Merge branch 'bs/maint-commit-options'
[git] / fast-import.c
1 /*
2 Format of STDIN stream:
3
4   stream ::= cmd*;
5
6   cmd ::= new_blob
7         | new_commit
8         | new_tag
9         | reset_branch
10         | checkpoint
11         | progress
12         ;
13
14   new_blob ::= 'blob' lf
15     mark?
16     file_content;
17   file_content ::= data;
18
19   new_commit ::= 'commit' sp ref_str lf
20     mark?
21     ('author' sp name '<' email '>' when lf)?
22     'committer' sp name '<' email '>' when lf
23     commit_msg
24     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
25     ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
26     file_change*
27     lf?;
28   commit_msg ::= data;
29
30   file_change ::= file_clr
31     | file_del
32     | file_rnm
33     | file_cpy
34     | file_obm
35     | file_inm;
36   file_clr ::= 'deleteall' lf;
37   file_del ::= 'D' sp path_str lf;
38   file_rnm ::= 'R' sp path_str sp path_str lf;
39   file_cpy ::= 'C' sp path_str sp path_str lf;
40   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
41   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
42     data;
43
44   new_tag ::= 'tag' sp tag_str lf
45     'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
46     'tagger' sp name '<' email '>' when lf
47     tag_msg;
48   tag_msg ::= data;
49
50   reset_branch ::= 'reset' sp ref_str lf
51     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
52     lf?;
53
54   checkpoint ::= 'checkpoint' lf
55     lf?;
56
57   progress ::= 'progress' sp not_lf* lf
58     lf?;
59
60      # note: the first idnum in a stream should be 1 and subsequent
61      # idnums should not have gaps between values as this will cause
62      # the stream parser to reserve space for the gapped values.  An
63      # idnum can be updated in the future to a new object by issuing
64      # a new mark directive with the old idnum.
65      #
66   mark ::= 'mark' sp idnum lf;
67   data ::= (delimited_data | exact_data)
68     lf?;
69
70     # note: delim may be any string but must not contain lf.
71     # data_line may contain any data but must not be exactly
72     # delim.
73   delimited_data ::= 'data' sp '<<' delim lf
74     (data_line lf)*
75     delim lf;
76
77      # note: declen indicates the length of binary_data in bytes.
78      # declen does not include the lf preceeding the binary data.
79      #
80   exact_data ::= 'data' sp declen lf
81     binary_data;
82
83      # note: quoted strings are C-style quoting supporting \c for
84      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
85      # is the signed byte value in octal.  Note that the only
86      # characters which must actually be escaped to protect the
87      # stream formatting is: \, " and LF.  Otherwise these values
88      # are UTF8.
89      #
90   ref_str     ::= ref;
91   sha1exp_str ::= sha1exp;
92   tag_str     ::= tag;
93   path_str    ::= path    | '"' quoted(path)    '"' ;
94   mode        ::= '100644' | '644'
95                 | '100755' | '755'
96                 | '120000'
97                 ;
98
99   declen ::= # unsigned 32 bit value, ascii base10 notation;
100   bigint ::= # unsigned integer value, ascii base10 notation;
101   binary_data ::= # file content, not interpreted;
102
103   when         ::= raw_when | rfc2822_when;
104   raw_when     ::= ts sp tz;
105   rfc2822_when ::= # Valid RFC 2822 date and time;
106
107   sp ::= # ASCII space character;
108   lf ::= # ASCII newline (LF) character;
109
110      # note: a colon (':') must precede the numerical value assigned to
111      # an idnum.  This is to distinguish it from a ref or tag name as
112      # GIT does not permit ':' in ref or tag strings.
113      #
114   idnum   ::= ':' bigint;
115   path    ::= # GIT style file path, e.g. "a/b/c";
116   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
117   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
118   sha1exp ::= # Any valid GIT SHA1 expression;
119   hexsha1 ::= # SHA1 in hexadecimal format;
120
121      # note: name and email are UTF8 strings, however name must not
122      # contain '<' or lf and email must not contain any of the
123      # following: '<', '>', lf.
124      #
125   name  ::= # valid GIT author/committer name;
126   email ::= # valid GIT author/committer email;
127   ts    ::= # time since the epoch in seconds, ascii base10 notation;
128   tz    ::= # GIT style timezone;
129
130      # note: comments may appear anywhere in the input, except
131      # within a data command.  Any form of the data command
132      # always escapes the related input from comment processing.
133      #
134      # In case it is not clear, the '#' that starts the comment
135      # must be the first character on that the line (an lf have
136      # preceeded it).
137      #
138   comment ::= '#' not_lf* lf;
139   not_lf  ::= # Any byte that is not ASCII newline (LF);
140 */
141
142 #include "builtin.h"
143 #include "cache.h"
144 #include "object.h"
145 #include "blob.h"
146 #include "tree.h"
147 #include "commit.h"
148 #include "delta.h"
149 #include "pack.h"
150 #include "refs.h"
151 #include "csum-file.h"
152 #include "quote.h"
153
154 #define PACK_ID_BITS 16
155 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
156 #define DEPTH_BITS 13
157 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
158
159 struct object_entry
160 {
161         struct object_entry *next;
162         uint32_t offset;
163         uint32_t type : TYPE_BITS,
164                 pack_id : PACK_ID_BITS,
165                 depth : DEPTH_BITS;
166         unsigned char sha1[20];
167 };
168
169 struct object_entry_pool
170 {
171         struct object_entry_pool *next_pool;
172         struct object_entry *next_free;
173         struct object_entry *end;
174         struct object_entry entries[FLEX_ARRAY]; /* more */
175 };
176
177 struct mark_set
178 {
179         union {
180                 struct object_entry *marked[1024];
181                 struct mark_set *sets[1024];
182         } data;
183         unsigned int shift;
184 };
185
186 struct last_object
187 {
188         struct strbuf data;
189         uint32_t offset;
190         unsigned int depth;
191         unsigned no_swap : 1;
192 };
193
194 struct mem_pool
195 {
196         struct mem_pool *next_pool;
197         char *next_free;
198         char *end;
199         char space[FLEX_ARRAY]; /* more */
200 };
201
202 struct atom_str
203 {
204         struct atom_str *next_atom;
205         unsigned short str_len;
206         char str_dat[FLEX_ARRAY]; /* more */
207 };
208
209 struct tree_content;
210 struct tree_entry
211 {
212         struct tree_content *tree;
213         struct atom_str* name;
214         struct tree_entry_ms
215         {
216                 uint16_t mode;
217                 unsigned char sha1[20];
218         } versions[2];
219 };
220
221 struct tree_content
222 {
223         unsigned int entry_capacity; /* must match avail_tree_content */
224         unsigned int entry_count;
225         unsigned int delta_depth;
226         struct tree_entry *entries[FLEX_ARRAY]; /* more */
227 };
228
229 struct avail_tree_content
230 {
231         unsigned int entry_capacity; /* must match tree_content */
232         struct avail_tree_content *next_avail;
233 };
234
235 struct branch
236 {
237         struct branch *table_next_branch;
238         struct branch *active_next_branch;
239         const char *name;
240         struct tree_entry branch_tree;
241         uintmax_t last_commit;
242         unsigned active : 1;
243         unsigned pack_id : PACK_ID_BITS;
244         unsigned char sha1[20];
245 };
246
247 struct tag
248 {
249         struct tag *next_tag;
250         const char *name;
251         unsigned int pack_id;
252         unsigned char sha1[20];
253 };
254
255 struct hash_list
256 {
257         struct hash_list *next;
258         unsigned char sha1[20];
259 };
260
261 typedef enum {
262         WHENSPEC_RAW = 1,
263         WHENSPEC_RFC2822,
264         WHENSPEC_NOW,
265 } whenspec_type;
266
267 struct recent_command
268 {
269         struct recent_command *prev;
270         struct recent_command *next;
271         char *buf;
272 };
273
274 /* Configured limits on output */
275 static unsigned long max_depth = 10;
276 static off_t max_packsize = (1LL << 32) - 1;
277 static int force_update;
278
279 /* Stats and misc. counters */
280 static uintmax_t alloc_count;
281 static uintmax_t marks_set_count;
282 static uintmax_t object_count_by_type[1 << TYPE_BITS];
283 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
284 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
285 static unsigned long object_count;
286 static unsigned long branch_count;
287 static unsigned long branch_load_count;
288 static int failure;
289 static FILE *pack_edges;
290
291 /* Memory pools */
292 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
293 static size_t total_allocd;
294 static struct mem_pool *mem_pool;
295
296 /* Atom management */
297 static unsigned int atom_table_sz = 4451;
298 static unsigned int atom_cnt;
299 static struct atom_str **atom_table;
300
301 /* The .pack file being generated */
302 static unsigned int pack_id;
303 static struct packed_git *pack_data;
304 static struct packed_git **all_packs;
305 static unsigned long pack_size;
306
307 /* Table of objects we've written. */
308 static unsigned int object_entry_alloc = 5000;
309 static struct object_entry_pool *blocks;
310 static struct object_entry *object_table[1 << 16];
311 static struct mark_set *marks;
312 static const char* mark_file;
313
314 /* Our last blob */
315 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
316
317 /* Tree management */
318 static unsigned int tree_entry_alloc = 1000;
319 static void *avail_tree_entry;
320 static unsigned int avail_tree_table_sz = 100;
321 static struct avail_tree_content **avail_tree_table;
322 static struct strbuf old_tree = STRBUF_INIT;
323 static struct strbuf new_tree = STRBUF_INIT;
324
325 /* Branch data */
326 static unsigned long max_active_branches = 5;
327 static unsigned long cur_active_branches;
328 static unsigned long branch_table_sz = 1039;
329 static struct branch **branch_table;
330 static struct branch *active_branches;
331
332 /* Tag data */
333 static struct tag *first_tag;
334 static struct tag *last_tag;
335
336 /* Input stream parsing */
337 static whenspec_type whenspec = WHENSPEC_RAW;
338 static struct strbuf command_buf = STRBUF_INIT;
339 static int unread_command_buf;
340 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
341 static struct recent_command *cmd_tail = &cmd_hist;
342 static struct recent_command *rc_free;
343 static unsigned int cmd_save = 100;
344 static uintmax_t next_mark;
345 static struct strbuf new_data = STRBUF_INIT;
346
347 static void write_branch_report(FILE *rpt, struct branch *b)
348 {
349         fprintf(rpt, "%s:\n", b->name);
350
351         fprintf(rpt, "  status      :");
352         if (b->active)
353                 fputs(" active", rpt);
354         if (b->branch_tree.tree)
355                 fputs(" loaded", rpt);
356         if (is_null_sha1(b->branch_tree.versions[1].sha1))
357                 fputs(" dirty", rpt);
358         fputc('\n', rpt);
359
360         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
361         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
362         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
363         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
364
365         fputs("  last pack   : ", rpt);
366         if (b->pack_id < MAX_PACK_ID)
367                 fprintf(rpt, "%u", b->pack_id);
368         fputc('\n', rpt);
369
370         fputc('\n', rpt);
371 }
372
373 static void write_crash_report(const char *err)
374 {
375         char *loc = git_path("fast_import_crash_%d", getpid());
376         FILE *rpt = fopen(loc, "w");
377         struct branch *b;
378         unsigned long lu;
379         struct recent_command *rc;
380
381         if (!rpt) {
382                 error("can't write crash report %s: %s", loc, strerror(errno));
383                 return;
384         }
385
386         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
387
388         fprintf(rpt, "fast-import crash report:\n");
389         fprintf(rpt, "    fast-import process: %d\n", getpid());
390         fprintf(rpt, "    parent process     : %d\n", getppid());
391         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
392         fputc('\n', rpt);
393
394         fputs("fatal: ", rpt);
395         fputs(err, rpt);
396         fputc('\n', rpt);
397
398         fputc('\n', rpt);
399         fputs("Most Recent Commands Before Crash\n", rpt);
400         fputs("---------------------------------\n", rpt);
401         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
402                 if (rc->next == &cmd_hist)
403                         fputs("* ", rpt);
404                 else
405                         fputs("  ", rpt);
406                 fputs(rc->buf, rpt);
407                 fputc('\n', rpt);
408         }
409
410         fputc('\n', rpt);
411         fputs("Active Branch LRU\n", rpt);
412         fputs("-----------------\n", rpt);
413         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
414                 cur_active_branches,
415                 max_active_branches);
416         fputc('\n', rpt);
417         fputs("  pos  clock name\n", rpt);
418         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
419         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
420                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
421                         ++lu, b->last_commit, b->name);
422
423         fputc('\n', rpt);
424         fputs("Inactive Branches\n", rpt);
425         fputs("-----------------\n", rpt);
426         for (lu = 0; lu < branch_table_sz; lu++) {
427                 for (b = branch_table[lu]; b; b = b->table_next_branch)
428                         write_branch_report(rpt, b);
429         }
430
431         fputc('\n', rpt);
432         fputs("-------------------\n", rpt);
433         fputs("END OF CRASH REPORT\n", rpt);
434         fclose(rpt);
435 }
436
437 static NORETURN void die_nicely(const char *err, va_list params)
438 {
439         static int zombie;
440         char message[2 * PATH_MAX];
441
442         vsnprintf(message, sizeof(message), err, params);
443         fputs("fatal: ", stderr);
444         fputs(message, stderr);
445         fputc('\n', stderr);
446
447         if (!zombie) {
448                 zombie = 1;
449                 write_crash_report(message);
450         }
451         exit(128);
452 }
453
454 static void alloc_objects(unsigned int cnt)
455 {
456         struct object_entry_pool *b;
457
458         b = xmalloc(sizeof(struct object_entry_pool)
459                 + cnt * sizeof(struct object_entry));
460         b->next_pool = blocks;
461         b->next_free = b->entries;
462         b->end = b->entries + cnt;
463         blocks = b;
464         alloc_count += cnt;
465 }
466
467 static struct object_entry *new_object(unsigned char *sha1)
468 {
469         struct object_entry *e;
470
471         if (blocks->next_free == blocks->end)
472                 alloc_objects(object_entry_alloc);
473
474         e = blocks->next_free++;
475         hashcpy(e->sha1, sha1);
476         return e;
477 }
478
479 static struct object_entry *find_object(unsigned char *sha1)
480 {
481         unsigned int h = sha1[0] << 8 | sha1[1];
482         struct object_entry *e;
483         for (e = object_table[h]; e; e = e->next)
484                 if (!hashcmp(sha1, e->sha1))
485                         return e;
486         return NULL;
487 }
488
489 static struct object_entry *insert_object(unsigned char *sha1)
490 {
491         unsigned int h = sha1[0] << 8 | sha1[1];
492         struct object_entry *e = object_table[h];
493         struct object_entry *p = NULL;
494
495         while (e) {
496                 if (!hashcmp(sha1, e->sha1))
497                         return e;
498                 p = e;
499                 e = e->next;
500         }
501
502         e = new_object(sha1);
503         e->next = NULL;
504         e->offset = 0;
505         if (p)
506                 p->next = e;
507         else
508                 object_table[h] = e;
509         return e;
510 }
511
512 static unsigned int hc_str(const char *s, size_t len)
513 {
514         unsigned int r = 0;
515         while (len-- > 0)
516                 r = r * 31 + *s++;
517         return r;
518 }
519
520 static void *pool_alloc(size_t len)
521 {
522         struct mem_pool *p;
523         void *r;
524
525         for (p = mem_pool; p; p = p->next_pool)
526                 if ((p->end - p->next_free >= len))
527                         break;
528
529         if (!p) {
530                 if (len >= (mem_pool_alloc/2)) {
531                         total_allocd += len;
532                         return xmalloc(len);
533                 }
534                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
535                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
536                 p->next_pool = mem_pool;
537                 p->next_free = p->space;
538                 p->end = p->next_free + mem_pool_alloc;
539                 mem_pool = p;
540         }
541
542         r = p->next_free;
543         /* round out to a pointer alignment */
544         if (len & (sizeof(void*) - 1))
545                 len += sizeof(void*) - (len & (sizeof(void*) - 1));
546         p->next_free += len;
547         return r;
548 }
549
550 static void *pool_calloc(size_t count, size_t size)
551 {
552         size_t len = count * size;
553         void *r = pool_alloc(len);
554         memset(r, 0, len);
555         return r;
556 }
557
558 static char *pool_strdup(const char *s)
559 {
560         char *r = pool_alloc(strlen(s) + 1);
561         strcpy(r, s);
562         return r;
563 }
564
565 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
566 {
567         struct mark_set *s = marks;
568         while ((idnum >> s->shift) >= 1024) {
569                 s = pool_calloc(1, sizeof(struct mark_set));
570                 s->shift = marks->shift + 10;
571                 s->data.sets[0] = marks;
572                 marks = s;
573         }
574         while (s->shift) {
575                 uintmax_t i = idnum >> s->shift;
576                 idnum -= i << s->shift;
577                 if (!s->data.sets[i]) {
578                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
579                         s->data.sets[i]->shift = s->shift - 10;
580                 }
581                 s = s->data.sets[i];
582         }
583         if (!s->data.marked[idnum])
584                 marks_set_count++;
585         s->data.marked[idnum] = oe;
586 }
587
588 static struct object_entry *find_mark(uintmax_t idnum)
589 {
590         uintmax_t orig_idnum = idnum;
591         struct mark_set *s = marks;
592         struct object_entry *oe = NULL;
593         if ((idnum >> s->shift) < 1024) {
594                 while (s && s->shift) {
595                         uintmax_t i = idnum >> s->shift;
596                         idnum -= i << s->shift;
597                         s = s->data.sets[i];
598                 }
599                 if (s)
600                         oe = s->data.marked[idnum];
601         }
602         if (!oe)
603                 die("mark :%" PRIuMAX " not declared", orig_idnum);
604         return oe;
605 }
606
607 static struct atom_str *to_atom(const char *s, unsigned short len)
608 {
609         unsigned int hc = hc_str(s, len) % atom_table_sz;
610         struct atom_str *c;
611
612         for (c = atom_table[hc]; c; c = c->next_atom)
613                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
614                         return c;
615
616         c = pool_alloc(sizeof(struct atom_str) + len + 1);
617         c->str_len = len;
618         strncpy(c->str_dat, s, len);
619         c->str_dat[len] = 0;
620         c->next_atom = atom_table[hc];
621         atom_table[hc] = c;
622         atom_cnt++;
623         return c;
624 }
625
626 static struct branch *lookup_branch(const char *name)
627 {
628         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
629         struct branch *b;
630
631         for (b = branch_table[hc]; b; b = b->table_next_branch)
632                 if (!strcmp(name, b->name))
633                         return b;
634         return NULL;
635 }
636
637 static struct branch *new_branch(const char *name)
638 {
639         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
640         struct branch* b = lookup_branch(name);
641
642         if (b)
643                 die("Invalid attempt to create duplicate branch: %s", name);
644         switch (check_ref_format(name)) {
645         case  0: break; /* its valid */
646         case -2: break; /* valid, but too few '/', allow anyway */
647         default:
648                 die("Branch name doesn't conform to GIT standards: %s", name);
649         }
650
651         b = pool_calloc(1, sizeof(struct branch));
652         b->name = pool_strdup(name);
653         b->table_next_branch = branch_table[hc];
654         b->branch_tree.versions[0].mode = S_IFDIR;
655         b->branch_tree.versions[1].mode = S_IFDIR;
656         b->active = 0;
657         b->pack_id = MAX_PACK_ID;
658         branch_table[hc] = b;
659         branch_count++;
660         return b;
661 }
662
663 static unsigned int hc_entries(unsigned int cnt)
664 {
665         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
666         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
667 }
668
669 static struct tree_content *new_tree_content(unsigned int cnt)
670 {
671         struct avail_tree_content *f, *l = NULL;
672         struct tree_content *t;
673         unsigned int hc = hc_entries(cnt);
674
675         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
676                 if (f->entry_capacity >= cnt)
677                         break;
678
679         if (f) {
680                 if (l)
681                         l->next_avail = f->next_avail;
682                 else
683                         avail_tree_table[hc] = f->next_avail;
684         } else {
685                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
686                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
687                 f->entry_capacity = cnt;
688         }
689
690         t = (struct tree_content*)f;
691         t->entry_count = 0;
692         t->delta_depth = 0;
693         return t;
694 }
695
696 static void release_tree_entry(struct tree_entry *e);
697 static void release_tree_content(struct tree_content *t)
698 {
699         struct avail_tree_content *f = (struct avail_tree_content*)t;
700         unsigned int hc = hc_entries(f->entry_capacity);
701         f->next_avail = avail_tree_table[hc];
702         avail_tree_table[hc] = f;
703 }
704
705 static void release_tree_content_recursive(struct tree_content *t)
706 {
707         unsigned int i;
708         for (i = 0; i < t->entry_count; i++)
709                 release_tree_entry(t->entries[i]);
710         release_tree_content(t);
711 }
712
713 static struct tree_content *grow_tree_content(
714         struct tree_content *t,
715         int amt)
716 {
717         struct tree_content *r = new_tree_content(t->entry_count + amt);
718         r->entry_count = t->entry_count;
719         r->delta_depth = t->delta_depth;
720         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
721         release_tree_content(t);
722         return r;
723 }
724
725 static struct tree_entry *new_tree_entry(void)
726 {
727         struct tree_entry *e;
728
729         if (!avail_tree_entry) {
730                 unsigned int n = tree_entry_alloc;
731                 total_allocd += n * sizeof(struct tree_entry);
732                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
733                 while (n-- > 1) {
734                         *((void**)e) = e + 1;
735                         e++;
736                 }
737                 *((void**)e) = NULL;
738         }
739
740         e = avail_tree_entry;
741         avail_tree_entry = *((void**)e);
742         return e;
743 }
744
745 static void release_tree_entry(struct tree_entry *e)
746 {
747         if (e->tree)
748                 release_tree_content_recursive(e->tree);
749         *((void**)e) = avail_tree_entry;
750         avail_tree_entry = e;
751 }
752
753 static struct tree_content *dup_tree_content(struct tree_content *s)
754 {
755         struct tree_content *d;
756         struct tree_entry *a, *b;
757         unsigned int i;
758
759         if (!s)
760                 return NULL;
761         d = new_tree_content(s->entry_count);
762         for (i = 0; i < s->entry_count; i++) {
763                 a = s->entries[i];
764                 b = new_tree_entry();
765                 memcpy(b, a, sizeof(*a));
766                 if (a->tree && is_null_sha1(b->versions[1].sha1))
767                         b->tree = dup_tree_content(a->tree);
768                 else
769                         b->tree = NULL;
770                 d->entries[i] = b;
771         }
772         d->entry_count = s->entry_count;
773         d->delta_depth = s->delta_depth;
774
775         return d;
776 }
777
778 static void start_packfile(void)
779 {
780         static char tmpfile[PATH_MAX];
781         struct packed_git *p;
782         struct pack_header hdr;
783         int pack_fd;
784
785         snprintf(tmpfile, sizeof(tmpfile),
786                 "%s/tmp_pack_XXXXXX", get_object_directory());
787         pack_fd = xmkstemp(tmpfile);
788         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
789         strcpy(p->pack_name, tmpfile);
790         p->pack_fd = pack_fd;
791
792         hdr.hdr_signature = htonl(PACK_SIGNATURE);
793         hdr.hdr_version = htonl(2);
794         hdr.hdr_entries = 0;
795         write_or_die(p->pack_fd, &hdr, sizeof(hdr));
796
797         pack_data = p;
798         pack_size = sizeof(hdr);
799         object_count = 0;
800
801         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
802         all_packs[pack_id] = p;
803 }
804
805 static int oecmp (const void *a_, const void *b_)
806 {
807         struct object_entry *a = *((struct object_entry**)a_);
808         struct object_entry *b = *((struct object_entry**)b_);
809         return hashcmp(a->sha1, b->sha1);
810 }
811
812 static char *create_index(void)
813 {
814         static char tmpfile[PATH_MAX];
815         SHA_CTX ctx;
816         struct sha1file *f;
817         struct object_entry **idx, **c, **last, *e;
818         struct object_entry_pool *o;
819         uint32_t array[256];
820         int i, idx_fd;
821
822         /* Build the sorted table of object IDs. */
823         idx = xmalloc(object_count * sizeof(struct object_entry*));
824         c = idx;
825         for (o = blocks; o; o = o->next_pool)
826                 for (e = o->next_free; e-- != o->entries;)
827                         if (pack_id == e->pack_id)
828                                 *c++ = e;
829         last = idx + object_count;
830         if (c != last)
831                 die("internal consistency error creating the index");
832         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
833
834         /* Generate the fan-out array. */
835         c = idx;
836         for (i = 0; i < 256; i++) {
837                 struct object_entry **next = c;;
838                 while (next < last) {
839                         if ((*next)->sha1[0] != i)
840                                 break;
841                         next++;
842                 }
843                 array[i] = htonl(next - idx);
844                 c = next;
845         }
846
847         snprintf(tmpfile, sizeof(tmpfile),
848                 "%s/tmp_idx_XXXXXX", get_object_directory());
849         idx_fd = xmkstemp(tmpfile);
850         f = sha1fd(idx_fd, tmpfile);
851         sha1write(f, array, 256 * sizeof(int));
852         SHA1_Init(&ctx);
853         for (c = idx; c != last; c++) {
854                 uint32_t offset = htonl((*c)->offset);
855                 sha1write(f, &offset, 4);
856                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
857                 SHA1_Update(&ctx, (*c)->sha1, 20);
858         }
859         sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
860         sha1close(f, NULL, 1);
861         free(idx);
862         SHA1_Final(pack_data->sha1, &ctx);
863         return tmpfile;
864 }
865
866 static char *keep_pack(char *curr_index_name)
867 {
868         static char name[PATH_MAX];
869         static const char *keep_msg = "fast-import";
870         int keep_fd;
871
872         chmod(pack_data->pack_name, 0444);
873         chmod(curr_index_name, 0444);
874
875         snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
876                  get_object_directory(), sha1_to_hex(pack_data->sha1));
877         keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
878         if (keep_fd < 0)
879                 die("cannot create keep file");
880         write(keep_fd, keep_msg, strlen(keep_msg));
881         close(keep_fd);
882
883         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
884                  get_object_directory(), sha1_to_hex(pack_data->sha1));
885         if (move_temp_to_file(pack_data->pack_name, name))
886                 die("cannot store pack file");
887
888         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
889                  get_object_directory(), sha1_to_hex(pack_data->sha1));
890         if (move_temp_to_file(curr_index_name, name))
891                 die("cannot store index file");
892         return name;
893 }
894
895 static void unkeep_all_packs(void)
896 {
897         static char name[PATH_MAX];
898         int k;
899
900         for (k = 0; k < pack_id; k++) {
901                 struct packed_git *p = all_packs[k];
902                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
903                          get_object_directory(), sha1_to_hex(p->sha1));
904                 unlink(name);
905         }
906 }
907
908 static void end_packfile(void)
909 {
910         struct packed_git *old_p = pack_data, *new_p;
911
912         if (object_count) {
913                 char *idx_name;
914                 int i;
915                 struct branch *b;
916                 struct tag *t;
917
918                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
919                                     pack_data->pack_name, object_count);
920                 close(pack_data->pack_fd);
921                 idx_name = keep_pack(create_index());
922
923                 /* Register the packfile with core git's machinary. */
924                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
925                 if (!new_p)
926                         die("core git rejected index %s", idx_name);
927                 new_p->windows = old_p->windows;
928                 all_packs[pack_id] = new_p;
929                 install_packed_git(new_p);
930
931                 /* Print the boundary */
932                 if (pack_edges) {
933                         fprintf(pack_edges, "%s:", new_p->pack_name);
934                         for (i = 0; i < branch_table_sz; i++) {
935                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
936                                         if (b->pack_id == pack_id)
937                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
938                                 }
939                         }
940                         for (t = first_tag; t; t = t->next_tag) {
941                                 if (t->pack_id == pack_id)
942                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
943                         }
944                         fputc('\n', pack_edges);
945                         fflush(pack_edges);
946                 }
947
948                 pack_id++;
949         }
950         else
951                 unlink(old_p->pack_name);
952         free(old_p);
953
954         /* We can't carry a delta across packfiles. */
955         strbuf_release(&last_blob.data);
956         last_blob.offset = 0;
957         last_blob.depth = 0;
958 }
959
960 static void cycle_packfile(void)
961 {
962         end_packfile();
963         start_packfile();
964 }
965
966 static size_t encode_header(
967         enum object_type type,
968         size_t size,
969         unsigned char *hdr)
970 {
971         int n = 1;
972         unsigned char c;
973
974         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
975                 die("bad type %d", type);
976
977         c = (type << 4) | (size & 15);
978         size >>= 4;
979         while (size) {
980                 *hdr++ = c | 0x80;
981                 c = size & 0x7f;
982                 size >>= 7;
983                 n++;
984         }
985         *hdr = c;
986         return n;
987 }
988
989 static int store_object(
990         enum object_type type,
991         struct strbuf *dat,
992         struct last_object *last,
993         unsigned char *sha1out,
994         uintmax_t mark)
995 {
996         void *out, *delta;
997         struct object_entry *e;
998         unsigned char hdr[96];
999         unsigned char sha1[20];
1000         unsigned long hdrlen, deltalen;
1001         SHA_CTX c;
1002         z_stream s;
1003
1004         hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1005                 (unsigned long)dat->len) + 1;
1006         SHA1_Init(&c);
1007         SHA1_Update(&c, hdr, hdrlen);
1008         SHA1_Update(&c, dat->buf, dat->len);
1009         SHA1_Final(sha1, &c);
1010         if (sha1out)
1011                 hashcpy(sha1out, sha1);
1012
1013         e = insert_object(sha1);
1014         if (mark)
1015                 insert_mark(mark, e);
1016         if (e->offset) {
1017                 duplicate_count_by_type[type]++;
1018                 return 1;
1019         } else if (find_sha1_pack(sha1, packed_git)) {
1020                 e->type = type;
1021                 e->pack_id = MAX_PACK_ID;
1022                 e->offset = 1; /* just not zero! */
1023                 duplicate_count_by_type[type]++;
1024                 return 1;
1025         }
1026
1027         if (last && last->data.buf && last->depth < max_depth) {
1028                 delta = diff_delta(last->data.buf, last->data.len,
1029                         dat->buf, dat->len,
1030                         &deltalen, 0);
1031                 if (delta && deltalen >= dat->len) {
1032                         free(delta);
1033                         delta = NULL;
1034                 }
1035         } else
1036                 delta = NULL;
1037
1038         memset(&s, 0, sizeof(s));
1039         deflateInit(&s, zlib_compression_level);
1040         if (delta) {
1041                 s.next_in = delta;
1042                 s.avail_in = deltalen;
1043         } else {
1044                 s.next_in = (void *)dat->buf;
1045                 s.avail_in = dat->len;
1046         }
1047         s.avail_out = deflateBound(&s, s.avail_in);
1048         s.next_out = out = xmalloc(s.avail_out);
1049         while (deflate(&s, Z_FINISH) == Z_OK)
1050                 /* nothing */;
1051         deflateEnd(&s);
1052
1053         /* Determine if we should auto-checkpoint. */
1054         if ((pack_size + 60 + s.total_out) > max_packsize
1055                 || (pack_size + 60 + s.total_out) < pack_size) {
1056
1057                 /* This new object needs to *not* have the current pack_id. */
1058                 e->pack_id = pack_id + 1;
1059                 cycle_packfile();
1060
1061                 /* We cannot carry a delta into the new pack. */
1062                 if (delta) {
1063                         free(delta);
1064                         delta = NULL;
1065
1066                         memset(&s, 0, sizeof(s));
1067                         deflateInit(&s, zlib_compression_level);
1068                         s.next_in = (void *)dat->buf;
1069                         s.avail_in = dat->len;
1070                         s.avail_out = deflateBound(&s, s.avail_in);
1071                         s.next_out = out = xrealloc(out, s.avail_out);
1072                         while (deflate(&s, Z_FINISH) == Z_OK)
1073                                 /* nothing */;
1074                         deflateEnd(&s);
1075                 }
1076         }
1077
1078         e->type = type;
1079         e->pack_id = pack_id;
1080         e->offset = pack_size;
1081         object_count++;
1082         object_count_by_type[type]++;
1083
1084         if (delta) {
1085                 unsigned long ofs = e->offset - last->offset;
1086                 unsigned pos = sizeof(hdr) - 1;
1087
1088                 delta_count_by_type[type]++;
1089                 e->depth = last->depth + 1;
1090
1091                 hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1092                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1093                 pack_size += hdrlen;
1094
1095                 hdr[pos] = ofs & 127;
1096                 while (ofs >>= 7)
1097                         hdr[--pos] = 128 | (--ofs & 127);
1098                 write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1099                 pack_size += sizeof(hdr) - pos;
1100         } else {
1101                 e->depth = 0;
1102                 hdrlen = encode_header(type, dat->len, hdr);
1103                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1104                 pack_size += hdrlen;
1105         }
1106
1107         write_or_die(pack_data->pack_fd, out, s.total_out);
1108         pack_size += s.total_out;
1109
1110         free(out);
1111         free(delta);
1112         if (last) {
1113                 if (last->no_swap) {
1114                         last->data = *dat;
1115                 } else {
1116                         strbuf_swap(&last->data, dat);
1117                 }
1118                 last->offset = e->offset;
1119                 last->depth = e->depth;
1120         }
1121         return 0;
1122 }
1123
1124 static void *gfi_unpack_entry(
1125         struct object_entry *oe,
1126         unsigned long *sizep)
1127 {
1128         enum object_type type;
1129         struct packed_git *p = all_packs[oe->pack_id];
1130         if (p == pack_data)
1131                 p->pack_size = pack_size + 20;
1132         return unpack_entry(p, oe->offset, &type, sizep);
1133 }
1134
1135 static const char *get_mode(const char *str, uint16_t *modep)
1136 {
1137         unsigned char c;
1138         uint16_t mode = 0;
1139
1140         while ((c = *str++) != ' ') {
1141                 if (c < '0' || c > '7')
1142                         return NULL;
1143                 mode = (mode << 3) + (c - '0');
1144         }
1145         *modep = mode;
1146         return str;
1147 }
1148
1149 static void load_tree(struct tree_entry *root)
1150 {
1151         unsigned char* sha1 = root->versions[1].sha1;
1152         struct object_entry *myoe;
1153         struct tree_content *t;
1154         unsigned long size;
1155         char *buf;
1156         const char *c;
1157
1158         root->tree = t = new_tree_content(8);
1159         if (is_null_sha1(sha1))
1160                 return;
1161
1162         myoe = find_object(sha1);
1163         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1164                 if (myoe->type != OBJ_TREE)
1165                         die("Not a tree: %s", sha1_to_hex(sha1));
1166                 t->delta_depth = myoe->depth;
1167                 buf = gfi_unpack_entry(myoe, &size);
1168         } else {
1169                 enum object_type type;
1170                 buf = read_sha1_file(sha1, &type, &size);
1171                 if (!buf || type != OBJ_TREE)
1172                         die("Can't load tree %s", sha1_to_hex(sha1));
1173         }
1174
1175         c = buf;
1176         while (c != (buf + size)) {
1177                 struct tree_entry *e = new_tree_entry();
1178
1179                 if (t->entry_count == t->entry_capacity)
1180                         root->tree = t = grow_tree_content(t, t->entry_count);
1181                 t->entries[t->entry_count++] = e;
1182
1183                 e->tree = NULL;
1184                 c = get_mode(c, &e->versions[1].mode);
1185                 if (!c)
1186                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1187                 e->versions[0].mode = e->versions[1].mode;
1188                 e->name = to_atom(c, strlen(c));
1189                 c += e->name->str_len + 1;
1190                 hashcpy(e->versions[0].sha1, (unsigned char*)c);
1191                 hashcpy(e->versions[1].sha1, (unsigned char*)c);
1192                 c += 20;
1193         }
1194         free(buf);
1195 }
1196
1197 static int tecmp0 (const void *_a, const void *_b)
1198 {
1199         struct tree_entry *a = *((struct tree_entry**)_a);
1200         struct tree_entry *b = *((struct tree_entry**)_b);
1201         return base_name_compare(
1202                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1203                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1204 }
1205
1206 static int tecmp1 (const void *_a, const void *_b)
1207 {
1208         struct tree_entry *a = *((struct tree_entry**)_a);
1209         struct tree_entry *b = *((struct tree_entry**)_b);
1210         return base_name_compare(
1211                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1212                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1213 }
1214
1215 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1216 {
1217         size_t maxlen = 0;
1218         unsigned int i;
1219
1220         if (!v)
1221                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1222         else
1223                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1224
1225         for (i = 0; i < t->entry_count; i++) {
1226                 if (t->entries[i]->versions[v].mode)
1227                         maxlen += t->entries[i]->name->str_len + 34;
1228         }
1229
1230         strbuf_reset(b);
1231         strbuf_grow(b, maxlen);
1232         for (i = 0; i < t->entry_count; i++) {
1233                 struct tree_entry *e = t->entries[i];
1234                 if (!e->versions[v].mode)
1235                         continue;
1236                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1237                                         e->name->str_dat, '\0');
1238                 strbuf_add(b, e->versions[v].sha1, 20);
1239         }
1240 }
1241
1242 static void store_tree(struct tree_entry *root)
1243 {
1244         struct tree_content *t = root->tree;
1245         unsigned int i, j, del;
1246         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1247         struct object_entry *le;
1248
1249         if (!is_null_sha1(root->versions[1].sha1))
1250                 return;
1251
1252         for (i = 0; i < t->entry_count; i++) {
1253                 if (t->entries[i]->tree)
1254                         store_tree(t->entries[i]);
1255         }
1256
1257         le = find_object(root->versions[0].sha1);
1258         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1259                 mktree(t, 0, &old_tree);
1260                 lo.data = old_tree;
1261                 lo.offset = le->offset;
1262                 lo.depth = t->delta_depth;
1263         }
1264
1265         mktree(t, 1, &new_tree);
1266         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1267
1268         t->delta_depth = lo.depth;
1269         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1270                 struct tree_entry *e = t->entries[i];
1271                 if (e->versions[1].mode) {
1272                         e->versions[0].mode = e->versions[1].mode;
1273                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1274                         t->entries[j++] = e;
1275                 } else {
1276                         release_tree_entry(e);
1277                         del++;
1278                 }
1279         }
1280         t->entry_count -= del;
1281 }
1282
1283 static int tree_content_set(
1284         struct tree_entry *root,
1285         const char *p,
1286         const unsigned char *sha1,
1287         const uint16_t mode,
1288         struct tree_content *subtree)
1289 {
1290         struct tree_content *t = root->tree;
1291         const char *slash1;
1292         unsigned int i, n;
1293         struct tree_entry *e;
1294
1295         slash1 = strchr(p, '/');
1296         if (slash1)
1297                 n = slash1 - p;
1298         else
1299                 n = strlen(p);
1300         if (!n)
1301                 die("Empty path component found in input");
1302         if (!slash1 && !S_ISDIR(mode) && subtree)
1303                 die("Non-directories cannot have subtrees");
1304
1305         for (i = 0; i < t->entry_count; i++) {
1306                 e = t->entries[i];
1307                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1308                         if (!slash1) {
1309                                 if (!S_ISDIR(mode)
1310                                                 && e->versions[1].mode == mode
1311                                                 && !hashcmp(e->versions[1].sha1, sha1))
1312                                         return 0;
1313                                 e->versions[1].mode = mode;
1314                                 hashcpy(e->versions[1].sha1, sha1);
1315                                 if (e->tree)
1316                                         release_tree_content_recursive(e->tree);
1317                                 e->tree = subtree;
1318                                 hashclr(root->versions[1].sha1);
1319                                 return 1;
1320                         }
1321                         if (!S_ISDIR(e->versions[1].mode)) {
1322                                 e->tree = new_tree_content(8);
1323                                 e->versions[1].mode = S_IFDIR;
1324                         }
1325                         if (!e->tree)
1326                                 load_tree(e);
1327                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1328                                 hashclr(root->versions[1].sha1);
1329                                 return 1;
1330                         }
1331                         return 0;
1332                 }
1333         }
1334
1335         if (t->entry_count == t->entry_capacity)
1336                 root->tree = t = grow_tree_content(t, t->entry_count);
1337         e = new_tree_entry();
1338         e->name = to_atom(p, n);
1339         e->versions[0].mode = 0;
1340         hashclr(e->versions[0].sha1);
1341         t->entries[t->entry_count++] = e;
1342         if (slash1) {
1343                 e->tree = new_tree_content(8);
1344                 e->versions[1].mode = S_IFDIR;
1345                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1346         } else {
1347                 e->tree = subtree;
1348                 e->versions[1].mode = mode;
1349                 hashcpy(e->versions[1].sha1, sha1);
1350         }
1351         hashclr(root->versions[1].sha1);
1352         return 1;
1353 }
1354
1355 static int tree_content_remove(
1356         struct tree_entry *root,
1357         const char *p,
1358         struct tree_entry *backup_leaf)
1359 {
1360         struct tree_content *t = root->tree;
1361         const char *slash1;
1362         unsigned int i, n;
1363         struct tree_entry *e;
1364
1365         slash1 = strchr(p, '/');
1366         if (slash1)
1367                 n = slash1 - p;
1368         else
1369                 n = strlen(p);
1370
1371         for (i = 0; i < t->entry_count; i++) {
1372                 e = t->entries[i];
1373                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1374                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1375                                 goto del_entry;
1376                         if (!e->tree)
1377                                 load_tree(e);
1378                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1379                                 for (n = 0; n < e->tree->entry_count; n++) {
1380                                         if (e->tree->entries[n]->versions[1].mode) {
1381                                                 hashclr(root->versions[1].sha1);
1382                                                 return 1;
1383                                         }
1384                                 }
1385                                 backup_leaf = NULL;
1386                                 goto del_entry;
1387                         }
1388                         return 0;
1389                 }
1390         }
1391         return 0;
1392
1393 del_entry:
1394         if (backup_leaf)
1395                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1396         else if (e->tree)
1397                 release_tree_content_recursive(e->tree);
1398         e->tree = NULL;
1399         e->versions[1].mode = 0;
1400         hashclr(e->versions[1].sha1);
1401         hashclr(root->versions[1].sha1);
1402         return 1;
1403 }
1404
1405 static int tree_content_get(
1406         struct tree_entry *root,
1407         const char *p,
1408         struct tree_entry *leaf)
1409 {
1410         struct tree_content *t = root->tree;
1411         const char *slash1;
1412         unsigned int i, n;
1413         struct tree_entry *e;
1414
1415         slash1 = strchr(p, '/');
1416         if (slash1)
1417                 n = slash1 - p;
1418         else
1419                 n = strlen(p);
1420
1421         for (i = 0; i < t->entry_count; i++) {
1422                 e = t->entries[i];
1423                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1424                         if (!slash1) {
1425                                 memcpy(leaf, e, sizeof(*leaf));
1426                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1427                                         leaf->tree = dup_tree_content(e->tree);
1428                                 else
1429                                         leaf->tree = NULL;
1430                                 return 1;
1431                         }
1432                         if (!S_ISDIR(e->versions[1].mode))
1433                                 return 0;
1434                         if (!e->tree)
1435                                 load_tree(e);
1436                         return tree_content_get(e, slash1 + 1, leaf);
1437                 }
1438         }
1439         return 0;
1440 }
1441
1442 static int update_branch(struct branch *b)
1443 {
1444         static const char *msg = "fast-import";
1445         struct ref_lock *lock;
1446         unsigned char old_sha1[20];
1447
1448         if (read_ref(b->name, old_sha1))
1449                 hashclr(old_sha1);
1450         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1451         if (!lock)
1452                 return error("Unable to lock %s", b->name);
1453         if (!force_update && !is_null_sha1(old_sha1)) {
1454                 struct commit *old_cmit, *new_cmit;
1455
1456                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1457                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1458                 if (!old_cmit || !new_cmit) {
1459                         unlock_ref(lock);
1460                         return error("Branch %s is missing commits.", b->name);
1461                 }
1462
1463                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1464                         unlock_ref(lock);
1465                         warning("Not updating %s"
1466                                 " (new tip %s does not contain %s)",
1467                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1468                         return -1;
1469                 }
1470         }
1471         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1472                 return error("Unable to update %s", b->name);
1473         return 0;
1474 }
1475
1476 static void dump_branches(void)
1477 {
1478         unsigned int i;
1479         struct branch *b;
1480
1481         for (i = 0; i < branch_table_sz; i++) {
1482                 for (b = branch_table[i]; b; b = b->table_next_branch)
1483                         failure |= update_branch(b);
1484         }
1485 }
1486
1487 static void dump_tags(void)
1488 {
1489         static const char *msg = "fast-import";
1490         struct tag *t;
1491         struct ref_lock *lock;
1492         char ref_name[PATH_MAX];
1493
1494         for (t = first_tag; t; t = t->next_tag) {
1495                 sprintf(ref_name, "tags/%s", t->name);
1496                 lock = lock_ref_sha1(ref_name, NULL);
1497                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1498                         failure |= error("Unable to update %s", ref_name);
1499         }
1500 }
1501
1502 static void dump_marks_helper(FILE *f,
1503         uintmax_t base,
1504         struct mark_set *m)
1505 {
1506         uintmax_t k;
1507         if (m->shift) {
1508                 for (k = 0; k < 1024; k++) {
1509                         if (m->data.sets[k])
1510                                 dump_marks_helper(f, (base + k) << m->shift,
1511                                         m->data.sets[k]);
1512                 }
1513         } else {
1514                 for (k = 0; k < 1024; k++) {
1515                         if (m->data.marked[k])
1516                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1517                                         sha1_to_hex(m->data.marked[k]->sha1));
1518                 }
1519         }
1520 }
1521
1522 static void dump_marks(void)
1523 {
1524         static struct lock_file mark_lock;
1525         int mark_fd;
1526         FILE *f;
1527
1528         if (!mark_file)
1529                 return;
1530
1531         mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1532         if (mark_fd < 0) {
1533                 failure |= error("Unable to write marks file %s: %s",
1534                         mark_file, strerror(errno));
1535                 return;
1536         }
1537
1538         f = fdopen(mark_fd, "w");
1539         if (!f) {
1540                 rollback_lock_file(&mark_lock);
1541                 failure |= error("Unable to write marks file %s: %s",
1542                         mark_file, strerror(errno));
1543                 return;
1544         }
1545
1546         dump_marks_helper(f, 0, marks);
1547         fclose(f);
1548         if (commit_lock_file(&mark_lock))
1549                 failure |= error("Unable to write marks file %s: %s",
1550                         mark_file, strerror(errno));
1551 }
1552
1553 static int read_next_command(void)
1554 {
1555         static int stdin_eof = 0;
1556
1557         if (stdin_eof) {
1558                 unread_command_buf = 0;
1559                 return EOF;
1560         }
1561
1562         do {
1563                 if (unread_command_buf) {
1564                         unread_command_buf = 0;
1565                 } else {
1566                         struct recent_command *rc;
1567
1568                         strbuf_detach(&command_buf, NULL);
1569                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1570                         if (stdin_eof)
1571                                 return EOF;
1572
1573                         rc = rc_free;
1574                         if (rc)
1575                                 rc_free = rc->next;
1576                         else {
1577                                 rc = cmd_hist.next;
1578                                 cmd_hist.next = rc->next;
1579                                 cmd_hist.next->prev = &cmd_hist;
1580                                 free(rc->buf);
1581                         }
1582
1583                         rc->buf = command_buf.buf;
1584                         rc->prev = cmd_tail;
1585                         rc->next = cmd_hist.prev;
1586                         rc->prev->next = rc;
1587                         cmd_tail = rc;
1588                 }
1589         } while (command_buf.buf[0] == '#');
1590
1591         return 0;
1592 }
1593
1594 static void skip_optional_lf(void)
1595 {
1596         int term_char = fgetc(stdin);
1597         if (term_char != '\n' && term_char != EOF)
1598                 ungetc(term_char, stdin);
1599 }
1600
1601 static void cmd_mark(void)
1602 {
1603         if (!prefixcmp(command_buf.buf, "mark :")) {
1604                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1605                 read_next_command();
1606         }
1607         else
1608                 next_mark = 0;
1609 }
1610
1611 static void cmd_data(struct strbuf *sb)
1612 {
1613         strbuf_reset(sb);
1614
1615         if (prefixcmp(command_buf.buf, "data "))
1616                 die("Expected 'data n' command, found: %s", command_buf.buf);
1617
1618         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1619                 char *term = xstrdup(command_buf.buf + 5 + 2);
1620                 size_t term_len = command_buf.len - 5 - 2;
1621
1622                 strbuf_detach(&command_buf, NULL);
1623                 for (;;) {
1624                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1625                                 die("EOF in data (terminator '%s' not found)", term);
1626                         if (term_len == command_buf.len
1627                                 && !strcmp(term, command_buf.buf))
1628                                 break;
1629                         strbuf_addbuf(sb, &command_buf);
1630                         strbuf_addch(sb, '\n');
1631                 }
1632                 free(term);
1633         }
1634         else {
1635                 size_t n = 0, length;
1636
1637                 length = strtoul(command_buf.buf + 5, NULL, 10);
1638
1639                 while (n < length) {
1640                         size_t s = strbuf_fread(sb, length - n, stdin);
1641                         if (!s && feof(stdin))
1642                                 die("EOF in data (%lu bytes remaining)",
1643                                         (unsigned long)(length - n));
1644                         n += s;
1645                 }
1646         }
1647
1648         skip_optional_lf();
1649 }
1650
1651 static int validate_raw_date(const char *src, char *result, int maxlen)
1652 {
1653         const char *orig_src = src;
1654         char *endp, sign;
1655
1656         strtoul(src, &endp, 10);
1657         if (endp == src || *endp != ' ')
1658                 return -1;
1659
1660         src = endp + 1;
1661         if (*src != '-' && *src != '+')
1662                 return -1;
1663         sign = *src;
1664
1665         strtoul(src + 1, &endp, 10);
1666         if (endp == src || *endp || (endp - orig_src) >= maxlen)
1667                 return -1;
1668
1669         strcpy(result, orig_src);
1670         return 0;
1671 }
1672
1673 static char *parse_ident(const char *buf)
1674 {
1675         const char *gt;
1676         size_t name_len;
1677         char *ident;
1678
1679         gt = strrchr(buf, '>');
1680         if (!gt)
1681                 die("Missing > in ident string: %s", buf);
1682         gt++;
1683         if (*gt != ' ')
1684                 die("Missing space after > in ident string: %s", buf);
1685         gt++;
1686         name_len = gt - buf;
1687         ident = xmalloc(name_len + 24);
1688         strncpy(ident, buf, name_len);
1689
1690         switch (whenspec) {
1691         case WHENSPEC_RAW:
1692                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1693                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1694                 break;
1695         case WHENSPEC_RFC2822:
1696                 if (parse_date(gt, ident + name_len, 24) < 0)
1697                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1698                 break;
1699         case WHENSPEC_NOW:
1700                 if (strcmp("now", gt))
1701                         die("Date in ident must be 'now': %s", buf);
1702                 datestamp(ident + name_len, 24);
1703                 break;
1704         }
1705
1706         return ident;
1707 }
1708
1709 static void cmd_new_blob(void)
1710 {
1711         static struct strbuf buf = STRBUF_INIT;
1712
1713         read_next_command();
1714         cmd_mark();
1715         cmd_data(&buf);
1716         store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1717 }
1718
1719 static void unload_one_branch(void)
1720 {
1721         while (cur_active_branches
1722                 && cur_active_branches >= max_active_branches) {
1723                 uintmax_t min_commit = ULONG_MAX;
1724                 struct branch *e, *l = NULL, *p = NULL;
1725
1726                 for (e = active_branches; e; e = e->active_next_branch) {
1727                         if (e->last_commit < min_commit) {
1728                                 p = l;
1729                                 min_commit = e->last_commit;
1730                         }
1731                         l = e;
1732                 }
1733
1734                 if (p) {
1735                         e = p->active_next_branch;
1736                         p->active_next_branch = e->active_next_branch;
1737                 } else {
1738                         e = active_branches;
1739                         active_branches = e->active_next_branch;
1740                 }
1741                 e->active = 0;
1742                 e->active_next_branch = NULL;
1743                 if (e->branch_tree.tree) {
1744                         release_tree_content_recursive(e->branch_tree.tree);
1745                         e->branch_tree.tree = NULL;
1746                 }
1747                 cur_active_branches--;
1748         }
1749 }
1750
1751 static void load_branch(struct branch *b)
1752 {
1753         load_tree(&b->branch_tree);
1754         if (!b->active) {
1755                 b->active = 1;
1756                 b->active_next_branch = active_branches;
1757                 active_branches = b;
1758                 cur_active_branches++;
1759                 branch_load_count++;
1760         }
1761 }
1762
1763 static void file_change_m(struct branch *b)
1764 {
1765         const char *p = command_buf.buf + 2;
1766         static struct strbuf uq = STRBUF_INIT;
1767         const char *endp;
1768         struct object_entry *oe = oe;
1769         unsigned char sha1[20];
1770         uint16_t mode, inline_data = 0;
1771
1772         p = get_mode(p, &mode);
1773         if (!p)
1774                 die("Corrupt mode: %s", command_buf.buf);
1775         switch (mode) {
1776         case S_IFREG | 0644:
1777         case S_IFREG | 0755:
1778         case S_IFLNK:
1779         case 0644:
1780         case 0755:
1781                 /* ok */
1782                 break;
1783         default:
1784                 die("Corrupt mode: %s", command_buf.buf);
1785         }
1786
1787         if (*p == ':') {
1788                 char *x;
1789                 oe = find_mark(strtoumax(p + 1, &x, 10));
1790                 hashcpy(sha1, oe->sha1);
1791                 p = x;
1792         } else if (!prefixcmp(p, "inline")) {
1793                 inline_data = 1;
1794                 p += 6;
1795         } else {
1796                 if (get_sha1_hex(p, sha1))
1797                         die("Invalid SHA1: %s", command_buf.buf);
1798                 oe = find_object(sha1);
1799                 p += 40;
1800         }
1801         if (*p++ != ' ')
1802                 die("Missing space after SHA1: %s", command_buf.buf);
1803
1804         strbuf_reset(&uq);
1805         if (!unquote_c_style(&uq, p, &endp)) {
1806                 if (*endp)
1807                         die("Garbage after path in: %s", command_buf.buf);
1808                 p = uq.buf;
1809         }
1810
1811         if (inline_data) {
1812                 static struct strbuf buf = STRBUF_INIT;
1813
1814                 if (p != uq.buf) {
1815                         strbuf_addstr(&uq, p);
1816                         p = uq.buf;
1817                 }
1818                 read_next_command();
1819                 cmd_data(&buf);
1820                 store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1821         } else if (oe) {
1822                 if (oe->type != OBJ_BLOB)
1823                         die("Not a blob (actually a %s): %s",
1824                                 typename(oe->type), command_buf.buf);
1825         } else {
1826                 enum object_type type = sha1_object_info(sha1, NULL);
1827                 if (type < 0)
1828                         die("Blob not found: %s", command_buf.buf);
1829                 if (type != OBJ_BLOB)
1830                         die("Not a blob (actually a %s): %s",
1831                             typename(type), command_buf.buf);
1832         }
1833
1834         tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode, NULL);
1835 }
1836
1837 static void file_change_d(struct branch *b)
1838 {
1839         const char *p = command_buf.buf + 2;
1840         static struct strbuf uq = STRBUF_INIT;
1841         const char *endp;
1842
1843         strbuf_reset(&uq);
1844         if (!unquote_c_style(&uq, p, &endp)) {
1845                 if (*endp)
1846                         die("Garbage after path in: %s", command_buf.buf);
1847                 p = uq.buf;
1848         }
1849         tree_content_remove(&b->branch_tree, p, NULL);
1850 }
1851
1852 static void file_change_cr(struct branch *b, int rename)
1853 {
1854         const char *s, *d;
1855         static struct strbuf s_uq = STRBUF_INIT;
1856         static struct strbuf d_uq = STRBUF_INIT;
1857         const char *endp;
1858         struct tree_entry leaf;
1859
1860         s = command_buf.buf + 2;
1861         strbuf_reset(&s_uq);
1862         if (!unquote_c_style(&s_uq, s, &endp)) {
1863                 if (*endp != ' ')
1864                         die("Missing space after source: %s", command_buf.buf);
1865         } else {
1866                 endp = strchr(s, ' ');
1867                 if (!endp)
1868                         die("Missing space after source: %s", command_buf.buf);
1869                 strbuf_add(&s_uq, s, endp - s);
1870         }
1871         s = s_uq.buf;
1872
1873         endp++;
1874         if (!*endp)
1875                 die("Missing dest: %s", command_buf.buf);
1876
1877         d = endp;
1878         strbuf_reset(&d_uq);
1879         if (!unquote_c_style(&d_uq, d, &endp)) {
1880                 if (*endp)
1881                         die("Garbage after dest in: %s", command_buf.buf);
1882                 d = d_uq.buf;
1883         }
1884
1885         memset(&leaf, 0, sizeof(leaf));
1886         if (rename)
1887                 tree_content_remove(&b->branch_tree, s, &leaf);
1888         else
1889                 tree_content_get(&b->branch_tree, s, &leaf);
1890         if (!leaf.versions[1].mode)
1891                 die("Path %s not in branch", s);
1892         tree_content_set(&b->branch_tree, d,
1893                 leaf.versions[1].sha1,
1894                 leaf.versions[1].mode,
1895                 leaf.tree);
1896 }
1897
1898 static void file_change_deleteall(struct branch *b)
1899 {
1900         release_tree_content_recursive(b->branch_tree.tree);
1901         hashclr(b->branch_tree.versions[0].sha1);
1902         hashclr(b->branch_tree.versions[1].sha1);
1903         load_tree(&b->branch_tree);
1904 }
1905
1906 static void cmd_from_commit(struct branch *b, char *buf, unsigned long size)
1907 {
1908         if (!buf || size < 46)
1909                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
1910         if (memcmp("tree ", buf, 5)
1911                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1912                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1913         hashcpy(b->branch_tree.versions[0].sha1,
1914                 b->branch_tree.versions[1].sha1);
1915 }
1916
1917 static void cmd_from_existing(struct branch *b)
1918 {
1919         if (is_null_sha1(b->sha1)) {
1920                 hashclr(b->branch_tree.versions[0].sha1);
1921                 hashclr(b->branch_tree.versions[1].sha1);
1922         } else {
1923                 unsigned long size;
1924                 char *buf;
1925
1926                 buf = read_object_with_reference(b->sha1,
1927                         commit_type, &size, b->sha1);
1928                 cmd_from_commit(b, buf, size);
1929                 free(buf);
1930         }
1931 }
1932
1933 static int cmd_from(struct branch *b)
1934 {
1935         const char *from;
1936         struct branch *s;
1937
1938         if (prefixcmp(command_buf.buf, "from "))
1939                 return 0;
1940
1941         if (b->branch_tree.tree) {
1942                 release_tree_content_recursive(b->branch_tree.tree);
1943                 b->branch_tree.tree = NULL;
1944         }
1945
1946         from = strchr(command_buf.buf, ' ') + 1;
1947         s = lookup_branch(from);
1948         if (b == s)
1949                 die("Can't create a branch from itself: %s", b->name);
1950         else if (s) {
1951                 unsigned char *t = s->branch_tree.versions[1].sha1;
1952                 hashcpy(b->sha1, s->sha1);
1953                 hashcpy(b->branch_tree.versions[0].sha1, t);
1954                 hashcpy(b->branch_tree.versions[1].sha1, t);
1955         } else if (*from == ':') {
1956                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
1957                 struct object_entry *oe = find_mark(idnum);
1958                 if (oe->type != OBJ_COMMIT)
1959                         die("Mark :%" PRIuMAX " not a commit", idnum);
1960                 hashcpy(b->sha1, oe->sha1);
1961                 if (oe->pack_id != MAX_PACK_ID) {
1962                         unsigned long size;
1963                         char *buf = gfi_unpack_entry(oe, &size);
1964                         cmd_from_commit(b, buf, size);
1965                         free(buf);
1966                 } else
1967                         cmd_from_existing(b);
1968         } else if (!get_sha1(from, b->sha1))
1969                 cmd_from_existing(b);
1970         else
1971                 die("Invalid ref name or SHA1 expression: %s", from);
1972
1973         read_next_command();
1974         return 1;
1975 }
1976
1977 static struct hash_list *cmd_merge(unsigned int *count)
1978 {
1979         struct hash_list *list = NULL, *n, *e = e;
1980         const char *from;
1981         struct branch *s;
1982
1983         *count = 0;
1984         while (!prefixcmp(command_buf.buf, "merge ")) {
1985                 from = strchr(command_buf.buf, ' ') + 1;
1986                 n = xmalloc(sizeof(*n));
1987                 s = lookup_branch(from);
1988                 if (s)
1989                         hashcpy(n->sha1, s->sha1);
1990                 else if (*from == ':') {
1991                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
1992                         struct object_entry *oe = find_mark(idnum);
1993                         if (oe->type != OBJ_COMMIT)
1994                                 die("Mark :%" PRIuMAX " not a commit", idnum);
1995                         hashcpy(n->sha1, oe->sha1);
1996                 } else if (!get_sha1(from, n->sha1)) {
1997                         unsigned long size;
1998                         char *buf = read_object_with_reference(n->sha1,
1999                                 commit_type, &size, n->sha1);
2000                         if (!buf || size < 46)
2001                                 die("Not a valid commit: %s", from);
2002                         free(buf);
2003                 } else
2004                         die("Invalid ref name or SHA1 expression: %s", from);
2005
2006                 n->next = NULL;
2007                 if (list)
2008                         e->next = n;
2009                 else
2010                         list = n;
2011                 e = n;
2012                 (*count)++;
2013                 read_next_command();
2014         }
2015         return list;
2016 }
2017
2018 static void cmd_new_commit(void)
2019 {
2020         static struct strbuf msg = STRBUF_INIT;
2021         struct branch *b;
2022         char *sp;
2023         char *author = NULL;
2024         char *committer = NULL;
2025         struct hash_list *merge_list = NULL;
2026         unsigned int merge_count;
2027
2028         /* Obtain the branch name from the rest of our command */
2029         sp = strchr(command_buf.buf, ' ') + 1;
2030         b = lookup_branch(sp);
2031         if (!b)
2032                 b = new_branch(sp);
2033
2034         read_next_command();
2035         cmd_mark();
2036         if (!prefixcmp(command_buf.buf, "author ")) {
2037                 author = parse_ident(command_buf.buf + 7);
2038                 read_next_command();
2039         }
2040         if (!prefixcmp(command_buf.buf, "committer ")) {
2041                 committer = parse_ident(command_buf.buf + 10);
2042                 read_next_command();
2043         }
2044         if (!committer)
2045                 die("Expected committer but didn't get one");
2046         cmd_data(&msg);
2047         read_next_command();
2048         cmd_from(b);
2049         merge_list = cmd_merge(&merge_count);
2050
2051         /* ensure the branch is active/loaded */
2052         if (!b->branch_tree.tree || !max_active_branches) {
2053                 unload_one_branch();
2054                 load_branch(b);
2055         }
2056
2057         /* file_change* */
2058         while (command_buf.len > 0) {
2059                 if (!prefixcmp(command_buf.buf, "M "))
2060                         file_change_m(b);
2061                 else if (!prefixcmp(command_buf.buf, "D "))
2062                         file_change_d(b);
2063                 else if (!prefixcmp(command_buf.buf, "R "))
2064                         file_change_cr(b, 1);
2065                 else if (!prefixcmp(command_buf.buf, "C "))
2066                         file_change_cr(b, 0);
2067                 else if (!strcmp("deleteall", command_buf.buf))
2068                         file_change_deleteall(b);
2069                 else {
2070                         unread_command_buf = 1;
2071                         break;
2072                 }
2073                 if (read_next_command() == EOF)
2074                         break;
2075         }
2076
2077         /* build the tree and the commit */
2078         store_tree(&b->branch_tree);
2079         hashcpy(b->branch_tree.versions[0].sha1,
2080                 b->branch_tree.versions[1].sha1);
2081
2082         strbuf_reset(&new_data);
2083         strbuf_addf(&new_data, "tree %s\n",
2084                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2085         if (!is_null_sha1(b->sha1))
2086                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2087         while (merge_list) {
2088                 struct hash_list *next = merge_list->next;
2089                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2090                 free(merge_list);
2091                 merge_list = next;
2092         }
2093         strbuf_addf(&new_data,
2094                 "author %s\n"
2095                 "committer %s\n"
2096                 "\n",
2097                 author ? author : committer, committer);
2098         strbuf_addbuf(&new_data, &msg);
2099         free(author);
2100         free(committer);
2101
2102         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2103                 b->pack_id = pack_id;
2104         b->last_commit = object_count_by_type[OBJ_COMMIT];
2105 }
2106
2107 static void cmd_new_tag(void)
2108 {
2109         static struct strbuf msg = STRBUF_INIT;
2110         char *sp;
2111         const char *from;
2112         char *tagger;
2113         struct branch *s;
2114         struct tag *t;
2115         uintmax_t from_mark = 0;
2116         unsigned char sha1[20];
2117
2118         /* Obtain the new tag name from the rest of our command */
2119         sp = strchr(command_buf.buf, ' ') + 1;
2120         t = pool_alloc(sizeof(struct tag));
2121         t->next_tag = NULL;
2122         t->name = pool_strdup(sp);
2123         if (last_tag)
2124                 last_tag->next_tag = t;
2125         else
2126                 first_tag = t;
2127         last_tag = t;
2128         read_next_command();
2129
2130         /* from ... */
2131         if (prefixcmp(command_buf.buf, "from "))
2132                 die("Expected from command, got %s", command_buf.buf);
2133         from = strchr(command_buf.buf, ' ') + 1;
2134         s = lookup_branch(from);
2135         if (s) {
2136                 hashcpy(sha1, s->sha1);
2137         } else if (*from == ':') {
2138                 struct object_entry *oe;
2139                 from_mark = strtoumax(from + 1, NULL, 10);
2140                 oe = find_mark(from_mark);
2141                 if (oe->type != OBJ_COMMIT)
2142                         die("Mark :%" PRIuMAX " not a commit", from_mark);
2143                 hashcpy(sha1, oe->sha1);
2144         } else if (!get_sha1(from, sha1)) {
2145                 unsigned long size;
2146                 char *buf;
2147
2148                 buf = read_object_with_reference(sha1,
2149                         commit_type, &size, sha1);
2150                 if (!buf || size < 46)
2151                         die("Not a valid commit: %s", from);
2152                 free(buf);
2153         } else
2154                 die("Invalid ref name or SHA1 expression: %s", from);
2155         read_next_command();
2156
2157         /* tagger ... */
2158         if (prefixcmp(command_buf.buf, "tagger "))
2159                 die("Expected tagger command, got %s", command_buf.buf);
2160         tagger = parse_ident(command_buf.buf + 7);
2161
2162         /* tag payload/message */
2163         read_next_command();
2164         cmd_data(&msg);
2165
2166         /* build the tag object */
2167         strbuf_reset(&new_data);
2168         strbuf_addf(&new_data,
2169                 "object %s\n"
2170                 "type %s\n"
2171                 "tag %s\n"
2172                 "tagger %s\n"
2173                 "\n",
2174                 sha1_to_hex(sha1), commit_type, t->name, tagger);
2175         strbuf_addbuf(&new_data, &msg);
2176         free(tagger);
2177
2178         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2179                 t->pack_id = MAX_PACK_ID;
2180         else
2181                 t->pack_id = pack_id;
2182 }
2183
2184 static void cmd_reset_branch(void)
2185 {
2186         struct branch *b;
2187         char *sp;
2188
2189         /* Obtain the branch name from the rest of our command */
2190         sp = strchr(command_buf.buf, ' ') + 1;
2191         b = lookup_branch(sp);
2192         if (b) {
2193                 hashclr(b->sha1);
2194                 hashclr(b->branch_tree.versions[0].sha1);
2195                 hashclr(b->branch_tree.versions[1].sha1);
2196                 if (b->branch_tree.tree) {
2197                         release_tree_content_recursive(b->branch_tree.tree);
2198                         b->branch_tree.tree = NULL;
2199                 }
2200         }
2201         else
2202                 b = new_branch(sp);
2203         read_next_command();
2204         if (!cmd_from(b) && command_buf.len > 0)
2205                 unread_command_buf = 1;
2206 }
2207
2208 static void cmd_checkpoint(void)
2209 {
2210         if (object_count) {
2211                 cycle_packfile();
2212                 dump_branches();
2213                 dump_tags();
2214                 dump_marks();
2215         }
2216         skip_optional_lf();
2217 }
2218
2219 static void cmd_progress(void)
2220 {
2221         fwrite(command_buf.buf, 1, command_buf.len, stdout);
2222         fputc('\n', stdout);
2223         fflush(stdout);
2224         skip_optional_lf();
2225 }
2226
2227 static void import_marks(const char *input_file)
2228 {
2229         char line[512];
2230         FILE *f = fopen(input_file, "r");
2231         if (!f)
2232                 die("cannot read %s: %s", input_file, strerror(errno));
2233         while (fgets(line, sizeof(line), f)) {
2234                 uintmax_t mark;
2235                 char *end;
2236                 unsigned char sha1[20];
2237                 struct object_entry *e;
2238
2239                 end = strchr(line, '\n');
2240                 if (line[0] != ':' || !end)
2241                         die("corrupt mark line: %s", line);
2242                 *end = 0;
2243                 mark = strtoumax(line + 1, &end, 10);
2244                 if (!mark || end == line + 1
2245                         || *end != ' ' || get_sha1(end + 1, sha1))
2246                         die("corrupt mark line: %s", line);
2247                 e = find_object(sha1);
2248                 if (!e) {
2249                         enum object_type type = sha1_object_info(sha1, NULL);
2250                         if (type < 0)
2251                                 die("object not found: %s", sha1_to_hex(sha1));
2252                         e = insert_object(sha1);
2253                         e->type = type;
2254                         e->pack_id = MAX_PACK_ID;
2255                         e->offset = 1; /* just not zero! */
2256                 }
2257                 insert_mark(mark, e);
2258         }
2259         fclose(f);
2260 }
2261
2262 static const char fast_import_usage[] =
2263 "git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2264
2265 int main(int argc, const char **argv)
2266 {
2267         unsigned int i, show_stats = 1;
2268
2269         git_config(git_default_config);
2270         alloc_objects(object_entry_alloc);
2271         strbuf_init(&command_buf, 0);
2272         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2273         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2274         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2275         marks = pool_calloc(1, sizeof(struct mark_set));
2276
2277         for (i = 1; i < argc; i++) {
2278                 const char *a = argv[i];
2279
2280                 if (*a != '-' || !strcmp(a, "--"))
2281                         break;
2282                 else if (!prefixcmp(a, "--date-format=")) {
2283                         const char *fmt = a + 14;
2284                         if (!strcmp(fmt, "raw"))
2285                                 whenspec = WHENSPEC_RAW;
2286                         else if (!strcmp(fmt, "rfc2822"))
2287                                 whenspec = WHENSPEC_RFC2822;
2288                         else if (!strcmp(fmt, "now"))
2289                                 whenspec = WHENSPEC_NOW;
2290                         else
2291                                 die("unknown --date-format argument %s", fmt);
2292                 }
2293                 else if (!prefixcmp(a, "--max-pack-size="))
2294                         max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2295                 else if (!prefixcmp(a, "--depth=")) {
2296                         max_depth = strtoul(a + 8, NULL, 0);
2297                         if (max_depth > MAX_DEPTH)
2298                                 die("--depth cannot exceed %u", MAX_DEPTH);
2299                 }
2300                 else if (!prefixcmp(a, "--active-branches="))
2301                         max_active_branches = strtoul(a + 18, NULL, 0);
2302                 else if (!prefixcmp(a, "--import-marks="))
2303                         import_marks(a + 15);
2304                 else if (!prefixcmp(a, "--export-marks="))
2305                         mark_file = a + 15;
2306                 else if (!prefixcmp(a, "--export-pack-edges=")) {
2307                         if (pack_edges)
2308                                 fclose(pack_edges);
2309                         pack_edges = fopen(a + 20, "a");
2310                         if (!pack_edges)
2311                                 die("Cannot open %s: %s", a + 20, strerror(errno));
2312                 } else if (!strcmp(a, "--force"))
2313                         force_update = 1;
2314                 else if (!strcmp(a, "--quiet"))
2315                         show_stats = 0;
2316                 else if (!strcmp(a, "--stats"))
2317                         show_stats = 1;
2318                 else
2319                         die("unknown option %s", a);
2320         }
2321         if (i != argc)
2322                 usage(fast_import_usage);
2323
2324         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2325         for (i = 0; i < (cmd_save - 1); i++)
2326                 rc_free[i].next = &rc_free[i + 1];
2327         rc_free[cmd_save - 1].next = NULL;
2328
2329         prepare_packed_git();
2330         start_packfile();
2331         set_die_routine(die_nicely);
2332         while (read_next_command() != EOF) {
2333                 if (!strcmp("blob", command_buf.buf))
2334                         cmd_new_blob();
2335                 else if (!prefixcmp(command_buf.buf, "commit "))
2336                         cmd_new_commit();
2337                 else if (!prefixcmp(command_buf.buf, "tag "))
2338                         cmd_new_tag();
2339                 else if (!prefixcmp(command_buf.buf, "reset "))
2340                         cmd_reset_branch();
2341                 else if (!strcmp("checkpoint", command_buf.buf))
2342                         cmd_checkpoint();
2343                 else if (!prefixcmp(command_buf.buf, "progress "))
2344                         cmd_progress();
2345                 else
2346                         die("Unsupported command: %s", command_buf.buf);
2347         }
2348         end_packfile();
2349
2350         dump_branches();
2351         dump_tags();
2352         unkeep_all_packs();
2353         dump_marks();
2354
2355         if (pack_edges)
2356                 fclose(pack_edges);
2357
2358         if (show_stats) {
2359                 uintmax_t total_count = 0, duplicate_count = 0;
2360                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2361                         total_count += object_count_by_type[i];
2362                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2363                         duplicate_count += duplicate_count_by_type[i];
2364
2365                 fprintf(stderr, "%s statistics:\n", argv[0]);
2366                 fprintf(stderr, "---------------------------------------------------------------------\n");
2367                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2368                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2369                 fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
2370                 fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
2371                 fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
2372                 fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
2373                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2374                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2375                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2376                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2377                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2378                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2379                 fprintf(stderr, "---------------------------------------------------------------------\n");
2380                 pack_report();
2381                 fprintf(stderr, "---------------------------------------------------------------------\n");
2382                 fprintf(stderr, "\n");
2383         }
2384
2385         return failure ? 1 : 0;
2386 }