Merge branch 'jc/checkout-orphan-warning'
[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 | ls)*
28     lf?;
29   commit_msg ::= data;
30
31   ls ::= 'ls' sp '"' quoted(path) '"' lf;
32
33   file_change ::= file_clr
34     | file_del
35     | file_rnm
36     | file_cpy
37     | file_obm
38     | file_inm;
39   file_clr ::= 'deleteall' lf;
40   file_del ::= 'D' sp path_str lf;
41   file_rnm ::= 'R' sp path_str sp path_str lf;
42   file_cpy ::= 'C' sp path_str sp path_str lf;
43   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
44   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
45     data;
46   note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
47   note_inm ::= 'N' sp 'inline' sp committish lf
48     data;
49
50   new_tag ::= 'tag' sp tag_str lf
51     'from' sp committish lf
52     ('tagger' (sp name)? sp '<' email '>' sp when lf)?
53     tag_msg;
54   tag_msg ::= data;
55
56   reset_branch ::= 'reset' sp ref_str lf
57     ('from' sp committish lf)?
58     lf?;
59
60   checkpoint ::= 'checkpoint' lf
61     lf?;
62
63   progress ::= 'progress' sp not_lf* lf
64     lf?;
65
66      # note: the first idnum in a stream should be 1 and subsequent
67      # idnums should not have gaps between values as this will cause
68      # the stream parser to reserve space for the gapped values.  An
69      # idnum can be updated in the future to a new object by issuing
70      # a new mark directive with the old idnum.
71      #
72   mark ::= 'mark' sp idnum lf;
73   data ::= (delimited_data | exact_data)
74     lf?;
75
76     # note: delim may be any string but must not contain lf.
77     # data_line may contain any data but must not be exactly
78     # delim.
79   delimited_data ::= 'data' sp '<<' delim lf
80     (data_line lf)*
81     delim lf;
82
83      # note: declen indicates the length of binary_data in bytes.
84      # declen does not include the lf preceding the binary data.
85      #
86   exact_data ::= 'data' sp declen lf
87     binary_data;
88
89      # note: quoted strings are C-style quoting supporting \c for
90      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
91      # is the signed byte value in octal.  Note that the only
92      # characters which must actually be escaped to protect the
93      # stream formatting is: \, " and LF.  Otherwise these values
94      # are UTF8.
95      #
96   committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
97   ref_str     ::= ref;
98   sha1exp_str ::= sha1exp;
99   tag_str     ::= tag;
100   path_str    ::= path    | '"' quoted(path)    '"' ;
101   mode        ::= '100644' | '644'
102                 | '100755' | '755'
103                 | '120000'
104                 ;
105
106   declen ::= # unsigned 32 bit value, ascii base10 notation;
107   bigint ::= # unsigned integer value, ascii base10 notation;
108   binary_data ::= # file content, not interpreted;
109
110   when         ::= raw_when | rfc2822_when;
111   raw_when     ::= ts sp tz;
112   rfc2822_when ::= # Valid RFC 2822 date and time;
113
114   sp ::= # ASCII space character;
115   lf ::= # ASCII newline (LF) character;
116
117      # note: a colon (':') must precede the numerical value assigned to
118      # an idnum.  This is to distinguish it from a ref or tag name as
119      # GIT does not permit ':' in ref or tag strings.
120      #
121   idnum   ::= ':' bigint;
122   path    ::= # GIT style file path, e.g. "a/b/c";
123   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
124   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
125   sha1exp ::= # Any valid GIT SHA1 expression;
126   hexsha1 ::= # SHA1 in hexadecimal format;
127
128      # note: name and email are UTF8 strings, however name must not
129      # contain '<' or lf and email must not contain any of the
130      # following: '<', '>', lf.
131      #
132   name  ::= # valid GIT author/committer name;
133   email ::= # valid GIT author/committer email;
134   ts    ::= # time since the epoch in seconds, ascii base10 notation;
135   tz    ::= # GIT style timezone;
136
137      # note: comments, ls and cat requests may appear anywhere
138      # in the input, except within a data command.  Any form
139      # of the data command always escapes the related input
140      # from comment processing.
141      #
142      # In case it is not clear, the '#' that starts the comment
143      # must be the first character on that line (an lf
144      # preceded it).
145      #
146
147   cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
148   ls_tree  ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
149
150   comment ::= '#' not_lf* lf;
151   not_lf  ::= # Any byte that is not ASCII newline (LF);
152 */
153
154 #include "builtin.h"
155 #include "cache.h"
156 #include "object.h"
157 #include "blob.h"
158 #include "tree.h"
159 #include "commit.h"
160 #include "delta.h"
161 #include "pack.h"
162 #include "refs.h"
163 #include "csum-file.h"
164 #include "quote.h"
165 #include "exec_cmd.h"
166 #include "dir.h"
167
168 #define PACK_ID_BITS 16
169 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
170 #define DEPTH_BITS 13
171 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
172
173 struct object_entry
174 {
175         struct pack_idx_entry idx;
176         struct object_entry *next;
177         uint32_t type : TYPE_BITS,
178                 pack_id : PACK_ID_BITS,
179                 depth : DEPTH_BITS;
180 };
181
182 struct object_entry_pool
183 {
184         struct object_entry_pool *next_pool;
185         struct object_entry *next_free;
186         struct object_entry *end;
187         struct object_entry entries[FLEX_ARRAY]; /* more */
188 };
189
190 struct mark_set
191 {
192         union {
193                 struct object_entry *marked[1024];
194                 struct mark_set *sets[1024];
195         } data;
196         unsigned int shift;
197 };
198
199 struct last_object
200 {
201         struct strbuf data;
202         off_t offset;
203         unsigned int depth;
204         unsigned no_swap : 1;
205 };
206
207 struct mem_pool
208 {
209         struct mem_pool *next_pool;
210         char *next_free;
211         char *end;
212         uintmax_t space[FLEX_ARRAY]; /* more */
213 };
214
215 struct atom_str
216 {
217         struct atom_str *next_atom;
218         unsigned short str_len;
219         char str_dat[FLEX_ARRAY]; /* more */
220 };
221
222 struct tree_content;
223 struct tree_entry
224 {
225         struct tree_content *tree;
226         struct atom_str *name;
227         struct tree_entry_ms
228         {
229                 uint16_t mode;
230                 unsigned char sha1[20];
231         } versions[2];
232 };
233
234 struct tree_content
235 {
236         unsigned int entry_capacity; /* must match avail_tree_content */
237         unsigned int entry_count;
238         unsigned int delta_depth;
239         struct tree_entry *entries[FLEX_ARRAY]; /* more */
240 };
241
242 struct avail_tree_content
243 {
244         unsigned int entry_capacity; /* must match tree_content */
245         struct avail_tree_content *next_avail;
246 };
247
248 struct branch
249 {
250         struct branch *table_next_branch;
251         struct branch *active_next_branch;
252         const char *name;
253         struct tree_entry branch_tree;
254         uintmax_t last_commit;
255         uintmax_t num_notes;
256         unsigned active : 1;
257         unsigned pack_id : PACK_ID_BITS;
258         unsigned char sha1[20];
259 };
260
261 struct tag
262 {
263         struct tag *next_tag;
264         const char *name;
265         unsigned int pack_id;
266         unsigned char sha1[20];
267 };
268
269 struct hash_list
270 {
271         struct hash_list *next;
272         unsigned char sha1[20];
273 };
274
275 typedef enum {
276         WHENSPEC_RAW = 1,
277         WHENSPEC_RFC2822,
278         WHENSPEC_NOW
279 } whenspec_type;
280
281 struct recent_command
282 {
283         struct recent_command *prev;
284         struct recent_command *next;
285         char *buf;
286 };
287
288 /* Configured limits on output */
289 static unsigned long max_depth = 10;
290 static off_t max_packsize;
291 static uintmax_t big_file_threshold = 512 * 1024 * 1024;
292 static int force_update;
293 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
294 static int pack_compression_seen;
295
296 /* Stats and misc. counters */
297 static uintmax_t alloc_count;
298 static uintmax_t marks_set_count;
299 static uintmax_t object_count_by_type[1 << TYPE_BITS];
300 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
301 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
302 static unsigned long object_count;
303 static unsigned long branch_count;
304 static unsigned long branch_load_count;
305 static int failure;
306 static FILE *pack_edges;
307 static unsigned int show_stats = 1;
308 static int global_argc;
309 static const char **global_argv;
310
311 /* Memory pools */
312 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
313 static size_t total_allocd;
314 static struct mem_pool *mem_pool;
315
316 /* Atom management */
317 static unsigned int atom_table_sz = 4451;
318 static unsigned int atom_cnt;
319 static struct atom_str **atom_table;
320
321 /* The .pack file being generated */
322 static unsigned int pack_id;
323 static struct sha1file *pack_file;
324 static struct packed_git *pack_data;
325 static struct packed_git **all_packs;
326 static off_t pack_size;
327
328 /* Table of objects we've written. */
329 static unsigned int object_entry_alloc = 5000;
330 static struct object_entry_pool *blocks;
331 static struct object_entry *object_table[1 << 16];
332 static struct mark_set *marks;
333 static const char *export_marks_file;
334 static const char *import_marks_file;
335 static int import_marks_file_from_stream;
336 static int import_marks_file_ignore_missing;
337 static int relative_marks_paths;
338
339 /* Our last blob */
340 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
341
342 /* Tree management */
343 static unsigned int tree_entry_alloc = 1000;
344 static void *avail_tree_entry;
345 static unsigned int avail_tree_table_sz = 100;
346 static struct avail_tree_content **avail_tree_table;
347 static struct strbuf old_tree = STRBUF_INIT;
348 static struct strbuf new_tree = STRBUF_INIT;
349
350 /* Branch data */
351 static unsigned long max_active_branches = 5;
352 static unsigned long cur_active_branches;
353 static unsigned long branch_table_sz = 1039;
354 static struct branch **branch_table;
355 static struct branch *active_branches;
356
357 /* Tag data */
358 static struct tag *first_tag;
359 static struct tag *last_tag;
360
361 /* Input stream parsing */
362 static whenspec_type whenspec = WHENSPEC_RAW;
363 static struct strbuf command_buf = STRBUF_INIT;
364 static int unread_command_buf;
365 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
366 static struct recent_command *cmd_tail = &cmd_hist;
367 static struct recent_command *rc_free;
368 static unsigned int cmd_save = 100;
369 static uintmax_t next_mark;
370 static struct strbuf new_data = STRBUF_INIT;
371 static int seen_data_command;
372
373 /* Signal handling */
374 static volatile sig_atomic_t checkpoint_requested;
375
376 /* Where to write output of cat-blob commands */
377 static int cat_blob_fd = STDOUT_FILENO;
378
379 static void parse_argv(void);
380 static void parse_cat_blob(void);
381 static void parse_ls(struct branch *b);
382
383 static void write_branch_report(FILE *rpt, struct branch *b)
384 {
385         fprintf(rpt, "%s:\n", b->name);
386
387         fprintf(rpt, "  status      :");
388         if (b->active)
389                 fputs(" active", rpt);
390         if (b->branch_tree.tree)
391                 fputs(" loaded", rpt);
392         if (is_null_sha1(b->branch_tree.versions[1].sha1))
393                 fputs(" dirty", rpt);
394         fputc('\n', rpt);
395
396         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
397         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
398         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
399         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
400
401         fputs("  last pack   : ", rpt);
402         if (b->pack_id < MAX_PACK_ID)
403                 fprintf(rpt, "%u", b->pack_id);
404         fputc('\n', rpt);
405
406         fputc('\n', rpt);
407 }
408
409 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
410
411 static void write_crash_report(const char *err)
412 {
413         char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
414         FILE *rpt = fopen(loc, "w");
415         struct branch *b;
416         unsigned long lu;
417         struct recent_command *rc;
418
419         if (!rpt) {
420                 error("can't write crash report %s: %s", loc, strerror(errno));
421                 return;
422         }
423
424         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
425
426         fprintf(rpt, "fast-import crash report:\n");
427         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
428         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
429         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
430         fputc('\n', rpt);
431
432         fputs("fatal: ", rpt);
433         fputs(err, rpt);
434         fputc('\n', rpt);
435
436         fputc('\n', rpt);
437         fputs("Most Recent Commands Before Crash\n", rpt);
438         fputs("---------------------------------\n", rpt);
439         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
440                 if (rc->next == &cmd_hist)
441                         fputs("* ", rpt);
442                 else
443                         fputs("  ", rpt);
444                 fputs(rc->buf, rpt);
445                 fputc('\n', rpt);
446         }
447
448         fputc('\n', rpt);
449         fputs("Active Branch LRU\n", rpt);
450         fputs("-----------------\n", rpt);
451         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
452                 cur_active_branches,
453                 max_active_branches);
454         fputc('\n', rpt);
455         fputs("  pos  clock name\n", rpt);
456         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
457         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
458                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
459                         ++lu, b->last_commit, b->name);
460
461         fputc('\n', rpt);
462         fputs("Inactive Branches\n", rpt);
463         fputs("-----------------\n", rpt);
464         for (lu = 0; lu < branch_table_sz; lu++) {
465                 for (b = branch_table[lu]; b; b = b->table_next_branch)
466                         write_branch_report(rpt, b);
467         }
468
469         if (first_tag) {
470                 struct tag *tg;
471                 fputc('\n', rpt);
472                 fputs("Annotated Tags\n", rpt);
473                 fputs("--------------\n", rpt);
474                 for (tg = first_tag; tg; tg = tg->next_tag) {
475                         fputs(sha1_to_hex(tg->sha1), rpt);
476                         fputc(' ', rpt);
477                         fputs(tg->name, rpt);
478                         fputc('\n', rpt);
479                 }
480         }
481
482         fputc('\n', rpt);
483         fputs("Marks\n", rpt);
484         fputs("-----\n", rpt);
485         if (export_marks_file)
486                 fprintf(rpt, "  exported to %s\n", export_marks_file);
487         else
488                 dump_marks_helper(rpt, 0, marks);
489
490         fputc('\n', rpt);
491         fputs("-------------------\n", rpt);
492         fputs("END OF CRASH REPORT\n", rpt);
493         fclose(rpt);
494 }
495
496 static void end_packfile(void);
497 static void unkeep_all_packs(void);
498 static void dump_marks(void);
499
500 static NORETURN void die_nicely(const char *err, va_list params)
501 {
502         static int zombie;
503         char message[2 * PATH_MAX];
504
505         vsnprintf(message, sizeof(message), err, params);
506         fputs("fatal: ", stderr);
507         fputs(message, stderr);
508         fputc('\n', stderr);
509
510         if (!zombie) {
511                 zombie = 1;
512                 write_crash_report(message);
513                 end_packfile();
514                 unkeep_all_packs();
515                 dump_marks();
516         }
517         exit(128);
518 }
519
520 #ifndef SIGUSR1 /* Windows, for example */
521
522 static void set_checkpoint_signal(void)
523 {
524 }
525
526 #else
527
528 static void checkpoint_signal(int signo)
529 {
530         checkpoint_requested = 1;
531 }
532
533 static void set_checkpoint_signal(void)
534 {
535         struct sigaction sa;
536
537         memset(&sa, 0, sizeof(sa));
538         sa.sa_handler = checkpoint_signal;
539         sigemptyset(&sa.sa_mask);
540         sa.sa_flags = SA_RESTART;
541         sigaction(SIGUSR1, &sa, NULL);
542 }
543
544 #endif
545
546 static void alloc_objects(unsigned int cnt)
547 {
548         struct object_entry_pool *b;
549
550         b = xmalloc(sizeof(struct object_entry_pool)
551                 + cnt * sizeof(struct object_entry));
552         b->next_pool = blocks;
553         b->next_free = b->entries;
554         b->end = b->entries + cnt;
555         blocks = b;
556         alloc_count += cnt;
557 }
558
559 static struct object_entry *new_object(unsigned char *sha1)
560 {
561         struct object_entry *e;
562
563         if (blocks->next_free == blocks->end)
564                 alloc_objects(object_entry_alloc);
565
566         e = blocks->next_free++;
567         hashcpy(e->idx.sha1, sha1);
568         return e;
569 }
570
571 static struct object_entry *find_object(unsigned char *sha1)
572 {
573         unsigned int h = sha1[0] << 8 | sha1[1];
574         struct object_entry *e;
575         for (e = object_table[h]; e; e = e->next)
576                 if (!hashcmp(sha1, e->idx.sha1))
577                         return e;
578         return NULL;
579 }
580
581 static struct object_entry *insert_object(unsigned char *sha1)
582 {
583         unsigned int h = sha1[0] << 8 | sha1[1];
584         struct object_entry *e = object_table[h];
585
586         while (e) {
587                 if (!hashcmp(sha1, e->idx.sha1))
588                         return e;
589                 e = e->next;
590         }
591
592         e = new_object(sha1);
593         e->next = object_table[h];
594         e->idx.offset = 0;
595         object_table[h] = e;
596         return e;
597 }
598
599 static unsigned int hc_str(const char *s, size_t len)
600 {
601         unsigned int r = 0;
602         while (len-- > 0)
603                 r = r * 31 + *s++;
604         return r;
605 }
606
607 static void *pool_alloc(size_t len)
608 {
609         struct mem_pool *p;
610         void *r;
611
612         /* round up to a 'uintmax_t' alignment */
613         if (len & (sizeof(uintmax_t) - 1))
614                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
615
616         for (p = mem_pool; p; p = p->next_pool)
617                 if ((p->end - p->next_free >= len))
618                         break;
619
620         if (!p) {
621                 if (len >= (mem_pool_alloc/2)) {
622                         total_allocd += len;
623                         return xmalloc(len);
624                 }
625                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
626                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
627                 p->next_pool = mem_pool;
628                 p->next_free = (char *) p->space;
629                 p->end = p->next_free + mem_pool_alloc;
630                 mem_pool = p;
631         }
632
633         r = p->next_free;
634         p->next_free += len;
635         return r;
636 }
637
638 static void *pool_calloc(size_t count, size_t size)
639 {
640         size_t len = count * size;
641         void *r = pool_alloc(len);
642         memset(r, 0, len);
643         return r;
644 }
645
646 static char *pool_strdup(const char *s)
647 {
648         char *r = pool_alloc(strlen(s) + 1);
649         strcpy(r, s);
650         return r;
651 }
652
653 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
654 {
655         struct mark_set *s = marks;
656         while ((idnum >> s->shift) >= 1024) {
657                 s = pool_calloc(1, sizeof(struct mark_set));
658                 s->shift = marks->shift + 10;
659                 s->data.sets[0] = marks;
660                 marks = s;
661         }
662         while (s->shift) {
663                 uintmax_t i = idnum >> s->shift;
664                 idnum -= i << s->shift;
665                 if (!s->data.sets[i]) {
666                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
667                         s->data.sets[i]->shift = s->shift - 10;
668                 }
669                 s = s->data.sets[i];
670         }
671         if (!s->data.marked[idnum])
672                 marks_set_count++;
673         s->data.marked[idnum] = oe;
674 }
675
676 static struct object_entry *find_mark(uintmax_t idnum)
677 {
678         uintmax_t orig_idnum = idnum;
679         struct mark_set *s = marks;
680         struct object_entry *oe = NULL;
681         if ((idnum >> s->shift) < 1024) {
682                 while (s && s->shift) {
683                         uintmax_t i = idnum >> s->shift;
684                         idnum -= i << s->shift;
685                         s = s->data.sets[i];
686                 }
687                 if (s)
688                         oe = s->data.marked[idnum];
689         }
690         if (!oe)
691                 die("mark :%" PRIuMAX " not declared", orig_idnum);
692         return oe;
693 }
694
695 static struct atom_str *to_atom(const char *s, unsigned short len)
696 {
697         unsigned int hc = hc_str(s, len) % atom_table_sz;
698         struct atom_str *c;
699
700         for (c = atom_table[hc]; c; c = c->next_atom)
701                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
702                         return c;
703
704         c = pool_alloc(sizeof(struct atom_str) + len + 1);
705         c->str_len = len;
706         strncpy(c->str_dat, s, len);
707         c->str_dat[len] = 0;
708         c->next_atom = atom_table[hc];
709         atom_table[hc] = c;
710         atom_cnt++;
711         return c;
712 }
713
714 static struct branch *lookup_branch(const char *name)
715 {
716         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
717         struct branch *b;
718
719         for (b = branch_table[hc]; b; b = b->table_next_branch)
720                 if (!strcmp(name, b->name))
721                         return b;
722         return NULL;
723 }
724
725 static struct branch *new_branch(const char *name)
726 {
727         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
728         struct branch *b = lookup_branch(name);
729
730         if (b)
731                 die("Invalid attempt to create duplicate branch: %s", name);
732         switch (check_ref_format(name)) {
733         case 0: break; /* its valid */
734         case CHECK_REF_FORMAT_ONELEVEL:
735                 break; /* valid, but too few '/', allow anyway */
736         default:
737                 die("Branch name doesn't conform to GIT standards: %s", name);
738         }
739
740         b = pool_calloc(1, sizeof(struct branch));
741         b->name = pool_strdup(name);
742         b->table_next_branch = branch_table[hc];
743         b->branch_tree.versions[0].mode = S_IFDIR;
744         b->branch_tree.versions[1].mode = S_IFDIR;
745         b->num_notes = 0;
746         b->active = 0;
747         b->pack_id = MAX_PACK_ID;
748         branch_table[hc] = b;
749         branch_count++;
750         return b;
751 }
752
753 static unsigned int hc_entries(unsigned int cnt)
754 {
755         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
756         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
757 }
758
759 static struct tree_content *new_tree_content(unsigned int cnt)
760 {
761         struct avail_tree_content *f, *l = NULL;
762         struct tree_content *t;
763         unsigned int hc = hc_entries(cnt);
764
765         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
766                 if (f->entry_capacity >= cnt)
767                         break;
768
769         if (f) {
770                 if (l)
771                         l->next_avail = f->next_avail;
772                 else
773                         avail_tree_table[hc] = f->next_avail;
774         } else {
775                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
776                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
777                 f->entry_capacity = cnt;
778         }
779
780         t = (struct tree_content*)f;
781         t->entry_count = 0;
782         t->delta_depth = 0;
783         return t;
784 }
785
786 static void release_tree_entry(struct tree_entry *e);
787 static void release_tree_content(struct tree_content *t)
788 {
789         struct avail_tree_content *f = (struct avail_tree_content*)t;
790         unsigned int hc = hc_entries(f->entry_capacity);
791         f->next_avail = avail_tree_table[hc];
792         avail_tree_table[hc] = f;
793 }
794
795 static void release_tree_content_recursive(struct tree_content *t)
796 {
797         unsigned int i;
798         for (i = 0; i < t->entry_count; i++)
799                 release_tree_entry(t->entries[i]);
800         release_tree_content(t);
801 }
802
803 static struct tree_content *grow_tree_content(
804         struct tree_content *t,
805         int amt)
806 {
807         struct tree_content *r = new_tree_content(t->entry_count + amt);
808         r->entry_count = t->entry_count;
809         r->delta_depth = t->delta_depth;
810         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
811         release_tree_content(t);
812         return r;
813 }
814
815 static struct tree_entry *new_tree_entry(void)
816 {
817         struct tree_entry *e;
818
819         if (!avail_tree_entry) {
820                 unsigned int n = tree_entry_alloc;
821                 total_allocd += n * sizeof(struct tree_entry);
822                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
823                 while (n-- > 1) {
824                         *((void**)e) = e + 1;
825                         e++;
826                 }
827                 *((void**)e) = NULL;
828         }
829
830         e = avail_tree_entry;
831         avail_tree_entry = *((void**)e);
832         return e;
833 }
834
835 static void release_tree_entry(struct tree_entry *e)
836 {
837         if (e->tree)
838                 release_tree_content_recursive(e->tree);
839         *((void**)e) = avail_tree_entry;
840         avail_tree_entry = e;
841 }
842
843 static struct tree_content *dup_tree_content(struct tree_content *s)
844 {
845         struct tree_content *d;
846         struct tree_entry *a, *b;
847         unsigned int i;
848
849         if (!s)
850                 return NULL;
851         d = new_tree_content(s->entry_count);
852         for (i = 0; i < s->entry_count; i++) {
853                 a = s->entries[i];
854                 b = new_tree_entry();
855                 memcpy(b, a, sizeof(*a));
856                 if (a->tree && is_null_sha1(b->versions[1].sha1))
857                         b->tree = dup_tree_content(a->tree);
858                 else
859                         b->tree = NULL;
860                 d->entries[i] = b;
861         }
862         d->entry_count = s->entry_count;
863         d->delta_depth = s->delta_depth;
864
865         return d;
866 }
867
868 static void start_packfile(void)
869 {
870         static char tmpfile[PATH_MAX];
871         struct packed_git *p;
872         struct pack_header hdr;
873         int pack_fd;
874
875         pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
876                               "pack/tmp_pack_XXXXXX");
877         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
878         strcpy(p->pack_name, tmpfile);
879         p->pack_fd = pack_fd;
880         pack_file = sha1fd(pack_fd, p->pack_name);
881
882         hdr.hdr_signature = htonl(PACK_SIGNATURE);
883         hdr.hdr_version = htonl(2);
884         hdr.hdr_entries = 0;
885         sha1write(pack_file, &hdr, sizeof(hdr));
886
887         pack_data = p;
888         pack_size = sizeof(hdr);
889         object_count = 0;
890
891         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
892         all_packs[pack_id] = p;
893 }
894
895 static const char *create_index(void)
896 {
897         const char *tmpfile;
898         struct pack_idx_entry **idx, **c, **last;
899         struct object_entry *e;
900         struct object_entry_pool *o;
901
902         /* Build the table of object IDs. */
903         idx = xmalloc(object_count * sizeof(*idx));
904         c = idx;
905         for (o = blocks; o; o = o->next_pool)
906                 for (e = o->next_free; e-- != o->entries;)
907                         if (pack_id == e->pack_id)
908                                 *c++ = &e->idx;
909         last = idx + object_count;
910         if (c != last)
911                 die("internal consistency error creating the index");
912
913         tmpfile = write_idx_file(NULL, idx, object_count, pack_data->sha1);
914         free(idx);
915         return tmpfile;
916 }
917
918 static char *keep_pack(const char *curr_index_name)
919 {
920         static char name[PATH_MAX];
921         static const char *keep_msg = "fast-import";
922         int keep_fd;
923
924         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
925         if (keep_fd < 0)
926                 die_errno("cannot create keep file");
927         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
928         if (close(keep_fd))
929                 die_errno("failed to write keep file");
930
931         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
932                  get_object_directory(), sha1_to_hex(pack_data->sha1));
933         if (move_temp_to_file(pack_data->pack_name, name))
934                 die("cannot store pack file");
935
936         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
937                  get_object_directory(), sha1_to_hex(pack_data->sha1));
938         if (move_temp_to_file(curr_index_name, name))
939                 die("cannot store index file");
940         free((void *)curr_index_name);
941         return name;
942 }
943
944 static void unkeep_all_packs(void)
945 {
946         static char name[PATH_MAX];
947         int k;
948
949         for (k = 0; k < pack_id; k++) {
950                 struct packed_git *p = all_packs[k];
951                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
952                          get_object_directory(), sha1_to_hex(p->sha1));
953                 unlink_or_warn(name);
954         }
955 }
956
957 static void end_packfile(void)
958 {
959         struct packed_git *old_p = pack_data, *new_p;
960
961         clear_delta_base_cache();
962         if (object_count) {
963                 unsigned char cur_pack_sha1[20];
964                 char *idx_name;
965                 int i;
966                 struct branch *b;
967                 struct tag *t;
968
969                 close_pack_windows(pack_data);
970                 sha1close(pack_file, cur_pack_sha1, 0);
971                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
972                                     pack_data->pack_name, object_count,
973                                     cur_pack_sha1, pack_size);
974                 close(pack_data->pack_fd);
975                 idx_name = keep_pack(create_index());
976
977                 /* Register the packfile with core git's machinery. */
978                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
979                 if (!new_p)
980                         die("core git rejected index %s", idx_name);
981                 all_packs[pack_id] = new_p;
982                 install_packed_git(new_p);
983
984                 /* Print the boundary */
985                 if (pack_edges) {
986                         fprintf(pack_edges, "%s:", new_p->pack_name);
987                         for (i = 0; i < branch_table_sz; i++) {
988                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
989                                         if (b->pack_id == pack_id)
990                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
991                                 }
992                         }
993                         for (t = first_tag; t; t = t->next_tag) {
994                                 if (t->pack_id == pack_id)
995                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
996                         }
997                         fputc('\n', pack_edges);
998                         fflush(pack_edges);
999                 }
1000
1001                 pack_id++;
1002         }
1003         else {
1004                 close(old_p->pack_fd);
1005                 unlink_or_warn(old_p->pack_name);
1006         }
1007         free(old_p);
1008
1009         /* We can't carry a delta across packfiles. */
1010         strbuf_release(&last_blob.data);
1011         last_blob.offset = 0;
1012         last_blob.depth = 0;
1013 }
1014
1015 static void cycle_packfile(void)
1016 {
1017         end_packfile();
1018         start_packfile();
1019 }
1020
1021 static int store_object(
1022         enum object_type type,
1023         struct strbuf *dat,
1024         struct last_object *last,
1025         unsigned char *sha1out,
1026         uintmax_t mark)
1027 {
1028         void *out, *delta;
1029         struct object_entry *e;
1030         unsigned char hdr[96];
1031         unsigned char sha1[20];
1032         unsigned long hdrlen, deltalen;
1033         git_SHA_CTX c;
1034         z_stream s;
1035
1036         hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1037                 (unsigned long)dat->len) + 1;
1038         git_SHA1_Init(&c);
1039         git_SHA1_Update(&c, hdr, hdrlen);
1040         git_SHA1_Update(&c, dat->buf, dat->len);
1041         git_SHA1_Final(sha1, &c);
1042         if (sha1out)
1043                 hashcpy(sha1out, sha1);
1044
1045         e = insert_object(sha1);
1046         if (mark)
1047                 insert_mark(mark, e);
1048         if (e->idx.offset) {
1049                 duplicate_count_by_type[type]++;
1050                 return 1;
1051         } else if (find_sha1_pack(sha1, packed_git)) {
1052                 e->type = type;
1053                 e->pack_id = MAX_PACK_ID;
1054                 e->idx.offset = 1; /* just not zero! */
1055                 duplicate_count_by_type[type]++;
1056                 return 1;
1057         }
1058
1059         if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1060                 delta = diff_delta(last->data.buf, last->data.len,
1061                         dat->buf, dat->len,
1062                         &deltalen, dat->len - 20);
1063         } else
1064                 delta = NULL;
1065
1066         memset(&s, 0, sizeof(s));
1067         deflateInit(&s, pack_compression_level);
1068         if (delta) {
1069                 s.next_in = delta;
1070                 s.avail_in = deltalen;
1071         } else {
1072                 s.next_in = (void *)dat->buf;
1073                 s.avail_in = dat->len;
1074         }
1075         s.avail_out = deflateBound(&s, s.avail_in);
1076         s.next_out = out = xmalloc(s.avail_out);
1077         while (deflate(&s, Z_FINISH) == Z_OK)
1078                 /* nothing */;
1079         deflateEnd(&s);
1080
1081         /* Determine if we should auto-checkpoint. */
1082         if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1083                 || (pack_size + 60 + s.total_out) < pack_size) {
1084
1085                 /* This new object needs to *not* have the current pack_id. */
1086                 e->pack_id = pack_id + 1;
1087                 cycle_packfile();
1088
1089                 /* We cannot carry a delta into the new pack. */
1090                 if (delta) {
1091                         free(delta);
1092                         delta = NULL;
1093
1094                         memset(&s, 0, sizeof(s));
1095                         deflateInit(&s, pack_compression_level);
1096                         s.next_in = (void *)dat->buf;
1097                         s.avail_in = dat->len;
1098                         s.avail_out = deflateBound(&s, s.avail_in);
1099                         s.next_out = out = xrealloc(out, s.avail_out);
1100                         while (deflate(&s, Z_FINISH) == Z_OK)
1101                                 /* nothing */;
1102                         deflateEnd(&s);
1103                 }
1104         }
1105
1106         e->type = type;
1107         e->pack_id = pack_id;
1108         e->idx.offset = pack_size;
1109         object_count++;
1110         object_count_by_type[type]++;
1111
1112         crc32_begin(pack_file);
1113
1114         if (delta) {
1115                 off_t ofs = e->idx.offset - last->offset;
1116                 unsigned pos = sizeof(hdr) - 1;
1117
1118                 delta_count_by_type[type]++;
1119                 e->depth = last->depth + 1;
1120
1121                 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1122                 sha1write(pack_file, hdr, hdrlen);
1123                 pack_size += hdrlen;
1124
1125                 hdr[pos] = ofs & 127;
1126                 while (ofs >>= 7)
1127                         hdr[--pos] = 128 | (--ofs & 127);
1128                 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1129                 pack_size += sizeof(hdr) - pos;
1130         } else {
1131                 e->depth = 0;
1132                 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1133                 sha1write(pack_file, hdr, hdrlen);
1134                 pack_size += hdrlen;
1135         }
1136
1137         sha1write(pack_file, out, s.total_out);
1138         pack_size += s.total_out;
1139
1140         e->idx.crc32 = crc32_end(pack_file);
1141
1142         free(out);
1143         free(delta);
1144         if (last) {
1145                 if (last->no_swap) {
1146                         last->data = *dat;
1147                 } else {
1148                         strbuf_swap(&last->data, dat);
1149                 }
1150                 last->offset = e->idx.offset;
1151                 last->depth = e->depth;
1152         }
1153         return 0;
1154 }
1155
1156 static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1157 {
1158         if (ftruncate(pack_data->pack_fd, to)
1159          || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1160                 die_errno("cannot truncate pack to skip duplicate");
1161         pack_size = to;
1162
1163         /* yes this is a layering violation */
1164         pack_file->total = to;
1165         pack_file->offset = 0;
1166         pack_file->ctx = *ctx;
1167 }
1168
1169 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1170 {
1171         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1172         unsigned char *in_buf = xmalloc(in_sz);
1173         unsigned char *out_buf = xmalloc(out_sz);
1174         struct object_entry *e;
1175         unsigned char sha1[20];
1176         unsigned long hdrlen;
1177         off_t offset;
1178         git_SHA_CTX c;
1179         git_SHA_CTX pack_file_ctx;
1180         z_stream s;
1181         int status = Z_OK;
1182
1183         /* Determine if we should auto-checkpoint. */
1184         if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1185                 || (pack_size + 60 + len) < pack_size)
1186                 cycle_packfile();
1187
1188         offset = pack_size;
1189
1190         /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1191         sha1flush(pack_file);
1192         pack_file_ctx = pack_file->ctx;
1193
1194         hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1195         if (out_sz <= hdrlen)
1196                 die("impossibly large object header");
1197
1198         git_SHA1_Init(&c);
1199         git_SHA1_Update(&c, out_buf, hdrlen);
1200
1201         crc32_begin(pack_file);
1202
1203         memset(&s, 0, sizeof(s));
1204         deflateInit(&s, pack_compression_level);
1205
1206         hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1207         if (out_sz <= hdrlen)
1208                 die("impossibly large object header");
1209
1210         s.next_out = out_buf + hdrlen;
1211         s.avail_out = out_sz - hdrlen;
1212
1213         while (status != Z_STREAM_END) {
1214                 if (0 < len && !s.avail_in) {
1215                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1216                         size_t n = fread(in_buf, 1, cnt, stdin);
1217                         if (!n && feof(stdin))
1218                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1219
1220                         git_SHA1_Update(&c, in_buf, n);
1221                         s.next_in = in_buf;
1222                         s.avail_in = n;
1223                         len -= n;
1224                 }
1225
1226                 status = deflate(&s, len ? 0 : Z_FINISH);
1227
1228                 if (!s.avail_out || status == Z_STREAM_END) {
1229                         size_t n = s.next_out - out_buf;
1230                         sha1write(pack_file, out_buf, n);
1231                         pack_size += n;
1232                         s.next_out = out_buf;
1233                         s.avail_out = out_sz;
1234                 }
1235
1236                 switch (status) {
1237                 case Z_OK:
1238                 case Z_BUF_ERROR:
1239                 case Z_STREAM_END:
1240                         continue;
1241                 default:
1242                         die("unexpected deflate failure: %d", status);
1243                 }
1244         }
1245         deflateEnd(&s);
1246         git_SHA1_Final(sha1, &c);
1247
1248         if (sha1out)
1249                 hashcpy(sha1out, sha1);
1250
1251         e = insert_object(sha1);
1252
1253         if (mark)
1254                 insert_mark(mark, e);
1255
1256         if (e->idx.offset) {
1257                 duplicate_count_by_type[OBJ_BLOB]++;
1258                 truncate_pack(offset, &pack_file_ctx);
1259
1260         } else if (find_sha1_pack(sha1, packed_git)) {
1261                 e->type = OBJ_BLOB;
1262                 e->pack_id = MAX_PACK_ID;
1263                 e->idx.offset = 1; /* just not zero! */
1264                 duplicate_count_by_type[OBJ_BLOB]++;
1265                 truncate_pack(offset, &pack_file_ctx);
1266
1267         } else {
1268                 e->depth = 0;
1269                 e->type = OBJ_BLOB;
1270                 e->pack_id = pack_id;
1271                 e->idx.offset = offset;
1272                 e->idx.crc32 = crc32_end(pack_file);
1273                 object_count++;
1274                 object_count_by_type[OBJ_BLOB]++;
1275         }
1276
1277         free(in_buf);
1278         free(out_buf);
1279 }
1280
1281 /* All calls must be guarded by find_object() or find_mark() to
1282  * ensure the 'struct object_entry' passed was written by this
1283  * process instance.  We unpack the entry by the offset, avoiding
1284  * the need for the corresponding .idx file.  This unpacking rule
1285  * works because we only use OBJ_REF_DELTA within the packfiles
1286  * created by fast-import.
1287  *
1288  * oe must not be NULL.  Such an oe usually comes from giving
1289  * an unknown SHA-1 to find_object() or an undefined mark to
1290  * find_mark().  Callers must test for this condition and use
1291  * the standard read_sha1_file() when it happens.
1292  *
1293  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1294  * find_mark(), where the mark was reloaded from an existing marks
1295  * file and is referencing an object that this fast-import process
1296  * instance did not write out to a packfile.  Callers must test for
1297  * this condition and use read_sha1_file() instead.
1298  */
1299 static void *gfi_unpack_entry(
1300         struct object_entry *oe,
1301         unsigned long *sizep)
1302 {
1303         enum object_type type;
1304         struct packed_git *p = all_packs[oe->pack_id];
1305         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1306                 /* The object is stored in the packfile we are writing to
1307                  * and we have modified it since the last time we scanned
1308                  * back to read a previously written object.  If an old
1309                  * window covered [p->pack_size, p->pack_size + 20) its
1310                  * data is stale and is not valid.  Closing all windows
1311                  * and updating the packfile length ensures we can read
1312                  * the newly written data.
1313                  */
1314                 close_pack_windows(p);
1315                 sha1flush(pack_file);
1316
1317                 /* We have to offer 20 bytes additional on the end of
1318                  * the packfile as the core unpacker code assumes the
1319                  * footer is present at the file end and must promise
1320                  * at least 20 bytes within any window it maps.  But
1321                  * we don't actually create the footer here.
1322                  */
1323                 p->pack_size = pack_size + 20;
1324         }
1325         return unpack_entry(p, oe->idx.offset, &type, sizep);
1326 }
1327
1328 static const char *get_mode(const char *str, uint16_t *modep)
1329 {
1330         unsigned char c;
1331         uint16_t mode = 0;
1332
1333         while ((c = *str++) != ' ') {
1334                 if (c < '0' || c > '7')
1335                         return NULL;
1336                 mode = (mode << 3) + (c - '0');
1337         }
1338         *modep = mode;
1339         return str;
1340 }
1341
1342 static void load_tree(struct tree_entry *root)
1343 {
1344         unsigned char *sha1 = root->versions[1].sha1;
1345         struct object_entry *myoe;
1346         struct tree_content *t;
1347         unsigned long size;
1348         char *buf;
1349         const char *c;
1350
1351         root->tree = t = new_tree_content(8);
1352         if (is_null_sha1(sha1))
1353                 return;
1354
1355         myoe = find_object(sha1);
1356         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1357                 if (myoe->type != OBJ_TREE)
1358                         die("Not a tree: %s", sha1_to_hex(sha1));
1359                 t->delta_depth = myoe->depth;
1360                 buf = gfi_unpack_entry(myoe, &size);
1361                 if (!buf)
1362                         die("Can't load tree %s", sha1_to_hex(sha1));
1363         } else {
1364                 enum object_type type;
1365                 buf = read_sha1_file(sha1, &type, &size);
1366                 if (!buf || type != OBJ_TREE)
1367                         die("Can't load tree %s", sha1_to_hex(sha1));
1368         }
1369
1370         c = buf;
1371         while (c != (buf + size)) {
1372                 struct tree_entry *e = new_tree_entry();
1373
1374                 if (t->entry_count == t->entry_capacity)
1375                         root->tree = t = grow_tree_content(t, t->entry_count);
1376                 t->entries[t->entry_count++] = e;
1377
1378                 e->tree = NULL;
1379                 c = get_mode(c, &e->versions[1].mode);
1380                 if (!c)
1381                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1382                 e->versions[0].mode = e->versions[1].mode;
1383                 e->name = to_atom(c, strlen(c));
1384                 c += e->name->str_len + 1;
1385                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1386                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1387                 c += 20;
1388         }
1389         free(buf);
1390 }
1391
1392 static int tecmp0 (const void *_a, const void *_b)
1393 {
1394         struct tree_entry *a = *((struct tree_entry**)_a);
1395         struct tree_entry *b = *((struct tree_entry**)_b);
1396         return base_name_compare(
1397                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1398                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1399 }
1400
1401 static int tecmp1 (const void *_a, const void *_b)
1402 {
1403         struct tree_entry *a = *((struct tree_entry**)_a);
1404         struct tree_entry *b = *((struct tree_entry**)_b);
1405         return base_name_compare(
1406                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1407                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1408 }
1409
1410 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1411 {
1412         size_t maxlen = 0;
1413         unsigned int i;
1414
1415         if (!v)
1416                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1417         else
1418                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1419
1420         for (i = 0; i < t->entry_count; i++) {
1421                 if (t->entries[i]->versions[v].mode)
1422                         maxlen += t->entries[i]->name->str_len + 34;
1423         }
1424
1425         strbuf_reset(b);
1426         strbuf_grow(b, maxlen);
1427         for (i = 0; i < t->entry_count; i++) {
1428                 struct tree_entry *e = t->entries[i];
1429                 if (!e->versions[v].mode)
1430                         continue;
1431                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1432                                         e->name->str_dat, '\0');
1433                 strbuf_add(b, e->versions[v].sha1, 20);
1434         }
1435 }
1436
1437 static void store_tree(struct tree_entry *root)
1438 {
1439         struct tree_content *t = root->tree;
1440         unsigned int i, j, del;
1441         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1442         struct object_entry *le;
1443
1444         if (!is_null_sha1(root->versions[1].sha1))
1445                 return;
1446
1447         for (i = 0; i < t->entry_count; i++) {
1448                 if (t->entries[i]->tree)
1449                         store_tree(t->entries[i]);
1450         }
1451
1452         le = find_object(root->versions[0].sha1);
1453         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1454                 mktree(t, 0, &old_tree);
1455                 lo.data = old_tree;
1456                 lo.offset = le->idx.offset;
1457                 lo.depth = t->delta_depth;
1458         }
1459
1460         mktree(t, 1, &new_tree);
1461         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1462
1463         t->delta_depth = lo.depth;
1464         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1465                 struct tree_entry *e = t->entries[i];
1466                 if (e->versions[1].mode) {
1467                         e->versions[0].mode = e->versions[1].mode;
1468                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1469                         t->entries[j++] = e;
1470                 } else {
1471                         release_tree_entry(e);
1472                         del++;
1473                 }
1474         }
1475         t->entry_count -= del;
1476 }
1477
1478 static void tree_content_replace(
1479         struct tree_entry *root,
1480         const unsigned char *sha1,
1481         const uint16_t mode,
1482         struct tree_content *newtree)
1483 {
1484         if (!S_ISDIR(mode))
1485                 die("Root cannot be a non-directory");
1486         hashcpy(root->versions[1].sha1, sha1);
1487         if (root->tree)
1488                 release_tree_content_recursive(root->tree);
1489         root->tree = newtree;
1490 }
1491
1492 static int tree_content_set(
1493         struct tree_entry *root,
1494         const char *p,
1495         const unsigned char *sha1,
1496         const uint16_t mode,
1497         struct tree_content *subtree)
1498 {
1499         struct tree_content *t;
1500         const char *slash1;
1501         unsigned int i, n;
1502         struct tree_entry *e;
1503
1504         slash1 = strchr(p, '/');
1505         if (slash1)
1506                 n = slash1 - p;
1507         else
1508                 n = strlen(p);
1509         if (!n)
1510                 die("Empty path component found in input");
1511         if (!slash1 && !S_ISDIR(mode) && subtree)
1512                 die("Non-directories cannot have subtrees");
1513
1514         if (!root->tree)
1515                 load_tree(root);
1516         t = root->tree;
1517         for (i = 0; i < t->entry_count; i++) {
1518                 e = t->entries[i];
1519                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1520                         if (!slash1) {
1521                                 if (!S_ISDIR(mode)
1522                                                 && e->versions[1].mode == mode
1523                                                 && !hashcmp(e->versions[1].sha1, sha1))
1524                                         return 0;
1525                                 e->versions[1].mode = mode;
1526                                 hashcpy(e->versions[1].sha1, sha1);
1527                                 if (e->tree)
1528                                         release_tree_content_recursive(e->tree);
1529                                 e->tree = subtree;
1530                                 hashclr(root->versions[1].sha1);
1531                                 return 1;
1532                         }
1533                         if (!S_ISDIR(e->versions[1].mode)) {
1534                                 e->tree = new_tree_content(8);
1535                                 e->versions[1].mode = S_IFDIR;
1536                         }
1537                         if (!e->tree)
1538                                 load_tree(e);
1539                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1540                                 hashclr(root->versions[1].sha1);
1541                                 return 1;
1542                         }
1543                         return 0;
1544                 }
1545         }
1546
1547         if (t->entry_count == t->entry_capacity)
1548                 root->tree = t = grow_tree_content(t, t->entry_count);
1549         e = new_tree_entry();
1550         e->name = to_atom(p, n);
1551         e->versions[0].mode = 0;
1552         hashclr(e->versions[0].sha1);
1553         t->entries[t->entry_count++] = e;
1554         if (slash1) {
1555                 e->tree = new_tree_content(8);
1556                 e->versions[1].mode = S_IFDIR;
1557                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1558         } else {
1559                 e->tree = subtree;
1560                 e->versions[1].mode = mode;
1561                 hashcpy(e->versions[1].sha1, sha1);
1562         }
1563         hashclr(root->versions[1].sha1);
1564         return 1;
1565 }
1566
1567 static int tree_content_remove(
1568         struct tree_entry *root,
1569         const char *p,
1570         struct tree_entry *backup_leaf)
1571 {
1572         struct tree_content *t;
1573         const char *slash1;
1574         unsigned int i, n;
1575         struct tree_entry *e;
1576
1577         slash1 = strchr(p, '/');
1578         if (slash1)
1579                 n = slash1 - p;
1580         else
1581                 n = strlen(p);
1582
1583         if (!root->tree)
1584                 load_tree(root);
1585         t = root->tree;
1586         for (i = 0; i < t->entry_count; i++) {
1587                 e = t->entries[i];
1588                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1589                         if (slash1 && !S_ISDIR(e->versions[1].mode))
1590                                 /*
1591                                  * If p names a file in some subdirectory, and a
1592                                  * file or symlink matching the name of the
1593                                  * parent directory of p exists, then p cannot
1594                                  * exist and need not be deleted.
1595                                  */
1596                                 return 1;
1597                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1598                                 goto del_entry;
1599                         if (!e->tree)
1600                                 load_tree(e);
1601                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1602                                 for (n = 0; n < e->tree->entry_count; n++) {
1603                                         if (e->tree->entries[n]->versions[1].mode) {
1604                                                 hashclr(root->versions[1].sha1);
1605                                                 return 1;
1606                                         }
1607                                 }
1608                                 backup_leaf = NULL;
1609                                 goto del_entry;
1610                         }
1611                         return 0;
1612                 }
1613         }
1614         return 0;
1615
1616 del_entry:
1617         if (backup_leaf)
1618                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1619         else if (e->tree)
1620                 release_tree_content_recursive(e->tree);
1621         e->tree = NULL;
1622         e->versions[1].mode = 0;
1623         hashclr(e->versions[1].sha1);
1624         hashclr(root->versions[1].sha1);
1625         return 1;
1626 }
1627
1628 static int tree_content_get(
1629         struct tree_entry *root,
1630         const char *p,
1631         struct tree_entry *leaf)
1632 {
1633         struct tree_content *t;
1634         const char *slash1;
1635         unsigned int i, n;
1636         struct tree_entry *e;
1637
1638         slash1 = strchr(p, '/');
1639         if (slash1)
1640                 n = slash1 - p;
1641         else
1642                 n = strlen(p);
1643
1644         if (!root->tree)
1645                 load_tree(root);
1646         t = root->tree;
1647         for (i = 0; i < t->entry_count; i++) {
1648                 e = t->entries[i];
1649                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1650                         if (!slash1) {
1651                                 memcpy(leaf, e, sizeof(*leaf));
1652                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1653                                         leaf->tree = dup_tree_content(e->tree);
1654                                 else
1655                                         leaf->tree = NULL;
1656                                 return 1;
1657                         }
1658                         if (!S_ISDIR(e->versions[1].mode))
1659                                 return 0;
1660                         if (!e->tree)
1661                                 load_tree(e);
1662                         return tree_content_get(e, slash1 + 1, leaf);
1663                 }
1664         }
1665         return 0;
1666 }
1667
1668 static int update_branch(struct branch *b)
1669 {
1670         static const char *msg = "fast-import";
1671         struct ref_lock *lock;
1672         unsigned char old_sha1[20];
1673
1674         if (is_null_sha1(b->sha1))
1675                 return 0;
1676         if (read_ref(b->name, old_sha1))
1677                 hashclr(old_sha1);
1678         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1679         if (!lock)
1680                 return error("Unable to lock %s", b->name);
1681         if (!force_update && !is_null_sha1(old_sha1)) {
1682                 struct commit *old_cmit, *new_cmit;
1683
1684                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1685                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1686                 if (!old_cmit || !new_cmit) {
1687                         unlock_ref(lock);
1688                         return error("Branch %s is missing commits.", b->name);
1689                 }
1690
1691                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1692                         unlock_ref(lock);
1693                         warning("Not updating %s"
1694                                 " (new tip %s does not contain %s)",
1695                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1696                         return -1;
1697                 }
1698         }
1699         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1700                 return error("Unable to update %s", b->name);
1701         return 0;
1702 }
1703
1704 static void dump_branches(void)
1705 {
1706         unsigned int i;
1707         struct branch *b;
1708
1709         for (i = 0; i < branch_table_sz; i++) {
1710                 for (b = branch_table[i]; b; b = b->table_next_branch)
1711                         failure |= update_branch(b);
1712         }
1713 }
1714
1715 static void dump_tags(void)
1716 {
1717         static const char *msg = "fast-import";
1718         struct tag *t;
1719         struct ref_lock *lock;
1720         char ref_name[PATH_MAX];
1721
1722         for (t = first_tag; t; t = t->next_tag) {
1723                 sprintf(ref_name, "tags/%s", t->name);
1724                 lock = lock_ref_sha1(ref_name, NULL);
1725                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1726                         failure |= error("Unable to update %s", ref_name);
1727         }
1728 }
1729
1730 static void dump_marks_helper(FILE *f,
1731         uintmax_t base,
1732         struct mark_set *m)
1733 {
1734         uintmax_t k;
1735         if (m->shift) {
1736                 for (k = 0; k < 1024; k++) {
1737                         if (m->data.sets[k])
1738                                 dump_marks_helper(f, base + (k << m->shift),
1739                                         m->data.sets[k]);
1740                 }
1741         } else {
1742                 for (k = 0; k < 1024; k++) {
1743                         if (m->data.marked[k])
1744                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1745                                         sha1_to_hex(m->data.marked[k]->idx.sha1));
1746                 }
1747         }
1748 }
1749
1750 static void dump_marks(void)
1751 {
1752         static struct lock_file mark_lock;
1753         int mark_fd;
1754         FILE *f;
1755
1756         if (!export_marks_file)
1757                 return;
1758
1759         mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1760         if (mark_fd < 0) {
1761                 failure |= error("Unable to write marks file %s: %s",
1762                         export_marks_file, strerror(errno));
1763                 return;
1764         }
1765
1766         f = fdopen(mark_fd, "w");
1767         if (!f) {
1768                 int saved_errno = errno;
1769                 rollback_lock_file(&mark_lock);
1770                 failure |= error("Unable to write marks file %s: %s",
1771                         export_marks_file, strerror(saved_errno));
1772                 return;
1773         }
1774
1775         /*
1776          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1777          * Assign -1 to the lock file descriptor so that commit_lock_file()
1778          * won't try to close() it.
1779          */
1780         mark_lock.fd = -1;
1781
1782         dump_marks_helper(f, 0, marks);
1783         if (ferror(f) || fclose(f)) {
1784                 int saved_errno = errno;
1785                 rollback_lock_file(&mark_lock);
1786                 failure |= error("Unable to write marks file %s: %s",
1787                         export_marks_file, strerror(saved_errno));
1788                 return;
1789         }
1790
1791         if (commit_lock_file(&mark_lock)) {
1792                 int saved_errno = errno;
1793                 rollback_lock_file(&mark_lock);
1794                 failure |= error("Unable to commit marks file %s: %s",
1795                         export_marks_file, strerror(saved_errno));
1796                 return;
1797         }
1798 }
1799
1800 static void read_marks(void)
1801 {
1802         char line[512];
1803         FILE *f = fopen(import_marks_file, "r");
1804         if (f)
1805                 ;
1806         else if (import_marks_file_ignore_missing && errno == ENOENT)
1807                 return; /* Marks file does not exist */
1808         else
1809                 die_errno("cannot read '%s'", import_marks_file);
1810         while (fgets(line, sizeof(line), f)) {
1811                 uintmax_t mark;
1812                 char *end;
1813                 unsigned char sha1[20];
1814                 struct object_entry *e;
1815
1816                 end = strchr(line, '\n');
1817                 if (line[0] != ':' || !end)
1818                         die("corrupt mark line: %s", line);
1819                 *end = 0;
1820                 mark = strtoumax(line + 1, &end, 10);
1821                 if (!mark || end == line + 1
1822                         || *end != ' ' || get_sha1(end + 1, sha1))
1823                         die("corrupt mark line: %s", line);
1824                 e = find_object(sha1);
1825                 if (!e) {
1826                         enum object_type type = sha1_object_info(sha1, NULL);
1827                         if (type < 0)
1828                                 die("object not found: %s", sha1_to_hex(sha1));
1829                         e = insert_object(sha1);
1830                         e->type = type;
1831                         e->pack_id = MAX_PACK_ID;
1832                         e->idx.offset = 1; /* just not zero! */
1833                 }
1834                 insert_mark(mark, e);
1835         }
1836         fclose(f);
1837 }
1838
1839
1840 static int read_next_command(void)
1841 {
1842         static int stdin_eof = 0;
1843
1844         if (stdin_eof) {
1845                 unread_command_buf = 0;
1846                 return EOF;
1847         }
1848
1849         for (;;) {
1850                 if (unread_command_buf) {
1851                         unread_command_buf = 0;
1852                 } else {
1853                         struct recent_command *rc;
1854
1855                         strbuf_detach(&command_buf, NULL);
1856                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1857                         if (stdin_eof)
1858                                 return EOF;
1859
1860                         if (!seen_data_command
1861                                 && prefixcmp(command_buf.buf, "feature ")
1862                                 && prefixcmp(command_buf.buf, "option ")) {
1863                                 parse_argv();
1864                         }
1865
1866                         rc = rc_free;
1867                         if (rc)
1868                                 rc_free = rc->next;
1869                         else {
1870                                 rc = cmd_hist.next;
1871                                 cmd_hist.next = rc->next;
1872                                 cmd_hist.next->prev = &cmd_hist;
1873                                 free(rc->buf);
1874                         }
1875
1876                         rc->buf = command_buf.buf;
1877                         rc->prev = cmd_tail;
1878                         rc->next = cmd_hist.prev;
1879                         rc->prev->next = rc;
1880                         cmd_tail = rc;
1881                 }
1882                 if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1883                         parse_cat_blob();
1884                         continue;
1885                 }
1886                 if (command_buf.buf[0] == '#')
1887                         continue;
1888                 return 0;
1889         }
1890 }
1891
1892 static void skip_optional_lf(void)
1893 {
1894         int term_char = fgetc(stdin);
1895         if (term_char != '\n' && term_char != EOF)
1896                 ungetc(term_char, stdin);
1897 }
1898
1899 static void parse_mark(void)
1900 {
1901         if (!prefixcmp(command_buf.buf, "mark :")) {
1902                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1903                 read_next_command();
1904         }
1905         else
1906                 next_mark = 0;
1907 }
1908
1909 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1910 {
1911         strbuf_reset(sb);
1912
1913         if (prefixcmp(command_buf.buf, "data "))
1914                 die("Expected 'data n' command, found: %s", command_buf.buf);
1915
1916         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1917                 char *term = xstrdup(command_buf.buf + 5 + 2);
1918                 size_t term_len = command_buf.len - 5 - 2;
1919
1920                 strbuf_detach(&command_buf, NULL);
1921                 for (;;) {
1922                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1923                                 die("EOF in data (terminator '%s' not found)", term);
1924                         if (term_len == command_buf.len
1925                                 && !strcmp(term, command_buf.buf))
1926                                 break;
1927                         strbuf_addbuf(sb, &command_buf);
1928                         strbuf_addch(sb, '\n');
1929                 }
1930                 free(term);
1931         }
1932         else {
1933                 uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1934                 size_t n = 0, length = (size_t)len;
1935
1936                 if (limit && limit < len) {
1937                         *len_res = len;
1938                         return 0;
1939                 }
1940                 if (length < len)
1941                         die("data is too large to use in this context");
1942
1943                 while (n < length) {
1944                         size_t s = strbuf_fread(sb, length - n, stdin);
1945                         if (!s && feof(stdin))
1946                                 die("EOF in data (%lu bytes remaining)",
1947                                         (unsigned long)(length - n));
1948                         n += s;
1949                 }
1950         }
1951
1952         skip_optional_lf();
1953         return 1;
1954 }
1955
1956 static int validate_raw_date(const char *src, char *result, int maxlen)
1957 {
1958         const char *orig_src = src;
1959         char *endp;
1960         unsigned long num;
1961
1962         errno = 0;
1963
1964         num = strtoul(src, &endp, 10);
1965         /* NEEDSWORK: perhaps check for reasonable values? */
1966         if (errno || endp == src || *endp != ' ')
1967                 return -1;
1968
1969         src = endp + 1;
1970         if (*src != '-' && *src != '+')
1971                 return -1;
1972
1973         num = strtoul(src + 1, &endp, 10);
1974         if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1975             1400 < num)
1976                 return -1;
1977
1978         strcpy(result, orig_src);
1979         return 0;
1980 }
1981
1982 static char *parse_ident(const char *buf)
1983 {
1984         const char *gt;
1985         size_t name_len;
1986         char *ident;
1987
1988         gt = strrchr(buf, '>');
1989         if (!gt)
1990                 die("Missing > in ident string: %s", buf);
1991         gt++;
1992         if (*gt != ' ')
1993                 die("Missing space after > in ident string: %s", buf);
1994         gt++;
1995         name_len = gt - buf;
1996         ident = xmalloc(name_len + 24);
1997         strncpy(ident, buf, name_len);
1998
1999         switch (whenspec) {
2000         case WHENSPEC_RAW:
2001                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
2002                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
2003                 break;
2004         case WHENSPEC_RFC2822:
2005                 if (parse_date(gt, ident + name_len, 24) < 0)
2006                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
2007                 break;
2008         case WHENSPEC_NOW:
2009                 if (strcmp("now", gt))
2010                         die("Date in ident must be 'now': %s", buf);
2011                 datestamp(ident + name_len, 24);
2012                 break;
2013         }
2014
2015         return ident;
2016 }
2017
2018 static void parse_and_store_blob(
2019         struct last_object *last,
2020         unsigned char *sha1out,
2021         uintmax_t mark)
2022 {
2023         static struct strbuf buf = STRBUF_INIT;
2024         uintmax_t len;
2025
2026         if (parse_data(&buf, big_file_threshold, &len))
2027                 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2028         else {
2029                 if (last) {
2030                         strbuf_release(&last->data);
2031                         last->offset = 0;
2032                         last->depth = 0;
2033                 }
2034                 stream_blob(len, sha1out, mark);
2035                 skip_optional_lf();
2036         }
2037 }
2038
2039 static void parse_new_blob(void)
2040 {
2041         read_next_command();
2042         parse_mark();
2043         parse_and_store_blob(&last_blob, NULL, next_mark);
2044 }
2045
2046 static void unload_one_branch(void)
2047 {
2048         while (cur_active_branches
2049                 && cur_active_branches >= max_active_branches) {
2050                 uintmax_t min_commit = ULONG_MAX;
2051                 struct branch *e, *l = NULL, *p = NULL;
2052
2053                 for (e = active_branches; e; e = e->active_next_branch) {
2054                         if (e->last_commit < min_commit) {
2055                                 p = l;
2056                                 min_commit = e->last_commit;
2057                         }
2058                         l = e;
2059                 }
2060
2061                 if (p) {
2062                         e = p->active_next_branch;
2063                         p->active_next_branch = e->active_next_branch;
2064                 } else {
2065                         e = active_branches;
2066                         active_branches = e->active_next_branch;
2067                 }
2068                 e->active = 0;
2069                 e->active_next_branch = NULL;
2070                 if (e->branch_tree.tree) {
2071                         release_tree_content_recursive(e->branch_tree.tree);
2072                         e->branch_tree.tree = NULL;
2073                 }
2074                 cur_active_branches--;
2075         }
2076 }
2077
2078 static void load_branch(struct branch *b)
2079 {
2080         load_tree(&b->branch_tree);
2081         if (!b->active) {
2082                 b->active = 1;
2083                 b->active_next_branch = active_branches;
2084                 active_branches = b;
2085                 cur_active_branches++;
2086                 branch_load_count++;
2087         }
2088 }
2089
2090 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2091 {
2092         unsigned char fanout = 0;
2093         while ((num_notes >>= 8))
2094                 fanout++;
2095         return fanout;
2096 }
2097
2098 static void construct_path_with_fanout(const char *hex_sha1,
2099                 unsigned char fanout, char *path)
2100 {
2101         unsigned int i = 0, j = 0;
2102         if (fanout >= 20)
2103                 die("Too large fanout (%u)", fanout);
2104         while (fanout) {
2105                 path[i++] = hex_sha1[j++];
2106                 path[i++] = hex_sha1[j++];
2107                 path[i++] = '/';
2108                 fanout--;
2109         }
2110         memcpy(path + i, hex_sha1 + j, 40 - j);
2111         path[i + 40 - j] = '\0';
2112 }
2113
2114 static uintmax_t do_change_note_fanout(
2115                 struct tree_entry *orig_root, struct tree_entry *root,
2116                 char *hex_sha1, unsigned int hex_sha1_len,
2117                 char *fullpath, unsigned int fullpath_len,
2118                 unsigned char fanout)
2119 {
2120         struct tree_content *t = root->tree;
2121         struct tree_entry *e, leaf;
2122         unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2123         uintmax_t num_notes = 0;
2124         unsigned char sha1[20];
2125         char realpath[60];
2126
2127         for (i = 0; t && i < t->entry_count; i++) {
2128                 e = t->entries[i];
2129                 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2130                 tmp_fullpath_len = fullpath_len;
2131
2132                 /*
2133                  * We're interested in EITHER existing note entries (entries
2134                  * with exactly 40 hex chars in path, not including directory
2135                  * separators), OR directory entries that may contain note
2136                  * entries (with < 40 hex chars in path).
2137                  * Also, each path component in a note entry must be a multiple
2138                  * of 2 chars.
2139                  */
2140                 if (!e->versions[1].mode ||
2141                     tmp_hex_sha1_len > 40 ||
2142                     e->name->str_len % 2)
2143                         continue;
2144
2145                 /* This _may_ be a note entry, or a subdir containing notes */
2146                 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2147                        e->name->str_len);
2148                 if (tmp_fullpath_len)
2149                         fullpath[tmp_fullpath_len++] = '/';
2150                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2151                        e->name->str_len);
2152                 tmp_fullpath_len += e->name->str_len;
2153                 fullpath[tmp_fullpath_len] = '\0';
2154
2155                 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2156                         /* This is a note entry */
2157                         construct_path_with_fanout(hex_sha1, fanout, realpath);
2158                         if (!strcmp(fullpath, realpath)) {
2159                                 /* Note entry is in correct location */
2160                                 num_notes++;
2161                                 continue;
2162                         }
2163
2164                         /* Rename fullpath to realpath */
2165                         if (!tree_content_remove(orig_root, fullpath, &leaf))
2166                                 die("Failed to remove path %s", fullpath);
2167                         tree_content_set(orig_root, realpath,
2168                                 leaf.versions[1].sha1,
2169                                 leaf.versions[1].mode,
2170                                 leaf.tree);
2171                 } else if (S_ISDIR(e->versions[1].mode)) {
2172                         /* This is a subdir that may contain note entries */
2173                         if (!e->tree)
2174                                 load_tree(e);
2175                         num_notes += do_change_note_fanout(orig_root, e,
2176                                 hex_sha1, tmp_hex_sha1_len,
2177                                 fullpath, tmp_fullpath_len, fanout);
2178                 }
2179
2180                 /* The above may have reallocated the current tree_content */
2181                 t = root->tree;
2182         }
2183         return num_notes;
2184 }
2185
2186 static uintmax_t change_note_fanout(struct tree_entry *root,
2187                 unsigned char fanout)
2188 {
2189         char hex_sha1[40], path[60];
2190         return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2191 }
2192
2193 static void file_change_m(struct branch *b)
2194 {
2195         const char *p = command_buf.buf + 2;
2196         static struct strbuf uq = STRBUF_INIT;
2197         const char *endp;
2198         struct object_entry *oe = oe;
2199         unsigned char sha1[20];
2200         uint16_t mode, inline_data = 0;
2201
2202         p = get_mode(p, &mode);
2203         if (!p)
2204                 die("Corrupt mode: %s", command_buf.buf);
2205         switch (mode) {
2206         case 0644:
2207         case 0755:
2208                 mode |= S_IFREG;
2209         case S_IFREG | 0644:
2210         case S_IFREG | 0755:
2211         case S_IFLNK:
2212         case S_IFDIR:
2213         case S_IFGITLINK:
2214                 /* ok */
2215                 break;
2216         default:
2217                 die("Corrupt mode: %s", command_buf.buf);
2218         }
2219
2220         if (*p == ':') {
2221                 char *x;
2222                 oe = find_mark(strtoumax(p + 1, &x, 10));
2223                 hashcpy(sha1, oe->idx.sha1);
2224                 p = x;
2225         } else if (!prefixcmp(p, "inline")) {
2226                 inline_data = 1;
2227                 p += 6;
2228         } else {
2229                 if (get_sha1_hex(p, sha1))
2230                         die("Invalid SHA1: %s", command_buf.buf);
2231                 oe = find_object(sha1);
2232                 p += 40;
2233         }
2234         if (*p++ != ' ')
2235                 die("Missing space after SHA1: %s", command_buf.buf);
2236
2237         strbuf_reset(&uq);
2238         if (!unquote_c_style(&uq, p, &endp)) {
2239                 if (*endp)
2240                         die("Garbage after path in: %s", command_buf.buf);
2241                 p = uq.buf;
2242         }
2243
2244         /* Git does not track empty, non-toplevel directories. */
2245         if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2246                 tree_content_remove(&b->branch_tree, p, NULL);
2247                 return;
2248         }
2249
2250         if (S_ISGITLINK(mode)) {
2251                 if (inline_data)
2252                         die("Git links cannot be specified 'inline': %s",
2253                                 command_buf.buf);
2254                 else if (oe) {
2255                         if (oe->type != OBJ_COMMIT)
2256                                 die("Not a commit (actually a %s): %s",
2257                                         typename(oe->type), command_buf.buf);
2258                 }
2259                 /*
2260                  * Accept the sha1 without checking; it expected to be in
2261                  * another repository.
2262                  */
2263         } else if (inline_data) {
2264                 if (S_ISDIR(mode))
2265                         die("Directories cannot be specified 'inline': %s",
2266                                 command_buf.buf);
2267                 if (p != uq.buf) {
2268                         strbuf_addstr(&uq, p);
2269                         p = uq.buf;
2270                 }
2271                 read_next_command();
2272                 parse_and_store_blob(&last_blob, sha1, 0);
2273         } else {
2274                 enum object_type expected = S_ISDIR(mode) ?
2275                                                 OBJ_TREE: OBJ_BLOB;
2276                 enum object_type type = oe ? oe->type :
2277                                         sha1_object_info(sha1, NULL);
2278                 if (type < 0)
2279                         die("%s not found: %s",
2280                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2281                                         command_buf.buf);
2282                 if (type != expected)
2283                         die("Not a %s (actually a %s): %s",
2284                                 typename(expected), typename(type),
2285                                 command_buf.buf);
2286         }
2287
2288         if (!*p) {
2289                 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2290                 return;
2291         }
2292         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2293 }
2294
2295 static void file_change_d(struct branch *b)
2296 {
2297         const char *p = command_buf.buf + 2;
2298         static struct strbuf uq = STRBUF_INIT;
2299         const char *endp;
2300
2301         strbuf_reset(&uq);
2302         if (!unquote_c_style(&uq, p, &endp)) {
2303                 if (*endp)
2304                         die("Garbage after path in: %s", command_buf.buf);
2305                 p = uq.buf;
2306         }
2307         tree_content_remove(&b->branch_tree, p, NULL);
2308 }
2309
2310 static void file_change_cr(struct branch *b, int rename)
2311 {
2312         const char *s, *d;
2313         static struct strbuf s_uq = STRBUF_INIT;
2314         static struct strbuf d_uq = STRBUF_INIT;
2315         const char *endp;
2316         struct tree_entry leaf;
2317
2318         s = command_buf.buf + 2;
2319         strbuf_reset(&s_uq);
2320         if (!unquote_c_style(&s_uq, s, &endp)) {
2321                 if (*endp != ' ')
2322                         die("Missing space after source: %s", command_buf.buf);
2323         } else {
2324                 endp = strchr(s, ' ');
2325                 if (!endp)
2326                         die("Missing space after source: %s", command_buf.buf);
2327                 strbuf_add(&s_uq, s, endp - s);
2328         }
2329         s = s_uq.buf;
2330
2331         endp++;
2332         if (!*endp)
2333                 die("Missing dest: %s", command_buf.buf);
2334
2335         d = endp;
2336         strbuf_reset(&d_uq);
2337         if (!unquote_c_style(&d_uq, d, &endp)) {
2338                 if (*endp)
2339                         die("Garbage after dest in: %s", command_buf.buf);
2340                 d = d_uq.buf;
2341         }
2342
2343         memset(&leaf, 0, sizeof(leaf));
2344         if (rename)
2345                 tree_content_remove(&b->branch_tree, s, &leaf);
2346         else
2347                 tree_content_get(&b->branch_tree, s, &leaf);
2348         if (!leaf.versions[1].mode)
2349                 die("Path %s not in branch", s);
2350         if (!*d) {      /* C "path/to/subdir" "" */
2351                 tree_content_replace(&b->branch_tree,
2352                         leaf.versions[1].sha1,
2353                         leaf.versions[1].mode,
2354                         leaf.tree);
2355                 return;
2356         }
2357         tree_content_set(&b->branch_tree, d,
2358                 leaf.versions[1].sha1,
2359                 leaf.versions[1].mode,
2360                 leaf.tree);
2361 }
2362
2363 static void note_change_n(struct branch *b, unsigned char old_fanout)
2364 {
2365         const char *p = command_buf.buf + 2;
2366         static struct strbuf uq = STRBUF_INIT;
2367         struct object_entry *oe = oe;
2368         struct branch *s;
2369         unsigned char sha1[20], commit_sha1[20];
2370         char path[60];
2371         uint16_t inline_data = 0;
2372         unsigned char new_fanout;
2373
2374         /* <dataref> or 'inline' */
2375         if (*p == ':') {
2376                 char *x;
2377                 oe = find_mark(strtoumax(p + 1, &x, 10));
2378                 hashcpy(sha1, oe->idx.sha1);
2379                 p = x;
2380         } else if (!prefixcmp(p, "inline")) {
2381                 inline_data = 1;
2382                 p += 6;
2383         } else {
2384                 if (get_sha1_hex(p, sha1))
2385                         die("Invalid SHA1: %s", command_buf.buf);
2386                 oe = find_object(sha1);
2387                 p += 40;
2388         }
2389         if (*p++ != ' ')
2390                 die("Missing space after SHA1: %s", command_buf.buf);
2391
2392         /* <committish> */
2393         s = lookup_branch(p);
2394         if (s) {
2395                 hashcpy(commit_sha1, s->sha1);
2396         } else if (*p == ':') {
2397                 uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2398                 struct object_entry *commit_oe = find_mark(commit_mark);
2399                 if (commit_oe->type != OBJ_COMMIT)
2400                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2401                 hashcpy(commit_sha1, commit_oe->idx.sha1);
2402         } else if (!get_sha1(p, commit_sha1)) {
2403                 unsigned long size;
2404                 char *buf = read_object_with_reference(commit_sha1,
2405                         commit_type, &size, commit_sha1);
2406                 if (!buf || size < 46)
2407                         die("Not a valid commit: %s", p);
2408                 free(buf);
2409         } else
2410                 die("Invalid ref name or SHA1 expression: %s", p);
2411
2412         if (inline_data) {
2413                 if (p != uq.buf) {
2414                         strbuf_addstr(&uq, p);
2415                         p = uq.buf;
2416                 }
2417                 read_next_command();
2418                 parse_and_store_blob(&last_blob, sha1, 0);
2419         } else if (oe) {
2420                 if (oe->type != OBJ_BLOB)
2421                         die("Not a blob (actually a %s): %s",
2422                                 typename(oe->type), command_buf.buf);
2423         } else if (!is_null_sha1(sha1)) {
2424                 enum object_type type = sha1_object_info(sha1, NULL);
2425                 if (type < 0)
2426                         die("Blob not found: %s", command_buf.buf);
2427                 if (type != OBJ_BLOB)
2428                         die("Not a blob (actually a %s): %s",
2429                             typename(type), command_buf.buf);
2430         }
2431
2432         construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
2433         if (tree_content_remove(&b->branch_tree, path, NULL))
2434                 b->num_notes--;
2435
2436         if (is_null_sha1(sha1))
2437                 return; /* nothing to insert */
2438
2439         b->num_notes++;
2440         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2441         construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2442         tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2443 }
2444
2445 static void file_change_deleteall(struct branch *b)
2446 {
2447         release_tree_content_recursive(b->branch_tree.tree);
2448         hashclr(b->branch_tree.versions[0].sha1);
2449         hashclr(b->branch_tree.versions[1].sha1);
2450         load_tree(&b->branch_tree);
2451         b->num_notes = 0;
2452 }
2453
2454 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2455 {
2456         if (!buf || size < 46)
2457                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2458         if (memcmp("tree ", buf, 5)
2459                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2460                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2461         hashcpy(b->branch_tree.versions[0].sha1,
2462                 b->branch_tree.versions[1].sha1);
2463 }
2464
2465 static void parse_from_existing(struct branch *b)
2466 {
2467         if (is_null_sha1(b->sha1)) {
2468                 hashclr(b->branch_tree.versions[0].sha1);
2469                 hashclr(b->branch_tree.versions[1].sha1);
2470         } else {
2471                 unsigned long size;
2472                 char *buf;
2473
2474                 buf = read_object_with_reference(b->sha1,
2475                         commit_type, &size, b->sha1);
2476                 parse_from_commit(b, buf, size);
2477                 free(buf);
2478         }
2479 }
2480
2481 static int parse_from(struct branch *b)
2482 {
2483         const char *from;
2484         struct branch *s;
2485
2486         if (prefixcmp(command_buf.buf, "from "))
2487                 return 0;
2488
2489         if (b->branch_tree.tree) {
2490                 release_tree_content_recursive(b->branch_tree.tree);
2491                 b->branch_tree.tree = NULL;
2492         }
2493
2494         from = strchr(command_buf.buf, ' ') + 1;
2495         s = lookup_branch(from);
2496         if (b == s)
2497                 die("Can't create a branch from itself: %s", b->name);
2498         else if (s) {
2499                 unsigned char *t = s->branch_tree.versions[1].sha1;
2500                 hashcpy(b->sha1, s->sha1);
2501                 hashcpy(b->branch_tree.versions[0].sha1, t);
2502                 hashcpy(b->branch_tree.versions[1].sha1, t);
2503         } else if (*from == ':') {
2504                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2505                 struct object_entry *oe = find_mark(idnum);
2506                 if (oe->type != OBJ_COMMIT)
2507                         die("Mark :%" PRIuMAX " not a commit", idnum);
2508                 hashcpy(b->sha1, oe->idx.sha1);
2509                 if (oe->pack_id != MAX_PACK_ID) {
2510                         unsigned long size;
2511                         char *buf = gfi_unpack_entry(oe, &size);
2512                         parse_from_commit(b, buf, size);
2513                         free(buf);
2514                 } else
2515                         parse_from_existing(b);
2516         } else if (!get_sha1(from, b->sha1))
2517                 parse_from_existing(b);
2518         else
2519                 die("Invalid ref name or SHA1 expression: %s", from);
2520
2521         read_next_command();
2522         return 1;
2523 }
2524
2525 static struct hash_list *parse_merge(unsigned int *count)
2526 {
2527         struct hash_list *list = NULL, *n, *e = e;
2528         const char *from;
2529         struct branch *s;
2530
2531         *count = 0;
2532         while (!prefixcmp(command_buf.buf, "merge ")) {
2533                 from = strchr(command_buf.buf, ' ') + 1;
2534                 n = xmalloc(sizeof(*n));
2535                 s = lookup_branch(from);
2536                 if (s)
2537                         hashcpy(n->sha1, s->sha1);
2538                 else if (*from == ':') {
2539                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2540                         struct object_entry *oe = find_mark(idnum);
2541                         if (oe->type != OBJ_COMMIT)
2542                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2543                         hashcpy(n->sha1, oe->idx.sha1);
2544                 } else if (!get_sha1(from, n->sha1)) {
2545                         unsigned long size;
2546                         char *buf = read_object_with_reference(n->sha1,
2547                                 commit_type, &size, n->sha1);
2548                         if (!buf || size < 46)
2549                                 die("Not a valid commit: %s", from);
2550                         free(buf);
2551                 } else
2552                         die("Invalid ref name or SHA1 expression: %s", from);
2553
2554                 n->next = NULL;
2555                 if (list)
2556                         e->next = n;
2557                 else
2558                         list = n;
2559                 e = n;
2560                 (*count)++;
2561                 read_next_command();
2562         }
2563         return list;
2564 }
2565
2566 static void parse_new_commit(void)
2567 {
2568         static struct strbuf msg = STRBUF_INIT;
2569         struct branch *b;
2570         char *sp;
2571         char *author = NULL;
2572         char *committer = NULL;
2573         struct hash_list *merge_list = NULL;
2574         unsigned int merge_count;
2575         unsigned char prev_fanout, new_fanout;
2576
2577         /* Obtain the branch name from the rest of our command */
2578         sp = strchr(command_buf.buf, ' ') + 1;
2579         b = lookup_branch(sp);
2580         if (!b)
2581                 b = new_branch(sp);
2582
2583         read_next_command();
2584         parse_mark();
2585         if (!prefixcmp(command_buf.buf, "author ")) {
2586                 author = parse_ident(command_buf.buf + 7);
2587                 read_next_command();
2588         }
2589         if (!prefixcmp(command_buf.buf, "committer ")) {
2590                 committer = parse_ident(command_buf.buf + 10);
2591                 read_next_command();
2592         }
2593         if (!committer)
2594                 die("Expected committer but didn't get one");
2595         parse_data(&msg, 0, NULL);
2596         read_next_command();
2597         parse_from(b);
2598         merge_list = parse_merge(&merge_count);
2599
2600         /* ensure the branch is active/loaded */
2601         if (!b->branch_tree.tree || !max_active_branches) {
2602                 unload_one_branch();
2603                 load_branch(b);
2604         }
2605
2606         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2607
2608         /* file_change* */
2609         while (command_buf.len > 0) {
2610                 if (!prefixcmp(command_buf.buf, "M "))
2611                         file_change_m(b);
2612                 else if (!prefixcmp(command_buf.buf, "D "))
2613                         file_change_d(b);
2614                 else if (!prefixcmp(command_buf.buf, "R "))
2615                         file_change_cr(b, 1);
2616                 else if (!prefixcmp(command_buf.buf, "C "))
2617                         file_change_cr(b, 0);
2618                 else if (!prefixcmp(command_buf.buf, "N "))
2619                         note_change_n(b, prev_fanout);
2620                 else if (!strcmp("deleteall", command_buf.buf))
2621                         file_change_deleteall(b);
2622                 else if (!prefixcmp(command_buf.buf, "ls "))
2623                         parse_ls(b);
2624                 else {
2625                         unread_command_buf = 1;
2626                         break;
2627                 }
2628                 if (read_next_command() == EOF)
2629                         break;
2630         }
2631
2632         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2633         if (new_fanout != prev_fanout)
2634                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2635
2636         /* build the tree and the commit */
2637         store_tree(&b->branch_tree);
2638         hashcpy(b->branch_tree.versions[0].sha1,
2639                 b->branch_tree.versions[1].sha1);
2640
2641         strbuf_reset(&new_data);
2642         strbuf_addf(&new_data, "tree %s\n",
2643                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2644         if (!is_null_sha1(b->sha1))
2645                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2646         while (merge_list) {
2647                 struct hash_list *next = merge_list->next;
2648                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2649                 free(merge_list);
2650                 merge_list = next;
2651         }
2652         strbuf_addf(&new_data,
2653                 "author %s\n"
2654                 "committer %s\n"
2655                 "\n",
2656                 author ? author : committer, committer);
2657         strbuf_addbuf(&new_data, &msg);
2658         free(author);
2659         free(committer);
2660
2661         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2662                 b->pack_id = pack_id;
2663         b->last_commit = object_count_by_type[OBJ_COMMIT];
2664 }
2665
2666 static void parse_new_tag(void)
2667 {
2668         static struct strbuf msg = STRBUF_INIT;
2669         char *sp;
2670         const char *from;
2671         char *tagger;
2672         struct branch *s;
2673         struct tag *t;
2674         uintmax_t from_mark = 0;
2675         unsigned char sha1[20];
2676         enum object_type type;
2677
2678         /* Obtain the new tag name from the rest of our command */
2679         sp = strchr(command_buf.buf, ' ') + 1;
2680         t = pool_alloc(sizeof(struct tag));
2681         t->next_tag = NULL;
2682         t->name = pool_strdup(sp);
2683         if (last_tag)
2684                 last_tag->next_tag = t;
2685         else
2686                 first_tag = t;
2687         last_tag = t;
2688         read_next_command();
2689
2690         /* from ... */
2691         if (prefixcmp(command_buf.buf, "from "))
2692                 die("Expected from command, got %s", command_buf.buf);
2693         from = strchr(command_buf.buf, ' ') + 1;
2694         s = lookup_branch(from);
2695         if (s) {
2696                 hashcpy(sha1, s->sha1);
2697                 type = OBJ_COMMIT;
2698         } else if (*from == ':') {
2699                 struct object_entry *oe;
2700                 from_mark = strtoumax(from + 1, NULL, 10);
2701                 oe = find_mark(from_mark);
2702                 type = oe->type;
2703                 hashcpy(sha1, oe->idx.sha1);
2704         } else if (!get_sha1(from, sha1)) {
2705                 unsigned long size;
2706                 char *buf;
2707
2708                 buf = read_sha1_file(sha1, &type, &size);
2709                 if (!buf || size < 46)
2710                         die("Not a valid commit: %s", from);
2711                 free(buf);
2712         } else
2713                 die("Invalid ref name or SHA1 expression: %s", from);
2714         read_next_command();
2715
2716         /* tagger ... */
2717         if (!prefixcmp(command_buf.buf, "tagger ")) {
2718                 tagger = parse_ident(command_buf.buf + 7);
2719                 read_next_command();
2720         } else
2721                 tagger = NULL;
2722
2723         /* tag payload/message */
2724         parse_data(&msg, 0, NULL);
2725
2726         /* build the tag object */
2727         strbuf_reset(&new_data);
2728
2729         strbuf_addf(&new_data,
2730                     "object %s\n"
2731                     "type %s\n"
2732                     "tag %s\n",
2733                     sha1_to_hex(sha1), typename(type), t->name);
2734         if (tagger)
2735                 strbuf_addf(&new_data,
2736                             "tagger %s\n", tagger);
2737         strbuf_addch(&new_data, '\n');
2738         strbuf_addbuf(&new_data, &msg);
2739         free(tagger);
2740
2741         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2742                 t->pack_id = MAX_PACK_ID;
2743         else
2744                 t->pack_id = pack_id;
2745 }
2746
2747 static void parse_reset_branch(void)
2748 {
2749         struct branch *b;
2750         char *sp;
2751
2752         /* Obtain the branch name from the rest of our command */
2753         sp = strchr(command_buf.buf, ' ') + 1;
2754         b = lookup_branch(sp);
2755         if (b) {
2756                 hashclr(b->sha1);
2757                 hashclr(b->branch_tree.versions[0].sha1);
2758                 hashclr(b->branch_tree.versions[1].sha1);
2759                 if (b->branch_tree.tree) {
2760                         release_tree_content_recursive(b->branch_tree.tree);
2761                         b->branch_tree.tree = NULL;
2762                 }
2763         }
2764         else
2765                 b = new_branch(sp);
2766         read_next_command();
2767         parse_from(b);
2768         if (command_buf.len > 0)
2769                 unread_command_buf = 1;
2770 }
2771
2772 static void cat_blob_write(const char *buf, unsigned long size)
2773 {
2774         if (write_in_full(cat_blob_fd, buf, size) != size)
2775                 die_errno("Write to frontend failed");
2776 }
2777
2778 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2779 {
2780         struct strbuf line = STRBUF_INIT;
2781         unsigned long size;
2782         enum object_type type = 0;
2783         char *buf;
2784
2785         if (!oe || oe->pack_id == MAX_PACK_ID) {
2786                 buf = read_sha1_file(sha1, &type, &size);
2787         } else {
2788                 type = oe->type;
2789                 buf = gfi_unpack_entry(oe, &size);
2790         }
2791
2792         /*
2793          * Output based on batch_one_object() from cat-file.c.
2794          */
2795         if (type <= 0) {
2796                 strbuf_reset(&line);
2797                 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2798                 cat_blob_write(line.buf, line.len);
2799                 strbuf_release(&line);
2800                 free(buf);
2801                 return;
2802         }
2803         if (!buf)
2804                 die("Can't read object %s", sha1_to_hex(sha1));
2805         if (type != OBJ_BLOB)
2806                 die("Object %s is a %s but a blob was expected.",
2807                     sha1_to_hex(sha1), typename(type));
2808         strbuf_reset(&line);
2809         strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2810                                                 typename(type), size);
2811         cat_blob_write(line.buf, line.len);
2812         strbuf_release(&line);
2813         cat_blob_write(buf, size);
2814         cat_blob_write("\n", 1);
2815         free(buf);
2816 }
2817
2818 static void parse_cat_blob(void)
2819 {
2820         const char *p;
2821         struct object_entry *oe = oe;
2822         unsigned char sha1[20];
2823
2824         /* cat-blob SP <object> LF */
2825         p = command_buf.buf + strlen("cat-blob ");
2826         if (*p == ':') {
2827                 char *x;
2828                 oe = find_mark(strtoumax(p + 1, &x, 10));
2829                 if (x == p + 1)
2830                         die("Invalid mark: %s", command_buf.buf);
2831                 if (!oe)
2832                         die("Unknown mark: %s", command_buf.buf);
2833                 if (*x)
2834                         die("Garbage after mark: %s", command_buf.buf);
2835                 hashcpy(sha1, oe->idx.sha1);
2836         } else {
2837                 if (get_sha1_hex(p, sha1))
2838                         die("Invalid SHA1: %s", command_buf.buf);
2839                 if (p[40])
2840                         die("Garbage after SHA1: %s", command_buf.buf);
2841                 oe = find_object(sha1);
2842         }
2843
2844         cat_blob(oe, sha1);
2845 }
2846
2847 static struct object_entry *dereference(struct object_entry *oe,
2848                                         unsigned char sha1[20])
2849 {
2850         unsigned long size;
2851         char *buf = NULL;
2852         if (!oe) {
2853                 enum object_type type = sha1_object_info(sha1, NULL);
2854                 if (type < 0)
2855                         die("object not found: %s", sha1_to_hex(sha1));
2856                 /* cache it! */
2857                 oe = insert_object(sha1);
2858                 oe->type = type;
2859                 oe->pack_id = MAX_PACK_ID;
2860                 oe->idx.offset = 1;
2861         }
2862         switch (oe->type) {
2863         case OBJ_TREE:  /* easy case. */
2864                 return oe;
2865         case OBJ_COMMIT:
2866         case OBJ_TAG:
2867                 break;
2868         default:
2869                 die("Not a treeish: %s", command_buf.buf);
2870         }
2871
2872         if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
2873                 buf = gfi_unpack_entry(oe, &size);
2874         } else {
2875                 enum object_type unused;
2876                 buf = read_sha1_file(sha1, &unused, &size);
2877         }
2878         if (!buf)
2879                 die("Can't load object %s", sha1_to_hex(sha1));
2880
2881         /* Peel one layer. */
2882         switch (oe->type) {
2883         case OBJ_TAG:
2884                 if (size < 40 + strlen("object ") ||
2885                     get_sha1_hex(buf + strlen("object "), sha1))
2886                         die("Invalid SHA1 in tag: %s", command_buf.buf);
2887                 break;
2888         case OBJ_COMMIT:
2889                 if (size < 40 + strlen("tree ") ||
2890                     get_sha1_hex(buf + strlen("tree "), sha1))
2891                         die("Invalid SHA1 in commit: %s", command_buf.buf);
2892         }
2893
2894         free(buf);
2895         return find_object(sha1);
2896 }
2897
2898 static struct object_entry *parse_treeish_dataref(const char **p)
2899 {
2900         unsigned char sha1[20];
2901         struct object_entry *e;
2902
2903         if (**p == ':') {       /* <mark> */
2904                 char *endptr;
2905                 e = find_mark(strtoumax(*p + 1, &endptr, 10));
2906                 if (endptr == *p + 1)
2907                         die("Invalid mark: %s", command_buf.buf);
2908                 if (!e)
2909                         die("Unknown mark: %s", command_buf.buf);
2910                 *p = endptr;
2911                 hashcpy(sha1, e->idx.sha1);
2912         } else {        /* <sha1> */
2913                 if (get_sha1_hex(*p, sha1))
2914                         die("Invalid SHA1: %s", command_buf.buf);
2915                 e = find_object(sha1);
2916                 *p += 40;
2917         }
2918
2919         while (!e || e->type != OBJ_TREE)
2920                 e = dereference(e, sha1);
2921         return e;
2922 }
2923
2924 static void print_ls(int mode, const unsigned char *sha1, const char *path)
2925 {
2926         static struct strbuf line = STRBUF_INIT;
2927
2928         /* See show_tree(). */
2929         const char *type =
2930                 S_ISGITLINK(mode) ? commit_type :
2931                 S_ISDIR(mode) ? tree_type :
2932                 blob_type;
2933
2934         if (!mode) {
2935                 /* missing SP path LF */
2936                 strbuf_reset(&line);
2937                 strbuf_addstr(&line, "missing ");
2938                 quote_c_style(path, &line, NULL, 0);
2939                 strbuf_addch(&line, '\n');
2940         } else {
2941                 /* mode SP type SP object_name TAB path LF */
2942                 strbuf_reset(&line);
2943                 strbuf_addf(&line, "%06o %s %s\t",
2944                                 mode, type, sha1_to_hex(sha1));
2945                 quote_c_style(path, &line, NULL, 0);
2946                 strbuf_addch(&line, '\n');
2947         }
2948         cat_blob_write(line.buf, line.len);
2949 }
2950
2951 static void parse_ls(struct branch *b)
2952 {
2953         const char *p;
2954         struct tree_entry *root = NULL;
2955         struct tree_entry leaf = {0};
2956
2957         /* ls SP (<treeish> SP)? <path> */
2958         p = command_buf.buf + strlen("ls ");
2959         if (*p == '"') {
2960                 if (!b)
2961                         die("Not in a commit: %s", command_buf.buf);
2962                 root = &b->branch_tree;
2963         } else {
2964                 struct object_entry *e = parse_treeish_dataref(&p);
2965                 root = new_tree_entry();
2966                 hashcpy(root->versions[1].sha1, e->idx.sha1);
2967                 load_tree(root);
2968                 if (*p++ != ' ')
2969                         die("Missing space after tree-ish: %s", command_buf.buf);
2970         }
2971         if (*p == '"') {
2972                 static struct strbuf uq = STRBUF_INIT;
2973                 const char *endp;
2974                 strbuf_reset(&uq);
2975                 if (unquote_c_style(&uq, p, &endp))
2976                         die("Invalid path: %s", command_buf.buf);
2977                 if (*endp)
2978                         die("Garbage after path in: %s", command_buf.buf);
2979                 p = uq.buf;
2980         }
2981         tree_content_get(root, p, &leaf);
2982         /*
2983          * A directory in preparation would have a sha1 of zero
2984          * until it is saved.  Save, for simplicity.
2985          */
2986         if (S_ISDIR(leaf.versions[1].mode))
2987                 store_tree(&leaf);
2988
2989         print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
2990         if (!b || root != &b->branch_tree)
2991                 release_tree_entry(root);
2992 }
2993
2994 static void checkpoint(void)
2995 {
2996         checkpoint_requested = 0;
2997         if (object_count) {
2998                 cycle_packfile();
2999                 dump_branches();
3000                 dump_tags();
3001                 dump_marks();
3002         }
3003 }
3004
3005 static void parse_checkpoint(void)
3006 {
3007         checkpoint_requested = 1;
3008         skip_optional_lf();
3009 }
3010
3011 static void parse_progress(void)
3012 {
3013         fwrite(command_buf.buf, 1, command_buf.len, stdout);
3014         fputc('\n', stdout);
3015         fflush(stdout);
3016         skip_optional_lf();
3017 }
3018
3019 static char* make_fast_import_path(const char *path)
3020 {
3021         struct strbuf abs_path = STRBUF_INIT;
3022
3023         if (!relative_marks_paths || is_absolute_path(path))
3024                 return xstrdup(path);
3025         strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
3026         return strbuf_detach(&abs_path, NULL);
3027 }
3028
3029 static void option_import_marks(const char *marks,
3030                                         int from_stream, int ignore_missing)
3031 {
3032         if (import_marks_file) {
3033                 if (from_stream)
3034                         die("Only one import-marks command allowed per stream");
3035
3036                 /* read previous mark file */
3037                 if(!import_marks_file_from_stream)
3038                         read_marks();
3039         }
3040
3041         import_marks_file = make_fast_import_path(marks);
3042         safe_create_leading_directories_const(import_marks_file);
3043         import_marks_file_from_stream = from_stream;
3044         import_marks_file_ignore_missing = ignore_missing;
3045 }
3046
3047 static void option_date_format(const char *fmt)
3048 {
3049         if (!strcmp(fmt, "raw"))
3050                 whenspec = WHENSPEC_RAW;
3051         else if (!strcmp(fmt, "rfc2822"))
3052                 whenspec = WHENSPEC_RFC2822;
3053         else if (!strcmp(fmt, "now"))
3054                 whenspec = WHENSPEC_NOW;
3055         else
3056                 die("unknown --date-format argument %s", fmt);
3057 }
3058
3059 static unsigned long ulong_arg(const char *option, const char *arg)
3060 {
3061         char *endptr;
3062         unsigned long rv = strtoul(arg, &endptr, 0);
3063         if (strchr(arg, '-') || endptr == arg || *endptr)
3064                 die("%s: argument must be a non-negative integer", option);
3065         return rv;
3066 }
3067
3068 static void option_depth(const char *depth)
3069 {
3070         max_depth = ulong_arg("--depth", depth);
3071         if (max_depth > MAX_DEPTH)
3072                 die("--depth cannot exceed %u", MAX_DEPTH);
3073 }
3074
3075 static void option_active_branches(const char *branches)
3076 {
3077         max_active_branches = ulong_arg("--active-branches", branches);
3078 }
3079
3080 static void option_export_marks(const char *marks)
3081 {
3082         export_marks_file = make_fast_import_path(marks);
3083         safe_create_leading_directories_const(export_marks_file);
3084 }
3085
3086 static void option_cat_blob_fd(const char *fd)
3087 {
3088         unsigned long n = ulong_arg("--cat-blob-fd", fd);
3089         if (n > (unsigned long) INT_MAX)
3090                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3091         cat_blob_fd = (int) n;
3092 }
3093
3094 static void option_export_pack_edges(const char *edges)
3095 {
3096         if (pack_edges)
3097                 fclose(pack_edges);
3098         pack_edges = fopen(edges, "a");
3099         if (!pack_edges)
3100                 die_errno("Cannot open '%s'", edges);
3101 }
3102
3103 static int parse_one_option(const char *option)
3104 {
3105         if (!prefixcmp(option, "max-pack-size=")) {
3106                 unsigned long v;
3107                 if (!git_parse_ulong(option + 14, &v))
3108                         return 0;
3109                 if (v < 8192) {
3110                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3111                         v *= 1024 * 1024;
3112                 } else if (v < 1024 * 1024) {
3113                         warning("minimum max-pack-size is 1 MiB");
3114                         v = 1024 * 1024;
3115                 }
3116                 max_packsize = v;
3117         } else if (!prefixcmp(option, "big-file-threshold=")) {
3118                 unsigned long v;
3119                 if (!git_parse_ulong(option + 19, &v))
3120                         return 0;
3121                 big_file_threshold = v;
3122         } else if (!prefixcmp(option, "depth=")) {
3123                 option_depth(option + 6);
3124         } else if (!prefixcmp(option, "active-branches=")) {
3125                 option_active_branches(option + 16);
3126         } else if (!prefixcmp(option, "export-pack-edges=")) {
3127                 option_export_pack_edges(option + 18);
3128         } else if (!prefixcmp(option, "quiet")) {
3129                 show_stats = 0;
3130         } else if (!prefixcmp(option, "stats")) {
3131                 show_stats = 1;
3132         } else {
3133                 return 0;
3134         }
3135
3136         return 1;
3137 }
3138
3139 static int parse_one_feature(const char *feature, int from_stream)
3140 {
3141         if (!prefixcmp(feature, "date-format=")) {
3142                 option_date_format(feature + 12);
3143         } else if (!prefixcmp(feature, "import-marks=")) {
3144                 option_import_marks(feature + 13, from_stream, 0);
3145         } else if (!prefixcmp(feature, "import-marks-if-exists=")) {
3146                 option_import_marks(feature + strlen("import-marks-if-exists="),
3147                                         from_stream, 1);
3148         } else if (!prefixcmp(feature, "export-marks=")) {
3149                 option_export_marks(feature + 13);
3150         } else if (!strcmp(feature, "cat-blob")) {
3151                 ; /* Don't die - this feature is supported */
3152         } else if (!prefixcmp(feature, "relative-marks")) {
3153                 relative_marks_paths = 1;
3154         } else if (!prefixcmp(feature, "no-relative-marks")) {
3155                 relative_marks_paths = 0;
3156         } else if (!prefixcmp(feature, "force")) {
3157                 force_update = 1;
3158         } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3159                 ; /* do nothing; we have the feature */
3160         } else {
3161                 return 0;
3162         }
3163
3164         return 1;
3165 }
3166
3167 static void parse_feature(void)
3168 {
3169         char *feature = command_buf.buf + 8;
3170
3171         if (seen_data_command)
3172                 die("Got feature command '%s' after data command", feature);
3173
3174         if (parse_one_feature(feature, 1))
3175                 return;
3176
3177         die("This version of fast-import does not support feature %s.", feature);
3178 }
3179
3180 static void parse_option(void)
3181 {
3182         char *option = command_buf.buf + 11;
3183
3184         if (seen_data_command)
3185                 die("Got option command '%s' after data command", option);
3186
3187         if (parse_one_option(option))
3188                 return;
3189
3190         die("This version of fast-import does not support option: %s", option);
3191 }
3192
3193 static int git_pack_config(const char *k, const char *v, void *cb)
3194 {
3195         if (!strcmp(k, "pack.depth")) {
3196                 max_depth = git_config_int(k, v);
3197                 if (max_depth > MAX_DEPTH)
3198                         max_depth = MAX_DEPTH;
3199                 return 0;
3200         }
3201         if (!strcmp(k, "pack.compression")) {
3202                 int level = git_config_int(k, v);
3203                 if (level == -1)
3204                         level = Z_DEFAULT_COMPRESSION;
3205                 else if (level < 0 || level > Z_BEST_COMPRESSION)
3206                         die("bad pack compression level %d", level);
3207                 pack_compression_level = level;
3208                 pack_compression_seen = 1;
3209                 return 0;
3210         }
3211         if (!strcmp(k, "pack.indexversion")) {
3212                 pack_idx_default_version = git_config_int(k, v);
3213                 if (pack_idx_default_version > 2)
3214                         die("bad pack.indexversion=%"PRIu32,
3215                             pack_idx_default_version);
3216                 return 0;
3217         }
3218         if (!strcmp(k, "pack.packsizelimit")) {
3219                 max_packsize = git_config_ulong(k, v);
3220                 return 0;
3221         }
3222         if (!strcmp(k, "core.bigfilethreshold")) {
3223                 long n = git_config_int(k, v);
3224                 big_file_threshold = 0 < n ? n : 0;
3225         }
3226         return git_default_config(k, v, cb);
3227 }
3228
3229 static const char fast_import_usage[] =
3230 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3231
3232 static void parse_argv(void)
3233 {
3234         unsigned int i;
3235
3236         for (i = 1; i < global_argc; i++) {
3237                 const char *a = global_argv[i];
3238
3239                 if (*a != '-' || !strcmp(a, "--"))
3240                         break;
3241
3242                 if (parse_one_option(a + 2))
3243                         continue;
3244
3245                 if (parse_one_feature(a + 2, 0))
3246                         continue;
3247
3248                 if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3249                         option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3250                         continue;
3251                 }
3252
3253                 die("unknown option %s", a);
3254         }
3255         if (i != global_argc)
3256                 usage(fast_import_usage);
3257
3258         seen_data_command = 1;
3259         if (import_marks_file)
3260                 read_marks();
3261 }
3262
3263 int main(int argc, const char **argv)
3264 {
3265         unsigned int i;
3266
3267         git_extract_argv0_path(argv[0]);
3268
3269         if (argc == 2 && !strcmp(argv[1], "-h"))
3270                 usage(fast_import_usage);
3271
3272         setup_git_directory();
3273         git_config(git_pack_config, NULL);
3274         if (!pack_compression_seen && core_compression_seen)
3275                 pack_compression_level = core_compression_level;
3276
3277         alloc_objects(object_entry_alloc);
3278         strbuf_init(&command_buf, 0);
3279         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3280         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3281         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3282         marks = pool_calloc(1, sizeof(struct mark_set));
3283
3284         global_argc = argc;
3285         global_argv = argv;
3286
3287         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3288         for (i = 0; i < (cmd_save - 1); i++)
3289                 rc_free[i].next = &rc_free[i + 1];
3290         rc_free[cmd_save - 1].next = NULL;
3291
3292         prepare_packed_git();
3293         start_packfile();
3294         set_die_routine(die_nicely);
3295         set_checkpoint_signal();
3296         while (read_next_command() != EOF) {
3297                 if (!strcmp("blob", command_buf.buf))
3298                         parse_new_blob();
3299                 else if (!prefixcmp(command_buf.buf, "ls "))
3300                         parse_ls(NULL);
3301                 else if (!prefixcmp(command_buf.buf, "commit "))
3302                         parse_new_commit();
3303                 else if (!prefixcmp(command_buf.buf, "tag "))
3304                         parse_new_tag();
3305                 else if (!prefixcmp(command_buf.buf, "reset "))
3306                         parse_reset_branch();
3307                 else if (!strcmp("checkpoint", command_buf.buf))
3308                         parse_checkpoint();
3309                 else if (!prefixcmp(command_buf.buf, "progress "))
3310                         parse_progress();
3311                 else if (!prefixcmp(command_buf.buf, "feature "))
3312                         parse_feature();
3313                 else if (!prefixcmp(command_buf.buf, "option git "))
3314                         parse_option();
3315                 else if (!prefixcmp(command_buf.buf, "option "))
3316                         /* ignore non-git options*/;
3317                 else
3318                         die("Unsupported command: %s", command_buf.buf);
3319
3320                 if (checkpoint_requested)
3321                         checkpoint();
3322         }
3323
3324         /* argv hasn't been parsed yet, do so */
3325         if (!seen_data_command)
3326                 parse_argv();
3327
3328         end_packfile();
3329
3330         dump_branches();
3331         dump_tags();
3332         unkeep_all_packs();
3333         dump_marks();
3334
3335         if (pack_edges)
3336                 fclose(pack_edges);
3337
3338         if (show_stats) {
3339                 uintmax_t total_count = 0, duplicate_count = 0;
3340                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3341                         total_count += object_count_by_type[i];
3342                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3343                         duplicate_count += duplicate_count_by_type[i];
3344
3345                 fprintf(stderr, "%s statistics:\n", argv[0]);
3346                 fprintf(stderr, "---------------------------------------------------------------------\n");
3347                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3348                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3349                 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]);
3350                 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]);
3351                 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]);
3352                 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]);
3353                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3354                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3355                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3356                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3357                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3358                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3359                 fprintf(stderr, "---------------------------------------------------------------------\n");
3360                 pack_report();
3361                 fprintf(stderr, "---------------------------------------------------------------------\n");
3362                 fprintf(stderr, "\n");
3363         }
3364
3365         return failure ? 1 : 0;
3366 }