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