use skip_prefix to avoid magic numbers
[git.git] / fast-import.c
bloba3ffe3063c3b89ad20af5d9e32903d3b76607e53
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 delete : 1;
252 unsigned pack_id : PACK_ID_BITS;
253 unsigned char sha1[20];
256 struct tag {
257 struct tag *next_tag;
258 const char *name;
259 unsigned int pack_id;
260 unsigned char sha1[20];
263 struct hash_list {
264 struct hash_list *next;
265 unsigned char sha1[20];
268 typedef enum {
269 WHENSPEC_RAW = 1,
270 WHENSPEC_RFC2822,
271 WHENSPEC_NOW
272 } whenspec_type;
274 struct recent_command {
275 struct recent_command *prev;
276 struct recent_command *next;
277 char *buf;
280 /* Configured limits on output */
281 static unsigned long max_depth = 10;
282 static off_t max_packsize;
283 static int force_update;
284 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
285 static int pack_compression_seen;
287 /* Stats and misc. counters */
288 static uintmax_t alloc_count;
289 static uintmax_t marks_set_count;
290 static uintmax_t object_count_by_type[1 << TYPE_BITS];
291 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
292 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
293 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
294 static unsigned long object_count;
295 static unsigned long branch_count;
296 static unsigned long branch_load_count;
297 static int failure;
298 static FILE *pack_edges;
299 static unsigned int show_stats = 1;
300 static int global_argc;
301 static char **global_argv;
303 /* Memory pools */
304 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
305 static size_t total_allocd;
306 static struct mem_pool *mem_pool;
308 /* Atom management */
309 static unsigned int atom_table_sz = 4451;
310 static unsigned int atom_cnt;
311 static struct atom_str **atom_table;
313 /* The .pack file being generated */
314 static struct pack_idx_option pack_idx_opts;
315 static unsigned int pack_id;
316 static struct sha1file *pack_file;
317 static struct packed_git *pack_data;
318 static struct packed_git **all_packs;
319 static off_t pack_size;
321 /* Table of objects we've written. */
322 static unsigned int object_entry_alloc = 5000;
323 static struct object_entry_pool *blocks;
324 static struct object_entry *object_table[1 << 16];
325 static struct mark_set *marks;
326 static const char *export_marks_file;
327 static const char *import_marks_file;
328 static int import_marks_file_from_stream;
329 static int import_marks_file_ignore_missing;
330 static int relative_marks_paths;
332 /* Our last blob */
333 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
335 /* Tree management */
336 static unsigned int tree_entry_alloc = 1000;
337 static void *avail_tree_entry;
338 static unsigned int avail_tree_table_sz = 100;
339 static struct avail_tree_content **avail_tree_table;
340 static struct strbuf old_tree = STRBUF_INIT;
341 static struct strbuf new_tree = STRBUF_INIT;
343 /* Branch data */
344 static unsigned long max_active_branches = 5;
345 static unsigned long cur_active_branches;
346 static unsigned long branch_table_sz = 1039;
347 static struct branch **branch_table;
348 static struct branch *active_branches;
350 /* Tag data */
351 static struct tag *first_tag;
352 static struct tag *last_tag;
354 /* Input stream parsing */
355 static whenspec_type whenspec = WHENSPEC_RAW;
356 static struct strbuf command_buf = STRBUF_INIT;
357 static int unread_command_buf;
358 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
359 static struct recent_command *cmd_tail = &cmd_hist;
360 static struct recent_command *rc_free;
361 static unsigned int cmd_save = 100;
362 static uintmax_t next_mark;
363 static struct strbuf new_data = STRBUF_INIT;
364 static int seen_data_command;
365 static int require_explicit_termination;
367 /* Signal handling */
368 static volatile sig_atomic_t checkpoint_requested;
370 /* Where to write output of cat-blob commands */
371 static int cat_blob_fd = STDOUT_FILENO;
373 static void parse_argv(void);
374 static void parse_cat_blob(void);
375 static void parse_ls(struct branch *b);
377 static void write_branch_report(FILE *rpt, struct branch *b)
379 fprintf(rpt, "%s:\n", b->name);
381 fprintf(rpt, " status :");
382 if (b->active)
383 fputs(" active", rpt);
384 if (b->branch_tree.tree)
385 fputs(" loaded", rpt);
386 if (is_null_sha1(b->branch_tree.versions[1].sha1))
387 fputs(" dirty", rpt);
388 fputc('\n', rpt);
390 fprintf(rpt, " tip commit : %s\n", sha1_to_hex(b->sha1));
391 fprintf(rpt, " old tree : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
392 fprintf(rpt, " cur tree : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
393 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
395 fputs(" last pack : ", rpt);
396 if (b->pack_id < MAX_PACK_ID)
397 fprintf(rpt, "%u", b->pack_id);
398 fputc('\n', rpt);
400 fputc('\n', rpt);
403 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
405 static void write_crash_report(const char *err)
407 char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
408 FILE *rpt = fopen(loc, "w");
409 struct branch *b;
410 unsigned long lu;
411 struct recent_command *rc;
413 if (!rpt) {
414 error("can't write crash report %s: %s", loc, strerror(errno));
415 return;
418 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
420 fprintf(rpt, "fast-import crash report:\n");
421 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
422 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
423 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
424 fputc('\n', rpt);
426 fputs("fatal: ", rpt);
427 fputs(err, rpt);
428 fputc('\n', rpt);
430 fputc('\n', rpt);
431 fputs("Most Recent Commands Before Crash\n", rpt);
432 fputs("---------------------------------\n", rpt);
433 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
434 if (rc->next == &cmd_hist)
435 fputs("* ", rpt);
436 else
437 fputs(" ", rpt);
438 fputs(rc->buf, rpt);
439 fputc('\n', rpt);
442 fputc('\n', rpt);
443 fputs("Active Branch LRU\n", rpt);
444 fputs("-----------------\n", rpt);
445 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
446 cur_active_branches,
447 max_active_branches);
448 fputc('\n', rpt);
449 fputs(" pos clock name\n", rpt);
450 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
451 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
452 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
453 ++lu, b->last_commit, b->name);
455 fputc('\n', rpt);
456 fputs("Inactive Branches\n", rpt);
457 fputs("-----------------\n", rpt);
458 for (lu = 0; lu < branch_table_sz; lu++) {
459 for (b = branch_table[lu]; b; b = b->table_next_branch)
460 write_branch_report(rpt, b);
463 if (first_tag) {
464 struct tag *tg;
465 fputc('\n', rpt);
466 fputs("Annotated Tags\n", rpt);
467 fputs("--------------\n", rpt);
468 for (tg = first_tag; tg; tg = tg->next_tag) {
469 fputs(sha1_to_hex(tg->sha1), rpt);
470 fputc(' ', rpt);
471 fputs(tg->name, rpt);
472 fputc('\n', rpt);
476 fputc('\n', rpt);
477 fputs("Marks\n", rpt);
478 fputs("-----\n", rpt);
479 if (export_marks_file)
480 fprintf(rpt, " exported to %s\n", export_marks_file);
481 else
482 dump_marks_helper(rpt, 0, marks);
484 fputc('\n', rpt);
485 fputs("-------------------\n", rpt);
486 fputs("END OF CRASH REPORT\n", rpt);
487 fclose(rpt);
490 static void end_packfile(void);
491 static void unkeep_all_packs(void);
492 static void dump_marks(void);
494 static NORETURN void die_nicely(const char *err, va_list params)
496 static int zombie;
497 char message[2 * PATH_MAX];
499 vsnprintf(message, sizeof(message), err, params);
500 fputs("fatal: ", stderr);
501 fputs(message, stderr);
502 fputc('\n', stderr);
504 if (!zombie) {
505 zombie = 1;
506 write_crash_report(message);
507 end_packfile();
508 unkeep_all_packs();
509 dump_marks();
511 exit(128);
514 #ifndef SIGUSR1 /* Windows, for example */
516 static void set_checkpoint_signal(void)
520 #else
522 static void checkpoint_signal(int signo)
524 checkpoint_requested = 1;
527 static void set_checkpoint_signal(void)
529 struct sigaction sa;
531 memset(&sa, 0, sizeof(sa));
532 sa.sa_handler = checkpoint_signal;
533 sigemptyset(&sa.sa_mask);
534 sa.sa_flags = SA_RESTART;
535 sigaction(SIGUSR1, &sa, NULL);
538 #endif
540 static void alloc_objects(unsigned int cnt)
542 struct object_entry_pool *b;
544 b = xmalloc(sizeof(struct object_entry_pool)
545 + cnt * sizeof(struct object_entry));
546 b->next_pool = blocks;
547 b->next_free = b->entries;
548 b->end = b->entries + cnt;
549 blocks = b;
550 alloc_count += cnt;
553 static struct object_entry *new_object(unsigned char *sha1)
555 struct object_entry *e;
557 if (blocks->next_free == blocks->end)
558 alloc_objects(object_entry_alloc);
560 e = blocks->next_free++;
561 hashcpy(e->idx.sha1, sha1);
562 return e;
565 static struct object_entry *find_object(unsigned char *sha1)
567 unsigned int h = sha1[0] << 8 | sha1[1];
568 struct object_entry *e;
569 for (e = object_table[h]; e; e = e->next)
570 if (!hashcmp(sha1, e->idx.sha1))
571 return e;
572 return NULL;
575 static struct object_entry *insert_object(unsigned char *sha1)
577 unsigned int h = sha1[0] << 8 | sha1[1];
578 struct object_entry *e = object_table[h];
580 while (e) {
581 if (!hashcmp(sha1, e->idx.sha1))
582 return e;
583 e = e->next;
586 e = new_object(sha1);
587 e->next = object_table[h];
588 e->idx.offset = 0;
589 object_table[h] = e;
590 return e;
593 static unsigned int hc_str(const char *s, size_t len)
595 unsigned int r = 0;
596 while (len-- > 0)
597 r = r * 31 + *s++;
598 return r;
601 static void *pool_alloc(size_t len)
603 struct mem_pool *p;
604 void *r;
606 /* round up to a 'uintmax_t' alignment */
607 if (len & (sizeof(uintmax_t) - 1))
608 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
610 for (p = mem_pool; p; p = p->next_pool)
611 if ((p->end - p->next_free >= len))
612 break;
614 if (!p) {
615 if (len >= (mem_pool_alloc/2)) {
616 total_allocd += len;
617 return xmalloc(len);
619 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
620 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
621 p->next_pool = mem_pool;
622 p->next_free = (char *) p->space;
623 p->end = p->next_free + mem_pool_alloc;
624 mem_pool = p;
627 r = p->next_free;
628 p->next_free += len;
629 return r;
632 static void *pool_calloc(size_t count, size_t size)
634 size_t len = count * size;
635 void *r = pool_alloc(len);
636 memset(r, 0, len);
637 return r;
640 static char *pool_strdup(const char *s)
642 char *r = pool_alloc(strlen(s) + 1);
643 strcpy(r, s);
644 return r;
647 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
649 struct mark_set *s = marks;
650 while ((idnum >> s->shift) >= 1024) {
651 s = pool_calloc(1, sizeof(struct mark_set));
652 s->shift = marks->shift + 10;
653 s->data.sets[0] = marks;
654 marks = s;
656 while (s->shift) {
657 uintmax_t i = idnum >> s->shift;
658 idnum -= i << s->shift;
659 if (!s->data.sets[i]) {
660 s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
661 s->data.sets[i]->shift = s->shift - 10;
663 s = s->data.sets[i];
665 if (!s->data.marked[idnum])
666 marks_set_count++;
667 s->data.marked[idnum] = oe;
670 static struct object_entry *find_mark(uintmax_t idnum)
672 uintmax_t orig_idnum = idnum;
673 struct mark_set *s = marks;
674 struct object_entry *oe = NULL;
675 if ((idnum >> s->shift) < 1024) {
676 while (s && s->shift) {
677 uintmax_t i = idnum >> s->shift;
678 idnum -= i << s->shift;
679 s = s->data.sets[i];
681 if (s)
682 oe = s->data.marked[idnum];
684 if (!oe)
685 die("mark :%" PRIuMAX " not declared", orig_idnum);
686 return oe;
689 static struct atom_str *to_atom(const char *s, unsigned short len)
691 unsigned int hc = hc_str(s, len) % atom_table_sz;
692 struct atom_str *c;
694 for (c = atom_table[hc]; c; c = c->next_atom)
695 if (c->str_len == len && !strncmp(s, c->str_dat, len))
696 return c;
698 c = pool_alloc(sizeof(struct atom_str) + len + 1);
699 c->str_len = len;
700 strncpy(c->str_dat, s, len);
701 c->str_dat[len] = 0;
702 c->next_atom = atom_table[hc];
703 atom_table[hc] = c;
704 atom_cnt++;
705 return c;
708 static struct branch *lookup_branch(const char *name)
710 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
711 struct branch *b;
713 for (b = branch_table[hc]; b; b = b->table_next_branch)
714 if (!strcmp(name, b->name))
715 return b;
716 return NULL;
719 static struct branch *new_branch(const char *name)
721 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
722 struct branch *b = lookup_branch(name);
724 if (b)
725 die("Invalid attempt to create duplicate branch: %s", name);
726 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
727 die("Branch name doesn't conform to GIT standards: %s", name);
729 b = pool_calloc(1, sizeof(struct branch));
730 b->name = pool_strdup(name);
731 b->table_next_branch = branch_table[hc];
732 b->branch_tree.versions[0].mode = S_IFDIR;
733 b->branch_tree.versions[1].mode = S_IFDIR;
734 b->num_notes = 0;
735 b->active = 0;
736 b->pack_id = MAX_PACK_ID;
737 branch_table[hc] = b;
738 branch_count++;
739 return b;
742 static unsigned int hc_entries(unsigned int cnt)
744 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
745 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
748 static struct tree_content *new_tree_content(unsigned int cnt)
750 struct avail_tree_content *f, *l = NULL;
751 struct tree_content *t;
752 unsigned int hc = hc_entries(cnt);
754 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
755 if (f->entry_capacity >= cnt)
756 break;
758 if (f) {
759 if (l)
760 l->next_avail = f->next_avail;
761 else
762 avail_tree_table[hc] = f->next_avail;
763 } else {
764 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
765 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
766 f->entry_capacity = cnt;
769 t = (struct tree_content*)f;
770 t->entry_count = 0;
771 t->delta_depth = 0;
772 return t;
775 static void release_tree_entry(struct tree_entry *e);
776 static void release_tree_content(struct tree_content *t)
778 struct avail_tree_content *f = (struct avail_tree_content*)t;
779 unsigned int hc = hc_entries(f->entry_capacity);
780 f->next_avail = avail_tree_table[hc];
781 avail_tree_table[hc] = f;
784 static void release_tree_content_recursive(struct tree_content *t)
786 unsigned int i;
787 for (i = 0; i < t->entry_count; i++)
788 release_tree_entry(t->entries[i]);
789 release_tree_content(t);
792 static struct tree_content *grow_tree_content(
793 struct tree_content *t,
794 int amt)
796 struct tree_content *r = new_tree_content(t->entry_count + amt);
797 r->entry_count = t->entry_count;
798 r->delta_depth = t->delta_depth;
799 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
800 release_tree_content(t);
801 return r;
804 static struct tree_entry *new_tree_entry(void)
806 struct tree_entry *e;
808 if (!avail_tree_entry) {
809 unsigned int n = tree_entry_alloc;
810 total_allocd += n * sizeof(struct tree_entry);
811 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
812 while (n-- > 1) {
813 *((void**)e) = e + 1;
814 e++;
816 *((void**)e) = NULL;
819 e = avail_tree_entry;
820 avail_tree_entry = *((void**)e);
821 return e;
824 static void release_tree_entry(struct tree_entry *e)
826 if (e->tree)
827 release_tree_content_recursive(e->tree);
828 *((void**)e) = avail_tree_entry;
829 avail_tree_entry = e;
832 static struct tree_content *dup_tree_content(struct tree_content *s)
834 struct tree_content *d;
835 struct tree_entry *a, *b;
836 unsigned int i;
838 if (!s)
839 return NULL;
840 d = new_tree_content(s->entry_count);
841 for (i = 0; i < s->entry_count; i++) {
842 a = s->entries[i];
843 b = new_tree_entry();
844 memcpy(b, a, sizeof(*a));
845 if (a->tree && is_null_sha1(b->versions[1].sha1))
846 b->tree = dup_tree_content(a->tree);
847 else
848 b->tree = NULL;
849 d->entries[i] = b;
851 d->entry_count = s->entry_count;
852 d->delta_depth = s->delta_depth;
854 return d;
857 static void start_packfile(void)
859 static char tmp_file[PATH_MAX];
860 struct packed_git *p;
861 struct pack_header hdr;
862 int pack_fd;
864 pack_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
865 "pack/tmp_pack_XXXXXX");
866 p = xcalloc(1, sizeof(*p) + strlen(tmp_file) + 2);
867 strcpy(p->pack_name, tmp_file);
868 p->pack_fd = pack_fd;
869 p->do_not_close = 1;
870 pack_file = sha1fd(pack_fd, p->pack_name);
872 hdr.hdr_signature = htonl(PACK_SIGNATURE);
873 hdr.hdr_version = htonl(2);
874 hdr.hdr_entries = 0;
875 sha1write(pack_file, &hdr, sizeof(hdr));
877 pack_data = p;
878 pack_size = sizeof(hdr);
879 object_count = 0;
881 all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
882 all_packs[pack_id] = p;
885 static const char *create_index(void)
887 const char *tmpfile;
888 struct pack_idx_entry **idx, **c, **last;
889 struct object_entry *e;
890 struct object_entry_pool *o;
892 /* Build the table of object IDs. */
893 idx = xmalloc(object_count * sizeof(*idx));
894 c = idx;
895 for (o = blocks; o; o = o->next_pool)
896 for (e = o->next_free; e-- != o->entries;)
897 if (pack_id == e->pack_id)
898 *c++ = &e->idx;
899 last = idx + object_count;
900 if (c != last)
901 die("internal consistency error creating the index");
903 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
904 free(idx);
905 return tmpfile;
908 static char *keep_pack(const char *curr_index_name)
910 static char name[PATH_MAX];
911 static const char *keep_msg = "fast-import";
912 int keep_fd;
914 keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
915 if (keep_fd < 0)
916 die_errno("cannot create keep file");
917 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
918 if (close(keep_fd))
919 die_errno("failed to write keep file");
921 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
922 get_object_directory(), sha1_to_hex(pack_data->sha1));
923 if (move_temp_to_file(pack_data->pack_name, name))
924 die("cannot store pack file");
926 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
927 get_object_directory(), sha1_to_hex(pack_data->sha1));
928 if (move_temp_to_file(curr_index_name, name))
929 die("cannot store index file");
930 free((void *)curr_index_name);
931 return name;
934 static void unkeep_all_packs(void)
936 static char name[PATH_MAX];
937 int k;
939 for (k = 0; k < pack_id; k++) {
940 struct packed_git *p = all_packs[k];
941 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
942 get_object_directory(), sha1_to_hex(p->sha1));
943 unlink_or_warn(name);
947 static void end_packfile(void)
949 struct packed_git *old_p = pack_data, *new_p;
951 clear_delta_base_cache();
952 if (object_count) {
953 unsigned char cur_pack_sha1[20];
954 char *idx_name;
955 int i;
956 struct branch *b;
957 struct tag *t;
959 close_pack_windows(pack_data);
960 sha1close(pack_file, cur_pack_sha1, 0);
961 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
962 pack_data->pack_name, object_count,
963 cur_pack_sha1, pack_size);
964 close(pack_data->pack_fd);
965 idx_name = keep_pack(create_index());
967 /* Register the packfile with core git's machinery. */
968 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
969 if (!new_p)
970 die("core git rejected index %s", idx_name);
971 all_packs[pack_id] = new_p;
972 install_packed_git(new_p);
974 /* Print the boundary */
975 if (pack_edges) {
976 fprintf(pack_edges, "%s:", new_p->pack_name);
977 for (i = 0; i < branch_table_sz; i++) {
978 for (b = branch_table[i]; b; b = b->table_next_branch) {
979 if (b->pack_id == pack_id)
980 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
983 for (t = first_tag; t; t = t->next_tag) {
984 if (t->pack_id == pack_id)
985 fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
987 fputc('\n', pack_edges);
988 fflush(pack_edges);
991 pack_id++;
993 else {
994 close(old_p->pack_fd);
995 unlink_or_warn(old_p->pack_name);
997 free(old_p);
999 /* We can't carry a delta across packfiles. */
1000 strbuf_release(&last_blob.data);
1001 last_blob.offset = 0;
1002 last_blob.depth = 0;
1005 static void cycle_packfile(void)
1007 end_packfile();
1008 start_packfile();
1011 static int store_object(
1012 enum object_type type,
1013 struct strbuf *dat,
1014 struct last_object *last,
1015 unsigned char *sha1out,
1016 uintmax_t mark)
1018 void *out, *delta;
1019 struct object_entry *e;
1020 unsigned char hdr[96];
1021 unsigned char sha1[20];
1022 unsigned long hdrlen, deltalen;
1023 git_SHA_CTX c;
1024 git_zstream s;
1026 hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1027 (unsigned long)dat->len) + 1;
1028 git_SHA1_Init(&c);
1029 git_SHA1_Update(&c, hdr, hdrlen);
1030 git_SHA1_Update(&c, dat->buf, dat->len);
1031 git_SHA1_Final(sha1, &c);
1032 if (sha1out)
1033 hashcpy(sha1out, sha1);
1035 e = insert_object(sha1);
1036 if (mark)
1037 insert_mark(mark, e);
1038 if (e->idx.offset) {
1039 duplicate_count_by_type[type]++;
1040 return 1;
1041 } else if (find_sha1_pack(sha1, packed_git)) {
1042 e->type = type;
1043 e->pack_id = MAX_PACK_ID;
1044 e->idx.offset = 1; /* just not zero! */
1045 duplicate_count_by_type[type]++;
1046 return 1;
1049 if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1050 delta_count_attempts_by_type[type]++;
1051 delta = diff_delta(last->data.buf, last->data.len,
1052 dat->buf, dat->len,
1053 &deltalen, dat->len - 20);
1054 } else
1055 delta = NULL;
1057 memset(&s, 0, sizeof(s));
1058 git_deflate_init(&s, pack_compression_level);
1059 if (delta) {
1060 s.next_in = delta;
1061 s.avail_in = deltalen;
1062 } else {
1063 s.next_in = (void *)dat->buf;
1064 s.avail_in = dat->len;
1066 s.avail_out = git_deflate_bound(&s, s.avail_in);
1067 s.next_out = out = xmalloc(s.avail_out);
1068 while (git_deflate(&s, Z_FINISH) == Z_OK)
1069 ; /* nothing */
1070 git_deflate_end(&s);
1072 /* Determine if we should auto-checkpoint. */
1073 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1074 || (pack_size + 60 + s.total_out) < pack_size) {
1076 /* This new object needs to *not* have the current pack_id. */
1077 e->pack_id = pack_id + 1;
1078 cycle_packfile();
1080 /* We cannot carry a delta into the new pack. */
1081 if (delta) {
1082 free(delta);
1083 delta = NULL;
1085 memset(&s, 0, sizeof(s));
1086 git_deflate_init(&s, pack_compression_level);
1087 s.next_in = (void *)dat->buf;
1088 s.avail_in = dat->len;
1089 s.avail_out = git_deflate_bound(&s, s.avail_in);
1090 s.next_out = out = xrealloc(out, s.avail_out);
1091 while (git_deflate(&s, Z_FINISH) == Z_OK)
1092 ; /* nothing */
1093 git_deflate_end(&s);
1097 e->type = type;
1098 e->pack_id = pack_id;
1099 e->idx.offset = pack_size;
1100 object_count++;
1101 object_count_by_type[type]++;
1103 crc32_begin(pack_file);
1105 if (delta) {
1106 off_t ofs = e->idx.offset - last->offset;
1107 unsigned pos = sizeof(hdr) - 1;
1109 delta_count_by_type[type]++;
1110 e->depth = last->depth + 1;
1112 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1113 sha1write(pack_file, hdr, hdrlen);
1114 pack_size += hdrlen;
1116 hdr[pos] = ofs & 127;
1117 while (ofs >>= 7)
1118 hdr[--pos] = 128 | (--ofs & 127);
1119 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1120 pack_size += sizeof(hdr) - pos;
1121 } else {
1122 e->depth = 0;
1123 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1124 sha1write(pack_file, hdr, hdrlen);
1125 pack_size += hdrlen;
1128 sha1write(pack_file, out, s.total_out);
1129 pack_size += s.total_out;
1131 e->idx.crc32 = crc32_end(pack_file);
1133 free(out);
1134 free(delta);
1135 if (last) {
1136 if (last->no_swap) {
1137 last->data = *dat;
1138 } else {
1139 strbuf_swap(&last->data, dat);
1141 last->offset = e->idx.offset;
1142 last->depth = e->depth;
1144 return 0;
1147 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1149 if (sha1file_truncate(pack_file, checkpoint))
1150 die_errno("cannot truncate pack to skip duplicate");
1151 pack_size = checkpoint->offset;
1154 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1156 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1157 unsigned char *in_buf = xmalloc(in_sz);
1158 unsigned char *out_buf = xmalloc(out_sz);
1159 struct object_entry *e;
1160 unsigned char sha1[20];
1161 unsigned long hdrlen;
1162 off_t offset;
1163 git_SHA_CTX c;
1164 git_zstream s;
1165 struct sha1file_checkpoint checkpoint;
1166 int status = Z_OK;
1168 /* Determine if we should auto-checkpoint. */
1169 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1170 || (pack_size + 60 + len) < pack_size)
1171 cycle_packfile();
1173 sha1file_checkpoint(pack_file, &checkpoint);
1174 offset = checkpoint.offset;
1176 hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1177 if (out_sz <= hdrlen)
1178 die("impossibly large object header");
1180 git_SHA1_Init(&c);
1181 git_SHA1_Update(&c, out_buf, hdrlen);
1183 crc32_begin(pack_file);
1185 memset(&s, 0, sizeof(s));
1186 git_deflate_init(&s, pack_compression_level);
1188 hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1189 if (out_sz <= hdrlen)
1190 die("impossibly large object header");
1192 s.next_out = out_buf + hdrlen;
1193 s.avail_out = out_sz - hdrlen;
1195 while (status != Z_STREAM_END) {
1196 if (0 < len && !s.avail_in) {
1197 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1198 size_t n = fread(in_buf, 1, cnt, stdin);
1199 if (!n && feof(stdin))
1200 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1202 git_SHA1_Update(&c, in_buf, n);
1203 s.next_in = in_buf;
1204 s.avail_in = n;
1205 len -= n;
1208 status = git_deflate(&s, len ? 0 : Z_FINISH);
1210 if (!s.avail_out || status == Z_STREAM_END) {
1211 size_t n = s.next_out - out_buf;
1212 sha1write(pack_file, out_buf, n);
1213 pack_size += n;
1214 s.next_out = out_buf;
1215 s.avail_out = out_sz;
1218 switch (status) {
1219 case Z_OK:
1220 case Z_BUF_ERROR:
1221 case Z_STREAM_END:
1222 continue;
1223 default:
1224 die("unexpected deflate failure: %d", status);
1227 git_deflate_end(&s);
1228 git_SHA1_Final(sha1, &c);
1230 if (sha1out)
1231 hashcpy(sha1out, sha1);
1233 e = insert_object(sha1);
1235 if (mark)
1236 insert_mark(mark, e);
1238 if (e->idx.offset) {
1239 duplicate_count_by_type[OBJ_BLOB]++;
1240 truncate_pack(&checkpoint);
1242 } else if (find_sha1_pack(sha1, packed_git)) {
1243 e->type = OBJ_BLOB;
1244 e->pack_id = MAX_PACK_ID;
1245 e->idx.offset = 1; /* just not zero! */
1246 duplicate_count_by_type[OBJ_BLOB]++;
1247 truncate_pack(&checkpoint);
1249 } else {
1250 e->depth = 0;
1251 e->type = OBJ_BLOB;
1252 e->pack_id = pack_id;
1253 e->idx.offset = offset;
1254 e->idx.crc32 = crc32_end(pack_file);
1255 object_count++;
1256 object_count_by_type[OBJ_BLOB]++;
1259 free(in_buf);
1260 free(out_buf);
1263 /* All calls must be guarded by find_object() or find_mark() to
1264 * ensure the 'struct object_entry' passed was written by this
1265 * process instance. We unpack the entry by the offset, avoiding
1266 * the need for the corresponding .idx file. This unpacking rule
1267 * works because we only use OBJ_REF_DELTA within the packfiles
1268 * created by fast-import.
1270 * oe must not be NULL. Such an oe usually comes from giving
1271 * an unknown SHA-1 to find_object() or an undefined mark to
1272 * find_mark(). Callers must test for this condition and use
1273 * the standard read_sha1_file() when it happens.
1275 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1276 * find_mark(), where the mark was reloaded from an existing marks
1277 * file and is referencing an object that this fast-import process
1278 * instance did not write out to a packfile. Callers must test for
1279 * this condition and use read_sha1_file() instead.
1281 static void *gfi_unpack_entry(
1282 struct object_entry *oe,
1283 unsigned long *sizep)
1285 enum object_type type;
1286 struct packed_git *p = all_packs[oe->pack_id];
1287 if (p == pack_data && p->pack_size < (pack_size + 20)) {
1288 /* The object is stored in the packfile we are writing to
1289 * and we have modified it since the last time we scanned
1290 * back to read a previously written object. If an old
1291 * window covered [p->pack_size, p->pack_size + 20) its
1292 * data is stale and is not valid. Closing all windows
1293 * and updating the packfile length ensures we can read
1294 * the newly written data.
1296 close_pack_windows(p);
1297 sha1flush(pack_file);
1299 /* We have to offer 20 bytes additional on the end of
1300 * the packfile as the core unpacker code assumes the
1301 * footer is present at the file end and must promise
1302 * at least 20 bytes within any window it maps. But
1303 * we don't actually create the footer here.
1305 p->pack_size = pack_size + 20;
1307 return unpack_entry(p, oe->idx.offset, &type, sizep);
1310 static const char *get_mode(const char *str, uint16_t *modep)
1312 unsigned char c;
1313 uint16_t mode = 0;
1315 while ((c = *str++) != ' ') {
1316 if (c < '0' || c > '7')
1317 return NULL;
1318 mode = (mode << 3) + (c - '0');
1320 *modep = mode;
1321 return str;
1324 static void load_tree(struct tree_entry *root)
1326 unsigned char *sha1 = root->versions[1].sha1;
1327 struct object_entry *myoe;
1328 struct tree_content *t;
1329 unsigned long size;
1330 char *buf;
1331 const char *c;
1333 root->tree = t = new_tree_content(8);
1334 if (is_null_sha1(sha1))
1335 return;
1337 myoe = find_object(sha1);
1338 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1339 if (myoe->type != OBJ_TREE)
1340 die("Not a tree: %s", sha1_to_hex(sha1));
1341 t->delta_depth = myoe->depth;
1342 buf = gfi_unpack_entry(myoe, &size);
1343 if (!buf)
1344 die("Can't load tree %s", sha1_to_hex(sha1));
1345 } else {
1346 enum object_type type;
1347 buf = read_sha1_file(sha1, &type, &size);
1348 if (!buf || type != OBJ_TREE)
1349 die("Can't load tree %s", sha1_to_hex(sha1));
1352 c = buf;
1353 while (c != (buf + size)) {
1354 struct tree_entry *e = new_tree_entry();
1356 if (t->entry_count == t->entry_capacity)
1357 root->tree = t = grow_tree_content(t, t->entry_count);
1358 t->entries[t->entry_count++] = e;
1360 e->tree = NULL;
1361 c = get_mode(c, &e->versions[1].mode);
1362 if (!c)
1363 die("Corrupt mode in %s", sha1_to_hex(sha1));
1364 e->versions[0].mode = e->versions[1].mode;
1365 e->name = to_atom(c, strlen(c));
1366 c += e->name->str_len + 1;
1367 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1368 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1369 c += 20;
1371 free(buf);
1374 static int tecmp0 (const void *_a, const void *_b)
1376 struct tree_entry *a = *((struct tree_entry**)_a);
1377 struct tree_entry *b = *((struct tree_entry**)_b);
1378 return base_name_compare(
1379 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1380 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1383 static int tecmp1 (const void *_a, const void *_b)
1385 struct tree_entry *a = *((struct tree_entry**)_a);
1386 struct tree_entry *b = *((struct tree_entry**)_b);
1387 return base_name_compare(
1388 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1389 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1392 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1394 size_t maxlen = 0;
1395 unsigned int i;
1397 if (!v)
1398 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1399 else
1400 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1402 for (i = 0; i < t->entry_count; i++) {
1403 if (t->entries[i]->versions[v].mode)
1404 maxlen += t->entries[i]->name->str_len + 34;
1407 strbuf_reset(b);
1408 strbuf_grow(b, maxlen);
1409 for (i = 0; i < t->entry_count; i++) {
1410 struct tree_entry *e = t->entries[i];
1411 if (!e->versions[v].mode)
1412 continue;
1413 strbuf_addf(b, "%o %s%c",
1414 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1415 e->name->str_dat, '\0');
1416 strbuf_add(b, e->versions[v].sha1, 20);
1420 static void store_tree(struct tree_entry *root)
1422 struct tree_content *t = root->tree;
1423 unsigned int i, j, del;
1424 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1425 struct object_entry *le = NULL;
1427 if (!is_null_sha1(root->versions[1].sha1))
1428 return;
1430 for (i = 0; i < t->entry_count; i++) {
1431 if (t->entries[i]->tree)
1432 store_tree(t->entries[i]);
1435 if (!(root->versions[0].mode & NO_DELTA))
1436 le = find_object(root->versions[0].sha1);
1437 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1438 mktree(t, 0, &old_tree);
1439 lo.data = old_tree;
1440 lo.offset = le->idx.offset;
1441 lo.depth = t->delta_depth;
1444 mktree(t, 1, &new_tree);
1445 store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1447 t->delta_depth = lo.depth;
1448 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1449 struct tree_entry *e = t->entries[i];
1450 if (e->versions[1].mode) {
1451 e->versions[0].mode = e->versions[1].mode;
1452 hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1453 t->entries[j++] = e;
1454 } else {
1455 release_tree_entry(e);
1456 del++;
1459 t->entry_count -= del;
1462 static void tree_content_replace(
1463 struct tree_entry *root,
1464 const unsigned char *sha1,
1465 const uint16_t mode,
1466 struct tree_content *newtree)
1468 if (!S_ISDIR(mode))
1469 die("Root cannot be a non-directory");
1470 hashclr(root->versions[0].sha1);
1471 hashcpy(root->versions[1].sha1, sha1);
1472 if (root->tree)
1473 release_tree_content_recursive(root->tree);
1474 root->tree = newtree;
1477 static int tree_content_set(
1478 struct tree_entry *root,
1479 const char *p,
1480 const unsigned char *sha1,
1481 const uint16_t mode,
1482 struct tree_content *subtree)
1484 struct tree_content *t;
1485 const char *slash1;
1486 unsigned int i, n;
1487 struct tree_entry *e;
1489 slash1 = strchrnul(p, '/');
1490 n = slash1 - p;
1491 if (!n)
1492 die("Empty path component found in input");
1493 if (!*slash1 && !S_ISDIR(mode) && subtree)
1494 die("Non-directories cannot have subtrees");
1496 if (!root->tree)
1497 load_tree(root);
1498 t = root->tree;
1499 for (i = 0; i < t->entry_count; i++) {
1500 e = t->entries[i];
1501 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1502 if (!*slash1) {
1503 if (!S_ISDIR(mode)
1504 && e->versions[1].mode == mode
1505 && !hashcmp(e->versions[1].sha1, sha1))
1506 return 0;
1507 e->versions[1].mode = mode;
1508 hashcpy(e->versions[1].sha1, sha1);
1509 if (e->tree)
1510 release_tree_content_recursive(e->tree);
1511 e->tree = subtree;
1514 * We need to leave e->versions[0].sha1 alone
1515 * to avoid modifying the preimage tree used
1516 * when writing out the parent directory.
1517 * But after replacing the subdir with a
1518 * completely different one, it's not a good
1519 * delta base any more, and besides, we've
1520 * thrown away the tree entries needed to
1521 * make a delta against it.
1523 * So let's just explicitly disable deltas
1524 * for the subtree.
1526 if (S_ISDIR(e->versions[0].mode))
1527 e->versions[0].mode |= NO_DELTA;
1529 hashclr(root->versions[1].sha1);
1530 return 1;
1532 if (!S_ISDIR(e->versions[1].mode)) {
1533 e->tree = new_tree_content(8);
1534 e->versions[1].mode = S_IFDIR;
1536 if (!e->tree)
1537 load_tree(e);
1538 if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1539 hashclr(root->versions[1].sha1);
1540 return 1;
1542 return 0;
1546 if (t->entry_count == t->entry_capacity)
1547 root->tree = t = grow_tree_content(t, t->entry_count);
1548 e = new_tree_entry();
1549 e->name = to_atom(p, n);
1550 e->versions[0].mode = 0;
1551 hashclr(e->versions[0].sha1);
1552 t->entries[t->entry_count++] = e;
1553 if (*slash1) {
1554 e->tree = new_tree_content(8);
1555 e->versions[1].mode = S_IFDIR;
1556 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1557 } else {
1558 e->tree = subtree;
1559 e->versions[1].mode = mode;
1560 hashcpy(e->versions[1].sha1, sha1);
1562 hashclr(root->versions[1].sha1);
1563 return 1;
1566 static int tree_content_remove(
1567 struct tree_entry *root,
1568 const char *p,
1569 struct tree_entry *backup_leaf,
1570 int allow_root)
1572 struct tree_content *t;
1573 const char *slash1;
1574 unsigned int i, n;
1575 struct tree_entry *e;
1577 slash1 = strchrnul(p, '/');
1578 n = slash1 - p;
1580 if (!root->tree)
1581 load_tree(root);
1583 if (!*p && allow_root) {
1584 e = root;
1585 goto del_entry;
1588 t = root->tree;
1589 for (i = 0; i < t->entry_count; i++) {
1590 e = t->entries[i];
1591 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1592 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1594 * If p names a file in some subdirectory, and a
1595 * file or symlink matching the name of the
1596 * parent directory of p exists, then p cannot
1597 * exist and need not be deleted.
1599 return 1;
1600 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1601 goto del_entry;
1602 if (!e->tree)
1603 load_tree(e);
1604 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1605 for (n = 0; n < e->tree->entry_count; n++) {
1606 if (e->tree->entries[n]->versions[1].mode) {
1607 hashclr(root->versions[1].sha1);
1608 return 1;
1611 backup_leaf = NULL;
1612 goto del_entry;
1614 return 0;
1617 return 0;
1619 del_entry:
1620 if (backup_leaf)
1621 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1622 else if (e->tree)
1623 release_tree_content_recursive(e->tree);
1624 e->tree = NULL;
1625 e->versions[1].mode = 0;
1626 hashclr(e->versions[1].sha1);
1627 hashclr(root->versions[1].sha1);
1628 return 1;
1631 static int tree_content_get(
1632 struct tree_entry *root,
1633 const char *p,
1634 struct tree_entry *leaf,
1635 int allow_root)
1637 struct tree_content *t;
1638 const char *slash1;
1639 unsigned int i, n;
1640 struct tree_entry *e;
1642 slash1 = strchrnul(p, '/');
1643 n = slash1 - p;
1644 if (!n && !allow_root)
1645 die("Empty path component found in input");
1647 if (!root->tree)
1648 load_tree(root);
1650 if (!n) {
1651 e = root;
1652 goto found_entry;
1655 t = root->tree;
1656 for (i = 0; i < t->entry_count; i++) {
1657 e = t->entries[i];
1658 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1659 if (!*slash1)
1660 goto found_entry;
1661 if (!S_ISDIR(e->versions[1].mode))
1662 return 0;
1663 if (!e->tree)
1664 load_tree(e);
1665 return tree_content_get(e, slash1 + 1, leaf, 0);
1668 return 0;
1670 found_entry:
1671 memcpy(leaf, e, sizeof(*leaf));
1672 if (e->tree && is_null_sha1(e->versions[1].sha1))
1673 leaf->tree = dup_tree_content(e->tree);
1674 else
1675 leaf->tree = NULL;
1676 return 1;
1679 static int update_branch(struct branch *b)
1681 static const char *msg = "fast-import";
1682 struct ref_lock *lock;
1683 unsigned char old_sha1[20];
1685 if (read_ref(b->name, old_sha1))
1686 hashclr(old_sha1);
1687 if (is_null_sha1(b->sha1)) {
1688 if (b->delete)
1689 delete_ref(b->name, old_sha1, 0);
1690 return 0;
1692 lock = lock_any_ref_for_update(b->name, old_sha1, 0, NULL);
1693 if (!lock)
1694 return error("Unable to lock %s", b->name);
1695 if (!force_update && !is_null_sha1(old_sha1)) {
1696 struct commit *old_cmit, *new_cmit;
1698 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1699 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1700 if (!old_cmit || !new_cmit) {
1701 unlock_ref(lock);
1702 return error("Branch %s is missing commits.", b->name);
1705 if (!in_merge_bases(old_cmit, new_cmit)) {
1706 unlock_ref(lock);
1707 warning("Not updating %s"
1708 " (new tip %s does not contain %s)",
1709 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1710 return -1;
1713 if (write_ref_sha1(lock, b->sha1, msg) < 0)
1714 return error("Unable to update %s", b->name);
1715 return 0;
1718 static void dump_branches(void)
1720 unsigned int i;
1721 struct branch *b;
1723 for (i = 0; i < branch_table_sz; i++) {
1724 for (b = branch_table[i]; b; b = b->table_next_branch)
1725 failure |= update_branch(b);
1729 static void dump_tags(void)
1731 static const char *msg = "fast-import";
1732 struct tag *t;
1733 struct ref_lock *lock;
1734 char ref_name[PATH_MAX];
1736 for (t = first_tag; t; t = t->next_tag) {
1737 sprintf(ref_name, "tags/%s", t->name);
1738 lock = lock_ref_sha1(ref_name, NULL);
1739 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1740 failure |= error("Unable to update %s", ref_name);
1744 static void dump_marks_helper(FILE *f,
1745 uintmax_t base,
1746 struct mark_set *m)
1748 uintmax_t k;
1749 if (m->shift) {
1750 for (k = 0; k < 1024; k++) {
1751 if (m->data.sets[k])
1752 dump_marks_helper(f, base + (k << m->shift),
1753 m->data.sets[k]);
1755 } else {
1756 for (k = 0; k < 1024; k++) {
1757 if (m->data.marked[k])
1758 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1759 sha1_to_hex(m->data.marked[k]->idx.sha1));
1764 static void dump_marks(void)
1766 static struct lock_file mark_lock;
1767 int mark_fd;
1768 FILE *f;
1770 if (!export_marks_file)
1771 return;
1773 mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1774 if (mark_fd < 0) {
1775 failure |= error("Unable to write marks file %s: %s",
1776 export_marks_file, strerror(errno));
1777 return;
1780 f = fdopen(mark_fd, "w");
1781 if (!f) {
1782 int saved_errno = errno;
1783 rollback_lock_file(&mark_lock);
1784 failure |= error("Unable to write marks file %s: %s",
1785 export_marks_file, strerror(saved_errno));
1786 return;
1790 * Since the lock file was fdopen()'ed, it should not be close()'ed.
1791 * Assign -1 to the lock file descriptor so that commit_lock_file()
1792 * won't try to close() it.
1794 mark_lock.fd = -1;
1796 dump_marks_helper(f, 0, marks);
1797 if (ferror(f) || fclose(f)) {
1798 int saved_errno = errno;
1799 rollback_lock_file(&mark_lock);
1800 failure |= error("Unable to write marks file %s: %s",
1801 export_marks_file, strerror(saved_errno));
1802 return;
1805 if (commit_lock_file(&mark_lock)) {
1806 int saved_errno = errno;
1807 rollback_lock_file(&mark_lock);
1808 failure |= error("Unable to commit marks file %s: %s",
1809 export_marks_file, strerror(saved_errno));
1810 return;
1814 static void read_marks(void)
1816 char line[512];
1817 FILE *f = fopen(import_marks_file, "r");
1818 if (f)
1820 else if (import_marks_file_ignore_missing && errno == ENOENT)
1821 return; /* Marks file does not exist */
1822 else
1823 die_errno("cannot read '%s'", import_marks_file);
1824 while (fgets(line, sizeof(line), f)) {
1825 uintmax_t mark;
1826 char *end;
1827 unsigned char sha1[20];
1828 struct object_entry *e;
1830 end = strchr(line, '\n');
1831 if (line[0] != ':' || !end)
1832 die("corrupt mark line: %s", line);
1833 *end = 0;
1834 mark = strtoumax(line + 1, &end, 10);
1835 if (!mark || end == line + 1
1836 || *end != ' ' || get_sha1_hex(end + 1, sha1))
1837 die("corrupt mark line: %s", line);
1838 e = find_object(sha1);
1839 if (!e) {
1840 enum object_type type = sha1_object_info(sha1, NULL);
1841 if (type < 0)
1842 die("object not found: %s", sha1_to_hex(sha1));
1843 e = insert_object(sha1);
1844 e->type = type;
1845 e->pack_id = MAX_PACK_ID;
1846 e->idx.offset = 1; /* just not zero! */
1848 insert_mark(mark, e);
1850 fclose(f);
1854 static int read_next_command(void)
1856 static int stdin_eof = 0;
1858 if (stdin_eof) {
1859 unread_command_buf = 0;
1860 return EOF;
1863 for (;;) {
1864 if (unread_command_buf) {
1865 unread_command_buf = 0;
1866 } else {
1867 struct recent_command *rc;
1869 strbuf_detach(&command_buf, NULL);
1870 stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1871 if (stdin_eof)
1872 return EOF;
1874 if (!seen_data_command
1875 && !starts_with(command_buf.buf, "feature ")
1876 && !starts_with(command_buf.buf, "option ")) {
1877 parse_argv();
1880 rc = rc_free;
1881 if (rc)
1882 rc_free = rc->next;
1883 else {
1884 rc = cmd_hist.next;
1885 cmd_hist.next = rc->next;
1886 cmd_hist.next->prev = &cmd_hist;
1887 free(rc->buf);
1890 rc->buf = command_buf.buf;
1891 rc->prev = cmd_tail;
1892 rc->next = cmd_hist.prev;
1893 rc->prev->next = rc;
1894 cmd_tail = rc;
1896 if (starts_with(command_buf.buf, "cat-blob ")) {
1897 parse_cat_blob();
1898 continue;
1900 if (command_buf.buf[0] == '#')
1901 continue;
1902 return 0;
1906 static void skip_optional_lf(void)
1908 int term_char = fgetc(stdin);
1909 if (term_char != '\n' && term_char != EOF)
1910 ungetc(term_char, stdin);
1913 static void parse_mark(void)
1915 const char *v;
1916 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1917 next_mark = strtoumax(v, NULL, 10);
1918 read_next_command();
1920 else
1921 next_mark = 0;
1924 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1926 const char *data;
1927 strbuf_reset(sb);
1929 if (!skip_prefix(command_buf.buf, "data ", &data))
1930 die("Expected 'data n' command, found: %s", command_buf.buf);
1932 if (skip_prefix(data, "<<", &data)) {
1933 char *term = xstrdup(data);
1934 size_t term_len = command_buf.len - (data - command_buf.buf);
1936 strbuf_detach(&command_buf, NULL);
1937 for (;;) {
1938 if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1939 die("EOF in data (terminator '%s' not found)", term);
1940 if (term_len == command_buf.len
1941 && !strcmp(term, command_buf.buf))
1942 break;
1943 strbuf_addbuf(sb, &command_buf);
1944 strbuf_addch(sb, '\n');
1946 free(term);
1948 else {
1949 uintmax_t len = strtoumax(data, NULL, 10);
1950 size_t n = 0, length = (size_t)len;
1952 if (limit && limit < len) {
1953 *len_res = len;
1954 return 0;
1956 if (length < len)
1957 die("data is too large to use in this context");
1959 while (n < length) {
1960 size_t s = strbuf_fread(sb, length - n, stdin);
1961 if (!s && feof(stdin))
1962 die("EOF in data (%lu bytes remaining)",
1963 (unsigned long)(length - n));
1964 n += s;
1968 skip_optional_lf();
1969 return 1;
1972 static int validate_raw_date(const char *src, char *result, int maxlen)
1974 const char *orig_src = src;
1975 char *endp;
1976 unsigned long num;
1978 errno = 0;
1980 num = strtoul(src, &endp, 10);
1981 /* NEEDSWORK: perhaps check for reasonable values? */
1982 if (errno || endp == src || *endp != ' ')
1983 return -1;
1985 src = endp + 1;
1986 if (*src != '-' && *src != '+')
1987 return -1;
1989 num = strtoul(src + 1, &endp, 10);
1990 if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1991 1400 < num)
1992 return -1;
1994 strcpy(result, orig_src);
1995 return 0;
1998 static char *parse_ident(const char *buf)
2000 const char *ltgt;
2001 size_t name_len;
2002 char *ident;
2004 /* ensure there is a space delimiter even if there is no name */
2005 if (*buf == '<')
2006 --buf;
2008 ltgt = buf + strcspn(buf, "<>");
2009 if (*ltgt != '<')
2010 die("Missing < in ident string: %s", buf);
2011 if (ltgt != buf && ltgt[-1] != ' ')
2012 die("Missing space before < in ident string: %s", buf);
2013 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2014 if (*ltgt != '>')
2015 die("Missing > in ident string: %s", buf);
2016 ltgt++;
2017 if (*ltgt != ' ')
2018 die("Missing space after > in ident string: %s", buf);
2019 ltgt++;
2020 name_len = ltgt - buf;
2021 ident = xmalloc(name_len + 24);
2022 strncpy(ident, buf, name_len);
2024 switch (whenspec) {
2025 case WHENSPEC_RAW:
2026 if (validate_raw_date(ltgt, ident + name_len, 24) < 0)
2027 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2028 break;
2029 case WHENSPEC_RFC2822:
2030 if (parse_date(ltgt, ident + name_len, 24) < 0)
2031 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2032 break;
2033 case WHENSPEC_NOW:
2034 if (strcmp("now", ltgt))
2035 die("Date in ident must be 'now': %s", buf);
2036 datestamp(ident + name_len, 24);
2037 break;
2040 return ident;
2043 static void parse_and_store_blob(
2044 struct last_object *last,
2045 unsigned char *sha1out,
2046 uintmax_t mark)
2048 static struct strbuf buf = STRBUF_INIT;
2049 uintmax_t len;
2051 if (parse_data(&buf, big_file_threshold, &len))
2052 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2053 else {
2054 if (last) {
2055 strbuf_release(&last->data);
2056 last->offset = 0;
2057 last->depth = 0;
2059 stream_blob(len, sha1out, mark);
2060 skip_optional_lf();
2064 static void parse_new_blob(void)
2066 read_next_command();
2067 parse_mark();
2068 parse_and_store_blob(&last_blob, NULL, next_mark);
2071 static void unload_one_branch(void)
2073 while (cur_active_branches
2074 && cur_active_branches >= max_active_branches) {
2075 uintmax_t min_commit = ULONG_MAX;
2076 struct branch *e, *l = NULL, *p = NULL;
2078 for (e = active_branches; e; e = e->active_next_branch) {
2079 if (e->last_commit < min_commit) {
2080 p = l;
2081 min_commit = e->last_commit;
2083 l = e;
2086 if (p) {
2087 e = p->active_next_branch;
2088 p->active_next_branch = e->active_next_branch;
2089 } else {
2090 e = active_branches;
2091 active_branches = e->active_next_branch;
2093 e->active = 0;
2094 e->active_next_branch = NULL;
2095 if (e->branch_tree.tree) {
2096 release_tree_content_recursive(e->branch_tree.tree);
2097 e->branch_tree.tree = NULL;
2099 cur_active_branches--;
2103 static void load_branch(struct branch *b)
2105 load_tree(&b->branch_tree);
2106 if (!b->active) {
2107 b->active = 1;
2108 b->active_next_branch = active_branches;
2109 active_branches = b;
2110 cur_active_branches++;
2111 branch_load_count++;
2115 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2117 unsigned char fanout = 0;
2118 while ((num_notes >>= 8))
2119 fanout++;
2120 return fanout;
2123 static void construct_path_with_fanout(const char *hex_sha1,
2124 unsigned char fanout, char *path)
2126 unsigned int i = 0, j = 0;
2127 if (fanout >= 20)
2128 die("Too large fanout (%u)", fanout);
2129 while (fanout) {
2130 path[i++] = hex_sha1[j++];
2131 path[i++] = hex_sha1[j++];
2132 path[i++] = '/';
2133 fanout--;
2135 memcpy(path + i, hex_sha1 + j, 40 - j);
2136 path[i + 40 - j] = '\0';
2139 static uintmax_t do_change_note_fanout(
2140 struct tree_entry *orig_root, struct tree_entry *root,
2141 char *hex_sha1, unsigned int hex_sha1_len,
2142 char *fullpath, unsigned int fullpath_len,
2143 unsigned char fanout)
2145 struct tree_content *t = root->tree;
2146 struct tree_entry *e, leaf;
2147 unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2148 uintmax_t num_notes = 0;
2149 unsigned char sha1[20];
2150 char realpath[60];
2152 for (i = 0; t && i < t->entry_count; i++) {
2153 e = t->entries[i];
2154 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2155 tmp_fullpath_len = fullpath_len;
2158 * We're interested in EITHER existing note entries (entries
2159 * with exactly 40 hex chars in path, not including directory
2160 * separators), OR directory entries that may contain note
2161 * entries (with < 40 hex chars in path).
2162 * Also, each path component in a note entry must be a multiple
2163 * of 2 chars.
2165 if (!e->versions[1].mode ||
2166 tmp_hex_sha1_len > 40 ||
2167 e->name->str_len % 2)
2168 continue;
2170 /* This _may_ be a note entry, or a subdir containing notes */
2171 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2172 e->name->str_len);
2173 if (tmp_fullpath_len)
2174 fullpath[tmp_fullpath_len++] = '/';
2175 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2176 e->name->str_len);
2177 tmp_fullpath_len += e->name->str_len;
2178 fullpath[tmp_fullpath_len] = '\0';
2180 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2181 /* This is a note entry */
2182 if (fanout == 0xff) {
2183 /* Counting mode, no rename */
2184 num_notes++;
2185 continue;
2187 construct_path_with_fanout(hex_sha1, fanout, realpath);
2188 if (!strcmp(fullpath, realpath)) {
2189 /* Note entry is in correct location */
2190 num_notes++;
2191 continue;
2194 /* Rename fullpath to realpath */
2195 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2196 die("Failed to remove path %s", fullpath);
2197 tree_content_set(orig_root, realpath,
2198 leaf.versions[1].sha1,
2199 leaf.versions[1].mode,
2200 leaf.tree);
2201 } else if (S_ISDIR(e->versions[1].mode)) {
2202 /* This is a subdir that may contain note entries */
2203 if (!e->tree)
2204 load_tree(e);
2205 num_notes += do_change_note_fanout(orig_root, e,
2206 hex_sha1, tmp_hex_sha1_len,
2207 fullpath, tmp_fullpath_len, fanout);
2210 /* The above may have reallocated the current tree_content */
2211 t = root->tree;
2213 return num_notes;
2216 static uintmax_t change_note_fanout(struct tree_entry *root,
2217 unsigned char fanout)
2219 char hex_sha1[40], path[60];
2220 return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2224 * Given a pointer into a string, parse a mark reference:
2226 * idnum ::= ':' bigint;
2228 * Return the first character after the value in *endptr.
2230 * Complain if the following character is not what is expected,
2231 * either a space or end of the string.
2233 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2235 uintmax_t mark;
2237 assert(*p == ':');
2238 p++;
2239 mark = strtoumax(p, endptr, 10);
2240 if (*endptr == p)
2241 die("No value after ':' in mark: %s", command_buf.buf);
2242 return mark;
2246 * Parse the mark reference, and complain if this is not the end of
2247 * the string.
2249 static uintmax_t parse_mark_ref_eol(const char *p)
2251 char *end;
2252 uintmax_t mark;
2254 mark = parse_mark_ref(p, &end);
2255 if (*end != '\0')
2256 die("Garbage after mark: %s", command_buf.buf);
2257 return mark;
2261 * Parse the mark reference, demanding a trailing space. Return a
2262 * pointer to the space.
2264 static uintmax_t parse_mark_ref_space(const char **p)
2266 uintmax_t mark;
2267 char *end;
2269 mark = parse_mark_ref(*p, &end);
2270 if (*end != ' ')
2271 die("Missing space after mark: %s", command_buf.buf);
2272 *p = end;
2273 return mark;
2276 static void file_change_m(struct branch *b)
2278 const char *p = command_buf.buf + 2;
2279 static struct strbuf uq = STRBUF_INIT;
2280 const char *endp;
2281 struct object_entry *oe;
2282 unsigned char sha1[20];
2283 uint16_t mode, inline_data = 0;
2285 p = get_mode(p, &mode);
2286 if (!p)
2287 die("Corrupt mode: %s", command_buf.buf);
2288 switch (mode) {
2289 case 0644:
2290 case 0755:
2291 mode |= S_IFREG;
2292 case S_IFREG | 0644:
2293 case S_IFREG | 0755:
2294 case S_IFLNK:
2295 case S_IFDIR:
2296 case S_IFGITLINK:
2297 /* ok */
2298 break;
2299 default:
2300 die("Corrupt mode: %s", command_buf.buf);
2303 if (*p == ':') {
2304 oe = find_mark(parse_mark_ref_space(&p));
2305 hashcpy(sha1, oe->idx.sha1);
2306 } else if (starts_with(p, "inline ")) {
2307 inline_data = 1;
2308 oe = NULL; /* not used with inline_data, but makes gcc happy */
2309 p += strlen("inline"); /* advance to space */
2310 } else {
2311 if (get_sha1_hex(p, sha1))
2312 die("Invalid dataref: %s", command_buf.buf);
2313 oe = find_object(sha1);
2314 p += 40;
2315 if (*p != ' ')
2316 die("Missing space after SHA1: %s", command_buf.buf);
2318 assert(*p == ' ');
2319 p++; /* skip space */
2321 strbuf_reset(&uq);
2322 if (!unquote_c_style(&uq, p, &endp)) {
2323 if (*endp)
2324 die("Garbage after path in: %s", command_buf.buf);
2325 p = uq.buf;
2328 /* Git does not track empty, non-toplevel directories. */
2329 if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2330 tree_content_remove(&b->branch_tree, p, NULL, 0);
2331 return;
2334 if (S_ISGITLINK(mode)) {
2335 if (inline_data)
2336 die("Git links cannot be specified 'inline': %s",
2337 command_buf.buf);
2338 else if (oe) {
2339 if (oe->type != OBJ_COMMIT)
2340 die("Not a commit (actually a %s): %s",
2341 typename(oe->type), command_buf.buf);
2344 * Accept the sha1 without checking; it expected to be in
2345 * another repository.
2347 } else if (inline_data) {
2348 if (S_ISDIR(mode))
2349 die("Directories cannot be specified 'inline': %s",
2350 command_buf.buf);
2351 if (p != uq.buf) {
2352 strbuf_addstr(&uq, p);
2353 p = uq.buf;
2355 read_next_command();
2356 parse_and_store_blob(&last_blob, sha1, 0);
2357 } else {
2358 enum object_type expected = S_ISDIR(mode) ?
2359 OBJ_TREE: OBJ_BLOB;
2360 enum object_type type = oe ? oe->type :
2361 sha1_object_info(sha1, NULL);
2362 if (type < 0)
2363 die("%s not found: %s",
2364 S_ISDIR(mode) ? "Tree" : "Blob",
2365 command_buf.buf);
2366 if (type != expected)
2367 die("Not a %s (actually a %s): %s",
2368 typename(expected), typename(type),
2369 command_buf.buf);
2372 if (!*p) {
2373 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2374 return;
2376 tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2379 static void file_change_d(struct branch *b)
2381 const char *p = command_buf.buf + 2;
2382 static struct strbuf uq = STRBUF_INIT;
2383 const char *endp;
2385 strbuf_reset(&uq);
2386 if (!unquote_c_style(&uq, p, &endp)) {
2387 if (*endp)
2388 die("Garbage after path in: %s", command_buf.buf);
2389 p = uq.buf;
2391 tree_content_remove(&b->branch_tree, p, NULL, 1);
2394 static void file_change_cr(struct branch *b, int rename)
2396 const char *s, *d;
2397 static struct strbuf s_uq = STRBUF_INIT;
2398 static struct strbuf d_uq = STRBUF_INIT;
2399 const char *endp;
2400 struct tree_entry leaf;
2402 s = command_buf.buf + 2;
2403 strbuf_reset(&s_uq);
2404 if (!unquote_c_style(&s_uq, s, &endp)) {
2405 if (*endp != ' ')
2406 die("Missing space after source: %s", command_buf.buf);
2407 } else {
2408 endp = strchr(s, ' ');
2409 if (!endp)
2410 die("Missing space after source: %s", command_buf.buf);
2411 strbuf_add(&s_uq, s, endp - s);
2413 s = s_uq.buf;
2415 endp++;
2416 if (!*endp)
2417 die("Missing dest: %s", command_buf.buf);
2419 d = endp;
2420 strbuf_reset(&d_uq);
2421 if (!unquote_c_style(&d_uq, d, &endp)) {
2422 if (*endp)
2423 die("Garbage after dest in: %s", command_buf.buf);
2424 d = d_uq.buf;
2427 memset(&leaf, 0, sizeof(leaf));
2428 if (rename)
2429 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2430 else
2431 tree_content_get(&b->branch_tree, s, &leaf, 1);
2432 if (!leaf.versions[1].mode)
2433 die("Path %s not in branch", s);
2434 if (!*d) { /* C "path/to/subdir" "" */
2435 tree_content_replace(&b->branch_tree,
2436 leaf.versions[1].sha1,
2437 leaf.versions[1].mode,
2438 leaf.tree);
2439 return;
2441 tree_content_set(&b->branch_tree, d,
2442 leaf.versions[1].sha1,
2443 leaf.versions[1].mode,
2444 leaf.tree);
2447 static void note_change_n(struct branch *b, unsigned char *old_fanout)
2449 const char *p = command_buf.buf + 2;
2450 static struct strbuf uq = STRBUF_INIT;
2451 struct object_entry *oe;
2452 struct branch *s;
2453 unsigned char sha1[20], commit_sha1[20];
2454 char path[60];
2455 uint16_t inline_data = 0;
2456 unsigned char new_fanout;
2459 * When loading a branch, we don't traverse its tree to count the real
2460 * number of notes (too expensive to do this for all non-note refs).
2461 * This means that recently loaded notes refs might incorrectly have
2462 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2464 * Fix this by traversing the tree and counting the number of notes
2465 * when b->num_notes == 0. If the notes tree is truly empty, the
2466 * calculation should not take long.
2468 if (b->num_notes == 0 && *old_fanout == 0) {
2469 /* Invoke change_note_fanout() in "counting mode". */
2470 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2471 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2474 /* Now parse the notemodify command. */
2475 /* <dataref> or 'inline' */
2476 if (*p == ':') {
2477 oe = find_mark(parse_mark_ref_space(&p));
2478 hashcpy(sha1, oe->idx.sha1);
2479 } else if (starts_with(p, "inline ")) {
2480 inline_data = 1;
2481 oe = NULL; /* not used with inline_data, but makes gcc happy */
2482 p += strlen("inline"); /* advance to space */
2483 } else {
2484 if (get_sha1_hex(p, sha1))
2485 die("Invalid dataref: %s", command_buf.buf);
2486 oe = find_object(sha1);
2487 p += 40;
2488 if (*p != ' ')
2489 die("Missing space after SHA1: %s", command_buf.buf);
2491 assert(*p == ' ');
2492 p++; /* skip space */
2494 /* <commit-ish> */
2495 s = lookup_branch(p);
2496 if (s) {
2497 if (is_null_sha1(s->sha1))
2498 die("Can't add a note on empty branch.");
2499 hashcpy(commit_sha1, s->sha1);
2500 } else if (*p == ':') {
2501 uintmax_t commit_mark = parse_mark_ref_eol(p);
2502 struct object_entry *commit_oe = find_mark(commit_mark);
2503 if (commit_oe->type != OBJ_COMMIT)
2504 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2505 hashcpy(commit_sha1, commit_oe->idx.sha1);
2506 } else if (!get_sha1(p, commit_sha1)) {
2507 unsigned long size;
2508 char *buf = read_object_with_reference(commit_sha1,
2509 commit_type, &size, commit_sha1);
2510 if (!buf || size < 46)
2511 die("Not a valid commit: %s", p);
2512 free(buf);
2513 } else
2514 die("Invalid ref name or SHA1 expression: %s", p);
2516 if (inline_data) {
2517 if (p != uq.buf) {
2518 strbuf_addstr(&uq, p);
2519 p = uq.buf;
2521 read_next_command();
2522 parse_and_store_blob(&last_blob, sha1, 0);
2523 } else if (oe) {
2524 if (oe->type != OBJ_BLOB)
2525 die("Not a blob (actually a %s): %s",
2526 typename(oe->type), command_buf.buf);
2527 } else if (!is_null_sha1(sha1)) {
2528 enum object_type type = sha1_object_info(sha1, NULL);
2529 if (type < 0)
2530 die("Blob not found: %s", command_buf.buf);
2531 if (type != OBJ_BLOB)
2532 die("Not a blob (actually a %s): %s",
2533 typename(type), command_buf.buf);
2536 construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2537 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2538 b->num_notes--;
2540 if (is_null_sha1(sha1))
2541 return; /* nothing to insert */
2543 b->num_notes++;
2544 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2545 construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2546 tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2549 static void file_change_deleteall(struct branch *b)
2551 release_tree_content_recursive(b->branch_tree.tree);
2552 hashclr(b->branch_tree.versions[0].sha1);
2553 hashclr(b->branch_tree.versions[1].sha1);
2554 load_tree(&b->branch_tree);
2555 b->num_notes = 0;
2558 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2560 if (!buf || size < 46)
2561 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2562 if (memcmp("tree ", buf, 5)
2563 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2564 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2565 hashcpy(b->branch_tree.versions[0].sha1,
2566 b->branch_tree.versions[1].sha1);
2569 static void parse_from_existing(struct branch *b)
2571 if (is_null_sha1(b->sha1)) {
2572 hashclr(b->branch_tree.versions[0].sha1);
2573 hashclr(b->branch_tree.versions[1].sha1);
2574 } else {
2575 unsigned long size;
2576 char *buf;
2578 buf = read_object_with_reference(b->sha1,
2579 commit_type, &size, b->sha1);
2580 parse_from_commit(b, buf, size);
2581 free(buf);
2585 static int parse_from(struct branch *b)
2587 const char *from;
2588 struct branch *s;
2590 if (!starts_with(command_buf.buf, "from "))
2591 return 0;
2593 if (b->branch_tree.tree) {
2594 release_tree_content_recursive(b->branch_tree.tree);
2595 b->branch_tree.tree = NULL;
2598 from = strchr(command_buf.buf, ' ') + 1;
2599 s = lookup_branch(from);
2600 if (b == s)
2601 die("Can't create a branch from itself: %s", b->name);
2602 else if (s) {
2603 unsigned char *t = s->branch_tree.versions[1].sha1;
2604 hashcpy(b->sha1, s->sha1);
2605 hashcpy(b->branch_tree.versions[0].sha1, t);
2606 hashcpy(b->branch_tree.versions[1].sha1, t);
2607 } else if (*from == ':') {
2608 uintmax_t idnum = parse_mark_ref_eol(from);
2609 struct object_entry *oe = find_mark(idnum);
2610 if (oe->type != OBJ_COMMIT)
2611 die("Mark :%" PRIuMAX " not a commit", idnum);
2612 hashcpy(b->sha1, oe->idx.sha1);
2613 if (oe->pack_id != MAX_PACK_ID) {
2614 unsigned long size;
2615 char *buf = gfi_unpack_entry(oe, &size);
2616 parse_from_commit(b, buf, size);
2617 free(buf);
2618 } else
2619 parse_from_existing(b);
2620 } else if (!get_sha1(from, b->sha1)) {
2621 parse_from_existing(b);
2622 if (is_null_sha1(b->sha1))
2623 b->delete = 1;
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 (starts_with(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;
2681 const char *v;
2683 /* Obtain the branch name from the rest of our command */
2684 sp = strchr(command_buf.buf, ' ') + 1;
2685 b = lookup_branch(sp);
2686 if (!b)
2687 b = new_branch(sp);
2689 read_next_command();
2690 parse_mark();
2691 if (skip_prefix(command_buf.buf, "author ", &v)) {
2692 author = parse_ident(v);
2693 read_next_command();
2695 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2696 committer = parse_ident(v);
2697 read_next_command();
2699 if (!committer)
2700 die("Expected committer but didn't get one");
2701 parse_data(&msg, 0, NULL);
2702 read_next_command();
2703 parse_from(b);
2704 merge_list = parse_merge(&merge_count);
2706 /* ensure the branch is active/loaded */
2707 if (!b->branch_tree.tree || !max_active_branches) {
2708 unload_one_branch();
2709 load_branch(b);
2712 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2714 /* file_change* */
2715 while (command_buf.len > 0) {
2716 if (starts_with(command_buf.buf, "M "))
2717 file_change_m(b);
2718 else if (starts_with(command_buf.buf, "D "))
2719 file_change_d(b);
2720 else if (starts_with(command_buf.buf, "R "))
2721 file_change_cr(b, 1);
2722 else if (starts_with(command_buf.buf, "C "))
2723 file_change_cr(b, 0);
2724 else if (starts_with(command_buf.buf, "N "))
2725 note_change_n(b, &prev_fanout);
2726 else if (!strcmp("deleteall", command_buf.buf))
2727 file_change_deleteall(b);
2728 else if (starts_with(command_buf.buf, "ls "))
2729 parse_ls(b);
2730 else {
2731 unread_command_buf = 1;
2732 break;
2734 if (read_next_command() == EOF)
2735 break;
2738 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2739 if (new_fanout != prev_fanout)
2740 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2742 /* build the tree and the commit */
2743 store_tree(&b->branch_tree);
2744 hashcpy(b->branch_tree.versions[0].sha1,
2745 b->branch_tree.versions[1].sha1);
2747 strbuf_reset(&new_data);
2748 strbuf_addf(&new_data, "tree %s\n",
2749 sha1_to_hex(b->branch_tree.versions[1].sha1));
2750 if (!is_null_sha1(b->sha1))
2751 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2752 while (merge_list) {
2753 struct hash_list *next = merge_list->next;
2754 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2755 free(merge_list);
2756 merge_list = next;
2758 strbuf_addf(&new_data,
2759 "author %s\n"
2760 "committer %s\n"
2761 "\n",
2762 author ? author : committer, committer);
2763 strbuf_addbuf(&new_data, &msg);
2764 free(author);
2765 free(committer);
2767 if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2768 b->pack_id = pack_id;
2769 b->last_commit = object_count_by_type[OBJ_COMMIT];
2772 static void parse_new_tag(void)
2774 static struct strbuf msg = STRBUF_INIT;
2775 char *sp;
2776 const char *from;
2777 char *tagger;
2778 struct branch *s;
2779 struct tag *t;
2780 uintmax_t from_mark = 0;
2781 unsigned char sha1[20];
2782 enum object_type type;
2783 const char *v;
2785 /* Obtain the new tag name from the rest of our command */
2786 sp = strchr(command_buf.buf, ' ') + 1;
2787 t = pool_alloc(sizeof(struct tag));
2788 memset(t, 0, sizeof(struct tag));
2789 t->name = pool_strdup(sp);
2790 if (last_tag)
2791 last_tag->next_tag = t;
2792 else
2793 first_tag = t;
2794 last_tag = t;
2795 read_next_command();
2797 /* from ... */
2798 if (!starts_with(command_buf.buf, "from "))
2799 die("Expected from command, got %s", command_buf.buf);
2800 from = strchr(command_buf.buf, ' ') + 1;
2801 s = lookup_branch(from);
2802 if (s) {
2803 if (is_null_sha1(s->sha1))
2804 die("Can't tag an empty branch.");
2805 hashcpy(sha1, s->sha1);
2806 type = OBJ_COMMIT;
2807 } else if (*from == ':') {
2808 struct object_entry *oe;
2809 from_mark = parse_mark_ref_eol(from);
2810 oe = find_mark(from_mark);
2811 type = oe->type;
2812 hashcpy(sha1, oe->idx.sha1);
2813 } else if (!get_sha1(from, sha1)) {
2814 struct object_entry *oe = find_object(sha1);
2815 if (!oe) {
2816 type = sha1_object_info(sha1, NULL);
2817 if (type < 0)
2818 die("Not a valid object: %s", from);
2819 } else
2820 type = oe->type;
2821 } else
2822 die("Invalid ref name or SHA1 expression: %s", from);
2823 read_next_command();
2825 /* tagger ... */
2826 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2827 tagger = parse_ident(v);
2828 read_next_command();
2829 } else
2830 tagger = NULL;
2832 /* tag payload/message */
2833 parse_data(&msg, 0, NULL);
2835 /* build the tag object */
2836 strbuf_reset(&new_data);
2838 strbuf_addf(&new_data,
2839 "object %s\n"
2840 "type %s\n"
2841 "tag %s\n",
2842 sha1_to_hex(sha1), typename(type), t->name);
2843 if (tagger)
2844 strbuf_addf(&new_data,
2845 "tagger %s\n", tagger);
2846 strbuf_addch(&new_data, '\n');
2847 strbuf_addbuf(&new_data, &msg);
2848 free(tagger);
2850 if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2851 t->pack_id = MAX_PACK_ID;
2852 else
2853 t->pack_id = pack_id;
2856 static void parse_reset_branch(void)
2858 struct branch *b;
2859 char *sp;
2861 /* Obtain the branch name from the rest of our command */
2862 sp = strchr(command_buf.buf, ' ') + 1;
2863 b = lookup_branch(sp);
2864 if (b) {
2865 hashclr(b->sha1);
2866 hashclr(b->branch_tree.versions[0].sha1);
2867 hashclr(b->branch_tree.versions[1].sha1);
2868 if (b->branch_tree.tree) {
2869 release_tree_content_recursive(b->branch_tree.tree);
2870 b->branch_tree.tree = NULL;
2873 else
2874 b = new_branch(sp);
2875 read_next_command();
2876 parse_from(b);
2877 if (command_buf.len > 0)
2878 unread_command_buf = 1;
2881 static void cat_blob_write(const char *buf, unsigned long size)
2883 if (write_in_full(cat_blob_fd, buf, size) != size)
2884 die_errno("Write to frontend failed");
2887 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2889 struct strbuf line = STRBUF_INIT;
2890 unsigned long size;
2891 enum object_type type = 0;
2892 char *buf;
2894 if (!oe || oe->pack_id == MAX_PACK_ID) {
2895 buf = read_sha1_file(sha1, &type, &size);
2896 } else {
2897 type = oe->type;
2898 buf = gfi_unpack_entry(oe, &size);
2902 * Output based on batch_one_object() from cat-file.c.
2904 if (type <= 0) {
2905 strbuf_reset(&line);
2906 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2907 cat_blob_write(line.buf, line.len);
2908 strbuf_release(&line);
2909 free(buf);
2910 return;
2912 if (!buf)
2913 die("Can't read object %s", sha1_to_hex(sha1));
2914 if (type != OBJ_BLOB)
2915 die("Object %s is a %s but a blob was expected.",
2916 sha1_to_hex(sha1), typename(type));
2917 strbuf_reset(&line);
2918 strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2919 typename(type), size);
2920 cat_blob_write(line.buf, line.len);
2921 strbuf_release(&line);
2922 cat_blob_write(buf, size);
2923 cat_blob_write("\n", 1);
2924 if (oe && oe->pack_id == pack_id) {
2925 last_blob.offset = oe->idx.offset;
2926 strbuf_attach(&last_blob.data, buf, size, size);
2927 last_blob.depth = oe->depth;
2928 } else
2929 free(buf);
2932 static void parse_cat_blob(void)
2934 const char *p;
2935 struct object_entry *oe = oe;
2936 unsigned char sha1[20];
2938 /* cat-blob SP <object> LF */
2939 p = command_buf.buf + strlen("cat-blob ");
2940 if (*p == ':') {
2941 oe = find_mark(parse_mark_ref_eol(p));
2942 if (!oe)
2943 die("Unknown mark: %s", command_buf.buf);
2944 hashcpy(sha1, oe->idx.sha1);
2945 } else {
2946 if (get_sha1_hex(p, sha1))
2947 die("Invalid dataref: %s", command_buf.buf);
2948 if (p[40])
2949 die("Garbage after SHA1: %s", command_buf.buf);
2950 oe = find_object(sha1);
2953 cat_blob(oe, sha1);
2956 static struct object_entry *dereference(struct object_entry *oe,
2957 unsigned char sha1[20])
2959 unsigned long size;
2960 char *buf = NULL;
2961 if (!oe) {
2962 enum object_type type = sha1_object_info(sha1, NULL);
2963 if (type < 0)
2964 die("object not found: %s", sha1_to_hex(sha1));
2965 /* cache it! */
2966 oe = insert_object(sha1);
2967 oe->type = type;
2968 oe->pack_id = MAX_PACK_ID;
2969 oe->idx.offset = 1;
2971 switch (oe->type) {
2972 case OBJ_TREE: /* easy case. */
2973 return oe;
2974 case OBJ_COMMIT:
2975 case OBJ_TAG:
2976 break;
2977 default:
2978 die("Not a tree-ish: %s", command_buf.buf);
2981 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
2982 buf = gfi_unpack_entry(oe, &size);
2983 } else {
2984 enum object_type unused;
2985 buf = read_sha1_file(sha1, &unused, &size);
2987 if (!buf)
2988 die("Can't load object %s", sha1_to_hex(sha1));
2990 /* Peel one layer. */
2991 switch (oe->type) {
2992 case OBJ_TAG:
2993 if (size < 40 + strlen("object ") ||
2994 get_sha1_hex(buf + strlen("object "), sha1))
2995 die("Invalid SHA1 in tag: %s", command_buf.buf);
2996 break;
2997 case OBJ_COMMIT:
2998 if (size < 40 + strlen("tree ") ||
2999 get_sha1_hex(buf + strlen("tree "), sha1))
3000 die("Invalid SHA1 in commit: %s", command_buf.buf);
3003 free(buf);
3004 return find_object(sha1);
3007 static struct object_entry *parse_treeish_dataref(const char **p)
3009 unsigned char sha1[20];
3010 struct object_entry *e;
3012 if (**p == ':') { /* <mark> */
3013 e = find_mark(parse_mark_ref_space(p));
3014 if (!e)
3015 die("Unknown mark: %s", command_buf.buf);
3016 hashcpy(sha1, e->idx.sha1);
3017 } else { /* <sha1> */
3018 if (get_sha1_hex(*p, sha1))
3019 die("Invalid dataref: %s", command_buf.buf);
3020 e = find_object(sha1);
3021 *p += 40;
3024 while (!e || e->type != OBJ_TREE)
3025 e = dereference(e, sha1);
3026 return e;
3029 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3031 static struct strbuf line = STRBUF_INIT;
3033 /* See show_tree(). */
3034 const char *type =
3035 S_ISGITLINK(mode) ? commit_type :
3036 S_ISDIR(mode) ? tree_type :
3037 blob_type;
3039 if (!mode) {
3040 /* missing SP path LF */
3041 strbuf_reset(&line);
3042 strbuf_addstr(&line, "missing ");
3043 quote_c_style(path, &line, NULL, 0);
3044 strbuf_addch(&line, '\n');
3045 } else {
3046 /* mode SP type SP object_name TAB path LF */
3047 strbuf_reset(&line);
3048 strbuf_addf(&line, "%06o %s %s\t",
3049 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3050 quote_c_style(path, &line, NULL, 0);
3051 strbuf_addch(&line, '\n');
3053 cat_blob_write(line.buf, line.len);
3056 static void parse_ls(struct branch *b)
3058 const char *p;
3059 struct tree_entry *root = NULL;
3060 struct tree_entry leaf = {NULL};
3062 /* ls SP (<tree-ish> SP)? <path> */
3063 p = command_buf.buf + strlen("ls ");
3064 if (*p == '"') {
3065 if (!b)
3066 die("Not in a commit: %s", command_buf.buf);
3067 root = &b->branch_tree;
3068 } else {
3069 struct object_entry *e = parse_treeish_dataref(&p);
3070 root = new_tree_entry();
3071 hashcpy(root->versions[1].sha1, e->idx.sha1);
3072 if (!is_null_sha1(root->versions[1].sha1))
3073 root->versions[1].mode = S_IFDIR;
3074 load_tree(root);
3075 if (*p++ != ' ')
3076 die("Missing space after tree-ish: %s", command_buf.buf);
3078 if (*p == '"') {
3079 static struct strbuf uq = STRBUF_INIT;
3080 const char *endp;
3081 strbuf_reset(&uq);
3082 if (unquote_c_style(&uq, p, &endp))
3083 die("Invalid path: %s", command_buf.buf);
3084 if (*endp)
3085 die("Garbage after path in: %s", command_buf.buf);
3086 p = uq.buf;
3088 tree_content_get(root, p, &leaf, 1);
3090 * A directory in preparation would have a sha1 of zero
3091 * until it is saved. Save, for simplicity.
3093 if (S_ISDIR(leaf.versions[1].mode))
3094 store_tree(&leaf);
3096 print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3097 if (leaf.tree)
3098 release_tree_content_recursive(leaf.tree);
3099 if (!b || root != &b->branch_tree)
3100 release_tree_entry(root);
3103 static void checkpoint(void)
3105 checkpoint_requested = 0;
3106 if (object_count) {
3107 cycle_packfile();
3108 dump_branches();
3109 dump_tags();
3110 dump_marks();
3114 static void parse_checkpoint(void)
3116 checkpoint_requested = 1;
3117 skip_optional_lf();
3120 static void parse_progress(void)
3122 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3123 fputc('\n', stdout);
3124 fflush(stdout);
3125 skip_optional_lf();
3128 static char* make_fast_import_path(const char *path)
3130 struct strbuf abs_path = STRBUF_INIT;
3132 if (!relative_marks_paths || is_absolute_path(path))
3133 return xstrdup(path);
3134 strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
3135 return strbuf_detach(&abs_path, NULL);
3138 static void option_import_marks(const char *marks,
3139 int from_stream, int ignore_missing)
3141 if (import_marks_file) {
3142 if (from_stream)
3143 die("Only one import-marks command allowed per stream");
3145 /* read previous mark file */
3146 if(!import_marks_file_from_stream)
3147 read_marks();
3150 import_marks_file = make_fast_import_path(marks);
3151 safe_create_leading_directories_const(import_marks_file);
3152 import_marks_file_from_stream = from_stream;
3153 import_marks_file_ignore_missing = ignore_missing;
3156 static void option_date_format(const char *fmt)
3158 if (!strcmp(fmt, "raw"))
3159 whenspec = WHENSPEC_RAW;
3160 else if (!strcmp(fmt, "rfc2822"))
3161 whenspec = WHENSPEC_RFC2822;
3162 else if (!strcmp(fmt, "now"))
3163 whenspec = WHENSPEC_NOW;
3164 else
3165 die("unknown --date-format argument %s", fmt);
3168 static unsigned long ulong_arg(const char *option, const char *arg)
3170 char *endptr;
3171 unsigned long rv = strtoul(arg, &endptr, 0);
3172 if (strchr(arg, '-') || endptr == arg || *endptr)
3173 die("%s: argument must be a non-negative integer", option);
3174 return rv;
3177 static void option_depth(const char *depth)
3179 max_depth = ulong_arg("--depth", depth);
3180 if (max_depth > MAX_DEPTH)
3181 die("--depth cannot exceed %u", MAX_DEPTH);
3184 static void option_active_branches(const char *branches)
3186 max_active_branches = ulong_arg("--active-branches", branches);
3189 static void option_export_marks(const char *marks)
3191 export_marks_file = make_fast_import_path(marks);
3192 safe_create_leading_directories_const(export_marks_file);
3195 static void option_cat_blob_fd(const char *fd)
3197 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3198 if (n > (unsigned long) INT_MAX)
3199 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3200 cat_blob_fd = (int) n;
3203 static void option_export_pack_edges(const char *edges)
3205 if (pack_edges)
3206 fclose(pack_edges);
3207 pack_edges = fopen(edges, "a");
3208 if (!pack_edges)
3209 die_errno("Cannot open '%s'", edges);
3212 static int parse_one_option(const char *option)
3214 if (skip_prefix(option, "max-pack-size=", &option)) {
3215 unsigned long v;
3216 if (!git_parse_ulong(option, &v))
3217 return 0;
3218 if (v < 8192) {
3219 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3220 v *= 1024 * 1024;
3221 } else if (v < 1024 * 1024) {
3222 warning("minimum max-pack-size is 1 MiB");
3223 v = 1024 * 1024;
3225 max_packsize = v;
3226 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3227 unsigned long v;
3228 if (!git_parse_ulong(option, &v))
3229 return 0;
3230 big_file_threshold = v;
3231 } else if (skip_prefix(option, "depth=", &option)) {
3232 option_depth(option);
3233 } else if (skip_prefix(option, "active-branches=", &option)) {
3234 option_active_branches(option);
3235 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3236 option_export_pack_edges(option);
3237 } else if (starts_with(option, "quiet")) {
3238 show_stats = 0;
3239 } else if (starts_with(option, "stats")) {
3240 show_stats = 1;
3241 } else {
3242 return 0;
3245 return 1;
3248 static int parse_one_feature(const char *feature, int from_stream)
3250 const char *arg;
3252 if (skip_prefix(feature, "date-format=", &arg)) {
3253 option_date_format(arg);
3254 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3255 option_import_marks(arg, from_stream, 0);
3256 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3257 option_import_marks(arg, from_stream, 1);
3258 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3259 option_export_marks(arg);
3260 } else if (!strcmp(feature, "cat-blob")) {
3261 ; /* Don't die - this feature is supported */
3262 } else if (!strcmp(feature, "relative-marks")) {
3263 relative_marks_paths = 1;
3264 } else if (!strcmp(feature, "no-relative-marks")) {
3265 relative_marks_paths = 0;
3266 } else if (!strcmp(feature, "done")) {
3267 require_explicit_termination = 1;
3268 } else if (!strcmp(feature, "force")) {
3269 force_update = 1;
3270 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3271 ; /* do nothing; we have the feature */
3272 } else {
3273 return 0;
3276 return 1;
3279 static void parse_feature(void)
3281 char *feature = command_buf.buf + 8;
3283 if (seen_data_command)
3284 die("Got feature command '%s' after data command", feature);
3286 if (parse_one_feature(feature, 1))
3287 return;
3289 die("This version of fast-import does not support feature %s.", feature);
3292 static void parse_option(void)
3294 char *option = command_buf.buf + 11;
3296 if (seen_data_command)
3297 die("Got option command '%s' after data command", option);
3299 if (parse_one_option(option))
3300 return;
3302 die("This version of fast-import does not support option: %s", option);
3305 static int git_pack_config(const char *k, const char *v, void *cb)
3307 if (!strcmp(k, "pack.depth")) {
3308 max_depth = git_config_int(k, v);
3309 if (max_depth > MAX_DEPTH)
3310 max_depth = MAX_DEPTH;
3311 return 0;
3313 if (!strcmp(k, "pack.compression")) {
3314 int level = git_config_int(k, v);
3315 if (level == -1)
3316 level = Z_DEFAULT_COMPRESSION;
3317 else if (level < 0 || level > Z_BEST_COMPRESSION)
3318 die("bad pack compression level %d", level);
3319 pack_compression_level = level;
3320 pack_compression_seen = 1;
3321 return 0;
3323 if (!strcmp(k, "pack.indexversion")) {
3324 pack_idx_opts.version = git_config_int(k, v);
3325 if (pack_idx_opts.version > 2)
3326 die("bad pack.indexversion=%"PRIu32,
3327 pack_idx_opts.version);
3328 return 0;
3330 if (!strcmp(k, "pack.packsizelimit")) {
3331 max_packsize = git_config_ulong(k, v);
3332 return 0;
3334 return git_default_config(k, v, cb);
3337 static const char fast_import_usage[] =
3338 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3340 static void parse_argv(void)
3342 unsigned int i;
3344 for (i = 1; i < global_argc; i++) {
3345 const char *a = global_argv[i];
3347 if (*a != '-' || !strcmp(a, "--"))
3348 break;
3350 if (!skip_prefix(a, "--", &a))
3351 die("unknown option %s", a);
3353 if (parse_one_option(a))
3354 continue;
3356 if (parse_one_feature(a, 0))
3357 continue;
3359 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3360 option_cat_blob_fd(a);
3361 continue;
3364 die("unknown option --%s", a);
3366 if (i != global_argc)
3367 usage(fast_import_usage);
3369 seen_data_command = 1;
3370 if (import_marks_file)
3371 read_marks();
3374 int main(int argc, char **argv)
3376 unsigned int i;
3378 git_extract_argv0_path(argv[0]);
3380 git_setup_gettext();
3382 if (argc == 2 && !strcmp(argv[1], "-h"))
3383 usage(fast_import_usage);
3385 setup_git_directory();
3386 reset_pack_idx_option(&pack_idx_opts);
3387 git_config(git_pack_config, NULL);
3388 if (!pack_compression_seen && core_compression_seen)
3389 pack_compression_level = core_compression_level;
3391 alloc_objects(object_entry_alloc);
3392 strbuf_init(&command_buf, 0);
3393 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3394 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3395 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3396 marks = pool_calloc(1, sizeof(struct mark_set));
3398 global_argc = argc;
3399 global_argv = argv;
3401 rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3402 for (i = 0; i < (cmd_save - 1); i++)
3403 rc_free[i].next = &rc_free[i + 1];
3404 rc_free[cmd_save - 1].next = NULL;
3406 prepare_packed_git();
3407 start_packfile();
3408 set_die_routine(die_nicely);
3409 set_checkpoint_signal();
3410 while (read_next_command() != EOF) {
3411 if (!strcmp("blob", command_buf.buf))
3412 parse_new_blob();
3413 else if (starts_with(command_buf.buf, "ls "))
3414 parse_ls(NULL);
3415 else if (starts_with(command_buf.buf, "commit "))
3416 parse_new_commit();
3417 else if (starts_with(command_buf.buf, "tag "))
3418 parse_new_tag();
3419 else if (starts_with(command_buf.buf, "reset "))
3420 parse_reset_branch();
3421 else if (!strcmp("checkpoint", command_buf.buf))
3422 parse_checkpoint();
3423 else if (!strcmp("done", command_buf.buf))
3424 break;
3425 else if (starts_with(command_buf.buf, "progress "))
3426 parse_progress();
3427 else if (starts_with(command_buf.buf, "feature "))
3428 parse_feature();
3429 else if (starts_with(command_buf.buf, "option git "))
3430 parse_option();
3431 else if (starts_with(command_buf.buf, "option "))
3432 /* ignore non-git options*/;
3433 else
3434 die("Unsupported command: %s", command_buf.buf);
3436 if (checkpoint_requested)
3437 checkpoint();
3440 /* argv hasn't been parsed yet, do so */
3441 if (!seen_data_command)
3442 parse_argv();
3444 if (require_explicit_termination && feof(stdin))
3445 die("stream ends early");
3447 end_packfile();
3449 dump_branches();
3450 dump_tags();
3451 unkeep_all_packs();
3452 dump_marks();
3454 if (pack_edges)
3455 fclose(pack_edges);
3457 if (show_stats) {
3458 uintmax_t total_count = 0, duplicate_count = 0;
3459 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3460 total_count += object_count_by_type[i];
3461 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3462 duplicate_count += duplicate_count_by_type[i];
3464 fprintf(stderr, "%s statistics:\n", argv[0]);
3465 fprintf(stderr, "---------------------------------------------------------------------\n");
3466 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3467 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3468 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]);
3469 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]);
3470 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]);
3471 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]);
3472 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3473 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3474 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3475 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3476 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)(total_allocd/1024));
3477 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3478 fprintf(stderr, "---------------------------------------------------------------------\n");
3479 pack_report();
3480 fprintf(stderr, "---------------------------------------------------------------------\n");
3481 fprintf(stderr, "\n");
3484 return failure ? 1 : 0;