Merge branch 'rs/daemon-plug-child-leak' into maint
[git/debian.git] / fast-import.c
blob6c7c3c9b669eb272f4da530aef459c9c39c4d294
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(LOCAL)));
428 fputc('\n', rpt);
430 fputs("fatal: ", rpt);
431 fputs(err, rpt);
432 fputc('\n', rpt);
434 fputc('\n', rpt);
435 fputs("Most Recent Commands Before Crash\n", rpt);
436 fputs("---------------------------------\n", rpt);
437 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
438 if (rc->next == &cmd_hist)
439 fputs("* ", rpt);
440 else
441 fputs(" ", rpt);
442 fputs(rc->buf, rpt);
443 fputc('\n', rpt);
446 fputc('\n', rpt);
447 fputs("Active Branch LRU\n", rpt);
448 fputs("-----------------\n", rpt);
449 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
450 cur_active_branches,
451 max_active_branches);
452 fputc('\n', rpt);
453 fputs(" pos clock name\n", rpt);
454 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
455 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
456 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
457 ++lu, b->last_commit, b->name);
459 fputc('\n', rpt);
460 fputs("Inactive Branches\n", rpt);
461 fputs("-----------------\n", rpt);
462 for (lu = 0; lu < branch_table_sz; lu++) {
463 for (b = branch_table[lu]; b; b = b->table_next_branch)
464 write_branch_report(rpt, b);
467 if (first_tag) {
468 struct tag *tg;
469 fputc('\n', rpt);
470 fputs("Annotated Tags\n", rpt);
471 fputs("--------------\n", rpt);
472 for (tg = first_tag; tg; tg = tg->next_tag) {
473 fputs(sha1_to_hex(tg->sha1), rpt);
474 fputc(' ', rpt);
475 fputs(tg->name, rpt);
476 fputc('\n', rpt);
480 fputc('\n', rpt);
481 fputs("Marks\n", rpt);
482 fputs("-----\n", rpt);
483 if (export_marks_file)
484 fprintf(rpt, " exported to %s\n", export_marks_file);
485 else
486 dump_marks_helper(rpt, 0, marks);
488 fputc('\n', rpt);
489 fputs("-------------------\n", rpt);
490 fputs("END OF CRASH REPORT\n", rpt);
491 fclose(rpt);
492 free(loc);
495 static void end_packfile(void);
496 static void unkeep_all_packs(void);
497 static void dump_marks(void);
499 static NORETURN void die_nicely(const char *err, va_list params)
501 static int zombie;
502 char message[2 * PATH_MAX];
504 vsnprintf(message, sizeof(message), err, params);
505 fputs("fatal: ", stderr);
506 fputs(message, stderr);
507 fputc('\n', stderr);
509 if (!zombie) {
510 zombie = 1;
511 write_crash_report(message);
512 end_packfile();
513 unkeep_all_packs();
514 dump_marks();
516 exit(128);
519 #ifndef SIGUSR1 /* Windows, for example */
521 static void set_checkpoint_signal(void)
525 #else
527 static void checkpoint_signal(int signo)
529 checkpoint_requested = 1;
532 static void set_checkpoint_signal(void)
534 struct sigaction sa;
536 memset(&sa, 0, sizeof(sa));
537 sa.sa_handler = checkpoint_signal;
538 sigemptyset(&sa.sa_mask);
539 sa.sa_flags = SA_RESTART;
540 sigaction(SIGUSR1, &sa, NULL);
543 #endif
545 static void alloc_objects(unsigned int cnt)
547 struct object_entry_pool *b;
549 b = xmalloc(sizeof(struct object_entry_pool)
550 + cnt * sizeof(struct object_entry));
551 b->next_pool = blocks;
552 b->next_free = b->entries;
553 b->end = b->entries + cnt;
554 blocks = b;
555 alloc_count += cnt;
558 static struct object_entry *new_object(unsigned char *sha1)
560 struct object_entry *e;
562 if (blocks->next_free == blocks->end)
563 alloc_objects(object_entry_alloc);
565 e = blocks->next_free++;
566 hashcpy(e->idx.sha1, sha1);
567 return e;
570 static struct object_entry *find_object(unsigned char *sha1)
572 unsigned int h = sha1[0] << 8 | sha1[1];
573 struct object_entry *e;
574 for (e = object_table[h]; e; e = e->next)
575 if (!hashcmp(sha1, e->idx.sha1))
576 return e;
577 return NULL;
580 static struct object_entry *insert_object(unsigned char *sha1)
582 unsigned int h = sha1[0] << 8 | sha1[1];
583 struct object_entry *e = object_table[h];
585 while (e) {
586 if (!hashcmp(sha1, e->idx.sha1))
587 return e;
588 e = e->next;
591 e = new_object(sha1);
592 e->next = object_table[h];
593 e->idx.offset = 0;
594 object_table[h] = e;
595 return e;
598 static unsigned int hc_str(const char *s, size_t len)
600 unsigned int r = 0;
601 while (len-- > 0)
602 r = r * 31 + *s++;
603 return r;
606 static void *pool_alloc(size_t len)
608 struct mem_pool *p;
609 void *r;
611 /* round up to a 'uintmax_t' alignment */
612 if (len & (sizeof(uintmax_t) - 1))
613 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
615 for (p = mem_pool; p; p = p->next_pool)
616 if ((p->end - p->next_free >= len))
617 break;
619 if (!p) {
620 if (len >= (mem_pool_alloc/2)) {
621 total_allocd += len;
622 return xmalloc(len);
624 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
625 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
626 p->next_pool = mem_pool;
627 p->next_free = (char *) p->space;
628 p->end = p->next_free + mem_pool_alloc;
629 mem_pool = p;
632 r = p->next_free;
633 p->next_free += len;
634 return r;
637 static void *pool_calloc(size_t count, size_t size)
639 size_t len = count * size;
640 void *r = pool_alloc(len);
641 memset(r, 0, len);
642 return r;
645 static char *pool_strdup(const char *s)
647 char *r = pool_alloc(strlen(s) + 1);
648 strcpy(r, s);
649 return r;
652 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
654 struct mark_set *s = marks;
655 while ((idnum >> s->shift) >= 1024) {
656 s = pool_calloc(1, sizeof(struct mark_set));
657 s->shift = marks->shift + 10;
658 s->data.sets[0] = marks;
659 marks = s;
661 while (s->shift) {
662 uintmax_t i = idnum >> s->shift;
663 idnum -= i << s->shift;
664 if (!s->data.sets[i]) {
665 s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
666 s->data.sets[i]->shift = s->shift - 10;
668 s = s->data.sets[i];
670 if (!s->data.marked[idnum])
671 marks_set_count++;
672 s->data.marked[idnum] = oe;
675 static struct object_entry *find_mark(uintmax_t idnum)
677 uintmax_t orig_idnum = idnum;
678 struct mark_set *s = marks;
679 struct object_entry *oe = NULL;
680 if ((idnum >> s->shift) < 1024) {
681 while (s && s->shift) {
682 uintmax_t i = idnum >> s->shift;
683 idnum -= i << s->shift;
684 s = s->data.sets[i];
686 if (s)
687 oe = s->data.marked[idnum];
689 if (!oe)
690 die("mark :%" PRIuMAX " not declared", orig_idnum);
691 return oe;
694 static struct atom_str *to_atom(const char *s, unsigned short len)
696 unsigned int hc = hc_str(s, len) % atom_table_sz;
697 struct atom_str *c;
699 for (c = atom_table[hc]; c; c = c->next_atom)
700 if (c->str_len == len && !strncmp(s, c->str_dat, len))
701 return c;
703 c = pool_alloc(sizeof(struct atom_str) + len + 1);
704 c->str_len = len;
705 strncpy(c->str_dat, s, len);
706 c->str_dat[len] = 0;
707 c->next_atom = atom_table[hc];
708 atom_table[hc] = c;
709 atom_cnt++;
710 return c;
713 static struct branch *lookup_branch(const char *name)
715 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
716 struct branch *b;
718 for (b = branch_table[hc]; b; b = b->table_next_branch)
719 if (!strcmp(name, b->name))
720 return b;
721 return NULL;
724 static struct branch *new_branch(const char *name)
726 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
727 struct branch *b = lookup_branch(name);
729 if (b)
730 die("Invalid attempt to create duplicate branch: %s", name);
731 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
732 die("Branch name doesn't conform to GIT standards: %s", name);
734 b = pool_calloc(1, sizeof(struct branch));
735 b->name = pool_strdup(name);
736 b->table_next_branch = branch_table[hc];
737 b->branch_tree.versions[0].mode = S_IFDIR;
738 b->branch_tree.versions[1].mode = S_IFDIR;
739 b->num_notes = 0;
740 b->active = 0;
741 b->pack_id = MAX_PACK_ID;
742 branch_table[hc] = b;
743 branch_count++;
744 return b;
747 static unsigned int hc_entries(unsigned int cnt)
749 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
750 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
753 static struct tree_content *new_tree_content(unsigned int cnt)
755 struct avail_tree_content *f, *l = NULL;
756 struct tree_content *t;
757 unsigned int hc = hc_entries(cnt);
759 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
760 if (f->entry_capacity >= cnt)
761 break;
763 if (f) {
764 if (l)
765 l->next_avail = f->next_avail;
766 else
767 avail_tree_table[hc] = f->next_avail;
768 } else {
769 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
770 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
771 f->entry_capacity = cnt;
774 t = (struct tree_content*)f;
775 t->entry_count = 0;
776 t->delta_depth = 0;
777 return t;
780 static void release_tree_entry(struct tree_entry *e);
781 static void release_tree_content(struct tree_content *t)
783 struct avail_tree_content *f = (struct avail_tree_content*)t;
784 unsigned int hc = hc_entries(f->entry_capacity);
785 f->next_avail = avail_tree_table[hc];
786 avail_tree_table[hc] = f;
789 static void release_tree_content_recursive(struct tree_content *t)
791 unsigned int i;
792 for (i = 0; i < t->entry_count; i++)
793 release_tree_entry(t->entries[i]);
794 release_tree_content(t);
797 static struct tree_content *grow_tree_content(
798 struct tree_content *t,
799 int amt)
801 struct tree_content *r = new_tree_content(t->entry_count + amt);
802 r->entry_count = t->entry_count;
803 r->delta_depth = t->delta_depth;
804 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
805 release_tree_content(t);
806 return r;
809 static struct tree_entry *new_tree_entry(void)
811 struct tree_entry *e;
813 if (!avail_tree_entry) {
814 unsigned int n = tree_entry_alloc;
815 total_allocd += n * sizeof(struct tree_entry);
816 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
817 while (n-- > 1) {
818 *((void**)e) = e + 1;
819 e++;
821 *((void**)e) = NULL;
824 e = avail_tree_entry;
825 avail_tree_entry = *((void**)e);
826 return e;
829 static void release_tree_entry(struct tree_entry *e)
831 if (e->tree)
832 release_tree_content_recursive(e->tree);
833 *((void**)e) = avail_tree_entry;
834 avail_tree_entry = e;
837 static struct tree_content *dup_tree_content(struct tree_content *s)
839 struct tree_content *d;
840 struct tree_entry *a, *b;
841 unsigned int i;
843 if (!s)
844 return NULL;
845 d = new_tree_content(s->entry_count);
846 for (i = 0; i < s->entry_count; i++) {
847 a = s->entries[i];
848 b = new_tree_entry();
849 memcpy(b, a, sizeof(*a));
850 if (a->tree && is_null_sha1(b->versions[1].sha1))
851 b->tree = dup_tree_content(a->tree);
852 else
853 b->tree = NULL;
854 d->entries[i] = b;
856 d->entry_count = s->entry_count;
857 d->delta_depth = s->delta_depth;
859 return d;
862 static void start_packfile(void)
864 static char tmp_file[PATH_MAX];
865 struct packed_git *p;
866 struct pack_header hdr;
867 int pack_fd;
869 pack_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
870 "pack/tmp_pack_XXXXXX");
871 p = xcalloc(1, sizeof(*p) + strlen(tmp_file) + 2);
872 strcpy(p->pack_name, tmp_file);
873 p->pack_fd = pack_fd;
874 p->do_not_close = 1;
875 pack_file = sha1fd(pack_fd, p->pack_name);
877 hdr.hdr_signature = htonl(PACK_SIGNATURE);
878 hdr.hdr_version = htonl(2);
879 hdr.hdr_entries = 0;
880 sha1write(pack_file, &hdr, sizeof(hdr));
882 pack_data = p;
883 pack_size = sizeof(hdr);
884 object_count = 0;
886 REALLOC_ARRAY(all_packs, pack_id + 1);
887 all_packs[pack_id] = p;
890 static const char *create_index(void)
892 const char *tmpfile;
893 struct pack_idx_entry **idx, **c, **last;
894 struct object_entry *e;
895 struct object_entry_pool *o;
897 /* Build the table of object IDs. */
898 idx = xmalloc(object_count * sizeof(*idx));
899 c = idx;
900 for (o = blocks; o; o = o->next_pool)
901 for (e = o->next_free; e-- != o->entries;)
902 if (pack_id == e->pack_id)
903 *c++ = &e->idx;
904 last = idx + object_count;
905 if (c != last)
906 die("internal consistency error creating the index");
908 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
909 free(idx);
910 return tmpfile;
913 static char *keep_pack(const char *curr_index_name)
915 static char name[PATH_MAX];
916 static const char *keep_msg = "fast-import";
917 int keep_fd;
919 keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
920 if (keep_fd < 0)
921 die_errno("cannot create keep file");
922 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
923 if (close(keep_fd))
924 die_errno("failed to write keep file");
926 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
927 get_object_directory(), sha1_to_hex(pack_data->sha1));
928 if (finalize_object_file(pack_data->pack_name, name))
929 die("cannot store pack file");
931 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
932 get_object_directory(), sha1_to_hex(pack_data->sha1));
933 if (finalize_object_file(curr_index_name, name))
934 die("cannot store index file");
935 free((void *)curr_index_name);
936 return name;
939 static void unkeep_all_packs(void)
941 static char name[PATH_MAX];
942 int k;
944 for (k = 0; k < pack_id; k++) {
945 struct packed_git *p = all_packs[k];
946 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
947 get_object_directory(), sha1_to_hex(p->sha1));
948 unlink_or_warn(name);
952 static void end_packfile(void)
954 static int running;
956 if (running || !pack_data)
957 return;
959 running = 1;
960 clear_delta_base_cache();
961 if (object_count) {
962 struct packed_git *new_p;
963 unsigned char cur_pack_sha1[20];
964 char *idx_name;
965 int i;
966 struct branch *b;
967 struct tag *t;
969 close_pack_windows(pack_data);
970 sha1close(pack_file, cur_pack_sha1, 0);
971 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
972 pack_data->pack_name, object_count,
973 cur_pack_sha1, pack_size);
974 close(pack_data->pack_fd);
975 idx_name = keep_pack(create_index());
977 /* Register the packfile with core git's machinery. */
978 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
979 if (!new_p)
980 die("core git rejected index %s", idx_name);
981 all_packs[pack_id] = new_p;
982 install_packed_git(new_p);
984 /* Print the boundary */
985 if (pack_edges) {
986 fprintf(pack_edges, "%s:", new_p->pack_name);
987 for (i = 0; i < branch_table_sz; i++) {
988 for (b = branch_table[i]; b; b = b->table_next_branch) {
989 if (b->pack_id == pack_id)
990 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
993 for (t = first_tag; t; t = t->next_tag) {
994 if (t->pack_id == pack_id)
995 fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
997 fputc('\n', pack_edges);
998 fflush(pack_edges);
1001 pack_id++;
1003 else {
1004 close(pack_data->pack_fd);
1005 unlink_or_warn(pack_data->pack_name);
1007 free(pack_data);
1008 pack_data = NULL;
1009 running = 0;
1011 /* We can't carry a delta across packfiles. */
1012 strbuf_release(&last_blob.data);
1013 last_blob.offset = 0;
1014 last_blob.depth = 0;
1017 static void cycle_packfile(void)
1019 end_packfile();
1020 start_packfile();
1023 static int store_object(
1024 enum object_type type,
1025 struct strbuf *dat,
1026 struct last_object *last,
1027 unsigned char *sha1out,
1028 uintmax_t mark)
1030 void *out, *delta;
1031 struct object_entry *e;
1032 unsigned char hdr[96];
1033 unsigned char sha1[20];
1034 unsigned long hdrlen, deltalen;
1035 git_SHA_CTX c;
1036 git_zstream s;
1038 hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1039 (unsigned long)dat->len) + 1;
1040 git_SHA1_Init(&c);
1041 git_SHA1_Update(&c, hdr, hdrlen);
1042 git_SHA1_Update(&c, dat->buf, dat->len);
1043 git_SHA1_Final(sha1, &c);
1044 if (sha1out)
1045 hashcpy(sha1out, sha1);
1047 e = insert_object(sha1);
1048 if (mark)
1049 insert_mark(mark, e);
1050 if (e->idx.offset) {
1051 duplicate_count_by_type[type]++;
1052 return 1;
1053 } else if (find_sha1_pack(sha1, packed_git)) {
1054 e->type = type;
1055 e->pack_id = MAX_PACK_ID;
1056 e->idx.offset = 1; /* just not zero! */
1057 duplicate_count_by_type[type]++;
1058 return 1;
1061 if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1062 delta_count_attempts_by_type[type]++;
1063 delta = diff_delta(last->data.buf, last->data.len,
1064 dat->buf, dat->len,
1065 &deltalen, dat->len - 20);
1066 } else
1067 delta = NULL;
1069 git_deflate_init(&s, pack_compression_level);
1070 if (delta) {
1071 s.next_in = delta;
1072 s.avail_in = deltalen;
1073 } else {
1074 s.next_in = (void *)dat->buf;
1075 s.avail_in = dat->len;
1077 s.avail_out = git_deflate_bound(&s, s.avail_in);
1078 s.next_out = out = xmalloc(s.avail_out);
1079 while (git_deflate(&s, Z_FINISH) == Z_OK)
1080 ; /* nothing */
1081 git_deflate_end(&s);
1083 /* Determine if we should auto-checkpoint. */
1084 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1085 || (pack_size + 60 + s.total_out) < pack_size) {
1087 /* This new object needs to *not* have the current pack_id. */
1088 e->pack_id = pack_id + 1;
1089 cycle_packfile();
1091 /* We cannot carry a delta into the new pack. */
1092 if (delta) {
1093 free(delta);
1094 delta = NULL;
1096 git_deflate_init(&s, pack_compression_level);
1097 s.next_in = (void *)dat->buf;
1098 s.avail_in = dat->len;
1099 s.avail_out = git_deflate_bound(&s, s.avail_in);
1100 s.next_out = out = xrealloc(out, s.avail_out);
1101 while (git_deflate(&s, Z_FINISH) == Z_OK)
1102 ; /* nothing */
1103 git_deflate_end(&s);
1107 e->type = type;
1108 e->pack_id = pack_id;
1109 e->idx.offset = pack_size;
1110 object_count++;
1111 object_count_by_type[type]++;
1113 crc32_begin(pack_file);
1115 if (delta) {
1116 off_t ofs = e->idx.offset - last->offset;
1117 unsigned pos = sizeof(hdr) - 1;
1119 delta_count_by_type[type]++;
1120 e->depth = last->depth + 1;
1122 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1123 sha1write(pack_file, hdr, hdrlen);
1124 pack_size += hdrlen;
1126 hdr[pos] = ofs & 127;
1127 while (ofs >>= 7)
1128 hdr[--pos] = 128 | (--ofs & 127);
1129 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1130 pack_size += sizeof(hdr) - pos;
1131 } else {
1132 e->depth = 0;
1133 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1134 sha1write(pack_file, hdr, hdrlen);
1135 pack_size += hdrlen;
1138 sha1write(pack_file, out, s.total_out);
1139 pack_size += s.total_out;
1141 e->idx.crc32 = crc32_end(pack_file);
1143 free(out);
1144 free(delta);
1145 if (last) {
1146 if (last->no_swap) {
1147 last->data = *dat;
1148 } else {
1149 strbuf_swap(&last->data, dat);
1151 last->offset = e->idx.offset;
1152 last->depth = e->depth;
1154 return 0;
1157 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1159 if (sha1file_truncate(pack_file, checkpoint))
1160 die_errno("cannot truncate pack to skip duplicate");
1161 pack_size = checkpoint->offset;
1164 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1166 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1167 unsigned char *in_buf = xmalloc(in_sz);
1168 unsigned char *out_buf = xmalloc(out_sz);
1169 struct object_entry *e;
1170 unsigned char sha1[20];
1171 unsigned long hdrlen;
1172 off_t offset;
1173 git_SHA_CTX c;
1174 git_zstream s;
1175 struct sha1file_checkpoint checkpoint;
1176 int status = Z_OK;
1178 /* Determine if we should auto-checkpoint. */
1179 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1180 || (pack_size + 60 + len) < pack_size)
1181 cycle_packfile();
1183 sha1file_checkpoint(pack_file, &checkpoint);
1184 offset = checkpoint.offset;
1186 hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1187 if (out_sz <= hdrlen)
1188 die("impossibly large object header");
1190 git_SHA1_Init(&c);
1191 git_SHA1_Update(&c, out_buf, hdrlen);
1193 crc32_begin(pack_file);
1195 git_deflate_init(&s, pack_compression_level);
1197 hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1198 if (out_sz <= hdrlen)
1199 die("impossibly large object header");
1201 s.next_out = out_buf + hdrlen;
1202 s.avail_out = out_sz - hdrlen;
1204 while (status != Z_STREAM_END) {
1205 if (0 < len && !s.avail_in) {
1206 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1207 size_t n = fread(in_buf, 1, cnt, stdin);
1208 if (!n && feof(stdin))
1209 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1211 git_SHA1_Update(&c, in_buf, n);
1212 s.next_in = in_buf;
1213 s.avail_in = n;
1214 len -= n;
1217 status = git_deflate(&s, len ? 0 : Z_FINISH);
1219 if (!s.avail_out || status == Z_STREAM_END) {
1220 size_t n = s.next_out - out_buf;
1221 sha1write(pack_file, out_buf, n);
1222 pack_size += n;
1223 s.next_out = out_buf;
1224 s.avail_out = out_sz;
1227 switch (status) {
1228 case Z_OK:
1229 case Z_BUF_ERROR:
1230 case Z_STREAM_END:
1231 continue;
1232 default:
1233 die("unexpected deflate failure: %d", status);
1236 git_deflate_end(&s);
1237 git_SHA1_Final(sha1, &c);
1239 if (sha1out)
1240 hashcpy(sha1out, sha1);
1242 e = insert_object(sha1);
1244 if (mark)
1245 insert_mark(mark, e);
1247 if (e->idx.offset) {
1248 duplicate_count_by_type[OBJ_BLOB]++;
1249 truncate_pack(&checkpoint);
1251 } else if (find_sha1_pack(sha1, packed_git)) {
1252 e->type = OBJ_BLOB;
1253 e->pack_id = MAX_PACK_ID;
1254 e->idx.offset = 1; /* just not zero! */
1255 duplicate_count_by_type[OBJ_BLOB]++;
1256 truncate_pack(&checkpoint);
1258 } else {
1259 e->depth = 0;
1260 e->type = OBJ_BLOB;
1261 e->pack_id = pack_id;
1262 e->idx.offset = offset;
1263 e->idx.crc32 = crc32_end(pack_file);
1264 object_count++;
1265 object_count_by_type[OBJ_BLOB]++;
1268 free(in_buf);
1269 free(out_buf);
1272 /* All calls must be guarded by find_object() or find_mark() to
1273 * ensure the 'struct object_entry' passed was written by this
1274 * process instance. We unpack the entry by the offset, avoiding
1275 * the need for the corresponding .idx file. This unpacking rule
1276 * works because we only use OBJ_REF_DELTA within the packfiles
1277 * created by fast-import.
1279 * oe must not be NULL. Such an oe usually comes from giving
1280 * an unknown SHA-1 to find_object() or an undefined mark to
1281 * find_mark(). Callers must test for this condition and use
1282 * the standard read_sha1_file() when it happens.
1284 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1285 * find_mark(), where the mark was reloaded from an existing marks
1286 * file and is referencing an object that this fast-import process
1287 * instance did not write out to a packfile. Callers must test for
1288 * this condition and use read_sha1_file() instead.
1290 static void *gfi_unpack_entry(
1291 struct object_entry *oe,
1292 unsigned long *sizep)
1294 enum object_type type;
1295 struct packed_git *p = all_packs[oe->pack_id];
1296 if (p == pack_data && p->pack_size < (pack_size + 20)) {
1297 /* The object is stored in the packfile we are writing to
1298 * and we have modified it since the last time we scanned
1299 * back to read a previously written object. If an old
1300 * window covered [p->pack_size, p->pack_size + 20) its
1301 * data is stale and is not valid. Closing all windows
1302 * and updating the packfile length ensures we can read
1303 * the newly written data.
1305 close_pack_windows(p);
1306 sha1flush(pack_file);
1308 /* We have to offer 20 bytes additional on the end of
1309 * the packfile as the core unpacker code assumes the
1310 * footer is present at the file end and must promise
1311 * at least 20 bytes within any window it maps. But
1312 * we don't actually create the footer here.
1314 p->pack_size = pack_size + 20;
1316 return unpack_entry(p, oe->idx.offset, &type, sizep);
1319 static const char *get_mode(const char *str, uint16_t *modep)
1321 unsigned char c;
1322 uint16_t mode = 0;
1324 while ((c = *str++) != ' ') {
1325 if (c < '0' || c > '7')
1326 return NULL;
1327 mode = (mode << 3) + (c - '0');
1329 *modep = mode;
1330 return str;
1333 static void load_tree(struct tree_entry *root)
1335 unsigned char *sha1 = root->versions[1].sha1;
1336 struct object_entry *myoe;
1337 struct tree_content *t;
1338 unsigned long size;
1339 char *buf;
1340 const char *c;
1342 root->tree = t = new_tree_content(8);
1343 if (is_null_sha1(sha1))
1344 return;
1346 myoe = find_object(sha1);
1347 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1348 if (myoe->type != OBJ_TREE)
1349 die("Not a tree: %s", sha1_to_hex(sha1));
1350 t->delta_depth = myoe->depth;
1351 buf = gfi_unpack_entry(myoe, &size);
1352 if (!buf)
1353 die("Can't load tree %s", sha1_to_hex(sha1));
1354 } else {
1355 enum object_type type;
1356 buf = read_sha1_file(sha1, &type, &size);
1357 if (!buf || type != OBJ_TREE)
1358 die("Can't load tree %s", sha1_to_hex(sha1));
1361 c = buf;
1362 while (c != (buf + size)) {
1363 struct tree_entry *e = new_tree_entry();
1365 if (t->entry_count == t->entry_capacity)
1366 root->tree = t = grow_tree_content(t, t->entry_count);
1367 t->entries[t->entry_count++] = e;
1369 e->tree = NULL;
1370 c = get_mode(c, &e->versions[1].mode);
1371 if (!c)
1372 die("Corrupt mode in %s", sha1_to_hex(sha1));
1373 e->versions[0].mode = e->versions[1].mode;
1374 e->name = to_atom(c, strlen(c));
1375 c += e->name->str_len + 1;
1376 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1377 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1378 c += 20;
1380 free(buf);
1383 static int tecmp0 (const void *_a, const void *_b)
1385 struct tree_entry *a = *((struct tree_entry**)_a);
1386 struct tree_entry *b = *((struct tree_entry**)_b);
1387 return base_name_compare(
1388 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1389 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1392 static int tecmp1 (const void *_a, const void *_b)
1394 struct tree_entry *a = *((struct tree_entry**)_a);
1395 struct tree_entry *b = *((struct tree_entry**)_b);
1396 return base_name_compare(
1397 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1398 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1401 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1403 size_t maxlen = 0;
1404 unsigned int i;
1406 if (!v)
1407 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1408 else
1409 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1411 for (i = 0; i < t->entry_count; i++) {
1412 if (t->entries[i]->versions[v].mode)
1413 maxlen += t->entries[i]->name->str_len + 34;
1416 strbuf_reset(b);
1417 strbuf_grow(b, maxlen);
1418 for (i = 0; i < t->entry_count; i++) {
1419 struct tree_entry *e = t->entries[i];
1420 if (!e->versions[v].mode)
1421 continue;
1422 strbuf_addf(b, "%o %s%c",
1423 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1424 e->name->str_dat, '\0');
1425 strbuf_add(b, e->versions[v].sha1, 20);
1429 static void store_tree(struct tree_entry *root)
1431 struct tree_content *t;
1432 unsigned int i, j, del;
1433 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1434 struct object_entry *le = NULL;
1436 if (!is_null_sha1(root->versions[1].sha1))
1437 return;
1439 if (!root->tree)
1440 load_tree(root);
1441 t = root->tree;
1443 for (i = 0; i < t->entry_count; i++) {
1444 if (t->entries[i]->tree)
1445 store_tree(t->entries[i]);
1448 if (!(root->versions[0].mode & NO_DELTA))
1449 le = find_object(root->versions[0].sha1);
1450 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1451 mktree(t, 0, &old_tree);
1452 lo.data = old_tree;
1453 lo.offset = le->idx.offset;
1454 lo.depth = t->delta_depth;
1457 mktree(t, 1, &new_tree);
1458 store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1460 t->delta_depth = lo.depth;
1461 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1462 struct tree_entry *e = t->entries[i];
1463 if (e->versions[1].mode) {
1464 e->versions[0].mode = e->versions[1].mode;
1465 hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1466 t->entries[j++] = e;
1467 } else {
1468 release_tree_entry(e);
1469 del++;
1472 t->entry_count -= del;
1475 static void tree_content_replace(
1476 struct tree_entry *root,
1477 const unsigned char *sha1,
1478 const uint16_t mode,
1479 struct tree_content *newtree)
1481 if (!S_ISDIR(mode))
1482 die("Root cannot be a non-directory");
1483 hashclr(root->versions[0].sha1);
1484 hashcpy(root->versions[1].sha1, sha1);
1485 if (root->tree)
1486 release_tree_content_recursive(root->tree);
1487 root->tree = newtree;
1490 static int tree_content_set(
1491 struct tree_entry *root,
1492 const char *p,
1493 const unsigned char *sha1,
1494 const uint16_t mode,
1495 struct tree_content *subtree)
1497 struct tree_content *t;
1498 const char *slash1;
1499 unsigned int i, n;
1500 struct tree_entry *e;
1502 slash1 = strchrnul(p, '/');
1503 n = slash1 - p;
1504 if (!n)
1505 die("Empty path component found in input");
1506 if (!*slash1 && !S_ISDIR(mode) && subtree)
1507 die("Non-directories cannot have subtrees");
1509 if (!root->tree)
1510 load_tree(root);
1511 t = root->tree;
1512 for (i = 0; i < t->entry_count; i++) {
1513 e = t->entries[i];
1514 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1515 if (!*slash1) {
1516 if (!S_ISDIR(mode)
1517 && e->versions[1].mode == mode
1518 && !hashcmp(e->versions[1].sha1, sha1))
1519 return 0;
1520 e->versions[1].mode = mode;
1521 hashcpy(e->versions[1].sha1, sha1);
1522 if (e->tree)
1523 release_tree_content_recursive(e->tree);
1524 e->tree = subtree;
1527 * We need to leave e->versions[0].sha1 alone
1528 * to avoid modifying the preimage tree used
1529 * when writing out the parent directory.
1530 * But after replacing the subdir with a
1531 * completely different one, it's not a good
1532 * delta base any more, and besides, we've
1533 * thrown away the tree entries needed to
1534 * make a delta against it.
1536 * So let's just explicitly disable deltas
1537 * for the subtree.
1539 if (S_ISDIR(e->versions[0].mode))
1540 e->versions[0].mode |= NO_DELTA;
1542 hashclr(root->versions[1].sha1);
1543 return 1;
1545 if (!S_ISDIR(e->versions[1].mode)) {
1546 e->tree = new_tree_content(8);
1547 e->versions[1].mode = S_IFDIR;
1549 if (!e->tree)
1550 load_tree(e);
1551 if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1552 hashclr(root->versions[1].sha1);
1553 return 1;
1555 return 0;
1559 if (t->entry_count == t->entry_capacity)
1560 root->tree = t = grow_tree_content(t, t->entry_count);
1561 e = new_tree_entry();
1562 e->name = to_atom(p, n);
1563 e->versions[0].mode = 0;
1564 hashclr(e->versions[0].sha1);
1565 t->entries[t->entry_count++] = e;
1566 if (*slash1) {
1567 e->tree = new_tree_content(8);
1568 e->versions[1].mode = S_IFDIR;
1569 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1570 } else {
1571 e->tree = subtree;
1572 e->versions[1].mode = mode;
1573 hashcpy(e->versions[1].sha1, sha1);
1575 hashclr(root->versions[1].sha1);
1576 return 1;
1579 static int tree_content_remove(
1580 struct tree_entry *root,
1581 const char *p,
1582 struct tree_entry *backup_leaf,
1583 int allow_root)
1585 struct tree_content *t;
1586 const char *slash1;
1587 unsigned int i, n;
1588 struct tree_entry *e;
1590 slash1 = strchrnul(p, '/');
1591 n = slash1 - p;
1593 if (!root->tree)
1594 load_tree(root);
1596 if (!*p && allow_root) {
1597 e = root;
1598 goto del_entry;
1601 t = root->tree;
1602 for (i = 0; i < t->entry_count; i++) {
1603 e = t->entries[i];
1604 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1605 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1607 * If p names a file in some subdirectory, and a
1608 * file or symlink matching the name of the
1609 * parent directory of p exists, then p cannot
1610 * exist and need not be deleted.
1612 return 1;
1613 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1614 goto del_entry;
1615 if (!e->tree)
1616 load_tree(e);
1617 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1618 for (n = 0; n < e->tree->entry_count; n++) {
1619 if (e->tree->entries[n]->versions[1].mode) {
1620 hashclr(root->versions[1].sha1);
1621 return 1;
1624 backup_leaf = NULL;
1625 goto del_entry;
1627 return 0;
1630 return 0;
1632 del_entry:
1633 if (backup_leaf)
1634 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1635 else if (e->tree)
1636 release_tree_content_recursive(e->tree);
1637 e->tree = NULL;
1638 e->versions[1].mode = 0;
1639 hashclr(e->versions[1].sha1);
1640 hashclr(root->versions[1].sha1);
1641 return 1;
1644 static int tree_content_get(
1645 struct tree_entry *root,
1646 const char *p,
1647 struct tree_entry *leaf,
1648 int allow_root)
1650 struct tree_content *t;
1651 const char *slash1;
1652 unsigned int i, n;
1653 struct tree_entry *e;
1655 slash1 = strchrnul(p, '/');
1656 n = slash1 - p;
1657 if (!n && !allow_root)
1658 die("Empty path component found in input");
1660 if (!root->tree)
1661 load_tree(root);
1663 if (!n) {
1664 e = root;
1665 goto found_entry;
1668 t = root->tree;
1669 for (i = 0; i < t->entry_count; i++) {
1670 e = t->entries[i];
1671 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1672 if (!*slash1)
1673 goto found_entry;
1674 if (!S_ISDIR(e->versions[1].mode))
1675 return 0;
1676 if (!e->tree)
1677 load_tree(e);
1678 return tree_content_get(e, slash1 + 1, leaf, 0);
1681 return 0;
1683 found_entry:
1684 memcpy(leaf, e, sizeof(*leaf));
1685 if (e->tree && is_null_sha1(e->versions[1].sha1))
1686 leaf->tree = dup_tree_content(e->tree);
1687 else
1688 leaf->tree = NULL;
1689 return 1;
1692 static int update_branch(struct branch *b)
1694 static const char *msg = "fast-import";
1695 struct ref_transaction *transaction;
1696 unsigned char old_sha1[20];
1697 struct strbuf err = STRBUF_INIT;
1699 if (is_null_sha1(b->sha1)) {
1700 if (b->delete)
1701 delete_ref(b->name, NULL, 0);
1702 return 0;
1704 if (read_ref(b->name, old_sha1))
1705 hashclr(old_sha1);
1706 if (!force_update && !is_null_sha1(old_sha1)) {
1707 struct commit *old_cmit, *new_cmit;
1709 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1710 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1711 if (!old_cmit || !new_cmit)
1712 return error("Branch %s is missing commits.", b->name);
1714 if (!in_merge_bases(old_cmit, new_cmit)) {
1715 warning("Not updating %s"
1716 " (new tip %s does not contain %s)",
1717 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1718 return -1;
1721 transaction = ref_transaction_begin(&err);
1722 if (!transaction ||
1723 ref_transaction_update(transaction, b->name, b->sha1, old_sha1,
1724 0, msg, &err) ||
1725 ref_transaction_commit(transaction, &err)) {
1726 ref_transaction_free(transaction);
1727 error("%s", err.buf);
1728 strbuf_release(&err);
1729 return -1;
1731 ref_transaction_free(transaction);
1732 strbuf_release(&err);
1733 return 0;
1736 static void dump_branches(void)
1738 unsigned int i;
1739 struct branch *b;
1741 for (i = 0; i < branch_table_sz; i++) {
1742 for (b = branch_table[i]; b; b = b->table_next_branch)
1743 failure |= update_branch(b);
1747 static void dump_tags(void)
1749 static const char *msg = "fast-import";
1750 struct tag *t;
1751 struct strbuf ref_name = STRBUF_INIT;
1752 struct strbuf err = STRBUF_INIT;
1753 struct ref_transaction *transaction;
1755 transaction = ref_transaction_begin(&err);
1756 if (!transaction) {
1757 failure |= error("%s", err.buf);
1758 goto cleanup;
1760 for (t = first_tag; t; t = t->next_tag) {
1761 strbuf_reset(&ref_name);
1762 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1764 if (ref_transaction_update(transaction, ref_name.buf,
1765 t->sha1, NULL, 0, msg, &err)) {
1766 failure |= error("%s", err.buf);
1767 goto cleanup;
1770 if (ref_transaction_commit(transaction, &err))
1771 failure |= error("%s", err.buf);
1773 cleanup:
1774 ref_transaction_free(transaction);
1775 strbuf_release(&ref_name);
1776 strbuf_release(&err);
1779 static void dump_marks_helper(FILE *f,
1780 uintmax_t base,
1781 struct mark_set *m)
1783 uintmax_t k;
1784 if (m->shift) {
1785 for (k = 0; k < 1024; k++) {
1786 if (m->data.sets[k])
1787 dump_marks_helper(f, base + (k << m->shift),
1788 m->data.sets[k]);
1790 } else {
1791 for (k = 0; k < 1024; k++) {
1792 if (m->data.marked[k])
1793 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1794 sha1_to_hex(m->data.marked[k]->idx.sha1));
1799 static void dump_marks(void)
1801 static struct lock_file mark_lock;
1802 FILE *f;
1804 if (!export_marks_file)
1805 return;
1807 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1808 failure |= error("Unable to write marks file %s: %s",
1809 export_marks_file, strerror(errno));
1810 return;
1813 f = fdopen_lock_file(&mark_lock, "w");
1814 if (!f) {
1815 int saved_errno = errno;
1816 rollback_lock_file(&mark_lock);
1817 failure |= error("Unable to write marks file %s: %s",
1818 export_marks_file, strerror(saved_errno));
1819 return;
1822 dump_marks_helper(f, 0, marks);
1823 if (commit_lock_file(&mark_lock)) {
1824 failure |= error("Unable to commit marks file %s: %s",
1825 export_marks_file, strerror(errno));
1826 return;
1830 static void read_marks(void)
1832 char line[512];
1833 FILE *f = fopen(import_marks_file, "r");
1834 if (f)
1836 else if (import_marks_file_ignore_missing && errno == ENOENT)
1837 return; /* Marks file does not exist */
1838 else
1839 die_errno("cannot read '%s'", import_marks_file);
1840 while (fgets(line, sizeof(line), f)) {
1841 uintmax_t mark;
1842 char *end;
1843 unsigned char sha1[20];
1844 struct object_entry *e;
1846 end = strchr(line, '\n');
1847 if (line[0] != ':' || !end)
1848 die("corrupt mark line: %s", line);
1849 *end = 0;
1850 mark = strtoumax(line + 1, &end, 10);
1851 if (!mark || end == line + 1
1852 || *end != ' ' || get_sha1_hex(end + 1, sha1))
1853 die("corrupt mark line: %s", line);
1854 e = find_object(sha1);
1855 if (!e) {
1856 enum object_type type = sha1_object_info(sha1, NULL);
1857 if (type < 0)
1858 die("object not found: %s", sha1_to_hex(sha1));
1859 e = insert_object(sha1);
1860 e->type = type;
1861 e->pack_id = MAX_PACK_ID;
1862 e->idx.offset = 1; /* just not zero! */
1864 insert_mark(mark, e);
1866 fclose(f);
1870 static int read_next_command(void)
1872 static int stdin_eof = 0;
1874 if (stdin_eof) {
1875 unread_command_buf = 0;
1876 return EOF;
1879 for (;;) {
1880 const char *p;
1882 if (unread_command_buf) {
1883 unread_command_buf = 0;
1884 } else {
1885 struct recent_command *rc;
1887 strbuf_detach(&command_buf, NULL);
1888 stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1889 if (stdin_eof)
1890 return EOF;
1892 if (!seen_data_command
1893 && !starts_with(command_buf.buf, "feature ")
1894 && !starts_with(command_buf.buf, "option ")) {
1895 parse_argv();
1898 rc = rc_free;
1899 if (rc)
1900 rc_free = rc->next;
1901 else {
1902 rc = cmd_hist.next;
1903 cmd_hist.next = rc->next;
1904 cmd_hist.next->prev = &cmd_hist;
1905 free(rc->buf);
1908 rc->buf = command_buf.buf;
1909 rc->prev = cmd_tail;
1910 rc->next = cmd_hist.prev;
1911 rc->prev->next = rc;
1912 cmd_tail = rc;
1914 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1915 parse_get_mark(p);
1916 continue;
1918 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1919 parse_cat_blob(p);
1920 continue;
1922 if (command_buf.buf[0] == '#')
1923 continue;
1924 return 0;
1928 static void skip_optional_lf(void)
1930 int term_char = fgetc(stdin);
1931 if (term_char != '\n' && term_char != EOF)
1932 ungetc(term_char, stdin);
1935 static void parse_mark(void)
1937 const char *v;
1938 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1939 next_mark = strtoumax(v, NULL, 10);
1940 read_next_command();
1942 else
1943 next_mark = 0;
1946 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1948 const char *data;
1949 strbuf_reset(sb);
1951 if (!skip_prefix(command_buf.buf, "data ", &data))
1952 die("Expected 'data n' command, found: %s", command_buf.buf);
1954 if (skip_prefix(data, "<<", &data)) {
1955 char *term = xstrdup(data);
1956 size_t term_len = command_buf.len - (data - command_buf.buf);
1958 strbuf_detach(&command_buf, NULL);
1959 for (;;) {
1960 if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1961 die("EOF in data (terminator '%s' not found)", term);
1962 if (term_len == command_buf.len
1963 && !strcmp(term, command_buf.buf))
1964 break;
1965 strbuf_addbuf(sb, &command_buf);
1966 strbuf_addch(sb, '\n');
1968 free(term);
1970 else {
1971 uintmax_t len = strtoumax(data, NULL, 10);
1972 size_t n = 0, length = (size_t)len;
1974 if (limit && limit < len) {
1975 *len_res = len;
1976 return 0;
1978 if (length < len)
1979 die("data is too large to use in this context");
1981 while (n < length) {
1982 size_t s = strbuf_fread(sb, length - n, stdin);
1983 if (!s && feof(stdin))
1984 die("EOF in data (%lu bytes remaining)",
1985 (unsigned long)(length - n));
1986 n += s;
1990 skip_optional_lf();
1991 return 1;
1994 static int validate_raw_date(const char *src, struct strbuf *result)
1996 const char *orig_src = src;
1997 char *endp;
1998 unsigned long num;
2000 errno = 0;
2002 num = strtoul(src, &endp, 10);
2003 /* NEEDSWORK: perhaps check for reasonable values? */
2004 if (errno || endp == src || *endp != ' ')
2005 return -1;
2007 src = endp + 1;
2008 if (*src != '-' && *src != '+')
2009 return -1;
2011 num = strtoul(src + 1, &endp, 10);
2012 if (errno || endp == src + 1 || *endp || 1400 < num)
2013 return -1;
2015 strbuf_addstr(result, orig_src);
2016 return 0;
2019 static char *parse_ident(const char *buf)
2021 const char *ltgt;
2022 size_t name_len;
2023 struct strbuf ident = STRBUF_INIT;
2025 /* ensure there is a space delimiter even if there is no name */
2026 if (*buf == '<')
2027 --buf;
2029 ltgt = buf + strcspn(buf, "<>");
2030 if (*ltgt != '<')
2031 die("Missing < in ident string: %s", buf);
2032 if (ltgt != buf && ltgt[-1] != ' ')
2033 die("Missing space before < in ident string: %s", buf);
2034 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2035 if (*ltgt != '>')
2036 die("Missing > in ident string: %s", buf);
2037 ltgt++;
2038 if (*ltgt != ' ')
2039 die("Missing space after > in ident string: %s", buf);
2040 ltgt++;
2041 name_len = ltgt - buf;
2042 strbuf_add(&ident, buf, name_len);
2044 switch (whenspec) {
2045 case WHENSPEC_RAW:
2046 if (validate_raw_date(ltgt, &ident) < 0)
2047 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2048 break;
2049 case WHENSPEC_RFC2822:
2050 if (parse_date(ltgt, &ident) < 0)
2051 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2052 break;
2053 case WHENSPEC_NOW:
2054 if (strcmp("now", ltgt))
2055 die("Date in ident must be 'now': %s", buf);
2056 datestamp(&ident);
2057 break;
2060 return strbuf_detach(&ident, NULL);
2063 static void parse_and_store_blob(
2064 struct last_object *last,
2065 unsigned char *sha1out,
2066 uintmax_t mark)
2068 static struct strbuf buf = STRBUF_INIT;
2069 uintmax_t len;
2071 if (parse_data(&buf, big_file_threshold, &len))
2072 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2073 else {
2074 if (last) {
2075 strbuf_release(&last->data);
2076 last->offset = 0;
2077 last->depth = 0;
2079 stream_blob(len, sha1out, mark);
2080 skip_optional_lf();
2084 static void parse_new_blob(void)
2086 read_next_command();
2087 parse_mark();
2088 parse_and_store_blob(&last_blob, NULL, next_mark);
2091 static void unload_one_branch(void)
2093 while (cur_active_branches
2094 && cur_active_branches >= max_active_branches) {
2095 uintmax_t min_commit = ULONG_MAX;
2096 struct branch *e, *l = NULL, *p = NULL;
2098 for (e = active_branches; e; e = e->active_next_branch) {
2099 if (e->last_commit < min_commit) {
2100 p = l;
2101 min_commit = e->last_commit;
2103 l = e;
2106 if (p) {
2107 e = p->active_next_branch;
2108 p->active_next_branch = e->active_next_branch;
2109 } else {
2110 e = active_branches;
2111 active_branches = e->active_next_branch;
2113 e->active = 0;
2114 e->active_next_branch = NULL;
2115 if (e->branch_tree.tree) {
2116 release_tree_content_recursive(e->branch_tree.tree);
2117 e->branch_tree.tree = NULL;
2119 cur_active_branches--;
2123 static void load_branch(struct branch *b)
2125 load_tree(&b->branch_tree);
2126 if (!b->active) {
2127 b->active = 1;
2128 b->active_next_branch = active_branches;
2129 active_branches = b;
2130 cur_active_branches++;
2131 branch_load_count++;
2135 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2137 unsigned char fanout = 0;
2138 while ((num_notes >>= 8))
2139 fanout++;
2140 return fanout;
2143 static void construct_path_with_fanout(const char *hex_sha1,
2144 unsigned char fanout, char *path)
2146 unsigned int i = 0, j = 0;
2147 if (fanout >= 20)
2148 die("Too large fanout (%u)", fanout);
2149 while (fanout) {
2150 path[i++] = hex_sha1[j++];
2151 path[i++] = hex_sha1[j++];
2152 path[i++] = '/';
2153 fanout--;
2155 memcpy(path + i, hex_sha1 + j, 40 - j);
2156 path[i + 40 - j] = '\0';
2159 static uintmax_t do_change_note_fanout(
2160 struct tree_entry *orig_root, struct tree_entry *root,
2161 char *hex_sha1, unsigned int hex_sha1_len,
2162 char *fullpath, unsigned int fullpath_len,
2163 unsigned char fanout)
2165 struct tree_content *t = root->tree;
2166 struct tree_entry *e, leaf;
2167 unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2168 uintmax_t num_notes = 0;
2169 unsigned char sha1[20];
2170 char realpath[60];
2172 for (i = 0; t && i < t->entry_count; i++) {
2173 e = t->entries[i];
2174 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2175 tmp_fullpath_len = fullpath_len;
2178 * We're interested in EITHER existing note entries (entries
2179 * with exactly 40 hex chars in path, not including directory
2180 * separators), OR directory entries that may contain note
2181 * entries (with < 40 hex chars in path).
2182 * Also, each path component in a note entry must be a multiple
2183 * of 2 chars.
2185 if (!e->versions[1].mode ||
2186 tmp_hex_sha1_len > 40 ||
2187 e->name->str_len % 2)
2188 continue;
2190 /* This _may_ be a note entry, or a subdir containing notes */
2191 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2192 e->name->str_len);
2193 if (tmp_fullpath_len)
2194 fullpath[tmp_fullpath_len++] = '/';
2195 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2196 e->name->str_len);
2197 tmp_fullpath_len += e->name->str_len;
2198 fullpath[tmp_fullpath_len] = '\0';
2200 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2201 /* This is a note entry */
2202 if (fanout == 0xff) {
2203 /* Counting mode, no rename */
2204 num_notes++;
2205 continue;
2207 construct_path_with_fanout(hex_sha1, fanout, realpath);
2208 if (!strcmp(fullpath, realpath)) {
2209 /* Note entry is in correct location */
2210 num_notes++;
2211 continue;
2214 /* Rename fullpath to realpath */
2215 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2216 die("Failed to remove path %s", fullpath);
2217 tree_content_set(orig_root, realpath,
2218 leaf.versions[1].sha1,
2219 leaf.versions[1].mode,
2220 leaf.tree);
2221 } else if (S_ISDIR(e->versions[1].mode)) {
2222 /* This is a subdir that may contain note entries */
2223 if (!e->tree)
2224 load_tree(e);
2225 num_notes += do_change_note_fanout(orig_root, e,
2226 hex_sha1, tmp_hex_sha1_len,
2227 fullpath, tmp_fullpath_len, fanout);
2230 /* The above may have reallocated the current tree_content */
2231 t = root->tree;
2233 return num_notes;
2236 static uintmax_t change_note_fanout(struct tree_entry *root,
2237 unsigned char fanout)
2239 char hex_sha1[40], path[60];
2240 return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2244 * Given a pointer into a string, parse a mark reference:
2246 * idnum ::= ':' bigint;
2248 * Return the first character after the value in *endptr.
2250 * Complain if the following character is not what is expected,
2251 * either a space or end of the string.
2253 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2255 uintmax_t mark;
2257 assert(*p == ':');
2258 p++;
2259 mark = strtoumax(p, endptr, 10);
2260 if (*endptr == p)
2261 die("No value after ':' in mark: %s", command_buf.buf);
2262 return mark;
2266 * Parse the mark reference, and complain if this is not the end of
2267 * the string.
2269 static uintmax_t parse_mark_ref_eol(const char *p)
2271 char *end;
2272 uintmax_t mark;
2274 mark = parse_mark_ref(p, &end);
2275 if (*end != '\0')
2276 die("Garbage after mark: %s", command_buf.buf);
2277 return mark;
2281 * Parse the mark reference, demanding a trailing space. Return a
2282 * pointer to the space.
2284 static uintmax_t parse_mark_ref_space(const char **p)
2286 uintmax_t mark;
2287 char *end;
2289 mark = parse_mark_ref(*p, &end);
2290 if (*end++ != ' ')
2291 die("Missing space after mark: %s", command_buf.buf);
2292 *p = end;
2293 return mark;
2296 static void file_change_m(const char *p, struct branch *b)
2298 static struct strbuf uq = STRBUF_INIT;
2299 const char *endp;
2300 struct object_entry *oe;
2301 unsigned char sha1[20];
2302 uint16_t mode, inline_data = 0;
2304 p = get_mode(p, &mode);
2305 if (!p)
2306 die("Corrupt mode: %s", command_buf.buf);
2307 switch (mode) {
2308 case 0644:
2309 case 0755:
2310 mode |= S_IFREG;
2311 case S_IFREG | 0644:
2312 case S_IFREG | 0755:
2313 case S_IFLNK:
2314 case S_IFDIR:
2315 case S_IFGITLINK:
2316 /* ok */
2317 break;
2318 default:
2319 die("Corrupt mode: %s", command_buf.buf);
2322 if (*p == ':') {
2323 oe = find_mark(parse_mark_ref_space(&p));
2324 hashcpy(sha1, oe->idx.sha1);
2325 } else if (skip_prefix(p, "inline ", &p)) {
2326 inline_data = 1;
2327 oe = NULL; /* not used with inline_data, but makes gcc happy */
2328 } else {
2329 if (get_sha1_hex(p, sha1))
2330 die("Invalid dataref: %s", command_buf.buf);
2331 oe = find_object(sha1);
2332 p += 40;
2333 if (*p++ != ' ')
2334 die("Missing space after SHA1: %s", command_buf.buf);
2337 strbuf_reset(&uq);
2338 if (!unquote_c_style(&uq, p, &endp)) {
2339 if (*endp)
2340 die("Garbage after path in: %s", command_buf.buf);
2341 p = uq.buf;
2344 /* Git does not track empty, non-toplevel directories. */
2345 if (S_ISDIR(mode) && !hashcmp(sha1, EMPTY_TREE_SHA1_BIN) && *p) {
2346 tree_content_remove(&b->branch_tree, p, NULL, 0);
2347 return;
2350 if (S_ISGITLINK(mode)) {
2351 if (inline_data)
2352 die("Git links cannot be specified 'inline': %s",
2353 command_buf.buf);
2354 else if (oe) {
2355 if (oe->type != OBJ_COMMIT)
2356 die("Not a commit (actually a %s): %s",
2357 typename(oe->type), command_buf.buf);
2360 * Accept the sha1 without checking; it expected to be in
2361 * another repository.
2363 } else if (inline_data) {
2364 if (S_ISDIR(mode))
2365 die("Directories cannot be specified 'inline': %s",
2366 command_buf.buf);
2367 if (p != uq.buf) {
2368 strbuf_addstr(&uq, p);
2369 p = uq.buf;
2371 read_next_command();
2372 parse_and_store_blob(&last_blob, sha1, 0);
2373 } else {
2374 enum object_type expected = S_ISDIR(mode) ?
2375 OBJ_TREE: OBJ_BLOB;
2376 enum object_type type = oe ? oe->type :
2377 sha1_object_info(sha1, NULL);
2378 if (type < 0)
2379 die("%s not found: %s",
2380 S_ISDIR(mode) ? "Tree" : "Blob",
2381 command_buf.buf);
2382 if (type != expected)
2383 die("Not a %s (actually a %s): %s",
2384 typename(expected), typename(type),
2385 command_buf.buf);
2388 if (!*p) {
2389 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2390 return;
2392 tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2395 static void file_change_d(const char *p, struct branch *b)
2397 static struct strbuf uq = STRBUF_INIT;
2398 const char *endp;
2400 strbuf_reset(&uq);
2401 if (!unquote_c_style(&uq, p, &endp)) {
2402 if (*endp)
2403 die("Garbage after path in: %s", command_buf.buf);
2404 p = uq.buf;
2406 tree_content_remove(&b->branch_tree, p, NULL, 1);
2409 static void file_change_cr(const char *s, struct branch *b, int rename)
2411 const char *d;
2412 static struct strbuf s_uq = STRBUF_INIT;
2413 static struct strbuf d_uq = STRBUF_INIT;
2414 const char *endp;
2415 struct tree_entry leaf;
2417 strbuf_reset(&s_uq);
2418 if (!unquote_c_style(&s_uq, s, &endp)) {
2419 if (*endp != ' ')
2420 die("Missing space after source: %s", command_buf.buf);
2421 } else {
2422 endp = strchr(s, ' ');
2423 if (!endp)
2424 die("Missing space after source: %s", command_buf.buf);
2425 strbuf_add(&s_uq, s, endp - s);
2427 s = s_uq.buf;
2429 endp++;
2430 if (!*endp)
2431 die("Missing dest: %s", command_buf.buf);
2433 d = endp;
2434 strbuf_reset(&d_uq);
2435 if (!unquote_c_style(&d_uq, d, &endp)) {
2436 if (*endp)
2437 die("Garbage after dest in: %s", command_buf.buf);
2438 d = d_uq.buf;
2441 memset(&leaf, 0, sizeof(leaf));
2442 if (rename)
2443 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2444 else
2445 tree_content_get(&b->branch_tree, s, &leaf, 1);
2446 if (!leaf.versions[1].mode)
2447 die("Path %s not in branch", s);
2448 if (!*d) { /* C "path/to/subdir" "" */
2449 tree_content_replace(&b->branch_tree,
2450 leaf.versions[1].sha1,
2451 leaf.versions[1].mode,
2452 leaf.tree);
2453 return;
2455 tree_content_set(&b->branch_tree, d,
2456 leaf.versions[1].sha1,
2457 leaf.versions[1].mode,
2458 leaf.tree);
2461 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2463 static struct strbuf uq = STRBUF_INIT;
2464 struct object_entry *oe;
2465 struct branch *s;
2466 unsigned char sha1[20], commit_sha1[20];
2467 char path[60];
2468 uint16_t inline_data = 0;
2469 unsigned char new_fanout;
2472 * When loading a branch, we don't traverse its tree to count the real
2473 * number of notes (too expensive to do this for all non-note refs).
2474 * This means that recently loaded notes refs might incorrectly have
2475 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2477 * Fix this by traversing the tree and counting the number of notes
2478 * when b->num_notes == 0. If the notes tree is truly empty, the
2479 * calculation should not take long.
2481 if (b->num_notes == 0 && *old_fanout == 0) {
2482 /* Invoke change_note_fanout() in "counting mode". */
2483 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2484 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2487 /* Now parse the notemodify command. */
2488 /* <dataref> or 'inline' */
2489 if (*p == ':') {
2490 oe = find_mark(parse_mark_ref_space(&p));
2491 hashcpy(sha1, oe->idx.sha1);
2492 } else if (skip_prefix(p, "inline ", &p)) {
2493 inline_data = 1;
2494 oe = NULL; /* not used with inline_data, but makes gcc happy */
2495 } else {
2496 if (get_sha1_hex(p, sha1))
2497 die("Invalid dataref: %s", command_buf.buf);
2498 oe = find_object(sha1);
2499 p += 40;
2500 if (*p++ != ' ')
2501 die("Missing space after SHA1: %s", command_buf.buf);
2504 /* <commit-ish> */
2505 s = lookup_branch(p);
2506 if (s) {
2507 if (is_null_sha1(s->sha1))
2508 die("Can't add a note on empty branch.");
2509 hashcpy(commit_sha1, s->sha1);
2510 } else if (*p == ':') {
2511 uintmax_t commit_mark = parse_mark_ref_eol(p);
2512 struct object_entry *commit_oe = find_mark(commit_mark);
2513 if (commit_oe->type != OBJ_COMMIT)
2514 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2515 hashcpy(commit_sha1, commit_oe->idx.sha1);
2516 } else if (!get_sha1(p, commit_sha1)) {
2517 unsigned long size;
2518 char *buf = read_object_with_reference(commit_sha1,
2519 commit_type, &size, commit_sha1);
2520 if (!buf || size < 46)
2521 die("Not a valid commit: %s", p);
2522 free(buf);
2523 } else
2524 die("Invalid ref name or SHA1 expression: %s", p);
2526 if (inline_data) {
2527 if (p != uq.buf) {
2528 strbuf_addstr(&uq, p);
2529 p = uq.buf;
2531 read_next_command();
2532 parse_and_store_blob(&last_blob, sha1, 0);
2533 } else if (oe) {
2534 if (oe->type != OBJ_BLOB)
2535 die("Not a blob (actually a %s): %s",
2536 typename(oe->type), command_buf.buf);
2537 } else if (!is_null_sha1(sha1)) {
2538 enum object_type type = sha1_object_info(sha1, NULL);
2539 if (type < 0)
2540 die("Blob not found: %s", command_buf.buf);
2541 if (type != OBJ_BLOB)
2542 die("Not a blob (actually a %s): %s",
2543 typename(type), command_buf.buf);
2546 construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2547 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2548 b->num_notes--;
2550 if (is_null_sha1(sha1))
2551 return; /* nothing to insert */
2553 b->num_notes++;
2554 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2555 construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2556 tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2559 static void file_change_deleteall(struct branch *b)
2561 release_tree_content_recursive(b->branch_tree.tree);
2562 hashclr(b->branch_tree.versions[0].sha1);
2563 hashclr(b->branch_tree.versions[1].sha1);
2564 load_tree(&b->branch_tree);
2565 b->num_notes = 0;
2568 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2570 if (!buf || size < 46)
2571 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2572 if (memcmp("tree ", buf, 5)
2573 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2574 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2575 hashcpy(b->branch_tree.versions[0].sha1,
2576 b->branch_tree.versions[1].sha1);
2579 static void parse_from_existing(struct branch *b)
2581 if (is_null_sha1(b->sha1)) {
2582 hashclr(b->branch_tree.versions[0].sha1);
2583 hashclr(b->branch_tree.versions[1].sha1);
2584 } else {
2585 unsigned long size;
2586 char *buf;
2588 buf = read_object_with_reference(b->sha1,
2589 commit_type, &size, b->sha1);
2590 parse_from_commit(b, buf, size);
2591 free(buf);
2595 static int parse_from(struct branch *b)
2597 const char *from;
2598 struct branch *s;
2599 unsigned char sha1[20];
2601 if (!skip_prefix(command_buf.buf, "from ", &from))
2602 return 0;
2604 hashcpy(sha1, b->branch_tree.versions[1].sha1);
2606 s = lookup_branch(from);
2607 if (b == s)
2608 die("Can't create a branch from itself: %s", b->name);
2609 else if (s) {
2610 unsigned char *t = s->branch_tree.versions[1].sha1;
2611 hashcpy(b->sha1, s->sha1);
2612 hashcpy(b->branch_tree.versions[0].sha1, t);
2613 hashcpy(b->branch_tree.versions[1].sha1, t);
2614 } else if (*from == ':') {
2615 uintmax_t idnum = parse_mark_ref_eol(from);
2616 struct object_entry *oe = find_mark(idnum);
2617 if (oe->type != OBJ_COMMIT)
2618 die("Mark :%" PRIuMAX " not a commit", idnum);
2619 if (hashcmp(b->sha1, oe->idx.sha1)) {
2620 hashcpy(b->sha1, oe->idx.sha1);
2621 if (oe->pack_id != MAX_PACK_ID) {
2622 unsigned long size;
2623 char *buf = gfi_unpack_entry(oe, &size);
2624 parse_from_commit(b, buf, size);
2625 free(buf);
2626 } else
2627 parse_from_existing(b);
2629 } else if (!get_sha1(from, b->sha1)) {
2630 parse_from_existing(b);
2631 if (is_null_sha1(b->sha1))
2632 b->delete = 1;
2634 else
2635 die("Invalid ref name or SHA1 expression: %s", from);
2637 if (b->branch_tree.tree && hashcmp(sha1, b->branch_tree.versions[1].sha1)) {
2638 release_tree_content_recursive(b->branch_tree.tree);
2639 b->branch_tree.tree = NULL;
2642 read_next_command();
2643 return 1;
2646 static struct hash_list *parse_merge(unsigned int *count)
2648 struct hash_list *list = NULL, **tail = &list, *n;
2649 const char *from;
2650 struct branch *s;
2652 *count = 0;
2653 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2654 n = xmalloc(sizeof(*n));
2655 s = lookup_branch(from);
2656 if (s)
2657 hashcpy(n->sha1, s->sha1);
2658 else if (*from == ':') {
2659 uintmax_t idnum = parse_mark_ref_eol(from);
2660 struct object_entry *oe = find_mark(idnum);
2661 if (oe->type != OBJ_COMMIT)
2662 die("Mark :%" PRIuMAX " not a commit", idnum);
2663 hashcpy(n->sha1, oe->idx.sha1);
2664 } else if (!get_sha1(from, n->sha1)) {
2665 unsigned long size;
2666 char *buf = read_object_with_reference(n->sha1,
2667 commit_type, &size, n->sha1);
2668 if (!buf || size < 46)
2669 die("Not a valid commit: %s", from);
2670 free(buf);
2671 } else
2672 die("Invalid ref name or SHA1 expression: %s", from);
2674 n->next = NULL;
2675 *tail = n;
2676 tail = &n->next;
2678 (*count)++;
2679 read_next_command();
2681 return list;
2684 static void parse_new_commit(const char *arg)
2686 static struct strbuf msg = STRBUF_INIT;
2687 struct branch *b;
2688 char *author = NULL;
2689 char *committer = NULL;
2690 struct hash_list *merge_list = NULL;
2691 unsigned int merge_count;
2692 unsigned char prev_fanout, new_fanout;
2693 const char *v;
2695 b = lookup_branch(arg);
2696 if (!b)
2697 b = new_branch(arg);
2699 read_next_command();
2700 parse_mark();
2701 if (skip_prefix(command_buf.buf, "author ", &v)) {
2702 author = parse_ident(v);
2703 read_next_command();
2705 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2706 committer = parse_ident(v);
2707 read_next_command();
2709 if (!committer)
2710 die("Expected committer but didn't get one");
2711 parse_data(&msg, 0, NULL);
2712 read_next_command();
2713 parse_from(b);
2714 merge_list = parse_merge(&merge_count);
2716 /* ensure the branch is active/loaded */
2717 if (!b->branch_tree.tree || !max_active_branches) {
2718 unload_one_branch();
2719 load_branch(b);
2722 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2724 /* file_change* */
2725 while (command_buf.len > 0) {
2726 if (skip_prefix(command_buf.buf, "M ", &v))
2727 file_change_m(v, b);
2728 else if (skip_prefix(command_buf.buf, "D ", &v))
2729 file_change_d(v, b);
2730 else if (skip_prefix(command_buf.buf, "R ", &v))
2731 file_change_cr(v, b, 1);
2732 else if (skip_prefix(command_buf.buf, "C ", &v))
2733 file_change_cr(v, b, 0);
2734 else if (skip_prefix(command_buf.buf, "N ", &v))
2735 note_change_n(v, b, &prev_fanout);
2736 else if (!strcmp("deleteall", command_buf.buf))
2737 file_change_deleteall(b);
2738 else if (skip_prefix(command_buf.buf, "ls ", &v))
2739 parse_ls(v, b);
2740 else {
2741 unread_command_buf = 1;
2742 break;
2744 if (read_next_command() == EOF)
2745 break;
2748 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2749 if (new_fanout != prev_fanout)
2750 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2752 /* build the tree and the commit */
2753 store_tree(&b->branch_tree);
2754 hashcpy(b->branch_tree.versions[0].sha1,
2755 b->branch_tree.versions[1].sha1);
2757 strbuf_reset(&new_data);
2758 strbuf_addf(&new_data, "tree %s\n",
2759 sha1_to_hex(b->branch_tree.versions[1].sha1));
2760 if (!is_null_sha1(b->sha1))
2761 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2762 while (merge_list) {
2763 struct hash_list *next = merge_list->next;
2764 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2765 free(merge_list);
2766 merge_list = next;
2768 strbuf_addf(&new_data,
2769 "author %s\n"
2770 "committer %s\n"
2771 "\n",
2772 author ? author : committer, committer);
2773 strbuf_addbuf(&new_data, &msg);
2774 free(author);
2775 free(committer);
2777 if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2778 b->pack_id = pack_id;
2779 b->last_commit = object_count_by_type[OBJ_COMMIT];
2782 static void parse_new_tag(const char *arg)
2784 static struct strbuf msg = STRBUF_INIT;
2785 const char *from;
2786 char *tagger;
2787 struct branch *s;
2788 struct tag *t;
2789 uintmax_t from_mark = 0;
2790 unsigned char sha1[20];
2791 enum object_type type;
2792 const char *v;
2794 t = pool_alloc(sizeof(struct tag));
2795 memset(t, 0, sizeof(struct tag));
2796 t->name = pool_strdup(arg);
2797 if (last_tag)
2798 last_tag->next_tag = t;
2799 else
2800 first_tag = t;
2801 last_tag = t;
2802 read_next_command();
2804 /* from ... */
2805 if (!skip_prefix(command_buf.buf, "from ", &from))
2806 die("Expected from command, got %s", command_buf.buf);
2807 s = lookup_branch(from);
2808 if (s) {
2809 if (is_null_sha1(s->sha1))
2810 die("Can't tag an empty branch.");
2811 hashcpy(sha1, s->sha1);
2812 type = OBJ_COMMIT;
2813 } else if (*from == ':') {
2814 struct object_entry *oe;
2815 from_mark = parse_mark_ref_eol(from);
2816 oe = find_mark(from_mark);
2817 type = oe->type;
2818 hashcpy(sha1, oe->idx.sha1);
2819 } else if (!get_sha1(from, sha1)) {
2820 struct object_entry *oe = find_object(sha1);
2821 if (!oe) {
2822 type = sha1_object_info(sha1, NULL);
2823 if (type < 0)
2824 die("Not a valid object: %s", from);
2825 } else
2826 type = oe->type;
2827 } else
2828 die("Invalid ref name or SHA1 expression: %s", from);
2829 read_next_command();
2831 /* tagger ... */
2832 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2833 tagger = parse_ident(v);
2834 read_next_command();
2835 } else
2836 tagger = NULL;
2838 /* tag payload/message */
2839 parse_data(&msg, 0, NULL);
2841 /* build the tag object */
2842 strbuf_reset(&new_data);
2844 strbuf_addf(&new_data,
2845 "object %s\n"
2846 "type %s\n"
2847 "tag %s\n",
2848 sha1_to_hex(sha1), typename(type), t->name);
2849 if (tagger)
2850 strbuf_addf(&new_data,
2851 "tagger %s\n", tagger);
2852 strbuf_addch(&new_data, '\n');
2853 strbuf_addbuf(&new_data, &msg);
2854 free(tagger);
2856 if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2857 t->pack_id = MAX_PACK_ID;
2858 else
2859 t->pack_id = pack_id;
2862 static void parse_reset_branch(const char *arg)
2864 struct branch *b;
2866 b = lookup_branch(arg);
2867 if (b) {
2868 hashclr(b->sha1);
2869 hashclr(b->branch_tree.versions[0].sha1);
2870 hashclr(b->branch_tree.versions[1].sha1);
2871 if (b->branch_tree.tree) {
2872 release_tree_content_recursive(b->branch_tree.tree);
2873 b->branch_tree.tree = NULL;
2876 else
2877 b = new_branch(arg);
2878 read_next_command();
2879 parse_from(b);
2880 if (command_buf.len > 0)
2881 unread_command_buf = 1;
2884 static void cat_blob_write(const char *buf, unsigned long size)
2886 if (write_in_full(cat_blob_fd, buf, size) != size)
2887 die_errno("Write to frontend failed");
2890 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2892 struct strbuf line = STRBUF_INIT;
2893 unsigned long size;
2894 enum object_type type = 0;
2895 char *buf;
2897 if (!oe || oe->pack_id == MAX_PACK_ID) {
2898 buf = read_sha1_file(sha1, &type, &size);
2899 } else {
2900 type = oe->type;
2901 buf = gfi_unpack_entry(oe, &size);
2905 * Output based on batch_one_object() from cat-file.c.
2907 if (type <= 0) {
2908 strbuf_reset(&line);
2909 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2910 cat_blob_write(line.buf, line.len);
2911 strbuf_release(&line);
2912 free(buf);
2913 return;
2915 if (!buf)
2916 die("Can't read object %s", sha1_to_hex(sha1));
2917 if (type != OBJ_BLOB)
2918 die("Object %s is a %s but a blob was expected.",
2919 sha1_to_hex(sha1), typename(type));
2920 strbuf_reset(&line);
2921 strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2922 typename(type), size);
2923 cat_blob_write(line.buf, line.len);
2924 strbuf_release(&line);
2925 cat_blob_write(buf, size);
2926 cat_blob_write("\n", 1);
2927 if (oe && oe->pack_id == pack_id) {
2928 last_blob.offset = oe->idx.offset;
2929 strbuf_attach(&last_blob.data, buf, size, size);
2930 last_blob.depth = oe->depth;
2931 } else
2932 free(buf);
2935 static void parse_get_mark(const char *p)
2937 struct object_entry *oe = oe;
2938 char output[42];
2940 /* get-mark SP <object> LF */
2941 if (*p != ':')
2942 die("Not a mark: %s", p);
2944 oe = find_mark(parse_mark_ref_eol(p));
2945 if (!oe)
2946 die("Unknown mark: %s", command_buf.buf);
2948 snprintf(output, sizeof(output), "%s\n", sha1_to_hex(oe->idx.sha1));
2949 cat_blob_write(output, 41);
2952 static void parse_cat_blob(const char *p)
2954 struct object_entry *oe = oe;
2955 unsigned char sha1[20];
2957 /* cat-blob SP <object> LF */
2958 if (*p == ':') {
2959 oe = find_mark(parse_mark_ref_eol(p));
2960 if (!oe)
2961 die("Unknown mark: %s", command_buf.buf);
2962 hashcpy(sha1, oe->idx.sha1);
2963 } else {
2964 if (get_sha1_hex(p, sha1))
2965 die("Invalid dataref: %s", command_buf.buf);
2966 if (p[40])
2967 die("Garbage after SHA1: %s", command_buf.buf);
2968 oe = find_object(sha1);
2971 cat_blob(oe, sha1);
2974 static struct object_entry *dereference(struct object_entry *oe,
2975 unsigned char sha1[20])
2977 unsigned long size;
2978 char *buf = NULL;
2979 if (!oe) {
2980 enum object_type type = sha1_object_info(sha1, NULL);
2981 if (type < 0)
2982 die("object not found: %s", sha1_to_hex(sha1));
2983 /* cache it! */
2984 oe = insert_object(sha1);
2985 oe->type = type;
2986 oe->pack_id = MAX_PACK_ID;
2987 oe->idx.offset = 1;
2989 switch (oe->type) {
2990 case OBJ_TREE: /* easy case. */
2991 return oe;
2992 case OBJ_COMMIT:
2993 case OBJ_TAG:
2994 break;
2995 default:
2996 die("Not a tree-ish: %s", command_buf.buf);
2999 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3000 buf = gfi_unpack_entry(oe, &size);
3001 } else {
3002 enum object_type unused;
3003 buf = read_sha1_file(sha1, &unused, &size);
3005 if (!buf)
3006 die("Can't load object %s", sha1_to_hex(sha1));
3008 /* Peel one layer. */
3009 switch (oe->type) {
3010 case OBJ_TAG:
3011 if (size < 40 + strlen("object ") ||
3012 get_sha1_hex(buf + strlen("object "), sha1))
3013 die("Invalid SHA1 in tag: %s", command_buf.buf);
3014 break;
3015 case OBJ_COMMIT:
3016 if (size < 40 + strlen("tree ") ||
3017 get_sha1_hex(buf + strlen("tree "), sha1))
3018 die("Invalid SHA1 in commit: %s", command_buf.buf);
3021 free(buf);
3022 return find_object(sha1);
3025 static struct object_entry *parse_treeish_dataref(const char **p)
3027 unsigned char sha1[20];
3028 struct object_entry *e;
3030 if (**p == ':') { /* <mark> */
3031 e = find_mark(parse_mark_ref_space(p));
3032 if (!e)
3033 die("Unknown mark: %s", command_buf.buf);
3034 hashcpy(sha1, e->idx.sha1);
3035 } else { /* <sha1> */
3036 if (get_sha1_hex(*p, sha1))
3037 die("Invalid dataref: %s", command_buf.buf);
3038 e = find_object(sha1);
3039 *p += 40;
3040 if (*(*p)++ != ' ')
3041 die("Missing space after tree-ish: %s", command_buf.buf);
3044 while (!e || e->type != OBJ_TREE)
3045 e = dereference(e, sha1);
3046 return e;
3049 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3051 static struct strbuf line = STRBUF_INIT;
3053 /* See show_tree(). */
3054 const char *type =
3055 S_ISGITLINK(mode) ? commit_type :
3056 S_ISDIR(mode) ? tree_type :
3057 blob_type;
3059 if (!mode) {
3060 /* missing SP path LF */
3061 strbuf_reset(&line);
3062 strbuf_addstr(&line, "missing ");
3063 quote_c_style(path, &line, NULL, 0);
3064 strbuf_addch(&line, '\n');
3065 } else {
3066 /* mode SP type SP object_name TAB path LF */
3067 strbuf_reset(&line);
3068 strbuf_addf(&line, "%06o %s %s\t",
3069 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3070 quote_c_style(path, &line, NULL, 0);
3071 strbuf_addch(&line, '\n');
3073 cat_blob_write(line.buf, line.len);
3076 static void parse_ls(const char *p, struct branch *b)
3078 struct tree_entry *root = NULL;
3079 struct tree_entry leaf = {NULL};
3081 /* ls SP (<tree-ish> SP)? <path> */
3082 if (*p == '"') {
3083 if (!b)
3084 die("Not in a commit: %s", command_buf.buf);
3085 root = &b->branch_tree;
3086 } else {
3087 struct object_entry *e = parse_treeish_dataref(&p);
3088 root = new_tree_entry();
3089 hashcpy(root->versions[1].sha1, e->idx.sha1);
3090 if (!is_null_sha1(root->versions[1].sha1))
3091 root->versions[1].mode = S_IFDIR;
3092 load_tree(root);
3094 if (*p == '"') {
3095 static struct strbuf uq = STRBUF_INIT;
3096 const char *endp;
3097 strbuf_reset(&uq);
3098 if (unquote_c_style(&uq, p, &endp))
3099 die("Invalid path: %s", command_buf.buf);
3100 if (*endp)
3101 die("Garbage after path in: %s", command_buf.buf);
3102 p = uq.buf;
3104 tree_content_get(root, p, &leaf, 1);
3106 * A directory in preparation would have a sha1 of zero
3107 * until it is saved. Save, for simplicity.
3109 if (S_ISDIR(leaf.versions[1].mode))
3110 store_tree(&leaf);
3112 print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3113 if (leaf.tree)
3114 release_tree_content_recursive(leaf.tree);
3115 if (!b || root != &b->branch_tree)
3116 release_tree_entry(root);
3119 static void checkpoint(void)
3121 checkpoint_requested = 0;
3122 if (object_count) {
3123 cycle_packfile();
3124 dump_branches();
3125 dump_tags();
3126 dump_marks();
3130 static void parse_checkpoint(void)
3132 checkpoint_requested = 1;
3133 skip_optional_lf();
3136 static void parse_progress(void)
3138 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3139 fputc('\n', stdout);
3140 fflush(stdout);
3141 skip_optional_lf();
3144 static char* make_fast_import_path(const char *path)
3146 if (!relative_marks_paths || is_absolute_path(path))
3147 return xstrdup(path);
3148 return xstrdup(git_path("info/fast-import/%s", path));
3151 static void option_import_marks(const char *marks,
3152 int from_stream, int ignore_missing)
3154 if (import_marks_file) {
3155 if (from_stream)
3156 die("Only one import-marks command allowed per stream");
3158 /* read previous mark file */
3159 if(!import_marks_file_from_stream)
3160 read_marks();
3163 import_marks_file = make_fast_import_path(marks);
3164 safe_create_leading_directories_const(import_marks_file);
3165 import_marks_file_from_stream = from_stream;
3166 import_marks_file_ignore_missing = ignore_missing;
3169 static void option_date_format(const char *fmt)
3171 if (!strcmp(fmt, "raw"))
3172 whenspec = WHENSPEC_RAW;
3173 else if (!strcmp(fmt, "rfc2822"))
3174 whenspec = WHENSPEC_RFC2822;
3175 else if (!strcmp(fmt, "now"))
3176 whenspec = WHENSPEC_NOW;
3177 else
3178 die("unknown --date-format argument %s", fmt);
3181 static unsigned long ulong_arg(const char *option, const char *arg)
3183 char *endptr;
3184 unsigned long rv = strtoul(arg, &endptr, 0);
3185 if (strchr(arg, '-') || endptr == arg || *endptr)
3186 die("%s: argument must be a non-negative integer", option);
3187 return rv;
3190 static void option_depth(const char *depth)
3192 max_depth = ulong_arg("--depth", depth);
3193 if (max_depth > MAX_DEPTH)
3194 die("--depth cannot exceed %u", MAX_DEPTH);
3197 static void option_active_branches(const char *branches)
3199 max_active_branches = ulong_arg("--active-branches", branches);
3202 static void option_export_marks(const char *marks)
3204 export_marks_file = make_fast_import_path(marks);
3205 safe_create_leading_directories_const(export_marks_file);
3208 static void option_cat_blob_fd(const char *fd)
3210 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3211 if (n > (unsigned long) INT_MAX)
3212 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3213 cat_blob_fd = (int) n;
3216 static void option_export_pack_edges(const char *edges)
3218 if (pack_edges)
3219 fclose(pack_edges);
3220 pack_edges = fopen(edges, "a");
3221 if (!pack_edges)
3222 die_errno("Cannot open '%s'", edges);
3225 static int parse_one_option(const char *option)
3227 if (skip_prefix(option, "max-pack-size=", &option)) {
3228 unsigned long v;
3229 if (!git_parse_ulong(option, &v))
3230 return 0;
3231 if (v < 8192) {
3232 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3233 v *= 1024 * 1024;
3234 } else if (v < 1024 * 1024) {
3235 warning("minimum max-pack-size is 1 MiB");
3236 v = 1024 * 1024;
3238 max_packsize = v;
3239 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3240 unsigned long v;
3241 if (!git_parse_ulong(option, &v))
3242 return 0;
3243 big_file_threshold = v;
3244 } else if (skip_prefix(option, "depth=", &option)) {
3245 option_depth(option);
3246 } else if (skip_prefix(option, "active-branches=", &option)) {
3247 option_active_branches(option);
3248 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3249 option_export_pack_edges(option);
3250 } else if (starts_with(option, "quiet")) {
3251 show_stats = 0;
3252 } else if (starts_with(option, "stats")) {
3253 show_stats = 1;
3254 } else {
3255 return 0;
3258 return 1;
3261 static int parse_one_feature(const char *feature, int from_stream)
3263 const char *arg;
3265 if (skip_prefix(feature, "date-format=", &arg)) {
3266 option_date_format(arg);
3267 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3268 option_import_marks(arg, from_stream, 0);
3269 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3270 option_import_marks(arg, from_stream, 1);
3271 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3272 option_export_marks(arg);
3273 } else if (!strcmp(feature, "get-mark")) {
3274 ; /* Don't die - this feature is supported */
3275 } else if (!strcmp(feature, "cat-blob")) {
3276 ; /* Don't die - this feature is supported */
3277 } else if (!strcmp(feature, "relative-marks")) {
3278 relative_marks_paths = 1;
3279 } else if (!strcmp(feature, "no-relative-marks")) {
3280 relative_marks_paths = 0;
3281 } else if (!strcmp(feature, "done")) {
3282 require_explicit_termination = 1;
3283 } else if (!strcmp(feature, "force")) {
3284 force_update = 1;
3285 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3286 ; /* do nothing; we have the feature */
3287 } else {
3288 return 0;
3291 return 1;
3294 static void parse_feature(const char *feature)
3296 if (seen_data_command)
3297 die("Got feature command '%s' after data command", feature);
3299 if (parse_one_feature(feature, 1))
3300 return;
3302 die("This version of fast-import does not support feature %s.", feature);
3305 static void parse_option(const char *option)
3307 if (seen_data_command)
3308 die("Got option command '%s' after data command", option);
3310 if (parse_one_option(option))
3311 return;
3313 die("This version of fast-import does not support option: %s", option);
3316 static void git_pack_config(void)
3318 int indexversion_value;
3319 unsigned long packsizelimit_value;
3321 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3322 if (max_depth > MAX_DEPTH)
3323 max_depth = MAX_DEPTH;
3325 if (!git_config_get_int("pack.compression", &pack_compression_level)) {
3326 if (pack_compression_level == -1)
3327 pack_compression_level = Z_DEFAULT_COMPRESSION;
3328 else if (pack_compression_level < 0 ||
3329 pack_compression_level > Z_BEST_COMPRESSION)
3330 git_die_config("pack.compression",
3331 "bad pack compression level %d", pack_compression_level);
3332 pack_compression_seen = 1;
3334 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3335 pack_idx_opts.version = indexversion_value;
3336 if (pack_idx_opts.version > 2)
3337 git_die_config("pack.indexversion",
3338 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3340 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3341 max_packsize = packsizelimit_value;
3343 git_config(git_default_config, NULL);
3346 static const char fast_import_usage[] =
3347 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3349 static void parse_argv(void)
3351 unsigned int i;
3353 for (i = 1; i < global_argc; i++) {
3354 const char *a = global_argv[i];
3356 if (*a != '-' || !strcmp(a, "--"))
3357 break;
3359 if (!skip_prefix(a, "--", &a))
3360 die("unknown option %s", a);
3362 if (parse_one_option(a))
3363 continue;
3365 if (parse_one_feature(a, 0))
3366 continue;
3368 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3369 option_cat_blob_fd(a);
3370 continue;
3373 die("unknown option --%s", a);
3375 if (i != global_argc)
3376 usage(fast_import_usage);
3378 seen_data_command = 1;
3379 if (import_marks_file)
3380 read_marks();
3383 int main(int argc, char **argv)
3385 unsigned int i;
3387 git_extract_argv0_path(argv[0]);
3389 git_setup_gettext();
3391 if (argc == 2 && !strcmp(argv[1], "-h"))
3392 usage(fast_import_usage);
3394 setup_git_directory();
3395 reset_pack_idx_option(&pack_idx_opts);
3396 git_pack_config();
3397 if (!pack_compression_seen && core_compression_seen)
3398 pack_compression_level = core_compression_level;
3400 alloc_objects(object_entry_alloc);
3401 strbuf_init(&command_buf, 0);
3402 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3403 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3404 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3405 marks = pool_calloc(1, sizeof(struct mark_set));
3407 global_argc = argc;
3408 global_argv = argv;
3410 rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3411 for (i = 0; i < (cmd_save - 1); i++)
3412 rc_free[i].next = &rc_free[i + 1];
3413 rc_free[cmd_save - 1].next = NULL;
3415 prepare_packed_git();
3416 start_packfile();
3417 set_die_routine(die_nicely);
3418 set_checkpoint_signal();
3419 while (read_next_command() != EOF) {
3420 const char *v;
3421 if (!strcmp("blob", command_buf.buf))
3422 parse_new_blob();
3423 else if (skip_prefix(command_buf.buf, "ls ", &v))
3424 parse_ls(v, NULL);
3425 else if (skip_prefix(command_buf.buf, "commit ", &v))
3426 parse_new_commit(v);
3427 else if (skip_prefix(command_buf.buf, "tag ", &v))
3428 parse_new_tag(v);
3429 else if (skip_prefix(command_buf.buf, "reset ", &v))
3430 parse_reset_branch(v);
3431 else if (!strcmp("checkpoint", command_buf.buf))
3432 parse_checkpoint();
3433 else if (!strcmp("done", command_buf.buf))
3434 break;
3435 else if (starts_with(command_buf.buf, "progress "))
3436 parse_progress();
3437 else if (skip_prefix(command_buf.buf, "feature ", &v))
3438 parse_feature(v);
3439 else if (skip_prefix(command_buf.buf, "option git ", &v))
3440 parse_option(v);
3441 else if (starts_with(command_buf.buf, "option "))
3442 /* ignore non-git options*/;
3443 else
3444 die("Unsupported command: %s", command_buf.buf);
3446 if (checkpoint_requested)
3447 checkpoint();
3450 /* argv hasn't been parsed yet, do so */
3451 if (!seen_data_command)
3452 parse_argv();
3454 if (require_explicit_termination && feof(stdin))
3455 die("stream ends early");
3457 end_packfile();
3459 dump_branches();
3460 dump_tags();
3461 unkeep_all_packs();
3462 dump_marks();
3464 if (pack_edges)
3465 fclose(pack_edges);
3467 if (show_stats) {
3468 uintmax_t total_count = 0, duplicate_count = 0;
3469 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3470 total_count += object_count_by_type[i];
3471 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3472 duplicate_count += duplicate_count_by_type[i];
3474 fprintf(stderr, "%s statistics:\n", argv[0]);
3475 fprintf(stderr, "---------------------------------------------------------------------\n");
3476 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3477 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3478 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]);
3479 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]);
3480 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]);
3481 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]);
3482 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3483 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3484 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3485 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3486 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)(total_allocd/1024));
3487 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3488 fprintf(stderr, "---------------------------------------------------------------------\n");
3489 pack_report();
3490 fprintf(stderr, "---------------------------------------------------------------------\n");
3491 fprintf(stderr, "\n");
3494 return failure ? 1 : 0;