = 0;
1770 mark = strtoumax(line + 1, &end, 10);
1771 if (!mark || end == line + 1
1772 || *end != ' '
1773 || get_oid_hex_any(end + 1, &oid) == GIT_HASH_UNKNOWN)
1774 die("corrupt mark line: %s", line);
1775 inserter(s, &oid, mark);
1779 static void read_marks(void)
1781 FILE *f = fopen(import_marks_file, "r");
1782 if (f)
1784 else if (import_marks_file_ignore_missing && errno == ENOENT)
1785 goto done; /* Marks file does not exist */
1786 else
1787 die_errno("cannot read '%s'", import_marks_file);
1788 read_mark_file(marks, f, insert_object_entry);
1789 fclose(f);
1790 done:
1791 import_marks_file_done = 1;
1795 static int read_next_command(void)
1797 static int stdin_eof = 0;
1799 if (stdin_eof) {
1800 unread_command_buf = 0;
1801 return EOF;
1804 for (;;) {
1805 if (unread_command_buf) {
1806 unread_command_buf = 0;
1807 } else {
1808 struct recent_command *rc;
1810 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1811 if (stdin_eof)
1812 return EOF;
1814 if (!seen_data_command
1815 && !starts_with(command_buf.buf, "feature ")
1816 && !starts_with(command_buf.buf, "option ")) {
1817 parse_argv();
1820 rc = rc_free;
1821 if (rc)
1822 rc_free = rc->next;
1823 else {
1824 rc = cmd_hist.next;
1825 cmd_hist.next = rc->next;
1826 cmd_hist.next->prev = &cmd_hist;
1827 free(rc->buf);
1830 rc->buf = xstrdup(command_buf.buf);
1831 rc->prev = cmd_tail;
1832 rc->next = cmd_hist.prev;
1833 rc->prev->next = rc;
1834 cmd_tail = rc;
1836 if (command_buf.buf[0] == '#')
1837 continue;
1838 return 0;
1842 static void skip_optional_lf(void)
1844 int term_char = fgetc(stdin);
1845 if (term_char != '\n' && term_char != EOF)
1846 ungetc(term_char, stdin);
1849 static void parse_mark(void)
1851 const char *v;
1852 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1853 next_mark = strtoumax(v, NULL, 10);
1854 read_next_command();
1856 else
1857 next_mark = 0;
1860 static void parse_original_identifier(void)
1862 const char *v;
1863 if (skip_prefix(command_buf.buf, "original-oid ", &v))
1864 read_next_command();
1867 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1869 const char *data;
1870 strbuf_reset(sb);
1872 if (!skip_prefix(command_buf.buf, "data ", &data))
1873 die("Expected 'data n' command, found: %s", command_buf.buf);
1875 if (skip_prefix(data, "<<", &data)) {
1876 char *term = xstrdup(data);
1877 size_t term_len = command_buf.len - (data - command_buf.buf);
1879 for (;;) {
1880 if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1881 die("EOF in data (terminator '%s' not found)", term);
1882 if (term_len == command_buf.len
1883 && !strcmp(term, command_buf.buf))
1884 break;
1885 strbuf_addbuf(sb, &command_buf);
1886 strbuf_addch(sb, '\n');
1888 free(term);
1890 else {
1891 uintmax_t len = strtoumax(data, NULL, 10);
1892 size_t n = 0, length = (size_t)len;
1894 if (limit && limit < len) {
1895 *len_res = len;
1896 return 0;
1898 if (length < len)
1899 die("data is too large to use in this context");
1901 while (n < length) {
1902 size_t s = strbuf_fread(sb, length - n, stdin);
1903 if (!s && feof(stdin))
1904 die("EOF in data (%lu bytes remaining)",
1905 (unsigned long)(length - n));
1906 n += s;
1910 skip_optional_lf();
1911 return 1;
1914 static int validate_raw_date(const char *src, struct strbuf *result)
1916 const char *orig_src = src;
1917 char *endp;
1918 unsigned long num;
1920 errno = 0;
1922 num = strtoul(src, &endp, 10);
1923 /* NEEDSWORK: perhaps check for reasonable values? */
1924 if (errno || endp == src || *endp != ' ')
1925 return -1;
1927 src = endp + 1;
1928 if (*src != '-' && *src != '+')
1929 return -1;
1931 num = strtoul(src + 1, &endp, 10);
1932 if (errno || endp == src + 1 || *endp || 1400 < num)
1933 return -1;
1935 strbuf_addstr(result, orig_src);
1936 return 0;
1939 static char *parse_ident(const char *buf)
1941 const char *ltgt;
1942 size_t name_len;
1943 struct strbuf ident = STRBUF_INIT;
1945 /* ensure there is a space delimiter even if there is no name */
1946 if (*buf == '<')
1947 --buf;
1949 ltgt = buf + strcspn(buf, "<>");
1950 if (*ltgt != '<')
1951 die("Missing < in ident string: %s", buf);
1952 if (ltgt != buf && ltgt[-1] != ' ')
1953 die("Missing space before < in ident string: %s", buf);
1954 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
1955 if (*ltgt != '>')
1956 die("Missing > in ident string: %s", buf);
1957 ltgt++;
1958 if (*ltgt != ' ')
1959 die("Missing space after > in ident string: %s", buf);
1960 ltgt++;
1961 name_len = ltgt - buf;
1962 strbuf_add(&ident, buf, name_len);
1964 switch (whenspec) {
1965 case WHENSPEC_RAW:
1966 if (validate_raw_date(ltgt, &ident) < 0)
1967 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1968 break;
1969 case WHENSPEC_RFC2822:
1970 if (parse_date(ltgt, &ident) < 0)
1971 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
1972 break;
1973 case WHENSPEC_NOW:
1974 if (strcmp("now", ltgt))
1975 die("Date in ident must be 'now': %s", buf);
1976 datestamp(&ident);
1977 break;
1980 return strbuf_detach(&ident, NULL);
1983 static void parse_and_store_blob(
1984 struct last_object *last,
1985 struct object_id *oidout,
1986 uintmax_t mark)
1988 static struct strbuf buf = STRBUF_INIT;
1989 uintmax_t len;
1991 if (parse_data(&buf, big_file_threshold, &len))
1992 store_object(OBJ_BLOB, &buf, last, oidout, mark);
1993 else {
1994 if (last) {
1995 strbuf_release(&last->data);
1996 last->offset = 0;
1997 last->depth = 0;
1999 stream_blob(len, oidout, mark);
2000 skip_optional_lf();
2004 static void parse_new_blob(void)
2006 read_next_command();
2007 parse_mark();
2008 parse_original_identifier();
2009 parse_and_store_blob(&last_blob, NULL, next_mark);
2012 static void unload_one_branch(void)
2014 while (cur_active_branches
2015 && cur_active_branches >= max_active_branches) {
2016 uintmax_t min_commit = ULONG_MAX;
2017 struct branch *e, *l = NULL, *p = NULL;
2019 for (e = active_branches; e; e = e->active_next_branch) {
2020 if (e->last_commit < min_commit) {
2021 p = l;
2022 min_commit = e->last_commit;
2024 l = e;
2027 if (p) {
2028 e = p->active_next_branch;
2029 p->active_next_branch = e->active_next_branch;
2030 } else {
2031 e = active_branches;
2032 active_branches = e->active_next_branch;
2034 e->active = 0;
2035 e->active_next_branch = NULL;
2036 if (e->branch_tree.tree) {
2037 release_tree_content_recursive(e->branch_tree.tree);
2038 e->branch_tree.tree = NULL;
2040 cur_active_branches--;
2044 static void load_branch(struct branch *b)
2046 load_tree(&b->branch_tree);
2047 if (!b->active) {
2048 b->active = 1;
2049 b->active_next_branch = active_branches;
2050 active_branches = b;
2051 cur_active_branches++;
2052 branch_load_count++;
2056 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2058 unsigned char fanout = 0;
2059 while ((num_notes >>= 8))
2060 fanout++;
2061 return fanout;
2064 static void construct_path_with_fanout(const char *hex_sha1,
2065 unsigned char fanout, char *path)
2067 unsigned int i = 0, j = 0;
2068 if (fanout >= the_hash_algo->rawsz)
2069 die("Too large fanout (%u)", fanout);
2070 while (fanout) {
2071 path[i++] = hex_sha1[j++];
2072 path[i++] = hex_sha1[j++];
2073 path[i++] = '/';
2074 fanout--;
2076 memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2077 path[i + the_hash_algo->hexsz - j] = '\0';
2080 static uintmax_t do_change_note_fanout(
2081 struct tree_entry *orig_root, struct tree_entry *root,
2082 char *hex_oid, unsigned int hex_oid_len,
2083 char *fullpath, unsigned int fullpath_len,
2084 unsigned char fanout)
2086 struct tree_content *t;
2087 struct tree_entry *e, leaf;
2088 unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2089 uintmax_t num_notes = 0;
2090 struct object_id oid;
2091 /* hex oid + '/' between each pair of hex digits + NUL */
2092 char realpath[GIT_MAX_HEXSZ + ((GIT_MAX_HEXSZ / 2) - 1) + 1];
2093 const unsigned hexsz = the_hash_algo->hexsz;
2095 if (!root->tree)
2096 load_tree(root);
2097 t = root->tree;
2099 for (i = 0; t && i < t->entry_count; i++) {
2100 e = t->entries[i];
2101 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2102 tmp_fullpath_len = fullpath_len;
2105 * We're interested in EITHER existing note entries (entries
2106 * with exactly 40 hex chars in path, not including directory
2107 * separators), OR directory entries that may contain note
2108 * entries (with < 40 hex chars in path).
2109 * Also, each path component in a note entry must be a multiple
2110 * of 2 chars.
2112 if (!e->versions[1].mode ||
2113 tmp_hex_oid_len > hexsz ||
2114 e->name->str_len % 2)
2115 continue;
2117 /* This _may_ be a note entry, or a subdir containing notes */
2118 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2119 e->name->str_len);
2120 if (tmp_fullpath_len)
2121 fullpath[tmp_fullpath_len++] = '/';
2122 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2123 e->name->str_len);
2124 tmp_fullpath_len += e->name->str_len;
2125 fullpath[tmp_fullpath_len] = '\0';
2127 if (tmp_hex_oid_len == hexsz && !get_oid_hex(hex_oid, &oid)) {
2128 /* This is a note entry */
2129 if (fanout == 0xff) {
2130 /* Counting mode, no rename */
2131 num_notes++;
2132 continue;
2134 construct_path_with_fanout(hex_oid, fanout, realpath);
2135 if (!strcmp(fullpath, realpath)) {
2136 /* Note entry is in correct location */
2137 num_notes++;
2138 continue;
2141 /* Rename fullpath to realpath */
2142 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2143 die("Failed to remove path %s", fullpath);
2144 tree_content_set(orig_root, realpath,
2145 &leaf.versions[1].oid,
2146 leaf.versions[1].mode,
2147 leaf.tree);
2148 } else if (S_ISDIR(e->versions[1].mode)) {
2149 /* This is a subdir that may contain note entries */
2150 num_notes += do_change_note_fanout(orig_root, e,
2151 hex_oid, tmp_hex_oid_len,
2152 fullpath, tmp_fullpath_len, fanout);
2155 /* The above may have reallocated the current tree_content */
2156 t = root->tree;
2158 return num_notes;
2161 static uintmax_t change_note_fanout(struct tree_entry *root,
2162 unsigned char fanout)
2165 * The size of path is due to one slash between every two hex digits,
2166 * plus the terminating NUL. Note that there is no slash at the end, so
2167 * the number of slashes is one less than half the number of hex
2168 * characters.
2170 char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2171 return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2174 static int parse_mapped_oid_hex(const char *hex, struct object_id *oid, const char **end)
2176 int algo;
2177 khiter_t it;
2179 /* Make SHA-1 object IDs have all-zero padding. */
2180 memset(oid->hash, 0, sizeof(oid->hash));
2182 algo = parse_oid_hex_any(hex, oid, end);
2183 if (algo == GIT_HASH_UNKNOWN)
2184 return -1;
2186 it = kh_get_oid_map(sub_oid_map, *oid);
2187 /* No such object? */
2188 if (it == kh_end(sub_oid_map)) {
2189 /* If we're using the same algorithm, pass it through. */
2190 if (hash_algos[algo].format_id == the_hash_algo->format_id)
2191 return 0;
2192 return -1;
2194 oidcpy(oid, kh_value(sub_oid_map, it));
2195 return 0;
2199 * Given a pointer into a string, parse a mark reference:
2201 * idnum ::= ':' bigint;
2203 * Return the first character after the value in *endptr.
2205 * Complain if the following character is not what is expected,
2206 * either a space or end of the string.
2208 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2210 uintmax_t mark;
2212 assert(*p == ':');
2213 p++;
2214 mark = strtoumax(p, endptr, 10);
2215 if (*endptr == p)
2216 die("No value after ':' in mark: %s", command_buf.buf);
2217 return mark;
2221 * Parse the mark reference, and complain if this is not the end of
2222 * the string.
2224 static uintmax_t parse_mark_ref_eol(const char *p)
2226 char *end;
2227 uintmax_t mark;
2229 mark = parse_mark_ref(p, &end);
2230 if (*end != '\0')
2231 die("Garbage after mark: %s", command_buf.buf);
2232 return mark;
2236 * Parse the mark reference, demanding a trailing space. Return a
2237 * pointer to the space.
2239 static uintmax_t parse_mark_ref_space(const char **p)
2241 uintmax_t mark;
2242 char *end;
2244 mark = parse_mark_ref(*p, &end);
2245 if (*end++ != ' ')
2246 die("Missing space after mark: %s", command_buf.buf);
2247 *p = end;
2248 return mark;
2251 static void file_change_m(const char *p, struct branch *b)
2253 static struct strbuf uq = STRBUF_INIT;
2254 const char *endp;
2255 struct object_entry *oe;
2256 struct object_id oid;
2257 uint16_t mode, inline_data = 0;
2259 p = get_mode(p, &mode);
2260 if (!p)
2261 die("Corrupt mode: %s", command_buf.buf);
2262 switch (mode) {
2263 case 0644:
2264 case 0755:
2265 mode |= S_IFREG;
2266 case S_IFREG | 0644:
2267 case S_IFREG | 0755:
2268 case S_IFLNK:
2269 case S_IFDIR:
2270 case S_IFGITLINK:
2271 /* ok */
2272 break;
2273 default:
2274 die("Corrupt mode: %s", command_buf.buf);
2277 if (*p == ':') {
2278 oe = find_mark(marks, parse_mark_ref_space(&p));
2279 oidcpy(&oid, &oe->idx.oid);
2280 } else if (skip_prefix(p, "inline ", &p)) {
2281 inline_data = 1;
2282 oe = NULL; /* not used with inline_data, but makes gcc happy */
2283 } else {
2284 if (parse_mapped_oid_hex(p, &oid, &p))
2285 die("Invalid dataref: %s", command_buf.buf);
2286 oe = find_object(&oid);
2287 if (*p++ != ' ')
2288 die("Missing space after SHA1: %s", command_buf.buf);
2291 strbuf_reset(&uq);
2292 if (!unquote_c_style(&uq, p, &endp)) {
2293 if (*endp)
2294 die("Garbage after path in: %s", command_buf.buf);
2295 p = uq.buf;
2298 /* Git does not track empty, non-toplevel directories. */
2299 if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2300 tree_content_remove(&b->branch_tree, p, NULL, 0);
2301 return;
2304 if (S_ISGITLINK(mode)) {
2305 if (inline_data)
2306 die("Git links cannot be specified 'inline': %s",
2307 command_buf.buf);
2308 else if (oe) {
2309 if (oe->type != OBJ_COMMIT)
2310 die("Not a commit (actually a %s): %s",
2311 type_name(oe->type), command_buf.buf);
2314 * Accept the sha1 without checking; it expected to be in
2315 * another repository.
2317 } else if (inline_data) {
2318 if (S_ISDIR(mode))
2319 die("Directories cannot be specified 'inline': %s",
2320 command_buf.buf);
2321 if (p != uq.buf) {
2322 strbuf_addstr(&uq, p);
2323 p = uq.buf;
2325 while (read_next_command() != EOF) {
2326 const char *v;
2327 if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2328 parse_cat_blob(v);
2329 else {
2330 parse_and_store_blob(&last_blob, &oid, 0);
2331 break;
2334 } else {
2335 enum object_type expected = S_ISDIR(mode) ?
2336 OBJ_TREE: OBJ_BLOB;
2337 enum object_type type = oe ? oe->type :
2338 oid_object_info(the_repository, &oid,
2339 NULL);
2340 if (type < 0)
2341 die("%s not found: %s",
2342 S_ISDIR(mode) ? "Tree" : "Blob",
2343 command_buf.buf);
2344 if (type != expected)
2345 die("Not a %s (actually a %s): %s",
2346 type_name(expected), type_name(type),
2347 command_buf.buf);
2350 if (!*p) {
2351 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2352 return;
2354 tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2357 static void file_change_d(const char *p, struct branch *b)
2359 static struct strbuf uq = STRBUF_INIT;
2360 const char *endp;
2362 strbuf_reset(&uq);
2363 if (!unquote_c_style(&uq, p, &endp)) {
2364 if (*endp)
2365 die("Garbage after path in: %s", command_buf.buf);
2366 p = uq.buf;
2368 tree_content_remove(&b->branch_tree, p, NULL, 1);
2371 static void file_change_cr(const char *s, struct branch *b, int rename)
2373 const char *d;
2374 static struct strbuf s_uq = STRBUF_INIT;
2375 static struct strbuf d_uq = STRBUF_INIT;
2376 const char *endp;
2377 struct tree_entry leaf;
2379 strbuf_reset(&s_uq);
2380 if (!unquote_c_style(&s_uq, s, &endp)) {
2381 if (*endp != ' ')
2382 die("Missing space after source: %s", command_buf.buf);
2383 } else {
2384 endp = strchr(s, ' ');
2385 if (!endp)
2386 die("Missing space after source: %s", command_buf.buf);
2387 strbuf_add(&s_uq, s, endp - s);
2389 s = s_uq.buf;
2391 endp++;
2392 if (!*endp)
2393 die("Missing dest: %s", command_buf.buf);
2395 d = endp;
2396 strbuf_reset(&d_uq);
2397 if (!unquote_c_style(&d_uq, d, &endp)) {
2398 if (*endp)
2399 die("Garbage after dest in: %s", command_buf.buf);
2400 d = d_uq.buf;
2403 memset(&leaf, 0, sizeof(leaf));
2404 if (rename)
2405 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2406 else
2407 tree_content_get(&b->branch_tree, s, &leaf, 1);
2408 if (!leaf.versions[1].mode)
2409 die("Path %s not in branch", s);
2410 if (!*d) { /* C "path/to/subdir" "" */
2411 tree_content_replace(&b->branch_tree,
2412 &leaf.versions[1].oid,
2413 leaf.versions[1].mode,
2414 leaf.tree);
2415 return;
2417 tree_content_set(&b->branch_tree, d,
2418 &leaf.versions[1].oid,
2419 leaf.versions[1].mode,
2420 leaf.tree);
2423 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2425 static struct strbuf uq = STRBUF_INIT;
2426 struct object_entry *oe;
2427 struct branch *s;
2428 struct object_id oid, commit_oid;
2429 char path[GIT_MAX_RAWSZ * 3];
2430 uint16_t inline_data = 0;
2431 unsigned char new_fanout;
2434 * When loading a branch, we don't traverse its tree to count the real
2435 * number of notes (too expensive to do this for all non-note refs).
2436 * This means that recently loaded notes refs might incorrectly have
2437 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2439 * Fix this by traversing the tree and counting the number of notes
2440 * when b->num_notes == 0. If the notes tree is truly empty, the
2441 * calculation should not take long.
2443 if (b->num_notes == 0 && *old_fanout == 0) {
2444 /* Invoke change_note_fanout() in "counting mode". */
2445 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2446 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2449 /* Now parse the notemodify command. */
2450 /* <dataref> or 'inline' */
2451 if (*p == ':') {
2452 oe = find_mark(marks, parse_mark_ref_space(&p));
2453 oidcpy(&oid, &oe->idx.oid);
2454 } else if (skip_prefix(p, "inline ", &p)) {
2455 inline_data = 1;
2456 oe = NULL; /* not used with inline_data, but makes gcc happy */
2457 } else {
2458 if (parse_mapped_oid_hex(p, &oid, &p))
2459 die("Invalid dataref: %s", command_buf.buf);
2460 oe = find_object(&oid);
2461 if (*p++ != ' ')
2462 die("Missing space after SHA1: %s", command_buf.buf);
2465 /* <commit-ish> */
2466 s = lookup_branch(p);
2467 if (s) {
2468 if (is_null_oid(&s->oid))
2469 die("Can't add a note on empty branch.");
2470 oidcpy(&commit_oid, &s->oid);
2471 } else if (*p == ':') {
2472 uintmax_t commit_mark = parse_mark_ref_eol(p);
2473 struct object_entry *commit_oe = find_mark(marks, commit_mark);
2474 if (commit_oe->type != OBJ_COMMIT)
2475 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2476 oidcpy(&commit_oid, &commit_oe->idx.oid);
2477 } else if (!get_oid(p, &commit_oid)) {
2478 unsigned long size;
2479 char *buf = read_object_with_reference(the_repository,
2480 &commit_oid,
2481 commit_type, &size,
2482 &commit_oid);
2483 if (!buf || size < the_hash_algo->hexsz + 6)
2484 die("Not a valid commit: %s", p);
2485 free(buf);
2486 } else
2487 die("Invalid ref name or SHA1 expression: %s", p);
2489 if (inline_data) {
2490 if (p != uq.buf) {
2491 strbuf_addstr(&uq, p);
2492 p = uq.buf;
2494 read_next_command();
2495 parse_and_store_blob(&last_blob, &oid, 0);
2496 } else if (oe) {
2497 if (oe->type != OBJ_BLOB)
2498 die("Not a blob (actually a %s): %s",
2499 type_name(oe->type), command_buf.buf);
2500 } else if (!is_null_oid(&oid)) {
2501 enum object_type type = oid_object_info(the_repository, &oid,
2502 NULL);
2503 if (type < 0)
2504 die("Blob not found: %s", command_buf.buf);
2505 if (type != OBJ_BLOB)
2506 die("Not a blob (actually a %s): %s",
2507 type_name(type), command_buf.buf);
2510 construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2511 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2512 b->num_notes--;
2514 if (is_null_oid(&oid))
2515 return; /* nothing to insert */
2517 b->num_notes++;
2518 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2519 construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2520 tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2523 static void file_change_deleteall(struct branch *b)
2525 release_tree_content_recursive(b->branch_tree.tree);
2526 oidclr(&b->branch_tree.versions[0].oid);
2527 oidclr(&b->branch_tree.versions[1].oid);
2528 load_tree(&b->branch_tree);
2529 b->num_notes = 0;
2532 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2534 if (!buf || size < the_hash_algo->hexsz + 6)
2535 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2536 if (memcmp("tree ", buf, 5)
2537 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2538 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2539 oidcpy(&b->branch_tree.versions[0].oid,
2540 &b->branch_tree.versions[1].oid);
2543 static void parse_from_existing(struct branch *b)
2545 if (is_null_oid(&b->oid)) {
2546 oidclr(&b->branch_tree.versions[0].oid);
2547 oidclr(&b->branch_tree.versions[1].oid);
2548 } else {
2549 unsigned long size;
2550 char *buf;
2552 buf = read_object_with_reference(the_repository,
2553 &b->oid, commit_type, &size,
2554 &b->oid);
2555 parse_from_commit(b, buf, size);
2556 free(buf);
2560 static int parse_objectish(struct branch *b, const char *objectish)
2562 struct branch *s;
2563 struct object_id oid;
2565 oidcpy(&oid, &b->branch_tree.versions[1].oid);
2567 s = lookup_branch(objectish);
2568 if (b == s)
2569 die("Can't create a branch from itself: %s", b->name);
2570 else if (s) {
2571 struct object_id *t = &s->branch_tree.versions[1].oid;
2572 oidcpy(&b->oid, &s->oid);
2573 oidcpy(&b->branch_tree.versions[0].oid, t);
2574 oidcpy(&b->branch_tree.versions[1].oid, t);
2575 } else if (*objectish == ':') {
2576 uintmax_t idnum = parse_mark_ref_eol(objectish);
2577 struct object_entry *oe = find_mark(marks, idnum);
2578 if (oe->type != OBJ_COMMIT)
2579 die("Mark :%" PRIuMAX " not a commit", idnum);
2580 if (!oideq(&b->oid, &oe->idx.oid)) {
2581 oidcpy(&b->oid, &oe->idx.oid);
2582 if (oe->pack_id != MAX_PACK_ID) {
2583 unsigned long size;
2584 char *buf = gfi_unpack_entry(oe, &size);
2585 parse_from_commit(b, buf, size);
2586 free(buf);
2587 } else
2588 parse_from_existing(b);
2590 } else if (!get_oid(objectish, &b->oid)) {
2591 parse_from_existing(b);
2592 if (is_null_oid(&b->oid))
2593 b->delete = 1;
2595 else
2596 die("Invalid ref name or SHA1 expression: %s", objectish);
2598 if (b->branch_tree.tree && !oideq(&oid, &b->branch_tree.versions[1].oid)) {
2599 release_tree_content_recursive(b->branch_tree.tree);
2600 b->branch_tree.tree = NULL;
2603 read_next_command();
2604 return 1;
2607 static int parse_from(struct branch *b)
2609 const char *from;
2611 if (!skip_prefix(command_buf.buf, "from ", &from))
2612 return 0;
2614 return parse_objectish(b, from);
2617 static int parse_objectish_with_prefix(struct branch *b, const char *prefix)
2619 const char *base;
2621 if (!skip_prefix(command_buf.buf, prefix, &base))
2622 return 0;
2624 return parse_objectish(b, base);
2627 static struct hash_list *parse_merge(unsigned int *count)
2629 struct hash_list *list = NULL, **tail = &list, *n;
2630 const char *from;
2631 struct branch *s;
2633 *count = 0;
2634 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2635 n = xmalloc(sizeof(*n));
2636 s = lookup_branch(from);
2637 if (s)
2638 oidcpy(&n->oid, &s->oid);
2639 else if (*from == ':') {
2640 uintmax_t idnum = parse_mark_ref_eol(from);
2641 struct object_entry *oe = find_mark(marks, idnum);
2642 if (oe->type != OBJ_COMMIT)
2643 die("Mark :%" PRIuMAX " not a commit", idnum);
2644 oidcpy(&n->oid, &oe->idx.oid);
2645 } else if (!get_oid(from, &n->oid)) {
2646 unsigned long size;
2647 char *buf = read_object_with_reference(the_repository,
2648 &n->oid,
2649 commit_type,
2650 &size, &n->oid);
2651 if (!buf || size < the_hash_algo->hexsz + 6)
2652 die("Not a valid commit: %s", from);
2653 free(buf);
2654 } else
2655 die("Invalid ref name or SHA1 expression: %s", from);
2657 n->next = NULL;
2658 *tail = n;
2659 tail = &n->next;
2661 (*count)++;
2662 read_next_command();
2664 return list;
2667 static void parse_new_commit(const char *arg)
2669 static struct strbuf msg = STRBUF_INIT;
2670 struct branch *b;
2671 char *author = NULL;
2672 char *committer = NULL;
2673 char *encoding = NULL;
2674 struct hash_list *merge_list = NULL;
2675 unsigned int merge_count;
2676 unsigned char prev_fanout, new_fanout;
2677 const char *v;
2679 b = lookup_branch(arg);
2680 if (!b)
2681 b = new_branch(arg);
2683 read_next_command();
2684 parse_mark();
2685 parse_original_identifier();
2686 if (skip_prefix(command_buf.buf, "author ", &v)) {
2687 author = parse_ident(v);
2688 read_next_command();
2690 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2691 committer = parse_ident(v);
2692 read_next_command();
2694 if (!committer)
2695 die("Expected committer but didn't get one");
2696 if (skip_prefix(command_buf.buf, "encoding ", &v)) {
2697 encoding = xstrdup(v);
2698 read_next_command();
2700 parse_data(&msg, 0, NULL);
2701 read_next_command();
2702 parse_from(b);
2703 merge_list = parse_merge(&merge_count);
2705 /* ensure the branch is active/loaded */
2706 if (!b->branch_tree.tree || !max_active_branches) {
2707 unload_one_branch();
2708 load_branch(b);
2711 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2713 /* file_change* */
2714 while (command_buf.len > 0) {
2715 if (skip_prefix(command_buf.buf, "M ", &v))
2716 file_change_m(v, b);
2717 else if (skip_prefix(command_buf.buf, "D ", &v))
2718 file_change_d(v, b);
2719 else if (skip_prefix(command_buf.buf, "R ", &v))
2720 file_change_cr(v, b, 1);
2721 else if (skip_prefix(command_buf.buf, "C ", &v))
2722 file_change_cr(v, b, 0);
2723 else if (skip_prefix(command_buf.buf, "N ", &v))
2724 note_change_n(v, b, &prev_fanout);
2725 else if (!strcmp("deleteall", command_buf.buf))
2726 file_change_deleteall(b);
2727 else if (skip_prefix(command_buf.buf, "ls ", &v))
2728 parse_ls(v, b);
2729 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2730 parse_cat_blob(v);
2731 else {
2732 unread_command_buf = 1;
2733 break;
2735 if (read_next_command() == EOF)
2736 break;
2739 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2740 if (new_fanout != prev_fanout)
2741 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2743 /* build the tree and the commit */
2744 store_tree(&b->branch_tree);
2745 oidcpy(&b->branch_tree.versions[0].oid,
2746 &b->branch_tree.versions[1].oid);
2748 strbuf_reset(&new_data);
2749 strbuf_addf(&new_data, "tree %s\n",
2750 oid_to_hex(&b->branch_tree.versions[1].oid));
2751 if (!is_null_oid(&b->oid))
2752 strbuf_addf(&new_data, "parent %s\n",
2753 oid_to_hex(&b->oid));
2754 while (merge_list) {
2755 struct hash_list *next = merge_list->next;
2756 strbuf_addf(&new_data, "parent %s\n",
2757 oid_to_hex(&merge_list->oid));
2758 free(merge_list);
2759 merge_list = next;
2761 strbuf_addf(&new_data,
2762 "author %s\n"
2763 "committer %s\n",
2764 author ? author : committer, committer);
2765 if (encoding)
2766 strbuf_addf(&new_data,
2767 "encoding %s\n",
2768 encoding);
2769 strbuf_addch(&new_data, '\n');
2770 strbuf_addbuf(&new_data, &msg);
2771 free(author);
2772 free(committer);
2773 free(encoding);
2775 if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2776 b->pack_id = pack_id;
2777 b->last_commit = object_count_by_type[OBJ_COMMIT];
2780 static void parse_new_tag(const char *arg)
2782 static struct strbuf msg = STRBUF_INIT;
2783 const char *from;
2784 char *tagger;
2785 struct branch *s;
2786 struct tag *t;
2787 uintmax_t from_mark = 0;
2788 struct object_id oid;
2789 enum object_type type;
2790 const char *v;
2792 t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2793 memset(t, 0, sizeof(struct tag));
2794 t->name = pool_strdup(arg);
2795 if (last_tag)
2796 last_tag->next_tag = t;
2797 else
2798 first_tag = t;
2799 last_tag = t;
2800 read_next_command();
2801 parse_mark();
2803 /* from ... */
2804 if (!skip_prefix(command_buf.buf, "from ", &from))
2805 die("Expected from command, got %s", command_buf.buf);
2806 s = lookup_branch(from);
2807 if (s) {
2808 if (is_null_oid(&s->oid))
2809 die("Can't tag an empty branch.");
2810 oidcpy(&oid, &s->oid);
2811 type = OBJ_COMMIT;
2812 } else if (*from == ':') {
2813 struct object_entry *oe;
2814 from_mark = parse_mark_ref_eol(from);
2815 oe = find_mark(marks, from_mark);
2816 type = oe->type;
2817 oidcpy(&oid, &oe->idx.oid);
2818 } else if (!get_oid(from, &oid)) {
2819 struct object_entry *oe = find_object(&oid);
2820 if (!oe) {
2821 type = oid_object_info(the_repository, &oid, NULL);
2822 if (type < 0)
2823 die("Not a valid object: %s", from);
2824 } else
2825 type = oe->type;
2826 } else
2827 die("Invalid ref name or SHA1 expression: %s", from);
2828 read_next_command();
2830 /* original-oid ... */
2831 parse_original_identifier();
2833 /* tagger ... */
2834 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2835 tagger = parse_ident(v);
2836 read_next_command();
2837 } else
2838 tagger = NULL;
2840 /* tag payload/message */
2841 parse_data(&msg, 0, NULL);
2843 /* build the tag object */
2844 strbuf_reset(&new_data);
2846 strbuf_addf(&new_data,
2847 "object %s\n"
2848 "type %s\n"
2849 "tag %s\n",
2850 oid_to_hex(&oid), type_name(type), t->name);
2851 if (tagger)
2852 strbuf_addf(&new_data,
2853 "tagger %s\n", tagger);
2854 strbuf_addch(&new_data, '\n');
2855 strbuf_addbuf(&new_data, &msg);
2856 free(tagger);
2858 if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, next_mark))
2859 t->pack_id = MAX_PACK_ID;
2860 else
2861 t->pack_id = pack_id;
2864 static void parse_reset_branch(const char *arg)
2866 struct branch *b;
2867 const char *tag_name;
2869 b = lookup_branch(arg);
2870 if (b) {
2871 oidclr(&b->oid);
2872 oidclr(&b->branch_tree.versions[0].oid);
2873 oidclr(&b->branch_tree.versions[1].oid);
2874 if (b->branch_tree.tree) {
2875 release_tree_content_recursive(b->branch_tree.tree);
2876 b->branch_tree.tree = NULL;
2879 else
2880 b = new_branch(arg);
2881 read_next_command();
2882 parse_from(b);
2883 if (b->delete && skip_prefix(b->name, "refs/tags/", &tag_name)) {
2885 * Elsewhere, we call dump_branches() before dump_tags(),
2886 * and dump_branches() will handle ref deletions first, so
2887 * in order to make sure the deletion actually takes effect,
2888 * we need to remove the tag from our list of tags to update.
2890 * NEEDSWORK: replace list of tags with hashmap for faster
2891 * deletion?
2893 struct tag *t, *prev = NULL;
2894 for (t = first_tag; t; t = t->next_tag) {
2895 if (!strcmp(t->name, tag_name))
2896 break;
2897 prev = t;
2899 if (t) {
2900 if (prev)
2901 prev->next_tag = t->next_tag;
2902 else
2903 first_tag = t->next_tag;
2904 if (!t->next_tag)
2905 last_tag = prev;
2906 /* There is no mem_pool_free(t) function to call. */
2909 if (command_buf.len > 0)
2910 unread_command_buf = 1;
2913 static void cat_blob_write(const char *buf, unsigned long size)
2915 if (write_in_full(cat_blob_fd, buf, size) < 0)
2916 die_errno("Write to frontend failed");
2919 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2921 struct strbuf line = STRBUF_INIT;
2922 unsigned long size;
2923 enum object_type type = 0;
2924 char *buf;
2926 if (!oe || oe->pack_id == MAX_PACK_ID) {
2927 buf = read_object_file(oid, &type, &size);
2928 } else {
2929 type = oe->type;
2930 buf = gfi_unpack_entry(oe, &size);
2934 * Output based on batch_one_object() from cat-file.c.
2936 if (type <= 0) {
2937 strbuf_reset(&line);
2938 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2939 cat_blob_write(line.buf, line.len);
2940 strbuf_release(&line);
2941 free(buf);
2942 return;
2944 if (!buf)
2945 die("Can't read object %s", oid_to_hex(oid));
2946 if (type != OBJ_BLOB)
2947 die("Object %s is a %s but a blob was expected.",
2948 oid_to_hex(oid), type_name(type));
2949 strbuf_reset(&line);
2950 strbuf_addf(&line, "%s %s %"PRIuMAX"\n", oid_to_hex(oid),
2951 type_name(type), (uintmax_t)size);
2952 cat_blob_write(line.buf, line.len);
2953 strbuf_release(&line);
2954 cat_blob_write(buf, size);
2955 cat_blob_write("\n", 1);
2956 if (oe && oe->pack_id == pack_id) {
2957 last_blob.offset = oe->idx.offset;
2958 strbuf_attach(&last_blob.data, buf, size, size);
2959 last_blob.depth = oe->depth;
2960 } else
2961 free(buf);
2964 static void parse_get_mark(const char *p)
2966 struct object_entry *oe;
2967 char output[GIT_MAX_HEXSZ + 2];
2969 /* get-mark SP <object> LF */
2970 if (*p != ':')
2971 die("Not a mark: %s", p);
2973 oe = find_mark(marks, parse_mark_ref_eol(p));
2974 if (!oe)
2975 die("Unknown mark: %s", command_buf.buf);
2977 xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2978 cat_blob_write(output, the_hash_algo->hexsz + 1);
2981 static void parse_cat_blob(const char *p)
2983 struct object_entry *oe;
2984 struct object_id oid;
2986 /* cat-blob SP <object> LF */
2987 if (*p == ':') {
2988 oe = find_mark(marks, parse_mark_ref_eol(p));
2989 if (!oe)
2990 die("Unknown mark: %s", command_buf.buf);
2991 oidcpy(&oid, &oe->idx.oid);
2992 } else {
2993 if (parse_mapped_oid_hex(p, &oid, &p))
2994 die("Invalid dataref: %s", command_buf.buf);
2995 if (*p)
2996 die("Garbage after SHA1: %s", command_buf.buf);
2997 oe = find_object(&oid);
3000 cat_blob(oe, &oid);
3003 static struct object_entry *dereference(struct object_entry *oe,
3004 struct object_id *oid)
3006 unsigned long size;
3007 char *buf = NULL;
3008 const unsigned hexsz = the_hash_algo->hexsz;
3010 if (!oe) {
3011 enum object_type type = oid_object_info(the_repository, oid,
3012 NULL);
3013 if (type < 0)
3014 die("object not found: %s", oid_to_hex(oid));
3015 /* cache it! */
3016 oe = insert_object(oid);
3017 oe->type = type;
3018 oe->pack_id = MAX_PACK_ID;
3019 oe->idx.offset = 1;
3021 switch (oe->type) {
3022 case OBJ_TREE: /* easy case. */
3023 return oe;
3024 case OBJ_COMMIT:
3025 case OBJ_TAG:
3026 break;
3027 default:
3028 die("Not a tree-ish: %s", command_buf.buf);
3031 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3032 buf = gfi_unpack_entry(oe, &size);
3033 } else {
3034 enum object_type unused;
3035 buf = read_object_file(oid, &unused, &size);
3037 if (!buf)
3038 die("Can't load object %s", oid_to_hex(oid));
3040 /* Peel one layer. */
3041 switch (oe->type) {
3042 case OBJ_TAG:
3043 if (size < hexsz + strlen("object ") ||
3044 get_oid_hex(buf + strlen("object "), oid))
3045 die("Invalid SHA1 in tag: %s", command_buf.buf);
3046 break;
3047 case OBJ_COMMIT:
3048 if (size < hexsz + strlen("tree ") ||
3049 get_oid_hex(buf + strlen("tree "), oid))
3050 die("Invalid SHA1 in commit: %s", command_buf.buf);
3053 free(buf);
3054 return find_object(oid);
3057 static void insert_mapped_mark(uintmax_t mark, void *object, void *cbp)
3059 struct object_id *fromoid = object;
3060 struct object_id *tooid = find_mark(cbp, mark);
3061 int ret;
3062 khiter_t it;
3064 it = kh_put_oid_map(sub_oid_map, *fromoid, &ret);
3065 /* We've already seen this object. */
3066 if (ret == 0)
3067 return;
3068 kh_value(sub_oid_map, it) = tooid;
3071 static void build_mark_map_one(struct mark_set *from, struct mark_set *to)
3073 for_each_mark(from, 0, insert_mapped_mark, to);
3076 static void build_mark_map(struct string_list *from, struct string_list *to)
3078 struct string_list_item *fromp, *top;
3080 sub_oid_map = kh_init_oid_map();
3082 for_each_string_list_item(fromp, from) {
3083 top = string_list_lookup(to, fromp->string);
3084 if (!fromp->util) {
3085 die(_("Missing from marks for submodule '%s'"), fromp->string);
3086 } else if (!top || !top->util) {
3087 die(_("Missing to marks for submodule '%s'"), fromp->string);
3089 build_mark_map_one(fromp->util, top->util);
3093 static struct object_entry *parse_treeish_dataref(const char **p)
3095 struct object_id oid;
3096 struct object_entry *e;
3098 if (**p == ':') { /* <mark> */
3099 e = find_mark(marks, parse_mark_ref_space(p));
3100 if (!e)
3101 die("Unknown mark: %s", command_buf.buf);
3102 oidcpy(&oid, &e->idx.oid);
3103 } else { /* <sha1> */
3104 if (parse_mapped_oid_hex(*p, &oid, p))
3105 die("Invalid dataref: %s", command_buf.buf);
3106 e = find_object(&oid);
3107 if (*(*p)++ != ' ')
3108 die("Missing space after tree-ish: %s", command_buf.buf);
3111 while (!e || e->type != OBJ_TREE)
3112 e = dereference(e, &oid);
3113 return e;
3116 static void print_ls(int mode, const unsigned char *hash, const char *path)
3118 static struct strbuf line = STRBUF_INIT;
3120 /* See show_tree(). */
3121 const char *type =
3122 S_ISGITLINK(mode) ? commit_type :
3123 S_ISDIR(mode) ? tree_type :
3124 blob_type;
3126 if (!mode) {
3127 /* missing SP path LF */
3128 strbuf_reset(&line);
3129 strbuf_addstr(&line, "missing ");
3130 quote_c_style(path, &line, NULL, 0);
3131 strbuf_addch(&line, '\n');
3132 } else {
3133 /* mode SP type SP object_name TAB path LF */
3134 strbuf_reset(&line);
3135 strbuf_addf(&line, "%06o %s %s\t",
3136 mode & ~NO_DELTA, type, hash_to_hex(hash));
3137 quote_c_style(path, &line, NULL, 0);
3138 strbuf_addch(&line, '\n');
3140 cat_blob_write(line.buf, line.len);
3143 static void parse_ls(const char *p, struct branch *b)
3145 struct tree_entry *root = NULL;
3146 struct tree_entry leaf = {NULL};
3148 /* ls SP (<tree-ish> SP)? <path> */
3149 if (*p == '"') {
3150 if (!b)
3151 die("Not in a commit: %s", command_buf.buf);
3152 root = &b->branch_tree;
3153 } else {
3154 struct object_entry *e = parse_treeish_dataref(&p);
3155 root = new_tree_entry();
3156 oidcpy(&root->versions[1].oid, &e->idx.oid);
3157 if (!is_null_oid(&root->versions[1].oid))
3158 root->versions[1].mode = S_IFDIR;
3159 load_tree(root);
3161 if (*p == '"') {
3162 static struct strbuf uq = STRBUF_INIT;
3163 const char *endp;
3164 strbuf_reset(&uq);
3165 if (unquote_c_style(&uq, p, &endp))
3166 die("Invalid path: %s", command_buf.buf);
3167 if (*endp)
3168 die("Garbage after path in: %s", command_buf.buf);
3169 p = uq.buf;
3171 tree_content_get(root, p, &leaf, 1);
3173 * A directory in preparation would have a sha1 of zero
3174 * until it is saved. Save, for simplicity.
3176 if (S_ISDIR(leaf.versions[1].mode))
3177 store_tree(&leaf);
3179 print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3180 if (leaf.tree)
3181 release_tree_content_recursive(leaf.tree);
3182 if (!b || root != &b->branch_tree)
3183 release_tree_entry(root);
3186 static void checkpoint(void)
3188 checkpoint_requested = 0;
3189 if (object_count) {
3190 cycle_packfile();
3192 dump_branches();
3193 dump_tags();
3194 dump_marks();
3197 static void parse_checkpoint(void)
3199 checkpoint_requested = 1;
3200 skip_optional_lf();
3203 static void parse_progress(void)
3205 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3206 fputc('\n', stdout);
3207 fflush(stdout);
3208 skip_optional_lf();
3211 static void parse_alias(void)
3213 struct object_entry *e;
3214 struct branch b;
3216 skip_optional_lf();
3217 read_next_command();
3219 /* mark ... */
3220 parse_mark();
3221 if (!next_mark)
3222 die(_("Expected 'mark' command, got %s"), command_buf.buf);
3224 /* to ... */
3225 memset(&b, 0, sizeof(b));
3226 if (!parse_objectish_with_prefix(&b, "to "))
3227 die(_("Expected 'to' command, got %s"), command_buf.buf);
3228 e = find_object(&b.oid);
3229 assert(e);
3230 insert_mark(marks, next_mark, e);
3233 static char* make_fast_import_path(const char *path)
3235 if (!relative_marks_paths || is_absolute_path(path))
3236 return xstrdup(path);
3237 return git_pathdup("info/fast-import/%s", path);
3240 static void option_import_marks(const char *marks,
3241 int from_stream, int ignore_missing)
3243 if (import_marks_file) {
3244 if (from_stream)
3245 die("Only one import-marks command allowed per stream");
3247 /* read previous mark file */
3248 if(!import_marks_file_from_stream)
3249 read_marks();
3252 import_marks_file = make_fast_import_path(marks);
3253 import_marks_file_from_stream = from_stream;
3254 import_marks_file_ignore_missing = ignore_missing;
3257 static void option_date_format(const char *fmt)
3259 if (!strcmp(fmt, "raw"))
3260 whenspec = WHENSPEC_RAW;
3261 else if (!strcmp(fmt, "rfc2822"))
3262 whenspec = WHENSPEC_RFC2822;
3263 else if (!strcmp(fmt, "now"))
3264 whenspec = WHENSPEC_NOW;
3265 else
3266 die("unknown --date-format argument %s", fmt);
3269 static unsigned long ulong_arg(const char *option, const char *arg)
3271 char *endptr;
3272 unsigned long rv = strtoul(arg, &endptr, 0);
3273 if (strchr(arg, '-') || endptr == arg || *endptr)
3274 die("%s: argument must be a non-negative integer", option);
3275 return rv;
3278 static void option_depth(const char *depth)
3280 max_depth = ulong_arg("--depth", depth);
3281 if (max_depth > MAX_DEPTH)
3282 die("--depth cannot exceed %u", MAX_DEPTH);
3285 static void option_active_branches(const char *branches)
3287 max_active_branches = ulong_arg("--active-branches", branches);
3290 static void option_export_marks(const char *marks)
3292 export_marks_file = make_fast_import_path(marks);
3295 static void option_cat_blob_fd(const char *fd)
3297 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3298 if (n > (unsigned long) INT_MAX)
3299 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3300 cat_blob_fd = (int) n;
3303 static void option_export_pack_edges(const char *edges)
3305 if (pack_edges)
3306 fclose(pack_edges);
3307 pack_edges = xfopen(edges, "a");
3310 static void option_rewrite_submodules(const char *arg, struct string_list *list)
3312 struct mark_set *ms;
3313 FILE *fp;
3314 char *s = xstrdup(arg);
3315 char *f = strchr(s, ':');
3316 if (!f)
3317 die(_("Expected format name:filename for submodule rewrite option"));
3318 *f = '\0';
3319 f++;
3320 ms = xcalloc(1, sizeof(*ms));
3321 string_list_insert(list, s)->util = ms;
3323 fp = fopen(f, "r");
3324 if (!fp)
3325 die_errno("cannot read '%s'", f);
3326 read_mark_file(ms, fp, insert_oid_entry);
3327 fclose(fp);
3330 static int parse_one_option(const char *option)
3332 if (skip_prefix(option, "max-pack-size=", &option)) {
3333 unsigned long v;
3334 if (!git_parse_ulong(option, &v))
3335 return 0;
3336 if (v < 8192) {
3337 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3338 v *= 1024 * 1024;
3339 } else if (v < 1024 * 1024) {
3340 warning("minimum max-pack-size is 1 MiB");
3341 v = 1024 * 1024;
3343 max_packsize = v;
3344 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3345 unsigned long v;
3346 if (!git_parse_ulong(option, &v))
3347 return 0;
3348 big_file_threshold = v;
3349 } else if (skip_prefix(option, "depth=", &option)) {
3350 option_depth(option);
3351 } else if (skip_prefix(option, "active-branches=", &option)) {
3352 option_active_branches(option);
3353 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3354 option_export_pack_edges(option);
3355 } else if (!strcmp(option, "quiet")) {
3356 show_stats = 0;
3357 } else if (!strcmp(option, "stats")) {
3358 show_stats = 1;
3359 } else if (!strcmp(option, "allow-unsafe-features")) {
3360 ; /* already handled during early option parsing */
3361 } else {
3362 return 0;
3365 return 1;
3368 static void check_unsafe_feature(const char *feature, int from_stream)
3370 if (from_stream && !allow_unsafe_features)
3371 die(_("feature '%s' forbidden in input without --allow-unsafe-features"),
3372 feature);
3375 static int parse_one_feature(const char *feature, int from_stream)
3377 const char *arg;
3379 if (skip_prefix(feature, "date-format=", &arg)) {
3380 option_date_format(arg);
3381 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3382 check_unsafe_feature("import-marks", from_stream);
3383 option_import_marks(arg, from_stream, 0);
3384 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3385 check_unsafe_feature("import-marks-if-exists", from_stream);
3386 option_import_marks(arg, from_stream, 1);
3387 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3388 check_unsafe_feature(feature, from_stream);
3389 option_export_marks(arg);
3390 } else if (!strcmp(feature, "alias")) {
3391 ; /* Don't die - this feature is supported */
3392 } else if (skip_prefix(feature, "rewrite-submodules-to=", &arg)) {
3393 option_rewrite_submodules(arg, &sub_marks_to);
3394 } else if (skip_prefix(feature, "rewrite-submodules-from=", &arg)) {
3395 option_rewrite_submodules(arg, &sub_marks_from);
3396 } else if (skip_prefix(feature, "rewrite-submodules-from=", &arg)) {
3397 } else if (!strcmp(feature, "get-mark")) {
3398 ; /* Don't die - this feature is supported */
3399 } else if (!strcmp(feature, "cat-blob")) {
3400 ; /* Don't die - this feature is supported */
3401 } else if (!strcmp(feature, "relative-marks")) {
3402 relative_marks_paths = 1;
3403 } else if (!strcmp(feature, "no-relative-marks")) {
3404 relative_marks_paths = 0;
3405 } else if (!strcmp(feature, "done")) {
3406 require_explicit_termination = 1;
3407 } else if (!strcmp(feature, "force")) {
3408 force_update = 1;
3409 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3410 ; /* do nothing; we have the feature */
3411 } else {
3412 return 0;
3415 return 1;
3418 static void parse_feature(const char *feature)
3420 if (seen_data_command)
3421 die("Got feature command '%s' after data command", feature);
3423 if (parse_one_feature(feature, 1))
3424 return;
3426 die("This version of fast-import does not support feature %s.", feature);
3429 static void parse_option(const char *option)
3431 if (seen_data_command)
3432 die("Got option command '%s' after data command", option);
3434 if (parse_one_option(option))
3435 return;
3437 die("This version of fast-import does not support option: %s", option);
3440 static void git_pack_config(void)
3442 int indexversion_value;
3443 int limit;
3444 unsigned long packsizelimit_value;
3446 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3447 if (max_depth > MAX_DEPTH)
3448 max_depth = MAX_DEPTH;
3450 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3451 pack_idx_opts.version = indexversion_value;
3452 if (pack_idx_opts.version > 2)
3453 git_die_config("pack.indexversion",
3454 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3456 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3457 max_packsize = packsizelimit_value;
3459 if (!git_config_get_int("fastimport.unpacklimit", &limit))
3460 unpack_limit = limit;
3461 else if (!git_config_get_int("transfer.unpacklimit", &limit))
3462 unpack_limit = limit;
3464 git_config(git_default_config, NULL);
3467 static const char fast_import_usage[] =
3468 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3470 static void parse_argv(void)
3472 unsigned int i;
3474 for (i = 1; i < global_argc; i++) {
3475 const char *a = global_argv[i];
3477 if (*a != '-' || !strcmp(a, "--"))
3478 break;
3480 if (!skip_prefix(a, "--", &a))
3481 die("unknown option %s", a);
3483 if (parse_one_option(a))
3484 continue;
3486 if (parse_one_feature(a, 0))
3487 continue;
3489 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3490 option_cat_blob_fd(a);
3491 continue;
3494 die("unknown option --%s", a);
3496 if (i != global_argc)
3497 usage(fast_import_usage);
3499 seen_data_command = 1;
3500 if (import_marks_file)
3501 read_marks();
3502 build_mark_map(&sub_marks_from, &sub_marks_to);
3505 int cmd_main(int argc, const char **argv)
3507 unsigned int i;
3509 if (argc == 2 && !strcmp(argv[1], "-h"))
3510 usage(fast_import_usage);
3512 setup_git_directory();
3513 reset_pack_idx_option(&pack_idx_opts);
3514 git_pack_config();
3516 alloc_objects(object_entry_alloc);
3517 strbuf_init(&command_buf, 0);
3518 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3519 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3520 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3521 marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3523 hashmap_init(&object_table, object_entry_hashcmp, NULL, 0);
3526 * We don't parse most options until after we've seen the set of
3527 * "feature" lines at the start of the stream (which allows the command
3528 * line to override stream data). But we must do an early parse of any
3529 * command-line options that impact how we interpret the feature lines.
3531 for (i = 1; i < argc; i++) {
3532 const char *arg = argv[i];
3533 if (*arg != '-' || !strcmp(arg, "--"))
3534 break;
3535 if (!strcmp(arg, "--allow-unsafe-features"))
3536 allow_unsafe_features = 1;
3539 global_argc = argc;
3540 global_argv = argv;
3542 rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3543 for (i = 0; i < (cmd_save - 1); i++)
3544 rc_free[i].next = &rc_free[i + 1];
3545 rc_free[cmd_save - 1].next = NULL;
3547 start_packfile();
3548 set_die_routine(die_nicely);
3549 set_checkpoint_signal();
3550 while (read_next_command() != EOF) {
3551 const char *v;
3552 if (!strcmp("blob", command_buf.buf))
3553 parse_new_blob();
3554 else if (skip_prefix(command_buf.buf, "commit ", &v))
3555 parse_new_commit(v);
3556 else if (skip_prefix(command_buf.buf, "tag ", &v))
3557 parse_new_tag(v);
3558 else if (skip_prefix(command_buf.buf, "reset ", &v))
3559 parse_reset_branch(v);
3560 else if (skip_prefix(command_buf.buf, "ls ", &v))
3561 parse_ls(v, NULL);
3562 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
3563 parse_cat_blob(v);
3564 else if (skip_prefix(command_buf.buf, "get-mark ", &v))
3565 parse_get_mark(v);
3566 else if (!strcmp("checkpoint", command_buf.buf))
3567 parse_checkpoint();
3568 else if (!strcmp("done", command_buf.buf))
3569 break;
3570 else if (!strcmp("alias", command_buf.buf))
3571 parse_alias();
3572 else if (starts_with(command_buf.buf, "progress "))
3573 parse_progress();
3574 else if (skip_prefix(command_buf.buf, "feature ", &v))
3575 parse_feature(v);
3576 else if (skip_prefix(command_buf.buf, "option git ", &v))
3577 parse_option(v);
3578 else if (starts_with(command_buf.buf, "option "))
3579 /* ignore non-git options*/;
3580 else
3581 die("Unsupported command: %s", command_buf.buf);
3583 if (checkpoint_requested)
3584 checkpoint();
3587 /* argv hasn't been parsed yet, do so */
3588 if (!seen_data_command)
3589 parse_argv();
3591 if (require_explicit_termination && feof(stdin))
3592 die("stream ends early");
3594 end_packfile();
3596 dump_branches();
3597 dump_tags();
3598 unkeep_all_packs();
3599 dump_marks();
3601 if (pack_edges)
3602 fclose(pack_edges);
3604 if (show_stats) {
3605 uintmax_t total_count = 0, duplicate_count = 0;
3606 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3607 total_count += object_count_by_type[i];
3608 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3609 duplicate_count += duplicate_count_by_type[i];
3611 fprintf(stderr, "%s statistics:\n", argv[0]);
3612 fprintf(stderr, "---------------------------------------------------------------------\n");
3613 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3614 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3615 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]);
3616 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]);
3617 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]);
3618 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]);
3619 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3620 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3621 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3622 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3623 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3624 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3625 fprintf(stderr, "---------------------------------------------------------------------\n");
3626 pack_report();
3627 fprintf(stderr, "---------------------------------------------------------------------\n");
3628 fprintf(stderr, "\n");
3631 return failure ? 1 : 0;