Merge branch 'jc/maint-1.6.1-checkout-m-custom-merge' into maint
[git] / fast-import.c
1 /*
2 (See Documentation/git-fast-import.txt for maintained documentation.)
3 Format of STDIN stream:
4
5   stream ::= cmd*;
6
7   cmd ::= new_blob
8         | new_commit
9         | new_tag
10         | reset_branch
11         | checkpoint
12         | progress
13         ;
14
15   new_blob ::= 'blob' lf
16     mark?
17     file_content;
18   file_content ::= data;
19
20   new_commit ::= 'commit' sp ref_str lf
21     mark?
22     ('author' (sp name)? sp '<' email '>' sp when lf)?
23     'committer' (sp name)? sp '<' email '>' sp when lf
24     commit_msg
25     ('from' sp committish lf)?
26     ('merge' sp committish lf)*
27     file_change*
28     lf?;
29   commit_msg ::= data;
30
31   file_change ::= file_clr
32     | file_del
33     | file_rnm
34     | file_cpy
35     | file_obm
36     | file_inm;
37   file_clr ::= 'deleteall' lf;
38   file_del ::= 'D' sp path_str lf;
39   file_rnm ::= 'R' sp path_str sp path_str lf;
40   file_cpy ::= 'C' sp path_str sp path_str lf;
41   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
42   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
43     data;
44   note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
45   note_inm ::= 'N' sp 'inline' sp committish lf
46     data;
47
48   new_tag ::= 'tag' sp tag_str lf
49     'from' sp committish lf
50     ('tagger' (sp name)? sp '<' email '>' sp when lf)?
51     tag_msg;
52   tag_msg ::= data;
53
54   reset_branch ::= 'reset' sp ref_str lf
55     ('from' sp committish lf)?
56     lf?;
57
58   checkpoint ::= 'checkpoint' lf
59     lf?;
60
61   progress ::= 'progress' sp not_lf* lf
62     lf?;
63
64      # note: the first idnum in a stream should be 1 and subsequent
65      # idnums should not have gaps between values as this will cause
66      # the stream parser to reserve space for the gapped values.  An
67      # idnum can be updated in the future to a new object by issuing
68      # a new mark directive with the old idnum.
69      #
70   mark ::= 'mark' sp idnum lf;
71   data ::= (delimited_data | exact_data)
72     lf?;
73
74     # note: delim may be any string but must not contain lf.
75     # data_line may contain any data but must not be exactly
76     # delim.
77   delimited_data ::= 'data' sp '<<' delim lf
78     (data_line lf)*
79     delim lf;
80
81      # note: declen indicates the length of binary_data in bytes.
82      # declen does not include the lf preceding the binary data.
83      #
84   exact_data ::= 'data' sp declen lf
85     binary_data;
86
87      # note: quoted strings are C-style quoting supporting \c for
88      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
89      # is the signed byte value in octal.  Note that the only
90      # characters which must actually be escaped to protect the
91      # stream formatting is: \, " and LF.  Otherwise these values
92      # are UTF8.
93      #
94   committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
95   ref_str     ::= ref;
96   sha1exp_str ::= sha1exp;
97   tag_str     ::= tag;
98   path_str    ::= path    | '"' quoted(path)    '"' ;
99   mode        ::= '100644' | '644'
100                 | '100755' | '755'
101                 | '120000'
102                 ;
103
104   declen ::= # unsigned 32 bit value, ascii base10 notation;
105   bigint ::= # unsigned integer value, ascii base10 notation;
106   binary_data ::= # file content, not interpreted;
107
108   when         ::= raw_when | rfc2822_when;
109   raw_when     ::= ts sp tz;
110   rfc2822_when ::= # Valid RFC 2822 date and time;
111
112   sp ::= # ASCII space character;
113   lf ::= # ASCII newline (LF) character;
114
115      # note: a colon (':') must precede the numerical value assigned to
116      # an idnum.  This is to distinguish it from a ref or tag name as
117      # GIT does not permit ':' in ref or tag strings.
118      #
119   idnum   ::= ':' bigint;
120   path    ::= # GIT style file path, e.g. "a/b/c";
121   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
122   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
123   sha1exp ::= # Any valid GIT SHA1 expression;
124   hexsha1 ::= # SHA1 in hexadecimal format;
125
126      # note: name and email are UTF8 strings, however name must not
127      # contain '<' or lf and email must not contain any of the
128      # following: '<', '>', lf.
129      #
130   name  ::= # valid GIT author/committer name;
131   email ::= # valid GIT author/committer email;
132   ts    ::= # time since the epoch in seconds, ascii base10 notation;
133   tz    ::= # GIT style timezone;
134
135      # note: comments may appear anywhere in the input, except
136      # within a data command.  Any form of the data command
137      # always escapes the related input from comment processing.
138      #
139      # In case it is not clear, the '#' that starts the comment
140      # must be the first character on that line (an lf
141      # preceded it).
142      #
143   comment ::= '#' not_lf* lf;
144   not_lf  ::= # Any byte that is not ASCII newline (LF);
145 */
146
147 #include "builtin.h"
148 #include "cache.h"
149 #include "object.h"
150 #include "blob.h"
151 #include "tree.h"
152 #include "commit.h"
153 #include "delta.h"
154 #include "pack.h"
155 #include "refs.h"
156 #include "csum-file.h"
157 #include "quote.h"
158 #include "exec_cmd.h"
159
160 #define PACK_ID_BITS 16
161 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
162 #define DEPTH_BITS 13
163 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
164
165 struct object_entry
166 {
167         struct object_entry *next;
168         uint32_t offset;
169         uint32_t type : TYPE_BITS,
170                 pack_id : PACK_ID_BITS,
171                 depth : DEPTH_BITS;
172         unsigned char sha1[20];
173 };
174
175 struct object_entry_pool
176 {
177         struct object_entry_pool *next_pool;
178         struct object_entry *next_free;
179         struct object_entry *end;
180         struct object_entry entries[FLEX_ARRAY]; /* more */
181 };
182
183 struct mark_set
184 {
185         union {
186                 struct object_entry *marked[1024];
187                 struct mark_set *sets[1024];
188         } data;
189         unsigned int shift;
190 };
191
192 struct last_object
193 {
194         struct strbuf data;
195         uint32_t offset;
196         unsigned int depth;
197         unsigned no_swap : 1;
198 };
199
200 struct mem_pool
201 {
202         struct mem_pool *next_pool;
203         char *next_free;
204         char *end;
205         uintmax_t space[FLEX_ARRAY]; /* more */
206 };
207
208 struct atom_str
209 {
210         struct atom_str *next_atom;
211         unsigned short str_len;
212         char str_dat[FLEX_ARRAY]; /* more */
213 };
214
215 struct tree_content;
216 struct tree_entry
217 {
218         struct tree_content *tree;
219         struct atom_str *name;
220         struct tree_entry_ms
221         {
222                 uint16_t mode;
223                 unsigned char sha1[20];
224         } versions[2];
225 };
226
227 struct tree_content
228 {
229         unsigned int entry_capacity; /* must match avail_tree_content */
230         unsigned int entry_count;
231         unsigned int delta_depth;
232         struct tree_entry *entries[FLEX_ARRAY]; /* more */
233 };
234
235 struct avail_tree_content
236 {
237         unsigned int entry_capacity; /* must match tree_content */
238         struct avail_tree_content *next_avail;
239 };
240
241 struct branch
242 {
243         struct branch *table_next_branch;
244         struct branch *active_next_branch;
245         const char *name;
246         struct tree_entry branch_tree;
247         uintmax_t last_commit;
248         unsigned active : 1;
249         unsigned pack_id : PACK_ID_BITS;
250         unsigned char sha1[20];
251 };
252
253 struct tag
254 {
255         struct tag *next_tag;
256         const char *name;
257         unsigned int pack_id;
258         unsigned char sha1[20];
259 };
260
261 struct hash_list
262 {
263         struct hash_list *next;
264         unsigned char sha1[20];
265 };
266
267 typedef enum {
268         WHENSPEC_RAW = 1,
269         WHENSPEC_RFC2822,
270         WHENSPEC_NOW,
271 } whenspec_type;
272
273 struct recent_command
274 {
275         struct recent_command *prev;
276         struct recent_command *next;
277         char *buf;
278 };
279
280 /* Configured limits on output */
281 static unsigned long max_depth = 10;
282 static off_t max_packsize = (1LL << 32) - 1;
283 static int force_update;
284 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
285 static int pack_compression_seen;
286
287 /* Stats and misc. counters */
288 static uintmax_t alloc_count;
289 static uintmax_t marks_set_count;
290 static uintmax_t object_count_by_type[1 << TYPE_BITS];
291 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
292 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
293 static unsigned long object_count;
294 static unsigned long branch_count;
295 static unsigned long branch_load_count;
296 static int failure;
297 static FILE *pack_edges;
298
299 /* Memory pools */
300 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
301 static size_t total_allocd;
302 static struct mem_pool *mem_pool;
303
304 /* Atom management */
305 static unsigned int atom_table_sz = 4451;
306 static unsigned int atom_cnt;
307 static struct atom_str **atom_table;
308
309 /* The .pack file being generated */
310 static unsigned int pack_id;
311 static struct packed_git *pack_data;
312 static struct packed_git **all_packs;
313 static unsigned long pack_size;
314
315 /* Table of objects we've written. */
316 static unsigned int object_entry_alloc = 5000;
317 static struct object_entry_pool *blocks;
318 static struct object_entry *object_table[1 << 16];
319 static struct mark_set *marks;
320 static const char *mark_file;
321
322 /* Our last blob */
323 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
324
325 /* Tree management */
326 static unsigned int tree_entry_alloc = 1000;
327 static void *avail_tree_entry;
328 static unsigned int avail_tree_table_sz = 100;
329 static struct avail_tree_content **avail_tree_table;
330 static struct strbuf old_tree = STRBUF_INIT;
331 static struct strbuf new_tree = STRBUF_INIT;
332
333 /* Branch data */
334 static unsigned long max_active_branches = 5;
335 static unsigned long cur_active_branches;
336 static unsigned long branch_table_sz = 1039;
337 static struct branch **branch_table;
338 static struct branch *active_branches;
339
340 /* Tag data */
341 static struct tag *first_tag;
342 static struct tag *last_tag;
343
344 /* Input stream parsing */
345 static whenspec_type whenspec = WHENSPEC_RAW;
346 static struct strbuf command_buf = STRBUF_INIT;
347 static int unread_command_buf;
348 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
349 static struct recent_command *cmd_tail = &cmd_hist;
350 static struct recent_command *rc_free;
351 static unsigned int cmd_save = 100;
352 static uintmax_t next_mark;
353 static struct strbuf new_data = STRBUF_INIT;
354
355 static void write_branch_report(FILE *rpt, struct branch *b)
356 {
357         fprintf(rpt, "%s:\n", b->name);
358
359         fprintf(rpt, "  status      :");
360         if (b->active)
361                 fputs(" active", rpt);
362         if (b->branch_tree.tree)
363                 fputs(" loaded", rpt);
364         if (is_null_sha1(b->branch_tree.versions[1].sha1))
365                 fputs(" dirty", rpt);
366         fputc('\n', rpt);
367
368         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
369         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
370         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
371         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
372
373         fputs("  last pack   : ", rpt);
374         if (b->pack_id < MAX_PACK_ID)
375                 fprintf(rpt, "%u", b->pack_id);
376         fputc('\n', rpt);
377
378         fputc('\n', rpt);
379 }
380
381 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
382
383 static void write_crash_report(const char *err)
384 {
385         char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
386         FILE *rpt = fopen(loc, "w");
387         struct branch *b;
388         unsigned long lu;
389         struct recent_command *rc;
390
391         if (!rpt) {
392                 error("can't write crash report %s: %s", loc, strerror(errno));
393                 return;
394         }
395
396         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
397
398         fprintf(rpt, "fast-import crash report:\n");
399         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
400         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
401         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
402         fputc('\n', rpt);
403
404         fputs("fatal: ", rpt);
405         fputs(err, rpt);
406         fputc('\n', rpt);
407
408         fputc('\n', rpt);
409         fputs("Most Recent Commands Before Crash\n", rpt);
410         fputs("---------------------------------\n", rpt);
411         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
412                 if (rc->next == &cmd_hist)
413                         fputs("* ", rpt);
414                 else
415                         fputs("  ", rpt);
416                 fputs(rc->buf, rpt);
417                 fputc('\n', rpt);
418         }
419
420         fputc('\n', rpt);
421         fputs("Active Branch LRU\n", rpt);
422         fputs("-----------------\n", rpt);
423         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
424                 cur_active_branches,
425                 max_active_branches);
426         fputc('\n', rpt);
427         fputs("  pos  clock name\n", rpt);
428         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
429         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
430                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
431                         ++lu, b->last_commit, b->name);
432
433         fputc('\n', rpt);
434         fputs("Inactive Branches\n", rpt);
435         fputs("-----------------\n", rpt);
436         for (lu = 0; lu < branch_table_sz; lu++) {
437                 for (b = branch_table[lu]; b; b = b->table_next_branch)
438                         write_branch_report(rpt, b);
439         }
440
441         if (first_tag) {
442                 struct tag *tg;
443                 fputc('\n', rpt);
444                 fputs("Annotated Tags\n", rpt);
445                 fputs("--------------\n", rpt);
446                 for (tg = first_tag; tg; tg = tg->next_tag) {
447                         fputs(sha1_to_hex(tg->sha1), rpt);
448                         fputc(' ', rpt);
449                         fputs(tg->name, rpt);
450                         fputc('\n', rpt);
451                 }
452         }
453
454         fputc('\n', rpt);
455         fputs("Marks\n", rpt);
456         fputs("-----\n", rpt);
457         if (mark_file)
458                 fprintf(rpt, "  exported to %s\n", mark_file);
459         else
460                 dump_marks_helper(rpt, 0, marks);
461
462         fputc('\n', rpt);
463         fputs("-------------------\n", rpt);
464         fputs("END OF CRASH REPORT\n", rpt);
465         fclose(rpt);
466 }
467
468 static void end_packfile(void);
469 static void unkeep_all_packs(void);
470 static void dump_marks(void);
471
472 static NORETURN void die_nicely(const char *err, va_list params)
473 {
474         static int zombie;
475         char message[2 * PATH_MAX];
476
477         vsnprintf(message, sizeof(message), err, params);
478         fputs("fatal: ", stderr);
479         fputs(message, stderr);
480         fputc('\n', stderr);
481
482         if (!zombie) {
483                 zombie = 1;
484                 write_crash_report(message);
485                 end_packfile();
486                 unkeep_all_packs();
487                 dump_marks();
488         }
489         exit(128);
490 }
491
492 static void alloc_objects(unsigned int cnt)
493 {
494         struct object_entry_pool *b;
495
496         b = xmalloc(sizeof(struct object_entry_pool)
497                 + cnt * sizeof(struct object_entry));
498         b->next_pool = blocks;
499         b->next_free = b->entries;
500         b->end = b->entries + cnt;
501         blocks = b;
502         alloc_count += cnt;
503 }
504
505 static struct object_entry *new_object(unsigned char *sha1)
506 {
507         struct object_entry *e;
508
509         if (blocks->next_free == blocks->end)
510                 alloc_objects(object_entry_alloc);
511
512         e = blocks->next_free++;
513         hashcpy(e->sha1, sha1);
514         return e;
515 }
516
517 static struct object_entry *find_object(unsigned char *sha1)
518 {
519         unsigned int h = sha1[0] << 8 | sha1[1];
520         struct object_entry *e;
521         for (e = object_table[h]; e; e = e->next)
522                 if (!hashcmp(sha1, e->sha1))
523                         return e;
524         return NULL;
525 }
526
527 static struct object_entry *insert_object(unsigned char *sha1)
528 {
529         unsigned int h = sha1[0] << 8 | sha1[1];
530         struct object_entry *e = object_table[h];
531         struct object_entry *p = NULL;
532
533         while (e) {
534                 if (!hashcmp(sha1, e->sha1))
535                         return e;
536                 p = e;
537                 e = e->next;
538         }
539
540         e = new_object(sha1);
541         e->next = NULL;
542         e->offset = 0;
543         if (p)
544                 p->next = e;
545         else
546                 object_table[h] = e;
547         return e;
548 }
549
550 static unsigned int hc_str(const char *s, size_t len)
551 {
552         unsigned int r = 0;
553         while (len-- > 0)
554                 r = r * 31 + *s++;
555         return r;
556 }
557
558 static void *pool_alloc(size_t len)
559 {
560         struct mem_pool *p;
561         void *r;
562
563         /* round up to a 'uintmax_t' alignment */
564         if (len & (sizeof(uintmax_t) - 1))
565                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
566
567         for (p = mem_pool; p; p = p->next_pool)
568                 if ((p->end - p->next_free >= len))
569                         break;
570
571         if (!p) {
572                 if (len >= (mem_pool_alloc/2)) {
573                         total_allocd += len;
574                         return xmalloc(len);
575                 }
576                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
577                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
578                 p->next_pool = mem_pool;
579                 p->next_free = (char *) p->space;
580                 p->end = p->next_free + mem_pool_alloc;
581                 mem_pool = p;
582         }
583
584         r = p->next_free;
585         p->next_free += len;
586         return r;
587 }
588
589 static void *pool_calloc(size_t count, size_t size)
590 {
591         size_t len = count * size;
592         void *r = pool_alloc(len);
593         memset(r, 0, len);
594         return r;
595 }
596
597 static char *pool_strdup(const char *s)
598 {
599         char *r = pool_alloc(strlen(s) + 1);
600         strcpy(r, s);
601         return r;
602 }
603
604 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
605 {
606         struct mark_set *s = marks;
607         while ((idnum >> s->shift) >= 1024) {
608                 s = pool_calloc(1, sizeof(struct mark_set));
609                 s->shift = marks->shift + 10;
610                 s->data.sets[0] = marks;
611                 marks = s;
612         }
613         while (s->shift) {
614                 uintmax_t i = idnum >> s->shift;
615                 idnum -= i << s->shift;
616                 if (!s->data.sets[i]) {
617                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
618                         s->data.sets[i]->shift = s->shift - 10;
619                 }
620                 s = s->data.sets[i];
621         }
622         if (!s->data.marked[idnum])
623                 marks_set_count++;
624         s->data.marked[idnum] = oe;
625 }
626
627 static struct object_entry *find_mark(uintmax_t idnum)
628 {
629         uintmax_t orig_idnum = idnum;
630         struct mark_set *s = marks;
631         struct object_entry *oe = NULL;
632         if ((idnum >> s->shift) < 1024) {
633                 while (s && s->shift) {
634                         uintmax_t i = idnum >> s->shift;
635                         idnum -= i << s->shift;
636                         s = s->data.sets[i];
637                 }
638                 if (s)
639                         oe = s->data.marked[idnum];
640         }
641         if (!oe)
642                 die("mark :%" PRIuMAX " not declared", orig_idnum);
643         return oe;
644 }
645
646 static struct atom_str *to_atom(const char *s, unsigned short len)
647 {
648         unsigned int hc = hc_str(s, len) % atom_table_sz;
649         struct atom_str *c;
650
651         for (c = atom_table[hc]; c; c = c->next_atom)
652                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
653                         return c;
654
655         c = pool_alloc(sizeof(struct atom_str) + len + 1);
656         c->str_len = len;
657         strncpy(c->str_dat, s, len);
658         c->str_dat[len] = 0;
659         c->next_atom = atom_table[hc];
660         atom_table[hc] = c;
661         atom_cnt++;
662         return c;
663 }
664
665 static struct branch *lookup_branch(const char *name)
666 {
667         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
668         struct branch *b;
669
670         for (b = branch_table[hc]; b; b = b->table_next_branch)
671                 if (!strcmp(name, b->name))
672                         return b;
673         return NULL;
674 }
675
676 static struct branch *new_branch(const char *name)
677 {
678         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
679         struct branch *b = lookup_branch(name);
680
681         if (b)
682                 die("Invalid attempt to create duplicate branch: %s", name);
683         switch (check_ref_format(name)) {
684         case 0: break; /* its valid */
685         case CHECK_REF_FORMAT_ONELEVEL:
686                 break; /* valid, but too few '/', allow anyway */
687         default:
688                 die("Branch name doesn't conform to GIT standards: %s", name);
689         }
690
691         b = pool_calloc(1, sizeof(struct branch));
692         b->name = pool_strdup(name);
693         b->table_next_branch = branch_table[hc];
694         b->branch_tree.versions[0].mode = S_IFDIR;
695         b->branch_tree.versions[1].mode = S_IFDIR;
696         b->active = 0;
697         b->pack_id = MAX_PACK_ID;
698         branch_table[hc] = b;
699         branch_count++;
700         return b;
701 }
702
703 static unsigned int hc_entries(unsigned int cnt)
704 {
705         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
706         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
707 }
708
709 static struct tree_content *new_tree_content(unsigned int cnt)
710 {
711         struct avail_tree_content *f, *l = NULL;
712         struct tree_content *t;
713         unsigned int hc = hc_entries(cnt);
714
715         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
716                 if (f->entry_capacity >= cnt)
717                         break;
718
719         if (f) {
720                 if (l)
721                         l->next_avail = f->next_avail;
722                 else
723                         avail_tree_table[hc] = f->next_avail;
724         } else {
725                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
726                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
727                 f->entry_capacity = cnt;
728         }
729
730         t = (struct tree_content*)f;
731         t->entry_count = 0;
732         t->delta_depth = 0;
733         return t;
734 }
735
736 static void release_tree_entry(struct tree_entry *e);
737 static void release_tree_content(struct tree_content *t)
738 {
739         struct avail_tree_content *f = (struct avail_tree_content*)t;
740         unsigned int hc = hc_entries(f->entry_capacity);
741         f->next_avail = avail_tree_table[hc];
742         avail_tree_table[hc] = f;
743 }
744
745 static void release_tree_content_recursive(struct tree_content *t)
746 {
747         unsigned int i;
748         for (i = 0; i < t->entry_count; i++)
749                 release_tree_entry(t->entries[i]);
750         release_tree_content(t);
751 }
752
753 static struct tree_content *grow_tree_content(
754         struct tree_content *t,
755         int amt)
756 {
757         struct tree_content *r = new_tree_content(t->entry_count + amt);
758         r->entry_count = t->entry_count;
759         r->delta_depth = t->delta_depth;
760         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
761         release_tree_content(t);
762         return r;
763 }
764
765 static struct tree_entry *new_tree_entry(void)
766 {
767         struct tree_entry *e;
768
769         if (!avail_tree_entry) {
770                 unsigned int n = tree_entry_alloc;
771                 total_allocd += n * sizeof(struct tree_entry);
772                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
773                 while (n-- > 1) {
774                         *((void**)e) = e + 1;
775                         e++;
776                 }
777                 *((void**)e) = NULL;
778         }
779
780         e = avail_tree_entry;
781         avail_tree_entry = *((void**)e);
782         return e;
783 }
784
785 static void release_tree_entry(struct tree_entry *e)
786 {
787         if (e->tree)
788                 release_tree_content_recursive(e->tree);
789         *((void**)e) = avail_tree_entry;
790         avail_tree_entry = e;
791 }
792
793 static struct tree_content *dup_tree_content(struct tree_content *s)
794 {
795         struct tree_content *d;
796         struct tree_entry *a, *b;
797         unsigned int i;
798
799         if (!s)
800                 return NULL;
801         d = new_tree_content(s->entry_count);
802         for (i = 0; i < s->entry_count; i++) {
803                 a = s->entries[i];
804                 b = new_tree_entry();
805                 memcpy(b, a, sizeof(*a));
806                 if (a->tree && is_null_sha1(b->versions[1].sha1))
807                         b->tree = dup_tree_content(a->tree);
808                 else
809                         b->tree = NULL;
810                 d->entries[i] = b;
811         }
812         d->entry_count = s->entry_count;
813         d->delta_depth = s->delta_depth;
814
815         return d;
816 }
817
818 static void start_packfile(void)
819 {
820         static char tmpfile[PATH_MAX];
821         struct packed_git *p;
822         struct pack_header hdr;
823         int pack_fd;
824
825         pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
826                               "pack/tmp_pack_XXXXXX");
827         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
828         strcpy(p->pack_name, tmpfile);
829         p->pack_fd = pack_fd;
830
831         hdr.hdr_signature = htonl(PACK_SIGNATURE);
832         hdr.hdr_version = htonl(2);
833         hdr.hdr_entries = 0;
834         write_or_die(p->pack_fd, &hdr, sizeof(hdr));
835
836         pack_data = p;
837         pack_size = sizeof(hdr);
838         object_count = 0;
839
840         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
841         all_packs[pack_id] = p;
842 }
843
844 static int oecmp (const void *a_, const void *b_)
845 {
846         struct object_entry *a = *((struct object_entry**)a_);
847         struct object_entry *b = *((struct object_entry**)b_);
848         return hashcmp(a->sha1, b->sha1);
849 }
850
851 static char *create_index(void)
852 {
853         static char tmpfile[PATH_MAX];
854         git_SHA_CTX ctx;
855         struct sha1file *f;
856         struct object_entry **idx, **c, **last, *e;
857         struct object_entry_pool *o;
858         uint32_t array[256];
859         int i, idx_fd;
860
861         /* Build the sorted table of object IDs. */
862         idx = xmalloc(object_count * sizeof(struct object_entry*));
863         c = idx;
864         for (o = blocks; o; o = o->next_pool)
865                 for (e = o->next_free; e-- != o->entries;)
866                         if (pack_id == e->pack_id)
867                                 *c++ = e;
868         last = idx + object_count;
869         if (c != last)
870                 die("internal consistency error creating the index");
871         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
872
873         /* Generate the fan-out array. */
874         c = idx;
875         for (i = 0; i < 256; i++) {
876                 struct object_entry **next = c;
877                 while (next < last) {
878                         if ((*next)->sha1[0] != i)
879                                 break;
880                         next++;
881                 }
882                 array[i] = htonl(next - idx);
883                 c = next;
884         }
885
886         idx_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
887                              "pack/tmp_idx_XXXXXX");
888         f = sha1fd(idx_fd, tmpfile);
889         sha1write(f, array, 256 * sizeof(int));
890         git_SHA1_Init(&ctx);
891         for (c = idx; c != last; c++) {
892                 uint32_t offset = htonl((*c)->offset);
893                 sha1write(f, &offset, 4);
894                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
895                 git_SHA1_Update(&ctx, (*c)->sha1, 20);
896         }
897         sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
898         sha1close(f, NULL, CSUM_FSYNC);
899         free(idx);
900         git_SHA1_Final(pack_data->sha1, &ctx);
901         return tmpfile;
902 }
903
904 static char *keep_pack(char *curr_index_name)
905 {
906         static char name[PATH_MAX];
907         static const char *keep_msg = "fast-import";
908         int keep_fd;
909
910         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
911         if (keep_fd < 0)
912                 die_errno("cannot create keep file");
913         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
914         if (close(keep_fd))
915                 die_errno("failed to write keep file");
916
917         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
918                  get_object_directory(), sha1_to_hex(pack_data->sha1));
919         if (move_temp_to_file(pack_data->pack_name, name))
920                 die("cannot store pack file");
921
922         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
923                  get_object_directory(), sha1_to_hex(pack_data->sha1));
924         if (move_temp_to_file(curr_index_name, name))
925                 die("cannot store index file");
926         return name;
927 }
928
929 static void unkeep_all_packs(void)
930 {
931         static char name[PATH_MAX];
932         int k;
933
934         for (k = 0; k < pack_id; k++) {
935                 struct packed_git *p = all_packs[k];
936                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
937                          get_object_directory(), sha1_to_hex(p->sha1));
938                 unlink_or_warn(name);
939         }
940 }
941
942 static void end_packfile(void)
943 {
944         struct packed_git *old_p = pack_data, *new_p;
945
946         clear_delta_base_cache();
947         if (object_count) {
948                 char *idx_name;
949                 int i;
950                 struct branch *b;
951                 struct tag *t;
952
953                 close_pack_windows(pack_data);
954                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
955                                     pack_data->pack_name, object_count,
956                                     NULL, 0);
957                 close(pack_data->pack_fd);
958                 idx_name = keep_pack(create_index());
959
960                 /* Register the packfile with core git's machinery. */
961                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
962                 if (!new_p)
963                         die("core git rejected index %s", idx_name);
964                 all_packs[pack_id] = new_p;
965                 install_packed_git(new_p);
966
967                 /* Print the boundary */
968                 if (pack_edges) {
969                         fprintf(pack_edges, "%s:", new_p->pack_name);
970                         for (i = 0; i < branch_table_sz; i++) {
971                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
972                                         if (b->pack_id == pack_id)
973                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
974                                 }
975                         }
976                         for (t = first_tag; t; t = t->next_tag) {
977                                 if (t->pack_id == pack_id)
978                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
979                         }
980                         fputc('\n', pack_edges);
981                         fflush(pack_edges);
982                 }
983
984                 pack_id++;
985         }
986         else {
987                 close(old_p->pack_fd);
988                 unlink_or_warn(old_p->pack_name);
989         }
990         free(old_p);
991
992         /* We can't carry a delta across packfiles. */
993         strbuf_release(&last_blob.data);
994         last_blob.offset = 0;
995         last_blob.depth = 0;
996 }
997
998 static void cycle_packfile(void)
999 {
1000         end_packfile();
1001         start_packfile();
1002 }
1003
1004 static size_t encode_header(
1005         enum object_type type,
1006         size_t size,
1007         unsigned char *hdr)
1008 {
1009         int n = 1;
1010         unsigned char c;
1011
1012         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
1013                 die("bad type %d", type);
1014
1015         c = (type << 4) | (size & 15);
1016         size >>= 4;
1017         while (size) {
1018                 *hdr++ = c | 0x80;
1019                 c = size & 0x7f;
1020                 size >>= 7;
1021                 n++;
1022         }
1023         *hdr = c;
1024         return n;
1025 }
1026
1027 static int store_object(
1028         enum object_type type,
1029         struct strbuf *dat,
1030         struct last_object *last,
1031         unsigned char *sha1out,
1032         uintmax_t mark)
1033 {
1034         void *out, *delta;
1035         struct object_entry *e;
1036         unsigned char hdr[96];
1037         unsigned char sha1[20];
1038         unsigned long hdrlen, deltalen;
1039         git_SHA_CTX c;
1040         z_stream s;
1041
1042         hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1043                 (unsigned long)dat->len) + 1;
1044         git_SHA1_Init(&c);
1045         git_SHA1_Update(&c, hdr, hdrlen);
1046         git_SHA1_Update(&c, dat->buf, dat->len);
1047         git_SHA1_Final(sha1, &c);
1048         if (sha1out)
1049                 hashcpy(sha1out, sha1);
1050
1051         e = insert_object(sha1);
1052         if (mark)
1053                 insert_mark(mark, e);
1054         if (e->offset) {
1055                 duplicate_count_by_type[type]++;
1056                 return 1;
1057         } else if (find_sha1_pack(sha1, packed_git)) {
1058                 e->type = type;
1059                 e->pack_id = MAX_PACK_ID;
1060                 e->offset = 1; /* just not zero! */
1061                 duplicate_count_by_type[type]++;
1062                 return 1;
1063         }
1064
1065         if (last && last->data.buf && last->depth < max_depth) {
1066                 delta = diff_delta(last->data.buf, last->data.len,
1067                         dat->buf, dat->len,
1068                         &deltalen, 0);
1069                 if (delta && deltalen >= dat->len) {
1070                         free(delta);
1071                         delta = NULL;
1072                 }
1073         } else
1074                 delta = NULL;
1075
1076         memset(&s, 0, sizeof(s));
1077         deflateInit(&s, pack_compression_level);
1078         if (delta) {
1079                 s.next_in = delta;
1080                 s.avail_in = deltalen;
1081         } else {
1082                 s.next_in = (void *)dat->buf;
1083                 s.avail_in = dat->len;
1084         }
1085         s.avail_out = deflateBound(&s, s.avail_in);
1086         s.next_out = out = xmalloc(s.avail_out);
1087         while (deflate(&s, Z_FINISH) == Z_OK)
1088                 /* nothing */;
1089         deflateEnd(&s);
1090
1091         /* Determine if we should auto-checkpoint. */
1092         if ((pack_size + 60 + s.total_out) > max_packsize
1093                 || (pack_size + 60 + s.total_out) < pack_size) {
1094
1095                 /* This new object needs to *not* have the current pack_id. */
1096                 e->pack_id = pack_id + 1;
1097                 cycle_packfile();
1098
1099                 /* We cannot carry a delta into the new pack. */
1100                 if (delta) {
1101                         free(delta);
1102                         delta = NULL;
1103
1104                         memset(&s, 0, sizeof(s));
1105                         deflateInit(&s, pack_compression_level);
1106                         s.next_in = (void *)dat->buf;
1107                         s.avail_in = dat->len;
1108                         s.avail_out = deflateBound(&s, s.avail_in);
1109                         s.next_out = out = xrealloc(out, s.avail_out);
1110                         while (deflate(&s, Z_FINISH) == Z_OK)
1111                                 /* nothing */;
1112                         deflateEnd(&s);
1113                 }
1114         }
1115
1116         e->type = type;
1117         e->pack_id = pack_id;
1118         e->offset = pack_size;
1119         object_count++;
1120         object_count_by_type[type]++;
1121
1122         if (delta) {
1123                 unsigned long ofs = e->offset - last->offset;
1124                 unsigned pos = sizeof(hdr) - 1;
1125
1126                 delta_count_by_type[type]++;
1127                 e->depth = last->depth + 1;
1128
1129                 hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1130                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1131                 pack_size += hdrlen;
1132
1133                 hdr[pos] = ofs & 127;
1134                 while (ofs >>= 7)
1135                         hdr[--pos] = 128 | (--ofs & 127);
1136                 write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1137                 pack_size += sizeof(hdr) - pos;
1138         } else {
1139                 e->depth = 0;
1140                 hdrlen = encode_header(type, dat->len, hdr);
1141                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1142                 pack_size += hdrlen;
1143         }
1144
1145         write_or_die(pack_data->pack_fd, out, s.total_out);
1146         pack_size += s.total_out;
1147
1148         free(out);
1149         free(delta);
1150         if (last) {
1151                 if (last->no_swap) {
1152                         last->data = *dat;
1153                 } else {
1154                         strbuf_swap(&last->data, dat);
1155                 }
1156                 last->offset = e->offset;
1157                 last->depth = e->depth;
1158         }
1159         return 0;
1160 }
1161
1162 /* All calls must be guarded by find_object() or find_mark() to
1163  * ensure the 'struct object_entry' passed was written by this
1164  * process instance.  We unpack the entry by the offset, avoiding
1165  * the need for the corresponding .idx file.  This unpacking rule
1166  * works because we only use OBJ_REF_DELTA within the packfiles
1167  * created by fast-import.
1168  *
1169  * oe must not be NULL.  Such an oe usually comes from giving
1170  * an unknown SHA-1 to find_object() or an undefined mark to
1171  * find_mark().  Callers must test for this condition and use
1172  * the standard read_sha1_file() when it happens.
1173  *
1174  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1175  * find_mark(), where the mark was reloaded from an existing marks
1176  * file and is referencing an object that this fast-import process
1177  * instance did not write out to a packfile.  Callers must test for
1178  * this condition and use read_sha1_file() instead.
1179  */
1180 static void *gfi_unpack_entry(
1181         struct object_entry *oe,
1182         unsigned long *sizep)
1183 {
1184         enum object_type type;
1185         struct packed_git *p = all_packs[oe->pack_id];
1186         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1187                 /* The object is stored in the packfile we are writing to
1188                  * and we have modified it since the last time we scanned
1189                  * back to read a previously written object.  If an old
1190                  * window covered [p->pack_size, p->pack_size + 20) its
1191                  * data is stale and is not valid.  Closing all windows
1192                  * and updating the packfile length ensures we can read
1193                  * the newly written data.
1194                  */
1195                 close_pack_windows(p);
1196
1197                 /* We have to offer 20 bytes additional on the end of
1198                  * the packfile as the core unpacker code assumes the
1199                  * footer is present at the file end and must promise
1200                  * at least 20 bytes within any window it maps.  But
1201                  * we don't actually create the footer here.
1202                  */
1203                 p->pack_size = pack_size + 20;
1204         }
1205         return unpack_entry(p, oe->offset, &type, sizep);
1206 }
1207
1208 static const char *get_mode(const char *str, uint16_t *modep)
1209 {
1210         unsigned char c;
1211         uint16_t mode = 0;
1212
1213         while ((c = *str++) != ' ') {
1214                 if (c < '0' || c > '7')
1215                         return NULL;
1216                 mode = (mode << 3) + (c - '0');
1217         }
1218         *modep = mode;
1219         return str;
1220 }
1221
1222 static void load_tree(struct tree_entry *root)
1223 {
1224         unsigned char *sha1 = root->versions[1].sha1;
1225         struct object_entry *myoe;
1226         struct tree_content *t;
1227         unsigned long size;
1228         char *buf;
1229         const char *c;
1230
1231         root->tree = t = new_tree_content(8);
1232         if (is_null_sha1(sha1))
1233                 return;
1234
1235         myoe = find_object(sha1);
1236         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1237                 if (myoe->type != OBJ_TREE)
1238                         die("Not a tree: %s", sha1_to_hex(sha1));
1239                 t->delta_depth = myoe->depth;
1240                 buf = gfi_unpack_entry(myoe, &size);
1241                 if (!buf)
1242                         die("Can't load tree %s", sha1_to_hex(sha1));
1243         } else {
1244                 enum object_type type;
1245                 buf = read_sha1_file(sha1, &type, &size);
1246                 if (!buf || type != OBJ_TREE)
1247                         die("Can't load tree %s", sha1_to_hex(sha1));
1248         }
1249
1250         c = buf;
1251         while (c != (buf + size)) {
1252                 struct tree_entry *e = new_tree_entry();
1253
1254                 if (t->entry_count == t->entry_capacity)
1255                         root->tree = t = grow_tree_content(t, t->entry_count);
1256                 t->entries[t->entry_count++] = e;
1257
1258                 e->tree = NULL;
1259                 c = get_mode(c, &e->versions[1].mode);
1260                 if (!c)
1261                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1262                 e->versions[0].mode = e->versions[1].mode;
1263                 e->name = to_atom(c, strlen(c));
1264                 c += e->name->str_len + 1;
1265                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1266                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1267                 c += 20;
1268         }
1269         free(buf);
1270 }
1271
1272 static int tecmp0 (const void *_a, const void *_b)
1273 {
1274         struct tree_entry *a = *((struct tree_entry**)_a);
1275         struct tree_entry *b = *((struct tree_entry**)_b);
1276         return base_name_compare(
1277                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1278                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1279 }
1280
1281 static int tecmp1 (const void *_a, const void *_b)
1282 {
1283         struct tree_entry *a = *((struct tree_entry**)_a);
1284         struct tree_entry *b = *((struct tree_entry**)_b);
1285         return base_name_compare(
1286                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1287                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1288 }
1289
1290 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1291 {
1292         size_t maxlen = 0;
1293         unsigned int i;
1294
1295         if (!v)
1296                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1297         else
1298                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1299
1300         for (i = 0; i < t->entry_count; i++) {
1301                 if (t->entries[i]->versions[v].mode)
1302                         maxlen += t->entries[i]->name->str_len + 34;
1303         }
1304
1305         strbuf_reset(b);
1306         strbuf_grow(b, maxlen);
1307         for (i = 0; i < t->entry_count; i++) {
1308                 struct tree_entry *e = t->entries[i];
1309                 if (!e->versions[v].mode)
1310                         continue;
1311                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1312                                         e->name->str_dat, '\0');
1313                 strbuf_add(b, e->versions[v].sha1, 20);
1314         }
1315 }
1316
1317 static void store_tree(struct tree_entry *root)
1318 {
1319         struct tree_content *t = root->tree;
1320         unsigned int i, j, del;
1321         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1322         struct object_entry *le;
1323
1324         if (!is_null_sha1(root->versions[1].sha1))
1325                 return;
1326
1327         for (i = 0; i < t->entry_count; i++) {
1328                 if (t->entries[i]->tree)
1329                         store_tree(t->entries[i]);
1330         }
1331
1332         le = find_object(root->versions[0].sha1);
1333         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1334                 mktree(t, 0, &old_tree);
1335                 lo.data = old_tree;
1336                 lo.offset = le->offset;
1337                 lo.depth = t->delta_depth;
1338         }
1339
1340         mktree(t, 1, &new_tree);
1341         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1342
1343         t->delta_depth = lo.depth;
1344         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1345                 struct tree_entry *e = t->entries[i];
1346                 if (e->versions[1].mode) {
1347                         e->versions[0].mode = e->versions[1].mode;
1348                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1349                         t->entries[j++] = e;
1350                 } else {
1351                         release_tree_entry(e);
1352                         del++;
1353                 }
1354         }
1355         t->entry_count -= del;
1356 }
1357
1358 static int tree_content_set(
1359         struct tree_entry *root,
1360         const char *p,
1361         const unsigned char *sha1,
1362         const uint16_t mode,
1363         struct tree_content *subtree)
1364 {
1365         struct tree_content *t = root->tree;
1366         const char *slash1;
1367         unsigned int i, n;
1368         struct tree_entry *e;
1369
1370         slash1 = strchr(p, '/');
1371         if (slash1)
1372                 n = slash1 - p;
1373         else
1374                 n = strlen(p);
1375         if (!n)
1376                 die("Empty path component found in input");
1377         if (!slash1 && !S_ISDIR(mode) && subtree)
1378                 die("Non-directories cannot have subtrees");
1379
1380         for (i = 0; i < t->entry_count; i++) {
1381                 e = t->entries[i];
1382                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1383                         if (!slash1) {
1384                                 if (!S_ISDIR(mode)
1385                                                 && e->versions[1].mode == mode
1386                                                 && !hashcmp(e->versions[1].sha1, sha1))
1387                                         return 0;
1388                                 e->versions[1].mode = mode;
1389                                 hashcpy(e->versions[1].sha1, sha1);
1390                                 if (e->tree)
1391                                         release_tree_content_recursive(e->tree);
1392                                 e->tree = subtree;
1393                                 hashclr(root->versions[1].sha1);
1394                                 return 1;
1395                         }
1396                         if (!S_ISDIR(e->versions[1].mode)) {
1397                                 e->tree = new_tree_content(8);
1398                                 e->versions[1].mode = S_IFDIR;
1399                         }
1400                         if (!e->tree)
1401                                 load_tree(e);
1402                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1403                                 hashclr(root->versions[1].sha1);
1404                                 return 1;
1405                         }
1406                         return 0;
1407                 }
1408         }
1409
1410         if (t->entry_count == t->entry_capacity)
1411                 root->tree = t = grow_tree_content(t, t->entry_count);
1412         e = new_tree_entry();
1413         e->name = to_atom(p, n);
1414         e->versions[0].mode = 0;
1415         hashclr(e->versions[0].sha1);
1416         t->entries[t->entry_count++] = e;
1417         if (slash1) {
1418                 e->tree = new_tree_content(8);
1419                 e->versions[1].mode = S_IFDIR;
1420                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1421         } else {
1422                 e->tree = subtree;
1423                 e->versions[1].mode = mode;
1424                 hashcpy(e->versions[1].sha1, sha1);
1425         }
1426         hashclr(root->versions[1].sha1);
1427         return 1;
1428 }
1429
1430 static int tree_content_remove(
1431         struct tree_entry *root,
1432         const char *p,
1433         struct tree_entry *backup_leaf)
1434 {
1435         struct tree_content *t = root->tree;
1436         const char *slash1;
1437         unsigned int i, n;
1438         struct tree_entry *e;
1439
1440         slash1 = strchr(p, '/');
1441         if (slash1)
1442                 n = slash1 - p;
1443         else
1444                 n = strlen(p);
1445
1446         for (i = 0; i < t->entry_count; i++) {
1447                 e = t->entries[i];
1448                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1449                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1450                                 goto del_entry;
1451                         if (!e->tree)
1452                                 load_tree(e);
1453                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1454                                 for (n = 0; n < e->tree->entry_count; n++) {
1455                                         if (e->tree->entries[n]->versions[1].mode) {
1456                                                 hashclr(root->versions[1].sha1);
1457                                                 return 1;
1458                                         }
1459                                 }
1460                                 backup_leaf = NULL;
1461                                 goto del_entry;
1462                         }
1463                         return 0;
1464                 }
1465         }
1466         return 0;
1467
1468 del_entry:
1469         if (backup_leaf)
1470                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1471         else if (e->tree)
1472                 release_tree_content_recursive(e->tree);
1473         e->tree = NULL;
1474         e->versions[1].mode = 0;
1475         hashclr(e->versions[1].sha1);
1476         hashclr(root->versions[1].sha1);
1477         return 1;
1478 }
1479
1480 static int tree_content_get(
1481         struct tree_entry *root,
1482         const char *p,
1483         struct tree_entry *leaf)
1484 {
1485         struct tree_content *t = root->tree;
1486         const char *slash1;
1487         unsigned int i, n;
1488         struct tree_entry *e;
1489
1490         slash1 = strchr(p, '/');
1491         if (slash1)
1492                 n = slash1 - p;
1493         else
1494                 n = strlen(p);
1495
1496         for (i = 0; i < t->entry_count; i++) {
1497                 e = t->entries[i];
1498                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1499                         if (!slash1) {
1500                                 memcpy(leaf, e, sizeof(*leaf));
1501                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1502                                         leaf->tree = dup_tree_content(e->tree);
1503                                 else
1504                                         leaf->tree = NULL;
1505                                 return 1;
1506                         }
1507                         if (!S_ISDIR(e->versions[1].mode))
1508                                 return 0;
1509                         if (!e->tree)
1510                                 load_tree(e);
1511                         return tree_content_get(e, slash1 + 1, leaf);
1512                 }
1513         }
1514         return 0;
1515 }
1516
1517 static int update_branch(struct branch *b)
1518 {
1519         static const char *msg = "fast-import";
1520         struct ref_lock *lock;
1521         unsigned char old_sha1[20];
1522
1523         if (is_null_sha1(b->sha1))
1524                 return 0;
1525         if (read_ref(b->name, old_sha1))
1526                 hashclr(old_sha1);
1527         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1528         if (!lock)
1529                 return error("Unable to lock %s", b->name);
1530         if (!force_update && !is_null_sha1(old_sha1)) {
1531                 struct commit *old_cmit, *new_cmit;
1532
1533                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1534                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1535                 if (!old_cmit || !new_cmit) {
1536                         unlock_ref(lock);
1537                         return error("Branch %s is missing commits.", b->name);
1538                 }
1539
1540                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1541                         unlock_ref(lock);
1542                         warning("Not updating %s"
1543                                 " (new tip %s does not contain %s)",
1544                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1545                         return -1;
1546                 }
1547         }
1548         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1549                 return error("Unable to update %s", b->name);
1550         return 0;
1551 }
1552
1553 static void dump_branches(void)
1554 {
1555         unsigned int i;
1556         struct branch *b;
1557
1558         for (i = 0; i < branch_table_sz; i++) {
1559                 for (b = branch_table[i]; b; b = b->table_next_branch)
1560                         failure |= update_branch(b);
1561         }
1562 }
1563
1564 static void dump_tags(void)
1565 {
1566         static const char *msg = "fast-import";
1567         struct tag *t;
1568         struct ref_lock *lock;
1569         char ref_name[PATH_MAX];
1570
1571         for (t = first_tag; t; t = t->next_tag) {
1572                 sprintf(ref_name, "tags/%s", t->name);
1573                 lock = lock_ref_sha1(ref_name, NULL);
1574                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1575                         failure |= error("Unable to update %s", ref_name);
1576         }
1577 }
1578
1579 static void dump_marks_helper(FILE *f,
1580         uintmax_t base,
1581         struct mark_set *m)
1582 {
1583         uintmax_t k;
1584         if (m->shift) {
1585                 for (k = 0; k < 1024; k++) {
1586                         if (m->data.sets[k])
1587                                 dump_marks_helper(f, (base + k) << m->shift,
1588                                         m->data.sets[k]);
1589                 }
1590         } else {
1591                 for (k = 0; k < 1024; k++) {
1592                         if (m->data.marked[k])
1593                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1594                                         sha1_to_hex(m->data.marked[k]->sha1));
1595                 }
1596         }
1597 }
1598
1599 static void dump_marks(void)
1600 {
1601         static struct lock_file mark_lock;
1602         int mark_fd;
1603         FILE *f;
1604
1605         if (!mark_file)
1606                 return;
1607
1608         mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1609         if (mark_fd < 0) {
1610                 failure |= error("Unable to write marks file %s: %s",
1611                         mark_file, strerror(errno));
1612                 return;
1613         }
1614
1615         f = fdopen(mark_fd, "w");
1616         if (!f) {
1617                 int saved_errno = errno;
1618                 rollback_lock_file(&mark_lock);
1619                 failure |= error("Unable to write marks file %s: %s",
1620                         mark_file, strerror(saved_errno));
1621                 return;
1622         }
1623
1624         /*
1625          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1626          * Assign -1 to the lock file descriptor so that commit_lock_file()
1627          * won't try to close() it.
1628          */
1629         mark_lock.fd = -1;
1630
1631         dump_marks_helper(f, 0, marks);
1632         if (ferror(f) || fclose(f)) {
1633                 int saved_errno = errno;
1634                 rollback_lock_file(&mark_lock);
1635                 failure |= error("Unable to write marks file %s: %s",
1636                         mark_file, strerror(saved_errno));
1637                 return;
1638         }
1639
1640         if (commit_lock_file(&mark_lock)) {
1641                 int saved_errno = errno;
1642                 rollback_lock_file(&mark_lock);
1643                 failure |= error("Unable to commit marks file %s: %s",
1644                         mark_file, strerror(saved_errno));
1645                 return;
1646         }
1647 }
1648
1649 static int read_next_command(void)
1650 {
1651         static int stdin_eof = 0;
1652
1653         if (stdin_eof) {
1654                 unread_command_buf = 0;
1655                 return EOF;
1656         }
1657
1658         do {
1659                 if (unread_command_buf) {
1660                         unread_command_buf = 0;
1661                 } else {
1662                         struct recent_command *rc;
1663
1664                         strbuf_detach(&command_buf, NULL);
1665                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1666                         if (stdin_eof)
1667                                 return EOF;
1668
1669                         rc = rc_free;
1670                         if (rc)
1671                                 rc_free = rc->next;
1672                         else {
1673                                 rc = cmd_hist.next;
1674                                 cmd_hist.next = rc->next;
1675                                 cmd_hist.next->prev = &cmd_hist;
1676                                 free(rc->buf);
1677                         }
1678
1679                         rc->buf = command_buf.buf;
1680                         rc->prev = cmd_tail;
1681                         rc->next = cmd_hist.prev;
1682                         rc->prev->next = rc;
1683                         cmd_tail = rc;
1684                 }
1685         } while (command_buf.buf[0] == '#');
1686
1687         return 0;
1688 }
1689
1690 static void skip_optional_lf(void)
1691 {
1692         int term_char = fgetc(stdin);
1693         if (term_char != '\n' && term_char != EOF)
1694                 ungetc(term_char, stdin);
1695 }
1696
1697 static void parse_mark(void)
1698 {
1699         if (!prefixcmp(command_buf.buf, "mark :")) {
1700                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1701                 read_next_command();
1702         }
1703         else
1704                 next_mark = 0;
1705 }
1706
1707 static void parse_data(struct strbuf *sb)
1708 {
1709         strbuf_reset(sb);
1710
1711         if (prefixcmp(command_buf.buf, "data "))
1712                 die("Expected 'data n' command, found: %s", command_buf.buf);
1713
1714         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1715                 char *term = xstrdup(command_buf.buf + 5 + 2);
1716                 size_t term_len = command_buf.len - 5 - 2;
1717
1718                 strbuf_detach(&command_buf, NULL);
1719                 for (;;) {
1720                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1721                                 die("EOF in data (terminator '%s' not found)", term);
1722                         if (term_len == command_buf.len
1723                                 && !strcmp(term, command_buf.buf))
1724                                 break;
1725                         strbuf_addbuf(sb, &command_buf);
1726                         strbuf_addch(sb, '\n');
1727                 }
1728                 free(term);
1729         }
1730         else {
1731                 size_t n = 0, length;
1732
1733                 length = strtoul(command_buf.buf + 5, NULL, 10);
1734
1735                 while (n < length) {
1736                         size_t s = strbuf_fread(sb, length - n, stdin);
1737                         if (!s && feof(stdin))
1738                                 die("EOF in data (%lu bytes remaining)",
1739                                         (unsigned long)(length - n));
1740                         n += s;
1741                 }
1742         }
1743
1744         skip_optional_lf();
1745 }
1746
1747 static int validate_raw_date(const char *src, char *result, int maxlen)
1748 {
1749         const char *orig_src = src;
1750         char *endp;
1751         unsigned long num;
1752
1753         errno = 0;
1754
1755         num = strtoul(src, &endp, 10);
1756         /* NEEDSWORK: perhaps check for reasonable values? */
1757         if (errno || endp == src || *endp != ' ')
1758                 return -1;
1759
1760         src = endp + 1;
1761         if (*src != '-' && *src != '+')
1762                 return -1;
1763
1764         num = strtoul(src + 1, &endp, 10);
1765         if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1766             1400 < num)
1767                 return -1;
1768
1769         strcpy(result, orig_src);
1770         return 0;
1771 }
1772
1773 static char *parse_ident(const char *buf)
1774 {
1775         const char *gt;
1776         size_t name_len;
1777         char *ident;
1778
1779         gt = strrchr(buf, '>');
1780         if (!gt)
1781                 die("Missing > in ident string: %s", buf);
1782         gt++;
1783         if (*gt != ' ')
1784                 die("Missing space after > in ident string: %s", buf);
1785         gt++;
1786         name_len = gt - buf;
1787         ident = xmalloc(name_len + 24);
1788         strncpy(ident, buf, name_len);
1789
1790         switch (whenspec) {
1791         case WHENSPEC_RAW:
1792                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1793                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1794                 break;
1795         case WHENSPEC_RFC2822:
1796                 if (parse_date(gt, ident + name_len, 24) < 0)
1797                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1798                 break;
1799         case WHENSPEC_NOW:
1800                 if (strcmp("now", gt))
1801                         die("Date in ident must be 'now': %s", buf);
1802                 datestamp(ident + name_len, 24);
1803                 break;
1804         }
1805
1806         return ident;
1807 }
1808
1809 static void parse_new_blob(void)
1810 {
1811         static struct strbuf buf = STRBUF_INIT;
1812
1813         read_next_command();
1814         parse_mark();
1815         parse_data(&buf);
1816         store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1817 }
1818
1819 static void unload_one_branch(void)
1820 {
1821         while (cur_active_branches
1822                 && cur_active_branches >= max_active_branches) {
1823                 uintmax_t min_commit = ULONG_MAX;
1824                 struct branch *e, *l = NULL, *p = NULL;
1825
1826                 for (e = active_branches; e; e = e->active_next_branch) {
1827                         if (e->last_commit < min_commit) {
1828                                 p = l;
1829                                 min_commit = e->last_commit;
1830                         }
1831                         l = e;
1832                 }
1833
1834                 if (p) {
1835                         e = p->active_next_branch;
1836                         p->active_next_branch = e->active_next_branch;
1837                 } else {
1838                         e = active_branches;
1839                         active_branches = e->active_next_branch;
1840                 }
1841                 e->active = 0;
1842                 e->active_next_branch = NULL;
1843                 if (e->branch_tree.tree) {
1844                         release_tree_content_recursive(e->branch_tree.tree);
1845                         e->branch_tree.tree = NULL;
1846                 }
1847                 cur_active_branches--;
1848         }
1849 }
1850
1851 static void load_branch(struct branch *b)
1852 {
1853         load_tree(&b->branch_tree);
1854         if (!b->active) {
1855                 b->active = 1;
1856                 b->active_next_branch = active_branches;
1857                 active_branches = b;
1858                 cur_active_branches++;
1859                 branch_load_count++;
1860         }
1861 }
1862
1863 static void file_change_m(struct branch *b)
1864 {
1865         const char *p = command_buf.buf + 2;
1866         static struct strbuf uq = STRBUF_INIT;
1867         const char *endp;
1868         struct object_entry *oe = oe;
1869         unsigned char sha1[20];
1870         uint16_t mode, inline_data = 0;
1871
1872         p = get_mode(p, &mode);
1873         if (!p)
1874                 die("Corrupt mode: %s", command_buf.buf);
1875         switch (mode) {
1876         case 0644:
1877         case 0755:
1878                 mode |= S_IFREG;
1879         case S_IFREG | 0644:
1880         case S_IFREG | 0755:
1881         case S_IFLNK:
1882         case S_IFGITLINK:
1883                 /* ok */
1884                 break;
1885         default:
1886                 die("Corrupt mode: %s", command_buf.buf);
1887         }
1888
1889         if (*p == ':') {
1890                 char *x;
1891                 oe = find_mark(strtoumax(p + 1, &x, 10));
1892                 hashcpy(sha1, oe->sha1);
1893                 p = x;
1894         } else if (!prefixcmp(p, "inline")) {
1895                 inline_data = 1;
1896                 p += 6;
1897         } else {
1898                 if (get_sha1_hex(p, sha1))
1899                         die("Invalid SHA1: %s", command_buf.buf);
1900                 oe = find_object(sha1);
1901                 p += 40;
1902         }
1903         if (*p++ != ' ')
1904                 die("Missing space after SHA1: %s", command_buf.buf);
1905
1906         strbuf_reset(&uq);
1907         if (!unquote_c_style(&uq, p, &endp)) {
1908                 if (*endp)
1909                         die("Garbage after path in: %s", command_buf.buf);
1910                 p = uq.buf;
1911         }
1912
1913         if (S_ISGITLINK(mode)) {
1914                 if (inline_data)
1915                         die("Git links cannot be specified 'inline': %s",
1916                                 command_buf.buf);
1917                 else if (oe) {
1918                         if (oe->type != OBJ_COMMIT)
1919                                 die("Not a commit (actually a %s): %s",
1920                                         typename(oe->type), command_buf.buf);
1921                 }
1922                 /*
1923                  * Accept the sha1 without checking; it expected to be in
1924                  * another repository.
1925                  */
1926         } else if (inline_data) {
1927                 static struct strbuf buf = STRBUF_INIT;
1928
1929                 if (p != uq.buf) {
1930                         strbuf_addstr(&uq, p);
1931                         p = uq.buf;
1932                 }
1933                 read_next_command();
1934                 parse_data(&buf);
1935                 store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1936         } else if (oe) {
1937                 if (oe->type != OBJ_BLOB)
1938                         die("Not a blob (actually a %s): %s",
1939                                 typename(oe->type), command_buf.buf);
1940         } else {
1941                 enum object_type type = sha1_object_info(sha1, NULL);
1942                 if (type < 0)
1943                         die("Blob not found: %s", command_buf.buf);
1944                 if (type != OBJ_BLOB)
1945                         die("Not a blob (actually a %s): %s",
1946                             typename(type), command_buf.buf);
1947         }
1948
1949         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
1950 }
1951
1952 static void file_change_d(struct branch *b)
1953 {
1954         const char *p = command_buf.buf + 2;
1955         static struct strbuf uq = STRBUF_INIT;
1956         const char *endp;
1957
1958         strbuf_reset(&uq);
1959         if (!unquote_c_style(&uq, p, &endp)) {
1960                 if (*endp)
1961                         die("Garbage after path in: %s", command_buf.buf);
1962                 p = uq.buf;
1963         }
1964         tree_content_remove(&b->branch_tree, p, NULL);
1965 }
1966
1967 static void file_change_cr(struct branch *b, int rename)
1968 {
1969         const char *s, *d;
1970         static struct strbuf s_uq = STRBUF_INIT;
1971         static struct strbuf d_uq = STRBUF_INIT;
1972         const char *endp;
1973         struct tree_entry leaf;
1974
1975         s = command_buf.buf + 2;
1976         strbuf_reset(&s_uq);
1977         if (!unquote_c_style(&s_uq, s, &endp)) {
1978                 if (*endp != ' ')
1979                         die("Missing space after source: %s", command_buf.buf);
1980         } else {
1981                 endp = strchr(s, ' ');
1982                 if (!endp)
1983                         die("Missing space after source: %s", command_buf.buf);
1984                 strbuf_add(&s_uq, s, endp - s);
1985         }
1986         s = s_uq.buf;
1987
1988         endp++;
1989         if (!*endp)
1990                 die("Missing dest: %s", command_buf.buf);
1991
1992         d = endp;
1993         strbuf_reset(&d_uq);
1994         if (!unquote_c_style(&d_uq, d, &endp)) {
1995                 if (*endp)
1996                         die("Garbage after dest in: %s", command_buf.buf);
1997                 d = d_uq.buf;
1998         }
1999
2000         memset(&leaf, 0, sizeof(leaf));
2001         if (rename)
2002                 tree_content_remove(&b->branch_tree, s, &leaf);
2003         else
2004                 tree_content_get(&b->branch_tree, s, &leaf);
2005         if (!leaf.versions[1].mode)
2006                 die("Path %s not in branch", s);
2007         tree_content_set(&b->branch_tree, d,
2008                 leaf.versions[1].sha1,
2009                 leaf.versions[1].mode,
2010                 leaf.tree);
2011 }
2012
2013 static void note_change_n(struct branch *b)
2014 {
2015         const char *p = command_buf.buf + 2;
2016         static struct strbuf uq = STRBUF_INIT;
2017         struct object_entry *oe = oe;
2018         struct branch *s;
2019         unsigned char sha1[20], commit_sha1[20];
2020         uint16_t inline_data = 0;
2021
2022         /* <dataref> or 'inline' */
2023         if (*p == ':') {
2024                 char *x;
2025                 oe = find_mark(strtoumax(p + 1, &x, 10));
2026                 hashcpy(sha1, oe->sha1);
2027                 p = x;
2028         } else if (!prefixcmp(p, "inline")) {
2029                 inline_data = 1;
2030                 p += 6;
2031         } else {
2032                 if (get_sha1_hex(p, sha1))
2033                         die("Invalid SHA1: %s", command_buf.buf);
2034                 oe = find_object(sha1);
2035                 p += 40;
2036         }
2037         if (*p++ != ' ')
2038                 die("Missing space after SHA1: %s", command_buf.buf);
2039
2040         /* <committish> */
2041         s = lookup_branch(p);
2042         if (s) {
2043                 hashcpy(commit_sha1, s->sha1);
2044         } else if (*p == ':') {
2045                 uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2046                 struct object_entry *commit_oe = find_mark(commit_mark);
2047                 if (commit_oe->type != OBJ_COMMIT)
2048                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2049                 hashcpy(commit_sha1, commit_oe->sha1);
2050         } else if (!get_sha1(p, commit_sha1)) {
2051                 unsigned long size;
2052                 char *buf = read_object_with_reference(commit_sha1,
2053                         commit_type, &size, commit_sha1);
2054                 if (!buf || size < 46)
2055                         die("Not a valid commit: %s", p);
2056                 free(buf);
2057         } else
2058                 die("Invalid ref name or SHA1 expression: %s", p);
2059
2060         if (inline_data) {
2061                 static struct strbuf buf = STRBUF_INIT;
2062
2063                 if (p != uq.buf) {
2064                         strbuf_addstr(&uq, p);
2065                         p = uq.buf;
2066                 }
2067                 read_next_command();
2068                 parse_data(&buf);
2069                 store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
2070         } else if (oe) {
2071                 if (oe->type != OBJ_BLOB)
2072                         die("Not a blob (actually a %s): %s",
2073                                 typename(oe->type), command_buf.buf);
2074         } else {
2075                 enum object_type type = sha1_object_info(sha1, NULL);
2076                 if (type < 0)
2077                         die("Blob not found: %s", command_buf.buf);
2078                 if (type != OBJ_BLOB)
2079                         die("Not a blob (actually a %s): %s",
2080                             typename(type), command_buf.buf);
2081         }
2082
2083         tree_content_set(&b->branch_tree, sha1_to_hex(commit_sha1), sha1,
2084                 S_IFREG | 0644, NULL);
2085 }
2086
2087 static void file_change_deleteall(struct branch *b)
2088 {
2089         release_tree_content_recursive(b->branch_tree.tree);
2090         hashclr(b->branch_tree.versions[0].sha1);
2091         hashclr(b->branch_tree.versions[1].sha1);
2092         load_tree(&b->branch_tree);
2093 }
2094
2095 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2096 {
2097         if (!buf || size < 46)
2098                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2099         if (memcmp("tree ", buf, 5)
2100                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2101                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2102         hashcpy(b->branch_tree.versions[0].sha1,
2103                 b->branch_tree.versions[1].sha1);
2104 }
2105
2106 static void parse_from_existing(struct branch *b)
2107 {
2108         if (is_null_sha1(b->sha1)) {
2109                 hashclr(b->branch_tree.versions[0].sha1);
2110                 hashclr(b->branch_tree.versions[1].sha1);
2111         } else {
2112                 unsigned long size;
2113                 char *buf;
2114
2115                 buf = read_object_with_reference(b->sha1,
2116                         commit_type, &size, b->sha1);
2117                 parse_from_commit(b, buf, size);
2118                 free(buf);
2119         }
2120 }
2121
2122 static int parse_from(struct branch *b)
2123 {
2124         const char *from;
2125         struct branch *s;
2126
2127         if (prefixcmp(command_buf.buf, "from "))
2128                 return 0;
2129
2130         if (b->branch_tree.tree) {
2131                 release_tree_content_recursive(b->branch_tree.tree);
2132                 b->branch_tree.tree = NULL;
2133         }
2134
2135         from = strchr(command_buf.buf, ' ') + 1;
2136         s = lookup_branch(from);
2137         if (b == s)
2138                 die("Can't create a branch from itself: %s", b->name);
2139         else if (s) {
2140                 unsigned char *t = s->branch_tree.versions[1].sha1;
2141                 hashcpy(b->sha1, s->sha1);
2142                 hashcpy(b->branch_tree.versions[0].sha1, t);
2143                 hashcpy(b->branch_tree.versions[1].sha1, t);
2144         } else if (*from == ':') {
2145                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2146                 struct object_entry *oe = find_mark(idnum);
2147                 if (oe->type != OBJ_COMMIT)
2148                         die("Mark :%" PRIuMAX " not a commit", idnum);
2149                 hashcpy(b->sha1, oe->sha1);
2150                 if (oe->pack_id != MAX_PACK_ID) {
2151                         unsigned long size;
2152                         char *buf = gfi_unpack_entry(oe, &size);
2153                         parse_from_commit(b, buf, size);
2154                         free(buf);
2155                 } else
2156                         parse_from_existing(b);
2157         } else if (!get_sha1(from, b->sha1))
2158                 parse_from_existing(b);
2159         else
2160                 die("Invalid ref name or SHA1 expression: %s", from);
2161
2162         read_next_command();
2163         return 1;
2164 }
2165
2166 static struct hash_list *parse_merge(unsigned int *count)
2167 {
2168         struct hash_list *list = NULL, *n, *e = e;
2169         const char *from;
2170         struct branch *s;
2171
2172         *count = 0;
2173         while (!prefixcmp(command_buf.buf, "merge ")) {
2174                 from = strchr(command_buf.buf, ' ') + 1;
2175                 n = xmalloc(sizeof(*n));
2176                 s = lookup_branch(from);
2177                 if (s)
2178                         hashcpy(n->sha1, s->sha1);
2179                 else if (*from == ':') {
2180                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2181                         struct object_entry *oe = find_mark(idnum);
2182                         if (oe->type != OBJ_COMMIT)
2183                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2184                         hashcpy(n->sha1, oe->sha1);
2185                 } else if (!get_sha1(from, n->sha1)) {
2186                         unsigned long size;
2187                         char *buf = read_object_with_reference(n->sha1,
2188                                 commit_type, &size, n->sha1);
2189                         if (!buf || size < 46)
2190                                 die("Not a valid commit: %s", from);
2191                         free(buf);
2192                 } else
2193                         die("Invalid ref name or SHA1 expression: %s", from);
2194
2195                 n->next = NULL;
2196                 if (list)
2197                         e->next = n;
2198                 else
2199                         list = n;
2200                 e = n;
2201                 (*count)++;
2202                 read_next_command();
2203         }
2204         return list;
2205 }
2206
2207 static void parse_new_commit(void)
2208 {
2209         static struct strbuf msg = STRBUF_INIT;
2210         struct branch *b;
2211         char *sp;
2212         char *author = NULL;
2213         char *committer = NULL;
2214         struct hash_list *merge_list = NULL;
2215         unsigned int merge_count;
2216
2217         /* Obtain the branch name from the rest of our command */
2218         sp = strchr(command_buf.buf, ' ') + 1;
2219         b = lookup_branch(sp);
2220         if (!b)
2221                 b = new_branch(sp);
2222
2223         read_next_command();
2224         parse_mark();
2225         if (!prefixcmp(command_buf.buf, "author ")) {
2226                 author = parse_ident(command_buf.buf + 7);
2227                 read_next_command();
2228         }
2229         if (!prefixcmp(command_buf.buf, "committer ")) {
2230                 committer = parse_ident(command_buf.buf + 10);
2231                 read_next_command();
2232         }
2233         if (!committer)
2234                 die("Expected committer but didn't get one");
2235         parse_data(&msg);
2236         read_next_command();
2237         parse_from(b);
2238         merge_list = parse_merge(&merge_count);
2239
2240         /* ensure the branch is active/loaded */
2241         if (!b->branch_tree.tree || !max_active_branches) {
2242                 unload_one_branch();
2243                 load_branch(b);
2244         }
2245
2246         /* file_change* */
2247         while (command_buf.len > 0) {
2248                 if (!prefixcmp(command_buf.buf, "M "))
2249                         file_change_m(b);
2250                 else if (!prefixcmp(command_buf.buf, "D "))
2251                         file_change_d(b);
2252                 else if (!prefixcmp(command_buf.buf, "R "))
2253                         file_change_cr(b, 1);
2254                 else if (!prefixcmp(command_buf.buf, "C "))
2255                         file_change_cr(b, 0);
2256                 else if (!prefixcmp(command_buf.buf, "N "))
2257                         note_change_n(b);
2258                 else if (!strcmp("deleteall", command_buf.buf))
2259                         file_change_deleteall(b);
2260                 else {
2261                         unread_command_buf = 1;
2262                         break;
2263                 }
2264                 if (read_next_command() == EOF)
2265                         break;
2266         }
2267
2268         /* build the tree and the commit */
2269         store_tree(&b->branch_tree);
2270         hashcpy(b->branch_tree.versions[0].sha1,
2271                 b->branch_tree.versions[1].sha1);
2272
2273         strbuf_reset(&new_data);
2274         strbuf_addf(&new_data, "tree %s\n",
2275                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2276         if (!is_null_sha1(b->sha1))
2277                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2278         while (merge_list) {
2279                 struct hash_list *next = merge_list->next;
2280                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2281                 free(merge_list);
2282                 merge_list = next;
2283         }
2284         strbuf_addf(&new_data,
2285                 "author %s\n"
2286                 "committer %s\n"
2287                 "\n",
2288                 author ? author : committer, committer);
2289         strbuf_addbuf(&new_data, &msg);
2290         free(author);
2291         free(committer);
2292
2293         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2294                 b->pack_id = pack_id;
2295         b->last_commit = object_count_by_type[OBJ_COMMIT];
2296 }
2297
2298 static void parse_new_tag(void)
2299 {
2300         static struct strbuf msg = STRBUF_INIT;
2301         char *sp;
2302         const char *from;
2303         char *tagger;
2304         struct branch *s;
2305         struct tag *t;
2306         uintmax_t from_mark = 0;
2307         unsigned char sha1[20];
2308         enum object_type type;
2309
2310         /* Obtain the new tag name from the rest of our command */
2311         sp = strchr(command_buf.buf, ' ') + 1;
2312         t = pool_alloc(sizeof(struct tag));
2313         t->next_tag = NULL;
2314         t->name = pool_strdup(sp);
2315         if (last_tag)
2316                 last_tag->next_tag = t;
2317         else
2318                 first_tag = t;
2319         last_tag = t;
2320         read_next_command();
2321
2322         /* from ... */
2323         if (prefixcmp(command_buf.buf, "from "))
2324                 die("Expected from command, got %s", command_buf.buf);
2325         from = strchr(command_buf.buf, ' ') + 1;
2326         s = lookup_branch(from);
2327         if (s) {
2328                 hashcpy(sha1, s->sha1);
2329                 type = OBJ_COMMIT;
2330         } else if (*from == ':') {
2331                 struct object_entry *oe;
2332                 from_mark = strtoumax(from + 1, NULL, 10);
2333                 oe = find_mark(from_mark);
2334                 type = oe->type;
2335                 hashcpy(sha1, oe->sha1);
2336         } else if (!get_sha1(from, sha1)) {
2337                 unsigned long size;
2338                 char *buf;
2339
2340                 buf = read_sha1_file(sha1, &type, &size);
2341                 if (!buf || size < 46)
2342                         die("Not a valid commit: %s", from);
2343                 free(buf);
2344         } else
2345                 die("Invalid ref name or SHA1 expression: %s", from);
2346         read_next_command();
2347
2348         /* tagger ... */
2349         if (!prefixcmp(command_buf.buf, "tagger ")) {
2350                 tagger = parse_ident(command_buf.buf + 7);
2351                 read_next_command();
2352         } else
2353                 tagger = NULL;
2354
2355         /* tag payload/message */
2356         parse_data(&msg);
2357
2358         /* build the tag object */
2359         strbuf_reset(&new_data);
2360
2361         strbuf_addf(&new_data,
2362                     "object %s\n"
2363                     "type %s\n"
2364                     "tag %s\n",
2365                     sha1_to_hex(sha1), typename(type), t->name);
2366         if (tagger)
2367                 strbuf_addf(&new_data,
2368                             "tagger %s\n", tagger);
2369         strbuf_addch(&new_data, '\n');
2370         strbuf_addbuf(&new_data, &msg);
2371         free(tagger);
2372
2373         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2374                 t->pack_id = MAX_PACK_ID;
2375         else
2376                 t->pack_id = pack_id;
2377 }
2378
2379 static void parse_reset_branch(void)
2380 {
2381         struct branch *b;
2382         char *sp;
2383
2384         /* Obtain the branch name from the rest of our command */
2385         sp = strchr(command_buf.buf, ' ') + 1;
2386         b = lookup_branch(sp);
2387         if (b) {
2388                 hashclr(b->sha1);
2389                 hashclr(b->branch_tree.versions[0].sha1);
2390                 hashclr(b->branch_tree.versions[1].sha1);
2391                 if (b->branch_tree.tree) {
2392                         release_tree_content_recursive(b->branch_tree.tree);
2393                         b->branch_tree.tree = NULL;
2394                 }
2395         }
2396         else
2397                 b = new_branch(sp);
2398         read_next_command();
2399         parse_from(b);
2400         if (command_buf.len > 0)
2401                 unread_command_buf = 1;
2402 }
2403
2404 static void parse_checkpoint(void)
2405 {
2406         if (object_count) {
2407                 cycle_packfile();
2408                 dump_branches();
2409                 dump_tags();
2410                 dump_marks();
2411         }
2412         skip_optional_lf();
2413 }
2414
2415 static void parse_progress(void)
2416 {
2417         fwrite(command_buf.buf, 1, command_buf.len, stdout);
2418         fputc('\n', stdout);
2419         fflush(stdout);
2420         skip_optional_lf();
2421 }
2422
2423 static void import_marks(const char *input_file)
2424 {
2425         char line[512];
2426         FILE *f = fopen(input_file, "r");
2427         if (!f)
2428                 die_errno("cannot read '%s'", input_file);
2429         while (fgets(line, sizeof(line), f)) {
2430                 uintmax_t mark;
2431                 char *end;
2432                 unsigned char sha1[20];
2433                 struct object_entry *e;
2434
2435                 end = strchr(line, '\n');
2436                 if (line[0] != ':' || !end)
2437                         die("corrupt mark line: %s", line);
2438                 *end = 0;
2439                 mark = strtoumax(line + 1, &end, 10);
2440                 if (!mark || end == line + 1
2441                         || *end != ' ' || get_sha1(end + 1, sha1))
2442                         die("corrupt mark line: %s", line);
2443                 e = find_object(sha1);
2444                 if (!e) {
2445                         enum object_type type = sha1_object_info(sha1, NULL);
2446                         if (type < 0)
2447                                 die("object not found: %s", sha1_to_hex(sha1));
2448                         e = insert_object(sha1);
2449                         e->type = type;
2450                         e->pack_id = MAX_PACK_ID;
2451                         e->offset = 1; /* just not zero! */
2452                 }
2453                 insert_mark(mark, e);
2454         }
2455         fclose(f);
2456 }
2457
2458 static int git_pack_config(const char *k, const char *v, void *cb)
2459 {
2460         if (!strcmp(k, "pack.depth")) {
2461                 max_depth = git_config_int(k, v);
2462                 if (max_depth > MAX_DEPTH)
2463                         max_depth = MAX_DEPTH;
2464                 return 0;
2465         }
2466         if (!strcmp(k, "pack.compression")) {
2467                 int level = git_config_int(k, v);
2468                 if (level == -1)
2469                         level = Z_DEFAULT_COMPRESSION;
2470                 else if (level < 0 || level > Z_BEST_COMPRESSION)
2471                         die("bad pack compression level %d", level);
2472                 pack_compression_level = level;
2473                 pack_compression_seen = 1;
2474                 return 0;
2475         }
2476         return git_default_config(k, v, cb);
2477 }
2478
2479 static const char fast_import_usage[] =
2480 "git fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2481
2482 int main(int argc, const char **argv)
2483 {
2484         unsigned int i, show_stats = 1;
2485
2486         git_extract_argv0_path(argv[0]);
2487
2488         if (argc == 2 && !strcmp(argv[1], "-h"))
2489                 usage(fast_import_usage);
2490
2491         setup_git_directory();
2492         git_config(git_pack_config, NULL);
2493         if (!pack_compression_seen && core_compression_seen)
2494                 pack_compression_level = core_compression_level;
2495
2496         alloc_objects(object_entry_alloc);
2497         strbuf_init(&command_buf, 0);
2498         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2499         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2500         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2501         marks = pool_calloc(1, sizeof(struct mark_set));
2502
2503         for (i = 1; i < argc; i++) {
2504                 const char *a = argv[i];
2505
2506                 if (*a != '-' || !strcmp(a, "--"))
2507                         break;
2508                 else if (!prefixcmp(a, "--date-format=")) {
2509                         const char *fmt = a + 14;
2510                         if (!strcmp(fmt, "raw"))
2511                                 whenspec = WHENSPEC_RAW;
2512                         else if (!strcmp(fmt, "rfc2822"))
2513                                 whenspec = WHENSPEC_RFC2822;
2514                         else if (!strcmp(fmt, "now"))
2515                                 whenspec = WHENSPEC_NOW;
2516                         else
2517                                 die("unknown --date-format argument %s", fmt);
2518                 }
2519                 else if (!prefixcmp(a, "--max-pack-size="))
2520                         max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2521                 else if (!prefixcmp(a, "--depth=")) {
2522                         max_depth = strtoul(a + 8, NULL, 0);
2523                         if (max_depth > MAX_DEPTH)
2524                                 die("--depth cannot exceed %u", MAX_DEPTH);
2525                 }
2526                 else if (!prefixcmp(a, "--active-branches="))
2527                         max_active_branches = strtoul(a + 18, NULL, 0);
2528                 else if (!prefixcmp(a, "--import-marks="))
2529                         import_marks(a + 15);
2530                 else if (!prefixcmp(a, "--export-marks="))
2531                         mark_file = a + 15;
2532                 else if (!prefixcmp(a, "--export-pack-edges=")) {
2533                         if (pack_edges)
2534                                 fclose(pack_edges);
2535                         pack_edges = fopen(a + 20, "a");
2536                         if (!pack_edges)
2537                                 die_errno("Cannot open '%s'", a + 20);
2538                 } else if (!strcmp(a, "--force"))
2539                         force_update = 1;
2540                 else if (!strcmp(a, "--quiet"))
2541                         show_stats = 0;
2542                 else if (!strcmp(a, "--stats"))
2543                         show_stats = 1;
2544                 else
2545                         die("unknown option %s", a);
2546         }
2547         if (i != argc)
2548                 usage(fast_import_usage);
2549
2550         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2551         for (i = 0; i < (cmd_save - 1); i++)
2552                 rc_free[i].next = &rc_free[i + 1];
2553         rc_free[cmd_save - 1].next = NULL;
2554
2555         prepare_packed_git();
2556         start_packfile();
2557         set_die_routine(die_nicely);
2558         while (read_next_command() != EOF) {
2559                 if (!strcmp("blob", command_buf.buf))
2560                         parse_new_blob();
2561                 else if (!prefixcmp(command_buf.buf, "commit "))
2562                         parse_new_commit();
2563                 else if (!prefixcmp(command_buf.buf, "tag "))
2564                         parse_new_tag();
2565                 else if (!prefixcmp(command_buf.buf, "reset "))
2566                         parse_reset_branch();
2567                 else if (!strcmp("checkpoint", command_buf.buf))
2568                         parse_checkpoint();
2569                 else if (!prefixcmp(command_buf.buf, "progress "))
2570                         parse_progress();
2571                 else
2572                         die("Unsupported command: %s", command_buf.buf);
2573         }
2574         end_packfile();
2575
2576         dump_branches();
2577         dump_tags();
2578         unkeep_all_packs();
2579         dump_marks();
2580
2581         if (pack_edges)
2582                 fclose(pack_edges);
2583
2584         if (show_stats) {
2585                 uintmax_t total_count = 0, duplicate_count = 0;
2586                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2587                         total_count += object_count_by_type[i];
2588                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2589                         duplicate_count += duplicate_count_by_type[i];
2590
2591                 fprintf(stderr, "%s statistics:\n", argv[0]);
2592                 fprintf(stderr, "---------------------------------------------------------------------\n");
2593                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2594                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2595                 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]);
2596                 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]);
2597                 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]);
2598                 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]);
2599                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2600                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2601                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2602                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2603                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2604                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2605                 fprintf(stderr, "---------------------------------------------------------------------\n");
2606                 pack_report();
2607                 fprintf(stderr, "---------------------------------------------------------------------\n");
2608                 fprintf(stderr, "\n");
2609         }
2610
2611         return failure ? 1 : 0;
2612 }