config: add test cases for empty value and no value config variables.
[git] / fast-import.c
1 /*
2 Format of STDIN stream:
3
4   stream ::= cmd*;
5
6   cmd ::= new_blob
7         | new_commit
8         | new_tag
9         | reset_branch
10         | checkpoint
11         | progress
12         ;
13
14   new_blob ::= 'blob' lf
15     mark?
16     file_content;
17   file_content ::= data;
18
19   new_commit ::= 'commit' sp ref_str lf
20     mark?
21     ('author' sp name '<' email '>' when lf)?
22     'committer' sp name '<' email '>' when lf
23     commit_msg
24     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
25     ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
26     file_change*
27     lf?;
28   commit_msg ::= data;
29
30   file_change ::= file_clr
31     | file_del
32     | file_rnm
33     | file_cpy
34     | file_obm
35     | file_inm;
36   file_clr ::= 'deleteall' lf;
37   file_del ::= 'D' sp path_str lf;
38   file_rnm ::= 'R' sp path_str sp path_str lf;
39   file_cpy ::= 'C' sp path_str sp path_str lf;
40   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
41   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
42     data;
43
44   new_tag ::= 'tag' sp tag_str lf
45     'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
46     'tagger' sp name '<' email '>' when lf
47     tag_msg;
48   tag_msg ::= data;
49
50   reset_branch ::= 'reset' sp ref_str lf
51     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
52     lf?;
53
54   checkpoint ::= 'checkpoint' lf
55     lf?;
56
57   progress ::= 'progress' sp not_lf* lf
58     lf?;
59
60      # note: the first idnum in a stream should be 1 and subsequent
61      # idnums should not have gaps between values as this will cause
62      # the stream parser to reserve space for the gapped values.  An
63      # idnum can be updated in the future to a new object by issuing
64      # a new mark directive with the old idnum.
65      #
66   mark ::= 'mark' sp idnum lf;
67   data ::= (delimited_data | exact_data)
68     lf?;
69
70     # note: delim may be any string but must not contain lf.
71     # data_line may contain any data but must not be exactly
72     # delim.
73   delimited_data ::= 'data' sp '<<' delim lf
74     (data_line lf)*
75     delim lf;
76
77      # note: declen indicates the length of binary_data in bytes.
78      # declen does not include the lf preceeding the binary data.
79      #
80   exact_data ::= 'data' sp declen lf
81     binary_data;
82
83      # note: quoted strings are C-style quoting supporting \c for
84      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
85      # is the signed byte value in octal.  Note that the only
86      # characters which must actually be escaped to protect the
87      # stream formatting is: \, " and LF.  Otherwise these values
88      # are UTF8.
89      #
90   ref_str     ::= ref;
91   sha1exp_str ::= sha1exp;
92   tag_str     ::= tag;
93   path_str    ::= path    | '"' quoted(path)    '"' ;
94   mode        ::= '100644' | '644'
95                 | '100755' | '755'
96                 | '120000'
97                 ;
98
99   declen ::= # unsigned 32 bit value, ascii base10 notation;
100   bigint ::= # unsigned integer value, ascii base10 notation;
101   binary_data ::= # file content, not interpreted;
102
103   when         ::= raw_when | rfc2822_when;
104   raw_when     ::= ts sp tz;
105   rfc2822_when ::= # Valid RFC 2822 date and time;
106
107   sp ::= # ASCII space character;
108   lf ::= # ASCII newline (LF) character;
109
110      # note: a colon (':') must precede the numerical value assigned to
111      # an idnum.  This is to distinguish it from a ref or tag name as
112      # GIT does not permit ':' in ref or tag strings.
113      #
114   idnum   ::= ':' bigint;
115   path    ::= # GIT style file path, e.g. "a/b/c";
116   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
117   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
118   sha1exp ::= # Any valid GIT SHA1 expression;
119   hexsha1 ::= # SHA1 in hexadecimal format;
120
121      # note: name and email are UTF8 strings, however name must not
122      # contain '<' or lf and email must not contain any of the
123      # following: '<', '>', lf.
124      #
125   name  ::= # valid GIT author/committer name;
126   email ::= # valid GIT author/committer email;
127   ts    ::= # time since the epoch in seconds, ascii base10 notation;
128   tz    ::= # GIT style timezone;
129
130      # note: comments may appear anywhere in the input, except
131      # within a data command.  Any form of the data command
132      # always escapes the related input from comment processing.
133      #
134      # In case it is not clear, the '#' that starts the comment
135      # must be the first character on that the line (an lf have
136      # preceeded it).
137      #
138   comment ::= '#' not_lf* lf;
139   not_lf  ::= # Any byte that is not ASCII newline (LF);
140 */
141
142 #include "builtin.h"
143 #include "cache.h"
144 #include "object.h"
145 #include "blob.h"
146 #include "tree.h"
147 #include "commit.h"
148 #include "delta.h"
149 #include "pack.h"
150 #include "refs.h"
151 #include "csum-file.h"
152 #include "quote.h"
153
154 #define PACK_ID_BITS 16
155 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
156 #define DEPTH_BITS 13
157 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
158
159 struct object_entry
160 {
161         struct object_entry *next;
162         uint32_t offset;
163         uint32_t type : TYPE_BITS,
164                 pack_id : PACK_ID_BITS,
165                 depth : DEPTH_BITS;
166         unsigned char sha1[20];
167 };
168
169 struct object_entry_pool
170 {
171         struct object_entry_pool *next_pool;
172         struct object_entry *next_free;
173         struct object_entry *end;
174         struct object_entry entries[FLEX_ARRAY]; /* more */
175 };
176
177 struct mark_set
178 {
179         union {
180                 struct object_entry *marked[1024];
181                 struct mark_set *sets[1024];
182         } data;
183         unsigned int shift;
184 };
185
186 struct last_object
187 {
188         struct strbuf data;
189         uint32_t offset;
190         unsigned int depth;
191         unsigned no_swap : 1;
192 };
193
194 struct mem_pool
195 {
196         struct mem_pool *next_pool;
197         char *next_free;
198         char *end;
199         uintmax_t space[FLEX_ARRAY]; /* more */
200 };
201
202 struct atom_str
203 {
204         struct atom_str *next_atom;
205         unsigned short str_len;
206         char str_dat[FLEX_ARRAY]; /* more */
207 };
208
209 struct tree_content;
210 struct tree_entry
211 {
212         struct tree_content *tree;
213         struct atom_str* name;
214         struct tree_entry_ms
215         {
216                 uint16_t mode;
217                 unsigned char sha1[20];
218         } versions[2];
219 };
220
221 struct tree_content
222 {
223         unsigned int entry_capacity; /* must match avail_tree_content */
224         unsigned int entry_count;
225         unsigned int delta_depth;
226         struct tree_entry *entries[FLEX_ARRAY]; /* more */
227 };
228
229 struct avail_tree_content
230 {
231         unsigned int entry_capacity; /* must match tree_content */
232         struct avail_tree_content *next_avail;
233 };
234
235 struct branch
236 {
237         struct branch *table_next_branch;
238         struct branch *active_next_branch;
239         const char *name;
240         struct tree_entry branch_tree;
241         uintmax_t last_commit;
242         unsigned active : 1;
243         unsigned pack_id : PACK_ID_BITS;
244         unsigned char sha1[20];
245 };
246
247 struct tag
248 {
249         struct tag *next_tag;
250         const char *name;
251         unsigned int pack_id;
252         unsigned char sha1[20];
253 };
254
255 struct hash_list
256 {
257         struct hash_list *next;
258         unsigned char sha1[20];
259 };
260
261 typedef enum {
262         WHENSPEC_RAW = 1,
263         WHENSPEC_RFC2822,
264         WHENSPEC_NOW,
265 } whenspec_type;
266
267 struct recent_command
268 {
269         struct recent_command *prev;
270         struct recent_command *next;
271         char *buf;
272 };
273
274 /* Configured limits on output */
275 static unsigned long max_depth = 10;
276 static off_t max_packsize = (1LL << 32) - 1;
277 static int force_update;
278 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
279 static int pack_compression_seen;
280
281 /* Stats and misc. counters */
282 static uintmax_t alloc_count;
283 static uintmax_t marks_set_count;
284 static uintmax_t object_count_by_type[1 << TYPE_BITS];
285 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
286 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
287 static unsigned long object_count;
288 static unsigned long branch_count;
289 static unsigned long branch_load_count;
290 static int failure;
291 static FILE *pack_edges;
292
293 /* Memory pools */
294 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
295 static size_t total_allocd;
296 static struct mem_pool *mem_pool;
297
298 /* Atom management */
299 static unsigned int atom_table_sz = 4451;
300 static unsigned int atom_cnt;
301 static struct atom_str **atom_table;
302
303 /* The .pack file being generated */
304 static unsigned int pack_id;
305 static struct packed_git *pack_data;
306 static struct packed_git **all_packs;
307 static unsigned long pack_size;
308
309 /* Table of objects we've written. */
310 static unsigned int object_entry_alloc = 5000;
311 static struct object_entry_pool *blocks;
312 static struct object_entry *object_table[1 << 16];
313 static struct mark_set *marks;
314 static const char* mark_file;
315
316 /* Our last blob */
317 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
318
319 /* Tree management */
320 static unsigned int tree_entry_alloc = 1000;
321 static void *avail_tree_entry;
322 static unsigned int avail_tree_table_sz = 100;
323 static struct avail_tree_content **avail_tree_table;
324 static struct strbuf old_tree = STRBUF_INIT;
325 static struct strbuf new_tree = STRBUF_INIT;
326
327 /* Branch data */
328 static unsigned long max_active_branches = 5;
329 static unsigned long cur_active_branches;
330 static unsigned long branch_table_sz = 1039;
331 static struct branch **branch_table;
332 static struct branch *active_branches;
333
334 /* Tag data */
335 static struct tag *first_tag;
336 static struct tag *last_tag;
337
338 /* Input stream parsing */
339 static whenspec_type whenspec = WHENSPEC_RAW;
340 static struct strbuf command_buf = STRBUF_INIT;
341 static int unread_command_buf;
342 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
343 static struct recent_command *cmd_tail = &cmd_hist;
344 static struct recent_command *rc_free;
345 static unsigned int cmd_save = 100;
346 static uintmax_t next_mark;
347 static struct strbuf new_data = STRBUF_INIT;
348
349 static void write_branch_report(FILE *rpt, struct branch *b)
350 {
351         fprintf(rpt, "%s:\n", b->name);
352
353         fprintf(rpt, "  status      :");
354         if (b->active)
355                 fputs(" active", rpt);
356         if (b->branch_tree.tree)
357                 fputs(" loaded", rpt);
358         if (is_null_sha1(b->branch_tree.versions[1].sha1))
359                 fputs(" dirty", rpt);
360         fputc('\n', rpt);
361
362         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
363         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
364         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
365         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
366
367         fputs("  last pack   : ", rpt);
368         if (b->pack_id < MAX_PACK_ID)
369                 fprintf(rpt, "%u", b->pack_id);
370         fputc('\n', rpt);
371
372         fputc('\n', rpt);
373 }
374
375 static void write_crash_report(const char *err)
376 {
377         char *loc = git_path("fast_import_crash_%d", getpid());
378         FILE *rpt = fopen(loc, "w");
379         struct branch *b;
380         unsigned long lu;
381         struct recent_command *rc;
382
383         if (!rpt) {
384                 error("can't write crash report %s: %s", loc, strerror(errno));
385                 return;
386         }
387
388         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
389
390         fprintf(rpt, "fast-import crash report:\n");
391         fprintf(rpt, "    fast-import process: %d\n", getpid());
392         fprintf(rpt, "    parent process     : %d\n", getppid());
393         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
394         fputc('\n', rpt);
395
396         fputs("fatal: ", rpt);
397         fputs(err, rpt);
398         fputc('\n', rpt);
399
400         fputc('\n', rpt);
401         fputs("Most Recent Commands Before Crash\n", rpt);
402         fputs("---------------------------------\n", rpt);
403         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
404                 if (rc->next == &cmd_hist)
405                         fputs("* ", rpt);
406                 else
407                         fputs("  ", rpt);
408                 fputs(rc->buf, rpt);
409                 fputc('\n', rpt);
410         }
411
412         fputc('\n', rpt);
413         fputs("Active Branch LRU\n", rpt);
414         fputs("-----------------\n", rpt);
415         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
416                 cur_active_branches,
417                 max_active_branches);
418         fputc('\n', rpt);
419         fputs("  pos  clock name\n", rpt);
420         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
421         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
422                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
423                         ++lu, b->last_commit, b->name);
424
425         fputc('\n', rpt);
426         fputs("Inactive Branches\n", rpt);
427         fputs("-----------------\n", rpt);
428         for (lu = 0; lu < branch_table_sz; lu++) {
429                 for (b = branch_table[lu]; b; b = b->table_next_branch)
430                         write_branch_report(rpt, b);
431         }
432
433         fputc('\n', rpt);
434         fputs("-------------------\n", rpt);
435         fputs("END OF CRASH REPORT\n", rpt);
436         fclose(rpt);
437 }
438
439 static NORETURN void die_nicely(const char *err, va_list params)
440 {
441         static int zombie;
442         char message[2 * PATH_MAX];
443
444         vsnprintf(message, sizeof(message), err, params);
445         fputs("fatal: ", stderr);
446         fputs(message, stderr);
447         fputc('\n', stderr);
448
449         if (!zombie) {
450                 zombie = 1;
451                 write_crash_report(message);
452         }
453         exit(128);
454 }
455
456 static void alloc_objects(unsigned int cnt)
457 {
458         struct object_entry_pool *b;
459
460         b = xmalloc(sizeof(struct object_entry_pool)
461                 + cnt * sizeof(struct object_entry));
462         b->next_pool = blocks;
463         b->next_free = b->entries;
464         b->end = b->entries + cnt;
465         blocks = b;
466         alloc_count += cnt;
467 }
468
469 static struct object_entry *new_object(unsigned char *sha1)
470 {
471         struct object_entry *e;
472
473         if (blocks->next_free == blocks->end)
474                 alloc_objects(object_entry_alloc);
475
476         e = blocks->next_free++;
477         hashcpy(e->sha1, sha1);
478         return e;
479 }
480
481 static struct object_entry *find_object(unsigned char *sha1)
482 {
483         unsigned int h = sha1[0] << 8 | sha1[1];
484         struct object_entry *e;
485         for (e = object_table[h]; e; e = e->next)
486                 if (!hashcmp(sha1, e->sha1))
487                         return e;
488         return NULL;
489 }
490
491 static struct object_entry *insert_object(unsigned char *sha1)
492 {
493         unsigned int h = sha1[0] << 8 | sha1[1];
494         struct object_entry *e = object_table[h];
495         struct object_entry *p = NULL;
496
497         while (e) {
498                 if (!hashcmp(sha1, e->sha1))
499                         return e;
500                 p = e;
501                 e = e->next;
502         }
503
504         e = new_object(sha1);
505         e->next = NULL;
506         e->offset = 0;
507         if (p)
508                 p->next = e;
509         else
510                 object_table[h] = e;
511         return e;
512 }
513
514 static unsigned int hc_str(const char *s, size_t len)
515 {
516         unsigned int r = 0;
517         while (len-- > 0)
518                 r = r * 31 + *s++;
519         return r;
520 }
521
522 static void *pool_alloc(size_t len)
523 {
524         struct mem_pool *p;
525         void *r;
526
527         for (p = mem_pool; p; p = p->next_pool)
528                 if ((p->end - p->next_free >= len))
529                         break;
530
531         if (!p) {
532                 if (len >= (mem_pool_alloc/2)) {
533                         total_allocd += len;
534                         return xmalloc(len);
535                 }
536                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
537                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
538                 p->next_pool = mem_pool;
539                 p->next_free = (char *) p->space;
540                 p->end = p->next_free + mem_pool_alloc;
541                 mem_pool = p;
542         }
543
544         r = p->next_free;
545         /* round out to a 'uintmax_t' alignment */
546         if (len & (sizeof(uintmax_t) - 1))
547                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
548         p->next_free += len;
549         return r;
550 }
551
552 static void *pool_calloc(size_t count, size_t size)
553 {
554         size_t len = count * size;
555         void *r = pool_alloc(len);
556         memset(r, 0, len);
557         return r;
558 }
559
560 static char *pool_strdup(const char *s)
561 {
562         char *r = pool_alloc(strlen(s) + 1);
563         strcpy(r, s);
564         return r;
565 }
566
567 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
568 {
569         struct mark_set *s = marks;
570         while ((idnum >> s->shift) >= 1024) {
571                 s = pool_calloc(1, sizeof(struct mark_set));
572                 s->shift = marks->shift + 10;
573                 s->data.sets[0] = marks;
574                 marks = s;
575         }
576         while (s->shift) {
577                 uintmax_t i = idnum >> s->shift;
578                 idnum -= i << s->shift;
579                 if (!s->data.sets[i]) {
580                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
581                         s->data.sets[i]->shift = s->shift - 10;
582                 }
583                 s = s->data.sets[i];
584         }
585         if (!s->data.marked[idnum])
586                 marks_set_count++;
587         s->data.marked[idnum] = oe;
588 }
589
590 static struct object_entry *find_mark(uintmax_t idnum)
591 {
592         uintmax_t orig_idnum = idnum;
593         struct mark_set *s = marks;
594         struct object_entry *oe = NULL;
595         if ((idnum >> s->shift) < 1024) {
596                 while (s && s->shift) {
597                         uintmax_t i = idnum >> s->shift;
598                         idnum -= i << s->shift;
599                         s = s->data.sets[i];
600                 }
601                 if (s)
602                         oe = s->data.marked[idnum];
603         }
604         if (!oe)
605                 die("mark :%" PRIuMAX " not declared", orig_idnum);
606         return oe;
607 }
608
609 static struct atom_str *to_atom(const char *s, unsigned short len)
610 {
611         unsigned int hc = hc_str(s, len) % atom_table_sz;
612         struct atom_str *c;
613
614         for (c = atom_table[hc]; c; c = c->next_atom)
615                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
616                         return c;
617
618         c = pool_alloc(sizeof(struct atom_str) + len + 1);
619         c->str_len = len;
620         strncpy(c->str_dat, s, len);
621         c->str_dat[len] = 0;
622         c->next_atom = atom_table[hc];
623         atom_table[hc] = c;
624         atom_cnt++;
625         return c;
626 }
627
628 static struct branch *lookup_branch(const char *name)
629 {
630         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
631         struct branch *b;
632
633         for (b = branch_table[hc]; b; b = b->table_next_branch)
634                 if (!strcmp(name, b->name))
635                         return b;
636         return NULL;
637 }
638
639 static struct branch *new_branch(const char *name)
640 {
641         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
642         struct branch* b = lookup_branch(name);
643
644         if (b)
645                 die("Invalid attempt to create duplicate branch: %s", name);
646         switch (check_ref_format(name)) {
647         case 0: break; /* its valid */
648         case CHECK_REF_FORMAT_ONELEVEL:
649                 break; /* valid, but too few '/', allow anyway */
650         default:
651                 die("Branch name doesn't conform to GIT standards: %s", name);
652         }
653
654         b = pool_calloc(1, sizeof(struct branch));
655         b->name = pool_strdup(name);
656         b->table_next_branch = branch_table[hc];
657         b->branch_tree.versions[0].mode = S_IFDIR;
658         b->branch_tree.versions[1].mode = S_IFDIR;
659         b->active = 0;
660         b->pack_id = MAX_PACK_ID;
661         branch_table[hc] = b;
662         branch_count++;
663         return b;
664 }
665
666 static unsigned int hc_entries(unsigned int cnt)
667 {
668         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
669         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
670 }
671
672 static struct tree_content *new_tree_content(unsigned int cnt)
673 {
674         struct avail_tree_content *f, *l = NULL;
675         struct tree_content *t;
676         unsigned int hc = hc_entries(cnt);
677
678         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
679                 if (f->entry_capacity >= cnt)
680                         break;
681
682         if (f) {
683                 if (l)
684                         l->next_avail = f->next_avail;
685                 else
686                         avail_tree_table[hc] = f->next_avail;
687         } else {
688                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
689                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
690                 f->entry_capacity = cnt;
691         }
692
693         t = (struct tree_content*)f;
694         t->entry_count = 0;
695         t->delta_depth = 0;
696         return t;
697 }
698
699 static void release_tree_entry(struct tree_entry *e);
700 static void release_tree_content(struct tree_content *t)
701 {
702         struct avail_tree_content *f = (struct avail_tree_content*)t;
703         unsigned int hc = hc_entries(f->entry_capacity);
704         f->next_avail = avail_tree_table[hc];
705         avail_tree_table[hc] = f;
706 }
707
708 static void release_tree_content_recursive(struct tree_content *t)
709 {
710         unsigned int i;
711         for (i = 0; i < t->entry_count; i++)
712                 release_tree_entry(t->entries[i]);
713         release_tree_content(t);
714 }
715
716 static struct tree_content *grow_tree_content(
717         struct tree_content *t,
718         int amt)
719 {
720         struct tree_content *r = new_tree_content(t->entry_count + amt);
721         r->entry_count = t->entry_count;
722         r->delta_depth = t->delta_depth;
723         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
724         release_tree_content(t);
725         return r;
726 }
727
728 static struct tree_entry *new_tree_entry(void)
729 {
730         struct tree_entry *e;
731
732         if (!avail_tree_entry) {
733                 unsigned int n = tree_entry_alloc;
734                 total_allocd += n * sizeof(struct tree_entry);
735                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
736                 while (n-- > 1) {
737                         *((void**)e) = e + 1;
738                         e++;
739                 }
740                 *((void**)e) = NULL;
741         }
742
743         e = avail_tree_entry;
744         avail_tree_entry = *((void**)e);
745         return e;
746 }
747
748 static void release_tree_entry(struct tree_entry *e)
749 {
750         if (e->tree)
751                 release_tree_content_recursive(e->tree);
752         *((void**)e) = avail_tree_entry;
753         avail_tree_entry = e;
754 }
755
756 static struct tree_content *dup_tree_content(struct tree_content *s)
757 {
758         struct tree_content *d;
759         struct tree_entry *a, *b;
760         unsigned int i;
761
762         if (!s)
763                 return NULL;
764         d = new_tree_content(s->entry_count);
765         for (i = 0; i < s->entry_count; i++) {
766                 a = s->entries[i];
767                 b = new_tree_entry();
768                 memcpy(b, a, sizeof(*a));
769                 if (a->tree && is_null_sha1(b->versions[1].sha1))
770                         b->tree = dup_tree_content(a->tree);
771                 else
772                         b->tree = NULL;
773                 d->entries[i] = b;
774         }
775         d->entry_count = s->entry_count;
776         d->delta_depth = s->delta_depth;
777
778         return d;
779 }
780
781 static void start_packfile(void)
782 {
783         static char tmpfile[PATH_MAX];
784         struct packed_git *p;
785         struct pack_header hdr;
786         int pack_fd;
787
788         snprintf(tmpfile, sizeof(tmpfile),
789                 "%s/tmp_pack_XXXXXX", get_object_directory());
790         pack_fd = xmkstemp(tmpfile);
791         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
792         strcpy(p->pack_name, tmpfile);
793         p->pack_fd = pack_fd;
794
795         hdr.hdr_signature = htonl(PACK_SIGNATURE);
796         hdr.hdr_version = htonl(2);
797         hdr.hdr_entries = 0;
798         write_or_die(p->pack_fd, &hdr, sizeof(hdr));
799
800         pack_data = p;
801         pack_size = sizeof(hdr);
802         object_count = 0;
803
804         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
805         all_packs[pack_id] = p;
806 }
807
808 static int oecmp (const void *a_, const void *b_)
809 {
810         struct object_entry *a = *((struct object_entry**)a_);
811         struct object_entry *b = *((struct object_entry**)b_);
812         return hashcmp(a->sha1, b->sha1);
813 }
814
815 static char *create_index(void)
816 {
817         static char tmpfile[PATH_MAX];
818         SHA_CTX ctx;
819         struct sha1file *f;
820         struct object_entry **idx, **c, **last, *e;
821         struct object_entry_pool *o;
822         uint32_t array[256];
823         int i, idx_fd;
824
825         /* Build the sorted table of object IDs. */
826         idx = xmalloc(object_count * sizeof(struct object_entry*));
827         c = idx;
828         for (o = blocks; o; o = o->next_pool)
829                 for (e = o->next_free; e-- != o->entries;)
830                         if (pack_id == e->pack_id)
831                                 *c++ = e;
832         last = idx + object_count;
833         if (c != last)
834                 die("internal consistency error creating the index");
835         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
836
837         /* Generate the fan-out array. */
838         c = idx;
839         for (i = 0; i < 256; i++) {
840                 struct object_entry **next = c;;
841                 while (next < last) {
842                         if ((*next)->sha1[0] != i)
843                                 break;
844                         next++;
845                 }
846                 array[i] = htonl(next - idx);
847                 c = next;
848         }
849
850         snprintf(tmpfile, sizeof(tmpfile),
851                 "%s/tmp_idx_XXXXXX", get_object_directory());
852         idx_fd = xmkstemp(tmpfile);
853         f = sha1fd(idx_fd, tmpfile);
854         sha1write(f, array, 256 * sizeof(int));
855         SHA1_Init(&ctx);
856         for (c = idx; c != last; c++) {
857                 uint32_t offset = htonl((*c)->offset);
858                 sha1write(f, &offset, 4);
859                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
860                 SHA1_Update(&ctx, (*c)->sha1, 20);
861         }
862         sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
863         sha1close(f, NULL, 1);
864         free(idx);
865         SHA1_Final(pack_data->sha1, &ctx);
866         return tmpfile;
867 }
868
869 static char *keep_pack(char *curr_index_name)
870 {
871         static char name[PATH_MAX];
872         static const char *keep_msg = "fast-import";
873         int keep_fd;
874
875         chmod(pack_data->pack_name, 0444);
876         chmod(curr_index_name, 0444);
877
878         snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
879                  get_object_directory(), sha1_to_hex(pack_data->sha1));
880         keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
881         if (keep_fd < 0)
882                 die("cannot create keep file");
883         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
884         if (close(keep_fd))
885                 die("failed to write keep file");
886
887         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
888                  get_object_directory(), sha1_to_hex(pack_data->sha1));
889         if (move_temp_to_file(pack_data->pack_name, name))
890                 die("cannot store pack file");
891
892         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
893                  get_object_directory(), sha1_to_hex(pack_data->sha1));
894         if (move_temp_to_file(curr_index_name, name))
895                 die("cannot store index file");
896         return name;
897 }
898
899 static void unkeep_all_packs(void)
900 {
901         static char name[PATH_MAX];
902         int k;
903
904         for (k = 0; k < pack_id; k++) {
905                 struct packed_git *p = all_packs[k];
906                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
907                          get_object_directory(), sha1_to_hex(p->sha1));
908                 unlink(name);
909         }
910 }
911
912 static void end_packfile(void)
913 {
914         struct packed_git *old_p = pack_data, *new_p;
915
916         if (object_count) {
917                 char *idx_name;
918                 int i;
919                 struct branch *b;
920                 struct tag *t;
921
922                 close_pack_windows(pack_data);
923                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
924                                     pack_data->pack_name, object_count);
925                 close(pack_data->pack_fd);
926                 idx_name = keep_pack(create_index());
927
928                 /* Register the packfile with core git's machinary. */
929                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
930                 if (!new_p)
931                         die("core git rejected index %s", idx_name);
932                 all_packs[pack_id] = new_p;
933                 install_packed_git(new_p);
934
935                 /* Print the boundary */
936                 if (pack_edges) {
937                         fprintf(pack_edges, "%s:", new_p->pack_name);
938                         for (i = 0; i < branch_table_sz; i++) {
939                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
940                                         if (b->pack_id == pack_id)
941                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
942                                 }
943                         }
944                         for (t = first_tag; t; t = t->next_tag) {
945                                 if (t->pack_id == pack_id)
946                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
947                         }
948                         fputc('\n', pack_edges);
949                         fflush(pack_edges);
950                 }
951
952                 pack_id++;
953         }
954         else
955                 unlink(old_p->pack_name);
956         free(old_p);
957
958         /* We can't carry a delta across packfiles. */
959         strbuf_release(&last_blob.data);
960         last_blob.offset = 0;
961         last_blob.depth = 0;
962 }
963
964 static void cycle_packfile(void)
965 {
966         end_packfile();
967         start_packfile();
968 }
969
970 static size_t encode_header(
971         enum object_type type,
972         size_t size,
973         unsigned char *hdr)
974 {
975         int n = 1;
976         unsigned char c;
977
978         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
979                 die("bad type %d", type);
980
981         c = (type << 4) | (size & 15);
982         size >>= 4;
983         while (size) {
984                 *hdr++ = c | 0x80;
985                 c = size & 0x7f;
986                 size >>= 7;
987                 n++;
988         }
989         *hdr = c;
990         return n;
991 }
992
993 static int store_object(
994         enum object_type type,
995         struct strbuf *dat,
996         struct last_object *last,
997         unsigned char *sha1out,
998         uintmax_t mark)
999 {
1000         void *out, *delta;
1001         struct object_entry *e;
1002         unsigned char hdr[96];
1003         unsigned char sha1[20];
1004         unsigned long hdrlen, deltalen;
1005         SHA_CTX c;
1006         z_stream s;
1007
1008         hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1009                 (unsigned long)dat->len) + 1;
1010         SHA1_Init(&c);
1011         SHA1_Update(&c, hdr, hdrlen);
1012         SHA1_Update(&c, dat->buf, dat->len);
1013         SHA1_Final(sha1, &c);
1014         if (sha1out)
1015                 hashcpy(sha1out, sha1);
1016
1017         e = insert_object(sha1);
1018         if (mark)
1019                 insert_mark(mark, e);
1020         if (e->offset) {
1021                 duplicate_count_by_type[type]++;
1022                 return 1;
1023         } else if (find_sha1_pack(sha1, packed_git)) {
1024                 e->type = type;
1025                 e->pack_id = MAX_PACK_ID;
1026                 e->offset = 1; /* just not zero! */
1027                 duplicate_count_by_type[type]++;
1028                 return 1;
1029         }
1030
1031         if (last && last->data.buf && last->depth < max_depth) {
1032                 delta = diff_delta(last->data.buf, last->data.len,
1033                         dat->buf, dat->len,
1034                         &deltalen, 0);
1035                 if (delta && deltalen >= dat->len) {
1036                         free(delta);
1037                         delta = NULL;
1038                 }
1039         } else
1040                 delta = NULL;
1041
1042         memset(&s, 0, sizeof(s));
1043         deflateInit(&s, pack_compression_level);
1044         if (delta) {
1045                 s.next_in = delta;
1046                 s.avail_in = deltalen;
1047         } else {
1048                 s.next_in = (void *)dat->buf;
1049                 s.avail_in = dat->len;
1050         }
1051         s.avail_out = deflateBound(&s, s.avail_in);
1052         s.next_out = out = xmalloc(s.avail_out);
1053         while (deflate(&s, Z_FINISH) == Z_OK)
1054                 /* nothing */;
1055         deflateEnd(&s);
1056
1057         /* Determine if we should auto-checkpoint. */
1058         if ((pack_size + 60 + s.total_out) > max_packsize
1059                 || (pack_size + 60 + s.total_out) < pack_size) {
1060
1061                 /* This new object needs to *not* have the current pack_id. */
1062                 e->pack_id = pack_id + 1;
1063                 cycle_packfile();
1064
1065                 /* We cannot carry a delta into the new pack. */
1066                 if (delta) {
1067                         free(delta);
1068                         delta = NULL;
1069
1070                         memset(&s, 0, sizeof(s));
1071                         deflateInit(&s, pack_compression_level);
1072                         s.next_in = (void *)dat->buf;
1073                         s.avail_in = dat->len;
1074                         s.avail_out = deflateBound(&s, s.avail_in);
1075                         s.next_out = out = xrealloc(out, s.avail_out);
1076                         while (deflate(&s, Z_FINISH) == Z_OK)
1077                                 /* nothing */;
1078                         deflateEnd(&s);
1079                 }
1080         }
1081
1082         e->type = type;
1083         e->pack_id = pack_id;
1084         e->offset = pack_size;
1085         object_count++;
1086         object_count_by_type[type]++;
1087
1088         if (delta) {
1089                 unsigned long ofs = e->offset - last->offset;
1090                 unsigned pos = sizeof(hdr) - 1;
1091
1092                 delta_count_by_type[type]++;
1093                 e->depth = last->depth + 1;
1094
1095                 hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1096                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1097                 pack_size += hdrlen;
1098
1099                 hdr[pos] = ofs & 127;
1100                 while (ofs >>= 7)
1101                         hdr[--pos] = 128 | (--ofs & 127);
1102                 write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1103                 pack_size += sizeof(hdr) - pos;
1104         } else {
1105                 e->depth = 0;
1106                 hdrlen = encode_header(type, dat->len, hdr);
1107                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
1108                 pack_size += hdrlen;
1109         }
1110
1111         write_or_die(pack_data->pack_fd, out, s.total_out);
1112         pack_size += s.total_out;
1113
1114         free(out);
1115         free(delta);
1116         if (last) {
1117                 if (last->no_swap) {
1118                         last->data = *dat;
1119                 } else {
1120                         strbuf_swap(&last->data, dat);
1121                 }
1122                 last->offset = e->offset;
1123                 last->depth = e->depth;
1124         }
1125         return 0;
1126 }
1127
1128 /* All calls must be guarded by find_object() or find_mark() to
1129  * ensure the 'struct object_entry' passed was written by this
1130  * process instance.  We unpack the entry by the offset, avoiding
1131  * the need for the corresponding .idx file.  This unpacking rule
1132  * works because we only use OBJ_REF_DELTA within the packfiles
1133  * created by fast-import.
1134  *
1135  * oe must not be NULL.  Such an oe usually comes from giving
1136  * an unknown SHA-1 to find_object() or an undefined mark to
1137  * find_mark().  Callers must test for this condition and use
1138  * the standard read_sha1_file() when it happens.
1139  *
1140  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1141  * find_mark(), where the mark was reloaded from an existing marks
1142  * file and is referencing an object that this fast-import process
1143  * instance did not write out to a packfile.  Callers must test for
1144  * this condition and use read_sha1_file() instead.
1145  */
1146 static void *gfi_unpack_entry(
1147         struct object_entry *oe,
1148         unsigned long *sizep)
1149 {
1150         enum object_type type;
1151         struct packed_git *p = all_packs[oe->pack_id];
1152         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1153                 /* The object is stored in the packfile we are writing to
1154                  * and we have modified it since the last time we scanned
1155                  * back to read a previously written object.  If an old
1156                  * window covered [p->pack_size, p->pack_size + 20) its
1157                  * data is stale and is not valid.  Closing all windows
1158                  * and updating the packfile length ensures we can read
1159                  * the newly written data.
1160                  */
1161                 close_pack_windows(p);
1162
1163                 /* We have to offer 20 bytes additional on the end of
1164                  * the packfile as the core unpacker code assumes the
1165                  * footer is present at the file end and must promise
1166                  * at least 20 bytes within any window it maps.  But
1167                  * we don't actually create the footer here.
1168                  */
1169                 p->pack_size = pack_size + 20;
1170         }
1171         return unpack_entry(p, oe->offset, &type, sizep);
1172 }
1173
1174 static const char *get_mode(const char *str, uint16_t *modep)
1175 {
1176         unsigned char c;
1177         uint16_t mode = 0;
1178
1179         while ((c = *str++) != ' ') {
1180                 if (c < '0' || c > '7')
1181                         return NULL;
1182                 mode = (mode << 3) + (c - '0');
1183         }
1184         *modep = mode;
1185         return str;
1186 }
1187
1188 static void load_tree(struct tree_entry *root)
1189 {
1190         unsigned char* sha1 = root->versions[1].sha1;
1191         struct object_entry *myoe;
1192         struct tree_content *t;
1193         unsigned long size;
1194         char *buf;
1195         const char *c;
1196
1197         root->tree = t = new_tree_content(8);
1198         if (is_null_sha1(sha1))
1199                 return;
1200
1201         myoe = find_object(sha1);
1202         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1203                 if (myoe->type != OBJ_TREE)
1204                         die("Not a tree: %s", sha1_to_hex(sha1));
1205                 t->delta_depth = myoe->depth;
1206                 buf = gfi_unpack_entry(myoe, &size);
1207         } else {
1208                 enum object_type type;
1209                 buf = read_sha1_file(sha1, &type, &size);
1210                 if (!buf || type != OBJ_TREE)
1211                         die("Can't load tree %s", sha1_to_hex(sha1));
1212         }
1213
1214         c = buf;
1215         while (c != (buf + size)) {
1216                 struct tree_entry *e = new_tree_entry();
1217
1218                 if (t->entry_count == t->entry_capacity)
1219                         root->tree = t = grow_tree_content(t, t->entry_count);
1220                 t->entries[t->entry_count++] = e;
1221
1222                 e->tree = NULL;
1223                 c = get_mode(c, &e->versions[1].mode);
1224                 if (!c)
1225                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1226                 e->versions[0].mode = e->versions[1].mode;
1227                 e->name = to_atom(c, strlen(c));
1228                 c += e->name->str_len + 1;
1229                 hashcpy(e->versions[0].sha1, (unsigned char*)c);
1230                 hashcpy(e->versions[1].sha1, (unsigned char*)c);
1231                 c += 20;
1232         }
1233         free(buf);
1234 }
1235
1236 static int tecmp0 (const void *_a, const void *_b)
1237 {
1238         struct tree_entry *a = *((struct tree_entry**)_a);
1239         struct tree_entry *b = *((struct tree_entry**)_b);
1240         return base_name_compare(
1241                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1242                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1243 }
1244
1245 static int tecmp1 (const void *_a, const void *_b)
1246 {
1247         struct tree_entry *a = *((struct tree_entry**)_a);
1248         struct tree_entry *b = *((struct tree_entry**)_b);
1249         return base_name_compare(
1250                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1251                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1252 }
1253
1254 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1255 {
1256         size_t maxlen = 0;
1257         unsigned int i;
1258
1259         if (!v)
1260                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1261         else
1262                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1263
1264         for (i = 0; i < t->entry_count; i++) {
1265                 if (t->entries[i]->versions[v].mode)
1266                         maxlen += t->entries[i]->name->str_len + 34;
1267         }
1268
1269         strbuf_reset(b);
1270         strbuf_grow(b, maxlen);
1271         for (i = 0; i < t->entry_count; i++) {
1272                 struct tree_entry *e = t->entries[i];
1273                 if (!e->versions[v].mode)
1274                         continue;
1275                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1276                                         e->name->str_dat, '\0');
1277                 strbuf_add(b, e->versions[v].sha1, 20);
1278         }
1279 }
1280
1281 static void store_tree(struct tree_entry *root)
1282 {
1283         struct tree_content *t = root->tree;
1284         unsigned int i, j, del;
1285         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1286         struct object_entry *le;
1287
1288         if (!is_null_sha1(root->versions[1].sha1))
1289                 return;
1290
1291         for (i = 0; i < t->entry_count; i++) {
1292                 if (t->entries[i]->tree)
1293                         store_tree(t->entries[i]);
1294         }
1295
1296         le = find_object(root->versions[0].sha1);
1297         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1298                 mktree(t, 0, &old_tree);
1299                 lo.data = old_tree;
1300                 lo.offset = le->offset;
1301                 lo.depth = t->delta_depth;
1302         }
1303
1304         mktree(t, 1, &new_tree);
1305         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1306
1307         t->delta_depth = lo.depth;
1308         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1309                 struct tree_entry *e = t->entries[i];
1310                 if (e->versions[1].mode) {
1311                         e->versions[0].mode = e->versions[1].mode;
1312                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1313                         t->entries[j++] = e;
1314                 } else {
1315                         release_tree_entry(e);
1316                         del++;
1317                 }
1318         }
1319         t->entry_count -= del;
1320 }
1321
1322 static int tree_content_set(
1323         struct tree_entry *root,
1324         const char *p,
1325         const unsigned char *sha1,
1326         const uint16_t mode,
1327         struct tree_content *subtree)
1328 {
1329         struct tree_content *t = root->tree;
1330         const char *slash1;
1331         unsigned int i, n;
1332         struct tree_entry *e;
1333
1334         slash1 = strchr(p, '/');
1335         if (slash1)
1336                 n = slash1 - p;
1337         else
1338                 n = strlen(p);
1339         if (!n)
1340                 die("Empty path component found in input");
1341         if (!slash1 && !S_ISDIR(mode) && subtree)
1342                 die("Non-directories cannot have subtrees");
1343
1344         for (i = 0; i < t->entry_count; i++) {
1345                 e = t->entries[i];
1346                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1347                         if (!slash1) {
1348                                 if (!S_ISDIR(mode)
1349                                                 && e->versions[1].mode == mode
1350                                                 && !hashcmp(e->versions[1].sha1, sha1))
1351                                         return 0;
1352                                 e->versions[1].mode = mode;
1353                                 hashcpy(e->versions[1].sha1, sha1);
1354                                 if (e->tree)
1355                                         release_tree_content_recursive(e->tree);
1356                                 e->tree = subtree;
1357                                 hashclr(root->versions[1].sha1);
1358                                 return 1;
1359                         }
1360                         if (!S_ISDIR(e->versions[1].mode)) {
1361                                 e->tree = new_tree_content(8);
1362                                 e->versions[1].mode = S_IFDIR;
1363                         }
1364                         if (!e->tree)
1365                                 load_tree(e);
1366                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1367                                 hashclr(root->versions[1].sha1);
1368                                 return 1;
1369                         }
1370                         return 0;
1371                 }
1372         }
1373
1374         if (t->entry_count == t->entry_capacity)
1375                 root->tree = t = grow_tree_content(t, t->entry_count);
1376         e = new_tree_entry();
1377         e->name = to_atom(p, n);
1378         e->versions[0].mode = 0;
1379         hashclr(e->versions[0].sha1);
1380         t->entries[t->entry_count++] = e;
1381         if (slash1) {
1382                 e->tree = new_tree_content(8);
1383                 e->versions[1].mode = S_IFDIR;
1384                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1385         } else {
1386                 e->tree = subtree;
1387                 e->versions[1].mode = mode;
1388                 hashcpy(e->versions[1].sha1, sha1);
1389         }
1390         hashclr(root->versions[1].sha1);
1391         return 1;
1392 }
1393
1394 static int tree_content_remove(
1395         struct tree_entry *root,
1396         const char *p,
1397         struct tree_entry *backup_leaf)
1398 {
1399         struct tree_content *t = root->tree;
1400         const char *slash1;
1401         unsigned int i, n;
1402         struct tree_entry *e;
1403
1404         slash1 = strchr(p, '/');
1405         if (slash1)
1406                 n = slash1 - p;
1407         else
1408                 n = strlen(p);
1409
1410         for (i = 0; i < t->entry_count; i++) {
1411                 e = t->entries[i];
1412                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1413                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1414                                 goto del_entry;
1415                         if (!e->tree)
1416                                 load_tree(e);
1417                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1418                                 for (n = 0; n < e->tree->entry_count; n++) {
1419                                         if (e->tree->entries[n]->versions[1].mode) {
1420                                                 hashclr(root->versions[1].sha1);
1421                                                 return 1;
1422                                         }
1423                                 }
1424                                 backup_leaf = NULL;
1425                                 goto del_entry;
1426                         }
1427                         return 0;
1428                 }
1429         }
1430         return 0;
1431
1432 del_entry:
1433         if (backup_leaf)
1434                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1435         else if (e->tree)
1436                 release_tree_content_recursive(e->tree);
1437         e->tree = NULL;
1438         e->versions[1].mode = 0;
1439         hashclr(e->versions[1].sha1);
1440         hashclr(root->versions[1].sha1);
1441         return 1;
1442 }
1443
1444 static int tree_content_get(
1445         struct tree_entry *root,
1446         const char *p,
1447         struct tree_entry *leaf)
1448 {
1449         struct tree_content *t = root->tree;
1450         const char *slash1;
1451         unsigned int i, n;
1452         struct tree_entry *e;
1453
1454         slash1 = strchr(p, '/');
1455         if (slash1)
1456                 n = slash1 - p;
1457         else
1458                 n = strlen(p);
1459
1460         for (i = 0; i < t->entry_count; i++) {
1461                 e = t->entries[i];
1462                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1463                         if (!slash1) {
1464                                 memcpy(leaf, e, sizeof(*leaf));
1465                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1466                                         leaf->tree = dup_tree_content(e->tree);
1467                                 else
1468                                         leaf->tree = NULL;
1469                                 return 1;
1470                         }
1471                         if (!S_ISDIR(e->versions[1].mode))
1472                                 return 0;
1473                         if (!e->tree)
1474                                 load_tree(e);
1475                         return tree_content_get(e, slash1 + 1, leaf);
1476                 }
1477         }
1478         return 0;
1479 }
1480
1481 static int update_branch(struct branch *b)
1482 {
1483         static const char *msg = "fast-import";
1484         struct ref_lock *lock;
1485         unsigned char old_sha1[20];
1486
1487         if (read_ref(b->name, old_sha1))
1488                 hashclr(old_sha1);
1489         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1490         if (!lock)
1491                 return error("Unable to lock %s", b->name);
1492         if (!force_update && !is_null_sha1(old_sha1)) {
1493                 struct commit *old_cmit, *new_cmit;
1494
1495                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1496                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1497                 if (!old_cmit || !new_cmit) {
1498                         unlock_ref(lock);
1499                         return error("Branch %s is missing commits.", b->name);
1500                 }
1501
1502                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1503                         unlock_ref(lock);
1504                         warning("Not updating %s"
1505                                 " (new tip %s does not contain %s)",
1506                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1507                         return -1;
1508                 }
1509         }
1510         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1511                 return error("Unable to update %s", b->name);
1512         return 0;
1513 }
1514
1515 static void dump_branches(void)
1516 {
1517         unsigned int i;
1518         struct branch *b;
1519
1520         for (i = 0; i < branch_table_sz; i++) {
1521                 for (b = branch_table[i]; b; b = b->table_next_branch)
1522                         failure |= update_branch(b);
1523         }
1524 }
1525
1526 static void dump_tags(void)
1527 {
1528         static const char *msg = "fast-import";
1529         struct tag *t;
1530         struct ref_lock *lock;
1531         char ref_name[PATH_MAX];
1532
1533         for (t = first_tag; t; t = t->next_tag) {
1534                 sprintf(ref_name, "tags/%s", t->name);
1535                 lock = lock_ref_sha1(ref_name, NULL);
1536                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1537                         failure |= error("Unable to update %s", ref_name);
1538         }
1539 }
1540
1541 static void dump_marks_helper(FILE *f,
1542         uintmax_t base,
1543         struct mark_set *m)
1544 {
1545         uintmax_t k;
1546         if (m->shift) {
1547                 for (k = 0; k < 1024; k++) {
1548                         if (m->data.sets[k])
1549                                 dump_marks_helper(f, (base + k) << m->shift,
1550                                         m->data.sets[k]);
1551                 }
1552         } else {
1553                 for (k = 0; k < 1024; k++) {
1554                         if (m->data.marked[k])
1555                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1556                                         sha1_to_hex(m->data.marked[k]->sha1));
1557                 }
1558         }
1559 }
1560
1561 static void dump_marks(void)
1562 {
1563         static struct lock_file mark_lock;
1564         int mark_fd;
1565         FILE *f;
1566
1567         if (!mark_file)
1568                 return;
1569
1570         mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1571         if (mark_fd < 0) {
1572                 failure |= error("Unable to write marks file %s: %s",
1573                         mark_file, strerror(errno));
1574                 return;
1575         }
1576
1577         f = fdopen(mark_fd, "w");
1578         if (!f) {
1579                 int saved_errno = errno;
1580                 rollback_lock_file(&mark_lock);
1581                 failure |= error("Unable to write marks file %s: %s",
1582                         mark_file, strerror(saved_errno));
1583                 return;
1584         }
1585
1586         /*
1587          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1588          * Assign -1 to the lock file descriptor so that commit_lock_file()
1589          * won't try to close() it.
1590          */
1591         mark_lock.fd = -1;
1592
1593         dump_marks_helper(f, 0, marks);
1594         if (ferror(f) || fclose(f)) {
1595                 int saved_errno = errno;
1596                 rollback_lock_file(&mark_lock);
1597                 failure |= error("Unable to write marks file %s: %s",
1598                         mark_file, strerror(saved_errno));
1599                 return;
1600         }
1601
1602         if (commit_lock_file(&mark_lock)) {
1603                 int saved_errno = errno;
1604                 rollback_lock_file(&mark_lock);
1605                 failure |= error("Unable to commit marks file %s: %s",
1606                         mark_file, strerror(saved_errno));
1607                 return;
1608         }
1609 }
1610
1611 static int read_next_command(void)
1612 {
1613         static int stdin_eof = 0;
1614
1615         if (stdin_eof) {
1616                 unread_command_buf = 0;
1617                 return EOF;
1618         }
1619
1620         do {
1621                 if (unread_command_buf) {
1622                         unread_command_buf = 0;
1623                 } else {
1624                         struct recent_command *rc;
1625
1626                         strbuf_detach(&command_buf, NULL);
1627                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1628                         if (stdin_eof)
1629                                 return EOF;
1630
1631                         rc = rc_free;
1632                         if (rc)
1633                                 rc_free = rc->next;
1634                         else {
1635                                 rc = cmd_hist.next;
1636                                 cmd_hist.next = rc->next;
1637                                 cmd_hist.next->prev = &cmd_hist;
1638                                 free(rc->buf);
1639                         }
1640
1641                         rc->buf = command_buf.buf;
1642                         rc->prev = cmd_tail;
1643                         rc->next = cmd_hist.prev;
1644                         rc->prev->next = rc;
1645                         cmd_tail = rc;
1646                 }
1647         } while (command_buf.buf[0] == '#');
1648
1649         return 0;
1650 }
1651
1652 static void skip_optional_lf(void)
1653 {
1654         int term_char = fgetc(stdin);
1655         if (term_char != '\n' && term_char != EOF)
1656                 ungetc(term_char, stdin);
1657 }
1658
1659 static void cmd_mark(void)
1660 {
1661         if (!prefixcmp(command_buf.buf, "mark :")) {
1662                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1663                 read_next_command();
1664         }
1665         else
1666                 next_mark = 0;
1667 }
1668
1669 static void cmd_data(struct strbuf *sb)
1670 {
1671         strbuf_reset(sb);
1672
1673         if (prefixcmp(command_buf.buf, "data "))
1674                 die("Expected 'data n' command, found: %s", command_buf.buf);
1675
1676         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1677                 char *term = xstrdup(command_buf.buf + 5 + 2);
1678                 size_t term_len = command_buf.len - 5 - 2;
1679
1680                 strbuf_detach(&command_buf, NULL);
1681                 for (;;) {
1682                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1683                                 die("EOF in data (terminator '%s' not found)", term);
1684                         if (term_len == command_buf.len
1685                                 && !strcmp(term, command_buf.buf))
1686                                 break;
1687                         strbuf_addbuf(sb, &command_buf);
1688                         strbuf_addch(sb, '\n');
1689                 }
1690                 free(term);
1691         }
1692         else {
1693                 size_t n = 0, length;
1694
1695                 length = strtoul(command_buf.buf + 5, NULL, 10);
1696
1697                 while (n < length) {
1698                         size_t s = strbuf_fread(sb, length - n, stdin);
1699                         if (!s && feof(stdin))
1700                                 die("EOF in data (%lu bytes remaining)",
1701                                         (unsigned long)(length - n));
1702                         n += s;
1703                 }
1704         }
1705
1706         skip_optional_lf();
1707 }
1708
1709 static int validate_raw_date(const char *src, char *result, int maxlen)
1710 {
1711         const char *orig_src = src;
1712         char *endp, sign;
1713
1714         strtoul(src, &endp, 10);
1715         if (endp == src || *endp != ' ')
1716                 return -1;
1717
1718         src = endp + 1;
1719         if (*src != '-' && *src != '+')
1720                 return -1;
1721         sign = *src;
1722
1723         strtoul(src + 1, &endp, 10);
1724         if (endp == src || *endp || (endp - orig_src) >= maxlen)
1725                 return -1;
1726
1727         strcpy(result, orig_src);
1728         return 0;
1729 }
1730
1731 static char *parse_ident(const char *buf)
1732 {
1733         const char *gt;
1734         size_t name_len;
1735         char *ident;
1736
1737         gt = strrchr(buf, '>');
1738         if (!gt)
1739                 die("Missing > in ident string: %s", buf);
1740         gt++;
1741         if (*gt != ' ')
1742                 die("Missing space after > in ident string: %s", buf);
1743         gt++;
1744         name_len = gt - buf;
1745         ident = xmalloc(name_len + 24);
1746         strncpy(ident, buf, name_len);
1747
1748         switch (whenspec) {
1749         case WHENSPEC_RAW:
1750                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1751                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1752                 break;
1753         case WHENSPEC_RFC2822:
1754                 if (parse_date(gt, ident + name_len, 24) < 0)
1755                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1756                 break;
1757         case WHENSPEC_NOW:
1758                 if (strcmp("now", gt))
1759                         die("Date in ident must be 'now': %s", buf);
1760                 datestamp(ident + name_len, 24);
1761                 break;
1762         }
1763
1764         return ident;
1765 }
1766
1767 static void cmd_new_blob(void)
1768 {
1769         static struct strbuf buf = STRBUF_INIT;
1770
1771         read_next_command();
1772         cmd_mark();
1773         cmd_data(&buf);
1774         store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1775 }
1776
1777 static void unload_one_branch(void)
1778 {
1779         while (cur_active_branches
1780                 && cur_active_branches >= max_active_branches) {
1781                 uintmax_t min_commit = ULONG_MAX;
1782                 struct branch *e, *l = NULL, *p = NULL;
1783
1784                 for (e = active_branches; e; e = e->active_next_branch) {
1785                         if (e->last_commit < min_commit) {
1786                                 p = l;
1787                                 min_commit = e->last_commit;
1788                         }
1789                         l = e;
1790                 }
1791
1792                 if (p) {
1793                         e = p->active_next_branch;
1794                         p->active_next_branch = e->active_next_branch;
1795                 } else {
1796                         e = active_branches;
1797                         active_branches = e->active_next_branch;
1798                 }
1799                 e->active = 0;
1800                 e->active_next_branch = NULL;
1801                 if (e->branch_tree.tree) {
1802                         release_tree_content_recursive(e->branch_tree.tree);
1803                         e->branch_tree.tree = NULL;
1804                 }
1805                 cur_active_branches--;
1806         }
1807 }
1808
1809 static void load_branch(struct branch *b)
1810 {
1811         load_tree(&b->branch_tree);
1812         if (!b->active) {
1813                 b->active = 1;
1814                 b->active_next_branch = active_branches;
1815                 active_branches = b;
1816                 cur_active_branches++;
1817                 branch_load_count++;
1818         }
1819 }
1820
1821 static void file_change_m(struct branch *b)
1822 {
1823         const char *p = command_buf.buf + 2;
1824         static struct strbuf uq = STRBUF_INIT;
1825         const char *endp;
1826         struct object_entry *oe = oe;
1827         unsigned char sha1[20];
1828         uint16_t mode, inline_data = 0;
1829
1830         p = get_mode(p, &mode);
1831         if (!p)
1832                 die("Corrupt mode: %s", command_buf.buf);
1833         switch (mode) {
1834         case S_IFREG | 0644:
1835         case S_IFREG | 0755:
1836         case S_IFLNK:
1837         case 0644:
1838         case 0755:
1839                 /* ok */
1840                 break;
1841         default:
1842                 die("Corrupt mode: %s", command_buf.buf);
1843         }
1844
1845         if (*p == ':') {
1846                 char *x;
1847                 oe = find_mark(strtoumax(p + 1, &x, 10));
1848                 hashcpy(sha1, oe->sha1);
1849                 p = x;
1850         } else if (!prefixcmp(p, "inline")) {
1851                 inline_data = 1;
1852                 p += 6;
1853         } else {
1854                 if (get_sha1_hex(p, sha1))
1855                         die("Invalid SHA1: %s", command_buf.buf);
1856                 oe = find_object(sha1);
1857                 p += 40;
1858         }
1859         if (*p++ != ' ')
1860                 die("Missing space after SHA1: %s", command_buf.buf);
1861
1862         strbuf_reset(&uq);
1863         if (!unquote_c_style(&uq, p, &endp)) {
1864                 if (*endp)
1865                         die("Garbage after path in: %s", command_buf.buf);
1866                 p = uq.buf;
1867         }
1868
1869         if (inline_data) {
1870                 static struct strbuf buf = STRBUF_INIT;
1871
1872                 if (p != uq.buf) {
1873                         strbuf_addstr(&uq, p);
1874                         p = uq.buf;
1875                 }
1876                 read_next_command();
1877                 cmd_data(&buf);
1878                 store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1879         } else if (oe) {
1880                 if (oe->type != OBJ_BLOB)
1881                         die("Not a blob (actually a %s): %s",
1882                                 typename(oe->type), command_buf.buf);
1883         } else {
1884                 enum object_type type = sha1_object_info(sha1, NULL);
1885                 if (type < 0)
1886                         die("Blob not found: %s", command_buf.buf);
1887                 if (type != OBJ_BLOB)
1888                         die("Not a blob (actually a %s): %s",
1889                             typename(type), command_buf.buf);
1890         }
1891
1892         tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode, NULL);
1893 }
1894
1895 static void file_change_d(struct branch *b)
1896 {
1897         const char *p = command_buf.buf + 2;
1898         static struct strbuf uq = STRBUF_INIT;
1899         const char *endp;
1900
1901         strbuf_reset(&uq);
1902         if (!unquote_c_style(&uq, p, &endp)) {
1903                 if (*endp)
1904                         die("Garbage after path in: %s", command_buf.buf);
1905                 p = uq.buf;
1906         }
1907         tree_content_remove(&b->branch_tree, p, NULL);
1908 }
1909
1910 static void file_change_cr(struct branch *b, int rename)
1911 {
1912         const char *s, *d;
1913         static struct strbuf s_uq = STRBUF_INIT;
1914         static struct strbuf d_uq = STRBUF_INIT;
1915         const char *endp;
1916         struct tree_entry leaf;
1917
1918         s = command_buf.buf + 2;
1919         strbuf_reset(&s_uq);
1920         if (!unquote_c_style(&s_uq, s, &endp)) {
1921                 if (*endp != ' ')
1922                         die("Missing space after source: %s", command_buf.buf);
1923         } else {
1924                 endp = strchr(s, ' ');
1925                 if (!endp)
1926                         die("Missing space after source: %s", command_buf.buf);
1927                 strbuf_add(&s_uq, s, endp - s);
1928         }
1929         s = s_uq.buf;
1930
1931         endp++;
1932         if (!*endp)
1933                 die("Missing dest: %s", command_buf.buf);
1934
1935         d = endp;
1936         strbuf_reset(&d_uq);
1937         if (!unquote_c_style(&d_uq, d, &endp)) {
1938                 if (*endp)
1939                         die("Garbage after dest in: %s", command_buf.buf);
1940                 d = d_uq.buf;
1941         }
1942
1943         memset(&leaf, 0, sizeof(leaf));
1944         if (rename)
1945                 tree_content_remove(&b->branch_tree, s, &leaf);
1946         else
1947                 tree_content_get(&b->branch_tree, s, &leaf);
1948         if (!leaf.versions[1].mode)
1949                 die("Path %s not in branch", s);
1950         tree_content_set(&b->branch_tree, d,
1951                 leaf.versions[1].sha1,
1952                 leaf.versions[1].mode,
1953                 leaf.tree);
1954 }
1955
1956 static void file_change_deleteall(struct branch *b)
1957 {
1958         release_tree_content_recursive(b->branch_tree.tree);
1959         hashclr(b->branch_tree.versions[0].sha1);
1960         hashclr(b->branch_tree.versions[1].sha1);
1961         load_tree(&b->branch_tree);
1962 }
1963
1964 static void cmd_from_commit(struct branch *b, char *buf, unsigned long size)
1965 {
1966         if (!buf || size < 46)
1967                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
1968         if (memcmp("tree ", buf, 5)
1969                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1970                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1971         hashcpy(b->branch_tree.versions[0].sha1,
1972                 b->branch_tree.versions[1].sha1);
1973 }
1974
1975 static void cmd_from_existing(struct branch *b)
1976 {
1977         if (is_null_sha1(b->sha1)) {
1978                 hashclr(b->branch_tree.versions[0].sha1);
1979                 hashclr(b->branch_tree.versions[1].sha1);
1980         } else {
1981                 unsigned long size;
1982                 char *buf;
1983
1984                 buf = read_object_with_reference(b->sha1,
1985                         commit_type, &size, b->sha1);
1986                 cmd_from_commit(b, buf, size);
1987                 free(buf);
1988         }
1989 }
1990
1991 static int cmd_from(struct branch *b)
1992 {
1993         const char *from;
1994         struct branch *s;
1995
1996         if (prefixcmp(command_buf.buf, "from "))
1997                 return 0;
1998
1999         if (b->branch_tree.tree) {
2000                 release_tree_content_recursive(b->branch_tree.tree);
2001                 b->branch_tree.tree = NULL;
2002         }
2003
2004         from = strchr(command_buf.buf, ' ') + 1;
2005         s = lookup_branch(from);
2006         if (b == s)
2007                 die("Can't create a branch from itself: %s", b->name);
2008         else if (s) {
2009                 unsigned char *t = s->branch_tree.versions[1].sha1;
2010                 hashcpy(b->sha1, s->sha1);
2011                 hashcpy(b->branch_tree.versions[0].sha1, t);
2012                 hashcpy(b->branch_tree.versions[1].sha1, t);
2013         } else if (*from == ':') {
2014                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2015                 struct object_entry *oe = find_mark(idnum);
2016                 if (oe->type != OBJ_COMMIT)
2017                         die("Mark :%" PRIuMAX " not a commit", idnum);
2018                 hashcpy(b->sha1, oe->sha1);
2019                 if (oe->pack_id != MAX_PACK_ID) {
2020                         unsigned long size;
2021                         char *buf = gfi_unpack_entry(oe, &size);
2022                         cmd_from_commit(b, buf, size);
2023                         free(buf);
2024                 } else
2025                         cmd_from_existing(b);
2026         } else if (!get_sha1(from, b->sha1))
2027                 cmd_from_existing(b);
2028         else
2029                 die("Invalid ref name or SHA1 expression: %s", from);
2030
2031         read_next_command();
2032         return 1;
2033 }
2034
2035 static struct hash_list *cmd_merge(unsigned int *count)
2036 {
2037         struct hash_list *list = NULL, *n, *e = e;
2038         const char *from;
2039         struct branch *s;
2040
2041         *count = 0;
2042         while (!prefixcmp(command_buf.buf, "merge ")) {
2043                 from = strchr(command_buf.buf, ' ') + 1;
2044                 n = xmalloc(sizeof(*n));
2045                 s = lookup_branch(from);
2046                 if (s)
2047                         hashcpy(n->sha1, s->sha1);
2048                 else if (*from == ':') {
2049                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2050                         struct object_entry *oe = find_mark(idnum);
2051                         if (oe->type != OBJ_COMMIT)
2052                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2053                         hashcpy(n->sha1, oe->sha1);
2054                 } else if (!get_sha1(from, n->sha1)) {
2055                         unsigned long size;
2056                         char *buf = read_object_with_reference(n->sha1,
2057                                 commit_type, &size, n->sha1);
2058                         if (!buf || size < 46)
2059                                 die("Not a valid commit: %s", from);
2060                         free(buf);
2061                 } else
2062                         die("Invalid ref name or SHA1 expression: %s", from);
2063
2064                 n->next = NULL;
2065                 if (list)
2066                         e->next = n;
2067                 else
2068                         list = n;
2069                 e = n;
2070                 (*count)++;
2071                 read_next_command();
2072         }
2073         return list;
2074 }
2075
2076 static void cmd_new_commit(void)
2077 {
2078         static struct strbuf msg = STRBUF_INIT;
2079         struct branch *b;
2080         char *sp;
2081         char *author = NULL;
2082         char *committer = NULL;
2083         struct hash_list *merge_list = NULL;
2084         unsigned int merge_count;
2085
2086         /* Obtain the branch name from the rest of our command */
2087         sp = strchr(command_buf.buf, ' ') + 1;
2088         b = lookup_branch(sp);
2089         if (!b)
2090                 b = new_branch(sp);
2091
2092         read_next_command();
2093         cmd_mark();
2094         if (!prefixcmp(command_buf.buf, "author ")) {
2095                 author = parse_ident(command_buf.buf + 7);
2096                 read_next_command();
2097         }
2098         if (!prefixcmp(command_buf.buf, "committer ")) {
2099                 committer = parse_ident(command_buf.buf + 10);
2100                 read_next_command();
2101         }
2102         if (!committer)
2103                 die("Expected committer but didn't get one");
2104         cmd_data(&msg);
2105         read_next_command();
2106         cmd_from(b);
2107         merge_list = cmd_merge(&merge_count);
2108
2109         /* ensure the branch is active/loaded */
2110         if (!b->branch_tree.tree || !max_active_branches) {
2111                 unload_one_branch();
2112                 load_branch(b);
2113         }
2114
2115         /* file_change* */
2116         while (command_buf.len > 0) {
2117                 if (!prefixcmp(command_buf.buf, "M "))
2118                         file_change_m(b);
2119                 else if (!prefixcmp(command_buf.buf, "D "))
2120                         file_change_d(b);
2121                 else if (!prefixcmp(command_buf.buf, "R "))
2122                         file_change_cr(b, 1);
2123                 else if (!prefixcmp(command_buf.buf, "C "))
2124                         file_change_cr(b, 0);
2125                 else if (!strcmp("deleteall", command_buf.buf))
2126                         file_change_deleteall(b);
2127                 else {
2128                         unread_command_buf = 1;
2129                         break;
2130                 }
2131                 if (read_next_command() == EOF)
2132                         break;
2133         }
2134
2135         /* build the tree and the commit */
2136         store_tree(&b->branch_tree);
2137         hashcpy(b->branch_tree.versions[0].sha1,
2138                 b->branch_tree.versions[1].sha1);
2139
2140         strbuf_reset(&new_data);
2141         strbuf_addf(&new_data, "tree %s\n",
2142                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2143         if (!is_null_sha1(b->sha1))
2144                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2145         while (merge_list) {
2146                 struct hash_list *next = merge_list->next;
2147                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2148                 free(merge_list);
2149                 merge_list = next;
2150         }
2151         strbuf_addf(&new_data,
2152                 "author %s\n"
2153                 "committer %s\n"
2154                 "\n",
2155                 author ? author : committer, committer);
2156         strbuf_addbuf(&new_data, &msg);
2157         free(author);
2158         free(committer);
2159
2160         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2161                 b->pack_id = pack_id;
2162         b->last_commit = object_count_by_type[OBJ_COMMIT];
2163 }
2164
2165 static void cmd_new_tag(void)
2166 {
2167         static struct strbuf msg = STRBUF_INIT;
2168         char *sp;
2169         const char *from;
2170         char *tagger;
2171         struct branch *s;
2172         struct tag *t;
2173         uintmax_t from_mark = 0;
2174         unsigned char sha1[20];
2175
2176         /* Obtain the new tag name from the rest of our command */
2177         sp = strchr(command_buf.buf, ' ') + 1;
2178         t = pool_alloc(sizeof(struct tag));
2179         t->next_tag = NULL;
2180         t->name = pool_strdup(sp);
2181         if (last_tag)
2182                 last_tag->next_tag = t;
2183         else
2184                 first_tag = t;
2185         last_tag = t;
2186         read_next_command();
2187
2188         /* from ... */
2189         if (prefixcmp(command_buf.buf, "from "))
2190                 die("Expected from command, got %s", command_buf.buf);
2191         from = strchr(command_buf.buf, ' ') + 1;
2192         s = lookup_branch(from);
2193         if (s) {
2194                 hashcpy(sha1, s->sha1);
2195         } else if (*from == ':') {
2196                 struct object_entry *oe;
2197                 from_mark = strtoumax(from + 1, NULL, 10);
2198                 oe = find_mark(from_mark);
2199                 if (oe->type != OBJ_COMMIT)
2200                         die("Mark :%" PRIuMAX " not a commit", from_mark);
2201                 hashcpy(sha1, oe->sha1);
2202         } else if (!get_sha1(from, sha1)) {
2203                 unsigned long size;
2204                 char *buf;
2205
2206                 buf = read_object_with_reference(sha1,
2207                         commit_type, &size, sha1);
2208                 if (!buf || size < 46)
2209                         die("Not a valid commit: %s", from);
2210                 free(buf);
2211         } else
2212                 die("Invalid ref name or SHA1 expression: %s", from);
2213         read_next_command();
2214
2215         /* tagger ... */
2216         if (prefixcmp(command_buf.buf, "tagger "))
2217                 die("Expected tagger command, got %s", command_buf.buf);
2218         tagger = parse_ident(command_buf.buf + 7);
2219
2220         /* tag payload/message */
2221         read_next_command();
2222         cmd_data(&msg);
2223
2224         /* build the tag object */
2225         strbuf_reset(&new_data);
2226         strbuf_addf(&new_data,
2227                 "object %s\n"
2228                 "type %s\n"
2229                 "tag %s\n"
2230                 "tagger %s\n"
2231                 "\n",
2232                 sha1_to_hex(sha1), commit_type, t->name, tagger);
2233         strbuf_addbuf(&new_data, &msg);
2234         free(tagger);
2235
2236         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2237                 t->pack_id = MAX_PACK_ID;
2238         else
2239                 t->pack_id = pack_id;
2240 }
2241
2242 static void cmd_reset_branch(void)
2243 {
2244         struct branch *b;
2245         char *sp;
2246
2247         /* Obtain the branch name from the rest of our command */
2248         sp = strchr(command_buf.buf, ' ') + 1;
2249         b = lookup_branch(sp);
2250         if (b) {
2251                 hashclr(b->sha1);
2252                 hashclr(b->branch_tree.versions[0].sha1);
2253                 hashclr(b->branch_tree.versions[1].sha1);
2254                 if (b->branch_tree.tree) {
2255                         release_tree_content_recursive(b->branch_tree.tree);
2256                         b->branch_tree.tree = NULL;
2257                 }
2258         }
2259         else
2260                 b = new_branch(sp);
2261         read_next_command();
2262         if (!cmd_from(b) && command_buf.len > 0)
2263                 unread_command_buf = 1;
2264 }
2265
2266 static void cmd_checkpoint(void)
2267 {
2268         if (object_count) {
2269                 cycle_packfile();
2270                 dump_branches();
2271                 dump_tags();
2272                 dump_marks();
2273         }
2274         skip_optional_lf();
2275 }
2276
2277 static void cmd_progress(void)
2278 {
2279         fwrite(command_buf.buf, 1, command_buf.len, stdout);
2280         fputc('\n', stdout);
2281         fflush(stdout);
2282         skip_optional_lf();
2283 }
2284
2285 static void import_marks(const char *input_file)
2286 {
2287         char line[512];
2288         FILE *f = fopen(input_file, "r");
2289         if (!f)
2290                 die("cannot read %s: %s", input_file, strerror(errno));
2291         while (fgets(line, sizeof(line), f)) {
2292                 uintmax_t mark;
2293                 char *end;
2294                 unsigned char sha1[20];
2295                 struct object_entry *e;
2296
2297                 end = strchr(line, '\n');
2298                 if (line[0] != ':' || !end)
2299                         die("corrupt mark line: %s", line);
2300                 *end = 0;
2301                 mark = strtoumax(line + 1, &end, 10);
2302                 if (!mark || end == line + 1
2303                         || *end != ' ' || get_sha1(end + 1, sha1))
2304                         die("corrupt mark line: %s", line);
2305                 e = find_object(sha1);
2306                 if (!e) {
2307                         enum object_type type = sha1_object_info(sha1, NULL);
2308                         if (type < 0)
2309                                 die("object not found: %s", sha1_to_hex(sha1));
2310                         e = insert_object(sha1);
2311                         e->type = type;
2312                         e->pack_id = MAX_PACK_ID;
2313                         e->offset = 1; /* just not zero! */
2314                 }
2315                 insert_mark(mark, e);
2316         }
2317         fclose(f);
2318 }
2319
2320 static int git_pack_config(const char *k, const char *v)
2321 {
2322         if (!strcmp(k, "pack.depth")) {
2323                 max_depth = git_config_int(k, v);
2324                 if (max_depth > MAX_DEPTH)
2325                         max_depth = MAX_DEPTH;
2326                 return 0;
2327         }
2328         if (!strcmp(k, "pack.compression")) {
2329                 int level = git_config_int(k, v);
2330                 if (level == -1)
2331                         level = Z_DEFAULT_COMPRESSION;
2332                 else if (level < 0 || level > Z_BEST_COMPRESSION)
2333                         die("bad pack compression level %d", level);
2334                 pack_compression_level = level;
2335                 pack_compression_seen = 1;
2336                 return 0;
2337         }
2338         return git_default_config(k, v);
2339 }
2340
2341 static const char fast_import_usage[] =
2342 "git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2343
2344 int main(int argc, const char **argv)
2345 {
2346         unsigned int i, show_stats = 1;
2347
2348         git_config(git_pack_config);
2349         if (!pack_compression_seen && core_compression_seen)
2350                 pack_compression_level = core_compression_level;
2351
2352         alloc_objects(object_entry_alloc);
2353         strbuf_init(&command_buf, 0);
2354         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2355         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2356         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2357         marks = pool_calloc(1, sizeof(struct mark_set));
2358
2359         for (i = 1; i < argc; i++) {
2360                 const char *a = argv[i];
2361
2362                 if (*a != '-' || !strcmp(a, "--"))
2363                         break;
2364                 else if (!prefixcmp(a, "--date-format=")) {
2365                         const char *fmt = a + 14;
2366                         if (!strcmp(fmt, "raw"))
2367                                 whenspec = WHENSPEC_RAW;
2368                         else if (!strcmp(fmt, "rfc2822"))
2369                                 whenspec = WHENSPEC_RFC2822;
2370                         else if (!strcmp(fmt, "now"))
2371                                 whenspec = WHENSPEC_NOW;
2372                         else
2373                                 die("unknown --date-format argument %s", fmt);
2374                 }
2375                 else if (!prefixcmp(a, "--max-pack-size="))
2376                         max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2377                 else if (!prefixcmp(a, "--depth=")) {
2378                         max_depth = strtoul(a + 8, NULL, 0);
2379                         if (max_depth > MAX_DEPTH)
2380                                 die("--depth cannot exceed %u", MAX_DEPTH);
2381                 }
2382                 else if (!prefixcmp(a, "--active-branches="))
2383                         max_active_branches = strtoul(a + 18, NULL, 0);
2384                 else if (!prefixcmp(a, "--import-marks="))
2385                         import_marks(a + 15);
2386                 else if (!prefixcmp(a, "--export-marks="))
2387                         mark_file = a + 15;
2388                 else if (!prefixcmp(a, "--export-pack-edges=")) {
2389                         if (pack_edges)
2390                                 fclose(pack_edges);
2391                         pack_edges = fopen(a + 20, "a");
2392                         if (!pack_edges)
2393                                 die("Cannot open %s: %s", a + 20, strerror(errno));
2394                 } else if (!strcmp(a, "--force"))
2395                         force_update = 1;
2396                 else if (!strcmp(a, "--quiet"))
2397                         show_stats = 0;
2398                 else if (!strcmp(a, "--stats"))
2399                         show_stats = 1;
2400                 else
2401                         die("unknown option %s", a);
2402         }
2403         if (i != argc)
2404                 usage(fast_import_usage);
2405
2406         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2407         for (i = 0; i < (cmd_save - 1); i++)
2408                 rc_free[i].next = &rc_free[i + 1];
2409         rc_free[cmd_save - 1].next = NULL;
2410
2411         prepare_packed_git();
2412         start_packfile();
2413         set_die_routine(die_nicely);
2414         while (read_next_command() != EOF) {
2415                 if (!strcmp("blob", command_buf.buf))
2416                         cmd_new_blob();
2417                 else if (!prefixcmp(command_buf.buf, "commit "))
2418                         cmd_new_commit();
2419                 else if (!prefixcmp(command_buf.buf, "tag "))
2420                         cmd_new_tag();
2421                 else if (!prefixcmp(command_buf.buf, "reset "))
2422                         cmd_reset_branch();
2423                 else if (!strcmp("checkpoint", command_buf.buf))
2424                         cmd_checkpoint();
2425                 else if (!prefixcmp(command_buf.buf, "progress "))
2426                         cmd_progress();
2427                 else
2428                         die("Unsupported command: %s", command_buf.buf);
2429         }
2430         end_packfile();
2431
2432         dump_branches();
2433         dump_tags();
2434         unkeep_all_packs();
2435         dump_marks();
2436
2437         if (pack_edges)
2438                 fclose(pack_edges);
2439
2440         if (show_stats) {
2441                 uintmax_t total_count = 0, duplicate_count = 0;
2442                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2443                         total_count += object_count_by_type[i];
2444                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2445                         duplicate_count += duplicate_count_by_type[i];
2446
2447                 fprintf(stderr, "%s statistics:\n", argv[0]);
2448                 fprintf(stderr, "---------------------------------------------------------------------\n");
2449                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2450                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2451                 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]);
2452                 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]);
2453                 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]);
2454                 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]);
2455                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2456                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2457                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2458                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2459                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2460                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2461                 fprintf(stderr, "---------------------------------------------------------------------\n");
2462                 pack_report();
2463                 fprintf(stderr, "---------------------------------------------------------------------\n");
2464                 fprintf(stderr, "\n");
2465         }
2466
2467         return failure ? 1 : 0;
2468 }