Merge branch 'sg/completion-clear-cached'
[git/debian.git] / fast-import.c
blob34edf3fb8f9012bf2fcaf231604ebb490e66940c
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 "repository.h"
158 #include "config.h"
159 #include "lockfile.h"
160 #include "object.h"
161 #include "blob.h"
162 #include "tree.h"
163 #include "commit.h"
164 #include "delta.h"
165 #include "pack.h"
166 #include "refs.h"
167 #include "csum-file.h"
168 #include "quote.h"
169 #include "dir.h"
170 #include "run-command.h"
171 #include "packfile.h"
172 #include "object-store.h"
173 #include "mem-pool.h"
175 #define PACK_ID_BITS 16
176 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
177 #define DEPTH_BITS 13
178 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
181 * We abuse the setuid bit on directories to mean "do not delta".
183 #define NO_DELTA S_ISUID
185 struct object_entry {
186 struct pack_idx_entry idx;
187 struct object_entry *next;
188 uint32_t type : TYPE_BITS,
189 pack_id : PACK_ID_BITS,
190 depth : DEPTH_BITS;
193 struct object_entry_pool {
194 struct object_entry_pool *next_pool;
195 struct object_entry *next_free;
196 struct object_entry *end;
197 struct object_entry entries[FLEX_ARRAY]; /* more */
200 struct mark_set {
201 union {
202 struct object_entry *marked[1024];
203 struct mark_set *sets[1024];
204 } data;
205 unsigned int shift;
208 struct last_object {
209 struct strbuf data;
210 off_t offset;
211 unsigned int depth;
212 unsigned no_swap : 1;
215 struct atom_str {
216 struct atom_str *next_atom;
217 unsigned short str_len;
218 char str_dat[FLEX_ARRAY]; /* more */
221 struct tree_content;
222 struct tree_entry {
223 struct tree_content *tree;
224 struct atom_str *name;
225 struct tree_entry_ms {
226 uint16_t mode;
227 struct object_id oid;
228 } versions[2];
231 struct tree_content {
232 unsigned int entry_capacity; /* must match avail_tree_content */
233 unsigned int entry_count;
234 unsigned int delta_depth;
235 struct tree_entry *entries[FLEX_ARRAY]; /* more */
238 struct avail_tree_content {
239 unsigned int entry_capacity; /* must match tree_content */
240 struct avail_tree_content *next_avail;
243 struct branch {
244 struct branch *table_next_branch;
245 struct branch *active_next_branch;
246 const char *name;
247 struct tree_entry branch_tree;
248 uintmax_t last_commit;
249 uintmax_t num_notes;
250 unsigned active : 1;
251 unsigned delete : 1;
252 unsigned pack_id : PACK_ID_BITS;
253 struct object_id oid;
256 struct tag {
257 struct tag *next_tag;
258 const char *name;
259 unsigned int pack_id;
260 struct object_id oid;
263 struct hash_list {
264 struct hash_list *next;
265 struct object_id oid;
268 typedef enum {
269 WHENSPEC_RAW = 1,
270 WHENSPEC_RFC2822,
271 WHENSPEC_NOW
272 } whenspec_type;
274 struct recent_command {
275 struct recent_command *prev;
276 struct recent_command *next;
277 char *buf;
280 /* Configured limits on output */
281 static unsigned long max_depth = 50;
282 static off_t max_packsize;
283 static int unpack_limit = 100;
284 static int force_update;
286 /* Stats and misc. counters */
287 static uintmax_t alloc_count;
288 static uintmax_t marks_set_count;
289 static uintmax_t object_count_by_type[1 << TYPE_BITS];
290 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
291 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
292 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
293 static unsigned long object_count;
294 static unsigned long branch_count;
295 static unsigned long branch_load_count;
296 static int failure;
297 static FILE *pack_edges;
298 static unsigned int show_stats = 1;
299 static int global_argc;
300 static const char **global_argv;
302 /* Memory pools */
303 static struct mem_pool fi_mem_pool = {NULL, 2*1024*1024 -
304 sizeof(struct mp_block), 0 };
306 /* Atom management */
307 static unsigned int atom_table_sz = 4451;
308 static unsigned int atom_cnt;
309 static struct atom_str **atom_table;
311 /* The .pack file being generated */
312 static struct pack_idx_option pack_idx_opts;
313 static unsigned int pack_id;
314 static struct hashfile *pack_file;
315 static struct packed_git *pack_data;
316 static struct packed_git **all_packs;
317 static off_t pack_size;
319 /* Table of objects we've written. */
320 static unsigned int object_entry_alloc = 5000;
321 static struct object_entry_pool *blocks;
322 static struct object_entry *object_table[1 << 16];
323 static struct mark_set *marks;
324 static const char *export_marks_file;
325 static const char *import_marks_file;
326 static int import_marks_file_from_stream;
327 static int import_marks_file_ignore_missing;
328 static int import_marks_file_done;
329 static int relative_marks_paths;
331 /* Our last blob */
332 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
334 /* Tree management */
335 static unsigned int tree_entry_alloc = 1000;
336 static void *avail_tree_entry;
337 static unsigned int avail_tree_table_sz = 100;
338 static struct avail_tree_content **avail_tree_table;
339 static size_t tree_entry_allocd;
340 static struct strbuf old_tree = STRBUF_INIT;
341 static struct strbuf new_tree = STRBUF_INIT;
343 /* Branch data */
344 static unsigned long max_active_branches = 5;
345 static unsigned long cur_active_branches;
346 static unsigned long branch_table_sz = 1039;
347 static struct branch **branch_table;
348 static struct branch *active_branches;
350 /* Tag data */
351 static struct tag *first_tag;
352 static struct tag *last_tag;
354 /* Input stream parsing */
355 static whenspec_type whenspec = WHENSPEC_RAW;
356 static struct strbuf command_buf = STRBUF_INIT;
357 static int unread_command_buf;
358 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
359 static struct recent_command *cmd_tail = &cmd_hist;
360 static struct recent_command *rc_free;
361 static unsigned int cmd_save = 100;
362 static uintmax_t next_mark;
363 static struct strbuf new_data = STRBUF_INIT;
364 static int seen_data_command;
365 static int require_explicit_termination;
367 /* Signal handling */
368 static volatile sig_atomic_t checkpoint_requested;
370 /* Where to write output of cat-blob commands */
371 static int cat_blob_fd = STDOUT_FILENO;
373 static void parse_argv(void);
374 static void parse_get_mark(const char *p);
375 static void parse_cat_blob(const char *p);
376 static void parse_ls(const char *p, struct branch *b);
378 static void write_branch_report(FILE *rpt, struct branch *b)
380 fprintf(rpt, "%s:\n", b->name);
382 fprintf(rpt, " status :");
383 if (b->active)
384 fputs(" active", rpt);
385 if (b->branch_tree.tree)
386 fputs(" loaded", rpt);
387 if (is_null_oid(&b->branch_tree.versions[1].oid))
388 fputs(" dirty", rpt);
389 fputc('\n', rpt);
391 fprintf(rpt, " tip commit : %s\n", oid_to_hex(&b->oid));
392 fprintf(rpt, " old tree : %s\n",
393 oid_to_hex(&b->branch_tree.versions[0].oid));
394 fprintf(rpt, " cur tree : %s\n",
395 oid_to_hex(&b->branch_tree.versions[1].oid));
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_errno("can't write crash report %s", loc);
418 free(loc);
419 return;
422 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
424 fprintf(rpt, "fast-import crash report:\n");
425 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
426 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
427 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
428 fputc('\n', rpt);
430 fputs("fatal: ", rpt);
431 fputs(err, rpt);
432 fputc('\n', rpt);
434 fputc('\n', rpt);
435 fputs("Most Recent Commands Before Crash\n", rpt);
436 fputs("---------------------------------\n", rpt);
437 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
438 if (rc->next == &cmd_hist)
439 fputs("* ", rpt);
440 else
441 fputs(" ", rpt);
442 fputs(rc->buf, rpt);
443 fputc('\n', rpt);
446 fputc('\n', rpt);
447 fputs("Active Branch LRU\n", rpt);
448 fputs("-----------------\n", rpt);
449 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
450 cur_active_branches,
451 max_active_branches);
452 fputc('\n', rpt);
453 fputs(" pos clock name\n", rpt);
454 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
455 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
456 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
457 ++lu, b->last_commit, b->name);
459 fputc('\n', rpt);
460 fputs("Inactive Branches\n", rpt);
461 fputs("-----------------\n", rpt);
462 for (lu = 0; lu < branch_table_sz; lu++) {
463 for (b = branch_table[lu]; b; b = b->table_next_branch)
464 write_branch_report(rpt, b);
467 if (first_tag) {
468 struct tag *tg;
469 fputc('\n', rpt);
470 fputs("Annotated Tags\n", rpt);
471 fputs("--------------\n", rpt);
472 for (tg = first_tag; tg; tg = tg->next_tag) {
473 fputs(oid_to_hex(&tg->oid), 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(struct object_id *oid)
560 struct object_entry *e;
562 if (blocks->next_free == blocks->end)
563 alloc_objects(object_entry_alloc);
565 e = blocks->next_free++;
566 oidcpy(&e->idx.oid, oid);
567 return e;
570 static struct object_entry *find_object(struct object_id *oid)
572 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
573 struct object_entry *e;
574 for (e = object_table[h]; e; e = e->next)
575 if (!oidcmp(oid, &e->idx.oid))
576 return e;
577 return NULL;
580 static struct object_entry *insert_object(struct object_id *oid)
582 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
583 struct object_entry *e = object_table[h];
585 while (e) {
586 if (!oidcmp(oid, &e->idx.oid))
587 return e;
588 e = e->next;
591 e = new_object(oid);
592 e->next = object_table[h];
593 e->idx.offset = 0;
594 object_table[h] = e;
595 return e;
598 static void invalidate_pack_id(unsigned int id)
600 unsigned int h;
601 unsigned long lu;
602 struct tag *t;
604 for (h = 0; h < ARRAY_SIZE(object_table); h++) {
605 struct object_entry *e;
607 for (e = object_table[h]; e; e = e->next)
608 if (e->pack_id == id)
609 e->pack_id = MAX_PACK_ID;
612 for (lu = 0; lu < branch_table_sz; lu++) {
613 struct branch *b;
615 for (b = branch_table[lu]; b; b = b->table_next_branch)
616 if (b->pack_id == id)
617 b->pack_id = MAX_PACK_ID;
620 for (t = first_tag; t; t = t->next_tag)
621 if (t->pack_id == id)
622 t->pack_id = MAX_PACK_ID;
625 static unsigned int hc_str(const char *s, size_t len)
627 unsigned int r = 0;
628 while (len-- > 0)
629 r = r * 31 + *s++;
630 return r;
633 static char *pool_strdup(const char *s)
635 size_t len = strlen(s) + 1;
636 char *r = mem_pool_alloc(&fi_mem_pool, len);
637 memcpy(r, s, len);
638 return r;
641 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
643 struct mark_set *s = marks;
644 while ((idnum >> s->shift) >= 1024) {
645 s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
646 s->shift = marks->shift + 10;
647 s->data.sets[0] = marks;
648 marks = s;
650 while (s->shift) {
651 uintmax_t i = idnum >> s->shift;
652 idnum -= i << s->shift;
653 if (!s->data.sets[i]) {
654 s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
655 s->data.sets[i]->shift = s->shift - 10;
657 s = s->data.sets[i];
659 if (!s->data.marked[idnum])
660 marks_set_count++;
661 s->data.marked[idnum] = oe;
664 static struct object_entry *find_mark(uintmax_t idnum)
666 uintmax_t orig_idnum = idnum;
667 struct mark_set *s = marks;
668 struct object_entry *oe = NULL;
669 if ((idnum >> s->shift) < 1024) {
670 while (s && s->shift) {
671 uintmax_t i = idnum >> s->shift;
672 idnum -= i << s->shift;
673 s = s->data.sets[i];
675 if (s)
676 oe = s->data.marked[idnum];
678 if (!oe)
679 die("mark :%" PRIuMAX " not declared", orig_idnum);
680 return oe;
683 static struct atom_str *to_atom(const char *s, unsigned short len)
685 unsigned int hc = hc_str(s, len) % atom_table_sz;
686 struct atom_str *c;
688 for (c = atom_table[hc]; c; c = c->next_atom)
689 if (c->str_len == len && !strncmp(s, c->str_dat, len))
690 return c;
692 c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
693 c->str_len = len;
694 memcpy(c->str_dat, s, len);
695 c->str_dat[len] = 0;
696 c->next_atom = atom_table[hc];
697 atom_table[hc] = c;
698 atom_cnt++;
699 return c;
702 static struct branch *lookup_branch(const char *name)
704 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
705 struct branch *b;
707 for (b = branch_table[hc]; b; b = b->table_next_branch)
708 if (!strcmp(name, b->name))
709 return b;
710 return NULL;
713 static struct branch *new_branch(const char *name)
715 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
716 struct branch *b = lookup_branch(name);
718 if (b)
719 die("Invalid attempt to create duplicate branch: %s", name);
720 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
721 die("Branch name doesn't conform to GIT standards: %s", name);
723 b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
724 b->name = pool_strdup(name);
725 b->table_next_branch = branch_table[hc];
726 b->branch_tree.versions[0].mode = S_IFDIR;
727 b->branch_tree.versions[1].mode = S_IFDIR;
728 b->num_notes = 0;
729 b->active = 0;
730 b->pack_id = MAX_PACK_ID;
731 branch_table[hc] = b;
732 branch_count++;
733 return b;
736 static unsigned int hc_entries(unsigned int cnt)
738 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
739 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
742 static struct tree_content *new_tree_content(unsigned int cnt)
744 struct avail_tree_content *f, *l = NULL;
745 struct tree_content *t;
746 unsigned int hc = hc_entries(cnt);
748 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
749 if (f->entry_capacity >= cnt)
750 break;
752 if (f) {
753 if (l)
754 l->next_avail = f->next_avail;
755 else
756 avail_tree_table[hc] = f->next_avail;
757 } else {
758 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
759 f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
760 f->entry_capacity = cnt;
763 t = (struct tree_content*)f;
764 t->entry_count = 0;
765 t->delta_depth = 0;
766 return t;
769 static void release_tree_entry(struct tree_entry *e);
770 static void release_tree_content(struct tree_content *t)
772 struct avail_tree_content *f = (struct avail_tree_content*)t;
773 unsigned int hc = hc_entries(f->entry_capacity);
774 f->next_avail = avail_tree_table[hc];
775 avail_tree_table[hc] = f;
778 static void release_tree_content_recursive(struct tree_content *t)
780 unsigned int i;
781 for (i = 0; i < t->entry_count; i++)
782 release_tree_entry(t->entries[i]);
783 release_tree_content(t);
786 static struct tree_content *grow_tree_content(
787 struct tree_content *t,
788 int amt)
790 struct tree_content *r = new_tree_content(t->entry_count + amt);
791 r->entry_count = t->entry_count;
792 r->delta_depth = t->delta_depth;
793 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
794 release_tree_content(t);
795 return r;
798 static struct tree_entry *new_tree_entry(void)
800 struct tree_entry *e;
802 if (!avail_tree_entry) {
803 unsigned int n = tree_entry_alloc;
804 tree_entry_allocd += n * sizeof(struct tree_entry);
805 ALLOC_ARRAY(e, n);
806 avail_tree_entry = e;
807 while (n-- > 1) {
808 *((void**)e) = e + 1;
809 e++;
811 *((void**)e) = NULL;
814 e = avail_tree_entry;
815 avail_tree_entry = *((void**)e);
816 return e;
819 static void release_tree_entry(struct tree_entry *e)
821 if (e->tree)
822 release_tree_content_recursive(e->tree);
823 *((void**)e) = avail_tree_entry;
824 avail_tree_entry = e;
827 static struct tree_content *dup_tree_content(struct tree_content *s)
829 struct tree_content *d;
830 struct tree_entry *a, *b;
831 unsigned int i;
833 if (!s)
834 return NULL;
835 d = new_tree_content(s->entry_count);
836 for (i = 0; i < s->entry_count; i++) {
837 a = s->entries[i];
838 b = new_tree_entry();
839 memcpy(b, a, sizeof(*a));
840 if (a->tree && is_null_oid(&b->versions[1].oid))
841 b->tree = dup_tree_content(a->tree);
842 else
843 b->tree = NULL;
844 d->entries[i] = b;
846 d->entry_count = s->entry_count;
847 d->delta_depth = s->delta_depth;
849 return d;
852 static void start_packfile(void)
854 struct strbuf tmp_file = STRBUF_INIT;
855 struct packed_git *p;
856 struct pack_header hdr;
857 int pack_fd;
859 pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
860 FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
861 strbuf_release(&tmp_file);
863 p->pack_fd = pack_fd;
864 p->do_not_close = 1;
865 pack_file = hashfd(pack_fd, p->pack_name);
867 hdr.hdr_signature = htonl(PACK_SIGNATURE);
868 hdr.hdr_version = htonl(2);
869 hdr.hdr_entries = 0;
870 hashwrite(pack_file, &hdr, sizeof(hdr));
872 pack_data = p;
873 pack_size = sizeof(hdr);
874 object_count = 0;
876 REALLOC_ARRAY(all_packs, pack_id + 1);
877 all_packs[pack_id] = p;
880 static const char *create_index(void)
882 const char *tmpfile;
883 struct pack_idx_entry **idx, **c, **last;
884 struct object_entry *e;
885 struct object_entry_pool *o;
887 /* Build the table of object IDs. */
888 ALLOC_ARRAY(idx, object_count);
889 c = idx;
890 for (o = blocks; o; o = o->next_pool)
891 for (e = o->next_free; e-- != o->entries;)
892 if (pack_id == e->pack_id)
893 *c++ = &e->idx;
894 last = idx + object_count;
895 if (c != last)
896 die("internal consistency error creating the index");
898 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
899 free(idx);
900 return tmpfile;
903 static char *keep_pack(const char *curr_index_name)
905 static const char *keep_msg = "fast-import";
906 struct strbuf name = STRBUF_INIT;
907 int keep_fd;
909 odb_pack_name(&name, pack_data->sha1, "keep");
910 keep_fd = odb_pack_keep(name.buf);
911 if (keep_fd < 0)
912 die_errno("cannot create keep file");
913 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
914 if (close(keep_fd))
915 die_errno("failed to write keep file");
917 odb_pack_name(&name, pack_data->sha1, "pack");
918 if (finalize_object_file(pack_data->pack_name, name.buf))
919 die("cannot store pack file");
921 odb_pack_name(&name, pack_data->sha1, "idx");
922 if (finalize_object_file(curr_index_name, name.buf))
923 die("cannot store index file");
924 free((void *)curr_index_name);
925 return strbuf_detach(&name, NULL);
928 static void unkeep_all_packs(void)
930 struct strbuf name = STRBUF_INIT;
931 int k;
933 for (k = 0; k < pack_id; k++) {
934 struct packed_git *p = all_packs[k];
935 odb_pack_name(&name, p->sha1, "keep");
936 unlink_or_warn(name.buf);
938 strbuf_release(&name);
941 static int loosen_small_pack(const struct packed_git *p)
943 struct child_process unpack = CHILD_PROCESS_INIT;
945 if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
946 die_errno("Failed seeking to start of '%s'", p->pack_name);
948 unpack.in = p->pack_fd;
949 unpack.git_cmd = 1;
950 unpack.stdout_to_stderr = 1;
951 argv_array_push(&unpack.args, "unpack-objects");
952 if (!show_stats)
953 argv_array_push(&unpack.args, "-q");
955 return run_command(&unpack);
958 static void end_packfile(void)
960 static int running;
962 if (running || !pack_data)
963 return;
965 running = 1;
966 clear_delta_base_cache();
967 if (object_count) {
968 struct packed_git *new_p;
969 struct object_id cur_pack_oid;
970 char *idx_name;
971 int i;
972 struct branch *b;
973 struct tag *t;
975 close_pack_windows(pack_data);
976 finalize_hashfile(pack_file, cur_pack_oid.hash, 0);
977 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
978 pack_data->pack_name, object_count,
979 cur_pack_oid.hash, pack_size);
981 if (object_count <= unpack_limit) {
982 if (!loosen_small_pack(pack_data)) {
983 invalidate_pack_id(pack_id);
984 goto discard_pack;
988 close(pack_data->pack_fd);
989 idx_name = keep_pack(create_index());
991 /* Register the packfile with core git's machinery. */
992 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
993 if (!new_p)
994 die("core git rejected index %s", idx_name);
995 all_packs[pack_id] = new_p;
996 install_packed_git(the_repository, new_p);
997 free(idx_name);
999 /* Print the boundary */
1000 if (pack_edges) {
1001 fprintf(pack_edges, "%s:", new_p->pack_name);
1002 for (i = 0; i < branch_table_sz; i++) {
1003 for (b = branch_table[i]; b; b = b->table_next_branch) {
1004 if (b->pack_id == pack_id)
1005 fprintf(pack_edges, " %s",
1006 oid_to_hex(&b->oid));
1009 for (t = first_tag; t; t = t->next_tag) {
1010 if (t->pack_id == pack_id)
1011 fprintf(pack_edges, " %s",
1012 oid_to_hex(&t->oid));
1014 fputc('\n', pack_edges);
1015 fflush(pack_edges);
1018 pack_id++;
1020 else {
1021 discard_pack:
1022 close(pack_data->pack_fd);
1023 unlink_or_warn(pack_data->pack_name);
1025 FREE_AND_NULL(pack_data);
1026 running = 0;
1028 /* We can't carry a delta across packfiles. */
1029 strbuf_release(&last_blob.data);
1030 last_blob.offset = 0;
1031 last_blob.depth = 0;
1034 static void cycle_packfile(void)
1036 end_packfile();
1037 start_packfile();
1040 static int store_object(
1041 enum object_type type,
1042 struct strbuf *dat,
1043 struct last_object *last,
1044 struct object_id *oidout,
1045 uintmax_t mark)
1047 void *out, *delta;
1048 struct object_entry *e;
1049 unsigned char hdr[96];
1050 struct object_id oid;
1051 unsigned long hdrlen, deltalen;
1052 git_hash_ctx c;
1053 git_zstream s;
1055 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1056 type_name(type), (unsigned long)dat->len) + 1;
1057 the_hash_algo->init_fn(&c);
1058 the_hash_algo->update_fn(&c, hdr, hdrlen);
1059 the_hash_algo->update_fn(&c, dat->buf, dat->len);
1060 the_hash_algo->final_fn(oid.hash, &c);
1061 if (oidout)
1062 oidcpy(oidout, &oid);
1064 e = insert_object(&oid);
1065 if (mark)
1066 insert_mark(mark, e);
1067 if (e->idx.offset) {
1068 duplicate_count_by_type[type]++;
1069 return 1;
1070 } else if (find_sha1_pack(oid.hash,
1071 get_packed_git(the_repository))) {
1072 e->type = type;
1073 e->pack_id = MAX_PACK_ID;
1074 e->idx.offset = 1; /* just not zero! */
1075 duplicate_count_by_type[type]++;
1076 return 1;
1079 if (last && last->data.buf && last->depth < max_depth
1080 && dat->len > the_hash_algo->rawsz) {
1082 delta_count_attempts_by_type[type]++;
1083 delta = diff_delta(last->data.buf, last->data.len,
1084 dat->buf, dat->len,
1085 &deltalen, dat->len - the_hash_algo->rawsz);
1086 } else
1087 delta = NULL;
1089 git_deflate_init(&s, pack_compression_level);
1090 if (delta) {
1091 s.next_in = delta;
1092 s.avail_in = deltalen;
1093 } else {
1094 s.next_in = (void *)dat->buf;
1095 s.avail_in = dat->len;
1097 s.avail_out = git_deflate_bound(&s, s.avail_in);
1098 s.next_out = out = xmalloc(s.avail_out);
1099 while (git_deflate(&s, Z_FINISH) == Z_OK)
1100 ; /* nothing */
1101 git_deflate_end(&s);
1103 /* Determine if we should auto-checkpoint. */
1104 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1105 || (pack_size + 60 + s.total_out) < pack_size) {
1107 /* This new object needs to *not* have the current pack_id. */
1108 e->pack_id = pack_id + 1;
1109 cycle_packfile();
1111 /* We cannot carry a delta into the new pack. */
1112 if (delta) {
1113 FREE_AND_NULL(delta);
1115 git_deflate_init(&s, pack_compression_level);
1116 s.next_in = (void *)dat->buf;
1117 s.avail_in = dat->len;
1118 s.avail_out = git_deflate_bound(&s, s.avail_in);
1119 s.next_out = out = xrealloc(out, s.avail_out);
1120 while (git_deflate(&s, Z_FINISH) == Z_OK)
1121 ; /* nothing */
1122 git_deflate_end(&s);
1126 e->type = type;
1127 e->pack_id = pack_id;
1128 e->idx.offset = pack_size;
1129 object_count++;
1130 object_count_by_type[type]++;
1132 crc32_begin(pack_file);
1134 if (delta) {
1135 off_t ofs = e->idx.offset - last->offset;
1136 unsigned pos = sizeof(hdr) - 1;
1138 delta_count_by_type[type]++;
1139 e->depth = last->depth + 1;
1141 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1142 OBJ_OFS_DELTA, deltalen);
1143 hashwrite(pack_file, hdr, hdrlen);
1144 pack_size += hdrlen;
1146 hdr[pos] = ofs & 127;
1147 while (ofs >>= 7)
1148 hdr[--pos] = 128 | (--ofs & 127);
1149 hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1150 pack_size += sizeof(hdr) - pos;
1151 } else {
1152 e->depth = 0;
1153 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1154 type, dat->len);
1155 hashwrite(pack_file, hdr, hdrlen);
1156 pack_size += hdrlen;
1159 hashwrite(pack_file, out, s.total_out);
1160 pack_size += s.total_out;
1162 e->idx.crc32 = crc32_end(pack_file);
1164 free(out);
1165 free(delta);
1166 if (last) {
1167 if (last->no_swap) {
1168 last->data = *dat;
1169 } else {
1170 strbuf_swap(&last->data, dat);
1172 last->offset = e->idx.offset;
1173 last->depth = e->depth;
1175 return 0;
1178 static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1180 if (hashfile_truncate(pack_file, checkpoint))
1181 die_errno("cannot truncate pack to skip duplicate");
1182 pack_size = checkpoint->offset;
1185 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1187 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1188 unsigned char *in_buf = xmalloc(in_sz);
1189 unsigned char *out_buf = xmalloc(out_sz);
1190 struct object_entry *e;
1191 struct object_id oid;
1192 unsigned long hdrlen;
1193 off_t offset;
1194 git_hash_ctx c;
1195 git_zstream s;
1196 struct hashfile_checkpoint checkpoint;
1197 int status = Z_OK;
1199 /* Determine if we should auto-checkpoint. */
1200 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1201 || (pack_size + 60 + len) < pack_size)
1202 cycle_packfile();
1204 hashfile_checkpoint(pack_file, &checkpoint);
1205 offset = checkpoint.offset;
1207 hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1209 the_hash_algo->init_fn(&c);
1210 the_hash_algo->update_fn(&c, out_buf, hdrlen);
1212 crc32_begin(pack_file);
1214 git_deflate_init(&s, pack_compression_level);
1216 hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1218 s.next_out = out_buf + hdrlen;
1219 s.avail_out = out_sz - hdrlen;
1221 while (status != Z_STREAM_END) {
1222 if (0 < len && !s.avail_in) {
1223 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1224 size_t n = fread(in_buf, 1, cnt, stdin);
1225 if (!n && feof(stdin))
1226 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1228 the_hash_algo->update_fn(&c, in_buf, n);
1229 s.next_in = in_buf;
1230 s.avail_in = n;
1231 len -= n;
1234 status = git_deflate(&s, len ? 0 : Z_FINISH);
1236 if (!s.avail_out || status == Z_STREAM_END) {
1237 size_t n = s.next_out - out_buf;
1238 hashwrite(pack_file, out_buf, n);
1239 pack_size += n;
1240 s.next_out = out_buf;
1241 s.avail_out = out_sz;
1244 switch (status) {
1245 case Z_OK:
1246 case Z_BUF_ERROR:
1247 case Z_STREAM_END:
1248 continue;
1249 default:
1250 die("unexpected deflate failure: %d", status);
1253 git_deflate_end(&s);
1254 the_hash_algo->final_fn(oid.hash, &c);
1256 if (oidout)
1257 oidcpy(oidout, &oid);
1259 e = insert_object(&oid);
1261 if (mark)
1262 insert_mark(mark, e);
1264 if (e->idx.offset) {
1265 duplicate_count_by_type[OBJ_BLOB]++;
1266 truncate_pack(&checkpoint);
1268 } else if (find_sha1_pack(oid.hash,
1269 get_packed_git(the_repository))) {
1270 e->type = OBJ_BLOB;
1271 e->pack_id = MAX_PACK_ID;
1272 e->idx.offset = 1; /* just not zero! */
1273 duplicate_count_by_type[OBJ_BLOB]++;
1274 truncate_pack(&checkpoint);
1276 } else {
1277 e->depth = 0;
1278 e->type = OBJ_BLOB;
1279 e->pack_id = pack_id;
1280 e->idx.offset = offset;
1281 e->idx.crc32 = crc32_end(pack_file);
1282 object_count++;
1283 object_count_by_type[OBJ_BLOB]++;
1286 free(in_buf);
1287 free(out_buf);
1290 /* All calls must be guarded by find_object() or find_mark() to
1291 * ensure the 'struct object_entry' passed was written by this
1292 * process instance. We unpack the entry by the offset, avoiding
1293 * the need for the corresponding .idx file. This unpacking rule
1294 * works because we only use OBJ_REF_DELTA within the packfiles
1295 * created by fast-import.
1297 * oe must not be NULL. Such an oe usually comes from giving
1298 * an unknown SHA-1 to find_object() or an undefined mark to
1299 * find_mark(). Callers must test for this condition and use
1300 * the standard read_sha1_file() when it happens.
1302 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1303 * find_mark(), where the mark was reloaded from an existing marks
1304 * file and is referencing an object that this fast-import process
1305 * instance did not write out to a packfile. Callers must test for
1306 * this condition and use read_sha1_file() instead.
1308 static void *gfi_unpack_entry(
1309 struct object_entry *oe,
1310 unsigned long *sizep)
1312 enum object_type type;
1313 struct packed_git *p = all_packs[oe->pack_id];
1314 if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1315 /* The object is stored in the packfile we are writing to
1316 * and we have modified it since the last time we scanned
1317 * back to read a previously written object. If an old
1318 * window covered [p->pack_size, p->pack_size + rawsz) its
1319 * data is stale and is not valid. Closing all windows
1320 * and updating the packfile length ensures we can read
1321 * the newly written data.
1323 close_pack_windows(p);
1324 hashflush(pack_file);
1326 /* We have to offer rawsz bytes additional on the end of
1327 * the packfile as the core unpacker code assumes the
1328 * footer is present at the file end and must promise
1329 * at least rawsz bytes within any window it maps. But
1330 * we don't actually create the footer here.
1332 p->pack_size = pack_size + the_hash_algo->rawsz;
1334 return unpack_entry(p, oe->idx.offset, &type, sizep);
1337 static const char *get_mode(const char *str, uint16_t *modep)
1339 unsigned char c;
1340 uint16_t mode = 0;
1342 while ((c = *str++) != ' ') {
1343 if (c < '0' || c > '7')
1344 return NULL;
1345 mode = (mode << 3) + (c - '0');
1347 *modep = mode;
1348 return str;
1351 static void load_tree(struct tree_entry *root)
1353 struct object_id *oid = &root->versions[1].oid;
1354 struct object_entry *myoe;
1355 struct tree_content *t;
1356 unsigned long size;
1357 char *buf;
1358 const char *c;
1360 root->tree = t = new_tree_content(8);
1361 if (is_null_oid(oid))
1362 return;
1364 myoe = find_object(oid);
1365 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1366 if (myoe->type != OBJ_TREE)
1367 die("Not a tree: %s", oid_to_hex(oid));
1368 t->delta_depth = myoe->depth;
1369 buf = gfi_unpack_entry(myoe, &size);
1370 if (!buf)
1371 die("Can't load tree %s", oid_to_hex(oid));
1372 } else {
1373 enum object_type type;
1374 buf = read_object_file(oid, &type, &size);
1375 if (!buf || type != OBJ_TREE)
1376 die("Can't load tree %s", oid_to_hex(oid));
1379 c = buf;
1380 while (c != (buf + size)) {
1381 struct tree_entry *e = new_tree_entry();
1383 if (t->entry_count == t->entry_capacity)
1384 root->tree = t = grow_tree_content(t, t->entry_count);
1385 t->entries[t->entry_count++] = e;
1387 e->tree = NULL;
1388 c = get_mode(c, &e->versions[1].mode);
1389 if (!c)
1390 die("Corrupt mode in %s", oid_to_hex(oid));
1391 e->versions[0].mode = e->versions[1].mode;
1392 e->name = to_atom(c, strlen(c));
1393 c += e->name->str_len + 1;
1394 hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1395 hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1396 c += GIT_SHA1_RAWSZ;
1398 free(buf);
1401 static int tecmp0 (const void *_a, const void *_b)
1403 struct tree_entry *a = *((struct tree_entry**)_a);
1404 struct tree_entry *b = *((struct tree_entry**)_b);
1405 return base_name_compare(
1406 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1407 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1410 static int tecmp1 (const void *_a, const void *_b)
1412 struct tree_entry *a = *((struct tree_entry**)_a);
1413 struct tree_entry *b = *((struct tree_entry**)_b);
1414 return base_name_compare(
1415 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1416 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1419 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1421 size_t maxlen = 0;
1422 unsigned int i;
1424 if (!v)
1425 QSORT(t->entries, t->entry_count, tecmp0);
1426 else
1427 QSORT(t->entries, t->entry_count, tecmp1);
1429 for (i = 0; i < t->entry_count; i++) {
1430 if (t->entries[i]->versions[v].mode)
1431 maxlen += t->entries[i]->name->str_len + 34;
1434 strbuf_reset(b);
1435 strbuf_grow(b, maxlen);
1436 for (i = 0; i < t->entry_count; i++) {
1437 struct tree_entry *e = t->entries[i];
1438 if (!e->versions[v].mode)
1439 continue;
1440 strbuf_addf(b, "%o %s%c",
1441 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1442 e->name->str_dat, '\0');
1443 strbuf_add(b, e->versions[v].oid.hash, GIT_SHA1_RAWSZ);
1447 static void store_tree(struct tree_entry *root)
1449 struct tree_content *t;
1450 unsigned int i, j, del;
1451 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1452 struct object_entry *le = NULL;
1454 if (!is_null_oid(&root->versions[1].oid))
1455 return;
1457 if (!root->tree)
1458 load_tree(root);
1459 t = root->tree;
1461 for (i = 0; i < t->entry_count; i++) {
1462 if (t->entries[i]->tree)
1463 store_tree(t->entries[i]);
1466 if (!(root->versions[0].mode & NO_DELTA))
1467 le = find_object(&root->versions[0].oid);
1468 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1469 mktree(t, 0, &old_tree);
1470 lo.data = old_tree;
1471 lo.offset = le->idx.offset;
1472 lo.depth = t->delta_depth;
1475 mktree(t, 1, &new_tree);
1476 store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1478 t->delta_depth = lo.depth;
1479 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1480 struct tree_entry *e = t->entries[i];
1481 if (e->versions[1].mode) {
1482 e->versions[0].mode = e->versions[1].mode;
1483 oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1484 t->entries[j++] = e;
1485 } else {
1486 release_tree_entry(e);
1487 del++;
1490 t->entry_count -= del;
1493 static void tree_content_replace(
1494 struct tree_entry *root,
1495 const struct object_id *oid,
1496 const uint16_t mode,
1497 struct tree_content *newtree)
1499 if (!S_ISDIR(mode))
1500 die("Root cannot be a non-directory");
1501 oidclr(&root->versions[0].oid);
1502 oidcpy(&root->versions[1].oid, oid);
1503 if (root->tree)
1504 release_tree_content_recursive(root->tree);
1505 root->tree = newtree;
1508 static int tree_content_set(
1509 struct tree_entry *root,
1510 const char *p,
1511 const struct object_id *oid,
1512 const uint16_t mode,
1513 struct tree_content *subtree)
1515 struct tree_content *t;
1516 const char *slash1;
1517 unsigned int i, n;
1518 struct tree_entry *e;
1520 slash1 = strchrnul(p, '/');
1521 n = slash1 - p;
1522 if (!n)
1523 die("Empty path component found in input");
1524 if (!*slash1 && !S_ISDIR(mode) && subtree)
1525 die("Non-directories cannot have subtrees");
1527 if (!root->tree)
1528 load_tree(root);
1529 t = root->tree;
1530 for (i = 0; i < t->entry_count; i++) {
1531 e = t->entries[i];
1532 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1533 if (!*slash1) {
1534 if (!S_ISDIR(mode)
1535 && e->versions[1].mode == mode
1536 && !oidcmp(&e->versions[1].oid, oid))
1537 return 0;
1538 e->versions[1].mode = mode;
1539 oidcpy(&e->versions[1].oid, oid);
1540 if (e->tree)
1541 release_tree_content_recursive(e->tree);
1542 e->tree = subtree;
1545 * We need to leave e->versions[0].sha1 alone
1546 * to avoid modifying the preimage tree used
1547 * when writing out the parent directory.
1548 * But after replacing the subdir with a
1549 * completely different one, it's not a good
1550 * delta base any more, and besides, we've
1551 * thrown away the tree entries needed to
1552 * make a delta against it.
1554 * So let's just explicitly disable deltas
1555 * for the subtree.
1557 if (S_ISDIR(e->versions[0].mode))
1558 e->versions[0].mode |= NO_DELTA;
1560 oidclr(&root->versions[1].oid);
1561 return 1;
1563 if (!S_ISDIR(e->versions[1].mode)) {
1564 e->tree = new_tree_content(8);
1565 e->versions[1].mode = S_IFDIR;
1567 if (!e->tree)
1568 load_tree(e);
1569 if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1570 oidclr(&root->versions[1].oid);
1571 return 1;
1573 return 0;
1577 if (t->entry_count == t->entry_capacity)
1578 root->tree = t = grow_tree_content(t, t->entry_count);
1579 e = new_tree_entry();
1580 e->name = to_atom(p, n);
1581 e->versions[0].mode = 0;
1582 oidclr(&e->versions[0].oid);
1583 t->entries[t->entry_count++] = e;
1584 if (*slash1) {
1585 e->tree = new_tree_content(8);
1586 e->versions[1].mode = S_IFDIR;
1587 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1588 } else {
1589 e->tree = subtree;
1590 e->versions[1].mode = mode;
1591 oidcpy(&e->versions[1].oid, oid);
1593 oidclr(&root->versions[1].oid);
1594 return 1;
1597 static int tree_content_remove(
1598 struct tree_entry *root,
1599 const char *p,
1600 struct tree_entry *backup_leaf,
1601 int allow_root)
1603 struct tree_content *t;
1604 const char *slash1;
1605 unsigned int i, n;
1606 struct tree_entry *e;
1608 slash1 = strchrnul(p, '/');
1609 n = slash1 - p;
1611 if (!root->tree)
1612 load_tree(root);
1614 if (!*p && allow_root) {
1615 e = root;
1616 goto del_entry;
1619 t = root->tree;
1620 for (i = 0; i < t->entry_count; i++) {
1621 e = t->entries[i];
1622 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1623 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1625 * If p names a file in some subdirectory, and a
1626 * file or symlink matching the name of the
1627 * parent directory of p exists, then p cannot
1628 * exist and need not be deleted.
1630 return 1;
1631 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1632 goto del_entry;
1633 if (!e->tree)
1634 load_tree(e);
1635 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1636 for (n = 0; n < e->tree->entry_count; n++) {
1637 if (e->tree->entries[n]->versions[1].mode) {
1638 oidclr(&root->versions[1].oid);
1639 return 1;
1642 backup_leaf = NULL;
1643 goto del_entry;
1645 return 0;
1648 return 0;
1650 del_entry:
1651 if (backup_leaf)
1652 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1653 else if (e->tree)
1654 release_tree_content_recursive(e->tree);
1655 e->tree = NULL;
1656 e->versions[1].mode = 0;
1657 oidclr(&e->versions[1].oid);
1658 oidclr(&root->versions[1].oid);
1659 return 1;
1662 static int tree_content_get(
1663 struct tree_entry *root,
1664 const char *p,
1665 struct tree_entry *leaf,
1666 int allow_root)
1668 struct tree_content *t;
1669 const char *slash1;
1670 unsigned int i, n;
1671 struct tree_entry *e;
1673 slash1 = strchrnul(p, '/');
1674 n = slash1 - p;
1675 if (!n && !allow_root)
1676 die("Empty path component found in input");
1678 if (!root->tree)
1679 load_tree(root);
1681 if (!n) {
1682 e = root;
1683 goto found_entry;
1686 t = root->tree;
1687 for (i = 0; i < t->entry_count; i++) {
1688 e = t->entries[i];
1689 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1690 if (!*slash1)
1691 goto found_entry;
1692 if (!S_ISDIR(e->versions[1].mode))
1693 return 0;
1694 if (!e->tree)
1695 load_tree(e);
1696 return tree_content_get(e, slash1 + 1, leaf, 0);
1699 return 0;
1701 found_entry:
1702 memcpy(leaf, e, sizeof(*leaf));
1703 if (e->tree && is_null_oid(&e->versions[1].oid))
1704 leaf->tree = dup_tree_content(e->tree);
1705 else
1706 leaf->tree = NULL;
1707 return 1;
1710 static int update_branch(struct branch *b)
1712 static const char *msg = "fast-import";
1713 struct ref_transaction *transaction;
1714 struct object_id old_oid;
1715 struct strbuf err = STRBUF_INIT;
1717 if (is_null_oid(&b->oid)) {
1718 if (b->delete)
1719 delete_ref(NULL, b->name, NULL, 0);
1720 return 0;
1722 if (read_ref(b->name, &old_oid))
1723 oidclr(&old_oid);
1724 if (!force_update && !is_null_oid(&old_oid)) {
1725 struct commit *old_cmit, *new_cmit;
1727 old_cmit = lookup_commit_reference_gently(&old_oid, 0);
1728 new_cmit = lookup_commit_reference_gently(&b->oid, 0);
1729 if (!old_cmit || !new_cmit)
1730 return error("Branch %s is missing commits.", b->name);
1732 if (!in_merge_bases(old_cmit, new_cmit)) {
1733 warning("Not updating %s"
1734 " (new tip %s does not contain %s)",
1735 b->name, oid_to_hex(&b->oid),
1736 oid_to_hex(&old_oid));
1737 return -1;
1740 transaction = ref_transaction_begin(&err);
1741 if (!transaction ||
1742 ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1743 0, msg, &err) ||
1744 ref_transaction_commit(transaction, &err)) {
1745 ref_transaction_free(transaction);
1746 error("%s", err.buf);
1747 strbuf_release(&err);
1748 return -1;
1750 ref_transaction_free(transaction);
1751 strbuf_release(&err);
1752 return 0;
1755 static void dump_branches(void)
1757 unsigned int i;
1758 struct branch *b;
1760 for (i = 0; i < branch_table_sz; i++) {
1761 for (b = branch_table[i]; b; b = b->table_next_branch)
1762 failure |= update_branch(b);
1766 static void dump_tags(void)
1768 static const char *msg = "fast-import";
1769 struct tag *t;
1770 struct strbuf ref_name = STRBUF_INIT;
1771 struct strbuf err = STRBUF_INIT;
1772 struct ref_transaction *transaction;
1774 transaction = ref_transaction_begin(&err);
1775 if (!transaction) {
1776 failure |= error("%s", err.buf);
1777 goto cleanup;
1779 for (t = first_tag; t; t = t->next_tag) {
1780 strbuf_reset(&ref_name);
1781 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1783 if (ref_transaction_update(transaction, ref_name.buf,
1784 &t->oid, NULL, 0, msg, &err)) {
1785 failure |= error("%s", err.buf);
1786 goto cleanup;
1789 if (ref_transaction_commit(transaction, &err))
1790 failure |= error("%s", err.buf);
1792 cleanup:
1793 ref_transaction_free(transaction);
1794 strbuf_release(&ref_name);
1795 strbuf_release(&err);
1798 static void dump_marks_helper(FILE *f,
1799 uintmax_t base,
1800 struct mark_set *m)
1802 uintmax_t k;
1803 if (m->shift) {
1804 for (k = 0; k < 1024; k++) {
1805 if (m->data.sets[k])
1806 dump_marks_helper(f, base + (k << m->shift),
1807 m->data.sets[k]);
1809 } else {
1810 for (k = 0; k < 1024; k++) {
1811 if (m->data.marked[k])
1812 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1813 oid_to_hex(&m->data.marked[k]->idx.oid));
1818 static void dump_marks(void)
1820 static struct lock_file mark_lock;
1821 FILE *f;
1823 if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1824 return;
1826 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1827 failure |= error_errno("Unable to write marks file %s",
1828 export_marks_file);
1829 return;
1832 f = fdopen_lock_file(&mark_lock, "w");
1833 if (!f) {
1834 int saved_errno = errno;
1835 rollback_lock_file(&mark_lock);
1836 failure |= error("Unable to write marks file %s: %s",
1837 export_marks_file, strerror(saved_errno));
1838 return;
1841 dump_marks_helper(f, 0, marks);
1842 if (commit_lock_file(&mark_lock)) {
1843 failure |= error_errno("Unable to write file %s",
1844 export_marks_file);
1845 return;
1849 static void read_marks(void)
1851 char line[512];
1852 FILE *f = fopen(import_marks_file, "r");
1853 if (f)
1855 else if (import_marks_file_ignore_missing && errno == ENOENT)
1856 goto done; /* Marks file does not exist */
1857 else
1858 die_errno("cannot read '%s'", import_marks_file);
1859 while (fgets(line, sizeof(line), f)) {
1860 uintmax_t mark;
1861 char *end;
1862 struct object_id oid;
1863 struct object_entry *e;
1865 end = strchr(line, '\n');
1866 if (line[0] != ':' || !end)
1867 die("corrupt mark line: %s", line);
1868 *end = 0;
1869 mark = strtoumax(line + 1, &end, 10);
1870 if (!mark || end == line + 1
1871 || *end != ' ' || get_oid_hex(end + 1, &oid))
1872 die("corrupt mark line: %s", line);
1873 e = find_object(&oid);
1874 if (!e) {
1875 enum object_type type = oid_object_info(&oid, NULL);
1876 if (type < 0)
1877 die("object not found: %s", oid_to_hex(&oid));
1878 e = insert_object(&oid);
1879 e->type = type;
1880 e->pack_id = MAX_PACK_ID;
1881 e->idx.offset = 1; /* just not zero! */
1883 insert_mark(mark, e);
1885 fclose(f);
1886 done:
1887 import_marks_file_done = 1;
1891 static int read_next_command(void)
1893 static int stdin_eof = 0;
1895 if (stdin_eof) {
1896 unread_command_buf = 0;
1897 return EOF;
1900 for (;;) {
1901 const char *p;
1903 if (unread_command_buf) {
1904 unread_command_buf = 0;
1905 } else {
1906 struct recent_command *rc;
1908 strbuf_detach(&command_buf, NULL);
1909 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1910 if (stdin_eof)
1911 return EOF;
1913 if (!seen_data_command
1914 && !starts_with(command_buf.buf, "feature ")
1915 && !starts_with(command_buf.buf, "option ")) {
1916 parse_argv();
1919 rc = rc_free;
1920 if (rc)
1921 rc_free = rc->next;
1922 else {
1923 rc = cmd_hist.next;
1924 cmd_hist.next = rc->next;
1925 cmd_hist.next->prev = &cmd_hist;
1926 free(rc->buf);
1929 rc->buf = command_buf.buf;
1930 rc->prev = cmd_tail;
1931 rc->next = cmd_hist.prev;
1932 rc->prev->next = rc;
1933 cmd_tail = rc;
1935 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1936 parse_get_mark(p);
1937 continue;
1939 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1940 parse_cat_blob(p);
1941 continue;
1943 if (command_buf.buf[0] == '#')
1944 continue;
1945 return 0;
1949 static void skip_optional_lf(void)
1951 int term_char = fgetc(stdin);
1952 if (term_char != '\n' && term_char != EOF)
1953 ungetc(term_char, stdin);
1956 static void parse_mark(void)
1958 const char *v;
1959 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1960 next_mark = strtoumax(v, NULL, 10);
1961 read_next_command();
1963 else
1964 next_mark = 0;
1967 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1969 const char *data;
1970 strbuf_reset(sb);
1972 if (!skip_prefix(command_buf.buf, "data ", &data))
1973 die("Expected 'data n' command, found: %s", command_buf.buf);
1975 if (skip_prefix(data, "<<", &data)) {
1976 char *term = xstrdup(data);
1977 size_t term_len = command_buf.len - (data - command_buf.buf);
1979 strbuf_detach(&command_buf, NULL);
1980 for (;;) {
1981 if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1982 die("EOF in data (terminator '%s' not found)", term);
1983 if (term_len == command_buf.len
1984 && !strcmp(term, command_buf.buf))
1985 break;
1986 strbuf_addbuf(sb, &command_buf);
1987 strbuf_addch(sb, '\n');
1989 free(term);
1991 else {
1992 uintmax_t len = strtoumax(data, NULL, 10);
1993 size_t n = 0, length = (size_t)len;
1995 if (limit && limit < len) {
1996 *len_res = len;
1997 return 0;
1999 if (length < len)
2000 die("data is too large to use in this context");
2002 while (n < length) {
2003 size_t s = strbuf_fread(sb, length - n, stdin);
2004 if (!s && feof(stdin))
2005 die("EOF in data (%lu bytes remaining)",
2006 (unsigned long)(length - n));
2007 n += s;
2011 skip_optional_lf();
2012 return 1;
2015 static int validate_raw_date(const char *src, struct strbuf *result)
2017 const char *orig_src = src;
2018 char *endp;
2019 unsigned long num;
2021 errno = 0;
2023 num = strtoul(src, &endp, 10);
2024 /* NEEDSWORK: perhaps check for reasonable values? */
2025 if (errno || endp == src || *endp != ' ')
2026 return -1;
2028 src = endp + 1;
2029 if (*src != '-' && *src != '+')
2030 return -1;
2032 num = strtoul(src + 1, &endp, 10);
2033 if (errno || endp == src + 1 || *endp || 1400 < num)
2034 return -1;
2036 strbuf_addstr(result, orig_src);
2037 return 0;
2040 static char *parse_ident(const char *buf)
2042 const char *ltgt;
2043 size_t name_len;
2044 struct strbuf ident = STRBUF_INIT;
2046 /* ensure there is a space delimiter even if there is no name */
2047 if (*buf == '<')
2048 --buf;
2050 ltgt = buf + strcspn(buf, "<>");
2051 if (*ltgt != '<')
2052 die("Missing < in ident string: %s", buf);
2053 if (ltgt != buf && ltgt[-1] != ' ')
2054 die("Missing space before < in ident string: %s", buf);
2055 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2056 if (*ltgt != '>')
2057 die("Missing > in ident string: %s", buf);
2058 ltgt++;
2059 if (*ltgt != ' ')
2060 die("Missing space after > in ident string: %s", buf);
2061 ltgt++;
2062 name_len = ltgt - buf;
2063 strbuf_add(&ident, buf, name_len);
2065 switch (whenspec) {
2066 case WHENSPEC_RAW:
2067 if (validate_raw_date(ltgt, &ident) < 0)
2068 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2069 break;
2070 case WHENSPEC_RFC2822:
2071 if (parse_date(ltgt, &ident) < 0)
2072 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2073 break;
2074 case WHENSPEC_NOW:
2075 if (strcmp("now", ltgt))
2076 die("Date in ident must be 'now': %s", buf);
2077 datestamp(&ident);
2078 break;
2081 return strbuf_detach(&ident, NULL);
2084 static void parse_and_store_blob(
2085 struct last_object *last,
2086 struct object_id *oidout,
2087 uintmax_t mark)
2089 static struct strbuf buf = STRBUF_INIT;
2090 uintmax_t len;
2092 if (parse_data(&buf, big_file_threshold, &len))
2093 store_object(OBJ_BLOB, &buf, last, oidout, mark);
2094 else {
2095 if (last) {
2096 strbuf_release(&last->data);
2097 last->offset = 0;
2098 last->depth = 0;
2100 stream_blob(len, oidout, mark);
2101 skip_optional_lf();
2105 static void parse_new_blob(void)
2107 read_next_command();
2108 parse_mark();
2109 parse_and_store_blob(&last_blob, NULL, next_mark);
2112 static void unload_one_branch(void)
2114 while (cur_active_branches
2115 && cur_active_branches >= max_active_branches) {
2116 uintmax_t min_commit = ULONG_MAX;
2117 struct branch *e, *l = NULL, *p = NULL;
2119 for (e = active_branches; e; e = e->active_next_branch) {
2120 if (e->last_commit < min_commit) {
2121 p = l;
2122 min_commit = e->last_commit;
2124 l = e;
2127 if (p) {
2128 e = p->active_next_branch;
2129 p->active_next_branch = e->active_next_branch;
2130 } else {
2131 e = active_branches;
2132 active_branches = e->active_next_branch;
2134 e->active = 0;
2135 e->active_next_branch = NULL;
2136 if (e->branch_tree.tree) {
2137 release_tree_content_recursive(e->branch_tree.tree);
2138 e->branch_tree.tree = NULL;
2140 cur_active_branches--;
2144 static void load_branch(struct branch *b)
2146 load_tree(&b->branch_tree);
2147 if (!b->active) {
2148 b->active = 1;
2149 b->active_next_branch = active_branches;
2150 active_branches = b;
2151 cur_active_branches++;
2152 branch_load_count++;
2156 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2158 unsigned char fanout = 0;
2159 while ((num_notes >>= 8))
2160 fanout++;
2161 return fanout;
2164 static void construct_path_with_fanout(const char *hex_sha1,
2165 unsigned char fanout, char *path)
2167 unsigned int i = 0, j = 0;
2168 if (fanout >= the_hash_algo->rawsz)
2169 die("Too large fanout (%u)", fanout);
2170 while (fanout) {
2171 path[i++] = hex_sha1[j++];
2172 path[i++] = hex_sha1[j++];
2173 path[i++] = '/';
2174 fanout--;
2176 memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2177 path[i + the_hash_algo->hexsz - j] = '\0';
2180 static uintmax_t do_change_note_fanout(
2181 struct tree_entry *orig_root, struct tree_entry *root,
2182 char *hex_oid, unsigned int hex_oid_len,
2183 char *fullpath, unsigned int fullpath_len,
2184 unsigned char fanout)
2186 struct tree_content *t;
2187 struct tree_entry *e, leaf;
2188 unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2189 uintmax_t num_notes = 0;
2190 struct object_id oid;
2191 char realpath[60];
2193 if (!root->tree)
2194 load_tree(root);
2195 t = root->tree;
2197 for (i = 0; t && i < t->entry_count; i++) {
2198 e = t->entries[i];
2199 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2200 tmp_fullpath_len = fullpath_len;
2203 * We're interested in EITHER existing note entries (entries
2204 * with exactly 40 hex chars in path, not including directory
2205 * separators), OR directory entries that may contain note
2206 * entries (with < 40 hex chars in path).
2207 * Also, each path component in a note entry must be a multiple
2208 * of 2 chars.
2210 if (!e->versions[1].mode ||
2211 tmp_hex_oid_len > GIT_SHA1_HEXSZ ||
2212 e->name->str_len % 2)
2213 continue;
2215 /* This _may_ be a note entry, or a subdir containing notes */
2216 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2217 e->name->str_len);
2218 if (tmp_fullpath_len)
2219 fullpath[tmp_fullpath_len++] = '/';
2220 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2221 e->name->str_len);
2222 tmp_fullpath_len += e->name->str_len;
2223 fullpath[tmp_fullpath_len] = '\0';
2225 if (tmp_hex_oid_len == GIT_SHA1_HEXSZ && !get_oid_hex(hex_oid, &oid)) {
2226 /* This is a note entry */
2227 if (fanout == 0xff) {
2228 /* Counting mode, no rename */
2229 num_notes++;
2230 continue;
2232 construct_path_with_fanout(hex_oid, fanout, realpath);
2233 if (!strcmp(fullpath, realpath)) {
2234 /* Note entry is in correct location */
2235 num_notes++;
2236 continue;
2239 /* Rename fullpath to realpath */
2240 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2241 die("Failed to remove path %s", fullpath);
2242 tree_content_set(orig_root, realpath,
2243 &leaf.versions[1].oid,
2244 leaf.versions[1].mode,
2245 leaf.tree);
2246 } else if (S_ISDIR(e->versions[1].mode)) {
2247 /* This is a subdir that may contain note entries */
2248 num_notes += do_change_note_fanout(orig_root, e,
2249 hex_oid, tmp_hex_oid_len,
2250 fullpath, tmp_fullpath_len, fanout);
2253 /* The above may have reallocated the current tree_content */
2254 t = root->tree;
2256 return num_notes;
2259 static uintmax_t change_note_fanout(struct tree_entry *root,
2260 unsigned char fanout)
2263 * The size of path is due to one slash between every two hex digits,
2264 * plus the terminating NUL. Note that there is no slash at the end, so
2265 * the number of slashes is one less than half the number of hex
2266 * characters.
2268 char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2269 return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2273 * Given a pointer into a string, parse a mark reference:
2275 * idnum ::= ':' bigint;
2277 * Return the first character after the value in *endptr.
2279 * Complain if the following character is not what is expected,
2280 * either a space or end of the string.
2282 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2284 uintmax_t mark;
2286 assert(*p == ':');
2287 p++;
2288 mark = strtoumax(p, endptr, 10);
2289 if (*endptr == p)
2290 die("No value after ':' in mark: %s", command_buf.buf);
2291 return mark;
2295 * Parse the mark reference, and complain if this is not the end of
2296 * the string.
2298 static uintmax_t parse_mark_ref_eol(const char *p)
2300 char *end;
2301 uintmax_t mark;
2303 mark = parse_mark_ref(p, &end);
2304 if (*end != '\0')
2305 die("Garbage after mark: %s", command_buf.buf);
2306 return mark;
2310 * Parse the mark reference, demanding a trailing space. Return a
2311 * pointer to the space.
2313 static uintmax_t parse_mark_ref_space(const char **p)
2315 uintmax_t mark;
2316 char *end;
2318 mark = parse_mark_ref(*p, &end);
2319 if (*end++ != ' ')
2320 die("Missing space after mark: %s", command_buf.buf);
2321 *p = end;
2322 return mark;
2325 static void file_change_m(const char *p, struct branch *b)
2327 static struct strbuf uq = STRBUF_INIT;
2328 const char *endp;
2329 struct object_entry *oe;
2330 struct object_id oid;
2331 uint16_t mode, inline_data = 0;
2333 p = get_mode(p, &mode);
2334 if (!p)
2335 die("Corrupt mode: %s", command_buf.buf);
2336 switch (mode) {
2337 case 0644:
2338 case 0755:
2339 mode |= S_IFREG;
2340 case S_IFREG | 0644:
2341 case S_IFREG | 0755:
2342 case S_IFLNK:
2343 case S_IFDIR:
2344 case S_IFGITLINK:
2345 /* ok */
2346 break;
2347 default:
2348 die("Corrupt mode: %s", command_buf.buf);
2351 if (*p == ':') {
2352 oe = find_mark(parse_mark_ref_space(&p));
2353 oidcpy(&oid, &oe->idx.oid);
2354 } else if (skip_prefix(p, "inline ", &p)) {
2355 inline_data = 1;
2356 oe = NULL; /* not used with inline_data, but makes gcc happy */
2357 } else {
2358 if (parse_oid_hex(p, &oid, &p))
2359 die("Invalid dataref: %s", command_buf.buf);
2360 oe = find_object(&oid);
2361 if (*p++ != ' ')
2362 die("Missing space after SHA1: %s", command_buf.buf);
2365 strbuf_reset(&uq);
2366 if (!unquote_c_style(&uq, p, &endp)) {
2367 if (*endp)
2368 die("Garbage after path in: %s", command_buf.buf);
2369 p = uq.buf;
2372 /* Git does not track empty, non-toplevel directories. */
2373 if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2374 tree_content_remove(&b->branch_tree, p, NULL, 0);
2375 return;
2378 if (S_ISGITLINK(mode)) {
2379 if (inline_data)
2380 die("Git links cannot be specified 'inline': %s",
2381 command_buf.buf);
2382 else if (oe) {
2383 if (oe->type != OBJ_COMMIT)
2384 die("Not a commit (actually a %s): %s",
2385 type_name(oe->type), command_buf.buf);
2388 * Accept the sha1 without checking; it expected to be in
2389 * another repository.
2391 } else if (inline_data) {
2392 if (S_ISDIR(mode))
2393 die("Directories cannot be specified 'inline': %s",
2394 command_buf.buf);
2395 if (p != uq.buf) {
2396 strbuf_addstr(&uq, p);
2397 p = uq.buf;
2399 read_next_command();
2400 parse_and_store_blob(&last_blob, &oid, 0);
2401 } else {
2402 enum object_type expected = S_ISDIR(mode) ?
2403 OBJ_TREE: OBJ_BLOB;
2404 enum object_type type = oe ? oe->type :
2405 oid_object_info(&oid, NULL);
2406 if (type < 0)
2407 die("%s not found: %s",
2408 S_ISDIR(mode) ? "Tree" : "Blob",
2409 command_buf.buf);
2410 if (type != expected)
2411 die("Not a %s (actually a %s): %s",
2412 type_name(expected), type_name(type),
2413 command_buf.buf);
2416 if (!*p) {
2417 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2418 return;
2420 tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2423 static void file_change_d(const char *p, struct branch *b)
2425 static struct strbuf uq = STRBUF_INIT;
2426 const char *endp;
2428 strbuf_reset(&uq);
2429 if (!unquote_c_style(&uq, p, &endp)) {
2430 if (*endp)
2431 die("Garbage after path in: %s", command_buf.buf);
2432 p = uq.buf;
2434 tree_content_remove(&b->branch_tree, p, NULL, 1);
2437 static void file_change_cr(const char *s, struct branch *b, int rename)
2439 const char *d;
2440 static struct strbuf s_uq = STRBUF_INIT;
2441 static struct strbuf d_uq = STRBUF_INIT;
2442 const char *endp;
2443 struct tree_entry leaf;
2445 strbuf_reset(&s_uq);
2446 if (!unquote_c_style(&s_uq, s, &endp)) {
2447 if (*endp != ' ')
2448 die("Missing space after source: %s", command_buf.buf);
2449 } else {
2450 endp = strchr(s, ' ');
2451 if (!endp)
2452 die("Missing space after source: %s", command_buf.buf);
2453 strbuf_add(&s_uq, s, endp - s);
2455 s = s_uq.buf;
2457 endp++;
2458 if (!*endp)
2459 die("Missing dest: %s", command_buf.buf);
2461 d = endp;
2462 strbuf_reset(&d_uq);
2463 if (!unquote_c_style(&d_uq, d, &endp)) {
2464 if (*endp)
2465 die("Garbage after dest in: %s", command_buf.buf);
2466 d = d_uq.buf;
2469 memset(&leaf, 0, sizeof(leaf));
2470 if (rename)
2471 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2472 else
2473 tree_content_get(&b->branch_tree, s, &leaf, 1);
2474 if (!leaf.versions[1].mode)
2475 die("Path %s not in branch", s);
2476 if (!*d) { /* C "path/to/subdir" "" */
2477 tree_content_replace(&b->branch_tree,
2478 &leaf.versions[1].oid,
2479 leaf.versions[1].mode,
2480 leaf.tree);
2481 return;
2483 tree_content_set(&b->branch_tree, d,
2484 &leaf.versions[1].oid,
2485 leaf.versions[1].mode,
2486 leaf.tree);
2489 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2491 static struct strbuf uq = STRBUF_INIT;
2492 struct object_entry *oe;
2493 struct branch *s;
2494 struct object_id oid, commit_oid;
2495 char path[60];
2496 uint16_t inline_data = 0;
2497 unsigned char new_fanout;
2500 * When loading a branch, we don't traverse its tree to count the real
2501 * number of notes (too expensive to do this for all non-note refs).
2502 * This means that recently loaded notes refs might incorrectly have
2503 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2505 * Fix this by traversing the tree and counting the number of notes
2506 * when b->num_notes == 0. If the notes tree is truly empty, the
2507 * calculation should not take long.
2509 if (b->num_notes == 0 && *old_fanout == 0) {
2510 /* Invoke change_note_fanout() in "counting mode". */
2511 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2512 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2515 /* Now parse the notemodify command. */
2516 /* <dataref> or 'inline' */
2517 if (*p == ':') {
2518 oe = find_mark(parse_mark_ref_space(&p));
2519 oidcpy(&oid, &oe->idx.oid);
2520 } else if (skip_prefix(p, "inline ", &p)) {
2521 inline_data = 1;
2522 oe = NULL; /* not used with inline_data, but makes gcc happy */
2523 } else {
2524 if (parse_oid_hex(p, &oid, &p))
2525 die("Invalid dataref: %s", command_buf.buf);
2526 oe = find_object(&oid);
2527 if (*p++ != ' ')
2528 die("Missing space after SHA1: %s", command_buf.buf);
2531 /* <commit-ish> */
2532 s = lookup_branch(p);
2533 if (s) {
2534 if (is_null_oid(&s->oid))
2535 die("Can't add a note on empty branch.");
2536 oidcpy(&commit_oid, &s->oid);
2537 } else if (*p == ':') {
2538 uintmax_t commit_mark = parse_mark_ref_eol(p);
2539 struct object_entry *commit_oe = find_mark(commit_mark);
2540 if (commit_oe->type != OBJ_COMMIT)
2541 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2542 oidcpy(&commit_oid, &commit_oe->idx.oid);
2543 } else if (!get_oid(p, &commit_oid)) {
2544 unsigned long size;
2545 char *buf = read_object_with_reference(&commit_oid,
2546 commit_type, &size,
2547 &commit_oid);
2548 if (!buf || size < 46)
2549 die("Not a valid commit: %s", p);
2550 free(buf);
2551 } else
2552 die("Invalid ref name or SHA1 expression: %s", p);
2554 if (inline_data) {
2555 if (p != uq.buf) {
2556 strbuf_addstr(&uq, p);
2557 p = uq.buf;
2559 read_next_command();
2560 parse_and_store_blob(&last_blob, &oid, 0);
2561 } else if (oe) {
2562 if (oe->type != OBJ_BLOB)
2563 die("Not a blob (actually a %s): %s",
2564 type_name(oe->type), command_buf.buf);
2565 } else if (!is_null_oid(&oid)) {
2566 enum object_type type = oid_object_info(&oid, NULL);
2567 if (type < 0)
2568 die("Blob not found: %s", command_buf.buf);
2569 if (type != OBJ_BLOB)
2570 die("Not a blob (actually a %s): %s",
2571 type_name(type), command_buf.buf);
2574 construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2575 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2576 b->num_notes--;
2578 if (is_null_oid(&oid))
2579 return; /* nothing to insert */
2581 b->num_notes++;
2582 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2583 construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2584 tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2587 static void file_change_deleteall(struct branch *b)
2589 release_tree_content_recursive(b->branch_tree.tree);
2590 oidclr(&b->branch_tree.versions[0].oid);
2591 oidclr(&b->branch_tree.versions[1].oid);
2592 load_tree(&b->branch_tree);
2593 b->num_notes = 0;
2596 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2598 if (!buf || size < GIT_SHA1_HEXSZ + 6)
2599 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2600 if (memcmp("tree ", buf, 5)
2601 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2602 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2603 oidcpy(&b->branch_tree.versions[0].oid,
2604 &b->branch_tree.versions[1].oid);
2607 static void parse_from_existing(struct branch *b)
2609 if (is_null_oid(&b->oid)) {
2610 oidclr(&b->branch_tree.versions[0].oid);
2611 oidclr(&b->branch_tree.versions[1].oid);
2612 } else {
2613 unsigned long size;
2614 char *buf;
2616 buf = read_object_with_reference(&b->oid, commit_type, &size,
2617 &b->oid);
2618 parse_from_commit(b, buf, size);
2619 free(buf);
2623 static int parse_from(struct branch *b)
2625 const char *from;
2626 struct branch *s;
2627 struct object_id oid;
2629 if (!skip_prefix(command_buf.buf, "from ", &from))
2630 return 0;
2632 oidcpy(&oid, &b->branch_tree.versions[1].oid);
2634 s = lookup_branch(from);
2635 if (b == s)
2636 die("Can't create a branch from itself: %s", b->name);
2637 else if (s) {
2638 struct object_id *t = &s->branch_tree.versions[1].oid;
2639 oidcpy(&b->oid, &s->oid);
2640 oidcpy(&b->branch_tree.versions[0].oid, t);
2641 oidcpy(&b->branch_tree.versions[1].oid, t);
2642 } else if (*from == ':') {
2643 uintmax_t idnum = parse_mark_ref_eol(from);
2644 struct object_entry *oe = find_mark(idnum);
2645 if (oe->type != OBJ_COMMIT)
2646 die("Mark :%" PRIuMAX " not a commit", idnum);
2647 if (oidcmp(&b->oid, &oe->idx.oid)) {
2648 oidcpy(&b->oid, &oe->idx.oid);
2649 if (oe->pack_id != MAX_PACK_ID) {
2650 unsigned long size;
2651 char *buf = gfi_unpack_entry(oe, &size);
2652 parse_from_commit(b, buf, size);
2653 free(buf);
2654 } else
2655 parse_from_existing(b);
2657 } else if (!get_oid(from, &b->oid)) {
2658 parse_from_existing(b);
2659 if (is_null_oid(&b->oid))
2660 b->delete = 1;
2662 else
2663 die("Invalid ref name or SHA1 expression: %s", from);
2665 if (b->branch_tree.tree && oidcmp(&oid, &b->branch_tree.versions[1].oid)) {
2666 release_tree_content_recursive(b->branch_tree.tree);
2667 b->branch_tree.tree = NULL;
2670 read_next_command();
2671 return 1;
2674 static struct hash_list *parse_merge(unsigned int *count)
2676 struct hash_list *list = NULL, **tail = &list, *n;
2677 const char *from;
2678 struct branch *s;
2680 *count = 0;
2681 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2682 n = xmalloc(sizeof(*n));
2683 s = lookup_branch(from);
2684 if (s)
2685 oidcpy(&n->oid, &s->oid);
2686 else if (*from == ':') {
2687 uintmax_t idnum = parse_mark_ref_eol(from);
2688 struct object_entry *oe = find_mark(idnum);
2689 if (oe->type != OBJ_COMMIT)
2690 die("Mark :%" PRIuMAX " not a commit", idnum);
2691 oidcpy(&n->oid, &oe->idx.oid);
2692 } else if (!get_oid(from, &n->oid)) {
2693 unsigned long size;
2694 char *buf = read_object_with_reference(&n->oid,
2695 commit_type,
2696 &size, &n->oid);
2697 if (!buf || size < 46)
2698 die("Not a valid commit: %s", from);
2699 free(buf);
2700 } else
2701 die("Invalid ref name or SHA1 expression: %s", from);
2703 n->next = NULL;
2704 *tail = n;
2705 tail = &n->next;
2707 (*count)++;
2708 read_next_command();
2710 return list;
2713 static void parse_new_commit(const char *arg)
2715 static struct strbuf msg = STRBUF_INIT;
2716 struct branch *b;
2717 char *author = NULL;
2718 char *committer = NULL;
2719 struct hash_list *merge_list = NULL;
2720 unsigned int merge_count;
2721 unsigned char prev_fanout, new_fanout;
2722 const char *v;
2724 b = lookup_branch(arg);
2725 if (!b)
2726 b = new_branch(arg);
2728 read_next_command();
2729 parse_mark();
2730 if (skip_prefix(command_buf.buf, "author ", &v)) {
2731 author = parse_ident(v);
2732 read_next_command();
2734 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2735 committer = parse_ident(v);
2736 read_next_command();
2738 if (!committer)
2739 die("Expected committer but didn't get one");
2740 parse_data(&msg, 0, NULL);
2741 read_next_command();
2742 parse_from(b);
2743 merge_list = parse_merge(&merge_count);
2745 /* ensure the branch is active/loaded */
2746 if (!b->branch_tree.tree || !max_active_branches) {
2747 unload_one_branch();
2748 load_branch(b);
2751 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2753 /* file_change* */
2754 while (command_buf.len > 0) {
2755 if (skip_prefix(command_buf.buf, "M ", &v))
2756 file_change_m(v, b);
2757 else if (skip_prefix(command_buf.buf, "D ", &v))
2758 file_change_d(v, b);
2759 else if (skip_prefix(command_buf.buf, "R ", &v))
2760 file_change_cr(v, b, 1);
2761 else if (skip_prefix(command_buf.buf, "C ", &v))
2762 file_change_cr(v, b, 0);
2763 else if (skip_prefix(command_buf.buf, "N ", &v))
2764 note_change_n(v, b, &prev_fanout);
2765 else if (!strcmp("deleteall", command_buf.buf))
2766 file_change_deleteall(b);
2767 else if (skip_prefix(command_buf.buf, "ls ", &v))
2768 parse_ls(v, b);
2769 else {
2770 unread_command_buf = 1;
2771 break;
2773 if (read_next_command() == EOF)
2774 break;
2777 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2778 if (new_fanout != prev_fanout)
2779 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2781 /* build the tree and the commit */
2782 store_tree(&b->branch_tree);
2783 oidcpy(&b->branch_tree.versions[0].oid,
2784 &b->branch_tree.versions[1].oid);
2786 strbuf_reset(&new_data);
2787 strbuf_addf(&new_data, "tree %s\n",
2788 oid_to_hex(&b->branch_tree.versions[1].oid));
2789 if (!is_null_oid(&b->oid))
2790 strbuf_addf(&new_data, "parent %s\n",
2791 oid_to_hex(&b->oid));
2792 while (merge_list) {
2793 struct hash_list *next = merge_list->next;
2794 strbuf_addf(&new_data, "parent %s\n",
2795 oid_to_hex(&merge_list->oid));
2796 free(merge_list);
2797 merge_list = next;
2799 strbuf_addf(&new_data,
2800 "author %s\n"
2801 "committer %s\n"
2802 "\n",
2803 author ? author : committer, committer);
2804 strbuf_addbuf(&new_data, &msg);
2805 free(author);
2806 free(committer);
2808 if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2809 b->pack_id = pack_id;
2810 b->last_commit = object_count_by_type[OBJ_COMMIT];
2813 static void parse_new_tag(const char *arg)
2815 static struct strbuf msg = STRBUF_INIT;
2816 const char *from;
2817 char *tagger;
2818 struct branch *s;
2819 struct tag *t;
2820 uintmax_t from_mark = 0;
2821 struct object_id oid;
2822 enum object_type type;
2823 const char *v;
2825 t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2826 memset(t, 0, sizeof(struct tag));
2827 t->name = pool_strdup(arg);
2828 if (last_tag)
2829 last_tag->next_tag = t;
2830 else
2831 first_tag = t;
2832 last_tag = t;
2833 read_next_command();
2835 /* from ... */
2836 if (!skip_prefix(command_buf.buf, "from ", &from))
2837 die("Expected from command, got %s", command_buf.buf);
2838 s = lookup_branch(from);
2839 if (s) {
2840 if (is_null_oid(&s->oid))
2841 die("Can't tag an empty branch.");
2842 oidcpy(&oid, &s->oid);
2843 type = OBJ_COMMIT;
2844 } else if (*from == ':') {
2845 struct object_entry *oe;
2846 from_mark = parse_mark_ref_eol(from);
2847 oe = find_mark(from_mark);
2848 type = oe->type;
2849 oidcpy(&oid, &oe->idx.oid);
2850 } else if (!get_oid(from, &oid)) {
2851 struct object_entry *oe = find_object(&oid);
2852 if (!oe) {
2853 type = oid_object_info(&oid, NULL);
2854 if (type < 0)
2855 die("Not a valid object: %s", from);
2856 } else
2857 type = oe->type;
2858 } else
2859 die("Invalid ref name or SHA1 expression: %s", from);
2860 read_next_command();
2862 /* tagger ... */
2863 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2864 tagger = parse_ident(v);
2865 read_next_command();
2866 } else
2867 tagger = NULL;
2869 /* tag payload/message */
2870 parse_data(&msg, 0, NULL);
2872 /* build the tag object */
2873 strbuf_reset(&new_data);
2875 strbuf_addf(&new_data,
2876 "object %s\n"
2877 "type %s\n"
2878 "tag %s\n",
2879 oid_to_hex(&oid), type_name(type), t->name);
2880 if (tagger)
2881 strbuf_addf(&new_data,
2882 "tagger %s\n", tagger);
2883 strbuf_addch(&new_data, '\n');
2884 strbuf_addbuf(&new_data, &msg);
2885 free(tagger);
2887 if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, 0))
2888 t->pack_id = MAX_PACK_ID;
2889 else
2890 t->pack_id = pack_id;
2893 static void parse_reset_branch(const char *arg)
2895 struct branch *b;
2897 b = lookup_branch(arg);
2898 if (b) {
2899 oidclr(&b->oid);
2900 oidclr(&b->branch_tree.versions[0].oid);
2901 oidclr(&b->branch_tree.versions[1].oid);
2902 if (b->branch_tree.tree) {
2903 release_tree_content_recursive(b->branch_tree.tree);
2904 b->branch_tree.tree = NULL;
2907 else
2908 b = new_branch(arg);
2909 read_next_command();
2910 parse_from(b);
2911 if (command_buf.len > 0)
2912 unread_command_buf = 1;
2915 static void cat_blob_write(const char *buf, unsigned long size)
2917 if (write_in_full(cat_blob_fd, buf, size) < 0)
2918 die_errno("Write to frontend failed");
2921 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2923 struct strbuf line = STRBUF_INIT;
2924 unsigned long size;
2925 enum object_type type = 0;
2926 char *buf;
2928 if (!oe || oe->pack_id == MAX_PACK_ID) {
2929 buf = read_object_file(oid, &type, &size);
2930 } else {
2931 type = oe->type;
2932 buf = gfi_unpack_entry(oe, &size);
2936 * Output based on batch_one_object() from cat-file.c.
2938 if (type <= 0) {
2939 strbuf_reset(&line);
2940 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2941 cat_blob_write(line.buf, line.len);
2942 strbuf_release(&line);
2943 free(buf);
2944 return;
2946 if (!buf)
2947 die("Can't read object %s", oid_to_hex(oid));
2948 if (type != OBJ_BLOB)
2949 die("Object %s is a %s but a blob was expected.",
2950 oid_to_hex(oid), type_name(type));
2951 strbuf_reset(&line);
2952 strbuf_addf(&line, "%s %s %lu\n", oid_to_hex(oid),
2953 type_name(type), size);
2954 cat_blob_write(line.buf, line.len);
2955 strbuf_release(&line);
2956 cat_blob_write(buf, size);
2957 cat_blob_write("\n", 1);
2958 if (oe && oe->pack_id == pack_id) {
2959 last_blob.offset = oe->idx.offset;
2960 strbuf_attach(&last_blob.data, buf, size, size);
2961 last_blob.depth = oe->depth;
2962 } else
2963 free(buf);
2966 static void parse_get_mark(const char *p)
2968 struct object_entry *oe;
2969 char output[GIT_MAX_HEXSZ + 2];
2971 /* get-mark SP <object> LF */
2972 if (*p != ':')
2973 die("Not a mark: %s", p);
2975 oe = find_mark(parse_mark_ref_eol(p));
2976 if (!oe)
2977 die("Unknown mark: %s", command_buf.buf);
2979 xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2980 cat_blob_write(output, GIT_SHA1_HEXSZ + 1);
2983 static void parse_cat_blob(const char *p)
2985 struct object_entry *oe;
2986 struct object_id oid;
2988 /* cat-blob SP <object> LF */
2989 if (*p == ':') {
2990 oe = find_mark(parse_mark_ref_eol(p));
2991 if (!oe)
2992 die("Unknown mark: %s", command_buf.buf);
2993 oidcpy(&oid, &oe->idx.oid);
2994 } else {
2995 if (parse_oid_hex(p, &oid, &p))
2996 die("Invalid dataref: %s", command_buf.buf);
2997 if (*p)
2998 die("Garbage after SHA1: %s", command_buf.buf);
2999 oe = find_object(&oid);
3002 cat_blob(oe, &oid);
3005 static struct object_entry *dereference(struct object_entry *oe,
3006 struct object_id *oid)
3008 unsigned long size;
3009 char *buf = NULL;
3010 if (!oe) {
3011 enum object_type type = oid_object_info(oid, NULL);
3012 if (type < 0)
3013 die("object not found: %s", oid_to_hex(oid));
3014 /* cache it! */
3015 oe = insert_object(oid);
3016 oe->type = type;
3017 oe->pack_id = MAX_PACK_ID;
3018 oe->idx.offset = 1;
3020 switch (oe->type) {
3021 case OBJ_TREE: /* easy case. */
3022 return oe;
3023 case OBJ_COMMIT:
3024 case OBJ_TAG:
3025 break;
3026 default:
3027 die("Not a tree-ish: %s", command_buf.buf);
3030 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
3031 buf = gfi_unpack_entry(oe, &size);
3032 } else {
3033 enum object_type unused;
3034 buf = read_object_file(oid, &unused, &size);
3036 if (!buf)
3037 die("Can't load object %s", oid_to_hex(oid));
3039 /* Peel one layer. */
3040 switch (oe->type) {
3041 case OBJ_TAG:
3042 if (size < GIT_SHA1_HEXSZ + strlen("object ") ||
3043 get_oid_hex(buf + strlen("object "), oid))
3044 die("Invalid SHA1 in tag: %s", command_buf.buf);
3045 break;
3046 case OBJ_COMMIT:
3047 if (size < GIT_SHA1_HEXSZ + strlen("tree ") ||
3048 get_oid_hex(buf + strlen("tree "), oid))
3049 die("Invalid SHA1 in commit: %s", command_buf.buf);
3052 free(buf);
3053 return find_object(oid);
3056 static struct object_entry *parse_treeish_dataref(const char **p)
3058 struct object_id oid;
3059 struct object_entry *e;
3061 if (**p == ':') { /* <mark> */
3062 e = find_mark(parse_mark_ref_space(p));
3063 if (!e)
3064 die("Unknown mark: %s", command_buf.buf);
3065 oidcpy(&oid, &e->idx.oid);
3066 } else { /* <sha1> */
3067 if (parse_oid_hex(*p, &oid, p))
3068 die("Invalid dataref: %s", command_buf.buf);
3069 e = find_object(&oid);
3070 if (*(*p)++ != ' ')
3071 die("Missing space after tree-ish: %s", command_buf.buf);
3074 while (!e || e->type != OBJ_TREE)
3075 e = dereference(e, &oid);
3076 return e;
3079 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3081 static struct strbuf line = STRBUF_INIT;
3083 /* See show_tree(). */
3084 const char *type =
3085 S_ISGITLINK(mode) ? commit_type :
3086 S_ISDIR(mode) ? tree_type :
3087 blob_type;
3089 if (!mode) {
3090 /* missing SP path LF */
3091 strbuf_reset(&line);
3092 strbuf_addstr(&line, "missing ");
3093 quote_c_style(path, &line, NULL, 0);
3094 strbuf_addch(&line, '\n');
3095 } else {
3096 /* mode SP type SP object_name TAB path LF */
3097 strbuf_reset(&line);
3098 strbuf_addf(&line, "%06o %s %s\t",
3099 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3100 quote_c_style(path, &line, NULL, 0);
3101 strbuf_addch(&line, '\n');
3103 cat_blob_write(line.buf, line.len);
3106 static void parse_ls(const char *p, struct branch *b)
3108 struct tree_entry *root = NULL;
3109 struct tree_entry leaf = {NULL};
3111 /* ls SP (<tree-ish> SP)? <path> */
3112 if (*p == '"') {
3113 if (!b)
3114 die("Not in a commit: %s", command_buf.buf);
3115 root = &b->branch_tree;
3116 } else {
3117 struct object_entry *e = parse_treeish_dataref(&p);
3118 root = new_tree_entry();
3119 oidcpy(&root->versions[1].oid, &e->idx.oid);
3120 if (!is_null_oid(&root->versions[1].oid))
3121 root->versions[1].mode = S_IFDIR;
3122 load_tree(root);
3124 if (*p == '"') {
3125 static struct strbuf uq = STRBUF_INIT;
3126 const char *endp;
3127 strbuf_reset(&uq);
3128 if (unquote_c_style(&uq, p, &endp))
3129 die("Invalid path: %s", command_buf.buf);
3130 if (*endp)
3131 die("Garbage after path in: %s", command_buf.buf);
3132 p = uq.buf;
3134 tree_content_get(root, p, &leaf, 1);
3136 * A directory in preparation would have a sha1 of zero
3137 * until it is saved. Save, for simplicity.
3139 if (S_ISDIR(leaf.versions[1].mode))
3140 store_tree(&leaf);
3142 print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3143 if (leaf.tree)
3144 release_tree_content_recursive(leaf.tree);
3145 if (!b || root != &b->branch_tree)
3146 release_tree_entry(root);
3149 static void checkpoint(void)
3151 checkpoint_requested = 0;
3152 if (object_count) {
3153 cycle_packfile();
3155 dump_branches();
3156 dump_tags();
3157 dump_marks();
3160 static void parse_checkpoint(void)
3162 checkpoint_requested = 1;
3163 skip_optional_lf();
3166 static void parse_progress(void)
3168 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3169 fputc('\n', stdout);
3170 fflush(stdout);
3171 skip_optional_lf();
3174 static char* make_fast_import_path(const char *path)
3176 if (!relative_marks_paths || is_absolute_path(path))
3177 return xstrdup(path);
3178 return git_pathdup("info/fast-import/%s", path);
3181 static void option_import_marks(const char *marks,
3182 int from_stream, int ignore_missing)
3184 if (import_marks_file) {
3185 if (from_stream)
3186 die("Only one import-marks command allowed per stream");
3188 /* read previous mark file */
3189 if(!import_marks_file_from_stream)
3190 read_marks();
3193 import_marks_file = make_fast_import_path(marks);
3194 safe_create_leading_directories_const(import_marks_file);
3195 import_marks_file_from_stream = from_stream;
3196 import_marks_file_ignore_missing = ignore_missing;
3199 static void option_date_format(const char *fmt)
3201 if (!strcmp(fmt, "raw"))
3202 whenspec = WHENSPEC_RAW;
3203 else if (!strcmp(fmt, "rfc2822"))
3204 whenspec = WHENSPEC_RFC2822;
3205 else if (!strcmp(fmt, "now"))
3206 whenspec = WHENSPEC_NOW;
3207 else
3208 die("unknown --date-format argument %s", fmt);
3211 static unsigned long ulong_arg(const char *option, const char *arg)
3213 char *endptr;
3214 unsigned long rv = strtoul(arg, &endptr, 0);
3215 if (strchr(arg, '-') || endptr == arg || *endptr)
3216 die("%s: argument must be a non-negative integer", option);
3217 return rv;
3220 static void option_depth(const char *depth)
3222 max_depth = ulong_arg("--depth", depth);
3223 if (max_depth > MAX_DEPTH)
3224 die("--depth cannot exceed %u", MAX_DEPTH);
3227 static void option_active_branches(const char *branches)
3229 max_active_branches = ulong_arg("--active-branches", branches);
3232 static void option_export_marks(const char *marks)
3234 export_marks_file = make_fast_import_path(marks);
3235 safe_create_leading_directories_const(export_marks_file);
3238 static void option_cat_blob_fd(const char *fd)
3240 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3241 if (n > (unsigned long) INT_MAX)
3242 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3243 cat_blob_fd = (int) n;
3246 static void option_export_pack_edges(const char *edges)
3248 if (pack_edges)
3249 fclose(pack_edges);
3250 pack_edges = xfopen(edges, "a");
3253 static int parse_one_option(const char *option)
3255 if (skip_prefix(option, "max-pack-size=", &option)) {
3256 unsigned long v;
3257 if (!git_parse_ulong(option, &v))
3258 return 0;
3259 if (v < 8192) {
3260 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3261 v *= 1024 * 1024;
3262 } else if (v < 1024 * 1024) {
3263 warning("minimum max-pack-size is 1 MiB");
3264 v = 1024 * 1024;
3266 max_packsize = v;
3267 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3268 unsigned long v;
3269 if (!git_parse_ulong(option, &v))
3270 return 0;
3271 big_file_threshold = v;
3272 } else if (skip_prefix(option, "depth=", &option)) {
3273 option_depth(option);
3274 } else if (skip_prefix(option, "active-branches=", &option)) {
3275 option_active_branches(option);
3276 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3277 option_export_pack_edges(option);
3278 } else if (starts_with(option, "quiet")) {
3279 show_stats = 0;
3280 } else if (starts_with(option, "stats")) {
3281 show_stats = 1;
3282 } else {
3283 return 0;
3286 return 1;
3289 static int parse_one_feature(const char *feature, int from_stream)
3291 const char *arg;
3293 if (skip_prefix(feature, "date-format=", &arg)) {
3294 option_date_format(arg);
3295 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3296 option_import_marks(arg, from_stream, 0);
3297 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3298 option_import_marks(arg, from_stream, 1);
3299 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3300 option_export_marks(arg);
3301 } else if (!strcmp(feature, "get-mark")) {
3302 ; /* Don't die - this feature is supported */
3303 } else if (!strcmp(feature, "cat-blob")) {
3304 ; /* Don't die - this feature is supported */
3305 } else if (!strcmp(feature, "relative-marks")) {
3306 relative_marks_paths = 1;
3307 } else if (!strcmp(feature, "no-relative-marks")) {
3308 relative_marks_paths = 0;
3309 } else if (!strcmp(feature, "done")) {
3310 require_explicit_termination = 1;
3311 } else if (!strcmp(feature, "force")) {
3312 force_update = 1;
3313 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3314 ; /* do nothing; we have the feature */
3315 } else {
3316 return 0;
3319 return 1;
3322 static void parse_feature(const char *feature)
3324 if (seen_data_command)
3325 die("Got feature command '%s' after data command", feature);
3327 if (parse_one_feature(feature, 1))
3328 return;
3330 die("This version of fast-import does not support feature %s.", feature);
3333 static void parse_option(const char *option)
3335 if (seen_data_command)
3336 die("Got option command '%s' after data command", option);
3338 if (parse_one_option(option))
3339 return;
3341 die("This version of fast-import does not support option: %s", option);
3344 static void git_pack_config(void)
3346 int indexversion_value;
3347 int limit;
3348 unsigned long packsizelimit_value;
3350 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3351 if (max_depth > MAX_DEPTH)
3352 max_depth = MAX_DEPTH;
3354 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3355 pack_idx_opts.version = indexversion_value;
3356 if (pack_idx_opts.version > 2)
3357 git_die_config("pack.indexversion",
3358 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3360 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3361 max_packsize = packsizelimit_value;
3363 if (!git_config_get_int("fastimport.unpacklimit", &limit))
3364 unpack_limit = limit;
3365 else if (!git_config_get_int("transfer.unpacklimit", &limit))
3366 unpack_limit = limit;
3368 git_config(git_default_config, NULL);
3371 static const char fast_import_usage[] =
3372 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3374 static void parse_argv(void)
3376 unsigned int i;
3378 for (i = 1; i < global_argc; i++) {
3379 const char *a = global_argv[i];
3381 if (*a != '-' || !strcmp(a, "--"))
3382 break;
3384 if (!skip_prefix(a, "--", &a))
3385 die("unknown option %s", a);
3387 if (parse_one_option(a))
3388 continue;
3390 if (parse_one_feature(a, 0))
3391 continue;
3393 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3394 option_cat_blob_fd(a);
3395 continue;
3398 die("unknown option --%s", a);
3400 if (i != global_argc)
3401 usage(fast_import_usage);
3403 seen_data_command = 1;
3404 if (import_marks_file)
3405 read_marks();
3408 int cmd_main(int argc, const char **argv)
3410 unsigned int i;
3412 if (argc == 2 && !strcmp(argv[1], "-h"))
3413 usage(fast_import_usage);
3415 setup_git_directory();
3416 reset_pack_idx_option(&pack_idx_opts);
3417 git_pack_config();
3419 alloc_objects(object_entry_alloc);
3420 strbuf_init(&command_buf, 0);
3421 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3422 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3423 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3424 marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3426 global_argc = argc;
3427 global_argv = argv;
3429 rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3430 for (i = 0; i < (cmd_save - 1); i++)
3431 rc_free[i].next = &rc_free[i + 1];
3432 rc_free[cmd_save - 1].next = NULL;
3434 start_packfile();
3435 set_die_routine(die_nicely);
3436 set_checkpoint_signal();
3437 while (read_next_command() != EOF) {
3438 const char *v;
3439 if (!strcmp("blob", command_buf.buf))
3440 parse_new_blob();
3441 else if (skip_prefix(command_buf.buf, "ls ", &v))
3442 parse_ls(v, NULL);
3443 else if (skip_prefix(command_buf.buf, "commit ", &v))
3444 parse_new_commit(v);
3445 else if (skip_prefix(command_buf.buf, "tag ", &v))
3446 parse_new_tag(v);
3447 else if (skip_prefix(command_buf.buf, "reset ", &v))
3448 parse_reset_branch(v);
3449 else if (!strcmp("checkpoint", command_buf.buf))
3450 parse_checkpoint();
3451 else if (!strcmp("done", command_buf.buf))
3452 break;
3453 else if (starts_with(command_buf.buf, "progress "))
3454 parse_progress();
3455 else if (skip_prefix(command_buf.buf, "feature ", &v))
3456 parse_feature(v);
3457 else if (skip_prefix(command_buf.buf, "option git ", &v))
3458 parse_option(v);
3459 else if (starts_with(command_buf.buf, "option "))
3460 /* ignore non-git options*/;
3461 else
3462 die("Unsupported command: %s", command_buf.buf);
3464 if (checkpoint_requested)
3465 checkpoint();
3468 /* argv hasn't been parsed yet, do so */
3469 if (!seen_data_command)
3470 parse_argv();
3472 if (require_explicit_termination && feof(stdin))
3473 die("stream ends early");
3475 end_packfile();
3477 dump_branches();
3478 dump_tags();
3479 unkeep_all_packs();
3480 dump_marks();
3482 if (pack_edges)
3483 fclose(pack_edges);
3485 if (show_stats) {
3486 uintmax_t total_count = 0, duplicate_count = 0;
3487 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3488 total_count += object_count_by_type[i];
3489 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3490 duplicate_count += duplicate_count_by_type[i];
3492 fprintf(stderr, "%s statistics:\n", argv[0]);
3493 fprintf(stderr, "---------------------------------------------------------------------\n");
3494 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3495 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3496 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]);
3497 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]);
3498 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]);
3499 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]);
3500 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3501 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3502 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3503 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3504 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3505 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3506 fprintf(stderr, "---------------------------------------------------------------------\n");
3507 pack_report();
3508 fprintf(stderr, "---------------------------------------------------------------------\n");
3509 fprintf(stderr, "\n");
3512 return failure ? 1 : 0;