doc: don't claim that cherry calls patch-id
[git/raj.git] / fast-import.c
blobf4d9969e5c81f9c1415f0908f090bf62b156e179
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, ls and cat requests may appear anywhere
138 # in the input, except within a data command. Any form
139 # of the data command always escapes the related input
140 # from 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 cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
148 ls_tree ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
150 comment ::= '#' not_lf* lf;
151 not_lf ::= # Any byte that is not ASCII newline (LF);
154 #include "builtin.h"
155 #include "cache.h"
156 #include "object.h"
157 #include "blob.h"
158 #include "tree.h"
159 #include "commit.h"
160 #include "delta.h"
161 #include "pack.h"
162 #include "refs.h"
163 #include "csum-file.h"
164 #include "quote.h"
165 #include "exec_cmd.h"
166 #include "dir.h"
168 #define PACK_ID_BITS 16
169 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
170 #define DEPTH_BITS 13
171 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
174 * We abuse the setuid bit on directories to mean "do not delta".
176 #define NO_DELTA S_ISUID
178 struct object_entry {
179 struct pack_idx_entry idx;
180 struct object_entry *next;
181 uint32_t type : TYPE_BITS,
182 pack_id : PACK_ID_BITS,
183 depth : DEPTH_BITS;
186 struct object_entry_pool {
187 struct object_entry_pool *next_pool;
188 struct object_entry *next_free;
189 struct object_entry *end;
190 struct object_entry entries[FLEX_ARRAY]; /* more */
193 struct mark_set {
194 union {
195 struct object_entry *marked[1024];
196 struct mark_set *sets[1024];
197 } data;
198 unsigned int shift;
201 struct last_object {
202 struct strbuf data;
203 off_t offset;
204 unsigned int depth;
205 unsigned no_swap : 1;
208 struct mem_pool {
209 struct mem_pool *next_pool;
210 char *next_free;
211 char *end;
212 uintmax_t space[FLEX_ARRAY]; /* more */
215 struct atom_str {
216 struct atom_str *next_atom;
217 unsigned short str_len;
218 char str_dat[FLEX_ARRAY]; /* more */
221 struct tree_content;
222 struct tree_entry {
223 struct tree_content *tree;
224 struct atom_str *name;
225 struct tree_entry_ms {
226 uint16_t mode;
227 unsigned char sha1[20];
228 } versions[2];
231 struct tree_content {
232 unsigned int entry_capacity; /* must match avail_tree_content */
233 unsigned int entry_count;
234 unsigned int delta_depth;
235 struct tree_entry *entries[FLEX_ARRAY]; /* more */
238 struct avail_tree_content {
239 unsigned int entry_capacity; /* must match tree_content */
240 struct avail_tree_content *next_avail;
243 struct branch {
244 struct branch *table_next_branch;
245 struct branch *active_next_branch;
246 const char *name;
247 struct tree_entry branch_tree;
248 uintmax_t last_commit;
249 uintmax_t num_notes;
250 unsigned active : 1;
251 unsigned pack_id : PACK_ID_BITS;
252 unsigned char sha1[20];
255 struct tag {
256 struct tag *next_tag;
257 const char *name;
258 unsigned int pack_id;
259 unsigned char sha1[20];
262 struct hash_list {
263 struct hash_list *next;
264 unsigned char sha1[20];
267 typedef enum {
268 WHENSPEC_RAW = 1,
269 WHENSPEC_RFC2822,
270 WHENSPEC_NOW
271 } whenspec_type;
273 struct recent_command {
274 struct recent_command *prev;
275 struct recent_command *next;
276 char *buf;
279 /* Configured limits on output */
280 static unsigned long max_depth = 10;
281 static off_t max_packsize;
282 static int force_update;
283 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
284 static int pack_compression_seen;
286 /* Stats and misc. counters */
287 static uintmax_t alloc_count;
288 static uintmax_t marks_set_count;
289 static uintmax_t object_count_by_type[1 << TYPE_BITS];
290 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
291 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
292 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
293 static unsigned long object_count;
294 static unsigned long branch_count;
295 static unsigned long branch_load_count;
296 static int failure;
297 static FILE *pack_edges;
298 static unsigned int show_stats = 1;
299 static int global_argc;
300 static char **global_argv;
302 /* Memory pools */
303 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
304 static size_t total_allocd;
305 static struct mem_pool *mem_pool;
307 /* Atom management */
308 static unsigned int atom_table_sz = 4451;
309 static unsigned int atom_cnt;
310 static struct atom_str **atom_table;
312 /* The .pack file being generated */
313 static struct pack_idx_option pack_idx_opts;
314 static unsigned int pack_id;
315 static struct sha1file *pack_file;
316 static struct packed_git *pack_data;
317 static struct packed_git **all_packs;
318 static off_t pack_size;
320 /* Table of objects we've written. */
321 static unsigned int object_entry_alloc = 5000;
322 static struct object_entry_pool *blocks;
323 static struct object_entry *object_table[1 << 16];
324 static struct mark_set *marks;
325 static const char *export_marks_file;
326 static const char *import_marks_file;
327 static int import_marks_file_from_stream;
328 static int import_marks_file_ignore_missing;
329 static int relative_marks_paths;
331 /* Our last blob */
332 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
334 /* Tree management */
335 static unsigned int tree_entry_alloc = 1000;
336 static void *avail_tree_entry;
337 static unsigned int avail_tree_table_sz = 100;
338 static struct avail_tree_content **avail_tree_table;
339 static struct strbuf old_tree = STRBUF_INIT;
340 static struct strbuf new_tree = STRBUF_INIT;
342 /* Branch data */
343 static unsigned long max_active_branches = 5;
344 static unsigned long cur_active_branches;
345 static unsigned long branch_table_sz = 1039;
346 static struct branch **branch_table;
347 static struct branch *active_branches;
349 /* Tag data */
350 static struct tag *first_tag;
351 static struct tag *last_tag;
353 /* Input stream parsing */
354 static whenspec_type whenspec = WHENSPEC_RAW;
355 static struct strbuf command_buf = STRBUF_INIT;
356 static int unread_command_buf;
357 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
358 static struct recent_command *cmd_tail = &cmd_hist;
359 static struct recent_command *rc_free;
360 static unsigned int cmd_save = 100;
361 static uintmax_t next_mark;
362 static struct strbuf new_data = STRBUF_INIT;
363 static int seen_data_command;
364 static int require_explicit_termination;
366 /* Signal handling */
367 static volatile sig_atomic_t checkpoint_requested;
369 /* Where to write output of cat-blob commands */
370 static int cat_blob_fd = STDOUT_FILENO;
372 static void parse_argv(void);
373 static void parse_cat_blob(void);
374 static void parse_ls(struct branch *b);
376 static void write_branch_report(FILE *rpt, struct branch *b)
378 fprintf(rpt, "%s:\n", b->name);
380 fprintf(rpt, " status :");
381 if (b->active)
382 fputs(" active", rpt);
383 if (b->branch_tree.tree)
384 fputs(" loaded", rpt);
385 if (is_null_sha1(b->branch_tree.versions[1].sha1))
386 fputs(" dirty", rpt);
387 fputc('\n', rpt);
389 fprintf(rpt, " tip commit : %s\n", sha1_to_hex(b->sha1));
390 fprintf(rpt, " old tree : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
391 fprintf(rpt, " cur tree : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
392 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
394 fputs(" last pack : ", rpt);
395 if (b->pack_id < MAX_PACK_ID)
396 fprintf(rpt, "%u", b->pack_id);
397 fputc('\n', rpt);
399 fputc('\n', rpt);
402 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
404 static void write_crash_report(const char *err)
406 char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
407 FILE *rpt = fopen(loc, "w");
408 struct branch *b;
409 unsigned long lu;
410 struct recent_command *rc;
412 if (!rpt) {
413 error("can't write crash report %s: %s", loc, strerror(errno));
414 return;
417 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
419 fprintf(rpt, "fast-import crash report:\n");
420 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
421 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
422 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
423 fputc('\n', rpt);
425 fputs("fatal: ", rpt);
426 fputs(err, rpt);
427 fputc('\n', rpt);
429 fputc('\n', rpt);
430 fputs("Most Recent Commands Before Crash\n", rpt);
431 fputs("---------------------------------\n", rpt);
432 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
433 if (rc->next == &cmd_hist)
434 fputs("* ", rpt);
435 else
436 fputs(" ", rpt);
437 fputs(rc->buf, rpt);
438 fputc('\n', rpt);
441 fputc('\n', rpt);
442 fputs("Active Branch LRU\n", rpt);
443 fputs("-----------------\n", rpt);
444 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
445 cur_active_branches,
446 max_active_branches);
447 fputc('\n', rpt);
448 fputs(" pos clock name\n", rpt);
449 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
450 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
451 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
452 ++lu, b->last_commit, b->name);
454 fputc('\n', rpt);
455 fputs("Inactive Branches\n", rpt);
456 fputs("-----------------\n", rpt);
457 for (lu = 0; lu < branch_table_sz; lu++) {
458 for (b = branch_table[lu]; b; b = b->table_next_branch)
459 write_branch_report(rpt, b);
462 if (first_tag) {
463 struct tag *tg;
464 fputc('\n', rpt);
465 fputs("Annotated Tags\n", rpt);
466 fputs("--------------\n", rpt);
467 for (tg = first_tag; tg; tg = tg->next_tag) {
468 fputs(sha1_to_hex(tg->sha1), rpt);
469 fputc(' ', rpt);
470 fputs(tg->name, rpt);
471 fputc('\n', rpt);
475 fputc('\n', rpt);
476 fputs("Marks\n", rpt);
477 fputs("-----\n", rpt);
478 if (export_marks_file)
479 fprintf(rpt, " exported to %s\n", export_marks_file);
480 else
481 dump_marks_helper(rpt, 0, marks);
483 fputc('\n', rpt);
484 fputs("-------------------\n", rpt);
485 fputs("END OF CRASH REPORT\n", rpt);
486 fclose(rpt);
489 static void end_packfile(void);
490 static void unkeep_all_packs(void);
491 static void dump_marks(void);
493 static NORETURN void die_nicely(const char *err, va_list params)
495 static int zombie;
496 char message[2 * PATH_MAX];
498 vsnprintf(message, sizeof(message), err, params);
499 fputs("fatal: ", stderr);
500 fputs(message, stderr);
501 fputc('\n', stderr);
503 if (!zombie) {
504 zombie = 1;
505 write_crash_report(message);
506 end_packfile();
507 unkeep_all_packs();
508 dump_marks();
510 exit(128);
513 #ifndef SIGUSR1 /* Windows, for example */
515 static void set_checkpoint_signal(void)
519 #else
521 static void checkpoint_signal(int signo)
523 checkpoint_requested = 1;
526 static void set_checkpoint_signal(void)
528 struct sigaction sa;
530 memset(&sa, 0, sizeof(sa));
531 sa.sa_handler = checkpoint_signal;
532 sigemptyset(&sa.sa_mask);
533 sa.sa_flags = SA_RESTART;
534 sigaction(SIGUSR1, &sa, NULL);
537 #endif
539 static void alloc_objects(unsigned int cnt)
541 struct object_entry_pool *b;
543 b = xmalloc(sizeof(struct object_entry_pool)
544 + cnt * sizeof(struct object_entry));
545 b->next_pool = blocks;
546 b->next_free = b->entries;
547 b->end = b->entries + cnt;
548 blocks = b;
549 alloc_count += cnt;
552 static struct object_entry *new_object(unsigned char *sha1)
554 struct object_entry *e;
556 if (blocks->next_free == blocks->end)
557 alloc_objects(object_entry_alloc);
559 e = blocks->next_free++;
560 hashcpy(e->idx.sha1, sha1);
561 return e;
564 static struct object_entry *find_object(unsigned char *sha1)
566 unsigned int h = sha1[0] << 8 | sha1[1];
567 struct object_entry *e;
568 for (e = object_table[h]; e; e = e->next)
569 if (!hashcmp(sha1, e->idx.sha1))
570 return e;
571 return NULL;
574 static struct object_entry *insert_object(unsigned char *sha1)
576 unsigned int h = sha1[0] << 8 | sha1[1];
577 struct object_entry *e = object_table[h];
579 while (e) {
580 if (!hashcmp(sha1, e->idx.sha1))
581 return e;
582 e = e->next;
585 e = new_object(sha1);
586 e->next = object_table[h];
587 e->idx.offset = 0;
588 object_table[h] = e;
589 return e;
592 static unsigned int hc_str(const char *s, size_t len)
594 unsigned int r = 0;
595 while (len-- > 0)
596 r = r * 31 + *s++;
597 return r;
600 static void *pool_alloc(size_t len)
602 struct mem_pool *p;
603 void *r;
605 /* round up to a 'uintmax_t' alignment */
606 if (len & (sizeof(uintmax_t) - 1))
607 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
609 for (p = mem_pool; p; p = p->next_pool)
610 if ((p->end - p->next_free >= len))
611 break;
613 if (!p) {
614 if (len >= (mem_pool_alloc/2)) {
615 total_allocd += len;
616 return xmalloc(len);
618 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
619 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
620 p->next_pool = mem_pool;
621 p->next_free = (char *) p->space;
622 p->end = p->next_free + mem_pool_alloc;
623 mem_pool = p;
626 r = p->next_free;
627 p->next_free += len;
628 return r;
631 static void *pool_calloc(size_t count, size_t size)
633 size_t len = count * size;
634 void *r = pool_alloc(len);
635 memset(r, 0, len);
636 return r;
639 static char *pool_strdup(const char *s)
641 char *r = pool_alloc(strlen(s) + 1);
642 strcpy(r, s);
643 return r;
646 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
648 struct mark_set *s = marks;
649 while ((idnum >> s->shift) >= 1024) {
650 s = pool_calloc(1, sizeof(struct mark_set));
651 s->shift = marks->shift + 10;
652 s->data.sets[0] = marks;
653 marks = s;
655 while (s->shift) {
656 uintmax_t i = idnum >> s->shift;
657 idnum -= i << s->shift;
658 if (!s->data.sets[i]) {
659 s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
660 s->data.sets[i]->shift = s->shift - 10;
662 s = s->data.sets[i];
664 if (!s->data.marked[idnum])
665 marks_set_count++;
666 s->data.marked[idnum] = oe;
669 static struct object_entry *find_mark(uintmax_t idnum)
671 uintmax_t orig_idnum = idnum;
672 struct mark_set *s = marks;
673 struct object_entry *oe = NULL;
674 if ((idnum >> s->shift) < 1024) {
675 while (s && s->shift) {
676 uintmax_t i = idnum >> s->shift;
677 idnum -= i << s->shift;
678 s = s->data.sets[i];
680 if (s)
681 oe = s->data.marked[idnum];
683 if (!oe)
684 die("mark :%" PRIuMAX " not declared", orig_idnum);
685 return oe;
688 static struct atom_str *to_atom(const char *s, unsigned short len)
690 unsigned int hc = hc_str(s, len) % atom_table_sz;
691 struct atom_str *c;
693 for (c = atom_table[hc]; c; c = c->next_atom)
694 if (c->str_len == len && !strncmp(s, c->str_dat, len))
695 return c;
697 c = pool_alloc(sizeof(struct atom_str) + len + 1);
698 c->str_len = len;
699 strncpy(c->str_dat, s, len);
700 c->str_dat[len] = 0;
701 c->next_atom = atom_table[hc];
702 atom_table[hc] = c;
703 atom_cnt++;
704 return c;
707 static struct branch *lookup_branch(const char *name)
709 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
710 struct branch *b;
712 for (b = branch_table[hc]; b; b = b->table_next_branch)
713 if (!strcmp(name, b->name))
714 return b;
715 return NULL;
718 static struct branch *new_branch(const char *name)
720 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
721 struct branch *b = lookup_branch(name);
723 if (b)
724 die("Invalid attempt to create duplicate branch: %s", name);
725 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
726 die("Branch name doesn't conform to GIT standards: %s", name);
728 b = pool_calloc(1, sizeof(struct branch));
729 b->name = pool_strdup(name);
730 b->table_next_branch = branch_table[hc];
731 b->branch_tree.versions[0].mode = S_IFDIR;
732 b->branch_tree.versions[1].mode = S_IFDIR;
733 b->num_notes = 0;
734 b->active = 0;
735 b->pack_id = MAX_PACK_ID;
736 branch_table[hc] = b;
737 branch_count++;
738 return b;
741 static unsigned int hc_entries(unsigned int cnt)
743 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
744 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
747 static struct tree_content *new_tree_content(unsigned int cnt)
749 struct avail_tree_content *f, *l = NULL;
750 struct tree_content *t;
751 unsigned int hc = hc_entries(cnt);
753 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
754 if (f->entry_capacity >= cnt)
755 break;
757 if (f) {
758 if (l)
759 l->next_avail = f->next_avail;
760 else
761 avail_tree_table[hc] = f->next_avail;
762 } else {
763 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
764 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
765 f->entry_capacity = cnt;
768 t = (struct tree_content*)f;
769 t->entry_count = 0;
770 t->delta_depth = 0;
771 return t;
774 static void release_tree_entry(struct tree_entry *e);
775 static void release_tree_content(struct tree_content *t)
777 struct avail_tree_content *f = (struct avail_tree_content*)t;
778 unsigned int hc = hc_entries(f->entry_capacity);
779 f->next_avail = avail_tree_table[hc];
780 avail_tree_table[hc] = f;
783 static void release_tree_content_recursive(struct tree_content *t)
785 unsigned int i;
786 for (i = 0; i < t->entry_count; i++)
787 release_tree_entry(t->entries[i]);
788 release_tree_content(t);
791 static struct tree_content *grow_tree_content(
792 struct tree_content *t,
793 int amt)
795 struct tree_content *r = new_tree_content(t->entry_count + amt);
796 r->entry_count = t->entry_count;
797 r->delta_depth = t->delta_depth;
798 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
799 release_tree_content(t);
800 return r;
803 static struct tree_entry *new_tree_entry(void)
805 struct tree_entry *e;
807 if (!avail_tree_entry) {
808 unsigned int n = tree_entry_alloc;
809 total_allocd += n * sizeof(struct tree_entry);
810 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
811 while (n-- > 1) {
812 *((void**)e) = e + 1;
813 e++;
815 *((void**)e) = NULL;
818 e = avail_tree_entry;
819 avail_tree_entry = *((void**)e);
820 return e;
823 static void release_tree_entry(struct tree_entry *e)
825 if (e->tree)
826 release_tree_content_recursive(e->tree);
827 *((void**)e) = avail_tree_entry;
828 avail_tree_entry = e;
831 static struct tree_content *dup_tree_content(struct tree_content *s)
833 struct tree_content *d;
834 struct tree_entry *a, *b;
835 unsigned int i;
837 if (!s)
838 return NULL;
839 d = new_tree_content(s->entry_count);
840 for (i = 0; i < s->entry_count; i++) {
841 a = s->entries[i];
842 b = new_tree_entry();
843 memcpy(b, a, sizeof(*a));
844 if (a->tree && is_null_sha1(b->versions[1].sha1))
845 b->tree = dup_tree_content(a->tree);
846 else
847 b->tree = NULL;
848 d->entries[i] = b;
850 d->entry_count = s->entry_count;
851 d->delta_depth = s->delta_depth;
853 return d;
856 static void start_packfile(void)
858 static char tmp_file[PATH_MAX];
859 struct packed_git *p;
860 struct pack_header hdr;
861 int pack_fd;
863 pack_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
864 "pack/tmp_pack_XXXXXX");
865 p = xcalloc(1, sizeof(*p) + strlen(tmp_file) + 2);
866 strcpy(p->pack_name, tmp_file);
867 p->pack_fd = pack_fd;
868 p->do_not_close = 1;
869 pack_file = sha1fd(pack_fd, p->pack_name);
871 hdr.hdr_signature = htonl(PACK_SIGNATURE);
872 hdr.hdr_version = htonl(2);
873 hdr.hdr_entries = 0;
874 sha1write(pack_file, &hdr, sizeof(hdr));
876 pack_data = p;
877 pack_size = sizeof(hdr);
878 object_count = 0;
880 all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
881 all_packs[pack_id] = p;
884 static const char *create_index(void)
886 const char *tmpfile;
887 struct pack_idx_entry **idx, **c, **last;
888 struct object_entry *e;
889 struct object_entry_pool *o;
891 /* Build the table of object IDs. */
892 idx = xmalloc(object_count * sizeof(*idx));
893 c = idx;
894 for (o = blocks; o; o = o->next_pool)
895 for (e = o->next_free; e-- != o->entries;)
896 if (pack_id == e->pack_id)
897 *c++ = &e->idx;
898 last = idx + object_count;
899 if (c != last)
900 die("internal consistency error creating the index");
902 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
903 free(idx);
904 return tmpfile;
907 static char *keep_pack(const char *curr_index_name)
909 static char name[PATH_MAX];
910 static const char *keep_msg = "fast-import";
911 int keep_fd;
913 keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
914 if (keep_fd < 0)
915 die_errno("cannot create keep file");
916 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
917 if (close(keep_fd))
918 die_errno("failed to write keep file");
920 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
921 get_object_directory(), sha1_to_hex(pack_data->sha1));
922 if (move_temp_to_file(pack_data->pack_name, name))
923 die("cannot store pack file");
925 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
926 get_object_directory(), sha1_to_hex(pack_data->sha1));
927 if (move_temp_to_file(curr_index_name, name))
928 die("cannot store index file");
929 free((void *)curr_index_name);
930 return name;
933 static void unkeep_all_packs(void)
935 static char name[PATH_MAX];
936 int k;
938 for (k = 0; k < pack_id; k++) {
939 struct packed_git *p = all_packs[k];
940 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
941 get_object_directory(), sha1_to_hex(p->sha1));
942 unlink_or_warn(name);
946 static void end_packfile(void)
948 struct packed_git *old_p = pack_data, *new_p;
950 clear_delta_base_cache();
951 if (object_count) {
952 unsigned char cur_pack_sha1[20];
953 char *idx_name;
954 int i;
955 struct branch *b;
956 struct tag *t;
958 close_pack_windows(pack_data);
959 sha1close(pack_file, cur_pack_sha1, 0);
960 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
961 pack_data->pack_name, object_count,
962 cur_pack_sha1, pack_size);
963 close(pack_data->pack_fd);
964 idx_name = keep_pack(create_index());
966 /* Register the packfile with core git's machinery. */
967 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
968 if (!new_p)
969 die("core git rejected index %s", idx_name);
970 all_packs[pack_id] = new_p;
971 install_packed_git(new_p);
973 /* Print the boundary */
974 if (pack_edges) {
975 fprintf(pack_edges, "%s:", new_p->pack_name);
976 for (i = 0; i < branch_table_sz; i++) {
977 for (b = branch_table[i]; b; b = b->table_next_branch) {
978 if (b->pack_id == pack_id)
979 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
982 for (t = first_tag; t; t = t->next_tag) {
983 if (t->pack_id == pack_id)
984 fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
986 fputc('\n', pack_edges);
987 fflush(pack_edges);
990 pack_id++;
992 else {
993 close(old_p->pack_fd);
994 unlink_or_warn(old_p->pack_name);
996 free(old_p);
998 /* We can't carry a delta across packfiles. */
999 strbuf_release(&last_blob.data);
1000 last_blob.offset = 0;
1001 last_blob.depth = 0;
1004 static void cycle_packfile(void)
1006 end_packfile();
1007 start_packfile();
1010 static int store_object(
1011 enum object_type type,
1012 struct strbuf *dat,
1013 struct last_object *last,
1014 unsigned char *sha1out,
1015 uintmax_t mark)
1017 void *out, *delta;
1018 struct object_entry *e;
1019 unsigned char hdr[96];
1020 unsigned char sha1[20];
1021 unsigned long hdrlen, deltalen;
1022 git_SHA_CTX c;
1023 git_zstream s;
1025 hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1026 (unsigned long)dat->len) + 1;
1027 git_SHA1_Init(&c);
1028 git_SHA1_Update(&c, hdr, hdrlen);
1029 git_SHA1_Update(&c, dat->buf, dat->len);
1030 git_SHA1_Final(sha1, &c);
1031 if (sha1out)
1032 hashcpy(sha1out, sha1);
1034 e = insert_object(sha1);
1035 if (mark)
1036 insert_mark(mark, e);
1037 if (e->idx.offset) {
1038 duplicate_count_by_type[type]++;
1039 return 1;
1040 } else if (find_sha1_pack(sha1, packed_git)) {
1041 e->type = type;
1042 e->pack_id = MAX_PACK_ID;
1043 e->idx.offset = 1; /* just not zero! */
1044 duplicate_count_by_type[type]++;
1045 return 1;
1048 if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1049 delta_count_attempts_by_type[type]++;
1050 delta = diff_delta(last->data.buf, last->data.len,
1051 dat->buf, dat->len,
1052 &deltalen, dat->len - 20);
1053 } else
1054 delta = NULL;
1056 memset(&s, 0, sizeof(s));
1057 git_deflate_init(&s, pack_compression_level);
1058 if (delta) {
1059 s.next_in = delta;
1060 s.avail_in = deltalen;
1061 } else {
1062 s.next_in = (void *)dat->buf;
1063 s.avail_in = dat->len;
1065 s.avail_out = git_deflate_bound(&s, s.avail_in);
1066 s.next_out = out = xmalloc(s.avail_out);
1067 while (git_deflate(&s, Z_FINISH) == Z_OK)
1068 ; /* nothing */
1069 git_deflate_end(&s);
1071 /* Determine if we should auto-checkpoint. */
1072 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1073 || (pack_size + 60 + s.total_out) < pack_size) {
1075 /* This new object needs to *not* have the current pack_id. */
1076 e->pack_id = pack_id + 1;
1077 cycle_packfile();
1079 /* We cannot carry a delta into the new pack. */
1080 if (delta) {
1081 free(delta);
1082 delta = NULL;
1084 memset(&s, 0, sizeof(s));
1085 git_deflate_init(&s, pack_compression_level);
1086 s.next_in = (void *)dat->buf;
1087 s.avail_in = dat->len;
1088 s.avail_out = git_deflate_bound(&s, s.avail_in);
1089 s.next_out = out = xrealloc(out, s.avail_out);
1090 while (git_deflate(&s, Z_FINISH) == Z_OK)
1091 ; /* nothing */
1092 git_deflate_end(&s);
1096 e->type = type;
1097 e->pack_id = pack_id;
1098 e->idx.offset = pack_size;
1099 object_count++;
1100 object_count_by_type[type]++;
1102 crc32_begin(pack_file);
1104 if (delta) {
1105 off_t ofs = e->idx.offset - last->offset;
1106 unsigned pos = sizeof(hdr) - 1;
1108 delta_count_by_type[type]++;
1109 e->depth = last->depth + 1;
1111 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1112 sha1write(pack_file, hdr, hdrlen);
1113 pack_size += hdrlen;
1115 hdr[pos] = ofs & 127;
1116 while (ofs >>= 7)
1117 hdr[--pos] = 128 | (--ofs & 127);
1118 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1119 pack_size += sizeof(hdr) - pos;
1120 } else {
1121 e->depth = 0;
1122 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1123 sha1write(pack_file, hdr, hdrlen);
1124 pack_size += hdrlen;
1127 sha1write(pack_file, out, s.total_out);
1128 pack_size += s.total_out;
1130 e->idx.crc32 = crc32_end(pack_file);
1132 free(out);
1133 free(delta);
1134 if (last) {
1135 if (last->no_swap) {
1136 last->data = *dat;
1137 } else {
1138 strbuf_swap(&last->data, dat);
1140 last->offset = e->idx.offset;
1141 last->depth = e->depth;
1143 return 0;
1146 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1148 if (sha1file_truncate(pack_file, checkpoint))
1149 die_errno("cannot truncate pack to skip duplicate");
1150 pack_size = checkpoint->offset;
1153 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1155 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1156 unsigned char *in_buf = xmalloc(in_sz);
1157 unsigned char *out_buf = xmalloc(out_sz);
1158 struct object_entry *e;
1159 unsigned char sha1[20];
1160 unsigned long hdrlen;
1161 off_t offset;
1162 git_SHA_CTX c;
1163 git_zstream s;
1164 struct sha1file_checkpoint checkpoint;
1165 int status = Z_OK;
1167 /* Determine if we should auto-checkpoint. */
1168 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1169 || (pack_size + 60 + len) < pack_size)
1170 cycle_packfile();
1172 sha1file_checkpoint(pack_file, &checkpoint);
1173 offset = checkpoint.offset;
1175 hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1176 if (out_sz <= hdrlen)
1177 die("impossibly large object header");
1179 git_SHA1_Init(&c);
1180 git_SHA1_Update(&c, out_buf, hdrlen);
1182 crc32_begin(pack_file);
1184 memset(&s, 0, sizeof(s));
1185 git_deflate_init(&s, pack_compression_level);
1187 hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1188 if (out_sz <= hdrlen)
1189 die("impossibly large object header");
1191 s.next_out = out_buf + hdrlen;
1192 s.avail_out = out_sz - hdrlen;
1194 while (status != Z_STREAM_END) {
1195 if (0 < len && !s.avail_in) {
1196 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1197 size_t n = fread(in_buf, 1, cnt, stdin);
1198 if (!n && feof(stdin))
1199 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1201 git_SHA1_Update(&c, in_buf, n);
1202 s.next_in = in_buf;
1203 s.avail_in = n;
1204 len -= n;
1207 status = git_deflate(&s, len ? 0 : Z_FINISH);
1209 if (!s.avail_out || status == Z_STREAM_END) {
1210 size_t n = s.next_out - out_buf;
1211 sha1write(pack_file, out_buf, n);
1212 pack_size += n;
1213 s.next_out = out_buf;
1214 s.avail_out = out_sz;
1217 switch (status) {
1218 case Z_OK:
1219 case Z_BUF_ERROR:
1220 case Z_STREAM_END:
1221 continue;
1222 default:
1223 die("unexpected deflate failure: %d", status);
1226 git_deflate_end(&s);
1227 git_SHA1_Final(sha1, &c);
1229 if (sha1out)
1230 hashcpy(sha1out, sha1);
1232 e = insert_object(sha1);
1234 if (mark)
1235 insert_mark(mark, e);
1237 if (e->idx.offset) {
1238 duplicate_count_by_type[OBJ_BLOB]++;
1239 truncate_pack(&checkpoint);
1241 } else if (find_sha1_pack(sha1, packed_git)) {
1242 e->type = OBJ_BLOB;
1243 e->pack_id = MAX_PACK_ID;
1244 e->idx.offset = 1; /* just not zero! */
1245 duplicate_count_by_type[OBJ_BLOB]++;
1246 truncate_pack(&checkpoint);
1248 } else {
1249 e->depth = 0;
1250 e->type = OBJ_BLOB;
1251 e->pack_id = pack_id;
1252 e->idx.offset = offset;
1253 e->idx.crc32 = crc32_end(pack_file);
1254 object_count++;
1255 object_count_by_type[OBJ_BLOB]++;
1258 free(in_buf);
1259 free(out_buf);
1262 /* All calls must be guarded by find_object() or find_mark() to
1263 * ensure the 'struct object_entry' passed was written by this
1264 * process instance. We unpack the entry by the offset, avoiding
1265 * the need for the corresponding .idx file. This unpacking rule
1266 * works because we only use OBJ_REF_DELTA within the packfiles
1267 * created by fast-import.
1269 * oe must not be NULL. Such an oe usually comes from giving
1270 * an unknown SHA-1 to find_object() or an undefined mark to
1271 * find_mark(). Callers must test for this condition and use
1272 * the standard read_sha1_file() when it happens.
1274 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1275 * find_mark(), where the mark was reloaded from an existing marks
1276 * file and is referencing an object that this fast-import process
1277 * instance did not write out to a packfile. Callers must test for
1278 * this condition and use read_sha1_file() instead.
1280 static void *gfi_unpack_entry(
1281 struct object_entry *oe,
1282 unsigned long *sizep)
1284 enum object_type type;
1285 struct packed_git *p = all_packs[oe->pack_id];
1286 if (p == pack_data && p->pack_size < (pack_size + 20)) {
1287 /* The object is stored in the packfile we are writing to
1288 * and we have modified it since the last time we scanned
1289 * back to read a previously written object. If an old
1290 * window covered [p->pack_size, p->pack_size + 20) its
1291 * data is stale and is not valid. Closing all windows
1292 * and updating the packfile length ensures we can read
1293 * the newly written data.
1295 close_pack_windows(p);
1296 sha1flush(pack_file);
1298 /* We have to offer 20 bytes additional on the end of
1299 * the packfile as the core unpacker code assumes the
1300 * footer is present at the file end and must promise
1301 * at least 20 bytes within any window it maps. But
1302 * we don't actually create the footer here.
1304 p->pack_size = pack_size + 20;
1306 return unpack_entry(p, oe->idx.offset, &type, sizep);
1309 static const char *get_mode(const char *str, uint16_t *modep)
1311 unsigned char c;
1312 uint16_t mode = 0;
1314 while ((c = *str++) != ' ') {
1315 if (c < '0' || c > '7')
1316 return NULL;
1317 mode = (mode << 3) + (c - '0');
1319 *modep = mode;
1320 return str;
1323 static void load_tree(struct tree_entry *root)
1325 unsigned char *sha1 = root->versions[1].sha1;
1326 struct object_entry *myoe;
1327 struct tree_content *t;
1328 unsigned long size;
1329 char *buf;
1330 const char *c;
1332 root->tree = t = new_tree_content(8);
1333 if (is_null_sha1(sha1))
1334 return;
1336 myoe = find_object(sha1);
1337 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1338 if (myoe->type != OBJ_TREE)
1339 die("Not a tree: %s", sha1_to_hex(sha1));
1340 t->delta_depth = myoe->depth;
1341 buf = gfi_unpack_entry(myoe, &size);
1342 if (!buf)
1343 die("Can't load tree %s", sha1_to_hex(sha1));
1344 } else {
1345 enum object_type type;
1346 buf = read_sha1_file(sha1, &type, &size);
1347 if (!buf || type != OBJ_TREE)
1348 die("Can't load tree %s", sha1_to_hex(sha1));
1351 c = buf;
1352 while (c != (buf + size)) {
1353 struct tree_entry *e = new_tree_entry();
1355 if (t->entry_count == t->entry_capacity)
1356 root->tree = t = grow_tree_content(t, t->entry_count);
1357 t->entries[t->entry_count++] = e;
1359 e->tree = NULL;
1360 c = get_mode(c, &e->versions[1].mode);
1361 if (!c)
1362 die("Corrupt mode in %s", sha1_to_hex(sha1));
1363 e->versions[0].mode = e->versions[1].mode;
1364 e->name = to_atom(c, strlen(c));
1365 c += e->name->str_len + 1;
1366 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1367 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1368 c += 20;
1370 free(buf);
1373 static int tecmp0 (const void *_a, const void *_b)
1375 struct tree_entry *a = *((struct tree_entry**)_a);
1376 struct tree_entry *b = *((struct tree_entry**)_b);
1377 return base_name_compare(
1378 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1379 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1382 static int tecmp1 (const void *_a, const void *_b)
1384 struct tree_entry *a = *((struct tree_entry**)_a);
1385 struct tree_entry *b = *((struct tree_entry**)_b);
1386 return base_name_compare(
1387 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1388 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1391 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1393 size_t maxlen = 0;
1394 unsigned int i;
1396 if (!v)
1397 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1398 else
1399 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1401 for (i = 0; i < t->entry_count; i++) {
1402 if (t->entries[i]->versions[v].mode)
1403 maxlen += t->entries[i]->name->str_len + 34;
1406 strbuf_reset(b);
1407 strbuf_grow(b, maxlen);
1408 for (i = 0; i < t->entry_count; i++) {
1409 struct tree_entry *e = t->entries[i];
1410 if (!e->versions[v].mode)
1411 continue;
1412 strbuf_addf(b, "%o %s%c",
1413 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1414 e->name->str_dat, '\0');
1415 strbuf_add(b, e->versions[v].sha1, 20);
1419 static void store_tree(struct tree_entry *root)
1421 struct tree_content *t = root->tree;
1422 unsigned int i, j, del;
1423 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1424 struct object_entry *le = NULL;
1426 if (!is_null_sha1(root->versions[1].sha1))
1427 return;
1429 for (i = 0; i < t->entry_count; i++) {
1430 if (t->entries[i]->tree)
1431 store_tree(t->entries[i]);
1434 if (!(root->versions[0].mode & NO_DELTA))
1435 le = find_object(root->versions[0].sha1);
1436 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1437 mktree(t, 0, &old_tree);
1438 lo.data = old_tree;
1439 lo.offset = le->idx.offset;
1440 lo.depth = t->delta_depth;
1443 mktree(t, 1, &new_tree);
1444 store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1446 t->delta_depth = lo.depth;
1447 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1448 struct tree_entry *e = t->entries[i];
1449 if (e->versions[1].mode) {
1450 e->versions[0].mode = e->versions[1].mode;
1451 hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1452 t->entries[j++] = e;
1453 } else {
1454 release_tree_entry(e);
1455 del++;
1458 t->entry_count -= del;
1461 static void tree_content_replace(
1462 struct tree_entry *root,
1463 const unsigned char *sha1,
1464 const uint16_t mode,
1465 struct tree_content *newtree)
1467 if (!S_ISDIR(mode))
1468 die("Root cannot be a non-directory");
1469 hashclr(root->versions[0].sha1);
1470 hashcpy(root->versions[1].sha1, sha1);
1471 if (root->tree)
1472 release_tree_content_recursive(root->tree);
1473 root->tree = newtree;
1476 static int tree_content_set(
1477 struct tree_entry *root,
1478 const char *p,
1479 const unsigned char *sha1,
1480 const uint16_t mode,
1481 struct tree_content *subtree)
1483 struct tree_content *t;
1484 const char *slash1;
1485 unsigned int i, n;
1486 struct tree_entry *e;
1488 slash1 = strchr(p, '/');
1489 if (slash1)
1490 n = slash1 - p;
1491 else
1492 n = strlen(p);
1493 if (!n)
1494 die("Empty path component found in input");
1495 if (!slash1 && !S_ISDIR(mode) && subtree)
1496 die("Non-directories cannot have subtrees");
1498 if (!root->tree)
1499 load_tree(root);
1500 t = root->tree;
1501 for (i = 0; i < t->entry_count; i++) {
1502 e = t->entries[i];
1503 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1504 if (!slash1) {
1505 if (!S_ISDIR(mode)
1506 && e->versions[1].mode == mode
1507 && !hashcmp(e->versions[1].sha1, sha1))
1508 return 0;
1509 e->versions[1].mode = mode;
1510 hashcpy(e->versions[1].sha1, sha1);
1511 if (e->tree)
1512 release_tree_content_recursive(e->tree);
1513 e->tree = subtree;
1516 * We need to leave e->versions[0].sha1 alone
1517 * to avoid modifying the preimage tree used
1518 * when writing out the parent directory.
1519 * But after replacing the subdir with a
1520 * completely different one, it's not a good
1521 * delta base any more, and besides, we've
1522 * thrown away the tree entries needed to
1523 * make a delta against it.
1525 * So let's just explicitly disable deltas
1526 * for the subtree.
1528 if (S_ISDIR(e->versions[0].mode))
1529 e->versions[0].mode |= NO_DELTA;
1531 hashclr(root->versions[1].sha1);
1532 return 1;
1534 if (!S_ISDIR(e->versions[1].mode)) {
1535 e->tree = new_tree_content(8);
1536 e->versions[1].mode = S_IFDIR;
1538 if (!e->tree)
1539 load_tree(e);
1540 if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1541 hashclr(root->versions[1].sha1);
1542 return 1;
1544 return 0;
1548 if (t->entry_count == t->entry_capacity)
1549 root->tree = t = grow_tree_content(t, t->entry_count);
1550 e = new_tree_entry();
1551 e->name = to_atom(p, n);
1552 e->versions[0].mode = 0;
1553 hashclr(e->versions[0].sha1);
1554 t->entries[t->entry_count++] = e;
1555 if (slash1) {
1556 e->tree = new_tree_content(8);
1557 e->versions[1].mode = S_IFDIR;
1558 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1559 } else {
1560 e->tree = subtree;
1561 e->versions[1].mode = mode;
1562 hashcpy(e->versions[1].sha1, sha1);
1564 hashclr(root->versions[1].sha1);
1565 return 1;
1568 static int tree_content_remove(
1569 struct tree_entry *root,
1570 const char *p,
1571 struct tree_entry *backup_leaf,
1572 int allow_root)
1574 struct tree_content *t;
1575 const char *slash1;
1576 unsigned int i, n;
1577 struct tree_entry *e;
1579 slash1 = strchr(p, '/');
1580 if (slash1)
1581 n = slash1 - p;
1582 else
1583 n = strlen(p);
1585 if (!root->tree)
1586 load_tree(root);
1588 if (!*p && allow_root) {
1589 e = root;
1590 goto del_entry;
1593 t = root->tree;
1594 for (i = 0; i < t->entry_count; i++) {
1595 e = t->entries[i];
1596 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1597 if (slash1 && !S_ISDIR(e->versions[1].mode))
1599 * If p names a file in some subdirectory, and a
1600 * file or symlink matching the name of the
1601 * parent directory of p exists, then p cannot
1602 * exist and need not be deleted.
1604 return 1;
1605 if (!slash1 || !S_ISDIR(e->versions[1].mode))
1606 goto del_entry;
1607 if (!e->tree)
1608 load_tree(e);
1609 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1610 for (n = 0; n < e->tree->entry_count; n++) {
1611 if (e->tree->entries[n]->versions[1].mode) {
1612 hashclr(root->versions[1].sha1);
1613 return 1;
1616 backup_leaf = NULL;
1617 goto del_entry;
1619 return 0;
1622 return 0;
1624 del_entry:
1625 if (backup_leaf)
1626 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1627 else if (e->tree)
1628 release_tree_content_recursive(e->tree);
1629 e->tree = NULL;
1630 e->versions[1].mode = 0;
1631 hashclr(e->versions[1].sha1);
1632 hashclr(root->versions[1].sha1);
1633 return 1;
1636 static int tree_content_get(
1637 struct tree_entry *root,
1638 const char *p,
1639 struct tree_entry *leaf,
1640 int allow_root)
1642 struct tree_content *t;
1643 const char *slash1;
1644 unsigned int i, n;
1645 struct tree_entry *e;
1647 slash1 = strchr(p, '/');
1648 if (slash1)
1649 n = slash1 - p;
1650 else
1651 n = strlen(p);
1652 if (!n && !allow_root)
1653 die("Empty path component found in input");
1655 if (!root->tree)
1656 load_tree(root);
1658 if (!n) {
1659 e = root;
1660 goto found_entry;
1663 t = root->tree;
1664 for (i = 0; i < t->entry_count; i++) {
1665 e = t->entries[i];
1666 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1667 if (!slash1)
1668 goto found_entry;
1669 if (!S_ISDIR(e->versions[1].mode))
1670 return 0;
1671 if (!e->tree)
1672 load_tree(e);
1673 return tree_content_get(e, slash1 + 1, leaf, 0);
1676 return 0;
1678 found_entry:
1679 memcpy(leaf, e, sizeof(*leaf));
1680 if (e->tree && is_null_sha1(e->versions[1].sha1))
1681 leaf->tree = dup_tree_content(e->tree);
1682 else
1683 leaf->tree = NULL;
1684 return 1;
1687 static int update_branch(struct branch *b)
1689 static const char *msg = "fast-import";
1690 struct ref_lock *lock;
1691 unsigned char old_sha1[20];
1693 if (is_null_sha1(b->sha1))
1694 return 0;
1695 if (read_ref(b->name, old_sha1))
1696 hashclr(old_sha1);
1697 lock = lock_any_ref_for_update(b->name, old_sha1, 0, NULL);
1698 if (!lock)
1699 return error("Unable to lock %s", b->name);
1700 if (!force_update && !is_null_sha1(old_sha1)) {
1701 struct commit *old_cmit, *new_cmit;
1703 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1704 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1705 if (!old_cmit || !new_cmit) {
1706 unlock_ref(lock);
1707 return error("Branch %s is missing commits.", b->name);
1710 if (!in_merge_bases(old_cmit, new_cmit)) {
1711 unlock_ref(lock);
1712 warning("Not updating %s"
1713 " (new tip %s does not contain %s)",
1714 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1715 return -1;
1718 if (write_ref_sha1(lock, b->sha1, msg) < 0)
1719 return error("Unable to update %s", b->name);
1720 return 0;
1723 static void dump_branches(void)
1725 unsigned int i;
1726 struct branch *b;
1728 for (i = 0; i < branch_table_sz; i++) {
1729 for (b = branch_table[i]; b; b = b->table_next_branch)
1730 failure |= update_branch(b);
1734 static void dump_tags(void)
1736 static const char *msg = "fast-import";
1737 struct tag *t;
1738 struct ref_lock *lock;
1739 char ref_name[PATH_MAX];
1741 for (t = first_tag; t; t = t->next_tag) {
1742 sprintf(ref_name, "tags/%s", t->name);
1743 lock = lock_ref_sha1(ref_name, NULL);
1744 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1745 failure |= error("Unable to update %s", ref_name);
1749 static void dump_marks_helper(FILE *f,
1750 uintmax_t base,
1751 struct mark_set *m)
1753 uintmax_t k;
1754 if (m->shift) {
1755 for (k = 0; k < 1024; k++) {
1756 if (m->data.sets[k])
1757 dump_marks_helper(f, base + (k << m->shift),
1758 m->data.sets[k]);
1760 } else {
1761 for (k = 0; k < 1024; k++) {
1762 if (m->data.marked[k])
1763 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1764 sha1_to_hex(m->data.marked[k]->idx.sha1));
1769 static void dump_marks(void)
1771 static struct lock_file mark_lock;
1772 int mark_fd;
1773 FILE *f;
1775 if (!export_marks_file)
1776 return;
1778 mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1779 if (mark_fd < 0) {
1780 failure |= error("Unable to write marks file %s: %s",
1781 export_marks_file, strerror(errno));
1782 return;
1785 f = fdopen(mark_fd, "w");
1786 if (!f) {
1787 int saved_errno = errno;
1788 rollback_lock_file(&mark_lock);
1789 failure |= error("Unable to write marks file %s: %s",
1790 export_marks_file, strerror(saved_errno));
1791 return;
1795 * Since the lock file was fdopen()'ed, it should not be close()'ed.
1796 * Assign -1 to the lock file descriptor so that commit_lock_file()
1797 * won't try to close() it.
1799 mark_lock.fd = -1;
1801 dump_marks_helper(f, 0, marks);
1802 if (ferror(f) || fclose(f)) {
1803 int saved_errno = errno;
1804 rollback_lock_file(&mark_lock);
1805 failure |= error("Unable to write marks file %s: %s",
1806 export_marks_file, strerror(saved_errno));
1807 return;
1810 if (commit_lock_file(&mark_lock)) {
1811 int saved_errno = errno;
1812 rollback_lock_file(&mark_lock);
1813 failure |= error("Unable to commit marks file %s: %s",
1814 export_marks_file, strerror(saved_errno));
1815 return;
1819 static void read_marks(void)
1821 char line[512];
1822 FILE *f = fopen(import_marks_file, "r");
1823 if (f)
1825 else if (import_marks_file_ignore_missing && errno == ENOENT)
1826 return; /* Marks file does not exist */
1827 else
1828 die_errno("cannot read '%s'", import_marks_file);
1829 while (fgets(line, sizeof(line), f)) {
1830 uintmax_t mark;
1831 char *end;
1832 unsigned char sha1[20];
1833 struct object_entry *e;
1835 end = strchr(line, '\n');
1836 if (line[0] != ':' || !end)
1837 die("corrupt mark line: %s", line);
1838 *end = 0;
1839 mark = strtoumax(line + 1, &end, 10);
1840 if (!mark || end == line + 1
1841 || *end != ' ' || get_sha1_hex(end + 1, sha1))
1842 die("corrupt mark line: %s", line);
1843 e = find_object(sha1);
1844 if (!e) {
1845 enum object_type type = sha1_object_info(sha1, NULL);
1846 if (type < 0)
1847 die("object not found: %s", sha1_to_hex(sha1));
1848 e = insert_object(sha1);
1849 e->type = type;
1850 e->pack_id = MAX_PACK_ID;
1851 e->idx.offset = 1; /* just not zero! */
1853 insert_mark(mark, e);
1855 fclose(f);
1859 static int read_next_command(void)
1861 static int stdin_eof = 0;
1863 if (stdin_eof) {
1864 unread_command_buf = 0;
1865 return EOF;
1868 for (;;) {
1869 if (unread_command_buf) {
1870 unread_command_buf = 0;
1871 } else {
1872 struct recent_command *rc;
1874 strbuf_detach(&command_buf, NULL);
1875 stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1876 if (stdin_eof)
1877 return EOF;
1879 if (!seen_data_command
1880 && prefixcmp(command_buf.buf, "feature ")
1881 && prefixcmp(command_buf.buf, "option ")) {
1882 parse_argv();
1885 rc = rc_free;
1886 if (rc)
1887 rc_free = rc->next;
1888 else {
1889 rc = cmd_hist.next;
1890 cmd_hist.next = rc->next;
1891 cmd_hist.next->prev = &cmd_hist;
1892 free(rc->buf);
1895 rc->buf = command_buf.buf;
1896 rc->prev = cmd_tail;
1897 rc->next = cmd_hist.prev;
1898 rc->prev->next = rc;
1899 cmd_tail = rc;
1901 if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1902 parse_cat_blob();
1903 continue;
1905 if (command_buf.buf[0] == '#')
1906 continue;
1907 return 0;
1911 static void skip_optional_lf(void)
1913 int term_char = fgetc(stdin);
1914 if (term_char != '\n' && term_char != EOF)
1915 ungetc(term_char, stdin);
1918 static void parse_mark(void)
1920 if (!prefixcmp(command_buf.buf, "mark :")) {
1921 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1922 read_next_command();
1924 else
1925 next_mark = 0;
1928 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1930 strbuf_reset(sb);
1932 if (prefixcmp(command_buf.buf, "data "))
1933 die("Expected 'data n' command, found: %s", command_buf.buf);
1935 if (!prefixcmp(command_buf.buf + 5, "<<")) {
1936 char *term = xstrdup(command_buf.buf + 5 + 2);
1937 size_t term_len = command_buf.len - 5 - 2;
1939 strbuf_detach(&command_buf, NULL);
1940 for (;;) {
1941 if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1942 die("EOF in data (terminator '%s' not found)", term);
1943 if (term_len == command_buf.len
1944 && !strcmp(term, command_buf.buf))
1945 break;
1946 strbuf_addbuf(sb, &command_buf);
1947 strbuf_addch(sb, '\n');
1949 free(term);
1951 else {
1952 uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1953 size_t n = 0, length = (size_t)len;
1955 if (limit && limit < len) {
1956 *len_res = len;
1957 return 0;
1959 if (length < len)
1960 die("data is too large to use in this context");
1962 while (n < length) {
1963 size_t s = strbuf_fread(sb, length - n, stdin);
1964 if (!s && feof(stdin))
1965 die("EOF in data (%lu bytes remaining)",
1966 (unsigned long)(length - n));
1967 n += s;
1971 skip_optional_lf();
1972 return 1;
1975 static int validate_raw_date(const char *src, char *result, int maxlen)
1977 const char *orig_src = src;
1978 char *endp;
1979 unsigned long num;
1981 errno = 0;
1983 num = strtoul(src, &endp, 10);
1984 /* NEEDSWORK: perhaps check for reasonable values? */
1985 if (errno || endp == src || *endp != ' ')
1986 return -1;
1988 src = endp + 1;
1989 if (*src != '-' && *src != '+')
1990 return -1;
1992 num = strtoul(src + 1, &endp, 10);
1993 if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1994 1400 < num)
1995 return -1;
1997 strcpy(result, orig_src);
1998 return 0;
2001 static char *parse_ident(const char *buf)
2003 const char *ltgt;
2004 size_t name_len;
2005 char *ident;
2007 /* ensure there is a space delimiter even if there is no name */
2008 if (*buf == '<')
2009 --buf;
2011 ltgt = buf + strcspn(buf, "<>");
2012 if (*ltgt != '<')
2013 die("Missing < in ident string: %s", buf);
2014 if (ltgt != buf && ltgt[-1] != ' ')
2015 die("Missing space before < in ident string: %s", buf);
2016 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2017 if (*ltgt != '>')
2018 die("Missing > in ident string: %s", buf);
2019 ltgt++;
2020 if (*ltgt != ' ')
2021 die("Missing space after > in ident string: %s", buf);
2022 ltgt++;
2023 name_len = ltgt - buf;
2024 ident = xmalloc(name_len + 24);
2025 strncpy(ident, buf, name_len);
2027 switch (whenspec) {
2028 case WHENSPEC_RAW:
2029 if (validate_raw_date(ltgt, ident + name_len, 24) < 0)
2030 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2031 break;
2032 case WHENSPEC_RFC2822:
2033 if (parse_date(ltgt, ident + name_len, 24) < 0)
2034 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2035 break;
2036 case WHENSPEC_NOW:
2037 if (strcmp("now", ltgt))
2038 die("Date in ident must be 'now': %s", buf);
2039 datestamp(ident + name_len, 24);
2040 break;
2043 return ident;
2046 static void parse_and_store_blob(
2047 struct last_object *last,
2048 unsigned char *sha1out,
2049 uintmax_t mark)
2051 static struct strbuf buf = STRBUF_INIT;
2052 uintmax_t len;
2054 if (parse_data(&buf, big_file_threshold, &len))
2055 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2056 else {
2057 if (last) {
2058 strbuf_release(&last->data);
2059 last->offset = 0;
2060 last->depth = 0;
2062 stream_blob(len, sha1out, mark);
2063 skip_optional_lf();
2067 static void parse_new_blob(void)
2069 read_next_command();
2070 parse_mark();
2071 parse_and_store_blob(&last_blob, NULL, next_mark);
2074 static void unload_one_branch(void)
2076 while (cur_active_branches
2077 && cur_active_branches >= max_active_branches) {
2078 uintmax_t min_commit = ULONG_MAX;
2079 struct branch *e, *l = NULL, *p = NULL;
2081 for (e = active_branches; e; e = e->active_next_branch) {
2082 if (e->last_commit < min_commit) {
2083 p = l;
2084 min_commit = e->last_commit;
2086 l = e;
2089 if (p) {
2090 e = p->active_next_branch;
2091 p->active_next_branch = e->active_next_branch;
2092 } else {
2093 e = active_branches;
2094 active_branches = e->active_next_branch;
2096 e->active = 0;
2097 e->active_next_branch = NULL;
2098 if (e->branch_tree.tree) {
2099 release_tree_content_recursive(e->branch_tree.tree);
2100 e->branch_tree.tree = NULL;
2102 cur_active_branches--;
2106 static void load_branch(struct branch *b)
2108 load_tree(&b->branch_tree);
2109 if (!b->active) {
2110 b->active = 1;
2111 b->active_next_branch = active_branches;
2112 active_branches = b;
2113 cur_active_branches++;
2114 branch_load_count++;
2118 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2120 unsigned char fanout = 0;
2121 while ((num_notes >>= 8))
2122 fanout++;
2123 return fanout;
2126 static void construct_path_with_fanout(const char *hex_sha1,
2127 unsigned char fanout, char *path)
2129 unsigned int i = 0, j = 0;
2130 if (fanout >= 20)
2131 die("Too large fanout (%u)", fanout);
2132 while (fanout) {
2133 path[i++] = hex_sha1[j++];
2134 path[i++] = hex_sha1[j++];
2135 path[i++] = '/';
2136 fanout--;
2138 memcpy(path + i, hex_sha1 + j, 40 - j);
2139 path[i + 40 - j] = '\0';
2142 static uintmax_t do_change_note_fanout(
2143 struct tree_entry *orig_root, struct tree_entry *root,
2144 char *hex_sha1, unsigned int hex_sha1_len,
2145 char *fullpath, unsigned int fullpath_len,
2146 unsigned char fanout)
2148 struct tree_content *t = root->tree;
2149 struct tree_entry *e, leaf;
2150 unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2151 uintmax_t num_notes = 0;
2152 unsigned char sha1[20];
2153 char realpath[60];
2155 for (i = 0; t && i < t->entry_count; i++) {
2156 e = t->entries[i];
2157 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2158 tmp_fullpath_len = fullpath_len;
2161 * We're interested in EITHER existing note entries (entries
2162 * with exactly 40 hex chars in path, not including directory
2163 * separators), OR directory entries that may contain note
2164 * entries (with < 40 hex chars in path).
2165 * Also, each path component in a note entry must be a multiple
2166 * of 2 chars.
2168 if (!e->versions[1].mode ||
2169 tmp_hex_sha1_len > 40 ||
2170 e->name->str_len % 2)
2171 continue;
2173 /* This _may_ be a note entry, or a subdir containing notes */
2174 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2175 e->name->str_len);
2176 if (tmp_fullpath_len)
2177 fullpath[tmp_fullpath_len++] = '/';
2178 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2179 e->name->str_len);
2180 tmp_fullpath_len += e->name->str_len;
2181 fullpath[tmp_fullpath_len] = '\0';
2183 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2184 /* This is a note entry */
2185 if (fanout == 0xff) {
2186 /* Counting mode, no rename */
2187 num_notes++;
2188 continue;
2190 construct_path_with_fanout(hex_sha1, fanout, realpath);
2191 if (!strcmp(fullpath, realpath)) {
2192 /* Note entry is in correct location */
2193 num_notes++;
2194 continue;
2197 /* Rename fullpath to realpath */
2198 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2199 die("Failed to remove path %s", fullpath);
2200 tree_content_set(orig_root, realpath,
2201 leaf.versions[1].sha1,
2202 leaf.versions[1].mode,
2203 leaf.tree);
2204 } else if (S_ISDIR(e->versions[1].mode)) {
2205 /* This is a subdir that may contain note entries */
2206 if (!e->tree)
2207 load_tree(e);
2208 num_notes += do_change_note_fanout(orig_root, e,
2209 hex_sha1, tmp_hex_sha1_len,
2210 fullpath, tmp_fullpath_len, fanout);
2213 /* The above may have reallocated the current tree_content */
2214 t = root->tree;
2216 return num_notes;
2219 static uintmax_t change_note_fanout(struct tree_entry *root,
2220 unsigned char fanout)
2222 char hex_sha1[40], path[60];
2223 return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2227 * Given a pointer into a string, parse a mark reference:
2229 * idnum ::= ':' bigint;
2231 * Return the first character after the value in *endptr.
2233 * Complain if the following character is not what is expected,
2234 * either a space or end of the string.
2236 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2238 uintmax_t mark;
2240 assert(*p == ':');
2241 p++;
2242 mark = strtoumax(p, endptr, 10);
2243 if (*endptr == p)
2244 die("No value after ':' in mark: %s", command_buf.buf);
2245 return mark;
2249 * Parse the mark reference, and complain if this is not the end of
2250 * the string.
2252 static uintmax_t parse_mark_ref_eol(const char *p)
2254 char *end;
2255 uintmax_t mark;
2257 mark = parse_mark_ref(p, &end);
2258 if (*end != '\0')
2259 die("Garbage after mark: %s", command_buf.buf);
2260 return mark;
2264 * Parse the mark reference, demanding a trailing space. Return a
2265 * pointer to the space.
2267 static uintmax_t parse_mark_ref_space(const char **p)
2269 uintmax_t mark;
2270 char *end;
2272 mark = parse_mark_ref(*p, &end);
2273 if (*end != ' ')
2274 die("Missing space after mark: %s", command_buf.buf);
2275 *p = end;
2276 return mark;
2279 static void file_change_m(struct branch *b)
2281 const char *p = command_buf.buf + 2;
2282 static struct strbuf uq = STRBUF_INIT;
2283 const char *endp;
2284 struct object_entry *oe;
2285 unsigned char sha1[20];
2286 uint16_t mode, inline_data = 0;
2288 p = get_mode(p, &mode);
2289 if (!p)
2290 die("Corrupt mode: %s", command_buf.buf);
2291 switch (mode) {
2292 case 0644:
2293 case 0755:
2294 mode |= S_IFREG;
2295 case S_IFREG | 0644:
2296 case S_IFREG | 0755:
2297 case S_IFLNK:
2298 case S_IFDIR:
2299 case S_IFGITLINK:
2300 /* ok */
2301 break;
2302 default:
2303 die("Corrupt mode: %s", command_buf.buf);
2306 if (*p == ':') {
2307 oe = find_mark(parse_mark_ref_space(&p));
2308 hashcpy(sha1, oe->idx.sha1);
2309 } else if (!prefixcmp(p, "inline ")) {
2310 inline_data = 1;
2311 oe = NULL; /* not used with inline_data, but makes gcc happy */
2312 p += strlen("inline"); /* advance to space */
2313 } else {
2314 if (get_sha1_hex(p, sha1))
2315 die("Invalid dataref: %s", command_buf.buf);
2316 oe = find_object(sha1);
2317 p += 40;
2318 if (*p != ' ')
2319 die("Missing space after SHA1: %s", command_buf.buf);
2321 assert(*p == ' ');
2322 p++; /* skip space */
2324 strbuf_reset(&uq);
2325 if (!unquote_c_style(&uq, p, &endp)) {
2326 if (*endp)
2327 die("Garbage after path in: %s", command_buf.buf);
2328 p = uq.buf;
2331 /* Git does not track empty, non-toplevel directories. */
2332 if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2333 tree_content_remove(&b->branch_tree, p, NULL, 0);
2334 return;
2337 if (S_ISGITLINK(mode)) {
2338 if (inline_data)
2339 die("Git links cannot be specified 'inline': %s",
2340 command_buf.buf);
2341 else if (oe) {
2342 if (oe->type != OBJ_COMMIT)
2343 die("Not a commit (actually a %s): %s",
2344 typename(oe->type), command_buf.buf);
2347 * Accept the sha1 without checking; it expected to be in
2348 * another repository.
2350 } else if (inline_data) {
2351 if (S_ISDIR(mode))
2352 die("Directories cannot be specified 'inline': %s",
2353 command_buf.buf);
2354 if (p != uq.buf) {
2355 strbuf_addstr(&uq, p);
2356 p = uq.buf;
2358 read_next_command();
2359 parse_and_store_blob(&last_blob, sha1, 0);
2360 } else {
2361 enum object_type expected = S_ISDIR(mode) ?
2362 OBJ_TREE: OBJ_BLOB;
2363 enum object_type type = oe ? oe->type :
2364 sha1_object_info(sha1, NULL);
2365 if (type < 0)
2366 die("%s not found: %s",
2367 S_ISDIR(mode) ? "Tree" : "Blob",
2368 command_buf.buf);
2369 if (type != expected)
2370 die("Not a %s (actually a %s): %s",
2371 typename(expected), typename(type),
2372 command_buf.buf);
2375 if (!*p) {
2376 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2377 return;
2379 tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2382 static void file_change_d(struct branch *b)
2384 const char *p = command_buf.buf + 2;
2385 static struct strbuf uq = STRBUF_INIT;
2386 const char *endp;
2388 strbuf_reset(&uq);
2389 if (!unquote_c_style(&uq, p, &endp)) {
2390 if (*endp)
2391 die("Garbage after path in: %s", command_buf.buf);
2392 p = uq.buf;
2394 tree_content_remove(&b->branch_tree, p, NULL, 1);
2397 static void file_change_cr(struct branch *b, int rename)
2399 const char *s, *d;
2400 static struct strbuf s_uq = STRBUF_INIT;
2401 static struct strbuf d_uq = STRBUF_INIT;
2402 const char *endp;
2403 struct tree_entry leaf;
2405 s = command_buf.buf + 2;
2406 strbuf_reset(&s_uq);
2407 if (!unquote_c_style(&s_uq, s, &endp)) {
2408 if (*endp != ' ')
2409 die("Missing space after source: %s", command_buf.buf);
2410 } else {
2411 endp = strchr(s, ' ');
2412 if (!endp)
2413 die("Missing space after source: %s", command_buf.buf);
2414 strbuf_add(&s_uq, s, endp - s);
2416 s = s_uq.buf;
2418 endp++;
2419 if (!*endp)
2420 die("Missing dest: %s", command_buf.buf);
2422 d = endp;
2423 strbuf_reset(&d_uq);
2424 if (!unquote_c_style(&d_uq, d, &endp)) {
2425 if (*endp)
2426 die("Garbage after dest in: %s", command_buf.buf);
2427 d = d_uq.buf;
2430 memset(&leaf, 0, sizeof(leaf));
2431 if (rename)
2432 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2433 else
2434 tree_content_get(&b->branch_tree, s, &leaf, 1);
2435 if (!leaf.versions[1].mode)
2436 die("Path %s not in branch", s);
2437 if (!*d) { /* C "path/to/subdir" "" */
2438 tree_content_replace(&b->branch_tree,
2439 leaf.versions[1].sha1,
2440 leaf.versions[1].mode,
2441 leaf.tree);
2442 return;
2444 tree_content_set(&b->branch_tree, d,
2445 leaf.versions[1].sha1,
2446 leaf.versions[1].mode,
2447 leaf.tree);
2450 static void note_change_n(struct branch *b, unsigned char *old_fanout)
2452 const char *p = command_buf.buf + 2;
2453 static struct strbuf uq = STRBUF_INIT;
2454 struct object_entry *oe;
2455 struct branch *s;
2456 unsigned char sha1[20], commit_sha1[20];
2457 char path[60];
2458 uint16_t inline_data = 0;
2459 unsigned char new_fanout;
2462 * When loading a branch, we don't traverse its tree to count the real
2463 * number of notes (too expensive to do this for all non-note refs).
2464 * This means that recently loaded notes refs might incorrectly have
2465 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2467 * Fix this by traversing the tree and counting the number of notes
2468 * when b->num_notes == 0. If the notes tree is truly empty, the
2469 * calculation should not take long.
2471 if (b->num_notes == 0 && *old_fanout == 0) {
2472 /* Invoke change_note_fanout() in "counting mode". */
2473 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2474 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2477 /* Now parse the notemodify command. */
2478 /* <dataref> or 'inline' */
2479 if (*p == ':') {
2480 oe = find_mark(parse_mark_ref_space(&p));
2481 hashcpy(sha1, oe->idx.sha1);
2482 } else if (!prefixcmp(p, "inline ")) {
2483 inline_data = 1;
2484 oe = NULL; /* not used with inline_data, but makes gcc happy */
2485 p += strlen("inline"); /* advance to space */
2486 } else {
2487 if (get_sha1_hex(p, sha1))
2488 die("Invalid dataref: %s", command_buf.buf);
2489 oe = find_object(sha1);
2490 p += 40;
2491 if (*p != ' ')
2492 die("Missing space after SHA1: %s", command_buf.buf);
2494 assert(*p == ' ');
2495 p++; /* skip space */
2497 /* <commit-ish> */
2498 s = lookup_branch(p);
2499 if (s) {
2500 if (is_null_sha1(s->sha1))
2501 die("Can't add a note on empty branch.");
2502 hashcpy(commit_sha1, s->sha1);
2503 } else if (*p == ':') {
2504 uintmax_t commit_mark = parse_mark_ref_eol(p);
2505 struct object_entry *commit_oe = find_mark(commit_mark);
2506 if (commit_oe->type != OBJ_COMMIT)
2507 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2508 hashcpy(commit_sha1, commit_oe->idx.sha1);
2509 } else if (!get_sha1(p, commit_sha1)) {
2510 unsigned long size;
2511 char *buf = read_object_with_reference(commit_sha1,
2512 commit_type, &size, commit_sha1);
2513 if (!buf || size < 46)
2514 die("Not a valid commit: %s", p);
2515 free(buf);
2516 } else
2517 die("Invalid ref name or SHA1 expression: %s", p);
2519 if (inline_data) {
2520 if (p != uq.buf) {
2521 strbuf_addstr(&uq, p);
2522 p = uq.buf;
2524 read_next_command();
2525 parse_and_store_blob(&last_blob, sha1, 0);
2526 } else if (oe) {
2527 if (oe->type != OBJ_BLOB)
2528 die("Not a blob (actually a %s): %s",
2529 typename(oe->type), command_buf.buf);
2530 } else if (!is_null_sha1(sha1)) {
2531 enum object_type type = sha1_object_info(sha1, NULL);
2532 if (type < 0)
2533 die("Blob not found: %s", command_buf.buf);
2534 if (type != OBJ_BLOB)
2535 die("Not a blob (actually a %s): %s",
2536 typename(type), command_buf.buf);
2539 construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2540 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2541 b->num_notes--;
2543 if (is_null_sha1(sha1))
2544 return; /* nothing to insert */
2546 b->num_notes++;
2547 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2548 construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2549 tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2552 static void file_change_deleteall(struct branch *b)
2554 release_tree_content_recursive(b->branch_tree.tree);
2555 hashclr(b->branch_tree.versions[0].sha1);
2556 hashclr(b->branch_tree.versions[1].sha1);
2557 load_tree(&b->branch_tree);
2558 b->num_notes = 0;
2561 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2563 if (!buf || size < 46)
2564 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2565 if (memcmp("tree ", buf, 5)
2566 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2567 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2568 hashcpy(b->branch_tree.versions[0].sha1,
2569 b->branch_tree.versions[1].sha1);
2572 static void parse_from_existing(struct branch *b)
2574 if (is_null_sha1(b->sha1)) {
2575 hashclr(b->branch_tree.versions[0].sha1);
2576 hashclr(b->branch_tree.versions[1].sha1);
2577 } else {
2578 unsigned long size;
2579 char *buf;
2581 buf = read_object_with_reference(b->sha1,
2582 commit_type, &size, b->sha1);
2583 parse_from_commit(b, buf, size);
2584 free(buf);
2588 static int parse_from(struct branch *b)
2590 const char *from;
2591 struct branch *s;
2593 if (prefixcmp(command_buf.buf, "from "))
2594 return 0;
2596 if (b->branch_tree.tree) {
2597 release_tree_content_recursive(b->branch_tree.tree);
2598 b->branch_tree.tree = NULL;
2601 from = strchr(command_buf.buf, ' ') + 1;
2602 s = lookup_branch(from);
2603 if (b == s)
2604 die("Can't create a branch from itself: %s", b->name);
2605 else if (s) {
2606 unsigned char *t = s->branch_tree.versions[1].sha1;
2607 hashcpy(b->sha1, s->sha1);
2608 hashcpy(b->branch_tree.versions[0].sha1, t);
2609 hashcpy(b->branch_tree.versions[1].sha1, t);
2610 } else if (*from == ':') {
2611 uintmax_t idnum = parse_mark_ref_eol(from);
2612 struct object_entry *oe = find_mark(idnum);
2613 if (oe->type != OBJ_COMMIT)
2614 die("Mark :%" PRIuMAX " not a commit", idnum);
2615 hashcpy(b->sha1, oe->idx.sha1);
2616 if (oe->pack_id != MAX_PACK_ID) {
2617 unsigned long size;
2618 char *buf = gfi_unpack_entry(oe, &size);
2619 parse_from_commit(b, buf, size);
2620 free(buf);
2621 } else
2622 parse_from_existing(b);
2623 } else if (!get_sha1(from, b->sha1))
2624 parse_from_existing(b);
2625 else
2626 die("Invalid ref name or SHA1 expression: %s", from);
2628 read_next_command();
2629 return 1;
2632 static struct hash_list *parse_merge(unsigned int *count)
2634 struct hash_list *list = NULL, **tail = &list, *n;
2635 const char *from;
2636 struct branch *s;
2638 *count = 0;
2639 while (!prefixcmp(command_buf.buf, "merge ")) {
2640 from = strchr(command_buf.buf, ' ') + 1;
2641 n = xmalloc(sizeof(*n));
2642 s = lookup_branch(from);
2643 if (s)
2644 hashcpy(n->sha1, s->sha1);
2645 else if (*from == ':') {
2646 uintmax_t idnum = parse_mark_ref_eol(from);
2647 struct object_entry *oe = find_mark(idnum);
2648 if (oe->type != OBJ_COMMIT)
2649 die("Mark :%" PRIuMAX " not a commit", idnum);
2650 hashcpy(n->sha1, oe->idx.sha1);
2651 } else if (!get_sha1(from, n->sha1)) {
2652 unsigned long size;
2653 char *buf = read_object_with_reference(n->sha1,
2654 commit_type, &size, n->sha1);
2655 if (!buf || size < 46)
2656 die("Not a valid commit: %s", from);
2657 free(buf);
2658 } else
2659 die("Invalid ref name or SHA1 expression: %s", from);
2661 n->next = NULL;
2662 *tail = n;
2663 tail = &n->next;
2665 (*count)++;
2666 read_next_command();
2668 return list;
2671 static void parse_new_commit(void)
2673 static struct strbuf msg = STRBUF_INIT;
2674 struct branch *b;
2675 char *sp;
2676 char *author = NULL;
2677 char *committer = NULL;
2678 struct hash_list *merge_list = NULL;
2679 unsigned int merge_count;
2680 unsigned char prev_fanout, new_fanout;
2682 /* Obtain the branch name from the rest of our command */
2683 sp = strchr(command_buf.buf, ' ') + 1;
2684 b = lookup_branch(sp);
2685 if (!b)
2686 b = new_branch(sp);
2688 read_next_command();
2689 parse_mark();
2690 if (!prefixcmp(command_buf.buf, "author ")) {
2691 author = parse_ident(command_buf.buf + 7);
2692 read_next_command();
2694 if (!prefixcmp(command_buf.buf, "committer ")) {
2695 committer = parse_ident(command_buf.buf + 10);
2696 read_next_command();
2698 if (!committer)
2699 die("Expected committer but didn't get one");
2700 parse_data(&msg, 0, NULL);
2701 read_next_command();
2702 parse_from(b);
2703 merge_list = parse_merge(&merge_count);
2705 /* ensure the branch is active/loaded */
2706 if (!b->branch_tree.tree || !max_active_branches) {
2707 unload_one_branch();
2708 load_branch(b);
2711 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2713 /* file_change* */
2714 while (command_buf.len > 0) {
2715 if (!prefixcmp(command_buf.buf, "M "))
2716 file_change_m(b);
2717 else if (!prefixcmp(command_buf.buf, "D "))
2718 file_change_d(b);
2719 else if (!prefixcmp(command_buf.buf, "R "))
2720 file_change_cr(b, 1);
2721 else if (!prefixcmp(command_buf.buf, "C "))
2722 file_change_cr(b, 0);
2723 else if (!prefixcmp(command_buf.buf, "N "))
2724 note_change_n(b, &prev_fanout);
2725 else if (!strcmp("deleteall", command_buf.buf))
2726 file_change_deleteall(b);
2727 else if (!prefixcmp(command_buf.buf, "ls "))
2728 parse_ls(b);
2729 else {
2730 unread_command_buf = 1;
2731 break;
2733 if (read_next_command() == EOF)
2734 break;
2737 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2738 if (new_fanout != prev_fanout)
2739 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2741 /* build the tree and the commit */
2742 store_tree(&b->branch_tree);
2743 hashcpy(b->branch_tree.versions[0].sha1,
2744 b->branch_tree.versions[1].sha1);
2746 strbuf_reset(&new_data);
2747 strbuf_addf(&new_data, "tree %s\n",
2748 sha1_to_hex(b->branch_tree.versions[1].sha1));
2749 if (!is_null_sha1(b->sha1))
2750 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2751 while (merge_list) {
2752 struct hash_list *next = merge_list->next;
2753 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2754 free(merge_list);
2755 merge_list = next;
2757 strbuf_addf(&new_data,
2758 "author %s\n"
2759 "committer %s\n"
2760 "\n",
2761 author ? author : committer, committer);
2762 strbuf_addbuf(&new_data, &msg);
2763 free(author);
2764 free(committer);
2766 if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2767 b->pack_id = pack_id;
2768 b->last_commit = object_count_by_type[OBJ_COMMIT];
2771 static void parse_new_tag(void)
2773 static struct strbuf msg = STRBUF_INIT;
2774 char *sp;
2775 const char *from;
2776 char *tagger;
2777 struct branch *s;
2778 struct tag *t;
2779 uintmax_t from_mark = 0;
2780 unsigned char sha1[20];
2781 enum object_type type;
2783 /* Obtain the new tag name from the rest of our command */
2784 sp = strchr(command_buf.buf, ' ') + 1;
2785 t = pool_alloc(sizeof(struct tag));
2786 memset(t, 0, sizeof(struct tag));
2787 t->name = pool_strdup(sp);
2788 if (last_tag)
2789 last_tag->next_tag = t;
2790 else
2791 first_tag = t;
2792 last_tag = t;
2793 read_next_command();
2795 /* from ... */
2796 if (prefixcmp(command_buf.buf, "from "))
2797 die("Expected from command, got %s", command_buf.buf);
2798 from = strchr(command_buf.buf, ' ') + 1;
2799 s = lookup_branch(from);
2800 if (s) {
2801 if (is_null_sha1(s->sha1))
2802 die("Can't tag an empty branch.");
2803 hashcpy(sha1, s->sha1);
2804 type = OBJ_COMMIT;
2805 } else if (*from == ':') {
2806 struct object_entry *oe;
2807 from_mark = parse_mark_ref_eol(from);
2808 oe = find_mark(from_mark);
2809 type = oe->type;
2810 hashcpy(sha1, oe->idx.sha1);
2811 } else if (!get_sha1(from, sha1)) {
2812 struct object_entry *oe = find_object(sha1);
2813 if (!oe) {
2814 type = sha1_object_info(sha1, NULL);
2815 if (type < 0)
2816 die("Not a valid object: %s", from);
2817 } else
2818 type = oe->type;
2819 } else
2820 die("Invalid ref name or SHA1 expression: %s", from);
2821 read_next_command();
2823 /* tagger ... */
2824 if (!prefixcmp(command_buf.buf, "tagger ")) {
2825 tagger = parse_ident(command_buf.buf + 7);
2826 read_next_command();
2827 } else
2828 tagger = NULL;
2830 /* tag payload/message */
2831 parse_data(&msg, 0, NULL);
2833 /* build the tag object */
2834 strbuf_reset(&new_data);
2836 strbuf_addf(&new_data,
2837 "object %s\n"
2838 "type %s\n"
2839 "tag %s\n",
2840 sha1_to_hex(sha1), typename(type), t->name);
2841 if (tagger)
2842 strbuf_addf(&new_data,
2843 "tagger %s\n", tagger);
2844 strbuf_addch(&new_data, '\n');
2845 strbuf_addbuf(&new_data, &msg);
2846 free(tagger);
2848 if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2849 t->pack_id = MAX_PACK_ID;
2850 else
2851 t->pack_id = pack_id;
2854 static void parse_reset_branch(void)
2856 struct branch *b;
2857 char *sp;
2859 /* Obtain the branch name from the rest of our command */
2860 sp = strchr(command_buf.buf, ' ') + 1;
2861 b = lookup_branch(sp);
2862 if (b) {
2863 hashclr(b->sha1);
2864 hashclr(b->branch_tree.versions[0].sha1);
2865 hashclr(b->branch_tree.versions[1].sha1);
2866 if (b->branch_tree.tree) {
2867 release_tree_content_recursive(b->branch_tree.tree);
2868 b->branch_tree.tree = NULL;
2871 else
2872 b = new_branch(sp);
2873 read_next_command();
2874 parse_from(b);
2875 if (command_buf.len > 0)
2876 unread_command_buf = 1;
2879 static void cat_blob_write(const char *buf, unsigned long size)
2881 if (write_in_full(cat_blob_fd, buf, size) != size)
2882 die_errno("Write to frontend failed");
2885 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2887 struct strbuf line = STRBUF_INIT;
2888 unsigned long size;
2889 enum object_type type = 0;
2890 char *buf;
2892 if (!oe || oe->pack_id == MAX_PACK_ID) {
2893 buf = read_sha1_file(sha1, &type, &size);
2894 } else {
2895 type = oe->type;
2896 buf = gfi_unpack_entry(oe, &size);
2900 * Output based on batch_one_object() from cat-file.c.
2902 if (type <= 0) {
2903 strbuf_reset(&line);
2904 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2905 cat_blob_write(line.buf, line.len);
2906 strbuf_release(&line);
2907 free(buf);
2908 return;
2910 if (!buf)
2911 die("Can't read object %s", sha1_to_hex(sha1));
2912 if (type != OBJ_BLOB)
2913 die("Object %s is a %s but a blob was expected.",
2914 sha1_to_hex(sha1), typename(type));
2915 strbuf_reset(&line);
2916 strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2917 typename(type), size);
2918 cat_blob_write(line.buf, line.len);
2919 strbuf_release(&line);
2920 cat_blob_write(buf, size);
2921 cat_blob_write("\n", 1);
2922 if (oe && oe->pack_id == pack_id) {
2923 last_blob.offset = oe->idx.offset;
2924 strbuf_attach(&last_blob.data, buf, size, size);
2925 last_blob.depth = oe->depth;
2926 } else
2927 free(buf);
2930 static void parse_cat_blob(void)
2932 const char *p;
2933 struct object_entry *oe = oe;
2934 unsigned char sha1[20];
2936 /* cat-blob SP <object> LF */
2937 p = command_buf.buf + strlen("cat-blob ");
2938 if (*p == ':') {
2939 oe = find_mark(parse_mark_ref_eol(p));
2940 if (!oe)
2941 die("Unknown mark: %s", command_buf.buf);
2942 hashcpy(sha1, oe->idx.sha1);
2943 } else {
2944 if (get_sha1_hex(p, sha1))
2945 die("Invalid dataref: %s", command_buf.buf);
2946 if (p[40])
2947 die("Garbage after SHA1: %s", command_buf.buf);
2948 oe = find_object(sha1);
2951 cat_blob(oe, sha1);
2954 static struct object_entry *dereference(struct object_entry *oe,
2955 unsigned char sha1[20])
2957 unsigned long size;
2958 char *buf = NULL;
2959 if (!oe) {
2960 enum object_type type = sha1_object_info(sha1, NULL);
2961 if (type < 0)
2962 die("object not found: %s", sha1_to_hex(sha1));
2963 /* cache it! */
2964 oe = insert_object(sha1);
2965 oe->type = type;
2966 oe->pack_id = MAX_PACK_ID;
2967 oe->idx.offset = 1;
2969 switch (oe->type) {
2970 case OBJ_TREE: /* easy case. */
2971 return oe;
2972 case OBJ_COMMIT:
2973 case OBJ_TAG:
2974 break;
2975 default:
2976 die("Not a tree-ish: %s", command_buf.buf);
2979 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
2980 buf = gfi_unpack_entry(oe, &size);
2981 } else {
2982 enum object_type unused;
2983 buf = read_sha1_file(sha1, &unused, &size);
2985 if (!buf)
2986 die("Can't load object %s", sha1_to_hex(sha1));
2988 /* Peel one layer. */
2989 switch (oe->type) {
2990 case OBJ_TAG:
2991 if (size < 40 + strlen("object ") ||
2992 get_sha1_hex(buf + strlen("object "), sha1))
2993 die("Invalid SHA1 in tag: %s", command_buf.buf);
2994 break;
2995 case OBJ_COMMIT:
2996 if (size < 40 + strlen("tree ") ||
2997 get_sha1_hex(buf + strlen("tree "), sha1))
2998 die("Invalid SHA1 in commit: %s", command_buf.buf);
3001 free(buf);
3002 return find_object(sha1);
3005 static struct object_entry *parse_treeish_dataref(const char **p)
3007 unsigned char sha1[20];
3008 struct object_entry *e;
3010 if (**p == ':') { /* <mark> */
3011 e = find_mark(parse_mark_ref_space(p));
3012 if (!e)
3013 die("Unknown mark: %s", command_buf.buf);
3014 hashcpy(sha1, e->idx.sha1);
3015 } else { /* <sha1> */
3016 if (get_sha1_hex(*p, sha1))
3017 die("Invalid dataref: %s", command_buf.buf);
3018 e = find_object(sha1);
3019 *p += 40;
3022 while (!e || e->type != OBJ_TREE)
3023 e = dereference(e, sha1);
3024 return e;
3027 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3029 static struct strbuf line = STRBUF_INIT;
3031 /* See show_tree(). */
3032 const char *type =
3033 S_ISGITLINK(mode) ? commit_type :
3034 S_ISDIR(mode) ? tree_type :
3035 blob_type;
3037 if (!mode) {
3038 /* missing SP path LF */
3039 strbuf_reset(&line);
3040 strbuf_addstr(&line, "missing ");
3041 quote_c_style(path, &line, NULL, 0);
3042 strbuf_addch(&line, '\n');
3043 } else {
3044 /* mode SP type SP object_name TAB path LF */
3045 strbuf_reset(&line);
3046 strbuf_addf(&line, "%06o %s %s\t",
3047 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3048 quote_c_style(path, &line, NULL, 0);
3049 strbuf_addch(&line, '\n');
3051 cat_blob_write(line.buf, line.len);
3054 static void parse_ls(struct branch *b)
3056 const char *p;
3057 struct tree_entry *root = NULL;
3058 struct tree_entry leaf = {NULL};
3060 /* ls SP (<tree-ish> SP)? <path> */
3061 p = command_buf.buf + strlen("ls ");
3062 if (*p == '"') {
3063 if (!b)
3064 die("Not in a commit: %s", command_buf.buf);
3065 root = &b->branch_tree;
3066 } else {
3067 struct object_entry *e = parse_treeish_dataref(&p);
3068 root = new_tree_entry();
3069 hashcpy(root->versions[1].sha1, e->idx.sha1);
3070 if (!is_null_sha1(root->versions[1].sha1))
3071 root->versions[1].mode = S_IFDIR;
3072 load_tree(root);
3073 if (*p++ != ' ')
3074 die("Missing space after tree-ish: %s", command_buf.buf);
3076 if (*p == '"') {
3077 static struct strbuf uq = STRBUF_INIT;
3078 const char *endp;
3079 strbuf_reset(&uq);
3080 if (unquote_c_style(&uq, p, &endp))
3081 die("Invalid path: %s", command_buf.buf);
3082 if (*endp)
3083 die("Garbage after path in: %s", command_buf.buf);
3084 p = uq.buf;
3086 tree_content_get(root, p, &leaf, 1);
3088 * A directory in preparation would have a sha1 of zero
3089 * until it is saved. Save, for simplicity.
3091 if (S_ISDIR(leaf.versions[1].mode))
3092 store_tree(&leaf);
3094 print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3095 if (leaf.tree)
3096 release_tree_content_recursive(leaf.tree);
3097 if (!b || root != &b->branch_tree)
3098 release_tree_entry(root);
3101 static void checkpoint(void)
3103 checkpoint_requested = 0;
3104 if (object_count) {
3105 cycle_packfile();
3106 dump_branches();
3107 dump_tags();
3108 dump_marks();
3112 static void parse_checkpoint(void)
3114 checkpoint_requested = 1;
3115 skip_optional_lf();
3118 static void parse_progress(void)
3120 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3121 fputc('\n', stdout);
3122 fflush(stdout);
3123 skip_optional_lf();
3126 static char* make_fast_import_path(const char *path)
3128 struct strbuf abs_path = STRBUF_INIT;
3130 if (!relative_marks_paths || is_absolute_path(path))
3131 return xstrdup(path);
3132 strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
3133 return strbuf_detach(&abs_path, NULL);
3136 static void option_import_marks(const char *marks,
3137 int from_stream, int ignore_missing)
3139 if (import_marks_file) {
3140 if (from_stream)
3141 die("Only one import-marks command allowed per stream");
3143 /* read previous mark file */
3144 if(!import_marks_file_from_stream)
3145 read_marks();
3148 import_marks_file = make_fast_import_path(marks);
3149 safe_create_leading_directories_const(import_marks_file);
3150 import_marks_file_from_stream = from_stream;
3151 import_marks_file_ignore_missing = ignore_missing;
3154 static void option_date_format(const char *fmt)
3156 if (!strcmp(fmt, "raw"))
3157 whenspec = WHENSPEC_RAW;
3158 else if (!strcmp(fmt, "rfc2822"))
3159 whenspec = WHENSPEC_RFC2822;
3160 else if (!strcmp(fmt, "now"))
3161 whenspec = WHENSPEC_NOW;
3162 else
3163 die("unknown --date-format argument %s", fmt);
3166 static unsigned long ulong_arg(const char *option, const char *arg)
3168 char *endptr;
3169 unsigned long rv = strtoul(arg, &endptr, 0);
3170 if (strchr(arg, '-') || endptr == arg || *endptr)
3171 die("%s: argument must be a non-negative integer", option);
3172 return rv;
3175 static void option_depth(const char *depth)
3177 max_depth = ulong_arg("--depth", depth);
3178 if (max_depth > MAX_DEPTH)
3179 die("--depth cannot exceed %u", MAX_DEPTH);
3182 static void option_active_branches(const char *branches)
3184 max_active_branches = ulong_arg("--active-branches", branches);
3187 static void option_export_marks(const char *marks)
3189 export_marks_file = make_fast_import_path(marks);
3190 safe_create_leading_directories_const(export_marks_file);
3193 static void option_cat_blob_fd(const char *fd)
3195 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3196 if (n > (unsigned long) INT_MAX)
3197 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3198 cat_blob_fd = (int) n;
3201 static void option_export_pack_edges(const char *edges)
3203 if (pack_edges)
3204 fclose(pack_edges);
3205 pack_edges = fopen(edges, "a");
3206 if (!pack_edges)
3207 die_errno("Cannot open '%s'", edges);
3210 static int parse_one_option(const char *option)
3212 if (!prefixcmp(option, "max-pack-size=")) {
3213 unsigned long v;
3214 if (!git_parse_ulong(option + 14, &v))
3215 return 0;
3216 if (v < 8192) {
3217 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3218 v *= 1024 * 1024;
3219 } else if (v < 1024 * 1024) {
3220 warning("minimum max-pack-size is 1 MiB");
3221 v = 1024 * 1024;
3223 max_packsize = v;
3224 } else if (!prefixcmp(option, "big-file-threshold=")) {
3225 unsigned long v;
3226 if (!git_parse_ulong(option + 19, &v))
3227 return 0;
3228 big_file_threshold = v;
3229 } else if (!prefixcmp(option, "depth=")) {
3230 option_depth(option + 6);
3231 } else if (!prefixcmp(option, "active-branches=")) {
3232 option_active_branches(option + 16);
3233 } else if (!prefixcmp(option, "export-pack-edges=")) {
3234 option_export_pack_edges(option + 18);
3235 } else if (!prefixcmp(option, "quiet")) {
3236 show_stats = 0;
3237 } else if (!prefixcmp(option, "stats")) {
3238 show_stats = 1;
3239 } else {
3240 return 0;
3243 return 1;
3246 static int parse_one_feature(const char *feature, int from_stream)
3248 if (!prefixcmp(feature, "date-format=")) {
3249 option_date_format(feature + 12);
3250 } else if (!prefixcmp(feature, "import-marks=")) {
3251 option_import_marks(feature + 13, from_stream, 0);
3252 } else if (!prefixcmp(feature, "import-marks-if-exists=")) {
3253 option_import_marks(feature + strlen("import-marks-if-exists="),
3254 from_stream, 1);
3255 } else if (!prefixcmp(feature, "export-marks=")) {
3256 option_export_marks(feature + 13);
3257 } else if (!strcmp(feature, "cat-blob")) {
3258 ; /* Don't die - this feature is supported */
3259 } else if (!strcmp(feature, "relative-marks")) {
3260 relative_marks_paths = 1;
3261 } else if (!strcmp(feature, "no-relative-marks")) {
3262 relative_marks_paths = 0;
3263 } else if (!strcmp(feature, "done")) {
3264 require_explicit_termination = 1;
3265 } else if (!strcmp(feature, "force")) {
3266 force_update = 1;
3267 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3268 ; /* do nothing; we have the feature */
3269 } else {
3270 return 0;
3273 return 1;
3276 static void parse_feature(void)
3278 char *feature = command_buf.buf + 8;
3280 if (seen_data_command)
3281 die("Got feature command '%s' after data command", feature);
3283 if (parse_one_feature(feature, 1))
3284 return;
3286 die("This version of fast-import does not support feature %s.", feature);
3289 static void parse_option(void)
3291 char *option = command_buf.buf + 11;
3293 if (seen_data_command)
3294 die("Got option command '%s' after data command", option);
3296 if (parse_one_option(option))
3297 return;
3299 die("This version of fast-import does not support option: %s", option);
3302 static int git_pack_config(const char *k, const char *v, void *cb)
3304 if (!strcmp(k, "pack.depth")) {
3305 max_depth = git_config_int(k, v);
3306 if (max_depth > MAX_DEPTH)
3307 max_depth = MAX_DEPTH;
3308 return 0;
3310 if (!strcmp(k, "pack.compression")) {
3311 int level = git_config_int(k, v);
3312 if (level == -1)
3313 level = Z_DEFAULT_COMPRESSION;
3314 else if (level < 0 || level > Z_BEST_COMPRESSION)
3315 die("bad pack compression level %d", level);
3316 pack_compression_level = level;
3317 pack_compression_seen = 1;
3318 return 0;
3320 if (!strcmp(k, "pack.indexversion")) {
3321 pack_idx_opts.version = git_config_int(k, v);
3322 if (pack_idx_opts.version > 2)
3323 die("bad pack.indexversion=%"PRIu32,
3324 pack_idx_opts.version);
3325 return 0;
3327 if (!strcmp(k, "pack.packsizelimit")) {
3328 max_packsize = git_config_ulong(k, v);
3329 return 0;
3331 return git_default_config(k, v, cb);
3334 static const char fast_import_usage[] =
3335 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3337 static void parse_argv(void)
3339 unsigned int i;
3341 for (i = 1; i < global_argc; i++) {
3342 const char *a = global_argv[i];
3344 if (*a != '-' || !strcmp(a, "--"))
3345 break;
3347 if (parse_one_option(a + 2))
3348 continue;
3350 if (parse_one_feature(a + 2, 0))
3351 continue;
3353 if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3354 option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3355 continue;
3358 die("unknown option %s", a);
3360 if (i != global_argc)
3361 usage(fast_import_usage);
3363 seen_data_command = 1;
3364 if (import_marks_file)
3365 read_marks();
3368 int main(int argc, char **argv)
3370 unsigned int i;
3372 git_extract_argv0_path(argv[0]);
3374 git_setup_gettext();
3376 if (argc == 2 && !strcmp(argv[1], "-h"))
3377 usage(fast_import_usage);
3379 setup_git_directory();
3380 reset_pack_idx_option(&pack_idx_opts);
3381 git_config(git_pack_config, NULL);
3382 if (!pack_compression_seen && core_compression_seen)
3383 pack_compression_level = core_compression_level;
3385 alloc_objects(object_entry_alloc);
3386 strbuf_init(&command_buf, 0);
3387 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3388 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3389 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3390 marks = pool_calloc(1, sizeof(struct mark_set));
3392 global_argc = argc;
3393 global_argv = argv;
3395 rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3396 for (i = 0; i < (cmd_save - 1); i++)
3397 rc_free[i].next = &rc_free[i + 1];
3398 rc_free[cmd_save - 1].next = NULL;
3400 prepare_packed_git();
3401 start_packfile();
3402 set_die_routine(die_nicely);
3403 set_checkpoint_signal();
3404 while (read_next_command() != EOF) {
3405 if (!strcmp("blob", command_buf.buf))
3406 parse_new_blob();
3407 else if (!prefixcmp(command_buf.buf, "ls "))
3408 parse_ls(NULL);
3409 else if (!prefixcmp(command_buf.buf, "commit "))
3410 parse_new_commit();
3411 else if (!prefixcmp(command_buf.buf, "tag "))
3412 parse_new_tag();
3413 else if (!prefixcmp(command_buf.buf, "reset "))
3414 parse_reset_branch();
3415 else if (!strcmp("checkpoint", command_buf.buf))
3416 parse_checkpoint();
3417 else if (!strcmp("done", command_buf.buf))
3418 break;
3419 else if (!prefixcmp(command_buf.buf, "progress "))
3420 parse_progress();
3421 else if (!prefixcmp(command_buf.buf, "feature "))
3422 parse_feature();
3423 else if (!prefixcmp(command_buf.buf, "option git "))
3424 parse_option();
3425 else if (!prefixcmp(command_buf.buf, "option "))
3426 /* ignore non-git options*/;
3427 else
3428 die("Unsupported command: %s", command_buf.buf);
3430 if (checkpoint_requested)
3431 checkpoint();
3434 /* argv hasn't been parsed yet, do so */
3435 if (!seen_data_command)
3436 parse_argv();
3438 if (require_explicit_termination && feof(stdin))
3439 die("stream ends early");
3441 end_packfile();
3443 dump_branches();
3444 dump_tags();
3445 unkeep_all_packs();
3446 dump_marks();
3448 if (pack_edges)
3449 fclose(pack_edges);
3451 if (show_stats) {
3452 uintmax_t total_count = 0, duplicate_count = 0;
3453 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3454 total_count += object_count_by_type[i];
3455 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3456 duplicate_count += duplicate_count_by_type[i];
3458 fprintf(stderr, "%s statistics:\n", argv[0]);
3459 fprintf(stderr, "---------------------------------------------------------------------\n");
3460 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3461 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3462 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]);
3463 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]);
3464 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]);
3465 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]);
3466 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3467 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3468 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3469 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3470 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)(total_allocd/1024));
3471 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3472 fprintf(stderr, "---------------------------------------------------------------------\n");
3473 pack_report();
3474 fprintf(stderr, "---------------------------------------------------------------------\n");
3475 fprintf(stderr, "\n");
3478 return failure ? 1 : 0;