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