Merge branch 'jk/reflog-date' into next
[git] / fast-import.c
1 /*
2 (See Documentation/git-fast-import.txt for maintained documentation.)
3 Format of STDIN stream:
4
5   stream ::= cmd*;
6
7   cmd ::= new_blob
8         | new_commit
9         | new_tag
10         | reset_branch
11         | checkpoint
12         | progress
13         ;
14
15   new_blob ::= 'blob' lf
16     mark?
17     file_content;
18   file_content ::= data;
19
20   new_commit ::= 'commit' sp ref_str lf
21     mark?
22     ('author' sp name sp '<' email '>' sp when lf)?
23     'committer' sp name sp '<' email '>' sp when lf
24     commit_msg
25     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
26     ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
27     file_change*
28     lf?;
29   commit_msg ::= data;
30
31   file_change ::= file_clr
32     | file_del
33     | file_rnm
34     | file_cpy
35     | file_obm
36     | file_inm;
37   file_clr ::= 'deleteall' lf;
38   file_del ::= 'D' sp path_str lf;
39   file_rnm ::= 'R' sp path_str sp path_str lf;
40   file_cpy ::= 'C' sp path_str sp path_str lf;
41   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
42   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
43     data;
44
45   new_tag ::= 'tag' sp tag_str lf
46     'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
47     ('tagger' sp name sp '<' email '>' sp when lf)?
48     tag_msg;
49   tag_msg ::= data;
50
51   reset_branch ::= 'reset' sp ref_str lf
52     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
53     lf?;
54
55   checkpoint ::= 'checkpoint' lf
56     lf?;
57
58   progress ::= 'progress' sp not_lf* lf
59     lf?;
60
61      # note: the first idnum in a stream should be 1 and subsequent
62      # idnums should not have gaps between values as this will cause
63      # the stream parser to reserve space for the gapped values.  An
64      # idnum can be updated in the future to a new object by issuing
65      # a new mark directive with the old idnum.
66      #
67   mark ::= 'mark' sp idnum lf;
68   data ::= (delimited_data | exact_data)
69     lf?;
70
71     # note: delim may be any string but must not contain lf.
72     # data_line may contain any data but must not be exactly
73     # delim.
74   delimited_data ::= 'data' sp '<<' delim lf
75     (data_line lf)*
76     delim lf;
77
78      # note: declen indicates the length of binary_data in bytes.
79      # declen does not include the lf preceding the binary data.
80      #
81   exact_data ::= 'data' sp declen lf
82     binary_data;
83
84      # note: quoted strings are C-style quoting supporting \c for
85      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
86      # is the signed byte value in octal.  Note that the only
87      # characters which must actually be escaped to protect the
88      # stream formatting is: \, " and LF.  Otherwise these values
89      # are UTF8.
90      #
91   ref_str     ::= ref;
92   sha1exp_str ::= sha1exp;
93   tag_str     ::= tag;
94   path_str    ::= path    | '"' quoted(path)    '"' ;
95   mode        ::= '100644' | '644'
96                 | '100755' | '755'
97                 | '120000'
98                 ;
99
100   declen ::= # unsigned 32 bit value, ascii base10 notation;
101   bigint ::= # unsigned integer value, ascii base10 notation;
102   binary_data ::= # file content, not interpreted;
103
104   when         ::= raw_when | rfc2822_when;
105   raw_when     ::= ts sp tz;
106   rfc2822_when ::= # Valid RFC 2822 date and time;
107
108   sp ::= # ASCII space character;
109   lf ::= # ASCII newline (LF) character;
110
111      # note: a colon (':') must precede the numerical value assigned to
112      # an idnum.  This is to distinguish it from a ref or tag name as
113      # GIT does not permit ':' in ref or tag strings.
114      #
115   idnum   ::= ':' bigint;
116   path    ::= # GIT style file path, e.g. "a/b/c";
117   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
118   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
119   sha1exp ::= # Any valid GIT SHA1 expression;
120   hexsha1 ::= # SHA1 in hexadecimal format;
121
122      # note: name and email are UTF8 strings, however name must not
123      # contain '<' or lf and email must not contain any of the
124      # following: '<', '>', lf.
125      #
126   name  ::= # valid GIT author/committer name;
127   email ::= # valid GIT author/committer email;
128   ts    ::= # time since the epoch in seconds, ascii base10 notation;
129   tz    ::= # GIT style timezone;
130
131      # note: comments may appear anywhere in the input, except
132      # within a data command.  Any form of the data command
133      # always escapes the related input from comment processing.
134      #
135      # In case it is not clear, the '#' that starts the comment
136      # must be the first character on that line (an lf
137      # preceded it).
138      #
139   comment ::= '#' not_lf* lf;
140   not_lf  ::= # Any byte that is not ASCII newline (LF);
141 */
142
143 #include "builtin.h"
144 #include "cache.h"
145 #include "object.h"
146 #include "blob.h"
147 #include "tree.h"
148 #include "commit.h"
149 #include "delta.h"
150 #include "pack.h"
151 #include "refs.h"
152 #include "csum-file.h"
153 #include "quote.h"
154 #include "exec_cmd.h"
155
156 #define PACK_ID_BITS 16
157 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
158 #define DEPTH_BITS 13
159 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
160
161 struct object_entry
162 {
163         struct object_entry *next;
164         uint32_t offset;
165         uint32_t type : TYPE_BITS,
166                 pack_id : PACK_ID_BITS,
167                 depth : DEPTH_BITS;
168         unsigned char sha1[20];
169 };
170
171 struct object_entry_pool
172 {
173         struct object_entry_pool *next_pool;
174         struct object_entry *next_free;
175         struct object_entry *end;
176         struct object_entry entries[FLEX_ARRAY]; /* more */
177 };
178
179 struct mark_set
180 {
181         union {
182                 struct object_entry *marked[1024];
183                 struct mark_set *sets[1024];
184         } data;
185         unsigned int shift;
186 };
187
188 struct last_object
189 {
190         struct strbuf data;
191         uint32_t offset;
192         unsigned int depth;
193         unsigned no_swap : 1;
194 };
195
196 struct mem_pool
197 {
198         struct mem_pool *next_pool;
199         char *next_free;
200         char *end;
201         uintmax_t space[FLEX_ARRAY]; /* more */
202 };
203
204 struct atom_str
205 {
206         struct atom_str *next_atom;
207         unsigned short str_len;
208         char str_dat[FLEX_ARRAY]; /* more */
209 };
210
211 struct tree_content;
212 struct tree_entry
213 {
214         struct tree_content *tree;
215         struct atom_str *name;
216         struct tree_entry_ms
217         {
218                 uint16_t mode;
219                 unsigned char sha1[20];
220         } versions[2];
221 };
222
223 struct tree_content
224 {
225         unsigned int entry_capacity; /* must match avail_tree_content */
226         unsigned int entry_count;
227         unsigned int delta_depth;
228         struct tree_entry *entries[FLEX_ARRAY]; /* more */
229 };
230
231 struct avail_tree_content
232 {
233         unsigned int entry_capacity; /* must match tree_content */
234         struct avail_tree_content *next_avail;
235 };
236
237 struct branch
238 {
239         struct branch *table_next_branch;
240         struct branch *active_next_branch;
241         const char *name;
242         struct tree_entry branch_tree;
243         uintmax_t last_commit;
244         unsigned active : 1;
245         unsigned pack_id : PACK_ID_BITS;
246         unsigned char sha1[20];
247 };
248
249 struct tag
250 {
251         struct tag *next_tag;
252         const char *name;
253         unsigned int pack_id;
254         unsigned char sha1[20];
255 };
256
257 struct hash_list
258 {
259         struct hash_list *next;
260         unsigned char sha1[20];
261 };
262
263 typedef enum {
264         WHENSPEC_RAW = 1,
265         WHENSPEC_RFC2822,
266         WHENSPEC_NOW,
267 } whenspec_type;
268
269 struct recent_command
270 {
271         struct recent_command *prev;
272         struct recent_command *next;
273         char *buf;
274 };
275
276 /* Configured limits on output */
277 static unsigned long max_depth = 10;
278 static off_t max_packsize = (1LL << 32) - 1;
279 static int force_update;
280 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
281 static int pack_compression_seen;
282
283 /* Stats and misc. counters */
284 static uintmax_t alloc_count;
285 static uintmax_t marks_set_count;
286 static uintmax_t object_count_by_type[1 << TYPE_BITS];
287 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
288 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
289 static unsigned long object_count;
290 static unsigned long branch_count;
291 static unsigned long branch_load_count;
292 static int failure;
293 static FILE *pack_edges;
294 static unsigned int show_stats = 1;
295 static int global_argc;
296 static const char **global_argv;
297
298 /* Memory pools */
299 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
300 static size_t total_allocd;
301 static struct mem_pool *mem_pool;
302
303 /* Atom management */
304 static unsigned int atom_table_sz = 4451;
305 static unsigned int atom_cnt;
306 static struct atom_str **atom_table;
307
308 /* The .pack file being generated */
309 static unsigned int pack_id;
310 static struct packed_git *pack_data;
311 static struct packed_git **all_packs;
312 static unsigned long pack_size;
313
314 /* Table of objects we've written. */
315 static unsigned int object_entry_alloc = 5000;
316 static struct object_entry_pool *blocks;
317 static struct object_entry *object_table[1 << 16];
318 static struct mark_set *marks;
319 static const char *mark_file;
320 static const char *input_file;
321
322 /* Our last blob */
323 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
324
325 /* Tree management */
326 static unsigned int tree_entry_alloc = 1000;
327 static void *avail_tree_entry;
328 static unsigned int avail_tree_table_sz = 100;
329 static struct avail_tree_content **avail_tree_table;
330 static struct strbuf old_tree = STRBUF_INIT;
331 static struct strbuf new_tree = STRBUF_INIT;
332
333 /* Branch data */
334 static unsigned long max_active_branches = 5;
335 static unsigned long cur_active_branches;
336 static unsigned long branch_table_sz = 1039;
337 static struct branch **branch_table;
338 static struct branch *active_branches;
339
340 /* Tag data */
341 static struct tag *first_tag;
342 static struct tag *last_tag;
343
344 /* Input stream parsing */
345 static whenspec_type whenspec = WHENSPEC_RAW;
346 static struct strbuf command_buf = STRBUF_INIT;
347 static int unread_command_buf;
348 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
349 static struct recent_command *cmd_tail = &cmd_hist;
350 static struct recent_command *rc_free;
351 static unsigned int cmd_save = 100;
352 static uintmax_t next_mark;
353 static struct strbuf new_data = STRBUF_INIT;
354 static int options_enabled;
355 static int seen_non_option_command;
356
357 static void parse_argv(void);
358
359 static void write_branch_report(FILE *rpt, struct branch *b)
360 {
361         fprintf(rpt, "%s:\n", b->name);
362
363         fprintf(rpt, "  status      :");
364         if (b->active)
365                 fputs(" active", rpt);
366         if (b->branch_tree.tree)
367                 fputs(" loaded", rpt);
368         if (is_null_sha1(b->branch_tree.versions[1].sha1))
369                 fputs(" dirty", rpt);
370         fputc('\n', rpt);
371
372         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
373         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
374         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
375         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
376
377         fputs("  last pack   : ", rpt);
378         if (b->pack_id < MAX_PACK_ID)
379                 fprintf(rpt, "%u", b->pack_id);
380         fputc('\n', rpt);
381
382         fputc('\n', rpt);
383 }
384
385 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
386
387 static void write_crash_report(const char *err)
388 {
389         char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
390         FILE *rpt = fopen(loc, "w");
391         struct branch *b;
392         unsigned long lu;
393         struct recent_command *rc;
394
395         if (!rpt) {
396                 error("can't write crash report %s: %s", loc, strerror(errno));
397                 return;
398         }
399
400         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
401
402         fprintf(rpt, "fast-import crash report:\n");
403         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
404         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
405         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
406         fputc('\n', rpt);
407
408         fputs("fatal: ", rpt);
409         fputs(err, rpt);
410         fputc('\n', rpt);
411
412         fputc('\n', rpt);
413         fputs("Most Recent Commands Before Crash\n", rpt);
414         fputs("---------------------------------\n", rpt);
415         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
416                 if (rc->next == &cmd_hist)
417                         fputs("* ", rpt);
418                 else
419                         fputs("  ", rpt);
420                 fputs(rc->buf, rpt);
421                 fputc('\n', rpt);
422         }
423
424         fputc('\n', rpt);
425         fputs("Active Branch LRU\n", rpt);
426         fputs("-----------------\n", rpt);
427         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
428                 cur_active_branches,
429                 max_active_branches);
430         fputc('\n', rpt);
431         fputs("  pos  clock name\n", rpt);
432         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
433         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
434                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
435                         ++lu, b->last_commit, b->name);
436
437         fputc('\n', rpt);
438         fputs("Inactive Branches\n", rpt);
439         fputs("-----------------\n", rpt);
440         for (lu = 0; lu < branch_table_sz; lu++) {
441                 for (b = branch_table[lu]; b; b = b->table_next_branch)
442                         write_branch_report(rpt, b);
443         }
444
445         if (first_tag) {
446                 struct tag *tg;
447                 fputc('\n', rpt);
448                 fputs("Annotated Tags\n", rpt);
449                 fputs("--------------\n", rpt);
450                 for (tg = first_tag; tg; tg = tg->next_tag) {
451                         fputs(sha1_to_hex(tg->sha1), rpt);
452                         fputc(' ', rpt);
453                         fputs(tg->name, rpt);
454                         fputc('\n', rpt);
455                 }
456         }
457
458         fputc('\n', rpt);
459         fputs("Marks\n", rpt);
460         fputs("-----\n", rpt);
461         if (mark_file)
462                 fprintf(rpt, "  exported to %s\n", mark_file);
463         else
464                 dump_marks_helper(rpt, 0, marks);
465
466         fputc('\n', rpt);
467         fputs("-------------------\n", rpt);
468         fputs("END OF CRASH REPORT\n", rpt);
469         fclose(rpt);
470 }
471
472 static void end_packfile(void);
473 static void unkeep_all_packs(void);
474 static void dump_marks(void);
475
476 static NORETURN void die_nicely(const char *err, va_list params)
477 {
478         static int zombie;
479         char message[2 * PATH_MAX];
480
481         vsnprintf(message, sizeof(message), err, params);
482         fputs("fatal: ", stderr);
483         fputs(message, stderr);
484         fputc('\n', stderr);
485
486         if (!zombie) {
487                 zombie = 1;
488                 write_crash_report(message);
489                 end_packfile();
490                 unkeep_all_packs();
491                 dump_marks();
492         }
493         exit(128);
494 }
495
496 static void alloc_objects(unsigned int cnt)
497 {
498         struct object_entry_pool *b;
499
500         b = xmalloc(sizeof(struct object_entry_pool)
501                 + cnt * sizeof(struct object_entry));
502         b->next_pool = blocks;
503         b->next_free = b->entries;
504         b->end = b->entries + cnt;
505         blocks = b;
506         alloc_count += cnt;
507 }
508
509 static struct object_entry *new_object(unsigned char *sha1)
510 {
511         struct object_entry *e;
512
513         if (blocks->next_free == blocks->end)
514                 alloc_objects(object_entry_alloc);
515
516         e = blocks->next_free++;
517         hashcpy(e->sha1, sha1);
518         return e;
519 }
520
521 static struct object_entry *find_object(unsigned char *sha1)
522 {
523         unsigned int h = sha1[0] << 8 | sha1[1];
524         struct object_entry *e;
525         for (e = object_table[h]; e; e = e->next)
526                 if (!hashcmp(sha1, e->sha1))
527                         return e;
528         return NULL;
529 }
530
531 static struct object_entry *insert_object(unsigned char *sha1)
532 {
533         unsigned int h = sha1[0] << 8 | sha1[1];
534         struct object_entry *e = object_table[h];
535         struct object_entry *p = NULL;
536
537         while (e) {
538                 if (!hashcmp(sha1, e->sha1))
539                         return e;
540                 p = e;
541                 e = e->next;
542         }
543
544         e = new_object(sha1);
545         e->next = NULL;
546         e->offset = 0;
547         if (p)
548                 p->next = e;
549         else
550                 object_table[h] = e;
551         return e;
552 }
553
554 static unsigned int hc_str(const char *s, size_t len)
555 {
556         unsigned int r = 0;
557         while (len-- > 0)
558                 r = r * 31 + *s++;
559         return r;
560 }
561
562 static void *pool_alloc(size_t len)
563 {
564         struct mem_pool *p;
565         void *r;
566
567         /* round up to a 'uintmax_t' alignment */
568         if (len & (sizeof(uintmax_t) - 1))
569                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
570
571         for (p = mem_pool; p; p = p->next_pool)
572                 if ((p->end - p->next_free >= len))
573                         break;
574
575         if (!p) {
576                 if (len >= (mem_pool_alloc/2)) {
577                         total_allocd += len;
578                         return xmalloc(len);
579                 }
580                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
581                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
582                 p->next_pool = mem_pool;
583                 p->next_free = (char *) p->space;
584                 p->end = p->next_free + mem_pool_alloc;
585                 mem_pool = p;
586         }
587
588         r = p->next_free;
589         p->next_free += len;
590         return r;
591 }
592
593 static void *pool_calloc(size_t count, size_t size)
594 {
595         size_t len = count * size;
596         void *r = pool_alloc(len);
597         memset(r, 0, len);
598         return r;
599 }
600
601 static char *pool_strdup(const char *s)
602 {
603         char *r = pool_alloc(strlen(s) + 1);
604         strcpy(r, s);
605         return r;
606 }
607
608 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
609 {
610         struct mark_set *s = marks;
611         while ((idnum >> s->shift) >= 1024) {
612                 s = pool_calloc(1, sizeof(struct mark_set));
613                 s->shift = marks->shift + 10;
614                 s->data.sets[0] = marks;
615                 marks = s;
616         }
617         while (s->shift) {
618                 uintmax_t i = idnum >> s->shift;
619                 idnum -= i << s->shift;
620                 if (!s->data.sets[i]) {
621                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
622                         s->data.sets[i]->shift = s->shift - 10;
623                 }
624                 s = s->data.sets[i];
625         }
626         if (!s->data.marked[idnum])
627                 marks_set_count++;
628         s->data.marked[idnum] = oe;
629 }
630
631 static struct object_entry *find_mark(uintmax_t idnum)
632 {
633         uintmax_t orig_idnum = idnum;
634         struct mark_set *s = marks;
635         struct object_entry *oe = NULL;
636         if ((idnum >> s->shift) < 1024) {
637                 while (s && s->shift) {
638                         uintmax_t i = idnum >> s->shift;
639                         idnum -= i << s->shift;
640                         s = s->data.sets[i];
641                 }
642                 if (s)
643                         oe = s->data.marked[idnum];
644         }
645         if (!oe)
646                 die("mark :%" PRIuMAX " not declared", orig_idnum);
647         return oe;
648 }
649
650 static struct atom_str *to_atom(const char *s, unsigned short len)
651 {
652         unsigned int hc = hc_str(s, len) % atom_table_sz;
653         struct atom_str *c;
654
655         for (c = atom_table[hc]; c; c = c->next_atom)
656                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
657                         return c;
658
659         c = pool_alloc(sizeof(struct atom_str) + len + 1);
660         c->str_len = len;
661         strncpy(c->str_dat, s, len);
662         c->str_dat[len] = 0;
663         c->next_atom = atom_table[hc];
664         atom_table[hc] = c;
665         atom_cnt++;
666         return c;
667 }
668
669 static struct branch *lookup_branch(const char *name)
670 {
671         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
672         struct branch *b;
673
674         for (b = branch_table[hc]; b; b = b->table_next_branch)
675                 if (!strcmp(name, b->name))
676                         return b;
677         return NULL;
678 }
679
680 static struct branch *new_branch(const char *name)
681 {
682         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
683         struct branch *b = lookup_branch(name);
684
685         if (b)
686                 die("Invalid attempt to create duplicate branch: %s", name);
687         switch (check_ref_format(name)) {
688         case 0: break; /* its valid */
689         case CHECK_REF_FORMAT_ONELEVEL:
690                 break; /* valid, but too few '/', allow anyway */
691         default:
692                 die("Branch name doesn't conform to GIT standards: %s", name);
693         }
694
695         b = pool_calloc(1, sizeof(struct branch));
696         b->name = pool_strdup(name);
697         b->table_next_branch = branch_table[hc];
698         b->branch_tree.versions[0].mode = S_IFDIR;
699         b->branch_tree.versions[1].mode = S_IFDIR;
700         b->active = 0;
701         b->pack_id = MAX_PACK_ID;
702         branch_table[hc] = b;
703         branch_count++;
704         return b;
705 }
706
707 static unsigned int hc_entries(unsigned int cnt)
708 {
709         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
710         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
711 }
712
713 static struct tree_content *new_tree_content(unsigned int cnt)
714 {
715         struct avail_tree_content *f, *l = NULL;
716         struct tree_content *t;
717         unsigned int hc = hc_entries(cnt);
718
719         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
720                 if (f->entry_capacity >= cnt)
721                         break;
722
723         if (f) {
724                 if (l)
725                         l->next_avail = f->next_avail;
726                 else
727                         avail_tree_table[hc] = f->next_avail;
728         } else {
729                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
730                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
731                 f->entry_capacity = cnt;
732         }
733
734         t = (struct tree_content*)f;
735         t->entry_count = 0;
736         t->delta_depth = 0;
737         return t;
738 }
739
740 static void release_tree_entry(struct tree_entry *e);
741 static void release_tree_content(struct tree_content *t)
742 {
743         struct avail_tree_content *f = (struct avail_tree_content*)t;
744         unsigned int hc = hc_entries(f->entry_capacity);
745         f->next_avail = avail_tree_table[hc];
746         avail_tree_table[hc] = f;
747 }
748
749 static void release_tree_content_recursive(struct tree_content *t)
750 {
751         unsigned int i;
752         for (i = 0; i < t->entry_count; i++)
753                 release_tree_entry(t->entries[i]);
754         release_tree_content(t);
755 }
756
757 static struct tree_content *grow_tree_content(
758         struct tree_content *t,
759         int amt)
760 {
761         struct tree_content *r = new_tree_content(t->entry_count + amt);
762         r->entry_count = t->entry_count;
763         r->delta_depth = t->delta_depth;
764         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
765         release_tree_content(t);
766         return r;
767 }
768
769 static struct tree_entry *new_tree_entry(void)
770 {
771         struct tree_entry *e;
772
773         if (!avail_tree_entry) {
774                 unsigned int n = tree_entry_alloc;
775                 total_allocd += n * sizeof(struct tree_entry);
776                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
777                 while (n-- > 1) {
778                         *((void**)e) = e + 1;
779                         e++;
780                 }
781                 *((void**)e) = NULL;
782         }
783
784         e = avail_tree_entry;
785         avail_tree_entry = *((void**)e);
786         return e;
787 }
788
789 static void release_tree_entry(struct tree_entry *e)
790 {
791         if (e->tree)
792                 release_tree_content_recursive(e->tree);
793         *((void**)e) = avail_tree_entry;
794         avail_tree_entry = e;
795 }
796
797 static struct tree_content *dup_tree_content(struct tree_content *s)
798 {
799         struct tree_content *d;
800         struct tree_entry *a, *b;
801         unsigned int i;
802
803         if (!s)
804                 return NULL;
805         d = new_tree_content(s->entry_count);
806         for (i = 0; i < s->entry_count; i++) {
807                 a = s->entries[i];
808                 b = new_tree_entry();
809                 memcpy(b, a, sizeof(*a));
810                 if (a->tree && is_null_sha1(b->versions[1].sha1))
811                         b->tree = dup_tree_content(a->tree);
812                 else
813                         b->tree = NULL;
814                 d->entries[i] = b;
815         }
816         d->entry_count = s->entry_count;
817         d->delta_depth = s->delta_depth;
818
819         return d;
820 }
821
822 static void start_packfile(void)
823 {
824         static char tmpfile[PATH_MAX];
825         struct packed_git *p;
826         struct pack_header hdr;
827         int pack_fd;
828
829         pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
830                               "pack/tmp_pack_XXXXXX");
831         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
832         strcpy(p->pack_name, tmpfile);
833         p->pack_fd = pack_fd;
834
835         hdr.hdr_signature = htonl(PACK_SIGNATURE);
836         hdr.hdr_version = htonl(2);
837         hdr.hdr_entries = 0;
838         write_or_die(p->pack_fd, &hdr, sizeof(hdr));
839
840         pack_data = p;
841         pack_size = sizeof(hdr);
842         object_count = 0;
843
844         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
845         all_packs[pack_id] = p;
846 }
847
848 static int oecmp (const void *a_, const void *b_)
849 {
850         struct object_entry *a = *((struct object_entry**)a_);
851         struct object_entry *b = *((struct object_entry**)b_);
852         return hashcmp(a->sha1, b->sha1);
853 }
854
855 static char *create_index(void)
856 {
857         static char tmpfile[PATH_MAX];
858         git_SHA_CTX ctx;
859         struct sha1file *f;
860         struct object_entry **idx, **c, **last, *e;
861         struct object_entry_pool *o;
862         uint32_t array[256];
863         int i, idx_fd;
864
865         /* Build the sorted table of object IDs. */
866         idx = xmalloc(object_count * sizeof(struct object_entry*));
867         c = idx;
868         for (o = blocks; o; o = o->next_pool)
869                 for (e = o->next_free; e-- != o->entries;)
870                         if (pack_id == e->pack_id)
871                                 *c++ = e;
872         last = idx + object_count;
873         if (c != last)
874                 die("internal consistency error creating the index");
875         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
876
877         /* Generate the fan-out array. */
878         c = idx;
879         for (i = 0; i < 256; i++) {
880                 struct object_entry **next = c;
881                 while (next < last) {
882                         if ((*next)->sha1[0] != i)
883                                 break;
884                         next++;
885                 }
886                 array[i] = htonl(next - idx);
887                 c = next;
888         }
889
890         idx_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
891                              "pack/tmp_idx_XXXXXX");
892         f = sha1fd(idx_fd, tmpfile);
893         sha1write(f, array, 256 * sizeof(int));
894         git_SHA1_Init(&ctx);
895         for (c = idx; c != last; c++) {
896                 uint32_t offset = htonl((*c)->offset);
897                 sha1write(f, &offset, 4);
898                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
899                 git_SHA1_Update(&ctx, (*c)->sha1, 20);
900         }
901         sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
902         sha1close(f, NULL, CSUM_FSYNC);
903         free(idx);
904         git_SHA1_Final(pack_data->sha1, &ctx);
905         return tmpfile;
906 }
907
908 static char *keep_pack(char *curr_index_name)
909 {
910         static char name[PATH_MAX];
911         static const char *keep_msg = "fast-import";
912         int keep_fd;
913
914         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
915         if (keep_fd < 0)
916                 die_errno("cannot create keep file");
917         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
918         if (close(keep_fd))
919                 die_errno("failed to write keep file");
920
921         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
922                  get_object_directory(), sha1_to_hex(pack_data->sha1));
923         if (move_temp_to_file(pack_data->pack_name, name))
924                 die("cannot store pack file");
925
926         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
927                  get_object_directory(), sha1_to_hex(pack_data->sha1));
928         if (move_temp_to_file(curr_index_name, name))
929                 die("cannot store index file");
930         return name;
931 }
932
933 static void unkeep_all_packs(void)
934 {
935         static char name[PATH_MAX];
936         int k;
937
938         for (k = 0; k < pack_id; k++) {
939                 struct packed_git *p = all_packs[k];
940                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
941                          get_object_directory(), sha1_to_hex(p->sha1));
942                 unlink_or_warn(name);
943         }
944 }
945
946 static void end_packfile(void)
947 {
948         struct packed_git *old_p = pack_data, *new_p;
949
950         clear_delta_base_cache();
951         if (object_count) {
952                 char *idx_name;
953                 int i;
954                 struct branch *b;
955                 struct tag *t;
956
957                 close_pack_windows(pack_data);
958                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
959                                     pack_data->pack_name, object_count,
960                                     NULL, 0);
961                 close(pack_data->pack_fd);
962                 idx_name = keep_pack(create_index());
963
964                 /* Register the packfile with core git's machinery. */
965                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
966                 if (!new_p)
967                         die("core git rejected index %s", idx_name);
968                 all_packs[pack_id] = new_p;
969                 install_packed_git(new_p);
970
971                 /* Print the boundary */
972                 if (pack_edges) {
973                         fprintf(pack_edges, "%s:", new_p->pack_name);
974                         for (i = 0; i < branch_table_sz; i++) {
975                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
976                                         if (b->pack_id == pack_id)
977                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
978                                 }
979                         }
980                         for (t = first_tag; t; t = t->next_tag) {
981                                 if (t->pack_id == pack_id)
982                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
983                         }
984                         fputc('\n', pack_edges);
985                         fflush(pack_edges);
986                 }
987
988                 pack_id++;
989         }
990         else {
991                 close(old_p->pack_fd);
992                 unlink_or_warn(old_p->pack_name);
993         }
994         free(old_p);
995
996         /* We can't carry a delta across packfiles. */
997         strbuf_release(&last_blob.data);
998         last_blob.offset = 0;
999         last_blob.depth = 0;
1000 }
1001
1002 static void cycle_packfile(void)
1003 {
1004         end_packfile();
1005         start_packfile();
1006 }
1007
1008 static size_t encode_header(
1009         enum object_type type,
1010         size_t size,
1011         unsigned char *hdr)
1012 {
1013         int n = 1;
1014         unsigned char c;
1015
1016         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
1017                 die("bad type %d", type);
1018
1019         c = (type << 4) | (size & 15);
1020         size >>= 4;
1021         while (size) {
1022                 *hdr++ = c | 0x80;
1023                 c = size & 0x7f;
1024                 size >>= 7;
1025                 n++;
1026         }
1027         *hdr = c;
1028         return n;
1029 }
1030
1031 static int store_object(
1032         enum object_type type,
1033         struct strbuf *dat,
1034         struct last_object *last,
1035         unsigned char *sha1out,
1036         uintmax_t mark)
1037 {
1038         void *out, *delta;
1039         struct object_entry *e;
1040         unsigned char hdr[96];
1041         unsigned char sha1[20];
1042         unsigned long hdrlen, deltalen;
1043         git_SHA_CTX c;
1044         z_stream s;
1045
1046         hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1047                 (unsigned long)dat->len) + 1;
1048         git_SHA1_Init(&c);
1049         git_SHA1_Update(&c, hdr, hdrlen);
1050         git_SHA1_Update(&c, dat->buf, dat->len);
1051         git_SHA1_Final(sha1, &c);
1052         if (sha1out)
1053                 hashcpy(sha1out, sha1);
1054
1055         e = insert_object(sha1);
1056         if (mark)
1057                 insert_mark(mark, e);
1058         if (e->offset) {
1059                 duplicate_count_by_type[type]++;
1060                 return 1;
1061         } else if (find_sha1_pack(sha1, packed_git)) {
1062                 e->type = type;
1063                 e->pack_id = MAX_PACK_ID;
1064                 e->offset = 1; /* just not zero! */
1065                 duplicate_count_by_type[type]++;
1066                 return 1;
1067         }
1068
1069         if (last && last->data.buf && last->depth < max_depth) {
1070                 delta = diff_delta(last->data.buf, last->data.len,
1071                         dat->buf, dat->len,
1072                         &deltalen, 0);
1073                 if (delta && deltalen >= dat->len) {
1074                         free(delta);
1075                         delta = NULL;
1076                 }
1077         } else
1078                 delta = NULL;
1079
1080         memset(&s, 0, sizeof(s));
1081         deflateInit(&s, pack_compression_level);
1082         if (delta) {
1083                 s.next_in = delta;
1084                 s.avail_in = deltalen;
1085         } else {
1086                 s.next_in = (void *)dat->buf;
1087                 s.avail_in = dat->len;
1088         }
1089         s.avail_out = deflateBound(&s, s.avail_in);
1090         s.next_out = out = xmalloc(s.avail_out);
1091         while (deflate(&s, Z_FINISH) == Z_OK)
1092                 /* nothing */;
1093         deflateEnd(&s);
1094
1095         /* Determine if we should auto-checkpoint. */
1096         if ((pack_size + 60 + s.total_out) > max_packsize
1097                 || (pack_size + 60 + s.total_out) < pack_size) {
1098
1099                 /* This new object needs to *not* have the current pack_id. */
1100                 e->pack_id = pack_id + 1;
1101                 cycle_packfile();
1102
1103                 /* We cannot carry a delta into the new pack. */
1104                 if (delta) {
1105                         free(delta);
1106                         delta = NULL;
1107
1108                         memset(&s, 0, sizeof(s));
1109                         deflateInit(&s, pack_compression_level);
1110                         s.next_in = (void *)dat->buf;
1111                         s.avail_in = dat->len;
1112                         s.avail_out = deflateBound(&s, s.avail_in);
1113                         s.next_out = out = xrealloc(out, s.avail_out);
1114                         while (deflate(&s, Z_FINISH) == Z_OK)
1115                                 /* nothing */;
1116                         deflateEnd(&s);
1117                 }
1118         }
1119
1120         e->type = type;
1121         e->pack_id = pack_id;
1122         e->offset = pack_size;
1123         object_count++;
1124         object_count_by_type[type]++;
1125
1126         if (delta) {
1127                 unsigned long ofs = e->offset - last->offset;
1128                 unsigned pos = sizeof(hdr) - 1;
1129
1130                 delta_count_by_type[type]++;
1131                 e->depth = last->depth + 1;
1132
1133                 hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1134                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1135                 pack_size += hdrlen;
1136
1137                 hdr[pos] = ofs & 127;
1138                 while (ofs >>= 7)
1139                         hdr[--pos] = 128 | (--ofs & 127);
1140                 write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1141                 pack_size += sizeof(hdr) - pos;
1142         } else {
1143                 e->depth = 0;
1144                 hdrlen = encode_header(type, dat->len, hdr);
1145                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1146                 pack_size += hdrlen;
1147         }
1148
1149         write_or_die(pack_data->pack_fd, out, s.total_out);
1150         pack_size += s.total_out;
1151
1152         free(out);
1153         free(delta);
1154         if (last) {
1155                 if (last->no_swap) {
1156                         last->data = *dat;
1157                 } else {
1158                         strbuf_swap(&last->data, dat);
1159                 }
1160                 last->offset = e->offset;
1161                 last->depth = e->depth;
1162         }
1163         return 0;
1164 }
1165
1166 /* All calls must be guarded by find_object() or find_mark() to
1167  * ensure the 'struct object_entry' passed was written by this
1168  * process instance.  We unpack the entry by the offset, avoiding
1169  * the need for the corresponding .idx file.  This unpacking rule
1170  * works because we only use OBJ_REF_DELTA within the packfiles
1171  * created by fast-import.
1172  *
1173  * oe must not be NULL.  Such an oe usually comes from giving
1174  * an unknown SHA-1 to find_object() or an undefined mark to
1175  * find_mark().  Callers must test for this condition and use
1176  * the standard read_sha1_file() when it happens.
1177  *
1178  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1179  * find_mark(), where the mark was reloaded from an existing marks
1180  * file and is referencing an object that this fast-import process
1181  * instance did not write out to a packfile.  Callers must test for
1182  * this condition and use read_sha1_file() instead.
1183  */
1184 static void *gfi_unpack_entry(
1185         struct object_entry *oe,
1186         unsigned long *sizep)
1187 {
1188         enum object_type type;
1189         struct packed_git *p = all_packs[oe->pack_id];
1190         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1191                 /* The object is stored in the packfile we are writing to
1192                  * and we have modified it since the last time we scanned
1193                  * back to read a previously written object.  If an old
1194                  * window covered [p->pack_size, p->pack_size + 20) its
1195                  * data is stale and is not valid.  Closing all windows
1196                  * and updating the packfile length ensures we can read
1197                  * the newly written data.
1198                  */
1199                 close_pack_windows(p);
1200
1201                 /* We have to offer 20 bytes additional on the end of
1202                  * the packfile as the core unpacker code assumes the
1203                  * footer is present at the file end and must promise
1204                  * at least 20 bytes within any window it maps.  But
1205                  * we don't actually create the footer here.
1206                  */
1207                 p->pack_size = pack_size + 20;
1208         }
1209         return unpack_entry(p, oe->offset, &type, sizep);
1210 }
1211
1212 static const char *get_mode(const char *str, uint16_t *modep)
1213 {
1214         unsigned char c;
1215         uint16_t mode = 0;
1216
1217         while ((c = *str++) != ' ') {
1218                 if (c < '0' || c > '7')
1219                         return NULL;
1220                 mode = (mode << 3) + (c - '0');
1221         }
1222         *modep = mode;
1223         return str;
1224 }
1225
1226 static void load_tree(struct tree_entry *root)
1227 {
1228         unsigned char *sha1 = root->versions[1].sha1;
1229         struct object_entry *myoe;
1230         struct tree_content *t;
1231         unsigned long size;
1232         char *buf;
1233         const char *c;
1234
1235         root->tree = t = new_tree_content(8);
1236         if (is_null_sha1(sha1))
1237                 return;
1238
1239         myoe = find_object(sha1);
1240         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1241                 if (myoe->type != OBJ_TREE)
1242                         die("Not a tree: %s", sha1_to_hex(sha1));
1243                 t->delta_depth = myoe->depth;
1244                 buf = gfi_unpack_entry(myoe, &size);
1245                 if (!buf)
1246                         die("Can't load tree %s", sha1_to_hex(sha1));
1247         } else {
1248                 enum object_type type;
1249                 buf = read_sha1_file(sha1, &type, &size);
1250                 if (!buf || type != OBJ_TREE)
1251                         die("Can't load tree %s", sha1_to_hex(sha1));
1252         }
1253
1254         c = buf;
1255         while (c != (buf + size)) {
1256                 struct tree_entry *e = new_tree_entry();
1257
1258                 if (t->entry_count == t->entry_capacity)
1259                         root->tree = t = grow_tree_content(t, t->entry_count);
1260                 t->entries[t->entry_count++] = e;
1261
1262                 e->tree = NULL;
1263                 c = get_mode(c, &e->versions[1].mode);
1264                 if (!c)
1265                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1266                 e->versions[0].mode = e->versions[1].mode;
1267                 e->name = to_atom(c, strlen(c));
1268                 c += e->name->str_len + 1;
1269                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1270                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1271                 c += 20;
1272         }
1273         free(buf);
1274 }
1275
1276 static int tecmp0 (const void *_a, const void *_b)
1277 {
1278         struct tree_entry *a = *((struct tree_entry**)_a);
1279         struct tree_entry *b = *((struct tree_entry**)_b);
1280         return base_name_compare(
1281                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1282                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1283 }
1284
1285 static int tecmp1 (const void *_a, const void *_b)
1286 {
1287         struct tree_entry *a = *((struct tree_entry**)_a);
1288         struct tree_entry *b = *((struct tree_entry**)_b);
1289         return base_name_compare(
1290                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1291                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1292 }
1293
1294 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1295 {
1296         size_t maxlen = 0;
1297         unsigned int i;
1298
1299         if (!v)
1300                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1301         else
1302                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1303
1304         for (i = 0; i < t->entry_count; i++) {
1305                 if (t->entries[i]->versions[v].mode)
1306                         maxlen += t->entries[i]->name->str_len + 34;
1307         }
1308
1309         strbuf_reset(b);
1310         strbuf_grow(b, maxlen);
1311         for (i = 0; i < t->entry_count; i++) {
1312                 struct tree_entry *e = t->entries[i];
1313                 if (!e->versions[v].mode)
1314                         continue;
1315                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1316                                         e->name->str_dat, '\0');
1317                 strbuf_add(b, e->versions[v].sha1, 20);
1318         }
1319 }
1320
1321 static void store_tree(struct tree_entry *root)
1322 {
1323         struct tree_content *t = root->tree;
1324         unsigned int i, j, del;
1325         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1326         struct object_entry *le;
1327
1328         if (!is_null_sha1(root->versions[1].sha1))
1329                 return;
1330
1331         for (i = 0; i < t->entry_count; i++) {
1332                 if (t->entries[i]->tree)
1333                         store_tree(t->entries[i]);
1334         }
1335
1336         le = find_object(root->versions[0].sha1);
1337         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1338                 mktree(t, 0, &old_tree);
1339                 lo.data = old_tree;
1340                 lo.offset = le->offset;
1341                 lo.depth = t->delta_depth;
1342         }
1343
1344         mktree(t, 1, &new_tree);
1345         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1346
1347         t->delta_depth = lo.depth;
1348         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1349                 struct tree_entry *e = t->entries[i];
1350                 if (e->versions[1].mode) {
1351                         e->versions[0].mode = e->versions[1].mode;
1352                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1353                         t->entries[j++] = e;
1354                 } else {
1355                         release_tree_entry(e);
1356                         del++;
1357                 }
1358         }
1359         t->entry_count -= del;
1360 }
1361
1362 static int tree_content_set(
1363         struct tree_entry *root,
1364         const char *p,
1365         const unsigned char *sha1,
1366         const uint16_t mode,
1367         struct tree_content *subtree)
1368 {
1369         struct tree_content *t = root->tree;
1370         const char *slash1;
1371         unsigned int i, n;
1372         struct tree_entry *e;
1373
1374         slash1 = strchr(p, '/');
1375         if (slash1)
1376                 n = slash1 - p;
1377         else
1378                 n = strlen(p);
1379         if (!n)
1380                 die("Empty path component found in input");
1381         if (!slash1 && !S_ISDIR(mode) && subtree)
1382                 die("Non-directories cannot have subtrees");
1383
1384         for (i = 0; i < t->entry_count; i++) {
1385                 e = t->entries[i];
1386                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1387                         if (!slash1) {
1388                                 if (!S_ISDIR(mode)
1389                                                 && e->versions[1].mode == mode
1390                                                 && !hashcmp(e->versions[1].sha1, sha1))
1391                                         return 0;
1392                                 e->versions[1].mode = mode;
1393                                 hashcpy(e->versions[1].sha1, sha1);
1394                                 if (e->tree)
1395                                         release_tree_content_recursive(e->tree);
1396                                 e->tree = subtree;
1397                                 hashclr(root->versions[1].sha1);
1398                                 return 1;
1399                         }
1400                         if (!S_ISDIR(e->versions[1].mode)) {
1401                                 e->tree = new_tree_content(8);
1402                                 e->versions[1].mode = S_IFDIR;
1403                         }
1404                         if (!e->tree)
1405                                 load_tree(e);
1406                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1407                                 hashclr(root->versions[1].sha1);
1408                                 return 1;
1409                         }
1410                         return 0;
1411                 }
1412         }
1413
1414         if (t->entry_count == t->entry_capacity)
1415                 root->tree = t = grow_tree_content(t, t->entry_count);
1416         e = new_tree_entry();
1417         e->name = to_atom(p, n);
1418         e->versions[0].mode = 0;
1419         hashclr(e->versions[0].sha1);
1420         t->entries[t->entry_count++] = e;
1421         if (slash1) {
1422                 e->tree = new_tree_content(8);
1423                 e->versions[1].mode = S_IFDIR;
1424                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1425         } else {
1426                 e->tree = subtree;
1427                 e->versions[1].mode = mode;
1428                 hashcpy(e->versions[1].sha1, sha1);
1429         }
1430         hashclr(root->versions[1].sha1);
1431         return 1;
1432 }
1433
1434 static int tree_content_remove(
1435         struct tree_entry *root,
1436         const char *p,
1437         struct tree_entry *backup_leaf)
1438 {
1439         struct tree_content *t = root->tree;
1440         const char *slash1;
1441         unsigned int i, n;
1442         struct tree_entry *e;
1443
1444         slash1 = strchr(p, '/');
1445         if (slash1)
1446                 n = slash1 - p;
1447         else
1448                 n = strlen(p);
1449
1450         for (i = 0; i < t->entry_count; i++) {
1451                 e = t->entries[i];
1452                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1453                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1454                                 goto del_entry;
1455                         if (!e->tree)
1456                                 load_tree(e);
1457                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1458                                 for (n = 0; n < e->tree->entry_count; n++) {
1459                                         if (e->tree->entries[n]->versions[1].mode) {
1460                                                 hashclr(root->versions[1].sha1);
1461                                                 return 1;
1462                                         }
1463                                 }
1464                                 backup_leaf = NULL;
1465                                 goto del_entry;
1466                         }
1467                         return 0;
1468                 }
1469         }
1470         return 0;
1471
1472 del_entry:
1473         if (backup_leaf)
1474                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1475         else if (e->tree)
1476                 release_tree_content_recursive(e->tree);
1477         e->tree = NULL;
1478         e->versions[1].mode = 0;
1479         hashclr(e->versions[1].sha1);
1480         hashclr(root->versions[1].sha1);
1481         return 1;
1482 }
1483
1484 static int tree_content_get(
1485         struct tree_entry *root,
1486         const char *p,
1487         struct tree_entry *leaf)
1488 {
1489         struct tree_content *t = root->tree;
1490         const char *slash1;
1491         unsigned int i, n;
1492         struct tree_entry *e;
1493
1494         slash1 = strchr(p, '/');
1495         if (slash1)
1496                 n = slash1 - p;
1497         else
1498                 n = strlen(p);
1499
1500         for (i = 0; i < t->entry_count; i++) {
1501                 e = t->entries[i];
1502                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1503                         if (!slash1) {
1504                                 memcpy(leaf, e, sizeof(*leaf));
1505                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1506                                         leaf->tree = dup_tree_content(e->tree);
1507                                 else
1508                                         leaf->tree = NULL;
1509                                 return 1;
1510                         }
1511                         if (!S_ISDIR(e->versions[1].mode))
1512                                 return 0;
1513                         if (!e->tree)
1514                                 load_tree(e);
1515                         return tree_content_get(e, slash1 + 1, leaf);
1516                 }
1517         }
1518         return 0;
1519 }
1520
1521 static int update_branch(struct branch *b)
1522 {
1523         static const char *msg = "fast-import";
1524         struct ref_lock *lock;
1525         unsigned char old_sha1[20];
1526
1527         if (is_null_sha1(b->sha1))
1528                 return 0;
1529         if (read_ref(b->name, old_sha1))
1530                 hashclr(old_sha1);
1531         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1532         if (!lock)
1533                 return error("Unable to lock %s", b->name);
1534         if (!force_update && !is_null_sha1(old_sha1)) {
1535                 struct commit *old_cmit, *new_cmit;
1536
1537                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1538                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1539                 if (!old_cmit || !new_cmit) {
1540                         unlock_ref(lock);
1541                         return error("Branch %s is missing commits.", b->name);
1542                 }
1543
1544                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1545                         unlock_ref(lock);
1546                         warning("Not updating %s"
1547                                 " (new tip %s does not contain %s)",
1548                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1549                         return -1;
1550                 }
1551         }
1552         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1553                 return error("Unable to update %s", b->name);
1554         return 0;
1555 }
1556
1557 static void dump_branches(void)
1558 {
1559         unsigned int i;
1560         struct branch *b;
1561
1562         for (i = 0; i < branch_table_sz; i++) {
1563                 for (b = branch_table[i]; b; b = b->table_next_branch)
1564                         failure |= update_branch(b);
1565         }
1566 }
1567
1568 static void dump_tags(void)
1569 {
1570         static const char *msg = "fast-import";
1571         struct tag *t;
1572         struct ref_lock *lock;
1573         char ref_name[PATH_MAX];
1574
1575         for (t = first_tag; t; t = t->next_tag) {
1576                 sprintf(ref_name, "tags/%s", t->name);
1577                 lock = lock_ref_sha1(ref_name, NULL);
1578                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1579                         failure |= error("Unable to update %s", ref_name);
1580         }
1581 }
1582
1583 static void dump_marks_helper(FILE *f,
1584         uintmax_t base,
1585         struct mark_set *m)
1586 {
1587         uintmax_t k;
1588         if (m->shift) {
1589                 for (k = 0; k < 1024; k++) {
1590                         if (m->data.sets[k])
1591                                 dump_marks_helper(f, (base + k) << m->shift,
1592                                         m->data.sets[k]);
1593                 }
1594         } else {
1595                 for (k = 0; k < 1024; k++) {
1596                         if (m->data.marked[k])
1597                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1598                                         sha1_to_hex(m->data.marked[k]->sha1));
1599                 }
1600         }
1601 }
1602
1603 static void dump_marks(void)
1604 {
1605         static struct lock_file mark_lock;
1606         int mark_fd;
1607         FILE *f;
1608
1609         if (!mark_file)
1610                 return;
1611
1612         mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1613         if (mark_fd < 0) {
1614                 failure |= error("Unable to write marks file %s: %s",
1615                         mark_file, strerror(errno));
1616                 return;
1617         }
1618
1619         f = fdopen(mark_fd, "w");
1620         if (!f) {
1621                 int saved_errno = errno;
1622                 rollback_lock_file(&mark_lock);
1623                 failure |= error("Unable to write marks file %s: %s",
1624                         mark_file, strerror(saved_errno));
1625                 return;
1626         }
1627
1628         /*
1629          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1630          * Assign -1 to the lock file descriptor so that commit_lock_file()
1631          * won't try to close() it.
1632          */
1633         mark_lock.fd = -1;
1634
1635         dump_marks_helper(f, 0, marks);
1636         if (ferror(f) || fclose(f)) {
1637                 int saved_errno = errno;
1638                 rollback_lock_file(&mark_lock);
1639                 failure |= error("Unable to write marks file %s: %s",
1640                         mark_file, strerror(saved_errno));
1641                 return;
1642         }
1643
1644         if (commit_lock_file(&mark_lock)) {
1645                 int saved_errno = errno;
1646                 rollback_lock_file(&mark_lock);
1647                 failure |= error("Unable to commit marks file %s: %s",
1648                         mark_file, strerror(saved_errno));
1649                 return;
1650         }
1651 }
1652
1653 static void read_marks(void)
1654 {
1655         char line[512];
1656         FILE *f = fopen(input_file, "r");
1657         if (!f)
1658                 die_errno("cannot read '%s'", input_file);
1659         while (fgets(line, sizeof(line), f)) {
1660                 uintmax_t mark;
1661                 char *end;
1662                 unsigned char sha1[20];
1663                 struct object_entry *e;
1664
1665                 end = strchr(line, '\n');
1666                 if (line[0] != ':' || !end)
1667                         die("corrupt mark line: %s", line);
1668                 *end = 0;
1669                 mark = strtoumax(line + 1, &end, 10);
1670                 if (!mark || end == line + 1
1671                         || *end != ' ' || get_sha1(end + 1, sha1))
1672                         die("corrupt mark line: %s", line);
1673                 e = find_object(sha1);
1674                 if (!e) {
1675                         enum object_type type = sha1_object_info(sha1, NULL);
1676                         if (type < 0)
1677                                 die("object not found: %s", sha1_to_hex(sha1));
1678                         e = insert_object(sha1);
1679                         e->type = type;
1680                         e->pack_id = MAX_PACK_ID;
1681                         e->offset = 1; /* just not zero! */
1682                 }
1683                 insert_mark(mark, e);
1684         }
1685         fclose(f);
1686 }
1687
1688
1689 static int read_next_command(void)
1690 {
1691         static int stdin_eof = 0;
1692
1693         if (stdin_eof) {
1694                 unread_command_buf = 0;
1695                 return EOF;
1696         }
1697
1698         do {
1699                 if (unread_command_buf) {
1700                         unread_command_buf = 0;
1701                 } else {
1702                         struct recent_command *rc;
1703
1704                         strbuf_detach(&command_buf, NULL);
1705                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1706                         if (stdin_eof)
1707                                 return EOF;
1708
1709                         if (!seen_non_option_command
1710                                 && prefixcmp(command_buf.buf, "feature ")
1711                                 && prefixcmp(command_buf.buf, "option ")) {
1712                                 parse_argv();
1713                         }
1714
1715                         rc = rc_free;
1716                         if (rc)
1717                                 rc_free = rc->next;
1718                         else {
1719                                 rc = cmd_hist.next;
1720                                 cmd_hist.next = rc->next;
1721                                 cmd_hist.next->prev = &cmd_hist;
1722                                 free(rc->buf);
1723                         }
1724
1725                         rc->buf = command_buf.buf;
1726                         rc->prev = cmd_tail;
1727                         rc->next = cmd_hist.prev;
1728                         rc->prev->next = rc;
1729                         cmd_tail = rc;
1730                 }
1731         } while (command_buf.buf[0] == '#');
1732
1733         return 0;
1734 }
1735
1736 static void skip_optional_lf(void)
1737 {
1738         int term_char = fgetc(stdin);
1739         if (term_char != '\n' && term_char != EOF)
1740                 ungetc(term_char, stdin);
1741 }
1742
1743 static void parse_mark(void)
1744 {
1745         if (!prefixcmp(command_buf.buf, "mark :")) {
1746                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1747                 read_next_command();
1748         }
1749         else
1750                 next_mark = 0;
1751 }
1752
1753 static void parse_data(struct strbuf *sb)
1754 {
1755         strbuf_reset(sb);
1756
1757         if (prefixcmp(command_buf.buf, "data "))
1758                 die("Expected 'data n' command, found: %s", command_buf.buf);
1759
1760         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1761                 char *term = xstrdup(command_buf.buf + 5 + 2);
1762                 size_t term_len = command_buf.len - 5 - 2;
1763
1764                 strbuf_detach(&command_buf, NULL);
1765                 for (;;) {
1766                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1767                                 die("EOF in data (terminator '%s' not found)", term);
1768                         if (term_len == command_buf.len
1769                                 && !strcmp(term, command_buf.buf))
1770                                 break;
1771                         strbuf_addbuf(sb, &command_buf);
1772                         strbuf_addch(sb, '\n');
1773                 }
1774                 free(term);
1775         }
1776         else {
1777                 size_t n = 0, length;
1778
1779                 length = strtoul(command_buf.buf + 5, NULL, 10);
1780
1781                 while (n < length) {
1782                         size_t s = strbuf_fread(sb, length - n, stdin);
1783                         if (!s && feof(stdin))
1784                                 die("EOF in data (%lu bytes remaining)",
1785                                         (unsigned long)(length - n));
1786                         n += s;
1787                 }
1788         }
1789
1790         skip_optional_lf();
1791 }
1792
1793 static int validate_raw_date(const char *src, char *result, int maxlen)
1794 {
1795         const char *orig_src = src;
1796         char *endp;
1797
1798         errno = 0;
1799
1800         strtoul(src, &endp, 10);
1801         if (errno || endp == src || *endp != ' ')
1802                 return -1;
1803
1804         src = endp + 1;
1805         if (*src != '-' && *src != '+')
1806                 return -1;
1807
1808         strtoul(src + 1, &endp, 10);
1809         if (errno || endp == src || *endp || (endp - orig_src) >= maxlen)
1810                 return -1;
1811
1812         strcpy(result, orig_src);
1813         return 0;
1814 }
1815
1816 static char *parse_ident(const char *buf)
1817 {
1818         const char *gt;
1819         size_t name_len;
1820         char *ident;
1821
1822         gt = strrchr(buf, '>');
1823         if (!gt)
1824                 die("Missing > in ident string: %s", buf);
1825         gt++;
1826         if (*gt != ' ')
1827                 die("Missing space after > in ident string: %s", buf);
1828         gt++;
1829         name_len = gt - buf;
1830         ident = xmalloc(name_len + 24);
1831         strncpy(ident, buf, name_len);
1832
1833         switch (whenspec) {
1834         case WHENSPEC_RAW:
1835                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1836                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1837                 break;
1838         case WHENSPEC_RFC2822:
1839                 if (parse_date(gt, ident + name_len, 24) < 0)
1840                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1841                 break;
1842         case WHENSPEC_NOW:
1843                 if (strcmp("now", gt))
1844                         die("Date in ident must be 'now': %s", buf);
1845                 datestamp(ident + name_len, 24);
1846                 break;
1847         }
1848
1849         return ident;
1850 }
1851
1852 static void parse_new_blob(void)
1853 {
1854         static struct strbuf buf = STRBUF_INIT;
1855
1856         read_next_command();
1857         parse_mark();
1858         parse_data(&buf);
1859         store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1860 }
1861
1862 static void unload_one_branch(void)
1863 {
1864         while (cur_active_branches
1865                 && cur_active_branches >= max_active_branches) {
1866                 uintmax_t min_commit = ULONG_MAX;
1867                 struct branch *e, *l = NULL, *p = NULL;
1868
1869                 for (e = active_branches; e; e = e->active_next_branch) {
1870                         if (e->last_commit < min_commit) {
1871                                 p = l;
1872                                 min_commit = e->last_commit;
1873                         }
1874                         l = e;
1875                 }
1876
1877                 if (p) {
1878                         e = p->active_next_branch;
1879                         p->active_next_branch = e->active_next_branch;
1880                 } else {
1881                         e = active_branches;
1882                         active_branches = e->active_next_branch;
1883                 }
1884                 e->active = 0;
1885                 e->active_next_branch = NULL;
1886                 if (e->branch_tree.tree) {
1887                         release_tree_content_recursive(e->branch_tree.tree);
1888                         e->branch_tree.tree = NULL;
1889                 }
1890                 cur_active_branches--;
1891         }
1892 }
1893
1894 static void load_branch(struct branch *b)
1895 {
1896         load_tree(&b->branch_tree);
1897         if (!b->active) {
1898                 b->active = 1;
1899                 b->active_next_branch = active_branches;
1900                 active_branches = b;
1901                 cur_active_branches++;
1902                 branch_load_count++;
1903         }
1904 }
1905
1906 static void file_change_m(struct branch *b)
1907 {
1908         const char *p = command_buf.buf + 2;
1909         static struct strbuf uq = STRBUF_INIT;
1910         const char *endp;
1911         struct object_entry *oe = oe;
1912         unsigned char sha1[20];
1913         uint16_t mode, inline_data = 0;
1914
1915         p = get_mode(p, &mode);
1916         if (!p)
1917                 die("Corrupt mode: %s", command_buf.buf);
1918         switch (mode) {
1919         case 0644:
1920         case 0755:
1921                 mode |= S_IFREG;
1922         case S_IFREG | 0644:
1923         case S_IFREG | 0755:
1924         case S_IFLNK:
1925         case S_IFGITLINK:
1926                 /* ok */
1927                 break;
1928         default:
1929                 die("Corrupt mode: %s", command_buf.buf);
1930         }
1931
1932         if (*p == ':') {
1933                 char *x;
1934                 oe = find_mark(strtoumax(p + 1, &x, 10));
1935                 hashcpy(sha1, oe->sha1);
1936                 p = x;
1937         } else if (!prefixcmp(p, "inline")) {
1938                 inline_data = 1;
1939                 p += 6;
1940         } else {
1941                 if (get_sha1_hex(p, sha1))
1942                         die("Invalid SHA1: %s", command_buf.buf);
1943                 oe = find_object(sha1);
1944                 p += 40;
1945         }
1946         if (*p++ != ' ')
1947                 die("Missing space after SHA1: %s", command_buf.buf);
1948
1949         strbuf_reset(&uq);
1950         if (!unquote_c_style(&uq, p, &endp)) {
1951                 if (*endp)
1952                         die("Garbage after path in: %s", command_buf.buf);
1953                 p = uq.buf;
1954         }
1955
1956         if (S_ISGITLINK(mode)) {
1957                 if (inline_data)
1958                         die("Git links cannot be specified 'inline': %s",
1959                                 command_buf.buf);
1960                 else if (oe) {
1961                         if (oe->type != OBJ_COMMIT)
1962                                 die("Not a commit (actually a %s): %s",
1963                                         typename(oe->type), command_buf.buf);
1964                 }
1965                 /*
1966                  * Accept the sha1 without checking; it expected to be in
1967                  * another repository.
1968                  */
1969         } else if (inline_data) {
1970                 static struct strbuf buf = STRBUF_INIT;
1971
1972                 if (p != uq.buf) {
1973                         strbuf_addstr(&uq, p);
1974                         p = uq.buf;
1975                 }
1976                 read_next_command();
1977                 parse_data(&buf);
1978                 store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1979         } else if (oe) {
1980                 if (oe->type != OBJ_BLOB)
1981                         die("Not a blob (actually a %s): %s",
1982                                 typename(oe->type), command_buf.buf);
1983         } else {
1984                 enum object_type type = sha1_object_info(sha1, NULL);
1985                 if (type < 0)
1986                         die("Blob not found: %s", command_buf.buf);
1987                 if (type != OBJ_BLOB)
1988                         die("Not a blob (actually a %s): %s",
1989                             typename(type), command_buf.buf);
1990         }
1991
1992         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
1993 }
1994
1995 static void file_change_d(struct branch *b)
1996 {
1997         const char *p = command_buf.buf + 2;
1998         static struct strbuf uq = STRBUF_INIT;
1999         const char *endp;
2000
2001         strbuf_reset(&uq);
2002         if (!unquote_c_style(&uq, p, &endp)) {
2003                 if (*endp)
2004                         die("Garbage after path in: %s", command_buf.buf);
2005                 p = uq.buf;
2006         }
2007         tree_content_remove(&b->branch_tree, p, NULL);
2008 }
2009
2010 static void file_change_cr(struct branch *b, int rename)
2011 {
2012         const char *s, *d;
2013         static struct strbuf s_uq = STRBUF_INIT;
2014         static struct strbuf d_uq = STRBUF_INIT;
2015         const char *endp;
2016         struct tree_entry leaf;
2017
2018         s = command_buf.buf + 2;
2019         strbuf_reset(&s_uq);
2020         if (!unquote_c_style(&s_uq, s, &endp)) {
2021                 if (*endp != ' ')
2022                         die("Missing space after source: %s", command_buf.buf);
2023         } else {
2024                 endp = strchr(s, ' ');
2025                 if (!endp)
2026                         die("Missing space after source: %s", command_buf.buf);
2027                 strbuf_add(&s_uq, s, endp - s);
2028         }
2029         s = s_uq.buf;
2030
2031         endp++;
2032         if (!*endp)
2033                 die("Missing dest: %s", command_buf.buf);
2034
2035         d = endp;
2036         strbuf_reset(&d_uq);
2037         if (!unquote_c_style(&d_uq, d, &endp)) {
2038                 if (*endp)
2039                         die("Garbage after dest in: %s", command_buf.buf);
2040                 d = d_uq.buf;
2041         }
2042
2043         memset(&leaf, 0, sizeof(leaf));
2044         if (rename)
2045                 tree_content_remove(&b->branch_tree, s, &leaf);
2046         else
2047                 tree_content_get(&b->branch_tree, s, &leaf);
2048         if (!leaf.versions[1].mode)
2049                 die("Path %s not in branch", s);
2050         tree_content_set(&b->branch_tree, d,
2051                 leaf.versions[1].sha1,
2052                 leaf.versions[1].mode,
2053                 leaf.tree);
2054 }
2055
2056 static void file_change_deleteall(struct branch *b)
2057 {
2058         release_tree_content_recursive(b->branch_tree.tree);
2059         hashclr(b->branch_tree.versions[0].sha1);
2060         hashclr(b->branch_tree.versions[1].sha1);
2061         load_tree(&b->branch_tree);
2062 }
2063
2064 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2065 {
2066         if (!buf || size < 46)
2067                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2068         if (memcmp("tree ", buf, 5)
2069                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2070                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2071         hashcpy(b->branch_tree.versions[0].sha1,
2072                 b->branch_tree.versions[1].sha1);
2073 }
2074
2075 static void parse_from_existing(struct branch *b)
2076 {
2077         if (is_null_sha1(b->sha1)) {
2078                 hashclr(b->branch_tree.versions[0].sha1);
2079                 hashclr(b->branch_tree.versions[1].sha1);
2080         } else {
2081                 unsigned long size;
2082                 char *buf;
2083
2084                 buf = read_object_with_reference(b->sha1,
2085                         commit_type, &size, b->sha1);
2086                 parse_from_commit(b, buf, size);
2087                 free(buf);
2088         }
2089 }
2090
2091 static int parse_from(struct branch *b)
2092 {
2093         const char *from;
2094         struct branch *s;
2095
2096         if (prefixcmp(command_buf.buf, "from "))
2097                 return 0;
2098
2099         if (b->branch_tree.tree) {
2100                 release_tree_content_recursive(b->branch_tree.tree);
2101                 b->branch_tree.tree = NULL;
2102         }
2103
2104         from = strchr(command_buf.buf, ' ') + 1;
2105         s = lookup_branch(from);
2106         if (b == s)
2107                 die("Can't create a branch from itself: %s", b->name);
2108         else if (s) {
2109                 unsigned char *t = s->branch_tree.versions[1].sha1;
2110                 hashcpy(b->sha1, s->sha1);
2111                 hashcpy(b->branch_tree.versions[0].sha1, t);
2112                 hashcpy(b->branch_tree.versions[1].sha1, t);
2113         } else if (*from == ':') {
2114                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2115                 struct object_entry *oe = find_mark(idnum);
2116                 if (oe->type != OBJ_COMMIT)
2117                         die("Mark :%" PRIuMAX " not a commit", idnum);
2118                 hashcpy(b->sha1, oe->sha1);
2119                 if (oe->pack_id != MAX_PACK_ID) {
2120                         unsigned long size;
2121                         char *buf = gfi_unpack_entry(oe, &size);
2122                         parse_from_commit(b, buf, size);
2123                         free(buf);
2124                 } else
2125                         parse_from_existing(b);
2126         } else if (!get_sha1(from, b->sha1))
2127                 parse_from_existing(b);
2128         else
2129                 die("Invalid ref name or SHA1 expression: %s", from);
2130
2131         read_next_command();
2132         return 1;
2133 }
2134
2135 static struct hash_list *parse_merge(unsigned int *count)
2136 {
2137         struct hash_list *list = NULL, *n, *e = e;
2138         const char *from;
2139         struct branch *s;
2140
2141         *count = 0;
2142         while (!prefixcmp(command_buf.buf, "merge ")) {
2143                 from = strchr(command_buf.buf, ' ') + 1;
2144                 n = xmalloc(sizeof(*n));
2145                 s = lookup_branch(from);
2146                 if (s)
2147                         hashcpy(n->sha1, s->sha1);
2148                 else if (*from == ':') {
2149                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2150                         struct object_entry *oe = find_mark(idnum);
2151                         if (oe->type != OBJ_COMMIT)
2152                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2153                         hashcpy(n->sha1, oe->sha1);
2154                 } else if (!get_sha1(from, n->sha1)) {
2155                         unsigned long size;
2156                         char *buf = read_object_with_reference(n->sha1,
2157                                 commit_type, &size, n->sha1);
2158                         if (!buf || size < 46)
2159                                 die("Not a valid commit: %s", from);
2160                         free(buf);
2161                 } else
2162                         die("Invalid ref name or SHA1 expression: %s", from);
2163
2164                 n->next = NULL;
2165                 if (list)
2166                         e->next = n;
2167                 else
2168                         list = n;
2169                 e = n;
2170                 (*count)++;
2171                 read_next_command();
2172         }
2173         return list;
2174 }
2175
2176 static void parse_new_commit(void)
2177 {
2178         static struct strbuf msg = STRBUF_INIT;
2179         struct branch *b;
2180         char *sp;
2181         char *author = NULL;
2182         char *committer = NULL;
2183         struct hash_list *merge_list = NULL;
2184         unsigned int merge_count;
2185
2186         /* Obtain the branch name from the rest of our command */
2187         sp = strchr(command_buf.buf, ' ') + 1;
2188         b = lookup_branch(sp);
2189         if (!b)
2190                 b = new_branch(sp);
2191
2192         read_next_command();
2193         parse_mark();
2194         if (!prefixcmp(command_buf.buf, "author ")) {
2195                 author = parse_ident(command_buf.buf + 7);
2196                 read_next_command();
2197         }
2198         if (!prefixcmp(command_buf.buf, "committer ")) {
2199                 committer = parse_ident(command_buf.buf + 10);
2200                 read_next_command();
2201         }
2202         if (!committer)
2203                 die("Expected committer but didn't get one");
2204         parse_data(&msg);
2205         read_next_command();
2206         parse_from(b);
2207         merge_list = parse_merge(&merge_count);
2208
2209         /* ensure the branch is active/loaded */
2210         if (!b->branch_tree.tree || !max_active_branches) {
2211                 unload_one_branch();
2212                 load_branch(b);
2213         }
2214
2215         /* file_change* */
2216         while (command_buf.len > 0) {
2217                 if (!prefixcmp(command_buf.buf, "M "))
2218                         file_change_m(b);
2219                 else if (!prefixcmp(command_buf.buf, "D "))
2220                         file_change_d(b);
2221                 else if (!prefixcmp(command_buf.buf, "R "))
2222                         file_change_cr(b, 1);
2223                 else if (!prefixcmp(command_buf.buf, "C "))
2224                         file_change_cr(b, 0);
2225                 else if (!strcmp("deleteall", command_buf.buf))
2226                         file_change_deleteall(b);
2227                 else {
2228                         unread_command_buf = 1;
2229                         break;
2230                 }
2231                 if (read_next_command() == EOF)
2232                         break;
2233         }
2234
2235         /* build the tree and the commit */
2236         store_tree(&b->branch_tree);
2237         hashcpy(b->branch_tree.versions[0].sha1,
2238                 b->branch_tree.versions[1].sha1);
2239
2240         strbuf_reset(&new_data);
2241         strbuf_addf(&new_data, "tree %s\n",
2242                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2243         if (!is_null_sha1(b->sha1))
2244                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2245         while (merge_list) {
2246                 struct hash_list *next = merge_list->next;
2247                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2248                 free(merge_list);
2249                 merge_list = next;
2250         }
2251         strbuf_addf(&new_data,
2252                 "author %s\n"
2253                 "committer %s\n"
2254                 "\n",
2255                 author ? author : committer, committer);
2256         strbuf_addbuf(&new_data, &msg);
2257         free(author);
2258         free(committer);
2259
2260         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2261                 b->pack_id = pack_id;
2262         b->last_commit = object_count_by_type[OBJ_COMMIT];
2263 }
2264
2265 static void parse_new_tag(void)
2266 {
2267         static struct strbuf msg = STRBUF_INIT;
2268         char *sp;
2269         const char *from;
2270         char *tagger;
2271         struct branch *s;
2272         struct tag *t;
2273         uintmax_t from_mark = 0;
2274         unsigned char sha1[20];
2275
2276         /* Obtain the new tag name from the rest of our command */
2277         sp = strchr(command_buf.buf, ' ') + 1;
2278         t = pool_alloc(sizeof(struct tag));
2279         t->next_tag = NULL;
2280         t->name = pool_strdup(sp);
2281         if (last_tag)
2282                 last_tag->next_tag = t;
2283         else
2284                 first_tag = t;
2285         last_tag = t;
2286         read_next_command();
2287
2288         /* from ... */
2289         if (prefixcmp(command_buf.buf, "from "))
2290                 die("Expected from command, got %s", command_buf.buf);
2291         from = strchr(command_buf.buf, ' ') + 1;
2292         s = lookup_branch(from);
2293         if (s) {
2294                 hashcpy(sha1, s->sha1);
2295         } else if (*from == ':') {
2296                 struct object_entry *oe;
2297                 from_mark = strtoumax(from + 1, NULL, 10);
2298                 oe = find_mark(from_mark);
2299                 if (oe->type != OBJ_COMMIT)
2300                         die("Mark :%" PRIuMAX " not a commit", from_mark);
2301                 hashcpy(sha1, oe->sha1);
2302         } else if (!get_sha1(from, sha1)) {
2303                 unsigned long size;
2304                 char *buf;
2305
2306                 buf = read_object_with_reference(sha1,
2307                         commit_type, &size, sha1);
2308                 if (!buf || size < 46)
2309                         die("Not a valid commit: %s", from);
2310                 free(buf);
2311         } else
2312                 die("Invalid ref name or SHA1 expression: %s", from);
2313         read_next_command();
2314
2315         /* tagger ... */
2316         if (!prefixcmp(command_buf.buf, "tagger ")) {
2317                 tagger = parse_ident(command_buf.buf + 7);
2318                 read_next_command();
2319         } else
2320                 tagger = NULL;
2321
2322         /* tag payload/message */
2323         parse_data(&msg);
2324
2325         /* build the tag object */
2326         strbuf_reset(&new_data);
2327
2328         strbuf_addf(&new_data,
2329                     "object %s\n"
2330                     "type %s\n"
2331                     "tag %s\n",
2332                     sha1_to_hex(sha1), commit_type, t->name);
2333         if (tagger)
2334                 strbuf_addf(&new_data,
2335                             "tagger %s\n", tagger);
2336         strbuf_addch(&new_data, '\n');
2337         strbuf_addbuf(&new_data, &msg);
2338         free(tagger);
2339
2340         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2341                 t->pack_id = MAX_PACK_ID;
2342         else
2343                 t->pack_id = pack_id;
2344 }
2345
2346 static void parse_reset_branch(void)
2347 {
2348         struct branch *b;
2349         char *sp;
2350
2351         /* Obtain the branch name from the rest of our command */
2352         sp = strchr(command_buf.buf, ' ') + 1;
2353         b = lookup_branch(sp);
2354         if (b) {
2355                 hashclr(b->sha1);
2356                 hashclr(b->branch_tree.versions[0].sha1);
2357                 hashclr(b->branch_tree.versions[1].sha1);
2358                 if (b->branch_tree.tree) {
2359                         release_tree_content_recursive(b->branch_tree.tree);
2360                         b->branch_tree.tree = NULL;
2361                 }
2362         }
2363         else
2364                 b = new_branch(sp);
2365         read_next_command();
2366         parse_from(b);
2367         if (command_buf.len > 0)
2368                 unread_command_buf = 1;
2369 }
2370
2371 static void parse_checkpoint(void)
2372 {
2373         if (object_count) {
2374                 cycle_packfile();
2375                 dump_branches();
2376                 dump_tags();
2377                 dump_marks();
2378         }
2379         skip_optional_lf();
2380 }
2381
2382 static void parse_progress(void)
2383 {
2384         fwrite(command_buf.buf, 1, command_buf.len, stdout);
2385         fputc('\n', stdout);
2386         fflush(stdout);
2387         skip_optional_lf();
2388 }
2389
2390 static void option_import_marks(const char *marks)
2391 {
2392         input_file = xstrdup(marks);
2393 }
2394
2395 static void option_date_format(const char *fmt)
2396 {
2397         if (!strcmp(fmt, "raw"))
2398                 whenspec = WHENSPEC_RAW;
2399         else if (!strcmp(fmt, "rfc2822"))
2400                 whenspec = WHENSPEC_RFC2822;
2401         else if (!strcmp(fmt, "now"))
2402                 whenspec = WHENSPEC_NOW;
2403         else
2404                 die("unknown --date-format argument %s", fmt);
2405 }
2406
2407 static void option_max_pack_size(const char *packsize)
2408 {
2409         max_packsize = strtoumax(packsize, NULL, 0) * 1024 * 1024;
2410 }
2411
2412 static void option_depth(const char *depth)
2413 {
2414         max_depth = strtoul(depth, NULL, 0);
2415         if (max_depth > MAX_DEPTH)
2416                 die("--depth cannot exceed %u", MAX_DEPTH);
2417 }
2418
2419 static void option_active_branches(const char *branches)
2420 {
2421         max_active_branches = strtoul(branches, NULL, 0);
2422 }
2423
2424 static void option_export_marks(const char *marks)
2425 {
2426         mark_file = xstrdup(marks);
2427 }
2428
2429 static void option_export_pack_edges(const char *edges)
2430 {
2431         if (pack_edges)
2432                 fclose(pack_edges);
2433         pack_edges = fopen(edges, "a");
2434         if (!pack_edges)
2435                 die_errno("Cannot open '%s'", edges);
2436 }
2437
2438 static void parse_one_option(const char *option, int optional)
2439 {
2440         if (!prefixcmp(option, "date-format=")) {
2441                 option_date_format(option + 12);
2442         } else if (!prefixcmp(option, "max-pack-size=")) {
2443                 option_max_pack_size(option + 14);
2444         } else if (!prefixcmp(option, "depth=")) {
2445                 option_depth(option + 6);
2446         } else if (!prefixcmp(option, "active-branches=")) {
2447                 option_active_branches(option + 16);
2448         } else if (!prefixcmp(option, "import-marks=")) {
2449                 option_import_marks(option + 13);
2450         } else if (!prefixcmp(option, "export-marks=")) {
2451                 option_export_marks(option + 13);
2452         } else if (!prefixcmp(option, "export-pack-edges=")) {
2453                 option_export_pack_edges(option + 18);
2454         } else if (!prefixcmp(option, "force")) {
2455                 force_update = 1;
2456         } else if (!prefixcmp(option, "quiet")) {
2457                 show_stats = 0;
2458         } else if (!prefixcmp(option, "stats")) {
2459                 show_stats = 1;
2460         } else if (!optional) {
2461                 die("Unsupported option: %s", option);
2462         }
2463 }
2464
2465 static void parse_feature(void)
2466 {
2467         char *feature = command_buf.buf + 8;
2468
2469         if (!prefixcmp(feature, "date-format=")) {
2470                 option_date_format(feature + 12);
2471         } else if (!strcmp("git-options", feature)) {
2472                 options_enabled = 1;
2473         } else {
2474                 die("This version of fast-import does not support feature %s.", feature);
2475         }
2476 }
2477
2478 static void parse_option(void)
2479 {
2480         char *option = command_buf.buf + 11;
2481
2482         if (!options_enabled)
2483                 die("Got option command '%s' before options feature'", option);
2484
2485         if (seen_non_option_command)
2486                 die("Got option command '%s' after non-option command", option);
2487
2488         /* ignore unknown options, instead of erroring */
2489         parse_one_option(option, 1);
2490 }
2491
2492 static void parse_nongit_option(void)
2493 {
2494         /* do nothing */
2495 }
2496
2497 static int git_pack_config(const char *k, const char *v, void *cb)
2498 {
2499         if (!strcmp(k, "pack.depth")) {
2500                 max_depth = git_config_int(k, v);
2501                 if (max_depth > MAX_DEPTH)
2502                         max_depth = MAX_DEPTH;
2503                 return 0;
2504         }
2505         if (!strcmp(k, "pack.compression")) {
2506                 int level = git_config_int(k, v);
2507                 if (level == -1)
2508                         level = Z_DEFAULT_COMPRESSION;
2509                 else if (level < 0 || level > Z_BEST_COMPRESSION)
2510                         die("bad pack compression level %d", level);
2511                 pack_compression_level = level;
2512                 pack_compression_seen = 1;
2513                 return 0;
2514         }
2515         return git_default_config(k, v, cb);
2516 }
2517
2518 static const char fast_import_usage[] =
2519 "git fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2520
2521 static void parse_argv(void)
2522 {
2523         unsigned int i;
2524
2525         for (i = 1; i < global_argc; i++) {
2526                 const char *a = global_argv[i];
2527
2528                 if (*a != '-' || !strcmp(a, "--"))
2529                         break;
2530
2531                 /* error on unknown options */
2532                 parse_one_option(a + 2, 0);
2533         }
2534         if (i != global_argc)
2535                 usage(fast_import_usage);
2536
2537         seen_non_option_command = 1;
2538         if (input_file)
2539                 read_marks();
2540 }
2541
2542 int main(int argc, const char **argv)
2543 {
2544         unsigned int i;
2545
2546         git_extract_argv0_path(argv[0]);
2547
2548         setup_git_directory();
2549         git_config(git_pack_config, NULL);
2550         if (!pack_compression_seen && core_compression_seen)
2551                 pack_compression_level = core_compression_level;
2552
2553         alloc_objects(object_entry_alloc);
2554         strbuf_init(&command_buf, 0);
2555         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2556         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2557         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2558         marks = pool_calloc(1, sizeof(struct mark_set));
2559
2560         global_argc = argc;
2561         global_argv = argv;
2562
2563         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2564         for (i = 0; i < (cmd_save - 1); i++)
2565                 rc_free[i].next = &rc_free[i + 1];
2566         rc_free[cmd_save - 1].next = NULL;
2567
2568         prepare_packed_git();
2569         start_packfile();
2570         set_die_routine(die_nicely);
2571         while (read_next_command() != EOF) {
2572                 if (!strcmp("blob", command_buf.buf))
2573                         parse_new_blob();
2574                 else if (!prefixcmp(command_buf.buf, "commit "))
2575                         parse_new_commit();
2576                 else if (!prefixcmp(command_buf.buf, "tag "))
2577                         parse_new_tag();
2578                 else if (!prefixcmp(command_buf.buf, "reset "))
2579                         parse_reset_branch();
2580                 else if (!strcmp("checkpoint", command_buf.buf))
2581                         parse_checkpoint();
2582                 else if (!prefixcmp(command_buf.buf, "progress "))
2583                         parse_progress();
2584                 else if (!prefixcmp(command_buf.buf, "feature "))
2585                         parse_feature();
2586                 else if (!prefixcmp(command_buf.buf, "option git "))
2587                         parse_option();
2588                 else if (!prefixcmp(command_buf.buf, "option "))
2589                         parse_nongit_option();
2590                 else
2591                         die("Unsupported command: %s", command_buf.buf);
2592         }
2593
2594         /* argv hasn't been parsed yet, do so */
2595         if (!seen_non_option_command)
2596                 parse_argv();
2597
2598         end_packfile();
2599
2600         dump_branches();
2601         dump_tags();
2602         unkeep_all_packs();
2603         dump_marks();
2604
2605         if (pack_edges)
2606                 fclose(pack_edges);
2607
2608         if (show_stats) {
2609                 uintmax_t total_count = 0, duplicate_count = 0;
2610                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2611                         total_count += object_count_by_type[i];
2612                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2613                         duplicate_count += duplicate_count_by_type[i];
2614
2615                 fprintf(stderr, "%s statistics:\n", argv[0]);
2616                 fprintf(stderr, "---------------------------------------------------------------------\n");
2617                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2618                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2619                 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]);
2620                 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]);
2621                 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]);
2622                 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]);
2623                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2624                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2625                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2626                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2627                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2628                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2629                 fprintf(stderr, "---------------------------------------------------------------------\n");
2630                 pack_report();
2631                 fprintf(stderr, "---------------------------------------------------------------------\n");
2632                 fprintf(stderr, "\n");
2633         }
2634
2635         return failure ? 1 : 0;
2636 }