lockfile.c: document the various states of lock_file objects
[git/debian.git] / fast-import.c
blob96b0f4236a2083044e4ae1dfa3b8180e3e586ba1
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(const char *p);
375 static void parse_ls(const char *p, 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 REALLOC_ARRAY(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 if (!pack_data)
950 return;
952 clear_delta_base_cache();
953 if (object_count) {
954 struct packed_git *new_p;
955 unsigned char cur_pack_sha1[20];
956 char *idx_name;
957 int i;
958 struct branch *b;
959 struct tag *t;
961 close_pack_windows(pack_data);
962 sha1close(pack_file, cur_pack_sha1, 0);
963 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
964 pack_data->pack_name, object_count,
965 cur_pack_sha1, pack_size);
966 close(pack_data->pack_fd);
967 idx_name = keep_pack(create_index());
969 /* Register the packfile with core git's machinery. */
970 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
971 if (!new_p)
972 die("core git rejected index %s", idx_name);
973 all_packs[pack_id] = new_p;
974 install_packed_git(new_p);
976 /* Print the boundary */
977 if (pack_edges) {
978 fprintf(pack_edges, "%s:", new_p->pack_name);
979 for (i = 0; i < branch_table_sz; i++) {
980 for (b = branch_table[i]; b; b = b->table_next_branch) {
981 if (b->pack_id == pack_id)
982 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
985 for (t = first_tag; t; t = t->next_tag) {
986 if (t->pack_id == pack_id)
987 fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
989 fputc('\n', pack_edges);
990 fflush(pack_edges);
993 pack_id++;
995 else {
996 close(pack_data->pack_fd);
997 unlink_or_warn(pack_data->pack_name);
999 free(pack_data);
1000 pack_data = NULL;
1002 /* We can't carry a delta across packfiles. */
1003 strbuf_release(&last_blob.data);
1004 last_blob.offset = 0;
1005 last_blob.depth = 0;
1008 static void cycle_packfile(void)
1010 end_packfile();
1011 start_packfile();
1014 static int store_object(
1015 enum object_type type,
1016 struct strbuf *dat,
1017 struct last_object *last,
1018 unsigned char *sha1out,
1019 uintmax_t mark)
1021 void *out, *delta;
1022 struct object_entry *e;
1023 unsigned char hdr[96];
1024 unsigned char sha1[20];
1025 unsigned long hdrlen, deltalen;
1026 git_SHA_CTX c;
1027 git_zstream s;
1029 hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1030 (unsigned long)dat->len) + 1;
1031 git_SHA1_Init(&c);
1032 git_SHA1_Update(&c, hdr, hdrlen);
1033 git_SHA1_Update(&c, dat->buf, dat->len);
1034 git_SHA1_Final(sha1, &c);
1035 if (sha1out)
1036 hashcpy(sha1out, sha1);
1038 e = insert_object(sha1);
1039 if (mark)
1040 insert_mark(mark, e);
1041 if (e->idx.offset) {
1042 duplicate_count_by_type[type]++;
1043 return 1;
1044 } else if (find_sha1_pack(sha1, packed_git)) {
1045 e->type = type;
1046 e->pack_id = MAX_PACK_ID;
1047 e->idx.offset = 1; /* just not zero! */
1048 duplicate_count_by_type[type]++;
1049 return 1;
1052 if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1053 delta_count_attempts_by_type[type]++;
1054 delta = diff_delta(last->data.buf, last->data.len,
1055 dat->buf, dat->len,
1056 &deltalen, dat->len - 20);
1057 } else
1058 delta = NULL;
1060 memset(&s, 0, sizeof(s));
1061 git_deflate_init(&s, pack_compression_level);
1062 if (delta) {
1063 s.next_in = delta;
1064 s.avail_in = deltalen;
1065 } else {
1066 s.next_in = (void *)dat->buf;
1067 s.avail_in = dat->len;
1069 s.avail_out = git_deflate_bound(&s, s.avail_in);
1070 s.next_out = out = xmalloc(s.avail_out);
1071 while (git_deflate(&s, Z_FINISH) == Z_OK)
1072 ; /* nothing */
1073 git_deflate_end(&s);
1075 /* Determine if we should auto-checkpoint. */
1076 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1077 || (pack_size + 60 + s.total_out) < pack_size) {
1079 /* This new object needs to *not* have the current pack_id. */
1080 e->pack_id = pack_id + 1;
1081 cycle_packfile();
1083 /* We cannot carry a delta into the new pack. */
1084 if (delta) {
1085 free(delta);
1086 delta = NULL;
1088 memset(&s, 0, sizeof(s));
1089 git_deflate_init(&s, pack_compression_level);
1090 s.next_in = (void *)dat->buf;
1091 s.avail_in = dat->len;
1092 s.avail_out = git_deflate_bound(&s, s.avail_in);
1093 s.next_out = out = xrealloc(out, s.avail_out);
1094 while (git_deflate(&s, Z_FINISH) == Z_OK)
1095 ; /* nothing */
1096 git_deflate_end(&s);
1100 e->type = type;
1101 e->pack_id = pack_id;
1102 e->idx.offset = pack_size;
1103 object_count++;
1104 object_count_by_type[type]++;
1106 crc32_begin(pack_file);
1108 if (delta) {
1109 off_t ofs = e->idx.offset - last->offset;
1110 unsigned pos = sizeof(hdr) - 1;
1112 delta_count_by_type[type]++;
1113 e->depth = last->depth + 1;
1115 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1116 sha1write(pack_file, hdr, hdrlen);
1117 pack_size += hdrlen;
1119 hdr[pos] = ofs & 127;
1120 while (ofs >>= 7)
1121 hdr[--pos] = 128 | (--ofs & 127);
1122 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1123 pack_size += sizeof(hdr) - pos;
1124 } else {
1125 e->depth = 0;
1126 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1127 sha1write(pack_file, hdr, hdrlen);
1128 pack_size += hdrlen;
1131 sha1write(pack_file, out, s.total_out);
1132 pack_size += s.total_out;
1134 e->idx.crc32 = crc32_end(pack_file);
1136 free(out);
1137 free(delta);
1138 if (last) {
1139 if (last->no_swap) {
1140 last->data = *dat;
1141 } else {
1142 strbuf_swap(&last->data, dat);
1144 last->offset = e->idx.offset;
1145 last->depth = e->depth;
1147 return 0;
1150 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1152 if (sha1file_truncate(pack_file, checkpoint))
1153 die_errno("cannot truncate pack to skip duplicate");
1154 pack_size = checkpoint->offset;
1157 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1159 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1160 unsigned char *in_buf = xmalloc(in_sz);
1161 unsigned char *out_buf = xmalloc(out_sz);
1162 struct object_entry *e;
1163 unsigned char sha1[20];
1164 unsigned long hdrlen;
1165 off_t offset;
1166 git_SHA_CTX c;
1167 git_zstream s;
1168 struct sha1file_checkpoint checkpoint;
1169 int status = Z_OK;
1171 /* Determine if we should auto-checkpoint. */
1172 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1173 || (pack_size + 60 + len) < pack_size)
1174 cycle_packfile();
1176 sha1file_checkpoint(pack_file, &checkpoint);
1177 offset = checkpoint.offset;
1179 hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1180 if (out_sz <= hdrlen)
1181 die("impossibly large object header");
1183 git_SHA1_Init(&c);
1184 git_SHA1_Update(&c, out_buf, hdrlen);
1186 crc32_begin(pack_file);
1188 memset(&s, 0, sizeof(s));
1189 git_deflate_init(&s, pack_compression_level);
1191 hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1192 if (out_sz <= hdrlen)
1193 die("impossibly large object header");
1195 s.next_out = out_buf + hdrlen;
1196 s.avail_out = out_sz - hdrlen;
1198 while (status != Z_STREAM_END) {
1199 if (0 < len && !s.avail_in) {
1200 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1201 size_t n = fread(in_buf, 1, cnt, stdin);
1202 if (!n && feof(stdin))
1203 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1205 git_SHA1_Update(&c, in_buf, n);
1206 s.next_in = in_buf;
1207 s.avail_in = n;
1208 len -= n;
1211 status = git_deflate(&s, len ? 0 : Z_FINISH);
1213 if (!s.avail_out || status == Z_STREAM_END) {
1214 size_t n = s.next_out - out_buf;
1215 sha1write(pack_file, out_buf, n);
1216 pack_size += n;
1217 s.next_out = out_buf;
1218 s.avail_out = out_sz;
1221 switch (status) {
1222 case Z_OK:
1223 case Z_BUF_ERROR:
1224 case Z_STREAM_END:
1225 continue;
1226 default:
1227 die("unexpected deflate failure: %d", status);
1230 git_deflate_end(&s);
1231 git_SHA1_Final(sha1, &c);
1233 if (sha1out)
1234 hashcpy(sha1out, sha1);
1236 e = insert_object(sha1);
1238 if (mark)
1239 insert_mark(mark, e);
1241 if (e->idx.offset) {
1242 duplicate_count_by_type[OBJ_BLOB]++;
1243 truncate_pack(&checkpoint);
1245 } else if (find_sha1_pack(sha1, packed_git)) {
1246 e->type = OBJ_BLOB;
1247 e->pack_id = MAX_PACK_ID;
1248 e->idx.offset = 1; /* just not zero! */
1249 duplicate_count_by_type[OBJ_BLOB]++;
1250 truncate_pack(&checkpoint);
1252 } else {
1253 e->depth = 0;
1254 e->type = OBJ_BLOB;
1255 e->pack_id = pack_id;
1256 e->idx.offset = offset;
1257 e->idx.crc32 = crc32_end(pack_file);
1258 object_count++;
1259 object_count_by_type[OBJ_BLOB]++;
1262 free(in_buf);
1263 free(out_buf);
1266 /* All calls must be guarded by find_object() or find_mark() to
1267 * ensure the 'struct object_entry' passed was written by this
1268 * process instance. We unpack the entry by the offset, avoiding
1269 * the need for the corresponding .idx file. This unpacking rule
1270 * works because we only use OBJ_REF_DELTA within the packfiles
1271 * created by fast-import.
1273 * oe must not be NULL. Such an oe usually comes from giving
1274 * an unknown SHA-1 to find_object() or an undefined mark to
1275 * find_mark(). Callers must test for this condition and use
1276 * the standard read_sha1_file() when it happens.
1278 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1279 * find_mark(), where the mark was reloaded from an existing marks
1280 * file and is referencing an object that this fast-import process
1281 * instance did not write out to a packfile. Callers must test for
1282 * this condition and use read_sha1_file() instead.
1284 static void *gfi_unpack_entry(
1285 struct object_entry *oe,
1286 unsigned long *sizep)
1288 enum object_type type;
1289 struct packed_git *p = all_packs[oe->pack_id];
1290 if (p == pack_data && p->pack_size < (pack_size + 20)) {
1291 /* The object is stored in the packfile we are writing to
1292 * and we have modified it since the last time we scanned
1293 * back to read a previously written object. If an old
1294 * window covered [p->pack_size, p->pack_size + 20) its
1295 * data is stale and is not valid. Closing all windows
1296 * and updating the packfile length ensures we can read
1297 * the newly written data.
1299 close_pack_windows(p);
1300 sha1flush(pack_file);
1302 /* We have to offer 20 bytes additional on the end of
1303 * the packfile as the core unpacker code assumes the
1304 * footer is present at the file end and must promise
1305 * at least 20 bytes within any window it maps. But
1306 * we don't actually create the footer here.
1308 p->pack_size = pack_size + 20;
1310 return unpack_entry(p, oe->idx.offset, &type, sizep);
1313 static const char *get_mode(const char *str, uint16_t *modep)
1315 unsigned char c;
1316 uint16_t mode = 0;
1318 while ((c = *str++) != ' ') {
1319 if (c < '0' || c > '7')
1320 return NULL;
1321 mode = (mode << 3) + (c - '0');
1323 *modep = mode;
1324 return str;
1327 static void load_tree(struct tree_entry *root)
1329 unsigned char *sha1 = root->versions[1].sha1;
1330 struct object_entry *myoe;
1331 struct tree_content *t;
1332 unsigned long size;
1333 char *buf;
1334 const char *c;
1336 root->tree = t = new_tree_content(8);
1337 if (is_null_sha1(sha1))
1338 return;
1340 myoe = find_object(sha1);
1341 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1342 if (myoe->type != OBJ_TREE)
1343 die("Not a tree: %s", sha1_to_hex(sha1));
1344 t->delta_depth = myoe->depth;
1345 buf = gfi_unpack_entry(myoe, &size);
1346 if (!buf)
1347 die("Can't load tree %s", sha1_to_hex(sha1));
1348 } else {
1349 enum object_type type;
1350 buf = read_sha1_file(sha1, &type, &size);
1351 if (!buf || type != OBJ_TREE)
1352 die("Can't load tree %s", sha1_to_hex(sha1));
1355 c = buf;
1356 while (c != (buf + size)) {
1357 struct tree_entry *e = new_tree_entry();
1359 if (t->entry_count == t->entry_capacity)
1360 root->tree = t = grow_tree_content(t, t->entry_count);
1361 t->entries[t->entry_count++] = e;
1363 e->tree = NULL;
1364 c = get_mode(c, &e->versions[1].mode);
1365 if (!c)
1366 die("Corrupt mode in %s", sha1_to_hex(sha1));
1367 e->versions[0].mode = e->versions[1].mode;
1368 e->name = to_atom(c, strlen(c));
1369 c += e->name->str_len + 1;
1370 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1371 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1372 c += 20;
1374 free(buf);
1377 static int tecmp0 (const void *_a, const void *_b)
1379 struct tree_entry *a = *((struct tree_entry**)_a);
1380 struct tree_entry *b = *((struct tree_entry**)_b);
1381 return base_name_compare(
1382 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1383 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1386 static int tecmp1 (const void *_a, const void *_b)
1388 struct tree_entry *a = *((struct tree_entry**)_a);
1389 struct tree_entry *b = *((struct tree_entry**)_b);
1390 return base_name_compare(
1391 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1392 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1395 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1397 size_t maxlen = 0;
1398 unsigned int i;
1400 if (!v)
1401 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1402 else
1403 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1405 for (i = 0; i < t->entry_count; i++) {
1406 if (t->entries[i]->versions[v].mode)
1407 maxlen += t->entries[i]->name->str_len + 34;
1410 strbuf_reset(b);
1411 strbuf_grow(b, maxlen);
1412 for (i = 0; i < t->entry_count; i++) {
1413 struct tree_entry *e = t->entries[i];
1414 if (!e->versions[v].mode)
1415 continue;
1416 strbuf_addf(b, "%o %s%c",
1417 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1418 e->name->str_dat, '\0');
1419 strbuf_add(b, e->versions[v].sha1, 20);
1423 static void store_tree(struct tree_entry *root)
1425 struct tree_content *t;
1426 unsigned int i, j, del;
1427 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1428 struct object_entry *le = NULL;
1430 if (!is_null_sha1(root->versions[1].sha1))
1431 return;
1433 if (!root->tree)
1434 load_tree(root);
1435 t = root->tree;
1437 for (i = 0; i < t->entry_count; i++) {
1438 if (t->entries[i]->tree)
1439 store_tree(t->entries[i]);
1442 if (!(root->versions[0].mode & NO_DELTA))
1443 le = find_object(root->versions[0].sha1);
1444 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1445 mktree(t, 0, &old_tree);
1446 lo.data = old_tree;
1447 lo.offset = le->idx.offset;
1448 lo.depth = t->delta_depth;
1451 mktree(t, 1, &new_tree);
1452 store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1454 t->delta_depth = lo.depth;
1455 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1456 struct tree_entry *e = t->entries[i];
1457 if (e->versions[1].mode) {
1458 e->versions[0].mode = e->versions[1].mode;
1459 hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1460 t->entries[j++] = e;
1461 } else {
1462 release_tree_entry(e);
1463 del++;
1466 t->entry_count -= del;
1469 static void tree_content_replace(
1470 struct tree_entry *root,
1471 const unsigned char *sha1,
1472 const uint16_t mode,
1473 struct tree_content *newtree)
1475 if (!S_ISDIR(mode))
1476 die("Root cannot be a non-directory");
1477 hashclr(root->versions[0].sha1);
1478 hashcpy(root->versions[1].sha1, sha1);
1479 if (root->tree)
1480 release_tree_content_recursive(root->tree);
1481 root->tree = newtree;
1484 static int tree_content_set(
1485 struct tree_entry *root,
1486 const char *p,
1487 const unsigned char *sha1,
1488 const uint16_t mode,
1489 struct tree_content *subtree)
1491 struct tree_content *t;
1492 const char *slash1;
1493 unsigned int i, n;
1494 struct tree_entry *e;
1496 slash1 = strchrnul(p, '/');
1497 n = slash1 - p;
1498 if (!n)
1499 die("Empty path component found in input");
1500 if (!*slash1 && !S_ISDIR(mode) && subtree)
1501 die("Non-directories cannot have subtrees");
1503 if (!root->tree)
1504 load_tree(root);
1505 t = root->tree;
1506 for (i = 0; i < t->entry_count; i++) {
1507 e = t->entries[i];
1508 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1509 if (!*slash1) {
1510 if (!S_ISDIR(mode)
1511 && e->versions[1].mode == mode
1512 && !hashcmp(e->versions[1].sha1, sha1))
1513 return 0;
1514 e->versions[1].mode = mode;
1515 hashcpy(e->versions[1].sha1, sha1);
1516 if (e->tree)
1517 release_tree_content_recursive(e->tree);
1518 e->tree = subtree;
1521 * We need to leave e->versions[0].sha1 alone
1522 * to avoid modifying the preimage tree used
1523 * when writing out the parent directory.
1524 * But after replacing the subdir with a
1525 * completely different one, it's not a good
1526 * delta base any more, and besides, we've
1527 * thrown away the tree entries needed to
1528 * make a delta against it.
1530 * So let's just explicitly disable deltas
1531 * for the subtree.
1533 if (S_ISDIR(e->versions[0].mode))
1534 e->versions[0].mode |= NO_DELTA;
1536 hashclr(root->versions[1].sha1);
1537 return 1;
1539 if (!S_ISDIR(e->versions[1].mode)) {
1540 e->tree = new_tree_content(8);
1541 e->versions[1].mode = S_IFDIR;
1543 if (!e->tree)
1544 load_tree(e);
1545 if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1546 hashclr(root->versions[1].sha1);
1547 return 1;
1549 return 0;
1553 if (t->entry_count == t->entry_capacity)
1554 root->tree = t = grow_tree_content(t, t->entry_count);
1555 e = new_tree_entry();
1556 e->name = to_atom(p, n);
1557 e->versions[0].mode = 0;
1558 hashclr(e->versions[0].sha1);
1559 t->entries[t->entry_count++] = e;
1560 if (*slash1) {
1561 e->tree = new_tree_content(8);
1562 e->versions[1].mode = S_IFDIR;
1563 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1564 } else {
1565 e->tree = subtree;
1566 e->versions[1].mode = mode;
1567 hashcpy(e->versions[1].sha1, sha1);
1569 hashclr(root->versions[1].sha1);
1570 return 1;
1573 static int tree_content_remove(
1574 struct tree_entry *root,
1575 const char *p,
1576 struct tree_entry *backup_leaf,
1577 int allow_root)
1579 struct tree_content *t;
1580 const char *slash1;
1581 unsigned int i, n;
1582 struct tree_entry *e;
1584 slash1 = strchrnul(p, '/');
1585 n = slash1 - p;
1587 if (!root->tree)
1588 load_tree(root);
1590 if (!*p && allow_root) {
1591 e = root;
1592 goto del_entry;
1595 t = root->tree;
1596 for (i = 0; i < t->entry_count; i++) {
1597 e = t->entries[i];
1598 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1599 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1601 * If p names a file in some subdirectory, and a
1602 * file or symlink matching the name of the
1603 * parent directory of p exists, then p cannot
1604 * exist and need not be deleted.
1606 return 1;
1607 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1608 goto del_entry;
1609 if (!e->tree)
1610 load_tree(e);
1611 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1612 for (n = 0; n < e->tree->entry_count; n++) {
1613 if (e->tree->entries[n]->versions[1].mode) {
1614 hashclr(root->versions[1].sha1);
1615 return 1;
1618 backup_leaf = NULL;
1619 goto del_entry;
1621 return 0;
1624 return 0;
1626 del_entry:
1627 if (backup_leaf)
1628 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1629 else if (e->tree)
1630 release_tree_content_recursive(e->tree);
1631 e->tree = NULL;
1632 e->versions[1].mode = 0;
1633 hashclr(e->versions[1].sha1);
1634 hashclr(root->versions[1].sha1);
1635 return 1;
1638 static int tree_content_get(
1639 struct tree_entry *root,
1640 const char *p,
1641 struct tree_entry *leaf,
1642 int allow_root)
1644 struct tree_content *t;
1645 const char *slash1;
1646 unsigned int i, n;
1647 struct tree_entry *e;
1649 slash1 = strchrnul(p, '/');
1650 n = slash1 - p;
1651 if (!n && !allow_root)
1652 die("Empty path component found in input");
1654 if (!root->tree)
1655 load_tree(root);
1657 if (!n) {
1658 e = root;
1659 goto found_entry;
1662 t = root->tree;
1663 for (i = 0; i < t->entry_count; i++) {
1664 e = t->entries[i];
1665 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1666 if (!*slash1)
1667 goto found_entry;
1668 if (!S_ISDIR(e->versions[1].mode))
1669 return 0;
1670 if (!e->tree)
1671 load_tree(e);
1672 return tree_content_get(e, slash1 + 1, leaf, 0);
1675 return 0;
1677 found_entry:
1678 memcpy(leaf, e, sizeof(*leaf));
1679 if (e->tree && is_null_sha1(e->versions[1].sha1))
1680 leaf->tree = dup_tree_content(e->tree);
1681 else
1682 leaf->tree = NULL;
1683 return 1;
1686 static int update_branch(struct branch *b)
1688 static const char *msg = "fast-import";
1689 struct ref_transaction *transaction;
1690 unsigned char old_sha1[20];
1691 struct strbuf err = STRBUF_INIT;
1693 if (read_ref(b->name, old_sha1))
1694 hashclr(old_sha1);
1695 if (is_null_sha1(b->sha1)) {
1696 if (b->delete)
1697 delete_ref(b->name, old_sha1, 0);
1698 return 0;
1700 if (!force_update && !is_null_sha1(old_sha1)) {
1701 struct commit *old_cmit, *new_cmit;
1703 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1704 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1705 if (!old_cmit || !new_cmit)
1706 return error("Branch %s is missing commits.", b->name);
1708 if (!in_merge_bases(old_cmit, new_cmit)) {
1709 warning("Not updating %s"
1710 " (new tip %s does not contain %s)",
1711 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1712 return -1;
1715 transaction = ref_transaction_begin(&err);
1716 if (!transaction ||
1717 ref_transaction_update(transaction, b->name, b->sha1, old_sha1,
1718 0, 1, &err) ||
1719 ref_transaction_commit(transaction, msg, &err)) {
1720 ref_transaction_free(transaction);
1721 error("%s", err.buf);
1722 strbuf_release(&err);
1723 return -1;
1725 ref_transaction_free(transaction);
1726 strbuf_release(&err);
1727 return 0;
1730 static void dump_branches(void)
1732 unsigned int i;
1733 struct branch *b;
1735 for (i = 0; i < branch_table_sz; i++) {
1736 for (b = branch_table[i]; b; b = b->table_next_branch)
1737 failure |= update_branch(b);
1741 static void dump_tags(void)
1743 static const char *msg = "fast-import";
1744 struct tag *t;
1745 struct strbuf ref_name = STRBUF_INIT;
1746 struct strbuf err = STRBUF_INIT;
1747 struct ref_transaction *transaction;
1749 transaction = ref_transaction_begin(&err);
1750 if (!transaction) {
1751 failure |= error("%s", err.buf);
1752 goto cleanup;
1754 for (t = first_tag; t; t = t->next_tag) {
1755 strbuf_reset(&ref_name);
1756 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1758 if (ref_transaction_update(transaction, ref_name.buf, t->sha1,
1759 NULL, 0, 0, &err)) {
1760 failure |= error("%s", err.buf);
1761 goto cleanup;
1764 if (ref_transaction_commit(transaction, msg, &err))
1765 failure |= error("%s", err.buf);
1767 cleanup:
1768 ref_transaction_free(transaction);
1769 strbuf_release(&ref_name);
1770 strbuf_release(&err);
1773 static void dump_marks_helper(FILE *f,
1774 uintmax_t base,
1775 struct mark_set *m)
1777 uintmax_t k;
1778 if (m->shift) {
1779 for (k = 0; k < 1024; k++) {
1780 if (m->data.sets[k])
1781 dump_marks_helper(f, base + (k << m->shift),
1782 m->data.sets[k]);
1784 } else {
1785 for (k = 0; k < 1024; k++) {
1786 if (m->data.marked[k])
1787 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1788 sha1_to_hex(m->data.marked[k]->idx.sha1));
1793 static void dump_marks(void)
1795 static struct lock_file mark_lock;
1796 int mark_fd;
1797 FILE *f;
1799 if (!export_marks_file)
1800 return;
1802 mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1803 if (mark_fd < 0) {
1804 failure |= error("Unable to write marks file %s: %s",
1805 export_marks_file, strerror(errno));
1806 return;
1809 f = fdopen(mark_fd, "w");
1810 if (!f) {
1811 int saved_errno = errno;
1812 rollback_lock_file(&mark_lock);
1813 failure |= error("Unable to write marks file %s: %s",
1814 export_marks_file, strerror(saved_errno));
1815 return;
1819 * Since the lock file was fdopen()'ed, it should not be close()'ed.
1820 * Assign -1 to the lock file descriptor so that commit_lock_file()
1821 * won't try to close() it.
1823 mark_lock.fd = -1;
1825 dump_marks_helper(f, 0, marks);
1826 if (ferror(f) || fclose(f)) {
1827 int saved_errno = errno;
1828 rollback_lock_file(&mark_lock);
1829 failure |= error("Unable to write marks file %s: %s",
1830 export_marks_file, strerror(saved_errno));
1831 return;
1834 if (commit_lock_file(&mark_lock)) {
1835 int saved_errno = errno;
1836 rollback_lock_file(&mark_lock);
1837 failure |= error("Unable to commit marks file %s: %s",
1838 export_marks_file, strerror(saved_errno));
1839 return;
1843 static void read_marks(void)
1845 char line[512];
1846 FILE *f = fopen(import_marks_file, "r");
1847 if (f)
1849 else if (import_marks_file_ignore_missing && errno == ENOENT)
1850 return; /* Marks file does not exist */
1851 else
1852 die_errno("cannot read '%s'", import_marks_file);
1853 while (fgets(line, sizeof(line), f)) {
1854 uintmax_t mark;
1855 char *end;
1856 unsigned char sha1[20];
1857 struct object_entry *e;
1859 end = strchr(line, '\n');
1860 if (line[0] != ':' || !end)
1861 die("corrupt mark line: %s", line);
1862 *end = 0;
1863 mark = strtoumax(line + 1, &end, 10);
1864 if (!mark || end == line + 1
1865 || *end != ' ' || get_sha1_hex(end + 1, sha1))
1866 die("corrupt mark line: %s", line);
1867 e = find_object(sha1);
1868 if (!e) {
1869 enum object_type type = sha1_object_info(sha1, NULL);
1870 if (type < 0)
1871 die("object not found: %s", sha1_to_hex(sha1));
1872 e = insert_object(sha1);
1873 e->type = type;
1874 e->pack_id = MAX_PACK_ID;
1875 e->idx.offset = 1; /* just not zero! */
1877 insert_mark(mark, e);
1879 fclose(f);
1883 static int read_next_command(void)
1885 static int stdin_eof = 0;
1887 if (stdin_eof) {
1888 unread_command_buf = 0;
1889 return EOF;
1892 for (;;) {
1893 const char *p;
1895 if (unread_command_buf) {
1896 unread_command_buf = 0;
1897 } else {
1898 struct recent_command *rc;
1900 strbuf_detach(&command_buf, NULL);
1901 stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1902 if (stdin_eof)
1903 return EOF;
1905 if (!seen_data_command
1906 && !starts_with(command_buf.buf, "feature ")
1907 && !starts_with(command_buf.buf, "option ")) {
1908 parse_argv();
1911 rc = rc_free;
1912 if (rc)
1913 rc_free = rc->next;
1914 else {
1915 rc = cmd_hist.next;
1916 cmd_hist.next = rc->next;
1917 cmd_hist.next->prev = &cmd_hist;
1918 free(rc->buf);
1921 rc->buf = command_buf.buf;
1922 rc->prev = cmd_tail;
1923 rc->next = cmd_hist.prev;
1924 rc->prev->next = rc;
1925 cmd_tail = rc;
1927 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1928 parse_cat_blob(p);
1929 continue;
1931 if (command_buf.buf[0] == '#')
1932 continue;
1933 return 0;
1937 static void skip_optional_lf(void)
1939 int term_char = fgetc(stdin);
1940 if (term_char != '\n' && term_char != EOF)
1941 ungetc(term_char, stdin);
1944 static void parse_mark(void)
1946 const char *v;
1947 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1948 next_mark = strtoumax(v, NULL, 10);
1949 read_next_command();
1951 else
1952 next_mark = 0;
1955 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1957 const char *data;
1958 strbuf_reset(sb);
1960 if (!skip_prefix(command_buf.buf, "data ", &data))
1961 die("Expected 'data n' command, found: %s", command_buf.buf);
1963 if (skip_prefix(data, "<<", &data)) {
1964 char *term = xstrdup(data);
1965 size_t term_len = command_buf.len - (data - command_buf.buf);
1967 strbuf_detach(&command_buf, NULL);
1968 for (;;) {
1969 if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1970 die("EOF in data (terminator '%s' not found)", term);
1971 if (term_len == command_buf.len
1972 && !strcmp(term, command_buf.buf))
1973 break;
1974 strbuf_addbuf(sb, &command_buf);
1975 strbuf_addch(sb, '\n');
1977 free(term);
1979 else {
1980 uintmax_t len = strtoumax(data, NULL, 10);
1981 size_t n = 0, length = (size_t)len;
1983 if (limit && limit < len) {
1984 *len_res = len;
1985 return 0;
1987 if (length < len)
1988 die("data is too large to use in this context");
1990 while (n < length) {
1991 size_t s = strbuf_fread(sb, length - n, stdin);
1992 if (!s && feof(stdin))
1993 die("EOF in data (%lu bytes remaining)",
1994 (unsigned long)(length - n));
1995 n += s;
1999 skip_optional_lf();
2000 return 1;
2003 static int validate_raw_date(const char *src, struct strbuf *result)
2005 const char *orig_src = src;
2006 char *endp;
2007 unsigned long num;
2009 errno = 0;
2011 num = strtoul(src, &endp, 10);
2012 /* NEEDSWORK: perhaps check for reasonable values? */
2013 if (errno || endp == src || *endp != ' ')
2014 return -1;
2016 src = endp + 1;
2017 if (*src != '-' && *src != '+')
2018 return -1;
2020 num = strtoul(src + 1, &endp, 10);
2021 if (errno || endp == src + 1 || *endp || 1400 < num)
2022 return -1;
2024 strbuf_addstr(result, orig_src);
2025 return 0;
2028 static char *parse_ident(const char *buf)
2030 const char *ltgt;
2031 size_t name_len;
2032 struct strbuf ident = STRBUF_INIT;
2034 /* ensure there is a space delimiter even if there is no name */
2035 if (*buf == '<')
2036 --buf;
2038 ltgt = buf + strcspn(buf, "<>");
2039 if (*ltgt != '<')
2040 die("Missing < in ident string: %s", buf);
2041 if (ltgt != buf && ltgt[-1] != ' ')
2042 die("Missing space before < in ident string: %s", buf);
2043 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2044 if (*ltgt != '>')
2045 die("Missing > in ident string: %s", buf);
2046 ltgt++;
2047 if (*ltgt != ' ')
2048 die("Missing space after > in ident string: %s", buf);
2049 ltgt++;
2050 name_len = ltgt - buf;
2051 strbuf_add(&ident, buf, name_len);
2053 switch (whenspec) {
2054 case WHENSPEC_RAW:
2055 if (validate_raw_date(ltgt, &ident) < 0)
2056 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2057 break;
2058 case WHENSPEC_RFC2822:
2059 if (parse_date(ltgt, &ident) < 0)
2060 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2061 break;
2062 case WHENSPEC_NOW:
2063 if (strcmp("now", ltgt))
2064 die("Date in ident must be 'now': %s", buf);
2065 datestamp(&ident);
2066 break;
2069 return strbuf_detach(&ident, NULL);
2072 static void parse_and_store_blob(
2073 struct last_object *last,
2074 unsigned char *sha1out,
2075 uintmax_t mark)
2077 static struct strbuf buf = STRBUF_INIT;
2078 uintmax_t len;
2080 if (parse_data(&buf, big_file_threshold, &len))
2081 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2082 else {
2083 if (last) {
2084 strbuf_release(&last->data);
2085 last->offset = 0;
2086 last->depth = 0;
2088 stream_blob(len, sha1out, mark);
2089 skip_optional_lf();
2093 static void parse_new_blob(void)
2095 read_next_command();
2096 parse_mark();
2097 parse_and_store_blob(&last_blob, NULL, next_mark);
2100 static void unload_one_branch(void)
2102 while (cur_active_branches
2103 && cur_active_branches >= max_active_branches) {
2104 uintmax_t min_commit = ULONG_MAX;
2105 struct branch *e, *l = NULL, *p = NULL;
2107 for (e = active_branches; e; e = e->active_next_branch) {
2108 if (e->last_commit < min_commit) {
2109 p = l;
2110 min_commit = e->last_commit;
2112 l = e;
2115 if (p) {
2116 e = p->active_next_branch;
2117 p->active_next_branch = e->active_next_branch;
2118 } else {
2119 e = active_branches;
2120 active_branches = e->active_next_branch;
2122 e->active = 0;
2123 e->active_next_branch = NULL;
2124 if (e->branch_tree.tree) {
2125 release_tree_content_recursive(e->branch_tree.tree);
2126 e->branch_tree.tree = NULL;
2128 cur_active_branches--;
2132 static void load_branch(struct branch *b)
2134 load_tree(&b->branch_tree);
2135 if (!b->active) {
2136 b->active = 1;
2137 b->active_next_branch = active_branches;
2138 active_branches = b;
2139 cur_active_branches++;
2140 branch_load_count++;
2144 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2146 unsigned char fanout = 0;
2147 while ((num_notes >>= 8))
2148 fanout++;
2149 return fanout;
2152 static void construct_path_with_fanout(const char *hex_sha1,
2153 unsigned char fanout, char *path)
2155 unsigned int i = 0, j = 0;
2156 if (fanout >= 20)
2157 die("Too large fanout (%u)", fanout);
2158 while (fanout) {
2159 path[i++] = hex_sha1[j++];
2160 path[i++] = hex_sha1[j++];
2161 path[i++] = '/';
2162 fanout--;
2164 memcpy(path + i, hex_sha1 + j, 40 - j);
2165 path[i + 40 - j] = '\0';
2168 static uintmax_t do_change_note_fanout(
2169 struct tree_entry *orig_root, struct tree_entry *root,
2170 char *hex_sha1, unsigned int hex_sha1_len,
2171 char *fullpath, unsigned int fullpath_len,
2172 unsigned char fanout)
2174 struct tree_content *t = root->tree;
2175 struct tree_entry *e, leaf;
2176 unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2177 uintmax_t num_notes = 0;
2178 unsigned char sha1[20];
2179 char realpath[60];
2181 for (i = 0; t && i < t->entry_count; i++) {
2182 e = t->entries[i];
2183 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2184 tmp_fullpath_len = fullpath_len;
2187 * We're interested in EITHER existing note entries (entries
2188 * with exactly 40 hex chars in path, not including directory
2189 * separators), OR directory entries that may contain note
2190 * entries (with < 40 hex chars in path).
2191 * Also, each path component in a note entry must be a multiple
2192 * of 2 chars.
2194 if (!e->versions[1].mode ||
2195 tmp_hex_sha1_len > 40 ||
2196 e->name->str_len % 2)
2197 continue;
2199 /* This _may_ be a note entry, or a subdir containing notes */
2200 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2201 e->name->str_len);
2202 if (tmp_fullpath_len)
2203 fullpath[tmp_fullpath_len++] = '/';
2204 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2205 e->name->str_len);
2206 tmp_fullpath_len += e->name->str_len;
2207 fullpath[tmp_fullpath_len] = '\0';
2209 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2210 /* This is a note entry */
2211 if (fanout == 0xff) {
2212 /* Counting mode, no rename */
2213 num_notes++;
2214 continue;
2216 construct_path_with_fanout(hex_sha1, fanout, realpath);
2217 if (!strcmp(fullpath, realpath)) {
2218 /* Note entry is in correct location */
2219 num_notes++;
2220 continue;
2223 /* Rename fullpath to realpath */
2224 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2225 die("Failed to remove path %s", fullpath);
2226 tree_content_set(orig_root, realpath,
2227 leaf.versions[1].sha1,
2228 leaf.versions[1].mode,
2229 leaf.tree);
2230 } else if (S_ISDIR(e->versions[1].mode)) {
2231 /* This is a subdir that may contain note entries */
2232 if (!e->tree)
2233 load_tree(e);
2234 num_notes += do_change_note_fanout(orig_root, e,
2235 hex_sha1, tmp_hex_sha1_len,
2236 fullpath, tmp_fullpath_len, fanout);
2239 /* The above may have reallocated the current tree_content */
2240 t = root->tree;
2242 return num_notes;
2245 static uintmax_t change_note_fanout(struct tree_entry *root,
2246 unsigned char fanout)
2248 char hex_sha1[40], path[60];
2249 return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2253 * Given a pointer into a string, parse a mark reference:
2255 * idnum ::= ':' bigint;
2257 * Return the first character after the value in *endptr.
2259 * Complain if the following character is not what is expected,
2260 * either a space or end of the string.
2262 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2264 uintmax_t mark;
2266 assert(*p == ':');
2267 p++;
2268 mark = strtoumax(p, endptr, 10);
2269 if (*endptr == p)
2270 die("No value after ':' in mark: %s", command_buf.buf);
2271 return mark;
2275 * Parse the mark reference, and complain if this is not the end of
2276 * the string.
2278 static uintmax_t parse_mark_ref_eol(const char *p)
2280 char *end;
2281 uintmax_t mark;
2283 mark = parse_mark_ref(p, &end);
2284 if (*end != '\0')
2285 die("Garbage after mark: %s", command_buf.buf);
2286 return mark;
2290 * Parse the mark reference, demanding a trailing space. Return a
2291 * pointer to the space.
2293 static uintmax_t parse_mark_ref_space(const char **p)
2295 uintmax_t mark;
2296 char *end;
2298 mark = parse_mark_ref(*p, &end);
2299 if (*end++ != ' ')
2300 die("Missing space after mark: %s", command_buf.buf);
2301 *p = end;
2302 return mark;
2305 static void file_change_m(const char *p, struct branch *b)
2307 static struct strbuf uq = STRBUF_INIT;
2308 const char *endp;
2309 struct object_entry *oe;
2310 unsigned char sha1[20];
2311 uint16_t mode, inline_data = 0;
2313 p = get_mode(p, &mode);
2314 if (!p)
2315 die("Corrupt mode: %s", command_buf.buf);
2316 switch (mode) {
2317 case 0644:
2318 case 0755:
2319 mode |= S_IFREG;
2320 case S_IFREG | 0644:
2321 case S_IFREG | 0755:
2322 case S_IFLNK:
2323 case S_IFDIR:
2324 case S_IFGITLINK:
2325 /* ok */
2326 break;
2327 default:
2328 die("Corrupt mode: %s", command_buf.buf);
2331 if (*p == ':') {
2332 oe = find_mark(parse_mark_ref_space(&p));
2333 hashcpy(sha1, oe->idx.sha1);
2334 } else if (skip_prefix(p, "inline ", &p)) {
2335 inline_data = 1;
2336 oe = NULL; /* not used with inline_data, but makes gcc happy */
2337 } else {
2338 if (get_sha1_hex(p, sha1))
2339 die("Invalid dataref: %s", command_buf.buf);
2340 oe = find_object(sha1);
2341 p += 40;
2342 if (*p++ != ' ')
2343 die("Missing space after SHA1: %s", command_buf.buf);
2346 strbuf_reset(&uq);
2347 if (!unquote_c_style(&uq, p, &endp)) {
2348 if (*endp)
2349 die("Garbage after path in: %s", command_buf.buf);
2350 p = uq.buf;
2353 /* Git does not track empty, non-toplevel directories. */
2354 if (S_ISDIR(mode) && !hashcmp(sha1, EMPTY_TREE_SHA1_BIN) && *p) {
2355 tree_content_remove(&b->branch_tree, p, NULL, 0);
2356 return;
2359 if (S_ISGITLINK(mode)) {
2360 if (inline_data)
2361 die("Git links cannot be specified 'inline': %s",
2362 command_buf.buf);
2363 else if (oe) {
2364 if (oe->type != OBJ_COMMIT)
2365 die("Not a commit (actually a %s): %s",
2366 typename(oe->type), command_buf.buf);
2369 * Accept the sha1 without checking; it expected to be in
2370 * another repository.
2372 } else if (inline_data) {
2373 if (S_ISDIR(mode))
2374 die("Directories cannot be specified 'inline': %s",
2375 command_buf.buf);
2376 if (p != uq.buf) {
2377 strbuf_addstr(&uq, p);
2378 p = uq.buf;
2380 read_next_command();
2381 parse_and_store_blob(&last_blob, sha1, 0);
2382 } else {
2383 enum object_type expected = S_ISDIR(mode) ?
2384 OBJ_TREE: OBJ_BLOB;
2385 enum object_type type = oe ? oe->type :
2386 sha1_object_info(sha1, NULL);
2387 if (type < 0)
2388 die("%s not found: %s",
2389 S_ISDIR(mode) ? "Tree" : "Blob",
2390 command_buf.buf);
2391 if (type != expected)
2392 die("Not a %s (actually a %s): %s",
2393 typename(expected), typename(type),
2394 command_buf.buf);
2397 if (!*p) {
2398 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2399 return;
2401 tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2404 static void file_change_d(const char *p, struct branch *b)
2406 static struct strbuf uq = STRBUF_INIT;
2407 const char *endp;
2409 strbuf_reset(&uq);
2410 if (!unquote_c_style(&uq, p, &endp)) {
2411 if (*endp)
2412 die("Garbage after path in: %s", command_buf.buf);
2413 p = uq.buf;
2415 tree_content_remove(&b->branch_tree, p, NULL, 1);
2418 static void file_change_cr(const char *s, struct branch *b, int rename)
2420 const char *d;
2421 static struct strbuf s_uq = STRBUF_INIT;
2422 static struct strbuf d_uq = STRBUF_INIT;
2423 const char *endp;
2424 struct tree_entry leaf;
2426 strbuf_reset(&s_uq);
2427 if (!unquote_c_style(&s_uq, s, &endp)) {
2428 if (*endp != ' ')
2429 die("Missing space after source: %s", command_buf.buf);
2430 } else {
2431 endp = strchr(s, ' ');
2432 if (!endp)
2433 die("Missing space after source: %s", command_buf.buf);
2434 strbuf_add(&s_uq, s, endp - s);
2436 s = s_uq.buf;
2438 endp++;
2439 if (!*endp)
2440 die("Missing dest: %s", command_buf.buf);
2442 d = endp;
2443 strbuf_reset(&d_uq);
2444 if (!unquote_c_style(&d_uq, d, &endp)) {
2445 if (*endp)
2446 die("Garbage after dest in: %s", command_buf.buf);
2447 d = d_uq.buf;
2450 memset(&leaf, 0, sizeof(leaf));
2451 if (rename)
2452 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2453 else
2454 tree_content_get(&b->branch_tree, s, &leaf, 1);
2455 if (!leaf.versions[1].mode)
2456 die("Path %s not in branch", s);
2457 if (!*d) { /* C "path/to/subdir" "" */
2458 tree_content_replace(&b->branch_tree,
2459 leaf.versions[1].sha1,
2460 leaf.versions[1].mode,
2461 leaf.tree);
2462 return;
2464 tree_content_set(&b->branch_tree, d,
2465 leaf.versions[1].sha1,
2466 leaf.versions[1].mode,
2467 leaf.tree);
2470 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2472 static struct strbuf uq = STRBUF_INIT;
2473 struct object_entry *oe;
2474 struct branch *s;
2475 unsigned char sha1[20], commit_sha1[20];
2476 char path[60];
2477 uint16_t inline_data = 0;
2478 unsigned char new_fanout;
2481 * When loading a branch, we don't traverse its tree to count the real
2482 * number of notes (too expensive to do this for all non-note refs).
2483 * This means that recently loaded notes refs might incorrectly have
2484 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2486 * Fix this by traversing the tree and counting the number of notes
2487 * when b->num_notes == 0. If the notes tree is truly empty, the
2488 * calculation should not take long.
2490 if (b->num_notes == 0 && *old_fanout == 0) {
2491 /* Invoke change_note_fanout() in "counting mode". */
2492 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2493 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2496 /* Now parse the notemodify command. */
2497 /* <dataref> or 'inline' */
2498 if (*p == ':') {
2499 oe = find_mark(parse_mark_ref_space(&p));
2500 hashcpy(sha1, oe->idx.sha1);
2501 } else if (skip_prefix(p, "inline ", &p)) {
2502 inline_data = 1;
2503 oe = NULL; /* not used with inline_data, but makes gcc happy */
2504 } else {
2505 if (get_sha1_hex(p, sha1))
2506 die("Invalid dataref: %s", command_buf.buf);
2507 oe = find_object(sha1);
2508 p += 40;
2509 if (*p++ != ' ')
2510 die("Missing space after SHA1: %s", command_buf.buf);
2513 /* <commit-ish> */
2514 s = lookup_branch(p);
2515 if (s) {
2516 if (is_null_sha1(s->sha1))
2517 die("Can't add a note on empty branch.");
2518 hashcpy(commit_sha1, s->sha1);
2519 } else if (*p == ':') {
2520 uintmax_t commit_mark = parse_mark_ref_eol(p);
2521 struct object_entry *commit_oe = find_mark(commit_mark);
2522 if (commit_oe->type != OBJ_COMMIT)
2523 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2524 hashcpy(commit_sha1, commit_oe->idx.sha1);
2525 } else if (!get_sha1(p, commit_sha1)) {
2526 unsigned long size;
2527 char *buf = read_object_with_reference(commit_sha1,
2528 commit_type, &size, commit_sha1);
2529 if (!buf || size < 46)
2530 die("Not a valid commit: %s", p);
2531 free(buf);
2532 } else
2533 die("Invalid ref name or SHA1 expression: %s", p);
2535 if (inline_data) {
2536 if (p != uq.buf) {
2537 strbuf_addstr(&uq, p);
2538 p = uq.buf;
2540 read_next_command();
2541 parse_and_store_blob(&last_blob, sha1, 0);
2542 } else if (oe) {
2543 if (oe->type != OBJ_BLOB)
2544 die("Not a blob (actually a %s): %s",
2545 typename(oe->type), command_buf.buf);
2546 } else if (!is_null_sha1(sha1)) {
2547 enum object_type type = sha1_object_info(sha1, NULL);
2548 if (type < 0)
2549 die("Blob not found: %s", command_buf.buf);
2550 if (type != OBJ_BLOB)
2551 die("Not a blob (actually a %s): %s",
2552 typename(type), command_buf.buf);
2555 construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2556 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2557 b->num_notes--;
2559 if (is_null_sha1(sha1))
2560 return; /* nothing to insert */
2562 b->num_notes++;
2563 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2564 construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2565 tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2568 static void file_change_deleteall(struct branch *b)
2570 release_tree_content_recursive(b->branch_tree.tree);
2571 hashclr(b->branch_tree.versions[0].sha1);
2572 hashclr(b->branch_tree.versions[1].sha1);
2573 load_tree(&b->branch_tree);
2574 b->num_notes = 0;
2577 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2579 if (!buf || size < 46)
2580 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2581 if (memcmp("tree ", buf, 5)
2582 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2583 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2584 hashcpy(b->branch_tree.versions[0].sha1,
2585 b->branch_tree.versions[1].sha1);
2588 static void parse_from_existing(struct branch *b)
2590 if (is_null_sha1(b->sha1)) {
2591 hashclr(b->branch_tree.versions[0].sha1);
2592 hashclr(b->branch_tree.versions[1].sha1);
2593 } else {
2594 unsigned long size;
2595 char *buf;
2597 buf = read_object_with_reference(b->sha1,
2598 commit_type, &size, b->sha1);
2599 parse_from_commit(b, buf, size);
2600 free(buf);
2604 static int parse_from(struct branch *b)
2606 const char *from;
2607 struct branch *s;
2609 if (!skip_prefix(command_buf.buf, "from ", &from))
2610 return 0;
2612 if (b->branch_tree.tree) {
2613 release_tree_content_recursive(b->branch_tree.tree);
2614 b->branch_tree.tree = NULL;
2617 s = lookup_branch(from);
2618 if (b == s)
2619 die("Can't create a branch from itself: %s", b->name);
2620 else if (s) {
2621 unsigned char *t = s->branch_tree.versions[1].sha1;
2622 hashcpy(b->sha1, s->sha1);
2623 hashcpy(b->branch_tree.versions[0].sha1, t);
2624 hashcpy(b->branch_tree.versions[1].sha1, t);
2625 } else if (*from == ':') {
2626 uintmax_t idnum = parse_mark_ref_eol(from);
2627 struct object_entry *oe = find_mark(idnum);
2628 if (oe->type != OBJ_COMMIT)
2629 die("Mark :%" PRIuMAX " not a commit", idnum);
2630 hashcpy(b->sha1, oe->idx.sha1);
2631 if (oe->pack_id != MAX_PACK_ID) {
2632 unsigned long size;
2633 char *buf = gfi_unpack_entry(oe, &size);
2634 parse_from_commit(b, buf, size);
2635 free(buf);
2636 } else
2637 parse_from_existing(b);
2638 } else if (!get_sha1(from, b->sha1)) {
2639 parse_from_existing(b);
2640 if (is_null_sha1(b->sha1))
2641 b->delete = 1;
2643 else
2644 die("Invalid ref name or SHA1 expression: %s", from);
2646 read_next_command();
2647 return 1;
2650 static struct hash_list *parse_merge(unsigned int *count)
2652 struct hash_list *list = NULL, **tail = &list, *n;
2653 const char *from;
2654 struct branch *s;
2656 *count = 0;
2657 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2658 n = xmalloc(sizeof(*n));
2659 s = lookup_branch(from);
2660 if (s)
2661 hashcpy(n->sha1, s->sha1);
2662 else if (*from == ':') {
2663 uintmax_t idnum = parse_mark_ref_eol(from);
2664 struct object_entry *oe = find_mark(idnum);
2665 if (oe->type != OBJ_COMMIT)
2666 die("Mark :%" PRIuMAX " not a commit", idnum);
2667 hashcpy(n->sha1, oe->idx.sha1);
2668 } else if (!get_sha1(from, n->sha1)) {
2669 unsigned long size;
2670 char *buf = read_object_with_reference(n->sha1,
2671 commit_type, &size, n->sha1);
2672 if (!buf || size < 46)
2673 die("Not a valid commit: %s", from);
2674 free(buf);
2675 } else
2676 die("Invalid ref name or SHA1 expression: %s", from);
2678 n->next = NULL;
2679 *tail = n;
2680 tail = &n->next;
2682 (*count)++;
2683 read_next_command();
2685 return list;
2688 static void parse_new_commit(const char *arg)
2690 static struct strbuf msg = STRBUF_INIT;
2691 struct branch *b;
2692 char *author = NULL;
2693 char *committer = NULL;
2694 struct hash_list *merge_list = NULL;
2695 unsigned int merge_count;
2696 unsigned char prev_fanout, new_fanout;
2697 const char *v;
2699 b = lookup_branch(arg);
2700 if (!b)
2701 b = new_branch(arg);
2703 read_next_command();
2704 parse_mark();
2705 if (skip_prefix(command_buf.buf, "author ", &v)) {
2706 author = parse_ident(v);
2707 read_next_command();
2709 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2710 committer = parse_ident(v);
2711 read_next_command();
2713 if (!committer)
2714 die("Expected committer but didn't get one");
2715 parse_data(&msg, 0, NULL);
2716 read_next_command();
2717 parse_from(b);
2718 merge_list = parse_merge(&merge_count);
2720 /* ensure the branch is active/loaded */
2721 if (!b->branch_tree.tree || !max_active_branches) {
2722 unload_one_branch();
2723 load_branch(b);
2726 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2728 /* file_change* */
2729 while (command_buf.len > 0) {
2730 if (skip_prefix(command_buf.buf, "M ", &v))
2731 file_change_m(v, b);
2732 else if (skip_prefix(command_buf.buf, "D ", &v))
2733 file_change_d(v, b);
2734 else if (skip_prefix(command_buf.buf, "R ", &v))
2735 file_change_cr(v, b, 1);
2736 else if (skip_prefix(command_buf.buf, "C ", &v))
2737 file_change_cr(v, b, 0);
2738 else if (skip_prefix(command_buf.buf, "N ", &v))
2739 note_change_n(v, b, &prev_fanout);
2740 else if (!strcmp("deleteall", command_buf.buf))
2741 file_change_deleteall(b);
2742 else if (skip_prefix(command_buf.buf, "ls ", &v))
2743 parse_ls(v, b);
2744 else {
2745 unread_command_buf = 1;
2746 break;
2748 if (read_next_command() == EOF)
2749 break;
2752 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2753 if (new_fanout != prev_fanout)
2754 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2756 /* build the tree and the commit */
2757 store_tree(&b->branch_tree);
2758 hashcpy(b->branch_tree.versions[0].sha1,
2759 b->branch_tree.versions[1].sha1);
2761 strbuf_reset(&new_data);
2762 strbuf_addf(&new_data, "tree %s\n",
2763 sha1_to_hex(b->branch_tree.versions[1].sha1));
2764 if (!is_null_sha1(b->sha1))
2765 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2766 while (merge_list) {
2767 struct hash_list *next = merge_list->next;
2768 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2769 free(merge_list);
2770 merge_list = next;
2772 strbuf_addf(&new_data,
2773 "author %s\n"
2774 "committer %s\n"
2775 "\n",
2776 author ? author : committer, committer);
2777 strbuf_addbuf(&new_data, &msg);
2778 free(author);
2779 free(committer);
2781 if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2782 b->pack_id = pack_id;
2783 b->last_commit = object_count_by_type[OBJ_COMMIT];
2786 static void parse_new_tag(const char *arg)
2788 static struct strbuf msg = STRBUF_INIT;
2789 const char *from;
2790 char *tagger;
2791 struct branch *s;
2792 struct tag *t;
2793 uintmax_t from_mark = 0;
2794 unsigned char sha1[20];
2795 enum object_type type;
2796 const char *v;
2798 t = pool_alloc(sizeof(struct tag));
2799 memset(t, 0, sizeof(struct tag));
2800 t->name = pool_strdup(arg);
2801 if (last_tag)
2802 last_tag->next_tag = t;
2803 else
2804 first_tag = t;
2805 last_tag = t;
2806 read_next_command();
2808 /* from ... */
2809 if (!skip_prefix(command_buf.buf, "from ", &from))
2810 die("Expected from command, got %s", command_buf.buf);
2811 s = lookup_branch(from);
2812 if (s) {
2813 if (is_null_sha1(s->sha1))
2814 die("Can't tag an empty branch.");
2815 hashcpy(sha1, s->sha1);
2816 type = OBJ_COMMIT;
2817 } else if (*from == ':') {
2818 struct object_entry *oe;
2819 from_mark = parse_mark_ref_eol(from);
2820 oe = find_mark(from_mark);
2821 type = oe->type;
2822 hashcpy(sha1, oe->idx.sha1);
2823 } else if (!get_sha1(from, sha1)) {
2824 struct object_entry *oe = find_object(sha1);
2825 if (!oe) {
2826 type = sha1_object_info(sha1, NULL);
2827 if (type < 0)
2828 die("Not a valid object: %s", from);
2829 } else
2830 type = oe->type;
2831 } else
2832 die("Invalid ref name or SHA1 expression: %s", from);
2833 read_next_command();
2835 /* tagger ... */
2836 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2837 tagger = parse_ident(v);
2838 read_next_command();
2839 } else
2840 tagger = NULL;
2842 /* tag payload/message */
2843 parse_data(&msg, 0, NULL);
2845 /* build the tag object */
2846 strbuf_reset(&new_data);
2848 strbuf_addf(&new_data,
2849 "object %s\n"
2850 "type %s\n"
2851 "tag %s\n",
2852 sha1_to_hex(sha1), typename(type), t->name);
2853 if (tagger)
2854 strbuf_addf(&new_data,
2855 "tagger %s\n", tagger);
2856 strbuf_addch(&new_data, '\n');
2857 strbuf_addbuf(&new_data, &msg);
2858 free(tagger);
2860 if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2861 t->pack_id = MAX_PACK_ID;
2862 else
2863 t->pack_id = pack_id;
2866 static void parse_reset_branch(const char *arg)
2868 struct branch *b;
2870 b = lookup_branch(arg);
2871 if (b) {
2872 hashclr(b->sha1);
2873 hashclr(b->branch_tree.versions[0].sha1);
2874 hashclr(b->branch_tree.versions[1].sha1);
2875 if (b->branch_tree.tree) {
2876 release_tree_content_recursive(b->branch_tree.tree);
2877 b->branch_tree.tree = NULL;
2880 else
2881 b = new_branch(arg);
2882 read_next_command();
2883 parse_from(b);
2884 if (command_buf.len > 0)
2885 unread_command_buf = 1;
2888 static void cat_blob_write(const char *buf, unsigned long size)
2890 if (write_in_full(cat_blob_fd, buf, size) != size)
2891 die_errno("Write to frontend failed");
2894 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2896 struct strbuf line = STRBUF_INIT;
2897 unsigned long size;
2898 enum object_type type = 0;
2899 char *buf;
2901 if (!oe || oe->pack_id == MAX_PACK_ID) {
2902 buf = read_sha1_file(sha1, &type, &size);
2903 } else {
2904 type = oe->type;
2905 buf = gfi_unpack_entry(oe, &size);
2909 * Output based on batch_one_object() from cat-file.c.
2911 if (type <= 0) {
2912 strbuf_reset(&line);
2913 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2914 cat_blob_write(line.buf, line.len);
2915 strbuf_release(&line);
2916 free(buf);
2917 return;
2919 if (!buf)
2920 die("Can't read object %s", sha1_to_hex(sha1));
2921 if (type != OBJ_BLOB)
2922 die("Object %s is a %s but a blob was expected.",
2923 sha1_to_hex(sha1), typename(type));
2924 strbuf_reset(&line);
2925 strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2926 typename(type), size);
2927 cat_blob_write(line.buf, line.len);
2928 strbuf_release(&line);
2929 cat_blob_write(buf, size);
2930 cat_blob_write("\n", 1);
2931 if (oe && oe->pack_id == pack_id) {
2932 last_blob.offset = oe->idx.offset;
2933 strbuf_attach(&last_blob.data, buf, size, size);
2934 last_blob.depth = oe->depth;
2935 } else
2936 free(buf);
2939 static void parse_cat_blob(const char *p)
2941 struct object_entry *oe = oe;
2942 unsigned char sha1[20];
2944 /* cat-blob SP <object> LF */
2945 if (*p == ':') {
2946 oe = find_mark(parse_mark_ref_eol(p));
2947 if (!oe)
2948 die("Unknown mark: %s", command_buf.buf);
2949 hashcpy(sha1, oe->idx.sha1);
2950 } else {
2951 if (get_sha1_hex(p, sha1))
2952 die("Invalid dataref: %s", command_buf.buf);
2953 if (p[40])
2954 die("Garbage after SHA1: %s", command_buf.buf);
2955 oe = find_object(sha1);
2958 cat_blob(oe, sha1);
2961 static struct object_entry *dereference(struct object_entry *oe,
2962 unsigned char sha1[20])
2964 unsigned long size;
2965 char *buf = NULL;
2966 if (!oe) {
2967 enum object_type type = sha1_object_info(sha1, NULL);
2968 if (type < 0)
2969 die("object not found: %s", sha1_to_hex(sha1));
2970 /* cache it! */
2971 oe = insert_object(sha1);
2972 oe->type = type;
2973 oe->pack_id = MAX_PACK_ID;
2974 oe->idx.offset = 1;
2976 switch (oe->type) {
2977 case OBJ_TREE: /* easy case. */
2978 return oe;
2979 case OBJ_COMMIT:
2980 case OBJ_TAG:
2981 break;
2982 default:
2983 die("Not a tree-ish: %s", command_buf.buf);
2986 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
2987 buf = gfi_unpack_entry(oe, &size);
2988 } else {
2989 enum object_type unused;
2990 buf = read_sha1_file(sha1, &unused, &size);
2992 if (!buf)
2993 die("Can't load object %s", sha1_to_hex(sha1));
2995 /* Peel one layer. */
2996 switch (oe->type) {
2997 case OBJ_TAG:
2998 if (size < 40 + strlen("object ") ||
2999 get_sha1_hex(buf + strlen("object "), sha1))
3000 die("Invalid SHA1 in tag: %s", command_buf.buf);
3001 break;
3002 case OBJ_COMMIT:
3003 if (size < 40 + strlen("tree ") ||
3004 get_sha1_hex(buf + strlen("tree "), sha1))
3005 die("Invalid SHA1 in commit: %s", command_buf.buf);
3008 free(buf);
3009 return find_object(sha1);
3012 static struct object_entry *parse_treeish_dataref(const char **p)
3014 unsigned char sha1[20];
3015 struct object_entry *e;
3017 if (**p == ':') { /* <mark> */
3018 e = find_mark(parse_mark_ref_space(p));
3019 if (!e)
3020 die("Unknown mark: %s", command_buf.buf);
3021 hashcpy(sha1, e->idx.sha1);
3022 } else { /* <sha1> */
3023 if (get_sha1_hex(*p, sha1))
3024 die("Invalid dataref: %s", command_buf.buf);
3025 e = find_object(sha1);
3026 *p += 40;
3027 if (*(*p)++ != ' ')
3028 die("Missing space after tree-ish: %s", command_buf.buf);
3031 while (!e || e->type != OBJ_TREE)
3032 e = dereference(e, sha1);
3033 return e;
3036 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3038 static struct strbuf line = STRBUF_INIT;
3040 /* See show_tree(). */
3041 const char *type =
3042 S_ISGITLINK(mode) ? commit_type :
3043 S_ISDIR(mode) ? tree_type :
3044 blob_type;
3046 if (!mode) {
3047 /* missing SP path LF */
3048 strbuf_reset(&line);
3049 strbuf_addstr(&line, "missing ");
3050 quote_c_style(path, &line, NULL, 0);
3051 strbuf_addch(&line, '\n');
3052 } else {
3053 /* mode SP type SP object_name TAB path LF */
3054 strbuf_reset(&line);
3055 strbuf_addf(&line, "%06o %s %s\t",
3056 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3057 quote_c_style(path, &line, NULL, 0);
3058 strbuf_addch(&line, '\n');
3060 cat_blob_write(line.buf, line.len);
3063 static void parse_ls(const char *p, struct branch *b)
3065 struct tree_entry *root = NULL;
3066 struct tree_entry leaf = {NULL};
3068 /* ls SP (<tree-ish> SP)? <path> */
3069 if (*p == '"') {
3070 if (!b)
3071 die("Not in a commit: %s", command_buf.buf);
3072 root = &b->branch_tree;
3073 } else {
3074 struct object_entry *e = parse_treeish_dataref(&p);
3075 root = new_tree_entry();
3076 hashcpy(root->versions[1].sha1, e->idx.sha1);
3077 if (!is_null_sha1(root->versions[1].sha1))
3078 root->versions[1].mode = S_IFDIR;
3079 load_tree(root);
3081 if (*p == '"') {
3082 static struct strbuf uq = STRBUF_INIT;
3083 const char *endp;
3084 strbuf_reset(&uq);
3085 if (unquote_c_style(&uq, p, &endp))
3086 die("Invalid path: %s", command_buf.buf);
3087 if (*endp)
3088 die("Garbage after path in: %s", command_buf.buf);
3089 p = uq.buf;
3091 tree_content_get(root, p, &leaf, 1);
3093 * A directory in preparation would have a sha1 of zero
3094 * until it is saved. Save, for simplicity.
3096 if (S_ISDIR(leaf.versions[1].mode))
3097 store_tree(&leaf);
3099 print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3100 if (leaf.tree)
3101 release_tree_content_recursive(leaf.tree);
3102 if (!b || root != &b->branch_tree)
3103 release_tree_entry(root);
3106 static void checkpoint(void)
3108 checkpoint_requested = 0;
3109 if (object_count) {
3110 cycle_packfile();
3111 dump_branches();
3112 dump_tags();
3113 dump_marks();
3117 static void parse_checkpoint(void)
3119 checkpoint_requested = 1;
3120 skip_optional_lf();
3123 static void parse_progress(void)
3125 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3126 fputc('\n', stdout);
3127 fflush(stdout);
3128 skip_optional_lf();
3131 static char* make_fast_import_path(const char *path)
3133 struct strbuf abs_path = STRBUF_INIT;
3135 if (!relative_marks_paths || is_absolute_path(path))
3136 return xstrdup(path);
3137 strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
3138 return strbuf_detach(&abs_path, NULL);
3141 static void option_import_marks(const char *marks,
3142 int from_stream, int ignore_missing)
3144 if (import_marks_file) {
3145 if (from_stream)
3146 die("Only one import-marks command allowed per stream");
3148 /* read previous mark file */
3149 if(!import_marks_file_from_stream)
3150 read_marks();
3153 import_marks_file = make_fast_import_path(marks);
3154 safe_create_leading_directories_const(import_marks_file);
3155 import_marks_file_from_stream = from_stream;
3156 import_marks_file_ignore_missing = ignore_missing;
3159 static void option_date_format(const char *fmt)
3161 if (!strcmp(fmt, "raw"))
3162 whenspec = WHENSPEC_RAW;
3163 else if (!strcmp(fmt, "rfc2822"))
3164 whenspec = WHENSPEC_RFC2822;
3165 else if (!strcmp(fmt, "now"))
3166 whenspec = WHENSPEC_NOW;
3167 else
3168 die("unknown --date-format argument %s", fmt);
3171 static unsigned long ulong_arg(const char *option, const char *arg)
3173 char *endptr;
3174 unsigned long rv = strtoul(arg, &endptr, 0);
3175 if (strchr(arg, '-') || endptr == arg || *endptr)
3176 die("%s: argument must be a non-negative integer", option);
3177 return rv;
3180 static void option_depth(const char *depth)
3182 max_depth = ulong_arg("--depth", depth);
3183 if (max_depth > MAX_DEPTH)
3184 die("--depth cannot exceed %u", MAX_DEPTH);
3187 static void option_active_branches(const char *branches)
3189 max_active_branches = ulong_arg("--active-branches", branches);
3192 static void option_export_marks(const char *marks)
3194 export_marks_file = make_fast_import_path(marks);
3195 safe_create_leading_directories_const(export_marks_file);
3198 static void option_cat_blob_fd(const char *fd)
3200 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3201 if (n > (unsigned long) INT_MAX)
3202 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3203 cat_blob_fd = (int) n;
3206 static void option_export_pack_edges(const char *edges)
3208 if (pack_edges)
3209 fclose(pack_edges);
3210 pack_edges = fopen(edges, "a");
3211 if (!pack_edges)
3212 die_errno("Cannot open '%s'", edges);
3215 static int parse_one_option(const char *option)
3217 if (skip_prefix(option, "max-pack-size=", &option)) {
3218 unsigned long v;
3219 if (!git_parse_ulong(option, &v))
3220 return 0;
3221 if (v < 8192) {
3222 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3223 v *= 1024 * 1024;
3224 } else if (v < 1024 * 1024) {
3225 warning("minimum max-pack-size is 1 MiB");
3226 v = 1024 * 1024;
3228 max_packsize = v;
3229 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3230 unsigned long v;
3231 if (!git_parse_ulong(option, &v))
3232 return 0;
3233 big_file_threshold = v;
3234 } else if (skip_prefix(option, "depth=", &option)) {
3235 option_depth(option);
3236 } else if (skip_prefix(option, "active-branches=", &option)) {
3237 option_active_branches(option);
3238 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3239 option_export_pack_edges(option);
3240 } else if (starts_with(option, "quiet")) {
3241 show_stats = 0;
3242 } else if (starts_with(option, "stats")) {
3243 show_stats = 1;
3244 } else {
3245 return 0;
3248 return 1;
3251 static int parse_one_feature(const char *feature, int from_stream)
3253 const char *arg;
3255 if (skip_prefix(feature, "date-format=", &arg)) {
3256 option_date_format(arg);
3257 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3258 option_import_marks(arg, from_stream, 0);
3259 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3260 option_import_marks(arg, from_stream, 1);
3261 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3262 option_export_marks(arg);
3263 } else if (!strcmp(feature, "cat-blob")) {
3264 ; /* Don't die - this feature is supported */
3265 } else if (!strcmp(feature, "relative-marks")) {
3266 relative_marks_paths = 1;
3267 } else if (!strcmp(feature, "no-relative-marks")) {
3268 relative_marks_paths = 0;
3269 } else if (!strcmp(feature, "done")) {
3270 require_explicit_termination = 1;
3271 } else if (!strcmp(feature, "force")) {
3272 force_update = 1;
3273 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3274 ; /* do nothing; we have the feature */
3275 } else {
3276 return 0;
3279 return 1;
3282 static void parse_feature(const char *feature)
3284 if (seen_data_command)
3285 die("Got feature command '%s' after data command", feature);
3287 if (parse_one_feature(feature, 1))
3288 return;
3290 die("This version of fast-import does not support feature %s.", feature);
3293 static void parse_option(const char *option)
3295 if (seen_data_command)
3296 die("Got option command '%s' after data command", option);
3298 if (parse_one_option(option))
3299 return;
3301 die("This version of fast-import does not support option: %s", option);
3304 static void git_pack_config(void)
3306 int indexversion_value;
3307 unsigned long packsizelimit_value;
3309 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3310 if (max_depth > MAX_DEPTH)
3311 max_depth = MAX_DEPTH;
3313 if (!git_config_get_int("pack.compression", &pack_compression_level)) {
3314 if (pack_compression_level == -1)
3315 pack_compression_level = Z_DEFAULT_COMPRESSION;
3316 else if (pack_compression_level < 0 ||
3317 pack_compression_level > Z_BEST_COMPRESSION)
3318 git_die_config("pack.compression",
3319 "bad pack compression level %d", pack_compression_level);
3320 pack_compression_seen = 1;
3322 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3323 pack_idx_opts.version = indexversion_value;
3324 if (pack_idx_opts.version > 2)
3325 git_die_config("pack.indexversion",
3326 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3328 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3329 max_packsize = packsizelimit_value;
3331 git_config(git_default_config, NULL);
3334 static const char fast_import_usage[] =
3335 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3337 static void parse_argv(void)
3339 unsigned int i;
3341 for (i = 1; i < global_argc; i++) {
3342 const char *a = global_argv[i];
3344 if (*a != '-' || !strcmp(a, "--"))
3345 break;
3347 if (!skip_prefix(a, "--", &a))
3348 die("unknown option %s", a);
3350 if (parse_one_option(a))
3351 continue;
3353 if (parse_one_feature(a, 0))
3354 continue;
3356 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3357 option_cat_blob_fd(a);
3358 continue;
3361 die("unknown option --%s", a);
3363 if (i != global_argc)
3364 usage(fast_import_usage);
3366 seen_data_command = 1;
3367 if (import_marks_file)
3368 read_marks();
3371 int main(int argc, char **argv)
3373 unsigned int i;
3375 git_extract_argv0_path(argv[0]);
3377 git_setup_gettext();
3379 if (argc == 2 && !strcmp(argv[1], "-h"))
3380 usage(fast_import_usage);
3382 setup_git_directory();
3383 reset_pack_idx_option(&pack_idx_opts);
3384 git_pack_config();
3385 if (!pack_compression_seen && core_compression_seen)
3386 pack_compression_level = core_compression_level;
3388 alloc_objects(object_entry_alloc);
3389 strbuf_init(&command_buf, 0);
3390 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3391 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3392 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3393 marks = pool_calloc(1, sizeof(struct mark_set));
3395 global_argc = argc;
3396 global_argv = argv;
3398 rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3399 for (i = 0; i < (cmd_save - 1); i++)
3400 rc_free[i].next = &rc_free[i + 1];
3401 rc_free[cmd_save - 1].next = NULL;
3403 prepare_packed_git();
3404 start_packfile();
3405 set_die_routine(die_nicely);
3406 set_checkpoint_signal();
3407 while (read_next_command() != EOF) {
3408 const char *v;
3409 if (!strcmp("blob", command_buf.buf))
3410 parse_new_blob();
3411 else if (skip_prefix(command_buf.buf, "ls ", &v))
3412 parse_ls(v, NULL);
3413 else if (skip_prefix(command_buf.buf, "commit ", &v))
3414 parse_new_commit(v);
3415 else if (skip_prefix(command_buf.buf, "tag ", &v))
3416 parse_new_tag(v);
3417 else if (skip_prefix(command_buf.buf, "reset ", &v))
3418 parse_reset_branch(v);
3419 else if (!strcmp("checkpoint", command_buf.buf))
3420 parse_checkpoint();
3421 else if (!strcmp("done", command_buf.buf))
3422 break;
3423 else if (starts_with(command_buf.buf, "progress "))
3424 parse_progress();
3425 else if (skip_prefix(command_buf.buf, "feature ", &v))
3426 parse_feature(v);
3427 else if (skip_prefix(command_buf.buf, "option git ", &v))
3428 parse_option(v);
3429 else if (starts_with(command_buf.buf, "option "))
3430 /* ignore non-git options*/;
3431 else
3432 die("Unsupported command: %s", command_buf.buf);
3434 if (checkpoint_requested)
3435 checkpoint();
3438 /* argv hasn't been parsed yet, do so */
3439 if (!seen_data_command)
3440 parse_argv();
3442 if (require_explicit_termination && feof(stdin))
3443 die("stream ends early");
3445 end_packfile();
3447 dump_branches();
3448 dump_tags();
3449 unkeep_all_packs();
3450 dump_marks();
3452 if (pack_edges)
3453 fclose(pack_edges);
3455 if (show_stats) {
3456 uintmax_t total_count = 0, duplicate_count = 0;
3457 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3458 total_count += object_count_by_type[i];
3459 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3460 duplicate_count += duplicate_count_by_type[i];
3462 fprintf(stderr, "%s statistics:\n", argv[0]);
3463 fprintf(stderr, "---------------------------------------------------------------------\n");
3464 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3465 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3466 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]);
3467 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]);
3468 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]);
3469 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]);
3470 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3471 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3472 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3473 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3474 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)(total_allocd/1024));
3475 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3476 fprintf(stderr, "---------------------------------------------------------------------\n");
3477 pack_report();
3478 fprintf(stderr, "---------------------------------------------------------------------\n");
3479 fprintf(stderr, "\n");
3482 return failure ? 1 : 0;