git-send-email: handle email address with quoted comma
[git] / fast-import.c
1 /*
2 Format of STDIN stream:
3
4   stream ::= cmd*;
5
6   cmd ::= new_blob
7         | new_commit
8         | new_tag
9         | reset_branch
10         | checkpoint
11         | progress
12         ;
13
14   new_blob ::= 'blob' lf
15     mark?
16     file_content;
17   file_content ::= data;
18
19   new_commit ::= 'commit' sp ref_str lf
20     mark?
21     ('author' sp name '<' email '>' when lf)?
22     'committer' sp name '<' email '>' when lf
23     commit_msg
24     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
25     ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
26     file_change*
27     lf?;
28   commit_msg ::= data;
29
30   file_change ::= file_clr
31     | file_del
32     | file_rnm
33     | file_cpy
34     | file_obm
35     | file_inm;
36   file_clr ::= 'deleteall' lf;
37   file_del ::= 'D' sp path_str lf;
38   file_rnm ::= 'R' sp path_str sp path_str lf;
39   file_cpy ::= 'C' sp path_str sp path_str lf;
40   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
41   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
42     data;
43
44   new_tag ::= 'tag' sp tag_str lf
45     'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
46     'tagger' sp name '<' email '>' when lf
47     tag_msg;
48   tag_msg ::= data;
49
50   reset_branch ::= 'reset' sp ref_str lf
51     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
52     lf?;
53
54   checkpoint ::= 'checkpoint' lf
55     lf?;
56
57   progress ::= 'progress' sp not_lf* lf
58     lf?;
59
60      # note: the first idnum in a stream should be 1 and subsequent
61      # idnums should not have gaps between values as this will cause
62      # the stream parser to reserve space for the gapped values.  An
63      # idnum can be updated in the future to a new object by issuing
64      # a new mark directive with the old idnum.
65      #
66   mark ::= 'mark' sp idnum lf;
67   data ::= (delimited_data | exact_data)
68     lf?;
69
70     # note: delim may be any string but must not contain lf.
71     # data_line may contain any data but must not be exactly
72     # delim.
73   delimited_data ::= 'data' sp '<<' delim lf
74     (data_line lf)*
75     delim lf;
76
77      # note: declen indicates the length of binary_data in bytes.
78      # declen does not include the lf preceeding the binary data.
79      #
80   exact_data ::= 'data' sp declen lf
81     binary_data;
82
83      # note: quoted strings are C-style quoting supporting \c for
84      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
85      # is the signed byte value in octal.  Note that the only
86      # characters which must actually be escaped to protect the
87      # stream formatting is: \, " and LF.  Otherwise these values
88      # are UTF8.
89      #
90   ref_str     ::= ref;
91   sha1exp_str ::= sha1exp;
92   tag_str     ::= tag;
93   path_str    ::= path    | '"' quoted(path)    '"' ;
94   mode        ::= '100644' | '644'
95                 | '100755' | '755'
96                 | '120000'
97                 ;
98
99   declen ::= # unsigned 32 bit value, ascii base10 notation;
100   bigint ::= # unsigned integer value, ascii base10 notation;
101   binary_data ::= # file content, not interpreted;
102
103   when         ::= raw_when | rfc2822_when;
104   raw_when     ::= ts sp tz;
105   rfc2822_when ::= # Valid RFC 2822 date and time;
106
107   sp ::= # ASCII space character;
108   lf ::= # ASCII newline (LF) character;
109
110      # note: a colon (':') must precede the numerical value assigned to
111      # an idnum.  This is to distinguish it from a ref or tag name as
112      # GIT does not permit ':' in ref or tag strings.
113      #
114   idnum   ::= ':' bigint;
115   path    ::= # GIT style file path, e.g. "a/b/c";
116   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
117   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
118   sha1exp ::= # Any valid GIT SHA1 expression;
119   hexsha1 ::= # SHA1 in hexadecimal format;
120
121      # note: name and email are UTF8 strings, however name must not
122      # contain '<' or lf and email must not contain any of the
123      # following: '<', '>', lf.
124      #
125   name  ::= # valid GIT author/committer name;
126   email ::= # valid GIT author/committer email;
127   ts    ::= # time since the epoch in seconds, ascii base10 notation;
128   tz    ::= # GIT style timezone;
129
130      # note: comments may appear anywhere in the input, except
131      # within a data command.  Any form of the data command
132      # always escapes the related input from comment processing.
133      #
134      # In case it is not clear, the '#' that starts the comment
135      # must be the first character on that the line (an lf have
136      # preceeded it).
137      #
138   comment ::= '#' not_lf* lf;
139   not_lf  ::= # Any byte that is not ASCII newline (LF);
140 */
141
142 #include "builtin.h"
143 #include "cache.h"
144 #include "object.h"
145 #include "blob.h"
146 #include "tree.h"
147 #include "commit.h"
148 #include "delta.h"
149 #include "pack.h"
150 #include "refs.h"
151 #include "csum-file.h"
152 #include "quote.h"
153
154 #define PACK_ID_BITS 16
155 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
156 #define DEPTH_BITS 13
157 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
158
159 struct object_entry
160 {
161         struct object_entry *next;
162         uint32_t offset;
163         uint32_t type : TYPE_BITS,
164                 pack_id : PACK_ID_BITS,
165                 depth : DEPTH_BITS;
166         unsigned char sha1[20];
167 };
168
169 struct object_entry_pool
170 {
171         struct object_entry_pool *next_pool;
172         struct object_entry *next_free;
173         struct object_entry *end;
174         struct object_entry entries[FLEX_ARRAY]; /* more */
175 };
176
177 struct mark_set
178 {
179         union {
180                 struct object_entry *marked[1024];
181                 struct mark_set *sets[1024];
182         } data;
183         unsigned int shift;
184 };
185
186 struct last_object
187 {
188         struct strbuf data;
189         uint32_t offset;
190         unsigned int depth;
191         unsigned no_swap : 1;
192 };
193
194 struct mem_pool
195 {
196         struct mem_pool *next_pool;
197         char *next_free;
198         char *end;
199         uintmax_t space[FLEX_ARRAY]; /* more */
200 };
201
202 struct atom_str
203 {
204         struct atom_str *next_atom;
205         unsigned short str_len;
206         char str_dat[FLEX_ARRAY]; /* more */
207 };
208
209 struct tree_content;
210 struct tree_entry
211 {
212         struct tree_content *tree;
213         struct atom_str* name;
214         struct tree_entry_ms
215         {
216                 uint16_t mode;
217                 unsigned char sha1[20];
218         } versions[2];
219 };
220
221 struct tree_content
222 {
223         unsigned int entry_capacity; /* must match avail_tree_content */
224         unsigned int entry_count;
225         unsigned int delta_depth;
226         struct tree_entry *entries[FLEX_ARRAY]; /* more */
227 };
228
229 struct avail_tree_content
230 {
231         unsigned int entry_capacity; /* must match tree_content */
232         struct avail_tree_content *next_avail;
233 };
234
235 struct branch
236 {
237         struct branch *table_next_branch;
238         struct branch *active_next_branch;
239         const char *name;
240         struct tree_entry branch_tree;
241         uintmax_t last_commit;
242         unsigned active : 1;
243         unsigned pack_id : PACK_ID_BITS;
244         unsigned char sha1[20];
245 };
246
247 struct tag
248 {
249         struct tag *next_tag;
250         const char *name;
251         unsigned int pack_id;
252         unsigned char sha1[20];
253 };
254
255 struct hash_list
256 {
257         struct hash_list *next;
258         unsigned char sha1[20];
259 };
260
261 typedef enum {
262         WHENSPEC_RAW = 1,
263         WHENSPEC_RFC2822,
264         WHENSPEC_NOW,
265 } whenspec_type;
266
267 struct recent_command
268 {
269         struct recent_command *prev;
270         struct recent_command *next;
271         char *buf;
272 };
273
274 /* Configured limits on output */
275 static unsigned long max_depth = 10;
276 static off_t max_packsize = (1LL << 32) - 1;
277 static int force_update;
278 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
279 static int pack_compression_seen;
280
281 /* Stats and misc. counters */
282 static uintmax_t alloc_count;
283 static uintmax_t marks_set_count;
284 static uintmax_t object_count_by_type[1 << TYPE_BITS];
285 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
286 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
287 static unsigned long object_count;
288 static unsigned long branch_count;
289 static unsigned long branch_load_count;
290 static int failure;
291 static FILE *pack_edges;
292
293 /* Memory pools */
294 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
295 static size_t total_allocd;
296 static struct mem_pool *mem_pool;
297
298 /* Atom management */
299 static unsigned int atom_table_sz = 4451;
300 static unsigned int atom_cnt;
301 static struct atom_str **atom_table;
302
303 /* The .pack file being generated */
304 static unsigned int pack_id;
305 static struct packed_git *pack_data;
306 static struct packed_git **all_packs;
307 static unsigned long pack_size;
308
309 /* Table of objects we've written. */
310 static unsigned int object_entry_alloc = 5000;
311 static struct object_entry_pool *blocks;
312 static struct object_entry *object_table[1 << 16];
313 static struct mark_set *marks;
314 static const char* mark_file;
315
316 /* Our last blob */
317 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
318
319 /* Tree management */
320 static unsigned int tree_entry_alloc = 1000;
321 static void *avail_tree_entry;
322 static unsigned int avail_tree_table_sz = 100;
323 static struct avail_tree_content **avail_tree_table;
324 static struct strbuf old_tree = STRBUF_INIT;
325 static struct strbuf new_tree = STRBUF_INIT;
326
327 /* Branch data */
328 static unsigned long max_active_branches = 5;
329 static unsigned long cur_active_branches;
330 static unsigned long branch_table_sz = 1039;
331 static struct branch **branch_table;
332 static struct branch *active_branches;
333
334 /* Tag data */
335 static struct tag *first_tag;
336 static struct tag *last_tag;
337
338 /* Input stream parsing */
339 static whenspec_type whenspec = WHENSPEC_RAW;
340 static struct strbuf command_buf = STRBUF_INIT;
341 static int unread_command_buf;
342 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
343 static struct recent_command *cmd_tail = &cmd_hist;
344 static struct recent_command *rc_free;
345 static unsigned int cmd_save = 100;
346 static uintmax_t next_mark;
347 static struct strbuf new_data = STRBUF_INIT;
348
349 static void write_branch_report(FILE *rpt, struct branch *b)
350 {
351         fprintf(rpt, "%s:\n", b->name);
352
353         fprintf(rpt, "  status      :");
354         if (b->active)
355                 fputs(" active", rpt);
356         if (b->branch_tree.tree)
357                 fputs(" loaded", rpt);
358         if (is_null_sha1(b->branch_tree.versions[1].sha1))
359                 fputs(" dirty", rpt);
360         fputc('\n', rpt);
361
362         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
363         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
364         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
365         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
366
367         fputs("  last pack   : ", rpt);
368         if (b->pack_id < MAX_PACK_ID)
369                 fprintf(rpt, "%u", b->pack_id);
370         fputc('\n', rpt);
371
372         fputc('\n', rpt);
373 }
374
375 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
376
377 static void write_crash_report(const char *err)
378 {
379         char *loc = git_path("fast_import_crash_%d", getpid());
380         FILE *rpt = fopen(loc, "w");
381         struct branch *b;
382         unsigned long lu;
383         struct recent_command *rc;
384
385         if (!rpt) {
386                 error("can't write crash report %s: %s", loc, strerror(errno));
387                 return;
388         }
389
390         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
391
392         fprintf(rpt, "fast-import crash report:\n");
393         fprintf(rpt, "    fast-import process: %d\n", getpid());
394         fprintf(rpt, "    parent process     : %d\n", getppid());
395         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
396         fputc('\n', rpt);
397
398         fputs("fatal: ", rpt);
399         fputs(err, rpt);
400         fputc('\n', rpt);
401
402         fputc('\n', rpt);
403         fputs("Most Recent Commands Before Crash\n", rpt);
404         fputs("---------------------------------\n", rpt);
405         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
406                 if (rc->next == &cmd_hist)
407                         fputs("* ", rpt);
408                 else
409                         fputs("  ", rpt);
410                 fputs(rc->buf, rpt);
411                 fputc('\n', rpt);
412         }
413
414         fputc('\n', rpt);
415         fputs("Active Branch LRU\n", rpt);
416         fputs("-----------------\n", rpt);
417         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
418                 cur_active_branches,
419                 max_active_branches);
420         fputc('\n', rpt);
421         fputs("  pos  clock name\n", rpt);
422         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
423         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
424                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
425                         ++lu, b->last_commit, b->name);
426
427         fputc('\n', rpt);
428         fputs("Inactive Branches\n", rpt);
429         fputs("-----------------\n", rpt);
430         for (lu = 0; lu < branch_table_sz; lu++) {
431                 for (b = branch_table[lu]; b; b = b->table_next_branch)
432                         write_branch_report(rpt, b);
433         }
434
435         if (first_tag) {
436                 struct tag *tg;
437                 fputc('\n', rpt);
438                 fputs("Annotated Tags\n", rpt);
439                 fputs("--------------\n", rpt);
440                 for (tg = first_tag; tg; tg = tg->next_tag) {
441                         fputs(sha1_to_hex(tg->sha1), rpt);
442                         fputc(' ', rpt);
443                         fputs(tg->name, rpt);
444                         fputc('\n', rpt);
445                 }
446         }
447
448         fputc('\n', rpt);
449         fputs("Marks\n", rpt);
450         fputs("-----\n", rpt);
451         if (mark_file)
452                 fprintf(rpt, "  exported to %s\n", mark_file);
453         else
454                 dump_marks_helper(rpt, 0, marks);
455
456         fputc('\n', rpt);
457         fputs("-------------------\n", rpt);
458         fputs("END OF CRASH REPORT\n", rpt);
459         fclose(rpt);
460 }
461
462 static void end_packfile(void);
463 static void unkeep_all_packs(void);
464 static void dump_marks(void);
465
466 static NORETURN void die_nicely(const char *err, va_list params)
467 {
468         static int zombie;
469         char message[2 * PATH_MAX];
470
471         vsnprintf(message, sizeof(message), err, params);
472         fputs("fatal: ", stderr);
473         fputs(message, stderr);
474         fputc('\n', stderr);
475
476         if (!zombie) {
477                 zombie = 1;
478                 write_crash_report(message);
479                 end_packfile();
480                 unkeep_all_packs();
481                 dump_marks();
482         }
483         exit(128);
484 }
485
486 static void alloc_objects(unsigned int cnt)
487 {
488         struct object_entry_pool *b;
489
490         b = xmalloc(sizeof(struct object_entry_pool)
491                 + cnt * sizeof(struct object_entry));
492         b->next_pool = blocks;
493         b->next_free = b->entries;
494         b->end = b->entries + cnt;
495         blocks = b;
496         alloc_count += cnt;
497 }
498
499 static struct object_entry *new_object(unsigned char *sha1)
500 {
501         struct object_entry *e;
502
503         if (blocks->next_free == blocks->end)
504                 alloc_objects(object_entry_alloc);
505
506         e = blocks->next_free++;
507         hashcpy(e->sha1, sha1);
508         return e;
509 }
510
511 static struct object_entry *find_object(unsigned char *sha1)
512 {
513         unsigned int h = sha1[0] << 8 | sha1[1];
514         struct object_entry *e;
515         for (e = object_table[h]; e; e = e->next)
516                 if (!hashcmp(sha1, e->sha1))
517                         return e;
518         return NULL;
519 }
520
521 static struct object_entry *insert_object(unsigned char *sha1)
522 {
523         unsigned int h = sha1[0] << 8 | sha1[1];
524         struct object_entry *e = object_table[h];
525         struct object_entry *p = NULL;
526
527         while (e) {
528                 if (!hashcmp(sha1, e->sha1))
529                         return e;
530                 p = e;
531                 e = e->next;
532         }
533
534         e = new_object(sha1);
535         e->next = NULL;
536         e->offset = 0;
537         if (p)
538                 p->next = e;
539         else
540                 object_table[h] = e;
541         return e;
542 }
543
544 static unsigned int hc_str(const char *s, size_t len)
545 {
546         unsigned int r = 0;
547         while (len-- > 0)
548                 r = r * 31 + *s++;
549         return r;
550 }
551
552 static void *pool_alloc(size_t len)
553 {
554         struct mem_pool *p;
555         void *r;
556
557         for (p = mem_pool; p; p = p->next_pool)
558                 if ((p->end - p->next_free >= len))
559                         break;
560
561         if (!p) {
562                 if (len >= (mem_pool_alloc/2)) {
563                         total_allocd += len;
564                         return xmalloc(len);
565                 }
566                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
567                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
568                 p->next_pool = mem_pool;
569                 p->next_free = (char *) p->space;
570                 p->end = p->next_free + mem_pool_alloc;
571                 mem_pool = p;
572         }
573
574         r = p->next_free;
575         /* round out to a 'uintmax_t' alignment */
576         if (len & (sizeof(uintmax_t) - 1))
577                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
578         p->next_free += len;
579         return r;
580 }
581
582 static void *pool_calloc(size_t count, size_t size)
583 {
584         size_t len = count * size;
585         void *r = pool_alloc(len);
586         memset(r, 0, len);
587         return r;
588 }
589
590 static char *pool_strdup(const char *s)
591 {
592         char *r = pool_alloc(strlen(s) + 1);
593         strcpy(r, s);
594         return r;
595 }
596
597 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
598 {
599         struct mark_set *s = marks;
600         while ((idnum >> s->shift) >= 1024) {
601                 s = pool_calloc(1, sizeof(struct mark_set));
602                 s->shift = marks->shift + 10;
603                 s->data.sets[0] = marks;
604                 marks = s;
605         }
606         while (s->shift) {
607                 uintmax_t i = idnum >> s->shift;
608                 idnum -= i << s->shift;
609                 if (!s->data.sets[i]) {
610                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
611                         s->data.sets[i]->shift = s->shift - 10;
612                 }
613                 s = s->data.sets[i];
614         }
615         if (!s->data.marked[idnum])
616                 marks_set_count++;
617         s->data.marked[idnum] = oe;
618 }
619
620 static struct object_entry *find_mark(uintmax_t idnum)
621 {
622         uintmax_t orig_idnum = idnum;
623         struct mark_set *s = marks;
624         struct object_entry *oe = NULL;
625         if ((idnum >> s->shift) < 1024) {
626                 while (s && s->shift) {
627                         uintmax_t i = idnum >> s->shift;
628                         idnum -= i << s->shift;
629                         s = s->data.sets[i];
630                 }
631                 if (s)
632                         oe = s->data.marked[idnum];
633         }
634         if (!oe)
635                 die("mark :%" PRIuMAX " not declared", orig_idnum);
636         return oe;
637 }
638
639 static struct atom_str *to_atom(const char *s, unsigned short len)
640 {
641         unsigned int hc = hc_str(s, len) % atom_table_sz;
642         struct atom_str *c;
643
644         for (c = atom_table[hc]; c; c = c->next_atom)
645                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
646                         return c;
647
648         c = pool_alloc(sizeof(struct atom_str) + len + 1);
649         c->str_len = len;
650         strncpy(c->str_dat, s, len);
651         c->str_dat[len] = 0;
652         c->next_atom = atom_table[hc];
653         atom_table[hc] = c;
654         atom_cnt++;
655         return c;
656 }
657
658 static struct branch *lookup_branch(const char *name)
659 {
660         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
661         struct branch *b;
662
663         for (b = branch_table[hc]; b; b = b->table_next_branch)
664                 if (!strcmp(name, b->name))
665                         return b;
666         return NULL;
667 }
668
669 static struct branch *new_branch(const char *name)
670 {
671         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
672         struct branch* b = lookup_branch(name);
673
674         if (b)
675                 die("Invalid attempt to create duplicate branch: %s", name);
676         switch (check_ref_format(name)) {
677         case 0: break; /* its valid */
678         case CHECK_REF_FORMAT_ONELEVEL:
679                 break; /* valid, but too few '/', allow anyway */
680         default:
681                 die("Branch name doesn't conform to GIT standards: %s", name);
682         }
683
684         b = pool_calloc(1, sizeof(struct branch));
685         b->name = pool_strdup(name);
686         b->table_next_branch = branch_table[hc];
687         b->branch_tree.versions[0].mode = S_IFDIR;
688         b->branch_tree.versions[1].mode = S_IFDIR;
689         b->active = 0;
690         b->pack_id = MAX_PACK_ID;
691         branch_table[hc] = b;
692         branch_count++;
693         return b;
694 }
695
696 static unsigned int hc_entries(unsigned int cnt)
697 {
698         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
699         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
700 }
701
702 static struct tree_content *new_tree_content(unsigned int cnt)
703 {
704         struct avail_tree_content *f, *l = NULL;
705         struct tree_content *t;
706         unsigned int hc = hc_entries(cnt);
707
708         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
709                 if (f->entry_capacity >= cnt)
710                         break;
711
712         if (f) {
713                 if (l)
714                         l->next_avail = f->next_avail;
715                 else
716                         avail_tree_table[hc] = f->next_avail;
717         } else {
718                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
719                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
720                 f->entry_capacity = cnt;
721         }
722
723         t = (struct tree_content*)f;
724         t->entry_count = 0;
725         t->delta_depth = 0;
726         return t;
727 }
728
729 static void release_tree_entry(struct tree_entry *e);
730 static void release_tree_content(struct tree_content *t)
731 {
732         struct avail_tree_content *f = (struct avail_tree_content*)t;
733         unsigned int hc = hc_entries(f->entry_capacity);
734         f->next_avail = avail_tree_table[hc];
735         avail_tree_table[hc] = f;
736 }
737
738 static void release_tree_content_recursive(struct tree_content *t)
739 {
740         unsigned int i;
741         for (i = 0; i < t->entry_count; i++)
742                 release_tree_entry(t->entries[i]);
743         release_tree_content(t);
744 }
745
746 static struct tree_content *grow_tree_content(
747         struct tree_content *t,
748         int amt)
749 {
750         struct tree_content *r = new_tree_content(t->entry_count + amt);
751         r->entry_count = t->entry_count;
752         r->delta_depth = t->delta_depth;
753         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
754         release_tree_content(t);
755         return r;
756 }
757
758 static struct tree_entry *new_tree_entry(void)
759 {
760         struct tree_entry *e;
761
762         if (!avail_tree_entry) {
763                 unsigned int n = tree_entry_alloc;
764                 total_allocd += n * sizeof(struct tree_entry);
765                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
766                 while (n-- > 1) {
767                         *((void**)e) = e + 1;
768                         e++;
769                 }
770                 *((void**)e) = NULL;
771         }
772
773         e = avail_tree_entry;
774         avail_tree_entry = *((void**)e);
775         return e;
776 }
777
778 static void release_tree_entry(struct tree_entry *e)
779 {
780         if (e->tree)
781                 release_tree_content_recursive(e->tree);
782         *((void**)e) = avail_tree_entry;
783         avail_tree_entry = e;
784 }
785
786 static struct tree_content *dup_tree_content(struct tree_content *s)
787 {
788         struct tree_content *d;
789         struct tree_entry *a, *b;
790         unsigned int i;
791
792         if (!s)
793                 return NULL;
794         d = new_tree_content(s->entry_count);
795         for (i = 0; i < s->entry_count; i++) {
796                 a = s->entries[i];
797                 b = new_tree_entry();
798                 memcpy(b, a, sizeof(*a));
799                 if (a->tree && is_null_sha1(b->versions[1].sha1))
800                         b->tree = dup_tree_content(a->tree);
801                 else
802                         b->tree = NULL;
803                 d->entries[i] = b;
804         }
805         d->entry_count = s->entry_count;
806         d->delta_depth = s->delta_depth;
807
808         return d;
809 }
810
811 static void start_packfile(void)
812 {
813         static char tmpfile[PATH_MAX];
814         struct packed_git *p;
815         struct pack_header hdr;
816         int pack_fd;
817
818         snprintf(tmpfile, sizeof(tmpfile),
819                 "%s/pack/tmp_pack_XXXXXX", get_object_directory());
820         pack_fd = xmkstemp(tmpfile);
821         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
822         strcpy(p->pack_name, tmpfile);
823         p->pack_fd = pack_fd;
824
825         hdr.hdr_signature = htonl(PACK_SIGNATURE);
826         hdr.hdr_version = htonl(2);
827         hdr.hdr_entries = 0;
828         write_or_die(p->pack_fd, &hdr, sizeof(hdr));
829
830         pack_data = p;
831         pack_size = sizeof(hdr);
832         object_count = 0;
833
834         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
835         all_packs[pack_id] = p;
836 }
837
838 static int oecmp (const void *a_, const void *b_)
839 {
840         struct object_entry *a = *((struct object_entry**)a_);
841         struct object_entry *b = *((struct object_entry**)b_);
842         return hashcmp(a->sha1, b->sha1);
843 }
844
845 static char *create_index(void)
846 {
847         static char tmpfile[PATH_MAX];
848         SHA_CTX ctx;
849         struct sha1file *f;
850         struct object_entry **idx, **c, **last, *e;
851         struct object_entry_pool *o;
852         uint32_t array[256];
853         int i, idx_fd;
854
855         /* Build the sorted table of object IDs. */
856         idx = xmalloc(object_count * sizeof(struct object_entry*));
857         c = idx;
858         for (o = blocks; o; o = o->next_pool)
859                 for (e = o->next_free; e-- != o->entries;)
860                         if (pack_id == e->pack_id)
861                                 *c++ = e;
862         last = idx + object_count;
863         if (c != last)
864                 die("internal consistency error creating the index");
865         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
866
867         /* Generate the fan-out array. */
868         c = idx;
869         for (i = 0; i < 256; i++) {
870                 struct object_entry **next = c;;
871                 while (next < last) {
872                         if ((*next)->sha1[0] != i)
873                                 break;
874                         next++;
875                 }
876                 array[i] = htonl(next - idx);
877                 c = next;
878         }
879
880         snprintf(tmpfile, sizeof(tmpfile),
881                 "%s/pack/tmp_idx_XXXXXX", get_object_directory());
882         idx_fd = xmkstemp(tmpfile);
883         f = sha1fd(idx_fd, tmpfile);
884         sha1write(f, array, 256 * sizeof(int));
885         SHA1_Init(&ctx);
886         for (c = idx; c != last; c++) {
887                 uint32_t offset = htonl((*c)->offset);
888                 sha1write(f, &offset, 4);
889                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
890                 SHA1_Update(&ctx, (*c)->sha1, 20);
891         }
892         sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
893         sha1close(f, NULL, CSUM_FSYNC);
894         free(idx);
895         SHA1_Final(pack_data->sha1, &ctx);
896         return tmpfile;
897 }
898
899 static char *keep_pack(char *curr_index_name)
900 {
901         static char name[PATH_MAX];
902         static const char *keep_msg = "fast-import";
903         int keep_fd;
904
905         chmod(pack_data->pack_name, 0444);
906         chmod(curr_index_name, 0444);
907
908         snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
909                  get_object_directory(), sha1_to_hex(pack_data->sha1));
910         keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
911         if (keep_fd < 0)
912                 die("cannot create keep file");
913         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
914         if (close(keep_fd))
915                 die("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(name);
939         }
940 }
941
942 static void end_packfile(void)
943 {
944         struct packed_git *old_p = pack_data, *new_p;
945
946         if (object_count) {
947                 char *idx_name;
948                 int i;
949                 struct branch *b;
950                 struct tag *t;
951
952                 close_pack_windows(pack_data);
953                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
954                                     pack_data->pack_name, object_count,
955                                     NULL, 0);
956                 close(pack_data->pack_fd);
957                 idx_name = keep_pack(create_index());
958
959                 /* Register the packfile with core git's machinary. */
960                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
961                 if (!new_p)
962                         die("core git rejected index %s", idx_name);
963                 all_packs[pack_id] = new_p;
964                 install_packed_git(new_p);
965
966                 /* Print the boundary */
967                 if (pack_edges) {
968                         fprintf(pack_edges, "%s:", new_p->pack_name);
969                         for (i = 0; i < branch_table_sz; i++) {
970                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
971                                         if (b->pack_id == pack_id)
972                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
973                                 }
974                         }
975                         for (t = first_tag; t; t = t->next_tag) {
976                                 if (t->pack_id == pack_id)
977                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
978                         }
979                         fputc('\n', pack_edges);
980                         fflush(pack_edges);
981                 }
982
983                 pack_id++;
984         }
985         else {
986                 close(old_p->pack_fd);
987                 unlink(old_p->pack_name);
988         }
989         free(old_p);
990
991         /* We can't carry a delta across packfiles. */
992         strbuf_release(&last_blob.data);
993         last_blob.offset = 0;
994         last_blob.depth = 0;
995 }
996
997 static void cycle_packfile(void)
998 {
999         end_packfile();
1000         start_packfile();
1001 }
1002
1003 static size_t encode_header(
1004         enum object_type type,
1005         size_t size,
1006         unsigned char *hdr)
1007 {
1008         int n = 1;
1009         unsigned char c;
1010
1011         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
1012                 die("bad type %d", type);
1013
1014         c = (type << 4) | (size & 15);
1015         size >>= 4;
1016         while (size) {
1017                 *hdr++ = c | 0x80;
1018                 c = size & 0x7f;
1019                 size >>= 7;
1020                 n++;
1021         }
1022         *hdr = c;
1023         return n;
1024 }
1025
1026 static int store_object(
1027         enum object_type type,
1028         struct strbuf *dat,
1029         struct last_object *last,
1030         unsigned char *sha1out,
1031         uintmax_t mark)
1032 {
1033         void *out, *delta;
1034         struct object_entry *e;
1035         unsigned char hdr[96];
1036         unsigned char sha1[20];
1037         unsigned long hdrlen, deltalen;
1038         SHA_CTX c;
1039         z_stream s;
1040
1041         hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1042                 (unsigned long)dat->len) + 1;
1043         SHA1_Init(&c);
1044         SHA1_Update(&c, hdr, hdrlen);
1045         SHA1_Update(&c, dat->buf, dat->len);
1046         SHA1_Final(sha1, &c);
1047         if (sha1out)
1048                 hashcpy(sha1out, sha1);
1049
1050         e = insert_object(sha1);
1051         if (mark)
1052                 insert_mark(mark, e);
1053         if (e->offset) {
1054                 duplicate_count_by_type[type]++;
1055                 return 1;
1056         } else if (find_sha1_pack(sha1, packed_git)) {
1057                 e->type = type;
1058                 e->pack_id = MAX_PACK_ID;
1059                 e->offset = 1; /* just not zero! */
1060                 duplicate_count_by_type[type]++;
1061                 return 1;
1062         }
1063
1064         if (last && last->data.buf && last->depth < max_depth) {
1065                 delta = diff_delta(last->data.buf, last->data.len,
1066                         dat->buf, dat->len,
1067                         &deltalen, 0);
1068                 if (delta && deltalen >= dat->len) {
1069                         free(delta);
1070                         delta = NULL;
1071                 }
1072         } else
1073                 delta = NULL;
1074
1075         memset(&s, 0, sizeof(s));
1076         deflateInit(&s, pack_compression_level);
1077         if (delta) {
1078                 s.next_in = delta;
1079                 s.avail_in = deltalen;
1080         } else {
1081                 s.next_in = (void *)dat->buf;
1082                 s.avail_in = dat->len;
1083         }
1084         s.avail_out = deflateBound(&s, s.avail_in);
1085         s.next_out = out = xmalloc(s.avail_out);
1086         while (deflate(&s, Z_FINISH) == Z_OK)
1087                 /* nothing */;
1088         deflateEnd(&s);
1089
1090         /* Determine if we should auto-checkpoint. */
1091         if ((pack_size + 60 + s.total_out) > max_packsize
1092                 || (pack_size + 60 + s.total_out) < pack_size) {
1093
1094                 /* This new object needs to *not* have the current pack_id. */
1095                 e->pack_id = pack_id + 1;
1096                 cycle_packfile();
1097
1098                 /* We cannot carry a delta into the new pack. */
1099                 if (delta) {
1100                         free(delta);
1101                         delta = NULL;
1102
1103                         memset(&s, 0, sizeof(s));
1104                         deflateInit(&s, pack_compression_level);
1105                         s.next_in = (void *)dat->buf;
1106                         s.avail_in = dat->len;
1107                         s.avail_out = deflateBound(&s, s.avail_in);
1108                         s.next_out = out = xrealloc(out, s.avail_out);
1109                         while (deflate(&s, Z_FINISH) == Z_OK)
1110                                 /* nothing */;
1111                         deflateEnd(&s);
1112                 }
1113         }
1114
1115         e->type = type;
1116         e->pack_id = pack_id;
1117         e->offset = pack_size;
1118         object_count++;
1119         object_count_by_type[type]++;
1120
1121         if (delta) {
1122                 unsigned long ofs = e->offset - last->offset;
1123                 unsigned pos = sizeof(hdr) - 1;
1124
1125                 delta_count_by_type[type]++;
1126                 e->depth = last->depth + 1;
1127
1128                 hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1129                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1130                 pack_size += hdrlen;
1131
1132                 hdr[pos] = ofs & 127;
1133                 while (ofs >>= 7)
1134                         hdr[--pos] = 128 | (--ofs & 127);
1135                 write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1136                 pack_size += sizeof(hdr) - pos;
1137         } else {
1138                 e->depth = 0;
1139                 hdrlen = encode_header(type, dat->len, hdr);
1140                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1141                 pack_size += hdrlen;
1142         }
1143
1144         write_or_die(pack_data->pack_fd, out, s.total_out);
1145         pack_size += s.total_out;
1146
1147         free(out);
1148         free(delta);
1149         if (last) {
1150                 if (last->no_swap) {
1151                         last->data = *dat;
1152                 } else {
1153                         strbuf_swap(&last->data, dat);
1154                 }
1155                 last->offset = e->offset;
1156                 last->depth = e->depth;
1157         }
1158         return 0;
1159 }
1160
1161 /* All calls must be guarded by find_object() or find_mark() to
1162  * ensure the 'struct object_entry' passed was written by this
1163  * process instance.  We unpack the entry by the offset, avoiding
1164  * the need for the corresponding .idx file.  This unpacking rule
1165  * works because we only use OBJ_REF_DELTA within the packfiles
1166  * created by fast-import.
1167  *
1168  * oe must not be NULL.  Such an oe usually comes from giving
1169  * an unknown SHA-1 to find_object() or an undefined mark to
1170  * find_mark().  Callers must test for this condition and use
1171  * the standard read_sha1_file() when it happens.
1172  *
1173  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1174  * find_mark(), where the mark was reloaded from an existing marks
1175  * file and is referencing an object that this fast-import process
1176  * instance did not write out to a packfile.  Callers must test for
1177  * this condition and use read_sha1_file() instead.
1178  */
1179 static void *gfi_unpack_entry(
1180         struct object_entry *oe,
1181         unsigned long *sizep)
1182 {
1183         enum object_type type;
1184         struct packed_git *p = all_packs[oe->pack_id];
1185         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1186                 /* The object is stored in the packfile we are writing to
1187                  * and we have modified it since the last time we scanned
1188                  * back to read a previously written object.  If an old
1189                  * window covered [p->pack_size, p->pack_size + 20) its
1190                  * data is stale and is not valid.  Closing all windows
1191                  * and updating the packfile length ensures we can read
1192                  * the newly written data.
1193                  */
1194                 close_pack_windows(p);
1195
1196                 /* We have to offer 20 bytes additional on the end of
1197                  * the packfile as the core unpacker code assumes the
1198                  * footer is present at the file end and must promise
1199                  * at least 20 bytes within any window it maps.  But
1200                  * we don't actually create the footer here.
1201                  */
1202                 p->pack_size = pack_size + 20;
1203         }
1204         return unpack_entry(p, oe->offset, &type, sizep);
1205 }
1206
1207 static const char *get_mode(const char *str, uint16_t *modep)
1208 {
1209         unsigned char c;
1210         uint16_t mode = 0;
1211
1212         while ((c = *str++) != ' ') {
1213                 if (c < '0' || c > '7')
1214                         return NULL;
1215                 mode = (mode << 3) + (c - '0');
1216         }
1217         *modep = mode;
1218         return str;
1219 }
1220
1221 static void load_tree(struct tree_entry *root)
1222 {
1223         unsigned char* sha1 = root->versions[1].sha1;
1224         struct object_entry *myoe;
1225         struct tree_content *t;
1226         unsigned long size;
1227         char *buf;
1228         const char *c;
1229
1230         root->tree = t = new_tree_content(8);
1231         if (is_null_sha1(sha1))
1232                 return;
1233
1234         myoe = find_object(sha1);
1235         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1236                 if (myoe->type != OBJ_TREE)
1237                         die("Not a tree: %s", sha1_to_hex(sha1));
1238                 t->delta_depth = myoe->depth;
1239                 buf = gfi_unpack_entry(myoe, &size);
1240                 if (!buf)
1241                         die("Can't load tree %s", sha1_to_hex(sha1));
1242         } else {
1243                 enum object_type type;
1244                 buf = read_sha1_file(sha1, &type, &size);
1245                 if (!buf || type != OBJ_TREE)
1246                         die("Can't load tree %s", sha1_to_hex(sha1));
1247         }
1248
1249         c = buf;
1250         while (c != (buf + size)) {
1251                 struct tree_entry *e = new_tree_entry();
1252
1253                 if (t->entry_count == t->entry_capacity)
1254                         root->tree = t = grow_tree_content(t, t->entry_count);
1255                 t->entries[t->entry_count++] = e;
1256
1257                 e->tree = NULL;
1258                 c = get_mode(c, &e->versions[1].mode);
1259                 if (!c)
1260                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1261                 e->versions[0].mode = e->versions[1].mode;
1262                 e->name = to_atom(c, strlen(c));
1263                 c += e->name->str_len + 1;
1264                 hashcpy(e->versions[0].sha1, (unsigned char*)c);
1265                 hashcpy(e->versions[1].sha1, (unsigned char*)c);
1266                 c += 20;
1267         }
1268         free(buf);
1269 }
1270
1271 static int tecmp0 (const void *_a, const void *_b)
1272 {
1273         struct tree_entry *a = *((struct tree_entry**)_a);
1274         struct tree_entry *b = *((struct tree_entry**)_b);
1275         return base_name_compare(
1276                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1277                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1278 }
1279
1280 static int tecmp1 (const void *_a, const void *_b)
1281 {
1282         struct tree_entry *a = *((struct tree_entry**)_a);
1283         struct tree_entry *b = *((struct tree_entry**)_b);
1284         return base_name_compare(
1285                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1286                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1287 }
1288
1289 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1290 {
1291         size_t maxlen = 0;
1292         unsigned int i;
1293
1294         if (!v)
1295                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1296         else
1297                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1298
1299         for (i = 0; i < t->entry_count; i++) {
1300                 if (t->entries[i]->versions[v].mode)
1301                         maxlen += t->entries[i]->name->str_len + 34;
1302         }
1303
1304         strbuf_reset(b);
1305         strbuf_grow(b, maxlen);
1306         for (i = 0; i < t->entry_count; i++) {
1307                 struct tree_entry *e = t->entries[i];
1308                 if (!e->versions[v].mode)
1309                         continue;
1310                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1311                                         e->name->str_dat, '\0');
1312                 strbuf_add(b, e->versions[v].sha1, 20);
1313         }
1314 }
1315
1316 static void store_tree(struct tree_entry *root)
1317 {
1318         struct tree_content *t = root->tree;
1319         unsigned int i, j, del;
1320         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1321         struct object_entry *le;
1322
1323         if (!is_null_sha1(root->versions[1].sha1))
1324                 return;
1325
1326         for (i = 0; i < t->entry_count; i++) {
1327                 if (t->entries[i]->tree)
1328                         store_tree(t->entries[i]);
1329         }
1330
1331         le = find_object(root->versions[0].sha1);
1332         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1333                 mktree(t, 0, &old_tree);
1334                 lo.data = old_tree;
1335                 lo.offset = le->offset;
1336                 lo.depth = t->delta_depth;
1337         }
1338
1339         mktree(t, 1, &new_tree);
1340         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1341
1342         t->delta_depth = lo.depth;
1343         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1344                 struct tree_entry *e = t->entries[i];
1345                 if (e->versions[1].mode) {
1346                         e->versions[0].mode = e->versions[1].mode;
1347                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1348                         t->entries[j++] = e;
1349                 } else {
1350                         release_tree_entry(e);
1351                         del++;
1352                 }
1353         }
1354         t->entry_count -= del;
1355 }
1356
1357 static int tree_content_set(
1358         struct tree_entry *root,
1359         const char *p,
1360         const unsigned char *sha1,
1361         const uint16_t mode,
1362         struct tree_content *subtree)
1363 {
1364         struct tree_content *t = root->tree;
1365         const char *slash1;
1366         unsigned int i, n;
1367         struct tree_entry *e;
1368
1369         slash1 = strchr(p, '/');
1370         if (slash1)
1371                 n = slash1 - p;
1372         else
1373                 n = strlen(p);
1374         if (!n)
1375                 die("Empty path component found in input");
1376         if (!slash1 && !S_ISDIR(mode) && subtree)
1377                 die("Non-directories cannot have subtrees");
1378
1379         for (i = 0; i < t->entry_count; i++) {
1380                 e = t->entries[i];
1381                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1382                         if (!slash1) {
1383                                 if (!S_ISDIR(mode)
1384                                                 && e->versions[1].mode == mode
1385                                                 && !hashcmp(e->versions[1].sha1, sha1))
1386                                         return 0;
1387                                 e->versions[1].mode = mode;
1388                                 hashcpy(e->versions[1].sha1, sha1);
1389                                 if (e->tree)
1390                                         release_tree_content_recursive(e->tree);
1391                                 e->tree = subtree;
1392                                 hashclr(root->versions[1].sha1);
1393                                 return 1;
1394                         }
1395                         if (!S_ISDIR(e->versions[1].mode)) {
1396                                 e->tree = new_tree_content(8);
1397                                 e->versions[1].mode = S_IFDIR;
1398                         }
1399                         if (!e->tree)
1400                                 load_tree(e);
1401                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1402                                 hashclr(root->versions[1].sha1);
1403                                 return 1;
1404                         }
1405                         return 0;
1406                 }
1407         }
1408
1409         if (t->entry_count == t->entry_capacity)
1410                 root->tree = t = grow_tree_content(t, t->entry_count);
1411         e = new_tree_entry();
1412         e->name = to_atom(p, n);
1413         e->versions[0].mode = 0;
1414         hashclr(e->versions[0].sha1);
1415         t->entries[t->entry_count++] = e;
1416         if (slash1) {
1417                 e->tree = new_tree_content(8);
1418                 e->versions[1].mode = S_IFDIR;
1419                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1420         } else {
1421                 e->tree = subtree;
1422                 e->versions[1].mode = mode;
1423                 hashcpy(e->versions[1].sha1, sha1);
1424         }
1425         hashclr(root->versions[1].sha1);
1426         return 1;
1427 }
1428
1429 static int tree_content_remove(
1430         struct tree_entry *root,
1431         const char *p,
1432         struct tree_entry *backup_leaf)
1433 {
1434         struct tree_content *t = root->tree;
1435         const char *slash1;
1436         unsigned int i, n;
1437         struct tree_entry *e;
1438
1439         slash1 = strchr(p, '/');
1440         if (slash1)
1441                 n = slash1 - p;
1442         else
1443                 n = strlen(p);
1444
1445         for (i = 0; i < t->entry_count; i++) {
1446                 e = t->entries[i];
1447                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1448                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1449                                 goto del_entry;
1450                         if (!e->tree)
1451                                 load_tree(e);
1452                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1453                                 for (n = 0; n < e->tree->entry_count; n++) {
1454                                         if (e->tree->entries[n]->versions[1].mode) {
1455                                                 hashclr(root->versions[1].sha1);
1456                                                 return 1;
1457                                         }
1458                                 }
1459                                 backup_leaf = NULL;
1460                                 goto del_entry;
1461                         }
1462                         return 0;
1463                 }
1464         }
1465         return 0;
1466
1467 del_entry:
1468         if (backup_leaf)
1469                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1470         else if (e->tree)
1471                 release_tree_content_recursive(e->tree);
1472         e->tree = NULL;
1473         e->versions[1].mode = 0;
1474         hashclr(e->versions[1].sha1);
1475         hashclr(root->versions[1].sha1);
1476         return 1;
1477 }
1478
1479 static int tree_content_get(
1480         struct tree_entry *root,
1481         const char *p,
1482         struct tree_entry *leaf)
1483 {
1484         struct tree_content *t = root->tree;
1485         const char *slash1;
1486         unsigned int i, n;
1487         struct tree_entry *e;
1488
1489         slash1 = strchr(p, '/');
1490         if (slash1)
1491                 n = slash1 - p;
1492         else
1493                 n = strlen(p);
1494
1495         for (i = 0; i < t->entry_count; i++) {
1496                 e = t->entries[i];
1497                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1498                         if (!slash1) {
1499                                 memcpy(leaf, e, sizeof(*leaf));
1500                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1501                                         leaf->tree = dup_tree_content(e->tree);
1502                                 else
1503                                         leaf->tree = NULL;
1504                                 return 1;
1505                         }
1506                         if (!S_ISDIR(e->versions[1].mode))
1507                                 return 0;
1508                         if (!e->tree)
1509                                 load_tree(e);
1510                         return tree_content_get(e, slash1 + 1, leaf);
1511                 }
1512         }
1513         return 0;
1514 }
1515
1516 static int update_branch(struct branch *b)
1517 {
1518         static const char *msg = "fast-import";
1519         struct ref_lock *lock;
1520         unsigned char old_sha1[20];
1521
1522         if (is_null_sha1(b->sha1))
1523                 return 0;
1524         if (read_ref(b->name, old_sha1))
1525                 hashclr(old_sha1);
1526         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1527         if (!lock)
1528                 return error("Unable to lock %s", b->name);
1529         if (!force_update && !is_null_sha1(old_sha1)) {
1530                 struct commit *old_cmit, *new_cmit;
1531
1532                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1533                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1534                 if (!old_cmit || !new_cmit) {
1535                         unlock_ref(lock);
1536                         return error("Branch %s is missing commits.", b->name);
1537                 }
1538
1539                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1540                         unlock_ref(lock);
1541                         warning("Not updating %s"
1542                                 " (new tip %s does not contain %s)",
1543                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1544                         return -1;
1545                 }
1546         }
1547         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1548                 return error("Unable to update %s", b->name);
1549         return 0;
1550 }
1551
1552 static void dump_branches(void)
1553 {
1554         unsigned int i;
1555         struct branch *b;
1556
1557         for (i = 0; i < branch_table_sz; i++) {
1558                 for (b = branch_table[i]; b; b = b->table_next_branch)
1559                         failure |= update_branch(b);
1560         }
1561 }
1562
1563 static void dump_tags(void)
1564 {
1565         static const char *msg = "fast-import";
1566         struct tag *t;
1567         struct ref_lock *lock;
1568         char ref_name[PATH_MAX];
1569
1570         for (t = first_tag; t; t = t->next_tag) {
1571                 sprintf(ref_name, "tags/%s", t->name);
1572                 lock = lock_ref_sha1(ref_name, NULL);
1573                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1574                         failure |= error("Unable to update %s", ref_name);
1575         }
1576 }
1577
1578 static void dump_marks_helper(FILE *f,
1579         uintmax_t base,
1580         struct mark_set *m)
1581 {
1582         uintmax_t k;
1583         if (m->shift) {
1584                 for (k = 0; k < 1024; k++) {
1585                         if (m->data.sets[k])
1586                                 dump_marks_helper(f, (base + k) << m->shift,
1587                                         m->data.sets[k]);
1588                 }
1589         } else {
1590                 for (k = 0; k < 1024; k++) {
1591                         if (m->data.marked[k])
1592                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1593                                         sha1_to_hex(m->data.marked[k]->sha1));
1594                 }
1595         }
1596 }
1597
1598 static void dump_marks(void)
1599 {
1600         static struct lock_file mark_lock;
1601         int mark_fd;
1602         FILE *f;
1603
1604         if (!mark_file)
1605                 return;
1606
1607         mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1608         if (mark_fd < 0) {
1609                 failure |= error("Unable to write marks file %s: %s",
1610                         mark_file, strerror(errno));
1611                 return;
1612         }
1613
1614         f = fdopen(mark_fd, "w");
1615         if (!f) {
1616                 int saved_errno = errno;
1617                 rollback_lock_file(&mark_lock);
1618                 failure |= error("Unable to write marks file %s: %s",
1619                         mark_file, strerror(saved_errno));
1620                 return;
1621         }
1622
1623         /*
1624          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1625          * Assign -1 to the lock file descriptor so that commit_lock_file()
1626          * won't try to close() it.
1627          */
1628         mark_lock.fd = -1;
1629
1630         dump_marks_helper(f, 0, marks);
1631         if (ferror(f) || fclose(f)) {
1632                 int saved_errno = errno;
1633                 rollback_lock_file(&mark_lock);
1634                 failure |= error("Unable to write marks file %s: %s",
1635                         mark_file, strerror(saved_errno));
1636                 return;
1637         }
1638
1639         if (commit_lock_file(&mark_lock)) {
1640                 int saved_errno = errno;
1641                 rollback_lock_file(&mark_lock);
1642                 failure |= error("Unable to commit marks file %s: %s",
1643                         mark_file, strerror(saved_errno));
1644                 return;
1645         }
1646 }
1647
1648 static int read_next_command(void)
1649 {
1650         static int stdin_eof = 0;
1651
1652         if (stdin_eof) {
1653                 unread_command_buf = 0;
1654                 return EOF;
1655         }
1656
1657         do {
1658                 if (unread_command_buf) {
1659                         unread_command_buf = 0;
1660                 } else {
1661                         struct recent_command *rc;
1662
1663                         strbuf_detach(&command_buf, NULL);
1664                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1665                         if (stdin_eof)
1666                                 return EOF;
1667
1668                         rc = rc_free;
1669                         if (rc)
1670                                 rc_free = rc->next;
1671                         else {
1672                                 rc = cmd_hist.next;
1673                                 cmd_hist.next = rc->next;
1674                                 cmd_hist.next->prev = &cmd_hist;
1675                                 free(rc->buf);
1676                         }
1677
1678                         rc->buf = command_buf.buf;
1679                         rc->prev = cmd_tail;
1680                         rc->next = cmd_hist.prev;
1681                         rc->prev->next = rc;
1682                         cmd_tail = rc;
1683                 }
1684         } while (command_buf.buf[0] == '#');
1685
1686         return 0;
1687 }
1688
1689 static void skip_optional_lf(void)
1690 {
1691         int term_char = fgetc(stdin);
1692         if (term_char != '\n' && term_char != EOF)
1693                 ungetc(term_char, stdin);
1694 }
1695
1696 static void parse_mark(void)
1697 {
1698         if (!prefixcmp(command_buf.buf, "mark :")) {
1699                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1700                 read_next_command();
1701         }
1702         else
1703                 next_mark = 0;
1704 }
1705
1706 static void parse_data(struct strbuf *sb)
1707 {
1708         strbuf_reset(sb);
1709
1710         if (prefixcmp(command_buf.buf, "data "))
1711                 die("Expected 'data n' command, found: %s", command_buf.buf);
1712
1713         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1714                 char *term = xstrdup(command_buf.buf + 5 + 2);
1715                 size_t term_len = command_buf.len - 5 - 2;
1716
1717                 strbuf_detach(&command_buf, NULL);
1718                 for (;;) {
1719                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1720                                 die("EOF in data (terminator '%s' not found)", term);
1721                         if (term_len == command_buf.len
1722                                 && !strcmp(term, command_buf.buf))
1723                                 break;
1724                         strbuf_addbuf(sb, &command_buf);
1725                         strbuf_addch(sb, '\n');
1726                 }
1727                 free(term);
1728         }
1729         else {
1730                 size_t n = 0, length;
1731
1732                 length = strtoul(command_buf.buf + 5, NULL, 10);
1733
1734                 while (n < length) {
1735                         size_t s = strbuf_fread(sb, length - n, stdin);
1736                         if (!s && feof(stdin))
1737                                 die("EOF in data (%lu bytes remaining)",
1738                                         (unsigned long)(length - n));
1739                         n += s;
1740                 }
1741         }
1742
1743         skip_optional_lf();
1744 }
1745
1746 static int validate_raw_date(const char *src, char *result, int maxlen)
1747 {
1748         const char *orig_src = src;
1749         char *endp, sign;
1750
1751         strtoul(src, &endp, 10);
1752         if (endp == src || *endp != ' ')
1753                 return -1;
1754
1755         src = endp + 1;
1756         if (*src != '-' && *src != '+')
1757                 return -1;
1758         sign = *src;
1759
1760         strtoul(src + 1, &endp, 10);
1761         if (endp == src || *endp || (endp - orig_src) >= maxlen)
1762                 return -1;
1763
1764         strcpy(result, orig_src);
1765         return 0;
1766 }
1767
1768 static char *parse_ident(const char *buf)
1769 {
1770         const char *gt;
1771         size_t name_len;
1772         char *ident;
1773
1774         gt = strrchr(buf, '>');
1775         if (!gt)
1776                 die("Missing > in ident string: %s", buf);
1777         gt++;
1778         if (*gt != ' ')
1779                 die("Missing space after > in ident string: %s", buf);
1780         gt++;
1781         name_len = gt - buf;
1782         ident = xmalloc(name_len + 24);
1783         strncpy(ident, buf, name_len);
1784
1785         switch (whenspec) {
1786         case WHENSPEC_RAW:
1787                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1788                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1789                 break;
1790         case WHENSPEC_RFC2822:
1791                 if (parse_date(gt, ident + name_len, 24) < 0)
1792                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1793                 break;
1794         case WHENSPEC_NOW:
1795                 if (strcmp("now", gt))
1796                         die("Date in ident must be 'now': %s", buf);
1797                 datestamp(ident + name_len, 24);
1798                 break;
1799         }
1800
1801         return ident;
1802 }
1803
1804 static void parse_new_blob(void)
1805 {
1806         static struct strbuf buf = STRBUF_INIT;
1807
1808         read_next_command();
1809         parse_mark();
1810         parse_data(&buf);
1811         store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1812 }
1813
1814 static void unload_one_branch(void)
1815 {
1816         while (cur_active_branches
1817                 && cur_active_branches >= max_active_branches) {
1818                 uintmax_t min_commit = ULONG_MAX;
1819                 struct branch *e, *l = NULL, *p = NULL;
1820
1821                 for (e = active_branches; e; e = e->active_next_branch) {
1822                         if (e->last_commit < min_commit) {
1823                                 p = l;
1824                                 min_commit = e->last_commit;
1825                         }
1826                         l = e;
1827                 }
1828
1829                 if (p) {
1830                         e = p->active_next_branch;
1831                         p->active_next_branch = e->active_next_branch;
1832                 } else {
1833                         e = active_branches;
1834                         active_branches = e->active_next_branch;
1835                 }
1836                 e->active = 0;
1837                 e->active_next_branch = NULL;
1838                 if (e->branch_tree.tree) {
1839                         release_tree_content_recursive(e->branch_tree.tree);
1840                         e->branch_tree.tree = NULL;
1841                 }
1842                 cur_active_branches--;
1843         }
1844 }
1845
1846 static void load_branch(struct branch *b)
1847 {
1848         load_tree(&b->branch_tree);
1849         if (!b->active) {
1850                 b->active = 1;
1851                 b->active_next_branch = active_branches;
1852                 active_branches = b;
1853                 cur_active_branches++;
1854                 branch_load_count++;
1855         }
1856 }
1857
1858 static void file_change_m(struct branch *b)
1859 {
1860         const char *p = command_buf.buf + 2;
1861         static struct strbuf uq = STRBUF_INIT;
1862         const char *endp;
1863         struct object_entry *oe = oe;
1864         unsigned char sha1[20];
1865         uint16_t mode, inline_data = 0;
1866
1867         p = get_mode(p, &mode);
1868         if (!p)
1869                 die("Corrupt mode: %s", command_buf.buf);
1870         switch (mode) {
1871         case S_IFREG | 0644:
1872         case S_IFREG | 0755:
1873         case S_IFLNK:
1874         case S_IFGITLINK:
1875         case 0644:
1876         case 0755:
1877                 /* ok */
1878                 break;
1879         default:
1880                 die("Corrupt mode: %s", command_buf.buf);
1881         }
1882
1883         if (*p == ':') {
1884                 char *x;
1885                 oe = find_mark(strtoumax(p + 1, &x, 10));
1886                 hashcpy(sha1, oe->sha1);
1887                 p = x;
1888         } else if (!prefixcmp(p, "inline")) {
1889                 inline_data = 1;
1890                 p += 6;
1891         } else {
1892                 if (get_sha1_hex(p, sha1))
1893                         die("Invalid SHA1: %s", command_buf.buf);
1894                 oe = find_object(sha1);
1895                 p += 40;
1896         }
1897         if (*p++ != ' ')
1898                 die("Missing space after SHA1: %s", command_buf.buf);
1899
1900         strbuf_reset(&uq);
1901         if (!unquote_c_style(&uq, p, &endp)) {
1902                 if (*endp)
1903                         die("Garbage after path in: %s", command_buf.buf);
1904                 p = uq.buf;
1905         }
1906
1907         if (S_ISGITLINK(mode)) {
1908                 if (inline_data)
1909                         die("Git links cannot be specified 'inline': %s",
1910                                 command_buf.buf);
1911                 else if (oe) {
1912                         if (oe->type != OBJ_COMMIT)
1913                                 die("Not a commit (actually a %s): %s",
1914                                         typename(oe->type), command_buf.buf);
1915                 }
1916                 /*
1917                  * Accept the sha1 without checking; it expected to be in
1918                  * another repository.
1919                  */
1920         } else if (inline_data) {
1921                 static struct strbuf buf = STRBUF_INIT;
1922
1923                 if (p != uq.buf) {
1924                         strbuf_addstr(&uq, p);
1925                         p = uq.buf;
1926                 }
1927                 read_next_command();
1928                 parse_data(&buf);
1929                 store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1930         } else if (oe) {
1931                 if (oe->type != OBJ_BLOB)
1932                         die("Not a blob (actually a %s): %s",
1933                                 typename(oe->type), command_buf.buf);
1934         } else {
1935                 enum object_type type = sha1_object_info(sha1, NULL);
1936                 if (type < 0)
1937                         die("Blob not found: %s", command_buf.buf);
1938                 if (type != OBJ_BLOB)
1939                         die("Not a blob (actually a %s): %s",
1940                             typename(type), command_buf.buf);
1941         }
1942
1943         tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode, NULL);
1944 }
1945
1946 static void file_change_d(struct branch *b)
1947 {
1948         const char *p = command_buf.buf + 2;
1949         static struct strbuf uq = STRBUF_INIT;
1950         const char *endp;
1951
1952         strbuf_reset(&uq);
1953         if (!unquote_c_style(&uq, p, &endp)) {
1954                 if (*endp)
1955                         die("Garbage after path in: %s", command_buf.buf);
1956                 p = uq.buf;
1957         }
1958         tree_content_remove(&b->branch_tree, p, NULL);
1959 }
1960
1961 static void file_change_cr(struct branch *b, int rename)
1962 {
1963         const char *s, *d;
1964         static struct strbuf s_uq = STRBUF_INIT;
1965         static struct strbuf d_uq = STRBUF_INIT;
1966         const char *endp;
1967         struct tree_entry leaf;
1968
1969         s = command_buf.buf + 2;
1970         strbuf_reset(&s_uq);
1971         if (!unquote_c_style(&s_uq, s, &endp)) {
1972                 if (*endp != ' ')
1973                         die("Missing space after source: %s", command_buf.buf);
1974         } else {
1975                 endp = strchr(s, ' ');
1976                 if (!endp)
1977                         die("Missing space after source: %s", command_buf.buf);
1978                 strbuf_add(&s_uq, s, endp - s);
1979         }
1980         s = s_uq.buf;
1981
1982         endp++;
1983         if (!*endp)
1984                 die("Missing dest: %s", command_buf.buf);
1985
1986         d = endp;
1987         strbuf_reset(&d_uq);
1988         if (!unquote_c_style(&d_uq, d, &endp)) {
1989                 if (*endp)
1990                         die("Garbage after dest in: %s", command_buf.buf);
1991                 d = d_uq.buf;
1992         }
1993
1994         memset(&leaf, 0, sizeof(leaf));
1995         if (rename)
1996                 tree_content_remove(&b->branch_tree, s, &leaf);
1997         else
1998                 tree_content_get(&b->branch_tree, s, &leaf);
1999         if (!leaf.versions[1].mode)
2000                 die("Path %s not in branch", s);
2001         tree_content_set(&b->branch_tree, d,
2002                 leaf.versions[1].sha1,
2003                 leaf.versions[1].mode,
2004                 leaf.tree);
2005 }
2006
2007 static void file_change_deleteall(struct branch *b)
2008 {
2009         release_tree_content_recursive(b->branch_tree.tree);
2010         hashclr(b->branch_tree.versions[0].sha1);
2011         hashclr(b->branch_tree.versions[1].sha1);
2012         load_tree(&b->branch_tree);
2013 }
2014
2015 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2016 {
2017         if (!buf || size < 46)
2018                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2019         if (memcmp("tree ", buf, 5)
2020                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2021                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2022         hashcpy(b->branch_tree.versions[0].sha1,
2023                 b->branch_tree.versions[1].sha1);
2024 }
2025
2026 static void parse_from_existing(struct branch *b)
2027 {
2028         if (is_null_sha1(b->sha1)) {
2029                 hashclr(b->branch_tree.versions[0].sha1);
2030                 hashclr(b->branch_tree.versions[1].sha1);
2031         } else {
2032                 unsigned long size;
2033                 char *buf;
2034
2035                 buf = read_object_with_reference(b->sha1,
2036                         commit_type, &size, b->sha1);
2037                 parse_from_commit(b, buf, size);
2038                 free(buf);
2039         }
2040 }
2041
2042 static int parse_from(struct branch *b)
2043 {
2044         const char *from;
2045         struct branch *s;
2046
2047         if (prefixcmp(command_buf.buf, "from "))
2048                 return 0;
2049
2050         if (b->branch_tree.tree) {
2051                 release_tree_content_recursive(b->branch_tree.tree);
2052                 b->branch_tree.tree = NULL;
2053         }
2054
2055         from = strchr(command_buf.buf, ' ') + 1;
2056         s = lookup_branch(from);
2057         if (b == s)
2058                 die("Can't create a branch from itself: %s", b->name);
2059         else if (s) {
2060                 unsigned char *t = s->branch_tree.versions[1].sha1;
2061                 hashcpy(b->sha1, s->sha1);
2062                 hashcpy(b->branch_tree.versions[0].sha1, t);
2063                 hashcpy(b->branch_tree.versions[1].sha1, t);
2064         } else if (*from == ':') {
2065                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2066                 struct object_entry *oe = find_mark(idnum);
2067                 if (oe->type != OBJ_COMMIT)
2068                         die("Mark :%" PRIuMAX " not a commit", idnum);
2069                 hashcpy(b->sha1, oe->sha1);
2070                 if (oe->pack_id != MAX_PACK_ID) {
2071                         unsigned long size;
2072                         char *buf = gfi_unpack_entry(oe, &size);
2073                         parse_from_commit(b, buf, size);
2074                         free(buf);
2075                 } else
2076                         parse_from_existing(b);
2077         } else if (!get_sha1(from, b->sha1))
2078                 parse_from_existing(b);
2079         else
2080                 die("Invalid ref name or SHA1 expression: %s", from);
2081
2082         read_next_command();
2083         return 1;
2084 }
2085
2086 static struct hash_list *parse_merge(unsigned int *count)
2087 {
2088         struct hash_list *list = NULL, *n, *e = e;
2089         const char *from;
2090         struct branch *s;
2091
2092         *count = 0;
2093         while (!prefixcmp(command_buf.buf, "merge ")) {
2094                 from = strchr(command_buf.buf, ' ') + 1;
2095                 n = xmalloc(sizeof(*n));
2096                 s = lookup_branch(from);
2097                 if (s)
2098                         hashcpy(n->sha1, s->sha1);
2099                 else if (*from == ':') {
2100                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2101                         struct object_entry *oe = find_mark(idnum);
2102                         if (oe->type != OBJ_COMMIT)
2103                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2104                         hashcpy(n->sha1, oe->sha1);
2105                 } else if (!get_sha1(from, n->sha1)) {
2106                         unsigned long size;
2107                         char *buf = read_object_with_reference(n->sha1,
2108                                 commit_type, &size, n->sha1);
2109                         if (!buf || size < 46)
2110                                 die("Not a valid commit: %s", from);
2111                         free(buf);
2112                 } else
2113                         die("Invalid ref name or SHA1 expression: %s", from);
2114
2115                 n->next = NULL;
2116                 if (list)
2117                         e->next = n;
2118                 else
2119                         list = n;
2120                 e = n;
2121                 (*count)++;
2122                 read_next_command();
2123         }
2124         return list;
2125 }
2126
2127 static void parse_new_commit(void)
2128 {
2129         static struct strbuf msg = STRBUF_INIT;
2130         struct branch *b;
2131         char *sp;
2132         char *author = NULL;
2133         char *committer = NULL;
2134         struct hash_list *merge_list = NULL;
2135         unsigned int merge_count;
2136
2137         /* Obtain the branch name from the rest of our command */
2138         sp = strchr(command_buf.buf, ' ') + 1;
2139         b = lookup_branch(sp);
2140         if (!b)
2141                 b = new_branch(sp);
2142
2143         read_next_command();
2144         parse_mark();
2145         if (!prefixcmp(command_buf.buf, "author ")) {
2146                 author = parse_ident(command_buf.buf + 7);
2147                 read_next_command();
2148         }
2149         if (!prefixcmp(command_buf.buf, "committer ")) {
2150                 committer = parse_ident(command_buf.buf + 10);
2151                 read_next_command();
2152         }
2153         if (!committer)
2154                 die("Expected committer but didn't get one");
2155         parse_data(&msg);
2156         read_next_command();
2157         parse_from(b);
2158         merge_list = parse_merge(&merge_count);
2159
2160         /* ensure the branch is active/loaded */
2161         if (!b->branch_tree.tree || !max_active_branches) {
2162                 unload_one_branch();
2163                 load_branch(b);
2164         }
2165
2166         /* file_change* */
2167         while (command_buf.len > 0) {
2168                 if (!prefixcmp(command_buf.buf, "M "))
2169                         file_change_m(b);
2170                 else if (!prefixcmp(command_buf.buf, "D "))
2171                         file_change_d(b);
2172                 else if (!prefixcmp(command_buf.buf, "R "))
2173                         file_change_cr(b, 1);
2174                 else if (!prefixcmp(command_buf.buf, "C "))
2175                         file_change_cr(b, 0);
2176                 else if (!strcmp("deleteall", command_buf.buf))
2177                         file_change_deleteall(b);
2178                 else {
2179                         unread_command_buf = 1;
2180                         break;
2181                 }
2182                 if (read_next_command() == EOF)
2183                         break;
2184         }
2185
2186         /* build the tree and the commit */
2187         store_tree(&b->branch_tree);
2188         hashcpy(b->branch_tree.versions[0].sha1,
2189                 b->branch_tree.versions[1].sha1);
2190
2191         strbuf_reset(&new_data);
2192         strbuf_addf(&new_data, "tree %s\n",
2193                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2194         if (!is_null_sha1(b->sha1))
2195                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2196         while (merge_list) {
2197                 struct hash_list *next = merge_list->next;
2198                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2199                 free(merge_list);
2200                 merge_list = next;
2201         }
2202         strbuf_addf(&new_data,
2203                 "author %s\n"
2204                 "committer %s\n"
2205                 "\n",
2206                 author ? author : committer, committer);
2207         strbuf_addbuf(&new_data, &msg);
2208         free(author);
2209         free(committer);
2210
2211         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2212                 b->pack_id = pack_id;
2213         b->last_commit = object_count_by_type[OBJ_COMMIT];
2214 }
2215
2216 static void parse_new_tag(void)
2217 {
2218         static struct strbuf msg = STRBUF_INIT;
2219         char *sp;
2220         const char *from;
2221         char *tagger;
2222         struct branch *s;
2223         struct tag *t;
2224         uintmax_t from_mark = 0;
2225         unsigned char sha1[20];
2226
2227         /* Obtain the new tag name from the rest of our command */
2228         sp = strchr(command_buf.buf, ' ') + 1;
2229         t = pool_alloc(sizeof(struct tag));
2230         t->next_tag = NULL;
2231         t->name = pool_strdup(sp);
2232         if (last_tag)
2233                 last_tag->next_tag = t;
2234         else
2235                 first_tag = t;
2236         last_tag = t;
2237         read_next_command();
2238
2239         /* from ... */
2240         if (prefixcmp(command_buf.buf, "from "))
2241                 die("Expected from command, got %s", command_buf.buf);
2242         from = strchr(command_buf.buf, ' ') + 1;
2243         s = lookup_branch(from);
2244         if (s) {
2245                 hashcpy(sha1, s->sha1);
2246         } else if (*from == ':') {
2247                 struct object_entry *oe;
2248                 from_mark = strtoumax(from + 1, NULL, 10);
2249                 oe = find_mark(from_mark);
2250                 if (oe->type != OBJ_COMMIT)
2251                         die("Mark :%" PRIuMAX " not a commit", from_mark);
2252                 hashcpy(sha1, oe->sha1);
2253         } else if (!get_sha1(from, sha1)) {
2254                 unsigned long size;
2255                 char *buf;
2256
2257                 buf = read_object_with_reference(sha1,
2258                         commit_type, &size, sha1);
2259                 if (!buf || size < 46)
2260                         die("Not a valid commit: %s", from);
2261                 free(buf);
2262         } else
2263                 die("Invalid ref name or SHA1 expression: %s", from);
2264         read_next_command();
2265
2266         /* tagger ... */
2267         if (prefixcmp(command_buf.buf, "tagger "))
2268                 die("Expected tagger command, got %s", command_buf.buf);
2269         tagger = parse_ident(command_buf.buf + 7);
2270
2271         /* tag payload/message */
2272         read_next_command();
2273         parse_data(&msg);
2274
2275         /* build the tag object */
2276         strbuf_reset(&new_data);
2277         strbuf_addf(&new_data,
2278                 "object %s\n"
2279                 "type %s\n"
2280                 "tag %s\n"
2281                 "tagger %s\n"
2282                 "\n",
2283                 sha1_to_hex(sha1), commit_type, t->name, tagger);
2284         strbuf_addbuf(&new_data, &msg);
2285         free(tagger);
2286
2287         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2288                 t->pack_id = MAX_PACK_ID;
2289         else
2290                 t->pack_id = pack_id;
2291 }
2292
2293 static void parse_reset_branch(void)
2294 {
2295         struct branch *b;
2296         char *sp;
2297
2298         /* Obtain the branch name from the rest of our command */
2299         sp = strchr(command_buf.buf, ' ') + 1;
2300         b = lookup_branch(sp);
2301         if (b) {
2302                 hashclr(b->sha1);
2303                 hashclr(b->branch_tree.versions[0].sha1);
2304                 hashclr(b->branch_tree.versions[1].sha1);
2305                 if (b->branch_tree.tree) {
2306                         release_tree_content_recursive(b->branch_tree.tree);
2307                         b->branch_tree.tree = NULL;
2308                 }
2309         }
2310         else
2311                 b = new_branch(sp);
2312         read_next_command();
2313         parse_from(b);
2314         if (command_buf.len > 0)
2315                 unread_command_buf = 1;
2316 }
2317
2318 static void parse_checkpoint(void)
2319 {
2320         if (object_count) {
2321                 cycle_packfile();
2322                 dump_branches();
2323                 dump_tags();
2324                 dump_marks();
2325         }
2326         skip_optional_lf();
2327 }
2328
2329 static void parse_progress(void)
2330 {
2331         fwrite(command_buf.buf, 1, command_buf.len, stdout);
2332         fputc('\n', stdout);
2333         fflush(stdout);
2334         skip_optional_lf();
2335 }
2336
2337 static void import_marks(const char *input_file)
2338 {
2339         char line[512];
2340         FILE *f = fopen(input_file, "r");
2341         if (!f)
2342                 die("cannot read %s: %s", input_file, strerror(errno));
2343         while (fgets(line, sizeof(line), f)) {
2344                 uintmax_t mark;
2345                 char *end;
2346                 unsigned char sha1[20];
2347                 struct object_entry *e;
2348
2349                 end = strchr(line, '\n');
2350                 if (line[0] != ':' || !end)
2351                         die("corrupt mark line: %s", line);
2352                 *end = 0;
2353                 mark = strtoumax(line + 1, &end, 10);
2354                 if (!mark || end == line + 1
2355                         || *end != ' ' || get_sha1(end + 1, sha1))
2356                         die("corrupt mark line: %s", line);
2357                 e = find_object(sha1);
2358                 if (!e) {
2359                         enum object_type type = sha1_object_info(sha1, NULL);
2360                         if (type < 0)
2361                                 die("object not found: %s", sha1_to_hex(sha1));
2362                         e = insert_object(sha1);
2363                         e->type = type;
2364                         e->pack_id = MAX_PACK_ID;
2365                         e->offset = 1; /* just not zero! */
2366                 }
2367                 insert_mark(mark, e);
2368         }
2369         fclose(f);
2370 }
2371
2372 static int git_pack_config(const char *k, const char *v, void *cb)
2373 {
2374         if (!strcmp(k, "pack.depth")) {
2375                 max_depth = git_config_int(k, v);
2376                 if (max_depth > MAX_DEPTH)
2377                         max_depth = MAX_DEPTH;
2378                 return 0;
2379         }
2380         if (!strcmp(k, "pack.compression")) {
2381                 int level = git_config_int(k, v);
2382                 if (level == -1)
2383                         level = Z_DEFAULT_COMPRESSION;
2384                 else if (level < 0 || level > Z_BEST_COMPRESSION)
2385                         die("bad pack compression level %d", level);
2386                 pack_compression_level = level;
2387                 pack_compression_seen = 1;
2388                 return 0;
2389         }
2390         return git_default_config(k, v, cb);
2391 }
2392
2393 static const char fast_import_usage[] =
2394 "git fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2395
2396 int main(int argc, const char **argv)
2397 {
2398         unsigned int i, show_stats = 1;
2399
2400         setup_git_directory();
2401         git_config(git_pack_config, NULL);
2402         if (!pack_compression_seen && core_compression_seen)
2403                 pack_compression_level = core_compression_level;
2404
2405         alloc_objects(object_entry_alloc);
2406         strbuf_init(&command_buf, 0);
2407         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2408         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2409         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2410         marks = pool_calloc(1, sizeof(struct mark_set));
2411
2412         for (i = 1; i < argc; i++) {
2413                 const char *a = argv[i];
2414
2415                 if (*a != '-' || !strcmp(a, "--"))
2416                         break;
2417                 else if (!prefixcmp(a, "--date-format=")) {
2418                         const char *fmt = a + 14;
2419                         if (!strcmp(fmt, "raw"))
2420                                 whenspec = WHENSPEC_RAW;
2421                         else if (!strcmp(fmt, "rfc2822"))
2422                                 whenspec = WHENSPEC_RFC2822;
2423                         else if (!strcmp(fmt, "now"))
2424                                 whenspec = WHENSPEC_NOW;
2425                         else
2426                                 die("unknown --date-format argument %s", fmt);
2427                 }
2428                 else if (!prefixcmp(a, "--max-pack-size="))
2429                         max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2430                 else if (!prefixcmp(a, "--depth=")) {
2431                         max_depth = strtoul(a + 8, NULL, 0);
2432                         if (max_depth > MAX_DEPTH)
2433                                 die("--depth cannot exceed %u", MAX_DEPTH);
2434                 }
2435                 else if (!prefixcmp(a, "--active-branches="))
2436                         max_active_branches = strtoul(a + 18, NULL, 0);
2437                 else if (!prefixcmp(a, "--import-marks="))
2438                         import_marks(a + 15);
2439                 else if (!prefixcmp(a, "--export-marks="))
2440                         mark_file = a + 15;
2441                 else if (!prefixcmp(a, "--export-pack-edges=")) {
2442                         if (pack_edges)
2443                                 fclose(pack_edges);
2444                         pack_edges = fopen(a + 20, "a");
2445                         if (!pack_edges)
2446                                 die("Cannot open %s: %s", a + 20, strerror(errno));
2447                 } else if (!strcmp(a, "--force"))
2448                         force_update = 1;
2449                 else if (!strcmp(a, "--quiet"))
2450                         show_stats = 0;
2451                 else if (!strcmp(a, "--stats"))
2452                         show_stats = 1;
2453                 else
2454                         die("unknown option %s", a);
2455         }
2456         if (i != argc)
2457                 usage(fast_import_usage);
2458
2459         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2460         for (i = 0; i < (cmd_save - 1); i++)
2461                 rc_free[i].next = &rc_free[i + 1];
2462         rc_free[cmd_save - 1].next = NULL;
2463
2464         prepare_packed_git();
2465         start_packfile();
2466         set_die_routine(die_nicely);
2467         while (read_next_command() != EOF) {
2468                 if (!strcmp("blob", command_buf.buf))
2469                         parse_new_blob();
2470                 else if (!prefixcmp(command_buf.buf, "commit "))
2471                         parse_new_commit();
2472                 else if (!prefixcmp(command_buf.buf, "tag "))
2473                         parse_new_tag();
2474                 else if (!prefixcmp(command_buf.buf, "reset "))
2475                         parse_reset_branch();
2476                 else if (!strcmp("checkpoint", command_buf.buf))
2477                         parse_checkpoint();
2478                 else if (!prefixcmp(command_buf.buf, "progress "))
2479                         parse_progress();
2480                 else
2481                         die("Unsupported command: %s", command_buf.buf);
2482         }
2483         end_packfile();
2484
2485         dump_branches();
2486         dump_tags();
2487         unkeep_all_packs();
2488         dump_marks();
2489
2490         if (pack_edges)
2491                 fclose(pack_edges);
2492
2493         if (show_stats) {
2494                 uintmax_t total_count = 0, duplicate_count = 0;
2495                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2496                         total_count += object_count_by_type[i];
2497                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2498                         duplicate_count += duplicate_count_by_type[i];
2499
2500                 fprintf(stderr, "%s statistics:\n", argv[0]);
2501                 fprintf(stderr, "---------------------------------------------------------------------\n");
2502                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2503                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2504                 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]);
2505                 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]);
2506                 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]);
2507                 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]);
2508                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2509                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2510                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2511                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2512                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2513                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2514                 fprintf(stderr, "---------------------------------------------------------------------\n");
2515                 pack_report();
2516                 fprintf(stderr, "---------------------------------------------------------------------\n");
2517                 fprintf(stderr, "\n");
2518         }
2519
2520         return failure ? 1 : 0;
2521 }