debian: new upstream release candidate
[git/debian.git] / fast-import.c
blobd5e4cf0bad411dad859dbe421ab7583cfb0f1561
1 /*
2 (See Documentation/git-fast-import.txt for maintained documentation.)
3 Format of STDIN stream:
5 stream ::= cmd*;
7 cmd ::= new_blob
8 | new_commit
9 | new_tag
10 | reset_branch
11 | checkpoint
12 | progress
15 new_blob ::= 'blob' lf
16 mark?
17 file_content;
18 file_content ::= data;
20 new_commit ::= 'commit' sp ref_str lf
21 mark?
22 ('author' (sp name)? sp '<' email '>' sp when lf)?
23 'committer' (sp name)? sp '<' email '>' sp when lf
24 commit_msg
25 ('from' sp commit-ish lf)?
26 ('merge' sp commit-ish lf)*
27 (file_change | ls)*
28 lf?;
29 commit_msg ::= data;
31 ls ::= 'ls' sp '"' quoted(path) '"' lf;
33 file_change ::= file_clr
34 | file_del
35 | file_rnm
36 | file_cpy
37 | file_obm
38 | file_inm;
39 file_clr ::= 'deleteall' lf;
40 file_del ::= 'D' sp path_str lf;
41 file_rnm ::= 'R' sp path_str sp path_str lf;
42 file_cpy ::= 'C' sp path_str sp path_str lf;
43 file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
44 file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
45 data;
46 note_obm ::= 'N' sp (hexsha1 | idnum) sp commit-ish lf;
47 note_inm ::= 'N' sp 'inline' sp commit-ish lf
48 data;
50 new_tag ::= 'tag' sp tag_str lf
51 'from' sp commit-ish lf
52 ('tagger' (sp name)? sp '<' email '>' sp when lf)?
53 tag_msg;
54 tag_msg ::= data;
56 reset_branch ::= 'reset' sp ref_str lf
57 ('from' sp commit-ish lf)?
58 lf?;
60 checkpoint ::= 'checkpoint' lf
61 lf?;
63 progress ::= 'progress' sp not_lf* lf
64 lf?;
66 # note: the first idnum in a stream should be 1 and subsequent
67 # idnums should not have gaps between values as this will cause
68 # the stream parser to reserve space for the gapped values. An
69 # idnum can be updated in the future to a new object by issuing
70 # a new mark directive with the old idnum.
72 mark ::= 'mark' sp idnum lf;
73 data ::= (delimited_data | exact_data)
74 lf?;
76 # note: delim may be any string but must not contain lf.
77 # data_line may contain any data but must not be exactly
78 # delim.
79 delimited_data ::= 'data' sp '<<' delim lf
80 (data_line lf)*
81 delim lf;
83 # note: declen indicates the length of binary_data in bytes.
84 # declen does not include the lf preceding the binary data.
86 exact_data ::= 'data' sp declen lf
87 binary_data;
89 # note: quoted strings are C-style quoting supporting \c for
90 # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
91 # is the signed byte value in octal. Note that the only
92 # characters which must actually be escaped to protect the
93 # stream formatting is: \, " and LF. Otherwise these values
94 # are UTF8.
96 commit-ish ::= (ref_str | hexsha1 | sha1exp_str | idnum);
97 ref_str ::= ref;
98 sha1exp_str ::= sha1exp;
99 tag_str ::= tag;
100 path_str ::= path | '"' quoted(path) '"' ;
101 mode ::= '100644' | '644'
102 | '100755' | '755'
103 | '120000'
106 declen ::= # unsigned 32 bit value, ascii base10 notation;
107 bigint ::= # unsigned integer value, ascii base10 notation;
108 binary_data ::= # file content, not interpreted;
110 when ::= raw_when | rfc2822_when;
111 raw_when ::= ts sp tz;
112 rfc2822_when ::= # Valid RFC 2822 date and time;
114 sp ::= # ASCII space character;
115 lf ::= # ASCII newline (LF) character;
117 # note: a colon (':') must precede the numerical value assigned to
118 # an idnum. This is to distinguish it from a ref or tag name as
119 # GIT does not permit ':' in ref or tag strings.
121 idnum ::= ':' bigint;
122 path ::= # GIT style file path, e.g. "a/b/c";
123 ref ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
124 tag ::= # GIT tag name, e.g. "FIREFOX_1_5";
125 sha1exp ::= # Any valid GIT SHA1 expression;
126 hexsha1 ::= # SHA1 in hexadecimal format;
128 # note: name and email are UTF8 strings, however name must not
129 # contain '<' or lf and email must not contain any of the
130 # following: '<', '>', lf.
132 name ::= # valid GIT author/committer name;
133 email ::= # valid GIT author/committer email;
134 ts ::= # time since the epoch in seconds, ascii base10 notation;
135 tz ::= # GIT style timezone;
137 # note: comments, get-mark, ls-tree, and cat-blob requests may
138 # appear anywhere in the input, except within a data command. Any
139 # form of the data command always escapes the related input from
140 # comment processing.
142 # In case it is not clear, the '#' that starts the comment
143 # must be the first character on that line (an lf
144 # preceded it).
147 get_mark ::= 'get-mark' sp idnum lf;
148 cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
149 ls_tree ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
151 comment ::= '#' not_lf* lf;
152 not_lf ::= # Any byte that is not ASCII newline (LF);
155 #include "builtin.h"
156 #include "cache.h"
157 #include "config.h"
158 #include "lockfile.h"
159 #include "object.h"
160 #include "blob.h"
161 #include "tree.h"
162 #include "commit.h"
163 #include "delta.h"
164 #include "pack.h"
165 #include "refs.h"
166 #include "csum-file.h"
167 #include "quote.h"
168 #include "dir.h"
169 #include "run-command.h"
170 #include "packfile.h"
172 #define PACK_ID_BITS 16
173 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
174 #define DEPTH_BITS 13
175 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
178 * We abuse the setuid bit on directories to mean "do not delta".
180 #define NO_DELTA S_ISUID
182 struct object_entry {
183 struct pack_idx_entry idx;
184 struct object_entry *next;
185 uint32_t type : TYPE_BITS,
186 pack_id : PACK_ID_BITS,
187 depth : DEPTH_BITS;
190 struct object_entry_pool {
191 struct object_entry_pool *next_pool;
192 struct object_entry *next_free;
193 struct object_entry *end;
194 struct object_entry entries[FLEX_ARRAY]; /* more */
197 struct mark_set {
198 union {
199 struct object_entry *marked[1024];
200 struct mark_set *sets[1024];
201 } data;
202 unsigned int shift;
205 struct last_object {
206 struct strbuf data;
207 off_t offset;
208 unsigned int depth;
209 unsigned no_swap : 1;
212 struct mem_pool {
213 struct mem_pool *next_pool;
214 char *next_free;
215 char *end;
216 uintmax_t space[FLEX_ARRAY]; /* more */
219 struct atom_str {
220 struct atom_str *next_atom;
221 unsigned short str_len;
222 char str_dat[FLEX_ARRAY]; /* more */
225 struct tree_content;
226 struct tree_entry {
227 struct tree_content *tree;
228 struct atom_str *name;
229 struct tree_entry_ms {
230 uint16_t mode;
231 struct object_id oid;
232 } versions[2];
235 struct tree_content {
236 unsigned int entry_capacity; /* must match avail_tree_content */
237 unsigned int entry_count;
238 unsigned int delta_depth;
239 struct tree_entry *entries[FLEX_ARRAY]; /* more */
242 struct avail_tree_content {
243 unsigned int entry_capacity; /* must match tree_content */
244 struct avail_tree_content *next_avail;
247 struct branch {
248 struct branch *table_next_branch;
249 struct branch *active_next_branch;
250 const char *name;
251 struct tree_entry branch_tree;
252 uintmax_t last_commit;
253 uintmax_t num_notes;
254 unsigned active : 1;
255 unsigned delete : 1;
256 unsigned pack_id : PACK_ID_BITS;
257 struct object_id oid;
260 struct tag {
261 struct tag *next_tag;
262 const char *name;
263 unsigned int pack_id;
264 struct object_id oid;
267 struct hash_list {
268 struct hash_list *next;
269 struct object_id oid;
272 typedef enum {
273 WHENSPEC_RAW = 1,
274 WHENSPEC_RFC2822,
275 WHENSPEC_NOW
276 } whenspec_type;
278 struct recent_command {
279 struct recent_command *prev;
280 struct recent_command *next;
281 char *buf;
284 /* Configured limits on output */
285 static unsigned long max_depth = 50;
286 static off_t max_packsize;
287 static int unpack_limit = 100;
288 static int force_update;
290 /* Stats and misc. counters */
291 static uintmax_t alloc_count;
292 static uintmax_t marks_set_count;
293 static uintmax_t object_count_by_type[1 << TYPE_BITS];
294 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
295 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
296 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
297 static unsigned long object_count;
298 static unsigned long branch_count;
299 static unsigned long branch_load_count;
300 static int failure;
301 static FILE *pack_edges;
302 static unsigned int show_stats = 1;
303 static int global_argc;
304 static const char **global_argv;
306 /* Memory pools */
307 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
308 static size_t total_allocd;
309 static struct mem_pool *mem_pool;
311 /* Atom management */
312 static unsigned int atom_table_sz = 4451;
313 static unsigned int atom_cnt;
314 static struct atom_str **atom_table;
316 /* The .pack file being generated */
317 static struct pack_idx_option pack_idx_opts;
318 static unsigned int pack_id;
319 static struct sha1file *pack_file;
320 static struct packed_git *pack_data;
321 static struct packed_git **all_packs;
322 static off_t pack_size;
324 /* Table of objects we've written. */
325 static unsigned int object_entry_alloc = 5000;
326 static struct object_entry_pool *blocks;
327 static struct object_entry *object_table[1 << 16];
328 static struct mark_set *marks;
329 static const char *export_marks_file;
330 static const char *import_marks_file;
331 static int import_marks_file_from_stream;
332 static int import_marks_file_ignore_missing;
333 static int import_marks_file_done;
334 static int relative_marks_paths;
336 /* Our last blob */
337 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
339 /* Tree management */
340 static unsigned int tree_entry_alloc = 1000;
341 static void *avail_tree_entry;
342 static unsigned int avail_tree_table_sz = 100;
343 static struct avail_tree_content **avail_tree_table;
344 static struct strbuf old_tree = STRBUF_INIT;
345 static struct strbuf new_tree = STRBUF_INIT;
347 /* Branch data */
348 static unsigned long max_active_branches = 5;
349 static unsigned long cur_active_branches;
350 static unsigned long branch_table_sz = 1039;
351 static struct branch **branch_table;
352 static struct branch *active_branches;
354 /* Tag data */
355 static struct tag *first_tag;
356 static struct tag *last_tag;
358 /* Input stream parsing */
359 static whenspec_type whenspec = WHENSPEC_RAW;
360 static struct strbuf command_buf = STRBUF_INIT;
361 static int unread_command_buf;
362 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
363 static struct recent_command *cmd_tail = &cmd_hist;
364 static struct recent_command *rc_free;
365 static unsigned int cmd_save = 100;
366 static uintmax_t next_mark;
367 static struct strbuf new_data = STRBUF_INIT;
368 static int seen_data_command;
369 static int require_explicit_termination;
371 /* Signal handling */
372 static volatile sig_atomic_t checkpoint_requested;
374 /* Where to write output of cat-blob commands */
375 static int cat_blob_fd = STDOUT_FILENO;
377 static void parse_argv(void);
378 static void parse_get_mark(const char *p);
379 static void parse_cat_blob(const char *p);
380 static void parse_ls(const char *p, struct branch *b);
382 static void write_branch_report(FILE *rpt, struct branch *b)
384 fprintf(rpt, "%s:\n", b->name);
386 fprintf(rpt, " status :");
387 if (b->active)
388 fputs(" active", rpt);
389 if (b->branch_tree.tree)
390 fputs(" loaded", rpt);
391 if (is_null_oid(&b->branch_tree.versions[1].oid))
392 fputs(" dirty", rpt);
393 fputc('\n', rpt);
395 fprintf(rpt, " tip commit : %s\n", oid_to_hex(&b->oid));
396 fprintf(rpt, " old tree : %s\n",
397 oid_to_hex(&b->branch_tree.versions[0].oid));
398 fprintf(rpt, " cur tree : %s\n",
399 oid_to_hex(&b->branch_tree.versions[1].oid));
400 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
402 fputs(" last pack : ", rpt);
403 if (b->pack_id < MAX_PACK_ID)
404 fprintf(rpt, "%u", b->pack_id);
405 fputc('\n', rpt);
407 fputc('\n', rpt);
410 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
412 static void write_crash_report(const char *err)
414 char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
415 FILE *rpt = fopen(loc, "w");
416 struct branch *b;
417 unsigned long lu;
418 struct recent_command *rc;
420 if (!rpt) {
421 error_errno("can't write crash report %s", loc);
422 free(loc);
423 return;
426 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
428 fprintf(rpt, "fast-import crash report:\n");
429 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
430 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
431 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
432 fputc('\n', rpt);
434 fputs("fatal: ", rpt);
435 fputs(err, rpt);
436 fputc('\n', rpt);
438 fputc('\n', rpt);
439 fputs("Most Recent Commands Before Crash\n", rpt);
440 fputs("---------------------------------\n", rpt);
441 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
442 if (rc->next == &cmd_hist)
443 fputs("* ", rpt);
444 else
445 fputs(" ", rpt);
446 fputs(rc->buf, rpt);
447 fputc('\n', rpt);
450 fputc('\n', rpt);
451 fputs("Active Branch LRU\n", rpt);
452 fputs("-----------------\n", rpt);
453 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
454 cur_active_branches,
455 max_active_branches);
456 fputc('\n', rpt);
457 fputs(" pos clock name\n", rpt);
458 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
459 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
460 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
461 ++lu, b->last_commit, b->name);
463 fputc('\n', rpt);
464 fputs("Inactive Branches\n", rpt);
465 fputs("-----------------\n", rpt);
466 for (lu = 0; lu < branch_table_sz; lu++) {
467 for (b = branch_table[lu]; b; b = b->table_next_branch)
468 write_branch_report(rpt, b);
471 if (first_tag) {
472 struct tag *tg;
473 fputc('\n', rpt);
474 fputs("Annotated Tags\n", rpt);
475 fputs("--------------\n", rpt);
476 for (tg = first_tag; tg; tg = tg->next_tag) {
477 fputs(oid_to_hex(&tg->oid), rpt);
478 fputc(' ', rpt);
479 fputs(tg->name, rpt);
480 fputc('\n', rpt);
484 fputc('\n', rpt);
485 fputs("Marks\n", rpt);
486 fputs("-----\n", rpt);
487 if (export_marks_file)
488 fprintf(rpt, " exported to %s\n", export_marks_file);
489 else
490 dump_marks_helper(rpt, 0, marks);
492 fputc('\n', rpt);
493 fputs("-------------------\n", rpt);
494 fputs("END OF CRASH REPORT\n", rpt);
495 fclose(rpt);
496 free(loc);
499 static void end_packfile(void);
500 static void unkeep_all_packs(void);
501 static void dump_marks(void);
503 static NORETURN void die_nicely(const char *err, va_list params)
505 static int zombie;
506 char message[2 * PATH_MAX];
508 vsnprintf(message, sizeof(message), err, params);
509 fputs("fatal: ", stderr);
510 fputs(message, stderr);
511 fputc('\n', stderr);
513 if (!zombie) {
514 zombie = 1;
515 write_crash_report(message);
516 end_packfile();
517 unkeep_all_packs();
518 dump_marks();
520 exit(128);
523 #ifndef SIGUSR1 /* Windows, for example */
525 static void set_checkpoint_signal(void)
529 #else
531 static void checkpoint_signal(int signo)
533 checkpoint_requested = 1;
536 static void set_checkpoint_signal(void)
538 struct sigaction sa;
540 memset(&sa, 0, sizeof(sa));
541 sa.sa_handler = checkpoint_signal;
542 sigemptyset(&sa.sa_mask);
543 sa.sa_flags = SA_RESTART;
544 sigaction(SIGUSR1, &sa, NULL);
547 #endif
549 static void alloc_objects(unsigned int cnt)
551 struct object_entry_pool *b;
553 b = xmalloc(sizeof(struct object_entry_pool)
554 + cnt * sizeof(struct object_entry));
555 b->next_pool = blocks;
556 b->next_free = b->entries;
557 b->end = b->entries + cnt;
558 blocks = b;
559 alloc_count += cnt;
562 static struct object_entry *new_object(struct object_id *oid)
564 struct object_entry *e;
566 if (blocks->next_free == blocks->end)
567 alloc_objects(object_entry_alloc);
569 e = blocks->next_free++;
570 oidcpy(&e->idx.oid, oid);
571 return e;
574 static struct object_entry *find_object(struct object_id *oid)
576 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
577 struct object_entry *e;
578 for (e = object_table[h]; e; e = e->next)
579 if (!oidcmp(oid, &e->idx.oid))
580 return e;
581 return NULL;
584 static struct object_entry *insert_object(struct object_id *oid)
586 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
587 struct object_entry *e = object_table[h];
589 while (e) {
590 if (!oidcmp(oid, &e->idx.oid))
591 return e;
592 e = e->next;
595 e = new_object(oid);
596 e->next = object_table[h];
597 e->idx.offset = 0;
598 object_table[h] = e;
599 return e;
602 static void invalidate_pack_id(unsigned int id)
604 unsigned int h;
605 unsigned long lu;
606 struct tag *t;
608 for (h = 0; h < ARRAY_SIZE(object_table); h++) {
609 struct object_entry *e;
611 for (e = object_table[h]; e; e = e->next)
612 if (e->pack_id == id)
613 e->pack_id = MAX_PACK_ID;
616 for (lu = 0; lu < branch_table_sz; lu++) {
617 struct branch *b;
619 for (b = branch_table[lu]; b; b = b->table_next_branch)
620 if (b->pack_id == id)
621 b->pack_id = MAX_PACK_ID;
624 for (t = first_tag; t; t = t->next_tag)
625 if (t->pack_id == id)
626 t->pack_id = MAX_PACK_ID;
629 static unsigned int hc_str(const char *s, size_t len)
631 unsigned int r = 0;
632 while (len-- > 0)
633 r = r * 31 + *s++;
634 return r;
637 static void *pool_alloc(size_t len)
639 struct mem_pool *p;
640 void *r;
642 /* round up to a 'uintmax_t' alignment */
643 if (len & (sizeof(uintmax_t) - 1))
644 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
646 for (p = mem_pool; p; p = p->next_pool)
647 if ((p->end - p->next_free >= len))
648 break;
650 if (!p) {
651 if (len >= (mem_pool_alloc/2)) {
652 total_allocd += len;
653 return xmalloc(len);
655 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
656 p = xmalloc(st_add(sizeof(struct mem_pool), mem_pool_alloc));
657 p->next_pool = mem_pool;
658 p->next_free = (char *) p->space;
659 p->end = p->next_free + mem_pool_alloc;
660 mem_pool = p;
663 r = p->next_free;
664 p->next_free += len;
665 return r;
668 static void *pool_calloc(size_t count, size_t size)
670 size_t len = count * size;
671 void *r = pool_alloc(len);
672 memset(r, 0, len);
673 return r;
676 static char *pool_strdup(const char *s)
678 size_t len = strlen(s) + 1;
679 char *r = pool_alloc(len);
680 memcpy(r, s, len);
681 return r;
684 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
686 struct mark_set *s = marks;
687 while ((idnum >> s->shift) >= 1024) {
688 s = pool_calloc(1, sizeof(struct mark_set));
689 s->shift = marks->shift + 10;
690 s->data.sets[0] = marks;
691 marks = s;
693 while (s->shift) {
694 uintmax_t i = idnum >> s->shift;
695 idnum -= i << s->shift;
696 if (!s->data.sets[i]) {
697 s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
698 s->data.sets[i]->shift = s->shift - 10;
700 s = s->data.sets[i];
702 if (!s->data.marked[idnum])
703 marks_set_count++;
704 s->data.marked[idnum] = oe;
707 static struct object_entry *find_mark(uintmax_t idnum)
709 uintmax_t orig_idnum = idnum;
710 struct mark_set *s = marks;
711 struct object_entry *oe = NULL;
712 if ((idnum >> s->shift) < 1024) {
713 while (s && s->shift) {
714 uintmax_t i = idnum >> s->shift;
715 idnum -= i << s->shift;
716 s = s->data.sets[i];
718 if (s)
719 oe = s->data.marked[idnum];
721 if (!oe)
722 die("mark :%" PRIuMAX " not declared", orig_idnum);
723 return oe;
726 static struct atom_str *to_atom(const char *s, unsigned short len)
728 unsigned int hc = hc_str(s, len) % atom_table_sz;
729 struct atom_str *c;
731 for (c = atom_table[hc]; c; c = c->next_atom)
732 if (c->str_len == len && !strncmp(s, c->str_dat, len))
733 return c;
735 c = pool_alloc(sizeof(struct atom_str) + len + 1);
736 c->str_len = len;
737 memcpy(c->str_dat, s, len);
738 c->str_dat[len] = 0;
739 c->next_atom = atom_table[hc];
740 atom_table[hc] = c;
741 atom_cnt++;
742 return c;
745 static struct branch *lookup_branch(const char *name)
747 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
748 struct branch *b;
750 for (b = branch_table[hc]; b; b = b->table_next_branch)
751 if (!strcmp(name, b->name))
752 return b;
753 return NULL;
756 static struct branch *new_branch(const char *name)
758 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
759 struct branch *b = lookup_branch(name);
761 if (b)
762 die("Invalid attempt to create duplicate branch: %s", name);
763 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
764 die("Branch name doesn't conform to GIT standards: %s", name);
766 b = pool_calloc(1, sizeof(struct branch));
767 b->name = pool_strdup(name);
768 b->table_next_branch = branch_table[hc];
769 b->branch_tree.versions[0].mode = S_IFDIR;
770 b->branch_tree.versions[1].mode = S_IFDIR;
771 b->num_notes = 0;
772 b->active = 0;
773 b->pack_id = MAX_PACK_ID;
774 branch_table[hc] = b;
775 branch_count++;
776 return b;
779 static unsigned int hc_entries(unsigned int cnt)
781 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
782 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
785 static struct tree_content *new_tree_content(unsigned int cnt)
787 struct avail_tree_content *f, *l = NULL;
788 struct tree_content *t;
789 unsigned int hc = hc_entries(cnt);
791 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
792 if (f->entry_capacity >= cnt)
793 break;
795 if (f) {
796 if (l)
797 l->next_avail = f->next_avail;
798 else
799 avail_tree_table[hc] = f->next_avail;
800 } else {
801 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
802 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
803 f->entry_capacity = cnt;
806 t = (struct tree_content*)f;
807 t->entry_count = 0;
808 t->delta_depth = 0;
809 return t;
812 static void release_tree_entry(struct tree_entry *e);
813 static void release_tree_content(struct tree_content *t)
815 struct avail_tree_content *f = (struct avail_tree_content*)t;
816 unsigned int hc = hc_entries(f->entry_capacity);
817 f->next_avail = avail_tree_table[hc];
818 avail_tree_table[hc] = f;
821 static void release_tree_content_recursive(struct tree_content *t)
823 unsigned int i;
824 for (i = 0; i < t->entry_count; i++)
825 release_tree_entry(t->entries[i]);
826 release_tree_content(t);
829 static struct tree_content *grow_tree_content(
830 struct tree_content *t,
831 int amt)
833 struct tree_content *r = new_tree_content(t->entry_count + amt);
834 r->entry_count = t->entry_count;
835 r->delta_depth = t->delta_depth;
836 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
837 release_tree_content(t);
838 return r;
841 static struct tree_entry *new_tree_entry(void)
843 struct tree_entry *e;
845 if (!avail_tree_entry) {
846 unsigned int n = tree_entry_alloc;
847 total_allocd += n * sizeof(struct tree_entry);
848 ALLOC_ARRAY(e, n);
849 avail_tree_entry = e;
850 while (n-- > 1) {
851 *((void**)e) = e + 1;
852 e++;
854 *((void**)e) = NULL;
857 e = avail_tree_entry;
858 avail_tree_entry = *((void**)e);
859 return e;
862 static void release_tree_entry(struct tree_entry *e)
864 if (e->tree)
865 release_tree_content_recursive(e->tree);
866 *((void**)e) = avail_tree_entry;
867 avail_tree_entry = e;
870 static struct tree_content *dup_tree_content(struct tree_content *s)
872 struct tree_content *d;
873 struct tree_entry *a, *b;
874 unsigned int i;
876 if (!s)
877 return NULL;
878 d = new_tree_content(s->entry_count);
879 for (i = 0; i < s->entry_count; i++) {
880 a = s->entries[i];
881 b = new_tree_entry();
882 memcpy(b, a, sizeof(*a));
883 if (a->tree && is_null_oid(&b->versions[1].oid))
884 b->tree = dup_tree_content(a->tree);
885 else
886 b->tree = NULL;
887 d->entries[i] = b;
889 d->entry_count = s->entry_count;
890 d->delta_depth = s->delta_depth;
892 return d;
895 static void start_packfile(void)
897 struct strbuf tmp_file = STRBUF_INIT;
898 struct packed_git *p;
899 struct pack_header hdr;
900 int pack_fd;
902 pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
903 FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
904 strbuf_release(&tmp_file);
906 p->pack_fd = pack_fd;
907 p->do_not_close = 1;
908 pack_file = sha1fd(pack_fd, p->pack_name);
910 hdr.hdr_signature = htonl(PACK_SIGNATURE);
911 hdr.hdr_version = htonl(2);
912 hdr.hdr_entries = 0;
913 sha1write(pack_file, &hdr, sizeof(hdr));
915 pack_data = p;
916 pack_size = sizeof(hdr);
917 object_count = 0;
919 REALLOC_ARRAY(all_packs, pack_id + 1);
920 all_packs[pack_id] = p;
923 static const char *create_index(void)
925 const char *tmpfile;
926 struct pack_idx_entry **idx, **c, **last;
927 struct object_entry *e;
928 struct object_entry_pool *o;
930 /* Build the table of object IDs. */
931 ALLOC_ARRAY(idx, object_count);
932 c = idx;
933 for (o = blocks; o; o = o->next_pool)
934 for (e = o->next_free; e-- != o->entries;)
935 if (pack_id == e->pack_id)
936 *c++ = &e->idx;
937 last = idx + object_count;
938 if (c != last)
939 die("internal consistency error creating the index");
941 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
942 free(idx);
943 return tmpfile;
946 static char *keep_pack(const char *curr_index_name)
948 static const char *keep_msg = "fast-import";
949 struct strbuf name = STRBUF_INIT;
950 int keep_fd;
952 odb_pack_name(&name, pack_data->sha1, "keep");
953 keep_fd = odb_pack_keep(name.buf);
954 if (keep_fd < 0)
955 die_errno("cannot create keep file");
956 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
957 if (close(keep_fd))
958 die_errno("failed to write keep file");
960 odb_pack_name(&name, pack_data->sha1, "pack");
961 if (finalize_object_file(pack_data->pack_name, name.buf))
962 die("cannot store pack file");
964 odb_pack_name(&name, pack_data->sha1, "idx");
965 if (finalize_object_file(curr_index_name, name.buf))
966 die("cannot store index file");
967 free((void *)curr_index_name);
968 return strbuf_detach(&name, NULL);
971 static void unkeep_all_packs(void)
973 struct strbuf name = STRBUF_INIT;
974 int k;
976 for (k = 0; k < pack_id; k++) {
977 struct packed_git *p = all_packs[k];
978 odb_pack_name(&name, p->sha1, "keep");
979 unlink_or_warn(name.buf);
981 strbuf_release(&name);
984 static int loosen_small_pack(const struct packed_git *p)
986 struct child_process unpack = CHILD_PROCESS_INIT;
988 if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
989 die_errno("Failed seeking to start of '%s'", p->pack_name);
991 unpack.in = p->pack_fd;
992 unpack.git_cmd = 1;
993 unpack.stdout_to_stderr = 1;
994 argv_array_push(&unpack.args, "unpack-objects");
995 if (!show_stats)
996 argv_array_push(&unpack.args, "-q");
998 return run_command(&unpack);
1001 static void end_packfile(void)
1003 static int running;
1005 if (running || !pack_data)
1006 return;
1008 running = 1;
1009 clear_delta_base_cache();
1010 if (object_count) {
1011 struct packed_git *new_p;
1012 struct object_id cur_pack_oid;
1013 char *idx_name;
1014 int i;
1015 struct branch *b;
1016 struct tag *t;
1018 close_pack_windows(pack_data);
1019 sha1close(pack_file, cur_pack_oid.hash, 0);
1020 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
1021 pack_data->pack_name, object_count,
1022 cur_pack_oid.hash, pack_size);
1024 if (object_count <= unpack_limit) {
1025 if (!loosen_small_pack(pack_data)) {
1026 invalidate_pack_id(pack_id);
1027 goto discard_pack;
1031 close(pack_data->pack_fd);
1032 idx_name = keep_pack(create_index());
1034 /* Register the packfile with core git's machinery. */
1035 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
1036 if (!new_p)
1037 die("core git rejected index %s", idx_name);
1038 all_packs[pack_id] = new_p;
1039 install_packed_git(new_p);
1040 free(idx_name);
1042 /* Print the boundary */
1043 if (pack_edges) {
1044 fprintf(pack_edges, "%s:", new_p->pack_name);
1045 for (i = 0; i < branch_table_sz; i++) {
1046 for (b = branch_table[i]; b; b = b->table_next_branch) {
1047 if (b->pack_id == pack_id)
1048 fprintf(pack_edges, " %s",
1049 oid_to_hex(&b->oid));
1052 for (t = first_tag; t; t = t->next_tag) {
1053 if (t->pack_id == pack_id)
1054 fprintf(pack_edges, " %s",
1055 oid_to_hex(&t->oid));
1057 fputc('\n', pack_edges);
1058 fflush(pack_edges);
1061 pack_id++;
1063 else {
1064 discard_pack:
1065 close(pack_data->pack_fd);
1066 unlink_or_warn(pack_data->pack_name);
1068 FREE_AND_NULL(pack_data);
1069 running = 0;
1071 /* We can't carry a delta across packfiles. */
1072 strbuf_release(&last_blob.data);
1073 last_blob.offset = 0;
1074 last_blob.depth = 0;
1077 static void cycle_packfile(void)
1079 end_packfile();
1080 start_packfile();
1083 static int store_object(
1084 enum object_type type,
1085 struct strbuf *dat,
1086 struct last_object *last,
1087 struct object_id *oidout,
1088 uintmax_t mark)
1090 void *out, *delta;
1091 struct object_entry *e;
1092 unsigned char hdr[96];
1093 struct object_id oid;
1094 unsigned long hdrlen, deltalen;
1095 git_SHA_CTX c;
1096 git_zstream s;
1098 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1099 typename(type), (unsigned long)dat->len) + 1;
1100 git_SHA1_Init(&c);
1101 git_SHA1_Update(&c, hdr, hdrlen);
1102 git_SHA1_Update(&c, dat->buf, dat->len);
1103 git_SHA1_Final(oid.hash, &c);
1104 if (oidout)
1105 oidcpy(oidout, &oid);
1107 e = insert_object(&oid);
1108 if (mark)
1109 insert_mark(mark, e);
1110 if (e->idx.offset) {
1111 duplicate_count_by_type[type]++;
1112 return 1;
1113 } else if (find_sha1_pack(oid.hash, packed_git)) {
1114 e->type = type;
1115 e->pack_id = MAX_PACK_ID;
1116 e->idx.offset = 1; /* just not zero! */
1117 duplicate_count_by_type[type]++;
1118 return 1;
1121 if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1122 delta_count_attempts_by_type[type]++;
1123 delta = diff_delta(last->data.buf, last->data.len,
1124 dat->buf, dat->len,
1125 &deltalen, dat->len - 20);
1126 } else
1127 delta = NULL;
1129 git_deflate_init(&s, pack_compression_level);
1130 if (delta) {
1131 s.next_in = delta;
1132 s.avail_in = deltalen;
1133 } else {
1134 s.next_in = (void *)dat->buf;
1135 s.avail_in = dat->len;
1137 s.avail_out = git_deflate_bound(&s, s.avail_in);
1138 s.next_out = out = xmalloc(s.avail_out);
1139 while (git_deflate(&s, Z_FINISH) == Z_OK)
1140 ; /* nothing */
1141 git_deflate_end(&s);
1143 /* Determine if we should auto-checkpoint. */
1144 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1145 || (pack_size + 60 + s.total_out) < pack_size) {
1147 /* This new object needs to *not* have the current pack_id. */
1148 e->pack_id = pack_id + 1;
1149 cycle_packfile();
1151 /* We cannot carry a delta into the new pack. */
1152 if (delta) {
1153 FREE_AND_NULL(delta);
1155 git_deflate_init(&s, pack_compression_level);
1156 s.next_in = (void *)dat->buf;
1157 s.avail_in = dat->len;
1158 s.avail_out = git_deflate_bound(&s, s.avail_in);
1159 s.next_out = out = xrealloc(out, s.avail_out);
1160 while (git_deflate(&s, Z_FINISH) == Z_OK)
1161 ; /* nothing */
1162 git_deflate_end(&s);
1166 e->type = type;
1167 e->pack_id = pack_id;
1168 e->idx.offset = pack_size;
1169 object_count++;
1170 object_count_by_type[type]++;
1172 crc32_begin(pack_file);
1174 if (delta) {
1175 off_t ofs = e->idx.offset - last->offset;
1176 unsigned pos = sizeof(hdr) - 1;
1178 delta_count_by_type[type]++;
1179 e->depth = last->depth + 1;
1181 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1182 OBJ_OFS_DELTA, deltalen);
1183 sha1write(pack_file, hdr, hdrlen);
1184 pack_size += hdrlen;
1186 hdr[pos] = ofs & 127;
1187 while (ofs >>= 7)
1188 hdr[--pos] = 128 | (--ofs & 127);
1189 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1190 pack_size += sizeof(hdr) - pos;
1191 } else {
1192 e->depth = 0;
1193 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1194 type, dat->len);
1195 sha1write(pack_file, hdr, hdrlen);
1196 pack_size += hdrlen;
1199 sha1write(pack_file, out, s.total_out);
1200 pack_size += s.total_out;
1202 e->idx.crc32 = crc32_end(pack_file);
1204 free(out);
1205 free(delta);
1206 if (last) {
1207 if (last->no_swap) {
1208 last->data = *dat;
1209 } else {
1210 strbuf_swap(&last->data, dat);
1212 last->offset = e->idx.offset;
1213 last->depth = e->depth;
1215 return 0;
1218 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1220 if (sha1file_truncate(pack_file, checkpoint))
1221 die_errno("cannot truncate pack to skip duplicate");
1222 pack_size = checkpoint->offset;
1225 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1227 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1228 unsigned char *in_buf = xmalloc(in_sz);
1229 unsigned char *out_buf = xmalloc(out_sz);
1230 struct object_entry *e;
1231 struct object_id oid;
1232 unsigned long hdrlen;
1233 off_t offset;
1234 git_SHA_CTX c;
1235 git_zstream s;
1236 struct sha1file_checkpoint checkpoint;
1237 int status = Z_OK;
1239 /* Determine if we should auto-checkpoint. */
1240 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1241 || (pack_size + 60 + len) < pack_size)
1242 cycle_packfile();
1244 sha1file_checkpoint(pack_file, &checkpoint);
1245 offset = checkpoint.offset;
1247 hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1249 git_SHA1_Init(&c);
1250 git_SHA1_Update(&c, out_buf, hdrlen);
1252 crc32_begin(pack_file);
1254 git_deflate_init(&s, pack_compression_level);
1256 hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1258 s.next_out = out_buf + hdrlen;
1259 s.avail_out = out_sz - hdrlen;
1261 while (status != Z_STREAM_END) {
1262 if (0 < len && !s.avail_in) {
1263 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1264 size_t n = fread(in_buf, 1, cnt, stdin);
1265 if (!n && feof(stdin))
1266 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1268 git_SHA1_Update(&c, in_buf, n);
1269 s.next_in = in_buf;
1270 s.avail_in = n;
1271 len -= n;
1274 status = git_deflate(&s, len ? 0 : Z_FINISH);
1276 if (!s.avail_out || status == Z_STREAM_END) {
1277 size_t n = s.next_out - out_buf;
1278 sha1write(pack_file, out_buf, n);
1279 pack_size += n;
1280 s.next_out = out_buf;
1281 s.avail_out = out_sz;
1284 switch (status) {
1285 case Z_OK:
1286 case Z_BUF_ERROR:
1287 case Z_STREAM_END:
1288 continue;
1289 default:
1290 die("unexpected deflate failure: %d", status);
1293 git_deflate_end(&s);
1294 git_SHA1_Final(oid.hash, &c);
1296 if (oidout)
1297 oidcpy(oidout, &oid);
1299 e = insert_object(&oid);
1301 if (mark)
1302 insert_mark(mark, e);
1304 if (e->idx.offset) {
1305 duplicate_count_by_type[OBJ_BLOB]++;
1306 truncate_pack(&checkpoint);
1308 } else if (find_sha1_pack(oid.hash, packed_git)) {
1309 e->type = OBJ_BLOB;
1310 e->pack_id = MAX_PACK_ID;
1311 e->idx.offset = 1; /* just not zero! */
1312 duplicate_count_by_type[OBJ_BLOB]++;
1313 truncate_pack(&checkpoint);
1315 } else {
1316 e->depth = 0;
1317 e->type = OBJ_BLOB;
1318 e->pack_id = pack_id;
1319 e->idx.offset = offset;
1320 e->idx.crc32 = crc32_end(pack_file);
1321 object_count++;
1322 object_count_by_type[OBJ_BLOB]++;
1325 free(in_buf);
1326 free(out_buf);
1329 /* All calls must be guarded by find_object() or find_mark() to
1330 * ensure the 'struct object_entry' passed was written by this
1331 * process instance. We unpack the entry by the offset, avoiding
1332 * the need for the corresponding .idx file. This unpacking rule
1333 * works because we only use OBJ_REF_DELTA within the packfiles
1334 * created by fast-import.
1336 * oe must not be NULL. Such an oe usually comes from giving
1337 * an unknown SHA-1 to find_object() or an undefined mark to
1338 * find_mark(). Callers must test for this condition and use
1339 * the standard read_sha1_file() when it happens.
1341 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1342 * find_mark(), where the mark was reloaded from an existing marks
1343 * file and is referencing an object that this fast-import process
1344 * instance did not write out to a packfile. Callers must test for
1345 * this condition and use read_sha1_file() instead.
1347 static void *gfi_unpack_entry(
1348 struct object_entry *oe,
1349 unsigned long *sizep)
1351 enum object_type type;
1352 struct packed_git *p = all_packs[oe->pack_id];
1353 if (p == pack_data && p->pack_size < (pack_size + 20)) {
1354 /* The object is stored in the packfile we are writing to
1355 * and we have modified it since the last time we scanned
1356 * back to read a previously written object. If an old
1357 * window covered [p->pack_size, p->pack_size + 20) its
1358 * data is stale and is not valid. Closing all windows
1359 * and updating the packfile length ensures we can read
1360 * the newly written data.
1362 close_pack_windows(p);
1363 sha1flush(pack_file);
1365 /* We have to offer 20 bytes additional on the end of
1366 * the packfile as the core unpacker code assumes the
1367 * footer is present at the file end and must promise
1368 * at least 20 bytes within any window it maps. But
1369 * we don't actually create the footer here.
1371 p->pack_size = pack_size + 20;
1373 return unpack_entry(p, oe->idx.offset, &type, sizep);
1376 static const char *get_mode(const char *str, uint16_t *modep)
1378 unsigned char c;
1379 uint16_t mode = 0;
1381 while ((c = *str++) != ' ') {
1382 if (c < '0' || c > '7')
1383 return NULL;
1384 mode = (mode << 3) + (c - '0');
1386 *modep = mode;
1387 return str;
1390 static void load_tree(struct tree_entry *root)
1392 struct object_id *oid = &root->versions[1].oid;
1393 struct object_entry *myoe;
1394 struct tree_content *t;
1395 unsigned long size;
1396 char *buf;
1397 const char *c;
1399 root->tree = t = new_tree_content(8);
1400 if (is_null_oid(oid))
1401 return;
1403 myoe = find_object(oid);
1404 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1405 if (myoe->type != OBJ_TREE)
1406 die("Not a tree: %s", oid_to_hex(oid));
1407 t->delta_depth = myoe->depth;
1408 buf = gfi_unpack_entry(myoe, &size);
1409 if (!buf)
1410 die("Can't load tree %s", oid_to_hex(oid));
1411 } else {
1412 enum object_type type;
1413 buf = read_sha1_file(oid->hash, &type, &size);
1414 if (!buf || type != OBJ_TREE)
1415 die("Can't load tree %s", oid_to_hex(oid));
1418 c = buf;
1419 while (c != (buf + size)) {
1420 struct tree_entry *e = new_tree_entry();
1422 if (t->entry_count == t->entry_capacity)
1423 root->tree = t = grow_tree_content(t, t->entry_count);
1424 t->entries[t->entry_count++] = e;
1426 e->tree = NULL;
1427 c = get_mode(c, &e->versions[1].mode);
1428 if (!c)
1429 die("Corrupt mode in %s", oid_to_hex(oid));
1430 e->versions[0].mode = e->versions[1].mode;
1431 e->name = to_atom(c, strlen(c));
1432 c += e->name->str_len + 1;
1433 hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1434 hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1435 c += GIT_SHA1_RAWSZ;
1437 free(buf);
1440 static int tecmp0 (const void *_a, const void *_b)
1442 struct tree_entry *a = *((struct tree_entry**)_a);
1443 struct tree_entry *b = *((struct tree_entry**)_b);
1444 return base_name_compare(
1445 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1446 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1449 static int tecmp1 (const void *_a, const void *_b)
1451 struct tree_entry *a = *((struct tree_entry**)_a);
1452 struct tree_entry *b = *((struct tree_entry**)_b);
1453 return base_name_compare(
1454 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1455 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1458 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1460 size_t maxlen = 0;
1461 unsigned int i;
1463 if (!v)
1464 QSORT(t->entries, t->entry_count, tecmp0);
1465 else
1466 QSORT(t->entries, t->entry_count, tecmp1);
1468 for (i = 0; i < t->entry_count; i++) {
1469 if (t->entries[i]->versions[v].mode)
1470 maxlen += t->entries[i]->name->str_len + 34;
1473 strbuf_reset(b);
1474 strbuf_grow(b, maxlen);
1475 for (i = 0; i < t->entry_count; i++) {
1476 struct tree_entry *e = t->entries[i];
1477 if (!e->versions[v].mode)
1478 continue;
1479 strbuf_addf(b, "%o %s%c",
1480 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1481 e->name->str_dat, '\0');
1482 strbuf_add(b, e->versions[v].oid.hash, GIT_SHA1_RAWSZ);
1486 static void store_tree(struct tree_entry *root)
1488 struct tree_content *t;
1489 unsigned int i, j, del;
1490 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1491 struct object_entry *le = NULL;
1493 if (!is_null_oid(&root->versions[1].oid))
1494 return;
1496 if (!root->tree)
1497 load_tree(root);
1498 t = root->tree;
1500 for (i = 0; i < t->entry_count; i++) {
1501 if (t->entries[i]->tree)
1502 store_tree(t->entries[i]);
1505 if (!(root->versions[0].mode & NO_DELTA))
1506 le = find_object(&root->versions[0].oid);
1507 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1508 mktree(t, 0, &old_tree);
1509 lo.data = old_tree;
1510 lo.offset = le->idx.offset;
1511 lo.depth = t->delta_depth;
1514 mktree(t, 1, &new_tree);
1515 store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1517 t->delta_depth = lo.depth;
1518 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1519 struct tree_entry *e = t->entries[i];
1520 if (e->versions[1].mode) {
1521 e->versions[0].mode = e->versions[1].mode;
1522 oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1523 t->entries[j++] = e;
1524 } else {
1525 release_tree_entry(e);
1526 del++;
1529 t->entry_count -= del;
1532 static void tree_content_replace(
1533 struct tree_entry *root,
1534 const struct object_id *oid,
1535 const uint16_t mode,
1536 struct tree_content *newtree)
1538 if (!S_ISDIR(mode))
1539 die("Root cannot be a non-directory");
1540 oidclr(&root->versions[0].oid);
1541 oidcpy(&root->versions[1].oid, oid);
1542 if (root->tree)
1543 release_tree_content_recursive(root->tree);
1544 root->tree = newtree;
1547 static int tree_content_set(
1548 struct tree_entry *root,
1549 const char *p,
1550 const struct object_id *oid,
1551 const uint16_t mode,
1552 struct tree_content *subtree)
1554 struct tree_content *t;
1555 const char *slash1;
1556 unsigned int i, n;
1557 struct tree_entry *e;
1559 slash1 = strchrnul(p, '/');
1560 n = slash1 - p;
1561 if (!n)
1562 die("Empty path component found in input");
1563 if (!*slash1 && !S_ISDIR(mode) && subtree)
1564 die("Non-directories cannot have subtrees");
1566 if (!root->tree)
1567 load_tree(root);
1568 t = root->tree;
1569 for (i = 0; i < t->entry_count; i++) {
1570 e = t->entries[i];
1571 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1572 if (!*slash1) {
1573 if (!S_ISDIR(mode)
1574 && e->versions[1].mode == mode
1575 && !oidcmp(&e->versions[1].oid, oid))
1576 return 0;
1577 e->versions[1].mode = mode;
1578 oidcpy(&e->versions[1].oid, oid);
1579 if (e->tree)
1580 release_tree_content_recursive(e->tree);
1581 e->tree = subtree;
1584 * We need to leave e->versions[0].sha1 alone
1585 * to avoid modifying the preimage tree used
1586 * when writing out the parent directory.
1587 * But after replacing the subdir with a
1588 * completely different one, it's not a good
1589 * delta base any more, and besides, we've
1590 * thrown away the tree entries needed to
1591 * make a delta against it.
1593 * So let's just explicitly disable deltas
1594 * for the subtree.
1596 if (S_ISDIR(e->versions[0].mode))
1597 e->versions[0].mode |= NO_DELTA;
1599 oidclr(&root->versions[1].oid);
1600 return 1;
1602 if (!S_ISDIR(e->versions[1].mode)) {
1603 e->tree = new_tree_content(8);
1604 e->versions[1].mode = S_IFDIR;
1606 if (!e->tree)
1607 load_tree(e);
1608 if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1609 oidclr(&root->versions[1].oid);
1610 return 1;
1612 return 0;
1616 if (t->entry_count == t->entry_capacity)
1617 root->tree = t = grow_tree_content(t, t->entry_count);
1618 e = new_tree_entry();
1619 e->name = to_atom(p, n);
1620 e->versions[0].mode = 0;
1621 oidclr(&e->versions[0].oid);
1622 t->entries[t->entry_count++] = e;
1623 if (*slash1) {
1624 e->tree = new_tree_content(8);
1625 e->versions[1].mode = S_IFDIR;
1626 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1627 } else {
1628 e->tree = subtree;
1629 e->versions[1].mode = mode;
1630 oidcpy(&e->versions[1].oid, oid);
1632 oidclr(&root->versions[1].oid);
1633 return 1;
1636 static int tree_content_remove(
1637 struct tree_entry *root,
1638 const char *p,
1639 struct tree_entry *backup_leaf,
1640 int allow_root)
1642 struct tree_content *t;
1643 const char *slash1;
1644 unsigned int i, n;
1645 struct tree_entry *e;
1647 slash1 = strchrnul(p, '/');
1648 n = slash1 - p;
1650 if (!root->tree)
1651 load_tree(root);
1653 if (!*p && allow_root) {
1654 e = root;
1655 goto del_entry;
1658 t = root->tree;
1659 for (i = 0; i < t->entry_count; i++) {
1660 e = t->entries[i];
1661 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1662 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1664 * If p names a file in some subdirectory, and a
1665 * file or symlink matching the name of the
1666 * parent directory of p exists, then p cannot
1667 * exist and need not be deleted.
1669 return 1;
1670 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1671 goto del_entry;
1672 if (!e->tree)
1673 load_tree(e);
1674 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1675 for (n = 0; n < e->tree->entry_count; n++) {
1676 if (e->tree->entries[n]->versions[1].mode) {
1677 oidclr(&root->versions[1].oid);
1678 return 1;
1681 backup_leaf = NULL;
1682 goto del_entry;
1684 return 0;
1687 return 0;
1689 del_entry:
1690 if (backup_leaf)
1691 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1692 else if (e->tree)
1693 release_tree_content_recursive(e->tree);
1694 e->tree = NULL;
1695 e->versions[1].mode = 0;
1696 oidclr(&e->versions[1].oid);
1697 oidclr(&root->versions[1].oid);
1698 return 1;
1701 static int tree_content_get(
1702 struct tree_entry *root,
1703 const char *p,
1704 struct tree_entry *leaf,
1705 int allow_root)
1707 struct tree_content *t;
1708 const char *slash1;
1709 unsigned int i, n;
1710 struct tree_entry *e;
1712 slash1 = strchrnul(p, '/');
1713 n = slash1 - p;
1714 if (!n && !allow_root)
1715 die("Empty path component found in input");
1717 if (!root->tree)
1718 load_tree(root);
1720 if (!n) {
1721 e = root;
1722 goto found_entry;
1725 t = root->tree;
1726 for (i = 0; i < t->entry_count; i++) {
1727 e = t->entries[i];
1728 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1729 if (!*slash1)
1730 goto found_entry;
1731 if (!S_ISDIR(e->versions[1].mode))
1732 return 0;
1733 if (!e->tree)
1734 load_tree(e);
1735 return tree_content_get(e, slash1 + 1, leaf, 0);
1738 return 0;
1740 found_entry:
1741 memcpy(leaf, e, sizeof(*leaf));
1742 if (e->tree && is_null_oid(&e->versions[1].oid))
1743 leaf->tree = dup_tree_content(e->tree);
1744 else
1745 leaf->tree = NULL;
1746 return 1;
1749 static int update_branch(struct branch *b)
1751 static const char *msg = "fast-import";
1752 struct ref_transaction *transaction;
1753 struct object_id old_oid;
1754 struct strbuf err = STRBUF_INIT;
1756 if (is_null_oid(&b->oid)) {
1757 if (b->delete)
1758 delete_ref(NULL, b->name, NULL, 0);
1759 return 0;
1761 if (read_ref(b->name, old_oid.hash))
1762 oidclr(&old_oid);
1763 if (!force_update && !is_null_oid(&old_oid)) {
1764 struct commit *old_cmit, *new_cmit;
1766 old_cmit = lookup_commit_reference_gently(&old_oid, 0);
1767 new_cmit = lookup_commit_reference_gently(&b->oid, 0);
1768 if (!old_cmit || !new_cmit)
1769 return error("Branch %s is missing commits.", b->name);
1771 if (!in_merge_bases(old_cmit, new_cmit)) {
1772 warning("Not updating %s"
1773 " (new tip %s does not contain %s)",
1774 b->name, oid_to_hex(&b->oid),
1775 oid_to_hex(&old_oid));
1776 return -1;
1779 transaction = ref_transaction_begin(&err);
1780 if (!transaction ||
1781 ref_transaction_update(transaction, b->name, b->oid.hash, old_oid.hash,
1782 0, msg, &err) ||
1783 ref_transaction_commit(transaction, &err)) {
1784 ref_transaction_free(transaction);
1785 error("%s", err.buf);
1786 strbuf_release(&err);
1787 return -1;
1789 ref_transaction_free(transaction);
1790 strbuf_release(&err);
1791 return 0;
1794 static void dump_branches(void)
1796 unsigned int i;
1797 struct branch *b;
1799 for (i = 0; i < branch_table_sz; i++) {
1800 for (b = branch_table[i]; b; b = b->table_next_branch)
1801 failure |= update_branch(b);
1805 static void dump_tags(void)
1807 static const char *msg = "fast-import";
1808 struct tag *t;
1809 struct strbuf ref_name = STRBUF_INIT;
1810 struct strbuf err = STRBUF_INIT;
1811 struct ref_transaction *transaction;
1813 transaction = ref_transaction_begin(&err);
1814 if (!transaction) {
1815 failure |= error("%s", err.buf);
1816 goto cleanup;
1818 for (t = first_tag; t; t = t->next_tag) {
1819 strbuf_reset(&ref_name);
1820 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1822 if (ref_transaction_update(transaction, ref_name.buf,
1823 t->oid.hash, NULL, 0, msg, &err)) {
1824 failure |= error("%s", err.buf);
1825 goto cleanup;
1828 if (ref_transaction_commit(transaction, &err))
1829 failure |= error("%s", err.buf);
1831 cleanup:
1832 ref_transaction_free(transaction);
1833 strbuf_release(&ref_name);
1834 strbuf_release(&err);
1837 static void dump_marks_helper(FILE *f,
1838 uintmax_t base,
1839 struct mark_set *m)
1841 uintmax_t k;
1842 if (m->shift) {
1843 for (k = 0; k < 1024; k++) {
1844 if (m->data.sets[k])
1845 dump_marks_helper(f, base + (k << m->shift),
1846 m->data.sets[k]);
1848 } else {
1849 for (k = 0; k < 1024; k++) {
1850 if (m->data.marked[k])
1851 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1852 oid_to_hex(&m->data.marked[k]->idx.oid));
1857 static void dump_marks(void)
1859 static struct lock_file mark_lock;
1860 FILE *f;
1862 if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1863 return;
1865 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1866 failure |= error_errno("Unable to write marks file %s",
1867 export_marks_file);
1868 return;
1871 f = fdopen_lock_file(&mark_lock, "w");
1872 if (!f) {
1873 int saved_errno = errno;
1874 rollback_lock_file(&mark_lock);
1875 failure |= error("Unable to write marks file %s: %s",
1876 export_marks_file, strerror(saved_errno));
1877 return;
1880 dump_marks_helper(f, 0, marks);
1881 if (commit_lock_file(&mark_lock)) {
1882 failure |= error_errno("Unable to write file %s",
1883 export_marks_file);
1884 return;
1888 static void read_marks(void)
1890 char line[512];
1891 FILE *f = fopen(import_marks_file, "r");
1892 if (f)
1894 else if (import_marks_file_ignore_missing && errno == ENOENT)
1895 goto done; /* Marks file does not exist */
1896 else
1897 die_errno("cannot read '%s'", import_marks_file);
1898 while (fgets(line, sizeof(line), f)) {
1899 uintmax_t mark;
1900 char *end;
1901 struct object_id oid;
1902 struct object_entry *e;
1904 end = strchr(line, '\n');
1905 if (line[0] != ':' || !end)
1906 die("corrupt mark line: %s", line);
1907 *end = 0;
1908 mark = strtoumax(line + 1, &end, 10);
1909 if (!mark || end == line + 1
1910 || *end != ' ' || get_oid_hex(end + 1, &oid))
1911 die("corrupt mark line: %s", line);
1912 e = find_object(&oid);
1913 if (!e) {
1914 enum object_type type = sha1_object_info(oid.hash, NULL);
1915 if (type < 0)
1916 die("object not found: %s", oid_to_hex(&oid));
1917 e = insert_object(&oid);
1918 e->type = type;
1919 e->pack_id = MAX_PACK_ID;
1920 e->idx.offset = 1; /* just not zero! */
1922 insert_mark(mark, e);
1924 fclose(f);
1925 done:
1926 import_marks_file_done = 1;
1930 static int read_next_command(void)
1932 static int stdin_eof = 0;
1934 if (stdin_eof) {
1935 unread_command_buf = 0;
1936 return EOF;
1939 for (;;) {
1940 const char *p;
1942 if (unread_command_buf) {
1943 unread_command_buf = 0;
1944 } else {
1945 struct recent_command *rc;
1947 strbuf_detach(&command_buf, NULL);
1948 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1949 if (stdin_eof)
1950 return EOF;
1952 if (!seen_data_command
1953 && !starts_with(command_buf.buf, "feature ")
1954 && !starts_with(command_buf.buf, "option ")) {
1955 parse_argv();
1958 rc = rc_free;
1959 if (rc)
1960 rc_free = rc->next;
1961 else {
1962 rc = cmd_hist.next;
1963 cmd_hist.next = rc->next;
1964 cmd_hist.next->prev = &cmd_hist;
1965 free(rc->buf);
1968 rc->buf = command_buf.buf;
1969 rc->prev = cmd_tail;
1970 rc->next = cmd_hist.prev;
1971 rc->prev->next = rc;
1972 cmd_tail = rc;
1974 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1975 parse_get_mark(p);
1976 continue;
1978 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1979 parse_cat_blob(p);
1980 continue;
1982 if (command_buf.buf[0] == '#')
1983 continue;
1984 return 0;
1988 static void skip_optional_lf(void)
1990 int term_char = fgetc(stdin);
1991 if (term_char != '\n' && term_char != EOF)
1992 ungetc(term_char, stdin);
1995 static void parse_mark(void)
1997 const char *v;
1998 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1999 next_mark = strtoumax(v, NULL, 10);
2000 read_next_command();
2002 else
2003 next_mark = 0;
2006 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
2008 const char *data;
2009 strbuf_reset(sb);
2011 if (!skip_prefix(command_buf.buf, "data ", &data))
2012 die("Expected 'data n' command, found: %s", command_buf.buf);
2014 if (skip_prefix(data, "<<", &data)) {
2015 char *term = xstrdup(data);
2016 size_t term_len = command_buf.len - (data - command_buf.buf);
2018 strbuf_detach(&command_buf, NULL);
2019 for (;;) {
2020 if (strbuf_getline_lf(&command_buf, stdin) == EOF)
2021 die("EOF in data (terminator '%s' not found)", term);
2022 if (term_len == command_buf.len
2023 && !strcmp(term, command_buf.buf))
2024 break;
2025 strbuf_addbuf(sb, &command_buf);
2026 strbuf_addch(sb, '\n');
2028 free(term);
2030 else {
2031 uintmax_t len = strtoumax(data, NULL, 10);
2032 size_t n = 0, length = (size_t)len;
2034 if (limit && limit < len) {
2035 *len_res = len;
2036 return 0;
2038 if (length < len)
2039 die("data is too large to use in this context");
2041 while (n < length) {
2042 size_t s = strbuf_fread(sb, length - n, stdin);
2043 if (!s && feof(stdin))
2044 die("EOF in data (%lu bytes remaining)",
2045 (unsigned long)(length - n));
2046 n += s;
2050 skip_optional_lf();
2051 return 1;
2054 static int validate_raw_date(const char *src, struct strbuf *result)
2056 const char *orig_src = src;
2057 char *endp;
2058 unsigned long num;
2060 errno = 0;
2062 num = strtoul(src, &endp, 10);
2063 /* NEEDSWORK: perhaps check for reasonable values? */
2064 if (errno || endp == src || *endp != ' ')
2065 return -1;
2067 src = endp + 1;
2068 if (*src != '-' && *src != '+')
2069 return -1;
2071 num = strtoul(src + 1, &endp, 10);
2072 if (errno || endp == src + 1 || *endp || 1400 < num)
2073 return -1;
2075 strbuf_addstr(result, orig_src);
2076 return 0;
2079 static char *parse_ident(const char *buf)
2081 const char *ltgt;
2082 size_t name_len;
2083 struct strbuf ident = STRBUF_INIT;
2085 /* ensure there is a space delimiter even if there is no name */
2086 if (*buf == '<')
2087 --buf;
2089 ltgt = buf + strcspn(buf, "<>");
2090 if (*ltgt != '<')
2091 die("Missing < in ident string: %s", buf);
2092 if (ltgt != buf && ltgt[-1] != ' ')
2093 die("Missing space before < in ident string: %s", buf);
2094 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2095 if (*ltgt != '>')
2096 die("Missing > in ident string: %s", buf);
2097 ltgt++;
2098 if (*ltgt != ' ')
2099 die("Missing space after > in ident string: %s", buf);
2100 ltgt++;
2101 name_len = ltgt - buf;
2102 strbuf_add(&ident, buf, name_len);
2104 switch (whenspec) {
2105 case WHENSPEC_RAW:
2106 if (validate_raw_date(ltgt, &ident) < 0)
2107 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2108 break;
2109 case WHENSPEC_RFC2822:
2110 if (parse_date(ltgt, &ident) < 0)
2111 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2112 break;
2113 case WHENSPEC_NOW:
2114 if (strcmp("now", ltgt))
2115 die("Date in ident must be 'now': %s", buf);
2116 datestamp(&ident);
2117 break;
2120 return strbuf_detach(&ident, NULL);
2123 static void parse_and_store_blob(
2124 struct last_object *last,
2125 struct object_id *oidout,
2126 uintmax_t mark)
2128 static struct strbuf buf = STRBUF_INIT;
2129 uintmax_t len;
2131 if (parse_data(&buf, big_file_threshold, &len))
2132 store_object(OBJ_BLOB, &buf, last, oidout, mark);
2133 else {
2134 if (last) {
2135 strbuf_release(&last->data);
2136 last->offset = 0;
2137 last->depth = 0;
2139 stream_blob(len, oidout, mark);
2140 skip_optional_lf();
2144 static void parse_new_blob(void)
2146 read_next_command();
2147 parse_mark();
2148 parse_and_store_blob(&last_blob, NULL, next_mark);
2151 static void unload_one_branch(void)
2153 while (cur_active_branches
2154 && cur_active_branches >= max_active_branches) {
2155 uintmax_t min_commit = ULONG_MAX;
2156 struct branch *e, *l = NULL, *p = NULL;
2158 for (e = active_branches; e; e = e->active_next_branch) {
2159 if (e->last_commit < min_commit) {
2160 p = l;
2161 min_commit = e->last_commit;
2163 l = e;
2166 if (p) {
2167 e = p->active_next_branch;
2168 p->active_next_branch = e->active_next_branch;
2169 } else {
2170 e = active_branches;
2171 active_branches = e->active_next_branch;
2173 e->active = 0;
2174 e->active_next_branch = NULL;
2175 if (e->branch_tree.tree) {
2176 release_tree_content_recursive(e->branch_tree.tree);
2177 e->branch_tree.tree = NULL;
2179 cur_active_branches--;
2183 static void load_branch(struct branch *b)
2185 load_tree(&b->branch_tree);
2186 if (!b->active) {
2187 b->active = 1;
2188 b->active_next_branch = active_branches;
2189 active_branches = b;
2190 cur_active_branches++;
2191 branch_load_count++;
2195 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2197 unsigned char fanout = 0;
2198 while ((num_notes >>= 8))
2199 fanout++;
2200 return fanout;
2203 static void construct_path_with_fanout(const char *hex_sha1,
2204 unsigned char fanout, char *path)
2206 unsigned int i = 0, j = 0;
2207 if (fanout >= 20)
2208 die("Too large fanout (%u)", fanout);
2209 while (fanout) {
2210 path[i++] = hex_sha1[j++];
2211 path[i++] = hex_sha1[j++];
2212 path[i++] = '/';
2213 fanout--;
2215 memcpy(path + i, hex_sha1 + j, GIT_SHA1_HEXSZ - j);
2216 path[i + GIT_SHA1_HEXSZ - j] = '\0';
2219 static uintmax_t do_change_note_fanout(
2220 struct tree_entry *orig_root, struct tree_entry *root,
2221 char *hex_oid, unsigned int hex_oid_len,
2222 char *fullpath, unsigned int fullpath_len,
2223 unsigned char fanout)
2225 struct tree_content *t;
2226 struct tree_entry *e, leaf;
2227 unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2228 uintmax_t num_notes = 0;
2229 struct object_id oid;
2230 char realpath[60];
2232 if (!root->tree)
2233 load_tree(root);
2234 t = root->tree;
2236 for (i = 0; t && i < t->entry_count; i++) {
2237 e = t->entries[i];
2238 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2239 tmp_fullpath_len = fullpath_len;
2242 * We're interested in EITHER existing note entries (entries
2243 * with exactly 40 hex chars in path, not including directory
2244 * separators), OR directory entries that may contain note
2245 * entries (with < 40 hex chars in path).
2246 * Also, each path component in a note entry must be a multiple
2247 * of 2 chars.
2249 if (!e->versions[1].mode ||
2250 tmp_hex_oid_len > GIT_SHA1_HEXSZ ||
2251 e->name->str_len % 2)
2252 continue;
2254 /* This _may_ be a note entry, or a subdir containing notes */
2255 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2256 e->name->str_len);
2257 if (tmp_fullpath_len)
2258 fullpath[tmp_fullpath_len++] = '/';
2259 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2260 e->name->str_len);
2261 tmp_fullpath_len += e->name->str_len;
2262 fullpath[tmp_fullpath_len] = '\0';
2264 if (tmp_hex_oid_len == GIT_SHA1_HEXSZ && !get_oid_hex(hex_oid, &oid)) {
2265 /* This is a note entry */
2266 if (fanout == 0xff) {
2267 /* Counting mode, no rename */
2268 num_notes++;
2269 continue;
2271 construct_path_with_fanout(hex_oid, fanout, realpath);
2272 if (!strcmp(fullpath, realpath)) {
2273 /* Note entry is in correct location */
2274 num_notes++;
2275 continue;
2278 /* Rename fullpath to realpath */
2279 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2280 die("Failed to remove path %s", fullpath);
2281 tree_content_set(orig_root, realpath,
2282 &leaf.versions[1].oid,
2283 leaf.versions[1].mode,
2284 leaf.tree);
2285 } else if (S_ISDIR(e->versions[1].mode)) {
2286 /* This is a subdir that may contain note entries */
2287 num_notes += do_change_note_fanout(orig_root, e,
2288 hex_oid, tmp_hex_oid_len,
2289 fullpath, tmp_fullpath_len, fanout);
2292 /* The above may have reallocated the current tree_content */
2293 t = root->tree;
2295 return num_notes;
2298 static uintmax_t change_note_fanout(struct tree_entry *root,
2299 unsigned char fanout)
2302 * The size of path is due to one slash between every two hex digits,
2303 * plus the terminating NUL. Note that there is no slash at the end, so
2304 * the number of slashes is one less than half the number of hex
2305 * characters.
2307 char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2308 return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2312 * Given a pointer into a string, parse a mark reference:
2314 * idnum ::= ':' bigint;
2316 * Return the first character after the value in *endptr.
2318 * Complain if the following character is not what is expected,
2319 * either a space or end of the string.
2321 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2323 uintmax_t mark;
2325 assert(*p == ':');
2326 p++;
2327 mark = strtoumax(p, endptr, 10);
2328 if (*endptr == p)
2329 die("No value after ':' in mark: %s", command_buf.buf);
2330 return mark;
2334 * Parse the mark reference, and complain if this is not the end of
2335 * the string.
2337 static uintmax_t parse_mark_ref_eol(const char *p)
2339 char *end;
2340 uintmax_t mark;
2342 mark = parse_mark_ref(p, &end);
2343 if (*end != '\0')
2344 die("Garbage after mark: %s", command_buf.buf);
2345 return mark;
2349 * Parse the mark reference, demanding a trailing space. Return a
2350 * pointer to the space.
2352 static uintmax_t parse_mark_ref_space(const char **p)
2354 uintmax_t mark;
2355 char *end;
2357 mark = parse_mark_ref(*p, &end);
2358 if (*end++ != ' ')
2359 die("Missing space after mark: %s", command_buf.buf);
2360 *p = end;
2361 return mark;
2364 static void file_change_m(const char *p, struct branch *b)
2366 static struct strbuf uq = STRBUF_INIT;
2367 const char *endp;
2368 struct object_entry *oe;
2369 struct object_id oid;
2370 uint16_t mode, inline_data = 0;
2372 p = get_mode(p, &mode);
2373 if (!p)
2374 die("Corrupt mode: %s", command_buf.buf);
2375 switch (mode) {
2376 case 0644:
2377 case 0755:
2378 mode |= S_IFREG;
2379 case S_IFREG | 0644:
2380 case S_IFREG | 0755:
2381 case S_IFLNK:
2382 case S_IFDIR:
2383 case S_IFGITLINK:
2384 /* ok */
2385 break;
2386 default:
2387 die("Corrupt mode: %s", command_buf.buf);
2390 if (*p == ':') {
2391 oe = find_mark(parse_mark_ref_space(&p));
2392 oidcpy(&oid, &oe->idx.oid);
2393 } else if (skip_prefix(p, "inline ", &p)) {
2394 inline_data = 1;
2395 oe = NULL; /* not used with inline_data, but makes gcc happy */
2396 } else {
2397 if (parse_oid_hex(p, &oid, &p))
2398 die("Invalid dataref: %s", command_buf.buf);
2399 oe = find_object(&oid);
2400 if (*p++ != ' ')
2401 die("Missing space after SHA1: %s", command_buf.buf);
2404 strbuf_reset(&uq);
2405 if (!unquote_c_style(&uq, p, &endp)) {
2406 if (*endp)
2407 die("Garbage after path in: %s", command_buf.buf);
2408 p = uq.buf;
2411 /* Git does not track empty, non-toplevel directories. */
2412 if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2413 tree_content_remove(&b->branch_tree, p, NULL, 0);
2414 return;
2417 if (S_ISGITLINK(mode)) {
2418 if (inline_data)
2419 die("Git links cannot be specified 'inline': %s",
2420 command_buf.buf);
2421 else if (oe) {
2422 if (oe->type != OBJ_COMMIT)
2423 die("Not a commit (actually a %s): %s",
2424 typename(oe->type), command_buf.buf);
2427 * Accept the sha1 without checking; it expected to be in
2428 * another repository.
2430 } else if (inline_data) {
2431 if (S_ISDIR(mode))
2432 die("Directories cannot be specified 'inline': %s",
2433 command_buf.buf);
2434 if (p != uq.buf) {
2435 strbuf_addstr(&uq, p);
2436 p = uq.buf;
2438 read_next_command();
2439 parse_and_store_blob(&last_blob, &oid, 0);
2440 } else {
2441 enum object_type expected = S_ISDIR(mode) ?
2442 OBJ_TREE: OBJ_BLOB;
2443 enum object_type type = oe ? oe->type :
2444 sha1_object_info(oid.hash, NULL);
2445 if (type < 0)
2446 die("%s not found: %s",
2447 S_ISDIR(mode) ? "Tree" : "Blob",
2448 command_buf.buf);
2449 if (type != expected)
2450 die("Not a %s (actually a %s): %s",
2451 typename(expected), typename(type),
2452 command_buf.buf);
2455 if (!*p) {
2456 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2457 return;
2459 tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2462 static void file_change_d(const char *p, struct branch *b)
2464 static struct strbuf uq = STRBUF_INIT;
2465 const char *endp;
2467 strbuf_reset(&uq);
2468 if (!unquote_c_style(&uq, p, &endp)) {
2469 if (*endp)
2470 die("Garbage after path in: %s", command_buf.buf);
2471 p = uq.buf;
2473 tree_content_remove(&b->branch_tree, p, NULL, 1);
2476 static void file_change_cr(const char *s, struct branch *b, int rename)
2478 const char *d;
2479 static struct strbuf s_uq = STRBUF_INIT;
2480 static struct strbuf d_uq = STRBUF_INIT;
2481 const char *endp;
2482 struct tree_entry leaf;
2484 strbuf_reset(&s_uq);
2485 if (!unquote_c_style(&s_uq, s, &endp)) {
2486 if (*endp != ' ')
2487 die("Missing space after source: %s", command_buf.buf);
2488 } else {
2489 endp = strchr(s, ' ');
2490 if (!endp)
2491 die("Missing space after source: %s", command_buf.buf);
2492 strbuf_add(&s_uq, s, endp - s);
2494 s = s_uq.buf;
2496 endp++;
2497 if (!*endp)
2498 die("Missing dest: %s", command_buf.buf);
2500 d = endp;
2501 strbuf_reset(&d_uq);
2502 if (!unquote_c_style(&d_uq, d, &endp)) {
2503 if (*endp)
2504 die("Garbage after dest in: %s", command_buf.buf);
2505 d = d_uq.buf;
2508 memset(&leaf, 0, sizeof(leaf));
2509 if (rename)
2510 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2511 else
2512 tree_content_get(&b->branch_tree, s, &leaf, 1);
2513 if (!leaf.versions[1].mode)
2514 die("Path %s not in branch", s);
2515 if (!*d) { /* C "path/to/subdir" "" */
2516 tree_content_replace(&b->branch_tree,
2517 &leaf.versions[1].oid,
2518 leaf.versions[1].mode,
2519 leaf.tree);
2520 return;
2522 tree_content_set(&b->branch_tree, d,
2523 &leaf.versions[1].oid,
2524 leaf.versions[1].mode,
2525 leaf.tree);
2528 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2530 static struct strbuf uq = STRBUF_INIT;
2531 struct object_entry *oe;
2532 struct branch *s;
2533 struct object_id oid, commit_oid;
2534 char path[60];
2535 uint16_t inline_data = 0;
2536 unsigned char new_fanout;
2539 * When loading a branch, we don't traverse its tree to count the real
2540 * number of notes (too expensive to do this for all non-note refs).
2541 * This means that recently loaded notes refs might incorrectly have
2542 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2544 * Fix this by traversing the tree and counting the number of notes
2545 * when b->num_notes == 0. If the notes tree is truly empty, the
2546 * calculation should not take long.
2548 if (b->num_notes == 0 && *old_fanout == 0) {
2549 /* Invoke change_note_fanout() in "counting mode". */
2550 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2551 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2554 /* Now parse the notemodify command. */
2555 /* <dataref> or 'inline' */
2556 if (*p == ':') {
2557 oe = find_mark(parse_mark_ref_space(&p));
2558 oidcpy(&oid, &oe->idx.oid);
2559 } else if (skip_prefix(p, "inline ", &p)) {
2560 inline_data = 1;
2561 oe = NULL; /* not used with inline_data, but makes gcc happy */
2562 } else {
2563 if (parse_oid_hex(p, &oid, &p))
2564 die("Invalid dataref: %s", command_buf.buf);
2565 oe = find_object(&oid);
2566 if (*p++ != ' ')
2567 die("Missing space after SHA1: %s", command_buf.buf);
2570 /* <commit-ish> */
2571 s = lookup_branch(p);
2572 if (s) {
2573 if (is_null_oid(&s->oid))
2574 die("Can't add a note on empty branch.");
2575 oidcpy(&commit_oid, &s->oid);
2576 } else if (*p == ':') {
2577 uintmax_t commit_mark = parse_mark_ref_eol(p);
2578 struct object_entry *commit_oe = find_mark(commit_mark);
2579 if (commit_oe->type != OBJ_COMMIT)
2580 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2581 oidcpy(&commit_oid, &commit_oe->idx.oid);
2582 } else if (!get_oid(p, &commit_oid)) {
2583 unsigned long size;
2584 char *buf = read_object_with_reference(commit_oid.hash,
2585 commit_type, &size, commit_oid.hash);
2586 if (!buf || size < 46)
2587 die("Not a valid commit: %s", p);
2588 free(buf);
2589 } else
2590 die("Invalid ref name or SHA1 expression: %s", p);
2592 if (inline_data) {
2593 if (p != uq.buf) {
2594 strbuf_addstr(&uq, p);
2595 p = uq.buf;
2597 read_next_command();
2598 parse_and_store_blob(&last_blob, &oid, 0);
2599 } else if (oe) {
2600 if (oe->type != OBJ_BLOB)
2601 die("Not a blob (actually a %s): %s",
2602 typename(oe->type), command_buf.buf);
2603 } else if (!is_null_oid(&oid)) {
2604 enum object_type type = sha1_object_info(oid.hash, NULL);
2605 if (type < 0)
2606 die("Blob not found: %s", command_buf.buf);
2607 if (type != OBJ_BLOB)
2608 die("Not a blob (actually a %s): %s",
2609 typename(type), command_buf.buf);
2612 construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2613 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2614 b->num_notes--;
2616 if (is_null_oid(&oid))
2617 return; /* nothing to insert */
2619 b->num_notes++;
2620 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2621 construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2622 tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2625 static void file_change_deleteall(struct branch *b)
2627 release_tree_content_recursive(b->branch_tree.tree);
2628 oidclr(&b->branch_tree.versions[0].oid);
2629 oidclr(&b->branch_tree.versions[1].oid);
2630 load_tree(&b->branch_tree);
2631 b->num_notes = 0;
2634 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2636 if (!buf || size < GIT_SHA1_HEXSZ + 6)
2637 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2638 if (memcmp("tree ", buf, 5)
2639 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2640 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2641 oidcpy(&b->branch_tree.versions[0].oid,
2642 &b->branch_tree.versions[1].oid);
2645 static void parse_from_existing(struct branch *b)
2647 if (is_null_oid(&b->oid)) {
2648 oidclr(&b->branch_tree.versions[0].oid);
2649 oidclr(&b->branch_tree.versions[1].oid);
2650 } else {
2651 unsigned long size;
2652 char *buf;
2654 buf = read_object_with_reference(b->oid.hash,
2655 commit_type, &size,
2656 b->oid.hash);
2657 parse_from_commit(b, buf, size);
2658 free(buf);
2662 static int parse_from(struct branch *b)
2664 const char *from;
2665 struct branch *s;
2666 struct object_id oid;
2668 if (!skip_prefix(command_buf.buf, "from ", &from))
2669 return 0;
2671 oidcpy(&oid, &b->branch_tree.versions[1].oid);
2673 s = lookup_branch(from);
2674 if (b == s)
2675 die("Can't create a branch from itself: %s", b->name);
2676 else if (s) {
2677 struct object_id *t = &s->branch_tree.versions[1].oid;
2678 oidcpy(&b->oid, &s->oid);
2679 oidcpy(&b->branch_tree.versions[0].oid, t);
2680 oidcpy(&b->branch_tree.versions[1].oid, t);
2681 } else if (*from == ':') {
2682 uintmax_t idnum = parse_mark_ref_eol(from);
2683 struct object_entry *oe = find_mark(idnum);
2684 if (oe->type != OBJ_COMMIT)
2685 die("Mark :%" PRIuMAX " not a commit", idnum);
2686 if (oidcmp(&b->oid, &oe->idx.oid)) {
2687 oidcpy(&b->oid, &oe->idx.oid);
2688 if (oe->pack_id != MAX_PACK_ID) {
2689 unsigned long size;
2690 char *buf = gfi_unpack_entry(oe, &size);
2691 parse_from_commit(b, buf, size);
2692 free(buf);
2693 } else
2694 parse_from_existing(b);
2696 } else if (!get_oid(from, &b->oid)) {
2697 parse_from_existing(b);
2698 if (is_null_oid(&b->oid))
2699 b->delete = 1;
2701 else
2702 die("Invalid ref name or SHA1 expression: %s", from);
2704 if (b->branch_tree.tree && oidcmp(&oid, &b->branch_tree.versions[1].oid)) {
2705 release_tree_content_recursive(b->branch_tree.tree);
2706 b->branch_tree.tree = NULL;
2709 read_next_command();
2710 return 1;
2713 static struct hash_list *parse_merge(unsigned int *count)
2715 struct hash_list *list = NULL, **tail = &list, *n;
2716 const char *from;
2717 struct branch *s;
2719 *count = 0;
2720 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2721 n = xmalloc(sizeof(*n));
2722 s = lookup_branch(from);
2723 if (s)
2724 oidcpy(&n->oid, &s->oid);
2725 else if (*from == ':') {
2726 uintmax_t idnum = parse_mark_ref_eol(from);
2727 struct object_entry *oe = find_mark(idnum);
2728 if (oe->type != OBJ_COMMIT)
2729 die("Mark :%" PRIuMAX " not a commit", idnum);
2730 oidcpy(&n->oid, &oe->idx.oid);
2731 } else if (!get_oid(from, &n->oid)) {
2732 unsigned long size;
2733 char *buf = read_object_with_reference(n->oid.hash,
2734 commit_type, &size, n->oid.hash);
2735 if (!buf || size < 46)
2736 die("Not a valid commit: %s", from);
2737 free(buf);
2738 } else
2739 die("Invalid ref name or SHA1 expression: %s", from);
2741 n->next = NULL;
2742 *tail = n;
2743 tail = &n->next;
2745 (*count)++;
2746 read_next_command();
2748 return list;
2751 static void parse_new_commit(const char *arg)
2753 static struct strbuf msg = STRBUF_INIT;
2754 struct branch *b;
2755 char *author = NULL;
2756 char *committer = NULL;
2757 struct hash_list *merge_list = NULL;
2758 unsigned int merge_count;
2759 unsigned char prev_fanout, new_fanout;
2760 const char *v;
2762 b = lookup_branch(arg);
2763 if (!b)
2764 b = new_branch(arg);
2766 read_next_command();
2767 parse_mark();
2768 if (skip_prefix(command_buf.buf, "author ", &v)) {
2769 author = parse_ident(v);
2770 read_next_command();
2772 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2773 committer = parse_ident(v);
2774 read_next_command();
2776 if (!committer)
2777 die("Expected committer but didn't get one");
2778 parse_data(&msg, 0, NULL);
2779 read_next_command();
2780 parse_from(b);
2781 merge_list = parse_merge(&merge_count);
2783 /* ensure the branch is active/loaded */
2784 if (!b->branch_tree.tree || !max_active_branches) {
2785 unload_one_branch();
2786 load_branch(b);
2789 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2791 /* file_change* */
2792 while (command_buf.len > 0) {
2793 if (skip_prefix(command_buf.buf, "M ", &v))
2794 file_change_m(v, b);
2795 else if (skip_prefix(command_buf.buf, "D ", &v))
2796 file_change_d(v, b);
2797 else if (skip_prefix(command_buf.buf, "R ", &v))
2798 file_change_cr(v, b, 1);
2799 else if (skip_prefix(command_buf.buf, "C ", &v))
2800 file_change_cr(v, b, 0);
2801 else if (skip_prefix(command_buf.buf, "N ", &v))
2802 note_change_n(v, b, &prev_fanout);
2803 else if (!strcmp("deleteall", command_buf.buf))
2804 file_change_deleteall(b);
2805 else if (skip_prefix(command_buf.buf, "ls ", &v))
2806 parse_ls(v, b);
2807 else {
2808 unread_command_buf = 1;
2809 break;
2811 if (read_next_command() == EOF)
2812 break;
2815 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2816 if (new_fanout != prev_fanout)
2817 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2819 /* build the tree and the commit */
2820 store_tree(&b->branch_tree);
2821 oidcpy(&b->branch_tree.versions[0].oid,
2822 &b->branch_tree.versions[1].oid);
2824 strbuf_reset(&new_data);
2825 strbuf_addf(&new_data, "tree %s\n",
2826 oid_to_hex(&b->branch_tree.versions[1].oid));
2827 if (!is_null_oid(&b->oid))
2828 strbuf_addf(&new_data, "parent %s\n",
2829 oid_to_hex(&b->oid));
2830 while (merge_list) {
2831 struct hash_list *next = merge_list->next;
2832 strbuf_addf(&new_data, "parent %s\n",
2833 oid_to_hex(&merge_list->oid));
2834 free(merge_list);
2835 merge_list = next;
2837 strbuf_addf(&new_data,
2838 "author %s\n"
2839 "committer %s\n"
2840 "\n",
2841 author ? author : committer, committer);
2842 strbuf_addbuf(&new_data, &msg);
2843 free(author);
2844 free(committer);
2846 if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2847 b->pack_id = pack_id;
2848 b->last_commit = object_count_by_type[OBJ_COMMIT];
2851 static void parse_new_tag(const char *arg)
2853 static struct strbuf msg = STRBUF_INIT;
2854 const char *from;
2855 char *tagger;
2856 struct branch *s;
2857 struct tag *t;
2858 uintmax_t from_mark = 0;
2859 struct object_id oid;
2860 enum object_type type;
2861 const char *v;
2863 t = pool_alloc(sizeof(struct tag));
2864 memset(t, 0, sizeof(struct tag));
2865 t->name = pool_strdup(arg);
2866 if (last_tag)
2867 last_tag->next_tag = t;
2868 else
2869 first_tag = t;
2870 last_tag = t;
2871 read_next_command();
2873 /* from ... */
2874 if (!skip_prefix(command_buf.buf, "from ", &from))
2875 die("Expected from command, got %s", command_buf.buf);
2876 s = lookup_branch(from);
2877 if (s) {
2878 if (is_null_oid(&s->oid))
2879 die("Can't tag an empty branch.");
2880 oidcpy(&oid, &s->oid);
2881 type = OBJ_COMMIT;
2882 } else if (*from == ':') {
2883 struct object_entry *oe;
2884 from_mark = parse_mark_ref_eol(from);
2885 oe = find_mark(from_mark);
2886 type = oe->type;
2887 oidcpy(&oid, &oe->idx.oid);
2888 } else if (!get_oid(from, &oid)) {
2889 struct object_entry *oe = find_object(&oid);
2890 if (!oe) {
2891 type = sha1_object_info(oid.hash, NULL);
2892 if (type < 0)
2893 die("Not a valid object: %s", from);
2894 } else
2895 type = oe->type;
2896 } else
2897 die("Invalid ref name or SHA1 expression: %s", from);
2898 read_next_command();
2900 /* tagger ... */
2901 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2902 tagger = parse_ident(v);
2903 read_next_command();
2904 } else
2905 tagger = NULL;
2907 /* tag payload/message */
2908 parse_data(&msg, 0, NULL);
2910 /* build the tag object */
2911 strbuf_reset(&new_data);
2913 strbuf_addf(&new_data,
2914 "object %s\n"
2915 "type %s\n"
2916 "tag %s\n",
2917 oid_to_hex(&oid), typename(type), t->name);
2918 if (tagger)
2919 strbuf_addf(&new_data,
2920 "tagger %s\n", tagger);
2921 strbuf_addch(&new_data, '\n');
2922 strbuf_addbuf(&new_data, &msg);
2923 free(tagger);
2925 if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, 0))
2926 t->pack_id = MAX_PACK_ID;
2927 else
2928 t->pack_id = pack_id;
2931 static void parse_reset_branch(const char *arg)
2933 struct branch *b;
2935 b = lookup_branch(arg);
2936 if (b) {
2937 oidclr(&b->oid);
2938 oidclr(&b->branch_tree.versions[0].oid);
2939 oidclr(&b->branch_tree.versions[1].oid);
2940 if (b->branch_tree.tree) {
2941 release_tree_content_recursive(b->branch_tree.tree);
2942 b->branch_tree.tree = NULL;
2945 else
2946 b = new_branch(arg);
2947 read_next_command();
2948 parse_from(b);
2949 if (command_buf.len > 0)
2950 unread_command_buf = 1;
2953 static void cat_blob_write(const char *buf, unsigned long size)
2955 if (write_in_full(cat_blob_fd, buf, size) < 0)
2956 die_errno("Write to frontend failed");
2959 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2961 struct strbuf line = STRBUF_INIT;
2962 unsigned long size;
2963 enum object_type type = 0;
2964 char *buf;
2966 if (!oe || oe->pack_id == MAX_PACK_ID) {
2967 buf = read_sha1_file(oid->hash, &type, &size);
2968 } else {
2969 type = oe->type;
2970 buf = gfi_unpack_entry(oe, &size);
2974 * Output based on batch_one_object() from cat-file.c.
2976 if (type <= 0) {
2977 strbuf_reset(&line);
2978 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2979 cat_blob_write(line.buf, line.len);
2980 strbuf_release(&line);
2981 free(buf);
2982 return;
2984 if (!buf)
2985 die("Can't read object %s", oid_to_hex(oid));
2986 if (type != OBJ_BLOB)
2987 die("Object %s is a %s but a blob was expected.",
2988 oid_to_hex(oid), typename(type));
2989 strbuf_reset(&line);
2990 strbuf_addf(&line, "%s %s %lu\n", oid_to_hex(oid),
2991 typename(type), size);
2992 cat_blob_write(line.buf, line.len);
2993 strbuf_release(&line);
2994 cat_blob_write(buf, size);
2995 cat_blob_write("\n", 1);
2996 if (oe && oe->pack_id == pack_id) {
2997 last_blob.offset = oe->idx.offset;
2998 strbuf_attach(&last_blob.data, buf, size, size);
2999 last_blob.depth = oe->depth;
3000 } else
3001 free(buf);
3004 static void parse_get_mark(const char *p)
3006 struct object_entry *oe = oe;
3007 char output[GIT_MAX_HEXSZ + 2];
3009 /* get-mark SP <object> LF */
3010 if (*p != ':')
3011 die("Not a mark: %s", p);
3013 oe = find_mark(parse_mark_ref_eol(p));
3014 if (!oe)
3015 die("Unknown mark: %s", command_buf.buf);
3017 xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
3018 cat_blob_write(output, GIT_SHA1_HEXSZ + 1);
3021 static void parse_cat_blob(const char *p)
3023 struct object_entry *oe = oe;
3024 struct object_id oid;
3026 /* cat-blob SP <object> LF */
3027 if (*p == ':') {
3028 oe = find_mark(parse_mark_ref_eol(p));
3029 if (!oe)
3030 die("Unknown mark: %s", command_buf.buf);
3031 oidcpy(&oid, &oe->idx.oid);
3032 } else {
3033 if (parse_oid_hex(p, &oid, &p))
3034 die("Invalid dataref: %s", command_buf.buf);
3035 if (*p)
3036 die("Garbage after SHA1: %s", command_buf.buf);
3037 oe = find_object(&oid);
3040 cat_blob(oe, &oid);
3043 static struct object_entry *dereference(struct object_entry *oe,
3044 struct object_id *oid)
3046 unsigned long size;
3047 char *buf = NULL;
3048 if (!oe) {
3049 enum object_type type = sha1_object_info(oid->hash, NULL);
3050 if (type < 0)
3051 die("object not found: %s", oid_to_hex(oid));
3052 /* cache it! */
3053 oe = insert_object(oid);
3054 oe->type = type;
3055 oe->pack_id = MAX_PACK_ID;
3056 oe->idx.offset = 1;
3058 switch (oe->type) {
3059 case OBJ_TREE: /* easy case. */
3060 return oe;
3061 case OBJ_COMMIT:
3062 case OBJ_TAG:
3063 break;
3064 default:
3065 die("Not a tree-ish: %s", command_buf.buf);
3068 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3069 buf = gfi_unpack_entry(oe, &size);
3070 } else {
3071 enum object_type unused;
3072 buf = read_sha1_file(oid->hash, &unused, &size);
3074 if (!buf)
3075 die("Can't load object %s", oid_to_hex(oid));
3077 /* Peel one layer. */
3078 switch (oe->type) {
3079 case OBJ_TAG:
3080 if (size < GIT_SHA1_HEXSZ + strlen("object ") ||
3081 get_oid_hex(buf + strlen("object "), oid))
3082 die("Invalid SHA1 in tag: %s", command_buf.buf);
3083 break;
3084 case OBJ_COMMIT:
3085 if (size < GIT_SHA1_HEXSZ + strlen("tree ") ||
3086 get_oid_hex(buf + strlen("tree "), oid))
3087 die("Invalid SHA1 in commit: %s", command_buf.buf);
3090 free(buf);
3091 return find_object(oid);
3094 static struct object_entry *parse_treeish_dataref(const char **p)
3096 struct object_id oid;
3097 struct object_entry *e;
3099 if (**p == ':') { /* <mark> */
3100 e = find_mark(parse_mark_ref_space(p));
3101 if (!e)
3102 die("Unknown mark: %s", command_buf.buf);
3103 oidcpy(&oid, &e->idx.oid);
3104 } else { /* <sha1> */
3105 if (parse_oid_hex(*p, &oid, p))
3106 die("Invalid dataref: %s", command_buf.buf);
3107 e = find_object(&oid);
3108 if (*(*p)++ != ' ')
3109 die("Missing space after tree-ish: %s", command_buf.buf);
3112 while (!e || e->type != OBJ_TREE)
3113 e = dereference(e, &oid);
3114 return e;
3117 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3119 static struct strbuf line = STRBUF_INIT;
3121 /* See show_tree(). */
3122 const char *type =
3123 S_ISGITLINK(mode) ? commit_type :
3124 S_ISDIR(mode) ? tree_type :
3125 blob_type;
3127 if (!mode) {
3128 /* missing SP path LF */
3129 strbuf_reset(&line);
3130 strbuf_addstr(&line, "missing ");
3131 quote_c_style(path, &line, NULL, 0);
3132 strbuf_addch(&line, '\n');
3133 } else {
3134 /* mode SP type SP object_name TAB path LF */
3135 strbuf_reset(&line);
3136 strbuf_addf(&line, "%06o %s %s\t",
3137 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3138 quote_c_style(path, &line, NULL, 0);
3139 strbuf_addch(&line, '\n');
3141 cat_blob_write(line.buf, line.len);
3144 static void parse_ls(const char *p, struct branch *b)
3146 struct tree_entry *root = NULL;
3147 struct tree_entry leaf = {NULL};
3149 /* ls SP (<tree-ish> SP)? <path> */
3150 if (*p == '"') {
3151 if (!b)
3152 die("Not in a commit: %s", command_buf.buf);
3153 root = &b->branch_tree;
3154 } else {
3155 struct object_entry *e = parse_treeish_dataref(&p);
3156 root = new_tree_entry();
3157 oidcpy(&root->versions[1].oid, &e->idx.oid);
3158 if (!is_null_oid(&root->versions[1].oid))
3159 root->versions[1].mode = S_IFDIR;
3160 load_tree(root);
3162 if (*p == '"') {
3163 static struct strbuf uq = STRBUF_INIT;
3164 const char *endp;
3165 strbuf_reset(&uq);
3166 if (unquote_c_style(&uq, p, &endp))
3167 die("Invalid path: %s", command_buf.buf);
3168 if (*endp)
3169 die("Garbage after path in: %s", command_buf.buf);
3170 p = uq.buf;
3172 tree_content_get(root, p, &leaf, 1);
3174 * A directory in preparation would have a sha1 of zero
3175 * until it is saved. Save, for simplicity.
3177 if (S_ISDIR(leaf.versions[1].mode))
3178 store_tree(&leaf);
3180 print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3181 if (leaf.tree)
3182 release_tree_content_recursive(leaf.tree);
3183 if (!b || root != &b->branch_tree)
3184 release_tree_entry(root);
3187 static void checkpoint(void)
3189 checkpoint_requested = 0;
3190 if (object_count) {
3191 cycle_packfile();
3193 dump_branches();
3194 dump_tags();
3195 dump_marks();
3198 static void parse_checkpoint(void)
3200 checkpoint_requested = 1;
3201 skip_optional_lf();
3204 static void parse_progress(void)
3206 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3207 fputc('\n', stdout);
3208 fflush(stdout);
3209 skip_optional_lf();
3212 static char* make_fast_import_path(const char *path)
3214 if (!relative_marks_paths || is_absolute_path(path))
3215 return xstrdup(path);
3216 return git_pathdup("info/fast-import/%s", path);
3219 static void option_import_marks(const char *marks,
3220 int from_stream, int ignore_missing)
3222 if (import_marks_file) {
3223 if (from_stream)
3224 die("Only one import-marks command allowed per stream");
3226 /* read previous mark file */
3227 if(!import_marks_file_from_stream)
3228 read_marks();
3231 import_marks_file = make_fast_import_path(marks);
3232 safe_create_leading_directories_const(import_marks_file);
3233 import_marks_file_from_stream = from_stream;
3234 import_marks_file_ignore_missing = ignore_missing;
3237 static void option_date_format(const char *fmt)
3239 if (!strcmp(fmt, "raw"))
3240 whenspec = WHENSPEC_RAW;
3241 else if (!strcmp(fmt, "rfc2822"))
3242 whenspec = WHENSPEC_RFC2822;
3243 else if (!strcmp(fmt, "now"))
3244 whenspec = WHENSPEC_NOW;
3245 else
3246 die("unknown --date-format argument %s", fmt);
3249 static unsigned long ulong_arg(const char *option, const char *arg)
3251 char *endptr;
3252 unsigned long rv = strtoul(arg, &endptr, 0);
3253 if (strchr(arg, '-') || endptr == arg || *endptr)
3254 die("%s: argument must be a non-negative integer", option);
3255 return rv;
3258 static void option_depth(const char *depth)
3260 max_depth = ulong_arg("--depth", depth);
3261 if (max_depth > MAX_DEPTH)
3262 die("--depth cannot exceed %u", MAX_DEPTH);
3265 static void option_active_branches(const char *branches)
3267 max_active_branches = ulong_arg("--active-branches", branches);
3270 static void option_export_marks(const char *marks)
3272 export_marks_file = make_fast_import_path(marks);
3273 safe_create_leading_directories_const(export_marks_file);
3276 static void option_cat_blob_fd(const char *fd)
3278 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3279 if (n > (unsigned long) INT_MAX)
3280 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3281 cat_blob_fd = (int) n;
3284 static void option_export_pack_edges(const char *edges)
3286 if (pack_edges)
3287 fclose(pack_edges);
3288 pack_edges = xfopen(edges, "a");
3291 static int parse_one_option(const char *option)
3293 if (skip_prefix(option, "max-pack-size=", &option)) {
3294 unsigned long v;
3295 if (!git_parse_ulong(option, &v))
3296 return 0;
3297 if (v < 8192) {
3298 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3299 v *= 1024 * 1024;
3300 } else if (v < 1024 * 1024) {
3301 warning("minimum max-pack-size is 1 MiB");
3302 v = 1024 * 1024;
3304 max_packsize = v;
3305 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3306 unsigned long v;
3307 if (!git_parse_ulong(option, &v))
3308 return 0;
3309 big_file_threshold = v;
3310 } else if (skip_prefix(option, "depth=", &option)) {
3311 option_depth(option);
3312 } else if (skip_prefix(option, "active-branches=", &option)) {
3313 option_active_branches(option);
3314 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3315 option_export_pack_edges(option);
3316 } else if (starts_with(option, "quiet")) {
3317 show_stats = 0;
3318 } else if (starts_with(option, "stats")) {
3319 show_stats = 1;
3320 } else {
3321 return 0;
3324 return 1;
3327 static int parse_one_feature(const char *feature, int from_stream)
3329 const char *arg;
3331 if (skip_prefix(feature, "date-format=", &arg)) {
3332 option_date_format(arg);
3333 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3334 option_import_marks(arg, from_stream, 0);
3335 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3336 option_import_marks(arg, from_stream, 1);
3337 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3338 option_export_marks(arg);
3339 } else if (!strcmp(feature, "get-mark")) {
3340 ; /* Don't die - this feature is supported */
3341 } else if (!strcmp(feature, "cat-blob")) {
3342 ; /* Don't die - this feature is supported */
3343 } else if (!strcmp(feature, "relative-marks")) {
3344 relative_marks_paths = 1;
3345 } else if (!strcmp(feature, "no-relative-marks")) {
3346 relative_marks_paths = 0;
3347 } else if (!strcmp(feature, "done")) {
3348 require_explicit_termination = 1;
3349 } else if (!strcmp(feature, "force")) {
3350 force_update = 1;
3351 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3352 ; /* do nothing; we have the feature */
3353 } else {
3354 return 0;
3357 return 1;
3360 static void parse_feature(const char *feature)
3362 if (seen_data_command)
3363 die("Got feature command '%s' after data command", feature);
3365 if (parse_one_feature(feature, 1))
3366 return;
3368 die("This version of fast-import does not support feature %s.", feature);
3371 static void parse_option(const char *option)
3373 if (seen_data_command)
3374 die("Got option command '%s' after data command", option);
3376 if (parse_one_option(option))
3377 return;
3379 die("This version of fast-import does not support option: %s", option);
3382 static void git_pack_config(void)
3384 int indexversion_value;
3385 int limit;
3386 unsigned long packsizelimit_value;
3388 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3389 if (max_depth > MAX_DEPTH)
3390 max_depth = MAX_DEPTH;
3392 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3393 pack_idx_opts.version = indexversion_value;
3394 if (pack_idx_opts.version > 2)
3395 git_die_config("pack.indexversion",
3396 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3398 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3399 max_packsize = packsizelimit_value;
3401 if (!git_config_get_int("fastimport.unpacklimit", &limit))
3402 unpack_limit = limit;
3403 else if (!git_config_get_int("transfer.unpacklimit", &limit))
3404 unpack_limit = limit;
3406 git_config(git_default_config, NULL);
3409 static const char fast_import_usage[] =
3410 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3412 static void parse_argv(void)
3414 unsigned int i;
3416 for (i = 1; i < global_argc; i++) {
3417 const char *a = global_argv[i];
3419 if (*a != '-' || !strcmp(a, "--"))
3420 break;
3422 if (!skip_prefix(a, "--", &a))
3423 die("unknown option %s", a);
3425 if (parse_one_option(a))
3426 continue;
3428 if (parse_one_feature(a, 0))
3429 continue;
3431 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3432 option_cat_blob_fd(a);
3433 continue;
3436 die("unknown option --%s", a);
3438 if (i != global_argc)
3439 usage(fast_import_usage);
3441 seen_data_command = 1;
3442 if (import_marks_file)
3443 read_marks();
3446 int cmd_main(int argc, const char **argv)
3448 unsigned int i;
3450 if (argc == 2 && !strcmp(argv[1], "-h"))
3451 usage(fast_import_usage);
3453 setup_git_directory();
3454 reset_pack_idx_option(&pack_idx_opts);
3455 git_pack_config();
3457 alloc_objects(object_entry_alloc);
3458 strbuf_init(&command_buf, 0);
3459 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3460 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3461 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3462 marks = pool_calloc(1, sizeof(struct mark_set));
3464 global_argc = argc;
3465 global_argv = argv;
3467 rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3468 for (i = 0; i < (cmd_save - 1); i++)
3469 rc_free[i].next = &rc_free[i + 1];
3470 rc_free[cmd_save - 1].next = NULL;
3472 prepare_packed_git();
3473 start_packfile();
3474 set_die_routine(die_nicely);
3475 set_checkpoint_signal();
3476 while (read_next_command() != EOF) {
3477 const char *v;
3478 if (!strcmp("blob", command_buf.buf))
3479 parse_new_blob();
3480 else if (skip_prefix(command_buf.buf, "ls ", &v))
3481 parse_ls(v, NULL);
3482 else if (skip_prefix(command_buf.buf, "commit ", &v))
3483 parse_new_commit(v);
3484 else if (skip_prefix(command_buf.buf, "tag ", &v))
3485 parse_new_tag(v);
3486 else if (skip_prefix(command_buf.buf, "reset ", &v))
3487 parse_reset_branch(v);
3488 else if (!strcmp("checkpoint", command_buf.buf))
3489 parse_checkpoint();
3490 else if (!strcmp("done", command_buf.buf))
3491 break;
3492 else if (starts_with(command_buf.buf, "progress "))
3493 parse_progress();
3494 else if (skip_prefix(command_buf.buf, "feature ", &v))
3495 parse_feature(v);
3496 else if (skip_prefix(command_buf.buf, "option git ", &v))
3497 parse_option(v);
3498 else if (starts_with(command_buf.buf, "option "))
3499 /* ignore non-git options*/;
3500 else
3501 die("Unsupported command: %s", command_buf.buf);
3503 if (checkpoint_requested)
3504 checkpoint();
3507 /* argv hasn't been parsed yet, do so */
3508 if (!seen_data_command)
3509 parse_argv();
3511 if (require_explicit_termination && feof(stdin))
3512 die("stream ends early");
3514 end_packfile();
3516 dump_branches();
3517 dump_tags();
3518 unkeep_all_packs();
3519 dump_marks();
3521 if (pack_edges)
3522 fclose(pack_edges);
3524 if (show_stats) {
3525 uintmax_t total_count = 0, duplicate_count = 0;
3526 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3527 total_count += object_count_by_type[i];
3528 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3529 duplicate_count += duplicate_count_by_type[i];
3531 fprintf(stderr, "%s statistics:\n", argv[0]);
3532 fprintf(stderr, "---------------------------------------------------------------------\n");
3533 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3534 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3535 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]);
3536 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]);
3537 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]);
3538 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]);
3539 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3540 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3541 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3542 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3543 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)(total_allocd/1024));
3544 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3545 fprintf(stderr, "---------------------------------------------------------------------\n");
3546 pack_report();
3547 fprintf(stderr, "---------------------------------------------------------------------\n");
3548 fprintf(stderr, "\n");
3551 return failure ? 1 : 0;