stash: use stash_push for no verb form
[git/raj.git] / fast-import.c
blobcb545d7df514b73a5ad9358b614ea9b4751e3c1e
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 "lockfile.h"
158 #include "object.h"
159 #include "blob.h"
160 #include "tree.h"
161 #include "commit.h"
162 #include "delta.h"
163 #include "pack.h"
164 #include "refs.h"
165 #include "csum-file.h"
166 #include "quote.h"
167 #include "dir.h"
168 #include "run-command.h"
170 #define PACK_ID_BITS 16
171 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
172 #define DEPTH_BITS 13
173 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
176 * We abuse the setuid bit on directories to mean "do not delta".
178 #define NO_DELTA S_ISUID
180 struct object_entry {
181 struct pack_idx_entry idx;
182 struct object_entry *next;
183 uint32_t type : TYPE_BITS,
184 pack_id : PACK_ID_BITS,
185 depth : DEPTH_BITS;
188 struct object_entry_pool {
189 struct object_entry_pool *next_pool;
190 struct object_entry *next_free;
191 struct object_entry *end;
192 struct object_entry entries[FLEX_ARRAY]; /* more */
195 struct mark_set {
196 union {
197 struct object_entry *marked[1024];
198 struct mark_set *sets[1024];
199 } data;
200 unsigned int shift;
203 struct last_object {
204 struct strbuf data;
205 off_t offset;
206 unsigned int depth;
207 unsigned no_swap : 1;
210 struct mem_pool {
211 struct mem_pool *next_pool;
212 char *next_free;
213 char *end;
214 uintmax_t space[FLEX_ARRAY]; /* more */
217 struct atom_str {
218 struct atom_str *next_atom;
219 unsigned short str_len;
220 char str_dat[FLEX_ARRAY]; /* more */
223 struct tree_content;
224 struct tree_entry {
225 struct tree_content *tree;
226 struct atom_str *name;
227 struct tree_entry_ms {
228 uint16_t mode;
229 unsigned char sha1[20];
230 } versions[2];
233 struct tree_content {
234 unsigned int entry_capacity; /* must match avail_tree_content */
235 unsigned int entry_count;
236 unsigned int delta_depth;
237 struct tree_entry *entries[FLEX_ARRAY]; /* more */
240 struct avail_tree_content {
241 unsigned int entry_capacity; /* must match tree_content */
242 struct avail_tree_content *next_avail;
245 struct branch {
246 struct branch *table_next_branch;
247 struct branch *active_next_branch;
248 const char *name;
249 struct tree_entry branch_tree;
250 uintmax_t last_commit;
251 uintmax_t num_notes;
252 unsigned active : 1;
253 unsigned delete : 1;
254 unsigned pack_id : PACK_ID_BITS;
255 unsigned char sha1[20];
258 struct tag {
259 struct tag *next_tag;
260 const char *name;
261 unsigned int pack_id;
262 unsigned char sha1[20];
265 struct hash_list {
266 struct hash_list *next;
267 unsigned char sha1[20];
270 typedef enum {
271 WHENSPEC_RAW = 1,
272 WHENSPEC_RFC2822,
273 WHENSPEC_NOW
274 } whenspec_type;
276 struct recent_command {
277 struct recent_command *prev;
278 struct recent_command *next;
279 char *buf;
282 /* Configured limits on output */
283 static unsigned long max_depth = 10;
284 static off_t max_packsize;
285 static int unpack_limit = 100;
286 static int force_update;
287 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
288 static int pack_compression_seen;
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_sha1(b->branch_tree.versions[1].sha1))
392 fputs(" dirty", rpt);
393 fputc('\n', rpt);
395 fprintf(rpt, " tip commit : %s\n", sha1_to_hex(b->sha1));
396 fprintf(rpt, " old tree : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
397 fprintf(rpt, " cur tree : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
398 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
400 fputs(" last pack : ", rpt);
401 if (b->pack_id < MAX_PACK_ID)
402 fprintf(rpt, "%u", b->pack_id);
403 fputc('\n', rpt);
405 fputc('\n', rpt);
408 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
410 static void write_crash_report(const char *err)
412 char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
413 FILE *rpt = fopen(loc, "w");
414 struct branch *b;
415 unsigned long lu;
416 struct recent_command *rc;
418 if (!rpt) {
419 error_errno("can't write crash report %s", loc);
420 free(loc);
421 return;
424 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
426 fprintf(rpt, "fast-import crash report:\n");
427 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
428 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
429 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
430 fputc('\n', rpt);
432 fputs("fatal: ", rpt);
433 fputs(err, rpt);
434 fputc('\n', rpt);
436 fputc('\n', rpt);
437 fputs("Most Recent Commands Before Crash\n", rpt);
438 fputs("---------------------------------\n", rpt);
439 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
440 if (rc->next == &cmd_hist)
441 fputs("* ", rpt);
442 else
443 fputs(" ", rpt);
444 fputs(rc->buf, rpt);
445 fputc('\n', rpt);
448 fputc('\n', rpt);
449 fputs("Active Branch LRU\n", rpt);
450 fputs("-----------------\n", rpt);
451 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
452 cur_active_branches,
453 max_active_branches);
454 fputc('\n', rpt);
455 fputs(" pos clock name\n", rpt);
456 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
457 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
458 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
459 ++lu, b->last_commit, b->name);
461 fputc('\n', rpt);
462 fputs("Inactive Branches\n", rpt);
463 fputs("-----------------\n", rpt);
464 for (lu = 0; lu < branch_table_sz; lu++) {
465 for (b = branch_table[lu]; b; b = b->table_next_branch)
466 write_branch_report(rpt, b);
469 if (first_tag) {
470 struct tag *tg;
471 fputc('\n', rpt);
472 fputs("Annotated Tags\n", rpt);
473 fputs("--------------\n", rpt);
474 for (tg = first_tag; tg; tg = tg->next_tag) {
475 fputs(sha1_to_hex(tg->sha1), rpt);
476 fputc(' ', rpt);
477 fputs(tg->name, rpt);
478 fputc('\n', rpt);
482 fputc('\n', rpt);
483 fputs("Marks\n", rpt);
484 fputs("-----\n", rpt);
485 if (export_marks_file)
486 fprintf(rpt, " exported to %s\n", export_marks_file);
487 else
488 dump_marks_helper(rpt, 0, marks);
490 fputc('\n', rpt);
491 fputs("-------------------\n", rpt);
492 fputs("END OF CRASH REPORT\n", rpt);
493 fclose(rpt);
494 free(loc);
497 static void end_packfile(void);
498 static void unkeep_all_packs(void);
499 static void dump_marks(void);
501 static NORETURN void die_nicely(const char *err, va_list params)
503 static int zombie;
504 char message[2 * PATH_MAX];
506 vsnprintf(message, sizeof(message), err, params);
507 fputs("fatal: ", stderr);
508 fputs(message, stderr);
509 fputc('\n', stderr);
511 if (!zombie) {
512 zombie = 1;
513 write_crash_report(message);
514 end_packfile();
515 unkeep_all_packs();
516 dump_marks();
518 exit(128);
521 #ifndef SIGUSR1 /* Windows, for example */
523 static void set_checkpoint_signal(void)
527 #else
529 static void checkpoint_signal(int signo)
531 checkpoint_requested = 1;
534 static void set_checkpoint_signal(void)
536 struct sigaction sa;
538 memset(&sa, 0, sizeof(sa));
539 sa.sa_handler = checkpoint_signal;
540 sigemptyset(&sa.sa_mask);
541 sa.sa_flags = SA_RESTART;
542 sigaction(SIGUSR1, &sa, NULL);
545 #endif
547 static void alloc_objects(unsigned int cnt)
549 struct object_entry_pool *b;
551 b = xmalloc(sizeof(struct object_entry_pool)
552 + cnt * sizeof(struct object_entry));
553 b->next_pool = blocks;
554 b->next_free = b->entries;
555 b->end = b->entries + cnt;
556 blocks = b;
557 alloc_count += cnt;
560 static struct object_entry *new_object(unsigned char *sha1)
562 struct object_entry *e;
564 if (blocks->next_free == blocks->end)
565 alloc_objects(object_entry_alloc);
567 e = blocks->next_free++;
568 hashcpy(e->idx.sha1, sha1);
569 return e;
572 static struct object_entry *find_object(unsigned char *sha1)
574 unsigned int h = sha1[0] << 8 | sha1[1];
575 struct object_entry *e;
576 for (e = object_table[h]; e; e = e->next)
577 if (!hashcmp(sha1, e->idx.sha1))
578 return e;
579 return NULL;
582 static struct object_entry *insert_object(unsigned char *sha1)
584 unsigned int h = sha1[0] << 8 | sha1[1];
585 struct object_entry *e = object_table[h];
587 while (e) {
588 if (!hashcmp(sha1, e->idx.sha1))
589 return e;
590 e = e->next;
593 e = new_object(sha1);
594 e->next = object_table[h];
595 e->idx.offset = 0;
596 object_table[h] = e;
597 return e;
600 static void invalidate_pack_id(unsigned int id)
602 unsigned int h;
603 unsigned long lu;
604 struct tag *t;
606 for (h = 0; h < ARRAY_SIZE(object_table); h++) {
607 struct object_entry *e;
609 for (e = object_table[h]; e; e = e->next)
610 if (e->pack_id == id)
611 e->pack_id = MAX_PACK_ID;
614 for (lu = 0; lu < branch_table_sz; lu++) {
615 struct branch *b;
617 for (b = branch_table[lu]; b; b = b->table_next_branch)
618 if (b->pack_id == id)
619 b->pack_id = MAX_PACK_ID;
622 for (t = first_tag; t; t = t->next_tag)
623 if (t->pack_id == id)
624 t->pack_id = MAX_PACK_ID;
627 static unsigned int hc_str(const char *s, size_t len)
629 unsigned int r = 0;
630 while (len-- > 0)
631 r = r * 31 + *s++;
632 return r;
635 static void *pool_alloc(size_t len)
637 struct mem_pool *p;
638 void *r;
640 /* round up to a 'uintmax_t' alignment */
641 if (len & (sizeof(uintmax_t) - 1))
642 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
644 for (p = mem_pool; p; p = p->next_pool)
645 if ((p->end - p->next_free >= len))
646 break;
648 if (!p) {
649 if (len >= (mem_pool_alloc/2)) {
650 total_allocd += len;
651 return xmalloc(len);
653 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
654 p = xmalloc(st_add(sizeof(struct mem_pool), mem_pool_alloc));
655 p->next_pool = mem_pool;
656 p->next_free = (char *) p->space;
657 p->end = p->next_free + mem_pool_alloc;
658 mem_pool = p;
661 r = p->next_free;
662 p->next_free += len;
663 return r;
666 static void *pool_calloc(size_t count, size_t size)
668 size_t len = count * size;
669 void *r = pool_alloc(len);
670 memset(r, 0, len);
671 return r;
674 static char *pool_strdup(const char *s)
676 size_t len = strlen(s) + 1;
677 char *r = pool_alloc(len);
678 memcpy(r, s, len);
679 return r;
682 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
684 struct mark_set *s = marks;
685 while ((idnum >> s->shift) >= 1024) {
686 s = pool_calloc(1, sizeof(struct mark_set));
687 s->shift = marks->shift + 10;
688 s->data.sets[0] = marks;
689 marks = s;
691 while (s->shift) {
692 uintmax_t i = idnum >> s->shift;
693 idnum -= i << s->shift;
694 if (!s->data.sets[i]) {
695 s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
696 s->data.sets[i]->shift = s->shift - 10;
698 s = s->data.sets[i];
700 if (!s->data.marked[idnum])
701 marks_set_count++;
702 s->data.marked[idnum] = oe;
705 static struct object_entry *find_mark(uintmax_t idnum)
707 uintmax_t orig_idnum = idnum;
708 struct mark_set *s = marks;
709 struct object_entry *oe = NULL;
710 if ((idnum >> s->shift) < 1024) {
711 while (s && s->shift) {
712 uintmax_t i = idnum >> s->shift;
713 idnum -= i << s->shift;
714 s = s->data.sets[i];
716 if (s)
717 oe = s->data.marked[idnum];
719 if (!oe)
720 die("mark :%" PRIuMAX " not declared", orig_idnum);
721 return oe;
724 static struct atom_str *to_atom(const char *s, unsigned short len)
726 unsigned int hc = hc_str(s, len) % atom_table_sz;
727 struct atom_str *c;
729 for (c = atom_table[hc]; c; c = c->next_atom)
730 if (c->str_len == len && !strncmp(s, c->str_dat, len))
731 return c;
733 c = pool_alloc(sizeof(struct atom_str) + len + 1);
734 c->str_len = len;
735 memcpy(c->str_dat, s, len);
736 c->str_dat[len] = 0;
737 c->next_atom = atom_table[hc];
738 atom_table[hc] = c;
739 atom_cnt++;
740 return c;
743 static struct branch *lookup_branch(const char *name)
745 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
746 struct branch *b;
748 for (b = branch_table[hc]; b; b = b->table_next_branch)
749 if (!strcmp(name, b->name))
750 return b;
751 return NULL;
754 static struct branch *new_branch(const char *name)
756 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
757 struct branch *b = lookup_branch(name);
759 if (b)
760 die("Invalid attempt to create duplicate branch: %s", name);
761 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
762 die("Branch name doesn't conform to GIT standards: %s", name);
764 b = pool_calloc(1, sizeof(struct branch));
765 b->name = pool_strdup(name);
766 b->table_next_branch = branch_table[hc];
767 b->branch_tree.versions[0].mode = S_IFDIR;
768 b->branch_tree.versions[1].mode = S_IFDIR;
769 b->num_notes = 0;
770 b->active = 0;
771 b->pack_id = MAX_PACK_ID;
772 branch_table[hc] = b;
773 branch_count++;
774 return b;
777 static unsigned int hc_entries(unsigned int cnt)
779 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
780 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
783 static struct tree_content *new_tree_content(unsigned int cnt)
785 struct avail_tree_content *f, *l = NULL;
786 struct tree_content *t;
787 unsigned int hc = hc_entries(cnt);
789 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
790 if (f->entry_capacity >= cnt)
791 break;
793 if (f) {
794 if (l)
795 l->next_avail = f->next_avail;
796 else
797 avail_tree_table[hc] = f->next_avail;
798 } else {
799 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
800 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
801 f->entry_capacity = cnt;
804 t = (struct tree_content*)f;
805 t->entry_count = 0;
806 t->delta_depth = 0;
807 return t;
810 static void release_tree_entry(struct tree_entry *e);
811 static void release_tree_content(struct tree_content *t)
813 struct avail_tree_content *f = (struct avail_tree_content*)t;
814 unsigned int hc = hc_entries(f->entry_capacity);
815 f->next_avail = avail_tree_table[hc];
816 avail_tree_table[hc] = f;
819 static void release_tree_content_recursive(struct tree_content *t)
821 unsigned int i;
822 for (i = 0; i < t->entry_count; i++)
823 release_tree_entry(t->entries[i]);
824 release_tree_content(t);
827 static struct tree_content *grow_tree_content(
828 struct tree_content *t,
829 int amt)
831 struct tree_content *r = new_tree_content(t->entry_count + amt);
832 r->entry_count = t->entry_count;
833 r->delta_depth = t->delta_depth;
834 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
835 release_tree_content(t);
836 return r;
839 static struct tree_entry *new_tree_entry(void)
841 struct tree_entry *e;
843 if (!avail_tree_entry) {
844 unsigned int n = tree_entry_alloc;
845 total_allocd += n * sizeof(struct tree_entry);
846 ALLOC_ARRAY(e, n);
847 avail_tree_entry = e;
848 while (n-- > 1) {
849 *((void**)e) = e + 1;
850 e++;
852 *((void**)e) = NULL;
855 e = avail_tree_entry;
856 avail_tree_entry = *((void**)e);
857 return e;
860 static void release_tree_entry(struct tree_entry *e)
862 if (e->tree)
863 release_tree_content_recursive(e->tree);
864 *((void**)e) = avail_tree_entry;
865 avail_tree_entry = e;
868 static struct tree_content *dup_tree_content(struct tree_content *s)
870 struct tree_content *d;
871 struct tree_entry *a, *b;
872 unsigned int i;
874 if (!s)
875 return NULL;
876 d = new_tree_content(s->entry_count);
877 for (i = 0; i < s->entry_count; i++) {
878 a = s->entries[i];
879 b = new_tree_entry();
880 memcpy(b, a, sizeof(*a));
881 if (a->tree && is_null_sha1(b->versions[1].sha1))
882 b->tree = dup_tree_content(a->tree);
883 else
884 b->tree = NULL;
885 d->entries[i] = b;
887 d->entry_count = s->entry_count;
888 d->delta_depth = s->delta_depth;
890 return d;
893 static void start_packfile(void)
895 static char tmp_file[PATH_MAX];
896 struct packed_git *p;
897 struct pack_header hdr;
898 int pack_fd;
900 pack_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
901 "pack/tmp_pack_XXXXXX");
902 FLEX_ALLOC_STR(p, pack_name, tmp_file);
903 p->pack_fd = pack_fd;
904 p->do_not_close = 1;
905 pack_file = sha1fd(pack_fd, p->pack_name);
907 hdr.hdr_signature = htonl(PACK_SIGNATURE);
908 hdr.hdr_version = htonl(2);
909 hdr.hdr_entries = 0;
910 sha1write(pack_file, &hdr, sizeof(hdr));
912 pack_data = p;
913 pack_size = sizeof(hdr);
914 object_count = 0;
916 REALLOC_ARRAY(all_packs, pack_id + 1);
917 all_packs[pack_id] = p;
920 static const char *create_index(void)
922 const char *tmpfile;
923 struct pack_idx_entry **idx, **c, **last;
924 struct object_entry *e;
925 struct object_entry_pool *o;
927 /* Build the table of object IDs. */
928 ALLOC_ARRAY(idx, object_count);
929 c = idx;
930 for (o = blocks; o; o = o->next_pool)
931 for (e = o->next_free; e-- != o->entries;)
932 if (pack_id == e->pack_id)
933 *c++ = &e->idx;
934 last = idx + object_count;
935 if (c != last)
936 die("internal consistency error creating the index");
938 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
939 free(idx);
940 return tmpfile;
943 static char *keep_pack(const char *curr_index_name)
945 static char name[PATH_MAX];
946 static const char *keep_msg = "fast-import";
947 int keep_fd;
949 keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
950 if (keep_fd < 0)
951 die_errno("cannot create keep file");
952 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
953 if (close(keep_fd))
954 die_errno("failed to write keep file");
956 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
957 get_object_directory(), sha1_to_hex(pack_data->sha1));
958 if (finalize_object_file(pack_data->pack_name, name))
959 die("cannot store pack file");
961 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
962 get_object_directory(), sha1_to_hex(pack_data->sha1));
963 if (finalize_object_file(curr_index_name, name))
964 die("cannot store index file");
965 free((void *)curr_index_name);
966 return name;
969 static void unkeep_all_packs(void)
971 static char name[PATH_MAX];
972 int k;
974 for (k = 0; k < pack_id; k++) {
975 struct packed_git *p = all_packs[k];
976 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
977 get_object_directory(), sha1_to_hex(p->sha1));
978 unlink_or_warn(name);
982 static int loosen_small_pack(const struct packed_git *p)
984 struct child_process unpack = CHILD_PROCESS_INIT;
986 if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
987 die_errno("Failed seeking to start of '%s'", p->pack_name);
989 unpack.in = p->pack_fd;
990 unpack.git_cmd = 1;
991 unpack.stdout_to_stderr = 1;
992 argv_array_push(&unpack.args, "unpack-objects");
993 if (!show_stats)
994 argv_array_push(&unpack.args, "-q");
996 return run_command(&unpack);
999 static void end_packfile(void)
1001 static int running;
1003 if (running || !pack_data)
1004 return;
1006 running = 1;
1007 clear_delta_base_cache();
1008 if (object_count) {
1009 struct packed_git *new_p;
1010 unsigned char cur_pack_sha1[20];
1011 char *idx_name;
1012 int i;
1013 struct branch *b;
1014 struct tag *t;
1016 close_pack_windows(pack_data);
1017 sha1close(pack_file, cur_pack_sha1, 0);
1018 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
1019 pack_data->pack_name, object_count,
1020 cur_pack_sha1, pack_size);
1022 if (object_count <= unpack_limit) {
1023 if (!loosen_small_pack(pack_data)) {
1024 invalidate_pack_id(pack_id);
1025 goto discard_pack;
1029 close(pack_data->pack_fd);
1030 idx_name = keep_pack(create_index());
1032 /* Register the packfile with core git's machinery. */
1033 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
1034 if (!new_p)
1035 die("core git rejected index %s", idx_name);
1036 all_packs[pack_id] = new_p;
1037 install_packed_git(new_p);
1039 /* Print the boundary */
1040 if (pack_edges) {
1041 fprintf(pack_edges, "%s:", new_p->pack_name);
1042 for (i = 0; i < branch_table_sz; i++) {
1043 for (b = branch_table[i]; b; b = b->table_next_branch) {
1044 if (b->pack_id == pack_id)
1045 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
1048 for (t = first_tag; t; t = t->next_tag) {
1049 if (t->pack_id == pack_id)
1050 fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
1052 fputc('\n', pack_edges);
1053 fflush(pack_edges);
1056 pack_id++;
1058 else {
1059 discard_pack:
1060 close(pack_data->pack_fd);
1061 unlink_or_warn(pack_data->pack_name);
1063 free(pack_data);
1064 pack_data = NULL;
1065 running = 0;
1067 /* We can't carry a delta across packfiles. */
1068 strbuf_release(&last_blob.data);
1069 last_blob.offset = 0;
1070 last_blob.depth = 0;
1073 static void cycle_packfile(void)
1075 end_packfile();
1076 start_packfile();
1079 static int store_object(
1080 enum object_type type,
1081 struct strbuf *dat,
1082 struct last_object *last,
1083 unsigned char *sha1out,
1084 uintmax_t mark)
1086 void *out, *delta;
1087 struct object_entry *e;
1088 unsigned char hdr[96];
1089 unsigned char sha1[20];
1090 unsigned long hdrlen, deltalen;
1091 git_SHA_CTX c;
1092 git_zstream s;
1094 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1095 typename(type), (unsigned long)dat->len) + 1;
1096 git_SHA1_Init(&c);
1097 git_SHA1_Update(&c, hdr, hdrlen);
1098 git_SHA1_Update(&c, dat->buf, dat->len);
1099 git_SHA1_Final(sha1, &c);
1100 if (sha1out)
1101 hashcpy(sha1out, sha1);
1103 e = insert_object(sha1);
1104 if (mark)
1105 insert_mark(mark, e);
1106 if (e->idx.offset) {
1107 duplicate_count_by_type[type]++;
1108 return 1;
1109 } else if (find_sha1_pack(sha1, packed_git)) {
1110 e->type = type;
1111 e->pack_id = MAX_PACK_ID;
1112 e->idx.offset = 1; /* just not zero! */
1113 duplicate_count_by_type[type]++;
1114 return 1;
1117 if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1118 delta_count_attempts_by_type[type]++;
1119 delta = diff_delta(last->data.buf, last->data.len,
1120 dat->buf, dat->len,
1121 &deltalen, dat->len - 20);
1122 } else
1123 delta = NULL;
1125 git_deflate_init(&s, pack_compression_level);
1126 if (delta) {
1127 s.next_in = delta;
1128 s.avail_in = deltalen;
1129 } else {
1130 s.next_in = (void *)dat->buf;
1131 s.avail_in = dat->len;
1133 s.avail_out = git_deflate_bound(&s, s.avail_in);
1134 s.next_out = out = xmalloc(s.avail_out);
1135 while (git_deflate(&s, Z_FINISH) == Z_OK)
1136 ; /* nothing */
1137 git_deflate_end(&s);
1139 /* Determine if we should auto-checkpoint. */
1140 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1141 || (pack_size + 60 + s.total_out) < pack_size) {
1143 /* This new object needs to *not* have the current pack_id. */
1144 e->pack_id = pack_id + 1;
1145 cycle_packfile();
1147 /* We cannot carry a delta into the new pack. */
1148 if (delta) {
1149 free(delta);
1150 delta = NULL;
1152 git_deflate_init(&s, pack_compression_level);
1153 s.next_in = (void *)dat->buf;
1154 s.avail_in = dat->len;
1155 s.avail_out = git_deflate_bound(&s, s.avail_in);
1156 s.next_out = out = xrealloc(out, s.avail_out);
1157 while (git_deflate(&s, Z_FINISH) == Z_OK)
1158 ; /* nothing */
1159 git_deflate_end(&s);
1163 e->type = type;
1164 e->pack_id = pack_id;
1165 e->idx.offset = pack_size;
1166 object_count++;
1167 object_count_by_type[type]++;
1169 crc32_begin(pack_file);
1171 if (delta) {
1172 off_t ofs = e->idx.offset - last->offset;
1173 unsigned pos = sizeof(hdr) - 1;
1175 delta_count_by_type[type]++;
1176 e->depth = last->depth + 1;
1178 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1179 sha1write(pack_file, hdr, hdrlen);
1180 pack_size += hdrlen;
1182 hdr[pos] = ofs & 127;
1183 while (ofs >>= 7)
1184 hdr[--pos] = 128 | (--ofs & 127);
1185 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1186 pack_size += sizeof(hdr) - pos;
1187 } else {
1188 e->depth = 0;
1189 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1190 sha1write(pack_file, hdr, hdrlen);
1191 pack_size += hdrlen;
1194 sha1write(pack_file, out, s.total_out);
1195 pack_size += s.total_out;
1197 e->idx.crc32 = crc32_end(pack_file);
1199 free(out);
1200 free(delta);
1201 if (last) {
1202 if (last->no_swap) {
1203 last->data = *dat;
1204 } else {
1205 strbuf_swap(&last->data, dat);
1207 last->offset = e->idx.offset;
1208 last->depth = e->depth;
1210 return 0;
1213 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1215 if (sha1file_truncate(pack_file, checkpoint))
1216 die_errno("cannot truncate pack to skip duplicate");
1217 pack_size = checkpoint->offset;
1220 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1222 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1223 unsigned char *in_buf = xmalloc(in_sz);
1224 unsigned char *out_buf = xmalloc(out_sz);
1225 struct object_entry *e;
1226 unsigned char sha1[20];
1227 unsigned long hdrlen;
1228 off_t offset;
1229 git_SHA_CTX c;
1230 git_zstream s;
1231 struct sha1file_checkpoint checkpoint;
1232 int status = Z_OK;
1234 /* Determine if we should auto-checkpoint. */
1235 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1236 || (pack_size + 60 + len) < pack_size)
1237 cycle_packfile();
1239 sha1file_checkpoint(pack_file, &checkpoint);
1240 offset = checkpoint.offset;
1242 hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1243 if (out_sz <= hdrlen)
1244 die("impossibly large object header");
1246 git_SHA1_Init(&c);
1247 git_SHA1_Update(&c, out_buf, hdrlen);
1249 crc32_begin(pack_file);
1251 git_deflate_init(&s, pack_compression_level);
1253 hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1254 if (out_sz <= hdrlen)
1255 die("impossibly large object header");
1257 s.next_out = out_buf + hdrlen;
1258 s.avail_out = out_sz - hdrlen;
1260 while (status != Z_STREAM_END) {
1261 if (0 < len && !s.avail_in) {
1262 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1263 size_t n = fread(in_buf, 1, cnt, stdin);
1264 if (!n && feof(stdin))
1265 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1267 git_SHA1_Update(&c, in_buf, n);
1268 s.next_in = in_buf;
1269 s.avail_in = n;
1270 len -= n;
1273 status = git_deflate(&s, len ? 0 : Z_FINISH);
1275 if (!s.avail_out || status == Z_STREAM_END) {
1276 size_t n = s.next_out - out_buf;
1277 sha1write(pack_file, out_buf, n);
1278 pack_size += n;
1279 s.next_out = out_buf;
1280 s.avail_out = out_sz;
1283 switch (status) {
1284 case Z_OK:
1285 case Z_BUF_ERROR:
1286 case Z_STREAM_END:
1287 continue;
1288 default:
1289 die("unexpected deflate failure: %d", status);
1292 git_deflate_end(&s);
1293 git_SHA1_Final(sha1, &c);
1295 if (sha1out)
1296 hashcpy(sha1out, sha1);
1298 e = insert_object(sha1);
1300 if (mark)
1301 insert_mark(mark, e);
1303 if (e->idx.offset) {
1304 duplicate_count_by_type[OBJ_BLOB]++;
1305 truncate_pack(&checkpoint);
1307 } else if (find_sha1_pack(sha1, packed_git)) {
1308 e->type = OBJ_BLOB;
1309 e->pack_id = MAX_PACK_ID;
1310 e->idx.offset = 1; /* just not zero! */
1311 duplicate_count_by_type[OBJ_BLOB]++;
1312 truncate_pack(&checkpoint);
1314 } else {
1315 e->depth = 0;
1316 e->type = OBJ_BLOB;
1317 e->pack_id = pack_id;
1318 e->idx.offset = offset;
1319 e->idx.crc32 = crc32_end(pack_file);
1320 object_count++;
1321 object_count_by_type[OBJ_BLOB]++;
1324 free(in_buf);
1325 free(out_buf);
1328 /* All calls must be guarded by find_object() or find_mark() to
1329 * ensure the 'struct object_entry' passed was written by this
1330 * process instance. We unpack the entry by the offset, avoiding
1331 * the need for the corresponding .idx file. This unpacking rule
1332 * works because we only use OBJ_REF_DELTA within the packfiles
1333 * created by fast-import.
1335 * oe must not be NULL. Such an oe usually comes from giving
1336 * an unknown SHA-1 to find_object() or an undefined mark to
1337 * find_mark(). Callers must test for this condition and use
1338 * the standard read_sha1_file() when it happens.
1340 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1341 * find_mark(), where the mark was reloaded from an existing marks
1342 * file and is referencing an object that this fast-import process
1343 * instance did not write out to a packfile. Callers must test for
1344 * this condition and use read_sha1_file() instead.
1346 static void *gfi_unpack_entry(
1347 struct object_entry *oe,
1348 unsigned long *sizep)
1350 enum object_type type;
1351 struct packed_git *p = all_packs[oe->pack_id];
1352 if (p == pack_data && p->pack_size < (pack_size + 20)) {
1353 /* The object is stored in the packfile we are writing to
1354 * and we have modified it since the last time we scanned
1355 * back to read a previously written object. If an old
1356 * window covered [p->pack_size, p->pack_size + 20) its
1357 * data is stale and is not valid. Closing all windows
1358 * and updating the packfile length ensures we can read
1359 * the newly written data.
1361 close_pack_windows(p);
1362 sha1flush(pack_file);
1364 /* We have to offer 20 bytes additional on the end of
1365 * the packfile as the core unpacker code assumes the
1366 * footer is present at the file end and must promise
1367 * at least 20 bytes within any window it maps. But
1368 * we don't actually create the footer here.
1370 p->pack_size = pack_size + 20;
1372 return unpack_entry(p, oe->idx.offset, &type, sizep);
1375 static const char *get_mode(const char *str, uint16_t *modep)
1377 unsigned char c;
1378 uint16_t mode = 0;
1380 while ((c = *str++) != ' ') {
1381 if (c < '0' || c > '7')
1382 return NULL;
1383 mode = (mode << 3) + (c - '0');
1385 *modep = mode;
1386 return str;
1389 static void load_tree(struct tree_entry *root)
1391 unsigned char *sha1 = root->versions[1].sha1;
1392 struct object_entry *myoe;
1393 struct tree_content *t;
1394 unsigned long size;
1395 char *buf;
1396 const char *c;
1398 root->tree = t = new_tree_content(8);
1399 if (is_null_sha1(sha1))
1400 return;
1402 myoe = find_object(sha1);
1403 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1404 if (myoe->type != OBJ_TREE)
1405 die("Not a tree: %s", sha1_to_hex(sha1));
1406 t->delta_depth = myoe->depth;
1407 buf = gfi_unpack_entry(myoe, &size);
1408 if (!buf)
1409 die("Can't load tree %s", sha1_to_hex(sha1));
1410 } else {
1411 enum object_type type;
1412 buf = read_sha1_file(sha1, &type, &size);
1413 if (!buf || type != OBJ_TREE)
1414 die("Can't load tree %s", sha1_to_hex(sha1));
1417 c = buf;
1418 while (c != (buf + size)) {
1419 struct tree_entry *e = new_tree_entry();
1421 if (t->entry_count == t->entry_capacity)
1422 root->tree = t = grow_tree_content(t, t->entry_count);
1423 t->entries[t->entry_count++] = e;
1425 e->tree = NULL;
1426 c = get_mode(c, &e->versions[1].mode);
1427 if (!c)
1428 die("Corrupt mode in %s", sha1_to_hex(sha1));
1429 e->versions[0].mode = e->versions[1].mode;
1430 e->name = to_atom(c, strlen(c));
1431 c += e->name->str_len + 1;
1432 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1433 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1434 c += 20;
1436 free(buf);
1439 static int tecmp0 (const void *_a, const void *_b)
1441 struct tree_entry *a = *((struct tree_entry**)_a);
1442 struct tree_entry *b = *((struct tree_entry**)_b);
1443 return base_name_compare(
1444 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1445 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1448 static int tecmp1 (const void *_a, const void *_b)
1450 struct tree_entry *a = *((struct tree_entry**)_a);
1451 struct tree_entry *b = *((struct tree_entry**)_b);
1452 return base_name_compare(
1453 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1454 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1457 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1459 size_t maxlen = 0;
1460 unsigned int i;
1462 if (!v)
1463 QSORT(t->entries, t->entry_count, tecmp0);
1464 else
1465 QSORT(t->entries, t->entry_count, tecmp1);
1467 for (i = 0; i < t->entry_count; i++) {
1468 if (t->entries[i]->versions[v].mode)
1469 maxlen += t->entries[i]->name->str_len + 34;
1472 strbuf_reset(b);
1473 strbuf_grow(b, maxlen);
1474 for (i = 0; i < t->entry_count; i++) {
1475 struct tree_entry *e = t->entries[i];
1476 if (!e->versions[v].mode)
1477 continue;
1478 strbuf_addf(b, "%o %s%c",
1479 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1480 e->name->str_dat, '\0');
1481 strbuf_add(b, e->versions[v].sha1, 20);
1485 static void store_tree(struct tree_entry *root)
1487 struct tree_content *t;
1488 unsigned int i, j, del;
1489 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1490 struct object_entry *le = NULL;
1492 if (!is_null_sha1(root->versions[1].sha1))
1493 return;
1495 if (!root->tree)
1496 load_tree(root);
1497 t = root->tree;
1499 for (i = 0; i < t->entry_count; i++) {
1500 if (t->entries[i]->tree)
1501 store_tree(t->entries[i]);
1504 if (!(root->versions[0].mode & NO_DELTA))
1505 le = find_object(root->versions[0].sha1);
1506 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1507 mktree(t, 0, &old_tree);
1508 lo.data = old_tree;
1509 lo.offset = le->idx.offset;
1510 lo.depth = t->delta_depth;
1513 mktree(t, 1, &new_tree);
1514 store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1516 t->delta_depth = lo.depth;
1517 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1518 struct tree_entry *e = t->entries[i];
1519 if (e->versions[1].mode) {
1520 e->versions[0].mode = e->versions[1].mode;
1521 hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1522 t->entries[j++] = e;
1523 } else {
1524 release_tree_entry(e);
1525 del++;
1528 t->entry_count -= del;
1531 static void tree_content_replace(
1532 struct tree_entry *root,
1533 const unsigned char *sha1,
1534 const uint16_t mode,
1535 struct tree_content *newtree)
1537 if (!S_ISDIR(mode))
1538 die("Root cannot be a non-directory");
1539 hashclr(root->versions[0].sha1);
1540 hashcpy(root->versions[1].sha1, sha1);
1541 if (root->tree)
1542 release_tree_content_recursive(root->tree);
1543 root->tree = newtree;
1546 static int tree_content_set(
1547 struct tree_entry *root,
1548 const char *p,
1549 const unsigned char *sha1,
1550 const uint16_t mode,
1551 struct tree_content *subtree)
1553 struct tree_content *t;
1554 const char *slash1;
1555 unsigned int i, n;
1556 struct tree_entry *e;
1558 slash1 = strchrnul(p, '/');
1559 n = slash1 - p;
1560 if (!n)
1561 die("Empty path component found in input");
1562 if (!*slash1 && !S_ISDIR(mode) && subtree)
1563 die("Non-directories cannot have subtrees");
1565 if (!root->tree)
1566 load_tree(root);
1567 t = root->tree;
1568 for (i = 0; i < t->entry_count; i++) {
1569 e = t->entries[i];
1570 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1571 if (!*slash1) {
1572 if (!S_ISDIR(mode)
1573 && e->versions[1].mode == mode
1574 && !hashcmp(e->versions[1].sha1, sha1))
1575 return 0;
1576 e->versions[1].mode = mode;
1577 hashcpy(e->versions[1].sha1, sha1);
1578 if (e->tree)
1579 release_tree_content_recursive(e->tree);
1580 e->tree = subtree;
1583 * We need to leave e->versions[0].sha1 alone
1584 * to avoid modifying the preimage tree used
1585 * when writing out the parent directory.
1586 * But after replacing the subdir with a
1587 * completely different one, it's not a good
1588 * delta base any more, and besides, we've
1589 * thrown away the tree entries needed to
1590 * make a delta against it.
1592 * So let's just explicitly disable deltas
1593 * for the subtree.
1595 if (S_ISDIR(e->versions[0].mode))
1596 e->versions[0].mode |= NO_DELTA;
1598 hashclr(root->versions[1].sha1);
1599 return 1;
1601 if (!S_ISDIR(e->versions[1].mode)) {
1602 e->tree = new_tree_content(8);
1603 e->versions[1].mode = S_IFDIR;
1605 if (!e->tree)
1606 load_tree(e);
1607 if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1608 hashclr(root->versions[1].sha1);
1609 return 1;
1611 return 0;
1615 if (t->entry_count == t->entry_capacity)
1616 root->tree = t = grow_tree_content(t, t->entry_count);
1617 e = new_tree_entry();
1618 e->name = to_atom(p, n);
1619 e->versions[0].mode = 0;
1620 hashclr(e->versions[0].sha1);
1621 t->entries[t->entry_count++] = e;
1622 if (*slash1) {
1623 e->tree = new_tree_content(8);
1624 e->versions[1].mode = S_IFDIR;
1625 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1626 } else {
1627 e->tree = subtree;
1628 e->versions[1].mode = mode;
1629 hashcpy(e->versions[1].sha1, sha1);
1631 hashclr(root->versions[1].sha1);
1632 return 1;
1635 static int tree_content_remove(
1636 struct tree_entry *root,
1637 const char *p,
1638 struct tree_entry *backup_leaf,
1639 int allow_root)
1641 struct tree_content *t;
1642 const char *slash1;
1643 unsigned int i, n;
1644 struct tree_entry *e;
1646 slash1 = strchrnul(p, '/');
1647 n = slash1 - p;
1649 if (!root->tree)
1650 load_tree(root);
1652 if (!*p && allow_root) {
1653 e = root;
1654 goto del_entry;
1657 t = root->tree;
1658 for (i = 0; i < t->entry_count; i++) {
1659 e = t->entries[i];
1660 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1661 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1663 * If p names a file in some subdirectory, and a
1664 * file or symlink matching the name of the
1665 * parent directory of p exists, then p cannot
1666 * exist and need not be deleted.
1668 return 1;
1669 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1670 goto del_entry;
1671 if (!e->tree)
1672 load_tree(e);
1673 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1674 for (n = 0; n < e->tree->entry_count; n++) {
1675 if (e->tree->entries[n]->versions[1].mode) {
1676 hashclr(root->versions[1].sha1);
1677 return 1;
1680 backup_leaf = NULL;
1681 goto del_entry;
1683 return 0;
1686 return 0;
1688 del_entry:
1689 if (backup_leaf)
1690 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1691 else if (e->tree)
1692 release_tree_content_recursive(e->tree);
1693 e->tree = NULL;
1694 e->versions[1].mode = 0;
1695 hashclr(e->versions[1].sha1);
1696 hashclr(root->versions[1].sha1);
1697 return 1;
1700 static int tree_content_get(
1701 struct tree_entry *root,
1702 const char *p,
1703 struct tree_entry *leaf,
1704 int allow_root)
1706 struct tree_content *t;
1707 const char *slash1;
1708 unsigned int i, n;
1709 struct tree_entry *e;
1711 slash1 = strchrnul(p, '/');
1712 n = slash1 - p;
1713 if (!n && !allow_root)
1714 die("Empty path component found in input");
1716 if (!root->tree)
1717 load_tree(root);
1719 if (!n) {
1720 e = root;
1721 goto found_entry;
1724 t = root->tree;
1725 for (i = 0; i < t->entry_count; i++) {
1726 e = t->entries[i];
1727 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1728 if (!*slash1)
1729 goto found_entry;
1730 if (!S_ISDIR(e->versions[1].mode))
1731 return 0;
1732 if (!e->tree)
1733 load_tree(e);
1734 return tree_content_get(e, slash1 + 1, leaf, 0);
1737 return 0;
1739 found_entry:
1740 memcpy(leaf, e, sizeof(*leaf));
1741 if (e->tree && is_null_sha1(e->versions[1].sha1))
1742 leaf->tree = dup_tree_content(e->tree);
1743 else
1744 leaf->tree = NULL;
1745 return 1;
1748 static int update_branch(struct branch *b)
1750 static const char *msg = "fast-import";
1751 struct ref_transaction *transaction;
1752 unsigned char old_sha1[20];
1753 struct strbuf err = STRBUF_INIT;
1755 if (is_null_sha1(b->sha1)) {
1756 if (b->delete)
1757 delete_ref(b->name, NULL, 0);
1758 return 0;
1760 if (read_ref(b->name, old_sha1))
1761 hashclr(old_sha1);
1762 if (!force_update && !is_null_sha1(old_sha1)) {
1763 struct commit *old_cmit, *new_cmit;
1765 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1766 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1767 if (!old_cmit || !new_cmit)
1768 return error("Branch %s is missing commits.", b->name);
1770 if (!in_merge_bases(old_cmit, new_cmit)) {
1771 warning("Not updating %s"
1772 " (new tip %s does not contain %s)",
1773 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1774 return -1;
1777 transaction = ref_transaction_begin(&err);
1778 if (!transaction ||
1779 ref_transaction_update(transaction, b->name, b->sha1, old_sha1,
1780 0, msg, &err) ||
1781 ref_transaction_commit(transaction, &err)) {
1782 ref_transaction_free(transaction);
1783 error("%s", err.buf);
1784 strbuf_release(&err);
1785 return -1;
1787 ref_transaction_free(transaction);
1788 strbuf_release(&err);
1789 return 0;
1792 static void dump_branches(void)
1794 unsigned int i;
1795 struct branch *b;
1797 for (i = 0; i < branch_table_sz; i++) {
1798 for (b = branch_table[i]; b; b = b->table_next_branch)
1799 failure |= update_branch(b);
1803 static void dump_tags(void)
1805 static const char *msg = "fast-import";
1806 struct tag *t;
1807 struct strbuf ref_name = STRBUF_INIT;
1808 struct strbuf err = STRBUF_INIT;
1809 struct ref_transaction *transaction;
1811 transaction = ref_transaction_begin(&err);
1812 if (!transaction) {
1813 failure |= error("%s", err.buf);
1814 goto cleanup;
1816 for (t = first_tag; t; t = t->next_tag) {
1817 strbuf_reset(&ref_name);
1818 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1820 if (ref_transaction_update(transaction, ref_name.buf,
1821 t->sha1, NULL, 0, msg, &err)) {
1822 failure |= error("%s", err.buf);
1823 goto cleanup;
1826 if (ref_transaction_commit(transaction, &err))
1827 failure |= error("%s", err.buf);
1829 cleanup:
1830 ref_transaction_free(transaction);
1831 strbuf_release(&ref_name);
1832 strbuf_release(&err);
1835 static void dump_marks_helper(FILE *f,
1836 uintmax_t base,
1837 struct mark_set *m)
1839 uintmax_t k;
1840 if (m->shift) {
1841 for (k = 0; k < 1024; k++) {
1842 if (m->data.sets[k])
1843 dump_marks_helper(f, base + (k << m->shift),
1844 m->data.sets[k]);
1846 } else {
1847 for (k = 0; k < 1024; k++) {
1848 if (m->data.marked[k])
1849 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1850 sha1_to_hex(m->data.marked[k]->idx.sha1));
1855 static void dump_marks(void)
1857 static struct lock_file mark_lock;
1858 FILE *f;
1860 if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1861 return;
1863 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1864 failure |= error_errno("Unable to write marks file %s",
1865 export_marks_file);
1866 return;
1869 f = fdopen_lock_file(&mark_lock, "w");
1870 if (!f) {
1871 int saved_errno = errno;
1872 rollback_lock_file(&mark_lock);
1873 failure |= error("Unable to write marks file %s: %s",
1874 export_marks_file, strerror(saved_errno));
1875 return;
1878 dump_marks_helper(f, 0, marks);
1879 if (commit_lock_file(&mark_lock)) {
1880 failure |= error_errno("Unable to write file %s",
1881 export_marks_file);
1882 return;
1886 static void read_marks(void)
1888 char line[512];
1889 FILE *f = fopen(import_marks_file, "r");
1890 if (f)
1892 else if (import_marks_file_ignore_missing && errno == ENOENT)
1893 goto done; /* Marks file does not exist */
1894 else
1895 die_errno("cannot read '%s'", import_marks_file);
1896 while (fgets(line, sizeof(line), f)) {
1897 uintmax_t mark;
1898 char *end;
1899 unsigned char sha1[20];
1900 struct object_entry *e;
1902 end = strchr(line, '\n');
1903 if (line[0] != ':' || !end)
1904 die("corrupt mark line: %s", line);
1905 *end = 0;
1906 mark = strtoumax(line + 1, &end, 10);
1907 if (!mark || end == line + 1
1908 || *end != ' ' || get_sha1_hex(end + 1, sha1))
1909 die("corrupt mark line: %s", line);
1910 e = find_object(sha1);
1911 if (!e) {
1912 enum object_type type = sha1_object_info(sha1, NULL);
1913 if (type < 0)
1914 die("object not found: %s", sha1_to_hex(sha1));
1915 e = insert_object(sha1);
1916 e->type = type;
1917 e->pack_id = MAX_PACK_ID;
1918 e->idx.offset = 1; /* just not zero! */
1920 insert_mark(mark, e);
1922 fclose(f);
1923 done:
1924 import_marks_file_done = 1;
1928 static int read_next_command(void)
1930 static int stdin_eof = 0;
1932 if (stdin_eof) {
1933 unread_command_buf = 0;
1934 return EOF;
1937 for (;;) {
1938 const char *p;
1940 if (unread_command_buf) {
1941 unread_command_buf = 0;
1942 } else {
1943 struct recent_command *rc;
1945 strbuf_detach(&command_buf, NULL);
1946 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1947 if (stdin_eof)
1948 return EOF;
1950 if (!seen_data_command
1951 && !starts_with(command_buf.buf, "feature ")
1952 && !starts_with(command_buf.buf, "option ")) {
1953 parse_argv();
1956 rc = rc_free;
1957 if (rc)
1958 rc_free = rc->next;
1959 else {
1960 rc = cmd_hist.next;
1961 cmd_hist.next = rc->next;
1962 cmd_hist.next->prev = &cmd_hist;
1963 free(rc->buf);
1966 rc->buf = command_buf.buf;
1967 rc->prev = cmd_tail;
1968 rc->next = cmd_hist.prev;
1969 rc->prev->next = rc;
1970 cmd_tail = rc;
1972 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1973 parse_get_mark(p);
1974 continue;
1976 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1977 parse_cat_blob(p);
1978 continue;
1980 if (command_buf.buf[0] == '#')
1981 continue;
1982 return 0;
1986 static void skip_optional_lf(void)
1988 int term_char = fgetc(stdin);
1989 if (term_char != '\n' && term_char != EOF)
1990 ungetc(term_char, stdin);
1993 static void parse_mark(void)
1995 const char *v;
1996 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1997 next_mark = strtoumax(v, NULL, 10);
1998 read_next_command();
2000 else
2001 next_mark = 0;
2004 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
2006 const char *data;
2007 strbuf_reset(sb);
2009 if (!skip_prefix(command_buf.buf, "data ", &data))
2010 die("Expected 'data n' command, found: %s", command_buf.buf);
2012 if (skip_prefix(data, "<<", &data)) {
2013 char *term = xstrdup(data);
2014 size_t term_len = command_buf.len - (data - command_buf.buf);
2016 strbuf_detach(&command_buf, NULL);
2017 for (;;) {
2018 if (strbuf_getline_lf(&command_buf, stdin) == EOF)
2019 die("EOF in data (terminator '%s' not found)", term);
2020 if (term_len == command_buf.len
2021 && !strcmp(term, command_buf.buf))
2022 break;
2023 strbuf_addbuf(sb, &command_buf);
2024 strbuf_addch(sb, '\n');
2026 free(term);
2028 else {
2029 uintmax_t len = strtoumax(data, NULL, 10);
2030 size_t n = 0, length = (size_t)len;
2032 if (limit && limit < len) {
2033 *len_res = len;
2034 return 0;
2036 if (length < len)
2037 die("data is too large to use in this context");
2039 while (n < length) {
2040 size_t s = strbuf_fread(sb, length - n, stdin);
2041 if (!s && feof(stdin))
2042 die("EOF in data (%lu bytes remaining)",
2043 (unsigned long)(length - n));
2044 n += s;
2048 skip_optional_lf();
2049 return 1;
2052 static int validate_raw_date(const char *src, struct strbuf *result)
2054 const char *orig_src = src;
2055 char *endp;
2056 unsigned long num;
2058 errno = 0;
2060 num = strtoul(src, &endp, 10);
2061 /* NEEDSWORK: perhaps check for reasonable values? */
2062 if (errno || endp == src || *endp != ' ')
2063 return -1;
2065 src = endp + 1;
2066 if (*src != '-' && *src != '+')
2067 return -1;
2069 num = strtoul(src + 1, &endp, 10);
2070 if (errno || endp == src + 1 || *endp || 1400 < num)
2071 return -1;
2073 strbuf_addstr(result, orig_src);
2074 return 0;
2077 static char *parse_ident(const char *buf)
2079 const char *ltgt;
2080 size_t name_len;
2081 struct strbuf ident = STRBUF_INIT;
2083 /* ensure there is a space delimiter even if there is no name */
2084 if (*buf == '<')
2085 --buf;
2087 ltgt = buf + strcspn(buf, "<>");
2088 if (*ltgt != '<')
2089 die("Missing < in ident string: %s", buf);
2090 if (ltgt != buf && ltgt[-1] != ' ')
2091 die("Missing space before < in ident string: %s", buf);
2092 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2093 if (*ltgt != '>')
2094 die("Missing > in ident string: %s", buf);
2095 ltgt++;
2096 if (*ltgt != ' ')
2097 die("Missing space after > in ident string: %s", buf);
2098 ltgt++;
2099 name_len = ltgt - buf;
2100 strbuf_add(&ident, buf, name_len);
2102 switch (whenspec) {
2103 case WHENSPEC_RAW:
2104 if (validate_raw_date(ltgt, &ident) < 0)
2105 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2106 break;
2107 case WHENSPEC_RFC2822:
2108 if (parse_date(ltgt, &ident) < 0)
2109 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2110 break;
2111 case WHENSPEC_NOW:
2112 if (strcmp("now", ltgt))
2113 die("Date in ident must be 'now': %s", buf);
2114 datestamp(&ident);
2115 break;
2118 return strbuf_detach(&ident, NULL);
2121 static void parse_and_store_blob(
2122 struct last_object *last,
2123 unsigned char *sha1out,
2124 uintmax_t mark)
2126 static struct strbuf buf = STRBUF_INIT;
2127 uintmax_t len;
2129 if (parse_data(&buf, big_file_threshold, &len))
2130 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2131 else {
2132 if (last) {
2133 strbuf_release(&last->data);
2134 last->offset = 0;
2135 last->depth = 0;
2137 stream_blob(len, sha1out, mark);
2138 skip_optional_lf();
2142 static void parse_new_blob(void)
2144 read_next_command();
2145 parse_mark();
2146 parse_and_store_blob(&last_blob, NULL, next_mark);
2149 static void unload_one_branch(void)
2151 while (cur_active_branches
2152 && cur_active_branches >= max_active_branches) {
2153 uintmax_t min_commit = ULONG_MAX;
2154 struct branch *e, *l = NULL, *p = NULL;
2156 for (e = active_branches; e; e = e->active_next_branch) {
2157 if (e->last_commit < min_commit) {
2158 p = l;
2159 min_commit = e->last_commit;
2161 l = e;
2164 if (p) {
2165 e = p->active_next_branch;
2166 p->active_next_branch = e->active_next_branch;
2167 } else {
2168 e = active_branches;
2169 active_branches = e->active_next_branch;
2171 e->active = 0;
2172 e->active_next_branch = NULL;
2173 if (e->branch_tree.tree) {
2174 release_tree_content_recursive(e->branch_tree.tree);
2175 e->branch_tree.tree = NULL;
2177 cur_active_branches--;
2181 static void load_branch(struct branch *b)
2183 load_tree(&b->branch_tree);
2184 if (!b->active) {
2185 b->active = 1;
2186 b->active_next_branch = active_branches;
2187 active_branches = b;
2188 cur_active_branches++;
2189 branch_load_count++;
2193 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2195 unsigned char fanout = 0;
2196 while ((num_notes >>= 8))
2197 fanout++;
2198 return fanout;
2201 static void construct_path_with_fanout(const char *hex_sha1,
2202 unsigned char fanout, char *path)
2204 unsigned int i = 0, j = 0;
2205 if (fanout >= 20)
2206 die("Too large fanout (%u)", fanout);
2207 while (fanout) {
2208 path[i++] = hex_sha1[j++];
2209 path[i++] = hex_sha1[j++];
2210 path[i++] = '/';
2211 fanout--;
2213 memcpy(path + i, hex_sha1 + j, 40 - j);
2214 path[i + 40 - j] = '\0';
2217 static uintmax_t do_change_note_fanout(
2218 struct tree_entry *orig_root, struct tree_entry *root,
2219 char *hex_sha1, unsigned int hex_sha1_len,
2220 char *fullpath, unsigned int fullpath_len,
2221 unsigned char fanout)
2223 struct tree_content *t = root->tree;
2224 struct tree_entry *e, leaf;
2225 unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2226 uintmax_t num_notes = 0;
2227 unsigned char sha1[20];
2228 char realpath[60];
2230 for (i = 0; t && i < t->entry_count; i++) {
2231 e = t->entries[i];
2232 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2233 tmp_fullpath_len = fullpath_len;
2236 * We're interested in EITHER existing note entries (entries
2237 * with exactly 40 hex chars in path, not including directory
2238 * separators), OR directory entries that may contain note
2239 * entries (with < 40 hex chars in path).
2240 * Also, each path component in a note entry must be a multiple
2241 * of 2 chars.
2243 if (!e->versions[1].mode ||
2244 tmp_hex_sha1_len > 40 ||
2245 e->name->str_len % 2)
2246 continue;
2248 /* This _may_ be a note entry, or a subdir containing notes */
2249 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2250 e->name->str_len);
2251 if (tmp_fullpath_len)
2252 fullpath[tmp_fullpath_len++] = '/';
2253 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2254 e->name->str_len);
2255 tmp_fullpath_len += e->name->str_len;
2256 fullpath[tmp_fullpath_len] = '\0';
2258 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2259 /* This is a note entry */
2260 if (fanout == 0xff) {
2261 /* Counting mode, no rename */
2262 num_notes++;
2263 continue;
2265 construct_path_with_fanout(hex_sha1, fanout, realpath);
2266 if (!strcmp(fullpath, realpath)) {
2267 /* Note entry is in correct location */
2268 num_notes++;
2269 continue;
2272 /* Rename fullpath to realpath */
2273 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2274 die("Failed to remove path %s", fullpath);
2275 tree_content_set(orig_root, realpath,
2276 leaf.versions[1].sha1,
2277 leaf.versions[1].mode,
2278 leaf.tree);
2279 } else if (S_ISDIR(e->versions[1].mode)) {
2280 /* This is a subdir that may contain note entries */
2281 if (!e->tree)
2282 load_tree(e);
2283 num_notes += do_change_note_fanout(orig_root, e,
2284 hex_sha1, tmp_hex_sha1_len,
2285 fullpath, tmp_fullpath_len, fanout);
2288 /* The above may have reallocated the current tree_content */
2289 t = root->tree;
2291 return num_notes;
2294 static uintmax_t change_note_fanout(struct tree_entry *root,
2295 unsigned char fanout)
2297 char hex_sha1[40], path[60];
2298 return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2302 * Given a pointer into a string, parse a mark reference:
2304 * idnum ::= ':' bigint;
2306 * Return the first character after the value in *endptr.
2308 * Complain if the following character is not what is expected,
2309 * either a space or end of the string.
2311 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2313 uintmax_t mark;
2315 assert(*p == ':');
2316 p++;
2317 mark = strtoumax(p, endptr, 10);
2318 if (*endptr == p)
2319 die("No value after ':' in mark: %s", command_buf.buf);
2320 return mark;
2324 * Parse the mark reference, and complain if this is not the end of
2325 * the string.
2327 static uintmax_t parse_mark_ref_eol(const char *p)
2329 char *end;
2330 uintmax_t mark;
2332 mark = parse_mark_ref(p, &end);
2333 if (*end != '\0')
2334 die("Garbage after mark: %s", command_buf.buf);
2335 return mark;
2339 * Parse the mark reference, demanding a trailing space. Return a
2340 * pointer to the space.
2342 static uintmax_t parse_mark_ref_space(const char **p)
2344 uintmax_t mark;
2345 char *end;
2347 mark = parse_mark_ref(*p, &end);
2348 if (*end++ != ' ')
2349 die("Missing space after mark: %s", command_buf.buf);
2350 *p = end;
2351 return mark;
2354 static void file_change_m(const char *p, struct branch *b)
2356 static struct strbuf uq = STRBUF_INIT;
2357 const char *endp;
2358 struct object_entry *oe;
2359 unsigned char sha1[20];
2360 uint16_t mode, inline_data = 0;
2362 p = get_mode(p, &mode);
2363 if (!p)
2364 die("Corrupt mode: %s", command_buf.buf);
2365 switch (mode) {
2366 case 0644:
2367 case 0755:
2368 mode |= S_IFREG;
2369 case S_IFREG | 0644:
2370 case S_IFREG | 0755:
2371 case S_IFLNK:
2372 case S_IFDIR:
2373 case S_IFGITLINK:
2374 /* ok */
2375 break;
2376 default:
2377 die("Corrupt mode: %s", command_buf.buf);
2380 if (*p == ':') {
2381 oe = find_mark(parse_mark_ref_space(&p));
2382 hashcpy(sha1, oe->idx.sha1);
2383 } else if (skip_prefix(p, "inline ", &p)) {
2384 inline_data = 1;
2385 oe = NULL; /* not used with inline_data, but makes gcc happy */
2386 } else {
2387 if (get_sha1_hex(p, sha1))
2388 die("Invalid dataref: %s", command_buf.buf);
2389 oe = find_object(sha1);
2390 p += 40;
2391 if (*p++ != ' ')
2392 die("Missing space after SHA1: %s", command_buf.buf);
2395 strbuf_reset(&uq);
2396 if (!unquote_c_style(&uq, p, &endp)) {
2397 if (*endp)
2398 die("Garbage after path in: %s", command_buf.buf);
2399 p = uq.buf;
2402 /* Git does not track empty, non-toplevel directories. */
2403 if (S_ISDIR(mode) && !hashcmp(sha1, EMPTY_TREE_SHA1_BIN) && *p) {
2404 tree_content_remove(&b->branch_tree, p, NULL, 0);
2405 return;
2408 if (S_ISGITLINK(mode)) {
2409 if (inline_data)
2410 die("Git links cannot be specified 'inline': %s",
2411 command_buf.buf);
2412 else if (oe) {
2413 if (oe->type != OBJ_COMMIT)
2414 die("Not a commit (actually a %s): %s",
2415 typename(oe->type), command_buf.buf);
2418 * Accept the sha1 without checking; it expected to be in
2419 * another repository.
2421 } else if (inline_data) {
2422 if (S_ISDIR(mode))
2423 die("Directories cannot be specified 'inline': %s",
2424 command_buf.buf);
2425 if (p != uq.buf) {
2426 strbuf_addstr(&uq, p);
2427 p = uq.buf;
2429 read_next_command();
2430 parse_and_store_blob(&last_blob, sha1, 0);
2431 } else {
2432 enum object_type expected = S_ISDIR(mode) ?
2433 OBJ_TREE: OBJ_BLOB;
2434 enum object_type type = oe ? oe->type :
2435 sha1_object_info(sha1, NULL);
2436 if (type < 0)
2437 die("%s not found: %s",
2438 S_ISDIR(mode) ? "Tree" : "Blob",
2439 command_buf.buf);
2440 if (type != expected)
2441 die("Not a %s (actually a %s): %s",
2442 typename(expected), typename(type),
2443 command_buf.buf);
2446 if (!*p) {
2447 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2448 return;
2450 tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2453 static void file_change_d(const char *p, struct branch *b)
2455 static struct strbuf uq = STRBUF_INIT;
2456 const char *endp;
2458 strbuf_reset(&uq);
2459 if (!unquote_c_style(&uq, p, &endp)) {
2460 if (*endp)
2461 die("Garbage after path in: %s", command_buf.buf);
2462 p = uq.buf;
2464 tree_content_remove(&b->branch_tree, p, NULL, 1);
2467 static void file_change_cr(const char *s, struct branch *b, int rename)
2469 const char *d;
2470 static struct strbuf s_uq = STRBUF_INIT;
2471 static struct strbuf d_uq = STRBUF_INIT;
2472 const char *endp;
2473 struct tree_entry leaf;
2475 strbuf_reset(&s_uq);
2476 if (!unquote_c_style(&s_uq, s, &endp)) {
2477 if (*endp != ' ')
2478 die("Missing space after source: %s", command_buf.buf);
2479 } else {
2480 endp = strchr(s, ' ');
2481 if (!endp)
2482 die("Missing space after source: %s", command_buf.buf);
2483 strbuf_add(&s_uq, s, endp - s);
2485 s = s_uq.buf;
2487 endp++;
2488 if (!*endp)
2489 die("Missing dest: %s", command_buf.buf);
2491 d = endp;
2492 strbuf_reset(&d_uq);
2493 if (!unquote_c_style(&d_uq, d, &endp)) {
2494 if (*endp)
2495 die("Garbage after dest in: %s", command_buf.buf);
2496 d = d_uq.buf;
2499 memset(&leaf, 0, sizeof(leaf));
2500 if (rename)
2501 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2502 else
2503 tree_content_get(&b->branch_tree, s, &leaf, 1);
2504 if (!leaf.versions[1].mode)
2505 die("Path %s not in branch", s);
2506 if (!*d) { /* C "path/to/subdir" "" */
2507 tree_content_replace(&b->branch_tree,
2508 leaf.versions[1].sha1,
2509 leaf.versions[1].mode,
2510 leaf.tree);
2511 return;
2513 tree_content_set(&b->branch_tree, d,
2514 leaf.versions[1].sha1,
2515 leaf.versions[1].mode,
2516 leaf.tree);
2519 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2521 static struct strbuf uq = STRBUF_INIT;
2522 struct object_entry *oe;
2523 struct branch *s;
2524 unsigned char sha1[20], commit_sha1[20];
2525 char path[60];
2526 uint16_t inline_data = 0;
2527 unsigned char new_fanout;
2530 * When loading a branch, we don't traverse its tree to count the real
2531 * number of notes (too expensive to do this for all non-note refs).
2532 * This means that recently loaded notes refs might incorrectly have
2533 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2535 * Fix this by traversing the tree and counting the number of notes
2536 * when b->num_notes == 0. If the notes tree is truly empty, the
2537 * calculation should not take long.
2539 if (b->num_notes == 0 && *old_fanout == 0) {
2540 /* Invoke change_note_fanout() in "counting mode". */
2541 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2542 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2545 /* Now parse the notemodify command. */
2546 /* <dataref> or 'inline' */
2547 if (*p == ':') {
2548 oe = find_mark(parse_mark_ref_space(&p));
2549 hashcpy(sha1, oe->idx.sha1);
2550 } else if (skip_prefix(p, "inline ", &p)) {
2551 inline_data = 1;
2552 oe = NULL; /* not used with inline_data, but makes gcc happy */
2553 } else {
2554 if (get_sha1_hex(p, sha1))
2555 die("Invalid dataref: %s", command_buf.buf);
2556 oe = find_object(sha1);
2557 p += 40;
2558 if (*p++ != ' ')
2559 die("Missing space after SHA1: %s", command_buf.buf);
2562 /* <commit-ish> */
2563 s = lookup_branch(p);
2564 if (s) {
2565 if (is_null_sha1(s->sha1))
2566 die("Can't add a note on empty branch.");
2567 hashcpy(commit_sha1, s->sha1);
2568 } else if (*p == ':') {
2569 uintmax_t commit_mark = parse_mark_ref_eol(p);
2570 struct object_entry *commit_oe = find_mark(commit_mark);
2571 if (commit_oe->type != OBJ_COMMIT)
2572 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2573 hashcpy(commit_sha1, commit_oe->idx.sha1);
2574 } else if (!get_sha1(p, commit_sha1)) {
2575 unsigned long size;
2576 char *buf = read_object_with_reference(commit_sha1,
2577 commit_type, &size, commit_sha1);
2578 if (!buf || size < 46)
2579 die("Not a valid commit: %s", p);
2580 free(buf);
2581 } else
2582 die("Invalid ref name or SHA1 expression: %s", p);
2584 if (inline_data) {
2585 if (p != uq.buf) {
2586 strbuf_addstr(&uq, p);
2587 p = uq.buf;
2589 read_next_command();
2590 parse_and_store_blob(&last_blob, sha1, 0);
2591 } else if (oe) {
2592 if (oe->type != OBJ_BLOB)
2593 die("Not a blob (actually a %s): %s",
2594 typename(oe->type), command_buf.buf);
2595 } else if (!is_null_sha1(sha1)) {
2596 enum object_type type = sha1_object_info(sha1, NULL);
2597 if (type < 0)
2598 die("Blob not found: %s", command_buf.buf);
2599 if (type != OBJ_BLOB)
2600 die("Not a blob (actually a %s): %s",
2601 typename(type), command_buf.buf);
2604 construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2605 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2606 b->num_notes--;
2608 if (is_null_sha1(sha1))
2609 return; /* nothing to insert */
2611 b->num_notes++;
2612 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2613 construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2614 tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2617 static void file_change_deleteall(struct branch *b)
2619 release_tree_content_recursive(b->branch_tree.tree);
2620 hashclr(b->branch_tree.versions[0].sha1);
2621 hashclr(b->branch_tree.versions[1].sha1);
2622 load_tree(&b->branch_tree);
2623 b->num_notes = 0;
2626 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2628 if (!buf || size < 46)
2629 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2630 if (memcmp("tree ", buf, 5)
2631 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2632 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2633 hashcpy(b->branch_tree.versions[0].sha1,
2634 b->branch_tree.versions[1].sha1);
2637 static void parse_from_existing(struct branch *b)
2639 if (is_null_sha1(b->sha1)) {
2640 hashclr(b->branch_tree.versions[0].sha1);
2641 hashclr(b->branch_tree.versions[1].sha1);
2642 } else {
2643 unsigned long size;
2644 char *buf;
2646 buf = read_object_with_reference(b->sha1,
2647 commit_type, &size, b->sha1);
2648 parse_from_commit(b, buf, size);
2649 free(buf);
2653 static int parse_from(struct branch *b)
2655 const char *from;
2656 struct branch *s;
2657 unsigned char sha1[20];
2659 if (!skip_prefix(command_buf.buf, "from ", &from))
2660 return 0;
2662 hashcpy(sha1, b->branch_tree.versions[1].sha1);
2664 s = lookup_branch(from);
2665 if (b == s)
2666 die("Can't create a branch from itself: %s", b->name);
2667 else if (s) {
2668 unsigned char *t = s->branch_tree.versions[1].sha1;
2669 hashcpy(b->sha1, s->sha1);
2670 hashcpy(b->branch_tree.versions[0].sha1, t);
2671 hashcpy(b->branch_tree.versions[1].sha1, t);
2672 } else if (*from == ':') {
2673 uintmax_t idnum = parse_mark_ref_eol(from);
2674 struct object_entry *oe = find_mark(idnum);
2675 if (oe->type != OBJ_COMMIT)
2676 die("Mark :%" PRIuMAX " not a commit", idnum);
2677 if (hashcmp(b->sha1, oe->idx.sha1)) {
2678 hashcpy(b->sha1, oe->idx.sha1);
2679 if (oe->pack_id != MAX_PACK_ID) {
2680 unsigned long size;
2681 char *buf = gfi_unpack_entry(oe, &size);
2682 parse_from_commit(b, buf, size);
2683 free(buf);
2684 } else
2685 parse_from_existing(b);
2687 } else if (!get_sha1(from, b->sha1)) {
2688 parse_from_existing(b);
2689 if (is_null_sha1(b->sha1))
2690 b->delete = 1;
2692 else
2693 die("Invalid ref name or SHA1 expression: %s", from);
2695 if (b->branch_tree.tree && hashcmp(sha1, b->branch_tree.versions[1].sha1)) {
2696 release_tree_content_recursive(b->branch_tree.tree);
2697 b->branch_tree.tree = NULL;
2700 read_next_command();
2701 return 1;
2704 static struct hash_list *parse_merge(unsigned int *count)
2706 struct hash_list *list = NULL, **tail = &list, *n;
2707 const char *from;
2708 struct branch *s;
2710 *count = 0;
2711 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2712 n = xmalloc(sizeof(*n));
2713 s = lookup_branch(from);
2714 if (s)
2715 hashcpy(n->sha1, s->sha1);
2716 else if (*from == ':') {
2717 uintmax_t idnum = parse_mark_ref_eol(from);
2718 struct object_entry *oe = find_mark(idnum);
2719 if (oe->type != OBJ_COMMIT)
2720 die("Mark :%" PRIuMAX " not a commit", idnum);
2721 hashcpy(n->sha1, oe->idx.sha1);
2722 } else if (!get_sha1(from, n->sha1)) {
2723 unsigned long size;
2724 char *buf = read_object_with_reference(n->sha1,
2725 commit_type, &size, n->sha1);
2726 if (!buf || size < 46)
2727 die("Not a valid commit: %s", from);
2728 free(buf);
2729 } else
2730 die("Invalid ref name or SHA1 expression: %s", from);
2732 n->next = NULL;
2733 *tail = n;
2734 tail = &n->next;
2736 (*count)++;
2737 read_next_command();
2739 return list;
2742 static void parse_new_commit(const char *arg)
2744 static struct strbuf msg = STRBUF_INIT;
2745 struct branch *b;
2746 char *author = NULL;
2747 char *committer = NULL;
2748 struct hash_list *merge_list = NULL;
2749 unsigned int merge_count;
2750 unsigned char prev_fanout, new_fanout;
2751 const char *v;
2753 b = lookup_branch(arg);
2754 if (!b)
2755 b = new_branch(arg);
2757 read_next_command();
2758 parse_mark();
2759 if (skip_prefix(command_buf.buf, "author ", &v)) {
2760 author = parse_ident(v);
2761 read_next_command();
2763 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2764 committer = parse_ident(v);
2765 read_next_command();
2767 if (!committer)
2768 die("Expected committer but didn't get one");
2769 parse_data(&msg, 0, NULL);
2770 read_next_command();
2771 parse_from(b);
2772 merge_list = parse_merge(&merge_count);
2774 /* ensure the branch is active/loaded */
2775 if (!b->branch_tree.tree || !max_active_branches) {
2776 unload_one_branch();
2777 load_branch(b);
2780 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2782 /* file_change* */
2783 while (command_buf.len > 0) {
2784 if (skip_prefix(command_buf.buf, "M ", &v))
2785 file_change_m(v, b);
2786 else if (skip_prefix(command_buf.buf, "D ", &v))
2787 file_change_d(v, b);
2788 else if (skip_prefix(command_buf.buf, "R ", &v))
2789 file_change_cr(v, b, 1);
2790 else if (skip_prefix(command_buf.buf, "C ", &v))
2791 file_change_cr(v, b, 0);
2792 else if (skip_prefix(command_buf.buf, "N ", &v))
2793 note_change_n(v, b, &prev_fanout);
2794 else if (!strcmp("deleteall", command_buf.buf))
2795 file_change_deleteall(b);
2796 else if (skip_prefix(command_buf.buf, "ls ", &v))
2797 parse_ls(v, b);
2798 else {
2799 unread_command_buf = 1;
2800 break;
2802 if (read_next_command() == EOF)
2803 break;
2806 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2807 if (new_fanout != prev_fanout)
2808 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2810 /* build the tree and the commit */
2811 store_tree(&b->branch_tree);
2812 hashcpy(b->branch_tree.versions[0].sha1,
2813 b->branch_tree.versions[1].sha1);
2815 strbuf_reset(&new_data);
2816 strbuf_addf(&new_data, "tree %s\n",
2817 sha1_to_hex(b->branch_tree.versions[1].sha1));
2818 if (!is_null_sha1(b->sha1))
2819 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2820 while (merge_list) {
2821 struct hash_list *next = merge_list->next;
2822 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2823 free(merge_list);
2824 merge_list = next;
2826 strbuf_addf(&new_data,
2827 "author %s\n"
2828 "committer %s\n"
2829 "\n",
2830 author ? author : committer, committer);
2831 strbuf_addbuf(&new_data, &msg);
2832 free(author);
2833 free(committer);
2835 if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2836 b->pack_id = pack_id;
2837 b->last_commit = object_count_by_type[OBJ_COMMIT];
2840 static void parse_new_tag(const char *arg)
2842 static struct strbuf msg = STRBUF_INIT;
2843 const char *from;
2844 char *tagger;
2845 struct branch *s;
2846 struct tag *t;
2847 uintmax_t from_mark = 0;
2848 unsigned char sha1[20];
2849 enum object_type type;
2850 const char *v;
2852 t = pool_alloc(sizeof(struct tag));
2853 memset(t, 0, sizeof(struct tag));
2854 t->name = pool_strdup(arg);
2855 if (last_tag)
2856 last_tag->next_tag = t;
2857 else
2858 first_tag = t;
2859 last_tag = t;
2860 read_next_command();
2862 /* from ... */
2863 if (!skip_prefix(command_buf.buf, "from ", &from))
2864 die("Expected from command, got %s", command_buf.buf);
2865 s = lookup_branch(from);
2866 if (s) {
2867 if (is_null_sha1(s->sha1))
2868 die("Can't tag an empty branch.");
2869 hashcpy(sha1, s->sha1);
2870 type = OBJ_COMMIT;
2871 } else if (*from == ':') {
2872 struct object_entry *oe;
2873 from_mark = parse_mark_ref_eol(from);
2874 oe = find_mark(from_mark);
2875 type = oe->type;
2876 hashcpy(sha1, oe->idx.sha1);
2877 } else if (!get_sha1(from, sha1)) {
2878 struct object_entry *oe = find_object(sha1);
2879 if (!oe) {
2880 type = sha1_object_info(sha1, NULL);
2881 if (type < 0)
2882 die("Not a valid object: %s", from);
2883 } else
2884 type = oe->type;
2885 } else
2886 die("Invalid ref name or SHA1 expression: %s", from);
2887 read_next_command();
2889 /* tagger ... */
2890 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2891 tagger = parse_ident(v);
2892 read_next_command();
2893 } else
2894 tagger = NULL;
2896 /* tag payload/message */
2897 parse_data(&msg, 0, NULL);
2899 /* build the tag object */
2900 strbuf_reset(&new_data);
2902 strbuf_addf(&new_data,
2903 "object %s\n"
2904 "type %s\n"
2905 "tag %s\n",
2906 sha1_to_hex(sha1), typename(type), t->name);
2907 if (tagger)
2908 strbuf_addf(&new_data,
2909 "tagger %s\n", tagger);
2910 strbuf_addch(&new_data, '\n');
2911 strbuf_addbuf(&new_data, &msg);
2912 free(tagger);
2914 if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2915 t->pack_id = MAX_PACK_ID;
2916 else
2917 t->pack_id = pack_id;
2920 static void parse_reset_branch(const char *arg)
2922 struct branch *b;
2924 b = lookup_branch(arg);
2925 if (b) {
2926 hashclr(b->sha1);
2927 hashclr(b->branch_tree.versions[0].sha1);
2928 hashclr(b->branch_tree.versions[1].sha1);
2929 if (b->branch_tree.tree) {
2930 release_tree_content_recursive(b->branch_tree.tree);
2931 b->branch_tree.tree = NULL;
2934 else
2935 b = new_branch(arg);
2936 read_next_command();
2937 parse_from(b);
2938 if (command_buf.len > 0)
2939 unread_command_buf = 1;
2942 static void cat_blob_write(const char *buf, unsigned long size)
2944 if (write_in_full(cat_blob_fd, buf, size) != size)
2945 die_errno("Write to frontend failed");
2948 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2950 struct strbuf line = STRBUF_INIT;
2951 unsigned long size;
2952 enum object_type type = 0;
2953 char *buf;
2955 if (!oe || oe->pack_id == MAX_PACK_ID) {
2956 buf = read_sha1_file(sha1, &type, &size);
2957 } else {
2958 type = oe->type;
2959 buf = gfi_unpack_entry(oe, &size);
2963 * Output based on batch_one_object() from cat-file.c.
2965 if (type <= 0) {
2966 strbuf_reset(&line);
2967 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2968 cat_blob_write(line.buf, line.len);
2969 strbuf_release(&line);
2970 free(buf);
2971 return;
2973 if (!buf)
2974 die("Can't read object %s", sha1_to_hex(sha1));
2975 if (type != OBJ_BLOB)
2976 die("Object %s is a %s but a blob was expected.",
2977 sha1_to_hex(sha1), typename(type));
2978 strbuf_reset(&line);
2979 strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2980 typename(type), size);
2981 cat_blob_write(line.buf, line.len);
2982 strbuf_release(&line);
2983 cat_blob_write(buf, size);
2984 cat_blob_write("\n", 1);
2985 if (oe && oe->pack_id == pack_id) {
2986 last_blob.offset = oe->idx.offset;
2987 strbuf_attach(&last_blob.data, buf, size, size);
2988 last_blob.depth = oe->depth;
2989 } else
2990 free(buf);
2993 static void parse_get_mark(const char *p)
2995 struct object_entry *oe = oe;
2996 char output[42];
2998 /* get-mark SP <object> LF */
2999 if (*p != ':')
3000 die("Not a mark: %s", p);
3002 oe = find_mark(parse_mark_ref_eol(p));
3003 if (!oe)
3004 die("Unknown mark: %s", command_buf.buf);
3006 snprintf(output, sizeof(output), "%s\n", sha1_to_hex(oe->idx.sha1));
3007 cat_blob_write(output, 41);
3010 static void parse_cat_blob(const char *p)
3012 struct object_entry *oe = oe;
3013 unsigned char sha1[20];
3015 /* cat-blob SP <object> LF */
3016 if (*p == ':') {
3017 oe = find_mark(parse_mark_ref_eol(p));
3018 if (!oe)
3019 die("Unknown mark: %s", command_buf.buf);
3020 hashcpy(sha1, oe->idx.sha1);
3021 } else {
3022 if (get_sha1_hex(p, sha1))
3023 die("Invalid dataref: %s", command_buf.buf);
3024 if (p[40])
3025 die("Garbage after SHA1: %s", command_buf.buf);
3026 oe = find_object(sha1);
3029 cat_blob(oe, sha1);
3032 static struct object_entry *dereference(struct object_entry *oe,
3033 unsigned char sha1[20])
3035 unsigned long size;
3036 char *buf = NULL;
3037 if (!oe) {
3038 enum object_type type = sha1_object_info(sha1, NULL);
3039 if (type < 0)
3040 die("object not found: %s", sha1_to_hex(sha1));
3041 /* cache it! */
3042 oe = insert_object(sha1);
3043 oe->type = type;
3044 oe->pack_id = MAX_PACK_ID;
3045 oe->idx.offset = 1;
3047 switch (oe->type) {
3048 case OBJ_TREE: /* easy case. */
3049 return oe;
3050 case OBJ_COMMIT:
3051 case OBJ_TAG:
3052 break;
3053 default:
3054 die("Not a tree-ish: %s", command_buf.buf);
3057 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3058 buf = gfi_unpack_entry(oe, &size);
3059 } else {
3060 enum object_type unused;
3061 buf = read_sha1_file(sha1, &unused, &size);
3063 if (!buf)
3064 die("Can't load object %s", sha1_to_hex(sha1));
3066 /* Peel one layer. */
3067 switch (oe->type) {
3068 case OBJ_TAG:
3069 if (size < 40 + strlen("object ") ||
3070 get_sha1_hex(buf + strlen("object "), sha1))
3071 die("Invalid SHA1 in tag: %s", command_buf.buf);
3072 break;
3073 case OBJ_COMMIT:
3074 if (size < 40 + strlen("tree ") ||
3075 get_sha1_hex(buf + strlen("tree "), sha1))
3076 die("Invalid SHA1 in commit: %s", command_buf.buf);
3079 free(buf);
3080 return find_object(sha1);
3083 static struct object_entry *parse_treeish_dataref(const char **p)
3085 unsigned char sha1[20];
3086 struct object_entry *e;
3088 if (**p == ':') { /* <mark> */
3089 e = find_mark(parse_mark_ref_space(p));
3090 if (!e)
3091 die("Unknown mark: %s", command_buf.buf);
3092 hashcpy(sha1, e->idx.sha1);
3093 } else { /* <sha1> */
3094 if (get_sha1_hex(*p, sha1))
3095 die("Invalid dataref: %s", command_buf.buf);
3096 e = find_object(sha1);
3097 *p += 40;
3098 if (*(*p)++ != ' ')
3099 die("Missing space after tree-ish: %s", command_buf.buf);
3102 while (!e || e->type != OBJ_TREE)
3103 e = dereference(e, sha1);
3104 return e;
3107 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3109 static struct strbuf line = STRBUF_INIT;
3111 /* See show_tree(). */
3112 const char *type =
3113 S_ISGITLINK(mode) ? commit_type :
3114 S_ISDIR(mode) ? tree_type :
3115 blob_type;
3117 if (!mode) {
3118 /* missing SP path LF */
3119 strbuf_reset(&line);
3120 strbuf_addstr(&line, "missing ");
3121 quote_c_style(path, &line, NULL, 0);
3122 strbuf_addch(&line, '\n');
3123 } else {
3124 /* mode SP type SP object_name TAB path LF */
3125 strbuf_reset(&line);
3126 strbuf_addf(&line, "%06o %s %s\t",
3127 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3128 quote_c_style(path, &line, NULL, 0);
3129 strbuf_addch(&line, '\n');
3131 cat_blob_write(line.buf, line.len);
3134 static void parse_ls(const char *p, struct branch *b)
3136 struct tree_entry *root = NULL;
3137 struct tree_entry leaf = {NULL};
3139 /* ls SP (<tree-ish> SP)? <path> */
3140 if (*p == '"') {
3141 if (!b)
3142 die("Not in a commit: %s", command_buf.buf);
3143 root = &b->branch_tree;
3144 } else {
3145 struct object_entry *e = parse_treeish_dataref(&p);
3146 root = new_tree_entry();
3147 hashcpy(root->versions[1].sha1, e->idx.sha1);
3148 if (!is_null_sha1(root->versions[1].sha1))
3149 root->versions[1].mode = S_IFDIR;
3150 load_tree(root);
3152 if (*p == '"') {
3153 static struct strbuf uq = STRBUF_INIT;
3154 const char *endp;
3155 strbuf_reset(&uq);
3156 if (unquote_c_style(&uq, p, &endp))
3157 die("Invalid path: %s", command_buf.buf);
3158 if (*endp)
3159 die("Garbage after path in: %s", command_buf.buf);
3160 p = uq.buf;
3162 tree_content_get(root, p, &leaf, 1);
3164 * A directory in preparation would have a sha1 of zero
3165 * until it is saved. Save, for simplicity.
3167 if (S_ISDIR(leaf.versions[1].mode))
3168 store_tree(&leaf);
3170 print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3171 if (leaf.tree)
3172 release_tree_content_recursive(leaf.tree);
3173 if (!b || root != &b->branch_tree)
3174 release_tree_entry(root);
3177 static void checkpoint(void)
3179 checkpoint_requested = 0;
3180 if (object_count) {
3181 cycle_packfile();
3182 dump_branches();
3183 dump_tags();
3184 dump_marks();
3188 static void parse_checkpoint(void)
3190 checkpoint_requested = 1;
3191 skip_optional_lf();
3194 static void parse_progress(void)
3196 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3197 fputc('\n', stdout);
3198 fflush(stdout);
3199 skip_optional_lf();
3202 static char* make_fast_import_path(const char *path)
3204 if (!relative_marks_paths || is_absolute_path(path))
3205 return xstrdup(path);
3206 return xstrdup(git_path("info/fast-import/%s", path));
3209 static void option_import_marks(const char *marks,
3210 int from_stream, int ignore_missing)
3212 if (import_marks_file) {
3213 if (from_stream)
3214 die("Only one import-marks command allowed per stream");
3216 /* read previous mark file */
3217 if(!import_marks_file_from_stream)
3218 read_marks();
3221 import_marks_file = make_fast_import_path(marks);
3222 safe_create_leading_directories_const(import_marks_file);
3223 import_marks_file_from_stream = from_stream;
3224 import_marks_file_ignore_missing = ignore_missing;
3227 static void option_date_format(const char *fmt)
3229 if (!strcmp(fmt, "raw"))
3230 whenspec = WHENSPEC_RAW;
3231 else if (!strcmp(fmt, "rfc2822"))
3232 whenspec = WHENSPEC_RFC2822;
3233 else if (!strcmp(fmt, "now"))
3234 whenspec = WHENSPEC_NOW;
3235 else
3236 die("unknown --date-format argument %s", fmt);
3239 static unsigned long ulong_arg(const char *option, const char *arg)
3241 char *endptr;
3242 unsigned long rv = strtoul(arg, &endptr, 0);
3243 if (strchr(arg, '-') || endptr == arg || *endptr)
3244 die("%s: argument must be a non-negative integer", option);
3245 return rv;
3248 static void option_depth(const char *depth)
3250 max_depth = ulong_arg("--depth", depth);
3251 if (max_depth > MAX_DEPTH)
3252 die("--depth cannot exceed %u", MAX_DEPTH);
3255 static void option_active_branches(const char *branches)
3257 max_active_branches = ulong_arg("--active-branches", branches);
3260 static void option_export_marks(const char *marks)
3262 export_marks_file = make_fast_import_path(marks);
3263 safe_create_leading_directories_const(export_marks_file);
3266 static void option_cat_blob_fd(const char *fd)
3268 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3269 if (n > (unsigned long) INT_MAX)
3270 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3271 cat_blob_fd = (int) n;
3274 static void option_export_pack_edges(const char *edges)
3276 if (pack_edges)
3277 fclose(pack_edges);
3278 pack_edges = fopen(edges, "a");
3279 if (!pack_edges)
3280 die_errno("Cannot open '%s'", edges);
3283 static int parse_one_option(const char *option)
3285 if (skip_prefix(option, "max-pack-size=", &option)) {
3286 unsigned long v;
3287 if (!git_parse_ulong(option, &v))
3288 return 0;
3289 if (v < 8192) {
3290 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3291 v *= 1024 * 1024;
3292 } else if (v < 1024 * 1024) {
3293 warning("minimum max-pack-size is 1 MiB");
3294 v = 1024 * 1024;
3296 max_packsize = v;
3297 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3298 unsigned long v;
3299 if (!git_parse_ulong(option, &v))
3300 return 0;
3301 big_file_threshold = v;
3302 } else if (skip_prefix(option, "depth=", &option)) {
3303 option_depth(option);
3304 } else if (skip_prefix(option, "active-branches=", &option)) {
3305 option_active_branches(option);
3306 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3307 option_export_pack_edges(option);
3308 } else if (starts_with(option, "quiet")) {
3309 show_stats = 0;
3310 } else if (starts_with(option, "stats")) {
3311 show_stats = 1;
3312 } else {
3313 return 0;
3316 return 1;
3319 static int parse_one_feature(const char *feature, int from_stream)
3321 const char *arg;
3323 if (skip_prefix(feature, "date-format=", &arg)) {
3324 option_date_format(arg);
3325 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3326 option_import_marks(arg, from_stream, 0);
3327 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3328 option_import_marks(arg, from_stream, 1);
3329 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3330 option_export_marks(arg);
3331 } else if (!strcmp(feature, "get-mark")) {
3332 ; /* Don't die - this feature is supported */
3333 } else if (!strcmp(feature, "cat-blob")) {
3334 ; /* Don't die - this feature is supported */
3335 } else if (!strcmp(feature, "relative-marks")) {
3336 relative_marks_paths = 1;
3337 } else if (!strcmp(feature, "no-relative-marks")) {
3338 relative_marks_paths = 0;
3339 } else if (!strcmp(feature, "done")) {
3340 require_explicit_termination = 1;
3341 } else if (!strcmp(feature, "force")) {
3342 force_update = 1;
3343 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3344 ; /* do nothing; we have the feature */
3345 } else {
3346 return 0;
3349 return 1;
3352 static void parse_feature(const char *feature)
3354 if (seen_data_command)
3355 die("Got feature command '%s' after data command", feature);
3357 if (parse_one_feature(feature, 1))
3358 return;
3360 die("This version of fast-import does not support feature %s.", feature);
3363 static void parse_option(const char *option)
3365 if (seen_data_command)
3366 die("Got option command '%s' after data command", option);
3368 if (parse_one_option(option))
3369 return;
3371 die("This version of fast-import does not support option: %s", option);
3374 static void git_pack_config(void)
3376 int indexversion_value;
3377 int limit;
3378 unsigned long packsizelimit_value;
3380 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3381 if (max_depth > MAX_DEPTH)
3382 max_depth = MAX_DEPTH;
3384 if (!git_config_get_int("pack.compression", &pack_compression_level)) {
3385 if (pack_compression_level == -1)
3386 pack_compression_level = Z_DEFAULT_COMPRESSION;
3387 else if (pack_compression_level < 0 ||
3388 pack_compression_level > Z_BEST_COMPRESSION)
3389 git_die_config("pack.compression",
3390 "bad pack compression level %d", pack_compression_level);
3391 pack_compression_seen = 1;
3393 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3394 pack_idx_opts.version = indexversion_value;
3395 if (pack_idx_opts.version > 2)
3396 git_die_config("pack.indexversion",
3397 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3399 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3400 max_packsize = packsizelimit_value;
3402 if (!git_config_get_int("fastimport.unpacklimit", &limit))
3403 unpack_limit = limit;
3404 else if (!git_config_get_int("transfer.unpacklimit", &limit))
3405 unpack_limit = limit;
3407 git_config(git_default_config, NULL);
3410 static const char fast_import_usage[] =
3411 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3413 static void parse_argv(void)
3415 unsigned int i;
3417 for (i = 1; i < global_argc; i++) {
3418 const char *a = global_argv[i];
3420 if (*a != '-' || !strcmp(a, "--"))
3421 break;
3423 if (!skip_prefix(a, "--", &a))
3424 die("unknown option %s", a);
3426 if (parse_one_option(a))
3427 continue;
3429 if (parse_one_feature(a, 0))
3430 continue;
3432 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3433 option_cat_blob_fd(a);
3434 continue;
3437 die("unknown option --%s", a);
3439 if (i != global_argc)
3440 usage(fast_import_usage);
3442 seen_data_command = 1;
3443 if (import_marks_file)
3444 read_marks();
3447 int cmd_main(int argc, const char **argv)
3449 unsigned int i;
3451 if (argc == 2 && !strcmp(argv[1], "-h"))
3452 usage(fast_import_usage);
3454 setup_git_directory();
3455 reset_pack_idx_option(&pack_idx_opts);
3456 git_pack_config();
3457 if (!pack_compression_seen && core_compression_seen)
3458 pack_compression_level = core_compression_level;
3460 alloc_objects(object_entry_alloc);
3461 strbuf_init(&command_buf, 0);
3462 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3463 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3464 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3465 marks = pool_calloc(1, sizeof(struct mark_set));
3467 global_argc = argc;
3468 global_argv = argv;
3470 rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3471 for (i = 0; i < (cmd_save - 1); i++)
3472 rc_free[i].next = &rc_free[i + 1];
3473 rc_free[cmd_save - 1].next = NULL;
3475 prepare_packed_git();
3476 start_packfile();
3477 set_die_routine(die_nicely);
3478 set_checkpoint_signal();
3479 while (read_next_command() != EOF) {
3480 const char *v;
3481 if (!strcmp("blob", command_buf.buf))
3482 parse_new_blob();
3483 else if (skip_prefix(command_buf.buf, "ls ", &v))
3484 parse_ls(v, NULL);
3485 else if (skip_prefix(command_buf.buf, "commit ", &v))
3486 parse_new_commit(v);
3487 else if (skip_prefix(command_buf.buf, "tag ", &v))
3488 parse_new_tag(v);
3489 else if (skip_prefix(command_buf.buf, "reset ", &v))
3490 parse_reset_branch(v);
3491 else if (!strcmp("checkpoint", command_buf.buf))
3492 parse_checkpoint();
3493 else if (!strcmp("done", command_buf.buf))
3494 break;
3495 else if (starts_with(command_buf.buf, "progress "))
3496 parse_progress();
3497 else if (skip_prefix(command_buf.buf, "feature ", &v))
3498 parse_feature(v);
3499 else if (skip_prefix(command_buf.buf, "option git ", &v))
3500 parse_option(v);
3501 else if (starts_with(command_buf.buf, "option "))
3502 /* ignore non-git options*/;
3503 else
3504 die("Unsupported command: %s", command_buf.buf);
3506 if (checkpoint_requested)
3507 checkpoint();
3510 /* argv hasn't been parsed yet, do so */
3511 if (!seen_data_command)
3512 parse_argv();
3514 if (require_explicit_termination && feof(stdin))
3515 die("stream ends early");
3517 end_packfile();
3519 dump_branches();
3520 dump_tags();
3521 unkeep_all_packs();
3522 dump_marks();
3524 if (pack_edges)
3525 fclose(pack_edges);
3527 if (show_stats) {
3528 uintmax_t total_count = 0, duplicate_count = 0;
3529 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3530 total_count += object_count_by_type[i];
3531 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3532 duplicate_count += duplicate_count_by_type[i];
3534 fprintf(stderr, "%s statistics:\n", argv[0]);
3535 fprintf(stderr, "---------------------------------------------------------------------\n");
3536 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3537 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3538 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]);
3539 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]);
3540 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]);
3541 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]);
3542 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3543 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3544 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3545 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3546 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)(total_allocd/1024));
3547 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3548 fprintf(stderr, "---------------------------------------------------------------------\n");
3549 pack_report();
3550 fprintf(stderr, "---------------------------------------------------------------------\n");
3551 fprintf(stderr, "\n");
3554 return failure ? 1 : 0;