commit: fix memory leak in `reduce_heads()`
[git.git] / fast-import.c
bloba959161b4631e134602aa9b5502105ee5c5bcfb8
1 /*
2 (See Documentation/git-fast-import.txt for maintained documentation.)
3 Format of STDIN stream:
5 stream ::= cmd*;
7 cmd ::= new_blob
8 | new_commit
9 | new_tag
10 | reset_branch
11 | checkpoint
12 | progress
15 new_blob ::= 'blob' lf
16 mark?
17 file_content;
18 file_content ::= data;
20 new_commit ::= 'commit' sp ref_str lf
21 mark?
22 ('author' (sp name)? sp '<' email '>' sp when lf)?
23 'committer' (sp name)? sp '<' email '>' sp when lf
24 commit_msg
25 ('from' sp commit-ish lf)?
26 ('merge' sp commit-ish lf)*
27 (file_change | ls)*
28 lf?;
29 commit_msg ::= data;
31 ls ::= 'ls' sp '"' quoted(path) '"' lf;
33 file_change ::= file_clr
34 | file_del
35 | file_rnm
36 | file_cpy
37 | file_obm
38 | file_inm;
39 file_clr ::= 'deleteall' lf;
40 file_del ::= 'D' sp path_str lf;
41 file_rnm ::= 'R' sp path_str sp path_str lf;
42 file_cpy ::= 'C' sp path_str sp path_str lf;
43 file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
44 file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
45 data;
46 note_obm ::= 'N' sp (hexsha1 | idnum) sp commit-ish lf;
47 note_inm ::= 'N' sp 'inline' sp commit-ish lf
48 data;
50 new_tag ::= 'tag' sp tag_str lf
51 'from' sp commit-ish lf
52 ('tagger' (sp name)? sp '<' email '>' sp when lf)?
53 tag_msg;
54 tag_msg ::= data;
56 reset_branch ::= 'reset' sp ref_str lf
57 ('from' sp commit-ish lf)?
58 lf?;
60 checkpoint ::= 'checkpoint' lf
61 lf?;
63 progress ::= 'progress' sp not_lf* lf
64 lf?;
66 # note: the first idnum in a stream should be 1 and subsequent
67 # idnums should not have gaps between values as this will cause
68 # the stream parser to reserve space for the gapped values. An
69 # idnum can be updated in the future to a new object by issuing
70 # a new mark directive with the old idnum.
72 mark ::= 'mark' sp idnum lf;
73 data ::= (delimited_data | exact_data)
74 lf?;
76 # note: delim may be any string but must not contain lf.
77 # data_line may contain any data but must not be exactly
78 # delim.
79 delimited_data ::= 'data' sp '<<' delim lf
80 (data_line lf)*
81 delim lf;
83 # note: declen indicates the length of binary_data in bytes.
84 # declen does not include the lf preceding the binary data.
86 exact_data ::= 'data' sp declen lf
87 binary_data;
89 # note: quoted strings are C-style quoting supporting \c for
90 # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
91 # is the signed byte value in octal. Note that the only
92 # characters which must actually be escaped to protect the
93 # stream formatting is: \, " and LF. Otherwise these values
94 # are UTF8.
96 commit-ish ::= (ref_str | hexsha1 | sha1exp_str | idnum);
97 ref_str ::= ref;
98 sha1exp_str ::= sha1exp;
99 tag_str ::= tag;
100 path_str ::= path | '"' quoted(path) '"' ;
101 mode ::= '100644' | '644'
102 | '100755' | '755'
103 | '120000'
106 declen ::= # unsigned 32 bit value, ascii base10 notation;
107 bigint ::= # unsigned integer value, ascii base10 notation;
108 binary_data ::= # file content, not interpreted;
110 when ::= raw_when | rfc2822_when;
111 raw_when ::= ts sp tz;
112 rfc2822_when ::= # Valid RFC 2822 date and time;
114 sp ::= # ASCII space character;
115 lf ::= # ASCII newline (LF) character;
117 # note: a colon (':') must precede the numerical value assigned to
118 # an idnum. This is to distinguish it from a ref or tag name as
119 # GIT does not permit ':' in ref or tag strings.
121 idnum ::= ':' bigint;
122 path ::= # GIT style file path, e.g. "a/b/c";
123 ref ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
124 tag ::= # GIT tag name, e.g. "FIREFOX_1_5";
125 sha1exp ::= # Any valid GIT SHA1 expression;
126 hexsha1 ::= # SHA1 in hexadecimal format;
128 # note: name and email are UTF8 strings, however name must not
129 # contain '<' or lf and email must not contain any of the
130 # following: '<', '>', lf.
132 name ::= # valid GIT author/committer name;
133 email ::= # valid GIT author/committer email;
134 ts ::= # time since the epoch in seconds, ascii base10 notation;
135 tz ::= # GIT style timezone;
137 # note: comments, get-mark, ls-tree, and cat-blob requests may
138 # appear anywhere in the input, except within a data command. Any
139 # form of the data command always escapes the related input from
140 # comment processing.
142 # In case it is not clear, the '#' that starts the comment
143 # must be the first character on that line (an lf
144 # preceded it).
147 get_mark ::= 'get-mark' sp idnum lf;
148 cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
149 ls_tree ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
151 comment ::= '#' not_lf* lf;
152 not_lf ::= # Any byte that is not ASCII newline (LF);
155 #include "builtin.h"
156 #include "cache.h"
157 #include "config.h"
158 #include "lockfile.h"
159 #include "object.h"
160 #include "blob.h"
161 #include "tree.h"
162 #include "commit.h"
163 #include "delta.h"
164 #include "pack.h"
165 #include "refs.h"
166 #include "csum-file.h"
167 #include "quote.h"
168 #include "dir.h"
169 #include "run-command.h"
171 #define PACK_ID_BITS 16
172 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
173 #define DEPTH_BITS 13
174 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
177 * We abuse the setuid bit on directories to mean "do not delta".
179 #define NO_DELTA S_ISUID
181 struct object_entry {
182 struct pack_idx_entry idx;
183 struct object_entry *next;
184 uint32_t type : TYPE_BITS,
185 pack_id : PACK_ID_BITS,
186 depth : DEPTH_BITS;
189 struct object_entry_pool {
190 struct object_entry_pool *next_pool;
191 struct object_entry *next_free;
192 struct object_entry *end;
193 struct object_entry entries[FLEX_ARRAY]; /* more */
196 struct mark_set {
197 union {
198 struct object_entry *marked[1024];
199 struct mark_set *sets[1024];
200 } data;
201 unsigned int shift;
204 struct last_object {
205 struct strbuf data;
206 off_t offset;
207 unsigned int depth;
208 unsigned no_swap : 1;
211 struct mem_pool {
212 struct mem_pool *next_pool;
213 char *next_free;
214 char *end;
215 uintmax_t space[FLEX_ARRAY]; /* more */
218 struct atom_str {
219 struct atom_str *next_atom;
220 unsigned short str_len;
221 char str_dat[FLEX_ARRAY]; /* more */
224 struct tree_content;
225 struct tree_entry {
226 struct tree_content *tree;
227 struct atom_str *name;
228 struct tree_entry_ms {
229 uint16_t mode;
230 struct object_id oid;
231 } versions[2];
234 struct tree_content {
235 unsigned int entry_capacity; /* must match avail_tree_content */
236 unsigned int entry_count;
237 unsigned int delta_depth;
238 struct tree_entry *entries[FLEX_ARRAY]; /* more */
241 struct avail_tree_content {
242 unsigned int entry_capacity; /* must match tree_content */
243 struct avail_tree_content *next_avail;
246 struct branch {
247 struct branch *table_next_branch;
248 struct branch *active_next_branch;
249 const char *name;
250 struct tree_entry branch_tree;
251 uintmax_t last_commit;
252 uintmax_t num_notes;
253 unsigned active : 1;
254 unsigned delete : 1;
255 unsigned pack_id : PACK_ID_BITS;
256 struct object_id oid;
259 struct tag {
260 struct tag *next_tag;
261 const char *name;
262 unsigned int pack_id;
263 struct object_id oid;
266 struct hash_list {
267 struct hash_list *next;
268 struct object_id oid;
271 typedef enum {
272 WHENSPEC_RAW = 1,
273 WHENSPEC_RFC2822,
274 WHENSPEC_NOW
275 } whenspec_type;
277 struct recent_command {
278 struct recent_command *prev;
279 struct recent_command *next;
280 char *buf;
283 /* Configured limits on output */
284 static unsigned long max_depth = 50;
285 static off_t max_packsize;
286 static int unpack_limit = 100;
287 static int force_update;
289 /* Stats and misc. counters */
290 static uintmax_t alloc_count;
291 static uintmax_t marks_set_count;
292 static uintmax_t object_count_by_type[1 << TYPE_BITS];
293 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
294 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
295 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
296 static unsigned long object_count;
297 static unsigned long branch_count;
298 static unsigned long branch_load_count;
299 static int failure;
300 static FILE *pack_edges;
301 static unsigned int show_stats = 1;
302 static int global_argc;
303 static const char **global_argv;
305 /* Memory pools */
306 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
307 static size_t total_allocd;
308 static struct mem_pool *mem_pool;
310 /* Atom management */
311 static unsigned int atom_table_sz = 4451;
312 static unsigned int atom_cnt;
313 static struct atom_str **atom_table;
315 /* The .pack file being generated */
316 static struct pack_idx_option pack_idx_opts;
317 static unsigned int pack_id;
318 static struct sha1file *pack_file;
319 static struct packed_git *pack_data;
320 static struct packed_git **all_packs;
321 static off_t pack_size;
323 /* Table of objects we've written. */
324 static unsigned int object_entry_alloc = 5000;
325 static struct object_entry_pool *blocks;
326 static struct object_entry *object_table[1 << 16];
327 static struct mark_set *marks;
328 static const char *export_marks_file;
329 static const char *import_marks_file;
330 static int import_marks_file_from_stream;
331 static int import_marks_file_ignore_missing;
332 static int import_marks_file_done;
333 static int relative_marks_paths;
335 /* Our last blob */
336 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
338 /* Tree management */
339 static unsigned int tree_entry_alloc = 1000;
340 static void *avail_tree_entry;
341 static unsigned int avail_tree_table_sz = 100;
342 static struct avail_tree_content **avail_tree_table;
343 static struct strbuf old_tree = STRBUF_INIT;
344 static struct strbuf new_tree = STRBUF_INIT;
346 /* Branch data */
347 static unsigned long max_active_branches = 5;
348 static unsigned long cur_active_branches;
349 static unsigned long branch_table_sz = 1039;
350 static struct branch **branch_table;
351 static struct branch *active_branches;
353 /* Tag data */
354 static struct tag *first_tag;
355 static struct tag *last_tag;
357 /* Input stream parsing */
358 static whenspec_type whenspec = WHENSPEC_RAW;
359 static struct strbuf command_buf = STRBUF_INIT;
360 static int unread_command_buf;
361 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
362 static struct recent_command *cmd_tail = &cmd_hist;
363 static struct recent_command *rc_free;
364 static unsigned int cmd_save = 100;
365 static uintmax_t next_mark;
366 static struct strbuf new_data = STRBUF_INIT;
367 static int seen_data_command;
368 static int require_explicit_termination;
370 /* Signal handling */
371 static volatile sig_atomic_t checkpoint_requested;
373 /* Where to write output of cat-blob commands */
374 static int cat_blob_fd = STDOUT_FILENO;
376 static void parse_argv(void);
377 static void parse_get_mark(const char *p);
378 static void parse_cat_blob(const char *p);
379 static void parse_ls(const char *p, struct branch *b);
381 static void write_branch_report(FILE *rpt, struct branch *b)
383 fprintf(rpt, "%s:\n", b->name);
385 fprintf(rpt, " status :");
386 if (b->active)
387 fputs(" active", rpt);
388 if (b->branch_tree.tree)
389 fputs(" loaded", rpt);
390 if (is_null_oid(&b->branch_tree.versions[1].oid))
391 fputs(" dirty", rpt);
392 fputc('\n', rpt);
394 fprintf(rpt, " tip commit : %s\n", oid_to_hex(&b->oid));
395 fprintf(rpt, " old tree : %s\n",
396 oid_to_hex(&b->branch_tree.versions[0].oid));
397 fprintf(rpt, " cur tree : %s\n",
398 oid_to_hex(&b->branch_tree.versions[1].oid));
399 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
401 fputs(" last pack : ", rpt);
402 if (b->pack_id < MAX_PACK_ID)
403 fprintf(rpt, "%u", b->pack_id);
404 fputc('\n', rpt);
406 fputc('\n', rpt);
409 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
411 static void write_crash_report(const char *err)
413 char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
414 FILE *rpt = fopen(loc, "w");
415 struct branch *b;
416 unsigned long lu;
417 struct recent_command *rc;
419 if (!rpt) {
420 error_errno("can't write crash report %s", loc);
421 free(loc);
422 return;
425 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
427 fprintf(rpt, "fast-import crash report:\n");
428 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
429 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
430 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
431 fputc('\n', rpt);
433 fputs("fatal: ", rpt);
434 fputs(err, rpt);
435 fputc('\n', rpt);
437 fputc('\n', rpt);
438 fputs("Most Recent Commands Before Crash\n", rpt);
439 fputs("---------------------------------\n", rpt);
440 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
441 if (rc->next == &cmd_hist)
442 fputs("* ", rpt);
443 else
444 fputs(" ", rpt);
445 fputs(rc->buf, rpt);
446 fputc('\n', rpt);
449 fputc('\n', rpt);
450 fputs("Active Branch LRU\n", rpt);
451 fputs("-----------------\n", rpt);
452 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
453 cur_active_branches,
454 max_active_branches);
455 fputc('\n', rpt);
456 fputs(" pos clock name\n", rpt);
457 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
458 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
459 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
460 ++lu, b->last_commit, b->name);
462 fputc('\n', rpt);
463 fputs("Inactive Branches\n", rpt);
464 fputs("-----------------\n", rpt);
465 for (lu = 0; lu < branch_table_sz; lu++) {
466 for (b = branch_table[lu]; b; b = b->table_next_branch)
467 write_branch_report(rpt, b);
470 if (first_tag) {
471 struct tag *tg;
472 fputc('\n', rpt);
473 fputs("Annotated Tags\n", rpt);
474 fputs("--------------\n", rpt);
475 for (tg = first_tag; tg; tg = tg->next_tag) {
476 fputs(oid_to_hex(&tg->oid), rpt);
477 fputc(' ', rpt);
478 fputs(tg->name, rpt);
479 fputc('\n', rpt);
483 fputc('\n', rpt);
484 fputs("Marks\n", rpt);
485 fputs("-----\n", rpt);
486 if (export_marks_file)
487 fprintf(rpt, " exported to %s\n", export_marks_file);
488 else
489 dump_marks_helper(rpt, 0, marks);
491 fputc('\n', rpt);
492 fputs("-------------------\n", rpt);
493 fputs("END OF CRASH REPORT\n", rpt);
494 fclose(rpt);
495 free(loc);
498 static void end_packfile(void);
499 static void unkeep_all_packs(void);
500 static void dump_marks(void);
502 static NORETURN void die_nicely(const char *err, va_list params)
504 static int zombie;
505 char message[2 * PATH_MAX];
507 vsnprintf(message, sizeof(message), err, params);
508 fputs("fatal: ", stderr);
509 fputs(message, stderr);
510 fputc('\n', stderr);
512 if (!zombie) {
513 zombie = 1;
514 write_crash_report(message);
515 end_packfile();
516 unkeep_all_packs();
517 dump_marks();
519 exit(128);
522 #ifndef SIGUSR1 /* Windows, for example */
524 static void set_checkpoint_signal(void)
528 #else
530 static void checkpoint_signal(int signo)
532 checkpoint_requested = 1;
535 static void set_checkpoint_signal(void)
537 struct sigaction sa;
539 memset(&sa, 0, sizeof(sa));
540 sa.sa_handler = checkpoint_signal;
541 sigemptyset(&sa.sa_mask);
542 sa.sa_flags = SA_RESTART;
543 sigaction(SIGUSR1, &sa, NULL);
546 #endif
548 static void alloc_objects(unsigned int cnt)
550 struct object_entry_pool *b;
552 b = xmalloc(sizeof(struct object_entry_pool)
553 + cnt * sizeof(struct object_entry));
554 b->next_pool = blocks;
555 b->next_free = b->entries;
556 b->end = b->entries + cnt;
557 blocks = b;
558 alloc_count += cnt;
561 static struct object_entry *new_object(struct object_id *oid)
563 struct object_entry *e;
565 if (blocks->next_free == blocks->end)
566 alloc_objects(object_entry_alloc);
568 e = blocks->next_free++;
569 oidcpy(&e->idx.oid, oid);
570 return e;
573 static struct object_entry *find_object(struct object_id *oid)
575 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
576 struct object_entry *e;
577 for (e = object_table[h]; e; e = e->next)
578 if (!oidcmp(oid, &e->idx.oid))
579 return e;
580 return NULL;
583 static struct object_entry *insert_object(struct object_id *oid)
585 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
586 struct object_entry *e = object_table[h];
588 while (e) {
589 if (!oidcmp(oid, &e->idx.oid))
590 return e;
591 e = e->next;
594 e = new_object(oid);
595 e->next = object_table[h];
596 e->idx.offset = 0;
597 object_table[h] = e;
598 return e;
601 static void invalidate_pack_id(unsigned int id)
603 unsigned int h;
604 unsigned long lu;
605 struct tag *t;
607 for (h = 0; h < ARRAY_SIZE(object_table); h++) {
608 struct object_entry *e;
610 for (e = object_table[h]; e; e = e->next)
611 if (e->pack_id == id)
612 e->pack_id = MAX_PACK_ID;
615 for (lu = 0; lu < branch_table_sz; lu++) {
616 struct branch *b;
618 for (b = branch_table[lu]; b; b = b->table_next_branch)
619 if (b->pack_id == id)
620 b->pack_id = MAX_PACK_ID;
623 for (t = first_tag; t; t = t->next_tag)
624 if (t->pack_id == id)
625 t->pack_id = MAX_PACK_ID;
628 static unsigned int hc_str(const char *s, size_t len)
630 unsigned int r = 0;
631 while (len-- > 0)
632 r = r * 31 + *s++;
633 return r;
636 static void *pool_alloc(size_t len)
638 struct mem_pool *p;
639 void *r;
641 /* round up to a 'uintmax_t' alignment */
642 if (len & (sizeof(uintmax_t) - 1))
643 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
645 for (p = mem_pool; p; p = p->next_pool)
646 if ((p->end - p->next_free >= len))
647 break;
649 if (!p) {
650 if (len >= (mem_pool_alloc/2)) {
651 total_allocd += len;
652 return xmalloc(len);
654 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
655 p = xmalloc(st_add(sizeof(struct mem_pool), mem_pool_alloc));
656 p->next_pool = mem_pool;
657 p->next_free = (char *) p->space;
658 p->end = p->next_free + mem_pool_alloc;
659 mem_pool = p;
662 r = p->next_free;
663 p->next_free += len;
664 return r;
667 static void *pool_calloc(size_t count, size_t size)
669 size_t len = count * size;
670 void *r = pool_alloc(len);
671 memset(r, 0, len);
672 return r;
675 static char *pool_strdup(const char *s)
677 size_t len = strlen(s) + 1;
678 char *r = pool_alloc(len);
679 memcpy(r, s, len);
680 return r;
683 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
685 struct mark_set *s = marks;
686 while ((idnum >> s->shift) >= 1024) {
687 s = pool_calloc(1, sizeof(struct mark_set));
688 s->shift = marks->shift + 10;
689 s->data.sets[0] = marks;
690 marks = s;
692 while (s->shift) {
693 uintmax_t i = idnum >> s->shift;
694 idnum -= i << s->shift;
695 if (!s->data.sets[i]) {
696 s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
697 s->data.sets[i]->shift = s->shift - 10;
699 s = s->data.sets[i];
701 if (!s->data.marked[idnum])
702 marks_set_count++;
703 s->data.marked[idnum] = oe;
706 static struct object_entry *find_mark(uintmax_t idnum)
708 uintmax_t orig_idnum = idnum;
709 struct mark_set *s = marks;
710 struct object_entry *oe = NULL;
711 if ((idnum >> s->shift) < 1024) {
712 while (s && s->shift) {
713 uintmax_t i = idnum >> s->shift;
714 idnum -= i << s->shift;
715 s = s->data.sets[i];
717 if (s)
718 oe = s->data.marked[idnum];
720 if (!oe)
721 die("mark :%" PRIuMAX " not declared", orig_idnum);
722 return oe;
725 static struct atom_str *to_atom(const char *s, unsigned short len)
727 unsigned int hc = hc_str(s, len) % atom_table_sz;
728 struct atom_str *c;
730 for (c = atom_table[hc]; c; c = c->next_atom)
731 if (c->str_len == len && !strncmp(s, c->str_dat, len))
732 return c;
734 c = pool_alloc(sizeof(struct atom_str) + len + 1);
735 c->str_len = len;
736 memcpy(c->str_dat, s, len);
737 c->str_dat[len] = 0;
738 c->next_atom = atom_table[hc];
739 atom_table[hc] = c;
740 atom_cnt++;
741 return c;
744 static struct branch *lookup_branch(const char *name)
746 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
747 struct branch *b;
749 for (b = branch_table[hc]; b; b = b->table_next_branch)
750 if (!strcmp(name, b->name))
751 return b;
752 return NULL;
755 static struct branch *new_branch(const char *name)
757 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
758 struct branch *b = lookup_branch(name);
760 if (b)
761 die("Invalid attempt to create duplicate branch: %s", name);
762 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
763 die("Branch name doesn't conform to GIT standards: %s", name);
765 b = pool_calloc(1, sizeof(struct branch));
766 b->name = pool_strdup(name);
767 b->table_next_branch = branch_table[hc];
768 b->branch_tree.versions[0].mode = S_IFDIR;
769 b->branch_tree.versions[1].mode = S_IFDIR;
770 b->num_notes = 0;
771 b->active = 0;
772 b->pack_id = MAX_PACK_ID;
773 branch_table[hc] = b;
774 branch_count++;
775 return b;
778 static unsigned int hc_entries(unsigned int cnt)
780 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
781 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
784 static struct tree_content *new_tree_content(unsigned int cnt)
786 struct avail_tree_content *f, *l = NULL;
787 struct tree_content *t;
788 unsigned int hc = hc_entries(cnt);
790 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
791 if (f->entry_capacity >= cnt)
792 break;
794 if (f) {
795 if (l)
796 l->next_avail = f->next_avail;
797 else
798 avail_tree_table[hc] = f->next_avail;
799 } else {
800 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
801 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
802 f->entry_capacity = cnt;
805 t = (struct tree_content*)f;
806 t->entry_count = 0;
807 t->delta_depth = 0;
808 return t;
811 static void release_tree_entry(struct tree_entry *e);
812 static void release_tree_content(struct tree_content *t)
814 struct avail_tree_content *f = (struct avail_tree_content*)t;
815 unsigned int hc = hc_entries(f->entry_capacity);
816 f->next_avail = avail_tree_table[hc];
817 avail_tree_table[hc] = f;
820 static void release_tree_content_recursive(struct tree_content *t)
822 unsigned int i;
823 for (i = 0; i < t->entry_count; i++)
824 release_tree_entry(t->entries[i]);
825 release_tree_content(t);
828 static struct tree_content *grow_tree_content(
829 struct tree_content *t,
830 int amt)
832 struct tree_content *r = new_tree_content(t->entry_count + amt);
833 r->entry_count = t->entry_count;
834 r->delta_depth = t->delta_depth;
835 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
836 release_tree_content(t);
837 return r;
840 static struct tree_entry *new_tree_entry(void)
842 struct tree_entry *e;
844 if (!avail_tree_entry) {
845 unsigned int n = tree_entry_alloc;
846 total_allocd += n * sizeof(struct tree_entry);
847 ALLOC_ARRAY(e, n);
848 avail_tree_entry = e;
849 while (n-- > 1) {
850 *((void**)e) = e + 1;
851 e++;
853 *((void**)e) = NULL;
856 e = avail_tree_entry;
857 avail_tree_entry = *((void**)e);
858 return e;
861 static void release_tree_entry(struct tree_entry *e)
863 if (e->tree)
864 release_tree_content_recursive(e->tree);
865 *((void**)e) = avail_tree_entry;
866 avail_tree_entry = e;
869 static struct tree_content *dup_tree_content(struct tree_content *s)
871 struct tree_content *d;
872 struct tree_entry *a, *b;
873 unsigned int i;
875 if (!s)
876 return NULL;
877 d = new_tree_content(s->entry_count);
878 for (i = 0; i < s->entry_count; i++) {
879 a = s->entries[i];
880 b = new_tree_entry();
881 memcpy(b, a, sizeof(*a));
882 if (a->tree && is_null_oid(&b->versions[1].oid))
883 b->tree = dup_tree_content(a->tree);
884 else
885 b->tree = NULL;
886 d->entries[i] = b;
888 d->entry_count = s->entry_count;
889 d->delta_depth = s->delta_depth;
891 return d;
894 static void start_packfile(void)
896 struct strbuf tmp_file = STRBUF_INIT;
897 struct packed_git *p;
898 struct pack_header hdr;
899 int pack_fd;
901 pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
902 FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
903 strbuf_release(&tmp_file);
905 p->pack_fd = pack_fd;
906 p->do_not_close = 1;
907 pack_file = sha1fd(pack_fd, p->pack_name);
909 hdr.hdr_signature = htonl(PACK_SIGNATURE);
910 hdr.hdr_version = htonl(2);
911 hdr.hdr_entries = 0;
912 sha1write(pack_file, &hdr, sizeof(hdr));
914 pack_data = p;
915 pack_size = sizeof(hdr);
916 object_count = 0;
918 REALLOC_ARRAY(all_packs, pack_id + 1);
919 all_packs[pack_id] = p;
922 static const char *create_index(void)
924 const char *tmpfile;
925 struct pack_idx_entry **idx, **c, **last;
926 struct object_entry *e;
927 struct object_entry_pool *o;
929 /* Build the table of object IDs. */
930 ALLOC_ARRAY(idx, object_count);
931 c = idx;
932 for (o = blocks; o; o = o->next_pool)
933 for (e = o->next_free; e-- != o->entries;)
934 if (pack_id == e->pack_id)
935 *c++ = &e->idx;
936 last = idx + object_count;
937 if (c != last)
938 die("internal consistency error creating the index");
940 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
941 free(idx);
942 return tmpfile;
945 static char *keep_pack(const char *curr_index_name)
947 static const char *keep_msg = "fast-import";
948 struct strbuf name = STRBUF_INIT;
949 int keep_fd;
951 odb_pack_name(&name, pack_data->sha1, "keep");
952 keep_fd = odb_pack_keep(name.buf);
953 if (keep_fd < 0)
954 die_errno("cannot create keep file");
955 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
956 if (close(keep_fd))
957 die_errno("failed to write keep file");
959 odb_pack_name(&name, pack_data->sha1, "pack");
960 if (finalize_object_file(pack_data->pack_name, name.buf))
961 die("cannot store pack file");
963 odb_pack_name(&name, pack_data->sha1, "idx");
964 if (finalize_object_file(curr_index_name, name.buf))
965 die("cannot store index file");
966 free((void *)curr_index_name);
967 return strbuf_detach(&name, NULL);
970 static void unkeep_all_packs(void)
972 struct strbuf name = STRBUF_INIT;
973 int k;
975 for (k = 0; k < pack_id; k++) {
976 struct packed_git *p = all_packs[k];
977 odb_pack_name(&name, p->sha1, "keep");
978 unlink_or_warn(name.buf);
980 strbuf_release(&name);
983 static int loosen_small_pack(const struct packed_git *p)
985 struct child_process unpack = CHILD_PROCESS_INIT;
987 if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
988 die_errno("Failed seeking to start of '%s'", p->pack_name);
990 unpack.in = p->pack_fd;
991 unpack.git_cmd = 1;
992 unpack.stdout_to_stderr = 1;
993 argv_array_push(&unpack.args, "unpack-objects");
994 if (!show_stats)
995 argv_array_push(&unpack.args, "-q");
997 return run_command(&unpack);
1000 static void end_packfile(void)
1002 static int running;
1004 if (running || !pack_data)
1005 return;
1007 running = 1;
1008 clear_delta_base_cache();
1009 if (object_count) {
1010 struct packed_git *new_p;
1011 struct object_id cur_pack_oid;
1012 char *idx_name;
1013 int i;
1014 struct branch *b;
1015 struct tag *t;
1017 close_pack_windows(pack_data);
1018 sha1close(pack_file, cur_pack_oid.hash, 0);
1019 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
1020 pack_data->pack_name, object_count,
1021 cur_pack_oid.hash, pack_size);
1023 if (object_count <= unpack_limit) {
1024 if (!loosen_small_pack(pack_data)) {
1025 invalidate_pack_id(pack_id);
1026 goto discard_pack;
1030 close(pack_data->pack_fd);
1031 idx_name = keep_pack(create_index());
1033 /* Register the packfile with core git's machinery. */
1034 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
1035 if (!new_p)
1036 die("core git rejected index %s", idx_name);
1037 all_packs[pack_id] = new_p;
1038 install_packed_git(new_p);
1039 free(idx_name);
1041 /* Print the boundary */
1042 if (pack_edges) {
1043 fprintf(pack_edges, "%s:", new_p->pack_name);
1044 for (i = 0; i < branch_table_sz; i++) {
1045 for (b = branch_table[i]; b; b = b->table_next_branch) {
1046 if (b->pack_id == pack_id)
1047 fprintf(pack_edges, " %s",
1048 oid_to_hex(&b->oid));
1051 for (t = first_tag; t; t = t->next_tag) {
1052 if (t->pack_id == pack_id)
1053 fprintf(pack_edges, " %s",
1054 oid_to_hex(&t->oid));
1056 fputc('\n', pack_edges);
1057 fflush(pack_edges);
1060 pack_id++;
1062 else {
1063 discard_pack:
1064 close(pack_data->pack_fd);
1065 unlink_or_warn(pack_data->pack_name);
1067 FREE_AND_NULL(pack_data);
1068 running = 0;
1070 /* We can't carry a delta across packfiles. */
1071 strbuf_release(&last_blob.data);
1072 last_blob.offset = 0;
1073 last_blob.depth = 0;
1076 static void cycle_packfile(void)
1078 end_packfile();
1079 start_packfile();
1082 static int store_object(
1083 enum object_type type,
1084 struct strbuf *dat,
1085 struct last_object *last,
1086 struct object_id *oidout,
1087 uintmax_t mark)
1089 void *out, *delta;
1090 struct object_entry *e;
1091 unsigned char hdr[96];
1092 struct object_id oid;
1093 unsigned long hdrlen, deltalen;
1094 git_SHA_CTX c;
1095 git_zstream s;
1097 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1098 typename(type), (unsigned long)dat->len) + 1;
1099 git_SHA1_Init(&c);
1100 git_SHA1_Update(&c, hdr, hdrlen);
1101 git_SHA1_Update(&c, dat->buf, dat->len);
1102 git_SHA1_Final(oid.hash, &c);
1103 if (oidout)
1104 oidcpy(oidout, &oid);
1106 e = insert_object(&oid);
1107 if (mark)
1108 insert_mark(mark, e);
1109 if (e->idx.offset) {
1110 duplicate_count_by_type[type]++;
1111 return 1;
1112 } else if (find_sha1_pack(oid.hash, packed_git)) {
1113 e->type = type;
1114 e->pack_id = MAX_PACK_ID;
1115 e->idx.offset = 1; /* just not zero! */
1116 duplicate_count_by_type[type]++;
1117 return 1;
1120 if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1121 delta_count_attempts_by_type[type]++;
1122 delta = diff_delta(last->data.buf, last->data.len,
1123 dat->buf, dat->len,
1124 &deltalen, dat->len - 20);
1125 } else
1126 delta = NULL;
1128 git_deflate_init(&s, pack_compression_level);
1129 if (delta) {
1130 s.next_in = delta;
1131 s.avail_in = deltalen;
1132 } else {
1133 s.next_in = (void *)dat->buf;
1134 s.avail_in = dat->len;
1136 s.avail_out = git_deflate_bound(&s, s.avail_in);
1137 s.next_out = out = xmalloc(s.avail_out);
1138 while (git_deflate(&s, Z_FINISH) == Z_OK)
1139 ; /* nothing */
1140 git_deflate_end(&s);
1142 /* Determine if we should auto-checkpoint. */
1143 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1144 || (pack_size + 60 + s.total_out) < pack_size) {
1146 /* This new object needs to *not* have the current pack_id. */
1147 e->pack_id = pack_id + 1;
1148 cycle_packfile();
1150 /* We cannot carry a delta into the new pack. */
1151 if (delta) {
1152 FREE_AND_NULL(delta);
1154 git_deflate_init(&s, pack_compression_level);
1155 s.next_in = (void *)dat->buf;
1156 s.avail_in = dat->len;
1157 s.avail_out = git_deflate_bound(&s, s.avail_in);
1158 s.next_out = out = xrealloc(out, s.avail_out);
1159 while (git_deflate(&s, Z_FINISH) == Z_OK)
1160 ; /* nothing */
1161 git_deflate_end(&s);
1165 e->type = type;
1166 e->pack_id = pack_id;
1167 e->idx.offset = pack_size;
1168 object_count++;
1169 object_count_by_type[type]++;
1171 crc32_begin(pack_file);
1173 if (delta) {
1174 off_t ofs = e->idx.offset - last->offset;
1175 unsigned pos = sizeof(hdr) - 1;
1177 delta_count_by_type[type]++;
1178 e->depth = last->depth + 1;
1180 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1181 OBJ_OFS_DELTA, deltalen);
1182 sha1write(pack_file, hdr, hdrlen);
1183 pack_size += hdrlen;
1185 hdr[pos] = ofs & 127;
1186 while (ofs >>= 7)
1187 hdr[--pos] = 128 | (--ofs & 127);
1188 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1189 pack_size += sizeof(hdr) - pos;
1190 } else {
1191 e->depth = 0;
1192 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1193 type, dat->len);
1194 sha1write(pack_file, hdr, hdrlen);
1195 pack_size += hdrlen;
1198 sha1write(pack_file, out, s.total_out);
1199 pack_size += s.total_out;
1201 e->idx.crc32 = crc32_end(pack_file);
1203 free(out);
1204 free(delta);
1205 if (last) {
1206 if (last->no_swap) {
1207 last->data = *dat;
1208 } else {
1209 strbuf_swap(&last->data, dat);
1211 last->offset = e->idx.offset;
1212 last->depth = e->depth;
1214 return 0;
1217 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1219 if (sha1file_truncate(pack_file, checkpoint))
1220 die_errno("cannot truncate pack to skip duplicate");
1221 pack_size = checkpoint->offset;
1224 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1226 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1227 unsigned char *in_buf = xmalloc(in_sz);
1228 unsigned char *out_buf = xmalloc(out_sz);
1229 struct object_entry *e;
1230 struct object_id oid;
1231 unsigned long hdrlen;
1232 off_t offset;
1233 git_SHA_CTX c;
1234 git_zstream s;
1235 struct sha1file_checkpoint checkpoint;
1236 int status = Z_OK;
1238 /* Determine if we should auto-checkpoint. */
1239 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1240 || (pack_size + 60 + len) < pack_size)
1241 cycle_packfile();
1243 sha1file_checkpoint(pack_file, &checkpoint);
1244 offset = checkpoint.offset;
1246 hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1248 git_SHA1_Init(&c);
1249 git_SHA1_Update(&c, out_buf, hdrlen);
1251 crc32_begin(pack_file);
1253 git_deflate_init(&s, pack_compression_level);
1255 hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1257 s.next_out = out_buf + hdrlen;
1258 s.avail_out = out_sz - hdrlen;
1260 while (status != Z_STREAM_END) {
1261 if (0 < len && !s.avail_in) {
1262 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1263 size_t n = fread(in_buf, 1, cnt, stdin);
1264 if (!n && feof(stdin))
1265 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1267 git_SHA1_Update(&c, in_buf, n);
1268 s.next_in = in_buf;
1269 s.avail_in = n;
1270 len -= n;
1273 status = git_deflate(&s, len ? 0 : Z_FINISH);
1275 if (!s.avail_out || status == Z_STREAM_END) {
1276 size_t n = s.next_out - out_buf;
1277 sha1write(pack_file, out_buf, n);
1278 pack_size += n;
1279 s.next_out = out_buf;
1280 s.avail_out = out_sz;
1283 switch (status) {
1284 case Z_OK:
1285 case Z_BUF_ERROR:
1286 case Z_STREAM_END:
1287 continue;
1288 default:
1289 die("unexpected deflate failure: %d", status);
1292 git_deflate_end(&s);
1293 git_SHA1_Final(oid.hash, &c);
1295 if (oidout)
1296 oidcpy(oidout, &oid);
1298 e = insert_object(&oid);
1300 if (mark)
1301 insert_mark(mark, e);
1303 if (e->idx.offset) {
1304 duplicate_count_by_type[OBJ_BLOB]++;
1305 truncate_pack(&checkpoint);
1307 } else if (find_sha1_pack(oid.hash, packed_git)) {
1308 e->type = OBJ_BLOB;
1309 e->pack_id = MAX_PACK_ID;
1310 e->idx.offset = 1; /* just not zero! */
1311 duplicate_count_by_type[OBJ_BLOB]++;
1312 truncate_pack(&checkpoint);
1314 } else {
1315 e->depth = 0;
1316 e->type = OBJ_BLOB;
1317 e->pack_id = pack_id;
1318 e->idx.offset = offset;
1319 e->idx.crc32 = crc32_end(pack_file);
1320 object_count++;
1321 object_count_by_type[OBJ_BLOB]++;
1324 free(in_buf);
1325 free(out_buf);
1328 /* All calls must be guarded by find_object() or find_mark() to
1329 * ensure the 'struct object_entry' passed was written by this
1330 * process instance. We unpack the entry by the offset, avoiding
1331 * the need for the corresponding .idx file. This unpacking rule
1332 * works because we only use OBJ_REF_DELTA within the packfiles
1333 * created by fast-import.
1335 * oe must not be NULL. Such an oe usually comes from giving
1336 * an unknown SHA-1 to find_object() or an undefined mark to
1337 * find_mark(). Callers must test for this condition and use
1338 * the standard read_sha1_file() when it happens.
1340 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1341 * find_mark(), where the mark was reloaded from an existing marks
1342 * file and is referencing an object that this fast-import process
1343 * instance did not write out to a packfile. Callers must test for
1344 * this condition and use read_sha1_file() instead.
1346 static void *gfi_unpack_entry(
1347 struct object_entry *oe,
1348 unsigned long *sizep)
1350 enum object_type type;
1351 struct packed_git *p = all_packs[oe->pack_id];
1352 if (p == pack_data && p->pack_size < (pack_size + 20)) {
1353 /* The object is stored in the packfile we are writing to
1354 * and we have modified it since the last time we scanned
1355 * back to read a previously written object. If an old
1356 * window covered [p->pack_size, p->pack_size + 20) its
1357 * data is stale and is not valid. Closing all windows
1358 * and updating the packfile length ensures we can read
1359 * the newly written data.
1361 close_pack_windows(p);
1362 sha1flush(pack_file);
1364 /* We have to offer 20 bytes additional on the end of
1365 * the packfile as the core unpacker code assumes the
1366 * footer is present at the file end and must promise
1367 * at least 20 bytes within any window it maps. But
1368 * we don't actually create the footer here.
1370 p->pack_size = pack_size + 20;
1372 return unpack_entry(p, oe->idx.offset, &type, sizep);
1375 static const char *get_mode(const char *str, uint16_t *modep)
1377 unsigned char c;
1378 uint16_t mode = 0;
1380 while ((c = *str++) != ' ') {
1381 if (c < '0' || c > '7')
1382 return NULL;
1383 mode = (mode << 3) + (c - '0');
1385 *modep = mode;
1386 return str;
1389 static void load_tree(struct tree_entry *root)
1391 struct object_id *oid = &root->versions[1].oid;
1392 struct object_entry *myoe;
1393 struct tree_content *t;
1394 unsigned long size;
1395 char *buf;
1396 const char *c;
1398 root->tree = t = new_tree_content(8);
1399 if (is_null_oid(oid))
1400 return;
1402 myoe = find_object(oid);
1403 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1404 if (myoe->type != OBJ_TREE)
1405 die("Not a tree: %s", oid_to_hex(oid));
1406 t->delta_depth = myoe->depth;
1407 buf = gfi_unpack_entry(myoe, &size);
1408 if (!buf)
1409 die("Can't load tree %s", oid_to_hex(oid));
1410 } else {
1411 enum object_type type;
1412 buf = read_sha1_file(oid->hash, &type, &size);
1413 if (!buf || type != OBJ_TREE)
1414 die("Can't load tree %s", oid_to_hex(oid));
1417 c = buf;
1418 while (c != (buf + size)) {
1419 struct tree_entry *e = new_tree_entry();
1421 if (t->entry_count == t->entry_capacity)
1422 root->tree = t = grow_tree_content(t, t->entry_count);
1423 t->entries[t->entry_count++] = e;
1425 e->tree = NULL;
1426 c = get_mode(c, &e->versions[1].mode);
1427 if (!c)
1428 die("Corrupt mode in %s", oid_to_hex(oid));
1429 e->versions[0].mode = e->versions[1].mode;
1430 e->name = to_atom(c, strlen(c));
1431 c += e->name->str_len + 1;
1432 hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1433 hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1434 c += GIT_SHA1_RAWSZ;
1436 free(buf);
1439 static int tecmp0 (const void *_a, const void *_b)
1441 struct tree_entry *a = *((struct tree_entry**)_a);
1442 struct tree_entry *b = *((struct tree_entry**)_b);
1443 return base_name_compare(
1444 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1445 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1448 static int tecmp1 (const void *_a, const void *_b)
1450 struct tree_entry *a = *((struct tree_entry**)_a);
1451 struct tree_entry *b = *((struct tree_entry**)_b);
1452 return base_name_compare(
1453 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1454 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1457 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1459 size_t maxlen = 0;
1460 unsigned int i;
1462 if (!v)
1463 QSORT(t->entries, t->entry_count, tecmp0);
1464 else
1465 QSORT(t->entries, t->entry_count, tecmp1);
1467 for (i = 0; i < t->entry_count; i++) {
1468 if (t->entries[i]->versions[v].mode)
1469 maxlen += t->entries[i]->name->str_len + 34;
1472 strbuf_reset(b);
1473 strbuf_grow(b, maxlen);
1474 for (i = 0; i < t->entry_count; i++) {
1475 struct tree_entry *e = t->entries[i];
1476 if (!e->versions[v].mode)
1477 continue;
1478 strbuf_addf(b, "%o %s%c",
1479 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1480 e->name->str_dat, '\0');
1481 strbuf_add(b, e->versions[v].oid.hash, GIT_SHA1_RAWSZ);
1485 static void store_tree(struct tree_entry *root)
1487 struct tree_content *t;
1488 unsigned int i, j, del;
1489 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1490 struct object_entry *le = NULL;
1492 if (!is_null_oid(&root->versions[1].oid))
1493 return;
1495 if (!root->tree)
1496 load_tree(root);
1497 t = root->tree;
1499 for (i = 0; i < t->entry_count; i++) {
1500 if (t->entries[i]->tree)
1501 store_tree(t->entries[i]);
1504 if (!(root->versions[0].mode & NO_DELTA))
1505 le = find_object(&root->versions[0].oid);
1506 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1507 mktree(t, 0, &old_tree);
1508 lo.data = old_tree;
1509 lo.offset = le->idx.offset;
1510 lo.depth = t->delta_depth;
1513 mktree(t, 1, &new_tree);
1514 store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1516 t->delta_depth = lo.depth;
1517 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1518 struct tree_entry *e = t->entries[i];
1519 if (e->versions[1].mode) {
1520 e->versions[0].mode = e->versions[1].mode;
1521 oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1522 t->entries[j++] = e;
1523 } else {
1524 release_tree_entry(e);
1525 del++;
1528 t->entry_count -= del;
1531 static void tree_content_replace(
1532 struct tree_entry *root,
1533 const struct object_id *oid,
1534 const uint16_t mode,
1535 struct tree_content *newtree)
1537 if (!S_ISDIR(mode))
1538 die("Root cannot be a non-directory");
1539 oidclr(&root->versions[0].oid);
1540 oidcpy(&root->versions[1].oid, oid);
1541 if (root->tree)
1542 release_tree_content_recursive(root->tree);
1543 root->tree = newtree;
1546 static int tree_content_set(
1547 struct tree_entry *root,
1548 const char *p,
1549 const struct object_id *oid,
1550 const uint16_t mode,
1551 struct tree_content *subtree)
1553 struct tree_content *t;
1554 const char *slash1;
1555 unsigned int i, n;
1556 struct tree_entry *e;
1558 slash1 = strchrnul(p, '/');
1559 n = slash1 - p;
1560 if (!n)
1561 die("Empty path component found in input");
1562 if (!*slash1 && !S_ISDIR(mode) && subtree)
1563 die("Non-directories cannot have subtrees");
1565 if (!root->tree)
1566 load_tree(root);
1567 t = root->tree;
1568 for (i = 0; i < t->entry_count; i++) {
1569 e = t->entries[i];
1570 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1571 if (!*slash1) {
1572 if (!S_ISDIR(mode)
1573 && e->versions[1].mode == mode
1574 && !oidcmp(&e->versions[1].oid, oid))
1575 return 0;
1576 e->versions[1].mode = mode;
1577 oidcpy(&e->versions[1].oid, oid);
1578 if (e->tree)
1579 release_tree_content_recursive(e->tree);
1580 e->tree = subtree;
1583 * We need to leave e->versions[0].sha1 alone
1584 * to avoid modifying the preimage tree used
1585 * when writing out the parent directory.
1586 * But after replacing the subdir with a
1587 * completely different one, it's not a good
1588 * delta base any more, and besides, we've
1589 * thrown away the tree entries needed to
1590 * make a delta against it.
1592 * So let's just explicitly disable deltas
1593 * for the subtree.
1595 if (S_ISDIR(e->versions[0].mode))
1596 e->versions[0].mode |= NO_DELTA;
1598 oidclr(&root->versions[1].oid);
1599 return 1;
1601 if (!S_ISDIR(e->versions[1].mode)) {
1602 e->tree = new_tree_content(8);
1603 e->versions[1].mode = S_IFDIR;
1605 if (!e->tree)
1606 load_tree(e);
1607 if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1608 oidclr(&root->versions[1].oid);
1609 return 1;
1611 return 0;
1615 if (t->entry_count == t->entry_capacity)
1616 root->tree = t = grow_tree_content(t, t->entry_count);
1617 e = new_tree_entry();
1618 e->name = to_atom(p, n);
1619 e->versions[0].mode = 0;
1620 oidclr(&e->versions[0].oid);
1621 t->entries[t->entry_count++] = e;
1622 if (*slash1) {
1623 e->tree = new_tree_content(8);
1624 e->versions[1].mode = S_IFDIR;
1625 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1626 } else {
1627 e->tree = subtree;
1628 e->versions[1].mode = mode;
1629 oidcpy(&e->versions[1].oid, oid);
1631 oidclr(&root->versions[1].oid);
1632 return 1;
1635 static int tree_content_remove(
1636 struct tree_entry *root,
1637 const char *p,
1638 struct tree_entry *backup_leaf,
1639 int allow_root)
1641 struct tree_content *t;
1642 const char *slash1;
1643 unsigned int i, n;
1644 struct tree_entry *e;
1646 slash1 = strchrnul(p, '/');
1647 n = slash1 - p;
1649 if (!root->tree)
1650 load_tree(root);
1652 if (!*p && allow_root) {
1653 e = root;
1654 goto del_entry;
1657 t = root->tree;
1658 for (i = 0; i < t->entry_count; i++) {
1659 e = t->entries[i];
1660 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1661 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1663 * If p names a file in some subdirectory, and a
1664 * file or symlink matching the name of the
1665 * parent directory of p exists, then p cannot
1666 * exist and need not be deleted.
1668 return 1;
1669 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1670 goto del_entry;
1671 if (!e->tree)
1672 load_tree(e);
1673 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1674 for (n = 0; n < e->tree->entry_count; n++) {
1675 if (e->tree->entries[n]->versions[1].mode) {
1676 oidclr(&root->versions[1].oid);
1677 return 1;
1680 backup_leaf = NULL;
1681 goto del_entry;
1683 return 0;
1686 return 0;
1688 del_entry:
1689 if (backup_leaf)
1690 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1691 else if (e->tree)
1692 release_tree_content_recursive(e->tree);
1693 e->tree = NULL;
1694 e->versions[1].mode = 0;
1695 oidclr(&e->versions[1].oid);
1696 oidclr(&root->versions[1].oid);
1697 return 1;
1700 static int tree_content_get(
1701 struct tree_entry *root,
1702 const char *p,
1703 struct tree_entry *leaf,
1704 int allow_root)
1706 struct tree_content *t;
1707 const char *slash1;
1708 unsigned int i, n;
1709 struct tree_entry *e;
1711 slash1 = strchrnul(p, '/');
1712 n = slash1 - p;
1713 if (!n && !allow_root)
1714 die("Empty path component found in input");
1716 if (!root->tree)
1717 load_tree(root);
1719 if (!n) {
1720 e = root;
1721 goto found_entry;
1724 t = root->tree;
1725 for (i = 0; i < t->entry_count; i++) {
1726 e = t->entries[i];
1727 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1728 if (!*slash1)
1729 goto found_entry;
1730 if (!S_ISDIR(e->versions[1].mode))
1731 return 0;
1732 if (!e->tree)
1733 load_tree(e);
1734 return tree_content_get(e, slash1 + 1, leaf, 0);
1737 return 0;
1739 found_entry:
1740 memcpy(leaf, e, sizeof(*leaf));
1741 if (e->tree && is_null_oid(&e->versions[1].oid))
1742 leaf->tree = dup_tree_content(e->tree);
1743 else
1744 leaf->tree = NULL;
1745 return 1;
1748 static int update_branch(struct branch *b)
1750 static const char *msg = "fast-import";
1751 struct ref_transaction *transaction;
1752 struct object_id old_oid;
1753 struct strbuf err = STRBUF_INIT;
1755 if (is_null_oid(&b->oid)) {
1756 if (b->delete)
1757 delete_ref(NULL, b->name, NULL, 0);
1758 return 0;
1760 if (read_ref(b->name, old_oid.hash))
1761 oidclr(&old_oid);
1762 if (!force_update && !is_null_oid(&old_oid)) {
1763 struct commit *old_cmit, *new_cmit;
1765 old_cmit = lookup_commit_reference_gently(&old_oid, 0);
1766 new_cmit = lookup_commit_reference_gently(&b->oid, 0);
1767 if (!old_cmit || !new_cmit)
1768 return error("Branch %s is missing commits.", b->name);
1770 if (!in_merge_bases(old_cmit, new_cmit)) {
1771 warning("Not updating %s"
1772 " (new tip %s does not contain %s)",
1773 b->name, oid_to_hex(&b->oid),
1774 oid_to_hex(&old_oid));
1775 return -1;
1778 transaction = ref_transaction_begin(&err);
1779 if (!transaction ||
1780 ref_transaction_update(transaction, b->name, b->oid.hash, old_oid.hash,
1781 0, msg, &err) ||
1782 ref_transaction_commit(transaction, &err)) {
1783 ref_transaction_free(transaction);
1784 error("%s", err.buf);
1785 strbuf_release(&err);
1786 return -1;
1788 ref_transaction_free(transaction);
1789 strbuf_release(&err);
1790 return 0;
1793 static void dump_branches(void)
1795 unsigned int i;
1796 struct branch *b;
1798 for (i = 0; i < branch_table_sz; i++) {
1799 for (b = branch_table[i]; b; b = b->table_next_branch)
1800 failure |= update_branch(b);
1804 static void dump_tags(void)
1806 static const char *msg = "fast-import";
1807 struct tag *t;
1808 struct strbuf ref_name = STRBUF_INIT;
1809 struct strbuf err = STRBUF_INIT;
1810 struct ref_transaction *transaction;
1812 transaction = ref_transaction_begin(&err);
1813 if (!transaction) {
1814 failure |= error("%s", err.buf);
1815 goto cleanup;
1817 for (t = first_tag; t; t = t->next_tag) {
1818 strbuf_reset(&ref_name);
1819 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1821 if (ref_transaction_update(transaction, ref_name.buf,
1822 t->oid.hash, NULL, 0, msg, &err)) {
1823 failure |= error("%s", err.buf);
1824 goto cleanup;
1827 if (ref_transaction_commit(transaction, &err))
1828 failure |= error("%s", err.buf);
1830 cleanup:
1831 ref_transaction_free(transaction);
1832 strbuf_release(&ref_name);
1833 strbuf_release(&err);
1836 static void dump_marks_helper(FILE *f,
1837 uintmax_t base,
1838 struct mark_set *m)
1840 uintmax_t k;
1841 if (m->shift) {
1842 for (k = 0; k < 1024; k++) {
1843 if (m->data.sets[k])
1844 dump_marks_helper(f, base + (k << m->shift),
1845 m->data.sets[k]);
1847 } else {
1848 for (k = 0; k < 1024; k++) {
1849 if (m->data.marked[k])
1850 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1851 oid_to_hex(&m->data.marked[k]->idx.oid));
1856 static void dump_marks(void)
1858 static struct lock_file mark_lock;
1859 FILE *f;
1861 if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1862 return;
1864 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1865 failure |= error_errno("Unable to write marks file %s",
1866 export_marks_file);
1867 return;
1870 f = fdopen_lock_file(&mark_lock, "w");
1871 if (!f) {
1872 int saved_errno = errno;
1873 rollback_lock_file(&mark_lock);
1874 failure |= error("Unable to write marks file %s: %s",
1875 export_marks_file, strerror(saved_errno));
1876 return;
1879 dump_marks_helper(f, 0, marks);
1880 if (commit_lock_file(&mark_lock)) {
1881 failure |= error_errno("Unable to write file %s",
1882 export_marks_file);
1883 return;
1887 static void read_marks(void)
1889 char line[512];
1890 FILE *f = fopen(import_marks_file, "r");
1891 if (f)
1893 else if (import_marks_file_ignore_missing && errno == ENOENT)
1894 goto done; /* Marks file does not exist */
1895 else
1896 die_errno("cannot read '%s'", import_marks_file);
1897 while (fgets(line, sizeof(line), f)) {
1898 uintmax_t mark;
1899 char *end;
1900 struct object_id oid;
1901 struct object_entry *e;
1903 end = strchr(line, '\n');
1904 if (line[0] != ':' || !end)
1905 die("corrupt mark line: %s", line);
1906 *end = 0;
1907 mark = strtoumax(line + 1, &end, 10);
1908 if (!mark || end == line + 1
1909 || *end != ' ' || get_oid_hex(end + 1, &oid))
1910 die("corrupt mark line: %s", line);
1911 e = find_object(&oid);
1912 if (!e) {
1913 enum object_type type = sha1_object_info(oid.hash, NULL);
1914 if (type < 0)
1915 die("object not found: %s", oid_to_hex(&oid));
1916 e = insert_object(&oid);
1917 e->type = type;
1918 e->pack_id = MAX_PACK_ID;
1919 e->idx.offset = 1; /* just not zero! */
1921 insert_mark(mark, e);
1923 fclose(f);
1924 done:
1925 import_marks_file_done = 1;
1929 static int read_next_command(void)
1931 static int stdin_eof = 0;
1933 if (stdin_eof) {
1934 unread_command_buf = 0;
1935 return EOF;
1938 for (;;) {
1939 const char *p;
1941 if (unread_command_buf) {
1942 unread_command_buf = 0;
1943 } else {
1944 struct recent_command *rc;
1946 strbuf_detach(&command_buf, NULL);
1947 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1948 if (stdin_eof)
1949 return EOF;
1951 if (!seen_data_command
1952 && !starts_with(command_buf.buf, "feature ")
1953 && !starts_with(command_buf.buf, "option ")) {
1954 parse_argv();
1957 rc = rc_free;
1958 if (rc)
1959 rc_free = rc->next;
1960 else {
1961 rc = cmd_hist.next;
1962 cmd_hist.next = rc->next;
1963 cmd_hist.next->prev = &cmd_hist;
1964 free(rc->buf);
1967 rc->buf = command_buf.buf;
1968 rc->prev = cmd_tail;
1969 rc->next = cmd_hist.prev;
1970 rc->prev->next = rc;
1971 cmd_tail = rc;
1973 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1974 parse_get_mark(p);
1975 continue;
1977 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1978 parse_cat_blob(p);
1979 continue;
1981 if (command_buf.buf[0] == '#')
1982 continue;
1983 return 0;
1987 static void skip_optional_lf(void)
1989 int term_char = fgetc(stdin);
1990 if (term_char != '\n' && term_char != EOF)
1991 ungetc(term_char, stdin);
1994 static void parse_mark(void)
1996 const char *v;
1997 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1998 next_mark = strtoumax(v, NULL, 10);
1999 read_next_command();
2001 else
2002 next_mark = 0;
2005 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
2007 const char *data;
2008 strbuf_reset(sb);
2010 if (!skip_prefix(command_buf.buf, "data ", &data))
2011 die("Expected 'data n' command, found: %s", command_buf.buf);
2013 if (skip_prefix(data, "<<", &data)) {
2014 char *term = xstrdup(data);
2015 size_t term_len = command_buf.len - (data - command_buf.buf);
2017 strbuf_detach(&command_buf, NULL);
2018 for (;;) {
2019 if (strbuf_getline_lf(&command_buf, stdin) == EOF)
2020 die("EOF in data (terminator '%s' not found)", term);
2021 if (term_len == command_buf.len
2022 && !strcmp(term, command_buf.buf))
2023 break;
2024 strbuf_addbuf(sb, &command_buf);
2025 strbuf_addch(sb, '\n');
2027 free(term);
2029 else {
2030 uintmax_t len = strtoumax(data, NULL, 10);
2031 size_t n = 0, length = (size_t)len;
2033 if (limit && limit < len) {
2034 *len_res = len;
2035 return 0;
2037 if (length < len)
2038 die("data is too large to use in this context");
2040 while (n < length) {
2041 size_t s = strbuf_fread(sb, length - n, stdin);
2042 if (!s && feof(stdin))
2043 die("EOF in data (%lu bytes remaining)",
2044 (unsigned long)(length - n));
2045 n += s;
2049 skip_optional_lf();
2050 return 1;
2053 static int validate_raw_date(const char *src, struct strbuf *result)
2055 const char *orig_src = src;
2056 char *endp;
2057 unsigned long num;
2059 errno = 0;
2061 num = strtoul(src, &endp, 10);
2062 /* NEEDSWORK: perhaps check for reasonable values? */
2063 if (errno || endp == src || *endp != ' ')
2064 return -1;
2066 src = endp + 1;
2067 if (*src != '-' && *src != '+')
2068 return -1;
2070 num = strtoul(src + 1, &endp, 10);
2071 if (errno || endp == src + 1 || *endp || 1400 < num)
2072 return -1;
2074 strbuf_addstr(result, orig_src);
2075 return 0;
2078 static char *parse_ident(const char *buf)
2080 const char *ltgt;
2081 size_t name_len;
2082 struct strbuf ident = STRBUF_INIT;
2084 /* ensure there is a space delimiter even if there is no name */
2085 if (*buf == '<')
2086 --buf;
2088 ltgt = buf + strcspn(buf, "<>");
2089 if (*ltgt != '<')
2090 die("Missing < in ident string: %s", buf);
2091 if (ltgt != buf && ltgt[-1] != ' ')
2092 die("Missing space before < in ident string: %s", buf);
2093 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2094 if (*ltgt != '>')
2095 die("Missing > in ident string: %s", buf);
2096 ltgt++;
2097 if (*ltgt != ' ')
2098 die("Missing space after > in ident string: %s", buf);
2099 ltgt++;
2100 name_len = ltgt - buf;
2101 strbuf_add(&ident, buf, name_len);
2103 switch (whenspec) {
2104 case WHENSPEC_RAW:
2105 if (validate_raw_date(ltgt, &ident) < 0)
2106 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2107 break;
2108 case WHENSPEC_RFC2822:
2109 if (parse_date(ltgt, &ident) < 0)
2110 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2111 break;
2112 case WHENSPEC_NOW:
2113 if (strcmp("now", ltgt))
2114 die("Date in ident must be 'now': %s", buf);
2115 datestamp(&ident);
2116 break;
2119 return strbuf_detach(&ident, NULL);
2122 static void parse_and_store_blob(
2123 struct last_object *last,
2124 struct object_id *oidout,
2125 uintmax_t mark)
2127 static struct strbuf buf = STRBUF_INIT;
2128 uintmax_t len;
2130 if (parse_data(&buf, big_file_threshold, &len))
2131 store_object(OBJ_BLOB, &buf, last, oidout, mark);
2132 else {
2133 if (last) {
2134 strbuf_release(&last->data);
2135 last->offset = 0;
2136 last->depth = 0;
2138 stream_blob(len, oidout, mark);
2139 skip_optional_lf();
2143 static void parse_new_blob(void)
2145 read_next_command();
2146 parse_mark();
2147 parse_and_store_blob(&last_blob, NULL, next_mark);
2150 static void unload_one_branch(void)
2152 while (cur_active_branches
2153 && cur_active_branches >= max_active_branches) {
2154 uintmax_t min_commit = ULONG_MAX;
2155 struct branch *e, *l = NULL, *p = NULL;
2157 for (e = active_branches; e; e = e->active_next_branch) {
2158 if (e->last_commit < min_commit) {
2159 p = l;
2160 min_commit = e->last_commit;
2162 l = e;
2165 if (p) {
2166 e = p->active_next_branch;
2167 p->active_next_branch = e->active_next_branch;
2168 } else {
2169 e = active_branches;
2170 active_branches = e->active_next_branch;
2172 e->active = 0;
2173 e->active_next_branch = NULL;
2174 if (e->branch_tree.tree) {
2175 release_tree_content_recursive(e->branch_tree.tree);
2176 e->branch_tree.tree = NULL;
2178 cur_active_branches--;
2182 static void load_branch(struct branch *b)
2184 load_tree(&b->branch_tree);
2185 if (!b->active) {
2186 b->active = 1;
2187 b->active_next_branch = active_branches;
2188 active_branches = b;
2189 cur_active_branches++;
2190 branch_load_count++;
2194 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2196 unsigned char fanout = 0;
2197 while ((num_notes >>= 8))
2198 fanout++;
2199 return fanout;
2202 static void construct_path_with_fanout(const char *hex_sha1,
2203 unsigned char fanout, char *path)
2205 unsigned int i = 0, j = 0;
2206 if (fanout >= 20)
2207 die("Too large fanout (%u)", fanout);
2208 while (fanout) {
2209 path[i++] = hex_sha1[j++];
2210 path[i++] = hex_sha1[j++];
2211 path[i++] = '/';
2212 fanout--;
2214 memcpy(path + i, hex_sha1 + j, GIT_SHA1_HEXSZ - j);
2215 path[i + GIT_SHA1_HEXSZ - j] = '\0';
2218 static uintmax_t do_change_note_fanout(
2219 struct tree_entry *orig_root, struct tree_entry *root,
2220 char *hex_oid, unsigned int hex_oid_len,
2221 char *fullpath, unsigned int fullpath_len,
2222 unsigned char fanout)
2224 struct tree_content *t;
2225 struct tree_entry *e, leaf;
2226 unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2227 uintmax_t num_notes = 0;
2228 struct object_id oid;
2229 char realpath[60];
2231 if (!root->tree)
2232 load_tree(root);
2233 t = root->tree;
2235 for (i = 0; t && i < t->entry_count; i++) {
2236 e = t->entries[i];
2237 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2238 tmp_fullpath_len = fullpath_len;
2241 * We're interested in EITHER existing note entries (entries
2242 * with exactly 40 hex chars in path, not including directory
2243 * separators), OR directory entries that may contain note
2244 * entries (with < 40 hex chars in path).
2245 * Also, each path component in a note entry must be a multiple
2246 * of 2 chars.
2248 if (!e->versions[1].mode ||
2249 tmp_hex_oid_len > GIT_SHA1_HEXSZ ||
2250 e->name->str_len % 2)
2251 continue;
2253 /* This _may_ be a note entry, or a subdir containing notes */
2254 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2255 e->name->str_len);
2256 if (tmp_fullpath_len)
2257 fullpath[tmp_fullpath_len++] = '/';
2258 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2259 e->name->str_len);
2260 tmp_fullpath_len += e->name->str_len;
2261 fullpath[tmp_fullpath_len] = '\0';
2263 if (tmp_hex_oid_len == GIT_SHA1_HEXSZ && !get_oid_hex(hex_oid, &oid)) {
2264 /* This is a note entry */
2265 if (fanout == 0xff) {
2266 /* Counting mode, no rename */
2267 num_notes++;
2268 continue;
2270 construct_path_with_fanout(hex_oid, fanout, realpath);
2271 if (!strcmp(fullpath, realpath)) {
2272 /* Note entry is in correct location */
2273 num_notes++;
2274 continue;
2277 /* Rename fullpath to realpath */
2278 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2279 die("Failed to remove path %s", fullpath);
2280 tree_content_set(orig_root, realpath,
2281 &leaf.versions[1].oid,
2282 leaf.versions[1].mode,
2283 leaf.tree);
2284 } else if (S_ISDIR(e->versions[1].mode)) {
2285 /* This is a subdir that may contain note entries */
2286 num_notes += do_change_note_fanout(orig_root, e,
2287 hex_oid, tmp_hex_oid_len,
2288 fullpath, tmp_fullpath_len, fanout);
2291 /* The above may have reallocated the current tree_content */
2292 t = root->tree;
2294 return num_notes;
2297 static uintmax_t change_note_fanout(struct tree_entry *root,
2298 unsigned char fanout)
2301 * The size of path is due to one slash between every two hex digits,
2302 * plus the terminating NUL. Note that there is no slash at the end, so
2303 * the number of slashes is one less than half the number of hex
2304 * characters.
2306 char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2307 return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2311 * Given a pointer into a string, parse a mark reference:
2313 * idnum ::= ':' bigint;
2315 * Return the first character after the value in *endptr.
2317 * Complain if the following character is not what is expected,
2318 * either a space or end of the string.
2320 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2322 uintmax_t mark;
2324 assert(*p == ':');
2325 p++;
2326 mark = strtoumax(p, endptr, 10);
2327 if (*endptr == p)
2328 die("No value after ':' in mark: %s", command_buf.buf);
2329 return mark;
2333 * Parse the mark reference, and complain if this is not the end of
2334 * the string.
2336 static uintmax_t parse_mark_ref_eol(const char *p)
2338 char *end;
2339 uintmax_t mark;
2341 mark = parse_mark_ref(p, &end);
2342 if (*end != '\0')
2343 die("Garbage after mark: %s", command_buf.buf);
2344 return mark;
2348 * Parse the mark reference, demanding a trailing space. Return a
2349 * pointer to the space.
2351 static uintmax_t parse_mark_ref_space(const char **p)
2353 uintmax_t mark;
2354 char *end;
2356 mark = parse_mark_ref(*p, &end);
2357 if (*end++ != ' ')
2358 die("Missing space after mark: %s", command_buf.buf);
2359 *p = end;
2360 return mark;
2363 static void file_change_m(const char *p, struct branch *b)
2365 static struct strbuf uq = STRBUF_INIT;
2366 const char *endp;
2367 struct object_entry *oe;
2368 struct object_id oid;
2369 uint16_t mode, inline_data = 0;
2371 p = get_mode(p, &mode);
2372 if (!p)
2373 die("Corrupt mode: %s", command_buf.buf);
2374 switch (mode) {
2375 case 0644:
2376 case 0755:
2377 mode |= S_IFREG;
2378 case S_IFREG | 0644:
2379 case S_IFREG | 0755:
2380 case S_IFLNK:
2381 case S_IFDIR:
2382 case S_IFGITLINK:
2383 /* ok */
2384 break;
2385 default:
2386 die("Corrupt mode: %s", command_buf.buf);
2389 if (*p == ':') {
2390 oe = find_mark(parse_mark_ref_space(&p));
2391 oidcpy(&oid, &oe->idx.oid);
2392 } else if (skip_prefix(p, "inline ", &p)) {
2393 inline_data = 1;
2394 oe = NULL; /* not used with inline_data, but makes gcc happy */
2395 } else {
2396 if (parse_oid_hex(p, &oid, &p))
2397 die("Invalid dataref: %s", command_buf.buf);
2398 oe = find_object(&oid);
2399 if (*p++ != ' ')
2400 die("Missing space after SHA1: %s", command_buf.buf);
2403 strbuf_reset(&uq);
2404 if (!unquote_c_style(&uq, p, &endp)) {
2405 if (*endp)
2406 die("Garbage after path in: %s", command_buf.buf);
2407 p = uq.buf;
2410 /* Git does not track empty, non-toplevel directories. */
2411 if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2412 tree_content_remove(&b->branch_tree, p, NULL, 0);
2413 return;
2416 if (S_ISGITLINK(mode)) {
2417 if (inline_data)
2418 die("Git links cannot be specified 'inline': %s",
2419 command_buf.buf);
2420 else if (oe) {
2421 if (oe->type != OBJ_COMMIT)
2422 die("Not a commit (actually a %s): %s",
2423 typename(oe->type), command_buf.buf);
2426 * Accept the sha1 without checking; it expected to be in
2427 * another repository.
2429 } else if (inline_data) {
2430 if (S_ISDIR(mode))
2431 die("Directories cannot be specified 'inline': %s",
2432 command_buf.buf);
2433 if (p != uq.buf) {
2434 strbuf_addstr(&uq, p);
2435 p = uq.buf;
2437 read_next_command();
2438 parse_and_store_blob(&last_blob, &oid, 0);
2439 } else {
2440 enum object_type expected = S_ISDIR(mode) ?
2441 OBJ_TREE: OBJ_BLOB;
2442 enum object_type type = oe ? oe->type :
2443 sha1_object_info(oid.hash, NULL);
2444 if (type < 0)
2445 die("%s not found: %s",
2446 S_ISDIR(mode) ? "Tree" : "Blob",
2447 command_buf.buf);
2448 if (type != expected)
2449 die("Not a %s (actually a %s): %s",
2450 typename(expected), typename(type),
2451 command_buf.buf);
2454 if (!*p) {
2455 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2456 return;
2458 tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2461 static void file_change_d(const char *p, struct branch *b)
2463 static struct strbuf uq = STRBUF_INIT;
2464 const char *endp;
2466 strbuf_reset(&uq);
2467 if (!unquote_c_style(&uq, p, &endp)) {
2468 if (*endp)
2469 die("Garbage after path in: %s", command_buf.buf);
2470 p = uq.buf;
2472 tree_content_remove(&b->branch_tree, p, NULL, 1);
2475 static void file_change_cr(const char *s, struct branch *b, int rename)
2477 const char *d;
2478 static struct strbuf s_uq = STRBUF_INIT;
2479 static struct strbuf d_uq = STRBUF_INIT;
2480 const char *endp;
2481 struct tree_entry leaf;
2483 strbuf_reset(&s_uq);
2484 if (!unquote_c_style(&s_uq, s, &endp)) {
2485 if (*endp != ' ')
2486 die("Missing space after source: %s", command_buf.buf);
2487 } else {
2488 endp = strchr(s, ' ');
2489 if (!endp)
2490 die("Missing space after source: %s", command_buf.buf);
2491 strbuf_add(&s_uq, s, endp - s);
2493 s = s_uq.buf;
2495 endp++;
2496 if (!*endp)
2497 die("Missing dest: %s", command_buf.buf);
2499 d = endp;
2500 strbuf_reset(&d_uq);
2501 if (!unquote_c_style(&d_uq, d, &endp)) {
2502 if (*endp)
2503 die("Garbage after dest in: %s", command_buf.buf);
2504 d = d_uq.buf;
2507 memset(&leaf, 0, sizeof(leaf));
2508 if (rename)
2509 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2510 else
2511 tree_content_get(&b->branch_tree, s, &leaf, 1);
2512 if (!leaf.versions[1].mode)
2513 die("Path %s not in branch", s);
2514 if (!*d) { /* C "path/to/subdir" "" */
2515 tree_content_replace(&b->branch_tree,
2516 &leaf.versions[1].oid,
2517 leaf.versions[1].mode,
2518 leaf.tree);
2519 return;
2521 tree_content_set(&b->branch_tree, d,
2522 &leaf.versions[1].oid,
2523 leaf.versions[1].mode,
2524 leaf.tree);
2527 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2529 static struct strbuf uq = STRBUF_INIT;
2530 struct object_entry *oe;
2531 struct branch *s;
2532 struct object_id oid, commit_oid;
2533 char path[60];
2534 uint16_t inline_data = 0;
2535 unsigned char new_fanout;
2538 * When loading a branch, we don't traverse its tree to count the real
2539 * number of notes (too expensive to do this for all non-note refs).
2540 * This means that recently loaded notes refs might incorrectly have
2541 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2543 * Fix this by traversing the tree and counting the number of notes
2544 * when b->num_notes == 0. If the notes tree is truly empty, the
2545 * calculation should not take long.
2547 if (b->num_notes == 0 && *old_fanout == 0) {
2548 /* Invoke change_note_fanout() in "counting mode". */
2549 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2550 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2553 /* Now parse the notemodify command. */
2554 /* <dataref> or 'inline' */
2555 if (*p == ':') {
2556 oe = find_mark(parse_mark_ref_space(&p));
2557 oidcpy(&oid, &oe->idx.oid);
2558 } else if (skip_prefix(p, "inline ", &p)) {
2559 inline_data = 1;
2560 oe = NULL; /* not used with inline_data, but makes gcc happy */
2561 } else {
2562 if (parse_oid_hex(p, &oid, &p))
2563 die("Invalid dataref: %s", command_buf.buf);
2564 oe = find_object(&oid);
2565 if (*p++ != ' ')
2566 die("Missing space after SHA1: %s", command_buf.buf);
2569 /* <commit-ish> */
2570 s = lookup_branch(p);
2571 if (s) {
2572 if (is_null_oid(&s->oid))
2573 die("Can't add a note on empty branch.");
2574 oidcpy(&commit_oid, &s->oid);
2575 } else if (*p == ':') {
2576 uintmax_t commit_mark = parse_mark_ref_eol(p);
2577 struct object_entry *commit_oe = find_mark(commit_mark);
2578 if (commit_oe->type != OBJ_COMMIT)
2579 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2580 oidcpy(&commit_oid, &commit_oe->idx.oid);
2581 } else if (!get_oid(p, &commit_oid)) {
2582 unsigned long size;
2583 char *buf = read_object_with_reference(commit_oid.hash,
2584 commit_type, &size, commit_oid.hash);
2585 if (!buf || size < 46)
2586 die("Not a valid commit: %s", p);
2587 free(buf);
2588 } else
2589 die("Invalid ref name or SHA1 expression: %s", p);
2591 if (inline_data) {
2592 if (p != uq.buf) {
2593 strbuf_addstr(&uq, p);
2594 p = uq.buf;
2596 read_next_command();
2597 parse_and_store_blob(&last_blob, &oid, 0);
2598 } else if (oe) {
2599 if (oe->type != OBJ_BLOB)
2600 die("Not a blob (actually a %s): %s",
2601 typename(oe->type), command_buf.buf);
2602 } else if (!is_null_oid(&oid)) {
2603 enum object_type type = sha1_object_info(oid.hash, NULL);
2604 if (type < 0)
2605 die("Blob not found: %s", command_buf.buf);
2606 if (type != OBJ_BLOB)
2607 die("Not a blob (actually a %s): %s",
2608 typename(type), command_buf.buf);
2611 construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2612 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2613 b->num_notes--;
2615 if (is_null_oid(&oid))
2616 return; /* nothing to insert */
2618 b->num_notes++;
2619 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2620 construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2621 tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2624 static void file_change_deleteall(struct branch *b)
2626 release_tree_content_recursive(b->branch_tree.tree);
2627 oidclr(&b->branch_tree.versions[0].oid);
2628 oidclr(&b->branch_tree.versions[1].oid);
2629 load_tree(&b->branch_tree);
2630 b->num_notes = 0;
2633 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2635 if (!buf || size < GIT_SHA1_HEXSZ + 6)
2636 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2637 if (memcmp("tree ", buf, 5)
2638 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2639 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2640 oidcpy(&b->branch_tree.versions[0].oid,
2641 &b->branch_tree.versions[1].oid);
2644 static void parse_from_existing(struct branch *b)
2646 if (is_null_oid(&b->oid)) {
2647 oidclr(&b->branch_tree.versions[0].oid);
2648 oidclr(&b->branch_tree.versions[1].oid);
2649 } else {
2650 unsigned long size;
2651 char *buf;
2653 buf = read_object_with_reference(b->oid.hash,
2654 commit_type, &size,
2655 b->oid.hash);
2656 parse_from_commit(b, buf, size);
2657 free(buf);
2661 static int parse_from(struct branch *b)
2663 const char *from;
2664 struct branch *s;
2665 struct object_id oid;
2667 if (!skip_prefix(command_buf.buf, "from ", &from))
2668 return 0;
2670 oidcpy(&oid, &b->branch_tree.versions[1].oid);
2672 s = lookup_branch(from);
2673 if (b == s)
2674 die("Can't create a branch from itself: %s", b->name);
2675 else if (s) {
2676 struct object_id *t = &s->branch_tree.versions[1].oid;
2677 oidcpy(&b->oid, &s->oid);
2678 oidcpy(&b->branch_tree.versions[0].oid, t);
2679 oidcpy(&b->branch_tree.versions[1].oid, t);
2680 } else if (*from == ':') {
2681 uintmax_t idnum = parse_mark_ref_eol(from);
2682 struct object_entry *oe = find_mark(idnum);
2683 if (oe->type != OBJ_COMMIT)
2684 die("Mark :%" PRIuMAX " not a commit", idnum);
2685 if (oidcmp(&b->oid, &oe->idx.oid)) {
2686 oidcpy(&b->oid, &oe->idx.oid);
2687 if (oe->pack_id != MAX_PACK_ID) {
2688 unsigned long size;
2689 char *buf = gfi_unpack_entry(oe, &size);
2690 parse_from_commit(b, buf, size);
2691 free(buf);
2692 } else
2693 parse_from_existing(b);
2695 } else if (!get_oid(from, &b->oid)) {
2696 parse_from_existing(b);
2697 if (is_null_oid(&b->oid))
2698 b->delete = 1;
2700 else
2701 die("Invalid ref name or SHA1 expression: %s", from);
2703 if (b->branch_tree.tree && oidcmp(&oid, &b->branch_tree.versions[1].oid)) {
2704 release_tree_content_recursive(b->branch_tree.tree);
2705 b->branch_tree.tree = NULL;
2708 read_next_command();
2709 return 1;
2712 static struct hash_list *parse_merge(unsigned int *count)
2714 struct hash_list *list = NULL, **tail = &list, *n;
2715 const char *from;
2716 struct branch *s;
2718 *count = 0;
2719 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2720 n = xmalloc(sizeof(*n));
2721 s = lookup_branch(from);
2722 if (s)
2723 oidcpy(&n->oid, &s->oid);
2724 else if (*from == ':') {
2725 uintmax_t idnum = parse_mark_ref_eol(from);
2726 struct object_entry *oe = find_mark(idnum);
2727 if (oe->type != OBJ_COMMIT)
2728 die("Mark :%" PRIuMAX " not a commit", idnum);
2729 oidcpy(&n->oid, &oe->idx.oid);
2730 } else if (!get_oid(from, &n->oid)) {
2731 unsigned long size;
2732 char *buf = read_object_with_reference(n->oid.hash,
2733 commit_type, &size, n->oid.hash);
2734 if (!buf || size < 46)
2735 die("Not a valid commit: %s", from);
2736 free(buf);
2737 } else
2738 die("Invalid ref name or SHA1 expression: %s", from);
2740 n->next = NULL;
2741 *tail = n;
2742 tail = &n->next;
2744 (*count)++;
2745 read_next_command();
2747 return list;
2750 static void parse_new_commit(const char *arg)
2752 static struct strbuf msg = STRBUF_INIT;
2753 struct branch *b;
2754 char *author = NULL;
2755 char *committer = NULL;
2756 struct hash_list *merge_list = NULL;
2757 unsigned int merge_count;
2758 unsigned char prev_fanout, new_fanout;
2759 const char *v;
2761 b = lookup_branch(arg);
2762 if (!b)
2763 b = new_branch(arg);
2765 read_next_command();
2766 parse_mark();
2767 if (skip_prefix(command_buf.buf, "author ", &v)) {
2768 author = parse_ident(v);
2769 read_next_command();
2771 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2772 committer = parse_ident(v);
2773 read_next_command();
2775 if (!committer)
2776 die("Expected committer but didn't get one");
2777 parse_data(&msg, 0, NULL);
2778 read_next_command();
2779 parse_from(b);
2780 merge_list = parse_merge(&merge_count);
2782 /* ensure the branch is active/loaded */
2783 if (!b->branch_tree.tree || !max_active_branches) {
2784 unload_one_branch();
2785 load_branch(b);
2788 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2790 /* file_change* */
2791 while (command_buf.len > 0) {
2792 if (skip_prefix(command_buf.buf, "M ", &v))
2793 file_change_m(v, b);
2794 else if (skip_prefix(command_buf.buf, "D ", &v))
2795 file_change_d(v, b);
2796 else if (skip_prefix(command_buf.buf, "R ", &v))
2797 file_change_cr(v, b, 1);
2798 else if (skip_prefix(command_buf.buf, "C ", &v))
2799 file_change_cr(v, b, 0);
2800 else if (skip_prefix(command_buf.buf, "N ", &v))
2801 note_change_n(v, b, &prev_fanout);
2802 else if (!strcmp("deleteall", command_buf.buf))
2803 file_change_deleteall(b);
2804 else if (skip_prefix(command_buf.buf, "ls ", &v))
2805 parse_ls(v, b);
2806 else {
2807 unread_command_buf = 1;
2808 break;
2810 if (read_next_command() == EOF)
2811 break;
2814 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2815 if (new_fanout != prev_fanout)
2816 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2818 /* build the tree and the commit */
2819 store_tree(&b->branch_tree);
2820 oidcpy(&b->branch_tree.versions[0].oid,
2821 &b->branch_tree.versions[1].oid);
2823 strbuf_reset(&new_data);
2824 strbuf_addf(&new_data, "tree %s\n",
2825 oid_to_hex(&b->branch_tree.versions[1].oid));
2826 if (!is_null_oid(&b->oid))
2827 strbuf_addf(&new_data, "parent %s\n",
2828 oid_to_hex(&b->oid));
2829 while (merge_list) {
2830 struct hash_list *next = merge_list->next;
2831 strbuf_addf(&new_data, "parent %s\n",
2832 oid_to_hex(&merge_list->oid));
2833 free(merge_list);
2834 merge_list = next;
2836 strbuf_addf(&new_data,
2837 "author %s\n"
2838 "committer %s\n"
2839 "\n",
2840 author ? author : committer, committer);
2841 strbuf_addbuf(&new_data, &msg);
2842 free(author);
2843 free(committer);
2845 if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2846 b->pack_id = pack_id;
2847 b->last_commit = object_count_by_type[OBJ_COMMIT];
2850 static void parse_new_tag(const char *arg)
2852 static struct strbuf msg = STRBUF_INIT;
2853 const char *from;
2854 char *tagger;
2855 struct branch *s;
2856 struct tag *t;
2857 uintmax_t from_mark = 0;
2858 struct object_id oid;
2859 enum object_type type;
2860 const char *v;
2862 t = pool_alloc(sizeof(struct tag));
2863 memset(t, 0, sizeof(struct tag));
2864 t->name = pool_strdup(arg);
2865 if (last_tag)
2866 last_tag->next_tag = t;
2867 else
2868 first_tag = t;
2869 last_tag = t;
2870 read_next_command();
2872 /* from ... */
2873 if (!skip_prefix(command_buf.buf, "from ", &from))
2874 die("Expected from command, got %s", command_buf.buf);
2875 s = lookup_branch(from);
2876 if (s) {
2877 if (is_null_oid(&s->oid))
2878 die("Can't tag an empty branch.");
2879 oidcpy(&oid, &s->oid);
2880 type = OBJ_COMMIT;
2881 } else if (*from == ':') {
2882 struct object_entry *oe;
2883 from_mark = parse_mark_ref_eol(from);
2884 oe = find_mark(from_mark);
2885 type = oe->type;
2886 oidcpy(&oid, &oe->idx.oid);
2887 } else if (!get_oid(from, &oid)) {
2888 struct object_entry *oe = find_object(&oid);
2889 if (!oe) {
2890 type = sha1_object_info(oid.hash, NULL);
2891 if (type < 0)
2892 die("Not a valid object: %s", from);
2893 } else
2894 type = oe->type;
2895 } else
2896 die("Invalid ref name or SHA1 expression: %s", from);
2897 read_next_command();
2899 /* tagger ... */
2900 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2901 tagger = parse_ident(v);
2902 read_next_command();
2903 } else
2904 tagger = NULL;
2906 /* tag payload/message */
2907 parse_data(&msg, 0, NULL);
2909 /* build the tag object */
2910 strbuf_reset(&new_data);
2912 strbuf_addf(&new_data,
2913 "object %s\n"
2914 "type %s\n"
2915 "tag %s\n",
2916 oid_to_hex(&oid), typename(type), t->name);
2917 if (tagger)
2918 strbuf_addf(&new_data,
2919 "tagger %s\n", tagger);
2920 strbuf_addch(&new_data, '\n');
2921 strbuf_addbuf(&new_data, &msg);
2922 free(tagger);
2924 if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, 0))
2925 t->pack_id = MAX_PACK_ID;
2926 else
2927 t->pack_id = pack_id;
2930 static void parse_reset_branch(const char *arg)
2932 struct branch *b;
2934 b = lookup_branch(arg);
2935 if (b) {
2936 oidclr(&b->oid);
2937 oidclr(&b->branch_tree.versions[0].oid);
2938 oidclr(&b->branch_tree.versions[1].oid);
2939 if (b->branch_tree.tree) {
2940 release_tree_content_recursive(b->branch_tree.tree);
2941 b->branch_tree.tree = NULL;
2944 else
2945 b = new_branch(arg);
2946 read_next_command();
2947 parse_from(b);
2948 if (command_buf.len > 0)
2949 unread_command_buf = 1;
2952 static void cat_blob_write(const char *buf, unsigned long size)
2954 if (write_in_full(cat_blob_fd, buf, size) != size)
2955 die_errno("Write to frontend failed");
2958 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2960 struct strbuf line = STRBUF_INIT;
2961 unsigned long size;
2962 enum object_type type = 0;
2963 char *buf;
2965 if (!oe || oe->pack_id == MAX_PACK_ID) {
2966 buf = read_sha1_file(oid->hash, &type, &size);
2967 } else {
2968 type = oe->type;
2969 buf = gfi_unpack_entry(oe, &size);
2973 * Output based on batch_one_object() from cat-file.c.
2975 if (type <= 0) {
2976 strbuf_reset(&line);
2977 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2978 cat_blob_write(line.buf, line.len);
2979 strbuf_release(&line);
2980 free(buf);
2981 return;
2983 if (!buf)
2984 die("Can't read object %s", oid_to_hex(oid));
2985 if (type != OBJ_BLOB)
2986 die("Object %s is a %s but a blob was expected.",
2987 oid_to_hex(oid), typename(type));
2988 strbuf_reset(&line);
2989 strbuf_addf(&line, "%s %s %lu\n", oid_to_hex(oid),
2990 typename(type), size);
2991 cat_blob_write(line.buf, line.len);
2992 strbuf_release(&line);
2993 cat_blob_write(buf, size);
2994 cat_blob_write("\n", 1);
2995 if (oe && oe->pack_id == pack_id) {
2996 last_blob.offset = oe->idx.offset;
2997 strbuf_attach(&last_blob.data, buf, size, size);
2998 last_blob.depth = oe->depth;
2999 } else
3000 free(buf);
3003 static void parse_get_mark(const char *p)
3005 struct object_entry *oe = oe;
3006 char output[GIT_MAX_HEXSZ + 2];
3008 /* get-mark SP <object> LF */
3009 if (*p != ':')
3010 die("Not a mark: %s", p);
3012 oe = find_mark(parse_mark_ref_eol(p));
3013 if (!oe)
3014 die("Unknown mark: %s", command_buf.buf);
3016 xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
3017 cat_blob_write(output, GIT_SHA1_HEXSZ + 1);
3020 static void parse_cat_blob(const char *p)
3022 struct object_entry *oe = oe;
3023 struct object_id oid;
3025 /* cat-blob SP <object> LF */
3026 if (*p == ':') {
3027 oe = find_mark(parse_mark_ref_eol(p));
3028 if (!oe)
3029 die("Unknown mark: %s", command_buf.buf);
3030 oidcpy(&oid, &oe->idx.oid);
3031 } else {
3032 if (parse_oid_hex(p, &oid, &p))
3033 die("Invalid dataref: %s", command_buf.buf);
3034 if (*p)
3035 die("Garbage after SHA1: %s", command_buf.buf);
3036 oe = find_object(&oid);
3039 cat_blob(oe, &oid);
3042 static struct object_entry *dereference(struct object_entry *oe,
3043 struct object_id *oid)
3045 unsigned long size;
3046 char *buf = NULL;
3047 if (!oe) {
3048 enum object_type type = sha1_object_info(oid->hash, NULL);
3049 if (type < 0)
3050 die("object not found: %s", oid_to_hex(oid));
3051 /* cache it! */
3052 oe = insert_object(oid);
3053 oe->type = type;
3054 oe->pack_id = MAX_PACK_ID;
3055 oe->idx.offset = 1;
3057 switch (oe->type) {
3058 case OBJ_TREE: /* easy case. */
3059 return oe;
3060 case OBJ_COMMIT:
3061 case OBJ_TAG:
3062 break;
3063 default:
3064 die("Not a tree-ish: %s", command_buf.buf);
3067 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3068 buf = gfi_unpack_entry(oe, &size);
3069 } else {
3070 enum object_type unused;
3071 buf = read_sha1_file(oid->hash, &unused, &size);
3073 if (!buf)
3074 die("Can't load object %s", oid_to_hex(oid));
3076 /* Peel one layer. */
3077 switch (oe->type) {
3078 case OBJ_TAG:
3079 if (size < GIT_SHA1_HEXSZ + strlen("object ") ||
3080 get_oid_hex(buf + strlen("object "), oid))
3081 die("Invalid SHA1 in tag: %s", command_buf.buf);
3082 break;
3083 case OBJ_COMMIT:
3084 if (size < GIT_SHA1_HEXSZ + strlen("tree ") ||
3085 get_oid_hex(buf + strlen("tree "), oid))
3086 die("Invalid SHA1 in commit: %s", command_buf.buf);
3089 free(buf);
3090 return find_object(oid);
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(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_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 *sha1, 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, sha1_to_hex(sha1));
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();
3191 dump_branches();
3192 dump_tags();
3193 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 char* make_fast_import_path(const char *path)
3213 if (!relative_marks_paths || is_absolute_path(path))
3214 return xstrdup(path);
3215 return git_pathdup("info/fast-import/%s", path);
3218 static void option_import_marks(const char *marks,
3219 int from_stream, int ignore_missing)
3221 if (import_marks_file) {
3222 if (from_stream)
3223 die("Only one import-marks command allowed per stream");
3225 /* read previous mark file */
3226 if(!import_marks_file_from_stream)
3227 read_marks();
3230 import_marks_file = make_fast_import_path(marks);
3231 safe_create_leading_directories_const(import_marks_file);
3232 import_marks_file_from_stream = from_stream;
3233 import_marks_file_ignore_missing = ignore_missing;
3236 static void option_date_format(const char *fmt)
3238 if (!strcmp(fmt, "raw"))
3239 whenspec = WHENSPEC_RAW;
3240 else if (!strcmp(fmt, "rfc2822"))
3241 whenspec = WHENSPEC_RFC2822;
3242 else if (!strcmp(fmt, "now"))
3243 whenspec = WHENSPEC_NOW;
3244 else
3245 die("unknown --date-format argument %s", fmt);
3248 static unsigned long ulong_arg(const char *option, const char *arg)
3250 char *endptr;
3251 unsigned long rv = strtoul(arg, &endptr, 0);
3252 if (strchr(arg, '-') || endptr == arg || *endptr)
3253 die("%s: argument must be a non-negative integer", option);
3254 return rv;
3257 static void option_depth(const char *depth)
3259 max_depth = ulong_arg("--depth", depth);
3260 if (max_depth > MAX_DEPTH)
3261 die("--depth cannot exceed %u", MAX_DEPTH);
3264 static void option_active_branches(const char *branches)
3266 max_active_branches = ulong_arg("--active-branches", branches);
3269 static void option_export_marks(const char *marks)
3271 export_marks_file = make_fast_import_path(marks);
3272 safe_create_leading_directories_const(export_marks_file);
3275 static void option_cat_blob_fd(const char *fd)
3277 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3278 if (n > (unsigned long) INT_MAX)
3279 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3280 cat_blob_fd = (int) n;
3283 static void option_export_pack_edges(const char *edges)
3285 if (pack_edges)
3286 fclose(pack_edges);
3287 pack_edges = xfopen(edges, "a");
3290 static int parse_one_option(const char *option)
3292 if (skip_prefix(option, "max-pack-size=", &option)) {
3293 unsigned long v;
3294 if (!git_parse_ulong(option, &v))
3295 return 0;
3296 if (v < 8192) {
3297 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3298 v *= 1024 * 1024;
3299 } else if (v < 1024 * 1024) {
3300 warning("minimum max-pack-size is 1 MiB");
3301 v = 1024 * 1024;
3303 max_packsize = v;
3304 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3305 unsigned long v;
3306 if (!git_parse_ulong(option, &v))
3307 return 0;
3308 big_file_threshold = v;
3309 } else if (skip_prefix(option, "depth=", &option)) {
3310 option_depth(option);
3311 } else if (skip_prefix(option, "active-branches=", &option)) {
3312 option_active_branches(option);
3313 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3314 option_export_pack_edges(option);
3315 } else if (starts_with(option, "quiet")) {
3316 show_stats = 0;
3317 } else if (starts_with(option, "stats")) {
3318 show_stats = 1;
3319 } else {
3320 return 0;
3323 return 1;
3326 static int parse_one_feature(const char *feature, int from_stream)
3328 const char *arg;
3330 if (skip_prefix(feature, "date-format=", &arg)) {
3331 option_date_format(arg);
3332 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3333 option_import_marks(arg, from_stream, 0);
3334 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3335 option_import_marks(arg, from_stream, 1);
3336 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3337 option_export_marks(arg);
3338 } else if (!strcmp(feature, "get-mark")) {
3339 ; /* Don't die - this feature is supported */
3340 } else if (!strcmp(feature, "cat-blob")) {
3341 ; /* Don't die - this feature is supported */
3342 } else if (!strcmp(feature, "relative-marks")) {
3343 relative_marks_paths = 1;
3344 } else if (!strcmp(feature, "no-relative-marks")) {
3345 relative_marks_paths = 0;
3346 } else if (!strcmp(feature, "done")) {
3347 require_explicit_termination = 1;
3348 } else if (!strcmp(feature, "force")) {
3349 force_update = 1;
3350 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3351 ; /* do nothing; we have the feature */
3352 } else {
3353 return 0;
3356 return 1;
3359 static void parse_feature(const char *feature)
3361 if (seen_data_command)
3362 die("Got feature command '%s' after data command", feature);
3364 if (parse_one_feature(feature, 1))
3365 return;
3367 die("This version of fast-import does not support feature %s.", feature);
3370 static void parse_option(const char *option)
3372 if (seen_data_command)
3373 die("Got option command '%s' after data command", option);
3375 if (parse_one_option(option))
3376 return;
3378 die("This version of fast-import does not support option: %s", option);
3381 static void git_pack_config(void)
3383 int indexversion_value;
3384 int limit;
3385 unsigned long packsizelimit_value;
3387 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3388 if (max_depth > MAX_DEPTH)
3389 max_depth = MAX_DEPTH;
3391 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3392 pack_idx_opts.version = indexversion_value;
3393 if (pack_idx_opts.version > 2)
3394 git_die_config("pack.indexversion",
3395 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3397 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3398 max_packsize = packsizelimit_value;
3400 if (!git_config_get_int("fastimport.unpacklimit", &limit))
3401 unpack_limit = limit;
3402 else if (!git_config_get_int("transfer.unpacklimit", &limit))
3403 unpack_limit = limit;
3405 git_config(git_default_config, NULL);
3408 static const char fast_import_usage[] =
3409 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3411 static void parse_argv(void)
3413 unsigned int i;
3415 for (i = 1; i < global_argc; i++) {
3416 const char *a = global_argv[i];
3418 if (*a != '-' || !strcmp(a, "--"))
3419 break;
3421 if (!skip_prefix(a, "--", &a))
3422 die("unknown option %s", a);
3424 if (parse_one_option(a))
3425 continue;
3427 if (parse_one_feature(a, 0))
3428 continue;
3430 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3431 option_cat_blob_fd(a);
3432 continue;
3435 die("unknown option --%s", a);
3437 if (i != global_argc)
3438 usage(fast_import_usage);
3440 seen_data_command = 1;
3441 if (import_marks_file)
3442 read_marks();
3445 int cmd_main(int argc, const char **argv)
3447 unsigned int i;
3449 if (argc == 2 && !strcmp(argv[1], "-h"))
3450 usage(fast_import_usage);
3452 setup_git_directory();
3453 reset_pack_idx_option(&pack_idx_opts);
3454 git_pack_config();
3456 alloc_objects(object_entry_alloc);
3457 strbuf_init(&command_buf, 0);
3458 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3459 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3460 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3461 marks = pool_calloc(1, sizeof(struct mark_set));
3463 global_argc = argc;
3464 global_argv = argv;
3466 rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3467 for (i = 0; i < (cmd_save - 1); i++)
3468 rc_free[i].next = &rc_free[i + 1];
3469 rc_free[cmd_save - 1].next = NULL;
3471 prepare_packed_git();
3472 start_packfile();
3473 set_die_routine(die_nicely);
3474 set_checkpoint_signal();
3475 while (read_next_command() != EOF) {
3476 const char *v;
3477 if (!strcmp("blob", command_buf.buf))
3478 parse_new_blob();
3479 else if (skip_prefix(command_buf.buf, "ls ", &v))
3480 parse_ls(v, NULL);
3481 else if (skip_prefix(command_buf.buf, "commit ", &v))
3482 parse_new_commit(v);
3483 else if (skip_prefix(command_buf.buf, "tag ", &v))
3484 parse_new_tag(v);
3485 else if (skip_prefix(command_buf.buf, "reset ", &v))
3486 parse_reset_branch(v);
3487 else if (!strcmp("checkpoint", command_buf.buf))
3488 parse_checkpoint();
3489 else if (!strcmp("done", command_buf.buf))
3490 break;
3491 else if (starts_with(command_buf.buf, "progress "))
3492 parse_progress();
3493 else if (skip_prefix(command_buf.buf, "feature ", &v))
3494 parse_feature(v);
3495 else if (skip_prefix(command_buf.buf, "option git ", &v))
3496 parse_option(v);
3497 else if (starts_with(command_buf.buf, "option "))
3498 /* ignore non-git options*/;
3499 else
3500 die("Unsupported command: %s", command_buf.buf);
3502 if (checkpoint_requested)
3503 checkpoint();
3506 /* argv hasn't been parsed yet, do so */
3507 if (!seen_data_command)
3508 parse_argv();
3510 if (require_explicit_termination && feof(stdin))
3511 die("stream ends early");
3513 end_packfile();
3515 dump_branches();
3516 dump_tags();
3517 unkeep_all_packs();
3518 dump_marks();
3520 if (pack_edges)
3521 fclose(pack_edges);
3523 if (show_stats) {
3524 uintmax_t total_count = 0, duplicate_count = 0;
3525 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3526 total_count += object_count_by_type[i];
3527 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3528 duplicate_count += duplicate_count_by_type[i];
3530 fprintf(stderr, "%s statistics:\n", argv[0]);
3531 fprintf(stderr, "---------------------------------------------------------------------\n");
3532 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3533 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3534 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]);
3535 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]);
3536 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]);
3537 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]);
3538 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3539 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3540 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3541 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3542 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)(total_allocd/1024));
3543 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3544 fprintf(stderr, "---------------------------------------------------------------------\n");
3545 pack_report();
3546 fprintf(stderr, "---------------------------------------------------------------------\n");
3547 fprintf(stderr, "\n");
3550 return failure ? 1 : 0;