GIT 1.6.5.1
[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
295 /* Memory pools */
296 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
297 static size_t total_allocd;
298 static struct mem_pool *mem_pool;
299
300 /* Atom management */
301 static unsigned int atom_table_sz = 4451;
302 static unsigned int atom_cnt;
303 static struct atom_str **atom_table;
304
305 /* The .pack file being generated */
306 static unsigned int pack_id;
307 static struct packed_git *pack_data;
308 static struct packed_git **all_packs;
309 static unsigned long pack_size;
310
311 /* Table of objects we've written. */
312 static unsigned int object_entry_alloc = 5000;
313 static struct object_entry_pool *blocks;
314 static struct object_entry *object_table[1 << 16];
315 static struct mark_set *marks;
316 static const char *mark_file;
317
318 /* Our last blob */
319 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
320
321 /* Tree management */
322 static unsigned int tree_entry_alloc = 1000;
323 static void *avail_tree_entry;
324 static unsigned int avail_tree_table_sz = 100;
325 static struct avail_tree_content **avail_tree_table;
326 static struct strbuf old_tree = STRBUF_INIT;
327 static struct strbuf new_tree = STRBUF_INIT;
328
329 /* Branch data */
330 static unsigned long max_active_branches = 5;
331 static unsigned long cur_active_branches;
332 static unsigned long branch_table_sz = 1039;
333 static struct branch **branch_table;
334 static struct branch *active_branches;
335
336 /* Tag data */
337 static struct tag *first_tag;
338 static struct tag *last_tag;
339
340 /* Input stream parsing */
341 static whenspec_type whenspec = WHENSPEC_RAW;
342 static struct strbuf command_buf = STRBUF_INIT;
343 static int unread_command_buf;
344 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
345 static struct recent_command *cmd_tail = &cmd_hist;
346 static struct recent_command *rc_free;
347 static unsigned int cmd_save = 100;
348 static uintmax_t next_mark;
349 static struct strbuf new_data = STRBUF_INIT;
350
351 static void write_branch_report(FILE *rpt, struct branch *b)
352 {
353         fprintf(rpt, "%s:\n", b->name);
354
355         fprintf(rpt, "  status      :");
356         if (b->active)
357                 fputs(" active", rpt);
358         if (b->branch_tree.tree)
359                 fputs(" loaded", rpt);
360         if (is_null_sha1(b->branch_tree.versions[1].sha1))
361                 fputs(" dirty", rpt);
362         fputc('\n', rpt);
363
364         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
365         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
366         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
367         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
368
369         fputs("  last pack   : ", rpt);
370         if (b->pack_id < MAX_PACK_ID)
371                 fprintf(rpt, "%u", b->pack_id);
372         fputc('\n', rpt);
373
374         fputc('\n', rpt);
375 }
376
377 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
378
379 static void write_crash_report(const char *err)
380 {
381         char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
382         FILE *rpt = fopen(loc, "w");
383         struct branch *b;
384         unsigned long lu;
385         struct recent_command *rc;
386
387         if (!rpt) {
388                 error("can't write crash report %s: %s", loc, strerror(errno));
389                 return;
390         }
391
392         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
393
394         fprintf(rpt, "fast-import crash report:\n");
395         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
396         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
397         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
398         fputc('\n', rpt);
399
400         fputs("fatal: ", rpt);
401         fputs(err, rpt);
402         fputc('\n', rpt);
403
404         fputc('\n', rpt);
405         fputs("Most Recent Commands Before Crash\n", rpt);
406         fputs("---------------------------------\n", rpt);
407         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
408                 if (rc->next == &cmd_hist)
409                         fputs("* ", rpt);
410                 else
411                         fputs("  ", rpt);
412                 fputs(rc->buf, rpt);
413                 fputc('\n', rpt);
414         }
415
416         fputc('\n', rpt);
417         fputs("Active Branch LRU\n", rpt);
418         fputs("-----------------\n", rpt);
419         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
420                 cur_active_branches,
421                 max_active_branches);
422         fputc('\n', rpt);
423         fputs("  pos  clock name\n", rpt);
424         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
425         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
426                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
427                         ++lu, b->last_commit, b->name);
428
429         fputc('\n', rpt);
430         fputs("Inactive Branches\n", rpt);
431         fputs("-----------------\n", rpt);
432         for (lu = 0; lu < branch_table_sz; lu++) {
433                 for (b = branch_table[lu]; b; b = b->table_next_branch)
434                         write_branch_report(rpt, b);
435         }
436
437         if (first_tag) {
438                 struct tag *tg;
439                 fputc('\n', rpt);
440                 fputs("Annotated Tags\n", rpt);
441                 fputs("--------------\n", rpt);
442                 for (tg = first_tag; tg; tg = tg->next_tag) {
443                         fputs(sha1_to_hex(tg->sha1), rpt);
444                         fputc(' ', rpt);
445                         fputs(tg->name, rpt);
446                         fputc('\n', rpt);
447                 }
448         }
449
450         fputc('\n', rpt);
451         fputs("Marks\n", rpt);
452         fputs("-----\n", rpt);
453         if (mark_file)
454                 fprintf(rpt, "  exported to %s\n", mark_file);
455         else
456                 dump_marks_helper(rpt, 0, marks);
457
458         fputc('\n', rpt);
459         fputs("-------------------\n", rpt);
460         fputs("END OF CRASH REPORT\n", rpt);
461         fclose(rpt);
462 }
463
464 static void end_packfile(void);
465 static void unkeep_all_packs(void);
466 static void dump_marks(void);
467
468 static NORETURN void die_nicely(const char *err, va_list params)
469 {
470         static int zombie;
471         char message[2 * PATH_MAX];
472
473         vsnprintf(message, sizeof(message), err, params);
474         fputs("fatal: ", stderr);
475         fputs(message, stderr);
476         fputc('\n', stderr);
477
478         if (!zombie) {
479                 zombie = 1;
480                 write_crash_report(message);
481                 end_packfile();
482                 unkeep_all_packs();
483                 dump_marks();
484         }
485         exit(128);
486 }
487
488 static void alloc_objects(unsigned int cnt)
489 {
490         struct object_entry_pool *b;
491
492         b = xmalloc(sizeof(struct object_entry_pool)
493                 + cnt * sizeof(struct object_entry));
494         b->next_pool = blocks;
495         b->next_free = b->entries;
496         b->end = b->entries + cnt;
497         blocks = b;
498         alloc_count += cnt;
499 }
500
501 static struct object_entry *new_object(unsigned char *sha1)
502 {
503         struct object_entry *e;
504
505         if (blocks->next_free == blocks->end)
506                 alloc_objects(object_entry_alloc);
507
508         e = blocks->next_free++;
509         hashcpy(e->sha1, sha1);
510         return e;
511 }
512
513 static struct object_entry *find_object(unsigned char *sha1)
514 {
515         unsigned int h = sha1[0] << 8 | sha1[1];
516         struct object_entry *e;
517         for (e = object_table[h]; e; e = e->next)
518                 if (!hashcmp(sha1, e->sha1))
519                         return e;
520         return NULL;
521 }
522
523 static struct object_entry *insert_object(unsigned char *sha1)
524 {
525         unsigned int h = sha1[0] << 8 | sha1[1];
526         struct object_entry *e = object_table[h];
527         struct object_entry *p = NULL;
528
529         while (e) {
530                 if (!hashcmp(sha1, e->sha1))
531                         return e;
532                 p = e;
533                 e = e->next;
534         }
535
536         e = new_object(sha1);
537         e->next = NULL;
538         e->offset = 0;
539         if (p)
540                 p->next = e;
541         else
542                 object_table[h] = e;
543         return e;
544 }
545
546 static unsigned int hc_str(const char *s, size_t len)
547 {
548         unsigned int r = 0;
549         while (len-- > 0)
550                 r = r * 31 + *s++;
551         return r;
552 }
553
554 static void *pool_alloc(size_t len)
555 {
556         struct mem_pool *p;
557         void *r;
558
559         /* round up to a 'uintmax_t' alignment */
560         if (len & (sizeof(uintmax_t) - 1))
561                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
562
563         for (p = mem_pool; p; p = p->next_pool)
564                 if ((p->end - p->next_free >= len))
565                         break;
566
567         if (!p) {
568                 if (len >= (mem_pool_alloc/2)) {
569                         total_allocd += len;
570                         return xmalloc(len);
571                 }
572                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
573                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
574                 p->next_pool = mem_pool;
575                 p->next_free = (char *) p->space;
576                 p->end = p->next_free + mem_pool_alloc;
577                 mem_pool = p;
578         }
579
580         r = p->next_free;
581         p->next_free += len;
582         return r;
583 }
584
585 static void *pool_calloc(size_t count, size_t size)
586 {
587         size_t len = count * size;
588         void *r = pool_alloc(len);
589         memset(r, 0, len);
590         return r;
591 }
592
593 static char *pool_strdup(const char *s)
594 {
595         char *r = pool_alloc(strlen(s) + 1);
596         strcpy(r, s);
597         return r;
598 }
599
600 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
601 {
602         struct mark_set *s = marks;
603         while ((idnum >> s->shift) >= 1024) {
604                 s = pool_calloc(1, sizeof(struct mark_set));
605                 s->shift = marks->shift + 10;
606                 s->data.sets[0] = marks;
607                 marks = s;
608         }
609         while (s->shift) {
610                 uintmax_t i = idnum >> s->shift;
611                 idnum -= i << s->shift;
612                 if (!s->data.sets[i]) {
613                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
614                         s->data.sets[i]->shift = s->shift - 10;
615                 }
616                 s = s->data.sets[i];
617         }
618         if (!s->data.marked[idnum])
619                 marks_set_count++;
620         s->data.marked[idnum] = oe;
621 }
622
623 static struct object_entry *find_mark(uintmax_t idnum)
624 {
625         uintmax_t orig_idnum = idnum;
626         struct mark_set *s = marks;
627         struct object_entry *oe = NULL;
628         if ((idnum >> s->shift) < 1024) {
629                 while (s && s->shift) {
630                         uintmax_t i = idnum >> s->shift;
631                         idnum -= i << s->shift;
632                         s = s->data.sets[i];
633                 }
634                 if (s)
635                         oe = s->data.marked[idnum];
636         }
637         if (!oe)
638                 die("mark :%" PRIuMAX " not declared", orig_idnum);
639         return oe;
640 }
641
642 static struct atom_str *to_atom(const char *s, unsigned short len)
643 {
644         unsigned int hc = hc_str(s, len) % atom_table_sz;
645         struct atom_str *c;
646
647         for (c = atom_table[hc]; c; c = c->next_atom)
648                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
649                         return c;
650
651         c = pool_alloc(sizeof(struct atom_str) + len + 1);
652         c->str_len = len;
653         strncpy(c->str_dat, s, len);
654         c->str_dat[len] = 0;
655         c->next_atom = atom_table[hc];
656         atom_table[hc] = c;
657         atom_cnt++;
658         return c;
659 }
660
661 static struct branch *lookup_branch(const char *name)
662 {
663         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
664         struct branch *b;
665
666         for (b = branch_table[hc]; b; b = b->table_next_branch)
667                 if (!strcmp(name, b->name))
668                         return b;
669         return NULL;
670 }
671
672 static struct branch *new_branch(const char *name)
673 {
674         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
675         struct branch *b = lookup_branch(name);
676
677         if (b)
678                 die("Invalid attempt to create duplicate branch: %s", name);
679         switch (check_ref_format(name)) {
680         case 0: break; /* its valid */
681         case CHECK_REF_FORMAT_ONELEVEL:
682                 break; /* valid, but too few '/', allow anyway */
683         default:
684                 die("Branch name doesn't conform to GIT standards: %s", name);
685         }
686
687         b = pool_calloc(1, sizeof(struct branch));
688         b->name = pool_strdup(name);
689         b->table_next_branch = branch_table[hc];
690         b->branch_tree.versions[0].mode = S_IFDIR;
691         b->branch_tree.versions[1].mode = S_IFDIR;
692         b->active = 0;
693         b->pack_id = MAX_PACK_ID;
694         branch_table[hc] = b;
695         branch_count++;
696         return b;
697 }
698
699 static unsigned int hc_entries(unsigned int cnt)
700 {
701         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
702         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
703 }
704
705 static struct tree_content *new_tree_content(unsigned int cnt)
706 {
707         struct avail_tree_content *f, *l = NULL;
708         struct tree_content *t;
709         unsigned int hc = hc_entries(cnt);
710
711         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
712                 if (f->entry_capacity >= cnt)
713                         break;
714
715         if (f) {
716                 if (l)
717                         l->next_avail = f->next_avail;
718                 else
719                         avail_tree_table[hc] = f->next_avail;
720         } else {
721                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
722                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
723                 f->entry_capacity = cnt;
724         }
725
726         t = (struct tree_content*)f;
727         t->entry_count = 0;
728         t->delta_depth = 0;
729         return t;
730 }
731
732 static void release_tree_entry(struct tree_entry *e);
733 static void release_tree_content(struct tree_content *t)
734 {
735         struct avail_tree_content *f = (struct avail_tree_content*)t;
736         unsigned int hc = hc_entries(f->entry_capacity);
737         f->next_avail = avail_tree_table[hc];
738         avail_tree_table[hc] = f;
739 }
740
741 static void release_tree_content_recursive(struct tree_content *t)
742 {
743         unsigned int i;
744         for (i = 0; i < t->entry_count; i++)
745                 release_tree_entry(t->entries[i]);
746         release_tree_content(t);
747 }
748
749 static struct tree_content *grow_tree_content(
750         struct tree_content *t,
751         int amt)
752 {
753         struct tree_content *r = new_tree_content(t->entry_count + amt);
754         r->entry_count = t->entry_count;
755         r->delta_depth = t->delta_depth;
756         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
757         release_tree_content(t);
758         return r;
759 }
760
761 static struct tree_entry *new_tree_entry(void)
762 {
763         struct tree_entry *e;
764
765         if (!avail_tree_entry) {
766                 unsigned int n = tree_entry_alloc;
767                 total_allocd += n * sizeof(struct tree_entry);
768                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
769                 while (n-- > 1) {
770                         *((void**)e) = e + 1;
771                         e++;
772                 }
773                 *((void**)e) = NULL;
774         }
775
776         e = avail_tree_entry;
777         avail_tree_entry = *((void**)e);
778         return e;
779 }
780
781 static void release_tree_entry(struct tree_entry *e)
782 {
783         if (e->tree)
784                 release_tree_content_recursive(e->tree);
785         *((void**)e) = avail_tree_entry;
786         avail_tree_entry = e;
787 }
788
789 static struct tree_content *dup_tree_content(struct tree_content *s)
790 {
791         struct tree_content *d;
792         struct tree_entry *a, *b;
793         unsigned int i;
794
795         if (!s)
796                 return NULL;
797         d = new_tree_content(s->entry_count);
798         for (i = 0; i < s->entry_count; i++) {
799                 a = s->entries[i];
800                 b = new_tree_entry();
801                 memcpy(b, a, sizeof(*a));
802                 if (a->tree && is_null_sha1(b->versions[1].sha1))
803                         b->tree = dup_tree_content(a->tree);
804                 else
805                         b->tree = NULL;
806                 d->entries[i] = b;
807         }
808         d->entry_count = s->entry_count;
809         d->delta_depth = s->delta_depth;
810
811         return d;
812 }
813
814 static void start_packfile(void)
815 {
816         static char tmpfile[PATH_MAX];
817         struct packed_git *p;
818         struct pack_header hdr;
819         int pack_fd;
820
821         pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
822                               "pack/tmp_pack_XXXXXX");
823         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
824         strcpy(p->pack_name, tmpfile);
825         p->pack_fd = pack_fd;
826
827         hdr.hdr_signature = htonl(PACK_SIGNATURE);
828         hdr.hdr_version = htonl(2);
829         hdr.hdr_entries = 0;
830         write_or_die(p->pack_fd, &hdr, sizeof(hdr));
831
832         pack_data = p;
833         pack_size = sizeof(hdr);
834         object_count = 0;
835
836         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
837         all_packs[pack_id] = p;
838 }
839
840 static int oecmp (const void *a_, const void *b_)
841 {
842         struct object_entry *a = *((struct object_entry**)a_);
843         struct object_entry *b = *((struct object_entry**)b_);
844         return hashcmp(a->sha1, b->sha1);
845 }
846
847 static char *create_index(void)
848 {
849         static char tmpfile[PATH_MAX];
850         git_SHA_CTX ctx;
851         struct sha1file *f;
852         struct object_entry **idx, **c, **last, *e;
853         struct object_entry_pool *o;
854         uint32_t array[256];
855         int i, idx_fd;
856
857         /* Build the sorted table of object IDs. */
858         idx = xmalloc(object_count * sizeof(struct object_entry*));
859         c = idx;
860         for (o = blocks; o; o = o->next_pool)
861                 for (e = o->next_free; e-- != o->entries;)
862                         if (pack_id == e->pack_id)
863                                 *c++ = e;
864         last = idx + object_count;
865         if (c != last)
866                 die("internal consistency error creating the index");
867         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
868
869         /* Generate the fan-out array. */
870         c = idx;
871         for (i = 0; i < 256; i++) {
872                 struct object_entry **next = c;
873                 while (next < last) {
874                         if ((*next)->sha1[0] != i)
875                                 break;
876                         next++;
877                 }
878                 array[i] = htonl(next - idx);
879                 c = next;
880         }
881
882         idx_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
883                              "pack/tmp_idx_XXXXXX");
884         f = sha1fd(idx_fd, tmpfile);
885         sha1write(f, array, 256 * sizeof(int));
886         git_SHA1_Init(&ctx);
887         for (c = idx; c != last; c++) {
888                 uint32_t offset = htonl((*c)->offset);
889                 sha1write(f, &offset, 4);
890                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
891                 git_SHA1_Update(&ctx, (*c)->sha1, 20);
892         }
893         sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
894         sha1close(f, NULL, CSUM_FSYNC);
895         free(idx);
896         git_SHA1_Final(pack_data->sha1, &ctx);
897         return tmpfile;
898 }
899
900 static char *keep_pack(char *curr_index_name)
901 {
902         static char name[PATH_MAX];
903         static const char *keep_msg = "fast-import";
904         int keep_fd;
905
906         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
907         if (keep_fd < 0)
908                 die_errno("cannot create keep file");
909         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
910         if (close(keep_fd))
911                 die_errno("failed to write keep file");
912
913         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
914                  get_object_directory(), sha1_to_hex(pack_data->sha1));
915         if (move_temp_to_file(pack_data->pack_name, name))
916                 die("cannot store pack file");
917
918         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
919                  get_object_directory(), sha1_to_hex(pack_data->sha1));
920         if (move_temp_to_file(curr_index_name, name))
921                 die("cannot store index file");
922         return name;
923 }
924
925 static void unkeep_all_packs(void)
926 {
927         static char name[PATH_MAX];
928         int k;
929
930         for (k = 0; k < pack_id; k++) {
931                 struct packed_git *p = all_packs[k];
932                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
933                          get_object_directory(), sha1_to_hex(p->sha1));
934                 unlink_or_warn(name);
935         }
936 }
937
938 static void end_packfile(void)
939 {
940         struct packed_git *old_p = pack_data, *new_p;
941
942         clear_delta_base_cache();
943         if (object_count) {
944                 char *idx_name;
945                 int i;
946                 struct branch *b;
947                 struct tag *t;
948
949                 close_pack_windows(pack_data);
950                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
951                                     pack_data->pack_name, object_count,
952                                     NULL, 0);
953                 close(pack_data->pack_fd);
954                 idx_name = keep_pack(create_index());
955
956                 /* Register the packfile with core git's machinery. */
957                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
958                 if (!new_p)
959                         die("core git rejected index %s", idx_name);
960                 all_packs[pack_id] = new_p;
961                 install_packed_git(new_p);
962
963                 /* Print the boundary */
964                 if (pack_edges) {
965                         fprintf(pack_edges, "%s:", new_p->pack_name);
966                         for (i = 0; i < branch_table_sz; i++) {
967                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
968                                         if (b->pack_id == pack_id)
969                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
970                                 }
971                         }
972                         for (t = first_tag; t; t = t->next_tag) {
973                                 if (t->pack_id == pack_id)
974                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
975                         }
976                         fputc('\n', pack_edges);
977                         fflush(pack_edges);
978                 }
979
980                 pack_id++;
981         }
982         else {
983                 close(old_p->pack_fd);
984                 unlink_or_warn(old_p->pack_name);
985         }
986         free(old_p);
987
988         /* We can't carry a delta across packfiles. */
989         strbuf_release(&last_blob.data);
990         last_blob.offset = 0;
991         last_blob.depth = 0;
992 }
993
994 static void cycle_packfile(void)
995 {
996         end_packfile();
997         start_packfile();
998 }
999
1000 static size_t encode_header(
1001         enum object_type type,
1002         size_t size,
1003         unsigned char *hdr)
1004 {
1005         int n = 1;
1006         unsigned char c;
1007
1008         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
1009                 die("bad type %d", type);
1010
1011         c = (type << 4) | (size & 15);
1012         size >>= 4;
1013         while (size) {
1014                 *hdr++ = c | 0x80;
1015                 c = size & 0x7f;
1016                 size >>= 7;
1017                 n++;
1018         }
1019         *hdr = c;
1020         return n;
1021 }
1022
1023 static int store_object(
1024         enum object_type type,
1025         struct strbuf *dat,
1026         struct last_object *last,
1027         unsigned char *sha1out,
1028         uintmax_t mark)
1029 {
1030         void *out, *delta;
1031         struct object_entry *e;
1032         unsigned char hdr[96];
1033         unsigned char sha1[20];
1034         unsigned long hdrlen, deltalen;
1035         git_SHA_CTX c;
1036         z_stream s;
1037
1038         hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1039                 (unsigned long)dat->len) + 1;
1040         git_SHA1_Init(&c);
1041         git_SHA1_Update(&c, hdr, hdrlen);
1042         git_SHA1_Update(&c, dat->buf, dat->len);
1043         git_SHA1_Final(sha1, &c);
1044         if (sha1out)
1045                 hashcpy(sha1out, sha1);
1046
1047         e = insert_object(sha1);
1048         if (mark)
1049                 insert_mark(mark, e);
1050         if (e->offset) {
1051                 duplicate_count_by_type[type]++;
1052                 return 1;
1053         } else if (find_sha1_pack(sha1, packed_git)) {
1054                 e->type = type;
1055                 e->pack_id = MAX_PACK_ID;
1056                 e->offset = 1; /* just not zero! */
1057                 duplicate_count_by_type[type]++;
1058                 return 1;
1059         }
1060
1061         if (last && last->data.buf && last->depth < max_depth) {
1062                 delta = diff_delta(last->data.buf, last->data.len,
1063                         dat->buf, dat->len,
1064                         &deltalen, 0);
1065                 if (delta && deltalen >= dat->len) {
1066                         free(delta);
1067                         delta = NULL;
1068                 }
1069         } else
1070                 delta = NULL;
1071
1072         memset(&s, 0, sizeof(s));
1073         deflateInit(&s, pack_compression_level);
1074         if (delta) {
1075                 s.next_in = delta;
1076                 s.avail_in = deltalen;
1077         } else {
1078                 s.next_in = (void *)dat->buf;
1079                 s.avail_in = dat->len;
1080         }
1081         s.avail_out = deflateBound(&s, s.avail_in);
1082         s.next_out = out = xmalloc(s.avail_out);
1083         while (deflate(&s, Z_FINISH) == Z_OK)
1084                 /* nothing */;
1085         deflateEnd(&s);
1086
1087         /* Determine if we should auto-checkpoint. */
1088         if ((pack_size + 60 + s.total_out) > max_packsize
1089                 || (pack_size + 60 + s.total_out) < pack_size) {
1090
1091                 /* This new object needs to *not* have the current pack_id. */
1092                 e->pack_id = pack_id + 1;
1093                 cycle_packfile();
1094
1095                 /* We cannot carry a delta into the new pack. */
1096                 if (delta) {
1097                         free(delta);
1098                         delta = NULL;
1099
1100                         memset(&s, 0, sizeof(s));
1101                         deflateInit(&s, pack_compression_level);
1102                         s.next_in = (void *)dat->buf;
1103                         s.avail_in = dat->len;
1104                         s.avail_out = deflateBound(&s, s.avail_in);
1105                         s.next_out = out = xrealloc(out, s.avail_out);
1106                         while (deflate(&s, Z_FINISH) == Z_OK)
1107                                 /* nothing */;
1108                         deflateEnd(&s);
1109                 }
1110         }
1111
1112         e->type = type;
1113         e->pack_id = pack_id;
1114         e->offset = pack_size;
1115         object_count++;
1116         object_count_by_type[type]++;
1117
1118         if (delta) {
1119                 unsigned long ofs = e->offset - last->offset;
1120                 unsigned pos = sizeof(hdr) - 1;
1121
1122                 delta_count_by_type[type]++;
1123                 e->depth = last->depth + 1;
1124
1125                 hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1126                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1127                 pack_size += hdrlen;
1128
1129                 hdr[pos] = ofs & 127;
1130                 while (ofs >>= 7)
1131                         hdr[--pos] = 128 | (--ofs & 127);
1132                 write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1133                 pack_size += sizeof(hdr) - pos;
1134         } else {
1135                 e->depth = 0;
1136                 hdrlen = encode_header(type, dat->len, hdr);
1137                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1138                 pack_size += hdrlen;
1139         }
1140
1141         write_or_die(pack_data->pack_fd, out, s.total_out);
1142         pack_size += s.total_out;
1143
1144         free(out);
1145         free(delta);
1146         if (last) {
1147                 if (last->no_swap) {
1148                         last->data = *dat;
1149                 } else {
1150                         strbuf_swap(&last->data, dat);
1151                 }
1152                 last->offset = e->offset;
1153                 last->depth = e->depth;
1154         }
1155         return 0;
1156 }
1157
1158 /* All calls must be guarded by find_object() or find_mark() to
1159  * ensure the 'struct object_entry' passed was written by this
1160  * process instance.  We unpack the entry by the offset, avoiding
1161  * the need for the corresponding .idx file.  This unpacking rule
1162  * works because we only use OBJ_REF_DELTA within the packfiles
1163  * created by fast-import.
1164  *
1165  * oe must not be NULL.  Such an oe usually comes from giving
1166  * an unknown SHA-1 to find_object() or an undefined mark to
1167  * find_mark().  Callers must test for this condition and use
1168  * the standard read_sha1_file() when it happens.
1169  *
1170  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1171  * find_mark(), where the mark was reloaded from an existing marks
1172  * file and is referencing an object that this fast-import process
1173  * instance did not write out to a packfile.  Callers must test for
1174  * this condition and use read_sha1_file() instead.
1175  */
1176 static void *gfi_unpack_entry(
1177         struct object_entry *oe,
1178         unsigned long *sizep)
1179 {
1180         enum object_type type;
1181         struct packed_git *p = all_packs[oe->pack_id];
1182         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1183                 /* The object is stored in the packfile we are writing to
1184                  * and we have modified it since the last time we scanned
1185                  * back to read a previously written object.  If an old
1186                  * window covered [p->pack_size, p->pack_size + 20) its
1187                  * data is stale and is not valid.  Closing all windows
1188                  * and updating the packfile length ensures we can read
1189                  * the newly written data.
1190                  */
1191                 close_pack_windows(p);
1192
1193                 /* We have to offer 20 bytes additional on the end of
1194                  * the packfile as the core unpacker code assumes the
1195                  * footer is present at the file end and must promise
1196                  * at least 20 bytes within any window it maps.  But
1197                  * we don't actually create the footer here.
1198                  */
1199                 p->pack_size = pack_size + 20;
1200         }
1201         return unpack_entry(p, oe->offset, &type, sizep);
1202 }
1203
1204 static const char *get_mode(const char *str, uint16_t *modep)
1205 {
1206         unsigned char c;
1207         uint16_t mode = 0;
1208
1209         while ((c = *str++) != ' ') {
1210                 if (c < '0' || c > '7')
1211                         return NULL;
1212                 mode = (mode << 3) + (c - '0');
1213         }
1214         *modep = mode;
1215         return str;
1216 }
1217
1218 static void load_tree(struct tree_entry *root)
1219 {
1220         unsigned char *sha1 = root->versions[1].sha1;
1221         struct object_entry *myoe;
1222         struct tree_content *t;
1223         unsigned long size;
1224         char *buf;
1225         const char *c;
1226
1227         root->tree = t = new_tree_content(8);
1228         if (is_null_sha1(sha1))
1229                 return;
1230
1231         myoe = find_object(sha1);
1232         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1233                 if (myoe->type != OBJ_TREE)
1234                         die("Not a tree: %s", sha1_to_hex(sha1));
1235                 t->delta_depth = myoe->depth;
1236                 buf = gfi_unpack_entry(myoe, &size);
1237                 if (!buf)
1238                         die("Can't load tree %s", sha1_to_hex(sha1));
1239         } else {
1240                 enum object_type type;
1241                 buf = read_sha1_file(sha1, &type, &size);
1242                 if (!buf || type != OBJ_TREE)
1243                         die("Can't load tree %s", sha1_to_hex(sha1));
1244         }
1245
1246         c = buf;
1247         while (c != (buf + size)) {
1248                 struct tree_entry *e = new_tree_entry();
1249
1250                 if (t->entry_count == t->entry_capacity)
1251                         root->tree = t = grow_tree_content(t, t->entry_count);
1252                 t->entries[t->entry_count++] = e;
1253
1254                 e->tree = NULL;
1255                 c = get_mode(c, &e->versions[1].mode);
1256                 if (!c)
1257                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1258                 e->versions[0].mode = e->versions[1].mode;
1259                 e->name = to_atom(c, strlen(c));
1260                 c += e->name->str_len + 1;
1261                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1262                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1263                 c += 20;
1264         }
1265         free(buf);
1266 }
1267
1268 static int tecmp0 (const void *_a, const void *_b)
1269 {
1270         struct tree_entry *a = *((struct tree_entry**)_a);
1271         struct tree_entry *b = *((struct tree_entry**)_b);
1272         return base_name_compare(
1273                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1274                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1275 }
1276
1277 static int tecmp1 (const void *_a, const void *_b)
1278 {
1279         struct tree_entry *a = *((struct tree_entry**)_a);
1280         struct tree_entry *b = *((struct tree_entry**)_b);
1281         return base_name_compare(
1282                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1283                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1284 }
1285
1286 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1287 {
1288         size_t maxlen = 0;
1289         unsigned int i;
1290
1291         if (!v)
1292                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1293         else
1294                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1295
1296         for (i = 0; i < t->entry_count; i++) {
1297                 if (t->entries[i]->versions[v].mode)
1298                         maxlen += t->entries[i]->name->str_len + 34;
1299         }
1300
1301         strbuf_reset(b);
1302         strbuf_grow(b, maxlen);
1303         for (i = 0; i < t->entry_count; i++) {
1304                 struct tree_entry *e = t->entries[i];
1305                 if (!e->versions[v].mode)
1306                         continue;
1307                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1308                                         e->name->str_dat, '\0');
1309                 strbuf_add(b, e->versions[v].sha1, 20);
1310         }
1311 }
1312
1313 static void store_tree(struct tree_entry *root)
1314 {
1315         struct tree_content *t = root->tree;
1316         unsigned int i, j, del;
1317         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1318         struct object_entry *le;
1319
1320         if (!is_null_sha1(root->versions[1].sha1))
1321                 return;
1322
1323         for (i = 0; i < t->entry_count; i++) {
1324                 if (t->entries[i]->tree)
1325                         store_tree(t->entries[i]);
1326         }
1327
1328         le = find_object(root->versions[0].sha1);
1329         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1330                 mktree(t, 0, &old_tree);
1331                 lo.data = old_tree;
1332                 lo.offset = le->offset;
1333                 lo.depth = t->delta_depth;
1334         }
1335
1336         mktree(t, 1, &new_tree);
1337         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1338
1339         t->delta_depth = lo.depth;
1340         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1341                 struct tree_entry *e = t->entries[i];
1342                 if (e->versions[1].mode) {
1343                         e->versions[0].mode = e->versions[1].mode;
1344                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1345                         t->entries[j++] = e;
1346                 } else {
1347                         release_tree_entry(e);
1348                         del++;
1349                 }
1350         }
1351         t->entry_count -= del;
1352 }
1353
1354 static int tree_content_set(
1355         struct tree_entry *root,
1356         const char *p,
1357         const unsigned char *sha1,
1358         const uint16_t mode,
1359         struct tree_content *subtree)
1360 {
1361         struct tree_content *t = root->tree;
1362         const char *slash1;
1363         unsigned int i, n;
1364         struct tree_entry *e;
1365
1366         slash1 = strchr(p, '/');
1367         if (slash1)
1368                 n = slash1 - p;
1369         else
1370                 n = strlen(p);
1371         if (!n)
1372                 die("Empty path component found in input");
1373         if (!slash1 && !S_ISDIR(mode) && subtree)
1374                 die("Non-directories cannot have subtrees");
1375
1376         for (i = 0; i < t->entry_count; i++) {
1377                 e = t->entries[i];
1378                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1379                         if (!slash1) {
1380                                 if (!S_ISDIR(mode)
1381                                                 && e->versions[1].mode == mode
1382                                                 && !hashcmp(e->versions[1].sha1, sha1))
1383                                         return 0;
1384                                 e->versions[1].mode = mode;
1385                                 hashcpy(e->versions[1].sha1, sha1);
1386                                 if (e->tree)
1387                                         release_tree_content_recursive(e->tree);
1388                                 e->tree = subtree;
1389                                 hashclr(root->versions[1].sha1);
1390                                 return 1;
1391                         }
1392                         if (!S_ISDIR(e->versions[1].mode)) {
1393                                 e->tree = new_tree_content(8);
1394                                 e->versions[1].mode = S_IFDIR;
1395                         }
1396                         if (!e->tree)
1397                                 load_tree(e);
1398                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1399                                 hashclr(root->versions[1].sha1);
1400                                 return 1;
1401                         }
1402                         return 0;
1403                 }
1404         }
1405
1406         if (t->entry_count == t->entry_capacity)
1407                 root->tree = t = grow_tree_content(t, t->entry_count);
1408         e = new_tree_entry();
1409         e->name = to_atom(p, n);
1410         e->versions[0].mode = 0;
1411         hashclr(e->versions[0].sha1);
1412         t->entries[t->entry_count++] = e;
1413         if (slash1) {
1414                 e->tree = new_tree_content(8);
1415                 e->versions[1].mode = S_IFDIR;
1416                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1417         } else {
1418                 e->tree = subtree;
1419                 e->versions[1].mode = mode;
1420                 hashcpy(e->versions[1].sha1, sha1);
1421         }
1422         hashclr(root->versions[1].sha1);
1423         return 1;
1424 }
1425
1426 static int tree_content_remove(
1427         struct tree_entry *root,
1428         const char *p,
1429         struct tree_entry *backup_leaf)
1430 {
1431         struct tree_content *t = root->tree;
1432         const char *slash1;
1433         unsigned int i, n;
1434         struct tree_entry *e;
1435
1436         slash1 = strchr(p, '/');
1437         if (slash1)
1438                 n = slash1 - p;
1439         else
1440                 n = strlen(p);
1441
1442         for (i = 0; i < t->entry_count; i++) {
1443                 e = t->entries[i];
1444                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1445                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1446                                 goto del_entry;
1447                         if (!e->tree)
1448                                 load_tree(e);
1449                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1450                                 for (n = 0; n < e->tree->entry_count; n++) {
1451                                         if (e->tree->entries[n]->versions[1].mode) {
1452                                                 hashclr(root->versions[1].sha1);
1453                                                 return 1;
1454                                         }
1455                                 }
1456                                 backup_leaf = NULL;
1457                                 goto del_entry;
1458                         }
1459                         return 0;
1460                 }
1461         }
1462         return 0;
1463
1464 del_entry:
1465         if (backup_leaf)
1466                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1467         else if (e->tree)
1468                 release_tree_content_recursive(e->tree);
1469         e->tree = NULL;
1470         e->versions[1].mode = 0;
1471         hashclr(e->versions[1].sha1);
1472         hashclr(root->versions[1].sha1);
1473         return 1;
1474 }
1475
1476 static int tree_content_get(
1477         struct tree_entry *root,
1478         const char *p,
1479         struct tree_entry *leaf)
1480 {
1481         struct tree_content *t = root->tree;
1482         const char *slash1;
1483         unsigned int i, n;
1484         struct tree_entry *e;
1485
1486         slash1 = strchr(p, '/');
1487         if (slash1)
1488                 n = slash1 - p;
1489         else
1490                 n = strlen(p);
1491
1492         for (i = 0; i < t->entry_count; i++) {
1493                 e = t->entries[i];
1494                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1495                         if (!slash1) {
1496                                 memcpy(leaf, e, sizeof(*leaf));
1497                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1498                                         leaf->tree = dup_tree_content(e->tree);
1499                                 else
1500                                         leaf->tree = NULL;
1501                                 return 1;
1502                         }
1503                         if (!S_ISDIR(e->versions[1].mode))
1504                                 return 0;
1505                         if (!e->tree)
1506                                 load_tree(e);
1507                         return tree_content_get(e, slash1 + 1, leaf);
1508                 }
1509         }
1510         return 0;
1511 }
1512
1513 static int update_branch(struct branch *b)
1514 {
1515         static const char *msg = "fast-import";
1516         struct ref_lock *lock;
1517         unsigned char old_sha1[20];
1518
1519         if (is_null_sha1(b->sha1))
1520                 return 0;
1521         if (read_ref(b->name, old_sha1))
1522                 hashclr(old_sha1);
1523         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1524         if (!lock)
1525                 return error("Unable to lock %s", b->name);
1526         if (!force_update && !is_null_sha1(old_sha1)) {
1527                 struct commit *old_cmit, *new_cmit;
1528
1529                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1530                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1531                 if (!old_cmit || !new_cmit) {
1532                         unlock_ref(lock);
1533                         return error("Branch %s is missing commits.", b->name);
1534                 }
1535
1536                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1537                         unlock_ref(lock);
1538                         warning("Not updating %s"
1539                                 " (new tip %s does not contain %s)",
1540                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1541                         return -1;
1542                 }
1543         }
1544         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1545                 return error("Unable to update %s", b->name);
1546         return 0;
1547 }
1548
1549 static void dump_branches(void)
1550 {
1551         unsigned int i;
1552         struct branch *b;
1553
1554         for (i = 0; i < branch_table_sz; i++) {
1555                 for (b = branch_table[i]; b; b = b->table_next_branch)
1556                         failure |= update_branch(b);
1557         }
1558 }
1559
1560 static void dump_tags(void)
1561 {
1562         static const char *msg = "fast-import";
1563         struct tag *t;
1564         struct ref_lock *lock;
1565         char ref_name[PATH_MAX];
1566
1567         for (t = first_tag; t; t = t->next_tag) {
1568                 sprintf(ref_name, "tags/%s", t->name);
1569                 lock = lock_ref_sha1(ref_name, NULL);
1570                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1571                         failure |= error("Unable to update %s", ref_name);
1572         }
1573 }
1574
1575 static void dump_marks_helper(FILE *f,
1576         uintmax_t base,
1577         struct mark_set *m)
1578 {
1579         uintmax_t k;
1580         if (m->shift) {
1581                 for (k = 0; k < 1024; k++) {
1582                         if (m->data.sets[k])
1583                                 dump_marks_helper(f, (base + k) << m->shift,
1584                                         m->data.sets[k]);
1585                 }
1586         } else {
1587                 for (k = 0; k < 1024; k++) {
1588                         if (m->data.marked[k])
1589                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1590                                         sha1_to_hex(m->data.marked[k]->sha1));
1591                 }
1592         }
1593 }
1594
1595 static void dump_marks(void)
1596 {
1597         static struct lock_file mark_lock;
1598         int mark_fd;
1599         FILE *f;
1600
1601         if (!mark_file)
1602                 return;
1603
1604         mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1605         if (mark_fd < 0) {
1606                 failure |= error("Unable to write marks file %s: %s",
1607                         mark_file, strerror(errno));
1608                 return;
1609         }
1610
1611         f = fdopen(mark_fd, "w");
1612         if (!f) {
1613                 int saved_errno = errno;
1614                 rollback_lock_file(&mark_lock);
1615                 failure |= error("Unable to write marks file %s: %s",
1616                         mark_file, strerror(saved_errno));
1617                 return;
1618         }
1619
1620         /*
1621          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1622          * Assign -1 to the lock file descriptor so that commit_lock_file()
1623          * won't try to close() it.
1624          */
1625         mark_lock.fd = -1;
1626
1627         dump_marks_helper(f, 0, marks);
1628         if (ferror(f) || fclose(f)) {
1629                 int saved_errno = errno;
1630                 rollback_lock_file(&mark_lock);
1631                 failure |= error("Unable to write marks file %s: %s",
1632                         mark_file, strerror(saved_errno));
1633                 return;
1634         }
1635
1636         if (commit_lock_file(&mark_lock)) {
1637                 int saved_errno = errno;
1638                 rollback_lock_file(&mark_lock);
1639                 failure |= error("Unable to commit marks file %s: %s",
1640                         mark_file, strerror(saved_errno));
1641                 return;
1642         }
1643 }
1644
1645 static int read_next_command(void)
1646 {
1647         static int stdin_eof = 0;
1648
1649         if (stdin_eof) {
1650                 unread_command_buf = 0;
1651                 return EOF;
1652         }
1653
1654         do {
1655                 if (unread_command_buf) {
1656                         unread_command_buf = 0;
1657                 } else {
1658                         struct recent_command *rc;
1659
1660                         strbuf_detach(&command_buf, NULL);
1661                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1662                         if (stdin_eof)
1663                                 return EOF;
1664
1665                         rc = rc_free;
1666                         if (rc)
1667                                 rc_free = rc->next;
1668                         else {
1669                                 rc = cmd_hist.next;
1670                                 cmd_hist.next = rc->next;
1671                                 cmd_hist.next->prev = &cmd_hist;
1672                                 free(rc->buf);
1673                         }
1674
1675                         rc->buf = command_buf.buf;
1676                         rc->prev = cmd_tail;
1677                         rc->next = cmd_hist.prev;
1678                         rc->prev->next = rc;
1679                         cmd_tail = rc;
1680                 }
1681         } while (command_buf.buf[0] == '#');
1682
1683         return 0;
1684 }
1685
1686 static void skip_optional_lf(void)
1687 {
1688         int term_char = fgetc(stdin);
1689         if (term_char != '\n' && term_char != EOF)
1690                 ungetc(term_char, stdin);
1691 }
1692
1693 static void parse_mark(void)
1694 {
1695         if (!prefixcmp(command_buf.buf, "mark :")) {
1696                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1697                 read_next_command();
1698         }
1699         else
1700                 next_mark = 0;
1701 }
1702
1703 static void parse_data(struct strbuf *sb)
1704 {
1705         strbuf_reset(sb);
1706
1707         if (prefixcmp(command_buf.buf, "data "))
1708                 die("Expected 'data n' command, found: %s", command_buf.buf);
1709
1710         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1711                 char *term = xstrdup(command_buf.buf + 5 + 2);
1712                 size_t term_len = command_buf.len - 5 - 2;
1713
1714                 strbuf_detach(&command_buf, NULL);
1715                 for (;;) {
1716                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1717                                 die("EOF in data (terminator '%s' not found)", term);
1718                         if (term_len == command_buf.len
1719                                 && !strcmp(term, command_buf.buf))
1720                                 break;
1721                         strbuf_addbuf(sb, &command_buf);
1722                         strbuf_addch(sb, '\n');
1723                 }
1724                 free(term);
1725         }
1726         else {
1727                 size_t n = 0, length;
1728
1729                 length = strtoul(command_buf.buf + 5, NULL, 10);
1730
1731                 while (n < length) {
1732                         size_t s = strbuf_fread(sb, length - n, stdin);
1733                         if (!s && feof(stdin))
1734                                 die("EOF in data (%lu bytes remaining)",
1735                                         (unsigned long)(length - n));
1736                         n += s;
1737                 }
1738         }
1739
1740         skip_optional_lf();
1741 }
1742
1743 static int validate_raw_date(const char *src, char *result, int maxlen)
1744 {
1745         const char *orig_src = src;
1746         char *endp;
1747         unsigned long num;
1748
1749         errno = 0;
1750
1751         num = strtoul(src, &endp, 10);
1752         /* NEEDSWORK: perhaps check for reasonable values? */
1753         if (errno || endp == src || *endp != ' ')
1754                 return -1;
1755
1756         src = endp + 1;
1757         if (*src != '-' && *src != '+')
1758                 return -1;
1759
1760         num = strtoul(src + 1, &endp, 10);
1761         if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1762             1400 < num)
1763                 return -1;
1764
1765         strcpy(result, orig_src);
1766         return 0;
1767 }
1768
1769 static char *parse_ident(const char *buf)
1770 {
1771         const char *gt;
1772         size_t name_len;
1773         char *ident;
1774
1775         gt = strrchr(buf, '>');
1776         if (!gt)
1777                 die("Missing > in ident string: %s", buf);
1778         gt++;
1779         if (*gt != ' ')
1780                 die("Missing space after > in ident string: %s", buf);
1781         gt++;
1782         name_len = gt - buf;
1783         ident = xmalloc(name_len + 24);
1784         strncpy(ident, buf, name_len);
1785
1786         switch (whenspec) {
1787         case WHENSPEC_RAW:
1788                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1789                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1790                 break;
1791         case WHENSPEC_RFC2822:
1792                 if (parse_date(gt, ident + name_len, 24) < 0)
1793                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1794                 break;
1795         case WHENSPEC_NOW:
1796                 if (strcmp("now", gt))
1797                         die("Date in ident must be 'now': %s", buf);
1798                 datestamp(ident + name_len, 24);
1799                 break;
1800         }
1801
1802         return ident;
1803 }
1804
1805 static void parse_new_blob(void)
1806 {
1807         static struct strbuf buf = STRBUF_INIT;
1808
1809         read_next_command();
1810         parse_mark();
1811         parse_data(&buf);
1812         store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1813 }
1814
1815 static void unload_one_branch(void)
1816 {
1817         while (cur_active_branches
1818                 && cur_active_branches >= max_active_branches) {
1819                 uintmax_t min_commit = ULONG_MAX;
1820                 struct branch *e, *l = NULL, *p = NULL;
1821
1822                 for (e = active_branches; e; e = e->active_next_branch) {
1823                         if (e->last_commit < min_commit) {
1824                                 p = l;
1825                                 min_commit = e->last_commit;
1826                         }
1827                         l = e;
1828                 }
1829
1830                 if (p) {
1831                         e = p->active_next_branch;
1832                         p->active_next_branch = e->active_next_branch;
1833                 } else {
1834                         e = active_branches;
1835                         active_branches = e->active_next_branch;
1836                 }
1837                 e->active = 0;
1838                 e->active_next_branch = NULL;
1839                 if (e->branch_tree.tree) {
1840                         release_tree_content_recursive(e->branch_tree.tree);
1841                         e->branch_tree.tree = NULL;
1842                 }
1843                 cur_active_branches--;
1844         }
1845 }
1846
1847 static void load_branch(struct branch *b)
1848 {
1849         load_tree(&b->branch_tree);
1850         if (!b->active) {
1851                 b->active = 1;
1852                 b->active_next_branch = active_branches;
1853                 active_branches = b;
1854                 cur_active_branches++;
1855                 branch_load_count++;
1856         }
1857 }
1858
1859 static void file_change_m(struct branch *b)
1860 {
1861         const char *p = command_buf.buf + 2;
1862         static struct strbuf uq = STRBUF_INIT;
1863         const char *endp;
1864         struct object_entry *oe = oe;
1865         unsigned char sha1[20];
1866         uint16_t mode, inline_data = 0;
1867
1868         p = get_mode(p, &mode);
1869         if (!p)
1870                 die("Corrupt mode: %s", command_buf.buf);
1871         switch (mode) {
1872         case 0644:
1873         case 0755:
1874                 mode |= S_IFREG;
1875         case S_IFREG | 0644:
1876         case S_IFREG | 0755:
1877         case S_IFLNK:
1878         case S_IFGITLINK:
1879                 /* ok */
1880                 break;
1881         default:
1882                 die("Corrupt mode: %s", command_buf.buf);
1883         }
1884
1885         if (*p == ':') {
1886                 char *x;
1887                 oe = find_mark(strtoumax(p + 1, &x, 10));
1888                 hashcpy(sha1, oe->sha1);
1889                 p = x;
1890         } else if (!prefixcmp(p, "inline")) {
1891                 inline_data = 1;
1892                 p += 6;
1893         } else {
1894                 if (get_sha1_hex(p, sha1))
1895                         die("Invalid SHA1: %s", command_buf.buf);
1896                 oe = find_object(sha1);
1897                 p += 40;
1898         }
1899         if (*p++ != ' ')
1900                 die("Missing space after SHA1: %s", command_buf.buf);
1901
1902         strbuf_reset(&uq);
1903         if (!unquote_c_style(&uq, p, &endp)) {
1904                 if (*endp)
1905                         die("Garbage after path in: %s", command_buf.buf);
1906                 p = uq.buf;
1907         }
1908
1909         if (S_ISGITLINK(mode)) {
1910                 if (inline_data)
1911                         die("Git links cannot be specified 'inline': %s",
1912                                 command_buf.buf);
1913                 else if (oe) {
1914                         if (oe->type != OBJ_COMMIT)
1915                                 die("Not a commit (actually a %s): %s",
1916                                         typename(oe->type), command_buf.buf);
1917                 }
1918                 /*
1919                  * Accept the sha1 without checking; it expected to be in
1920                  * another repository.
1921                  */
1922         } else if (inline_data) {
1923                 static struct strbuf buf = STRBUF_INIT;
1924
1925                 if (p != uq.buf) {
1926                         strbuf_addstr(&uq, p);
1927                         p = uq.buf;
1928                 }
1929                 read_next_command();
1930                 parse_data(&buf);
1931                 store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1932         } else if (oe) {
1933                 if (oe->type != OBJ_BLOB)
1934                         die("Not a blob (actually a %s): %s",
1935                                 typename(oe->type), command_buf.buf);
1936         } else {
1937                 enum object_type type = sha1_object_info(sha1, NULL);
1938                 if (type < 0)
1939                         die("Blob not found: %s", command_buf.buf);
1940                 if (type != OBJ_BLOB)
1941                         die("Not a blob (actually a %s): %s",
1942                             typename(type), command_buf.buf);
1943         }
1944
1945         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
1946 }
1947
1948 static void file_change_d(struct branch *b)
1949 {
1950         const char *p = command_buf.buf + 2;
1951         static struct strbuf uq = STRBUF_INIT;
1952         const char *endp;
1953
1954         strbuf_reset(&uq);
1955         if (!unquote_c_style(&uq, p, &endp)) {
1956                 if (*endp)
1957                         die("Garbage after path in: %s", command_buf.buf);
1958                 p = uq.buf;
1959         }
1960         tree_content_remove(&b->branch_tree, p, NULL);
1961 }
1962
1963 static void file_change_cr(struct branch *b, int rename)
1964 {
1965         const char *s, *d;
1966         static struct strbuf s_uq = STRBUF_INIT;
1967         static struct strbuf d_uq = STRBUF_INIT;
1968         const char *endp;
1969         struct tree_entry leaf;
1970
1971         s = command_buf.buf + 2;
1972         strbuf_reset(&s_uq);
1973         if (!unquote_c_style(&s_uq, s, &endp)) {
1974                 if (*endp != ' ')
1975                         die("Missing space after source: %s", command_buf.buf);
1976         } else {
1977                 endp = strchr(s, ' ');
1978                 if (!endp)
1979                         die("Missing space after source: %s", command_buf.buf);
1980                 strbuf_add(&s_uq, s, endp - s);
1981         }
1982         s = s_uq.buf;
1983
1984         endp++;
1985         if (!*endp)
1986                 die("Missing dest: %s", command_buf.buf);
1987
1988         d = endp;
1989         strbuf_reset(&d_uq);
1990         if (!unquote_c_style(&d_uq, d, &endp)) {
1991                 if (*endp)
1992                         die("Garbage after dest in: %s", command_buf.buf);
1993                 d = d_uq.buf;
1994         }
1995
1996         memset(&leaf, 0, sizeof(leaf));
1997         if (rename)
1998                 tree_content_remove(&b->branch_tree, s, &leaf);
1999         else
2000                 tree_content_get(&b->branch_tree, s, &leaf);
2001         if (!leaf.versions[1].mode)
2002                 die("Path %s not in branch", s);
2003         tree_content_set(&b->branch_tree, d,
2004                 leaf.versions[1].sha1,
2005                 leaf.versions[1].mode,
2006                 leaf.tree);
2007 }
2008
2009 static void file_change_deleteall(struct branch *b)
2010 {
2011         release_tree_content_recursive(b->branch_tree.tree);
2012         hashclr(b->branch_tree.versions[0].sha1);
2013         hashclr(b->branch_tree.versions[1].sha1);
2014         load_tree(&b->branch_tree);
2015 }
2016
2017 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2018 {
2019         if (!buf || size < 46)
2020                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2021         if (memcmp("tree ", buf, 5)
2022                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2023                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2024         hashcpy(b->branch_tree.versions[0].sha1,
2025                 b->branch_tree.versions[1].sha1);
2026 }
2027
2028 static void parse_from_existing(struct branch *b)
2029 {
2030         if (is_null_sha1(b->sha1)) {
2031                 hashclr(b->branch_tree.versions[0].sha1);
2032                 hashclr(b->branch_tree.versions[1].sha1);
2033         } else {
2034                 unsigned long size;
2035                 char *buf;
2036
2037                 buf = read_object_with_reference(b->sha1,
2038                         commit_type, &size, b->sha1);
2039                 parse_from_commit(b, buf, size);
2040                 free(buf);
2041         }
2042 }
2043
2044 static int parse_from(struct branch *b)
2045 {
2046         const char *from;
2047         struct branch *s;
2048
2049         if (prefixcmp(command_buf.buf, "from "))
2050                 return 0;
2051
2052         if (b->branch_tree.tree) {
2053                 release_tree_content_recursive(b->branch_tree.tree);
2054                 b->branch_tree.tree = NULL;
2055         }
2056
2057         from = strchr(command_buf.buf, ' ') + 1;
2058         s = lookup_branch(from);
2059         if (b == s)
2060                 die("Can't create a branch from itself: %s", b->name);
2061         else if (s) {
2062                 unsigned char *t = s->branch_tree.versions[1].sha1;
2063                 hashcpy(b->sha1, s->sha1);
2064                 hashcpy(b->branch_tree.versions[0].sha1, t);
2065                 hashcpy(b->branch_tree.versions[1].sha1, t);
2066         } else if (*from == ':') {
2067                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2068                 struct object_entry *oe = find_mark(idnum);
2069                 if (oe->type != OBJ_COMMIT)
2070                         die("Mark :%" PRIuMAX " not a commit", idnum);
2071                 hashcpy(b->sha1, oe->sha1);
2072                 if (oe->pack_id != MAX_PACK_ID) {
2073                         unsigned long size;
2074                         char *buf = gfi_unpack_entry(oe, &size);
2075                         parse_from_commit(b, buf, size);
2076                         free(buf);
2077                 } else
2078                         parse_from_existing(b);
2079         } else if (!get_sha1(from, b->sha1))
2080                 parse_from_existing(b);
2081         else
2082                 die("Invalid ref name or SHA1 expression: %s", from);
2083
2084         read_next_command();
2085         return 1;
2086 }
2087
2088 static struct hash_list *parse_merge(unsigned int *count)
2089 {
2090         struct hash_list *list = NULL, *n, *e = e;
2091         const char *from;
2092         struct branch *s;
2093
2094         *count = 0;
2095         while (!prefixcmp(command_buf.buf, "merge ")) {
2096                 from = strchr(command_buf.buf, ' ') + 1;
2097                 n = xmalloc(sizeof(*n));
2098                 s = lookup_branch(from);
2099                 if (s)
2100                         hashcpy(n->sha1, s->sha1);
2101                 else if (*from == ':') {
2102                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2103                         struct object_entry *oe = find_mark(idnum);
2104                         if (oe->type != OBJ_COMMIT)
2105                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2106                         hashcpy(n->sha1, oe->sha1);
2107                 } else if (!get_sha1(from, n->sha1)) {
2108                         unsigned long size;
2109                         char *buf = read_object_with_reference(n->sha1,
2110                                 commit_type, &size, n->sha1);
2111                         if (!buf || size < 46)
2112                                 die("Not a valid commit: %s", from);
2113                         free(buf);
2114                 } else
2115                         die("Invalid ref name or SHA1 expression: %s", from);
2116
2117                 n->next = NULL;
2118                 if (list)
2119                         e->next = n;
2120                 else
2121                         list = n;
2122                 e = n;
2123                 (*count)++;
2124                 read_next_command();
2125         }
2126         return list;
2127 }
2128
2129 static void parse_new_commit(void)
2130 {
2131         static struct strbuf msg = STRBUF_INIT;
2132         struct branch *b;
2133         char *sp;
2134         char *author = NULL;
2135         char *committer = NULL;
2136         struct hash_list *merge_list = NULL;
2137         unsigned int merge_count;
2138
2139         /* Obtain the branch name from the rest of our command */
2140         sp = strchr(command_buf.buf, ' ') + 1;
2141         b = lookup_branch(sp);
2142         if (!b)
2143                 b = new_branch(sp);
2144
2145         read_next_command();
2146         parse_mark();
2147         if (!prefixcmp(command_buf.buf, "author ")) {
2148                 author = parse_ident(command_buf.buf + 7);
2149                 read_next_command();
2150         }
2151         if (!prefixcmp(command_buf.buf, "committer ")) {
2152                 committer = parse_ident(command_buf.buf + 10);
2153                 read_next_command();
2154         }
2155         if (!committer)
2156                 die("Expected committer but didn't get one");
2157         parse_data(&msg);
2158         read_next_command();
2159         parse_from(b);
2160         merge_list = parse_merge(&merge_count);
2161
2162         /* ensure the branch is active/loaded */
2163         if (!b->branch_tree.tree || !max_active_branches) {
2164                 unload_one_branch();
2165                 load_branch(b);
2166         }
2167
2168         /* file_change* */
2169         while (command_buf.len > 0) {
2170                 if (!prefixcmp(command_buf.buf, "M "))
2171                         file_change_m(b);
2172                 else if (!prefixcmp(command_buf.buf, "D "))
2173                         file_change_d(b);
2174                 else if (!prefixcmp(command_buf.buf, "R "))
2175                         file_change_cr(b, 1);
2176                 else if (!prefixcmp(command_buf.buf, "C "))
2177                         file_change_cr(b, 0);
2178                 else if (!strcmp("deleteall", command_buf.buf))
2179                         file_change_deleteall(b);
2180                 else {
2181                         unread_command_buf = 1;
2182                         break;
2183                 }
2184                 if (read_next_command() == EOF)
2185                         break;
2186         }
2187
2188         /* build the tree and the commit */
2189         store_tree(&b->branch_tree);
2190         hashcpy(b->branch_tree.versions[0].sha1,
2191                 b->branch_tree.versions[1].sha1);
2192
2193         strbuf_reset(&new_data);
2194         strbuf_addf(&new_data, "tree %s\n",
2195                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2196         if (!is_null_sha1(b->sha1))
2197                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2198         while (merge_list) {
2199                 struct hash_list *next = merge_list->next;
2200                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2201                 free(merge_list);
2202                 merge_list = next;
2203         }
2204         strbuf_addf(&new_data,
2205                 "author %s\n"
2206                 "committer %s\n"
2207                 "\n",
2208                 author ? author : committer, committer);
2209         strbuf_addbuf(&new_data, &msg);
2210         free(author);
2211         free(committer);
2212
2213         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2214                 b->pack_id = pack_id;
2215         b->last_commit = object_count_by_type[OBJ_COMMIT];
2216 }
2217
2218 static void parse_new_tag(void)
2219 {
2220         static struct strbuf msg = STRBUF_INIT;
2221         char *sp;
2222         const char *from;
2223         char *tagger;
2224         struct branch *s;
2225         struct tag *t;
2226         uintmax_t from_mark = 0;
2227         unsigned char sha1[20];
2228
2229         /* Obtain the new tag name from the rest of our command */
2230         sp = strchr(command_buf.buf, ' ') + 1;
2231         t = pool_alloc(sizeof(struct tag));
2232         t->next_tag = NULL;
2233         t->name = pool_strdup(sp);
2234         if (last_tag)
2235                 last_tag->next_tag = t;
2236         else
2237                 first_tag = t;
2238         last_tag = t;
2239         read_next_command();
2240
2241         /* from ... */
2242         if (prefixcmp(command_buf.buf, "from "))
2243                 die("Expected from command, got %s", command_buf.buf);
2244         from = strchr(command_buf.buf, ' ') + 1;
2245         s = lookup_branch(from);
2246         if (s) {
2247                 hashcpy(sha1, s->sha1);
2248         } else if (*from == ':') {
2249                 struct object_entry *oe;
2250                 from_mark = strtoumax(from + 1, NULL, 10);
2251                 oe = find_mark(from_mark);
2252                 if (oe->type != OBJ_COMMIT)
2253                         die("Mark :%" PRIuMAX " not a commit", from_mark);
2254                 hashcpy(sha1, oe->sha1);
2255         } else if (!get_sha1(from, sha1)) {
2256                 unsigned long size;
2257                 char *buf;
2258
2259                 buf = read_object_with_reference(sha1,
2260                         commit_type, &size, sha1);
2261                 if (!buf || size < 46)
2262                         die("Not a valid commit: %s", from);
2263                 free(buf);
2264         } else
2265                 die("Invalid ref name or SHA1 expression: %s", from);
2266         read_next_command();
2267
2268         /* tagger ... */
2269         if (!prefixcmp(command_buf.buf, "tagger ")) {
2270                 tagger = parse_ident(command_buf.buf + 7);
2271                 read_next_command();
2272         } else
2273                 tagger = NULL;
2274
2275         /* tag payload/message */
2276         parse_data(&msg);
2277
2278         /* build the tag object */
2279         strbuf_reset(&new_data);
2280
2281         strbuf_addf(&new_data,
2282                     "object %s\n"
2283                     "type %s\n"
2284                     "tag %s\n",
2285                     sha1_to_hex(sha1), commit_type, t->name);
2286         if (tagger)
2287                 strbuf_addf(&new_data,
2288                             "tagger %s\n", tagger);
2289         strbuf_addch(&new_data, '\n');
2290         strbuf_addbuf(&new_data, &msg);
2291         free(tagger);
2292
2293         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2294                 t->pack_id = MAX_PACK_ID;
2295         else
2296                 t->pack_id = pack_id;
2297 }
2298
2299 static void parse_reset_branch(void)
2300 {
2301         struct branch *b;
2302         char *sp;
2303
2304         /* Obtain the branch name from the rest of our command */
2305         sp = strchr(command_buf.buf, ' ') + 1;
2306         b = lookup_branch(sp);
2307         if (b) {
2308                 hashclr(b->sha1);
2309                 hashclr(b->branch_tree.versions[0].sha1);
2310                 hashclr(b->branch_tree.versions[1].sha1);
2311                 if (b->branch_tree.tree) {
2312                         release_tree_content_recursive(b->branch_tree.tree);
2313                         b->branch_tree.tree = NULL;
2314                 }
2315         }
2316         else
2317                 b = new_branch(sp);
2318         read_next_command();
2319         parse_from(b);
2320         if (command_buf.len > 0)
2321                 unread_command_buf = 1;
2322 }
2323
2324 static void parse_checkpoint(void)
2325 {
2326         if (object_count) {
2327                 cycle_packfile();
2328                 dump_branches();
2329                 dump_tags();
2330                 dump_marks();
2331         }
2332         skip_optional_lf();
2333 }
2334
2335 static void parse_progress(void)
2336 {
2337         fwrite(command_buf.buf, 1, command_buf.len, stdout);
2338         fputc('\n', stdout);
2339         fflush(stdout);
2340         skip_optional_lf();
2341 }
2342
2343 static void import_marks(const char *input_file)
2344 {
2345         char line[512];
2346         FILE *f = fopen(input_file, "r");
2347         if (!f)
2348                 die_errno("cannot read '%s'", input_file);
2349         while (fgets(line, sizeof(line), f)) {
2350                 uintmax_t mark;
2351                 char *end;
2352                 unsigned char sha1[20];
2353                 struct object_entry *e;
2354
2355                 end = strchr(line, '\n');
2356                 if (line[0] != ':' || !end)
2357                         die("corrupt mark line: %s", line);
2358                 *end = 0;
2359                 mark = strtoumax(line + 1, &end, 10);
2360                 if (!mark || end == line + 1
2361                         || *end != ' ' || get_sha1(end + 1, sha1))
2362                         die("corrupt mark line: %s", line);
2363                 e = find_object(sha1);
2364                 if (!e) {
2365                         enum object_type type = sha1_object_info(sha1, NULL);
2366                         if (type < 0)
2367                                 die("object not found: %s", sha1_to_hex(sha1));
2368                         e = insert_object(sha1);
2369                         e->type = type;
2370                         e->pack_id = MAX_PACK_ID;
2371                         e->offset = 1; /* just not zero! */
2372                 }
2373                 insert_mark(mark, e);
2374         }
2375         fclose(f);
2376 }
2377
2378 static int git_pack_config(const char *k, const char *v, void *cb)
2379 {
2380         if (!strcmp(k, "pack.depth")) {
2381                 max_depth = git_config_int(k, v);
2382                 if (max_depth > MAX_DEPTH)
2383                         max_depth = MAX_DEPTH;
2384                 return 0;
2385         }
2386         if (!strcmp(k, "pack.compression")) {
2387                 int level = git_config_int(k, v);
2388                 if (level == -1)
2389                         level = Z_DEFAULT_COMPRESSION;
2390                 else if (level < 0 || level > Z_BEST_COMPRESSION)
2391                         die("bad pack compression level %d", level);
2392                 pack_compression_level = level;
2393                 pack_compression_seen = 1;
2394                 return 0;
2395         }
2396         return git_default_config(k, v, cb);
2397 }
2398
2399 static const char fast_import_usage[] =
2400 "git fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2401
2402 int main(int argc, const char **argv)
2403 {
2404         unsigned int i, show_stats = 1;
2405
2406         git_extract_argv0_path(argv[0]);
2407
2408         setup_git_directory();
2409         git_config(git_pack_config, NULL);
2410         if (!pack_compression_seen && core_compression_seen)
2411                 pack_compression_level = core_compression_level;
2412
2413         alloc_objects(object_entry_alloc);
2414         strbuf_init(&command_buf, 0);
2415         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2416         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2417         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2418         marks = pool_calloc(1, sizeof(struct mark_set));
2419
2420         for (i = 1; i < argc; i++) {
2421                 const char *a = argv[i];
2422
2423                 if (*a != '-' || !strcmp(a, "--"))
2424                         break;
2425                 else if (!prefixcmp(a, "--date-format=")) {
2426                         const char *fmt = a + 14;
2427                         if (!strcmp(fmt, "raw"))
2428                                 whenspec = WHENSPEC_RAW;
2429                         else if (!strcmp(fmt, "rfc2822"))
2430                                 whenspec = WHENSPEC_RFC2822;
2431                         else if (!strcmp(fmt, "now"))
2432                                 whenspec = WHENSPEC_NOW;
2433                         else
2434                                 die("unknown --date-format argument %s", fmt);
2435                 }
2436                 else if (!prefixcmp(a, "--max-pack-size="))
2437                         max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2438                 else if (!prefixcmp(a, "--depth=")) {
2439                         max_depth = strtoul(a + 8, NULL, 0);
2440                         if (max_depth > MAX_DEPTH)
2441                                 die("--depth cannot exceed %u", MAX_DEPTH);
2442                 }
2443                 else if (!prefixcmp(a, "--active-branches="))
2444                         max_active_branches = strtoul(a + 18, NULL, 0);
2445                 else if (!prefixcmp(a, "--import-marks="))
2446                         import_marks(a + 15);
2447                 else if (!prefixcmp(a, "--export-marks="))
2448                         mark_file = a + 15;
2449                 else if (!prefixcmp(a, "--export-pack-edges=")) {
2450                         if (pack_edges)
2451                                 fclose(pack_edges);
2452                         pack_edges = fopen(a + 20, "a");
2453                         if (!pack_edges)
2454                                 die_errno("Cannot open '%s'", a + 20);
2455                 } else if (!strcmp(a, "--force"))
2456                         force_update = 1;
2457                 else if (!strcmp(a, "--quiet"))
2458                         show_stats = 0;
2459                 else if (!strcmp(a, "--stats"))
2460                         show_stats = 1;
2461                 else
2462                         die("unknown option %s", a);
2463         }
2464         if (i != argc)
2465                 usage(fast_import_usage);
2466
2467         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2468         for (i = 0; i < (cmd_save - 1); i++)
2469                 rc_free[i].next = &rc_free[i + 1];
2470         rc_free[cmd_save - 1].next = NULL;
2471
2472         prepare_packed_git();
2473         start_packfile();
2474         set_die_routine(die_nicely);
2475         while (read_next_command() != EOF) {
2476                 if (!strcmp("blob", command_buf.buf))
2477                         parse_new_blob();
2478                 else if (!prefixcmp(command_buf.buf, "commit "))
2479                         parse_new_commit();
2480                 else if (!prefixcmp(command_buf.buf, "tag "))
2481                         parse_new_tag();
2482                 else if (!prefixcmp(command_buf.buf, "reset "))
2483                         parse_reset_branch();
2484                 else if (!strcmp("checkpoint", command_buf.buf))
2485                         parse_checkpoint();
2486                 else if (!prefixcmp(command_buf.buf, "progress "))
2487                         parse_progress();
2488                 else
2489                         die("Unsupported command: %s", command_buf.buf);
2490         }
2491         end_packfile();
2492
2493         dump_branches();
2494         dump_tags();
2495         unkeep_all_packs();
2496         dump_marks();
2497
2498         if (pack_edges)
2499                 fclose(pack_edges);
2500
2501         if (show_stats) {
2502                 uintmax_t total_count = 0, duplicate_count = 0;
2503                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2504                         total_count += object_count_by_type[i];
2505                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2506                         duplicate_count += duplicate_count_by_type[i];
2507
2508                 fprintf(stderr, "%s statistics:\n", argv[0]);
2509                 fprintf(stderr, "---------------------------------------------------------------------\n");
2510                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2511                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2512                 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]);
2513                 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]);
2514                 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]);
2515                 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]);
2516                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2517                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2518                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2519                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2520                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2521                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2522                 fprintf(stderr, "---------------------------------------------------------------------\n");
2523                 pack_report();
2524                 fprintf(stderr, "---------------------------------------------------------------------\n");
2525                 fprintf(stderr, "\n");
2526         }
2527
2528         return failure ? 1 : 0;
2529 }