convert trivial cases to ALLOC_ARRAY
[git/debian.git] / fast-import.c
blob36dbaa5b16513434fdf0729777d16e621b289c8f
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 "exec_cmd.h"
168 #include "dir.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 force_update;
286 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
287 static int pack_compression_seen;
289 /* Stats and misc. counters */
290 static uintmax_t alloc_count;
291 static uintmax_t marks_set_count;
292 static uintmax_t object_count_by_type[1 << TYPE_BITS];
293 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
294 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
295 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
296 static unsigned long object_count;
297 static unsigned long branch_count;
298 static unsigned long branch_load_count;
299 static int failure;
300 static FILE *pack_edges;
301 static unsigned int show_stats = 1;
302 static int global_argc;
303 static char **global_argv;
305 /* Memory pools */
306 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
307 static size_t total_allocd;
308 static struct mem_pool *mem_pool;
310 /* Atom management */
311 static unsigned int atom_table_sz = 4451;
312 static unsigned int atom_cnt;
313 static struct atom_str **atom_table;
315 /* The .pack file being generated */
316 static struct pack_idx_option pack_idx_opts;
317 static unsigned int pack_id;
318 static struct sha1file *pack_file;
319 static struct packed_git *pack_data;
320 static struct packed_git **all_packs;
321 static off_t pack_size;
323 /* Table of objects we've written. */
324 static unsigned int object_entry_alloc = 5000;
325 static struct object_entry_pool *blocks;
326 static struct object_entry *object_table[1 << 16];
327 static struct mark_set *marks;
328 static const char *export_marks_file;
329 static const char *import_marks_file;
330 static int import_marks_file_from_stream;
331 static int import_marks_file_ignore_missing;
332 static int relative_marks_paths;
334 /* Our last blob */
335 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
337 /* Tree management */
338 static unsigned int tree_entry_alloc = 1000;
339 static void *avail_tree_entry;
340 static unsigned int avail_tree_table_sz = 100;
341 static struct avail_tree_content **avail_tree_table;
342 static struct strbuf old_tree = STRBUF_INIT;
343 static struct strbuf new_tree = STRBUF_INIT;
345 /* Branch data */
346 static unsigned long max_active_branches = 5;
347 static unsigned long cur_active_branches;
348 static unsigned long branch_table_sz = 1039;
349 static struct branch **branch_table;
350 static struct branch *active_branches;
352 /* Tag data */
353 static struct tag *first_tag;
354 static struct tag *last_tag;
356 /* Input stream parsing */
357 static whenspec_type whenspec = WHENSPEC_RAW;
358 static struct strbuf command_buf = STRBUF_INIT;
359 static int unread_command_buf;
360 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
361 static struct recent_command *cmd_tail = &cmd_hist;
362 static struct recent_command *rc_free;
363 static unsigned int cmd_save = 100;
364 static uintmax_t next_mark;
365 static struct strbuf new_data = STRBUF_INIT;
366 static int seen_data_command;
367 static int require_explicit_termination;
369 /* Signal handling */
370 static volatile sig_atomic_t checkpoint_requested;
372 /* Where to write output of cat-blob commands */
373 static int cat_blob_fd = STDOUT_FILENO;
375 static void parse_argv(void);
376 static void parse_get_mark(const char *p);
377 static void parse_cat_blob(const char *p);
378 static void parse_ls(const char *p, struct branch *b);
380 static void write_branch_report(FILE *rpt, struct branch *b)
382 fprintf(rpt, "%s:\n", b->name);
384 fprintf(rpt, " status :");
385 if (b->active)
386 fputs(" active", rpt);
387 if (b->branch_tree.tree)
388 fputs(" loaded", rpt);
389 if (is_null_sha1(b->branch_tree.versions[1].sha1))
390 fputs(" dirty", rpt);
391 fputc('\n', rpt);
393 fprintf(rpt, " tip commit : %s\n", sha1_to_hex(b->sha1));
394 fprintf(rpt, " old tree : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
395 fprintf(rpt, " cur tree : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
396 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
398 fputs(" last pack : ", rpt);
399 if (b->pack_id < MAX_PACK_ID)
400 fprintf(rpt, "%u", b->pack_id);
401 fputc('\n', rpt);
403 fputc('\n', rpt);
406 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
408 static void write_crash_report(const char *err)
410 char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
411 FILE *rpt = fopen(loc, "w");
412 struct branch *b;
413 unsigned long lu;
414 struct recent_command *rc;
416 if (!rpt) {
417 error("can't write crash report %s: %s", loc, strerror(errno));
418 free(loc);
419 return;
422 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
424 fprintf(rpt, "fast-import crash report:\n");
425 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
426 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
427 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
428 fputc('\n', rpt);
430 fputs("fatal: ", rpt);
431 fputs(err, rpt);
432 fputc('\n', rpt);
434 fputc('\n', rpt);
435 fputs("Most Recent Commands Before Crash\n", rpt);
436 fputs("---------------------------------\n", rpt);
437 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
438 if (rc->next == &cmd_hist)
439 fputs("* ", rpt);
440 else
441 fputs(" ", rpt);
442 fputs(rc->buf, rpt);
443 fputc('\n', rpt);
446 fputc('\n', rpt);
447 fputs("Active Branch LRU\n", rpt);
448 fputs("-----------------\n", rpt);
449 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
450 cur_active_branches,
451 max_active_branches);
452 fputc('\n', rpt);
453 fputs(" pos clock name\n", rpt);
454 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
455 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
456 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
457 ++lu, b->last_commit, b->name);
459 fputc('\n', rpt);
460 fputs("Inactive Branches\n", rpt);
461 fputs("-----------------\n", rpt);
462 for (lu = 0; lu < branch_table_sz; lu++) {
463 for (b = branch_table[lu]; b; b = b->table_next_branch)
464 write_branch_report(rpt, b);
467 if (first_tag) {
468 struct tag *tg;
469 fputc('\n', rpt);
470 fputs("Annotated Tags\n", rpt);
471 fputs("--------------\n", rpt);
472 for (tg = first_tag; tg; tg = tg->next_tag) {
473 fputs(sha1_to_hex(tg->sha1), rpt);
474 fputc(' ', rpt);
475 fputs(tg->name, rpt);
476 fputc('\n', rpt);
480 fputc('\n', rpt);
481 fputs("Marks\n", rpt);
482 fputs("-----\n", rpt);
483 if (export_marks_file)
484 fprintf(rpt, " exported to %s\n", export_marks_file);
485 else
486 dump_marks_helper(rpt, 0, marks);
488 fputc('\n', rpt);
489 fputs("-------------------\n", rpt);
490 fputs("END OF CRASH REPORT\n", rpt);
491 fclose(rpt);
492 free(loc);
495 static void end_packfile(void);
496 static void unkeep_all_packs(void);
497 static void dump_marks(void);
499 static NORETURN void die_nicely(const char *err, va_list params)
501 static int zombie;
502 char message[2 * PATH_MAX];
504 vsnprintf(message, sizeof(message), err, params);
505 fputs("fatal: ", stderr);
506 fputs(message, stderr);
507 fputc('\n', stderr);
509 if (!zombie) {
510 zombie = 1;
511 write_crash_report(message);
512 end_packfile();
513 unkeep_all_packs();
514 dump_marks();
516 exit(128);
519 #ifndef SIGUSR1 /* Windows, for example */
521 static void set_checkpoint_signal(void)
525 #else
527 static void checkpoint_signal(int signo)
529 checkpoint_requested = 1;
532 static void set_checkpoint_signal(void)
534 struct sigaction sa;
536 memset(&sa, 0, sizeof(sa));
537 sa.sa_handler = checkpoint_signal;
538 sigemptyset(&sa.sa_mask);
539 sa.sa_flags = SA_RESTART;
540 sigaction(SIGUSR1, &sa, NULL);
543 #endif
545 static void alloc_objects(unsigned int cnt)
547 struct object_entry_pool *b;
549 b = xmalloc(sizeof(struct object_entry_pool)
550 + cnt * sizeof(struct object_entry));
551 b->next_pool = blocks;
552 b->next_free = b->entries;
553 b->end = b->entries + cnt;
554 blocks = b;
555 alloc_count += cnt;
558 static struct object_entry *new_object(unsigned char *sha1)
560 struct object_entry *e;
562 if (blocks->next_free == blocks->end)
563 alloc_objects(object_entry_alloc);
565 e = blocks->next_free++;
566 hashcpy(e->idx.sha1, sha1);
567 return e;
570 static struct object_entry *find_object(unsigned char *sha1)
572 unsigned int h = sha1[0] << 8 | sha1[1];
573 struct object_entry *e;
574 for (e = object_table[h]; e; e = e->next)
575 if (!hashcmp(sha1, e->idx.sha1))
576 return e;
577 return NULL;
580 static struct object_entry *insert_object(unsigned char *sha1)
582 unsigned int h = sha1[0] << 8 | sha1[1];
583 struct object_entry *e = object_table[h];
585 while (e) {
586 if (!hashcmp(sha1, e->idx.sha1))
587 return e;
588 e = e->next;
591 e = new_object(sha1);
592 e->next = object_table[h];
593 e->idx.offset = 0;
594 object_table[h] = e;
595 return e;
598 static unsigned int hc_str(const char *s, size_t len)
600 unsigned int r = 0;
601 while (len-- > 0)
602 r = r * 31 + *s++;
603 return r;
606 static void *pool_alloc(size_t len)
608 struct mem_pool *p;
609 void *r;
611 /* round up to a 'uintmax_t' alignment */
612 if (len & (sizeof(uintmax_t) - 1))
613 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
615 for (p = mem_pool; p; p = p->next_pool)
616 if ((p->end - p->next_free >= len))
617 break;
619 if (!p) {
620 if (len >= (mem_pool_alloc/2)) {
621 total_allocd += len;
622 return xmalloc(len);
624 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
625 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
626 p->next_pool = mem_pool;
627 p->next_free = (char *) p->space;
628 p->end = p->next_free + mem_pool_alloc;
629 mem_pool = p;
632 r = p->next_free;
633 p->next_free += len;
634 return r;
637 static void *pool_calloc(size_t count, size_t size)
639 size_t len = count * size;
640 void *r = pool_alloc(len);
641 memset(r, 0, len);
642 return r;
645 static char *pool_strdup(const char *s)
647 size_t len = strlen(s) + 1;
648 char *r = pool_alloc(len);
649 memcpy(r, s, len);
650 return r;
653 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
655 struct mark_set *s = marks;
656 while ((idnum >> s->shift) >= 1024) {
657 s = pool_calloc(1, sizeof(struct mark_set));
658 s->shift = marks->shift + 10;
659 s->data.sets[0] = marks;
660 marks = s;
662 while (s->shift) {
663 uintmax_t i = idnum >> s->shift;
664 idnum -= i << s->shift;
665 if (!s->data.sets[i]) {
666 s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
667 s->data.sets[i]->shift = s->shift - 10;
669 s = s->data.sets[i];
671 if (!s->data.marked[idnum])
672 marks_set_count++;
673 s->data.marked[idnum] = oe;
676 static struct object_entry *find_mark(uintmax_t idnum)
678 uintmax_t orig_idnum = idnum;
679 struct mark_set *s = marks;
680 struct object_entry *oe = NULL;
681 if ((idnum >> s->shift) < 1024) {
682 while (s && s->shift) {
683 uintmax_t i = idnum >> s->shift;
684 idnum -= i << s->shift;
685 s = s->data.sets[i];
687 if (s)
688 oe = s->data.marked[idnum];
690 if (!oe)
691 die("mark :%" PRIuMAX " not declared", orig_idnum);
692 return oe;
695 static struct atom_str *to_atom(const char *s, unsigned short len)
697 unsigned int hc = hc_str(s, len) % atom_table_sz;
698 struct atom_str *c;
700 for (c = atom_table[hc]; c; c = c->next_atom)
701 if (c->str_len == len && !strncmp(s, c->str_dat, len))
702 return c;
704 c = pool_alloc(sizeof(struct atom_str) + len + 1);
705 c->str_len = len;
706 memcpy(c->str_dat, s, len);
707 c->str_dat[len] = 0;
708 c->next_atom = atom_table[hc];
709 atom_table[hc] = c;
710 atom_cnt++;
711 return c;
714 static struct branch *lookup_branch(const char *name)
716 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
717 struct branch *b;
719 for (b = branch_table[hc]; b; b = b->table_next_branch)
720 if (!strcmp(name, b->name))
721 return b;
722 return NULL;
725 static struct branch *new_branch(const char *name)
727 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
728 struct branch *b = lookup_branch(name);
730 if (b)
731 die("Invalid attempt to create duplicate branch: %s", name);
732 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
733 die("Branch name doesn't conform to GIT standards: %s", name);
735 b = pool_calloc(1, sizeof(struct branch));
736 b->name = pool_strdup(name);
737 b->table_next_branch = branch_table[hc];
738 b->branch_tree.versions[0].mode = S_IFDIR;
739 b->branch_tree.versions[1].mode = S_IFDIR;
740 b->num_notes = 0;
741 b->active = 0;
742 b->pack_id = MAX_PACK_ID;
743 branch_table[hc] = b;
744 branch_count++;
745 return b;
748 static unsigned int hc_entries(unsigned int cnt)
750 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
751 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
754 static struct tree_content *new_tree_content(unsigned int cnt)
756 struct avail_tree_content *f, *l = NULL;
757 struct tree_content *t;
758 unsigned int hc = hc_entries(cnt);
760 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
761 if (f->entry_capacity >= cnt)
762 break;
764 if (f) {
765 if (l)
766 l->next_avail = f->next_avail;
767 else
768 avail_tree_table[hc] = f->next_avail;
769 } else {
770 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
771 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
772 f->entry_capacity = cnt;
775 t = (struct tree_content*)f;
776 t->entry_count = 0;
777 t->delta_depth = 0;
778 return t;
781 static void release_tree_entry(struct tree_entry *e);
782 static void release_tree_content(struct tree_content *t)
784 struct avail_tree_content *f = (struct avail_tree_content*)t;
785 unsigned int hc = hc_entries(f->entry_capacity);
786 f->next_avail = avail_tree_table[hc];
787 avail_tree_table[hc] = f;
790 static void release_tree_content_recursive(struct tree_content *t)
792 unsigned int i;
793 for (i = 0; i < t->entry_count; i++)
794 release_tree_entry(t->entries[i]);
795 release_tree_content(t);
798 static struct tree_content *grow_tree_content(
799 struct tree_content *t,
800 int amt)
802 struct tree_content *r = new_tree_content(t->entry_count + amt);
803 r->entry_count = t->entry_count;
804 r->delta_depth = t->delta_depth;
805 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
806 release_tree_content(t);
807 return r;
810 static struct tree_entry *new_tree_entry(void)
812 struct tree_entry *e;
814 if (!avail_tree_entry) {
815 unsigned int n = tree_entry_alloc;
816 total_allocd += n * sizeof(struct tree_entry);
817 ALLOC_ARRAY(e, n);
818 avail_tree_entry = e;
819 while (n-- > 1) {
820 *((void**)e) = e + 1;
821 e++;
823 *((void**)e) = NULL;
826 e = avail_tree_entry;
827 avail_tree_entry = *((void**)e);
828 return e;
831 static void release_tree_entry(struct tree_entry *e)
833 if (e->tree)
834 release_tree_content_recursive(e->tree);
835 *((void**)e) = avail_tree_entry;
836 avail_tree_entry = e;
839 static struct tree_content *dup_tree_content(struct tree_content *s)
841 struct tree_content *d;
842 struct tree_entry *a, *b;
843 unsigned int i;
845 if (!s)
846 return NULL;
847 d = new_tree_content(s->entry_count);
848 for (i = 0; i < s->entry_count; i++) {
849 a = s->entries[i];
850 b = new_tree_entry();
851 memcpy(b, a, sizeof(*a));
852 if (a->tree && is_null_sha1(b->versions[1].sha1))
853 b->tree = dup_tree_content(a->tree);
854 else
855 b->tree = NULL;
856 d->entries[i] = b;
858 d->entry_count = s->entry_count;
859 d->delta_depth = s->delta_depth;
861 return d;
864 static void start_packfile(void)
866 static char tmp_file[PATH_MAX];
867 struct packed_git *p;
868 int namelen;
869 struct pack_header hdr;
870 int pack_fd;
872 pack_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
873 "pack/tmp_pack_XXXXXX");
874 namelen = strlen(tmp_file) + 2;
875 p = xcalloc(1, sizeof(*p) + namelen);
876 xsnprintf(p->pack_name, namelen, "%s", tmp_file);
877 p->pack_fd = pack_fd;
878 p->do_not_close = 1;
879 pack_file = sha1fd(pack_fd, p->pack_name);
881 hdr.hdr_signature = htonl(PACK_SIGNATURE);
882 hdr.hdr_version = htonl(2);
883 hdr.hdr_entries = 0;
884 sha1write(pack_file, &hdr, sizeof(hdr));
886 pack_data = p;
887 pack_size = sizeof(hdr);
888 object_count = 0;
890 REALLOC_ARRAY(all_packs, pack_id + 1);
891 all_packs[pack_id] = p;
894 static const char *create_index(void)
896 const char *tmpfile;
897 struct pack_idx_entry **idx, **c, **last;
898 struct object_entry *e;
899 struct object_entry_pool *o;
901 /* Build the table of object IDs. */
902 ALLOC_ARRAY(idx, object_count);
903 c = idx;
904 for (o = blocks; o; o = o->next_pool)
905 for (e = o->next_free; e-- != o->entries;)
906 if (pack_id == e->pack_id)
907 *c++ = &e->idx;
908 last = idx + object_count;
909 if (c != last)
910 die("internal consistency error creating the index");
912 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
913 free(idx);
914 return tmpfile;
917 static char *keep_pack(const char *curr_index_name)
919 static char name[PATH_MAX];
920 static const char *keep_msg = "fast-import";
921 int keep_fd;
923 keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
924 if (keep_fd < 0)
925 die_errno("cannot create keep file");
926 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
927 if (close(keep_fd))
928 die_errno("failed to write keep file");
930 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
931 get_object_directory(), sha1_to_hex(pack_data->sha1));
932 if (finalize_object_file(pack_data->pack_name, name))
933 die("cannot store pack file");
935 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
936 get_object_directory(), sha1_to_hex(pack_data->sha1));
937 if (finalize_object_file(curr_index_name, name))
938 die("cannot store index file");
939 free((void *)curr_index_name);
940 return name;
943 static void unkeep_all_packs(void)
945 static char name[PATH_MAX];
946 int k;
948 for (k = 0; k < pack_id; k++) {
949 struct packed_git *p = all_packs[k];
950 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
951 get_object_directory(), sha1_to_hex(p->sha1));
952 unlink_or_warn(name);
956 static void end_packfile(void)
958 static int running;
960 if (running || !pack_data)
961 return;
963 running = 1;
964 clear_delta_base_cache();
965 if (object_count) {
966 struct packed_git *new_p;
967 unsigned char cur_pack_sha1[20];
968 char *idx_name;
969 int i;
970 struct branch *b;
971 struct tag *t;
973 close_pack_windows(pack_data);
974 sha1close(pack_file, cur_pack_sha1, 0);
975 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
976 pack_data->pack_name, object_count,
977 cur_pack_sha1, pack_size);
978 close(pack_data->pack_fd);
979 idx_name = keep_pack(create_index());
981 /* Register the packfile with core git's machinery. */
982 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
983 if (!new_p)
984 die("core git rejected index %s", idx_name);
985 all_packs[pack_id] = new_p;
986 install_packed_git(new_p);
988 /* Print the boundary */
989 if (pack_edges) {
990 fprintf(pack_edges, "%s:", new_p->pack_name);
991 for (i = 0; i < branch_table_sz; i++) {
992 for (b = branch_table[i]; b; b = b->table_next_branch) {
993 if (b->pack_id == pack_id)
994 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
997 for (t = first_tag; t; t = t->next_tag) {
998 if (t->pack_id == pack_id)
999 fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
1001 fputc('\n', pack_edges);
1002 fflush(pack_edges);
1005 pack_id++;
1007 else {
1008 close(pack_data->pack_fd);
1009 unlink_or_warn(pack_data->pack_name);
1011 free(pack_data);
1012 pack_data = NULL;
1013 running = 0;
1015 /* We can't carry a delta across packfiles. */
1016 strbuf_release(&last_blob.data);
1017 last_blob.offset = 0;
1018 last_blob.depth = 0;
1021 static void cycle_packfile(void)
1023 end_packfile();
1024 start_packfile();
1027 static int store_object(
1028 enum object_type type,
1029 struct strbuf *dat,
1030 struct last_object *last,
1031 unsigned char *sha1out,
1032 uintmax_t mark)
1034 void *out, *delta;
1035 struct object_entry *e;
1036 unsigned char hdr[96];
1037 unsigned char sha1[20];
1038 unsigned long hdrlen, deltalen;
1039 git_SHA_CTX c;
1040 git_zstream s;
1042 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1043 typename(type), (unsigned long)dat->len) + 1;
1044 git_SHA1_Init(&c);
1045 git_SHA1_Update(&c, hdr, hdrlen);
1046 git_SHA1_Update(&c, dat->buf, dat->len);
1047 git_SHA1_Final(sha1, &c);
1048 if (sha1out)
1049 hashcpy(sha1out, sha1);
1051 e = insert_object(sha1);
1052 if (mark)
1053 insert_mark(mark, e);
1054 if (e->idx.offset) {
1055 duplicate_count_by_type[type]++;
1056 return 1;
1057 } else if (find_sha1_pack(sha1, packed_git)) {
1058 e->type = type;
1059 e->pack_id = MAX_PACK_ID;
1060 e->idx.offset = 1; /* just not zero! */
1061 duplicate_count_by_type[type]++;
1062 return 1;
1065 if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1066 delta_count_attempts_by_type[type]++;
1067 delta = diff_delta(last->data.buf, last->data.len,
1068 dat->buf, dat->len,
1069 &deltalen, dat->len - 20);
1070 } else
1071 delta = NULL;
1073 git_deflate_init(&s, pack_compression_level);
1074 if (delta) {
1075 s.next_in = delta;
1076 s.avail_in = deltalen;
1077 } else {
1078 s.next_in = (void *)dat->buf;
1079 s.avail_in = dat->len;
1081 s.avail_out = git_deflate_bound(&s, s.avail_in);
1082 s.next_out = out = xmalloc(s.avail_out);
1083 while (git_deflate(&s, Z_FINISH) == Z_OK)
1084 ; /* nothing */
1085 git_deflate_end(&s);
1087 /* Determine if we should auto-checkpoint. */
1088 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1089 || (pack_size + 60 + s.total_out) < pack_size) {
1091 /* This new object needs to *not* have the current pack_id. */
1092 e->pack_id = pack_id + 1;
1093 cycle_packfile();
1095 /* We cannot carry a delta into the new pack. */
1096 if (delta) {
1097 free(delta);
1098 delta = NULL;
1100 git_deflate_init(&s, pack_compression_level);
1101 s.next_in = (void *)dat->buf;
1102 s.avail_in = dat->len;
1103 s.avail_out = git_deflate_bound(&s, s.avail_in);
1104 s.next_out = out = xrealloc(out, s.avail_out);
1105 while (git_deflate(&s, Z_FINISH) == Z_OK)
1106 ; /* nothing */
1107 git_deflate_end(&s);
1111 e->type = type;
1112 e->pack_id = pack_id;
1113 e->idx.offset = pack_size;
1114 object_count++;
1115 object_count_by_type[type]++;
1117 crc32_begin(pack_file);
1119 if (delta) {
1120 off_t ofs = e->idx.offset - last->offset;
1121 unsigned pos = sizeof(hdr) - 1;
1123 delta_count_by_type[type]++;
1124 e->depth = last->depth + 1;
1126 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1127 sha1write(pack_file, hdr, hdrlen);
1128 pack_size += hdrlen;
1130 hdr[pos] = ofs & 127;
1131 while (ofs >>= 7)
1132 hdr[--pos] = 128 | (--ofs & 127);
1133 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1134 pack_size += sizeof(hdr) - pos;
1135 } else {
1136 e->depth = 0;
1137 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1138 sha1write(pack_file, hdr, hdrlen);
1139 pack_size += hdrlen;
1142 sha1write(pack_file, out, s.total_out);
1143 pack_size += s.total_out;
1145 e->idx.crc32 = crc32_end(pack_file);
1147 free(out);
1148 free(delta);
1149 if (last) {
1150 if (last->no_swap) {
1151 last->data = *dat;
1152 } else {
1153 strbuf_swap(&last->data, dat);
1155 last->offset = e->idx.offset;
1156 last->depth = e->depth;
1158 return 0;
1161 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1163 if (sha1file_truncate(pack_file, checkpoint))
1164 die_errno("cannot truncate pack to skip duplicate");
1165 pack_size = checkpoint->offset;
1168 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1170 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1171 unsigned char *in_buf = xmalloc(in_sz);
1172 unsigned char *out_buf = xmalloc(out_sz);
1173 struct object_entry *e;
1174 unsigned char sha1[20];
1175 unsigned long hdrlen;
1176 off_t offset;
1177 git_SHA_CTX c;
1178 git_zstream s;
1179 struct sha1file_checkpoint checkpoint;
1180 int status = Z_OK;
1182 /* Determine if we should auto-checkpoint. */
1183 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1184 || (pack_size + 60 + len) < pack_size)
1185 cycle_packfile();
1187 sha1file_checkpoint(pack_file, &checkpoint);
1188 offset = checkpoint.offset;
1190 hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1191 if (out_sz <= hdrlen)
1192 die("impossibly large object header");
1194 git_SHA1_Init(&c);
1195 git_SHA1_Update(&c, out_buf, hdrlen);
1197 crc32_begin(pack_file);
1199 git_deflate_init(&s, pack_compression_level);
1201 hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1202 if (out_sz <= hdrlen)
1203 die("impossibly large object header");
1205 s.next_out = out_buf + hdrlen;
1206 s.avail_out = out_sz - hdrlen;
1208 while (status != Z_STREAM_END) {
1209 if (0 < len && !s.avail_in) {
1210 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1211 size_t n = fread(in_buf, 1, cnt, stdin);
1212 if (!n && feof(stdin))
1213 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1215 git_SHA1_Update(&c, in_buf, n);
1216 s.next_in = in_buf;
1217 s.avail_in = n;
1218 len -= n;
1221 status = git_deflate(&s, len ? 0 : Z_FINISH);
1223 if (!s.avail_out || status == Z_STREAM_END) {
1224 size_t n = s.next_out - out_buf;
1225 sha1write(pack_file, out_buf, n);
1226 pack_size += n;
1227 s.next_out = out_buf;
1228 s.avail_out = out_sz;
1231 switch (status) {
1232 case Z_OK:
1233 case Z_BUF_ERROR:
1234 case Z_STREAM_END:
1235 continue;
1236 default:
1237 die("unexpected deflate failure: %d", status);
1240 git_deflate_end(&s);
1241 git_SHA1_Final(sha1, &c);
1243 if (sha1out)
1244 hashcpy(sha1out, sha1);
1246 e = insert_object(sha1);
1248 if (mark)
1249 insert_mark(mark, e);
1251 if (e->idx.offset) {
1252 duplicate_count_by_type[OBJ_BLOB]++;
1253 truncate_pack(&checkpoint);
1255 } else if (find_sha1_pack(sha1, packed_git)) {
1256 e->type = OBJ_BLOB;
1257 e->pack_id = MAX_PACK_ID;
1258 e->idx.offset = 1; /* just not zero! */
1259 duplicate_count_by_type[OBJ_BLOB]++;
1260 truncate_pack(&checkpoint);
1262 } else {
1263 e->depth = 0;
1264 e->type = OBJ_BLOB;
1265 e->pack_id = pack_id;
1266 e->idx.offset = offset;
1267 e->idx.crc32 = crc32_end(pack_file);
1268 object_count++;
1269 object_count_by_type[OBJ_BLOB]++;
1272 free(in_buf);
1273 free(out_buf);
1276 /* All calls must be guarded by find_object() or find_mark() to
1277 * ensure the 'struct object_entry' passed was written by this
1278 * process instance. We unpack the entry by the offset, avoiding
1279 * the need for the corresponding .idx file. This unpacking rule
1280 * works because we only use OBJ_REF_DELTA within the packfiles
1281 * created by fast-import.
1283 * oe must not be NULL. Such an oe usually comes from giving
1284 * an unknown SHA-1 to find_object() or an undefined mark to
1285 * find_mark(). Callers must test for this condition and use
1286 * the standard read_sha1_file() when it happens.
1288 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1289 * find_mark(), where the mark was reloaded from an existing marks
1290 * file and is referencing an object that this fast-import process
1291 * instance did not write out to a packfile. Callers must test for
1292 * this condition and use read_sha1_file() instead.
1294 static void *gfi_unpack_entry(
1295 struct object_entry *oe,
1296 unsigned long *sizep)
1298 enum object_type type;
1299 struct packed_git *p = all_packs[oe->pack_id];
1300 if (p == pack_data && p->pack_size < (pack_size + 20)) {
1301 /* The object is stored in the packfile we are writing to
1302 * and we have modified it since the last time we scanned
1303 * back to read a previously written object. If an old
1304 * window covered [p->pack_size, p->pack_size + 20) its
1305 * data is stale and is not valid. Closing all windows
1306 * and updating the packfile length ensures we can read
1307 * the newly written data.
1309 close_pack_windows(p);
1310 sha1flush(pack_file);
1312 /* We have to offer 20 bytes additional on the end of
1313 * the packfile as the core unpacker code assumes the
1314 * footer is present at the file end and must promise
1315 * at least 20 bytes within any window it maps. But
1316 * we don't actually create the footer here.
1318 p->pack_size = pack_size + 20;
1320 return unpack_entry(p, oe->idx.offset, &type, sizep);
1323 static const char *get_mode(const char *str, uint16_t *modep)
1325 unsigned char c;
1326 uint16_t mode = 0;
1328 while ((c = *str++) != ' ') {
1329 if (c < '0' || c > '7')
1330 return NULL;
1331 mode = (mode << 3) + (c - '0');
1333 *modep = mode;
1334 return str;
1337 static void load_tree(struct tree_entry *root)
1339 unsigned char *sha1 = root->versions[1].sha1;
1340 struct object_entry *myoe;
1341 struct tree_content *t;
1342 unsigned long size;
1343 char *buf;
1344 const char *c;
1346 root->tree = t = new_tree_content(8);
1347 if (is_null_sha1(sha1))
1348 return;
1350 myoe = find_object(sha1);
1351 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1352 if (myoe->type != OBJ_TREE)
1353 die("Not a tree: %s", sha1_to_hex(sha1));
1354 t->delta_depth = myoe->depth;
1355 buf = gfi_unpack_entry(myoe, &size);
1356 if (!buf)
1357 die("Can't load tree %s", sha1_to_hex(sha1));
1358 } else {
1359 enum object_type type;
1360 buf = read_sha1_file(sha1, &type, &size);
1361 if (!buf || type != OBJ_TREE)
1362 die("Can't load tree %s", sha1_to_hex(sha1));
1365 c = buf;
1366 while (c != (buf + size)) {
1367 struct tree_entry *e = new_tree_entry();
1369 if (t->entry_count == t->entry_capacity)
1370 root->tree = t = grow_tree_content(t, t->entry_count);
1371 t->entries[t->entry_count++] = e;
1373 e->tree = NULL;
1374 c = get_mode(c, &e->versions[1].mode);
1375 if (!c)
1376 die("Corrupt mode in %s", sha1_to_hex(sha1));
1377 e->versions[0].mode = e->versions[1].mode;
1378 e->name = to_atom(c, strlen(c));
1379 c += e->name->str_len + 1;
1380 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1381 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1382 c += 20;
1384 free(buf);
1387 static int tecmp0 (const void *_a, const void *_b)
1389 struct tree_entry *a = *((struct tree_entry**)_a);
1390 struct tree_entry *b = *((struct tree_entry**)_b);
1391 return base_name_compare(
1392 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1393 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1396 static int tecmp1 (const void *_a, const void *_b)
1398 struct tree_entry *a = *((struct tree_entry**)_a);
1399 struct tree_entry *b = *((struct tree_entry**)_b);
1400 return base_name_compare(
1401 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1402 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1405 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1407 size_t maxlen = 0;
1408 unsigned int i;
1410 if (!v)
1411 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1412 else
1413 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1415 for (i = 0; i < t->entry_count; i++) {
1416 if (t->entries[i]->versions[v].mode)
1417 maxlen += t->entries[i]->name->str_len + 34;
1420 strbuf_reset(b);
1421 strbuf_grow(b, maxlen);
1422 for (i = 0; i < t->entry_count; i++) {
1423 struct tree_entry *e = t->entries[i];
1424 if (!e->versions[v].mode)
1425 continue;
1426 strbuf_addf(b, "%o %s%c",
1427 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1428 e->name->str_dat, '\0');
1429 strbuf_add(b, e->versions[v].sha1, 20);
1433 static void store_tree(struct tree_entry *root)
1435 struct tree_content *t;
1436 unsigned int i, j, del;
1437 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1438 struct object_entry *le = NULL;
1440 if (!is_null_sha1(root->versions[1].sha1))
1441 return;
1443 if (!root->tree)
1444 load_tree(root);
1445 t = root->tree;
1447 for (i = 0; i < t->entry_count; i++) {
1448 if (t->entries[i]->tree)
1449 store_tree(t->entries[i]);
1452 if (!(root->versions[0].mode & NO_DELTA))
1453 le = find_object(root->versions[0].sha1);
1454 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1455 mktree(t, 0, &old_tree);
1456 lo.data = old_tree;
1457 lo.offset = le->idx.offset;
1458 lo.depth = t->delta_depth;
1461 mktree(t, 1, &new_tree);
1462 store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1464 t->delta_depth = lo.depth;
1465 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1466 struct tree_entry *e = t->entries[i];
1467 if (e->versions[1].mode) {
1468 e->versions[0].mode = e->versions[1].mode;
1469 hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1470 t->entries[j++] = e;
1471 } else {
1472 release_tree_entry(e);
1473 del++;
1476 t->entry_count -= del;
1479 static void tree_content_replace(
1480 struct tree_entry *root,
1481 const unsigned char *sha1,
1482 const uint16_t mode,
1483 struct tree_content *newtree)
1485 if (!S_ISDIR(mode))
1486 die("Root cannot be a non-directory");
1487 hashclr(root->versions[0].sha1);
1488 hashcpy(root->versions[1].sha1, sha1);
1489 if (root->tree)
1490 release_tree_content_recursive(root->tree);
1491 root->tree = newtree;
1494 static int tree_content_set(
1495 struct tree_entry *root,
1496 const char *p,
1497 const unsigned char *sha1,
1498 const uint16_t mode,
1499 struct tree_content *subtree)
1501 struct tree_content *t;
1502 const char *slash1;
1503 unsigned int i, n;
1504 struct tree_entry *e;
1506 slash1 = strchrnul(p, '/');
1507 n = slash1 - p;
1508 if (!n)
1509 die("Empty path component found in input");
1510 if (!*slash1 && !S_ISDIR(mode) && subtree)
1511 die("Non-directories cannot have subtrees");
1513 if (!root->tree)
1514 load_tree(root);
1515 t = root->tree;
1516 for (i = 0; i < t->entry_count; i++) {
1517 e = t->entries[i];
1518 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1519 if (!*slash1) {
1520 if (!S_ISDIR(mode)
1521 && e->versions[1].mode == mode
1522 && !hashcmp(e->versions[1].sha1, sha1))
1523 return 0;
1524 e->versions[1].mode = mode;
1525 hashcpy(e->versions[1].sha1, sha1);
1526 if (e->tree)
1527 release_tree_content_recursive(e->tree);
1528 e->tree = subtree;
1531 * We need to leave e->versions[0].sha1 alone
1532 * to avoid modifying the preimage tree used
1533 * when writing out the parent directory.
1534 * But after replacing the subdir with a
1535 * completely different one, it's not a good
1536 * delta base any more, and besides, we've
1537 * thrown away the tree entries needed to
1538 * make a delta against it.
1540 * So let's just explicitly disable deltas
1541 * for the subtree.
1543 if (S_ISDIR(e->versions[0].mode))
1544 e->versions[0].mode |= NO_DELTA;
1546 hashclr(root->versions[1].sha1);
1547 return 1;
1549 if (!S_ISDIR(e->versions[1].mode)) {
1550 e->tree = new_tree_content(8);
1551 e->versions[1].mode = S_IFDIR;
1553 if (!e->tree)
1554 load_tree(e);
1555 if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1556 hashclr(root->versions[1].sha1);
1557 return 1;
1559 return 0;
1563 if (t->entry_count == t->entry_capacity)
1564 root->tree = t = grow_tree_content(t, t->entry_count);
1565 e = new_tree_entry();
1566 e->name = to_atom(p, n);
1567 e->versions[0].mode = 0;
1568 hashclr(e->versions[0].sha1);
1569 t->entries[t->entry_count++] = e;
1570 if (*slash1) {
1571 e->tree = new_tree_content(8);
1572 e->versions[1].mode = S_IFDIR;
1573 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1574 } else {
1575 e->tree = subtree;
1576 e->versions[1].mode = mode;
1577 hashcpy(e->versions[1].sha1, sha1);
1579 hashclr(root->versions[1].sha1);
1580 return 1;
1583 static int tree_content_remove(
1584 struct tree_entry *root,
1585 const char *p,
1586 struct tree_entry *backup_leaf,
1587 int allow_root)
1589 struct tree_content *t;
1590 const char *slash1;
1591 unsigned int i, n;
1592 struct tree_entry *e;
1594 slash1 = strchrnul(p, '/');
1595 n = slash1 - p;
1597 if (!root->tree)
1598 load_tree(root);
1600 if (!*p && allow_root) {
1601 e = root;
1602 goto del_entry;
1605 t = root->tree;
1606 for (i = 0; i < t->entry_count; i++) {
1607 e = t->entries[i];
1608 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1609 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1611 * If p names a file in some subdirectory, and a
1612 * file or symlink matching the name of the
1613 * parent directory of p exists, then p cannot
1614 * exist and need not be deleted.
1616 return 1;
1617 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1618 goto del_entry;
1619 if (!e->tree)
1620 load_tree(e);
1621 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1622 for (n = 0; n < e->tree->entry_count; n++) {
1623 if (e->tree->entries[n]->versions[1].mode) {
1624 hashclr(root->versions[1].sha1);
1625 return 1;
1628 backup_leaf = NULL;
1629 goto del_entry;
1631 return 0;
1634 return 0;
1636 del_entry:
1637 if (backup_leaf)
1638 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1639 else if (e->tree)
1640 release_tree_content_recursive(e->tree);
1641 e->tree = NULL;
1642 e->versions[1].mode = 0;
1643 hashclr(e->versions[1].sha1);
1644 hashclr(root->versions[1].sha1);
1645 return 1;
1648 static int tree_content_get(
1649 struct tree_entry *root,
1650 const char *p,
1651 struct tree_entry *leaf,
1652 int allow_root)
1654 struct tree_content *t;
1655 const char *slash1;
1656 unsigned int i, n;
1657 struct tree_entry *e;
1659 slash1 = strchrnul(p, '/');
1660 n = slash1 - p;
1661 if (!n && !allow_root)
1662 die("Empty path component found in input");
1664 if (!root->tree)
1665 load_tree(root);
1667 if (!n) {
1668 e = root;
1669 goto found_entry;
1672 t = root->tree;
1673 for (i = 0; i < t->entry_count; i++) {
1674 e = t->entries[i];
1675 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1676 if (!*slash1)
1677 goto found_entry;
1678 if (!S_ISDIR(e->versions[1].mode))
1679 return 0;
1680 if (!e->tree)
1681 load_tree(e);
1682 return tree_content_get(e, slash1 + 1, leaf, 0);
1685 return 0;
1687 found_entry:
1688 memcpy(leaf, e, sizeof(*leaf));
1689 if (e->tree && is_null_sha1(e->versions[1].sha1))
1690 leaf->tree = dup_tree_content(e->tree);
1691 else
1692 leaf->tree = NULL;
1693 return 1;
1696 static int update_branch(struct branch *b)
1698 static const char *msg = "fast-import";
1699 struct ref_transaction *transaction;
1700 unsigned char old_sha1[20];
1701 struct strbuf err = STRBUF_INIT;
1703 if (is_null_sha1(b->sha1)) {
1704 if (b->delete)
1705 delete_ref(b->name, NULL, 0);
1706 return 0;
1708 if (read_ref(b->name, old_sha1))
1709 hashclr(old_sha1);
1710 if (!force_update && !is_null_sha1(old_sha1)) {
1711 struct commit *old_cmit, *new_cmit;
1713 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1714 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1715 if (!old_cmit || !new_cmit)
1716 return error("Branch %s is missing commits.", b->name);
1718 if (!in_merge_bases(old_cmit, new_cmit)) {
1719 warning("Not updating %s"
1720 " (new tip %s does not contain %s)",
1721 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1722 return -1;
1725 transaction = ref_transaction_begin(&err);
1726 if (!transaction ||
1727 ref_transaction_update(transaction, b->name, b->sha1, old_sha1,
1728 0, msg, &err) ||
1729 ref_transaction_commit(transaction, &err)) {
1730 ref_transaction_free(transaction);
1731 error("%s", err.buf);
1732 strbuf_release(&err);
1733 return -1;
1735 ref_transaction_free(transaction);
1736 strbuf_release(&err);
1737 return 0;
1740 static void dump_branches(void)
1742 unsigned int i;
1743 struct branch *b;
1745 for (i = 0; i < branch_table_sz; i++) {
1746 for (b = branch_table[i]; b; b = b->table_next_branch)
1747 failure |= update_branch(b);
1751 static void dump_tags(void)
1753 static const char *msg = "fast-import";
1754 struct tag *t;
1755 struct strbuf ref_name = STRBUF_INIT;
1756 struct strbuf err = STRBUF_INIT;
1757 struct ref_transaction *transaction;
1759 transaction = ref_transaction_begin(&err);
1760 if (!transaction) {
1761 failure |= error("%s", err.buf);
1762 goto cleanup;
1764 for (t = first_tag; t; t = t->next_tag) {
1765 strbuf_reset(&ref_name);
1766 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1768 if (ref_transaction_update(transaction, ref_name.buf,
1769 t->sha1, NULL, 0, msg, &err)) {
1770 failure |= error("%s", err.buf);
1771 goto cleanup;
1774 if (ref_transaction_commit(transaction, &err))
1775 failure |= error("%s", err.buf);
1777 cleanup:
1778 ref_transaction_free(transaction);
1779 strbuf_release(&ref_name);
1780 strbuf_release(&err);
1783 static void dump_marks_helper(FILE *f,
1784 uintmax_t base,
1785 struct mark_set *m)
1787 uintmax_t k;
1788 if (m->shift) {
1789 for (k = 0; k < 1024; k++) {
1790 if (m->data.sets[k])
1791 dump_marks_helper(f, base + (k << m->shift),
1792 m->data.sets[k]);
1794 } else {
1795 for (k = 0; k < 1024; k++) {
1796 if (m->data.marked[k])
1797 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1798 sha1_to_hex(m->data.marked[k]->idx.sha1));
1803 static void dump_marks(void)
1805 static struct lock_file mark_lock;
1806 FILE *f;
1808 if (!export_marks_file)
1809 return;
1811 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1812 failure |= error("Unable to write marks file %s: %s",
1813 export_marks_file, strerror(errno));
1814 return;
1817 f = fdopen_lock_file(&mark_lock, "w");
1818 if (!f) {
1819 int saved_errno = errno;
1820 rollback_lock_file(&mark_lock);
1821 failure |= error("Unable to write marks file %s: %s",
1822 export_marks_file, strerror(saved_errno));
1823 return;
1826 dump_marks_helper(f, 0, marks);
1827 if (commit_lock_file(&mark_lock)) {
1828 failure |= error("Unable to write file %s: %s",
1829 export_marks_file, strerror(errno));
1830 return;
1834 static void read_marks(void)
1836 char line[512];
1837 FILE *f = fopen(import_marks_file, "r");
1838 if (f)
1840 else if (import_marks_file_ignore_missing && errno == ENOENT)
1841 return; /* Marks file does not exist */
1842 else
1843 die_errno("cannot read '%s'", import_marks_file);
1844 while (fgets(line, sizeof(line), f)) {
1845 uintmax_t mark;
1846 char *end;
1847 unsigned char sha1[20];
1848 struct object_entry *e;
1850 end = strchr(line, '\n');
1851 if (line[0] != ':' || !end)
1852 die("corrupt mark line: %s", line);
1853 *end = 0;
1854 mark = strtoumax(line + 1, &end, 10);
1855 if (!mark || end == line + 1
1856 || *end != ' ' || get_sha1_hex(end + 1, sha1))
1857 die("corrupt mark line: %s", line);
1858 e = find_object(sha1);
1859 if (!e) {
1860 enum object_type type = sha1_object_info(sha1, NULL);
1861 if (type < 0)
1862 die("object not found: %s", sha1_to_hex(sha1));
1863 e = insert_object(sha1);
1864 e->type = type;
1865 e->pack_id = MAX_PACK_ID;
1866 e->idx.offset = 1; /* just not zero! */
1868 insert_mark(mark, e);
1870 fclose(f);
1874 static int read_next_command(void)
1876 static int stdin_eof = 0;
1878 if (stdin_eof) {
1879 unread_command_buf = 0;
1880 return EOF;
1883 for (;;) {
1884 const char *p;
1886 if (unread_command_buf) {
1887 unread_command_buf = 0;
1888 } else {
1889 struct recent_command *rc;
1891 strbuf_detach(&command_buf, NULL);
1892 stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1893 if (stdin_eof)
1894 return EOF;
1896 if (!seen_data_command
1897 && !starts_with(command_buf.buf, "feature ")
1898 && !starts_with(command_buf.buf, "option ")) {
1899 parse_argv();
1902 rc = rc_free;
1903 if (rc)
1904 rc_free = rc->next;
1905 else {
1906 rc = cmd_hist.next;
1907 cmd_hist.next = rc->next;
1908 cmd_hist.next->prev = &cmd_hist;
1909 free(rc->buf);
1912 rc->buf = command_buf.buf;
1913 rc->prev = cmd_tail;
1914 rc->next = cmd_hist.prev;
1915 rc->prev->next = rc;
1916 cmd_tail = rc;
1918 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1919 parse_get_mark(p);
1920 continue;
1922 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1923 parse_cat_blob(p);
1924 continue;
1926 if (command_buf.buf[0] == '#')
1927 continue;
1928 return 0;
1932 static void skip_optional_lf(void)
1934 int term_char = fgetc(stdin);
1935 if (term_char != '\n' && term_char != EOF)
1936 ungetc(term_char, stdin);
1939 static void parse_mark(void)
1941 const char *v;
1942 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1943 next_mark = strtoumax(v, NULL, 10);
1944 read_next_command();
1946 else
1947 next_mark = 0;
1950 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1952 const char *data;
1953 strbuf_reset(sb);
1955 if (!skip_prefix(command_buf.buf, "data ", &data))
1956 die("Expected 'data n' command, found: %s", command_buf.buf);
1958 if (skip_prefix(data, "<<", &data)) {
1959 char *term = xstrdup(data);
1960 size_t term_len = command_buf.len - (data - command_buf.buf);
1962 strbuf_detach(&command_buf, NULL);
1963 for (;;) {
1964 if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1965 die("EOF in data (terminator '%s' not found)", term);
1966 if (term_len == command_buf.len
1967 && !strcmp(term, command_buf.buf))
1968 break;
1969 strbuf_addbuf(sb, &command_buf);
1970 strbuf_addch(sb, '\n');
1972 free(term);
1974 else {
1975 uintmax_t len = strtoumax(data, NULL, 10);
1976 size_t n = 0, length = (size_t)len;
1978 if (limit && limit < len) {
1979 *len_res = len;
1980 return 0;
1982 if (length < len)
1983 die("data is too large to use in this context");
1985 while (n < length) {
1986 size_t s = strbuf_fread(sb, length - n, stdin);
1987 if (!s && feof(stdin))
1988 die("EOF in data (%lu bytes remaining)",
1989 (unsigned long)(length - n));
1990 n += s;
1994 skip_optional_lf();
1995 return 1;
1998 static int validate_raw_date(const char *src, struct strbuf *result)
2000 const char *orig_src = src;
2001 char *endp;
2002 unsigned long num;
2004 errno = 0;
2006 num = strtoul(src, &endp, 10);
2007 /* NEEDSWORK: perhaps check for reasonable values? */
2008 if (errno || endp == src || *endp != ' ')
2009 return -1;
2011 src = endp + 1;
2012 if (*src != '-' && *src != '+')
2013 return -1;
2015 num = strtoul(src + 1, &endp, 10);
2016 if (errno || endp == src + 1 || *endp || 1400 < num)
2017 return -1;
2019 strbuf_addstr(result, orig_src);
2020 return 0;
2023 static char *parse_ident(const char *buf)
2025 const char *ltgt;
2026 size_t name_len;
2027 struct strbuf ident = STRBUF_INIT;
2029 /* ensure there is a space delimiter even if there is no name */
2030 if (*buf == '<')
2031 --buf;
2033 ltgt = buf + strcspn(buf, "<>");
2034 if (*ltgt != '<')
2035 die("Missing < in ident string: %s", buf);
2036 if (ltgt != buf && ltgt[-1] != ' ')
2037 die("Missing space before < in ident string: %s", buf);
2038 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2039 if (*ltgt != '>')
2040 die("Missing > in ident string: %s", buf);
2041 ltgt++;
2042 if (*ltgt != ' ')
2043 die("Missing space after > in ident string: %s", buf);
2044 ltgt++;
2045 name_len = ltgt - buf;
2046 strbuf_add(&ident, buf, name_len);
2048 switch (whenspec) {
2049 case WHENSPEC_RAW:
2050 if (validate_raw_date(ltgt, &ident) < 0)
2051 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2052 break;
2053 case WHENSPEC_RFC2822:
2054 if (parse_date(ltgt, &ident) < 0)
2055 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2056 break;
2057 case WHENSPEC_NOW:
2058 if (strcmp("now", ltgt))
2059 die("Date in ident must be 'now': %s", buf);
2060 datestamp(&ident);
2061 break;
2064 return strbuf_detach(&ident, NULL);
2067 static void parse_and_store_blob(
2068 struct last_object *last,
2069 unsigned char *sha1out,
2070 uintmax_t mark)
2072 static struct strbuf buf = STRBUF_INIT;
2073 uintmax_t len;
2075 if (parse_data(&buf, big_file_threshold, &len))
2076 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2077 else {
2078 if (last) {
2079 strbuf_release(&last->data);
2080 last->offset = 0;
2081 last->depth = 0;
2083 stream_blob(len, sha1out, mark);
2084 skip_optional_lf();
2088 static void parse_new_blob(void)
2090 read_next_command();
2091 parse_mark();
2092 parse_and_store_blob(&last_blob, NULL, next_mark);
2095 static void unload_one_branch(void)
2097 while (cur_active_branches
2098 && cur_active_branches >= max_active_branches) {
2099 uintmax_t min_commit = ULONG_MAX;
2100 struct branch *e, *l = NULL, *p = NULL;
2102 for (e = active_branches; e; e = e->active_next_branch) {
2103 if (e->last_commit < min_commit) {
2104 p = l;
2105 min_commit = e->last_commit;
2107 l = e;
2110 if (p) {
2111 e = p->active_next_branch;
2112 p->active_next_branch = e->active_next_branch;
2113 } else {
2114 e = active_branches;
2115 active_branches = e->active_next_branch;
2117 e->active = 0;
2118 e->active_next_branch = NULL;
2119 if (e->branch_tree.tree) {
2120 release_tree_content_recursive(e->branch_tree.tree);
2121 e->branch_tree.tree = NULL;
2123 cur_active_branches--;
2127 static void load_branch(struct branch *b)
2129 load_tree(&b->branch_tree);
2130 if (!b->active) {
2131 b->active = 1;
2132 b->active_next_branch = active_branches;
2133 active_branches = b;
2134 cur_active_branches++;
2135 branch_load_count++;
2139 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2141 unsigned char fanout = 0;
2142 while ((num_notes >>= 8))
2143 fanout++;
2144 return fanout;
2147 static void construct_path_with_fanout(const char *hex_sha1,
2148 unsigned char fanout, char *path)
2150 unsigned int i = 0, j = 0;
2151 if (fanout >= 20)
2152 die("Too large fanout (%u)", fanout);
2153 while (fanout) {
2154 path[i++] = hex_sha1[j++];
2155 path[i++] = hex_sha1[j++];
2156 path[i++] = '/';
2157 fanout--;
2159 memcpy(path + i, hex_sha1 + j, 40 - j);
2160 path[i + 40 - j] = '\0';
2163 static uintmax_t do_change_note_fanout(
2164 struct tree_entry *orig_root, struct tree_entry *root,
2165 char *hex_sha1, unsigned int hex_sha1_len,
2166 char *fullpath, unsigned int fullpath_len,
2167 unsigned char fanout)
2169 struct tree_content *t = root->tree;
2170 struct tree_entry *e, leaf;
2171 unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2172 uintmax_t num_notes = 0;
2173 unsigned char sha1[20];
2174 char realpath[60];
2176 for (i = 0; t && i < t->entry_count; i++) {
2177 e = t->entries[i];
2178 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2179 tmp_fullpath_len = fullpath_len;
2182 * We're interested in EITHER existing note entries (entries
2183 * with exactly 40 hex chars in path, not including directory
2184 * separators), OR directory entries that may contain note
2185 * entries (with < 40 hex chars in path).
2186 * Also, each path component in a note entry must be a multiple
2187 * of 2 chars.
2189 if (!e->versions[1].mode ||
2190 tmp_hex_sha1_len > 40 ||
2191 e->name->str_len % 2)
2192 continue;
2194 /* This _may_ be a note entry, or a subdir containing notes */
2195 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2196 e->name->str_len);
2197 if (tmp_fullpath_len)
2198 fullpath[tmp_fullpath_len++] = '/';
2199 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2200 e->name->str_len);
2201 tmp_fullpath_len += e->name->str_len;
2202 fullpath[tmp_fullpath_len] = '\0';
2204 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2205 /* This is a note entry */
2206 if (fanout == 0xff) {
2207 /* Counting mode, no rename */
2208 num_notes++;
2209 continue;
2211 construct_path_with_fanout(hex_sha1, fanout, realpath);
2212 if (!strcmp(fullpath, realpath)) {
2213 /* Note entry is in correct location */
2214 num_notes++;
2215 continue;
2218 /* Rename fullpath to realpath */
2219 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2220 die("Failed to remove path %s", fullpath);
2221 tree_content_set(orig_root, realpath,
2222 leaf.versions[1].sha1,
2223 leaf.versions[1].mode,
2224 leaf.tree);
2225 } else if (S_ISDIR(e->versions[1].mode)) {
2226 /* This is a subdir that may contain note entries */
2227 if (!e->tree)
2228 load_tree(e);
2229 num_notes += do_change_note_fanout(orig_root, e,
2230 hex_sha1, tmp_hex_sha1_len,
2231 fullpath, tmp_fullpath_len, fanout);
2234 /* The above may have reallocated the current tree_content */
2235 t = root->tree;
2237 return num_notes;
2240 static uintmax_t change_note_fanout(struct tree_entry *root,
2241 unsigned char fanout)
2243 char hex_sha1[40], path[60];
2244 return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2248 * Given a pointer into a string, parse a mark reference:
2250 * idnum ::= ':' bigint;
2252 * Return the first character after the value in *endptr.
2254 * Complain if the following character is not what is expected,
2255 * either a space or end of the string.
2257 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2259 uintmax_t mark;
2261 assert(*p == ':');
2262 p++;
2263 mark = strtoumax(p, endptr, 10);
2264 if (*endptr == p)
2265 die("No value after ':' in mark: %s", command_buf.buf);
2266 return mark;
2270 * Parse the mark reference, and complain if this is not the end of
2271 * the string.
2273 static uintmax_t parse_mark_ref_eol(const char *p)
2275 char *end;
2276 uintmax_t mark;
2278 mark = parse_mark_ref(p, &end);
2279 if (*end != '\0')
2280 die("Garbage after mark: %s", command_buf.buf);
2281 return mark;
2285 * Parse the mark reference, demanding a trailing space. Return a
2286 * pointer to the space.
2288 static uintmax_t parse_mark_ref_space(const char **p)
2290 uintmax_t mark;
2291 char *end;
2293 mark = parse_mark_ref(*p, &end);
2294 if (*end++ != ' ')
2295 die("Missing space after mark: %s", command_buf.buf);
2296 *p = end;
2297 return mark;
2300 static void file_change_m(const char *p, struct branch *b)
2302 static struct strbuf uq = STRBUF_INIT;
2303 const char *endp;
2304 struct object_entry *oe;
2305 unsigned char sha1[20];
2306 uint16_t mode, inline_data = 0;
2308 p = get_mode(p, &mode);
2309 if (!p)
2310 die("Corrupt mode: %s", command_buf.buf);
2311 switch (mode) {
2312 case 0644:
2313 case 0755:
2314 mode |= S_IFREG;
2315 case S_IFREG | 0644:
2316 case S_IFREG | 0755:
2317 case S_IFLNK:
2318 case S_IFDIR:
2319 case S_IFGITLINK:
2320 /* ok */
2321 break;
2322 default:
2323 die("Corrupt mode: %s", command_buf.buf);
2326 if (*p == ':') {
2327 oe = find_mark(parse_mark_ref_space(&p));
2328 hashcpy(sha1, oe->idx.sha1);
2329 } else if (skip_prefix(p, "inline ", &p)) {
2330 inline_data = 1;
2331 oe = NULL; /* not used with inline_data, but makes gcc happy */
2332 } else {
2333 if (get_sha1_hex(p, sha1))
2334 die("Invalid dataref: %s", command_buf.buf);
2335 oe = find_object(sha1);
2336 p += 40;
2337 if (*p++ != ' ')
2338 die("Missing space after SHA1: %s", command_buf.buf);
2341 strbuf_reset(&uq);
2342 if (!unquote_c_style(&uq, p, &endp)) {
2343 if (*endp)
2344 die("Garbage after path in: %s", command_buf.buf);
2345 p = uq.buf;
2348 /* Git does not track empty, non-toplevel directories. */
2349 if (S_ISDIR(mode) && !hashcmp(sha1, EMPTY_TREE_SHA1_BIN) && *p) {
2350 tree_content_remove(&b->branch_tree, p, NULL, 0);
2351 return;
2354 if (S_ISGITLINK(mode)) {
2355 if (inline_data)
2356 die("Git links cannot be specified 'inline': %s",
2357 command_buf.buf);
2358 else if (oe) {
2359 if (oe->type != OBJ_COMMIT)
2360 die("Not a commit (actually a %s): %s",
2361 typename(oe->type), command_buf.buf);
2364 * Accept the sha1 without checking; it expected to be in
2365 * another repository.
2367 } else if (inline_data) {
2368 if (S_ISDIR(mode))
2369 die("Directories cannot be specified 'inline': %s",
2370 command_buf.buf);
2371 if (p != uq.buf) {
2372 strbuf_addstr(&uq, p);
2373 p = uq.buf;
2375 read_next_command();
2376 parse_and_store_blob(&last_blob, sha1, 0);
2377 } else {
2378 enum object_type expected = S_ISDIR(mode) ?
2379 OBJ_TREE: OBJ_BLOB;
2380 enum object_type type = oe ? oe->type :
2381 sha1_object_info(sha1, NULL);
2382 if (type < 0)
2383 die("%s not found: %s",
2384 S_ISDIR(mode) ? "Tree" : "Blob",
2385 command_buf.buf);
2386 if (type != expected)
2387 die("Not a %s (actually a %s): %s",
2388 typename(expected), typename(type),
2389 command_buf.buf);
2392 if (!*p) {
2393 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2394 return;
2396 tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2399 static void file_change_d(const char *p, struct branch *b)
2401 static struct strbuf uq = STRBUF_INIT;
2402 const char *endp;
2404 strbuf_reset(&uq);
2405 if (!unquote_c_style(&uq, p, &endp)) {
2406 if (*endp)
2407 die("Garbage after path in: %s", command_buf.buf);
2408 p = uq.buf;
2410 tree_content_remove(&b->branch_tree, p, NULL, 1);
2413 static void file_change_cr(const char *s, struct branch *b, int rename)
2415 const char *d;
2416 static struct strbuf s_uq = STRBUF_INIT;
2417 static struct strbuf d_uq = STRBUF_INIT;
2418 const char *endp;
2419 struct tree_entry leaf;
2421 strbuf_reset(&s_uq);
2422 if (!unquote_c_style(&s_uq, s, &endp)) {
2423 if (*endp != ' ')
2424 die("Missing space after source: %s", command_buf.buf);
2425 } else {
2426 endp = strchr(s, ' ');
2427 if (!endp)
2428 die("Missing space after source: %s", command_buf.buf);
2429 strbuf_add(&s_uq, s, endp - s);
2431 s = s_uq.buf;
2433 endp++;
2434 if (!*endp)
2435 die("Missing dest: %s", command_buf.buf);
2437 d = endp;
2438 strbuf_reset(&d_uq);
2439 if (!unquote_c_style(&d_uq, d, &endp)) {
2440 if (*endp)
2441 die("Garbage after dest in: %s", command_buf.buf);
2442 d = d_uq.buf;
2445 memset(&leaf, 0, sizeof(leaf));
2446 if (rename)
2447 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2448 else
2449 tree_content_get(&b->branch_tree, s, &leaf, 1);
2450 if (!leaf.versions[1].mode)
2451 die("Path %s not in branch", s);
2452 if (!*d) { /* C "path/to/subdir" "" */
2453 tree_content_replace(&b->branch_tree,
2454 leaf.versions[1].sha1,
2455 leaf.versions[1].mode,
2456 leaf.tree);
2457 return;
2459 tree_content_set(&b->branch_tree, d,
2460 leaf.versions[1].sha1,
2461 leaf.versions[1].mode,
2462 leaf.tree);
2465 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2467 static struct strbuf uq = STRBUF_INIT;
2468 struct object_entry *oe;
2469 struct branch *s;
2470 unsigned char sha1[20], commit_sha1[20];
2471 char path[60];
2472 uint16_t inline_data = 0;
2473 unsigned char new_fanout;
2476 * When loading a branch, we don't traverse its tree to count the real
2477 * number of notes (too expensive to do this for all non-note refs).
2478 * This means that recently loaded notes refs might incorrectly have
2479 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2481 * Fix this by traversing the tree and counting the number of notes
2482 * when b->num_notes == 0. If the notes tree is truly empty, the
2483 * calculation should not take long.
2485 if (b->num_notes == 0 && *old_fanout == 0) {
2486 /* Invoke change_note_fanout() in "counting mode". */
2487 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2488 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2491 /* Now parse the notemodify command. */
2492 /* <dataref> or 'inline' */
2493 if (*p == ':') {
2494 oe = find_mark(parse_mark_ref_space(&p));
2495 hashcpy(sha1, oe->idx.sha1);
2496 } else if (skip_prefix(p, "inline ", &p)) {
2497 inline_data = 1;
2498 oe = NULL; /* not used with inline_data, but makes gcc happy */
2499 } else {
2500 if (get_sha1_hex(p, sha1))
2501 die("Invalid dataref: %s", command_buf.buf);
2502 oe = find_object(sha1);
2503 p += 40;
2504 if (*p++ != ' ')
2505 die("Missing space after SHA1: %s", command_buf.buf);
2508 /* <commit-ish> */
2509 s = lookup_branch(p);
2510 if (s) {
2511 if (is_null_sha1(s->sha1))
2512 die("Can't add a note on empty branch.");
2513 hashcpy(commit_sha1, s->sha1);
2514 } else if (*p == ':') {
2515 uintmax_t commit_mark = parse_mark_ref_eol(p);
2516 struct object_entry *commit_oe = find_mark(commit_mark);
2517 if (commit_oe->type != OBJ_COMMIT)
2518 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2519 hashcpy(commit_sha1, commit_oe->idx.sha1);
2520 } else if (!get_sha1(p, commit_sha1)) {
2521 unsigned long size;
2522 char *buf = read_object_with_reference(commit_sha1,
2523 commit_type, &size, commit_sha1);
2524 if (!buf || size < 46)
2525 die("Not a valid commit: %s", p);
2526 free(buf);
2527 } else
2528 die("Invalid ref name or SHA1 expression: %s", p);
2530 if (inline_data) {
2531 if (p != uq.buf) {
2532 strbuf_addstr(&uq, p);
2533 p = uq.buf;
2535 read_next_command();
2536 parse_and_store_blob(&last_blob, sha1, 0);
2537 } else if (oe) {
2538 if (oe->type != OBJ_BLOB)
2539 die("Not a blob (actually a %s): %s",
2540 typename(oe->type), command_buf.buf);
2541 } else if (!is_null_sha1(sha1)) {
2542 enum object_type type = sha1_object_info(sha1, NULL);
2543 if (type < 0)
2544 die("Blob not found: %s", command_buf.buf);
2545 if (type != OBJ_BLOB)
2546 die("Not a blob (actually a %s): %s",
2547 typename(type), command_buf.buf);
2550 construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2551 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2552 b->num_notes--;
2554 if (is_null_sha1(sha1))
2555 return; /* nothing to insert */
2557 b->num_notes++;
2558 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2559 construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2560 tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2563 static void file_change_deleteall(struct branch *b)
2565 release_tree_content_recursive(b->branch_tree.tree);
2566 hashclr(b->branch_tree.versions[0].sha1);
2567 hashclr(b->branch_tree.versions[1].sha1);
2568 load_tree(&b->branch_tree);
2569 b->num_notes = 0;
2572 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2574 if (!buf || size < 46)
2575 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2576 if (memcmp("tree ", buf, 5)
2577 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2578 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2579 hashcpy(b->branch_tree.versions[0].sha1,
2580 b->branch_tree.versions[1].sha1);
2583 static void parse_from_existing(struct branch *b)
2585 if (is_null_sha1(b->sha1)) {
2586 hashclr(b->branch_tree.versions[0].sha1);
2587 hashclr(b->branch_tree.versions[1].sha1);
2588 } else {
2589 unsigned long size;
2590 char *buf;
2592 buf = read_object_with_reference(b->sha1,
2593 commit_type, &size, b->sha1);
2594 parse_from_commit(b, buf, size);
2595 free(buf);
2599 static int parse_from(struct branch *b)
2601 const char *from;
2602 struct branch *s;
2603 unsigned char sha1[20];
2605 if (!skip_prefix(command_buf.buf, "from ", &from))
2606 return 0;
2608 hashcpy(sha1, b->branch_tree.versions[1].sha1);
2610 s = lookup_branch(from);
2611 if (b == s)
2612 die("Can't create a branch from itself: %s", b->name);
2613 else if (s) {
2614 unsigned char *t = s->branch_tree.versions[1].sha1;
2615 hashcpy(b->sha1, s->sha1);
2616 hashcpy(b->branch_tree.versions[0].sha1, t);
2617 hashcpy(b->branch_tree.versions[1].sha1, t);
2618 } else if (*from == ':') {
2619 uintmax_t idnum = parse_mark_ref_eol(from);
2620 struct object_entry *oe = find_mark(idnum);
2621 if (oe->type != OBJ_COMMIT)
2622 die("Mark :%" PRIuMAX " not a commit", idnum);
2623 if (hashcmp(b->sha1, oe->idx.sha1)) {
2624 hashcpy(b->sha1, oe->idx.sha1);
2625 if (oe->pack_id != MAX_PACK_ID) {
2626 unsigned long size;
2627 char *buf = gfi_unpack_entry(oe, &size);
2628 parse_from_commit(b, buf, size);
2629 free(buf);
2630 } else
2631 parse_from_existing(b);
2633 } else if (!get_sha1(from, b->sha1)) {
2634 parse_from_existing(b);
2635 if (is_null_sha1(b->sha1))
2636 b->delete = 1;
2638 else
2639 die("Invalid ref name or SHA1 expression: %s", from);
2641 if (b->branch_tree.tree && hashcmp(sha1, b->branch_tree.versions[1].sha1)) {
2642 release_tree_content_recursive(b->branch_tree.tree);
2643 b->branch_tree.tree = NULL;
2646 read_next_command();
2647 return 1;
2650 static struct hash_list *parse_merge(unsigned int *count)
2652 struct hash_list *list = NULL, **tail = &list, *n;
2653 const char *from;
2654 struct branch *s;
2656 *count = 0;
2657 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2658 n = xmalloc(sizeof(*n));
2659 s = lookup_branch(from);
2660 if (s)
2661 hashcpy(n->sha1, s->sha1);
2662 else if (*from == ':') {
2663 uintmax_t idnum = parse_mark_ref_eol(from);
2664 struct object_entry *oe = find_mark(idnum);
2665 if (oe->type != OBJ_COMMIT)
2666 die("Mark :%" PRIuMAX " not a commit", idnum);
2667 hashcpy(n->sha1, oe->idx.sha1);
2668 } else if (!get_sha1(from, n->sha1)) {
2669 unsigned long size;
2670 char *buf = read_object_with_reference(n->sha1,
2671 commit_type, &size, n->sha1);
2672 if (!buf || size < 46)
2673 die("Not a valid commit: %s", from);
2674 free(buf);
2675 } else
2676 die("Invalid ref name or SHA1 expression: %s", from);
2678 n->next = NULL;
2679 *tail = n;
2680 tail = &n->next;
2682 (*count)++;
2683 read_next_command();
2685 return list;
2688 static void parse_new_commit(const char *arg)
2690 static struct strbuf msg = STRBUF_INIT;
2691 struct branch *b;
2692 char *author = NULL;
2693 char *committer = NULL;
2694 struct hash_list *merge_list = NULL;
2695 unsigned int merge_count;
2696 unsigned char prev_fanout, new_fanout;
2697 const char *v;
2699 b = lookup_branch(arg);
2700 if (!b)
2701 b = new_branch(arg);
2703 read_next_command();
2704 parse_mark();
2705 if (skip_prefix(command_buf.buf, "author ", &v)) {
2706 author = parse_ident(v);
2707 read_next_command();
2709 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2710 committer = parse_ident(v);
2711 read_next_command();
2713 if (!committer)
2714 die("Expected committer but didn't get one");
2715 parse_data(&msg, 0, NULL);
2716 read_next_command();
2717 parse_from(b);
2718 merge_list = parse_merge(&merge_count);
2720 /* ensure the branch is active/loaded */
2721 if (!b->branch_tree.tree || !max_active_branches) {
2722 unload_one_branch();
2723 load_branch(b);
2726 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2728 /* file_change* */
2729 while (command_buf.len > 0) {
2730 if (skip_prefix(command_buf.buf, "M ", &v))
2731 file_change_m(v, b);
2732 else if (skip_prefix(command_buf.buf, "D ", &v))
2733 file_change_d(v, b);
2734 else if (skip_prefix(command_buf.buf, "R ", &v))
2735 file_change_cr(v, b, 1);
2736 else if (skip_prefix(command_buf.buf, "C ", &v))
2737 file_change_cr(v, b, 0);
2738 else if (skip_prefix(command_buf.buf, "N ", &v))
2739 note_change_n(v, b, &prev_fanout);
2740 else if (!strcmp("deleteall", command_buf.buf))
2741 file_change_deleteall(b);
2742 else if (skip_prefix(command_buf.buf, "ls ", &v))
2743 parse_ls(v, b);
2744 else {
2745 unread_command_buf = 1;
2746 break;
2748 if (read_next_command() == EOF)
2749 break;
2752 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2753 if (new_fanout != prev_fanout)
2754 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2756 /* build the tree and the commit */
2757 store_tree(&b->branch_tree);
2758 hashcpy(b->branch_tree.versions[0].sha1,
2759 b->branch_tree.versions[1].sha1);
2761 strbuf_reset(&new_data);
2762 strbuf_addf(&new_data, "tree %s\n",
2763 sha1_to_hex(b->branch_tree.versions[1].sha1));
2764 if (!is_null_sha1(b->sha1))
2765 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2766 while (merge_list) {
2767 struct hash_list *next = merge_list->next;
2768 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2769 free(merge_list);
2770 merge_list = next;
2772 strbuf_addf(&new_data,
2773 "author %s\n"
2774 "committer %s\n"
2775 "\n",
2776 author ? author : committer, committer);
2777 strbuf_addbuf(&new_data, &msg);
2778 free(author);
2779 free(committer);
2781 if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2782 b->pack_id = pack_id;
2783 b->last_commit = object_count_by_type[OBJ_COMMIT];
2786 static void parse_new_tag(const char *arg)
2788 static struct strbuf msg = STRBUF_INIT;
2789 const char *from;
2790 char *tagger;
2791 struct branch *s;
2792 struct tag *t;
2793 uintmax_t from_mark = 0;
2794 unsigned char sha1[20];
2795 enum object_type type;
2796 const char *v;
2798 t = pool_alloc(sizeof(struct tag));
2799 memset(t, 0, sizeof(struct tag));
2800 t->name = pool_strdup(arg);
2801 if (last_tag)
2802 last_tag->next_tag = t;
2803 else
2804 first_tag = t;
2805 last_tag = t;
2806 read_next_command();
2808 /* from ... */
2809 if (!skip_prefix(command_buf.buf, "from ", &from))
2810 die("Expected from command, got %s", command_buf.buf);
2811 s = lookup_branch(from);
2812 if (s) {
2813 if (is_null_sha1(s->sha1))
2814 die("Can't tag an empty branch.");
2815 hashcpy(sha1, s->sha1);
2816 type = OBJ_COMMIT;
2817 } else if (*from == ':') {
2818 struct object_entry *oe;
2819 from_mark = parse_mark_ref_eol(from);
2820 oe = find_mark(from_mark);
2821 type = oe->type;
2822 hashcpy(sha1, oe->idx.sha1);
2823 } else if (!get_sha1(from, sha1)) {
2824 struct object_entry *oe = find_object(sha1);
2825 if (!oe) {
2826 type = sha1_object_info(sha1, NULL);
2827 if (type < 0)
2828 die("Not a valid object: %s", from);
2829 } else
2830 type = oe->type;
2831 } else
2832 die("Invalid ref name or SHA1 expression: %s", from);
2833 read_next_command();
2835 /* tagger ... */
2836 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2837 tagger = parse_ident(v);
2838 read_next_command();
2839 } else
2840 tagger = NULL;
2842 /* tag payload/message */
2843 parse_data(&msg, 0, NULL);
2845 /* build the tag object */
2846 strbuf_reset(&new_data);
2848 strbuf_addf(&new_data,
2849 "object %s\n"
2850 "type %s\n"
2851 "tag %s\n",
2852 sha1_to_hex(sha1), typename(type), t->name);
2853 if (tagger)
2854 strbuf_addf(&new_data,
2855 "tagger %s\n", tagger);
2856 strbuf_addch(&new_data, '\n');
2857 strbuf_addbuf(&new_data, &msg);
2858 free(tagger);
2860 if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2861 t->pack_id = MAX_PACK_ID;
2862 else
2863 t->pack_id = pack_id;
2866 static void parse_reset_branch(const char *arg)
2868 struct branch *b;
2870 b = lookup_branch(arg);
2871 if (b) {
2872 hashclr(b->sha1);
2873 hashclr(b->branch_tree.versions[0].sha1);
2874 hashclr(b->branch_tree.versions[1].sha1);
2875 if (b->branch_tree.tree) {
2876 release_tree_content_recursive(b->branch_tree.tree);
2877 b->branch_tree.tree = NULL;
2880 else
2881 b = new_branch(arg);
2882 read_next_command();
2883 parse_from(b);
2884 if (command_buf.len > 0)
2885 unread_command_buf = 1;
2888 static void cat_blob_write(const char *buf, unsigned long size)
2890 if (write_in_full(cat_blob_fd, buf, size) != size)
2891 die_errno("Write to frontend failed");
2894 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2896 struct strbuf line = STRBUF_INIT;
2897 unsigned long size;
2898 enum object_type type = 0;
2899 char *buf;
2901 if (!oe || oe->pack_id == MAX_PACK_ID) {
2902 buf = read_sha1_file(sha1, &type, &size);
2903 } else {
2904 type = oe->type;
2905 buf = gfi_unpack_entry(oe, &size);
2909 * Output based on batch_one_object() from cat-file.c.
2911 if (type <= 0) {
2912 strbuf_reset(&line);
2913 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2914 cat_blob_write(line.buf, line.len);
2915 strbuf_release(&line);
2916 free(buf);
2917 return;
2919 if (!buf)
2920 die("Can't read object %s", sha1_to_hex(sha1));
2921 if (type != OBJ_BLOB)
2922 die("Object %s is a %s but a blob was expected.",
2923 sha1_to_hex(sha1), typename(type));
2924 strbuf_reset(&line);
2925 strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2926 typename(type), size);
2927 cat_blob_write(line.buf, line.len);
2928 strbuf_release(&line);
2929 cat_blob_write(buf, size);
2930 cat_blob_write("\n", 1);
2931 if (oe && oe->pack_id == pack_id) {
2932 last_blob.offset = oe->idx.offset;
2933 strbuf_attach(&last_blob.data, buf, size, size);
2934 last_blob.depth = oe->depth;
2935 } else
2936 free(buf);
2939 static void parse_get_mark(const char *p)
2941 struct object_entry *oe = oe;
2942 char output[42];
2944 /* get-mark SP <object> LF */
2945 if (*p != ':')
2946 die("Not a mark: %s", p);
2948 oe = find_mark(parse_mark_ref_eol(p));
2949 if (!oe)
2950 die("Unknown mark: %s", command_buf.buf);
2952 snprintf(output, sizeof(output), "%s\n", sha1_to_hex(oe->idx.sha1));
2953 cat_blob_write(output, 41);
2956 static void parse_cat_blob(const char *p)
2958 struct object_entry *oe = oe;
2959 unsigned char sha1[20];
2961 /* cat-blob SP <object> LF */
2962 if (*p == ':') {
2963 oe = find_mark(parse_mark_ref_eol(p));
2964 if (!oe)
2965 die("Unknown mark: %s", command_buf.buf);
2966 hashcpy(sha1, oe->idx.sha1);
2967 } else {
2968 if (get_sha1_hex(p, sha1))
2969 die("Invalid dataref: %s", command_buf.buf);
2970 if (p[40])
2971 die("Garbage after SHA1: %s", command_buf.buf);
2972 oe = find_object(sha1);
2975 cat_blob(oe, sha1);
2978 static struct object_entry *dereference(struct object_entry *oe,
2979 unsigned char sha1[20])
2981 unsigned long size;
2982 char *buf = NULL;
2983 if (!oe) {
2984 enum object_type type = sha1_object_info(sha1, NULL);
2985 if (type < 0)
2986 die("object not found: %s", sha1_to_hex(sha1));
2987 /* cache it! */
2988 oe = insert_object(sha1);
2989 oe->type = type;
2990 oe->pack_id = MAX_PACK_ID;
2991 oe->idx.offset = 1;
2993 switch (oe->type) {
2994 case OBJ_TREE: /* easy case. */
2995 return oe;
2996 case OBJ_COMMIT:
2997 case OBJ_TAG:
2998 break;
2999 default:
3000 die("Not a tree-ish: %s", command_buf.buf);
3003 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3004 buf = gfi_unpack_entry(oe, &size);
3005 } else {
3006 enum object_type unused;
3007 buf = read_sha1_file(sha1, &unused, &size);
3009 if (!buf)
3010 die("Can't load object %s", sha1_to_hex(sha1));
3012 /* Peel one layer. */
3013 switch (oe->type) {
3014 case OBJ_TAG:
3015 if (size < 40 + strlen("object ") ||
3016 get_sha1_hex(buf + strlen("object "), sha1))
3017 die("Invalid SHA1 in tag: %s", command_buf.buf);
3018 break;
3019 case OBJ_COMMIT:
3020 if (size < 40 + strlen("tree ") ||
3021 get_sha1_hex(buf + strlen("tree "), sha1))
3022 die("Invalid SHA1 in commit: %s", command_buf.buf);
3025 free(buf);
3026 return find_object(sha1);
3029 static struct object_entry *parse_treeish_dataref(const char **p)
3031 unsigned char sha1[20];
3032 struct object_entry *e;
3034 if (**p == ':') { /* <mark> */
3035 e = find_mark(parse_mark_ref_space(p));
3036 if (!e)
3037 die("Unknown mark: %s", command_buf.buf);
3038 hashcpy(sha1, e->idx.sha1);
3039 } else { /* <sha1> */
3040 if (get_sha1_hex(*p, sha1))
3041 die("Invalid dataref: %s", command_buf.buf);
3042 e = find_object(sha1);
3043 *p += 40;
3044 if (*(*p)++ != ' ')
3045 die("Missing space after tree-ish: %s", command_buf.buf);
3048 while (!e || e->type != OBJ_TREE)
3049 e = dereference(e, sha1);
3050 return e;
3053 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3055 static struct strbuf line = STRBUF_INIT;
3057 /* See show_tree(). */
3058 const char *type =
3059 S_ISGITLINK(mode) ? commit_type :
3060 S_ISDIR(mode) ? tree_type :
3061 blob_type;
3063 if (!mode) {
3064 /* missing SP path LF */
3065 strbuf_reset(&line);
3066 strbuf_addstr(&line, "missing ");
3067 quote_c_style(path, &line, NULL, 0);
3068 strbuf_addch(&line, '\n');
3069 } else {
3070 /* mode SP type SP object_name TAB path LF */
3071 strbuf_reset(&line);
3072 strbuf_addf(&line, "%06o %s %s\t",
3073 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3074 quote_c_style(path, &line, NULL, 0);
3075 strbuf_addch(&line, '\n');
3077 cat_blob_write(line.buf, line.len);
3080 static void parse_ls(const char *p, struct branch *b)
3082 struct tree_entry *root = NULL;
3083 struct tree_entry leaf = {NULL};
3085 /* ls SP (<tree-ish> SP)? <path> */
3086 if (*p == '"') {
3087 if (!b)
3088 die("Not in a commit: %s", command_buf.buf);
3089 root = &b->branch_tree;
3090 } else {
3091 struct object_entry *e = parse_treeish_dataref(&p);
3092 root = new_tree_entry();
3093 hashcpy(root->versions[1].sha1, e->idx.sha1);
3094 if (!is_null_sha1(root->versions[1].sha1))
3095 root->versions[1].mode = S_IFDIR;
3096 load_tree(root);
3098 if (*p == '"') {
3099 static struct strbuf uq = STRBUF_INIT;
3100 const char *endp;
3101 strbuf_reset(&uq);
3102 if (unquote_c_style(&uq, p, &endp))
3103 die("Invalid path: %s", command_buf.buf);
3104 if (*endp)
3105 die("Garbage after path in: %s", command_buf.buf);
3106 p = uq.buf;
3108 tree_content_get(root, p, &leaf, 1);
3110 * A directory in preparation would have a sha1 of zero
3111 * until it is saved. Save, for simplicity.
3113 if (S_ISDIR(leaf.versions[1].mode))
3114 store_tree(&leaf);
3116 print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3117 if (leaf.tree)
3118 release_tree_content_recursive(leaf.tree);
3119 if (!b || root != &b->branch_tree)
3120 release_tree_entry(root);
3123 static void checkpoint(void)
3125 checkpoint_requested = 0;
3126 if (object_count) {
3127 cycle_packfile();
3128 dump_branches();
3129 dump_tags();
3130 dump_marks();
3134 static void parse_checkpoint(void)
3136 checkpoint_requested = 1;
3137 skip_optional_lf();
3140 static void parse_progress(void)
3142 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3143 fputc('\n', stdout);
3144 fflush(stdout);
3145 skip_optional_lf();
3148 static char* make_fast_import_path(const char *path)
3150 if (!relative_marks_paths || is_absolute_path(path))
3151 return xstrdup(path);
3152 return xstrdup(git_path("info/fast-import/%s", path));
3155 static void option_import_marks(const char *marks,
3156 int from_stream, int ignore_missing)
3158 if (import_marks_file) {
3159 if (from_stream)
3160 die("Only one import-marks command allowed per stream");
3162 /* read previous mark file */
3163 if(!import_marks_file_from_stream)
3164 read_marks();
3167 import_marks_file = make_fast_import_path(marks);
3168 safe_create_leading_directories_const(import_marks_file);
3169 import_marks_file_from_stream = from_stream;
3170 import_marks_file_ignore_missing = ignore_missing;
3173 static void option_date_format(const char *fmt)
3175 if (!strcmp(fmt, "raw"))
3176 whenspec = WHENSPEC_RAW;
3177 else if (!strcmp(fmt, "rfc2822"))
3178 whenspec = WHENSPEC_RFC2822;
3179 else if (!strcmp(fmt, "now"))
3180 whenspec = WHENSPEC_NOW;
3181 else
3182 die("unknown --date-format argument %s", fmt);
3185 static unsigned long ulong_arg(const char *option, const char *arg)
3187 char *endptr;
3188 unsigned long rv = strtoul(arg, &endptr, 0);
3189 if (strchr(arg, '-') || endptr == arg || *endptr)
3190 die("%s: argument must be a non-negative integer", option);
3191 return rv;
3194 static void option_depth(const char *depth)
3196 max_depth = ulong_arg("--depth", depth);
3197 if (max_depth > MAX_DEPTH)
3198 die("--depth cannot exceed %u", MAX_DEPTH);
3201 static void option_active_branches(const char *branches)
3203 max_active_branches = ulong_arg("--active-branches", branches);
3206 static void option_export_marks(const char *marks)
3208 export_marks_file = make_fast_import_path(marks);
3209 safe_create_leading_directories_const(export_marks_file);
3212 static void option_cat_blob_fd(const char *fd)
3214 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3215 if (n > (unsigned long) INT_MAX)
3216 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3217 cat_blob_fd = (int) n;
3220 static void option_export_pack_edges(const char *edges)
3222 if (pack_edges)
3223 fclose(pack_edges);
3224 pack_edges = fopen(edges, "a");
3225 if (!pack_edges)
3226 die_errno("Cannot open '%s'", edges);
3229 static int parse_one_option(const char *option)
3231 if (skip_prefix(option, "max-pack-size=", &option)) {
3232 unsigned long v;
3233 if (!git_parse_ulong(option, &v))
3234 return 0;
3235 if (v < 8192) {
3236 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3237 v *= 1024 * 1024;
3238 } else if (v < 1024 * 1024) {
3239 warning("minimum max-pack-size is 1 MiB");
3240 v = 1024 * 1024;
3242 max_packsize = v;
3243 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3244 unsigned long v;
3245 if (!git_parse_ulong(option, &v))
3246 return 0;
3247 big_file_threshold = v;
3248 } else if (skip_prefix(option, "depth=", &option)) {
3249 option_depth(option);
3250 } else if (skip_prefix(option, "active-branches=", &option)) {
3251 option_active_branches(option);
3252 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3253 option_export_pack_edges(option);
3254 } else if (starts_with(option, "quiet")) {
3255 show_stats = 0;
3256 } else if (starts_with(option, "stats")) {
3257 show_stats = 1;
3258 } else {
3259 return 0;
3262 return 1;
3265 static int parse_one_feature(const char *feature, int from_stream)
3267 const char *arg;
3269 if (skip_prefix(feature, "date-format=", &arg)) {
3270 option_date_format(arg);
3271 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3272 option_import_marks(arg, from_stream, 0);
3273 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3274 option_import_marks(arg, from_stream, 1);
3275 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3276 option_export_marks(arg);
3277 } else if (!strcmp(feature, "get-mark")) {
3278 ; /* Don't die - this feature is supported */
3279 } else if (!strcmp(feature, "cat-blob")) {
3280 ; /* Don't die - this feature is supported */
3281 } else if (!strcmp(feature, "relative-marks")) {
3282 relative_marks_paths = 1;
3283 } else if (!strcmp(feature, "no-relative-marks")) {
3284 relative_marks_paths = 0;
3285 } else if (!strcmp(feature, "done")) {
3286 require_explicit_termination = 1;
3287 } else if (!strcmp(feature, "force")) {
3288 force_update = 1;
3289 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3290 ; /* do nothing; we have the feature */
3291 } else {
3292 return 0;
3295 return 1;
3298 static void parse_feature(const char *feature)
3300 if (seen_data_command)
3301 die("Got feature command '%s' after data command", feature);
3303 if (parse_one_feature(feature, 1))
3304 return;
3306 die("This version of fast-import does not support feature %s.", feature);
3309 static void parse_option(const char *option)
3311 if (seen_data_command)
3312 die("Got option command '%s' after data command", option);
3314 if (parse_one_option(option))
3315 return;
3317 die("This version of fast-import does not support option: %s", option);
3320 static void git_pack_config(void)
3322 int indexversion_value;
3323 unsigned long packsizelimit_value;
3325 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3326 if (max_depth > MAX_DEPTH)
3327 max_depth = MAX_DEPTH;
3329 if (!git_config_get_int("pack.compression", &pack_compression_level)) {
3330 if (pack_compression_level == -1)
3331 pack_compression_level = Z_DEFAULT_COMPRESSION;
3332 else if (pack_compression_level < 0 ||
3333 pack_compression_level > Z_BEST_COMPRESSION)
3334 git_die_config("pack.compression",
3335 "bad pack compression level %d", pack_compression_level);
3336 pack_compression_seen = 1;
3338 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3339 pack_idx_opts.version = indexversion_value;
3340 if (pack_idx_opts.version > 2)
3341 git_die_config("pack.indexversion",
3342 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3344 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3345 max_packsize = packsizelimit_value;
3347 git_config(git_default_config, NULL);
3350 static const char fast_import_usage[] =
3351 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3353 static void parse_argv(void)
3355 unsigned int i;
3357 for (i = 1; i < global_argc; i++) {
3358 const char *a = global_argv[i];
3360 if (*a != '-' || !strcmp(a, "--"))
3361 break;
3363 if (!skip_prefix(a, "--", &a))
3364 die("unknown option %s", a);
3366 if (parse_one_option(a))
3367 continue;
3369 if (parse_one_feature(a, 0))
3370 continue;
3372 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3373 option_cat_blob_fd(a);
3374 continue;
3377 die("unknown option --%s", a);
3379 if (i != global_argc)
3380 usage(fast_import_usage);
3382 seen_data_command = 1;
3383 if (import_marks_file)
3384 read_marks();
3387 int main(int argc, char **argv)
3389 unsigned int i;
3391 git_extract_argv0_path(argv[0]);
3393 git_setup_gettext();
3395 if (argc == 2 && !strcmp(argv[1], "-h"))
3396 usage(fast_import_usage);
3398 setup_git_directory();
3399 reset_pack_idx_option(&pack_idx_opts);
3400 git_pack_config();
3401 if (!pack_compression_seen && core_compression_seen)
3402 pack_compression_level = core_compression_level;
3404 alloc_objects(object_entry_alloc);
3405 strbuf_init(&command_buf, 0);
3406 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3407 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3408 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3409 marks = pool_calloc(1, sizeof(struct mark_set));
3411 global_argc = argc;
3412 global_argv = argv;
3414 rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3415 for (i = 0; i < (cmd_save - 1); i++)
3416 rc_free[i].next = &rc_free[i + 1];
3417 rc_free[cmd_save - 1].next = NULL;
3419 prepare_packed_git();
3420 start_packfile();
3421 set_die_routine(die_nicely);
3422 set_checkpoint_signal();
3423 while (read_next_command() != EOF) {
3424 const char *v;
3425 if (!strcmp("blob", command_buf.buf))
3426 parse_new_blob();
3427 else if (skip_prefix(command_buf.buf, "ls ", &v))
3428 parse_ls(v, NULL);
3429 else if (skip_prefix(command_buf.buf, "commit ", &v))
3430 parse_new_commit(v);
3431 else if (skip_prefix(command_buf.buf, "tag ", &v))
3432 parse_new_tag(v);
3433 else if (skip_prefix(command_buf.buf, "reset ", &v))
3434 parse_reset_branch(v);
3435 else if (!strcmp("checkpoint", command_buf.buf))
3436 parse_checkpoint();
3437 else if (!strcmp("done", command_buf.buf))
3438 break;
3439 else if (starts_with(command_buf.buf, "progress "))
3440 parse_progress();
3441 else if (skip_prefix(command_buf.buf, "feature ", &v))
3442 parse_feature(v);
3443 else if (skip_prefix(command_buf.buf, "option git ", &v))
3444 parse_option(v);
3445 else if (starts_with(command_buf.buf, "option "))
3446 /* ignore non-git options*/;
3447 else
3448 die("Unsupported command: %s", command_buf.buf);
3450 if (checkpoint_requested)
3451 checkpoint();
3454 /* argv hasn't been parsed yet, do so */
3455 if (!seen_data_command)
3456 parse_argv();
3458 if (require_explicit_termination && feof(stdin))
3459 die("stream ends early");
3461 end_packfile();
3463 dump_branches();
3464 dump_tags();
3465 unkeep_all_packs();
3466 dump_marks();
3468 if (pack_edges)
3469 fclose(pack_edges);
3471 if (show_stats) {
3472 uintmax_t total_count = 0, duplicate_count = 0;
3473 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3474 total_count += object_count_by_type[i];
3475 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3476 duplicate_count += duplicate_count_by_type[i];
3478 fprintf(stderr, "%s statistics:\n", argv[0]);
3479 fprintf(stderr, "---------------------------------------------------------------------\n");
3480 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3481 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3482 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]);
3483 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]);
3484 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]);
3485 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]);
3486 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3487 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3488 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3489 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3490 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)(total_allocd/1024));
3491 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3492 fprintf(stderr, "---------------------------------------------------------------------\n");
3493 pack_report();
3494 fprintf(stderr, "---------------------------------------------------------------------\n");
3495 fprintf(stderr, "\n");
3498 return failure ? 1 : 0;