Move sha1_file_to_archive into libgit
[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         uintmax_t 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 = (char *) 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 'uintmax_t' alignment */
544         if (len & (sizeof(uintmax_t) - 1))
545                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 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 CHECK_REF_FORMAT_ONELEVEL:
647                 break; /* valid, but too few '/', allow anyway */
648         default:
649                 die("Branch name doesn't conform to GIT standards: %s", name);
650         }
651
652         b = pool_calloc(1, sizeof(struct branch));
653         b->name = pool_strdup(name);
654         b->table_next_branch = branch_table[hc];
655         b->branch_tree.versions[0].mode = S_IFDIR;
656         b->branch_tree.versions[1].mode = S_IFDIR;
657         b->active = 0;
658         b->pack_id = MAX_PACK_ID;
659         branch_table[hc] = b;
660         branch_count++;
661         return b;
662 }
663
664 static unsigned int hc_entries(unsigned int cnt)
665 {
666         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
667         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
668 }
669
670 static struct tree_content *new_tree_content(unsigned int cnt)
671 {
672         struct avail_tree_content *f, *l = NULL;
673         struct tree_content *t;
674         unsigned int hc = hc_entries(cnt);
675
676         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
677                 if (f->entry_capacity >= cnt)
678                         break;
679
680         if (f) {
681                 if (l)
682                         l->next_avail = f->next_avail;
683                 else
684                         avail_tree_table[hc] = f->next_avail;
685         } else {
686                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
687                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
688                 f->entry_capacity = cnt;
689         }
690
691         t = (struct tree_content*)f;
692         t->entry_count = 0;
693         t->delta_depth = 0;
694         return t;
695 }
696
697 static void release_tree_entry(struct tree_entry *e);
698 static void release_tree_content(struct tree_content *t)
699 {
700         struct avail_tree_content *f = (struct avail_tree_content*)t;
701         unsigned int hc = hc_entries(f->entry_capacity);
702         f->next_avail = avail_tree_table[hc];
703         avail_tree_table[hc] = f;
704 }
705
706 static void release_tree_content_recursive(struct tree_content *t)
707 {
708         unsigned int i;
709         for (i = 0; i < t->entry_count; i++)
710                 release_tree_entry(t->entries[i]);
711         release_tree_content(t);
712 }
713
714 static struct tree_content *grow_tree_content(
715         struct tree_content *t,
716         int amt)
717 {
718         struct tree_content *r = new_tree_content(t->entry_count + amt);
719         r->entry_count = t->entry_count;
720         r->delta_depth = t->delta_depth;
721         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
722         release_tree_content(t);
723         return r;
724 }
725
726 static struct tree_entry *new_tree_entry(void)
727 {
728         struct tree_entry *e;
729
730         if (!avail_tree_entry) {
731                 unsigned int n = tree_entry_alloc;
732                 total_allocd += n * sizeof(struct tree_entry);
733                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
734                 while (n-- > 1) {
735                         *((void**)e) = e + 1;
736                         e++;
737                 }
738                 *((void**)e) = NULL;
739         }
740
741         e = avail_tree_entry;
742         avail_tree_entry = *((void**)e);
743         return e;
744 }
745
746 static void release_tree_entry(struct tree_entry *e)
747 {
748         if (e->tree)
749                 release_tree_content_recursive(e->tree);
750         *((void**)e) = avail_tree_entry;
751         avail_tree_entry = e;
752 }
753
754 static struct tree_content *dup_tree_content(struct tree_content *s)
755 {
756         struct tree_content *d;
757         struct tree_entry *a, *b;
758         unsigned int i;
759
760         if (!s)
761                 return NULL;
762         d = new_tree_content(s->entry_count);
763         for (i = 0; i < s->entry_count; i++) {
764                 a = s->entries[i];
765                 b = new_tree_entry();
766                 memcpy(b, a, sizeof(*a));
767                 if (a->tree && is_null_sha1(b->versions[1].sha1))
768                         b->tree = dup_tree_content(a->tree);
769                 else
770                         b->tree = NULL;
771                 d->entries[i] = b;
772         }
773         d->entry_count = s->entry_count;
774         d->delta_depth = s->delta_depth;
775
776         return d;
777 }
778
779 static void start_packfile(void)
780 {
781         static char tmpfile[PATH_MAX];
782         struct packed_git *p;
783         struct pack_header hdr;
784         int pack_fd;
785
786         snprintf(tmpfile, sizeof(tmpfile),
787                 "%s/tmp_pack_XXXXXX", get_object_directory());
788         pack_fd = xmkstemp(tmpfile);
789         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
790         strcpy(p->pack_name, tmpfile);
791         p->pack_fd = pack_fd;
792
793         hdr.hdr_signature = htonl(PACK_SIGNATURE);
794         hdr.hdr_version = htonl(2);
795         hdr.hdr_entries = 0;
796         write_or_die(p->pack_fd, &hdr, sizeof(hdr));
797
798         pack_data = p;
799         pack_size = sizeof(hdr);
800         object_count = 0;
801
802         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
803         all_packs[pack_id] = p;
804 }
805
806 static int oecmp (const void *a_, const void *b_)
807 {
808         struct object_entry *a = *((struct object_entry**)a_);
809         struct object_entry *b = *((struct object_entry**)b_);
810         return hashcmp(a->sha1, b->sha1);
811 }
812
813 static char *create_index(void)
814 {
815         static char tmpfile[PATH_MAX];
816         SHA_CTX ctx;
817         struct sha1file *f;
818         struct object_entry **idx, **c, **last, *e;
819         struct object_entry_pool *o;
820         uint32_t array[256];
821         int i, idx_fd;
822
823         /* Build the sorted table of object IDs. */
824         idx = xmalloc(object_count * sizeof(struct object_entry*));
825         c = idx;
826         for (o = blocks; o; o = o->next_pool)
827                 for (e = o->next_free; e-- != o->entries;)
828                         if (pack_id == e->pack_id)
829                                 *c++ = e;
830         last = idx + object_count;
831         if (c != last)
832                 die("internal consistency error creating the index");
833         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
834
835         /* Generate the fan-out array. */
836         c = idx;
837         for (i = 0; i < 256; i++) {
838                 struct object_entry **next = c;;
839                 while (next < last) {
840                         if ((*next)->sha1[0] != i)
841                                 break;
842                         next++;
843                 }
844                 array[i] = htonl(next - idx);
845                 c = next;
846         }
847
848         snprintf(tmpfile, sizeof(tmpfile),
849                 "%s/tmp_idx_XXXXXX", get_object_directory());
850         idx_fd = xmkstemp(tmpfile);
851         f = sha1fd(idx_fd, tmpfile);
852         sha1write(f, array, 256 * sizeof(int));
853         SHA1_Init(&ctx);
854         for (c = idx; c != last; c++) {
855                 uint32_t offset = htonl((*c)->offset);
856                 sha1write(f, &offset, 4);
857                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
858                 SHA1_Update(&ctx, (*c)->sha1, 20);
859         }
860         sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
861         sha1close(f, NULL, 1);
862         free(idx);
863         SHA1_Final(pack_data->sha1, &ctx);
864         return tmpfile;
865 }
866
867 static char *keep_pack(char *curr_index_name)
868 {
869         static char name[PATH_MAX];
870         static const char *keep_msg = "fast-import";
871         int keep_fd;
872
873         chmod(pack_data->pack_name, 0444);
874         chmod(curr_index_name, 0444);
875
876         snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
877                  get_object_directory(), sha1_to_hex(pack_data->sha1));
878         keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
879         if (keep_fd < 0)
880                 die("cannot create keep file");
881         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
882         if (close(keep_fd))
883                 die("failed to write keep file");
884
885         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
886                  get_object_directory(), sha1_to_hex(pack_data->sha1));
887         if (move_temp_to_file(pack_data->pack_name, name))
888                 die("cannot store pack file");
889
890         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
891                  get_object_directory(), sha1_to_hex(pack_data->sha1));
892         if (move_temp_to_file(curr_index_name, name))
893                 die("cannot store index file");
894         return name;
895 }
896
897 static void unkeep_all_packs(void)
898 {
899         static char name[PATH_MAX];
900         int k;
901
902         for (k = 0; k < pack_id; k++) {
903                 struct packed_git *p = all_packs[k];
904                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
905                          get_object_directory(), sha1_to_hex(p->sha1));
906                 unlink(name);
907         }
908 }
909
910 static void end_packfile(void)
911 {
912         struct packed_git *old_p = pack_data, *new_p;
913
914         if (object_count) {
915                 char *idx_name;
916                 int i;
917                 struct branch *b;
918                 struct tag *t;
919
920                 close_pack_windows(pack_data);
921                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
922                                     pack_data->pack_name, object_count);
923                 close(pack_data->pack_fd);
924                 idx_name = keep_pack(create_index());
925
926                 /* Register the packfile with core git's machinary. */
927                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
928                 if (!new_p)
929                         die("core git rejected index %s", idx_name);
930                 all_packs[pack_id] = new_p;
931                 install_packed_git(new_p);
932
933                 /* Print the boundary */
934                 if (pack_edges) {
935                         fprintf(pack_edges, "%s:", new_p->pack_name);
936                         for (i = 0; i < branch_table_sz; i++) {
937                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
938                                         if (b->pack_id == pack_id)
939                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
940                                 }
941                         }
942                         for (t = first_tag; t; t = t->next_tag) {
943                                 if (t->pack_id == pack_id)
944                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
945                         }
946                         fputc('\n', pack_edges);
947                         fflush(pack_edges);
948                 }
949
950                 pack_id++;
951         }
952         else
953                 unlink(old_p->pack_name);
954         free(old_p);
955
956         /* We can't carry a delta across packfiles. */
957         strbuf_release(&last_blob.data);
958         last_blob.offset = 0;
959         last_blob.depth = 0;
960 }
961
962 static void cycle_packfile(void)
963 {
964         end_packfile();
965         start_packfile();
966 }
967
968 static size_t encode_header(
969         enum object_type type,
970         size_t size,
971         unsigned char *hdr)
972 {
973         int n = 1;
974         unsigned char c;
975
976         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
977                 die("bad type %d", type);
978
979         c = (type << 4) | (size & 15);
980         size >>= 4;
981         while (size) {
982                 *hdr++ = c | 0x80;
983                 c = size & 0x7f;
984                 size >>= 7;
985                 n++;
986         }
987         *hdr = c;
988         return n;
989 }
990
991 static int store_object(
992         enum object_type type,
993         struct strbuf *dat,
994         struct last_object *last,
995         unsigned char *sha1out,
996         uintmax_t mark)
997 {
998         void *out, *delta;
999         struct object_entry *e;
1000         unsigned char hdr[96];
1001         unsigned char sha1[20];
1002         unsigned long hdrlen, deltalen;
1003         SHA_CTX c;
1004         z_stream s;
1005
1006         hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1007                 (unsigned long)dat->len) + 1;
1008         SHA1_Init(&c);
1009         SHA1_Update(&c, hdr, hdrlen);
1010         SHA1_Update(&c, dat->buf, dat->len);
1011         SHA1_Final(sha1, &c);
1012         if (sha1out)
1013                 hashcpy(sha1out, sha1);
1014
1015         e = insert_object(sha1);
1016         if (mark)
1017                 insert_mark(mark, e);
1018         if (e->offset) {
1019                 duplicate_count_by_type[type]++;
1020                 return 1;
1021         } else if (find_sha1_pack(sha1, packed_git)) {
1022                 e->type = type;
1023                 e->pack_id = MAX_PACK_ID;
1024                 e->offset = 1; /* just not zero! */
1025                 duplicate_count_by_type[type]++;
1026                 return 1;
1027         }
1028
1029         if (last && last->data.buf && last->depth < max_depth) {
1030                 delta = diff_delta(last->data.buf, last->data.len,
1031                         dat->buf, dat->len,
1032                         &deltalen, 0);
1033                 if (delta && deltalen >= dat->len) {
1034                         free(delta);
1035                         delta = NULL;
1036                 }
1037         } else
1038                 delta = NULL;
1039
1040         memset(&s, 0, sizeof(s));
1041         deflateInit(&s, zlib_compression_level);
1042         if (delta) {
1043                 s.next_in = delta;
1044                 s.avail_in = deltalen;
1045         } else {
1046                 s.next_in = (void *)dat->buf;
1047                 s.avail_in = dat->len;
1048         }
1049         s.avail_out = deflateBound(&s, s.avail_in);
1050         s.next_out = out = xmalloc(s.avail_out);
1051         while (deflate(&s, Z_FINISH) == Z_OK)
1052                 /* nothing */;
1053         deflateEnd(&s);
1054
1055         /* Determine if we should auto-checkpoint. */
1056         if ((pack_size + 60 + s.total_out) > max_packsize
1057                 || (pack_size + 60 + s.total_out) < pack_size) {
1058
1059                 /* This new object needs to *not* have the current pack_id. */
1060                 e->pack_id = pack_id + 1;
1061                 cycle_packfile();
1062
1063                 /* We cannot carry a delta into the new pack. */
1064                 if (delta) {
1065                         free(delta);
1066                         delta = NULL;
1067
1068                         memset(&s, 0, sizeof(s));
1069                         deflateInit(&s, zlib_compression_level);
1070                         s.next_in = (void *)dat->buf;
1071                         s.avail_in = dat->len;
1072                         s.avail_out = deflateBound(&s, s.avail_in);
1073                         s.next_out = out = xrealloc(out, s.avail_out);
1074                         while (deflate(&s, Z_FINISH) == Z_OK)
1075                                 /* nothing */;
1076                         deflateEnd(&s);
1077                 }
1078         }
1079
1080         e->type = type;
1081         e->pack_id = pack_id;
1082         e->offset = pack_size;
1083         object_count++;
1084         object_count_by_type[type]++;
1085
1086         if (delta) {
1087                 unsigned long ofs = e->offset - last->offset;
1088                 unsigned pos = sizeof(hdr) - 1;
1089
1090                 delta_count_by_type[type]++;
1091                 e->depth = last->depth + 1;
1092
1093                 hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1094                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1095                 pack_size += hdrlen;
1096
1097                 hdr[pos] = ofs & 127;
1098                 while (ofs >>= 7)
1099                         hdr[--pos] = 128 | (--ofs & 127);
1100                 write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1101                 pack_size += sizeof(hdr) - pos;
1102         } else {
1103                 e->depth = 0;
1104                 hdrlen = encode_header(type, dat->len, hdr);
1105                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1106                 pack_size += hdrlen;
1107         }
1108
1109         write_or_die(pack_data->pack_fd, out, s.total_out);
1110         pack_size += s.total_out;
1111
1112         free(out);
1113         free(delta);
1114         if (last) {
1115                 if (last->no_swap) {
1116                         last->data = *dat;
1117                 } else {
1118                         strbuf_swap(&last->data, dat);
1119                 }
1120                 last->offset = e->offset;
1121                 last->depth = e->depth;
1122         }
1123         return 0;
1124 }
1125
1126 static void *gfi_unpack_entry(
1127         struct object_entry *oe,
1128         unsigned long *sizep)
1129 {
1130         enum object_type type;
1131         struct packed_git *p = all_packs[oe->pack_id];
1132         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1133                 close_pack_windows(p);
1134                 p->pack_size = pack_size + 20;
1135         }
1136         return unpack_entry(p, oe->offset, &type, sizep);
1137 }
1138
1139 static const char *get_mode(const char *str, uint16_t *modep)
1140 {
1141         unsigned char c;
1142         uint16_t mode = 0;
1143
1144         while ((c = *str++) != ' ') {
1145                 if (c < '0' || c > '7')
1146                         return NULL;
1147                 mode = (mode << 3) + (c - '0');
1148         }
1149         *modep = mode;
1150         return str;
1151 }
1152
1153 static void load_tree(struct tree_entry *root)
1154 {
1155         unsigned char* sha1 = root->versions[1].sha1;
1156         struct object_entry *myoe;
1157         struct tree_content *t;
1158         unsigned long size;
1159         char *buf;
1160         const char *c;
1161
1162         root->tree = t = new_tree_content(8);
1163         if (is_null_sha1(sha1))
1164                 return;
1165
1166         myoe = find_object(sha1);
1167         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1168                 if (myoe->type != OBJ_TREE)
1169                         die("Not a tree: %s", sha1_to_hex(sha1));
1170                 t->delta_depth = myoe->depth;
1171                 buf = gfi_unpack_entry(myoe, &size);
1172         } else {
1173                 enum object_type type;
1174                 buf = read_sha1_file(sha1, &type, &size);
1175                 if (!buf || type != OBJ_TREE)
1176                         die("Can't load tree %s", sha1_to_hex(sha1));
1177         }
1178
1179         c = buf;
1180         while (c != (buf + size)) {
1181                 struct tree_entry *e = new_tree_entry();
1182
1183                 if (t->entry_count == t->entry_capacity)
1184                         root->tree = t = grow_tree_content(t, t->entry_count);
1185                 t->entries[t->entry_count++] = e;
1186
1187                 e->tree = NULL;
1188                 c = get_mode(c, &e->versions[1].mode);
1189                 if (!c)
1190                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1191                 e->versions[0].mode = e->versions[1].mode;
1192                 e->name = to_atom(c, strlen(c));
1193                 c += e->name->str_len + 1;
1194                 hashcpy(e->versions[0].sha1, (unsigned char*)c);
1195                 hashcpy(e->versions[1].sha1, (unsigned char*)c);
1196                 c += 20;
1197         }
1198         free(buf);
1199 }
1200
1201 static int tecmp0 (const void *_a, const void *_b)
1202 {
1203         struct tree_entry *a = *((struct tree_entry**)_a);
1204         struct tree_entry *b = *((struct tree_entry**)_b);
1205         return base_name_compare(
1206                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1207                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1208 }
1209
1210 static int tecmp1 (const void *_a, const void *_b)
1211 {
1212         struct tree_entry *a = *((struct tree_entry**)_a);
1213         struct tree_entry *b = *((struct tree_entry**)_b);
1214         return base_name_compare(
1215                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1216                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1217 }
1218
1219 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1220 {
1221         size_t maxlen = 0;
1222         unsigned int i;
1223
1224         if (!v)
1225                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1226         else
1227                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1228
1229         for (i = 0; i < t->entry_count; i++) {
1230                 if (t->entries[i]->versions[v].mode)
1231                         maxlen += t->entries[i]->name->str_len + 34;
1232         }
1233
1234         strbuf_reset(b);
1235         strbuf_grow(b, maxlen);
1236         for (i = 0; i < t->entry_count; i++) {
1237                 struct tree_entry *e = t->entries[i];
1238                 if (!e->versions[v].mode)
1239                         continue;
1240                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1241                                         e->name->str_dat, '\0');
1242                 strbuf_add(b, e->versions[v].sha1, 20);
1243         }
1244 }
1245
1246 static void store_tree(struct tree_entry *root)
1247 {
1248         struct tree_content *t = root->tree;
1249         unsigned int i, j, del;
1250         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1251         struct object_entry *le;
1252
1253         if (!is_null_sha1(root->versions[1].sha1))
1254                 return;
1255
1256         for (i = 0; i < t->entry_count; i++) {
1257                 if (t->entries[i]->tree)
1258                         store_tree(t->entries[i]);
1259         }
1260
1261         le = find_object(root->versions[0].sha1);
1262         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1263                 mktree(t, 0, &old_tree);
1264                 lo.data = old_tree;
1265                 lo.offset = le->offset;
1266                 lo.depth = t->delta_depth;
1267         }
1268
1269         mktree(t, 1, &new_tree);
1270         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1271
1272         t->delta_depth = lo.depth;
1273         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1274                 struct tree_entry *e = t->entries[i];
1275                 if (e->versions[1].mode) {
1276                         e->versions[0].mode = e->versions[1].mode;
1277                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1278                         t->entries[j++] = e;
1279                 } else {
1280                         release_tree_entry(e);
1281                         del++;
1282                 }
1283         }
1284         t->entry_count -= del;
1285 }
1286
1287 static int tree_content_set(
1288         struct tree_entry *root,
1289         const char *p,
1290         const unsigned char *sha1,
1291         const uint16_t mode,
1292         struct tree_content *subtree)
1293 {
1294         struct tree_content *t = root->tree;
1295         const char *slash1;
1296         unsigned int i, n;
1297         struct tree_entry *e;
1298
1299         slash1 = strchr(p, '/');
1300         if (slash1)
1301                 n = slash1 - p;
1302         else
1303                 n = strlen(p);
1304         if (!n)
1305                 die("Empty path component found in input");
1306         if (!slash1 && !S_ISDIR(mode) && subtree)
1307                 die("Non-directories cannot have subtrees");
1308
1309         for (i = 0; i < t->entry_count; i++) {
1310                 e = t->entries[i];
1311                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1312                         if (!slash1) {
1313                                 if (!S_ISDIR(mode)
1314                                                 && e->versions[1].mode == mode
1315                                                 && !hashcmp(e->versions[1].sha1, sha1))
1316                                         return 0;
1317                                 e->versions[1].mode = mode;
1318                                 hashcpy(e->versions[1].sha1, sha1);
1319                                 if (e->tree)
1320                                         release_tree_content_recursive(e->tree);
1321                                 e->tree = subtree;
1322                                 hashclr(root->versions[1].sha1);
1323                                 return 1;
1324                         }
1325                         if (!S_ISDIR(e->versions[1].mode)) {
1326                                 e->tree = new_tree_content(8);
1327                                 e->versions[1].mode = S_IFDIR;
1328                         }
1329                         if (!e->tree)
1330                                 load_tree(e);
1331                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1332                                 hashclr(root->versions[1].sha1);
1333                                 return 1;
1334                         }
1335                         return 0;
1336                 }
1337         }
1338
1339         if (t->entry_count == t->entry_capacity)
1340                 root->tree = t = grow_tree_content(t, t->entry_count);
1341         e = new_tree_entry();
1342         e->name = to_atom(p, n);
1343         e->versions[0].mode = 0;
1344         hashclr(e->versions[0].sha1);
1345         t->entries[t->entry_count++] = e;
1346         if (slash1) {
1347                 e->tree = new_tree_content(8);
1348                 e->versions[1].mode = S_IFDIR;
1349                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1350         } else {
1351                 e->tree = subtree;
1352                 e->versions[1].mode = mode;
1353                 hashcpy(e->versions[1].sha1, sha1);
1354         }
1355         hashclr(root->versions[1].sha1);
1356         return 1;
1357 }
1358
1359 static int tree_content_remove(
1360         struct tree_entry *root,
1361         const char *p,
1362         struct tree_entry *backup_leaf)
1363 {
1364         struct tree_content *t = root->tree;
1365         const char *slash1;
1366         unsigned int i, n;
1367         struct tree_entry *e;
1368
1369         slash1 = strchr(p, '/');
1370         if (slash1)
1371                 n = slash1 - p;
1372         else
1373                 n = strlen(p);
1374
1375         for (i = 0; i < t->entry_count; i++) {
1376                 e = t->entries[i];
1377                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1378                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1379                                 goto del_entry;
1380                         if (!e->tree)
1381                                 load_tree(e);
1382                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1383                                 for (n = 0; n < e->tree->entry_count; n++) {
1384                                         if (e->tree->entries[n]->versions[1].mode) {
1385                                                 hashclr(root->versions[1].sha1);
1386                                                 return 1;
1387                                         }
1388                                 }
1389                                 backup_leaf = NULL;
1390                                 goto del_entry;
1391                         }
1392                         return 0;
1393                 }
1394         }
1395         return 0;
1396
1397 del_entry:
1398         if (backup_leaf)
1399                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1400         else if (e->tree)
1401                 release_tree_content_recursive(e->tree);
1402         e->tree = NULL;
1403         e->versions[1].mode = 0;
1404         hashclr(e->versions[1].sha1);
1405         hashclr(root->versions[1].sha1);
1406         return 1;
1407 }
1408
1409 static int tree_content_get(
1410         struct tree_entry *root,
1411         const char *p,
1412         struct tree_entry *leaf)
1413 {
1414         struct tree_content *t = root->tree;
1415         const char *slash1;
1416         unsigned int i, n;
1417         struct tree_entry *e;
1418
1419         slash1 = strchr(p, '/');
1420         if (slash1)
1421                 n = slash1 - p;
1422         else
1423                 n = strlen(p);
1424
1425         for (i = 0; i < t->entry_count; i++) {
1426                 e = t->entries[i];
1427                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1428                         if (!slash1) {
1429                                 memcpy(leaf, e, sizeof(*leaf));
1430                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1431                                         leaf->tree = dup_tree_content(e->tree);
1432                                 else
1433                                         leaf->tree = NULL;
1434                                 return 1;
1435                         }
1436                         if (!S_ISDIR(e->versions[1].mode))
1437                                 return 0;
1438                         if (!e->tree)
1439                                 load_tree(e);
1440                         return tree_content_get(e, slash1 + 1, leaf);
1441                 }
1442         }
1443         return 0;
1444 }
1445
1446 static int update_branch(struct branch *b)
1447 {
1448         static const char *msg = "fast-import";
1449         struct ref_lock *lock;
1450         unsigned char old_sha1[20];
1451
1452         if (read_ref(b->name, old_sha1))
1453                 hashclr(old_sha1);
1454         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1455         if (!lock)
1456                 return error("Unable to lock %s", b->name);
1457         if (!force_update && !is_null_sha1(old_sha1)) {
1458                 struct commit *old_cmit, *new_cmit;
1459
1460                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1461                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1462                 if (!old_cmit || !new_cmit) {
1463                         unlock_ref(lock);
1464                         return error("Branch %s is missing commits.", b->name);
1465                 }
1466
1467                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1468                         unlock_ref(lock);
1469                         warning("Not updating %s"
1470                                 " (new tip %s does not contain %s)",
1471                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1472                         return -1;
1473                 }
1474         }
1475         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1476                 return error("Unable to update %s", b->name);
1477         return 0;
1478 }
1479
1480 static void dump_branches(void)
1481 {
1482         unsigned int i;
1483         struct branch *b;
1484
1485         for (i = 0; i < branch_table_sz; i++) {
1486                 for (b = branch_table[i]; b; b = b->table_next_branch)
1487                         failure |= update_branch(b);
1488         }
1489 }
1490
1491 static void dump_tags(void)
1492 {
1493         static const char *msg = "fast-import";
1494         struct tag *t;
1495         struct ref_lock *lock;
1496         char ref_name[PATH_MAX];
1497
1498         for (t = first_tag; t; t = t->next_tag) {
1499                 sprintf(ref_name, "tags/%s", t->name);
1500                 lock = lock_ref_sha1(ref_name, NULL);
1501                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1502                         failure |= error("Unable to update %s", ref_name);
1503         }
1504 }
1505
1506 static void dump_marks_helper(FILE *f,
1507         uintmax_t base,
1508         struct mark_set *m)
1509 {
1510         uintmax_t k;
1511         if (m->shift) {
1512                 for (k = 0; k < 1024; k++) {
1513                         if (m->data.sets[k])
1514                                 dump_marks_helper(f, (base + k) << m->shift,
1515                                         m->data.sets[k]);
1516                 }
1517         } else {
1518                 for (k = 0; k < 1024; k++) {
1519                         if (m->data.marked[k])
1520                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1521                                         sha1_to_hex(m->data.marked[k]->sha1));
1522                 }
1523         }
1524 }
1525
1526 static void dump_marks(void)
1527 {
1528         static struct lock_file mark_lock;
1529         int mark_fd;
1530         FILE *f;
1531
1532         if (!mark_file)
1533                 return;
1534
1535         mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1536         if (mark_fd < 0) {
1537                 failure |= error("Unable to write marks file %s: %s",
1538                         mark_file, strerror(errno));
1539                 return;
1540         }
1541
1542         f = fdopen(mark_fd, "w");
1543         if (!f) {
1544                 rollback_lock_file(&mark_lock);
1545                 failure |= error("Unable to write marks file %s: %s",
1546                         mark_file, strerror(errno));
1547                 return;
1548         }
1549
1550         /*
1551          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1552          * Assign -1 to the lock file descriptor so that commit_lock_file()
1553          * won't try to close() it.
1554          */
1555         mark_lock.fd = -1;
1556
1557         dump_marks_helper(f, 0, marks);
1558         if (ferror(f) || fclose(f)) {
1559                 rollback_lock_file(&mark_lock);
1560                 failure |= error("Unable to write marks file %s: %s",
1561                         mark_file, strerror(errno));
1562                 return;
1563         }
1564
1565         if (commit_lock_file(&mark_lock)) {
1566                 rollback_lock_file(&mark_lock);
1567                 failure |= error("Unable to commit marks file %s: %s",
1568                         mark_file, strerror(errno));
1569                 return;
1570         }
1571 }
1572
1573 static int read_next_command(void)
1574 {
1575         static int stdin_eof = 0;
1576
1577         if (stdin_eof) {
1578                 unread_command_buf = 0;
1579                 return EOF;
1580         }
1581
1582         do {
1583                 if (unread_command_buf) {
1584                         unread_command_buf = 0;
1585                 } else {
1586                         struct recent_command *rc;
1587
1588                         strbuf_detach(&command_buf, NULL);
1589                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1590                         if (stdin_eof)
1591                                 return EOF;
1592
1593                         rc = rc_free;
1594                         if (rc)
1595                                 rc_free = rc->next;
1596                         else {
1597                                 rc = cmd_hist.next;
1598                                 cmd_hist.next = rc->next;
1599                                 cmd_hist.next->prev = &cmd_hist;
1600                                 free(rc->buf);
1601                         }
1602
1603                         rc->buf = command_buf.buf;
1604                         rc->prev = cmd_tail;
1605                         rc->next = cmd_hist.prev;
1606                         rc->prev->next = rc;
1607                         cmd_tail = rc;
1608                 }
1609         } while (command_buf.buf[0] == '#');
1610
1611         return 0;
1612 }
1613
1614 static void skip_optional_lf(void)
1615 {
1616         int term_char = fgetc(stdin);
1617         if (term_char != '\n' && term_char != EOF)
1618                 ungetc(term_char, stdin);
1619 }
1620
1621 static void cmd_mark(void)
1622 {
1623         if (!prefixcmp(command_buf.buf, "mark :")) {
1624                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1625                 read_next_command();
1626         }
1627         else
1628                 next_mark = 0;
1629 }
1630
1631 static void cmd_data(struct strbuf *sb)
1632 {
1633         strbuf_reset(sb);
1634
1635         if (prefixcmp(command_buf.buf, "data "))
1636                 die("Expected 'data n' command, found: %s", command_buf.buf);
1637
1638         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1639                 char *term = xstrdup(command_buf.buf + 5 + 2);
1640                 size_t term_len = command_buf.len - 5 - 2;
1641
1642                 strbuf_detach(&command_buf, NULL);
1643                 for (;;) {
1644                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1645                                 die("EOF in data (terminator '%s' not found)", term);
1646                         if (term_len == command_buf.len
1647                                 && !strcmp(term, command_buf.buf))
1648                                 break;
1649                         strbuf_addbuf(sb, &command_buf);
1650                         strbuf_addch(sb, '\n');
1651                 }
1652                 free(term);
1653         }
1654         else {
1655                 size_t n = 0, length;
1656
1657                 length = strtoul(command_buf.buf + 5, NULL, 10);
1658
1659                 while (n < length) {
1660                         size_t s = strbuf_fread(sb, length - n, stdin);
1661                         if (!s && feof(stdin))
1662                                 die("EOF in data (%lu bytes remaining)",
1663                                         (unsigned long)(length - n));
1664                         n += s;
1665                 }
1666         }
1667
1668         skip_optional_lf();
1669 }
1670
1671 static int validate_raw_date(const char *src, char *result, int maxlen)
1672 {
1673         const char *orig_src = src;
1674         char *endp, sign;
1675
1676         strtoul(src, &endp, 10);
1677         if (endp == src || *endp != ' ')
1678                 return -1;
1679
1680         src = endp + 1;
1681         if (*src != '-' && *src != '+')
1682                 return -1;
1683         sign = *src;
1684
1685         strtoul(src + 1, &endp, 10);
1686         if (endp == src || *endp || (endp - orig_src) >= maxlen)
1687                 return -1;
1688
1689         strcpy(result, orig_src);
1690         return 0;
1691 }
1692
1693 static char *parse_ident(const char *buf)
1694 {
1695         const char *gt;
1696         size_t name_len;
1697         char *ident;
1698
1699         gt = strrchr(buf, '>');
1700         if (!gt)
1701                 die("Missing > in ident string: %s", buf);
1702         gt++;
1703         if (*gt != ' ')
1704                 die("Missing space after > in ident string: %s", buf);
1705         gt++;
1706         name_len = gt - buf;
1707         ident = xmalloc(name_len + 24);
1708         strncpy(ident, buf, name_len);
1709
1710         switch (whenspec) {
1711         case WHENSPEC_RAW:
1712                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1713                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1714                 break;
1715         case WHENSPEC_RFC2822:
1716                 if (parse_date(gt, ident + name_len, 24) < 0)
1717                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1718                 break;
1719         case WHENSPEC_NOW:
1720                 if (strcmp("now", gt))
1721                         die("Date in ident must be 'now': %s", buf);
1722                 datestamp(ident + name_len, 24);
1723                 break;
1724         }
1725
1726         return ident;
1727 }
1728
1729 static void cmd_new_blob(void)
1730 {
1731         static struct strbuf buf = STRBUF_INIT;
1732
1733         read_next_command();
1734         cmd_mark();
1735         cmd_data(&buf);
1736         store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1737 }
1738
1739 static void unload_one_branch(void)
1740 {
1741         while (cur_active_branches
1742                 && cur_active_branches >= max_active_branches) {
1743                 uintmax_t min_commit = ULONG_MAX;
1744                 struct branch *e, *l = NULL, *p = NULL;
1745
1746                 for (e = active_branches; e; e = e->active_next_branch) {
1747                         if (e->last_commit < min_commit) {
1748                                 p = l;
1749                                 min_commit = e->last_commit;
1750                         }
1751                         l = e;
1752                 }
1753
1754                 if (p) {
1755                         e = p->active_next_branch;
1756                         p->active_next_branch = e->active_next_branch;
1757                 } else {
1758                         e = active_branches;
1759                         active_branches = e->active_next_branch;
1760                 }
1761                 e->active = 0;
1762                 e->active_next_branch = NULL;
1763                 if (e->branch_tree.tree) {
1764                         release_tree_content_recursive(e->branch_tree.tree);
1765                         e->branch_tree.tree = NULL;
1766                 }
1767                 cur_active_branches--;
1768         }
1769 }
1770
1771 static void load_branch(struct branch *b)
1772 {
1773         load_tree(&b->branch_tree);
1774         if (!b->active) {
1775                 b->active = 1;
1776                 b->active_next_branch = active_branches;
1777                 active_branches = b;
1778                 cur_active_branches++;
1779                 branch_load_count++;
1780         }
1781 }
1782
1783 static void file_change_m(struct branch *b)
1784 {
1785         const char *p = command_buf.buf + 2;
1786         static struct strbuf uq = STRBUF_INIT;
1787         const char *endp;
1788         struct object_entry *oe = oe;
1789         unsigned char sha1[20];
1790         uint16_t mode, inline_data = 0;
1791
1792         p = get_mode(p, &mode);
1793         if (!p)
1794                 die("Corrupt mode: %s", command_buf.buf);
1795         switch (mode) {
1796         case S_IFREG | 0644:
1797         case S_IFREG | 0755:
1798         case S_IFLNK:
1799         case 0644:
1800         case 0755:
1801                 /* ok */
1802                 break;
1803         default:
1804                 die("Corrupt mode: %s", command_buf.buf);
1805         }
1806
1807         if (*p == ':') {
1808                 char *x;
1809                 oe = find_mark(strtoumax(p + 1, &x, 10));
1810                 hashcpy(sha1, oe->sha1);
1811                 p = x;
1812         } else if (!prefixcmp(p, "inline")) {
1813                 inline_data = 1;
1814                 p += 6;
1815         } else {
1816                 if (get_sha1_hex(p, sha1))
1817                         die("Invalid SHA1: %s", command_buf.buf);
1818                 oe = find_object(sha1);
1819                 p += 40;
1820         }
1821         if (*p++ != ' ')
1822                 die("Missing space after SHA1: %s", command_buf.buf);
1823
1824         strbuf_reset(&uq);
1825         if (!unquote_c_style(&uq, p, &endp)) {
1826                 if (*endp)
1827                         die("Garbage after path in: %s", command_buf.buf);
1828                 p = uq.buf;
1829         }
1830
1831         if (inline_data) {
1832                 static struct strbuf buf = STRBUF_INIT;
1833
1834                 if (p != uq.buf) {
1835                         strbuf_addstr(&uq, p);
1836                         p = uq.buf;
1837                 }
1838                 read_next_command();
1839                 cmd_data(&buf);
1840                 store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1841         } else if (oe) {
1842                 if (oe->type != OBJ_BLOB)
1843                         die("Not a blob (actually a %s): %s",
1844                                 typename(oe->type), command_buf.buf);
1845         } else {
1846                 enum object_type type = sha1_object_info(sha1, NULL);
1847                 if (type < 0)
1848                         die("Blob not found: %s", command_buf.buf);
1849                 if (type != OBJ_BLOB)
1850                         die("Not a blob (actually a %s): %s",
1851                             typename(type), command_buf.buf);
1852         }
1853
1854         tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode, NULL);
1855 }
1856
1857 static void file_change_d(struct branch *b)
1858 {
1859         const char *p = command_buf.buf + 2;
1860         static struct strbuf uq = STRBUF_INIT;
1861         const char *endp;
1862
1863         strbuf_reset(&uq);
1864         if (!unquote_c_style(&uq, p, &endp)) {
1865                 if (*endp)
1866                         die("Garbage after path in: %s", command_buf.buf);
1867                 p = uq.buf;
1868         }
1869         tree_content_remove(&b->branch_tree, p, NULL);
1870 }
1871
1872 static void file_change_cr(struct branch *b, int rename)
1873 {
1874         const char *s, *d;
1875         static struct strbuf s_uq = STRBUF_INIT;
1876         static struct strbuf d_uq = STRBUF_INIT;
1877         const char *endp;
1878         struct tree_entry leaf;
1879
1880         s = command_buf.buf + 2;
1881         strbuf_reset(&s_uq);
1882         if (!unquote_c_style(&s_uq, s, &endp)) {
1883                 if (*endp != ' ')
1884                         die("Missing space after source: %s", command_buf.buf);
1885         } else {
1886                 endp = strchr(s, ' ');
1887                 if (!endp)
1888                         die("Missing space after source: %s", command_buf.buf);
1889                 strbuf_add(&s_uq, s, endp - s);
1890         }
1891         s = s_uq.buf;
1892
1893         endp++;
1894         if (!*endp)
1895                 die("Missing dest: %s", command_buf.buf);
1896
1897         d = endp;
1898         strbuf_reset(&d_uq);
1899         if (!unquote_c_style(&d_uq, d, &endp)) {
1900                 if (*endp)
1901                         die("Garbage after dest in: %s", command_buf.buf);
1902                 d = d_uq.buf;
1903         }
1904
1905         memset(&leaf, 0, sizeof(leaf));
1906         if (rename)
1907                 tree_content_remove(&b->branch_tree, s, &leaf);
1908         else
1909                 tree_content_get(&b->branch_tree, s, &leaf);
1910         if (!leaf.versions[1].mode)
1911                 die("Path %s not in branch", s);
1912         tree_content_set(&b->branch_tree, d,
1913                 leaf.versions[1].sha1,
1914                 leaf.versions[1].mode,
1915                 leaf.tree);
1916 }
1917
1918 static void file_change_deleteall(struct branch *b)
1919 {
1920         release_tree_content_recursive(b->branch_tree.tree);
1921         hashclr(b->branch_tree.versions[0].sha1);
1922         hashclr(b->branch_tree.versions[1].sha1);
1923         load_tree(&b->branch_tree);
1924 }
1925
1926 static void cmd_from_commit(struct branch *b, char *buf, unsigned long size)
1927 {
1928         if (!buf || size < 46)
1929                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
1930         if (memcmp("tree ", buf, 5)
1931                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1932                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1933         hashcpy(b->branch_tree.versions[0].sha1,
1934                 b->branch_tree.versions[1].sha1);
1935 }
1936
1937 static void cmd_from_existing(struct branch *b)
1938 {
1939         if (is_null_sha1(b->sha1)) {
1940                 hashclr(b->branch_tree.versions[0].sha1);
1941                 hashclr(b->branch_tree.versions[1].sha1);
1942         } else {
1943                 unsigned long size;
1944                 char *buf;
1945
1946                 buf = read_object_with_reference(b->sha1,
1947                         commit_type, &size, b->sha1);
1948                 cmd_from_commit(b, buf, size);
1949                 free(buf);
1950         }
1951 }
1952
1953 static int cmd_from(struct branch *b)
1954 {
1955         const char *from;
1956         struct branch *s;
1957
1958         if (prefixcmp(command_buf.buf, "from "))
1959                 return 0;
1960
1961         if (b->branch_tree.tree) {
1962                 release_tree_content_recursive(b->branch_tree.tree);
1963                 b->branch_tree.tree = NULL;
1964         }
1965
1966         from = strchr(command_buf.buf, ' ') + 1;
1967         s = lookup_branch(from);
1968         if (b == s)
1969                 die("Can't create a branch from itself: %s", b->name);
1970         else if (s) {
1971                 unsigned char *t = s->branch_tree.versions[1].sha1;
1972                 hashcpy(b->sha1, s->sha1);
1973                 hashcpy(b->branch_tree.versions[0].sha1, t);
1974                 hashcpy(b->branch_tree.versions[1].sha1, t);
1975         } else if (*from == ':') {
1976                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
1977                 struct object_entry *oe = find_mark(idnum);
1978                 if (oe->type != OBJ_COMMIT)
1979                         die("Mark :%" PRIuMAX " not a commit", idnum);
1980                 hashcpy(b->sha1, oe->sha1);
1981                 if (oe->pack_id != MAX_PACK_ID) {
1982                         unsigned long size;
1983                         char *buf = gfi_unpack_entry(oe, &size);
1984                         cmd_from_commit(b, buf, size);
1985                         free(buf);
1986                 } else
1987                         cmd_from_existing(b);
1988         } else if (!get_sha1(from, b->sha1))
1989                 cmd_from_existing(b);
1990         else
1991                 die("Invalid ref name or SHA1 expression: %s", from);
1992
1993         read_next_command();
1994         return 1;
1995 }
1996
1997 static struct hash_list *cmd_merge(unsigned int *count)
1998 {
1999         struct hash_list *list = NULL, *n, *e = e;
2000         const char *from;
2001         struct branch *s;
2002
2003         *count = 0;
2004         while (!prefixcmp(command_buf.buf, "merge ")) {
2005                 from = strchr(command_buf.buf, ' ') + 1;
2006                 n = xmalloc(sizeof(*n));
2007                 s = lookup_branch(from);
2008                 if (s)
2009                         hashcpy(n->sha1, s->sha1);
2010                 else if (*from == ':') {
2011                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2012                         struct object_entry *oe = find_mark(idnum);
2013                         if (oe->type != OBJ_COMMIT)
2014                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2015                         hashcpy(n->sha1, oe->sha1);
2016                 } else if (!get_sha1(from, n->sha1)) {
2017                         unsigned long size;
2018                         char *buf = read_object_with_reference(n->sha1,
2019                                 commit_type, &size, n->sha1);
2020                         if (!buf || size < 46)
2021                                 die("Not a valid commit: %s", from);
2022                         free(buf);
2023                 } else
2024                         die("Invalid ref name or SHA1 expression: %s", from);
2025
2026                 n->next = NULL;
2027                 if (list)
2028                         e->next = n;
2029                 else
2030                         list = n;
2031                 e = n;
2032                 (*count)++;
2033                 read_next_command();
2034         }
2035         return list;
2036 }
2037
2038 static void cmd_new_commit(void)
2039 {
2040         static struct strbuf msg = STRBUF_INIT;
2041         struct branch *b;
2042         char *sp;
2043         char *author = NULL;
2044         char *committer = NULL;
2045         struct hash_list *merge_list = NULL;
2046         unsigned int merge_count;
2047
2048         /* Obtain the branch name from the rest of our command */
2049         sp = strchr(command_buf.buf, ' ') + 1;
2050         b = lookup_branch(sp);
2051         if (!b)
2052                 b = new_branch(sp);
2053
2054         read_next_command();
2055         cmd_mark();
2056         if (!prefixcmp(command_buf.buf, "author ")) {
2057                 author = parse_ident(command_buf.buf + 7);
2058                 read_next_command();
2059         }
2060         if (!prefixcmp(command_buf.buf, "committer ")) {
2061                 committer = parse_ident(command_buf.buf + 10);
2062                 read_next_command();
2063         }
2064         if (!committer)
2065                 die("Expected committer but didn't get one");
2066         cmd_data(&msg);
2067         read_next_command();
2068         cmd_from(b);
2069         merge_list = cmd_merge(&merge_count);
2070
2071         /* ensure the branch is active/loaded */
2072         if (!b->branch_tree.tree || !max_active_branches) {
2073                 unload_one_branch();
2074                 load_branch(b);
2075         }
2076
2077         /* file_change* */
2078         while (command_buf.len > 0) {
2079                 if (!prefixcmp(command_buf.buf, "M "))
2080                         file_change_m(b);
2081                 else if (!prefixcmp(command_buf.buf, "D "))
2082                         file_change_d(b);
2083                 else if (!prefixcmp(command_buf.buf, "R "))
2084                         file_change_cr(b, 1);
2085                 else if (!prefixcmp(command_buf.buf, "C "))
2086                         file_change_cr(b, 0);
2087                 else if (!strcmp("deleteall", command_buf.buf))
2088                         file_change_deleteall(b);
2089                 else {
2090                         unread_command_buf = 1;
2091                         break;
2092                 }
2093                 if (read_next_command() == EOF)
2094                         break;
2095         }
2096
2097         /* build the tree and the commit */
2098         store_tree(&b->branch_tree);
2099         hashcpy(b->branch_tree.versions[0].sha1,
2100                 b->branch_tree.versions[1].sha1);
2101
2102         strbuf_reset(&new_data);
2103         strbuf_addf(&new_data, "tree %s\n",
2104                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2105         if (!is_null_sha1(b->sha1))
2106                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2107         while (merge_list) {
2108                 struct hash_list *next = merge_list->next;
2109                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2110                 free(merge_list);
2111                 merge_list = next;
2112         }
2113         strbuf_addf(&new_data,
2114                 "author %s\n"
2115                 "committer %s\n"
2116                 "\n",
2117                 author ? author : committer, committer);
2118         strbuf_addbuf(&new_data, &msg);
2119         free(author);
2120         free(committer);
2121
2122         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2123                 b->pack_id = pack_id;
2124         b->last_commit = object_count_by_type[OBJ_COMMIT];
2125 }
2126
2127 static void cmd_new_tag(void)
2128 {
2129         static struct strbuf msg = STRBUF_INIT;
2130         char *sp;
2131         const char *from;
2132         char *tagger;
2133         struct branch *s;
2134         struct tag *t;
2135         uintmax_t from_mark = 0;
2136         unsigned char sha1[20];
2137
2138         /* Obtain the new tag name from the rest of our command */
2139         sp = strchr(command_buf.buf, ' ') + 1;
2140         t = pool_alloc(sizeof(struct tag));
2141         t->next_tag = NULL;
2142         t->name = pool_strdup(sp);
2143         if (last_tag)
2144                 last_tag->next_tag = t;
2145         else
2146                 first_tag = t;
2147         last_tag = t;
2148         read_next_command();
2149
2150         /* from ... */
2151         if (prefixcmp(command_buf.buf, "from "))
2152                 die("Expected from command, got %s", command_buf.buf);
2153         from = strchr(command_buf.buf, ' ') + 1;
2154         s = lookup_branch(from);
2155         if (s) {
2156                 hashcpy(sha1, s->sha1);
2157         } else if (*from == ':') {
2158                 struct object_entry *oe;
2159                 from_mark = strtoumax(from + 1, NULL, 10);
2160                 oe = find_mark(from_mark);
2161                 if (oe->type != OBJ_COMMIT)
2162                         die("Mark :%" PRIuMAX " not a commit", from_mark);
2163                 hashcpy(sha1, oe->sha1);
2164         } else if (!get_sha1(from, sha1)) {
2165                 unsigned long size;
2166                 char *buf;
2167
2168                 buf = read_object_with_reference(sha1,
2169                         commit_type, &size, sha1);
2170                 if (!buf || size < 46)
2171                         die("Not a valid commit: %s", from);
2172                 free(buf);
2173         } else
2174                 die("Invalid ref name or SHA1 expression: %s", from);
2175         read_next_command();
2176
2177         /* tagger ... */
2178         if (prefixcmp(command_buf.buf, "tagger "))
2179                 die("Expected tagger command, got %s", command_buf.buf);
2180         tagger = parse_ident(command_buf.buf + 7);
2181
2182         /* tag payload/message */
2183         read_next_command();
2184         cmd_data(&msg);
2185
2186         /* build the tag object */
2187         strbuf_reset(&new_data);
2188         strbuf_addf(&new_data,
2189                 "object %s\n"
2190                 "type %s\n"
2191                 "tag %s\n"
2192                 "tagger %s\n"
2193                 "\n",
2194                 sha1_to_hex(sha1), commit_type, t->name, tagger);
2195         strbuf_addbuf(&new_data, &msg);
2196         free(tagger);
2197
2198         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2199                 t->pack_id = MAX_PACK_ID;
2200         else
2201                 t->pack_id = pack_id;
2202 }
2203
2204 static void cmd_reset_branch(void)
2205 {
2206         struct branch *b;
2207         char *sp;
2208
2209         /* Obtain the branch name from the rest of our command */
2210         sp = strchr(command_buf.buf, ' ') + 1;
2211         b = lookup_branch(sp);
2212         if (b) {
2213                 hashclr(b->sha1);
2214                 hashclr(b->branch_tree.versions[0].sha1);
2215                 hashclr(b->branch_tree.versions[1].sha1);
2216                 if (b->branch_tree.tree) {
2217                         release_tree_content_recursive(b->branch_tree.tree);
2218                         b->branch_tree.tree = NULL;
2219                 }
2220         }
2221         else
2222                 b = new_branch(sp);
2223         read_next_command();
2224         if (!cmd_from(b) && command_buf.len > 0)
2225                 unread_command_buf = 1;
2226 }
2227
2228 static void cmd_checkpoint(void)
2229 {
2230         if (object_count) {
2231                 cycle_packfile();
2232                 dump_branches();
2233                 dump_tags();
2234                 dump_marks();
2235         }
2236         skip_optional_lf();
2237 }
2238
2239 static void cmd_progress(void)
2240 {
2241         fwrite(command_buf.buf, 1, command_buf.len, stdout);
2242         fputc('\n', stdout);
2243         fflush(stdout);
2244         skip_optional_lf();
2245 }
2246
2247 static void import_marks(const char *input_file)
2248 {
2249         char line[512];
2250         FILE *f = fopen(input_file, "r");
2251         if (!f)
2252                 die("cannot read %s: %s", input_file, strerror(errno));
2253         while (fgets(line, sizeof(line), f)) {
2254                 uintmax_t mark;
2255                 char *end;
2256                 unsigned char sha1[20];
2257                 struct object_entry *e;
2258
2259                 end = strchr(line, '\n');
2260                 if (line[0] != ':' || !end)
2261                         die("corrupt mark line: %s", line);
2262                 *end = 0;
2263                 mark = strtoumax(line + 1, &end, 10);
2264                 if (!mark || end == line + 1
2265                         || *end != ' ' || get_sha1(end + 1, sha1))
2266                         die("corrupt mark line: %s", line);
2267                 e = find_object(sha1);
2268                 if (!e) {
2269                         enum object_type type = sha1_object_info(sha1, NULL);
2270                         if (type < 0)
2271                                 die("object not found: %s", sha1_to_hex(sha1));
2272                         e = insert_object(sha1);
2273                         e->type = type;
2274                         e->pack_id = MAX_PACK_ID;
2275                         e->offset = 1; /* just not zero! */
2276                 }
2277                 insert_mark(mark, e);
2278         }
2279         fclose(f);
2280 }
2281
2282 static const char fast_import_usage[] =
2283 "git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2284
2285 int main(int argc, const char **argv)
2286 {
2287         unsigned int i, show_stats = 1;
2288
2289         git_config(git_default_config);
2290         alloc_objects(object_entry_alloc);
2291         strbuf_init(&command_buf, 0);
2292         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2293         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2294         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2295         marks = pool_calloc(1, sizeof(struct mark_set));
2296
2297         for (i = 1; i < argc; i++) {
2298                 const char *a = argv[i];
2299
2300                 if (*a != '-' || !strcmp(a, "--"))
2301                         break;
2302                 else if (!prefixcmp(a, "--date-format=")) {
2303                         const char *fmt = a + 14;
2304                         if (!strcmp(fmt, "raw"))
2305                                 whenspec = WHENSPEC_RAW;
2306                         else if (!strcmp(fmt, "rfc2822"))
2307                                 whenspec = WHENSPEC_RFC2822;
2308                         else if (!strcmp(fmt, "now"))
2309                                 whenspec = WHENSPEC_NOW;
2310                         else
2311                                 die("unknown --date-format argument %s", fmt);
2312                 }
2313                 else if (!prefixcmp(a, "--max-pack-size="))
2314                         max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2315                 else if (!prefixcmp(a, "--depth=")) {
2316                         max_depth = strtoul(a + 8, NULL, 0);
2317                         if (max_depth > MAX_DEPTH)
2318                                 die("--depth cannot exceed %u", MAX_DEPTH);
2319                 }
2320                 else if (!prefixcmp(a, "--active-branches="))
2321                         max_active_branches = strtoul(a + 18, NULL, 0);
2322                 else if (!prefixcmp(a, "--import-marks="))
2323                         import_marks(a + 15);
2324                 else if (!prefixcmp(a, "--export-marks="))
2325                         mark_file = a + 15;
2326                 else if (!prefixcmp(a, "--export-pack-edges=")) {
2327                         if (pack_edges)
2328                                 fclose(pack_edges);
2329                         pack_edges = fopen(a + 20, "a");
2330                         if (!pack_edges)
2331                                 die("Cannot open %s: %s", a + 20, strerror(errno));
2332                 } else if (!strcmp(a, "--force"))
2333                         force_update = 1;
2334                 else if (!strcmp(a, "--quiet"))
2335                         show_stats = 0;
2336                 else if (!strcmp(a, "--stats"))
2337                         show_stats = 1;
2338                 else
2339                         die("unknown option %s", a);
2340         }
2341         if (i != argc)
2342                 usage(fast_import_usage);
2343
2344         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2345         for (i = 0; i < (cmd_save - 1); i++)
2346                 rc_free[i].next = &rc_free[i + 1];
2347         rc_free[cmd_save - 1].next = NULL;
2348
2349         prepare_packed_git();
2350         start_packfile();
2351         set_die_routine(die_nicely);
2352         while (read_next_command() != EOF) {
2353                 if (!strcmp("blob", command_buf.buf))
2354                         cmd_new_blob();
2355                 else if (!prefixcmp(command_buf.buf, "commit "))
2356                         cmd_new_commit();
2357                 else if (!prefixcmp(command_buf.buf, "tag "))
2358                         cmd_new_tag();
2359                 else if (!prefixcmp(command_buf.buf, "reset "))
2360                         cmd_reset_branch();
2361                 else if (!strcmp("checkpoint", command_buf.buf))
2362                         cmd_checkpoint();
2363                 else if (!prefixcmp(command_buf.buf, "progress "))
2364                         cmd_progress();
2365                 else
2366                         die("Unsupported command: %s", command_buf.buf);
2367         }
2368         end_packfile();
2369
2370         dump_branches();
2371         dump_tags();
2372         unkeep_all_packs();
2373         dump_marks();
2374
2375         if (pack_edges)
2376                 fclose(pack_edges);
2377
2378         if (show_stats) {
2379                 uintmax_t total_count = 0, duplicate_count = 0;
2380                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2381                         total_count += object_count_by_type[i];
2382                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2383                         duplicate_count += duplicate_count_by_type[i];
2384
2385                 fprintf(stderr, "%s statistics:\n", argv[0]);
2386                 fprintf(stderr, "---------------------------------------------------------------------\n");
2387                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2388                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2389                 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]);
2390                 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]);
2391                 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]);
2392                 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]);
2393                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2394                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2395                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2396                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2397                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2398                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2399                 fprintf(stderr, "---------------------------------------------------------------------\n");
2400                 pack_report();
2401                 fprintf(stderr, "---------------------------------------------------------------------\n");
2402                 fprintf(stderr, "\n");
2403         }
2404
2405         return failure ? 1 : 0;
2406 }