Merge branch 'tk/apply-dev-null-verify-name-fix'
[git] / fast-import.c
1 /*
2 (See Documentation/git-fast-import.txt for maintained documentation.)
3 Format of STDIN stream:
4
5   stream ::= cmd*;
6
7   cmd ::= new_blob
8         | new_commit
9         | new_tag
10         | reset_branch
11         | checkpoint
12         | progress
13         ;
14
15   new_blob ::= 'blob' lf
16     mark?
17     file_content;
18   file_content ::= data;
19
20   new_commit ::= 'commit' sp ref_str lf
21     mark?
22     ('author' (sp name)? sp '<' email '>' sp when lf)?
23     'committer' (sp name)? sp '<' email '>' sp when lf
24     commit_msg
25     ('from' sp commit-ish lf)?
26     ('merge' sp commit-ish lf)*
27     (file_change | ls)*
28     lf?;
29   commit_msg ::= data;
30
31   ls ::= 'ls' sp '"' quoted(path) '"' lf;
32
33   file_change ::= file_clr
34     | file_del
35     | file_rnm
36     | file_cpy
37     | file_obm
38     | file_inm;
39   file_clr ::= 'deleteall' lf;
40   file_del ::= 'D' sp path_str lf;
41   file_rnm ::= 'R' sp path_str sp path_str lf;
42   file_cpy ::= 'C' sp path_str sp path_str lf;
43   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
44   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
45     data;
46   note_obm ::= 'N' sp (hexsha1 | idnum) sp commit-ish lf;
47   note_inm ::= 'N' sp 'inline' sp commit-ish lf
48     data;
49
50   new_tag ::= 'tag' sp tag_str lf
51     'from' sp commit-ish lf
52     ('tagger' (sp name)? sp '<' email '>' sp when lf)?
53     tag_msg;
54   tag_msg ::= data;
55
56   reset_branch ::= 'reset' sp ref_str lf
57     ('from' sp commit-ish lf)?
58     lf?;
59
60   checkpoint ::= 'checkpoint' lf
61     lf?;
62
63   progress ::= 'progress' sp not_lf* lf
64     lf?;
65
66      # note: the first idnum in a stream should be 1 and subsequent
67      # idnums should not have gaps between values as this will cause
68      # the stream parser to reserve space for the gapped values.  An
69      # idnum can be updated in the future to a new object by issuing
70      # a new mark directive with the old idnum.
71      #
72   mark ::= 'mark' sp idnum lf;
73   data ::= (delimited_data | exact_data)
74     lf?;
75
76     # note: delim may be any string but must not contain lf.
77     # data_line may contain any data but must not be exactly
78     # delim.
79   delimited_data ::= 'data' sp '<<' delim lf
80     (data_line lf)*
81     delim lf;
82
83      # note: declen indicates the length of binary_data in bytes.
84      # declen does not include the lf preceding the binary data.
85      #
86   exact_data ::= 'data' sp declen lf
87     binary_data;
88
89      # note: quoted strings are C-style quoting supporting \c for
90      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
91      # is the signed byte value in octal.  Note that the only
92      # characters which must actually be escaped to protect the
93      # stream formatting is: \, " and LF.  Otherwise these values
94      # are UTF8.
95      #
96   commit-ish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
97   ref_str     ::= ref;
98   sha1exp_str ::= sha1exp;
99   tag_str     ::= tag;
100   path_str    ::= path    | '"' quoted(path)    '"' ;
101   mode        ::= '100644' | '644'
102                 | '100755' | '755'
103                 | '120000'
104                 ;
105
106   declen ::= # unsigned 32 bit value, ascii base10 notation;
107   bigint ::= # unsigned integer value, ascii base10 notation;
108   binary_data ::= # file content, not interpreted;
109
110   when         ::= raw_when | rfc2822_when;
111   raw_when     ::= ts sp tz;
112   rfc2822_when ::= # Valid RFC 2822 date and time;
113
114   sp ::= # ASCII space character;
115   lf ::= # ASCII newline (LF) character;
116
117      # note: a colon (':') must precede the numerical value assigned to
118      # an idnum.  This is to distinguish it from a ref or tag name as
119      # GIT does not permit ':' in ref or tag strings.
120      #
121   idnum   ::= ':' bigint;
122   path    ::= # GIT style file path, e.g. "a/b/c";
123   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
124   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
125   sha1exp ::= # Any valid GIT SHA1 expression;
126   hexsha1 ::= # SHA1 in hexadecimal format;
127
128      # note: name and email are UTF8 strings, however name must not
129      # contain '<' or lf and email must not contain any of the
130      # following: '<', '>', lf.
131      #
132   name  ::= # valid GIT author/committer name;
133   email ::= # valid GIT author/committer email;
134   ts    ::= # time since the epoch in seconds, ascii base10 notation;
135   tz    ::= # GIT style timezone;
136
137      # note: comments, get-mark, ls-tree, and cat-blob requests may
138      # appear anywhere in the input, except within a data command. Any
139      # form of the data command always escapes the related input from
140      # comment processing.
141      #
142      # In case it is not clear, the '#' that starts the comment
143      # must be the first character on that line (an lf
144      # preceded it).
145      #
146
147   get_mark ::= 'get-mark' sp idnum lf;
148   cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
149   ls_tree  ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
150
151   comment ::= '#' not_lf* lf;
152   not_lf  ::= # Any byte that is not ASCII newline (LF);
153 */
154
155 #include "builtin.h"
156 #include "cache.h"
157 #include "config.h"
158 #include "lockfile.h"
159 #include "object.h"
160 #include "blob.h"
161 #include "tree.h"
162 #include "commit.h"
163 #include "delta.h"
164 #include "pack.h"
165 #include "refs.h"
166 #include "csum-file.h"
167 #include "quote.h"
168 #include "dir.h"
169 #include "run-command.h"
170 #include "packfile.h"
171
172 #define PACK_ID_BITS 16
173 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
174 #define DEPTH_BITS 13
175 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
176
177 /*
178  * We abuse the setuid bit on directories to mean "do not delta".
179  */
180 #define NO_DELTA S_ISUID
181
182 struct object_entry {
183         struct pack_idx_entry idx;
184         struct object_entry *next;
185         uint32_t type : TYPE_BITS,
186                 pack_id : PACK_ID_BITS,
187                 depth : DEPTH_BITS;
188 };
189
190 struct object_entry_pool {
191         struct object_entry_pool *next_pool;
192         struct object_entry *next_free;
193         struct object_entry *end;
194         struct object_entry entries[FLEX_ARRAY]; /* more */
195 };
196
197 struct mark_set {
198         union {
199                 struct object_entry *marked[1024];
200                 struct mark_set *sets[1024];
201         } data;
202         unsigned int shift;
203 };
204
205 struct last_object {
206         struct strbuf data;
207         off_t offset;
208         unsigned int depth;
209         unsigned no_swap : 1;
210 };
211
212 struct mem_pool {
213         struct mem_pool *next_pool;
214         char *next_free;
215         char *end;
216         uintmax_t space[FLEX_ARRAY]; /* more */
217 };
218
219 struct atom_str {
220         struct atom_str *next_atom;
221         unsigned short str_len;
222         char str_dat[FLEX_ARRAY]; /* more */
223 };
224
225 struct tree_content;
226 struct tree_entry {
227         struct tree_content *tree;
228         struct atom_str *name;
229         struct tree_entry_ms {
230                 uint16_t mode;
231                 struct object_id oid;
232         } versions[2];
233 };
234
235 struct tree_content {
236         unsigned int entry_capacity; /* must match avail_tree_content */
237         unsigned int entry_count;
238         unsigned int delta_depth;
239         struct tree_entry *entries[FLEX_ARRAY]; /* more */
240 };
241
242 struct avail_tree_content {
243         unsigned int entry_capacity; /* must match tree_content */
244         struct avail_tree_content *next_avail;
245 };
246
247 struct branch {
248         struct branch *table_next_branch;
249         struct branch *active_next_branch;
250         const char *name;
251         struct tree_entry branch_tree;
252         uintmax_t last_commit;
253         uintmax_t num_notes;
254         unsigned active : 1;
255         unsigned delete : 1;
256         unsigned pack_id : PACK_ID_BITS;
257         struct object_id oid;
258 };
259
260 struct tag {
261         struct tag *next_tag;
262         const char *name;
263         unsigned int pack_id;
264         struct object_id oid;
265 };
266
267 struct hash_list {
268         struct hash_list *next;
269         struct object_id oid;
270 };
271
272 typedef enum {
273         WHENSPEC_RAW = 1,
274         WHENSPEC_RFC2822,
275         WHENSPEC_NOW
276 } whenspec_type;
277
278 struct recent_command {
279         struct recent_command *prev;
280         struct recent_command *next;
281         char *buf;
282 };
283
284 /* Configured limits on output */
285 static unsigned long max_depth = 50;
286 static off_t max_packsize;
287 static int unpack_limit = 100;
288 static int force_update;
289
290 /* Stats and misc. counters */
291 static uintmax_t alloc_count;
292 static uintmax_t marks_set_count;
293 static uintmax_t object_count_by_type[1 << TYPE_BITS];
294 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
295 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
296 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
297 static unsigned long object_count;
298 static unsigned long branch_count;
299 static unsigned long branch_load_count;
300 static int failure;
301 static FILE *pack_edges;
302 static unsigned int show_stats = 1;
303 static int global_argc;
304 static const char **global_argv;
305
306 /* Memory pools */
307 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
308 static size_t total_allocd;
309 static struct mem_pool *mem_pool;
310
311 /* Atom management */
312 static unsigned int atom_table_sz = 4451;
313 static unsigned int atom_cnt;
314 static struct atom_str **atom_table;
315
316 /* The .pack file being generated */
317 static struct pack_idx_option pack_idx_opts;
318 static unsigned int pack_id;
319 static struct hashfile *pack_file;
320 static struct packed_git *pack_data;
321 static struct packed_git **all_packs;
322 static off_t pack_size;
323
324 /* Table of objects we've written. */
325 static unsigned int object_entry_alloc = 5000;
326 static struct object_entry_pool *blocks;
327 static struct object_entry *object_table[1 << 16];
328 static struct mark_set *marks;
329 static const char *export_marks_file;
330 static const char *import_marks_file;
331 static int import_marks_file_from_stream;
332 static int import_marks_file_ignore_missing;
333 static int import_marks_file_done;
334 static int relative_marks_paths;
335
336 /* Our last blob */
337 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
338
339 /* Tree management */
340 static unsigned int tree_entry_alloc = 1000;
341 static void *avail_tree_entry;
342 static unsigned int avail_tree_table_sz = 100;
343 static struct avail_tree_content **avail_tree_table;
344 static struct strbuf old_tree = STRBUF_INIT;
345 static struct strbuf new_tree = STRBUF_INIT;
346
347 /* Branch data */
348 static unsigned long max_active_branches = 5;
349 static unsigned long cur_active_branches;
350 static unsigned long branch_table_sz = 1039;
351 static struct branch **branch_table;
352 static struct branch *active_branches;
353
354 /* Tag data */
355 static struct tag *first_tag;
356 static struct tag *last_tag;
357
358 /* Input stream parsing */
359 static whenspec_type whenspec = WHENSPEC_RAW;
360 static struct strbuf command_buf = STRBUF_INIT;
361 static int unread_command_buf;
362 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
363 static struct recent_command *cmd_tail = &cmd_hist;
364 static struct recent_command *rc_free;
365 static unsigned int cmd_save = 100;
366 static uintmax_t next_mark;
367 static struct strbuf new_data = STRBUF_INIT;
368 static int seen_data_command;
369 static int require_explicit_termination;
370
371 /* Signal handling */
372 static volatile sig_atomic_t checkpoint_requested;
373
374 /* Where to write output of cat-blob commands */
375 static int cat_blob_fd = STDOUT_FILENO;
376
377 static void parse_argv(void);
378 static void parse_get_mark(const char *p);
379 static void parse_cat_blob(const char *p);
380 static void parse_ls(const char *p, struct branch *b);
381
382 static void write_branch_report(FILE *rpt, struct branch *b)
383 {
384         fprintf(rpt, "%s:\n", b->name);
385
386         fprintf(rpt, "  status      :");
387         if (b->active)
388                 fputs(" active", rpt);
389         if (b->branch_tree.tree)
390                 fputs(" loaded", rpt);
391         if (is_null_oid(&b->branch_tree.versions[1].oid))
392                 fputs(" dirty", rpt);
393         fputc('\n', rpt);
394
395         fprintf(rpt, "  tip commit  : %s\n", oid_to_hex(&b->oid));
396         fprintf(rpt, "  old tree    : %s\n",
397                 oid_to_hex(&b->branch_tree.versions[0].oid));
398         fprintf(rpt, "  cur tree    : %s\n",
399                 oid_to_hex(&b->branch_tree.versions[1].oid));
400         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
401
402         fputs("  last pack   : ", rpt);
403         if (b->pack_id < MAX_PACK_ID)
404                 fprintf(rpt, "%u", b->pack_id);
405         fputc('\n', rpt);
406
407         fputc('\n', rpt);
408 }
409
410 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
411
412 static void write_crash_report(const char *err)
413 {
414         char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
415         FILE *rpt = fopen(loc, "w");
416         struct branch *b;
417         unsigned long lu;
418         struct recent_command *rc;
419
420         if (!rpt) {
421                 error_errno("can't write crash report %s", loc);
422                 free(loc);
423                 return;
424         }
425
426         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
427
428         fprintf(rpt, "fast-import crash report:\n");
429         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
430         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
431         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
432         fputc('\n', rpt);
433
434         fputs("fatal: ", rpt);
435         fputs(err, rpt);
436         fputc('\n', rpt);
437
438         fputc('\n', rpt);
439         fputs("Most Recent Commands Before Crash\n", rpt);
440         fputs("---------------------------------\n", rpt);
441         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
442                 if (rc->next == &cmd_hist)
443                         fputs("* ", rpt);
444                 else
445                         fputs("  ", rpt);
446                 fputs(rc->buf, rpt);
447                 fputc('\n', rpt);
448         }
449
450         fputc('\n', rpt);
451         fputs("Active Branch LRU\n", rpt);
452         fputs("-----------------\n", rpt);
453         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
454                 cur_active_branches,
455                 max_active_branches);
456         fputc('\n', rpt);
457         fputs("  pos  clock name\n", rpt);
458         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
459         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
460                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
461                         ++lu, b->last_commit, b->name);
462
463         fputc('\n', rpt);
464         fputs("Inactive Branches\n", rpt);
465         fputs("-----------------\n", rpt);
466         for (lu = 0; lu < branch_table_sz; lu++) {
467                 for (b = branch_table[lu]; b; b = b->table_next_branch)
468                         write_branch_report(rpt, b);
469         }
470
471         if (first_tag) {
472                 struct tag *tg;
473                 fputc('\n', rpt);
474                 fputs("Annotated Tags\n", rpt);
475                 fputs("--------------\n", rpt);
476                 for (tg = first_tag; tg; tg = tg->next_tag) {
477                         fputs(oid_to_hex(&tg->oid), rpt);
478                         fputc(' ', rpt);
479                         fputs(tg->name, rpt);
480                         fputc('\n', rpt);
481                 }
482         }
483
484         fputc('\n', rpt);
485         fputs("Marks\n", rpt);
486         fputs("-----\n", rpt);
487         if (export_marks_file)
488                 fprintf(rpt, "  exported to %s\n", export_marks_file);
489         else
490                 dump_marks_helper(rpt, 0, marks);
491
492         fputc('\n', rpt);
493         fputs("-------------------\n", rpt);
494         fputs("END OF CRASH REPORT\n", rpt);
495         fclose(rpt);
496         free(loc);
497 }
498
499 static void end_packfile(void);
500 static void unkeep_all_packs(void);
501 static void dump_marks(void);
502
503 static NORETURN void die_nicely(const char *err, va_list params)
504 {
505         static int zombie;
506         char message[2 * PATH_MAX];
507
508         vsnprintf(message, sizeof(message), err, params);
509         fputs("fatal: ", stderr);
510         fputs(message, stderr);
511         fputc('\n', stderr);
512
513         if (!zombie) {
514                 zombie = 1;
515                 write_crash_report(message);
516                 end_packfile();
517                 unkeep_all_packs();
518                 dump_marks();
519         }
520         exit(128);
521 }
522
523 #ifndef SIGUSR1 /* Windows, for example */
524
525 static void set_checkpoint_signal(void)
526 {
527 }
528
529 #else
530
531 static void checkpoint_signal(int signo)
532 {
533         checkpoint_requested = 1;
534 }
535
536 static void set_checkpoint_signal(void)
537 {
538         struct sigaction sa;
539
540         memset(&sa, 0, sizeof(sa));
541         sa.sa_handler = checkpoint_signal;
542         sigemptyset(&sa.sa_mask);
543         sa.sa_flags = SA_RESTART;
544         sigaction(SIGUSR1, &sa, NULL);
545 }
546
547 #endif
548
549 static void alloc_objects(unsigned int cnt)
550 {
551         struct object_entry_pool *b;
552
553         b = xmalloc(sizeof(struct object_entry_pool)
554                 + cnt * sizeof(struct object_entry));
555         b->next_pool = blocks;
556         b->next_free = b->entries;
557         b->end = b->entries + cnt;
558         blocks = b;
559         alloc_count += cnt;
560 }
561
562 static struct object_entry *new_object(struct object_id *oid)
563 {
564         struct object_entry *e;
565
566         if (blocks->next_free == blocks->end)
567                 alloc_objects(object_entry_alloc);
568
569         e = blocks->next_free++;
570         oidcpy(&e->idx.oid, oid);
571         return e;
572 }
573
574 static struct object_entry *find_object(struct object_id *oid)
575 {
576         unsigned int h = oid->hash[0] << 8 | oid->hash[1];
577         struct object_entry *e;
578         for (e = object_table[h]; e; e = e->next)
579                 if (!oidcmp(oid, &e->idx.oid))
580                         return e;
581         return NULL;
582 }
583
584 static struct object_entry *insert_object(struct object_id *oid)
585 {
586         unsigned int h = oid->hash[0] << 8 | oid->hash[1];
587         struct object_entry *e = object_table[h];
588
589         while (e) {
590                 if (!oidcmp(oid, &e->idx.oid))
591                         return e;
592                 e = e->next;
593         }
594
595         e = new_object(oid);
596         e->next = object_table[h];
597         e->idx.offset = 0;
598         object_table[h] = e;
599         return e;
600 }
601
602 static void invalidate_pack_id(unsigned int id)
603 {
604         unsigned int h;
605         unsigned long lu;
606         struct tag *t;
607
608         for (h = 0; h < ARRAY_SIZE(object_table); h++) {
609                 struct object_entry *e;
610
611                 for (e = object_table[h]; e; e = e->next)
612                         if (e->pack_id == id)
613                                 e->pack_id = MAX_PACK_ID;
614         }
615
616         for (lu = 0; lu < branch_table_sz; lu++) {
617                 struct branch *b;
618
619                 for (b = branch_table[lu]; b; b = b->table_next_branch)
620                         if (b->pack_id == id)
621                                 b->pack_id = MAX_PACK_ID;
622         }
623
624         for (t = first_tag; t; t = t->next_tag)
625                 if (t->pack_id == id)
626                         t->pack_id = MAX_PACK_ID;
627 }
628
629 static unsigned int hc_str(const char *s, size_t len)
630 {
631         unsigned int r = 0;
632         while (len-- > 0)
633                 r = r * 31 + *s++;
634         return r;
635 }
636
637 static void *pool_alloc(size_t len)
638 {
639         struct mem_pool *p;
640         void *r;
641
642         /* round up to a 'uintmax_t' alignment */
643         if (len & (sizeof(uintmax_t) - 1))
644                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
645
646         for (p = mem_pool; p; p = p->next_pool)
647                 if ((p->end - p->next_free >= len))
648                         break;
649
650         if (!p) {
651                 if (len >= (mem_pool_alloc/2)) {
652                         total_allocd += len;
653                         return xmalloc(len);
654                 }
655                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
656                 p = xmalloc(st_add(sizeof(struct mem_pool), mem_pool_alloc));
657                 p->next_pool = mem_pool;
658                 p->next_free = (char *) p->space;
659                 p->end = p->next_free + mem_pool_alloc;
660                 mem_pool = p;
661         }
662
663         r = p->next_free;
664         p->next_free += len;
665         return r;
666 }
667
668 static void *pool_calloc(size_t count, size_t size)
669 {
670         size_t len = count * size;
671         void *r = pool_alloc(len);
672         memset(r, 0, len);
673         return r;
674 }
675
676 static char *pool_strdup(const char *s)
677 {
678         size_t len = strlen(s) + 1;
679         char *r = pool_alloc(len);
680         memcpy(r, s, len);
681         return r;
682 }
683
684 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
685 {
686         struct mark_set *s = marks;
687         while ((idnum >> s->shift) >= 1024) {
688                 s = pool_calloc(1, sizeof(struct mark_set));
689                 s->shift = marks->shift + 10;
690                 s->data.sets[0] = marks;
691                 marks = s;
692         }
693         while (s->shift) {
694                 uintmax_t i = idnum >> s->shift;
695                 idnum -= i << s->shift;
696                 if (!s->data.sets[i]) {
697                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
698                         s->data.sets[i]->shift = s->shift - 10;
699                 }
700                 s = s->data.sets[i];
701         }
702         if (!s->data.marked[idnum])
703                 marks_set_count++;
704         s->data.marked[idnum] = oe;
705 }
706
707 static struct object_entry *find_mark(uintmax_t idnum)
708 {
709         uintmax_t orig_idnum = idnum;
710         struct mark_set *s = marks;
711         struct object_entry *oe = NULL;
712         if ((idnum >> s->shift) < 1024) {
713                 while (s && s->shift) {
714                         uintmax_t i = idnum >> s->shift;
715                         idnum -= i << s->shift;
716                         s = s->data.sets[i];
717                 }
718                 if (s)
719                         oe = s->data.marked[idnum];
720         }
721         if (!oe)
722                 die("mark :%" PRIuMAX " not declared", orig_idnum);
723         return oe;
724 }
725
726 static struct atom_str *to_atom(const char *s, unsigned short len)
727 {
728         unsigned int hc = hc_str(s, len) % atom_table_sz;
729         struct atom_str *c;
730
731         for (c = atom_table[hc]; c; c = c->next_atom)
732                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
733                         return c;
734
735         c = pool_alloc(sizeof(struct atom_str) + len + 1);
736         c->str_len = len;
737         memcpy(c->str_dat, s, len);
738         c->str_dat[len] = 0;
739         c->next_atom = atom_table[hc];
740         atom_table[hc] = c;
741         atom_cnt++;
742         return c;
743 }
744
745 static struct branch *lookup_branch(const char *name)
746 {
747         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
748         struct branch *b;
749
750         for (b = branch_table[hc]; b; b = b->table_next_branch)
751                 if (!strcmp(name, b->name))
752                         return b;
753         return NULL;
754 }
755
756 static struct branch *new_branch(const char *name)
757 {
758         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
759         struct branch *b = lookup_branch(name);
760
761         if (b)
762                 die("Invalid attempt to create duplicate branch: %s", name);
763         if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
764                 die("Branch name doesn't conform to GIT standards: %s", name);
765
766         b = pool_calloc(1, sizeof(struct branch));
767         b->name = pool_strdup(name);
768         b->table_next_branch = branch_table[hc];
769         b->branch_tree.versions[0].mode = S_IFDIR;
770         b->branch_tree.versions[1].mode = S_IFDIR;
771         b->num_notes = 0;
772         b->active = 0;
773         b->pack_id = MAX_PACK_ID;
774         branch_table[hc] = b;
775         branch_count++;
776         return b;
777 }
778
779 static unsigned int hc_entries(unsigned int cnt)
780 {
781         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
782         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
783 }
784
785 static struct tree_content *new_tree_content(unsigned int cnt)
786 {
787         struct avail_tree_content *f, *l = NULL;
788         struct tree_content *t;
789         unsigned int hc = hc_entries(cnt);
790
791         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
792                 if (f->entry_capacity >= cnt)
793                         break;
794
795         if (f) {
796                 if (l)
797                         l->next_avail = f->next_avail;
798                 else
799                         avail_tree_table[hc] = f->next_avail;
800         } else {
801                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
802                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
803                 f->entry_capacity = cnt;
804         }
805
806         t = (struct tree_content*)f;
807         t->entry_count = 0;
808         t->delta_depth = 0;
809         return t;
810 }
811
812 static void release_tree_entry(struct tree_entry *e);
813 static void release_tree_content(struct tree_content *t)
814 {
815         struct avail_tree_content *f = (struct avail_tree_content*)t;
816         unsigned int hc = hc_entries(f->entry_capacity);
817         f->next_avail = avail_tree_table[hc];
818         avail_tree_table[hc] = f;
819 }
820
821 static void release_tree_content_recursive(struct tree_content *t)
822 {
823         unsigned int i;
824         for (i = 0; i < t->entry_count; i++)
825                 release_tree_entry(t->entries[i]);
826         release_tree_content(t);
827 }
828
829 static struct tree_content *grow_tree_content(
830         struct tree_content *t,
831         int amt)
832 {
833         struct tree_content *r = new_tree_content(t->entry_count + amt);
834         r->entry_count = t->entry_count;
835         r->delta_depth = t->delta_depth;
836         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
837         release_tree_content(t);
838         return r;
839 }
840
841 static struct tree_entry *new_tree_entry(void)
842 {
843         struct tree_entry *e;
844
845         if (!avail_tree_entry) {
846                 unsigned int n = tree_entry_alloc;
847                 total_allocd += n * sizeof(struct tree_entry);
848                 ALLOC_ARRAY(e, n);
849                 avail_tree_entry = e;
850                 while (n-- > 1) {
851                         *((void**)e) = e + 1;
852                         e++;
853                 }
854                 *((void**)e) = NULL;
855         }
856
857         e = avail_tree_entry;
858         avail_tree_entry = *((void**)e);
859         return e;
860 }
861
862 static void release_tree_entry(struct tree_entry *e)
863 {
864         if (e->tree)
865                 release_tree_content_recursive(e->tree);
866         *((void**)e) = avail_tree_entry;
867         avail_tree_entry = e;
868 }
869
870 static struct tree_content *dup_tree_content(struct tree_content *s)
871 {
872         struct tree_content *d;
873         struct tree_entry *a, *b;
874         unsigned int i;
875
876         if (!s)
877                 return NULL;
878         d = new_tree_content(s->entry_count);
879         for (i = 0; i < s->entry_count; i++) {
880                 a = s->entries[i];
881                 b = new_tree_entry();
882                 memcpy(b, a, sizeof(*a));
883                 if (a->tree && is_null_oid(&b->versions[1].oid))
884                         b->tree = dup_tree_content(a->tree);
885                 else
886                         b->tree = NULL;
887                 d->entries[i] = b;
888         }
889         d->entry_count = s->entry_count;
890         d->delta_depth = s->delta_depth;
891
892         return d;
893 }
894
895 static void start_packfile(void)
896 {
897         struct strbuf tmp_file = STRBUF_INIT;
898         struct packed_git *p;
899         struct pack_header hdr;
900         int pack_fd;
901
902         pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
903         FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
904         strbuf_release(&tmp_file);
905
906         p->pack_fd = pack_fd;
907         p->do_not_close = 1;
908         pack_file = hashfd(pack_fd, p->pack_name);
909
910         hdr.hdr_signature = htonl(PACK_SIGNATURE);
911         hdr.hdr_version = htonl(2);
912         hdr.hdr_entries = 0;
913         hashwrite(pack_file, &hdr, sizeof(hdr));
914
915         pack_data = p;
916         pack_size = sizeof(hdr);
917         object_count = 0;
918
919         REALLOC_ARRAY(all_packs, pack_id + 1);
920         all_packs[pack_id] = p;
921 }
922
923 static const char *create_index(void)
924 {
925         const char *tmpfile;
926         struct pack_idx_entry **idx, **c, **last;
927         struct object_entry *e;
928         struct object_entry_pool *o;
929
930         /* Build the table of object IDs. */
931         ALLOC_ARRAY(idx, object_count);
932         c = idx;
933         for (o = blocks; o; o = o->next_pool)
934                 for (e = o->next_free; e-- != o->entries;)
935                         if (pack_id == e->pack_id)
936                                 *c++ = &e->idx;
937         last = idx + object_count;
938         if (c != last)
939                 die("internal consistency error creating the index");
940
941         tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
942         free(idx);
943         return tmpfile;
944 }
945
946 static char *keep_pack(const char *curr_index_name)
947 {
948         static const char *keep_msg = "fast-import";
949         struct strbuf name = STRBUF_INIT;
950         int keep_fd;
951
952         odb_pack_name(&name, pack_data->sha1, "keep");
953         keep_fd = odb_pack_keep(name.buf);
954         if (keep_fd < 0)
955                 die_errno("cannot create keep file");
956         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
957         if (close(keep_fd))
958                 die_errno("failed to write keep file");
959
960         odb_pack_name(&name, pack_data->sha1, "pack");
961         if (finalize_object_file(pack_data->pack_name, name.buf))
962                 die("cannot store pack file");
963
964         odb_pack_name(&name, pack_data->sha1, "idx");
965         if (finalize_object_file(curr_index_name, name.buf))
966                 die("cannot store index file");
967         free((void *)curr_index_name);
968         return strbuf_detach(&name, NULL);
969 }
970
971 static void unkeep_all_packs(void)
972 {
973         struct strbuf name = STRBUF_INIT;
974         int k;
975
976         for (k = 0; k < pack_id; k++) {
977                 struct packed_git *p = all_packs[k];
978                 odb_pack_name(&name, p->sha1, "keep");
979                 unlink_or_warn(name.buf);
980         }
981         strbuf_release(&name);
982 }
983
984 static int loosen_small_pack(const struct packed_git *p)
985 {
986         struct child_process unpack = CHILD_PROCESS_INIT;
987
988         if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
989                 die_errno("Failed seeking to start of '%s'", p->pack_name);
990
991         unpack.in = p->pack_fd;
992         unpack.git_cmd = 1;
993         unpack.stdout_to_stderr = 1;
994         argv_array_push(&unpack.args, "unpack-objects");
995         if (!show_stats)
996                 argv_array_push(&unpack.args, "-q");
997
998         return run_command(&unpack);
999 }
1000
1001 static void end_packfile(void)
1002 {
1003         static int running;
1004
1005         if (running || !pack_data)
1006                 return;
1007
1008         running = 1;
1009         clear_delta_base_cache();
1010         if (object_count) {
1011                 struct packed_git *new_p;
1012                 struct object_id cur_pack_oid;
1013                 char *idx_name;
1014                 int i;
1015                 struct branch *b;
1016                 struct tag *t;
1017
1018                 close_pack_windows(pack_data);
1019                 hashclose(pack_file, cur_pack_oid.hash, 0);
1020                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
1021                                     pack_data->pack_name, object_count,
1022                                     cur_pack_oid.hash, pack_size);
1023
1024                 if (object_count <= unpack_limit) {
1025                         if (!loosen_small_pack(pack_data)) {
1026                                 invalidate_pack_id(pack_id);
1027                                 goto discard_pack;
1028                         }
1029                 }
1030
1031                 close(pack_data->pack_fd);
1032                 idx_name = keep_pack(create_index());
1033
1034                 /* Register the packfile with core git's machinery. */
1035                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
1036                 if (!new_p)
1037                         die("core git rejected index %s", idx_name);
1038                 all_packs[pack_id] = new_p;
1039                 install_packed_git(new_p);
1040                 free(idx_name);
1041
1042                 /* Print the boundary */
1043                 if (pack_edges) {
1044                         fprintf(pack_edges, "%s:", new_p->pack_name);
1045                         for (i = 0; i < branch_table_sz; i++) {
1046                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
1047                                         if (b->pack_id == pack_id)
1048                                                 fprintf(pack_edges, " %s",
1049                                                         oid_to_hex(&b->oid));
1050                                 }
1051                         }
1052                         for (t = first_tag; t; t = t->next_tag) {
1053                                 if (t->pack_id == pack_id)
1054                                         fprintf(pack_edges, " %s",
1055                                                 oid_to_hex(&t->oid));
1056                         }
1057                         fputc('\n', pack_edges);
1058                         fflush(pack_edges);
1059                 }
1060
1061                 pack_id++;
1062         }
1063         else {
1064 discard_pack:
1065                 close(pack_data->pack_fd);
1066                 unlink_or_warn(pack_data->pack_name);
1067         }
1068         FREE_AND_NULL(pack_data);
1069         running = 0;
1070
1071         /* We can't carry a delta across packfiles. */
1072         strbuf_release(&last_blob.data);
1073         last_blob.offset = 0;
1074         last_blob.depth = 0;
1075 }
1076
1077 static void cycle_packfile(void)
1078 {
1079         end_packfile();
1080         start_packfile();
1081 }
1082
1083 static int store_object(
1084         enum object_type type,
1085         struct strbuf *dat,
1086         struct last_object *last,
1087         struct object_id *oidout,
1088         uintmax_t mark)
1089 {
1090         void *out, *delta;
1091         struct object_entry *e;
1092         unsigned char hdr[96];
1093         struct object_id oid;
1094         unsigned long hdrlen, deltalen;
1095         git_hash_ctx c;
1096         git_zstream s;
1097
1098         hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1099                            typename(type), (unsigned long)dat->len) + 1;
1100         the_hash_algo->init_fn(&c);
1101         the_hash_algo->update_fn(&c, hdr, hdrlen);
1102         the_hash_algo->update_fn(&c, dat->buf, dat->len);
1103         the_hash_algo->final_fn(oid.hash, &c);
1104         if (oidout)
1105                 oidcpy(oidout, &oid);
1106
1107         e = insert_object(&oid);
1108         if (mark)
1109                 insert_mark(mark, e);
1110         if (e->idx.offset) {
1111                 duplicate_count_by_type[type]++;
1112                 return 1;
1113         } else if (find_sha1_pack(oid.hash, packed_git)) {
1114                 e->type = type;
1115                 e->pack_id = MAX_PACK_ID;
1116                 e->idx.offset = 1; /* just not zero! */
1117                 duplicate_count_by_type[type]++;
1118                 return 1;
1119         }
1120
1121         if (last && last->data.buf && last->depth < max_depth
1122                 && dat->len > the_hash_algo->rawsz) {
1123
1124                 delta_count_attempts_by_type[type]++;
1125                 delta = diff_delta(last->data.buf, last->data.len,
1126                         dat->buf, dat->len,
1127                         &deltalen, dat->len - the_hash_algo->rawsz);
1128         } else
1129                 delta = NULL;
1130
1131         git_deflate_init(&s, pack_compression_level);
1132         if (delta) {
1133                 s.next_in = delta;
1134                 s.avail_in = deltalen;
1135         } else {
1136                 s.next_in = (void *)dat->buf;
1137                 s.avail_in = dat->len;
1138         }
1139         s.avail_out = git_deflate_bound(&s, s.avail_in);
1140         s.next_out = out = xmalloc(s.avail_out);
1141         while (git_deflate(&s, Z_FINISH) == Z_OK)
1142                 ; /* nothing */
1143         git_deflate_end(&s);
1144
1145         /* Determine if we should auto-checkpoint. */
1146         if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1147                 || (pack_size + 60 + s.total_out) < pack_size) {
1148
1149                 /* This new object needs to *not* have the current pack_id. */
1150                 e->pack_id = pack_id + 1;
1151                 cycle_packfile();
1152
1153                 /* We cannot carry a delta into the new pack. */
1154                 if (delta) {
1155                         FREE_AND_NULL(delta);
1156
1157                         git_deflate_init(&s, pack_compression_level);
1158                         s.next_in = (void *)dat->buf;
1159                         s.avail_in = dat->len;
1160                         s.avail_out = git_deflate_bound(&s, s.avail_in);
1161                         s.next_out = out = xrealloc(out, s.avail_out);
1162                         while (git_deflate(&s, Z_FINISH) == Z_OK)
1163                                 ; /* nothing */
1164                         git_deflate_end(&s);
1165                 }
1166         }
1167
1168         e->type = type;
1169         e->pack_id = pack_id;
1170         e->idx.offset = pack_size;
1171         object_count++;
1172         object_count_by_type[type]++;
1173
1174         crc32_begin(pack_file);
1175
1176         if (delta) {
1177                 off_t ofs = e->idx.offset - last->offset;
1178                 unsigned pos = sizeof(hdr) - 1;
1179
1180                 delta_count_by_type[type]++;
1181                 e->depth = last->depth + 1;
1182
1183                 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1184                                                       OBJ_OFS_DELTA, deltalen);
1185                 hashwrite(pack_file, hdr, hdrlen);
1186                 pack_size += hdrlen;
1187
1188                 hdr[pos] = ofs & 127;
1189                 while (ofs >>= 7)
1190                         hdr[--pos] = 128 | (--ofs & 127);
1191                 hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1192                 pack_size += sizeof(hdr) - pos;
1193         } else {
1194                 e->depth = 0;
1195                 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1196                                                       type, dat->len);
1197                 hashwrite(pack_file, hdr, hdrlen);
1198                 pack_size += hdrlen;
1199         }
1200
1201         hashwrite(pack_file, out, s.total_out);
1202         pack_size += s.total_out;
1203
1204         e->idx.crc32 = crc32_end(pack_file);
1205
1206         free(out);
1207         free(delta);
1208         if (last) {
1209                 if (last->no_swap) {
1210                         last->data = *dat;
1211                 } else {
1212                         strbuf_swap(&last->data, dat);
1213                 }
1214                 last->offset = e->idx.offset;
1215                 last->depth = e->depth;
1216         }
1217         return 0;
1218 }
1219
1220 static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1221 {
1222         if (hashfile_truncate(pack_file, checkpoint))
1223                 die_errno("cannot truncate pack to skip duplicate");
1224         pack_size = checkpoint->offset;
1225 }
1226
1227 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1228 {
1229         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1230         unsigned char *in_buf = xmalloc(in_sz);
1231         unsigned char *out_buf = xmalloc(out_sz);
1232         struct object_entry *e;
1233         struct object_id oid;
1234         unsigned long hdrlen;
1235         off_t offset;
1236         git_hash_ctx c;
1237         git_zstream s;
1238         struct hashfile_checkpoint checkpoint;
1239         int status = Z_OK;
1240
1241         /* Determine if we should auto-checkpoint. */
1242         if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1243                 || (pack_size + 60 + len) < pack_size)
1244                 cycle_packfile();
1245
1246         hashfile_checkpoint(pack_file, &checkpoint);
1247         offset = checkpoint.offset;
1248
1249         hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1250
1251         the_hash_algo->init_fn(&c);
1252         the_hash_algo->update_fn(&c, out_buf, hdrlen);
1253
1254         crc32_begin(pack_file);
1255
1256         git_deflate_init(&s, pack_compression_level);
1257
1258         hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1259
1260         s.next_out = out_buf + hdrlen;
1261         s.avail_out = out_sz - hdrlen;
1262
1263         while (status != Z_STREAM_END) {
1264                 if (0 < len && !s.avail_in) {
1265                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1266                         size_t n = fread(in_buf, 1, cnt, stdin);
1267                         if (!n && feof(stdin))
1268                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1269
1270                         the_hash_algo->update_fn(&c, in_buf, n);
1271                         s.next_in = in_buf;
1272                         s.avail_in = n;
1273                         len -= n;
1274                 }
1275
1276                 status = git_deflate(&s, len ? 0 : Z_FINISH);
1277
1278                 if (!s.avail_out || status == Z_STREAM_END) {
1279                         size_t n = s.next_out - out_buf;
1280                         hashwrite(pack_file, out_buf, n);
1281                         pack_size += n;
1282                         s.next_out = out_buf;
1283                         s.avail_out = out_sz;
1284                 }
1285
1286                 switch (status) {
1287                 case Z_OK:
1288                 case Z_BUF_ERROR:
1289                 case Z_STREAM_END:
1290                         continue;
1291                 default:
1292                         die("unexpected deflate failure: %d", status);
1293                 }
1294         }
1295         git_deflate_end(&s);
1296         the_hash_algo->final_fn(oid.hash, &c);
1297
1298         if (oidout)
1299                 oidcpy(oidout, &oid);
1300
1301         e = insert_object(&oid);
1302
1303         if (mark)
1304                 insert_mark(mark, e);
1305
1306         if (e->idx.offset) {
1307                 duplicate_count_by_type[OBJ_BLOB]++;
1308                 truncate_pack(&checkpoint);
1309
1310         } else if (find_sha1_pack(oid.hash, packed_git)) {
1311                 e->type = OBJ_BLOB;
1312                 e->pack_id = MAX_PACK_ID;
1313                 e->idx.offset = 1; /* just not zero! */
1314                 duplicate_count_by_type[OBJ_BLOB]++;
1315                 truncate_pack(&checkpoint);
1316
1317         } else {
1318                 e->depth = 0;
1319                 e->type = OBJ_BLOB;
1320                 e->pack_id = pack_id;
1321                 e->idx.offset = offset;
1322                 e->idx.crc32 = crc32_end(pack_file);
1323                 object_count++;
1324                 object_count_by_type[OBJ_BLOB]++;
1325         }
1326
1327         free(in_buf);
1328         free(out_buf);
1329 }
1330
1331 /* All calls must be guarded by find_object() or find_mark() to
1332  * ensure the 'struct object_entry' passed was written by this
1333  * process instance.  We unpack the entry by the offset, avoiding
1334  * the need for the corresponding .idx file.  This unpacking rule
1335  * works because we only use OBJ_REF_DELTA within the packfiles
1336  * created by fast-import.
1337  *
1338  * oe must not be NULL.  Such an oe usually comes from giving
1339  * an unknown SHA-1 to find_object() or an undefined mark to
1340  * find_mark().  Callers must test for this condition and use
1341  * the standard read_sha1_file() when it happens.
1342  *
1343  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1344  * find_mark(), where the mark was reloaded from an existing marks
1345  * file and is referencing an object that this fast-import process
1346  * instance did not write out to a packfile.  Callers must test for
1347  * this condition and use read_sha1_file() instead.
1348  */
1349 static void *gfi_unpack_entry(
1350         struct object_entry *oe,
1351         unsigned long *sizep)
1352 {
1353         enum object_type type;
1354         struct packed_git *p = all_packs[oe->pack_id];
1355         if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1356                 /* The object is stored in the packfile we are writing to
1357                  * and we have modified it since the last time we scanned
1358                  * back to read a previously written object.  If an old
1359                  * window covered [p->pack_size, p->pack_size + rawsz) its
1360                  * data is stale and is not valid.  Closing all windows
1361                  * and updating the packfile length ensures we can read
1362                  * the newly written data.
1363                  */
1364                 close_pack_windows(p);
1365                 hashflush(pack_file);
1366
1367                 /* We have to offer rawsz bytes additional on the end of
1368                  * the packfile as the core unpacker code assumes the
1369                  * footer is present at the file end and must promise
1370                  * at least rawsz bytes within any window it maps.  But
1371                  * we don't actually create the footer here.
1372                  */
1373                 p->pack_size = pack_size + the_hash_algo->rawsz;
1374         }
1375         return unpack_entry(p, oe->idx.offset, &type, sizep);
1376 }
1377
1378 static const char *get_mode(const char *str, uint16_t *modep)
1379 {
1380         unsigned char c;
1381         uint16_t mode = 0;
1382
1383         while ((c = *str++) != ' ') {
1384                 if (c < '0' || c > '7')
1385                         return NULL;
1386                 mode = (mode << 3) + (c - '0');
1387         }
1388         *modep = mode;
1389         return str;
1390 }
1391
1392 static void load_tree(struct tree_entry *root)
1393 {
1394         struct object_id *oid = &root->versions[1].oid;
1395         struct object_entry *myoe;
1396         struct tree_content *t;
1397         unsigned long size;
1398         char *buf;
1399         const char *c;
1400
1401         root->tree = t = new_tree_content(8);
1402         if (is_null_oid(oid))
1403                 return;
1404
1405         myoe = find_object(oid);
1406         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1407                 if (myoe->type != OBJ_TREE)
1408                         die("Not a tree: %s", oid_to_hex(oid));
1409                 t->delta_depth = myoe->depth;
1410                 buf = gfi_unpack_entry(myoe, &size);
1411                 if (!buf)
1412                         die("Can't load tree %s", oid_to_hex(oid));
1413         } else {
1414                 enum object_type type;
1415                 buf = read_sha1_file(oid->hash, &type, &size);
1416                 if (!buf || type != OBJ_TREE)
1417                         die("Can't load tree %s", oid_to_hex(oid));
1418         }
1419
1420         c = buf;
1421         while (c != (buf + size)) {
1422                 struct tree_entry *e = new_tree_entry();
1423
1424                 if (t->entry_count == t->entry_capacity)
1425                         root->tree = t = grow_tree_content(t, t->entry_count);
1426                 t->entries[t->entry_count++] = e;
1427
1428                 e->tree = NULL;
1429                 c = get_mode(c, &e->versions[1].mode);
1430                 if (!c)
1431                         die("Corrupt mode in %s", oid_to_hex(oid));
1432                 e->versions[0].mode = e->versions[1].mode;
1433                 e->name = to_atom(c, strlen(c));
1434                 c += e->name->str_len + 1;
1435                 hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1436                 hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1437                 c += GIT_SHA1_RAWSZ;
1438         }
1439         free(buf);
1440 }
1441
1442 static int tecmp0 (const void *_a, const void *_b)
1443 {
1444         struct tree_entry *a = *((struct tree_entry**)_a);
1445         struct tree_entry *b = *((struct tree_entry**)_b);
1446         return base_name_compare(
1447                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1448                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1449 }
1450
1451 static int tecmp1 (const void *_a, const void *_b)
1452 {
1453         struct tree_entry *a = *((struct tree_entry**)_a);
1454         struct tree_entry *b = *((struct tree_entry**)_b);
1455         return base_name_compare(
1456                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1457                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1458 }
1459
1460 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1461 {
1462         size_t maxlen = 0;
1463         unsigned int i;
1464
1465         if (!v)
1466                 QSORT(t->entries, t->entry_count, tecmp0);
1467         else
1468                 QSORT(t->entries, t->entry_count, tecmp1);
1469
1470         for (i = 0; i < t->entry_count; i++) {
1471                 if (t->entries[i]->versions[v].mode)
1472                         maxlen += t->entries[i]->name->str_len + 34;
1473         }
1474
1475         strbuf_reset(b);
1476         strbuf_grow(b, maxlen);
1477         for (i = 0; i < t->entry_count; i++) {
1478                 struct tree_entry *e = t->entries[i];
1479                 if (!e->versions[v].mode)
1480                         continue;
1481                 strbuf_addf(b, "%o %s%c",
1482                         (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1483                         e->name->str_dat, '\0');
1484                 strbuf_add(b, e->versions[v].oid.hash, GIT_SHA1_RAWSZ);
1485         }
1486 }
1487
1488 static void store_tree(struct tree_entry *root)
1489 {
1490         struct tree_content *t;
1491         unsigned int i, j, del;
1492         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1493         struct object_entry *le = NULL;
1494
1495         if (!is_null_oid(&root->versions[1].oid))
1496                 return;
1497
1498         if (!root->tree)
1499                 load_tree(root);
1500         t = root->tree;
1501
1502         for (i = 0; i < t->entry_count; i++) {
1503                 if (t->entries[i]->tree)
1504                         store_tree(t->entries[i]);
1505         }
1506
1507         if (!(root->versions[0].mode & NO_DELTA))
1508                 le = find_object(&root->versions[0].oid);
1509         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1510                 mktree(t, 0, &old_tree);
1511                 lo.data = old_tree;
1512                 lo.offset = le->idx.offset;
1513                 lo.depth = t->delta_depth;
1514         }
1515
1516         mktree(t, 1, &new_tree);
1517         store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1518
1519         t->delta_depth = lo.depth;
1520         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1521                 struct tree_entry *e = t->entries[i];
1522                 if (e->versions[1].mode) {
1523                         e->versions[0].mode = e->versions[1].mode;
1524                         oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1525                         t->entries[j++] = e;
1526                 } else {
1527                         release_tree_entry(e);
1528                         del++;
1529                 }
1530         }
1531         t->entry_count -= del;
1532 }
1533
1534 static void tree_content_replace(
1535         struct tree_entry *root,
1536         const struct object_id *oid,
1537         const uint16_t mode,
1538         struct tree_content *newtree)
1539 {
1540         if (!S_ISDIR(mode))
1541                 die("Root cannot be a non-directory");
1542         oidclr(&root->versions[0].oid);
1543         oidcpy(&root->versions[1].oid, oid);
1544         if (root->tree)
1545                 release_tree_content_recursive(root->tree);
1546         root->tree = newtree;
1547 }
1548
1549 static int tree_content_set(
1550         struct tree_entry *root,
1551         const char *p,
1552         const struct object_id *oid,
1553         const uint16_t mode,
1554         struct tree_content *subtree)
1555 {
1556         struct tree_content *t;
1557         const char *slash1;
1558         unsigned int i, n;
1559         struct tree_entry *e;
1560
1561         slash1 = strchrnul(p, '/');
1562         n = slash1 - p;
1563         if (!n)
1564                 die("Empty path component found in input");
1565         if (!*slash1 && !S_ISDIR(mode) && subtree)
1566                 die("Non-directories cannot have subtrees");
1567
1568         if (!root->tree)
1569                 load_tree(root);
1570         t = root->tree;
1571         for (i = 0; i < t->entry_count; i++) {
1572                 e = t->entries[i];
1573                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1574                         if (!*slash1) {
1575                                 if (!S_ISDIR(mode)
1576                                                 && e->versions[1].mode == mode
1577                                                 && !oidcmp(&e->versions[1].oid, oid))
1578                                         return 0;
1579                                 e->versions[1].mode = mode;
1580                                 oidcpy(&e->versions[1].oid, oid);
1581                                 if (e->tree)
1582                                         release_tree_content_recursive(e->tree);
1583                                 e->tree = subtree;
1584
1585                                 /*
1586                                  * We need to leave e->versions[0].sha1 alone
1587                                  * to avoid modifying the preimage tree used
1588                                  * when writing out the parent directory.
1589                                  * But after replacing the subdir with a
1590                                  * completely different one, it's not a good
1591                                  * delta base any more, and besides, we've
1592                                  * thrown away the tree entries needed to
1593                                  * make a delta against it.
1594                                  *
1595                                  * So let's just explicitly disable deltas
1596                                  * for the subtree.
1597                                  */
1598                                 if (S_ISDIR(e->versions[0].mode))
1599                                         e->versions[0].mode |= NO_DELTA;
1600
1601                                 oidclr(&root->versions[1].oid);
1602                                 return 1;
1603                         }
1604                         if (!S_ISDIR(e->versions[1].mode)) {
1605                                 e->tree = new_tree_content(8);
1606                                 e->versions[1].mode = S_IFDIR;
1607                         }
1608                         if (!e->tree)
1609                                 load_tree(e);
1610                         if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1611                                 oidclr(&root->versions[1].oid);
1612                                 return 1;
1613                         }
1614                         return 0;
1615                 }
1616         }
1617
1618         if (t->entry_count == t->entry_capacity)
1619                 root->tree = t = grow_tree_content(t, t->entry_count);
1620         e = new_tree_entry();
1621         e->name = to_atom(p, n);
1622         e->versions[0].mode = 0;
1623         oidclr(&e->versions[0].oid);
1624         t->entries[t->entry_count++] = e;
1625         if (*slash1) {
1626                 e->tree = new_tree_content(8);
1627                 e->versions[1].mode = S_IFDIR;
1628                 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1629         } else {
1630                 e->tree = subtree;
1631                 e->versions[1].mode = mode;
1632                 oidcpy(&e->versions[1].oid, oid);
1633         }
1634         oidclr(&root->versions[1].oid);
1635         return 1;
1636 }
1637
1638 static int tree_content_remove(
1639         struct tree_entry *root,
1640         const char *p,
1641         struct tree_entry *backup_leaf,
1642         int allow_root)
1643 {
1644         struct tree_content *t;
1645         const char *slash1;
1646         unsigned int i, n;
1647         struct tree_entry *e;
1648
1649         slash1 = strchrnul(p, '/');
1650         n = slash1 - p;
1651
1652         if (!root->tree)
1653                 load_tree(root);
1654
1655         if (!*p && allow_root) {
1656                 e = root;
1657                 goto del_entry;
1658         }
1659
1660         t = root->tree;
1661         for (i = 0; i < t->entry_count; i++) {
1662                 e = t->entries[i];
1663                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1664                         if (*slash1 && !S_ISDIR(e->versions[1].mode))
1665                                 /*
1666                                  * If p names a file in some subdirectory, and a
1667                                  * file or symlink matching the name of the
1668                                  * parent directory of p exists, then p cannot
1669                                  * exist and need not be deleted.
1670                                  */
1671                                 return 1;
1672                         if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1673                                 goto del_entry;
1674                         if (!e->tree)
1675                                 load_tree(e);
1676                         if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1677                                 for (n = 0; n < e->tree->entry_count; n++) {
1678                                         if (e->tree->entries[n]->versions[1].mode) {
1679                                                 oidclr(&root->versions[1].oid);
1680                                                 return 1;
1681                                         }
1682                                 }
1683                                 backup_leaf = NULL;
1684                                 goto del_entry;
1685                         }
1686                         return 0;
1687                 }
1688         }
1689         return 0;
1690
1691 del_entry:
1692         if (backup_leaf)
1693                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1694         else if (e->tree)
1695                 release_tree_content_recursive(e->tree);
1696         e->tree = NULL;
1697         e->versions[1].mode = 0;
1698         oidclr(&e->versions[1].oid);
1699         oidclr(&root->versions[1].oid);
1700         return 1;
1701 }
1702
1703 static int tree_content_get(
1704         struct tree_entry *root,
1705         const char *p,
1706         struct tree_entry *leaf,
1707         int allow_root)
1708 {
1709         struct tree_content *t;
1710         const char *slash1;
1711         unsigned int i, n;
1712         struct tree_entry *e;
1713
1714         slash1 = strchrnul(p, '/');
1715         n = slash1 - p;
1716         if (!n && !allow_root)
1717                 die("Empty path component found in input");
1718
1719         if (!root->tree)
1720                 load_tree(root);
1721
1722         if (!n) {
1723                 e = root;
1724                 goto found_entry;
1725         }
1726
1727         t = root->tree;
1728         for (i = 0; i < t->entry_count; i++) {
1729                 e = t->entries[i];
1730                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1731                         if (!*slash1)
1732                                 goto found_entry;
1733                         if (!S_ISDIR(e->versions[1].mode))
1734                                 return 0;
1735                         if (!e->tree)
1736                                 load_tree(e);
1737                         return tree_content_get(e, slash1 + 1, leaf, 0);
1738                 }
1739         }
1740         return 0;
1741
1742 found_entry:
1743         memcpy(leaf, e, sizeof(*leaf));
1744         if (e->tree && is_null_oid(&e->versions[1].oid))
1745                 leaf->tree = dup_tree_content(e->tree);
1746         else
1747                 leaf->tree = NULL;
1748         return 1;
1749 }
1750
1751 static int update_branch(struct branch *b)
1752 {
1753         static const char *msg = "fast-import";
1754         struct ref_transaction *transaction;
1755         struct object_id old_oid;
1756         struct strbuf err = STRBUF_INIT;
1757
1758         if (is_null_oid(&b->oid)) {
1759                 if (b->delete)
1760                         delete_ref(NULL, b->name, NULL, 0);
1761                 return 0;
1762         }
1763         if (read_ref(b->name, &old_oid))
1764                 oidclr(&old_oid);
1765         if (!force_update && !is_null_oid(&old_oid)) {
1766                 struct commit *old_cmit, *new_cmit;
1767
1768                 old_cmit = lookup_commit_reference_gently(&old_oid, 0);
1769                 new_cmit = lookup_commit_reference_gently(&b->oid, 0);
1770                 if (!old_cmit || !new_cmit)
1771                         return error("Branch %s is missing commits.", b->name);
1772
1773                 if (!in_merge_bases(old_cmit, new_cmit)) {
1774                         warning("Not updating %s"
1775                                 " (new tip %s does not contain %s)",
1776                                 b->name, oid_to_hex(&b->oid),
1777                                 oid_to_hex(&old_oid));
1778                         return -1;
1779                 }
1780         }
1781         transaction = ref_transaction_begin(&err);
1782         if (!transaction ||
1783             ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1784                                    0, msg, &err) ||
1785             ref_transaction_commit(transaction, &err)) {
1786                 ref_transaction_free(transaction);
1787                 error("%s", err.buf);
1788                 strbuf_release(&err);
1789                 return -1;
1790         }
1791         ref_transaction_free(transaction);
1792         strbuf_release(&err);
1793         return 0;
1794 }
1795
1796 static void dump_branches(void)
1797 {
1798         unsigned int i;
1799         struct branch *b;
1800
1801         for (i = 0; i < branch_table_sz; i++) {
1802                 for (b = branch_table[i]; b; b = b->table_next_branch)
1803                         failure |= update_branch(b);
1804         }
1805 }
1806
1807 static void dump_tags(void)
1808 {
1809         static const char *msg = "fast-import";
1810         struct tag *t;
1811         struct strbuf ref_name = STRBUF_INIT;
1812         struct strbuf err = STRBUF_INIT;
1813         struct ref_transaction *transaction;
1814
1815         transaction = ref_transaction_begin(&err);
1816         if (!transaction) {
1817                 failure |= error("%s", err.buf);
1818                 goto cleanup;
1819         }
1820         for (t = first_tag; t; t = t->next_tag) {
1821                 strbuf_reset(&ref_name);
1822                 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1823
1824                 if (ref_transaction_update(transaction, ref_name.buf,
1825                                            &t->oid, NULL, 0, msg, &err)) {
1826                         failure |= error("%s", err.buf);
1827                         goto cleanup;
1828                 }
1829         }
1830         if (ref_transaction_commit(transaction, &err))
1831                 failure |= error("%s", err.buf);
1832
1833  cleanup:
1834         ref_transaction_free(transaction);
1835         strbuf_release(&ref_name);
1836         strbuf_release(&err);
1837 }
1838
1839 static void dump_marks_helper(FILE *f,
1840         uintmax_t base,
1841         struct mark_set *m)
1842 {
1843         uintmax_t k;
1844         if (m->shift) {
1845                 for (k = 0; k < 1024; k++) {
1846                         if (m->data.sets[k])
1847                                 dump_marks_helper(f, base + (k << m->shift),
1848                                         m->data.sets[k]);
1849                 }
1850         } else {
1851                 for (k = 0; k < 1024; k++) {
1852                         if (m->data.marked[k])
1853                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1854                                         oid_to_hex(&m->data.marked[k]->idx.oid));
1855                 }
1856         }
1857 }
1858
1859 static void dump_marks(void)
1860 {
1861         static struct lock_file mark_lock;
1862         FILE *f;
1863
1864         if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1865                 return;
1866
1867         if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1868                 failure |= error_errno("Unable to write marks file %s",
1869                                        export_marks_file);
1870                 return;
1871         }
1872
1873         f = fdopen_lock_file(&mark_lock, "w");
1874         if (!f) {
1875                 int saved_errno = errno;
1876                 rollback_lock_file(&mark_lock);
1877                 failure |= error("Unable to write marks file %s: %s",
1878                         export_marks_file, strerror(saved_errno));
1879                 return;
1880         }
1881
1882         dump_marks_helper(f, 0, marks);
1883         if (commit_lock_file(&mark_lock)) {
1884                 failure |= error_errno("Unable to write file %s",
1885                                        export_marks_file);
1886                 return;
1887         }
1888 }
1889
1890 static void read_marks(void)
1891 {
1892         char line[512];
1893         FILE *f = fopen(import_marks_file, "r");
1894         if (f)
1895                 ;
1896         else if (import_marks_file_ignore_missing && errno == ENOENT)
1897                 goto done; /* Marks file does not exist */
1898         else
1899                 die_errno("cannot read '%s'", import_marks_file);
1900         while (fgets(line, sizeof(line), f)) {
1901                 uintmax_t mark;
1902                 char *end;
1903                 struct object_id oid;
1904                 struct object_entry *e;
1905
1906                 end = strchr(line, '\n');
1907                 if (line[0] != ':' || !end)
1908                         die("corrupt mark line: %s", line);
1909                 *end = 0;
1910                 mark = strtoumax(line + 1, &end, 10);
1911                 if (!mark || end == line + 1
1912                         || *end != ' ' || get_oid_hex(end + 1, &oid))
1913                         die("corrupt mark line: %s", line);
1914                 e = find_object(&oid);
1915                 if (!e) {
1916                         enum object_type type = sha1_object_info(oid.hash, NULL);
1917                         if (type < 0)
1918                                 die("object not found: %s", oid_to_hex(&oid));
1919                         e = insert_object(&oid);
1920                         e->type = type;
1921                         e->pack_id = MAX_PACK_ID;
1922                         e->idx.offset = 1; /* just not zero! */
1923                 }
1924                 insert_mark(mark, e);
1925         }
1926         fclose(f);
1927 done:
1928         import_marks_file_done = 1;
1929 }
1930
1931
1932 static int read_next_command(void)
1933 {
1934         static int stdin_eof = 0;
1935
1936         if (stdin_eof) {
1937                 unread_command_buf = 0;
1938                 return EOF;
1939         }
1940
1941         for (;;) {
1942                 const char *p;
1943
1944                 if (unread_command_buf) {
1945                         unread_command_buf = 0;
1946                 } else {
1947                         struct recent_command *rc;
1948
1949                         strbuf_detach(&command_buf, NULL);
1950                         stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1951                         if (stdin_eof)
1952                                 return EOF;
1953
1954                         if (!seen_data_command
1955                                 && !starts_with(command_buf.buf, "feature ")
1956                                 && !starts_with(command_buf.buf, "option ")) {
1957                                 parse_argv();
1958                         }
1959
1960                         rc = rc_free;
1961                         if (rc)
1962                                 rc_free = rc->next;
1963                         else {
1964                                 rc = cmd_hist.next;
1965                                 cmd_hist.next = rc->next;
1966                                 cmd_hist.next->prev = &cmd_hist;
1967                                 free(rc->buf);
1968                         }
1969
1970                         rc->buf = command_buf.buf;
1971                         rc->prev = cmd_tail;
1972                         rc->next = cmd_hist.prev;
1973                         rc->prev->next = rc;
1974                         cmd_tail = rc;
1975                 }
1976                 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1977                         parse_get_mark(p);
1978                         continue;
1979                 }
1980                 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1981                         parse_cat_blob(p);
1982                         continue;
1983                 }
1984                 if (command_buf.buf[0] == '#')
1985                         continue;
1986                 return 0;
1987         }
1988 }
1989
1990 static void skip_optional_lf(void)
1991 {
1992         int term_char = fgetc(stdin);
1993         if (term_char != '\n' && term_char != EOF)
1994                 ungetc(term_char, stdin);
1995 }
1996
1997 static void parse_mark(void)
1998 {
1999         const char *v;
2000         if (skip_prefix(command_buf.buf, "mark :", &v)) {
2001                 next_mark = strtoumax(v, NULL, 10);
2002                 read_next_command();
2003         }
2004         else
2005                 next_mark = 0;
2006 }
2007
2008 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
2009 {
2010         const char *data;
2011         strbuf_reset(sb);
2012
2013         if (!skip_prefix(command_buf.buf, "data ", &data))
2014                 die("Expected 'data n' command, found: %s", command_buf.buf);
2015
2016         if (skip_prefix(data, "<<", &data)) {
2017                 char *term = xstrdup(data);
2018                 size_t term_len = command_buf.len - (data - command_buf.buf);
2019
2020                 strbuf_detach(&command_buf, NULL);
2021                 for (;;) {
2022                         if (strbuf_getline_lf(&command_buf, stdin) == EOF)
2023                                 die("EOF in data (terminator '%s' not found)", term);
2024                         if (term_len == command_buf.len
2025                                 && !strcmp(term, command_buf.buf))
2026                                 break;
2027                         strbuf_addbuf(sb, &command_buf);
2028                         strbuf_addch(sb, '\n');
2029                 }
2030                 free(term);
2031         }
2032         else {
2033                 uintmax_t len = strtoumax(data, NULL, 10);
2034                 size_t n = 0, length = (size_t)len;
2035
2036                 if (limit && limit < len) {
2037                         *len_res = len;
2038                         return 0;
2039                 }
2040                 if (length < len)
2041                         die("data is too large to use in this context");
2042
2043                 while (n < length) {
2044                         size_t s = strbuf_fread(sb, length - n, stdin);
2045                         if (!s && feof(stdin))
2046                                 die("EOF in data (%lu bytes remaining)",
2047                                         (unsigned long)(length - n));
2048                         n += s;
2049                 }
2050         }
2051
2052         skip_optional_lf();
2053         return 1;
2054 }
2055
2056 static int validate_raw_date(const char *src, struct strbuf *result)
2057 {
2058         const char *orig_src = src;
2059         char *endp;
2060         unsigned long num;
2061
2062         errno = 0;
2063
2064         num = strtoul(src, &endp, 10);
2065         /* NEEDSWORK: perhaps check for reasonable values? */
2066         if (errno || endp == src || *endp != ' ')
2067                 return -1;
2068
2069         src = endp + 1;
2070         if (*src != '-' && *src != '+')
2071                 return -1;
2072
2073         num = strtoul(src + 1, &endp, 10);
2074         if (errno || endp == src + 1 || *endp || 1400 < num)
2075                 return -1;
2076
2077         strbuf_addstr(result, orig_src);
2078         return 0;
2079 }
2080
2081 static char *parse_ident(const char *buf)
2082 {
2083         const char *ltgt;
2084         size_t name_len;
2085         struct strbuf ident = STRBUF_INIT;
2086
2087         /* ensure there is a space delimiter even if there is no name */
2088         if (*buf == '<')
2089                 --buf;
2090
2091         ltgt = buf + strcspn(buf, "<>");
2092         if (*ltgt != '<')
2093                 die("Missing < in ident string: %s", buf);
2094         if (ltgt != buf && ltgt[-1] != ' ')
2095                 die("Missing space before < in ident string: %s", buf);
2096         ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2097         if (*ltgt != '>')
2098                 die("Missing > in ident string: %s", buf);
2099         ltgt++;
2100         if (*ltgt != ' ')
2101                 die("Missing space after > in ident string: %s", buf);
2102         ltgt++;
2103         name_len = ltgt - buf;
2104         strbuf_add(&ident, buf, name_len);
2105
2106         switch (whenspec) {
2107         case WHENSPEC_RAW:
2108                 if (validate_raw_date(ltgt, &ident) < 0)
2109                         die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2110                 break;
2111         case WHENSPEC_RFC2822:
2112                 if (parse_date(ltgt, &ident) < 0)
2113                         die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2114                 break;
2115         case WHENSPEC_NOW:
2116                 if (strcmp("now", ltgt))
2117                         die("Date in ident must be 'now': %s", buf);
2118                 datestamp(&ident);
2119                 break;
2120         }
2121
2122         return strbuf_detach(&ident, NULL);
2123 }
2124
2125 static void parse_and_store_blob(
2126         struct last_object *last,
2127         struct object_id *oidout,
2128         uintmax_t mark)
2129 {
2130         static struct strbuf buf = STRBUF_INIT;
2131         uintmax_t len;
2132
2133         if (parse_data(&buf, big_file_threshold, &len))
2134                 store_object(OBJ_BLOB, &buf, last, oidout, mark);
2135         else {
2136                 if (last) {
2137                         strbuf_release(&last->data);
2138                         last->offset = 0;
2139                         last->depth = 0;
2140                 }
2141                 stream_blob(len, oidout, mark);
2142                 skip_optional_lf();
2143         }
2144 }
2145
2146 static void parse_new_blob(void)
2147 {
2148         read_next_command();
2149         parse_mark();
2150         parse_and_store_blob(&last_blob, NULL, next_mark);
2151 }
2152
2153 static void unload_one_branch(void)
2154 {
2155         while (cur_active_branches
2156                 && cur_active_branches >= max_active_branches) {
2157                 uintmax_t min_commit = ULONG_MAX;
2158                 struct branch *e, *l = NULL, *p = NULL;
2159
2160                 for (e = active_branches; e; e = e->active_next_branch) {
2161                         if (e->last_commit < min_commit) {
2162                                 p = l;
2163                                 min_commit = e->last_commit;
2164                         }
2165                         l = e;
2166                 }
2167
2168                 if (p) {
2169                         e = p->active_next_branch;
2170                         p->active_next_branch = e->active_next_branch;
2171                 } else {
2172                         e = active_branches;
2173                         active_branches = e->active_next_branch;
2174                 }
2175                 e->active = 0;
2176                 e->active_next_branch = NULL;
2177                 if (e->branch_tree.tree) {
2178                         release_tree_content_recursive(e->branch_tree.tree);
2179                         e->branch_tree.tree = NULL;
2180                 }
2181                 cur_active_branches--;
2182         }
2183 }
2184
2185 static void load_branch(struct branch *b)
2186 {
2187         load_tree(&b->branch_tree);
2188         if (!b->active) {
2189                 b->active = 1;
2190                 b->active_next_branch = active_branches;
2191                 active_branches = b;
2192                 cur_active_branches++;
2193                 branch_load_count++;
2194         }
2195 }
2196
2197 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2198 {
2199         unsigned char fanout = 0;
2200         while ((num_notes >>= 8))
2201                 fanout++;
2202         return fanout;
2203 }
2204
2205 static void construct_path_with_fanout(const char *hex_sha1,
2206                 unsigned char fanout, char *path)
2207 {
2208         unsigned int i = 0, j = 0;
2209         if (fanout >= the_hash_algo->rawsz)
2210                 die("Too large fanout (%u)", fanout);
2211         while (fanout) {
2212                 path[i++] = hex_sha1[j++];
2213                 path[i++] = hex_sha1[j++];
2214                 path[i++] = '/';
2215                 fanout--;
2216         }
2217         memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2218         path[i + the_hash_algo->hexsz - j] = '\0';
2219 }
2220
2221 static uintmax_t do_change_note_fanout(
2222                 struct tree_entry *orig_root, struct tree_entry *root,
2223                 char *hex_oid, unsigned int hex_oid_len,
2224                 char *fullpath, unsigned int fullpath_len,
2225                 unsigned char fanout)
2226 {
2227         struct tree_content *t;
2228         struct tree_entry *e, leaf;
2229         unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2230         uintmax_t num_notes = 0;
2231         struct object_id oid;
2232         char realpath[60];
2233
2234         if (!root->tree)
2235                 load_tree(root);
2236         t = root->tree;
2237
2238         for (i = 0; t && i < t->entry_count; i++) {
2239                 e = t->entries[i];
2240                 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2241                 tmp_fullpath_len = fullpath_len;
2242
2243                 /*
2244                  * We're interested in EITHER existing note entries (entries
2245                  * with exactly 40 hex chars in path, not including directory
2246                  * separators), OR directory entries that may contain note
2247                  * entries (with < 40 hex chars in path).
2248                  * Also, each path component in a note entry must be a multiple
2249                  * of 2 chars.
2250                  */
2251                 if (!e->versions[1].mode ||
2252                     tmp_hex_oid_len > GIT_SHA1_HEXSZ ||
2253                     e->name->str_len % 2)
2254                         continue;
2255
2256                 /* This _may_ be a note entry, or a subdir containing notes */
2257                 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2258                        e->name->str_len);
2259                 if (tmp_fullpath_len)
2260                         fullpath[tmp_fullpath_len++] = '/';
2261                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2262                        e->name->str_len);
2263                 tmp_fullpath_len += e->name->str_len;
2264                 fullpath[tmp_fullpath_len] = '\0';
2265
2266                 if (tmp_hex_oid_len == GIT_SHA1_HEXSZ && !get_oid_hex(hex_oid, &oid)) {
2267                         /* This is a note entry */
2268                         if (fanout == 0xff) {
2269                                 /* Counting mode, no rename */
2270                                 num_notes++;
2271                                 continue;
2272                         }
2273                         construct_path_with_fanout(hex_oid, fanout, realpath);
2274                         if (!strcmp(fullpath, realpath)) {
2275                                 /* Note entry is in correct location */
2276                                 num_notes++;
2277                                 continue;
2278                         }
2279
2280                         /* Rename fullpath to realpath */
2281                         if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2282                                 die("Failed to remove path %s", fullpath);
2283                         tree_content_set(orig_root, realpath,
2284                                 &leaf.versions[1].oid,
2285                                 leaf.versions[1].mode,
2286                                 leaf.tree);
2287                 } else if (S_ISDIR(e->versions[1].mode)) {
2288                         /* This is a subdir that may contain note entries */
2289                         num_notes += do_change_note_fanout(orig_root, e,
2290                                 hex_oid, tmp_hex_oid_len,
2291                                 fullpath, tmp_fullpath_len, fanout);
2292                 }
2293
2294                 /* The above may have reallocated the current tree_content */
2295                 t = root->tree;
2296         }
2297         return num_notes;
2298 }
2299
2300 static uintmax_t change_note_fanout(struct tree_entry *root,
2301                 unsigned char fanout)
2302 {
2303         /*
2304          * The size of path is due to one slash between every two hex digits,
2305          * plus the terminating NUL.  Note that there is no slash at the end, so
2306          * the number of slashes is one less than half the number of hex
2307          * characters.
2308          */
2309         char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2310         return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2311 }
2312
2313 /*
2314  * Given a pointer into a string, parse a mark reference:
2315  *
2316  *   idnum ::= ':' bigint;
2317  *
2318  * Return the first character after the value in *endptr.
2319  *
2320  * Complain if the following character is not what is expected,
2321  * either a space or end of the string.
2322  */
2323 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2324 {
2325         uintmax_t mark;
2326
2327         assert(*p == ':');
2328         p++;
2329         mark = strtoumax(p, endptr, 10);
2330         if (*endptr == p)
2331                 die("No value after ':' in mark: %s", command_buf.buf);
2332         return mark;
2333 }
2334
2335 /*
2336  * Parse the mark reference, and complain if this is not the end of
2337  * the string.
2338  */
2339 static uintmax_t parse_mark_ref_eol(const char *p)
2340 {
2341         char *end;
2342         uintmax_t mark;
2343
2344         mark = parse_mark_ref(p, &end);
2345         if (*end != '\0')
2346                 die("Garbage after mark: %s", command_buf.buf);
2347         return mark;
2348 }
2349
2350 /*
2351  * Parse the mark reference, demanding a trailing space.  Return a
2352  * pointer to the space.
2353  */
2354 static uintmax_t parse_mark_ref_space(const char **p)
2355 {
2356         uintmax_t mark;
2357         char *end;
2358
2359         mark = parse_mark_ref(*p, &end);
2360         if (*end++ != ' ')
2361                 die("Missing space after mark: %s", command_buf.buf);
2362         *p = end;
2363         return mark;
2364 }
2365
2366 static void file_change_m(const char *p, struct branch *b)
2367 {
2368         static struct strbuf uq = STRBUF_INIT;
2369         const char *endp;
2370         struct object_entry *oe;
2371         struct object_id oid;
2372         uint16_t mode, inline_data = 0;
2373
2374         p = get_mode(p, &mode);
2375         if (!p)
2376                 die("Corrupt mode: %s", command_buf.buf);
2377         switch (mode) {
2378         case 0644:
2379         case 0755:
2380                 mode |= S_IFREG;
2381         case S_IFREG | 0644:
2382         case S_IFREG | 0755:
2383         case S_IFLNK:
2384         case S_IFDIR:
2385         case S_IFGITLINK:
2386                 /* ok */
2387                 break;
2388         default:
2389                 die("Corrupt mode: %s", command_buf.buf);
2390         }
2391
2392         if (*p == ':') {
2393                 oe = find_mark(parse_mark_ref_space(&p));
2394                 oidcpy(&oid, &oe->idx.oid);
2395         } else if (skip_prefix(p, "inline ", &p)) {
2396                 inline_data = 1;
2397                 oe = NULL; /* not used with inline_data, but makes gcc happy */
2398         } else {
2399                 if (parse_oid_hex(p, &oid, &p))
2400                         die("Invalid dataref: %s", command_buf.buf);
2401                 oe = find_object(&oid);
2402                 if (*p++ != ' ')
2403                         die("Missing space after SHA1: %s", command_buf.buf);
2404         }
2405
2406         strbuf_reset(&uq);
2407         if (!unquote_c_style(&uq, p, &endp)) {
2408                 if (*endp)
2409                         die("Garbage after path in: %s", command_buf.buf);
2410                 p = uq.buf;
2411         }
2412
2413         /* Git does not track empty, non-toplevel directories. */
2414         if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2415                 tree_content_remove(&b->branch_tree, p, NULL, 0);
2416                 return;
2417         }
2418
2419         if (S_ISGITLINK(mode)) {
2420                 if (inline_data)
2421                         die("Git links cannot be specified 'inline': %s",
2422                                 command_buf.buf);
2423                 else if (oe) {
2424                         if (oe->type != OBJ_COMMIT)
2425                                 die("Not a commit (actually a %s): %s",
2426                                         typename(oe->type), command_buf.buf);
2427                 }
2428                 /*
2429                  * Accept the sha1 without checking; it expected to be in
2430                  * another repository.
2431                  */
2432         } else if (inline_data) {
2433                 if (S_ISDIR(mode))
2434                         die("Directories cannot be specified 'inline': %s",
2435                                 command_buf.buf);
2436                 if (p != uq.buf) {
2437                         strbuf_addstr(&uq, p);
2438                         p = uq.buf;
2439                 }
2440                 read_next_command();
2441                 parse_and_store_blob(&last_blob, &oid, 0);
2442         } else {
2443                 enum object_type expected = S_ISDIR(mode) ?
2444                                                 OBJ_TREE: OBJ_BLOB;
2445                 enum object_type type = oe ? oe->type :
2446                                         sha1_object_info(oid.hash, NULL);
2447                 if (type < 0)
2448                         die("%s not found: %s",
2449                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2450                                         command_buf.buf);
2451                 if (type != expected)
2452                         die("Not a %s (actually a %s): %s",
2453                                 typename(expected), typename(type),
2454                                 command_buf.buf);
2455         }
2456
2457         if (!*p) {
2458                 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2459                 return;
2460         }
2461         tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2462 }
2463
2464 static void file_change_d(const char *p, struct branch *b)
2465 {
2466         static struct strbuf uq = STRBUF_INIT;
2467         const char *endp;
2468
2469         strbuf_reset(&uq);
2470         if (!unquote_c_style(&uq, p, &endp)) {
2471                 if (*endp)
2472                         die("Garbage after path in: %s", command_buf.buf);
2473                 p = uq.buf;
2474         }
2475         tree_content_remove(&b->branch_tree, p, NULL, 1);
2476 }
2477
2478 static void file_change_cr(const char *s, struct branch *b, int rename)
2479 {
2480         const char *d;
2481         static struct strbuf s_uq = STRBUF_INIT;
2482         static struct strbuf d_uq = STRBUF_INIT;
2483         const char *endp;
2484         struct tree_entry leaf;
2485
2486         strbuf_reset(&s_uq);
2487         if (!unquote_c_style(&s_uq, s, &endp)) {
2488                 if (*endp != ' ')
2489                         die("Missing space after source: %s", command_buf.buf);
2490         } else {
2491                 endp = strchr(s, ' ');
2492                 if (!endp)
2493                         die("Missing space after source: %s", command_buf.buf);
2494                 strbuf_add(&s_uq, s, endp - s);
2495         }
2496         s = s_uq.buf;
2497
2498         endp++;
2499         if (!*endp)
2500                 die("Missing dest: %s", command_buf.buf);
2501
2502         d = endp;
2503         strbuf_reset(&d_uq);
2504         if (!unquote_c_style(&d_uq, d, &endp)) {
2505                 if (*endp)
2506                         die("Garbage after dest in: %s", command_buf.buf);
2507                 d = d_uq.buf;
2508         }
2509
2510         memset(&leaf, 0, sizeof(leaf));
2511         if (rename)
2512                 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2513         else
2514                 tree_content_get(&b->branch_tree, s, &leaf, 1);
2515         if (!leaf.versions[1].mode)
2516                 die("Path %s not in branch", s);
2517         if (!*d) {      /* C "path/to/subdir" "" */
2518                 tree_content_replace(&b->branch_tree,
2519                         &leaf.versions[1].oid,
2520                         leaf.versions[1].mode,
2521                         leaf.tree);
2522                 return;
2523         }
2524         tree_content_set(&b->branch_tree, d,
2525                 &leaf.versions[1].oid,
2526                 leaf.versions[1].mode,
2527                 leaf.tree);
2528 }
2529
2530 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2531 {
2532         static struct strbuf uq = STRBUF_INIT;
2533         struct object_entry *oe;
2534         struct branch *s;
2535         struct object_id oid, commit_oid;
2536         char path[60];
2537         uint16_t inline_data = 0;
2538         unsigned char new_fanout;
2539
2540         /*
2541          * When loading a branch, we don't traverse its tree to count the real
2542          * number of notes (too expensive to do this for all non-note refs).
2543          * This means that recently loaded notes refs might incorrectly have
2544          * b->num_notes == 0, and consequently, old_fanout might be wrong.
2545          *
2546          * Fix this by traversing the tree and counting the number of notes
2547          * when b->num_notes == 0. If the notes tree is truly empty, the
2548          * calculation should not take long.
2549          */
2550         if (b->num_notes == 0 && *old_fanout == 0) {
2551                 /* Invoke change_note_fanout() in "counting mode". */
2552                 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2553                 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2554         }
2555
2556         /* Now parse the notemodify command. */
2557         /* <dataref> or 'inline' */
2558         if (*p == ':') {
2559                 oe = find_mark(parse_mark_ref_space(&p));
2560                 oidcpy(&oid, &oe->idx.oid);
2561         } else if (skip_prefix(p, "inline ", &p)) {
2562                 inline_data = 1;
2563                 oe = NULL; /* not used with inline_data, but makes gcc happy */
2564         } else {
2565                 if (parse_oid_hex(p, &oid, &p))
2566                         die("Invalid dataref: %s", command_buf.buf);
2567                 oe = find_object(&oid);
2568                 if (*p++ != ' ')
2569                         die("Missing space after SHA1: %s", command_buf.buf);
2570         }
2571
2572         /* <commit-ish> */
2573         s = lookup_branch(p);
2574         if (s) {
2575                 if (is_null_oid(&s->oid))
2576                         die("Can't add a note on empty branch.");
2577                 oidcpy(&commit_oid, &s->oid);
2578         } else if (*p == ':') {
2579                 uintmax_t commit_mark = parse_mark_ref_eol(p);
2580                 struct object_entry *commit_oe = find_mark(commit_mark);
2581                 if (commit_oe->type != OBJ_COMMIT)
2582                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2583                 oidcpy(&commit_oid, &commit_oe->idx.oid);
2584         } else if (!get_oid(p, &commit_oid)) {
2585                 unsigned long size;
2586                 char *buf = read_object_with_reference(commit_oid.hash,
2587                         commit_type, &size, commit_oid.hash);
2588                 if (!buf || size < 46)
2589                         die("Not a valid commit: %s", p);
2590                 free(buf);
2591         } else
2592                 die("Invalid ref name or SHA1 expression: %s", p);
2593
2594         if (inline_data) {
2595                 if (p != uq.buf) {
2596                         strbuf_addstr(&uq, p);
2597                         p = uq.buf;
2598                 }
2599                 read_next_command();
2600                 parse_and_store_blob(&last_blob, &oid, 0);
2601         } else if (oe) {
2602                 if (oe->type != OBJ_BLOB)
2603                         die("Not a blob (actually a %s): %s",
2604                                 typename(oe->type), command_buf.buf);
2605         } else if (!is_null_oid(&oid)) {
2606                 enum object_type type = sha1_object_info(oid.hash, NULL);
2607                 if (type < 0)
2608                         die("Blob not found: %s", command_buf.buf);
2609                 if (type != OBJ_BLOB)
2610                         die("Not a blob (actually a %s): %s",
2611                             typename(type), command_buf.buf);
2612         }
2613
2614         construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2615         if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2616                 b->num_notes--;
2617
2618         if (is_null_oid(&oid))
2619                 return; /* nothing to insert */
2620
2621         b->num_notes++;
2622         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2623         construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2624         tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2625 }
2626
2627 static void file_change_deleteall(struct branch *b)
2628 {
2629         release_tree_content_recursive(b->branch_tree.tree);
2630         oidclr(&b->branch_tree.versions[0].oid);
2631         oidclr(&b->branch_tree.versions[1].oid);
2632         load_tree(&b->branch_tree);
2633         b->num_notes = 0;
2634 }
2635
2636 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2637 {
2638         if (!buf || size < GIT_SHA1_HEXSZ + 6)
2639                 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2640         if (memcmp("tree ", buf, 5)
2641                 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2642                 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2643         oidcpy(&b->branch_tree.versions[0].oid,
2644                &b->branch_tree.versions[1].oid);
2645 }
2646
2647 static void parse_from_existing(struct branch *b)
2648 {
2649         if (is_null_oid(&b->oid)) {
2650                 oidclr(&b->branch_tree.versions[0].oid);
2651                 oidclr(&b->branch_tree.versions[1].oid);
2652         } else {
2653                 unsigned long size;
2654                 char *buf;
2655
2656                 buf = read_object_with_reference(b->oid.hash,
2657                                                  commit_type, &size,
2658                                                  b->oid.hash);
2659                 parse_from_commit(b, buf, size);
2660                 free(buf);
2661         }
2662 }
2663
2664 static int parse_from(struct branch *b)
2665 {
2666         const char *from;
2667         struct branch *s;
2668         struct object_id oid;
2669
2670         if (!skip_prefix(command_buf.buf, "from ", &from))
2671                 return 0;
2672
2673         oidcpy(&oid, &b->branch_tree.versions[1].oid);
2674
2675         s = lookup_branch(from);
2676         if (b == s)
2677                 die("Can't create a branch from itself: %s", b->name);
2678         else if (s) {
2679                 struct object_id *t = &s->branch_tree.versions[1].oid;
2680                 oidcpy(&b->oid, &s->oid);
2681                 oidcpy(&b->branch_tree.versions[0].oid, t);
2682                 oidcpy(&b->branch_tree.versions[1].oid, t);
2683         } else if (*from == ':') {
2684                 uintmax_t idnum = parse_mark_ref_eol(from);
2685                 struct object_entry *oe = find_mark(idnum);
2686                 if (oe->type != OBJ_COMMIT)
2687                         die("Mark :%" PRIuMAX " not a commit", idnum);
2688                 if (oidcmp(&b->oid, &oe->idx.oid)) {
2689                         oidcpy(&b->oid, &oe->idx.oid);
2690                         if (oe->pack_id != MAX_PACK_ID) {
2691                                 unsigned long size;
2692                                 char *buf = gfi_unpack_entry(oe, &size);
2693                                 parse_from_commit(b, buf, size);
2694                                 free(buf);
2695                         } else
2696                                 parse_from_existing(b);
2697                 }
2698         } else if (!get_oid(from, &b->oid)) {
2699                 parse_from_existing(b);
2700                 if (is_null_oid(&b->oid))
2701                         b->delete = 1;
2702         }
2703         else
2704                 die("Invalid ref name or SHA1 expression: %s", from);
2705
2706         if (b->branch_tree.tree && oidcmp(&oid, &b->branch_tree.versions[1].oid)) {
2707                 release_tree_content_recursive(b->branch_tree.tree);
2708                 b->branch_tree.tree = NULL;
2709         }
2710
2711         read_next_command();
2712         return 1;
2713 }
2714
2715 static struct hash_list *parse_merge(unsigned int *count)
2716 {
2717         struct hash_list *list = NULL, **tail = &list, *n;
2718         const char *from;
2719         struct branch *s;
2720
2721         *count = 0;
2722         while (skip_prefix(command_buf.buf, "merge ", &from)) {
2723                 n = xmalloc(sizeof(*n));
2724                 s = lookup_branch(from);
2725                 if (s)
2726                         oidcpy(&n->oid, &s->oid);
2727                 else if (*from == ':') {
2728                         uintmax_t idnum = parse_mark_ref_eol(from);
2729                         struct object_entry *oe = find_mark(idnum);
2730                         if (oe->type != OBJ_COMMIT)
2731                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2732                         oidcpy(&n->oid, &oe->idx.oid);
2733                 } else if (!get_oid(from, &n->oid)) {
2734                         unsigned long size;
2735                         char *buf = read_object_with_reference(n->oid.hash,
2736                                 commit_type, &size, n->oid.hash);
2737                         if (!buf || size < 46)
2738                                 die("Not a valid commit: %s", from);
2739                         free(buf);
2740                 } else
2741                         die("Invalid ref name or SHA1 expression: %s", from);
2742
2743                 n->next = NULL;
2744                 *tail = n;
2745                 tail = &n->next;
2746
2747                 (*count)++;
2748                 read_next_command();
2749         }
2750         return list;
2751 }
2752
2753 static void parse_new_commit(const char *arg)
2754 {
2755         static struct strbuf msg = STRBUF_INIT;
2756         struct branch *b;
2757         char *author = NULL;
2758         char *committer = NULL;
2759         struct hash_list *merge_list = NULL;
2760         unsigned int merge_count;
2761         unsigned char prev_fanout, new_fanout;
2762         const char *v;
2763
2764         b = lookup_branch(arg);
2765         if (!b)
2766                 b = new_branch(arg);
2767
2768         read_next_command();
2769         parse_mark();
2770         if (skip_prefix(command_buf.buf, "author ", &v)) {
2771                 author = parse_ident(v);
2772                 read_next_command();
2773         }
2774         if (skip_prefix(command_buf.buf, "committer ", &v)) {
2775                 committer = parse_ident(v);
2776                 read_next_command();
2777         }
2778         if (!committer)
2779                 die("Expected committer but didn't get one");
2780         parse_data(&msg, 0, NULL);
2781         read_next_command();
2782         parse_from(b);
2783         merge_list = parse_merge(&merge_count);
2784
2785         /* ensure the branch is active/loaded */
2786         if (!b->branch_tree.tree || !max_active_branches) {
2787                 unload_one_branch();
2788                 load_branch(b);
2789         }
2790
2791         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2792
2793         /* file_change* */
2794         while (command_buf.len > 0) {
2795                 if (skip_prefix(command_buf.buf, "M ", &v))
2796                         file_change_m(v, b);
2797                 else if (skip_prefix(command_buf.buf, "D ", &v))
2798                         file_change_d(v, b);
2799                 else if (skip_prefix(command_buf.buf, "R ", &v))
2800                         file_change_cr(v, b, 1);
2801                 else if (skip_prefix(command_buf.buf, "C ", &v))
2802                         file_change_cr(v, b, 0);
2803                 else if (skip_prefix(command_buf.buf, "N ", &v))
2804                         note_change_n(v, b, &prev_fanout);
2805                 else if (!strcmp("deleteall", command_buf.buf))
2806                         file_change_deleteall(b);
2807                 else if (skip_prefix(command_buf.buf, "ls ", &v))
2808                         parse_ls(v, b);
2809                 else {
2810                         unread_command_buf = 1;
2811                         break;
2812                 }
2813                 if (read_next_command() == EOF)
2814                         break;
2815         }
2816
2817         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2818         if (new_fanout != prev_fanout)
2819                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2820
2821         /* build the tree and the commit */
2822         store_tree(&b->branch_tree);
2823         oidcpy(&b->branch_tree.versions[0].oid,
2824                &b->branch_tree.versions[1].oid);
2825
2826         strbuf_reset(&new_data);
2827         strbuf_addf(&new_data, "tree %s\n",
2828                 oid_to_hex(&b->branch_tree.versions[1].oid));
2829         if (!is_null_oid(&b->oid))
2830                 strbuf_addf(&new_data, "parent %s\n",
2831                             oid_to_hex(&b->oid));
2832         while (merge_list) {
2833                 struct hash_list *next = merge_list->next;
2834                 strbuf_addf(&new_data, "parent %s\n",
2835                             oid_to_hex(&merge_list->oid));
2836                 free(merge_list);
2837                 merge_list = next;
2838         }
2839         strbuf_addf(&new_data,
2840                 "author %s\n"
2841                 "committer %s\n"
2842                 "\n",
2843                 author ? author : committer, committer);
2844         strbuf_addbuf(&new_data, &msg);
2845         free(author);
2846         free(committer);
2847
2848         if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2849                 b->pack_id = pack_id;
2850         b->last_commit = object_count_by_type[OBJ_COMMIT];
2851 }
2852
2853 static void parse_new_tag(const char *arg)
2854 {
2855         static struct strbuf msg = STRBUF_INIT;
2856         const char *from;
2857         char *tagger;
2858         struct branch *s;
2859         struct tag *t;
2860         uintmax_t from_mark = 0;
2861         struct object_id oid;
2862         enum object_type type;
2863         const char *v;
2864
2865         t = pool_alloc(sizeof(struct tag));
2866         memset(t, 0, sizeof(struct tag));
2867         t->name = pool_strdup(arg);
2868         if (last_tag)
2869                 last_tag->next_tag = t;
2870         else
2871                 first_tag = t;
2872         last_tag = t;
2873         read_next_command();
2874
2875         /* from ... */
2876         if (!skip_prefix(command_buf.buf, "from ", &from))
2877                 die("Expected from command, got %s", command_buf.buf);
2878         s = lookup_branch(from);
2879         if (s) {
2880                 if (is_null_oid(&s->oid))
2881                         die("Can't tag an empty branch.");
2882                 oidcpy(&oid, &s->oid);
2883                 type = OBJ_COMMIT;
2884         } else if (*from == ':') {
2885                 struct object_entry *oe;
2886                 from_mark = parse_mark_ref_eol(from);
2887                 oe = find_mark(from_mark);
2888                 type = oe->type;
2889                 oidcpy(&oid, &oe->idx.oid);
2890         } else if (!get_oid(from, &oid)) {
2891                 struct object_entry *oe = find_object(&oid);
2892                 if (!oe) {
2893                         type = sha1_object_info(oid.hash, NULL);
2894                         if (type < 0)
2895                                 die("Not a valid object: %s", from);
2896                 } else
2897                         type = oe->type;
2898         } else
2899                 die("Invalid ref name or SHA1 expression: %s", from);
2900         read_next_command();
2901
2902         /* tagger ... */
2903         if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2904                 tagger = parse_ident(v);
2905                 read_next_command();
2906         } else
2907                 tagger = NULL;
2908
2909         /* tag payload/message */
2910         parse_data(&msg, 0, NULL);
2911
2912         /* build the tag object */
2913         strbuf_reset(&new_data);
2914
2915         strbuf_addf(&new_data,
2916                     "object %s\n"
2917                     "type %s\n"
2918                     "tag %s\n",
2919                     oid_to_hex(&oid), typename(type), t->name);
2920         if (tagger)
2921                 strbuf_addf(&new_data,
2922                             "tagger %s\n", tagger);
2923         strbuf_addch(&new_data, '\n');
2924         strbuf_addbuf(&new_data, &msg);
2925         free(tagger);
2926
2927         if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, 0))
2928                 t->pack_id = MAX_PACK_ID;
2929         else
2930                 t->pack_id = pack_id;
2931 }
2932
2933 static void parse_reset_branch(const char *arg)
2934 {
2935         struct branch *b;
2936
2937         b = lookup_branch(arg);
2938         if (b) {
2939                 oidclr(&b->oid);
2940                 oidclr(&b->branch_tree.versions[0].oid);
2941                 oidclr(&b->branch_tree.versions[1].oid);
2942                 if (b->branch_tree.tree) {
2943                         release_tree_content_recursive(b->branch_tree.tree);
2944                         b->branch_tree.tree = NULL;
2945                 }
2946         }
2947         else
2948                 b = new_branch(arg);
2949         read_next_command();
2950         parse_from(b);
2951         if (command_buf.len > 0)
2952                 unread_command_buf = 1;
2953 }
2954
2955 static void cat_blob_write(const char *buf, unsigned long size)
2956 {
2957         if (write_in_full(cat_blob_fd, buf, size) < 0)
2958                 die_errno("Write to frontend failed");
2959 }
2960
2961 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2962 {
2963         struct strbuf line = STRBUF_INIT;
2964         unsigned long size;
2965         enum object_type type = 0;
2966         char *buf;
2967
2968         if (!oe || oe->pack_id == MAX_PACK_ID) {
2969                 buf = read_sha1_file(oid->hash, &type, &size);
2970         } else {
2971                 type = oe->type;
2972                 buf = gfi_unpack_entry(oe, &size);
2973         }
2974
2975         /*
2976          * Output based on batch_one_object() from cat-file.c.
2977          */
2978         if (type <= 0) {
2979                 strbuf_reset(&line);
2980                 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2981                 cat_blob_write(line.buf, line.len);
2982                 strbuf_release(&line);
2983                 free(buf);
2984                 return;
2985         }
2986         if (!buf)
2987                 die("Can't read object %s", oid_to_hex(oid));
2988         if (type != OBJ_BLOB)
2989                 die("Object %s is a %s but a blob was expected.",
2990                     oid_to_hex(oid), typename(type));
2991         strbuf_reset(&line);
2992         strbuf_addf(&line, "%s %s %lu\n", oid_to_hex(oid),
2993                                                 typename(type), size);
2994         cat_blob_write(line.buf, line.len);
2995         strbuf_release(&line);
2996         cat_blob_write(buf, size);
2997         cat_blob_write("\n", 1);
2998         if (oe && oe->pack_id == pack_id) {
2999                 last_blob.offset = oe->idx.offset;
3000                 strbuf_attach(&last_blob.data, buf, size, size);
3001                 last_blob.depth = oe->depth;
3002         } else
3003                 free(buf);
3004 }
3005
3006 static void parse_get_mark(const char *p)
3007 {
3008         struct object_entry *oe = oe;
3009         char output[GIT_MAX_HEXSZ + 2];
3010
3011         /* get-mark SP <object> LF */
3012         if (*p != ':')
3013                 die("Not a mark: %s", p);
3014
3015         oe = find_mark(parse_mark_ref_eol(p));
3016         if (!oe)
3017                 die("Unknown mark: %s", command_buf.buf);
3018
3019         xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
3020         cat_blob_write(output, GIT_SHA1_HEXSZ + 1);
3021 }
3022
3023 static void parse_cat_blob(const char *p)
3024 {
3025         struct object_entry *oe = oe;
3026         struct object_id oid;
3027
3028         /* cat-blob SP <object> LF */
3029         if (*p == ':') {
3030                 oe = find_mark(parse_mark_ref_eol(p));
3031                 if (!oe)
3032                         die("Unknown mark: %s", command_buf.buf);
3033                 oidcpy(&oid, &oe->idx.oid);
3034         } else {
3035                 if (parse_oid_hex(p, &oid, &p))
3036                         die("Invalid dataref: %s", command_buf.buf);
3037                 if (*p)
3038                         die("Garbage after SHA1: %s", command_buf.buf);
3039                 oe = find_object(&oid);
3040         }
3041
3042         cat_blob(oe, &oid);
3043 }
3044
3045 static struct object_entry *dereference(struct object_entry *oe,
3046                                         struct object_id *oid)
3047 {
3048         unsigned long size;
3049         char *buf = NULL;
3050         if (!oe) {
3051                 enum object_type type = sha1_object_info(oid->hash, NULL);
3052                 if (type < 0)
3053                         die("object not found: %s", oid_to_hex(oid));
3054                 /* cache it! */
3055                 oe = insert_object(oid);
3056                 oe->type = type;
3057                 oe->pack_id = MAX_PACK_ID;
3058                 oe->idx.offset = 1;
3059         }
3060         switch (oe->type) {
3061         case OBJ_TREE:  /* easy case. */
3062                 return oe;
3063         case OBJ_COMMIT:
3064         case OBJ_TAG:
3065                 break;
3066         default:
3067                 die("Not a tree-ish: %s", command_buf.buf);
3068         }
3069
3070         if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
3071                 buf = gfi_unpack_entry(oe, &size);
3072         } else {
3073                 enum object_type unused;
3074                 buf = read_sha1_file(oid->hash, &unused, &size);
3075         }
3076         if (!buf)
3077                 die("Can't load object %s", oid_to_hex(oid));
3078
3079         /* Peel one layer. */
3080         switch (oe->type) {
3081         case OBJ_TAG:
3082                 if (size < GIT_SHA1_HEXSZ + strlen("object ") ||
3083                     get_oid_hex(buf + strlen("object "), oid))
3084                         die("Invalid SHA1 in tag: %s", command_buf.buf);
3085                 break;
3086         case OBJ_COMMIT:
3087                 if (size < GIT_SHA1_HEXSZ + strlen("tree ") ||
3088                     get_oid_hex(buf + strlen("tree "), oid))
3089                         die("Invalid SHA1 in commit: %s", command_buf.buf);
3090         }
3091
3092         free(buf);
3093         return find_object(oid);
3094 }
3095
3096 static struct object_entry *parse_treeish_dataref(const char **p)
3097 {
3098         struct object_id oid;
3099         struct object_entry *e;
3100
3101         if (**p == ':') {       /* <mark> */
3102                 e = find_mark(parse_mark_ref_space(p));
3103                 if (!e)
3104                         die("Unknown mark: %s", command_buf.buf);
3105                 oidcpy(&oid, &e->idx.oid);
3106         } else {        /* <sha1> */
3107                 if (parse_oid_hex(*p, &oid, p))
3108                         die("Invalid dataref: %s", command_buf.buf);
3109                 e = find_object(&oid);
3110                 if (*(*p)++ != ' ')
3111                         die("Missing space after tree-ish: %s", command_buf.buf);
3112         }
3113
3114         while (!e || e->type != OBJ_TREE)
3115                 e = dereference(e, &oid);
3116         return e;
3117 }
3118
3119 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3120 {
3121         static struct strbuf line = STRBUF_INIT;
3122
3123         /* See show_tree(). */
3124         const char *type =
3125                 S_ISGITLINK(mode) ? commit_type :
3126                 S_ISDIR(mode) ? tree_type :
3127                 blob_type;
3128
3129         if (!mode) {
3130                 /* missing SP path LF */
3131                 strbuf_reset(&line);
3132                 strbuf_addstr(&line, "missing ");
3133                 quote_c_style(path, &line, NULL, 0);
3134                 strbuf_addch(&line, '\n');
3135         } else {
3136                 /* mode SP type SP object_name TAB path LF */
3137                 strbuf_reset(&line);
3138                 strbuf_addf(&line, "%06o %s %s\t",
3139                                 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3140                 quote_c_style(path, &line, NULL, 0);
3141                 strbuf_addch(&line, '\n');
3142         }
3143         cat_blob_write(line.buf, line.len);
3144 }
3145
3146 static void parse_ls(const char *p, struct branch *b)
3147 {
3148         struct tree_entry *root = NULL;
3149         struct tree_entry leaf = {NULL};
3150
3151         /* ls SP (<tree-ish> SP)? <path> */
3152         if (*p == '"') {
3153                 if (!b)
3154                         die("Not in a commit: %s", command_buf.buf);
3155                 root = &b->branch_tree;
3156         } else {
3157                 struct object_entry *e = parse_treeish_dataref(&p);
3158                 root = new_tree_entry();
3159                 oidcpy(&root->versions[1].oid, &e->idx.oid);
3160                 if (!is_null_oid(&root->versions[1].oid))
3161                         root->versions[1].mode = S_IFDIR;
3162                 load_tree(root);
3163         }
3164         if (*p == '"') {
3165                 static struct strbuf uq = STRBUF_INIT;
3166                 const char *endp;
3167                 strbuf_reset(&uq);
3168                 if (unquote_c_style(&uq, p, &endp))
3169                         die("Invalid path: %s", command_buf.buf);
3170                 if (*endp)
3171                         die("Garbage after path in: %s", command_buf.buf);
3172                 p = uq.buf;
3173         }
3174         tree_content_get(root, p, &leaf, 1);
3175         /*
3176          * A directory in preparation would have a sha1 of zero
3177          * until it is saved.  Save, for simplicity.
3178          */
3179         if (S_ISDIR(leaf.versions[1].mode))
3180                 store_tree(&leaf);
3181
3182         print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3183         if (leaf.tree)
3184                 release_tree_content_recursive(leaf.tree);
3185         if (!b || root != &b->branch_tree)
3186                 release_tree_entry(root);
3187 }
3188
3189 static void checkpoint(void)
3190 {
3191         checkpoint_requested = 0;
3192         if (object_count) {
3193                 cycle_packfile();
3194         }
3195         dump_branches();
3196         dump_tags();
3197         dump_marks();
3198 }
3199
3200 static void parse_checkpoint(void)
3201 {
3202         checkpoint_requested = 1;
3203         skip_optional_lf();
3204 }
3205
3206 static void parse_progress(void)
3207 {
3208         fwrite(command_buf.buf, 1, command_buf.len, stdout);
3209         fputc('\n', stdout);
3210         fflush(stdout);
3211         skip_optional_lf();
3212 }
3213
3214 static char* make_fast_import_path(const char *path)
3215 {
3216         if (!relative_marks_paths || is_absolute_path(path))
3217                 return xstrdup(path);
3218         return git_pathdup("info/fast-import/%s", path);
3219 }
3220
3221 static void option_import_marks(const char *marks,
3222                                         int from_stream, int ignore_missing)
3223 {
3224         if (import_marks_file) {
3225                 if (from_stream)
3226                         die("Only one import-marks command allowed per stream");
3227
3228                 /* read previous mark file */
3229                 if(!import_marks_file_from_stream)
3230                         read_marks();
3231         }
3232
3233         import_marks_file = make_fast_import_path(marks);
3234         safe_create_leading_directories_const(import_marks_file);
3235         import_marks_file_from_stream = from_stream;
3236         import_marks_file_ignore_missing = ignore_missing;
3237 }
3238
3239 static void option_date_format(const char *fmt)
3240 {
3241         if (!strcmp(fmt, "raw"))
3242                 whenspec = WHENSPEC_RAW;
3243         else if (!strcmp(fmt, "rfc2822"))
3244                 whenspec = WHENSPEC_RFC2822;
3245         else if (!strcmp(fmt, "now"))
3246                 whenspec = WHENSPEC_NOW;
3247         else
3248                 die("unknown --date-format argument %s", fmt);
3249 }
3250
3251 static unsigned long ulong_arg(const char *option, const char *arg)
3252 {
3253         char *endptr;
3254         unsigned long rv = strtoul(arg, &endptr, 0);
3255         if (strchr(arg, '-') || endptr == arg || *endptr)
3256                 die("%s: argument must be a non-negative integer", option);
3257         return rv;
3258 }
3259
3260 static void option_depth(const char *depth)
3261 {
3262         max_depth = ulong_arg("--depth", depth);
3263         if (max_depth > MAX_DEPTH)
3264                 die("--depth cannot exceed %u", MAX_DEPTH);
3265 }
3266
3267 static void option_active_branches(const char *branches)
3268 {
3269         max_active_branches = ulong_arg("--active-branches", branches);
3270 }
3271
3272 static void option_export_marks(const char *marks)
3273 {
3274         export_marks_file = make_fast_import_path(marks);
3275         safe_create_leading_directories_const(export_marks_file);
3276 }
3277
3278 static void option_cat_blob_fd(const char *fd)
3279 {
3280         unsigned long n = ulong_arg("--cat-blob-fd", fd);
3281         if (n > (unsigned long) INT_MAX)
3282                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3283         cat_blob_fd = (int) n;
3284 }
3285
3286 static void option_export_pack_edges(const char *edges)
3287 {
3288         if (pack_edges)
3289                 fclose(pack_edges);
3290         pack_edges = xfopen(edges, "a");
3291 }
3292
3293 static int parse_one_option(const char *option)
3294 {
3295         if (skip_prefix(option, "max-pack-size=", &option)) {
3296                 unsigned long v;
3297                 if (!git_parse_ulong(option, &v))
3298                         return 0;
3299                 if (v < 8192) {
3300                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3301                         v *= 1024 * 1024;
3302                 } else if (v < 1024 * 1024) {
3303                         warning("minimum max-pack-size is 1 MiB");
3304                         v = 1024 * 1024;
3305                 }
3306                 max_packsize = v;
3307         } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3308                 unsigned long v;
3309                 if (!git_parse_ulong(option, &v))
3310                         return 0;
3311                 big_file_threshold = v;
3312         } else if (skip_prefix(option, "depth=", &option)) {
3313                 option_depth(option);
3314         } else if (skip_prefix(option, "active-branches=", &option)) {
3315                 option_active_branches(option);
3316         } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3317                 option_export_pack_edges(option);
3318         } else if (starts_with(option, "quiet")) {
3319                 show_stats = 0;
3320         } else if (starts_with(option, "stats")) {
3321                 show_stats = 1;
3322         } else {
3323                 return 0;
3324         }
3325
3326         return 1;
3327 }
3328
3329 static int parse_one_feature(const char *feature, int from_stream)
3330 {
3331         const char *arg;
3332
3333         if (skip_prefix(feature, "date-format=", &arg)) {
3334                 option_date_format(arg);
3335         } else if (skip_prefix(feature, "import-marks=", &arg)) {
3336                 option_import_marks(arg, from_stream, 0);
3337         } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3338                 option_import_marks(arg, from_stream, 1);
3339         } else if (skip_prefix(feature, "export-marks=", &arg)) {
3340                 option_export_marks(arg);
3341         } else if (!strcmp(feature, "get-mark")) {
3342                 ; /* Don't die - this feature is supported */
3343         } else if (!strcmp(feature, "cat-blob")) {
3344                 ; /* Don't die - this feature is supported */
3345         } else if (!strcmp(feature, "relative-marks")) {
3346                 relative_marks_paths = 1;
3347         } else if (!strcmp(feature, "no-relative-marks")) {
3348                 relative_marks_paths = 0;
3349         } else if (!strcmp(feature, "done")) {
3350                 require_explicit_termination = 1;
3351         } else if (!strcmp(feature, "force")) {
3352                 force_update = 1;
3353         } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3354                 ; /* do nothing; we have the feature */
3355         } else {
3356                 return 0;
3357         }
3358
3359         return 1;
3360 }
3361
3362 static void parse_feature(const char *feature)
3363 {
3364         if (seen_data_command)
3365                 die("Got feature command '%s' after data command", feature);
3366
3367         if (parse_one_feature(feature, 1))
3368                 return;
3369
3370         die("This version of fast-import does not support feature %s.", feature);
3371 }
3372
3373 static void parse_option(const char *option)
3374 {
3375         if (seen_data_command)
3376                 die("Got option command '%s' after data command", option);
3377
3378         if (parse_one_option(option))
3379                 return;
3380
3381         die("This version of fast-import does not support option: %s", option);
3382 }
3383
3384 static void git_pack_config(void)
3385 {
3386         int indexversion_value;
3387         int limit;
3388         unsigned long packsizelimit_value;
3389
3390         if (!git_config_get_ulong("pack.depth", &max_depth)) {
3391                 if (max_depth > MAX_DEPTH)
3392                         max_depth = MAX_DEPTH;
3393         }
3394         if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3395                 pack_idx_opts.version = indexversion_value;
3396                 if (pack_idx_opts.version > 2)
3397                         git_die_config("pack.indexversion",
3398                                         "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3399         }
3400         if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3401                 max_packsize = packsizelimit_value;
3402
3403         if (!git_config_get_int("fastimport.unpacklimit", &limit))
3404                 unpack_limit = limit;
3405         else if (!git_config_get_int("transfer.unpacklimit", &limit))
3406                 unpack_limit = limit;
3407
3408         git_config(git_default_config, NULL);
3409 }
3410
3411 static const char fast_import_usage[] =
3412 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3413
3414 static void parse_argv(void)
3415 {
3416         unsigned int i;
3417
3418         for (i = 1; i < global_argc; i++) {
3419                 const char *a = global_argv[i];
3420
3421                 if (*a != '-' || !strcmp(a, "--"))
3422                         break;
3423
3424                 if (!skip_prefix(a, "--", &a))
3425                         die("unknown option %s", a);
3426
3427                 if (parse_one_option(a))
3428                         continue;
3429
3430                 if (parse_one_feature(a, 0))
3431                         continue;
3432
3433                 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3434                         option_cat_blob_fd(a);
3435                         continue;
3436                 }
3437
3438                 die("unknown option --%s", a);
3439         }
3440         if (i != global_argc)
3441                 usage(fast_import_usage);
3442
3443         seen_data_command = 1;
3444         if (import_marks_file)
3445                 read_marks();
3446 }
3447
3448 int cmd_main(int argc, const char **argv)
3449 {
3450         unsigned int i;
3451
3452         if (argc == 2 && !strcmp(argv[1], "-h"))
3453                 usage(fast_import_usage);
3454
3455         setup_git_directory();
3456         reset_pack_idx_option(&pack_idx_opts);
3457         git_pack_config();
3458
3459         alloc_objects(object_entry_alloc);
3460         strbuf_init(&command_buf, 0);
3461         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3462         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3463         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3464         marks = pool_calloc(1, sizeof(struct mark_set));
3465
3466         global_argc = argc;
3467         global_argv = argv;
3468
3469         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3470         for (i = 0; i < (cmd_save - 1); i++)
3471                 rc_free[i].next = &rc_free[i + 1];
3472         rc_free[cmd_save - 1].next = NULL;
3473
3474         prepare_packed_git();
3475         start_packfile();
3476         set_die_routine(die_nicely);
3477         set_checkpoint_signal();
3478         while (read_next_command() != EOF) {
3479                 const char *v;
3480                 if (!strcmp("blob", command_buf.buf))
3481                         parse_new_blob();
3482                 else if (skip_prefix(command_buf.buf, "ls ", &v))
3483                         parse_ls(v, NULL);
3484                 else if (skip_prefix(command_buf.buf, "commit ", &v))
3485                         parse_new_commit(v);
3486                 else if (skip_prefix(command_buf.buf, "tag ", &v))
3487                         parse_new_tag(v);
3488                 else if (skip_prefix(command_buf.buf, "reset ", &v))
3489                         parse_reset_branch(v);
3490                 else if (!strcmp("checkpoint", command_buf.buf))
3491                         parse_checkpoint();
3492                 else if (!strcmp("done", command_buf.buf))
3493                         break;
3494                 else if (starts_with(command_buf.buf, "progress "))
3495                         parse_progress();
3496                 else if (skip_prefix(command_buf.buf, "feature ", &v))
3497                         parse_feature(v);
3498                 else if (skip_prefix(command_buf.buf, "option git ", &v))
3499                         parse_option(v);
3500                 else if (starts_with(command_buf.buf, "option "))
3501                         /* ignore non-git options*/;
3502                 else
3503                         die("Unsupported command: %s", command_buf.buf);
3504
3505                 if (checkpoint_requested)
3506                         checkpoint();
3507         }
3508
3509         /* argv hasn't been parsed yet, do so */
3510         if (!seen_data_command)
3511                 parse_argv();
3512
3513         if (require_explicit_termination && feof(stdin))
3514                 die("stream ends early");
3515
3516         end_packfile();
3517
3518         dump_branches();
3519         dump_tags();
3520         unkeep_all_packs();
3521         dump_marks();
3522
3523         if (pack_edges)
3524                 fclose(pack_edges);
3525
3526         if (show_stats) {
3527                 uintmax_t total_count = 0, duplicate_count = 0;
3528                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3529                         total_count += object_count_by_type[i];
3530                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3531                         duplicate_count += duplicate_count_by_type[i];
3532
3533                 fprintf(stderr, "%s statistics:\n", argv[0]);
3534                 fprintf(stderr, "---------------------------------------------------------------------\n");
3535                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3536                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3537                 fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB], delta_count_attempts_by_type[OBJ_BLOB]);
3538                 fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE], delta_count_attempts_by_type[OBJ_TREE]);
3539                 fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT], delta_count_attempts_by_type[OBJ_COMMIT]);
3540                 fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG], delta_count_attempts_by_type[OBJ_TAG]);
3541                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3542                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3543                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3544                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3545                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3546                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3547                 fprintf(stderr, "---------------------------------------------------------------------\n");
3548                 pack_report();
3549                 fprintf(stderr, "---------------------------------------------------------------------\n");
3550                 fprintf(stderr, "\n");
3551         }
3552
3553         return failure ? 1 : 0;
3554 }