Merge branch 'sp/maint-fd-limit'
[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         p->do_not_close = 1;
881         pack_file = sha1fd(pack_fd, p->pack_name);
882
883         hdr.hdr_signature = htonl(PACK_SIGNATURE);
884         hdr.hdr_version = htonl(2);
885         hdr.hdr_entries = 0;
886         sha1write(pack_file, &hdr, sizeof(hdr));
887
888         pack_data = p;
889         pack_size = sizeof(hdr);
890         object_count = 0;
891
892         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
893         all_packs[pack_id] = p;
894 }
895
896 static const char *create_index(void)
897 {
898         const char *tmpfile;
899         struct pack_idx_entry **idx, **c, **last;
900         struct object_entry *e;
901         struct object_entry_pool *o;
902
903         /* Build the table of object IDs. */
904         idx = xmalloc(object_count * sizeof(*idx));
905         c = idx;
906         for (o = blocks; o; o = o->next_pool)
907                 for (e = o->next_free; e-- != o->entries;)
908                         if (pack_id == e->pack_id)
909                                 *c++ = &e->idx;
910         last = idx + object_count;
911         if (c != last)
912                 die("internal consistency error creating the index");
913
914         tmpfile = write_idx_file(NULL, idx, object_count, pack_data->sha1);
915         free(idx);
916         return tmpfile;
917 }
918
919 static char *keep_pack(const char *curr_index_name)
920 {
921         static char name[PATH_MAX];
922         static const char *keep_msg = "fast-import";
923         int keep_fd;
924
925         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
926         if (keep_fd < 0)
927                 die_errno("cannot create keep file");
928         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
929         if (close(keep_fd))
930                 die_errno("failed to write keep file");
931
932         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
933                  get_object_directory(), sha1_to_hex(pack_data->sha1));
934         if (move_temp_to_file(pack_data->pack_name, name))
935                 die("cannot store pack file");
936
937         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
938                  get_object_directory(), sha1_to_hex(pack_data->sha1));
939         if (move_temp_to_file(curr_index_name, name))
940                 die("cannot store index file");
941         free((void *)curr_index_name);
942         return name;
943 }
944
945 static void unkeep_all_packs(void)
946 {
947         static char name[PATH_MAX];
948         int k;
949
950         for (k = 0; k < pack_id; k++) {
951                 struct packed_git *p = all_packs[k];
952                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
953                          get_object_directory(), sha1_to_hex(p->sha1));
954                 unlink_or_warn(name);
955         }
956 }
957
958 static void end_packfile(void)
959 {
960         struct packed_git *old_p = pack_data, *new_p;
961
962         clear_delta_base_cache();
963         if (object_count) {
964                 unsigned char cur_pack_sha1[20];
965                 char *idx_name;
966                 int i;
967                 struct branch *b;
968                 struct tag *t;
969
970                 close_pack_windows(pack_data);
971                 sha1close(pack_file, cur_pack_sha1, 0);
972                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
973                                     pack_data->pack_name, object_count,
974                                     cur_pack_sha1, pack_size);
975                 close(pack_data->pack_fd);
976                 idx_name = keep_pack(create_index());
977
978                 /* Register the packfile with core git's machinery. */
979                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
980                 if (!new_p)
981                         die("core git rejected index %s", idx_name);
982                 all_packs[pack_id] = new_p;
983                 install_packed_git(new_p);
984
985                 /* Print the boundary */
986                 if (pack_edges) {
987                         fprintf(pack_edges, "%s:", new_p->pack_name);
988                         for (i = 0; i < branch_table_sz; i++) {
989                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
990                                         if (b->pack_id == pack_id)
991                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
992                                 }
993                         }
994                         for (t = first_tag; t; t = t->next_tag) {
995                                 if (t->pack_id == pack_id)
996                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
997                         }
998                         fputc('\n', pack_edges);
999                         fflush(pack_edges);
1000                 }
1001
1002                 pack_id++;
1003         }
1004         else {
1005                 close(old_p->pack_fd);
1006                 unlink_or_warn(old_p->pack_name);
1007         }
1008         free(old_p);
1009
1010         /* We can't carry a delta across packfiles. */
1011         strbuf_release(&last_blob.data);
1012         last_blob.offset = 0;
1013         last_blob.depth = 0;
1014 }
1015
1016 static void cycle_packfile(void)
1017 {
1018         end_packfile();
1019         start_packfile();
1020 }
1021
1022 static int store_object(
1023         enum object_type type,
1024         struct strbuf *dat,
1025         struct last_object *last,
1026         unsigned char *sha1out,
1027         uintmax_t mark)
1028 {
1029         void *out, *delta;
1030         struct object_entry *e;
1031         unsigned char hdr[96];
1032         unsigned char sha1[20];
1033         unsigned long hdrlen, deltalen;
1034         git_SHA_CTX c;
1035         z_stream s;
1036
1037         hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1038                 (unsigned long)dat->len) + 1;
1039         git_SHA1_Init(&c);
1040         git_SHA1_Update(&c, hdr, hdrlen);
1041         git_SHA1_Update(&c, dat->buf, dat->len);
1042         git_SHA1_Final(sha1, &c);
1043         if (sha1out)
1044                 hashcpy(sha1out, sha1);
1045
1046         e = insert_object(sha1);
1047         if (mark)
1048                 insert_mark(mark, e);
1049         if (e->idx.offset) {
1050                 duplicate_count_by_type[type]++;
1051                 return 1;
1052         } else if (find_sha1_pack(sha1, packed_git)) {
1053                 e->type = type;
1054                 e->pack_id = MAX_PACK_ID;
1055                 e->idx.offset = 1; /* just not zero! */
1056                 duplicate_count_by_type[type]++;
1057                 return 1;
1058         }
1059
1060         if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1061                 delta = diff_delta(last->data.buf, last->data.len,
1062                         dat->buf, dat->len,
1063                         &deltalen, dat->len - 20);
1064         } else
1065                 delta = NULL;
1066
1067         memset(&s, 0, sizeof(s));
1068         deflateInit(&s, pack_compression_level);
1069         if (delta) {
1070                 s.next_in = delta;
1071                 s.avail_in = deltalen;
1072         } else {
1073                 s.next_in = (void *)dat->buf;
1074                 s.avail_in = dat->len;
1075         }
1076         s.avail_out = deflateBound(&s, s.avail_in);
1077         s.next_out = out = xmalloc(s.avail_out);
1078         while (deflate(&s, Z_FINISH) == Z_OK)
1079                 /* nothing */;
1080         deflateEnd(&s);
1081
1082         /* Determine if we should auto-checkpoint. */
1083         if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1084                 || (pack_size + 60 + s.total_out) < pack_size) {
1085
1086                 /* This new object needs to *not* have the current pack_id. */
1087                 e->pack_id = pack_id + 1;
1088                 cycle_packfile();
1089
1090                 /* We cannot carry a delta into the new pack. */
1091                 if (delta) {
1092                         free(delta);
1093                         delta = NULL;
1094
1095                         memset(&s, 0, sizeof(s));
1096                         deflateInit(&s, pack_compression_level);
1097                         s.next_in = (void *)dat->buf;
1098                         s.avail_in = dat->len;
1099                         s.avail_out = deflateBound(&s, s.avail_in);
1100                         s.next_out = out = xrealloc(out, s.avail_out);
1101                         while (deflate(&s, Z_FINISH) == Z_OK)
1102                                 /* nothing */;
1103                         deflateEnd(&s);
1104                 }
1105         }
1106
1107         e->type = type;
1108         e->pack_id = pack_id;
1109         e->idx.offset = pack_size;
1110         object_count++;
1111         object_count_by_type[type]++;
1112
1113         crc32_begin(pack_file);
1114
1115         if (delta) {
1116                 off_t ofs = e->idx.offset - last->offset;
1117                 unsigned pos = sizeof(hdr) - 1;
1118
1119                 delta_count_by_type[type]++;
1120                 e->depth = last->depth + 1;
1121
1122                 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1123                 sha1write(pack_file, hdr, hdrlen);
1124                 pack_size += hdrlen;
1125
1126                 hdr[pos] = ofs & 127;
1127                 while (ofs >>= 7)
1128                         hdr[--pos] = 128 | (--ofs & 127);
1129                 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1130                 pack_size += sizeof(hdr) - pos;
1131         } else {
1132                 e->depth = 0;
1133                 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1134                 sha1write(pack_file, hdr, hdrlen);
1135                 pack_size += hdrlen;
1136         }
1137
1138         sha1write(pack_file, out, s.total_out);
1139         pack_size += s.total_out;
1140
1141         e->idx.crc32 = crc32_end(pack_file);
1142
1143         free(out);
1144         free(delta);
1145         if (last) {
1146                 if (last->no_swap) {
1147                         last->data = *dat;
1148                 } else {
1149                         strbuf_swap(&last->data, dat);
1150                 }
1151                 last->offset = e->idx.offset;
1152                 last->depth = e->depth;
1153         }
1154         return 0;
1155 }
1156
1157 static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1158 {
1159         if (ftruncate(pack_data->pack_fd, to)
1160          || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1161                 die_errno("cannot truncate pack to skip duplicate");
1162         pack_size = to;
1163
1164         /* yes this is a layering violation */
1165         pack_file->total = to;
1166         pack_file->offset = 0;
1167         pack_file->ctx = *ctx;
1168 }
1169
1170 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1171 {
1172         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1173         unsigned char *in_buf = xmalloc(in_sz);
1174         unsigned char *out_buf = xmalloc(out_sz);
1175         struct object_entry *e;
1176         unsigned char sha1[20];
1177         unsigned long hdrlen;
1178         off_t offset;
1179         git_SHA_CTX c;
1180         git_SHA_CTX pack_file_ctx;
1181         z_stream s;
1182         int status = Z_OK;
1183
1184         /* Determine if we should auto-checkpoint. */
1185         if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1186                 || (pack_size + 60 + len) < pack_size)
1187                 cycle_packfile();
1188
1189         offset = pack_size;
1190
1191         /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1192         sha1flush(pack_file);
1193         pack_file_ctx = pack_file->ctx;
1194
1195         hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1196         if (out_sz <= hdrlen)
1197                 die("impossibly large object header");
1198
1199         git_SHA1_Init(&c);
1200         git_SHA1_Update(&c, out_buf, hdrlen);
1201
1202         crc32_begin(pack_file);
1203
1204         memset(&s, 0, sizeof(s));
1205         deflateInit(&s, pack_compression_level);
1206
1207         hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1208         if (out_sz <= hdrlen)
1209                 die("impossibly large object header");
1210
1211         s.next_out = out_buf + hdrlen;
1212         s.avail_out = out_sz - hdrlen;
1213
1214         while (status != Z_STREAM_END) {
1215                 if (0 < len && !s.avail_in) {
1216                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1217                         size_t n = fread(in_buf, 1, cnt, stdin);
1218                         if (!n && feof(stdin))
1219                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1220
1221                         git_SHA1_Update(&c, in_buf, n);
1222                         s.next_in = in_buf;
1223                         s.avail_in = n;
1224                         len -= n;
1225                 }
1226
1227                 status = deflate(&s, len ? 0 : Z_FINISH);
1228
1229                 if (!s.avail_out || status == Z_STREAM_END) {
1230                         size_t n = s.next_out - out_buf;
1231                         sha1write(pack_file, out_buf, n);
1232                         pack_size += n;
1233                         s.next_out = out_buf;
1234                         s.avail_out = out_sz;
1235                 }
1236
1237                 switch (status) {
1238                 case Z_OK:
1239                 case Z_BUF_ERROR:
1240                 case Z_STREAM_END:
1241                         continue;
1242                 default:
1243                         die("unexpected deflate failure: %d", status);
1244                 }
1245         }
1246         deflateEnd(&s);
1247         git_SHA1_Final(sha1, &c);
1248
1249         if (sha1out)
1250                 hashcpy(sha1out, sha1);
1251
1252         e = insert_object(sha1);
1253
1254         if (mark)
1255                 insert_mark(mark, e);
1256
1257         if (e->idx.offset) {
1258                 duplicate_count_by_type[OBJ_BLOB]++;
1259                 truncate_pack(offset, &pack_file_ctx);
1260
1261         } else if (find_sha1_pack(sha1, packed_git)) {
1262                 e->type = OBJ_BLOB;
1263                 e->pack_id = MAX_PACK_ID;
1264                 e->idx.offset = 1; /* just not zero! */
1265                 duplicate_count_by_type[OBJ_BLOB]++;
1266                 truncate_pack(offset, &pack_file_ctx);
1267
1268         } else {
1269                 e->depth = 0;
1270                 e->type = OBJ_BLOB;
1271                 e->pack_id = pack_id;
1272                 e->idx.offset = offset;
1273                 e->idx.crc32 = crc32_end(pack_file);
1274                 object_count++;
1275                 object_count_by_type[OBJ_BLOB]++;
1276         }
1277
1278         free(in_buf);
1279         free(out_buf);
1280 }
1281
1282 /* All calls must be guarded by find_object() or find_mark() to
1283  * ensure the 'struct object_entry' passed was written by this
1284  * process instance.  We unpack the entry by the offset, avoiding
1285  * the need for the corresponding .idx file.  This unpacking rule
1286  * works because we only use OBJ_REF_DELTA within the packfiles
1287  * created by fast-import.
1288  *
1289  * oe must not be NULL.  Such an oe usually comes from giving
1290  * an unknown SHA-1 to find_object() or an undefined mark to
1291  * find_mark().  Callers must test for this condition and use
1292  * the standard read_sha1_file() when it happens.
1293  *
1294  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1295  * find_mark(), where the mark was reloaded from an existing marks
1296  * file and is referencing an object that this fast-import process
1297  * instance did not write out to a packfile.  Callers must test for
1298  * this condition and use read_sha1_file() instead.
1299  */
1300 static void *gfi_unpack_entry(
1301         struct object_entry *oe,
1302         unsigned long *sizep)
1303 {
1304         enum object_type type;
1305         struct packed_git *p = all_packs[oe->pack_id];
1306         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1307                 /* The object is stored in the packfile we are writing to
1308                  * and we have modified it since the last time we scanned
1309                  * back to read a previously written object.  If an old
1310                  * window covered [p->pack_size, p->pack_size + 20) its
1311                  * data is stale and is not valid.  Closing all windows
1312                  * and updating the packfile length ensures we can read
1313                  * the newly written data.
1314                  */
1315                 close_pack_windows(p);
1316                 sha1flush(pack_file);
1317
1318                 /* We have to offer 20 bytes additional on the end of
1319                  * the packfile as the core unpacker code assumes the
1320                  * footer is present at the file end and must promise
1321                  * at least 20 bytes within any window it maps.  But
1322                  * we don't actually create the footer here.
1323                  */
1324                 p->pack_size = pack_size + 20;
1325         }
1326         return unpack_entry(p, oe->idx.offset, &type, sizep);
1327 }
1328
1329 static const char *get_mode(const char *str, uint16_t *modep)
1330 {
1331         unsigned char c;
1332         uint16_t mode = 0;
1333
1334         while ((c = *str++) != ' ') {
1335                 if (c < '0' || c > '7')
1336                         return NULL;
1337                 mode = (mode << 3) + (c - '0');
1338         }
1339         *modep = mode;
1340         return str;
1341 }
1342
1343 static void load_tree(struct tree_entry *root)
1344 {
1345         unsigned char *sha1 = root->versions[1].sha1;
1346         struct object_entry *myoe;
1347         struct tree_content *t;
1348         unsigned long size;
1349         char *buf;
1350         const char *c;
1351
1352         root->tree = t = new_tree_content(8);
1353         if (is_null_sha1(sha1))
1354                 return;
1355
1356         myoe = find_object(sha1);
1357         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1358                 if (myoe->type != OBJ_TREE)
1359                         die("Not a tree: %s", sha1_to_hex(sha1));
1360                 t->delta_depth = myoe->depth;
1361                 buf = gfi_unpack_entry(myoe, &size);
1362                 if (!buf)
1363                         die("Can't load tree %s", sha1_to_hex(sha1));
1364         } else {
1365                 enum object_type type;
1366                 buf = read_sha1_file(sha1, &type, &size);
1367                 if (!buf || type != OBJ_TREE)
1368                         die("Can't load tree %s", sha1_to_hex(sha1));
1369         }
1370
1371         c = buf;
1372         while (c != (buf + size)) {
1373                 struct tree_entry *e = new_tree_entry();
1374
1375                 if (t->entry_count == t->entry_capacity)
1376                         root->tree = t = grow_tree_content(t, t->entry_count);
1377                 t->entries[t->entry_count++] = e;
1378
1379                 e->tree = NULL;
1380                 c = get_mode(c, &e->versions[1].mode);
1381                 if (!c)
1382                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1383                 e->versions[0].mode = e->versions[1].mode;
1384                 e->name = to_atom(c, strlen(c));
1385                 c += e->name->str_len + 1;
1386                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1387                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1388                 c += 20;
1389         }
1390         free(buf);
1391 }
1392
1393 static int tecmp0 (const void *_a, const void *_b)
1394 {
1395         struct tree_entry *a = *((struct tree_entry**)_a);
1396         struct tree_entry *b = *((struct tree_entry**)_b);
1397         return base_name_compare(
1398                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1399                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1400 }
1401
1402 static int tecmp1 (const void *_a, const void *_b)
1403 {
1404         struct tree_entry *a = *((struct tree_entry**)_a);
1405         struct tree_entry *b = *((struct tree_entry**)_b);
1406         return base_name_compare(
1407                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1408                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1409 }
1410
1411 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1412 {
1413         size_t maxlen = 0;
1414         unsigned int i;
1415
1416         if (!v)
1417                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1418         else
1419                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1420
1421         for (i = 0; i < t->entry_count; i++) {
1422                 if (t->entries[i]->versions[v].mode)
1423                         maxlen += t->entries[i]->name->str_len + 34;
1424         }
1425
1426         strbuf_reset(b);
1427         strbuf_grow(b, maxlen);
1428         for (i = 0; i < t->entry_count; i++) {
1429                 struct tree_entry *e = t->entries[i];
1430                 if (!e->versions[v].mode)
1431                         continue;
1432                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1433                                         e->name->str_dat, '\0');
1434                 strbuf_add(b, e->versions[v].sha1, 20);
1435         }
1436 }
1437
1438 static void store_tree(struct tree_entry *root)
1439 {
1440         struct tree_content *t = root->tree;
1441         unsigned int i, j, del;
1442         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1443         struct object_entry *le;
1444
1445         if (!is_null_sha1(root->versions[1].sha1))
1446                 return;
1447
1448         for (i = 0; i < t->entry_count; i++) {
1449                 if (t->entries[i]->tree)
1450                         store_tree(t->entries[i]);
1451         }
1452
1453         le = find_object(root->versions[0].sha1);
1454         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1455                 mktree(t, 0, &old_tree);
1456                 lo.data = old_tree;
1457                 lo.offset = le->idx.offset;
1458                 lo.depth = t->delta_depth;
1459         }
1460
1461         mktree(t, 1, &new_tree);
1462         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1463
1464         t->delta_depth = lo.depth;
1465         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1466                 struct tree_entry *e = t->entries[i];
1467                 if (e->versions[1].mode) {
1468                         e->versions[0].mode = e->versions[1].mode;
1469                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1470                         t->entries[j++] = e;
1471                 } else {
1472                         release_tree_entry(e);
1473                         del++;
1474                 }
1475         }
1476         t->entry_count -= del;
1477 }
1478
1479 static void tree_content_replace(
1480         struct tree_entry *root,
1481         const unsigned char *sha1,
1482         const uint16_t mode,
1483         struct tree_content *newtree)
1484 {
1485         if (!S_ISDIR(mode))
1486                 die("Root cannot be a non-directory");
1487         hashcpy(root->versions[1].sha1, sha1);
1488         if (root->tree)
1489                 release_tree_content_recursive(root->tree);
1490         root->tree = newtree;
1491 }
1492
1493 static int tree_content_set(
1494         struct tree_entry *root,
1495         const char *p,
1496         const unsigned char *sha1,
1497         const uint16_t mode,
1498         struct tree_content *subtree)
1499 {
1500         struct tree_content *t;
1501         const char *slash1;
1502         unsigned int i, n;
1503         struct tree_entry *e;
1504
1505         slash1 = strchr(p, '/');
1506         if (slash1)
1507                 n = slash1 - p;
1508         else
1509                 n = strlen(p);
1510         if (!n)
1511                 die("Empty path component found in input");
1512         if (!slash1 && !S_ISDIR(mode) && subtree)
1513                 die("Non-directories cannot have subtrees");
1514
1515         if (!root->tree)
1516                 load_tree(root);
1517         t = root->tree;
1518         for (i = 0; i < t->entry_count; i++) {
1519                 e = t->entries[i];
1520                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1521                         if (!slash1) {
1522                                 if (!S_ISDIR(mode)
1523                                                 && e->versions[1].mode == mode
1524                                                 && !hashcmp(e->versions[1].sha1, sha1))
1525                                         return 0;
1526                                 e->versions[1].mode = mode;
1527                                 hashcpy(e->versions[1].sha1, sha1);
1528                                 if (e->tree)
1529                                         release_tree_content_recursive(e->tree);
1530                                 e->tree = subtree;
1531                                 hashclr(root->versions[1].sha1);
1532                                 return 1;
1533                         }
1534                         if (!S_ISDIR(e->versions[1].mode)) {
1535                                 e->tree = new_tree_content(8);
1536                                 e->versions[1].mode = S_IFDIR;
1537                         }
1538                         if (!e->tree)
1539                                 load_tree(e);
1540                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1541                                 hashclr(root->versions[1].sha1);
1542                                 return 1;
1543                         }
1544                         return 0;
1545                 }
1546         }
1547
1548         if (t->entry_count == t->entry_capacity)
1549                 root->tree = t = grow_tree_content(t, t->entry_count);
1550         e = new_tree_entry();
1551         e->name = to_atom(p, n);
1552         e->versions[0].mode = 0;
1553         hashclr(e->versions[0].sha1);
1554         t->entries[t->entry_count++] = e;
1555         if (slash1) {
1556                 e->tree = new_tree_content(8);
1557                 e->versions[1].mode = S_IFDIR;
1558                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1559         } else {
1560                 e->tree = subtree;
1561                 e->versions[1].mode = mode;
1562                 hashcpy(e->versions[1].sha1, sha1);
1563         }
1564         hashclr(root->versions[1].sha1);
1565         return 1;
1566 }
1567
1568 static int tree_content_remove(
1569         struct tree_entry *root,
1570         const char *p,
1571         struct tree_entry *backup_leaf)
1572 {
1573         struct tree_content *t;
1574         const char *slash1;
1575         unsigned int i, n;
1576         struct tree_entry *e;
1577
1578         slash1 = strchr(p, '/');
1579         if (slash1)
1580                 n = slash1 - p;
1581         else
1582                 n = strlen(p);
1583
1584         if (!root->tree)
1585                 load_tree(root);
1586         t = root->tree;
1587         for (i = 0; i < t->entry_count; i++) {
1588                 e = t->entries[i];
1589                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1590                         if (slash1 && !S_ISDIR(e->versions[1].mode))
1591                                 /*
1592                                  * If p names a file in some subdirectory, and a
1593                                  * file or symlink matching the name of the
1594                                  * parent directory of p exists, then p cannot
1595                                  * exist and need not be deleted.
1596                                  */
1597                                 return 1;
1598                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1599                                 goto del_entry;
1600                         if (!e->tree)
1601                                 load_tree(e);
1602                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1603                                 for (n = 0; n < e->tree->entry_count; n++) {
1604                                         if (e->tree->entries[n]->versions[1].mode) {
1605                                                 hashclr(root->versions[1].sha1);
1606                                                 return 1;
1607                                         }
1608                                 }
1609                                 backup_leaf = NULL;
1610                                 goto del_entry;
1611                         }
1612                         return 0;
1613                 }
1614         }
1615         return 0;
1616
1617 del_entry:
1618         if (backup_leaf)
1619                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1620         else if (e->tree)
1621                 release_tree_content_recursive(e->tree);
1622         e->tree = NULL;
1623         e->versions[1].mode = 0;
1624         hashclr(e->versions[1].sha1);
1625         hashclr(root->versions[1].sha1);
1626         return 1;
1627 }
1628
1629 static int tree_content_get(
1630         struct tree_entry *root,
1631         const char *p,
1632         struct tree_entry *leaf)
1633 {
1634         struct tree_content *t;
1635         const char *slash1;
1636         unsigned int i, n;
1637         struct tree_entry *e;
1638
1639         slash1 = strchr(p, '/');
1640         if (slash1)
1641                 n = slash1 - p;
1642         else
1643                 n = strlen(p);
1644
1645         if (!root->tree)
1646                 load_tree(root);
1647         t = root->tree;
1648         for (i = 0; i < t->entry_count; i++) {
1649                 e = t->entries[i];
1650                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1651                         if (!slash1) {
1652                                 memcpy(leaf, e, sizeof(*leaf));
1653                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1654                                         leaf->tree = dup_tree_content(e->tree);
1655                                 else
1656                                         leaf->tree = NULL;
1657                                 return 1;
1658                         }
1659                         if (!S_ISDIR(e->versions[1].mode))
1660                                 return 0;
1661                         if (!e->tree)
1662                                 load_tree(e);
1663                         return tree_content_get(e, slash1 + 1, leaf);
1664                 }
1665         }
1666         return 0;
1667 }
1668
1669 static int update_branch(struct branch *b)
1670 {
1671         static const char *msg = "fast-import";
1672         struct ref_lock *lock;
1673         unsigned char old_sha1[20];
1674
1675         if (is_null_sha1(b->sha1))
1676                 return 0;
1677         if (read_ref(b->name, old_sha1))
1678                 hashclr(old_sha1);
1679         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1680         if (!lock)
1681                 return error("Unable to lock %s", b->name);
1682         if (!force_update && !is_null_sha1(old_sha1)) {
1683                 struct commit *old_cmit, *new_cmit;
1684
1685                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1686                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1687                 if (!old_cmit || !new_cmit) {
1688                         unlock_ref(lock);
1689                         return error("Branch %s is missing commits.", b->name);
1690                 }
1691
1692                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1693                         unlock_ref(lock);
1694                         warning("Not updating %s"
1695                                 " (new tip %s does not contain %s)",
1696                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1697                         return -1;
1698                 }
1699         }
1700         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1701                 return error("Unable to update %s", b->name);
1702         return 0;
1703 }
1704
1705 static void dump_branches(void)
1706 {
1707         unsigned int i;
1708         struct branch *b;
1709
1710         for (i = 0; i < branch_table_sz; i++) {
1711                 for (b = branch_table[i]; b; b = b->table_next_branch)
1712                         failure |= update_branch(b);
1713         }
1714 }
1715
1716 static void dump_tags(void)
1717 {
1718         static const char *msg = "fast-import";
1719         struct tag *t;
1720         struct ref_lock *lock;
1721         char ref_name[PATH_MAX];
1722
1723         for (t = first_tag; t; t = t->next_tag) {
1724                 sprintf(ref_name, "tags/%s", t->name);
1725                 lock = lock_ref_sha1(ref_name, NULL);
1726                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1727                         failure |= error("Unable to update %s", ref_name);
1728         }
1729 }
1730
1731 static void dump_marks_helper(FILE *f,
1732         uintmax_t base,
1733         struct mark_set *m)
1734 {
1735         uintmax_t k;
1736         if (m->shift) {
1737                 for (k = 0; k < 1024; k++) {
1738                         if (m->data.sets[k])
1739                                 dump_marks_helper(f, base + (k << m->shift),
1740                                         m->data.sets[k]);
1741                 }
1742         } else {
1743                 for (k = 0; k < 1024; k++) {
1744                         if (m->data.marked[k])
1745                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1746                                         sha1_to_hex(m->data.marked[k]->idx.sha1));
1747                 }
1748         }
1749 }
1750
1751 static void dump_marks(void)
1752 {
1753         static struct lock_file mark_lock;
1754         int mark_fd;
1755         FILE *f;
1756
1757         if (!export_marks_file)
1758                 return;
1759
1760         mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1761         if (mark_fd < 0) {
1762                 failure |= error("Unable to write marks file %s: %s",
1763                         export_marks_file, strerror(errno));
1764                 return;
1765         }
1766
1767         f = fdopen(mark_fd, "w");
1768         if (!f) {
1769                 int saved_errno = errno;
1770                 rollback_lock_file(&mark_lock);
1771                 failure |= error("Unable to write marks file %s: %s",
1772                         export_marks_file, strerror(saved_errno));
1773                 return;
1774         }
1775
1776         /*
1777          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1778          * Assign -1 to the lock file descriptor so that commit_lock_file()
1779          * won't try to close() it.
1780          */
1781         mark_lock.fd = -1;
1782
1783         dump_marks_helper(f, 0, marks);
1784         if (ferror(f) || fclose(f)) {
1785                 int saved_errno = errno;
1786                 rollback_lock_file(&mark_lock);
1787                 failure |= error("Unable to write marks file %s: %s",
1788                         export_marks_file, strerror(saved_errno));
1789                 return;
1790         }
1791
1792         if (commit_lock_file(&mark_lock)) {
1793                 int saved_errno = errno;
1794                 rollback_lock_file(&mark_lock);
1795                 failure |= error("Unable to commit marks file %s: %s",
1796                         export_marks_file, strerror(saved_errno));
1797                 return;
1798         }
1799 }
1800
1801 static void read_marks(void)
1802 {
1803         char line[512];
1804         FILE *f = fopen(import_marks_file, "r");
1805         if (f)
1806                 ;
1807         else if (import_marks_file_ignore_missing && errno == ENOENT)
1808                 return; /* Marks file does not exist */
1809         else
1810                 die_errno("cannot read '%s'", import_marks_file);
1811         while (fgets(line, sizeof(line), f)) {
1812                 uintmax_t mark;
1813                 char *end;
1814                 unsigned char sha1[20];
1815                 struct object_entry *e;
1816
1817                 end = strchr(line, '\n');
1818                 if (line[0] != ':' || !end)
1819                         die("corrupt mark line: %s", line);
1820                 *end = 0;
1821                 mark = strtoumax(line + 1, &end, 10);
1822                 if (!mark || end == line + 1
1823                         || *end != ' ' || get_sha1(end + 1, sha1))
1824                         die("corrupt mark line: %s", line);
1825                 e = find_object(sha1);
1826                 if (!e) {
1827                         enum object_type type = sha1_object_info(sha1, NULL);
1828                         if (type < 0)
1829                                 die("object not found: %s", sha1_to_hex(sha1));
1830                         e = insert_object(sha1);
1831                         e->type = type;
1832                         e->pack_id = MAX_PACK_ID;
1833                         e->idx.offset = 1; /* just not zero! */
1834                 }
1835                 insert_mark(mark, e);
1836         }
1837         fclose(f);
1838 }
1839
1840
1841 static int read_next_command(void)
1842 {
1843         static int stdin_eof = 0;
1844
1845         if (stdin_eof) {
1846                 unread_command_buf = 0;
1847                 return EOF;
1848         }
1849
1850         for (;;) {
1851                 if (unread_command_buf) {
1852                         unread_command_buf = 0;
1853                 } else {
1854                         struct recent_command *rc;
1855
1856                         strbuf_detach(&command_buf, NULL);
1857                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1858                         if (stdin_eof)
1859                                 return EOF;
1860
1861                         if (!seen_data_command
1862                                 && prefixcmp(command_buf.buf, "feature ")
1863                                 && prefixcmp(command_buf.buf, "option ")) {
1864                                 parse_argv();
1865                         }
1866
1867                         rc = rc_free;
1868                         if (rc)
1869                                 rc_free = rc->next;
1870                         else {
1871                                 rc = cmd_hist.next;
1872                                 cmd_hist.next = rc->next;
1873                                 cmd_hist.next->prev = &cmd_hist;
1874                                 free(rc->buf);
1875                         }
1876
1877                         rc->buf = command_buf.buf;
1878                         rc->prev = cmd_tail;
1879                         rc->next = cmd_hist.prev;
1880                         rc->prev->next = rc;
1881                         cmd_tail = rc;
1882                 }
1883                 if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1884                         parse_cat_blob();
1885                         continue;
1886                 }
1887                 if (command_buf.buf[0] == '#')
1888                         continue;
1889                 return 0;
1890         }
1891 }
1892
1893 static void skip_optional_lf(void)
1894 {
1895         int term_char = fgetc(stdin);
1896         if (term_char != '\n' && term_char != EOF)
1897                 ungetc(term_char, stdin);
1898 }
1899
1900 static void parse_mark(void)
1901 {
1902         if (!prefixcmp(command_buf.buf, "mark :")) {
1903                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1904                 read_next_command();
1905         }
1906         else
1907                 next_mark = 0;
1908 }
1909
1910 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1911 {
1912         strbuf_reset(sb);
1913
1914         if (prefixcmp(command_buf.buf, "data "))
1915                 die("Expected 'data n' command, found: %s", command_buf.buf);
1916
1917         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1918                 char *term = xstrdup(command_buf.buf + 5 + 2);
1919                 size_t term_len = command_buf.len - 5 - 2;
1920
1921                 strbuf_detach(&command_buf, NULL);
1922                 for (;;) {
1923                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1924                                 die("EOF in data (terminator '%s' not found)", term);
1925                         if (term_len == command_buf.len
1926                                 && !strcmp(term, command_buf.buf))
1927                                 break;
1928                         strbuf_addbuf(sb, &command_buf);
1929                         strbuf_addch(sb, '\n');
1930                 }
1931                 free(term);
1932         }
1933         else {
1934                 uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1935                 size_t n = 0, length = (size_t)len;
1936
1937                 if (limit && limit < len) {
1938                         *len_res = len;
1939                         return 0;
1940                 }
1941                 if (length < len)
1942                         die("data is too large to use in this context");
1943
1944                 while (n < length) {
1945                         size_t s = strbuf_fread(sb, length - n, stdin);
1946                         if (!s && feof(stdin))
1947                                 die("EOF in data (%lu bytes remaining)",
1948                                         (unsigned long)(length - n));
1949                         n += s;
1950                 }
1951         }
1952
1953         skip_optional_lf();
1954         return 1;
1955 }
1956
1957 static int validate_raw_date(const char *src, char *result, int maxlen)
1958 {
1959         const char *orig_src = src;
1960         char *endp;
1961         unsigned long num;
1962
1963         errno = 0;
1964
1965         num = strtoul(src, &endp, 10);
1966         /* NEEDSWORK: perhaps check for reasonable values? */
1967         if (errno || endp == src || *endp != ' ')
1968                 return -1;
1969
1970         src = endp + 1;
1971         if (*src != '-' && *src != '+')
1972                 return -1;
1973
1974         num = strtoul(src + 1, &endp, 10);
1975         if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1976             1400 < num)
1977                 return -1;
1978
1979         strcpy(result, orig_src);
1980         return 0;
1981 }
1982
1983 static char *parse_ident(const char *buf)
1984 {
1985         const char *gt;
1986         size_t name_len;
1987         char *ident;
1988
1989         gt = strrchr(buf, '>');
1990         if (!gt)
1991                 die("Missing > in ident string: %s", buf);
1992         gt++;
1993         if (*gt != ' ')
1994                 die("Missing space after > in ident string: %s", buf);
1995         gt++;
1996         name_len = gt - buf;
1997         ident = xmalloc(name_len + 24);
1998         strncpy(ident, buf, name_len);
1999
2000         switch (whenspec) {
2001         case WHENSPEC_RAW:
2002                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
2003                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
2004                 break;
2005         case WHENSPEC_RFC2822:
2006                 if (parse_date(gt, ident + name_len, 24) < 0)
2007                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
2008                 break;
2009         case WHENSPEC_NOW:
2010                 if (strcmp("now", gt))
2011                         die("Date in ident must be 'now': %s", buf);
2012                 datestamp(ident + name_len, 24);
2013                 break;
2014         }
2015
2016         return ident;
2017 }
2018
2019 static void parse_and_store_blob(
2020         struct last_object *last,
2021         unsigned char *sha1out,
2022         uintmax_t mark)
2023 {
2024         static struct strbuf buf = STRBUF_INIT;
2025         uintmax_t len;
2026
2027         if (parse_data(&buf, big_file_threshold, &len))
2028                 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2029         else {
2030                 if (last) {
2031                         strbuf_release(&last->data);
2032                         last->offset = 0;
2033                         last->depth = 0;
2034                 }
2035                 stream_blob(len, sha1out, mark);
2036                 skip_optional_lf();
2037         }
2038 }
2039
2040 static void parse_new_blob(void)
2041 {
2042         read_next_command();
2043         parse_mark();
2044         parse_and_store_blob(&last_blob, NULL, next_mark);
2045 }
2046
2047 static void unload_one_branch(void)
2048 {
2049         while (cur_active_branches
2050                 && cur_active_branches >= max_active_branches) {
2051                 uintmax_t min_commit = ULONG_MAX;
2052                 struct branch *e, *l = NULL, *p = NULL;
2053
2054                 for (e = active_branches; e; e = e->active_next_branch) {
2055                         if (e->last_commit < min_commit) {
2056                                 p = l;
2057                                 min_commit = e->last_commit;
2058                         }
2059                         l = e;
2060                 }
2061
2062                 if (p) {
2063                         e = p->active_next_branch;
2064                         p->active_next_branch = e->active_next_branch;
2065                 } else {
2066                         e = active_branches;
2067                         active_branches = e->active_next_branch;
2068                 }
2069                 e->active = 0;
2070                 e->active_next_branch = NULL;
2071                 if (e->branch_tree.tree) {
2072                         release_tree_content_recursive(e->branch_tree.tree);
2073                         e->branch_tree.tree = NULL;
2074                 }
2075                 cur_active_branches--;
2076         }
2077 }
2078
2079 static void load_branch(struct branch *b)
2080 {
2081         load_tree(&b->branch_tree);
2082         if (!b->active) {
2083                 b->active = 1;
2084                 b->active_next_branch = active_branches;
2085                 active_branches = b;
2086                 cur_active_branches++;
2087                 branch_load_count++;
2088         }
2089 }
2090
2091 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2092 {
2093         unsigned char fanout = 0;
2094         while ((num_notes >>= 8))
2095                 fanout++;
2096         return fanout;
2097 }
2098
2099 static void construct_path_with_fanout(const char *hex_sha1,
2100                 unsigned char fanout, char *path)
2101 {
2102         unsigned int i = 0, j = 0;
2103         if (fanout >= 20)
2104                 die("Too large fanout (%u)", fanout);
2105         while (fanout) {
2106                 path[i++] = hex_sha1[j++];
2107                 path[i++] = hex_sha1[j++];
2108                 path[i++] = '/';
2109                 fanout--;
2110         }
2111         memcpy(path + i, hex_sha1 + j, 40 - j);
2112         path[i + 40 - j] = '\0';
2113 }
2114
2115 static uintmax_t do_change_note_fanout(
2116                 struct tree_entry *orig_root, struct tree_entry *root,
2117                 char *hex_sha1, unsigned int hex_sha1_len,
2118                 char *fullpath, unsigned int fullpath_len,
2119                 unsigned char fanout)
2120 {
2121         struct tree_content *t = root->tree;
2122         struct tree_entry *e, leaf;
2123         unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2124         uintmax_t num_notes = 0;
2125         unsigned char sha1[20];
2126         char realpath[60];
2127
2128         for (i = 0; t && i < t->entry_count; i++) {
2129                 e = t->entries[i];
2130                 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2131                 tmp_fullpath_len = fullpath_len;
2132
2133                 /*
2134                  * We're interested in EITHER existing note entries (entries
2135                  * with exactly 40 hex chars in path, not including directory
2136                  * separators), OR directory entries that may contain note
2137                  * entries (with < 40 hex chars in path).
2138                  * Also, each path component in a note entry must be a multiple
2139                  * of 2 chars.
2140                  */
2141                 if (!e->versions[1].mode ||
2142                     tmp_hex_sha1_len > 40 ||
2143                     e->name->str_len % 2)
2144                         continue;
2145
2146                 /* This _may_ be a note entry, or a subdir containing notes */
2147                 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2148                        e->name->str_len);
2149                 if (tmp_fullpath_len)
2150                         fullpath[tmp_fullpath_len++] = '/';
2151                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2152                        e->name->str_len);
2153                 tmp_fullpath_len += e->name->str_len;
2154                 fullpath[tmp_fullpath_len] = '\0';
2155
2156                 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2157                         /* This is a note entry */
2158                         construct_path_with_fanout(hex_sha1, fanout, realpath);
2159                         if (!strcmp(fullpath, realpath)) {
2160                                 /* Note entry is in correct location */
2161                                 num_notes++;
2162                                 continue;
2163                         }
2164
2165                         /* Rename fullpath to realpath */
2166                         if (!tree_content_remove(orig_root, fullpath, &leaf))
2167                                 die("Failed to remove path %s", fullpath);
2168                         tree_content_set(orig_root, realpath,
2169                                 leaf.versions[1].sha1,
2170                                 leaf.versions[1].mode,
2171                                 leaf.tree);
2172                 } else if (S_ISDIR(e->versions[1].mode)) {
2173                         /* This is a subdir that may contain note entries */
2174                         if (!e->tree)
2175                                 load_tree(e);
2176                         num_notes += do_change_note_fanout(orig_root, e,
2177                                 hex_sha1, tmp_hex_sha1_len,
2178                                 fullpath, tmp_fullpath_len, fanout);
2179                 }
2180
2181                 /* The above may have reallocated the current tree_content */
2182                 t = root->tree;
2183         }
2184         return num_notes;
2185 }
2186
2187 static uintmax_t change_note_fanout(struct tree_entry *root,
2188                 unsigned char fanout)
2189 {
2190         char hex_sha1[40], path[60];
2191         return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2192 }
2193
2194 static void file_change_m(struct branch *b)
2195 {
2196         const char *p = command_buf.buf + 2;
2197         static struct strbuf uq = STRBUF_INIT;
2198         const char *endp;
2199         struct object_entry *oe = oe;
2200         unsigned char sha1[20];
2201         uint16_t mode, inline_data = 0;
2202
2203         p = get_mode(p, &mode);
2204         if (!p)
2205                 die("Corrupt mode: %s", command_buf.buf);
2206         switch (mode) {
2207         case 0644:
2208         case 0755:
2209                 mode |= S_IFREG;
2210         case S_IFREG | 0644:
2211         case S_IFREG | 0755:
2212         case S_IFLNK:
2213         case S_IFDIR:
2214         case S_IFGITLINK:
2215                 /* ok */
2216                 break;
2217         default:
2218                 die("Corrupt mode: %s", command_buf.buf);
2219         }
2220
2221         if (*p == ':') {
2222                 char *x;
2223                 oe = find_mark(strtoumax(p + 1, &x, 10));
2224                 hashcpy(sha1, oe->idx.sha1);
2225                 p = x;
2226         } else if (!prefixcmp(p, "inline")) {
2227                 inline_data = 1;
2228                 p += 6;
2229         } else {
2230                 if (get_sha1_hex(p, sha1))
2231                         die("Invalid SHA1: %s", command_buf.buf);
2232                 oe = find_object(sha1);
2233                 p += 40;
2234         }
2235         if (*p++ != ' ')
2236                 die("Missing space after SHA1: %s", command_buf.buf);
2237
2238         strbuf_reset(&uq);
2239         if (!unquote_c_style(&uq, p, &endp)) {
2240                 if (*endp)
2241                         die("Garbage after path in: %s", command_buf.buf);
2242                 p = uq.buf;
2243         }
2244
2245         /* Git does not track empty, non-toplevel directories. */
2246         if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2247                 tree_content_remove(&b->branch_tree, p, NULL);
2248                 return;
2249         }
2250
2251         if (S_ISGITLINK(mode)) {
2252                 if (inline_data)
2253                         die("Git links cannot be specified 'inline': %s",
2254                                 command_buf.buf);
2255                 else if (oe) {
2256                         if (oe->type != OBJ_COMMIT)
2257                                 die("Not a commit (actually a %s): %s",
2258                                         typename(oe->type), command_buf.buf);
2259                 }
2260                 /*
2261                  * Accept the sha1 without checking; it expected to be in
2262                  * another repository.
2263                  */
2264         } else if (inline_data) {
2265                 if (S_ISDIR(mode))
2266                         die("Directories cannot be specified 'inline': %s",
2267                                 command_buf.buf);
2268                 if (p != uq.buf) {
2269                         strbuf_addstr(&uq, p);
2270                         p = uq.buf;
2271                 }
2272                 read_next_command();
2273                 parse_and_store_blob(&last_blob, sha1, 0);
2274         } else {
2275                 enum object_type expected = S_ISDIR(mode) ?
2276                                                 OBJ_TREE: OBJ_BLOB;
2277                 enum object_type type = oe ? oe->type :
2278                                         sha1_object_info(sha1, NULL);
2279                 if (type < 0)
2280                         die("%s not found: %s",
2281                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2282                                         command_buf.buf);
2283                 if (type != expected)
2284                         die("Not a %s (actually a %s): %s",
2285                                 typename(expected), typename(type),
2286                                 command_buf.buf);
2287         }
2288
2289         if (!*p) {
2290                 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2291                 return;
2292         }
2293         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2294 }
2295
2296 static void file_change_d(struct branch *b)
2297 {
2298         const char *p = command_buf.buf + 2;
2299         static struct strbuf uq = STRBUF_INIT;
2300         const char *endp;
2301
2302         strbuf_reset(&uq);
2303         if (!unquote_c_style(&uq, p, &endp)) {
2304                 if (*endp)
2305                         die("Garbage after path in: %s", command_buf.buf);
2306                 p = uq.buf;
2307         }
2308         tree_content_remove(&b->branch_tree, p, NULL);
2309 }
2310
2311 static void file_change_cr(struct branch *b, int rename)
2312 {
2313         const char *s, *d;
2314         static struct strbuf s_uq = STRBUF_INIT;
2315         static struct strbuf d_uq = STRBUF_INIT;
2316         const char *endp;
2317         struct tree_entry leaf;
2318
2319         s = command_buf.buf + 2;
2320         strbuf_reset(&s_uq);
2321         if (!unquote_c_style(&s_uq, s, &endp)) {
2322                 if (*endp != ' ')
2323                         die("Missing space after source: %s", command_buf.buf);
2324         } else {
2325                 endp = strchr(s, ' ');
2326                 if (!endp)
2327                         die("Missing space after source: %s", command_buf.buf);
2328                 strbuf_add(&s_uq, s, endp - s);
2329         }
2330         s = s_uq.buf;
2331
2332         endp++;
2333         if (!*endp)
2334                 die("Missing dest: %s", command_buf.buf);
2335
2336         d = endp;
2337         strbuf_reset(&d_uq);
2338         if (!unquote_c_style(&d_uq, d, &endp)) {
2339                 if (*endp)
2340                         die("Garbage after dest in: %s", command_buf.buf);
2341                 d = d_uq.buf;
2342         }
2343
2344         memset(&leaf, 0, sizeof(leaf));
2345         if (rename)
2346                 tree_content_remove(&b->branch_tree, s, &leaf);
2347         else
2348                 tree_content_get(&b->branch_tree, s, &leaf);
2349         if (!leaf.versions[1].mode)
2350                 die("Path %s not in branch", s);
2351         if (!*d) {      /* C "path/to/subdir" "" */
2352                 tree_content_replace(&b->branch_tree,
2353                         leaf.versions[1].sha1,
2354                         leaf.versions[1].mode,
2355                         leaf.tree);
2356                 return;
2357         }
2358         tree_content_set(&b->branch_tree, d,
2359                 leaf.versions[1].sha1,
2360                 leaf.versions[1].mode,
2361                 leaf.tree);
2362 }
2363
2364 static void note_change_n(struct branch *b, unsigned char old_fanout)
2365 {
2366         const char *p = command_buf.buf + 2;
2367         static struct strbuf uq = STRBUF_INIT;
2368         struct object_entry *oe = oe;
2369         struct branch *s;
2370         unsigned char sha1[20], commit_sha1[20];
2371         char path[60];
2372         uint16_t inline_data = 0;
2373         unsigned char new_fanout;
2374
2375         /* <dataref> or 'inline' */
2376         if (*p == ':') {
2377                 char *x;
2378                 oe = find_mark(strtoumax(p + 1, &x, 10));
2379                 hashcpy(sha1, oe->idx.sha1);
2380                 p = x;
2381         } else if (!prefixcmp(p, "inline")) {
2382                 inline_data = 1;
2383                 p += 6;
2384         } else {
2385                 if (get_sha1_hex(p, sha1))
2386                         die("Invalid SHA1: %s", command_buf.buf);
2387                 oe = find_object(sha1);
2388                 p += 40;
2389         }
2390         if (*p++ != ' ')
2391                 die("Missing space after SHA1: %s", command_buf.buf);
2392
2393         /* <committish> */
2394         s = lookup_branch(p);
2395         if (s) {
2396                 hashcpy(commit_sha1, s->sha1);
2397         } else if (*p == ':') {
2398                 uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2399                 struct object_entry *commit_oe = find_mark(commit_mark);
2400                 if (commit_oe->type != OBJ_COMMIT)
2401                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2402                 hashcpy(commit_sha1, commit_oe->idx.sha1);
2403         } else if (!get_sha1(p, commit_sha1)) {
2404                 unsigned long size;
2405                 char *buf = read_object_with_reference(commit_sha1,
2406                         commit_type, &size, commit_sha1);
2407                 if (!buf || size < 46)
2408                         die("Not a valid commit: %s", p);
2409                 free(buf);
2410         } else
2411                 die("Invalid ref name or SHA1 expression: %s", p);
2412
2413         if (inline_data) {
2414                 if (p != uq.buf) {
2415                         strbuf_addstr(&uq, p);
2416                         p = uq.buf;
2417                 }
2418                 read_next_command();
2419                 parse_and_store_blob(&last_blob, sha1, 0);
2420         } else if (oe) {
2421                 if (oe->type != OBJ_BLOB)
2422                         die("Not a blob (actually a %s): %s",
2423                                 typename(oe->type), command_buf.buf);
2424         } else if (!is_null_sha1(sha1)) {
2425                 enum object_type type = sha1_object_info(sha1, NULL);
2426                 if (type < 0)
2427                         die("Blob not found: %s", command_buf.buf);
2428                 if (type != OBJ_BLOB)
2429                         die("Not a blob (actually a %s): %s",
2430                             typename(type), command_buf.buf);
2431         }
2432
2433         construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
2434         if (tree_content_remove(&b->branch_tree, path, NULL))
2435                 b->num_notes--;
2436
2437         if (is_null_sha1(sha1))
2438                 return; /* nothing to insert */
2439
2440         b->num_notes++;
2441         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2442         construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2443         tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2444 }
2445
2446 static void file_change_deleteall(struct branch *b)
2447 {
2448         release_tree_content_recursive(b->branch_tree.tree);
2449         hashclr(b->branch_tree.versions[0].sha1);
2450         hashclr(b->branch_tree.versions[1].sha1);
2451         load_tree(&b->branch_tree);
2452         b->num_notes = 0;
2453 }
2454
2455 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2456 {
2457         if (!buf || size < 46)
2458                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2459         if (memcmp("tree ", buf, 5)
2460                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2461                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2462         hashcpy(b->branch_tree.versions[0].sha1,
2463                 b->branch_tree.versions[1].sha1);
2464 }
2465
2466 static void parse_from_existing(struct branch *b)
2467 {
2468         if (is_null_sha1(b->sha1)) {
2469                 hashclr(b->branch_tree.versions[0].sha1);
2470                 hashclr(b->branch_tree.versions[1].sha1);
2471         } else {
2472                 unsigned long size;
2473                 char *buf;
2474
2475                 buf = read_object_with_reference(b->sha1,
2476                         commit_type, &size, b->sha1);
2477                 parse_from_commit(b, buf, size);
2478                 free(buf);
2479         }
2480 }
2481
2482 static int parse_from(struct branch *b)
2483 {
2484         const char *from;
2485         struct branch *s;
2486
2487         if (prefixcmp(command_buf.buf, "from "))
2488                 return 0;
2489
2490         if (b->branch_tree.tree) {
2491                 release_tree_content_recursive(b->branch_tree.tree);
2492                 b->branch_tree.tree = NULL;
2493         }
2494
2495         from = strchr(command_buf.buf, ' ') + 1;
2496         s = lookup_branch(from);
2497         if (b == s)
2498                 die("Can't create a branch from itself: %s", b->name);
2499         else if (s) {
2500                 unsigned char *t = s->branch_tree.versions[1].sha1;
2501                 hashcpy(b->sha1, s->sha1);
2502                 hashcpy(b->branch_tree.versions[0].sha1, t);
2503                 hashcpy(b->branch_tree.versions[1].sha1, t);
2504         } else if (*from == ':') {
2505                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2506                 struct object_entry *oe = find_mark(idnum);
2507                 if (oe->type != OBJ_COMMIT)
2508                         die("Mark :%" PRIuMAX " not a commit", idnum);
2509                 hashcpy(b->sha1, oe->idx.sha1);
2510                 if (oe->pack_id != MAX_PACK_ID) {
2511                         unsigned long size;
2512                         char *buf = gfi_unpack_entry(oe, &size);
2513                         parse_from_commit(b, buf, size);
2514                         free(buf);
2515                 } else
2516                         parse_from_existing(b);
2517         } else if (!get_sha1(from, b->sha1))
2518                 parse_from_existing(b);
2519         else
2520                 die("Invalid ref name or SHA1 expression: %s", from);
2521
2522         read_next_command();
2523         return 1;
2524 }
2525
2526 static struct hash_list *parse_merge(unsigned int *count)
2527 {
2528         struct hash_list *list = NULL, *n, *e = e;
2529         const char *from;
2530         struct branch *s;
2531
2532         *count = 0;
2533         while (!prefixcmp(command_buf.buf, "merge ")) {
2534                 from = strchr(command_buf.buf, ' ') + 1;
2535                 n = xmalloc(sizeof(*n));
2536                 s = lookup_branch(from);
2537                 if (s)
2538                         hashcpy(n->sha1, s->sha1);
2539                 else if (*from == ':') {
2540                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2541                         struct object_entry *oe = find_mark(idnum);
2542                         if (oe->type != OBJ_COMMIT)
2543                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2544                         hashcpy(n->sha1, oe->idx.sha1);
2545                 } else if (!get_sha1(from, n->sha1)) {
2546                         unsigned long size;
2547                         char *buf = read_object_with_reference(n->sha1,
2548                                 commit_type, &size, n->sha1);
2549                         if (!buf || size < 46)
2550                                 die("Not a valid commit: %s", from);
2551                         free(buf);
2552                 } else
2553                         die("Invalid ref name or SHA1 expression: %s", from);
2554
2555                 n->next = NULL;
2556                 if (list)
2557                         e->next = n;
2558                 else
2559                         list = n;
2560                 e = n;
2561                 (*count)++;
2562                 read_next_command();
2563         }
2564         return list;
2565 }
2566
2567 static void parse_new_commit(void)
2568 {
2569         static struct strbuf msg = STRBUF_INIT;
2570         struct branch *b;
2571         char *sp;
2572         char *author = NULL;
2573         char *committer = NULL;
2574         struct hash_list *merge_list = NULL;
2575         unsigned int merge_count;
2576         unsigned char prev_fanout, new_fanout;
2577
2578         /* Obtain the branch name from the rest of our command */
2579         sp = strchr(command_buf.buf, ' ') + 1;
2580         b = lookup_branch(sp);
2581         if (!b)
2582                 b = new_branch(sp);
2583
2584         read_next_command();
2585         parse_mark();
2586         if (!prefixcmp(command_buf.buf, "author ")) {
2587                 author = parse_ident(command_buf.buf + 7);
2588                 read_next_command();
2589         }
2590         if (!prefixcmp(command_buf.buf, "committer ")) {
2591                 committer = parse_ident(command_buf.buf + 10);
2592                 read_next_command();
2593         }
2594         if (!committer)
2595                 die("Expected committer but didn't get one");
2596         parse_data(&msg, 0, NULL);
2597         read_next_command();
2598         parse_from(b);
2599         merge_list = parse_merge(&merge_count);
2600
2601         /* ensure the branch is active/loaded */
2602         if (!b->branch_tree.tree || !max_active_branches) {
2603                 unload_one_branch();
2604                 load_branch(b);
2605         }
2606
2607         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2608
2609         /* file_change* */
2610         while (command_buf.len > 0) {
2611                 if (!prefixcmp(command_buf.buf, "M "))
2612                         file_change_m(b);
2613                 else if (!prefixcmp(command_buf.buf, "D "))
2614                         file_change_d(b);
2615                 else if (!prefixcmp(command_buf.buf, "R "))
2616                         file_change_cr(b, 1);
2617                 else if (!prefixcmp(command_buf.buf, "C "))
2618                         file_change_cr(b, 0);
2619                 else if (!prefixcmp(command_buf.buf, "N "))
2620                         note_change_n(b, prev_fanout);
2621                 else if (!strcmp("deleteall", command_buf.buf))
2622                         file_change_deleteall(b);
2623                 else if (!prefixcmp(command_buf.buf, "ls "))
2624                         parse_ls(b);
2625                 else {
2626                         unread_command_buf = 1;
2627                         break;
2628                 }
2629                 if (read_next_command() == EOF)
2630                         break;
2631         }
2632
2633         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2634         if (new_fanout != prev_fanout)
2635                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2636
2637         /* build the tree and the commit */
2638         store_tree(&b->branch_tree);
2639         hashcpy(b->branch_tree.versions[0].sha1,
2640                 b->branch_tree.versions[1].sha1);
2641
2642         strbuf_reset(&new_data);
2643         strbuf_addf(&new_data, "tree %s\n",
2644                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2645         if (!is_null_sha1(b->sha1))
2646                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2647         while (merge_list) {
2648                 struct hash_list *next = merge_list->next;
2649                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2650                 free(merge_list);
2651                 merge_list = next;
2652         }
2653         strbuf_addf(&new_data,
2654                 "author %s\n"
2655                 "committer %s\n"
2656                 "\n",
2657                 author ? author : committer, committer);
2658         strbuf_addbuf(&new_data, &msg);
2659         free(author);
2660         free(committer);
2661
2662         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2663                 b->pack_id = pack_id;
2664         b->last_commit = object_count_by_type[OBJ_COMMIT];
2665 }
2666
2667 static void parse_new_tag(void)
2668 {
2669         static struct strbuf msg = STRBUF_INIT;
2670         char *sp;
2671         const char *from;
2672         char *tagger;
2673         struct branch *s;
2674         struct tag *t;
2675         uintmax_t from_mark = 0;
2676         unsigned char sha1[20];
2677         enum object_type type;
2678
2679         /* Obtain the new tag name from the rest of our command */
2680         sp = strchr(command_buf.buf, ' ') + 1;
2681         t = pool_alloc(sizeof(struct tag));
2682         t->next_tag = NULL;
2683         t->name = pool_strdup(sp);
2684         if (last_tag)
2685                 last_tag->next_tag = t;
2686         else
2687                 first_tag = t;
2688         last_tag = t;
2689         read_next_command();
2690
2691         /* from ... */
2692         if (prefixcmp(command_buf.buf, "from "))
2693                 die("Expected from command, got %s", command_buf.buf);
2694         from = strchr(command_buf.buf, ' ') + 1;
2695         s = lookup_branch(from);
2696         if (s) {
2697                 hashcpy(sha1, s->sha1);
2698                 type = OBJ_COMMIT;
2699         } else if (*from == ':') {
2700                 struct object_entry *oe;
2701                 from_mark = strtoumax(from + 1, NULL, 10);
2702                 oe = find_mark(from_mark);
2703                 type = oe->type;
2704                 hashcpy(sha1, oe->idx.sha1);
2705         } else if (!get_sha1(from, sha1)) {
2706                 unsigned long size;
2707                 char *buf;
2708
2709                 buf = read_sha1_file(sha1, &type, &size);
2710                 if (!buf || size < 46)
2711                         die("Not a valid commit: %s", from);
2712                 free(buf);
2713         } else
2714                 die("Invalid ref name or SHA1 expression: %s", from);
2715         read_next_command();
2716
2717         /* tagger ... */
2718         if (!prefixcmp(command_buf.buf, "tagger ")) {
2719                 tagger = parse_ident(command_buf.buf + 7);
2720                 read_next_command();
2721         } else
2722                 tagger = NULL;
2723
2724         /* tag payload/message */
2725         parse_data(&msg, 0, NULL);
2726
2727         /* build the tag object */
2728         strbuf_reset(&new_data);
2729
2730         strbuf_addf(&new_data,
2731                     "object %s\n"
2732                     "type %s\n"
2733                     "tag %s\n",
2734                     sha1_to_hex(sha1), typename(type), t->name);
2735         if (tagger)
2736                 strbuf_addf(&new_data,
2737                             "tagger %s\n", tagger);
2738         strbuf_addch(&new_data, '\n');
2739         strbuf_addbuf(&new_data, &msg);
2740         free(tagger);
2741
2742         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2743                 t->pack_id = MAX_PACK_ID;
2744         else
2745                 t->pack_id = pack_id;
2746 }
2747
2748 static void parse_reset_branch(void)
2749 {
2750         struct branch *b;
2751         char *sp;
2752
2753         /* Obtain the branch name from the rest of our command */
2754         sp = strchr(command_buf.buf, ' ') + 1;
2755         b = lookup_branch(sp);
2756         if (b) {
2757                 hashclr(b->sha1);
2758                 hashclr(b->branch_tree.versions[0].sha1);
2759                 hashclr(b->branch_tree.versions[1].sha1);
2760                 if (b->branch_tree.tree) {
2761                         release_tree_content_recursive(b->branch_tree.tree);
2762                         b->branch_tree.tree = NULL;
2763                 }
2764         }
2765         else
2766                 b = new_branch(sp);
2767         read_next_command();
2768         parse_from(b);
2769         if (command_buf.len > 0)
2770                 unread_command_buf = 1;
2771 }
2772
2773 static void cat_blob_write(const char *buf, unsigned long size)
2774 {
2775         if (write_in_full(cat_blob_fd, buf, size) != size)
2776                 die_errno("Write to frontend failed");
2777 }
2778
2779 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2780 {
2781         struct strbuf line = STRBUF_INIT;
2782         unsigned long size;
2783         enum object_type type = 0;
2784         char *buf;
2785
2786         if (!oe || oe->pack_id == MAX_PACK_ID) {
2787                 buf = read_sha1_file(sha1, &type, &size);
2788         } else {
2789                 type = oe->type;
2790                 buf = gfi_unpack_entry(oe, &size);
2791         }
2792
2793         /*
2794          * Output based on batch_one_object() from cat-file.c.
2795          */
2796         if (type <= 0) {
2797                 strbuf_reset(&line);
2798                 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2799                 cat_blob_write(line.buf, line.len);
2800                 strbuf_release(&line);
2801                 free(buf);
2802                 return;
2803         }
2804         if (!buf)
2805                 die("Can't read object %s", sha1_to_hex(sha1));
2806         if (type != OBJ_BLOB)
2807                 die("Object %s is a %s but a blob was expected.",
2808                     sha1_to_hex(sha1), typename(type));
2809         strbuf_reset(&line);
2810         strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2811                                                 typename(type), size);
2812         cat_blob_write(line.buf, line.len);
2813         strbuf_release(&line);
2814         cat_blob_write(buf, size);
2815         cat_blob_write("\n", 1);
2816         free(buf);
2817 }
2818
2819 static void parse_cat_blob(void)
2820 {
2821         const char *p;
2822         struct object_entry *oe = oe;
2823         unsigned char sha1[20];
2824
2825         /* cat-blob SP <object> LF */
2826         p = command_buf.buf + strlen("cat-blob ");
2827         if (*p == ':') {
2828                 char *x;
2829                 oe = find_mark(strtoumax(p + 1, &x, 10));
2830                 if (x == p + 1)
2831                         die("Invalid mark: %s", command_buf.buf);
2832                 if (!oe)
2833                         die("Unknown mark: %s", command_buf.buf);
2834                 if (*x)
2835                         die("Garbage after mark: %s", command_buf.buf);
2836                 hashcpy(sha1, oe->idx.sha1);
2837         } else {
2838                 if (get_sha1_hex(p, sha1))
2839                         die("Invalid SHA1: %s", command_buf.buf);
2840                 if (p[40])
2841                         die("Garbage after SHA1: %s", command_buf.buf);
2842                 oe = find_object(sha1);
2843         }
2844
2845         cat_blob(oe, sha1);
2846 }
2847
2848 static struct object_entry *dereference(struct object_entry *oe,
2849                                         unsigned char sha1[20])
2850 {
2851         unsigned long size;
2852         char *buf = NULL;
2853         if (!oe) {
2854                 enum object_type type = sha1_object_info(sha1, NULL);
2855                 if (type < 0)
2856                         die("object not found: %s", sha1_to_hex(sha1));
2857                 /* cache it! */
2858                 oe = insert_object(sha1);
2859                 oe->type = type;
2860                 oe->pack_id = MAX_PACK_ID;
2861                 oe->idx.offset = 1;
2862         }
2863         switch (oe->type) {
2864         case OBJ_TREE:  /* easy case. */
2865                 return oe;
2866         case OBJ_COMMIT:
2867         case OBJ_TAG:
2868                 break;
2869         default:
2870                 die("Not a treeish: %s", command_buf.buf);
2871         }
2872
2873         if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
2874                 buf = gfi_unpack_entry(oe, &size);
2875         } else {
2876                 enum object_type unused;
2877                 buf = read_sha1_file(sha1, &unused, &size);
2878         }
2879         if (!buf)
2880                 die("Can't load object %s", sha1_to_hex(sha1));
2881
2882         /* Peel one layer. */
2883         switch (oe->type) {
2884         case OBJ_TAG:
2885                 if (size < 40 + strlen("object ") ||
2886                     get_sha1_hex(buf + strlen("object "), sha1))
2887                         die("Invalid SHA1 in tag: %s", command_buf.buf);
2888                 break;
2889         case OBJ_COMMIT:
2890                 if (size < 40 + strlen("tree ") ||
2891                     get_sha1_hex(buf + strlen("tree "), sha1))
2892                         die("Invalid SHA1 in commit: %s", command_buf.buf);
2893         }
2894
2895         free(buf);
2896         return find_object(sha1);
2897 }
2898
2899 static struct object_entry *parse_treeish_dataref(const char **p)
2900 {
2901         unsigned char sha1[20];
2902         struct object_entry *e;
2903
2904         if (**p == ':') {       /* <mark> */
2905                 char *endptr;
2906                 e = find_mark(strtoumax(*p + 1, &endptr, 10));
2907                 if (endptr == *p + 1)
2908                         die("Invalid mark: %s", command_buf.buf);
2909                 if (!e)
2910                         die("Unknown mark: %s", command_buf.buf);
2911                 *p = endptr;
2912                 hashcpy(sha1, e->idx.sha1);
2913         } else {        /* <sha1> */
2914                 if (get_sha1_hex(*p, sha1))
2915                         die("Invalid SHA1: %s", command_buf.buf);
2916                 e = find_object(sha1);
2917                 *p += 40;
2918         }
2919
2920         while (!e || e->type != OBJ_TREE)
2921                 e = dereference(e, sha1);
2922         return e;
2923 }
2924
2925 static void print_ls(int mode, const unsigned char *sha1, const char *path)
2926 {
2927         static struct strbuf line = STRBUF_INIT;
2928
2929         /* See show_tree(). */
2930         const char *type =
2931                 S_ISGITLINK(mode) ? commit_type :
2932                 S_ISDIR(mode) ? tree_type :
2933                 blob_type;
2934
2935         if (!mode) {
2936                 /* missing SP path LF */
2937                 strbuf_reset(&line);
2938                 strbuf_addstr(&line, "missing ");
2939                 quote_c_style(path, &line, NULL, 0);
2940                 strbuf_addch(&line, '\n');
2941         } else {
2942                 /* mode SP type SP object_name TAB path LF */
2943                 strbuf_reset(&line);
2944                 strbuf_addf(&line, "%06o %s %s\t",
2945                                 mode, type, sha1_to_hex(sha1));
2946                 quote_c_style(path, &line, NULL, 0);
2947                 strbuf_addch(&line, '\n');
2948         }
2949         cat_blob_write(line.buf, line.len);
2950 }
2951
2952 static void parse_ls(struct branch *b)
2953 {
2954         const char *p;
2955         struct tree_entry *root = NULL;
2956         struct tree_entry leaf = {0};
2957
2958         /* ls SP (<treeish> SP)? <path> */
2959         p = command_buf.buf + strlen("ls ");
2960         if (*p == '"') {
2961                 if (!b)
2962                         die("Not in a commit: %s", command_buf.buf);
2963                 root = &b->branch_tree;
2964         } else {
2965                 struct object_entry *e = parse_treeish_dataref(&p);
2966                 root = new_tree_entry();
2967                 hashcpy(root->versions[1].sha1, e->idx.sha1);
2968                 load_tree(root);
2969                 if (*p++ != ' ')
2970                         die("Missing space after tree-ish: %s", command_buf.buf);
2971         }
2972         if (*p == '"') {
2973                 static struct strbuf uq = STRBUF_INIT;
2974                 const char *endp;
2975                 strbuf_reset(&uq);
2976                 if (unquote_c_style(&uq, p, &endp))
2977                         die("Invalid path: %s", command_buf.buf);
2978                 if (*endp)
2979                         die("Garbage after path in: %s", command_buf.buf);
2980                 p = uq.buf;
2981         }
2982         tree_content_get(root, p, &leaf);
2983         /*
2984          * A directory in preparation would have a sha1 of zero
2985          * until it is saved.  Save, for simplicity.
2986          */
2987         if (S_ISDIR(leaf.versions[1].mode))
2988                 store_tree(&leaf);
2989
2990         print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
2991         if (!b || root != &b->branch_tree)
2992                 release_tree_entry(root);
2993 }
2994
2995 static void checkpoint(void)
2996 {
2997         checkpoint_requested = 0;
2998         if (object_count) {
2999                 cycle_packfile();
3000                 dump_branches();
3001                 dump_tags();
3002                 dump_marks();
3003         }
3004 }
3005
3006 static void parse_checkpoint(void)
3007 {
3008         checkpoint_requested = 1;
3009         skip_optional_lf();
3010 }
3011
3012 static void parse_progress(void)
3013 {
3014         fwrite(command_buf.buf, 1, command_buf.len, stdout);
3015         fputc('\n', stdout);
3016         fflush(stdout);
3017         skip_optional_lf();
3018 }
3019
3020 static char* make_fast_import_path(const char *path)
3021 {
3022         struct strbuf abs_path = STRBUF_INIT;
3023
3024         if (!relative_marks_paths || is_absolute_path(path))
3025                 return xstrdup(path);
3026         strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
3027         return strbuf_detach(&abs_path, NULL);
3028 }
3029
3030 static void option_import_marks(const char *marks,
3031                                         int from_stream, int ignore_missing)
3032 {
3033         if (import_marks_file) {
3034                 if (from_stream)
3035                         die("Only one import-marks command allowed per stream");
3036
3037                 /* read previous mark file */
3038                 if(!import_marks_file_from_stream)
3039                         read_marks();
3040         }
3041
3042         import_marks_file = make_fast_import_path(marks);
3043         safe_create_leading_directories_const(import_marks_file);
3044         import_marks_file_from_stream = from_stream;
3045         import_marks_file_ignore_missing = ignore_missing;
3046 }
3047
3048 static void option_date_format(const char *fmt)
3049 {
3050         if (!strcmp(fmt, "raw"))
3051                 whenspec = WHENSPEC_RAW;
3052         else if (!strcmp(fmt, "rfc2822"))
3053                 whenspec = WHENSPEC_RFC2822;
3054         else if (!strcmp(fmt, "now"))
3055                 whenspec = WHENSPEC_NOW;
3056         else
3057                 die("unknown --date-format argument %s", fmt);
3058 }
3059
3060 static unsigned long ulong_arg(const char *option, const char *arg)
3061 {
3062         char *endptr;
3063         unsigned long rv = strtoul(arg, &endptr, 0);
3064         if (strchr(arg, '-') || endptr == arg || *endptr)
3065                 die("%s: argument must be a non-negative integer", option);
3066         return rv;
3067 }
3068
3069 static void option_depth(const char *depth)
3070 {
3071         max_depth = ulong_arg("--depth", depth);
3072         if (max_depth > MAX_DEPTH)
3073                 die("--depth cannot exceed %u", MAX_DEPTH);
3074 }
3075
3076 static void option_active_branches(const char *branches)
3077 {
3078         max_active_branches = ulong_arg("--active-branches", branches);
3079 }
3080
3081 static void option_export_marks(const char *marks)
3082 {
3083         export_marks_file = make_fast_import_path(marks);
3084         safe_create_leading_directories_const(export_marks_file);
3085 }
3086
3087 static void option_cat_blob_fd(const char *fd)
3088 {
3089         unsigned long n = ulong_arg("--cat-blob-fd", fd);
3090         if (n > (unsigned long) INT_MAX)
3091                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3092         cat_blob_fd = (int) n;
3093 }
3094
3095 static void option_export_pack_edges(const char *edges)
3096 {
3097         if (pack_edges)
3098                 fclose(pack_edges);
3099         pack_edges = fopen(edges, "a");
3100         if (!pack_edges)
3101                 die_errno("Cannot open '%s'", edges);
3102 }
3103
3104 static int parse_one_option(const char *option)
3105 {
3106         if (!prefixcmp(option, "max-pack-size=")) {
3107                 unsigned long v;
3108                 if (!git_parse_ulong(option + 14, &v))
3109                         return 0;
3110                 if (v < 8192) {
3111                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3112                         v *= 1024 * 1024;
3113                 } else if (v < 1024 * 1024) {
3114                         warning("minimum max-pack-size is 1 MiB");
3115                         v = 1024 * 1024;
3116                 }
3117                 max_packsize = v;
3118         } else if (!prefixcmp(option, "big-file-threshold=")) {
3119                 unsigned long v;
3120                 if (!git_parse_ulong(option + 19, &v))
3121                         return 0;
3122                 big_file_threshold = v;
3123         } else if (!prefixcmp(option, "depth=")) {
3124                 option_depth(option + 6);
3125         } else if (!prefixcmp(option, "active-branches=")) {
3126                 option_active_branches(option + 16);
3127         } else if (!prefixcmp(option, "export-pack-edges=")) {
3128                 option_export_pack_edges(option + 18);
3129         } else if (!prefixcmp(option, "quiet")) {
3130                 show_stats = 0;
3131         } else if (!prefixcmp(option, "stats")) {
3132                 show_stats = 1;
3133         } else {
3134                 return 0;
3135         }
3136
3137         return 1;
3138 }
3139
3140 static int parse_one_feature(const char *feature, int from_stream)
3141 {
3142         if (!prefixcmp(feature, "date-format=")) {
3143                 option_date_format(feature + 12);
3144         } else if (!prefixcmp(feature, "import-marks=")) {
3145                 option_import_marks(feature + 13, from_stream, 0);
3146         } else if (!prefixcmp(feature, "import-marks-if-exists=")) {
3147                 option_import_marks(feature + strlen("import-marks-if-exists="),
3148                                         from_stream, 1);
3149         } else if (!prefixcmp(feature, "export-marks=")) {
3150                 option_export_marks(feature + 13);
3151         } else if (!strcmp(feature, "cat-blob")) {
3152                 ; /* Don't die - this feature is supported */
3153         } else if (!prefixcmp(feature, "relative-marks")) {
3154                 relative_marks_paths = 1;
3155         } else if (!prefixcmp(feature, "no-relative-marks")) {
3156                 relative_marks_paths = 0;
3157         } else if (!prefixcmp(feature, "force")) {
3158                 force_update = 1;
3159         } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3160                 ; /* do nothing; we have the feature */
3161         } else {
3162                 return 0;
3163         }
3164
3165         return 1;
3166 }
3167
3168 static void parse_feature(void)
3169 {
3170         char *feature = command_buf.buf + 8;
3171
3172         if (seen_data_command)
3173                 die("Got feature command '%s' after data command", feature);
3174
3175         if (parse_one_feature(feature, 1))
3176                 return;
3177
3178         die("This version of fast-import does not support feature %s.", feature);
3179 }
3180
3181 static void parse_option(void)
3182 {
3183         char *option = command_buf.buf + 11;
3184
3185         if (seen_data_command)
3186                 die("Got option command '%s' after data command", option);
3187
3188         if (parse_one_option(option))
3189                 return;
3190
3191         die("This version of fast-import does not support option: %s", option);
3192 }
3193
3194 static int git_pack_config(const char *k, const char *v, void *cb)
3195 {
3196         if (!strcmp(k, "pack.depth")) {
3197                 max_depth = git_config_int(k, v);
3198                 if (max_depth > MAX_DEPTH)
3199                         max_depth = MAX_DEPTH;
3200                 return 0;
3201         }
3202         if (!strcmp(k, "pack.compression")) {
3203                 int level = git_config_int(k, v);
3204                 if (level == -1)
3205                         level = Z_DEFAULT_COMPRESSION;
3206                 else if (level < 0 || level > Z_BEST_COMPRESSION)
3207                         die("bad pack compression level %d", level);
3208                 pack_compression_level = level;
3209                 pack_compression_seen = 1;
3210                 return 0;
3211         }
3212         if (!strcmp(k, "pack.indexversion")) {
3213                 pack_idx_default_version = git_config_int(k, v);
3214                 if (pack_idx_default_version > 2)
3215                         die("bad pack.indexversion=%"PRIu32,
3216                             pack_idx_default_version);
3217                 return 0;
3218         }
3219         if (!strcmp(k, "pack.packsizelimit")) {
3220                 max_packsize = git_config_ulong(k, v);
3221                 return 0;
3222         }
3223         if (!strcmp(k, "core.bigfilethreshold")) {
3224                 long n = git_config_int(k, v);
3225                 big_file_threshold = 0 < n ? n : 0;
3226         }
3227         return git_default_config(k, v, cb);
3228 }
3229
3230 static const char fast_import_usage[] =
3231 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3232
3233 static void parse_argv(void)
3234 {
3235         unsigned int i;
3236
3237         for (i = 1; i < global_argc; i++) {
3238                 const char *a = global_argv[i];
3239
3240                 if (*a != '-' || !strcmp(a, "--"))
3241                         break;
3242
3243                 if (parse_one_option(a + 2))
3244                         continue;
3245
3246                 if (parse_one_feature(a + 2, 0))
3247                         continue;
3248
3249                 if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3250                         option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3251                         continue;
3252                 }
3253
3254                 die("unknown option %s", a);
3255         }
3256         if (i != global_argc)
3257                 usage(fast_import_usage);
3258
3259         seen_data_command = 1;
3260         if (import_marks_file)
3261                 read_marks();
3262 }
3263
3264 int main(int argc, const char **argv)
3265 {
3266         unsigned int i;
3267
3268         git_extract_argv0_path(argv[0]);
3269
3270         if (argc == 2 && !strcmp(argv[1], "-h"))
3271                 usage(fast_import_usage);
3272
3273         setup_git_directory();
3274         git_config(git_pack_config, NULL);
3275         if (!pack_compression_seen && core_compression_seen)
3276                 pack_compression_level = core_compression_level;
3277
3278         alloc_objects(object_entry_alloc);
3279         strbuf_init(&command_buf, 0);
3280         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3281         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3282         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3283         marks = pool_calloc(1, sizeof(struct mark_set));
3284
3285         global_argc = argc;
3286         global_argv = argv;
3287
3288         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3289         for (i = 0; i < (cmd_save - 1); i++)
3290                 rc_free[i].next = &rc_free[i + 1];
3291         rc_free[cmd_save - 1].next = NULL;
3292
3293         prepare_packed_git();
3294         start_packfile();
3295         set_die_routine(die_nicely);
3296         set_checkpoint_signal();
3297         while (read_next_command() != EOF) {
3298                 if (!strcmp("blob", command_buf.buf))
3299                         parse_new_blob();
3300                 else if (!prefixcmp(command_buf.buf, "ls "))
3301                         parse_ls(NULL);
3302                 else if (!prefixcmp(command_buf.buf, "commit "))
3303                         parse_new_commit();
3304                 else if (!prefixcmp(command_buf.buf, "tag "))
3305                         parse_new_tag();
3306                 else if (!prefixcmp(command_buf.buf, "reset "))
3307                         parse_reset_branch();
3308                 else if (!strcmp("checkpoint", command_buf.buf))
3309                         parse_checkpoint();
3310                 else if (!prefixcmp(command_buf.buf, "progress "))
3311                         parse_progress();
3312                 else if (!prefixcmp(command_buf.buf, "feature "))
3313                         parse_feature();
3314                 else if (!prefixcmp(command_buf.buf, "option git "))
3315                         parse_option();
3316                 else if (!prefixcmp(command_buf.buf, "option "))
3317                         /* ignore non-git options*/;
3318                 else
3319                         die("Unsupported command: %s", command_buf.buf);
3320
3321                 if (checkpoint_requested)
3322                         checkpoint();
3323         }
3324
3325         /* argv hasn't been parsed yet, do so */
3326         if (!seen_data_command)
3327                 parse_argv();
3328
3329         end_packfile();
3330
3331         dump_branches();
3332         dump_tags();
3333         unkeep_all_packs();
3334         dump_marks();
3335
3336         if (pack_edges)
3337                 fclose(pack_edges);
3338
3339         if (show_stats) {
3340                 uintmax_t total_count = 0, duplicate_count = 0;
3341                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3342                         total_count += object_count_by_type[i];
3343                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3344                         duplicate_count += duplicate_count_by_type[i];
3345
3346                 fprintf(stderr, "%s statistics:\n", argv[0]);
3347                 fprintf(stderr, "---------------------------------------------------------------------\n");
3348                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3349                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3350                 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]);
3351                 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]);
3352                 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]);
3353                 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]);
3354                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3355                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3356                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3357                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3358                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3359                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3360                 fprintf(stderr, "---------------------------------------------------------------------\n");
3361                 pack_report();
3362                 fprintf(stderr, "---------------------------------------------------------------------\n");
3363                 fprintf(stderr, "\n");
3364         }
3365
3366         return failure ? 1 : 0;
3367 }