Merge branch 'sb/clean-test-fix'
[git/debian.git] / fast-import.c
blob9fc7093406b1e2085a3b2180a8506bcd6d486e06
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(st_add(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 struct pack_header hdr;
869 int pack_fd;
871 pack_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
872 "pack/tmp_pack_XXXXXX");
873 FLEX_ALLOC_STR(p, pack_name, tmp_file);
874 p->pack_fd = pack_fd;
875 p->do_not_close = 1;
876 pack_file = sha1fd(pack_fd, p->pack_name);
878 hdr.hdr_signature = htonl(PACK_SIGNATURE);
879 hdr.hdr_version = htonl(2);
880 hdr.hdr_entries = 0;
881 sha1write(pack_file, &hdr, sizeof(hdr));
883 pack_data = p;
884 pack_size = sizeof(hdr);
885 object_count = 0;
887 REALLOC_ARRAY(all_packs, pack_id + 1);
888 all_packs[pack_id] = p;
891 static const char *create_index(void)
893 const char *tmpfile;
894 struct pack_idx_entry **idx, **c, **last;
895 struct object_entry *e;
896 struct object_entry_pool *o;
898 /* Build the table of object IDs. */
899 ALLOC_ARRAY(idx, object_count);
900 c = idx;
901 for (o = blocks; o; o = o->next_pool)
902 for (e = o->next_free; e-- != o->entries;)
903 if (pack_id == e->pack_id)
904 *c++ = &e->idx;
905 last = idx + object_count;
906 if (c != last)
907 die("internal consistency error creating the index");
909 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
910 free(idx);
911 return tmpfile;
914 static char *keep_pack(const char *curr_index_name)
916 static char name[PATH_MAX];
917 static const char *keep_msg = "fast-import";
918 int keep_fd;
920 keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
921 if (keep_fd < 0)
922 die_errno("cannot create keep file");
923 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
924 if (close(keep_fd))
925 die_errno("failed to write keep file");
927 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
928 get_object_directory(), sha1_to_hex(pack_data->sha1));
929 if (finalize_object_file(pack_data->pack_name, name))
930 die("cannot store pack file");
932 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
933 get_object_directory(), sha1_to_hex(pack_data->sha1));
934 if (finalize_object_file(curr_index_name, name))
935 die("cannot store index file");
936 free((void *)curr_index_name);
937 return name;
940 static void unkeep_all_packs(void)
942 static char name[PATH_MAX];
943 int k;
945 for (k = 0; k < pack_id; k++) {
946 struct packed_git *p = all_packs[k];
947 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
948 get_object_directory(), sha1_to_hex(p->sha1));
949 unlink_or_warn(name);
953 static void end_packfile(void)
955 static int running;
957 if (running || !pack_data)
958 return;
960 running = 1;
961 clear_delta_base_cache();
962 if (object_count) {
963 struct packed_git *new_p;
964 unsigned char cur_pack_sha1[20];
965 char *idx_name;
966 int i;
967 struct branch *b;
968 struct tag *t;
970 close_pack_windows(pack_data);
971 sha1close(pack_file, cur_pack_sha1, 0);
972 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
973 pack_data->pack_name, object_count,
974 cur_pack_sha1, pack_size);
975 close(pack_data->pack_fd);
976 idx_name = keep_pack(create_index());
978 /* Register the packfile with core git's machinery. */
979 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
980 if (!new_p)
981 die("core git rejected index %s", idx_name);
982 all_packs[pack_id] = new_p;
983 install_packed_git(new_p);
985 /* Print the boundary */
986 if (pack_edges) {
987 fprintf(pack_edges, "%s:", new_p->pack_name);
988 for (i = 0; i < branch_table_sz; i++) {
989 for (b = branch_table[i]; b; b = b->table_next_branch) {
990 if (b->pack_id == pack_id)
991 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
994 for (t = first_tag; t; t = t->next_tag) {
995 if (t->pack_id == pack_id)
996 fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
998 fputc('\n', pack_edges);
999 fflush(pack_edges);
1002 pack_id++;
1004 else {
1005 close(pack_data->pack_fd);
1006 unlink_or_warn(pack_data->pack_name);
1008 free(pack_data);
1009 pack_data = NULL;
1010 running = 0;
1012 /* We can't carry a delta across packfiles. */
1013 strbuf_release(&last_blob.data);
1014 last_blob.offset = 0;
1015 last_blob.depth = 0;
1018 static void cycle_packfile(void)
1020 end_packfile();
1021 start_packfile();
1024 static int store_object(
1025 enum object_type type,
1026 struct strbuf *dat,
1027 struct last_object *last,
1028 unsigned char *sha1out,
1029 uintmax_t mark)
1031 void *out, *delta;
1032 struct object_entry *e;
1033 unsigned char hdr[96];
1034 unsigned char sha1[20];
1035 unsigned long hdrlen, deltalen;
1036 git_SHA_CTX c;
1037 git_zstream s;
1039 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1040 typename(type), (unsigned long)dat->len) + 1;
1041 git_SHA1_Init(&c);
1042 git_SHA1_Update(&c, hdr, hdrlen);
1043 git_SHA1_Update(&c, dat->buf, dat->len);
1044 git_SHA1_Final(sha1, &c);
1045 if (sha1out)
1046 hashcpy(sha1out, sha1);
1048 e = insert_object(sha1);
1049 if (mark)
1050 insert_mark(mark, e);
1051 if (e->idx.offset) {
1052 duplicate_count_by_type[type]++;
1053 return 1;
1054 } else if (find_sha1_pack(sha1, packed_git)) {
1055 e->type = type;
1056 e->pack_id = MAX_PACK_ID;
1057 e->idx.offset = 1; /* just not zero! */
1058 duplicate_count_by_type[type]++;
1059 return 1;
1062 if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1063 delta_count_attempts_by_type[type]++;
1064 delta = diff_delta(last->data.buf, last->data.len,
1065 dat->buf, dat->len,
1066 &deltalen, dat->len - 20);
1067 } else
1068 delta = NULL;
1070 git_deflate_init(&s, pack_compression_level);
1071 if (delta) {
1072 s.next_in = delta;
1073 s.avail_in = deltalen;
1074 } else {
1075 s.next_in = (void *)dat->buf;
1076 s.avail_in = dat->len;
1078 s.avail_out = git_deflate_bound(&s, s.avail_in);
1079 s.next_out = out = xmalloc(s.avail_out);
1080 while (git_deflate(&s, Z_FINISH) == Z_OK)
1081 ; /* nothing */
1082 git_deflate_end(&s);
1084 /* Determine if we should auto-checkpoint. */
1085 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1086 || (pack_size + 60 + s.total_out) < pack_size) {
1088 /* This new object needs to *not* have the current pack_id. */
1089 e->pack_id = pack_id + 1;
1090 cycle_packfile();
1092 /* We cannot carry a delta into the new pack. */
1093 if (delta) {
1094 free(delta);
1095 delta = NULL;
1097 git_deflate_init(&s, pack_compression_level);
1098 s.next_in = (void *)dat->buf;
1099 s.avail_in = dat->len;
1100 s.avail_out = git_deflate_bound(&s, s.avail_in);
1101 s.next_out = out = xrealloc(out, s.avail_out);
1102 while (git_deflate(&s, Z_FINISH) == Z_OK)
1103 ; /* nothing */
1104 git_deflate_end(&s);
1108 e->type = type;
1109 e->pack_id = pack_id;
1110 e->idx.offset = pack_size;
1111 object_count++;
1112 object_count_by_type[type]++;
1114 crc32_begin(pack_file);
1116 if (delta) {
1117 off_t ofs = e->idx.offset - last->offset;
1118 unsigned pos = sizeof(hdr) - 1;
1120 delta_count_by_type[type]++;
1121 e->depth = last->depth + 1;
1123 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1124 sha1write(pack_file, hdr, hdrlen);
1125 pack_size += hdrlen;
1127 hdr[pos] = ofs & 127;
1128 while (ofs >>= 7)
1129 hdr[--pos] = 128 | (--ofs & 127);
1130 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1131 pack_size += sizeof(hdr) - pos;
1132 } else {
1133 e->depth = 0;
1134 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1135 sha1write(pack_file, hdr, hdrlen);
1136 pack_size += hdrlen;
1139 sha1write(pack_file, out, s.total_out);
1140 pack_size += s.total_out;
1142 e->idx.crc32 = crc32_end(pack_file);
1144 free(out);
1145 free(delta);
1146 if (last) {
1147 if (last->no_swap) {
1148 last->data = *dat;
1149 } else {
1150 strbuf_swap(&last->data, dat);
1152 last->offset = e->idx.offset;
1153 last->depth = e->depth;
1155 return 0;
1158 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1160 if (sha1file_truncate(pack_file, checkpoint))
1161 die_errno("cannot truncate pack to skip duplicate");
1162 pack_size = checkpoint->offset;
1165 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1167 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1168 unsigned char *in_buf = xmalloc(in_sz);
1169 unsigned char *out_buf = xmalloc(out_sz);
1170 struct object_entry *e;
1171 unsigned char sha1[20];
1172 unsigned long hdrlen;
1173 off_t offset;
1174 git_SHA_CTX c;
1175 git_zstream s;
1176 struct sha1file_checkpoint checkpoint;
1177 int status = Z_OK;
1179 /* Determine if we should auto-checkpoint. */
1180 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1181 || (pack_size + 60 + len) < pack_size)
1182 cycle_packfile();
1184 sha1file_checkpoint(pack_file, &checkpoint);
1185 offset = checkpoint.offset;
1187 hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1188 if (out_sz <= hdrlen)
1189 die("impossibly large object header");
1191 git_SHA1_Init(&c);
1192 git_SHA1_Update(&c, out_buf, hdrlen);
1194 crc32_begin(pack_file);
1196 git_deflate_init(&s, pack_compression_level);
1198 hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1199 if (out_sz <= hdrlen)
1200 die("impossibly large object header");
1202 s.next_out = out_buf + hdrlen;
1203 s.avail_out = out_sz - hdrlen;
1205 while (status != Z_STREAM_END) {
1206 if (0 < len && !s.avail_in) {
1207 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1208 size_t n = fread(in_buf, 1, cnt, stdin);
1209 if (!n && feof(stdin))
1210 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1212 git_SHA1_Update(&c, in_buf, n);
1213 s.next_in = in_buf;
1214 s.avail_in = n;
1215 len -= n;
1218 status = git_deflate(&s, len ? 0 : Z_FINISH);
1220 if (!s.avail_out || status == Z_STREAM_END) {
1221 size_t n = s.next_out - out_buf;
1222 sha1write(pack_file, out_buf, n);
1223 pack_size += n;
1224 s.next_out = out_buf;
1225 s.avail_out = out_sz;
1228 switch (status) {
1229 case Z_OK:
1230 case Z_BUF_ERROR:
1231 case Z_STREAM_END:
1232 continue;
1233 default:
1234 die("unexpected deflate failure: %d", status);
1237 git_deflate_end(&s);
1238 git_SHA1_Final(sha1, &c);
1240 if (sha1out)
1241 hashcpy(sha1out, sha1);
1243 e = insert_object(sha1);
1245 if (mark)
1246 insert_mark(mark, e);
1248 if (e->idx.offset) {
1249 duplicate_count_by_type[OBJ_BLOB]++;
1250 truncate_pack(&checkpoint);
1252 } else if (find_sha1_pack(sha1, packed_git)) {
1253 e->type = OBJ_BLOB;
1254 e->pack_id = MAX_PACK_ID;
1255 e->idx.offset = 1; /* just not zero! */
1256 duplicate_count_by_type[OBJ_BLOB]++;
1257 truncate_pack(&checkpoint);
1259 } else {
1260 e->depth = 0;
1261 e->type = OBJ_BLOB;
1262 e->pack_id = pack_id;
1263 e->idx.offset = offset;
1264 e->idx.crc32 = crc32_end(pack_file);
1265 object_count++;
1266 object_count_by_type[OBJ_BLOB]++;
1269 free(in_buf);
1270 free(out_buf);
1273 /* All calls must be guarded by find_object() or find_mark() to
1274 * ensure the 'struct object_entry' passed was written by this
1275 * process instance. We unpack the entry by the offset, avoiding
1276 * the need for the corresponding .idx file. This unpacking rule
1277 * works because we only use OBJ_REF_DELTA within the packfiles
1278 * created by fast-import.
1280 * oe must not be NULL. Such an oe usually comes from giving
1281 * an unknown SHA-1 to find_object() or an undefined mark to
1282 * find_mark(). Callers must test for this condition and use
1283 * the standard read_sha1_file() when it happens.
1285 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1286 * find_mark(), where the mark was reloaded from an existing marks
1287 * file and is referencing an object that this fast-import process
1288 * instance did not write out to a packfile. Callers must test for
1289 * this condition and use read_sha1_file() instead.
1291 static void *gfi_unpack_entry(
1292 struct object_entry *oe,
1293 unsigned long *sizep)
1295 enum object_type type;
1296 struct packed_git *p = all_packs[oe->pack_id];
1297 if (p == pack_data && p->pack_size < (pack_size + 20)) {
1298 /* The object is stored in the packfile we are writing to
1299 * and we have modified it since the last time we scanned
1300 * back to read a previously written object. If an old
1301 * window covered [p->pack_size, p->pack_size + 20) its
1302 * data is stale and is not valid. Closing all windows
1303 * and updating the packfile length ensures we can read
1304 * the newly written data.
1306 close_pack_windows(p);
1307 sha1flush(pack_file);
1309 /* We have to offer 20 bytes additional on the end of
1310 * the packfile as the core unpacker code assumes the
1311 * footer is present at the file end and must promise
1312 * at least 20 bytes within any window it maps. But
1313 * we don't actually create the footer here.
1315 p->pack_size = pack_size + 20;
1317 return unpack_entry(p, oe->idx.offset, &type, sizep);
1320 static const char *get_mode(const char *str, uint16_t *modep)
1322 unsigned char c;
1323 uint16_t mode = 0;
1325 while ((c = *str++) != ' ') {
1326 if (c < '0' || c > '7')
1327 return NULL;
1328 mode = (mode << 3) + (c - '0');
1330 *modep = mode;
1331 return str;
1334 static void load_tree(struct tree_entry *root)
1336 unsigned char *sha1 = root->versions[1].sha1;
1337 struct object_entry *myoe;
1338 struct tree_content *t;
1339 unsigned long size;
1340 char *buf;
1341 const char *c;
1343 root->tree = t = new_tree_content(8);
1344 if (is_null_sha1(sha1))
1345 return;
1347 myoe = find_object(sha1);
1348 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1349 if (myoe->type != OBJ_TREE)
1350 die("Not a tree: %s", sha1_to_hex(sha1));
1351 t->delta_depth = myoe->depth;
1352 buf = gfi_unpack_entry(myoe, &size);
1353 if (!buf)
1354 die("Can't load tree %s", sha1_to_hex(sha1));
1355 } else {
1356 enum object_type type;
1357 buf = read_sha1_file(sha1, &type, &size);
1358 if (!buf || type != OBJ_TREE)
1359 die("Can't load tree %s", sha1_to_hex(sha1));
1362 c = buf;
1363 while (c != (buf + size)) {
1364 struct tree_entry *e = new_tree_entry();
1366 if (t->entry_count == t->entry_capacity)
1367 root->tree = t = grow_tree_content(t, t->entry_count);
1368 t->entries[t->entry_count++] = e;
1370 e->tree = NULL;
1371 c = get_mode(c, &e->versions[1].mode);
1372 if (!c)
1373 die("Corrupt mode in %s", sha1_to_hex(sha1));
1374 e->versions[0].mode = e->versions[1].mode;
1375 e->name = to_atom(c, strlen(c));
1376 c += e->name->str_len + 1;
1377 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1378 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1379 c += 20;
1381 free(buf);
1384 static int tecmp0 (const void *_a, const void *_b)
1386 struct tree_entry *a = *((struct tree_entry**)_a);
1387 struct tree_entry *b = *((struct tree_entry**)_b);
1388 return base_name_compare(
1389 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1390 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1393 static int tecmp1 (const void *_a, const void *_b)
1395 struct tree_entry *a = *((struct tree_entry**)_a);
1396 struct tree_entry *b = *((struct tree_entry**)_b);
1397 return base_name_compare(
1398 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1399 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1402 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1404 size_t maxlen = 0;
1405 unsigned int i;
1407 if (!v)
1408 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1409 else
1410 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1412 for (i = 0; i < t->entry_count; i++) {
1413 if (t->entries[i]->versions[v].mode)
1414 maxlen += t->entries[i]->name->str_len + 34;
1417 strbuf_reset(b);
1418 strbuf_grow(b, maxlen);
1419 for (i = 0; i < t->entry_count; i++) {
1420 struct tree_entry *e = t->entries[i];
1421 if (!e->versions[v].mode)
1422 continue;
1423 strbuf_addf(b, "%o %s%c",
1424 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1425 e->name->str_dat, '\0');
1426 strbuf_add(b, e->versions[v].sha1, 20);
1430 static void store_tree(struct tree_entry *root)
1432 struct tree_content *t;
1433 unsigned int i, j, del;
1434 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1435 struct object_entry *le = NULL;
1437 if (!is_null_sha1(root->versions[1].sha1))
1438 return;
1440 if (!root->tree)
1441 load_tree(root);
1442 t = root->tree;
1444 for (i = 0; i < t->entry_count; i++) {
1445 if (t->entries[i]->tree)
1446 store_tree(t->entries[i]);
1449 if (!(root->versions[0].mode & NO_DELTA))
1450 le = find_object(root->versions[0].sha1);
1451 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1452 mktree(t, 0, &old_tree);
1453 lo.data = old_tree;
1454 lo.offset = le->idx.offset;
1455 lo.depth = t->delta_depth;
1458 mktree(t, 1, &new_tree);
1459 store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1461 t->delta_depth = lo.depth;
1462 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1463 struct tree_entry *e = t->entries[i];
1464 if (e->versions[1].mode) {
1465 e->versions[0].mode = e->versions[1].mode;
1466 hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1467 t->entries[j++] = e;
1468 } else {
1469 release_tree_entry(e);
1470 del++;
1473 t->entry_count -= del;
1476 static void tree_content_replace(
1477 struct tree_entry *root,
1478 const unsigned char *sha1,
1479 const uint16_t mode,
1480 struct tree_content *newtree)
1482 if (!S_ISDIR(mode))
1483 die("Root cannot be a non-directory");
1484 hashclr(root->versions[0].sha1);
1485 hashcpy(root->versions[1].sha1, sha1);
1486 if (root->tree)
1487 release_tree_content_recursive(root->tree);
1488 root->tree = newtree;
1491 static int tree_content_set(
1492 struct tree_entry *root,
1493 const char *p,
1494 const unsigned char *sha1,
1495 const uint16_t mode,
1496 struct tree_content *subtree)
1498 struct tree_content *t;
1499 const char *slash1;
1500 unsigned int i, n;
1501 struct tree_entry *e;
1503 slash1 = strchrnul(p, '/');
1504 n = slash1 - p;
1505 if (!n)
1506 die("Empty path component found in input");
1507 if (!*slash1 && !S_ISDIR(mode) && subtree)
1508 die("Non-directories cannot have subtrees");
1510 if (!root->tree)
1511 load_tree(root);
1512 t = root->tree;
1513 for (i = 0; i < t->entry_count; i++) {
1514 e = t->entries[i];
1515 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1516 if (!*slash1) {
1517 if (!S_ISDIR(mode)
1518 && e->versions[1].mode == mode
1519 && !hashcmp(e->versions[1].sha1, sha1))
1520 return 0;
1521 e->versions[1].mode = mode;
1522 hashcpy(e->versions[1].sha1, sha1);
1523 if (e->tree)
1524 release_tree_content_recursive(e->tree);
1525 e->tree = subtree;
1528 * We need to leave e->versions[0].sha1 alone
1529 * to avoid modifying the preimage tree used
1530 * when writing out the parent directory.
1531 * But after replacing the subdir with a
1532 * completely different one, it's not a good
1533 * delta base any more, and besides, we've
1534 * thrown away the tree entries needed to
1535 * make a delta against it.
1537 * So let's just explicitly disable deltas
1538 * for the subtree.
1540 if (S_ISDIR(e->versions[0].mode))
1541 e->versions[0].mode |= NO_DELTA;
1543 hashclr(root->versions[1].sha1);
1544 return 1;
1546 if (!S_ISDIR(e->versions[1].mode)) {
1547 e->tree = new_tree_content(8);
1548 e->versions[1].mode = S_IFDIR;
1550 if (!e->tree)
1551 load_tree(e);
1552 if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1553 hashclr(root->versions[1].sha1);
1554 return 1;
1556 return 0;
1560 if (t->entry_count == t->entry_capacity)
1561 root->tree = t = grow_tree_content(t, t->entry_count);
1562 e = new_tree_entry();
1563 e->name = to_atom(p, n);
1564 e->versions[0].mode = 0;
1565 hashclr(e->versions[0].sha1);
1566 t->entries[t->entry_count++] = e;
1567 if (*slash1) {
1568 e->tree = new_tree_content(8);
1569 e->versions[1].mode = S_IFDIR;
1570 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1571 } else {
1572 e->tree = subtree;
1573 e->versions[1].mode = mode;
1574 hashcpy(e->versions[1].sha1, sha1);
1576 hashclr(root->versions[1].sha1);
1577 return 1;
1580 static int tree_content_remove(
1581 struct tree_entry *root,
1582 const char *p,
1583 struct tree_entry *backup_leaf,
1584 int allow_root)
1586 struct tree_content *t;
1587 const char *slash1;
1588 unsigned int i, n;
1589 struct tree_entry *e;
1591 slash1 = strchrnul(p, '/');
1592 n = slash1 - p;
1594 if (!root->tree)
1595 load_tree(root);
1597 if (!*p && allow_root) {
1598 e = root;
1599 goto del_entry;
1602 t = root->tree;
1603 for (i = 0; i < t->entry_count; i++) {
1604 e = t->entries[i];
1605 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1606 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1608 * If p names a file in some subdirectory, and a
1609 * file or symlink matching the name of the
1610 * parent directory of p exists, then p cannot
1611 * exist and need not be deleted.
1613 return 1;
1614 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1615 goto del_entry;
1616 if (!e->tree)
1617 load_tree(e);
1618 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1619 for (n = 0; n < e->tree->entry_count; n++) {
1620 if (e->tree->entries[n]->versions[1].mode) {
1621 hashclr(root->versions[1].sha1);
1622 return 1;
1625 backup_leaf = NULL;
1626 goto del_entry;
1628 return 0;
1631 return 0;
1633 del_entry:
1634 if (backup_leaf)
1635 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1636 else if (e->tree)
1637 release_tree_content_recursive(e->tree);
1638 e->tree = NULL;
1639 e->versions[1].mode = 0;
1640 hashclr(e->versions[1].sha1);
1641 hashclr(root->versions[1].sha1);
1642 return 1;
1645 static int tree_content_get(
1646 struct tree_entry *root,
1647 const char *p,
1648 struct tree_entry *leaf,
1649 int allow_root)
1651 struct tree_content *t;
1652 const char *slash1;
1653 unsigned int i, n;
1654 struct tree_entry *e;
1656 slash1 = strchrnul(p, '/');
1657 n = slash1 - p;
1658 if (!n && !allow_root)
1659 die("Empty path component found in input");
1661 if (!root->tree)
1662 load_tree(root);
1664 if (!n) {
1665 e = root;
1666 goto found_entry;
1669 t = root->tree;
1670 for (i = 0; i < t->entry_count; i++) {
1671 e = t->entries[i];
1672 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1673 if (!*slash1)
1674 goto found_entry;
1675 if (!S_ISDIR(e->versions[1].mode))
1676 return 0;
1677 if (!e->tree)
1678 load_tree(e);
1679 return tree_content_get(e, slash1 + 1, leaf, 0);
1682 return 0;
1684 found_entry:
1685 memcpy(leaf, e, sizeof(*leaf));
1686 if (e->tree && is_null_sha1(e->versions[1].sha1))
1687 leaf->tree = dup_tree_content(e->tree);
1688 else
1689 leaf->tree = NULL;
1690 return 1;
1693 static int update_branch(struct branch *b)
1695 static const char *msg = "fast-import";
1696 struct ref_transaction *transaction;
1697 unsigned char old_sha1[20];
1698 struct strbuf err = STRBUF_INIT;
1700 if (is_null_sha1(b->sha1)) {
1701 if (b->delete)
1702 delete_ref(b->name, NULL, 0);
1703 return 0;
1705 if (read_ref(b->name, old_sha1))
1706 hashclr(old_sha1);
1707 if (!force_update && !is_null_sha1(old_sha1)) {
1708 struct commit *old_cmit, *new_cmit;
1710 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1711 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1712 if (!old_cmit || !new_cmit)
1713 return error("Branch %s is missing commits.", b->name);
1715 if (!in_merge_bases(old_cmit, new_cmit)) {
1716 warning("Not updating %s"
1717 " (new tip %s does not contain %s)",
1718 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1719 return -1;
1722 transaction = ref_transaction_begin(&err);
1723 if (!transaction ||
1724 ref_transaction_update(transaction, b->name, b->sha1, old_sha1,
1725 0, msg, &err) ||
1726 ref_transaction_commit(transaction, &err)) {
1727 ref_transaction_free(transaction);
1728 error("%s", err.buf);
1729 strbuf_release(&err);
1730 return -1;
1732 ref_transaction_free(transaction);
1733 strbuf_release(&err);
1734 return 0;
1737 static void dump_branches(void)
1739 unsigned int i;
1740 struct branch *b;
1742 for (i = 0; i < branch_table_sz; i++) {
1743 for (b = branch_table[i]; b; b = b->table_next_branch)
1744 failure |= update_branch(b);
1748 static void dump_tags(void)
1750 static const char *msg = "fast-import";
1751 struct tag *t;
1752 struct strbuf ref_name = STRBUF_INIT;
1753 struct strbuf err = STRBUF_INIT;
1754 struct ref_transaction *transaction;
1756 transaction = ref_transaction_begin(&err);
1757 if (!transaction) {
1758 failure |= error("%s", err.buf);
1759 goto cleanup;
1761 for (t = first_tag; t; t = t->next_tag) {
1762 strbuf_reset(&ref_name);
1763 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1765 if (ref_transaction_update(transaction, ref_name.buf,
1766 t->sha1, NULL, 0, msg, &err)) {
1767 failure |= error("%s", err.buf);
1768 goto cleanup;
1771 if (ref_transaction_commit(transaction, &err))
1772 failure |= error("%s", err.buf);
1774 cleanup:
1775 ref_transaction_free(transaction);
1776 strbuf_release(&ref_name);
1777 strbuf_release(&err);
1780 static void dump_marks_helper(FILE *f,
1781 uintmax_t base,
1782 struct mark_set *m)
1784 uintmax_t k;
1785 if (m->shift) {
1786 for (k = 0; k < 1024; k++) {
1787 if (m->data.sets[k])
1788 dump_marks_helper(f, base + (k << m->shift),
1789 m->data.sets[k]);
1791 } else {
1792 for (k = 0; k < 1024; k++) {
1793 if (m->data.marked[k])
1794 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1795 sha1_to_hex(m->data.marked[k]->idx.sha1));
1800 static void dump_marks(void)
1802 static struct lock_file mark_lock;
1803 FILE *f;
1805 if (!export_marks_file)
1806 return;
1808 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1809 failure |= error("Unable to write marks file %s: %s",
1810 export_marks_file, strerror(errno));
1811 return;
1814 f = fdopen_lock_file(&mark_lock, "w");
1815 if (!f) {
1816 int saved_errno = errno;
1817 rollback_lock_file(&mark_lock);
1818 failure |= error("Unable to write marks file %s: %s",
1819 export_marks_file, strerror(saved_errno));
1820 return;
1823 dump_marks_helper(f, 0, marks);
1824 if (commit_lock_file(&mark_lock)) {
1825 failure |= error("Unable to write file %s: %s",
1826 export_marks_file, strerror(errno));
1827 return;
1831 static void read_marks(void)
1833 char line[512];
1834 FILE *f = fopen(import_marks_file, "r");
1835 if (f)
1837 else if (import_marks_file_ignore_missing && errno == ENOENT)
1838 return; /* Marks file does not exist */
1839 else
1840 die_errno("cannot read '%s'", import_marks_file);
1841 while (fgets(line, sizeof(line), f)) {
1842 uintmax_t mark;
1843 char *end;
1844 unsigned char sha1[20];
1845 struct object_entry *e;
1847 end = strchr(line, '\n');
1848 if (line[0] != ':' || !end)
1849 die("corrupt mark line: %s", line);
1850 *end = 0;
1851 mark = strtoumax(line + 1, &end, 10);
1852 if (!mark || end == line + 1
1853 || *end != ' ' || get_sha1_hex(end + 1, sha1))
1854 die("corrupt mark line: %s", line);
1855 e = find_object(sha1);
1856 if (!e) {
1857 enum object_type type = sha1_object_info(sha1, NULL);
1858 if (type < 0)
1859 die("object not found: %s", sha1_to_hex(sha1));
1860 e = insert_object(sha1);
1861 e->type = type;
1862 e->pack_id = MAX_PACK_ID;
1863 e->idx.offset = 1; /* just not zero! */
1865 insert_mark(mark, e);
1867 fclose(f);
1871 static int read_next_command(void)
1873 static int stdin_eof = 0;
1875 if (stdin_eof) {
1876 unread_command_buf = 0;
1877 return EOF;
1880 for (;;) {
1881 const char *p;
1883 if (unread_command_buf) {
1884 unread_command_buf = 0;
1885 } else {
1886 struct recent_command *rc;
1888 strbuf_detach(&command_buf, NULL);
1889 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1890 if (stdin_eof)
1891 return EOF;
1893 if (!seen_data_command
1894 && !starts_with(command_buf.buf, "feature ")
1895 && !starts_with(command_buf.buf, "option ")) {
1896 parse_argv();
1899 rc = rc_free;
1900 if (rc)
1901 rc_free = rc->next;
1902 else {
1903 rc = cmd_hist.next;
1904 cmd_hist.next = rc->next;
1905 cmd_hist.next->prev = &cmd_hist;
1906 free(rc->buf);
1909 rc->buf = command_buf.buf;
1910 rc->prev = cmd_tail;
1911 rc->next = cmd_hist.prev;
1912 rc->prev->next = rc;
1913 cmd_tail = rc;
1915 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1916 parse_get_mark(p);
1917 continue;
1919 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1920 parse_cat_blob(p);
1921 continue;
1923 if (command_buf.buf[0] == '#')
1924 continue;
1925 return 0;
1929 static void skip_optional_lf(void)
1931 int term_char = fgetc(stdin);
1932 if (term_char != '\n' && term_char != EOF)
1933 ungetc(term_char, stdin);
1936 static void parse_mark(void)
1938 const char *v;
1939 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1940 next_mark = strtoumax(v, NULL, 10);
1941 read_next_command();
1943 else
1944 next_mark = 0;
1947 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1949 const char *data;
1950 strbuf_reset(sb);
1952 if (!skip_prefix(command_buf.buf, "data ", &data))
1953 die("Expected 'data n' command, found: %s", command_buf.buf);
1955 if (skip_prefix(data, "<<", &data)) {
1956 char *term = xstrdup(data);
1957 size_t term_len = command_buf.len - (data - command_buf.buf);
1959 strbuf_detach(&command_buf, NULL);
1960 for (;;) {
1961 if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1962 die("EOF in data (terminator '%s' not found)", term);
1963 if (term_len == command_buf.len
1964 && !strcmp(term, command_buf.buf))
1965 break;
1966 strbuf_addbuf(sb, &command_buf);
1967 strbuf_addch(sb, '\n');
1969 free(term);
1971 else {
1972 uintmax_t len = strtoumax(data, NULL, 10);
1973 size_t n = 0, length = (size_t)len;
1975 if (limit && limit < len) {
1976 *len_res = len;
1977 return 0;
1979 if (length < len)
1980 die("data is too large to use in this context");
1982 while (n < length) {
1983 size_t s = strbuf_fread(sb, length - n, stdin);
1984 if (!s && feof(stdin))
1985 die("EOF in data (%lu bytes remaining)",
1986 (unsigned long)(length - n));
1987 n += s;
1991 skip_optional_lf();
1992 return 1;
1995 static int validate_raw_date(const char *src, struct strbuf *result)
1997 const char *orig_src = src;
1998 char *endp;
1999 unsigned long num;
2001 errno = 0;
2003 num = strtoul(src, &endp, 10);
2004 /* NEEDSWORK: perhaps check for reasonable values? */
2005 if (errno || endp == src || *endp != ' ')
2006 return -1;
2008 src = endp + 1;
2009 if (*src != '-' && *src != '+')
2010 return -1;
2012 num = strtoul(src + 1, &endp, 10);
2013 if (errno || endp == src + 1 || *endp || 1400 < num)
2014 return -1;
2016 strbuf_addstr(result, orig_src);
2017 return 0;
2020 static char *parse_ident(const char *buf)
2022 const char *ltgt;
2023 size_t name_len;
2024 struct strbuf ident = STRBUF_INIT;
2026 /* ensure there is a space delimiter even if there is no name */
2027 if (*buf == '<')
2028 --buf;
2030 ltgt = buf + strcspn(buf, "<>");
2031 if (*ltgt != '<')
2032 die("Missing < in ident string: %s", buf);
2033 if (ltgt != buf && ltgt[-1] != ' ')
2034 die("Missing space before < in ident string: %s", buf);
2035 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2036 if (*ltgt != '>')
2037 die("Missing > in ident string: %s", buf);
2038 ltgt++;
2039 if (*ltgt != ' ')
2040 die("Missing space after > in ident string: %s", buf);
2041 ltgt++;
2042 name_len = ltgt - buf;
2043 strbuf_add(&ident, buf, name_len);
2045 switch (whenspec) {
2046 case WHENSPEC_RAW:
2047 if (validate_raw_date(ltgt, &ident) < 0)
2048 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2049 break;
2050 case WHENSPEC_RFC2822:
2051 if (parse_date(ltgt, &ident) < 0)
2052 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2053 break;
2054 case WHENSPEC_NOW:
2055 if (strcmp("now", ltgt))
2056 die("Date in ident must be 'now': %s", buf);
2057 datestamp(&ident);
2058 break;
2061 return strbuf_detach(&ident, NULL);
2064 static void parse_and_store_blob(
2065 struct last_object *last,
2066 unsigned char *sha1out,
2067 uintmax_t mark)
2069 static struct strbuf buf = STRBUF_INIT;
2070 uintmax_t len;
2072 if (parse_data(&buf, big_file_threshold, &len))
2073 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2074 else {
2075 if (last) {
2076 strbuf_release(&last->data);
2077 last->offset = 0;
2078 last->depth = 0;
2080 stream_blob(len, sha1out, mark);
2081 skip_optional_lf();
2085 static void parse_new_blob(void)
2087 read_next_command();
2088 parse_mark();
2089 parse_and_store_blob(&last_blob, NULL, next_mark);
2092 static void unload_one_branch(void)
2094 while (cur_active_branches
2095 && cur_active_branches >= max_active_branches) {
2096 uintmax_t min_commit = ULONG_MAX;
2097 struct branch *e, *l = NULL, *p = NULL;
2099 for (e = active_branches; e; e = e->active_next_branch) {
2100 if (e->last_commit < min_commit) {
2101 p = l;
2102 min_commit = e->last_commit;
2104 l = e;
2107 if (p) {
2108 e = p->active_next_branch;
2109 p->active_next_branch = e->active_next_branch;
2110 } else {
2111 e = active_branches;
2112 active_branches = e->active_next_branch;
2114 e->active = 0;
2115 e->active_next_branch = NULL;
2116 if (e->branch_tree.tree) {
2117 release_tree_content_recursive(e->branch_tree.tree);
2118 e->branch_tree.tree = NULL;
2120 cur_active_branches--;
2124 static void load_branch(struct branch *b)
2126 load_tree(&b->branch_tree);
2127 if (!b->active) {
2128 b->active = 1;
2129 b->active_next_branch = active_branches;
2130 active_branches = b;
2131 cur_active_branches++;
2132 branch_load_count++;
2136 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2138 unsigned char fanout = 0;
2139 while ((num_notes >>= 8))
2140 fanout++;
2141 return fanout;
2144 static void construct_path_with_fanout(const char *hex_sha1,
2145 unsigned char fanout, char *path)
2147 unsigned int i = 0, j = 0;
2148 if (fanout >= 20)
2149 die("Too large fanout (%u)", fanout);
2150 while (fanout) {
2151 path[i++] = hex_sha1[j++];
2152 path[i++] = hex_sha1[j++];
2153 path[i++] = '/';
2154 fanout--;
2156 memcpy(path + i, hex_sha1 + j, 40 - j);
2157 path[i + 40 - j] = '\0';
2160 static uintmax_t do_change_note_fanout(
2161 struct tree_entry *orig_root, struct tree_entry *root,
2162 char *hex_sha1, unsigned int hex_sha1_len,
2163 char *fullpath, unsigned int fullpath_len,
2164 unsigned char fanout)
2166 struct tree_content *t = root->tree;
2167 struct tree_entry *e, leaf;
2168 unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2169 uintmax_t num_notes = 0;
2170 unsigned char sha1[20];
2171 char realpath[60];
2173 for (i = 0; t && i < t->entry_count; i++) {
2174 e = t->entries[i];
2175 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2176 tmp_fullpath_len = fullpath_len;
2179 * We're interested in EITHER existing note entries (entries
2180 * with exactly 40 hex chars in path, not including directory
2181 * separators), OR directory entries that may contain note
2182 * entries (with < 40 hex chars in path).
2183 * Also, each path component in a note entry must be a multiple
2184 * of 2 chars.
2186 if (!e->versions[1].mode ||
2187 tmp_hex_sha1_len > 40 ||
2188 e->name->str_len % 2)
2189 continue;
2191 /* This _may_ be a note entry, or a subdir containing notes */
2192 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2193 e->name->str_len);
2194 if (tmp_fullpath_len)
2195 fullpath[tmp_fullpath_len++] = '/';
2196 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2197 e->name->str_len);
2198 tmp_fullpath_len += e->name->str_len;
2199 fullpath[tmp_fullpath_len] = '\0';
2201 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2202 /* This is a note entry */
2203 if (fanout == 0xff) {
2204 /* Counting mode, no rename */
2205 num_notes++;
2206 continue;
2208 construct_path_with_fanout(hex_sha1, fanout, realpath);
2209 if (!strcmp(fullpath, realpath)) {
2210 /* Note entry is in correct location */
2211 num_notes++;
2212 continue;
2215 /* Rename fullpath to realpath */
2216 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2217 die("Failed to remove path %s", fullpath);
2218 tree_content_set(orig_root, realpath,
2219 leaf.versions[1].sha1,
2220 leaf.versions[1].mode,
2221 leaf.tree);
2222 } else if (S_ISDIR(e->versions[1].mode)) {
2223 /* This is a subdir that may contain note entries */
2224 if (!e->tree)
2225 load_tree(e);
2226 num_notes += do_change_note_fanout(orig_root, e,
2227 hex_sha1, tmp_hex_sha1_len,
2228 fullpath, tmp_fullpath_len, fanout);
2231 /* The above may have reallocated the current tree_content */
2232 t = root->tree;
2234 return num_notes;
2237 static uintmax_t change_note_fanout(struct tree_entry *root,
2238 unsigned char fanout)
2240 char hex_sha1[40], path[60];
2241 return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2245 * Given a pointer into a string, parse a mark reference:
2247 * idnum ::= ':' bigint;
2249 * Return the first character after the value in *endptr.
2251 * Complain if the following character is not what is expected,
2252 * either a space or end of the string.
2254 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2256 uintmax_t mark;
2258 assert(*p == ':');
2259 p++;
2260 mark = strtoumax(p, endptr, 10);
2261 if (*endptr == p)
2262 die("No value after ':' in mark: %s", command_buf.buf);
2263 return mark;
2267 * Parse the mark reference, and complain if this is not the end of
2268 * the string.
2270 static uintmax_t parse_mark_ref_eol(const char *p)
2272 char *end;
2273 uintmax_t mark;
2275 mark = parse_mark_ref(p, &end);
2276 if (*end != '\0')
2277 die("Garbage after mark: %s", command_buf.buf);
2278 return mark;
2282 * Parse the mark reference, demanding a trailing space. Return a
2283 * pointer to the space.
2285 static uintmax_t parse_mark_ref_space(const char **p)
2287 uintmax_t mark;
2288 char *end;
2290 mark = parse_mark_ref(*p, &end);
2291 if (*end++ != ' ')
2292 die("Missing space after mark: %s", command_buf.buf);
2293 *p = end;
2294 return mark;
2297 static void file_change_m(const char *p, struct branch *b)
2299 static struct strbuf uq = STRBUF_INIT;
2300 const char *endp;
2301 struct object_entry *oe;
2302 unsigned char sha1[20];
2303 uint16_t mode, inline_data = 0;
2305 p = get_mode(p, &mode);
2306 if (!p)
2307 die("Corrupt mode: %s", command_buf.buf);
2308 switch (mode) {
2309 case 0644:
2310 case 0755:
2311 mode |= S_IFREG;
2312 case S_IFREG | 0644:
2313 case S_IFREG | 0755:
2314 case S_IFLNK:
2315 case S_IFDIR:
2316 case S_IFGITLINK:
2317 /* ok */
2318 break;
2319 default:
2320 die("Corrupt mode: %s", command_buf.buf);
2323 if (*p == ':') {
2324 oe = find_mark(parse_mark_ref_space(&p));
2325 hashcpy(sha1, oe->idx.sha1);
2326 } else if (skip_prefix(p, "inline ", &p)) {
2327 inline_data = 1;
2328 oe = NULL; /* not used with inline_data, but makes gcc happy */
2329 } else {
2330 if (get_sha1_hex(p, sha1))
2331 die("Invalid dataref: %s", command_buf.buf);
2332 oe = find_object(sha1);
2333 p += 40;
2334 if (*p++ != ' ')
2335 die("Missing space after SHA1: %s", command_buf.buf);
2338 strbuf_reset(&uq);
2339 if (!unquote_c_style(&uq, p, &endp)) {
2340 if (*endp)
2341 die("Garbage after path in: %s", command_buf.buf);
2342 p = uq.buf;
2345 /* Git does not track empty, non-toplevel directories. */
2346 if (S_ISDIR(mode) && !hashcmp(sha1, EMPTY_TREE_SHA1_BIN) && *p) {
2347 tree_content_remove(&b->branch_tree, p, NULL, 0);
2348 return;
2351 if (S_ISGITLINK(mode)) {
2352 if (inline_data)
2353 die("Git links cannot be specified 'inline': %s",
2354 command_buf.buf);
2355 else if (oe) {
2356 if (oe->type != OBJ_COMMIT)
2357 die("Not a commit (actually a %s): %s",
2358 typename(oe->type), command_buf.buf);
2361 * Accept the sha1 without checking; it expected to be in
2362 * another repository.
2364 } else if (inline_data) {
2365 if (S_ISDIR(mode))
2366 die("Directories cannot be specified 'inline': %s",
2367 command_buf.buf);
2368 if (p != uq.buf) {
2369 strbuf_addstr(&uq, p);
2370 p = uq.buf;
2372 read_next_command();
2373 parse_and_store_blob(&last_blob, sha1, 0);
2374 } else {
2375 enum object_type expected = S_ISDIR(mode) ?
2376 OBJ_TREE: OBJ_BLOB;
2377 enum object_type type = oe ? oe->type :
2378 sha1_object_info(sha1, NULL);
2379 if (type < 0)
2380 die("%s not found: %s",
2381 S_ISDIR(mode) ? "Tree" : "Blob",
2382 command_buf.buf);
2383 if (type != expected)
2384 die("Not a %s (actually a %s): %s",
2385 typename(expected), typename(type),
2386 command_buf.buf);
2389 if (!*p) {
2390 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2391 return;
2393 tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2396 static void file_change_d(const char *p, struct branch *b)
2398 static struct strbuf uq = STRBUF_INIT;
2399 const char *endp;
2401 strbuf_reset(&uq);
2402 if (!unquote_c_style(&uq, p, &endp)) {
2403 if (*endp)
2404 die("Garbage after path in: %s", command_buf.buf);
2405 p = uq.buf;
2407 tree_content_remove(&b->branch_tree, p, NULL, 1);
2410 static void file_change_cr(const char *s, struct branch *b, int rename)
2412 const char *d;
2413 static struct strbuf s_uq = STRBUF_INIT;
2414 static struct strbuf d_uq = STRBUF_INIT;
2415 const char *endp;
2416 struct tree_entry leaf;
2418 strbuf_reset(&s_uq);
2419 if (!unquote_c_style(&s_uq, s, &endp)) {
2420 if (*endp != ' ')
2421 die("Missing space after source: %s", command_buf.buf);
2422 } else {
2423 endp = strchr(s, ' ');
2424 if (!endp)
2425 die("Missing space after source: %s", command_buf.buf);
2426 strbuf_add(&s_uq, s, endp - s);
2428 s = s_uq.buf;
2430 endp++;
2431 if (!*endp)
2432 die("Missing dest: %s", command_buf.buf);
2434 d = endp;
2435 strbuf_reset(&d_uq);
2436 if (!unquote_c_style(&d_uq, d, &endp)) {
2437 if (*endp)
2438 die("Garbage after dest in: %s", command_buf.buf);
2439 d = d_uq.buf;
2442 memset(&leaf, 0, sizeof(leaf));
2443 if (rename)
2444 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2445 else
2446 tree_content_get(&b->branch_tree, s, &leaf, 1);
2447 if (!leaf.versions[1].mode)
2448 die("Path %s not in branch", s);
2449 if (!*d) { /* C "path/to/subdir" "" */
2450 tree_content_replace(&b->branch_tree,
2451 leaf.versions[1].sha1,
2452 leaf.versions[1].mode,
2453 leaf.tree);
2454 return;
2456 tree_content_set(&b->branch_tree, d,
2457 leaf.versions[1].sha1,
2458 leaf.versions[1].mode,
2459 leaf.tree);
2462 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2464 static struct strbuf uq = STRBUF_INIT;
2465 struct object_entry *oe;
2466 struct branch *s;
2467 unsigned char sha1[20], commit_sha1[20];
2468 char path[60];
2469 uint16_t inline_data = 0;
2470 unsigned char new_fanout;
2473 * When loading a branch, we don't traverse its tree to count the real
2474 * number of notes (too expensive to do this for all non-note refs).
2475 * This means that recently loaded notes refs might incorrectly have
2476 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2478 * Fix this by traversing the tree and counting the number of notes
2479 * when b->num_notes == 0. If the notes tree is truly empty, the
2480 * calculation should not take long.
2482 if (b->num_notes == 0 && *old_fanout == 0) {
2483 /* Invoke change_note_fanout() in "counting mode". */
2484 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2485 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2488 /* Now parse the notemodify command. */
2489 /* <dataref> or 'inline' */
2490 if (*p == ':') {
2491 oe = find_mark(parse_mark_ref_space(&p));
2492 hashcpy(sha1, oe->idx.sha1);
2493 } else if (skip_prefix(p, "inline ", &p)) {
2494 inline_data = 1;
2495 oe = NULL; /* not used with inline_data, but makes gcc happy */
2496 } else {
2497 if (get_sha1_hex(p, sha1))
2498 die("Invalid dataref: %s", command_buf.buf);
2499 oe = find_object(sha1);
2500 p += 40;
2501 if (*p++ != ' ')
2502 die("Missing space after SHA1: %s", command_buf.buf);
2505 /* <commit-ish> */
2506 s = lookup_branch(p);
2507 if (s) {
2508 if (is_null_sha1(s->sha1))
2509 die("Can't add a note on empty branch.");
2510 hashcpy(commit_sha1, s->sha1);
2511 } else if (*p == ':') {
2512 uintmax_t commit_mark = parse_mark_ref_eol(p);
2513 struct object_entry *commit_oe = find_mark(commit_mark);
2514 if (commit_oe->type != OBJ_COMMIT)
2515 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2516 hashcpy(commit_sha1, commit_oe->idx.sha1);
2517 } else if (!get_sha1(p, commit_sha1)) {
2518 unsigned long size;
2519 char *buf = read_object_with_reference(commit_sha1,
2520 commit_type, &size, commit_sha1);
2521 if (!buf || size < 46)
2522 die("Not a valid commit: %s", p);
2523 free(buf);
2524 } else
2525 die("Invalid ref name or SHA1 expression: %s", p);
2527 if (inline_data) {
2528 if (p != uq.buf) {
2529 strbuf_addstr(&uq, p);
2530 p = uq.buf;
2532 read_next_command();
2533 parse_and_store_blob(&last_blob, sha1, 0);
2534 } else if (oe) {
2535 if (oe->type != OBJ_BLOB)
2536 die("Not a blob (actually a %s): %s",
2537 typename(oe->type), command_buf.buf);
2538 } else if (!is_null_sha1(sha1)) {
2539 enum object_type type = sha1_object_info(sha1, NULL);
2540 if (type < 0)
2541 die("Blob not found: %s", command_buf.buf);
2542 if (type != OBJ_BLOB)
2543 die("Not a blob (actually a %s): %s",
2544 typename(type), command_buf.buf);
2547 construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2548 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2549 b->num_notes--;
2551 if (is_null_sha1(sha1))
2552 return; /* nothing to insert */
2554 b->num_notes++;
2555 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2556 construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2557 tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2560 static void file_change_deleteall(struct branch *b)
2562 release_tree_content_recursive(b->branch_tree.tree);
2563 hashclr(b->branch_tree.versions[0].sha1);
2564 hashclr(b->branch_tree.versions[1].sha1);
2565 load_tree(&b->branch_tree);
2566 b->num_notes = 0;
2569 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2571 if (!buf || size < 46)
2572 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2573 if (memcmp("tree ", buf, 5)
2574 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2575 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2576 hashcpy(b->branch_tree.versions[0].sha1,
2577 b->branch_tree.versions[1].sha1);
2580 static void parse_from_existing(struct branch *b)
2582 if (is_null_sha1(b->sha1)) {
2583 hashclr(b->branch_tree.versions[0].sha1);
2584 hashclr(b->branch_tree.versions[1].sha1);
2585 } else {
2586 unsigned long size;
2587 char *buf;
2589 buf = read_object_with_reference(b->sha1,
2590 commit_type, &size, b->sha1);
2591 parse_from_commit(b, buf, size);
2592 free(buf);
2596 static int parse_from(struct branch *b)
2598 const char *from;
2599 struct branch *s;
2600 unsigned char sha1[20];
2602 if (!skip_prefix(command_buf.buf, "from ", &from))
2603 return 0;
2605 hashcpy(sha1, b->branch_tree.versions[1].sha1);
2607 s = lookup_branch(from);
2608 if (b == s)
2609 die("Can't create a branch from itself: %s", b->name);
2610 else if (s) {
2611 unsigned char *t = s->branch_tree.versions[1].sha1;
2612 hashcpy(b->sha1, s->sha1);
2613 hashcpy(b->branch_tree.versions[0].sha1, t);
2614 hashcpy(b->branch_tree.versions[1].sha1, t);
2615 } else if (*from == ':') {
2616 uintmax_t idnum = parse_mark_ref_eol(from);
2617 struct object_entry *oe = find_mark(idnum);
2618 if (oe->type != OBJ_COMMIT)
2619 die("Mark :%" PRIuMAX " not a commit", idnum);
2620 if (hashcmp(b->sha1, oe->idx.sha1)) {
2621 hashcpy(b->sha1, oe->idx.sha1);
2622 if (oe->pack_id != MAX_PACK_ID) {
2623 unsigned long size;
2624 char *buf = gfi_unpack_entry(oe, &size);
2625 parse_from_commit(b, buf, size);
2626 free(buf);
2627 } else
2628 parse_from_existing(b);
2630 } else if (!get_sha1(from, b->sha1)) {
2631 parse_from_existing(b);
2632 if (is_null_sha1(b->sha1))
2633 b->delete = 1;
2635 else
2636 die("Invalid ref name or SHA1 expression: %s", from);
2638 if (b->branch_tree.tree && hashcmp(sha1, b->branch_tree.versions[1].sha1)) {
2639 release_tree_content_recursive(b->branch_tree.tree);
2640 b->branch_tree.tree = NULL;
2643 read_next_command();
2644 return 1;
2647 static struct hash_list *parse_merge(unsigned int *count)
2649 struct hash_list *list = NULL, **tail = &list, *n;
2650 const char *from;
2651 struct branch *s;
2653 *count = 0;
2654 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2655 n = xmalloc(sizeof(*n));
2656 s = lookup_branch(from);
2657 if (s)
2658 hashcpy(n->sha1, s->sha1);
2659 else if (*from == ':') {
2660 uintmax_t idnum = parse_mark_ref_eol(from);
2661 struct object_entry *oe = find_mark(idnum);
2662 if (oe->type != OBJ_COMMIT)
2663 die("Mark :%" PRIuMAX " not a commit", idnum);
2664 hashcpy(n->sha1, oe->idx.sha1);
2665 } else if (!get_sha1(from, n->sha1)) {
2666 unsigned long size;
2667 char *buf = read_object_with_reference(n->sha1,
2668 commit_type, &size, n->sha1);
2669 if (!buf || size < 46)
2670 die("Not a valid commit: %s", from);
2671 free(buf);
2672 } else
2673 die("Invalid ref name or SHA1 expression: %s", from);
2675 n->next = NULL;
2676 *tail = n;
2677 tail = &n->next;
2679 (*count)++;
2680 read_next_command();
2682 return list;
2685 static void parse_new_commit(const char *arg)
2687 static struct strbuf msg = STRBUF_INIT;
2688 struct branch *b;
2689 char *author = NULL;
2690 char *committer = NULL;
2691 struct hash_list *merge_list = NULL;
2692 unsigned int merge_count;
2693 unsigned char prev_fanout, new_fanout;
2694 const char *v;
2696 b = lookup_branch(arg);
2697 if (!b)
2698 b = new_branch(arg);
2700 read_next_command();
2701 parse_mark();
2702 if (skip_prefix(command_buf.buf, "author ", &v)) {
2703 author = parse_ident(v);
2704 read_next_command();
2706 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2707 committer = parse_ident(v);
2708 read_next_command();
2710 if (!committer)
2711 die("Expected committer but didn't get one");
2712 parse_data(&msg, 0, NULL);
2713 read_next_command();
2714 parse_from(b);
2715 merge_list = parse_merge(&merge_count);
2717 /* ensure the branch is active/loaded */
2718 if (!b->branch_tree.tree || !max_active_branches) {
2719 unload_one_branch();
2720 load_branch(b);
2723 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2725 /* file_change* */
2726 while (command_buf.len > 0) {
2727 if (skip_prefix(command_buf.buf, "M ", &v))
2728 file_change_m(v, b);
2729 else if (skip_prefix(command_buf.buf, "D ", &v))
2730 file_change_d(v, b);
2731 else if (skip_prefix(command_buf.buf, "R ", &v))
2732 file_change_cr(v, b, 1);
2733 else if (skip_prefix(command_buf.buf, "C ", &v))
2734 file_change_cr(v, b, 0);
2735 else if (skip_prefix(command_buf.buf, "N ", &v))
2736 note_change_n(v, b, &prev_fanout);
2737 else if (!strcmp("deleteall", command_buf.buf))
2738 file_change_deleteall(b);
2739 else if (skip_prefix(command_buf.buf, "ls ", &v))
2740 parse_ls(v, b);
2741 else {
2742 unread_command_buf = 1;
2743 break;
2745 if (read_next_command() == EOF)
2746 break;
2749 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2750 if (new_fanout != prev_fanout)
2751 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2753 /* build the tree and the commit */
2754 store_tree(&b->branch_tree);
2755 hashcpy(b->branch_tree.versions[0].sha1,
2756 b->branch_tree.versions[1].sha1);
2758 strbuf_reset(&new_data);
2759 strbuf_addf(&new_data, "tree %s\n",
2760 sha1_to_hex(b->branch_tree.versions[1].sha1));
2761 if (!is_null_sha1(b->sha1))
2762 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2763 while (merge_list) {
2764 struct hash_list *next = merge_list->next;
2765 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2766 free(merge_list);
2767 merge_list = next;
2769 strbuf_addf(&new_data,
2770 "author %s\n"
2771 "committer %s\n"
2772 "\n",
2773 author ? author : committer, committer);
2774 strbuf_addbuf(&new_data, &msg);
2775 free(author);
2776 free(committer);
2778 if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2779 b->pack_id = pack_id;
2780 b->last_commit = object_count_by_type[OBJ_COMMIT];
2783 static void parse_new_tag(const char *arg)
2785 static struct strbuf msg = STRBUF_INIT;
2786 const char *from;
2787 char *tagger;
2788 struct branch *s;
2789 struct tag *t;
2790 uintmax_t from_mark = 0;
2791 unsigned char sha1[20];
2792 enum object_type type;
2793 const char *v;
2795 t = pool_alloc(sizeof(struct tag));
2796 memset(t, 0, sizeof(struct tag));
2797 t->name = pool_strdup(arg);
2798 if (last_tag)
2799 last_tag->next_tag = t;
2800 else
2801 first_tag = t;
2802 last_tag = t;
2803 read_next_command();
2805 /* from ... */
2806 if (!skip_prefix(command_buf.buf, "from ", &from))
2807 die("Expected from command, got %s", command_buf.buf);
2808 s = lookup_branch(from);
2809 if (s) {
2810 if (is_null_sha1(s->sha1))
2811 die("Can't tag an empty branch.");
2812 hashcpy(sha1, s->sha1);
2813 type = OBJ_COMMIT;
2814 } else if (*from == ':') {
2815 struct object_entry *oe;
2816 from_mark = parse_mark_ref_eol(from);
2817 oe = find_mark(from_mark);
2818 type = oe->type;
2819 hashcpy(sha1, oe->idx.sha1);
2820 } else if (!get_sha1(from, sha1)) {
2821 struct object_entry *oe = find_object(sha1);
2822 if (!oe) {
2823 type = sha1_object_info(sha1, NULL);
2824 if (type < 0)
2825 die("Not a valid object: %s", from);
2826 } else
2827 type = oe->type;
2828 } else
2829 die("Invalid ref name or SHA1 expression: %s", from);
2830 read_next_command();
2832 /* tagger ... */
2833 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2834 tagger = parse_ident(v);
2835 read_next_command();
2836 } else
2837 tagger = NULL;
2839 /* tag payload/message */
2840 parse_data(&msg, 0, NULL);
2842 /* build the tag object */
2843 strbuf_reset(&new_data);
2845 strbuf_addf(&new_data,
2846 "object %s\n"
2847 "type %s\n"
2848 "tag %s\n",
2849 sha1_to_hex(sha1), typename(type), t->name);
2850 if (tagger)
2851 strbuf_addf(&new_data,
2852 "tagger %s\n", tagger);
2853 strbuf_addch(&new_data, '\n');
2854 strbuf_addbuf(&new_data, &msg);
2855 free(tagger);
2857 if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2858 t->pack_id = MAX_PACK_ID;
2859 else
2860 t->pack_id = pack_id;
2863 static void parse_reset_branch(const char *arg)
2865 struct branch *b;
2867 b = lookup_branch(arg);
2868 if (b) {
2869 hashclr(b->sha1);
2870 hashclr(b->branch_tree.versions[0].sha1);
2871 hashclr(b->branch_tree.versions[1].sha1);
2872 if (b->branch_tree.tree) {
2873 release_tree_content_recursive(b->branch_tree.tree);
2874 b->branch_tree.tree = NULL;
2877 else
2878 b = new_branch(arg);
2879 read_next_command();
2880 parse_from(b);
2881 if (command_buf.len > 0)
2882 unread_command_buf = 1;
2885 static void cat_blob_write(const char *buf, unsigned long size)
2887 if (write_in_full(cat_blob_fd, buf, size) != size)
2888 die_errno("Write to frontend failed");
2891 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2893 struct strbuf line = STRBUF_INIT;
2894 unsigned long size;
2895 enum object_type type = 0;
2896 char *buf;
2898 if (!oe || oe->pack_id == MAX_PACK_ID) {
2899 buf = read_sha1_file(sha1, &type, &size);
2900 } else {
2901 type = oe->type;
2902 buf = gfi_unpack_entry(oe, &size);
2906 * Output based on batch_one_object() from cat-file.c.
2908 if (type <= 0) {
2909 strbuf_reset(&line);
2910 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2911 cat_blob_write(line.buf, line.len);
2912 strbuf_release(&line);
2913 free(buf);
2914 return;
2916 if (!buf)
2917 die("Can't read object %s", sha1_to_hex(sha1));
2918 if (type != OBJ_BLOB)
2919 die("Object %s is a %s but a blob was expected.",
2920 sha1_to_hex(sha1), typename(type));
2921 strbuf_reset(&line);
2922 strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2923 typename(type), size);
2924 cat_blob_write(line.buf, line.len);
2925 strbuf_release(&line);
2926 cat_blob_write(buf, size);
2927 cat_blob_write("\n", 1);
2928 if (oe && oe->pack_id == pack_id) {
2929 last_blob.offset = oe->idx.offset;
2930 strbuf_attach(&last_blob.data, buf, size, size);
2931 last_blob.depth = oe->depth;
2932 } else
2933 free(buf);
2936 static void parse_get_mark(const char *p)
2938 struct object_entry *oe = oe;
2939 char output[42];
2941 /* get-mark SP <object> LF */
2942 if (*p != ':')
2943 die("Not a mark: %s", p);
2945 oe = find_mark(parse_mark_ref_eol(p));
2946 if (!oe)
2947 die("Unknown mark: %s", command_buf.buf);
2949 snprintf(output, sizeof(output), "%s\n", sha1_to_hex(oe->idx.sha1));
2950 cat_blob_write(output, 41);
2953 static void parse_cat_blob(const char *p)
2955 struct object_entry *oe = oe;
2956 unsigned char sha1[20];
2958 /* cat-blob SP <object> LF */
2959 if (*p == ':') {
2960 oe = find_mark(parse_mark_ref_eol(p));
2961 if (!oe)
2962 die("Unknown mark: %s", command_buf.buf);
2963 hashcpy(sha1, oe->idx.sha1);
2964 } else {
2965 if (get_sha1_hex(p, sha1))
2966 die("Invalid dataref: %s", command_buf.buf);
2967 if (p[40])
2968 die("Garbage after SHA1: %s", command_buf.buf);
2969 oe = find_object(sha1);
2972 cat_blob(oe, sha1);
2975 static struct object_entry *dereference(struct object_entry *oe,
2976 unsigned char sha1[20])
2978 unsigned long size;
2979 char *buf = NULL;
2980 if (!oe) {
2981 enum object_type type = sha1_object_info(sha1, NULL);
2982 if (type < 0)
2983 die("object not found: %s", sha1_to_hex(sha1));
2984 /* cache it! */
2985 oe = insert_object(sha1);
2986 oe->type = type;
2987 oe->pack_id = MAX_PACK_ID;
2988 oe->idx.offset = 1;
2990 switch (oe->type) {
2991 case OBJ_TREE: /* easy case. */
2992 return oe;
2993 case OBJ_COMMIT:
2994 case OBJ_TAG:
2995 break;
2996 default:
2997 die("Not a tree-ish: %s", command_buf.buf);
3000 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3001 buf = gfi_unpack_entry(oe, &size);
3002 } else {
3003 enum object_type unused;
3004 buf = read_sha1_file(sha1, &unused, &size);
3006 if (!buf)
3007 die("Can't load object %s", sha1_to_hex(sha1));
3009 /* Peel one layer. */
3010 switch (oe->type) {
3011 case OBJ_TAG:
3012 if (size < 40 + strlen("object ") ||
3013 get_sha1_hex(buf + strlen("object "), sha1))
3014 die("Invalid SHA1 in tag: %s", command_buf.buf);
3015 break;
3016 case OBJ_COMMIT:
3017 if (size < 40 + strlen("tree ") ||
3018 get_sha1_hex(buf + strlen("tree "), sha1))
3019 die("Invalid SHA1 in commit: %s", command_buf.buf);
3022 free(buf);
3023 return find_object(sha1);
3026 static struct object_entry *parse_treeish_dataref(const char **p)
3028 unsigned char sha1[20];
3029 struct object_entry *e;
3031 if (**p == ':') { /* <mark> */
3032 e = find_mark(parse_mark_ref_space(p));
3033 if (!e)
3034 die("Unknown mark: %s", command_buf.buf);
3035 hashcpy(sha1, e->idx.sha1);
3036 } else { /* <sha1> */
3037 if (get_sha1_hex(*p, sha1))
3038 die("Invalid dataref: %s", command_buf.buf);
3039 e = find_object(sha1);
3040 *p += 40;
3041 if (*(*p)++ != ' ')
3042 die("Missing space after tree-ish: %s", command_buf.buf);
3045 while (!e || e->type != OBJ_TREE)
3046 e = dereference(e, sha1);
3047 return e;
3050 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3052 static struct strbuf line = STRBUF_INIT;
3054 /* See show_tree(). */
3055 const char *type =
3056 S_ISGITLINK(mode) ? commit_type :
3057 S_ISDIR(mode) ? tree_type :
3058 blob_type;
3060 if (!mode) {
3061 /* missing SP path LF */
3062 strbuf_reset(&line);
3063 strbuf_addstr(&line, "missing ");
3064 quote_c_style(path, &line, NULL, 0);
3065 strbuf_addch(&line, '\n');
3066 } else {
3067 /* mode SP type SP object_name TAB path LF */
3068 strbuf_reset(&line);
3069 strbuf_addf(&line, "%06o %s %s\t",
3070 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3071 quote_c_style(path, &line, NULL, 0);
3072 strbuf_addch(&line, '\n');
3074 cat_blob_write(line.buf, line.len);
3077 static void parse_ls(const char *p, struct branch *b)
3079 struct tree_entry *root = NULL;
3080 struct tree_entry leaf = {NULL};
3082 /* ls SP (<tree-ish> SP)? <path> */
3083 if (*p == '"') {
3084 if (!b)
3085 die("Not in a commit: %s", command_buf.buf);
3086 root = &b->branch_tree;
3087 } else {
3088 struct object_entry *e = parse_treeish_dataref(&p);
3089 root = new_tree_entry();
3090 hashcpy(root->versions[1].sha1, e->idx.sha1);
3091 if (!is_null_sha1(root->versions[1].sha1))
3092 root->versions[1].mode = S_IFDIR;
3093 load_tree(root);
3095 if (*p == '"') {
3096 static struct strbuf uq = STRBUF_INIT;
3097 const char *endp;
3098 strbuf_reset(&uq);
3099 if (unquote_c_style(&uq, p, &endp))
3100 die("Invalid path: %s", command_buf.buf);
3101 if (*endp)
3102 die("Garbage after path in: %s", command_buf.buf);
3103 p = uq.buf;
3105 tree_content_get(root, p, &leaf, 1);
3107 * A directory in preparation would have a sha1 of zero
3108 * until it is saved. Save, for simplicity.
3110 if (S_ISDIR(leaf.versions[1].mode))
3111 store_tree(&leaf);
3113 print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3114 if (leaf.tree)
3115 release_tree_content_recursive(leaf.tree);
3116 if (!b || root != &b->branch_tree)
3117 release_tree_entry(root);
3120 static void checkpoint(void)
3122 checkpoint_requested = 0;
3123 if (object_count) {
3124 cycle_packfile();
3125 dump_branches();
3126 dump_tags();
3127 dump_marks();
3131 static void parse_checkpoint(void)
3133 checkpoint_requested = 1;
3134 skip_optional_lf();
3137 static void parse_progress(void)
3139 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3140 fputc('\n', stdout);
3141 fflush(stdout);
3142 skip_optional_lf();
3145 static char* make_fast_import_path(const char *path)
3147 if (!relative_marks_paths || is_absolute_path(path))
3148 return xstrdup(path);
3149 return xstrdup(git_path("info/fast-import/%s", path));
3152 static void option_import_marks(const char *marks,
3153 int from_stream, int ignore_missing)
3155 if (import_marks_file) {
3156 if (from_stream)
3157 die("Only one import-marks command allowed per stream");
3159 /* read previous mark file */
3160 if(!import_marks_file_from_stream)
3161 read_marks();
3164 import_marks_file = make_fast_import_path(marks);
3165 safe_create_leading_directories_const(import_marks_file);
3166 import_marks_file_from_stream = from_stream;
3167 import_marks_file_ignore_missing = ignore_missing;
3170 static void option_date_format(const char *fmt)
3172 if (!strcmp(fmt, "raw"))
3173 whenspec = WHENSPEC_RAW;
3174 else if (!strcmp(fmt, "rfc2822"))
3175 whenspec = WHENSPEC_RFC2822;
3176 else if (!strcmp(fmt, "now"))
3177 whenspec = WHENSPEC_NOW;
3178 else
3179 die("unknown --date-format argument %s", fmt);
3182 static unsigned long ulong_arg(const char *option, const char *arg)
3184 char *endptr;
3185 unsigned long rv = strtoul(arg, &endptr, 0);
3186 if (strchr(arg, '-') || endptr == arg || *endptr)
3187 die("%s: argument must be a non-negative integer", option);
3188 return rv;
3191 static void option_depth(const char *depth)
3193 max_depth = ulong_arg("--depth", depth);
3194 if (max_depth > MAX_DEPTH)
3195 die("--depth cannot exceed %u", MAX_DEPTH);
3198 static void option_active_branches(const char *branches)
3200 max_active_branches = ulong_arg("--active-branches", branches);
3203 static void option_export_marks(const char *marks)
3205 export_marks_file = make_fast_import_path(marks);
3206 safe_create_leading_directories_const(export_marks_file);
3209 static void option_cat_blob_fd(const char *fd)
3211 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3212 if (n > (unsigned long) INT_MAX)
3213 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3214 cat_blob_fd = (int) n;
3217 static void option_export_pack_edges(const char *edges)
3219 if (pack_edges)
3220 fclose(pack_edges);
3221 pack_edges = fopen(edges, "a");
3222 if (!pack_edges)
3223 die_errno("Cannot open '%s'", edges);
3226 static int parse_one_option(const char *option)
3228 if (skip_prefix(option, "max-pack-size=", &option)) {
3229 unsigned long v;
3230 if (!git_parse_ulong(option, &v))
3231 return 0;
3232 if (v < 8192) {
3233 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3234 v *= 1024 * 1024;
3235 } else if (v < 1024 * 1024) {
3236 warning("minimum max-pack-size is 1 MiB");
3237 v = 1024 * 1024;
3239 max_packsize = v;
3240 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3241 unsigned long v;
3242 if (!git_parse_ulong(option, &v))
3243 return 0;
3244 big_file_threshold = v;
3245 } else if (skip_prefix(option, "depth=", &option)) {
3246 option_depth(option);
3247 } else if (skip_prefix(option, "active-branches=", &option)) {
3248 option_active_branches(option);
3249 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3250 option_export_pack_edges(option);
3251 } else if (starts_with(option, "quiet")) {
3252 show_stats = 0;
3253 } else if (starts_with(option, "stats")) {
3254 show_stats = 1;
3255 } else {
3256 return 0;
3259 return 1;
3262 static int parse_one_feature(const char *feature, int from_stream)
3264 const char *arg;
3266 if (skip_prefix(feature, "date-format=", &arg)) {
3267 option_date_format(arg);
3268 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3269 option_import_marks(arg, from_stream, 0);
3270 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3271 option_import_marks(arg, from_stream, 1);
3272 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3273 option_export_marks(arg);
3274 } else if (!strcmp(feature, "get-mark")) {
3275 ; /* Don't die - this feature is supported */
3276 } else if (!strcmp(feature, "cat-blob")) {
3277 ; /* Don't die - this feature is supported */
3278 } else if (!strcmp(feature, "relative-marks")) {
3279 relative_marks_paths = 1;
3280 } else if (!strcmp(feature, "no-relative-marks")) {
3281 relative_marks_paths = 0;
3282 } else if (!strcmp(feature, "done")) {
3283 require_explicit_termination = 1;
3284 } else if (!strcmp(feature, "force")) {
3285 force_update = 1;
3286 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3287 ; /* do nothing; we have the feature */
3288 } else {
3289 return 0;
3292 return 1;
3295 static void parse_feature(const char *feature)
3297 if (seen_data_command)
3298 die("Got feature command '%s' after data command", feature);
3300 if (parse_one_feature(feature, 1))
3301 return;
3303 die("This version of fast-import does not support feature %s.", feature);
3306 static void parse_option(const char *option)
3308 if (seen_data_command)
3309 die("Got option command '%s' after data command", option);
3311 if (parse_one_option(option))
3312 return;
3314 die("This version of fast-import does not support option: %s", option);
3317 static void git_pack_config(void)
3319 int indexversion_value;
3320 unsigned long packsizelimit_value;
3322 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3323 if (max_depth > MAX_DEPTH)
3324 max_depth = MAX_DEPTH;
3326 if (!git_config_get_int("pack.compression", &pack_compression_level)) {
3327 if (pack_compression_level == -1)
3328 pack_compression_level = Z_DEFAULT_COMPRESSION;
3329 else if (pack_compression_level < 0 ||
3330 pack_compression_level > Z_BEST_COMPRESSION)
3331 git_die_config("pack.compression",
3332 "bad pack compression level %d", pack_compression_level);
3333 pack_compression_seen = 1;
3335 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3336 pack_idx_opts.version = indexversion_value;
3337 if (pack_idx_opts.version > 2)
3338 git_die_config("pack.indexversion",
3339 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3341 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3342 max_packsize = packsizelimit_value;
3344 git_config(git_default_config, NULL);
3347 static const char fast_import_usage[] =
3348 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3350 static void parse_argv(void)
3352 unsigned int i;
3354 for (i = 1; i < global_argc; i++) {
3355 const char *a = global_argv[i];
3357 if (*a != '-' || !strcmp(a, "--"))
3358 break;
3360 if (!skip_prefix(a, "--", &a))
3361 die("unknown option %s", a);
3363 if (parse_one_option(a))
3364 continue;
3366 if (parse_one_feature(a, 0))
3367 continue;
3369 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3370 option_cat_blob_fd(a);
3371 continue;
3374 die("unknown option --%s", a);
3376 if (i != global_argc)
3377 usage(fast_import_usage);
3379 seen_data_command = 1;
3380 if (import_marks_file)
3381 read_marks();
3384 int main(int argc, char **argv)
3386 unsigned int i;
3388 git_extract_argv0_path(argv[0]);
3390 git_setup_gettext();
3392 if (argc == 2 && !strcmp(argv[1], "-h"))
3393 usage(fast_import_usage);
3395 setup_git_directory();
3396 reset_pack_idx_option(&pack_idx_opts);
3397 git_pack_config();
3398 if (!pack_compression_seen && core_compression_seen)
3399 pack_compression_level = core_compression_level;
3401 alloc_objects(object_entry_alloc);
3402 strbuf_init(&command_buf, 0);
3403 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3404 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3405 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3406 marks = pool_calloc(1, sizeof(struct mark_set));
3408 global_argc = argc;
3409 global_argv = argv;
3411 rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3412 for (i = 0; i < (cmd_save - 1); i++)
3413 rc_free[i].next = &rc_free[i + 1];
3414 rc_free[cmd_save - 1].next = NULL;
3416 prepare_packed_git();
3417 start_packfile();
3418 set_die_routine(die_nicely);
3419 set_checkpoint_signal();
3420 while (read_next_command() != EOF) {
3421 const char *v;
3422 if (!strcmp("blob", command_buf.buf))
3423 parse_new_blob();
3424 else if (skip_prefix(command_buf.buf, "ls ", &v))
3425 parse_ls(v, NULL);
3426 else if (skip_prefix(command_buf.buf, "commit ", &v))
3427 parse_new_commit(v);
3428 else if (skip_prefix(command_buf.buf, "tag ", &v))
3429 parse_new_tag(v);
3430 else if (skip_prefix(command_buf.buf, "reset ", &v))
3431 parse_reset_branch(v);
3432 else if (!strcmp("checkpoint", command_buf.buf))
3433 parse_checkpoint();
3434 else if (!strcmp("done", command_buf.buf))
3435 break;
3436 else if (starts_with(command_buf.buf, "progress "))
3437 parse_progress();
3438 else if (skip_prefix(command_buf.buf, "feature ", &v))
3439 parse_feature(v);
3440 else if (skip_prefix(command_buf.buf, "option git ", &v))
3441 parse_option(v);
3442 else if (starts_with(command_buf.buf, "option "))
3443 /* ignore non-git options*/;
3444 else
3445 die("Unsupported command: %s", command_buf.buf);
3447 if (checkpoint_requested)
3448 checkpoint();
3451 /* argv hasn't been parsed yet, do so */
3452 if (!seen_data_command)
3453 parse_argv();
3455 if (require_explicit_termination && feof(stdin))
3456 die("stream ends early");
3458 end_packfile();
3460 dump_branches();
3461 dump_tags();
3462 unkeep_all_packs();
3463 dump_marks();
3465 if (pack_edges)
3466 fclose(pack_edges);
3468 if (show_stats) {
3469 uintmax_t total_count = 0, duplicate_count = 0;
3470 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3471 total_count += object_count_by_type[i];
3472 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3473 duplicate_count += duplicate_count_by_type[i];
3475 fprintf(stderr, "%s statistics:\n", argv[0]);
3476 fprintf(stderr, "---------------------------------------------------------------------\n");
3477 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3478 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3479 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]);
3480 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]);
3481 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]);
3482 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]);
3483 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3484 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3485 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3486 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3487 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)(total_allocd/1024));
3488 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3489 fprintf(stderr, "---------------------------------------------------------------------\n");
3490 pack_report();
3491 fprintf(stderr, "---------------------------------------------------------------------\n");
3492 fprintf(stderr, "\n");
3495 return failure ? 1 : 0;