web--browse: look at the BROWSER env var
[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 committish lf)?
26     ('merge' sp committish lf)*
27     file_change*
28     lf?;
29   commit_msg ::= data;
30
31   file_change ::= file_clr
32     | file_del
33     | file_rnm
34     | file_cpy
35     | file_obm
36     | file_inm;
37   file_clr ::= 'deleteall' lf;
38   file_del ::= 'D' sp path_str lf;
39   file_rnm ::= 'R' sp path_str sp path_str lf;
40   file_cpy ::= 'C' sp path_str sp path_str lf;
41   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
42   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
43     data;
44   note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
45   note_inm ::= 'N' sp 'inline' sp committish lf
46     data;
47
48   new_tag ::= 'tag' sp tag_str lf
49     'from' sp committish lf
50     ('tagger' (sp name)? sp '<' email '>' sp when lf)?
51     tag_msg;
52   tag_msg ::= data;
53
54   reset_branch ::= 'reset' sp ref_str lf
55     ('from' sp committish lf)?
56     lf?;
57
58   checkpoint ::= 'checkpoint' lf
59     lf?;
60
61   progress ::= 'progress' sp not_lf* lf
62     lf?;
63
64      # note: the first idnum in a stream should be 1 and subsequent
65      # idnums should not have gaps between values as this will cause
66      # the stream parser to reserve space for the gapped values.  An
67      # idnum can be updated in the future to a new object by issuing
68      # a new mark directive with the old idnum.
69      #
70   mark ::= 'mark' sp idnum lf;
71   data ::= (delimited_data | exact_data)
72     lf?;
73
74     # note: delim may be any string but must not contain lf.
75     # data_line may contain any data but must not be exactly
76     # delim.
77   delimited_data ::= 'data' sp '<<' delim lf
78     (data_line lf)*
79     delim lf;
80
81      # note: declen indicates the length of binary_data in bytes.
82      # declen does not include the lf preceding the binary data.
83      #
84   exact_data ::= 'data' sp declen lf
85     binary_data;
86
87      # note: quoted strings are C-style quoting supporting \c for
88      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
89      # is the signed byte value in octal.  Note that the only
90      # characters which must actually be escaped to protect the
91      # stream formatting is: \, " and LF.  Otherwise these values
92      # are UTF8.
93      #
94   committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
95   ref_str     ::= ref;
96   sha1exp_str ::= sha1exp;
97   tag_str     ::= tag;
98   path_str    ::= path    | '"' quoted(path)    '"' ;
99   mode        ::= '100644' | '644'
100                 | '100755' | '755'
101                 | '120000'
102                 ;
103
104   declen ::= # unsigned 32 bit value, ascii base10 notation;
105   bigint ::= # unsigned integer value, ascii base10 notation;
106   binary_data ::= # file content, not interpreted;
107
108   when         ::= raw_when | rfc2822_when;
109   raw_when     ::= ts sp tz;
110   rfc2822_when ::= # Valid RFC 2822 date and time;
111
112   sp ::= # ASCII space character;
113   lf ::= # ASCII newline (LF) character;
114
115      # note: a colon (':') must precede the numerical value assigned to
116      # an idnum.  This is to distinguish it from a ref or tag name as
117      # GIT does not permit ':' in ref or tag strings.
118      #
119   idnum   ::= ':' bigint;
120   path    ::= # GIT style file path, e.g. "a/b/c";
121   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
122   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
123   sha1exp ::= # Any valid GIT SHA1 expression;
124   hexsha1 ::= # SHA1 in hexadecimal format;
125
126      # note: name and email are UTF8 strings, however name must not
127      # contain '<' or lf and email must not contain any of the
128      # following: '<', '>', lf.
129      #
130   name  ::= # valid GIT author/committer name;
131   email ::= # valid GIT author/committer email;
132   ts    ::= # time since the epoch in seconds, ascii base10 notation;
133   tz    ::= # GIT style timezone;
134
135      # note: comments and cat requests may appear anywhere
136      # in the input, except within a data command.  Any form
137      # of the data command always escapes the related input
138      # from comment processing.
139      #
140      # In case it is not clear, the '#' that starts the comment
141      # must be the first character on that line (an lf
142      # preceded it).
143      #
144   cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
145
146   comment ::= '#' not_lf* lf;
147   not_lf  ::= # Any byte that is not ASCII newline (LF);
148 */
149
150 #include "builtin.h"
151 #include "cache.h"
152 #include "object.h"
153 #include "blob.h"
154 #include "tree.h"
155 #include "commit.h"
156 #include "delta.h"
157 #include "pack.h"
158 #include "refs.h"
159 #include "csum-file.h"
160 #include "quote.h"
161 #include "exec_cmd.h"
162 #include "dir.h"
163
164 #define PACK_ID_BITS 16
165 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
166 #define DEPTH_BITS 13
167 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
168
169 struct object_entry
170 {
171         struct pack_idx_entry idx;
172         struct object_entry *next;
173         uint32_t type : TYPE_BITS,
174                 pack_id : PACK_ID_BITS,
175                 depth : DEPTH_BITS;
176 };
177
178 struct object_entry_pool
179 {
180         struct object_entry_pool *next_pool;
181         struct object_entry *next_free;
182         struct object_entry *end;
183         struct object_entry entries[FLEX_ARRAY]; /* more */
184 };
185
186 struct mark_set
187 {
188         union {
189                 struct object_entry *marked[1024];
190                 struct mark_set *sets[1024];
191         } data;
192         unsigned int shift;
193 };
194
195 struct last_object
196 {
197         struct strbuf data;
198         off_t offset;
199         unsigned int depth;
200         unsigned no_swap : 1;
201 };
202
203 struct mem_pool
204 {
205         struct mem_pool *next_pool;
206         char *next_free;
207         char *end;
208         uintmax_t space[FLEX_ARRAY]; /* more */
209 };
210
211 struct atom_str
212 {
213         struct atom_str *next_atom;
214         unsigned short str_len;
215         char str_dat[FLEX_ARRAY]; /* more */
216 };
217
218 struct tree_content;
219 struct tree_entry
220 {
221         struct tree_content *tree;
222         struct atom_str *name;
223         struct tree_entry_ms
224         {
225                 uint16_t mode;
226                 unsigned char sha1[20];
227         } versions[2];
228 };
229
230 struct tree_content
231 {
232         unsigned int entry_capacity; /* must match avail_tree_content */
233         unsigned int entry_count;
234         unsigned int delta_depth;
235         struct tree_entry *entries[FLEX_ARRAY]; /* more */
236 };
237
238 struct avail_tree_content
239 {
240         unsigned int entry_capacity; /* must match tree_content */
241         struct avail_tree_content *next_avail;
242 };
243
244 struct branch
245 {
246         struct branch *table_next_branch;
247         struct branch *active_next_branch;
248         const char *name;
249         struct tree_entry branch_tree;
250         uintmax_t last_commit;
251         uintmax_t num_notes;
252         unsigned active : 1;
253         unsigned pack_id : PACK_ID_BITS;
254         unsigned char sha1[20];
255 };
256
257 struct tag
258 {
259         struct tag *next_tag;
260         const char *name;
261         unsigned int pack_id;
262         unsigned char sha1[20];
263 };
264
265 struct hash_list
266 {
267         struct hash_list *next;
268         unsigned char sha1[20];
269 };
270
271 typedef enum {
272         WHENSPEC_RAW = 1,
273         WHENSPEC_RFC2822,
274         WHENSPEC_NOW
275 } whenspec_type;
276
277 struct recent_command
278 {
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 = 10;
286 static off_t max_packsize;
287 static uintmax_t big_file_threshold = 512 * 1024 * 1024;
288 static int force_update;
289 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
290 static int pack_compression_seen;
291
292 /* Stats and misc. counters */
293 static uintmax_t alloc_count;
294 static uintmax_t marks_set_count;
295 static uintmax_t object_count_by_type[1 << TYPE_BITS];
296 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
297 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
298 static unsigned long object_count;
299 static unsigned long branch_count;
300 static unsigned long branch_load_count;
301 static int failure;
302 static FILE *pack_edges;
303 static unsigned int show_stats = 1;
304 static int global_argc;
305 static const char **global_argv;
306
307 /* Memory pools */
308 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
309 static size_t total_allocd;
310 static struct mem_pool *mem_pool;
311
312 /* Atom management */
313 static unsigned int atom_table_sz = 4451;
314 static unsigned int atom_cnt;
315 static struct atom_str **atom_table;
316
317 /* The .pack file being generated */
318 static unsigned int pack_id;
319 static struct sha1file *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 relative_marks_paths;
333
334 /* Our last blob */
335 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
336
337 /* Tree management */
338 static unsigned int tree_entry_alloc = 1000;
339 static void *avail_tree_entry;
340 static unsigned int avail_tree_table_sz = 100;
341 static struct avail_tree_content **avail_tree_table;
342 static struct strbuf old_tree = STRBUF_INIT;
343 static struct strbuf new_tree = STRBUF_INIT;
344
345 /* Branch data */
346 static unsigned long max_active_branches = 5;
347 static unsigned long cur_active_branches;
348 static unsigned long branch_table_sz = 1039;
349 static struct branch **branch_table;
350 static struct branch *active_branches;
351
352 /* Tag data */
353 static struct tag *first_tag;
354 static struct tag *last_tag;
355
356 /* Input stream parsing */
357 static whenspec_type whenspec = WHENSPEC_RAW;
358 static struct strbuf command_buf = STRBUF_INIT;
359 static int unread_command_buf;
360 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
361 static struct recent_command *cmd_tail = &cmd_hist;
362 static struct recent_command *rc_free;
363 static unsigned int cmd_save = 100;
364 static uintmax_t next_mark;
365 static struct strbuf new_data = STRBUF_INIT;
366 static int seen_data_command;
367
368 /* Signal handling */
369 static volatile sig_atomic_t checkpoint_requested;
370
371 /* Where to write output of cat-blob commands */
372 static int cat_blob_fd = STDOUT_FILENO;
373
374 static void parse_argv(void);
375 static void parse_cat_blob(void);
376
377 static void write_branch_report(FILE *rpt, struct branch *b)
378 {
379         fprintf(rpt, "%s:\n", b->name);
380
381         fprintf(rpt, "  status      :");
382         if (b->active)
383                 fputs(" active", rpt);
384         if (b->branch_tree.tree)
385                 fputs(" loaded", rpt);
386         if (is_null_sha1(b->branch_tree.versions[1].sha1))
387                 fputs(" dirty", rpt);
388         fputc('\n', rpt);
389
390         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
391         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
392         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
393         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
394
395         fputs("  last pack   : ", rpt);
396         if (b->pack_id < MAX_PACK_ID)
397                 fprintf(rpt, "%u", b->pack_id);
398         fputc('\n', rpt);
399
400         fputc('\n', rpt);
401 }
402
403 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
404
405 static void write_crash_report(const char *err)
406 {
407         char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
408         FILE *rpt = fopen(loc, "w");
409         struct branch *b;
410         unsigned long lu;
411         struct recent_command *rc;
412
413         if (!rpt) {
414                 error("can't write crash report %s: %s", loc, strerror(errno));
415                 return;
416         }
417
418         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
419
420         fprintf(rpt, "fast-import crash report:\n");
421         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
422         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
423         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
424         fputc('\n', rpt);
425
426         fputs("fatal: ", rpt);
427         fputs(err, rpt);
428         fputc('\n', rpt);
429
430         fputc('\n', rpt);
431         fputs("Most Recent Commands Before Crash\n", rpt);
432         fputs("---------------------------------\n", rpt);
433         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
434                 if (rc->next == &cmd_hist)
435                         fputs("* ", rpt);
436                 else
437                         fputs("  ", rpt);
438                 fputs(rc->buf, rpt);
439                 fputc('\n', rpt);
440         }
441
442         fputc('\n', rpt);
443         fputs("Active Branch LRU\n", rpt);
444         fputs("-----------------\n", rpt);
445         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
446                 cur_active_branches,
447                 max_active_branches);
448         fputc('\n', rpt);
449         fputs("  pos  clock name\n", rpt);
450         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
451         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
452                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
453                         ++lu, b->last_commit, b->name);
454
455         fputc('\n', rpt);
456         fputs("Inactive Branches\n", rpt);
457         fputs("-----------------\n", rpt);
458         for (lu = 0; lu < branch_table_sz; lu++) {
459                 for (b = branch_table[lu]; b; b = b->table_next_branch)
460                         write_branch_report(rpt, b);
461         }
462
463         if (first_tag) {
464                 struct tag *tg;
465                 fputc('\n', rpt);
466                 fputs("Annotated Tags\n", rpt);
467                 fputs("--------------\n", rpt);
468                 for (tg = first_tag; tg; tg = tg->next_tag) {
469                         fputs(sha1_to_hex(tg->sha1), rpt);
470                         fputc(' ', rpt);
471                         fputs(tg->name, rpt);
472                         fputc('\n', rpt);
473                 }
474         }
475
476         fputc('\n', rpt);
477         fputs("Marks\n", rpt);
478         fputs("-----\n", rpt);
479         if (export_marks_file)
480                 fprintf(rpt, "  exported to %s\n", export_marks_file);
481         else
482                 dump_marks_helper(rpt, 0, marks);
483
484         fputc('\n', rpt);
485         fputs("-------------------\n", rpt);
486         fputs("END OF CRASH REPORT\n", rpt);
487         fclose(rpt);
488 }
489
490 static void end_packfile(void);
491 static void unkeep_all_packs(void);
492 static void dump_marks(void);
493
494 static NORETURN void die_nicely(const char *err, va_list params)
495 {
496         static int zombie;
497         char message[2 * PATH_MAX];
498
499         vsnprintf(message, sizeof(message), err, params);
500         fputs("fatal: ", stderr);
501         fputs(message, stderr);
502         fputc('\n', stderr);
503
504         if (!zombie) {
505                 zombie = 1;
506                 write_crash_report(message);
507                 end_packfile();
508                 unkeep_all_packs();
509                 dump_marks();
510         }
511         exit(128);
512 }
513
514 #ifndef SIGUSR1 /* Windows, for example */
515
516 static void set_checkpoint_signal(void)
517 {
518 }
519
520 #else
521
522 static void checkpoint_signal(int signo)
523 {
524         checkpoint_requested = 1;
525 }
526
527 static void set_checkpoint_signal(void)
528 {
529         struct sigaction sa;
530
531         memset(&sa, 0, sizeof(sa));
532         sa.sa_handler = checkpoint_signal;
533         sigemptyset(&sa.sa_mask);
534         sa.sa_flags = SA_RESTART;
535         sigaction(SIGUSR1, &sa, NULL);
536 }
537
538 #endif
539
540 static void alloc_objects(unsigned int cnt)
541 {
542         struct object_entry_pool *b;
543
544         b = xmalloc(sizeof(struct object_entry_pool)
545                 + cnt * sizeof(struct object_entry));
546         b->next_pool = blocks;
547         b->next_free = b->entries;
548         b->end = b->entries + cnt;
549         blocks = b;
550         alloc_count += cnt;
551 }
552
553 static struct object_entry *new_object(unsigned char *sha1)
554 {
555         struct object_entry *e;
556
557         if (blocks->next_free == blocks->end)
558                 alloc_objects(object_entry_alloc);
559
560         e = blocks->next_free++;
561         hashcpy(e->idx.sha1, sha1);
562         return e;
563 }
564
565 static struct object_entry *find_object(unsigned char *sha1)
566 {
567         unsigned int h = sha1[0] << 8 | sha1[1];
568         struct object_entry *e;
569         for (e = object_table[h]; e; e = e->next)
570                 if (!hashcmp(sha1, e->idx.sha1))
571                         return e;
572         return NULL;
573 }
574
575 static struct object_entry *insert_object(unsigned char *sha1)
576 {
577         unsigned int h = sha1[0] << 8 | sha1[1];
578         struct object_entry *e = object_table[h];
579
580         while (e) {
581                 if (!hashcmp(sha1, e->idx.sha1))
582                         return e;
583                 e = e->next;
584         }
585
586         e = new_object(sha1);
587         e->next = object_table[h];
588         e->idx.offset = 0;
589         object_table[h] = e;
590         return e;
591 }
592
593 static unsigned int hc_str(const char *s, size_t len)
594 {
595         unsigned int r = 0;
596         while (len-- > 0)
597                 r = r * 31 + *s++;
598         return r;
599 }
600
601 static void *pool_alloc(size_t len)
602 {
603         struct mem_pool *p;
604         void *r;
605
606         /* round up to a 'uintmax_t' alignment */
607         if (len & (sizeof(uintmax_t) - 1))
608                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
609
610         for (p = mem_pool; p; p = p->next_pool)
611                 if ((p->end - p->next_free >= len))
612                         break;
613
614         if (!p) {
615                 if (len >= (mem_pool_alloc/2)) {
616                         total_allocd += len;
617                         return xmalloc(len);
618                 }
619                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
620                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
621                 p->next_pool = mem_pool;
622                 p->next_free = (char *) p->space;
623                 p->end = p->next_free + mem_pool_alloc;
624                 mem_pool = p;
625         }
626
627         r = p->next_free;
628         p->next_free += len;
629         return r;
630 }
631
632 static void *pool_calloc(size_t count, size_t size)
633 {
634         size_t len = count * size;
635         void *r = pool_alloc(len);
636         memset(r, 0, len);
637         return r;
638 }
639
640 static char *pool_strdup(const char *s)
641 {
642         char *r = pool_alloc(strlen(s) + 1);
643         strcpy(r, s);
644         return r;
645 }
646
647 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
648 {
649         struct mark_set *s = marks;
650         while ((idnum >> s->shift) >= 1024) {
651                 s = pool_calloc(1, sizeof(struct mark_set));
652                 s->shift = marks->shift + 10;
653                 s->data.sets[0] = marks;
654                 marks = s;
655         }
656         while (s->shift) {
657                 uintmax_t i = idnum >> s->shift;
658                 idnum -= i << s->shift;
659                 if (!s->data.sets[i]) {
660                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
661                         s->data.sets[i]->shift = s->shift - 10;
662                 }
663                 s = s->data.sets[i];
664         }
665         if (!s->data.marked[idnum])
666                 marks_set_count++;
667         s->data.marked[idnum] = oe;
668 }
669
670 static struct object_entry *find_mark(uintmax_t idnum)
671 {
672         uintmax_t orig_idnum = idnum;
673         struct mark_set *s = marks;
674         struct object_entry *oe = NULL;
675         if ((idnum >> s->shift) < 1024) {
676                 while (s && s->shift) {
677                         uintmax_t i = idnum >> s->shift;
678                         idnum -= i << s->shift;
679                         s = s->data.sets[i];
680                 }
681                 if (s)
682                         oe = s->data.marked[idnum];
683         }
684         if (!oe)
685                 die("mark :%" PRIuMAX " not declared", orig_idnum);
686         return oe;
687 }
688
689 static struct atom_str *to_atom(const char *s, unsigned short len)
690 {
691         unsigned int hc = hc_str(s, len) % atom_table_sz;
692         struct atom_str *c;
693
694         for (c = atom_table[hc]; c; c = c->next_atom)
695                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
696                         return c;
697
698         c = pool_alloc(sizeof(struct atom_str) + len + 1);
699         c->str_len = len;
700         strncpy(c->str_dat, s, len);
701         c->str_dat[len] = 0;
702         c->next_atom = atom_table[hc];
703         atom_table[hc] = c;
704         atom_cnt++;
705         return c;
706 }
707
708 static struct branch *lookup_branch(const char *name)
709 {
710         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
711         struct branch *b;
712
713         for (b = branch_table[hc]; b; b = b->table_next_branch)
714                 if (!strcmp(name, b->name))
715                         return b;
716         return NULL;
717 }
718
719 static struct branch *new_branch(const char *name)
720 {
721         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
722         struct branch *b = lookup_branch(name);
723
724         if (b)
725                 die("Invalid attempt to create duplicate branch: %s", name);
726         switch (check_ref_format(name)) {
727         case 0: break; /* its valid */
728         case CHECK_REF_FORMAT_ONELEVEL:
729                 break; /* valid, but too few '/', allow anyway */
730         default:
731                 die("Branch name doesn't conform to GIT standards: %s", name);
732         }
733
734         b = pool_calloc(1, sizeof(struct branch));
735         b->name = pool_strdup(name);
736         b->table_next_branch = branch_table[hc];
737         b->branch_tree.versions[0].mode = S_IFDIR;
738         b->branch_tree.versions[1].mode = S_IFDIR;
739         b->num_notes = 0;
740         b->active = 0;
741         b->pack_id = MAX_PACK_ID;
742         branch_table[hc] = b;
743         branch_count++;
744         return b;
745 }
746
747 static unsigned int hc_entries(unsigned int cnt)
748 {
749         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
750         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
751 }
752
753 static struct tree_content *new_tree_content(unsigned int cnt)
754 {
755         struct avail_tree_content *f, *l = NULL;
756         struct tree_content *t;
757         unsigned int hc = hc_entries(cnt);
758
759         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
760                 if (f->entry_capacity >= cnt)
761                         break;
762
763         if (f) {
764                 if (l)
765                         l->next_avail = f->next_avail;
766                 else
767                         avail_tree_table[hc] = f->next_avail;
768         } else {
769                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
770                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
771                 f->entry_capacity = cnt;
772         }
773
774         t = (struct tree_content*)f;
775         t->entry_count = 0;
776         t->delta_depth = 0;
777         return t;
778 }
779
780 static void release_tree_entry(struct tree_entry *e);
781 static void release_tree_content(struct tree_content *t)
782 {
783         struct avail_tree_content *f = (struct avail_tree_content*)t;
784         unsigned int hc = hc_entries(f->entry_capacity);
785         f->next_avail = avail_tree_table[hc];
786         avail_tree_table[hc] = f;
787 }
788
789 static void release_tree_content_recursive(struct tree_content *t)
790 {
791         unsigned int i;
792         for (i = 0; i < t->entry_count; i++)
793                 release_tree_entry(t->entries[i]);
794         release_tree_content(t);
795 }
796
797 static struct tree_content *grow_tree_content(
798         struct tree_content *t,
799         int amt)
800 {
801         struct tree_content *r = new_tree_content(t->entry_count + amt);
802         r->entry_count = t->entry_count;
803         r->delta_depth = t->delta_depth;
804         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
805         release_tree_content(t);
806         return r;
807 }
808
809 static struct tree_entry *new_tree_entry(void)
810 {
811         struct tree_entry *e;
812
813         if (!avail_tree_entry) {
814                 unsigned int n = tree_entry_alloc;
815                 total_allocd += n * sizeof(struct tree_entry);
816                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
817                 while (n-- > 1) {
818                         *((void**)e) = e + 1;
819                         e++;
820                 }
821                 *((void**)e) = NULL;
822         }
823
824         e = avail_tree_entry;
825         avail_tree_entry = *((void**)e);
826         return e;
827 }
828
829 static void release_tree_entry(struct tree_entry *e)
830 {
831         if (e->tree)
832                 release_tree_content_recursive(e->tree);
833         *((void**)e) = avail_tree_entry;
834         avail_tree_entry = e;
835 }
836
837 static struct tree_content *dup_tree_content(struct tree_content *s)
838 {
839         struct tree_content *d;
840         struct tree_entry *a, *b;
841         unsigned int i;
842
843         if (!s)
844                 return NULL;
845         d = new_tree_content(s->entry_count);
846         for (i = 0; i < s->entry_count; i++) {
847                 a = s->entries[i];
848                 b = new_tree_entry();
849                 memcpy(b, a, sizeof(*a));
850                 if (a->tree && is_null_sha1(b->versions[1].sha1))
851                         b->tree = dup_tree_content(a->tree);
852                 else
853                         b->tree = NULL;
854                 d->entries[i] = b;
855         }
856         d->entry_count = s->entry_count;
857         d->delta_depth = s->delta_depth;
858
859         return d;
860 }
861
862 static void start_packfile(void)
863 {
864         static char tmpfile[PATH_MAX];
865         struct packed_git *p;
866         struct pack_header hdr;
867         int pack_fd;
868
869         pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
870                               "pack/tmp_pack_XXXXXX");
871         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
872         strcpy(p->pack_name, tmpfile);
873         p->pack_fd = pack_fd;
874         pack_file = sha1fd(pack_fd, p->pack_name);
875
876         hdr.hdr_signature = htonl(PACK_SIGNATURE);
877         hdr.hdr_version = htonl(2);
878         hdr.hdr_entries = 0;
879         sha1write(pack_file, &hdr, sizeof(hdr));
880
881         pack_data = p;
882         pack_size = sizeof(hdr);
883         object_count = 0;
884
885         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
886         all_packs[pack_id] = p;
887 }
888
889 static const char *create_index(void)
890 {
891         const char *tmpfile;
892         struct pack_idx_entry **idx, **c, **last;
893         struct object_entry *e;
894         struct object_entry_pool *o;
895
896         /* Build the table of object IDs. */
897         idx = xmalloc(object_count * sizeof(*idx));
898         c = idx;
899         for (o = blocks; o; o = o->next_pool)
900                 for (e = o->next_free; e-- != o->entries;)
901                         if (pack_id == e->pack_id)
902                                 *c++ = &e->idx;
903         last = idx + object_count;
904         if (c != last)
905                 die("internal consistency error creating the index");
906
907         tmpfile = write_idx_file(NULL, idx, object_count, pack_data->sha1);
908         free(idx);
909         return tmpfile;
910 }
911
912 static char *keep_pack(const char *curr_index_name)
913 {
914         static char name[PATH_MAX];
915         static const char *keep_msg = "fast-import";
916         int keep_fd;
917
918         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
919         if (keep_fd < 0)
920                 die_errno("cannot create keep file");
921         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
922         if (close(keep_fd))
923                 die_errno("failed to write keep file");
924
925         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
926                  get_object_directory(), sha1_to_hex(pack_data->sha1));
927         if (move_temp_to_file(pack_data->pack_name, name))
928                 die("cannot store pack file");
929
930         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
931                  get_object_directory(), sha1_to_hex(pack_data->sha1));
932         if (move_temp_to_file(curr_index_name, name))
933                 die("cannot store index file");
934         free((void *)curr_index_name);
935         return name;
936 }
937
938 static void unkeep_all_packs(void)
939 {
940         static char name[PATH_MAX];
941         int k;
942
943         for (k = 0; k < pack_id; k++) {
944                 struct packed_git *p = all_packs[k];
945                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
946                          get_object_directory(), sha1_to_hex(p->sha1));
947                 unlink_or_warn(name);
948         }
949 }
950
951 static void end_packfile(void)
952 {
953         struct packed_git *old_p = pack_data, *new_p;
954
955         clear_delta_base_cache();
956         if (object_count) {
957                 unsigned char cur_pack_sha1[20];
958                 char *idx_name;
959                 int i;
960                 struct branch *b;
961                 struct tag *t;
962
963                 close_pack_windows(pack_data);
964                 sha1close(pack_file, cur_pack_sha1, 0);
965                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
966                                     pack_data->pack_name, object_count,
967                                     cur_pack_sha1, pack_size);
968                 close(pack_data->pack_fd);
969                 idx_name = keep_pack(create_index());
970
971                 /* Register the packfile with core git's machinery. */
972                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
973                 if (!new_p)
974                         die("core git rejected index %s", idx_name);
975                 all_packs[pack_id] = new_p;
976                 install_packed_git(new_p);
977
978                 /* Print the boundary */
979                 if (pack_edges) {
980                         fprintf(pack_edges, "%s:", new_p->pack_name);
981                         for (i = 0; i < branch_table_sz; i++) {
982                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
983                                         if (b->pack_id == pack_id)
984                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
985                                 }
986                         }
987                         for (t = first_tag; t; t = t->next_tag) {
988                                 if (t->pack_id == pack_id)
989                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
990                         }
991                         fputc('\n', pack_edges);
992                         fflush(pack_edges);
993                 }
994
995                 pack_id++;
996         }
997         else {
998                 close(old_p->pack_fd);
999                 unlink_or_warn(old_p->pack_name);
1000         }
1001         free(old_p);
1002
1003         /* We can't carry a delta across packfiles. */
1004         strbuf_release(&last_blob.data);
1005         last_blob.offset = 0;
1006         last_blob.depth = 0;
1007 }
1008
1009 static void cycle_packfile(void)
1010 {
1011         end_packfile();
1012         start_packfile();
1013 }
1014
1015 static int store_object(
1016         enum object_type type,
1017         struct strbuf *dat,
1018         struct last_object *last,
1019         unsigned char *sha1out,
1020         uintmax_t mark)
1021 {
1022         void *out, *delta;
1023         struct object_entry *e;
1024         unsigned char hdr[96];
1025         unsigned char sha1[20];
1026         unsigned long hdrlen, deltalen;
1027         git_SHA_CTX c;
1028         z_stream s;
1029
1030         hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1031                 (unsigned long)dat->len) + 1;
1032         git_SHA1_Init(&c);
1033         git_SHA1_Update(&c, hdr, hdrlen);
1034         git_SHA1_Update(&c, dat->buf, dat->len);
1035         git_SHA1_Final(sha1, &c);
1036         if (sha1out)
1037                 hashcpy(sha1out, sha1);
1038
1039         e = insert_object(sha1);
1040         if (mark)
1041                 insert_mark(mark, e);
1042         if (e->idx.offset) {
1043                 duplicate_count_by_type[type]++;
1044                 return 1;
1045         } else if (find_sha1_pack(sha1, packed_git)) {
1046                 e->type = type;
1047                 e->pack_id = MAX_PACK_ID;
1048                 e->idx.offset = 1; /* just not zero! */
1049                 duplicate_count_by_type[type]++;
1050                 return 1;
1051         }
1052
1053         if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1054                 delta = diff_delta(last->data.buf, last->data.len,
1055                         dat->buf, dat->len,
1056                         &deltalen, dat->len - 20);
1057         } else
1058                 delta = NULL;
1059
1060         memset(&s, 0, sizeof(s));
1061         deflateInit(&s, pack_compression_level);
1062         if (delta) {
1063                 s.next_in = delta;
1064                 s.avail_in = deltalen;
1065         } else {
1066                 s.next_in = (void *)dat->buf;
1067                 s.avail_in = dat->len;
1068         }
1069         s.avail_out = deflateBound(&s, s.avail_in);
1070         s.next_out = out = xmalloc(s.avail_out);
1071         while (deflate(&s, Z_FINISH) == Z_OK)
1072                 /* nothing */;
1073         deflateEnd(&s);
1074
1075         /* Determine if we should auto-checkpoint. */
1076         if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1077                 || (pack_size + 60 + s.total_out) < pack_size) {
1078
1079                 /* This new object needs to *not* have the current pack_id. */
1080                 e->pack_id = pack_id + 1;
1081                 cycle_packfile();
1082
1083                 /* We cannot carry a delta into the new pack. */
1084                 if (delta) {
1085                         free(delta);
1086                         delta = NULL;
1087
1088                         memset(&s, 0, sizeof(s));
1089                         deflateInit(&s, pack_compression_level);
1090                         s.next_in = (void *)dat->buf;
1091                         s.avail_in = dat->len;
1092                         s.avail_out = deflateBound(&s, s.avail_in);
1093                         s.next_out = out = xrealloc(out, s.avail_out);
1094                         while (deflate(&s, Z_FINISH) == Z_OK)
1095                                 /* nothing */;
1096                         deflateEnd(&s);
1097                 }
1098         }
1099
1100         e->type = type;
1101         e->pack_id = pack_id;
1102         e->idx.offset = pack_size;
1103         object_count++;
1104         object_count_by_type[type]++;
1105
1106         crc32_begin(pack_file);
1107
1108         if (delta) {
1109                 off_t ofs = e->idx.offset - last->offset;
1110                 unsigned pos = sizeof(hdr) - 1;
1111
1112                 delta_count_by_type[type]++;
1113                 e->depth = last->depth + 1;
1114
1115                 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1116                 sha1write(pack_file, hdr, hdrlen);
1117                 pack_size += hdrlen;
1118
1119                 hdr[pos] = ofs & 127;
1120                 while (ofs >>= 7)
1121                         hdr[--pos] = 128 | (--ofs & 127);
1122                 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1123                 pack_size += sizeof(hdr) - pos;
1124         } else {
1125                 e->depth = 0;
1126                 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1127                 sha1write(pack_file, hdr, hdrlen);
1128                 pack_size += hdrlen;
1129         }
1130
1131         sha1write(pack_file, out, s.total_out);
1132         pack_size += s.total_out;
1133
1134         e->idx.crc32 = crc32_end(pack_file);
1135
1136         free(out);
1137         free(delta);
1138         if (last) {
1139                 if (last->no_swap) {
1140                         last->data = *dat;
1141                 } else {
1142                         strbuf_swap(&last->data, dat);
1143                 }
1144                 last->offset = e->idx.offset;
1145                 last->depth = e->depth;
1146         }
1147         return 0;
1148 }
1149
1150 static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1151 {
1152         if (ftruncate(pack_data->pack_fd, to)
1153          || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1154                 die_errno("cannot truncate pack to skip duplicate");
1155         pack_size = to;
1156
1157         /* yes this is a layering violation */
1158         pack_file->total = to;
1159         pack_file->offset = 0;
1160         pack_file->ctx = *ctx;
1161 }
1162
1163 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1164 {
1165         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1166         unsigned char *in_buf = xmalloc(in_sz);
1167         unsigned char *out_buf = xmalloc(out_sz);
1168         struct object_entry *e;
1169         unsigned char sha1[20];
1170         unsigned long hdrlen;
1171         off_t offset;
1172         git_SHA_CTX c;
1173         git_SHA_CTX pack_file_ctx;
1174         z_stream s;
1175         int status = Z_OK;
1176
1177         /* Determine if we should auto-checkpoint. */
1178         if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1179                 || (pack_size + 60 + len) < pack_size)
1180                 cycle_packfile();
1181
1182         offset = pack_size;
1183
1184         /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1185         sha1flush(pack_file);
1186         pack_file_ctx = pack_file->ctx;
1187
1188         hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1189         if (out_sz <= hdrlen)
1190                 die("impossibly large object header");
1191
1192         git_SHA1_Init(&c);
1193         git_SHA1_Update(&c, out_buf, hdrlen);
1194
1195         crc32_begin(pack_file);
1196
1197         memset(&s, 0, sizeof(s));
1198         deflateInit(&s, pack_compression_level);
1199
1200         hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1201         if (out_sz <= hdrlen)
1202                 die("impossibly large object header");
1203
1204         s.next_out = out_buf + hdrlen;
1205         s.avail_out = out_sz - hdrlen;
1206
1207         while (status != Z_STREAM_END) {
1208                 if (0 < len && !s.avail_in) {
1209                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1210                         size_t n = fread(in_buf, 1, cnt, stdin);
1211                         if (!n && feof(stdin))
1212                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1213
1214                         git_SHA1_Update(&c, in_buf, n);
1215                         s.next_in = in_buf;
1216                         s.avail_in = n;
1217                         len -= n;
1218                 }
1219
1220                 status = deflate(&s, len ? 0 : Z_FINISH);
1221
1222                 if (!s.avail_out || status == Z_STREAM_END) {
1223                         size_t n = s.next_out - out_buf;
1224                         sha1write(pack_file, out_buf, n);
1225                         pack_size += n;
1226                         s.next_out = out_buf;
1227                         s.avail_out = out_sz;
1228                 }
1229
1230                 switch (status) {
1231                 case Z_OK:
1232                 case Z_BUF_ERROR:
1233                 case Z_STREAM_END:
1234                         continue;
1235                 default:
1236                         die("unexpected deflate failure: %d", status);
1237                 }
1238         }
1239         deflateEnd(&s);
1240         git_SHA1_Final(sha1, &c);
1241
1242         if (sha1out)
1243                 hashcpy(sha1out, sha1);
1244
1245         e = insert_object(sha1);
1246
1247         if (mark)
1248                 insert_mark(mark, e);
1249
1250         if (e->idx.offset) {
1251                 duplicate_count_by_type[OBJ_BLOB]++;
1252                 truncate_pack(offset, &pack_file_ctx);
1253
1254         } else if (find_sha1_pack(sha1, packed_git)) {
1255                 e->type = OBJ_BLOB;
1256                 e->pack_id = MAX_PACK_ID;
1257                 e->idx.offset = 1; /* just not zero! */
1258                 duplicate_count_by_type[OBJ_BLOB]++;
1259                 truncate_pack(offset, &pack_file_ctx);
1260
1261         } else {
1262                 e->depth = 0;
1263                 e->type = OBJ_BLOB;
1264                 e->pack_id = pack_id;
1265                 e->idx.offset = offset;
1266                 e->idx.crc32 = crc32_end(pack_file);
1267                 object_count++;
1268                 object_count_by_type[OBJ_BLOB]++;
1269         }
1270
1271         free(in_buf);
1272         free(out_buf);
1273 }
1274
1275 /* All calls must be guarded by find_object() or find_mark() to
1276  * ensure the 'struct object_entry' passed was written by this
1277  * process instance.  We unpack the entry by the offset, avoiding
1278  * the need for the corresponding .idx file.  This unpacking rule
1279  * works because we only use OBJ_REF_DELTA within the packfiles
1280  * created by fast-import.
1281  *
1282  * oe must not be NULL.  Such an oe usually comes from giving
1283  * an unknown SHA-1 to find_object() or an undefined mark to
1284  * find_mark().  Callers must test for this condition and use
1285  * the standard read_sha1_file() when it happens.
1286  *
1287  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1288  * find_mark(), where the mark was reloaded from an existing marks
1289  * file and is referencing an object that this fast-import process
1290  * instance did not write out to a packfile.  Callers must test for
1291  * this condition and use read_sha1_file() instead.
1292  */
1293 static void *gfi_unpack_entry(
1294         struct object_entry *oe,
1295         unsigned long *sizep)
1296 {
1297         enum object_type type;
1298         struct packed_git *p = all_packs[oe->pack_id];
1299         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1300                 /* The object is stored in the packfile we are writing to
1301                  * and we have modified it since the last time we scanned
1302                  * back to read a previously written object.  If an old
1303                  * window covered [p->pack_size, p->pack_size + 20) its
1304                  * data is stale and is not valid.  Closing all windows
1305                  * and updating the packfile length ensures we can read
1306                  * the newly written data.
1307                  */
1308                 close_pack_windows(p);
1309                 sha1flush(pack_file);
1310
1311                 /* We have to offer 20 bytes additional on the end of
1312                  * the packfile as the core unpacker code assumes the
1313                  * footer is present at the file end and must promise
1314                  * at least 20 bytes within any window it maps.  But
1315                  * we don't actually create the footer here.
1316                  */
1317                 p->pack_size = pack_size + 20;
1318         }
1319         return unpack_entry(p, oe->idx.offset, &type, sizep);
1320 }
1321
1322 static const char *get_mode(const char *str, uint16_t *modep)
1323 {
1324         unsigned char c;
1325         uint16_t mode = 0;
1326
1327         while ((c = *str++) != ' ') {
1328                 if (c < '0' || c > '7')
1329                         return NULL;
1330                 mode = (mode << 3) + (c - '0');
1331         }
1332         *modep = mode;
1333         return str;
1334 }
1335
1336 static void load_tree(struct tree_entry *root)
1337 {
1338         unsigned char *sha1 = root->versions[1].sha1;
1339         struct object_entry *myoe;
1340         struct tree_content *t;
1341         unsigned long size;
1342         char *buf;
1343         const char *c;
1344
1345         root->tree = t = new_tree_content(8);
1346         if (is_null_sha1(sha1))
1347                 return;
1348
1349         myoe = find_object(sha1);
1350         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1351                 if (myoe->type != OBJ_TREE)
1352                         die("Not a tree: %s", sha1_to_hex(sha1));
1353                 t->delta_depth = myoe->depth;
1354                 buf = gfi_unpack_entry(myoe, &size);
1355                 if (!buf)
1356                         die("Can't load tree %s", sha1_to_hex(sha1));
1357         } else {
1358                 enum object_type type;
1359                 buf = read_sha1_file(sha1, &type, &size);
1360                 if (!buf || type != OBJ_TREE)
1361                         die("Can't load tree %s", sha1_to_hex(sha1));
1362         }
1363
1364         c = buf;
1365         while (c != (buf + size)) {
1366                 struct tree_entry *e = new_tree_entry();
1367
1368                 if (t->entry_count == t->entry_capacity)
1369                         root->tree = t = grow_tree_content(t, t->entry_count);
1370                 t->entries[t->entry_count++] = e;
1371
1372                 e->tree = NULL;
1373                 c = get_mode(c, &e->versions[1].mode);
1374                 if (!c)
1375                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1376                 e->versions[0].mode = e->versions[1].mode;
1377                 e->name = to_atom(c, strlen(c));
1378                 c += e->name->str_len + 1;
1379                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1380                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1381                 c += 20;
1382         }
1383         free(buf);
1384 }
1385
1386 static int tecmp0 (const void *_a, const void *_b)
1387 {
1388         struct tree_entry *a = *((struct tree_entry**)_a);
1389         struct tree_entry *b = *((struct tree_entry**)_b);
1390         return base_name_compare(
1391                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1392                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1393 }
1394
1395 static int tecmp1 (const void *_a, const void *_b)
1396 {
1397         struct tree_entry *a = *((struct tree_entry**)_a);
1398         struct tree_entry *b = *((struct tree_entry**)_b);
1399         return base_name_compare(
1400                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1401                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1402 }
1403
1404 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1405 {
1406         size_t maxlen = 0;
1407         unsigned int i;
1408
1409         if (!v)
1410                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1411         else
1412                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1413
1414         for (i = 0; i < t->entry_count; i++) {
1415                 if (t->entries[i]->versions[v].mode)
1416                         maxlen += t->entries[i]->name->str_len + 34;
1417         }
1418
1419         strbuf_reset(b);
1420         strbuf_grow(b, maxlen);
1421         for (i = 0; i < t->entry_count; i++) {
1422                 struct tree_entry *e = t->entries[i];
1423                 if (!e->versions[v].mode)
1424                         continue;
1425                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1426                                         e->name->str_dat, '\0');
1427                 strbuf_add(b, e->versions[v].sha1, 20);
1428         }
1429 }
1430
1431 static void store_tree(struct tree_entry *root)
1432 {
1433         struct tree_content *t = root->tree;
1434         unsigned int i, j, del;
1435         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1436         struct object_entry *le;
1437
1438         if (!is_null_sha1(root->versions[1].sha1))
1439                 return;
1440
1441         for (i = 0; i < t->entry_count; i++) {
1442                 if (t->entries[i]->tree)
1443                         store_tree(t->entries[i]);
1444         }
1445
1446         le = find_object(root->versions[0].sha1);
1447         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1448                 mktree(t, 0, &old_tree);
1449                 lo.data = old_tree;
1450                 lo.offset = le->idx.offset;
1451                 lo.depth = t->delta_depth;
1452         }
1453
1454         mktree(t, 1, &new_tree);
1455         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1456
1457         t->delta_depth = lo.depth;
1458         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1459                 struct tree_entry *e = t->entries[i];
1460                 if (e->versions[1].mode) {
1461                         e->versions[0].mode = e->versions[1].mode;
1462                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1463                         t->entries[j++] = e;
1464                 } else {
1465                         release_tree_entry(e);
1466                         del++;
1467                 }
1468         }
1469         t->entry_count -= del;
1470 }
1471
1472 static void tree_content_replace(
1473         struct tree_entry *root,
1474         const unsigned char *sha1,
1475         const uint16_t mode,
1476         struct tree_content *newtree)
1477 {
1478         if (!S_ISDIR(mode))
1479                 die("Root cannot be a non-directory");
1480         hashcpy(root->versions[1].sha1, sha1);
1481         if (root->tree)
1482                 release_tree_content_recursive(root->tree);
1483         root->tree = newtree;
1484 }
1485
1486 static int tree_content_set(
1487         struct tree_entry *root,
1488         const char *p,
1489         const unsigned char *sha1,
1490         const uint16_t mode,
1491         struct tree_content *subtree)
1492 {
1493         struct tree_content *t;
1494         const char *slash1;
1495         unsigned int i, n;
1496         struct tree_entry *e;
1497
1498         slash1 = strchr(p, '/');
1499         if (slash1)
1500                 n = slash1 - p;
1501         else
1502                 n = strlen(p);
1503         if (!n)
1504                 die("Empty path component found in input");
1505         if (!slash1 && !S_ISDIR(mode) && subtree)
1506                 die("Non-directories cannot have subtrees");
1507
1508         if (!root->tree)
1509                 load_tree(root);
1510         t = root->tree;
1511         for (i = 0; i < t->entry_count; i++) {
1512                 e = t->entries[i];
1513                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1514                         if (!slash1) {
1515                                 if (!S_ISDIR(mode)
1516                                                 && e->versions[1].mode == mode
1517                                                 && !hashcmp(e->versions[1].sha1, sha1))
1518                                         return 0;
1519                                 e->versions[1].mode = mode;
1520                                 hashcpy(e->versions[1].sha1, sha1);
1521                                 if (e->tree)
1522                                         release_tree_content_recursive(e->tree);
1523                                 e->tree = subtree;
1524                                 hashclr(root->versions[1].sha1);
1525                                 return 1;
1526                         }
1527                         if (!S_ISDIR(e->versions[1].mode)) {
1528                                 e->tree = new_tree_content(8);
1529                                 e->versions[1].mode = S_IFDIR;
1530                         }
1531                         if (!e->tree)
1532                                 load_tree(e);
1533                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1534                                 hashclr(root->versions[1].sha1);
1535                                 return 1;
1536                         }
1537                         return 0;
1538                 }
1539         }
1540
1541         if (t->entry_count == t->entry_capacity)
1542                 root->tree = t = grow_tree_content(t, t->entry_count);
1543         e = new_tree_entry();
1544         e->name = to_atom(p, n);
1545         e->versions[0].mode = 0;
1546         hashclr(e->versions[0].sha1);
1547         t->entries[t->entry_count++] = e;
1548         if (slash1) {
1549                 e->tree = new_tree_content(8);
1550                 e->versions[1].mode = S_IFDIR;
1551                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1552         } else {
1553                 e->tree = subtree;
1554                 e->versions[1].mode = mode;
1555                 hashcpy(e->versions[1].sha1, sha1);
1556         }
1557         hashclr(root->versions[1].sha1);
1558         return 1;
1559 }
1560
1561 static int tree_content_remove(
1562         struct tree_entry *root,
1563         const char *p,
1564         struct tree_entry *backup_leaf)
1565 {
1566         struct tree_content *t;
1567         const char *slash1;
1568         unsigned int i, n;
1569         struct tree_entry *e;
1570
1571         slash1 = strchr(p, '/');
1572         if (slash1)
1573                 n = slash1 - p;
1574         else
1575                 n = strlen(p);
1576
1577         if (!root->tree)
1578                 load_tree(root);
1579         t = root->tree;
1580         for (i = 0; i < t->entry_count; i++) {
1581                 e = t->entries[i];
1582                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1583                         if (slash1 && !S_ISDIR(e->versions[1].mode))
1584                                 /*
1585                                  * If p names a file in some subdirectory, and a
1586                                  * file or symlink matching the name of the
1587                                  * parent directory of p exists, then p cannot
1588                                  * exist and need not be deleted.
1589                                  */
1590                                 return 1;
1591                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1592                                 goto del_entry;
1593                         if (!e->tree)
1594                                 load_tree(e);
1595                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1596                                 for (n = 0; n < e->tree->entry_count; n++) {
1597                                         if (e->tree->entries[n]->versions[1].mode) {
1598                                                 hashclr(root->versions[1].sha1);
1599                                                 return 1;
1600                                         }
1601                                 }
1602                                 backup_leaf = NULL;
1603                                 goto del_entry;
1604                         }
1605                         return 0;
1606                 }
1607         }
1608         return 0;
1609
1610 del_entry:
1611         if (backup_leaf)
1612                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1613         else if (e->tree)
1614                 release_tree_content_recursive(e->tree);
1615         e->tree = NULL;
1616         e->versions[1].mode = 0;
1617         hashclr(e->versions[1].sha1);
1618         hashclr(root->versions[1].sha1);
1619         return 1;
1620 }
1621
1622 static int tree_content_get(
1623         struct tree_entry *root,
1624         const char *p,
1625         struct tree_entry *leaf)
1626 {
1627         struct tree_content *t;
1628         const char *slash1;
1629         unsigned int i, n;
1630         struct tree_entry *e;
1631
1632         slash1 = strchr(p, '/');
1633         if (slash1)
1634                 n = slash1 - p;
1635         else
1636                 n = strlen(p);
1637
1638         if (!root->tree)
1639                 load_tree(root);
1640         t = root->tree;
1641         for (i = 0; i < t->entry_count; i++) {
1642                 e = t->entries[i];
1643                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1644                         if (!slash1) {
1645                                 memcpy(leaf, e, sizeof(*leaf));
1646                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1647                                         leaf->tree = dup_tree_content(e->tree);
1648                                 else
1649                                         leaf->tree = NULL;
1650                                 return 1;
1651                         }
1652                         if (!S_ISDIR(e->versions[1].mode))
1653                                 return 0;
1654                         if (!e->tree)
1655                                 load_tree(e);
1656                         return tree_content_get(e, slash1 + 1, leaf);
1657                 }
1658         }
1659         return 0;
1660 }
1661
1662 static int update_branch(struct branch *b)
1663 {
1664         static const char *msg = "fast-import";
1665         struct ref_lock *lock;
1666         unsigned char old_sha1[20];
1667
1668         if (is_null_sha1(b->sha1))
1669                 return 0;
1670         if (read_ref(b->name, old_sha1))
1671                 hashclr(old_sha1);
1672         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1673         if (!lock)
1674                 return error("Unable to lock %s", b->name);
1675         if (!force_update && !is_null_sha1(old_sha1)) {
1676                 struct commit *old_cmit, *new_cmit;
1677
1678                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1679                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1680                 if (!old_cmit || !new_cmit) {
1681                         unlock_ref(lock);
1682                         return error("Branch %s is missing commits.", b->name);
1683                 }
1684
1685                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1686                         unlock_ref(lock);
1687                         warning("Not updating %s"
1688                                 " (new tip %s does not contain %s)",
1689                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1690                         return -1;
1691                 }
1692         }
1693         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1694                 return error("Unable to update %s", b->name);
1695         return 0;
1696 }
1697
1698 static void dump_branches(void)
1699 {
1700         unsigned int i;
1701         struct branch *b;
1702
1703         for (i = 0; i < branch_table_sz; i++) {
1704                 for (b = branch_table[i]; b; b = b->table_next_branch)
1705                         failure |= update_branch(b);
1706         }
1707 }
1708
1709 static void dump_tags(void)
1710 {
1711         static const char *msg = "fast-import";
1712         struct tag *t;
1713         struct ref_lock *lock;
1714         char ref_name[PATH_MAX];
1715
1716         for (t = first_tag; t; t = t->next_tag) {
1717                 sprintf(ref_name, "tags/%s", t->name);
1718                 lock = lock_ref_sha1(ref_name, NULL);
1719                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1720                         failure |= error("Unable to update %s", ref_name);
1721         }
1722 }
1723
1724 static void dump_marks_helper(FILE *f,
1725         uintmax_t base,
1726         struct mark_set *m)
1727 {
1728         uintmax_t k;
1729         if (m->shift) {
1730                 for (k = 0; k < 1024; k++) {
1731                         if (m->data.sets[k])
1732                                 dump_marks_helper(f, base + (k << m->shift),
1733                                         m->data.sets[k]);
1734                 }
1735         } else {
1736                 for (k = 0; k < 1024; k++) {
1737                         if (m->data.marked[k])
1738                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1739                                         sha1_to_hex(m->data.marked[k]->idx.sha1));
1740                 }
1741         }
1742 }
1743
1744 static void dump_marks(void)
1745 {
1746         static struct lock_file mark_lock;
1747         int mark_fd;
1748         FILE *f;
1749
1750         if (!export_marks_file)
1751                 return;
1752
1753         mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1754         if (mark_fd < 0) {
1755                 failure |= error("Unable to write marks file %s: %s",
1756                         export_marks_file, strerror(errno));
1757                 return;
1758         }
1759
1760         f = fdopen(mark_fd, "w");
1761         if (!f) {
1762                 int saved_errno = errno;
1763                 rollback_lock_file(&mark_lock);
1764                 failure |= error("Unable to write marks file %s: %s",
1765                         export_marks_file, strerror(saved_errno));
1766                 return;
1767         }
1768
1769         /*
1770          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1771          * Assign -1 to the lock file descriptor so that commit_lock_file()
1772          * won't try to close() it.
1773          */
1774         mark_lock.fd = -1;
1775
1776         dump_marks_helper(f, 0, marks);
1777         if (ferror(f) || fclose(f)) {
1778                 int saved_errno = errno;
1779                 rollback_lock_file(&mark_lock);
1780                 failure |= error("Unable to write marks file %s: %s",
1781                         export_marks_file, strerror(saved_errno));
1782                 return;
1783         }
1784
1785         if (commit_lock_file(&mark_lock)) {
1786                 int saved_errno = errno;
1787                 rollback_lock_file(&mark_lock);
1788                 failure |= error("Unable to commit marks file %s: %s",
1789                         export_marks_file, strerror(saved_errno));
1790                 return;
1791         }
1792 }
1793
1794 static void read_marks(void)
1795 {
1796         char line[512];
1797         FILE *f = fopen(import_marks_file, "r");
1798         if (!f)
1799                 die_errno("cannot read '%s'", import_marks_file);
1800         while (fgets(line, sizeof(line), f)) {
1801                 uintmax_t mark;
1802                 char *end;
1803                 unsigned char sha1[20];
1804                 struct object_entry *e;
1805
1806                 end = strchr(line, '\n');
1807                 if (line[0] != ':' || !end)
1808                         die("corrupt mark line: %s", line);
1809                 *end = 0;
1810                 mark = strtoumax(line + 1, &end, 10);
1811                 if (!mark || end == line + 1
1812                         || *end != ' ' || get_sha1(end + 1, sha1))
1813                         die("corrupt mark line: %s", line);
1814                 e = find_object(sha1);
1815                 if (!e) {
1816                         enum object_type type = sha1_object_info(sha1, NULL);
1817                         if (type < 0)
1818                                 die("object not found: %s", sha1_to_hex(sha1));
1819                         e = insert_object(sha1);
1820                         e->type = type;
1821                         e->pack_id = MAX_PACK_ID;
1822                         e->idx.offset = 1; /* just not zero! */
1823                 }
1824                 insert_mark(mark, e);
1825         }
1826         fclose(f);
1827 }
1828
1829
1830 static int read_next_command(void)
1831 {
1832         static int stdin_eof = 0;
1833
1834         if (stdin_eof) {
1835                 unread_command_buf = 0;
1836                 return EOF;
1837         }
1838
1839         for (;;) {
1840                 if (unread_command_buf) {
1841                         unread_command_buf = 0;
1842                 } else {
1843                         struct recent_command *rc;
1844
1845                         strbuf_detach(&command_buf, NULL);
1846                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1847                         if (stdin_eof)
1848                                 return EOF;
1849
1850                         if (!seen_data_command
1851                                 && prefixcmp(command_buf.buf, "feature ")
1852                                 && prefixcmp(command_buf.buf, "option ")) {
1853                                 parse_argv();
1854                         }
1855
1856                         rc = rc_free;
1857                         if (rc)
1858                                 rc_free = rc->next;
1859                         else {
1860                                 rc = cmd_hist.next;
1861                                 cmd_hist.next = rc->next;
1862                                 cmd_hist.next->prev = &cmd_hist;
1863                                 free(rc->buf);
1864                         }
1865
1866                         rc->buf = command_buf.buf;
1867                         rc->prev = cmd_tail;
1868                         rc->next = cmd_hist.prev;
1869                         rc->prev->next = rc;
1870                         cmd_tail = rc;
1871                 }
1872                 if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1873                         parse_cat_blob();
1874                         continue;
1875                 }
1876                 if (command_buf.buf[0] == '#')
1877                         continue;
1878                 return 0;
1879         }
1880 }
1881
1882 static void skip_optional_lf(void)
1883 {
1884         int term_char = fgetc(stdin);
1885         if (term_char != '\n' && term_char != EOF)
1886                 ungetc(term_char, stdin);
1887 }
1888
1889 static void parse_mark(void)
1890 {
1891         if (!prefixcmp(command_buf.buf, "mark :")) {
1892                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1893                 read_next_command();
1894         }
1895         else
1896                 next_mark = 0;
1897 }
1898
1899 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1900 {
1901         strbuf_reset(sb);
1902
1903         if (prefixcmp(command_buf.buf, "data "))
1904                 die("Expected 'data n' command, found: %s", command_buf.buf);
1905
1906         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1907                 char *term = xstrdup(command_buf.buf + 5 + 2);
1908                 size_t term_len = command_buf.len - 5 - 2;
1909
1910                 strbuf_detach(&command_buf, NULL);
1911                 for (;;) {
1912                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1913                                 die("EOF in data (terminator '%s' not found)", term);
1914                         if (term_len == command_buf.len
1915                                 && !strcmp(term, command_buf.buf))
1916                                 break;
1917                         strbuf_addbuf(sb, &command_buf);
1918                         strbuf_addch(sb, '\n');
1919                 }
1920                 free(term);
1921         }
1922         else {
1923                 uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1924                 size_t n = 0, length = (size_t)len;
1925
1926                 if (limit && limit < len) {
1927                         *len_res = len;
1928                         return 0;
1929                 }
1930                 if (length < len)
1931                         die("data is too large to use in this context");
1932
1933                 while (n < length) {
1934                         size_t s = strbuf_fread(sb, length - n, stdin);
1935                         if (!s && feof(stdin))
1936                                 die("EOF in data (%lu bytes remaining)",
1937                                         (unsigned long)(length - n));
1938                         n += s;
1939                 }
1940         }
1941
1942         skip_optional_lf();
1943         return 1;
1944 }
1945
1946 static int validate_raw_date(const char *src, char *result, int maxlen)
1947 {
1948         const char *orig_src = src;
1949         char *endp;
1950         unsigned long num;
1951
1952         errno = 0;
1953
1954         num = strtoul(src, &endp, 10);
1955         /* NEEDSWORK: perhaps check for reasonable values? */
1956         if (errno || endp == src || *endp != ' ')
1957                 return -1;
1958
1959         src = endp + 1;
1960         if (*src != '-' && *src != '+')
1961                 return -1;
1962
1963         num = strtoul(src + 1, &endp, 10);
1964         if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1965             1400 < num)
1966                 return -1;
1967
1968         strcpy(result, orig_src);
1969         return 0;
1970 }
1971
1972 static char *parse_ident(const char *buf)
1973 {
1974         const char *gt;
1975         size_t name_len;
1976         char *ident;
1977
1978         gt = strrchr(buf, '>');
1979         if (!gt)
1980                 die("Missing > in ident string: %s", buf);
1981         gt++;
1982         if (*gt != ' ')
1983                 die("Missing space after > in ident string: %s", buf);
1984         gt++;
1985         name_len = gt - buf;
1986         ident = xmalloc(name_len + 24);
1987         strncpy(ident, buf, name_len);
1988
1989         switch (whenspec) {
1990         case WHENSPEC_RAW:
1991                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1992                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1993                 break;
1994         case WHENSPEC_RFC2822:
1995                 if (parse_date(gt, ident + name_len, 24) < 0)
1996                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1997                 break;
1998         case WHENSPEC_NOW:
1999                 if (strcmp("now", gt))
2000                         die("Date in ident must be 'now': %s", buf);
2001                 datestamp(ident + name_len, 24);
2002                 break;
2003         }
2004
2005         return ident;
2006 }
2007
2008 static void parse_and_store_blob(
2009         struct last_object *last,
2010         unsigned char *sha1out,
2011         uintmax_t mark)
2012 {
2013         static struct strbuf buf = STRBUF_INIT;
2014         uintmax_t len;
2015
2016         if (parse_data(&buf, big_file_threshold, &len))
2017                 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2018         else {
2019                 if (last) {
2020                         strbuf_release(&last->data);
2021                         last->offset = 0;
2022                         last->depth = 0;
2023                 }
2024                 stream_blob(len, sha1out, mark);
2025                 skip_optional_lf();
2026         }
2027 }
2028
2029 static void parse_new_blob(void)
2030 {
2031         read_next_command();
2032         parse_mark();
2033         parse_and_store_blob(&last_blob, NULL, next_mark);
2034 }
2035
2036 static void unload_one_branch(void)
2037 {
2038         while (cur_active_branches
2039                 && cur_active_branches >= max_active_branches) {
2040                 uintmax_t min_commit = ULONG_MAX;
2041                 struct branch *e, *l = NULL, *p = NULL;
2042
2043                 for (e = active_branches; e; e = e->active_next_branch) {
2044                         if (e->last_commit < min_commit) {
2045                                 p = l;
2046                                 min_commit = e->last_commit;
2047                         }
2048                         l = e;
2049                 }
2050
2051                 if (p) {
2052                         e = p->active_next_branch;
2053                         p->active_next_branch = e->active_next_branch;
2054                 } else {
2055                         e = active_branches;
2056                         active_branches = e->active_next_branch;
2057                 }
2058                 e->active = 0;
2059                 e->active_next_branch = NULL;
2060                 if (e->branch_tree.tree) {
2061                         release_tree_content_recursive(e->branch_tree.tree);
2062                         e->branch_tree.tree = NULL;
2063                 }
2064                 cur_active_branches--;
2065         }
2066 }
2067
2068 static void load_branch(struct branch *b)
2069 {
2070         load_tree(&b->branch_tree);
2071         if (!b->active) {
2072                 b->active = 1;
2073                 b->active_next_branch = active_branches;
2074                 active_branches = b;
2075                 cur_active_branches++;
2076                 branch_load_count++;
2077         }
2078 }
2079
2080 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2081 {
2082         unsigned char fanout = 0;
2083         while ((num_notes >>= 8))
2084                 fanout++;
2085         return fanout;
2086 }
2087
2088 static void construct_path_with_fanout(const char *hex_sha1,
2089                 unsigned char fanout, char *path)
2090 {
2091         unsigned int i = 0, j = 0;
2092         if (fanout >= 20)
2093                 die("Too large fanout (%u)", fanout);
2094         while (fanout) {
2095                 path[i++] = hex_sha1[j++];
2096                 path[i++] = hex_sha1[j++];
2097                 path[i++] = '/';
2098                 fanout--;
2099         }
2100         memcpy(path + i, hex_sha1 + j, 40 - j);
2101         path[i + 40 - j] = '\0';
2102 }
2103
2104 static uintmax_t do_change_note_fanout(
2105                 struct tree_entry *orig_root, struct tree_entry *root,
2106                 char *hex_sha1, unsigned int hex_sha1_len,
2107                 char *fullpath, unsigned int fullpath_len,
2108                 unsigned char fanout)
2109 {
2110         struct tree_content *t = root->tree;
2111         struct tree_entry *e, leaf;
2112         unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2113         uintmax_t num_notes = 0;
2114         unsigned char sha1[20];
2115         char realpath[60];
2116
2117         for (i = 0; t && i < t->entry_count; i++) {
2118                 e = t->entries[i];
2119                 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2120                 tmp_fullpath_len = fullpath_len;
2121
2122                 /*
2123                  * We're interested in EITHER existing note entries (entries
2124                  * with exactly 40 hex chars in path, not including directory
2125                  * separators), OR directory entries that may contain note
2126                  * entries (with < 40 hex chars in path).
2127                  * Also, each path component in a note entry must be a multiple
2128                  * of 2 chars.
2129                  */
2130                 if (!e->versions[1].mode ||
2131                     tmp_hex_sha1_len > 40 ||
2132                     e->name->str_len % 2)
2133                         continue;
2134
2135                 /* This _may_ be a note entry, or a subdir containing notes */
2136                 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2137                        e->name->str_len);
2138                 if (tmp_fullpath_len)
2139                         fullpath[tmp_fullpath_len++] = '/';
2140                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2141                        e->name->str_len);
2142                 tmp_fullpath_len += e->name->str_len;
2143                 fullpath[tmp_fullpath_len] = '\0';
2144
2145                 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2146                         /* This is a note entry */
2147                         construct_path_with_fanout(hex_sha1, fanout, realpath);
2148                         if (!strcmp(fullpath, realpath)) {
2149                                 /* Note entry is in correct location */
2150                                 num_notes++;
2151                                 continue;
2152                         }
2153
2154                         /* Rename fullpath to realpath */
2155                         if (!tree_content_remove(orig_root, fullpath, &leaf))
2156                                 die("Failed to remove path %s", fullpath);
2157                         tree_content_set(orig_root, realpath,
2158                                 leaf.versions[1].sha1,
2159                                 leaf.versions[1].mode,
2160                                 leaf.tree);
2161                 } else if (S_ISDIR(e->versions[1].mode)) {
2162                         /* This is a subdir that may contain note entries */
2163                         if (!e->tree)
2164                                 load_tree(e);
2165                         num_notes += do_change_note_fanout(orig_root, e,
2166                                 hex_sha1, tmp_hex_sha1_len,
2167                                 fullpath, tmp_fullpath_len, fanout);
2168                 }
2169
2170                 /* The above may have reallocated the current tree_content */
2171                 t = root->tree;
2172         }
2173         return num_notes;
2174 }
2175
2176 static uintmax_t change_note_fanout(struct tree_entry *root,
2177                 unsigned char fanout)
2178 {
2179         char hex_sha1[40], path[60];
2180         return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2181 }
2182
2183 static void file_change_m(struct branch *b)
2184 {
2185         const char *p = command_buf.buf + 2;
2186         static struct strbuf uq = STRBUF_INIT;
2187         const char *endp;
2188         struct object_entry *oe = oe;
2189         unsigned char sha1[20];
2190         uint16_t mode, inline_data = 0;
2191
2192         p = get_mode(p, &mode);
2193         if (!p)
2194                 die("Corrupt mode: %s", command_buf.buf);
2195         switch (mode) {
2196         case 0644:
2197         case 0755:
2198                 mode |= S_IFREG;
2199         case S_IFREG | 0644:
2200         case S_IFREG | 0755:
2201         case S_IFLNK:
2202         case S_IFDIR:
2203         case S_IFGITLINK:
2204                 /* ok */
2205                 break;
2206         default:
2207                 die("Corrupt mode: %s", command_buf.buf);
2208         }
2209
2210         if (*p == ':') {
2211                 char *x;
2212                 oe = find_mark(strtoumax(p + 1, &x, 10));
2213                 hashcpy(sha1, oe->idx.sha1);
2214                 p = x;
2215         } else if (!prefixcmp(p, "inline")) {
2216                 inline_data = 1;
2217                 p += 6;
2218         } else {
2219                 if (get_sha1_hex(p, sha1))
2220                         die("Invalid SHA1: %s", command_buf.buf);
2221                 oe = find_object(sha1);
2222                 p += 40;
2223         }
2224         if (*p++ != ' ')
2225                 die("Missing space after SHA1: %s", command_buf.buf);
2226
2227         strbuf_reset(&uq);
2228         if (!unquote_c_style(&uq, p, &endp)) {
2229                 if (*endp)
2230                         die("Garbage after path in: %s", command_buf.buf);
2231                 p = uq.buf;
2232         }
2233
2234         if (S_ISGITLINK(mode)) {
2235                 if (inline_data)
2236                         die("Git links cannot be specified 'inline': %s",
2237                                 command_buf.buf);
2238                 else if (oe) {
2239                         if (oe->type != OBJ_COMMIT)
2240                                 die("Not a commit (actually a %s): %s",
2241                                         typename(oe->type), command_buf.buf);
2242                 }
2243                 /*
2244                  * Accept the sha1 without checking; it expected to be in
2245                  * another repository.
2246                  */
2247         } else if (inline_data) {
2248                 if (S_ISDIR(mode))
2249                         die("Directories cannot be specified 'inline': %s",
2250                                 command_buf.buf);
2251                 if (p != uq.buf) {
2252                         strbuf_addstr(&uq, p);
2253                         p = uq.buf;
2254                 }
2255                 read_next_command();
2256                 parse_and_store_blob(&last_blob, sha1, 0);
2257         } else {
2258                 enum object_type expected = S_ISDIR(mode) ?
2259                                                 OBJ_TREE: OBJ_BLOB;
2260                 enum object_type type = oe ? oe->type :
2261                                         sha1_object_info(sha1, NULL);
2262                 if (type < 0)
2263                         die("%s not found: %s",
2264                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2265                                         command_buf.buf);
2266                 if (type != expected)
2267                         die("Not a %s (actually a %s): %s",
2268                                 typename(expected), typename(type),
2269                                 command_buf.buf);
2270         }
2271
2272         if (!*p) {
2273                 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2274                 return;
2275         }
2276         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2277 }
2278
2279 static void file_change_d(struct branch *b)
2280 {
2281         const char *p = command_buf.buf + 2;
2282         static struct strbuf uq = STRBUF_INIT;
2283         const char *endp;
2284
2285         strbuf_reset(&uq);
2286         if (!unquote_c_style(&uq, p, &endp)) {
2287                 if (*endp)
2288                         die("Garbage after path in: %s", command_buf.buf);
2289                 p = uq.buf;
2290         }
2291         tree_content_remove(&b->branch_tree, p, NULL);
2292 }
2293
2294 static void file_change_cr(struct branch *b, int rename)
2295 {
2296         const char *s, *d;
2297         static struct strbuf s_uq = STRBUF_INIT;
2298         static struct strbuf d_uq = STRBUF_INIT;
2299         const char *endp;
2300         struct tree_entry leaf;
2301
2302         s = command_buf.buf + 2;
2303         strbuf_reset(&s_uq);
2304         if (!unquote_c_style(&s_uq, s, &endp)) {
2305                 if (*endp != ' ')
2306                         die("Missing space after source: %s", command_buf.buf);
2307         } else {
2308                 endp = strchr(s, ' ');
2309                 if (!endp)
2310                         die("Missing space after source: %s", command_buf.buf);
2311                 strbuf_add(&s_uq, s, endp - s);
2312         }
2313         s = s_uq.buf;
2314
2315         endp++;
2316         if (!*endp)
2317                 die("Missing dest: %s", command_buf.buf);
2318
2319         d = endp;
2320         strbuf_reset(&d_uq);
2321         if (!unquote_c_style(&d_uq, d, &endp)) {
2322                 if (*endp)
2323                         die("Garbage after dest in: %s", command_buf.buf);
2324                 d = d_uq.buf;
2325         }
2326
2327         memset(&leaf, 0, sizeof(leaf));
2328         if (rename)
2329                 tree_content_remove(&b->branch_tree, s, &leaf);
2330         else
2331                 tree_content_get(&b->branch_tree, s, &leaf);
2332         if (!leaf.versions[1].mode)
2333                 die("Path %s not in branch", s);
2334         if (!*d) {      /* C "path/to/subdir" "" */
2335                 tree_content_replace(&b->branch_tree,
2336                         leaf.versions[1].sha1,
2337                         leaf.versions[1].mode,
2338                         leaf.tree);
2339                 return;
2340         }
2341         tree_content_set(&b->branch_tree, d,
2342                 leaf.versions[1].sha1,
2343                 leaf.versions[1].mode,
2344                 leaf.tree);
2345 }
2346
2347 static void note_change_n(struct branch *b, unsigned char old_fanout)
2348 {
2349         const char *p = command_buf.buf + 2;
2350         static struct strbuf uq = STRBUF_INIT;
2351         struct object_entry *oe = oe;
2352         struct branch *s;
2353         unsigned char sha1[20], commit_sha1[20];
2354         char path[60];
2355         uint16_t inline_data = 0;
2356         unsigned char new_fanout;
2357
2358         /* <dataref> or 'inline' */
2359         if (*p == ':') {
2360                 char *x;
2361                 oe = find_mark(strtoumax(p + 1, &x, 10));
2362                 hashcpy(sha1, oe->idx.sha1);
2363                 p = x;
2364         } else if (!prefixcmp(p, "inline")) {
2365                 inline_data = 1;
2366                 p += 6;
2367         } else {
2368                 if (get_sha1_hex(p, sha1))
2369                         die("Invalid SHA1: %s", command_buf.buf);
2370                 oe = find_object(sha1);
2371                 p += 40;
2372         }
2373         if (*p++ != ' ')
2374                 die("Missing space after SHA1: %s", command_buf.buf);
2375
2376         /* <committish> */
2377         s = lookup_branch(p);
2378         if (s) {
2379                 hashcpy(commit_sha1, s->sha1);
2380         } else if (*p == ':') {
2381                 uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2382                 struct object_entry *commit_oe = find_mark(commit_mark);
2383                 if (commit_oe->type != OBJ_COMMIT)
2384                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2385                 hashcpy(commit_sha1, commit_oe->idx.sha1);
2386         } else if (!get_sha1(p, commit_sha1)) {
2387                 unsigned long size;
2388                 char *buf = read_object_with_reference(commit_sha1,
2389                         commit_type, &size, commit_sha1);
2390                 if (!buf || size < 46)
2391                         die("Not a valid commit: %s", p);
2392                 free(buf);
2393         } else
2394                 die("Invalid ref name or SHA1 expression: %s", p);
2395
2396         if (inline_data) {
2397                 if (p != uq.buf) {
2398                         strbuf_addstr(&uq, p);
2399                         p = uq.buf;
2400                 }
2401                 read_next_command();
2402                 parse_and_store_blob(&last_blob, sha1, 0);
2403         } else if (oe) {
2404                 if (oe->type != OBJ_BLOB)
2405                         die("Not a blob (actually a %s): %s",
2406                                 typename(oe->type), command_buf.buf);
2407         } else if (!is_null_sha1(sha1)) {
2408                 enum object_type type = sha1_object_info(sha1, NULL);
2409                 if (type < 0)
2410                         die("Blob not found: %s", command_buf.buf);
2411                 if (type != OBJ_BLOB)
2412                         die("Not a blob (actually a %s): %s",
2413                             typename(type), command_buf.buf);
2414         }
2415
2416         construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
2417         if (tree_content_remove(&b->branch_tree, path, NULL))
2418                 b->num_notes--;
2419
2420         if (is_null_sha1(sha1))
2421                 return; /* nothing to insert */
2422
2423         b->num_notes++;
2424         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2425         construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2426         tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2427 }
2428
2429 static void file_change_deleteall(struct branch *b)
2430 {
2431         release_tree_content_recursive(b->branch_tree.tree);
2432         hashclr(b->branch_tree.versions[0].sha1);
2433         hashclr(b->branch_tree.versions[1].sha1);
2434         load_tree(&b->branch_tree);
2435         b->num_notes = 0;
2436 }
2437
2438 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2439 {
2440         if (!buf || size < 46)
2441                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2442         if (memcmp("tree ", buf, 5)
2443                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2444                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2445         hashcpy(b->branch_tree.versions[0].sha1,
2446                 b->branch_tree.versions[1].sha1);
2447 }
2448
2449 static void parse_from_existing(struct branch *b)
2450 {
2451         if (is_null_sha1(b->sha1)) {
2452                 hashclr(b->branch_tree.versions[0].sha1);
2453                 hashclr(b->branch_tree.versions[1].sha1);
2454         } else {
2455                 unsigned long size;
2456                 char *buf;
2457
2458                 buf = read_object_with_reference(b->sha1,
2459                         commit_type, &size, b->sha1);
2460                 parse_from_commit(b, buf, size);
2461                 free(buf);
2462         }
2463 }
2464
2465 static int parse_from(struct branch *b)
2466 {
2467         const char *from;
2468         struct branch *s;
2469
2470         if (prefixcmp(command_buf.buf, "from "))
2471                 return 0;
2472
2473         if (b->branch_tree.tree) {
2474                 release_tree_content_recursive(b->branch_tree.tree);
2475                 b->branch_tree.tree = NULL;
2476         }
2477
2478         from = strchr(command_buf.buf, ' ') + 1;
2479         s = lookup_branch(from);
2480         if (b == s)
2481                 die("Can't create a branch from itself: %s", b->name);
2482         else if (s) {
2483                 unsigned char *t = s->branch_tree.versions[1].sha1;
2484                 hashcpy(b->sha1, s->sha1);
2485                 hashcpy(b->branch_tree.versions[0].sha1, t);
2486                 hashcpy(b->branch_tree.versions[1].sha1, t);
2487         } else if (*from == ':') {
2488                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2489                 struct object_entry *oe = find_mark(idnum);
2490                 if (oe->type != OBJ_COMMIT)
2491                         die("Mark :%" PRIuMAX " not a commit", idnum);
2492                 hashcpy(b->sha1, oe->idx.sha1);
2493                 if (oe->pack_id != MAX_PACK_ID) {
2494                         unsigned long size;
2495                         char *buf = gfi_unpack_entry(oe, &size);
2496                         parse_from_commit(b, buf, size);
2497                         free(buf);
2498                 } else
2499                         parse_from_existing(b);
2500         } else if (!get_sha1(from, b->sha1))
2501                 parse_from_existing(b);
2502         else
2503                 die("Invalid ref name or SHA1 expression: %s", from);
2504
2505         read_next_command();
2506         return 1;
2507 }
2508
2509 static struct hash_list *parse_merge(unsigned int *count)
2510 {
2511         struct hash_list *list = NULL, *n, *e = e;
2512         const char *from;
2513         struct branch *s;
2514
2515         *count = 0;
2516         while (!prefixcmp(command_buf.buf, "merge ")) {
2517                 from = strchr(command_buf.buf, ' ') + 1;
2518                 n = xmalloc(sizeof(*n));
2519                 s = lookup_branch(from);
2520                 if (s)
2521                         hashcpy(n->sha1, s->sha1);
2522                 else if (*from == ':') {
2523                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2524                         struct object_entry *oe = find_mark(idnum);
2525                         if (oe->type != OBJ_COMMIT)
2526                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2527                         hashcpy(n->sha1, oe->idx.sha1);
2528                 } else if (!get_sha1(from, n->sha1)) {
2529                         unsigned long size;
2530                         char *buf = read_object_with_reference(n->sha1,
2531                                 commit_type, &size, n->sha1);
2532                         if (!buf || size < 46)
2533                                 die("Not a valid commit: %s", from);
2534                         free(buf);
2535                 } else
2536                         die("Invalid ref name or SHA1 expression: %s", from);
2537
2538                 n->next = NULL;
2539                 if (list)
2540                         e->next = n;
2541                 else
2542                         list = n;
2543                 e = n;
2544                 (*count)++;
2545                 read_next_command();
2546         }
2547         return list;
2548 }
2549
2550 static void parse_new_commit(void)
2551 {
2552         static struct strbuf msg = STRBUF_INIT;
2553         struct branch *b;
2554         char *sp;
2555         char *author = NULL;
2556         char *committer = NULL;
2557         struct hash_list *merge_list = NULL;
2558         unsigned int merge_count;
2559         unsigned char prev_fanout, new_fanout;
2560
2561         /* Obtain the branch name from the rest of our command */
2562         sp = strchr(command_buf.buf, ' ') + 1;
2563         b = lookup_branch(sp);
2564         if (!b)
2565                 b = new_branch(sp);
2566
2567         read_next_command();
2568         parse_mark();
2569         if (!prefixcmp(command_buf.buf, "author ")) {
2570                 author = parse_ident(command_buf.buf + 7);
2571                 read_next_command();
2572         }
2573         if (!prefixcmp(command_buf.buf, "committer ")) {
2574                 committer = parse_ident(command_buf.buf + 10);
2575                 read_next_command();
2576         }
2577         if (!committer)
2578                 die("Expected committer but didn't get one");
2579         parse_data(&msg, 0, NULL);
2580         read_next_command();
2581         parse_from(b);
2582         merge_list = parse_merge(&merge_count);
2583
2584         /* ensure the branch is active/loaded */
2585         if (!b->branch_tree.tree || !max_active_branches) {
2586                 unload_one_branch();
2587                 load_branch(b);
2588         }
2589
2590         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2591
2592         /* file_change* */
2593         while (command_buf.len > 0) {
2594                 if (!prefixcmp(command_buf.buf, "M "))
2595                         file_change_m(b);
2596                 else if (!prefixcmp(command_buf.buf, "D "))
2597                         file_change_d(b);
2598                 else if (!prefixcmp(command_buf.buf, "R "))
2599                         file_change_cr(b, 1);
2600                 else if (!prefixcmp(command_buf.buf, "C "))
2601                         file_change_cr(b, 0);
2602                 else if (!prefixcmp(command_buf.buf, "N "))
2603                         note_change_n(b, prev_fanout);
2604                 else if (!strcmp("deleteall", command_buf.buf))
2605                         file_change_deleteall(b);
2606                 else {
2607                         unread_command_buf = 1;
2608                         break;
2609                 }
2610                 if (read_next_command() == EOF)
2611                         break;
2612         }
2613
2614         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2615         if (new_fanout != prev_fanout)
2616                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2617
2618         /* build the tree and the commit */
2619         store_tree(&b->branch_tree);
2620         hashcpy(b->branch_tree.versions[0].sha1,
2621                 b->branch_tree.versions[1].sha1);
2622
2623         strbuf_reset(&new_data);
2624         strbuf_addf(&new_data, "tree %s\n",
2625                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2626         if (!is_null_sha1(b->sha1))
2627                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2628         while (merge_list) {
2629                 struct hash_list *next = merge_list->next;
2630                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2631                 free(merge_list);
2632                 merge_list = next;
2633         }
2634         strbuf_addf(&new_data,
2635                 "author %s\n"
2636                 "committer %s\n"
2637                 "\n",
2638                 author ? author : committer, committer);
2639         strbuf_addbuf(&new_data, &msg);
2640         free(author);
2641         free(committer);
2642
2643         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2644                 b->pack_id = pack_id;
2645         b->last_commit = object_count_by_type[OBJ_COMMIT];
2646 }
2647
2648 static void parse_new_tag(void)
2649 {
2650         static struct strbuf msg = STRBUF_INIT;
2651         char *sp;
2652         const char *from;
2653         char *tagger;
2654         struct branch *s;
2655         struct tag *t;
2656         uintmax_t from_mark = 0;
2657         unsigned char sha1[20];
2658         enum object_type type;
2659
2660         /* Obtain the new tag name from the rest of our command */
2661         sp = strchr(command_buf.buf, ' ') + 1;
2662         t = pool_alloc(sizeof(struct tag));
2663         t->next_tag = NULL;
2664         t->name = pool_strdup(sp);
2665         if (last_tag)
2666                 last_tag->next_tag = t;
2667         else
2668                 first_tag = t;
2669         last_tag = t;
2670         read_next_command();
2671
2672         /* from ... */
2673         if (prefixcmp(command_buf.buf, "from "))
2674                 die("Expected from command, got %s", command_buf.buf);
2675         from = strchr(command_buf.buf, ' ') + 1;
2676         s = lookup_branch(from);
2677         if (s) {
2678                 hashcpy(sha1, s->sha1);
2679                 type = OBJ_COMMIT;
2680         } else if (*from == ':') {
2681                 struct object_entry *oe;
2682                 from_mark = strtoumax(from + 1, NULL, 10);
2683                 oe = find_mark(from_mark);
2684                 type = oe->type;
2685                 hashcpy(sha1, oe->idx.sha1);
2686         } else if (!get_sha1(from, sha1)) {
2687                 unsigned long size;
2688                 char *buf;
2689
2690                 buf = read_sha1_file(sha1, &type, &size);
2691                 if (!buf || size < 46)
2692                         die("Not a valid commit: %s", from);
2693                 free(buf);
2694         } else
2695                 die("Invalid ref name or SHA1 expression: %s", from);
2696         read_next_command();
2697
2698         /* tagger ... */
2699         if (!prefixcmp(command_buf.buf, "tagger ")) {
2700                 tagger = parse_ident(command_buf.buf + 7);
2701                 read_next_command();
2702         } else
2703                 tagger = NULL;
2704
2705         /* tag payload/message */
2706         parse_data(&msg, 0, NULL);
2707
2708         /* build the tag object */
2709         strbuf_reset(&new_data);
2710
2711         strbuf_addf(&new_data,
2712                     "object %s\n"
2713                     "type %s\n"
2714                     "tag %s\n",
2715                     sha1_to_hex(sha1), typename(type), t->name);
2716         if (tagger)
2717                 strbuf_addf(&new_data,
2718                             "tagger %s\n", tagger);
2719         strbuf_addch(&new_data, '\n');
2720         strbuf_addbuf(&new_data, &msg);
2721         free(tagger);
2722
2723         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2724                 t->pack_id = MAX_PACK_ID;
2725         else
2726                 t->pack_id = pack_id;
2727 }
2728
2729 static void parse_reset_branch(void)
2730 {
2731         struct branch *b;
2732         char *sp;
2733
2734         /* Obtain the branch name from the rest of our command */
2735         sp = strchr(command_buf.buf, ' ') + 1;
2736         b = lookup_branch(sp);
2737         if (b) {
2738                 hashclr(b->sha1);
2739                 hashclr(b->branch_tree.versions[0].sha1);
2740                 hashclr(b->branch_tree.versions[1].sha1);
2741                 if (b->branch_tree.tree) {
2742                         release_tree_content_recursive(b->branch_tree.tree);
2743                         b->branch_tree.tree = NULL;
2744                 }
2745         }
2746         else
2747                 b = new_branch(sp);
2748         read_next_command();
2749         parse_from(b);
2750         if (command_buf.len > 0)
2751                 unread_command_buf = 1;
2752 }
2753
2754 static void cat_blob_write(const char *buf, unsigned long size)
2755 {
2756         if (write_in_full(cat_blob_fd, buf, size) != size)
2757                 die_errno("Write to frontend failed");
2758 }
2759
2760 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2761 {
2762         struct strbuf line = STRBUF_INIT;
2763         unsigned long size;
2764         enum object_type type = 0;
2765         char *buf;
2766
2767         if (!oe || oe->pack_id == MAX_PACK_ID) {
2768                 buf = read_sha1_file(sha1, &type, &size);
2769         } else {
2770                 type = oe->type;
2771                 buf = gfi_unpack_entry(oe, &size);
2772         }
2773
2774         /*
2775          * Output based on batch_one_object() from cat-file.c.
2776          */
2777         if (type <= 0) {
2778                 strbuf_reset(&line);
2779                 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2780                 cat_blob_write(line.buf, line.len);
2781                 strbuf_release(&line);
2782                 free(buf);
2783                 return;
2784         }
2785         if (!buf)
2786                 die("Can't read object %s", sha1_to_hex(sha1));
2787         if (type != OBJ_BLOB)
2788                 die("Object %s is a %s but a blob was expected.",
2789                     sha1_to_hex(sha1), typename(type));
2790         strbuf_reset(&line);
2791         strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2792                                                 typename(type), size);
2793         cat_blob_write(line.buf, line.len);
2794         strbuf_release(&line);
2795         cat_blob_write(buf, size);
2796         cat_blob_write("\n", 1);
2797         free(buf);
2798 }
2799
2800 static void parse_cat_blob(void)
2801 {
2802         const char *p;
2803         struct object_entry *oe = oe;
2804         unsigned char sha1[20];
2805
2806         /* cat-blob SP <object> LF */
2807         p = command_buf.buf + strlen("cat-blob ");
2808         if (*p == ':') {
2809                 char *x;
2810                 oe = find_mark(strtoumax(p + 1, &x, 10));
2811                 if (x == p + 1)
2812                         die("Invalid mark: %s", command_buf.buf);
2813                 if (!oe)
2814                         die("Unknown mark: %s", command_buf.buf);
2815                 if (*x)
2816                         die("Garbage after mark: %s", command_buf.buf);
2817                 hashcpy(sha1, oe->idx.sha1);
2818         } else {
2819                 if (get_sha1_hex(p, sha1))
2820                         die("Invalid SHA1: %s", command_buf.buf);
2821                 if (p[40])
2822                         die("Garbage after SHA1: %s", command_buf.buf);
2823                 oe = find_object(sha1);
2824         }
2825
2826         cat_blob(oe, sha1);
2827 }
2828
2829 static void checkpoint(void)
2830 {
2831         checkpoint_requested = 0;
2832         if (object_count) {
2833                 cycle_packfile();
2834                 dump_branches();
2835                 dump_tags();
2836                 dump_marks();
2837         }
2838 }
2839
2840 static void parse_checkpoint(void)
2841 {
2842         checkpoint_requested = 1;
2843         skip_optional_lf();
2844 }
2845
2846 static void parse_progress(void)
2847 {
2848         fwrite(command_buf.buf, 1, command_buf.len, stdout);
2849         fputc('\n', stdout);
2850         fflush(stdout);
2851         skip_optional_lf();
2852 }
2853
2854 static char* make_fast_import_path(const char *path)
2855 {
2856         struct strbuf abs_path = STRBUF_INIT;
2857
2858         if (!relative_marks_paths || is_absolute_path(path))
2859                 return xstrdup(path);
2860         strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
2861         return strbuf_detach(&abs_path, NULL);
2862 }
2863
2864 static void option_import_marks(const char *marks, int from_stream)
2865 {
2866         if (import_marks_file) {
2867                 if (from_stream)
2868                         die("Only one import-marks command allowed per stream");
2869
2870                 /* read previous mark file */
2871                 if(!import_marks_file_from_stream)
2872                         read_marks();
2873         }
2874
2875         import_marks_file = make_fast_import_path(marks);
2876         safe_create_leading_directories_const(import_marks_file);
2877         import_marks_file_from_stream = from_stream;
2878 }
2879
2880 static void option_date_format(const char *fmt)
2881 {
2882         if (!strcmp(fmt, "raw"))
2883                 whenspec = WHENSPEC_RAW;
2884         else if (!strcmp(fmt, "rfc2822"))
2885                 whenspec = WHENSPEC_RFC2822;
2886         else if (!strcmp(fmt, "now"))
2887                 whenspec = WHENSPEC_NOW;
2888         else
2889                 die("unknown --date-format argument %s", fmt);
2890 }
2891
2892 static unsigned long ulong_arg(const char *option, const char *arg)
2893 {
2894         char *endptr;
2895         unsigned long rv = strtoul(arg, &endptr, 0);
2896         if (strchr(arg, '-') || endptr == arg || *endptr)
2897                 die("%s: argument must be a non-negative integer", option);
2898         return rv;
2899 }
2900
2901 static void option_depth(const char *depth)
2902 {
2903         max_depth = ulong_arg("--depth", depth);
2904         if (max_depth > MAX_DEPTH)
2905                 die("--depth cannot exceed %u", MAX_DEPTH);
2906 }
2907
2908 static void option_active_branches(const char *branches)
2909 {
2910         max_active_branches = ulong_arg("--active-branches", branches);
2911 }
2912
2913 static void option_export_marks(const char *marks)
2914 {
2915         export_marks_file = make_fast_import_path(marks);
2916         safe_create_leading_directories_const(export_marks_file);
2917 }
2918
2919 static void option_cat_blob_fd(const char *fd)
2920 {
2921         unsigned long n = ulong_arg("--cat-blob-fd", fd);
2922         if (n > (unsigned long) INT_MAX)
2923                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
2924         cat_blob_fd = (int) n;
2925 }
2926
2927 static void option_export_pack_edges(const char *edges)
2928 {
2929         if (pack_edges)
2930                 fclose(pack_edges);
2931         pack_edges = fopen(edges, "a");
2932         if (!pack_edges)
2933                 die_errno("Cannot open '%s'", edges);
2934 }
2935
2936 static int parse_one_option(const char *option)
2937 {
2938         if (!prefixcmp(option, "max-pack-size=")) {
2939                 unsigned long v;
2940                 if (!git_parse_ulong(option + 14, &v))
2941                         return 0;
2942                 if (v < 8192) {
2943                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
2944                         v *= 1024 * 1024;
2945                 } else if (v < 1024 * 1024) {
2946                         warning("minimum max-pack-size is 1 MiB");
2947                         v = 1024 * 1024;
2948                 }
2949                 max_packsize = v;
2950         } else if (!prefixcmp(option, "big-file-threshold=")) {
2951                 unsigned long v;
2952                 if (!git_parse_ulong(option + 19, &v))
2953                         return 0;
2954                 big_file_threshold = v;
2955         } else if (!prefixcmp(option, "depth=")) {
2956                 option_depth(option + 6);
2957         } else if (!prefixcmp(option, "active-branches=")) {
2958                 option_active_branches(option + 16);
2959         } else if (!prefixcmp(option, "export-pack-edges=")) {
2960                 option_export_pack_edges(option + 18);
2961         } else if (!prefixcmp(option, "quiet")) {
2962                 show_stats = 0;
2963         } else if (!prefixcmp(option, "stats")) {
2964                 show_stats = 1;
2965         } else {
2966                 return 0;
2967         }
2968
2969         return 1;
2970 }
2971
2972 static int parse_one_feature(const char *feature, int from_stream)
2973 {
2974         if (!prefixcmp(feature, "date-format=")) {
2975                 option_date_format(feature + 12);
2976         } else if (!prefixcmp(feature, "import-marks=")) {
2977                 option_import_marks(feature + 13, from_stream);
2978         } else if (!prefixcmp(feature, "export-marks=")) {
2979                 option_export_marks(feature + 13);
2980         } else if (!strcmp(feature, "cat-blob")) {
2981                 ; /* Don't die - this feature is supported */
2982         } else if (!prefixcmp(feature, "relative-marks")) {
2983                 relative_marks_paths = 1;
2984         } else if (!prefixcmp(feature, "no-relative-marks")) {
2985                 relative_marks_paths = 0;
2986         } else if (!prefixcmp(feature, "force")) {
2987                 force_update = 1;
2988         } else {
2989                 return 0;
2990         }
2991
2992         return 1;
2993 }
2994
2995 static void parse_feature(void)
2996 {
2997         char *feature = command_buf.buf + 8;
2998
2999         if (seen_data_command)
3000                 die("Got feature command '%s' after data command", feature);
3001
3002         if (parse_one_feature(feature, 1))
3003                 return;
3004
3005         die("This version of fast-import does not support feature %s.", feature);
3006 }
3007
3008 static void parse_option(void)
3009 {
3010         char *option = command_buf.buf + 11;
3011
3012         if (seen_data_command)
3013                 die("Got option command '%s' after data command", option);
3014
3015         if (parse_one_option(option))
3016                 return;
3017
3018         die("This version of fast-import does not support option: %s", option);
3019 }
3020
3021 static int git_pack_config(const char *k, const char *v, void *cb)
3022 {
3023         if (!strcmp(k, "pack.depth")) {
3024                 max_depth = git_config_int(k, v);
3025                 if (max_depth > MAX_DEPTH)
3026                         max_depth = MAX_DEPTH;
3027                 return 0;
3028         }
3029         if (!strcmp(k, "pack.compression")) {
3030                 int level = git_config_int(k, v);
3031                 if (level == -1)
3032                         level = Z_DEFAULT_COMPRESSION;
3033                 else if (level < 0 || level > Z_BEST_COMPRESSION)
3034                         die("bad pack compression level %d", level);
3035                 pack_compression_level = level;
3036                 pack_compression_seen = 1;
3037                 return 0;
3038         }
3039         if (!strcmp(k, "pack.indexversion")) {
3040                 pack_idx_default_version = git_config_int(k, v);
3041                 if (pack_idx_default_version > 2)
3042                         die("bad pack.indexversion=%"PRIu32,
3043                             pack_idx_default_version);
3044                 return 0;
3045         }
3046         if (!strcmp(k, "pack.packsizelimit")) {
3047                 max_packsize = git_config_ulong(k, v);
3048                 return 0;
3049         }
3050         if (!strcmp(k, "core.bigfilethreshold")) {
3051                 long n = git_config_int(k, v);
3052                 big_file_threshold = 0 < n ? n : 0;
3053         }
3054         return git_default_config(k, v, cb);
3055 }
3056
3057 static const char fast_import_usage[] =
3058 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3059
3060 static void parse_argv(void)
3061 {
3062         unsigned int i;
3063
3064         for (i = 1; i < global_argc; i++) {
3065                 const char *a = global_argv[i];
3066
3067                 if (*a != '-' || !strcmp(a, "--"))
3068                         break;
3069
3070                 if (parse_one_option(a + 2))
3071                         continue;
3072
3073                 if (parse_one_feature(a + 2, 0))
3074                         continue;
3075
3076                 if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3077                         option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3078                         continue;
3079                 }
3080
3081                 die("unknown option %s", a);
3082         }
3083         if (i != global_argc)
3084                 usage(fast_import_usage);
3085
3086         seen_data_command = 1;
3087         if (import_marks_file)
3088                 read_marks();
3089 }
3090
3091 int main(int argc, const char **argv)
3092 {
3093         unsigned int i;
3094
3095         git_extract_argv0_path(argv[0]);
3096
3097         if (argc == 2 && !strcmp(argv[1], "-h"))
3098                 usage(fast_import_usage);
3099
3100         setup_git_directory();
3101         git_config(git_pack_config, NULL);
3102         if (!pack_compression_seen && core_compression_seen)
3103                 pack_compression_level = core_compression_level;
3104
3105         alloc_objects(object_entry_alloc);
3106         strbuf_init(&command_buf, 0);
3107         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3108         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3109         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3110         marks = pool_calloc(1, sizeof(struct mark_set));
3111
3112         global_argc = argc;
3113         global_argv = argv;
3114
3115         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3116         for (i = 0; i < (cmd_save - 1); i++)
3117                 rc_free[i].next = &rc_free[i + 1];
3118         rc_free[cmd_save - 1].next = NULL;
3119
3120         prepare_packed_git();
3121         start_packfile();
3122         set_die_routine(die_nicely);
3123         set_checkpoint_signal();
3124         while (read_next_command() != EOF) {
3125                 if (!strcmp("blob", command_buf.buf))
3126                         parse_new_blob();
3127                 else if (!prefixcmp(command_buf.buf, "commit "))
3128                         parse_new_commit();
3129                 else if (!prefixcmp(command_buf.buf, "tag "))
3130                         parse_new_tag();
3131                 else if (!prefixcmp(command_buf.buf, "reset "))
3132                         parse_reset_branch();
3133                 else if (!strcmp("checkpoint", command_buf.buf))
3134                         parse_checkpoint();
3135                 else if (!prefixcmp(command_buf.buf, "progress "))
3136                         parse_progress();
3137                 else if (!prefixcmp(command_buf.buf, "feature "))
3138                         parse_feature();
3139                 else if (!prefixcmp(command_buf.buf, "option git "))
3140                         parse_option();
3141                 else if (!prefixcmp(command_buf.buf, "option "))
3142                         /* ignore non-git options*/;
3143                 else
3144                         die("Unsupported command: %s", command_buf.buf);
3145
3146                 if (checkpoint_requested)
3147                         checkpoint();
3148         }
3149
3150         /* argv hasn't been parsed yet, do so */
3151         if (!seen_data_command)
3152                 parse_argv();
3153
3154         end_packfile();
3155
3156         dump_branches();
3157         dump_tags();
3158         unkeep_all_packs();
3159         dump_marks();
3160
3161         if (pack_edges)
3162                 fclose(pack_edges);
3163
3164         if (show_stats) {
3165                 uintmax_t total_count = 0, duplicate_count = 0;
3166                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3167                         total_count += object_count_by_type[i];
3168                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3169                         duplicate_count += duplicate_count_by_type[i];
3170
3171                 fprintf(stderr, "%s statistics:\n", argv[0]);
3172                 fprintf(stderr, "---------------------------------------------------------------------\n");
3173                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3174                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3175                 fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
3176                 fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
3177                 fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
3178                 fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
3179                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3180                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3181                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3182                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3183                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3184                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3185                 fprintf(stderr, "---------------------------------------------------------------------\n");
3186                 pack_report();
3187                 fprintf(stderr, "---------------------------------------------------------------------\n");
3188                 fprintf(stderr, "\n");
3189         }
3190
3191         return failure ? 1 : 0;
3192 }