rebase -i: use the rebase--helper builtin
[git] / fast-import.c
1 /*
2 (See Documentation/git-fast-import.txt for maintained documentation.)
3 Format of STDIN stream:
4
5   stream ::= cmd*;
6
7   cmd ::= new_blob
8         | new_commit
9         | new_tag
10         | reset_branch
11         | checkpoint
12         | progress
13         ;
14
15   new_blob ::= 'blob' lf
16     mark?
17     file_content;
18   file_content ::= data;
19
20   new_commit ::= 'commit' sp ref_str lf
21     mark?
22     ('author' (sp name)? sp '<' email '>' sp when lf)?
23     'committer' (sp name)? sp '<' email '>' sp when lf
24     commit_msg
25     ('from' sp commit-ish lf)?
26     ('merge' sp commit-ish lf)*
27     (file_change | ls)*
28     lf?;
29   commit_msg ::= data;
30
31   ls ::= 'ls' sp '"' quoted(path) '"' lf;
32
33   file_change ::= file_clr
34     | file_del
35     | file_rnm
36     | file_cpy
37     | file_obm
38     | file_inm;
39   file_clr ::= 'deleteall' lf;
40   file_del ::= 'D' sp path_str lf;
41   file_rnm ::= 'R' sp path_str sp path_str lf;
42   file_cpy ::= 'C' sp path_str sp path_str lf;
43   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
44   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
45     data;
46   note_obm ::= 'N' sp (hexsha1 | idnum) sp commit-ish lf;
47   note_inm ::= 'N' sp 'inline' sp commit-ish lf
48     data;
49
50   new_tag ::= 'tag' sp tag_str lf
51     'from' sp commit-ish lf
52     ('tagger' (sp name)? sp '<' email '>' sp when lf)?
53     tag_msg;
54   tag_msg ::= data;
55
56   reset_branch ::= 'reset' sp ref_str lf
57     ('from' sp commit-ish lf)?
58     lf?;
59
60   checkpoint ::= 'checkpoint' lf
61     lf?;
62
63   progress ::= 'progress' sp not_lf* lf
64     lf?;
65
66      # note: the first idnum in a stream should be 1 and subsequent
67      # idnums should not have gaps between values as this will cause
68      # the stream parser to reserve space for the gapped values.  An
69      # idnum can be updated in the future to a new object by issuing
70      # a new mark directive with the old idnum.
71      #
72   mark ::= 'mark' sp idnum lf;
73   data ::= (delimited_data | exact_data)
74     lf?;
75
76     # note: delim may be any string but must not contain lf.
77     # data_line may contain any data but must not be exactly
78     # delim.
79   delimited_data ::= 'data' sp '<<' delim lf
80     (data_line lf)*
81     delim lf;
82
83      # note: declen indicates the length of binary_data in bytes.
84      # declen does not include the lf preceding the binary data.
85      #
86   exact_data ::= 'data' sp declen lf
87     binary_data;
88
89      # note: quoted strings are C-style quoting supporting \c for
90      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
91      # is the signed byte value in octal.  Note that the only
92      # characters which must actually be escaped to protect the
93      # stream formatting is: \, " and LF.  Otherwise these values
94      # are UTF8.
95      #
96   commit-ish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
97   ref_str     ::= ref;
98   sha1exp_str ::= sha1exp;
99   tag_str     ::= tag;
100   path_str    ::= path    | '"' quoted(path)    '"' ;
101   mode        ::= '100644' | '644'
102                 | '100755' | '755'
103                 | '120000'
104                 ;
105
106   declen ::= # unsigned 32 bit value, ascii base10 notation;
107   bigint ::= # unsigned integer value, ascii base10 notation;
108   binary_data ::= # file content, not interpreted;
109
110   when         ::= raw_when | rfc2822_when;
111   raw_when     ::= ts sp tz;
112   rfc2822_when ::= # Valid RFC 2822 date and time;
113
114   sp ::= # ASCII space character;
115   lf ::= # ASCII newline (LF) character;
116
117      # note: a colon (':') must precede the numerical value assigned to
118      # an idnum.  This is to distinguish it from a ref or tag name as
119      # GIT does not permit ':' in ref or tag strings.
120      #
121   idnum   ::= ':' bigint;
122   path    ::= # GIT style file path, e.g. "a/b/c";
123   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
124   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
125   sha1exp ::= # Any valid GIT SHA1 expression;
126   hexsha1 ::= # SHA1 in hexadecimal format;
127
128      # note: name and email are UTF8 strings, however name must not
129      # contain '<' or lf and email must not contain any of the
130      # following: '<', '>', lf.
131      #
132   name  ::= # valid GIT author/committer name;
133   email ::= # valid GIT author/committer email;
134   ts    ::= # time since the epoch in seconds, ascii base10 notation;
135   tz    ::= # GIT style timezone;
136
137      # note: comments, get-mark, ls-tree, and cat-blob requests may
138      # appear anywhere in the input, except within a data command. Any
139      # form of the data command always escapes the related input from
140      # comment processing.
141      #
142      # In case it is not clear, the '#' that starts the comment
143      # must be the first character on that line (an lf
144      # preceded it).
145      #
146
147   get_mark ::= 'get-mark' sp idnum lf;
148   cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
149   ls_tree  ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
150
151   comment ::= '#' not_lf* lf;
152   not_lf  ::= # Any byte that is not ASCII newline (LF);
153 */
154
155 #include "builtin.h"
156 #include "cache.h"
157 #include "lockfile.h"
158 #include "object.h"
159 #include "blob.h"
160 #include "tree.h"
161 #include "commit.h"
162 #include "delta.h"
163 #include "pack.h"
164 #include "refs.h"
165 #include "csum-file.h"
166 #include "quote.h"
167 #include "dir.h"
168 #include "run-command.h"
169
170 #define PACK_ID_BITS 16
171 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
172 #define DEPTH_BITS 13
173 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
174
175 /*
176  * We abuse the setuid bit on directories to mean "do not delta".
177  */
178 #define NO_DELTA S_ISUID
179
180 struct object_entry {
181         struct pack_idx_entry idx;
182         struct object_entry *next;
183         uint32_t type : TYPE_BITS,
184                 pack_id : PACK_ID_BITS,
185                 depth : DEPTH_BITS;
186 };
187
188 struct object_entry_pool {
189         struct object_entry_pool *next_pool;
190         struct object_entry *next_free;
191         struct object_entry *end;
192         struct object_entry entries[FLEX_ARRAY]; /* more */
193 };
194
195 struct mark_set {
196         union {
197                 struct object_entry *marked[1024];
198                 struct mark_set *sets[1024];
199         } data;
200         unsigned int shift;
201 };
202
203 struct last_object {
204         struct strbuf data;
205         off_t offset;
206         unsigned int depth;
207         unsigned no_swap : 1;
208 };
209
210 struct mem_pool {
211         struct mem_pool *next_pool;
212         char *next_free;
213         char *end;
214         uintmax_t space[FLEX_ARRAY]; /* more */
215 };
216
217 struct atom_str {
218         struct atom_str *next_atom;
219         unsigned short str_len;
220         char str_dat[FLEX_ARRAY]; /* more */
221 };
222
223 struct tree_content;
224 struct tree_entry {
225         struct tree_content *tree;
226         struct atom_str *name;
227         struct tree_entry_ms {
228                 uint16_t mode;
229                 unsigned char sha1[20];
230         } versions[2];
231 };
232
233 struct tree_content {
234         unsigned int entry_capacity; /* must match avail_tree_content */
235         unsigned int entry_count;
236         unsigned int delta_depth;
237         struct tree_entry *entries[FLEX_ARRAY]; /* more */
238 };
239
240 struct avail_tree_content {
241         unsigned int entry_capacity; /* must match tree_content */
242         struct avail_tree_content *next_avail;
243 };
244
245 struct branch {
246         struct branch *table_next_branch;
247         struct branch *active_next_branch;
248         const char *name;
249         struct tree_entry branch_tree;
250         uintmax_t last_commit;
251         uintmax_t num_notes;
252         unsigned active : 1;
253         unsigned delete : 1;
254         unsigned pack_id : PACK_ID_BITS;
255         unsigned char sha1[20];
256 };
257
258 struct tag {
259         struct tag *next_tag;
260         const char *name;
261         unsigned int pack_id;
262         unsigned char sha1[20];
263 };
264
265 struct hash_list {
266         struct hash_list *next;
267         unsigned char sha1[20];
268 };
269
270 typedef enum {
271         WHENSPEC_RAW = 1,
272         WHENSPEC_RFC2822,
273         WHENSPEC_NOW
274 } whenspec_type;
275
276 struct recent_command {
277         struct recent_command *prev;
278         struct recent_command *next;
279         char *buf;
280 };
281
282 /* Configured limits on output */
283 static unsigned long max_depth = 10;
284 static off_t max_packsize;
285 static int unpack_limit = 100;
286 static int force_update;
287
288 /* Stats and misc. counters */
289 static uintmax_t alloc_count;
290 static uintmax_t marks_set_count;
291 static uintmax_t object_count_by_type[1 << TYPE_BITS];
292 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
293 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
294 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
295 static unsigned long object_count;
296 static unsigned long branch_count;
297 static unsigned long branch_load_count;
298 static int failure;
299 static FILE *pack_edges;
300 static unsigned int show_stats = 1;
301 static int global_argc;
302 static const char **global_argv;
303
304 /* Memory pools */
305 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
306 static size_t total_allocd;
307 static struct mem_pool *mem_pool;
308
309 /* Atom management */
310 static unsigned int atom_table_sz = 4451;
311 static unsigned int atom_cnt;
312 static struct atom_str **atom_table;
313
314 /* The .pack file being generated */
315 static struct pack_idx_option pack_idx_opts;
316 static unsigned int pack_id;
317 static struct sha1file *pack_file;
318 static struct packed_git *pack_data;
319 static struct packed_git **all_packs;
320 static off_t pack_size;
321
322 /* Table of objects we've written. */
323 static unsigned int object_entry_alloc = 5000;
324 static struct object_entry_pool *blocks;
325 static struct object_entry *object_table[1 << 16];
326 static struct mark_set *marks;
327 static const char *export_marks_file;
328 static const char *import_marks_file;
329 static int import_marks_file_from_stream;
330 static int import_marks_file_ignore_missing;
331 static int import_marks_file_done;
332 static int relative_marks_paths;
333
334 /* Our last blob */
335 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
336
337 /* Tree management */
338 static unsigned int tree_entry_alloc = 1000;
339 static void *avail_tree_entry;
340 static unsigned int avail_tree_table_sz = 100;
341 static struct avail_tree_content **avail_tree_table;
342 static struct strbuf old_tree = STRBUF_INIT;
343 static struct strbuf new_tree = STRBUF_INIT;
344
345 /* Branch data */
346 static unsigned long max_active_branches = 5;
347 static unsigned long cur_active_branches;
348 static unsigned long branch_table_sz = 1039;
349 static struct branch **branch_table;
350 static struct branch *active_branches;
351
352 /* Tag data */
353 static struct tag *first_tag;
354 static struct tag *last_tag;
355
356 /* Input stream parsing */
357 static whenspec_type whenspec = WHENSPEC_RAW;
358 static struct strbuf command_buf = STRBUF_INIT;
359 static int unread_command_buf;
360 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
361 static struct recent_command *cmd_tail = &cmd_hist;
362 static struct recent_command *rc_free;
363 static unsigned int cmd_save = 100;
364 static uintmax_t next_mark;
365 static struct strbuf new_data = STRBUF_INIT;
366 static int seen_data_command;
367 static int require_explicit_termination;
368
369 /* Signal handling */
370 static volatile sig_atomic_t checkpoint_requested;
371
372 /* Where to write output of cat-blob commands */
373 static int cat_blob_fd = STDOUT_FILENO;
374
375 static void parse_argv(void);
376 static void parse_get_mark(const char *p);
377 static void parse_cat_blob(const char *p);
378 static void parse_ls(const char *p, struct branch *b);
379
380 static void write_branch_report(FILE *rpt, struct branch *b)
381 {
382         fprintf(rpt, "%s:\n", b->name);
383
384         fprintf(rpt, "  status      :");
385         if (b->active)
386                 fputs(" active", rpt);
387         if (b->branch_tree.tree)
388                 fputs(" loaded", rpt);
389         if (is_null_sha1(b->branch_tree.versions[1].sha1))
390                 fputs(" dirty", rpt);
391         fputc('\n', rpt);
392
393         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
394         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
395         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
396         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
397
398         fputs("  last pack   : ", rpt);
399         if (b->pack_id < MAX_PACK_ID)
400                 fprintf(rpt, "%u", b->pack_id);
401         fputc('\n', rpt);
402
403         fputc('\n', rpt);
404 }
405
406 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
407
408 static void write_crash_report(const char *err)
409 {
410         char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
411         FILE *rpt = fopen(loc, "w");
412         struct branch *b;
413         unsigned long lu;
414         struct recent_command *rc;
415
416         if (!rpt) {
417                 error_errno("can't write crash report %s", loc);
418                 free(loc);
419                 return;
420         }
421
422         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
423
424         fprintf(rpt, "fast-import crash report:\n");
425         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
426         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
427         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
428         fputc('\n', rpt);
429
430         fputs("fatal: ", rpt);
431         fputs(err, rpt);
432         fputc('\n', rpt);
433
434         fputc('\n', rpt);
435         fputs("Most Recent Commands Before Crash\n", rpt);
436         fputs("---------------------------------\n", rpt);
437         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
438                 if (rc->next == &cmd_hist)
439                         fputs("* ", rpt);
440                 else
441                         fputs("  ", rpt);
442                 fputs(rc->buf, rpt);
443                 fputc('\n', rpt);
444         }
445
446         fputc('\n', rpt);
447         fputs("Active Branch LRU\n", rpt);
448         fputs("-----------------\n", rpt);
449         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
450                 cur_active_branches,
451                 max_active_branches);
452         fputc('\n', rpt);
453         fputs("  pos  clock name\n", rpt);
454         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
455         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
456                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
457                         ++lu, b->last_commit, b->name);
458
459         fputc('\n', rpt);
460         fputs("Inactive Branches\n", rpt);
461         fputs("-----------------\n", rpt);
462         for (lu = 0; lu < branch_table_sz; lu++) {
463                 for (b = branch_table[lu]; b; b = b->table_next_branch)
464                         write_branch_report(rpt, b);
465         }
466
467         if (first_tag) {
468                 struct tag *tg;
469                 fputc('\n', rpt);
470                 fputs("Annotated Tags\n", rpt);
471                 fputs("--------------\n", rpt);
472                 for (tg = first_tag; tg; tg = tg->next_tag) {
473                         fputs(sha1_to_hex(tg->sha1), rpt);
474                         fputc(' ', rpt);
475                         fputs(tg->name, rpt);
476                         fputc('\n', rpt);
477                 }
478         }
479
480         fputc('\n', rpt);
481         fputs("Marks\n", rpt);
482         fputs("-----\n", rpt);
483         if (export_marks_file)
484                 fprintf(rpt, "  exported to %s\n", export_marks_file);
485         else
486                 dump_marks_helper(rpt, 0, marks);
487
488         fputc('\n', rpt);
489         fputs("-------------------\n", rpt);
490         fputs("END OF CRASH REPORT\n", rpt);
491         fclose(rpt);
492         free(loc);
493 }
494
495 static void end_packfile(void);
496 static void unkeep_all_packs(void);
497 static void dump_marks(void);
498
499 static NORETURN void die_nicely(const char *err, va_list params)
500 {
501         static int zombie;
502         char message[2 * PATH_MAX];
503
504         vsnprintf(message, sizeof(message), err, params);
505         fputs("fatal: ", stderr);
506         fputs(message, stderr);
507         fputc('\n', stderr);
508
509         if (!zombie) {
510                 zombie = 1;
511                 write_crash_report(message);
512                 end_packfile();
513                 unkeep_all_packs();
514                 dump_marks();
515         }
516         exit(128);
517 }
518
519 #ifndef SIGUSR1 /* Windows, for example */
520
521 static void set_checkpoint_signal(void)
522 {
523 }
524
525 #else
526
527 static void checkpoint_signal(int signo)
528 {
529         checkpoint_requested = 1;
530 }
531
532 static void set_checkpoint_signal(void)
533 {
534         struct sigaction sa;
535
536         memset(&sa, 0, sizeof(sa));
537         sa.sa_handler = checkpoint_signal;
538         sigemptyset(&sa.sa_mask);
539         sa.sa_flags = SA_RESTART;
540         sigaction(SIGUSR1, &sa, NULL);
541 }
542
543 #endif
544
545 static void alloc_objects(unsigned int cnt)
546 {
547         struct object_entry_pool *b;
548
549         b = xmalloc(sizeof(struct object_entry_pool)
550                 + cnt * sizeof(struct object_entry));
551         b->next_pool = blocks;
552         b->next_free = b->entries;
553         b->end = b->entries + cnt;
554         blocks = b;
555         alloc_count += cnt;
556 }
557
558 static struct object_entry *new_object(unsigned char *sha1)
559 {
560         struct object_entry *e;
561
562         if (blocks->next_free == blocks->end)
563                 alloc_objects(object_entry_alloc);
564
565         e = blocks->next_free++;
566         hashcpy(e->idx.sha1, sha1);
567         return e;
568 }
569
570 static struct object_entry *find_object(unsigned char *sha1)
571 {
572         unsigned int h = sha1[0] << 8 | sha1[1];
573         struct object_entry *e;
574         for (e = object_table[h]; e; e = e->next)
575                 if (!hashcmp(sha1, e->idx.sha1))
576                         return e;
577         return NULL;
578 }
579
580 static struct object_entry *insert_object(unsigned char *sha1)
581 {
582         unsigned int h = sha1[0] << 8 | sha1[1];
583         struct object_entry *e = object_table[h];
584
585         while (e) {
586                 if (!hashcmp(sha1, e->idx.sha1))
587                         return e;
588                 e = e->next;
589         }
590
591         e = new_object(sha1);
592         e->next = object_table[h];
593         e->idx.offset = 0;
594         object_table[h] = e;
595         return e;
596 }
597
598 static void invalidate_pack_id(unsigned int id)
599 {
600         unsigned int h;
601         unsigned long lu;
602         struct tag *t;
603
604         for (h = 0; h < ARRAY_SIZE(object_table); h++) {
605                 struct object_entry *e;
606
607                 for (e = object_table[h]; e; e = e->next)
608                         if (e->pack_id == id)
609                                 e->pack_id = MAX_PACK_ID;
610         }
611
612         for (lu = 0; lu < branch_table_sz; lu++) {
613                 struct branch *b;
614
615                 for (b = branch_table[lu]; b; b = b->table_next_branch)
616                         if (b->pack_id == id)
617                                 b->pack_id = MAX_PACK_ID;
618         }
619
620         for (t = first_tag; t; t = t->next_tag)
621                 if (t->pack_id == id)
622                         t->pack_id = MAX_PACK_ID;
623 }
624
625 static unsigned int hc_str(const char *s, size_t len)
626 {
627         unsigned int r = 0;
628         while (len-- > 0)
629                 r = r * 31 + *s++;
630         return r;
631 }
632
633 static void *pool_alloc(size_t len)
634 {
635         struct mem_pool *p;
636         void *r;
637
638         /* round up to a 'uintmax_t' alignment */
639         if (len & (sizeof(uintmax_t) - 1))
640                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
641
642         for (p = mem_pool; p; p = p->next_pool)
643                 if ((p->end - p->next_free >= len))
644                         break;
645
646         if (!p) {
647                 if (len >= (mem_pool_alloc/2)) {
648                         total_allocd += len;
649                         return xmalloc(len);
650                 }
651                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
652                 p = xmalloc(st_add(sizeof(struct mem_pool), mem_pool_alloc));
653                 p->next_pool = mem_pool;
654                 p->next_free = (char *) p->space;
655                 p->end = p->next_free + mem_pool_alloc;
656                 mem_pool = p;
657         }
658
659         r = p->next_free;
660         p->next_free += len;
661         return r;
662 }
663
664 static void *pool_calloc(size_t count, size_t size)
665 {
666         size_t len = count * size;
667         void *r = pool_alloc(len);
668         memset(r, 0, len);
669         return r;
670 }
671
672 static char *pool_strdup(const char *s)
673 {
674         size_t len = strlen(s) + 1;
675         char *r = pool_alloc(len);
676         memcpy(r, s, len);
677         return r;
678 }
679
680 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
681 {
682         struct mark_set *s = marks;
683         while ((idnum >> s->shift) >= 1024) {
684                 s = pool_calloc(1, sizeof(struct mark_set));
685                 s->shift = marks->shift + 10;
686                 s->data.sets[0] = marks;
687                 marks = s;
688         }
689         while (s->shift) {
690                 uintmax_t i = idnum >> s->shift;
691                 idnum -= i << s->shift;
692                 if (!s->data.sets[i]) {
693                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
694                         s->data.sets[i]->shift = s->shift - 10;
695                 }
696                 s = s->data.sets[i];
697         }
698         if (!s->data.marked[idnum])
699                 marks_set_count++;
700         s->data.marked[idnum] = oe;
701 }
702
703 static struct object_entry *find_mark(uintmax_t idnum)
704 {
705         uintmax_t orig_idnum = idnum;
706         struct mark_set *s = marks;
707         struct object_entry *oe = NULL;
708         if ((idnum >> s->shift) < 1024) {
709                 while (s && s->shift) {
710                         uintmax_t i = idnum >> s->shift;
711                         idnum -= i << s->shift;
712                         s = s->data.sets[i];
713                 }
714                 if (s)
715                         oe = s->data.marked[idnum];
716         }
717         if (!oe)
718                 die("mark :%" PRIuMAX " not declared", orig_idnum);
719         return oe;
720 }
721
722 static struct atom_str *to_atom(const char *s, unsigned short len)
723 {
724         unsigned int hc = hc_str(s, len) % atom_table_sz;
725         struct atom_str *c;
726
727         for (c = atom_table[hc]; c; c = c->next_atom)
728                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
729                         return c;
730
731         c = pool_alloc(sizeof(struct atom_str) + len + 1);
732         c->str_len = len;
733         memcpy(c->str_dat, s, len);
734         c->str_dat[len] = 0;
735         c->next_atom = atom_table[hc];
736         atom_table[hc] = c;
737         atom_cnt++;
738         return c;
739 }
740
741 static struct branch *lookup_branch(const char *name)
742 {
743         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
744         struct branch *b;
745
746         for (b = branch_table[hc]; b; b = b->table_next_branch)
747                 if (!strcmp(name, b->name))
748                         return b;
749         return NULL;
750 }
751
752 static struct branch *new_branch(const char *name)
753 {
754         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
755         struct branch *b = lookup_branch(name);
756
757         if (b)
758                 die("Invalid attempt to create duplicate branch: %s", name);
759         if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
760                 die("Branch name doesn't conform to GIT standards: %s", name);
761
762         b = pool_calloc(1, sizeof(struct branch));
763         b->name = pool_strdup(name);
764         b->table_next_branch = branch_table[hc];
765         b->branch_tree.versions[0].mode = S_IFDIR;
766         b->branch_tree.versions[1].mode = S_IFDIR;
767         b->num_notes = 0;
768         b->active = 0;
769         b->pack_id = MAX_PACK_ID;
770         branch_table[hc] = b;
771         branch_count++;
772         return b;
773 }
774
775 static unsigned int hc_entries(unsigned int cnt)
776 {
777         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
778         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
779 }
780
781 static struct tree_content *new_tree_content(unsigned int cnt)
782 {
783         struct avail_tree_content *f, *l = NULL;
784         struct tree_content *t;
785         unsigned int hc = hc_entries(cnt);
786
787         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
788                 if (f->entry_capacity >= cnt)
789                         break;
790
791         if (f) {
792                 if (l)
793                         l->next_avail = f->next_avail;
794                 else
795                         avail_tree_table[hc] = f->next_avail;
796         } else {
797                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
798                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
799                 f->entry_capacity = cnt;
800         }
801
802         t = (struct tree_content*)f;
803         t->entry_count = 0;
804         t->delta_depth = 0;
805         return t;
806 }
807
808 static void release_tree_entry(struct tree_entry *e);
809 static void release_tree_content(struct tree_content *t)
810 {
811         struct avail_tree_content *f = (struct avail_tree_content*)t;
812         unsigned int hc = hc_entries(f->entry_capacity);
813         f->next_avail = avail_tree_table[hc];
814         avail_tree_table[hc] = f;
815 }
816
817 static void release_tree_content_recursive(struct tree_content *t)
818 {
819         unsigned int i;
820         for (i = 0; i < t->entry_count; i++)
821                 release_tree_entry(t->entries[i]);
822         release_tree_content(t);
823 }
824
825 static struct tree_content *grow_tree_content(
826         struct tree_content *t,
827         int amt)
828 {
829         struct tree_content *r = new_tree_content(t->entry_count + amt);
830         r->entry_count = t->entry_count;
831         r->delta_depth = t->delta_depth;
832         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
833         release_tree_content(t);
834         return r;
835 }
836
837 static struct tree_entry *new_tree_entry(void)
838 {
839         struct tree_entry *e;
840
841         if (!avail_tree_entry) {
842                 unsigned int n = tree_entry_alloc;
843                 total_allocd += n * sizeof(struct tree_entry);
844                 ALLOC_ARRAY(e, n);
845                 avail_tree_entry = e;
846                 while (n-- > 1) {
847                         *((void**)e) = e + 1;
848                         e++;
849                 }
850                 *((void**)e) = NULL;
851         }
852
853         e = avail_tree_entry;
854         avail_tree_entry = *((void**)e);
855         return e;
856 }
857
858 static void release_tree_entry(struct tree_entry *e)
859 {
860         if (e->tree)
861                 release_tree_content_recursive(e->tree);
862         *((void**)e) = avail_tree_entry;
863         avail_tree_entry = e;
864 }
865
866 static struct tree_content *dup_tree_content(struct tree_content *s)
867 {
868         struct tree_content *d;
869         struct tree_entry *a, *b;
870         unsigned int i;
871
872         if (!s)
873                 return NULL;
874         d = new_tree_content(s->entry_count);
875         for (i = 0; i < s->entry_count; i++) {
876                 a = s->entries[i];
877                 b = new_tree_entry();
878                 memcpy(b, a, sizeof(*a));
879                 if (a->tree && is_null_sha1(b->versions[1].sha1))
880                         b->tree = dup_tree_content(a->tree);
881                 else
882                         b->tree = NULL;
883                 d->entries[i] = b;
884         }
885         d->entry_count = s->entry_count;
886         d->delta_depth = s->delta_depth;
887
888         return d;
889 }
890
891 static void start_packfile(void)
892 {
893         static char tmp_file[PATH_MAX];
894         struct packed_git *p;
895         struct pack_header hdr;
896         int pack_fd;
897
898         pack_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
899                               "pack/tmp_pack_XXXXXX");
900         FLEX_ALLOC_STR(p, pack_name, tmp_file);
901         p->pack_fd = pack_fd;
902         p->do_not_close = 1;
903         pack_file = sha1fd(pack_fd, p->pack_name);
904
905         hdr.hdr_signature = htonl(PACK_SIGNATURE);
906         hdr.hdr_version = htonl(2);
907         hdr.hdr_entries = 0;
908         sha1write(pack_file, &hdr, sizeof(hdr));
909
910         pack_data = p;
911         pack_size = sizeof(hdr);
912         object_count = 0;
913
914         REALLOC_ARRAY(all_packs, pack_id + 1);
915         all_packs[pack_id] = p;
916 }
917
918 static const char *create_index(void)
919 {
920         const char *tmpfile;
921         struct pack_idx_entry **idx, **c, **last;
922         struct object_entry *e;
923         struct object_entry_pool *o;
924
925         /* Build the table of object IDs. */
926         ALLOC_ARRAY(idx, object_count);
927         c = idx;
928         for (o = blocks; o; o = o->next_pool)
929                 for (e = o->next_free; e-- != o->entries;)
930                         if (pack_id == e->pack_id)
931                                 *c++ = &e->idx;
932         last = idx + object_count;
933         if (c != last)
934                 die("internal consistency error creating the index");
935
936         tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
937         free(idx);
938         return tmpfile;
939 }
940
941 static char *keep_pack(const char *curr_index_name)
942 {
943         static char name[PATH_MAX];
944         static const char *keep_msg = "fast-import";
945         int keep_fd;
946
947         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
948         if (keep_fd < 0)
949                 die_errno("cannot create keep file");
950         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
951         if (close(keep_fd))
952                 die_errno("failed to write keep file");
953
954         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
955                  get_object_directory(), sha1_to_hex(pack_data->sha1));
956         if (finalize_object_file(pack_data->pack_name, name))
957                 die("cannot store pack file");
958
959         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
960                  get_object_directory(), sha1_to_hex(pack_data->sha1));
961         if (finalize_object_file(curr_index_name, name))
962                 die("cannot store index file");
963         free((void *)curr_index_name);
964         return name;
965 }
966
967 static void unkeep_all_packs(void)
968 {
969         static char name[PATH_MAX];
970         int k;
971
972         for (k = 0; k < pack_id; k++) {
973                 struct packed_git *p = all_packs[k];
974                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
975                          get_object_directory(), sha1_to_hex(p->sha1));
976                 unlink_or_warn(name);
977         }
978 }
979
980 static int loosen_small_pack(const struct packed_git *p)
981 {
982         struct child_process unpack = CHILD_PROCESS_INIT;
983
984         if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
985                 die_errno("Failed seeking to start of '%s'", p->pack_name);
986
987         unpack.in = p->pack_fd;
988         unpack.git_cmd = 1;
989         unpack.stdout_to_stderr = 1;
990         argv_array_push(&unpack.args, "unpack-objects");
991         if (!show_stats)
992                 argv_array_push(&unpack.args, "-q");
993
994         return run_command(&unpack);
995 }
996
997 static void end_packfile(void)
998 {
999         static int running;
1000
1001         if (running || !pack_data)
1002                 return;
1003
1004         running = 1;
1005         clear_delta_base_cache();
1006         if (object_count) {
1007                 struct packed_git *new_p;
1008                 unsigned char cur_pack_sha1[20];
1009                 char *idx_name;
1010                 int i;
1011                 struct branch *b;
1012                 struct tag *t;
1013
1014                 close_pack_windows(pack_data);
1015                 sha1close(pack_file, cur_pack_sha1, 0);
1016                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
1017                                     pack_data->pack_name, object_count,
1018                                     cur_pack_sha1, pack_size);
1019
1020                 if (object_count <= unpack_limit) {
1021                         if (!loosen_small_pack(pack_data)) {
1022                                 invalidate_pack_id(pack_id);
1023                                 goto discard_pack;
1024                         }
1025                 }
1026
1027                 close(pack_data->pack_fd);
1028                 idx_name = keep_pack(create_index());
1029
1030                 /* Register the packfile with core git's machinery. */
1031                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
1032                 if (!new_p)
1033                         die("core git rejected index %s", idx_name);
1034                 all_packs[pack_id] = new_p;
1035                 install_packed_git(new_p);
1036
1037                 /* Print the boundary */
1038                 if (pack_edges) {
1039                         fprintf(pack_edges, "%s:", new_p->pack_name);
1040                         for (i = 0; i < branch_table_sz; i++) {
1041                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
1042                                         if (b->pack_id == pack_id)
1043                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
1044                                 }
1045                         }
1046                         for (t = first_tag; t; t = t->next_tag) {
1047                                 if (t->pack_id == pack_id)
1048                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
1049                         }
1050                         fputc('\n', pack_edges);
1051                         fflush(pack_edges);
1052                 }
1053
1054                 pack_id++;
1055         }
1056         else {
1057 discard_pack:
1058                 close(pack_data->pack_fd);
1059                 unlink_or_warn(pack_data->pack_name);
1060         }
1061         free(pack_data);
1062         pack_data = NULL;
1063         running = 0;
1064
1065         /* We can't carry a delta across packfiles. */
1066         strbuf_release(&last_blob.data);
1067         last_blob.offset = 0;
1068         last_blob.depth = 0;
1069 }
1070
1071 static void cycle_packfile(void)
1072 {
1073         end_packfile();
1074         start_packfile();
1075 }
1076
1077 static int store_object(
1078         enum object_type type,
1079         struct strbuf *dat,
1080         struct last_object *last,
1081         unsigned char *sha1out,
1082         uintmax_t mark)
1083 {
1084         void *out, *delta;
1085         struct object_entry *e;
1086         unsigned char hdr[96];
1087         unsigned char sha1[20];
1088         unsigned long hdrlen, deltalen;
1089         git_SHA_CTX c;
1090         git_zstream s;
1091
1092         hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1093                            typename(type), (unsigned long)dat->len) + 1;
1094         git_SHA1_Init(&c);
1095         git_SHA1_Update(&c, hdr, hdrlen);
1096         git_SHA1_Update(&c, dat->buf, dat->len);
1097         git_SHA1_Final(sha1, &c);
1098         if (sha1out)
1099                 hashcpy(sha1out, sha1);
1100
1101         e = insert_object(sha1);
1102         if (mark)
1103                 insert_mark(mark, e);
1104         if (e->idx.offset) {
1105                 duplicate_count_by_type[type]++;
1106                 return 1;
1107         } else if (find_sha1_pack(sha1, packed_git)) {
1108                 e->type = type;
1109                 e->pack_id = MAX_PACK_ID;
1110                 e->idx.offset = 1; /* just not zero! */
1111                 duplicate_count_by_type[type]++;
1112                 return 1;
1113         }
1114
1115         if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1116                 delta_count_attempts_by_type[type]++;
1117                 delta = diff_delta(last->data.buf, last->data.len,
1118                         dat->buf, dat->len,
1119                         &deltalen, dat->len - 20);
1120         } else
1121                 delta = NULL;
1122
1123         git_deflate_init(&s, pack_compression_level);
1124         if (delta) {
1125                 s.next_in = delta;
1126                 s.avail_in = deltalen;
1127         } else {
1128                 s.next_in = (void *)dat->buf;
1129                 s.avail_in = dat->len;
1130         }
1131         s.avail_out = git_deflate_bound(&s, s.avail_in);
1132         s.next_out = out = xmalloc(s.avail_out);
1133         while (git_deflate(&s, Z_FINISH) == Z_OK)
1134                 ; /* nothing */
1135         git_deflate_end(&s);
1136
1137         /* Determine if we should auto-checkpoint. */
1138         if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1139                 || (pack_size + 60 + s.total_out) < pack_size) {
1140
1141                 /* This new object needs to *not* have the current pack_id. */
1142                 e->pack_id = pack_id + 1;
1143                 cycle_packfile();
1144
1145                 /* We cannot carry a delta into the new pack. */
1146                 if (delta) {
1147                         free(delta);
1148                         delta = NULL;
1149
1150                         git_deflate_init(&s, pack_compression_level);
1151                         s.next_in = (void *)dat->buf;
1152                         s.avail_in = dat->len;
1153                         s.avail_out = git_deflate_bound(&s, s.avail_in);
1154                         s.next_out = out = xrealloc(out, s.avail_out);
1155                         while (git_deflate(&s, Z_FINISH) == Z_OK)
1156                                 ; /* nothing */
1157                         git_deflate_end(&s);
1158                 }
1159         }
1160
1161         e->type = type;
1162         e->pack_id = pack_id;
1163         e->idx.offset = pack_size;
1164         object_count++;
1165         object_count_by_type[type]++;
1166
1167         crc32_begin(pack_file);
1168
1169         if (delta) {
1170                 off_t ofs = e->idx.offset - last->offset;
1171                 unsigned pos = sizeof(hdr) - 1;
1172
1173                 delta_count_by_type[type]++;
1174                 e->depth = last->depth + 1;
1175
1176                 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1177                 sha1write(pack_file, hdr, hdrlen);
1178                 pack_size += hdrlen;
1179
1180                 hdr[pos] = ofs & 127;
1181                 while (ofs >>= 7)
1182                         hdr[--pos] = 128 | (--ofs & 127);
1183                 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1184                 pack_size += sizeof(hdr) - pos;
1185         } else {
1186                 e->depth = 0;
1187                 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1188                 sha1write(pack_file, hdr, hdrlen);
1189                 pack_size += hdrlen;
1190         }
1191
1192         sha1write(pack_file, out, s.total_out);
1193         pack_size += s.total_out;
1194
1195         e->idx.crc32 = crc32_end(pack_file);
1196
1197         free(out);
1198         free(delta);
1199         if (last) {
1200                 if (last->no_swap) {
1201                         last->data = *dat;
1202                 } else {
1203                         strbuf_swap(&last->data, dat);
1204                 }
1205                 last->offset = e->idx.offset;
1206                 last->depth = e->depth;
1207         }
1208         return 0;
1209 }
1210
1211 static void truncate_pack(struct sha1file_checkpoint *checkpoint)
1212 {
1213         if (sha1file_truncate(pack_file, checkpoint))
1214                 die_errno("cannot truncate pack to skip duplicate");
1215         pack_size = checkpoint->offset;
1216 }
1217
1218 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1219 {
1220         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1221         unsigned char *in_buf = xmalloc(in_sz);
1222         unsigned char *out_buf = xmalloc(out_sz);
1223         struct object_entry *e;
1224         unsigned char sha1[20];
1225         unsigned long hdrlen;
1226         off_t offset;
1227         git_SHA_CTX c;
1228         git_zstream s;
1229         struct sha1file_checkpoint checkpoint;
1230         int status = Z_OK;
1231
1232         /* Determine if we should auto-checkpoint. */
1233         if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1234                 || (pack_size + 60 + len) < pack_size)
1235                 cycle_packfile();
1236
1237         sha1file_checkpoint(pack_file, &checkpoint);
1238         offset = checkpoint.offset;
1239
1240         hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1241         if (out_sz <= hdrlen)
1242                 die("impossibly large object header");
1243
1244         git_SHA1_Init(&c);
1245         git_SHA1_Update(&c, out_buf, hdrlen);
1246
1247         crc32_begin(pack_file);
1248
1249         git_deflate_init(&s, pack_compression_level);
1250
1251         hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1252         if (out_sz <= hdrlen)
1253                 die("impossibly large object header");
1254
1255         s.next_out = out_buf + hdrlen;
1256         s.avail_out = out_sz - hdrlen;
1257
1258         while (status != Z_STREAM_END) {
1259                 if (0 < len && !s.avail_in) {
1260                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1261                         size_t n = fread(in_buf, 1, cnt, stdin);
1262                         if (!n && feof(stdin))
1263                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1264
1265                         git_SHA1_Update(&c, in_buf, n);
1266                         s.next_in = in_buf;
1267                         s.avail_in = n;
1268                         len -= n;
1269                 }
1270
1271                 status = git_deflate(&s, len ? 0 : Z_FINISH);
1272
1273                 if (!s.avail_out || status == Z_STREAM_END) {
1274                         size_t n = s.next_out - out_buf;
1275                         sha1write(pack_file, out_buf, n);
1276                         pack_size += n;
1277                         s.next_out = out_buf;
1278                         s.avail_out = out_sz;
1279                 }
1280
1281                 switch (status) {
1282                 case Z_OK:
1283                 case Z_BUF_ERROR:
1284                 case Z_STREAM_END:
1285                         continue;
1286                 default:
1287                         die("unexpected deflate failure: %d", status);
1288                 }
1289         }
1290         git_deflate_end(&s);
1291         git_SHA1_Final(sha1, &c);
1292
1293         if (sha1out)
1294                 hashcpy(sha1out, sha1);
1295
1296         e = insert_object(sha1);
1297
1298         if (mark)
1299                 insert_mark(mark, e);
1300
1301         if (e->idx.offset) {
1302                 duplicate_count_by_type[OBJ_BLOB]++;
1303                 truncate_pack(&checkpoint);
1304
1305         } else if (find_sha1_pack(sha1, packed_git)) {
1306                 e->type = OBJ_BLOB;
1307                 e->pack_id = MAX_PACK_ID;
1308                 e->idx.offset = 1; /* just not zero! */
1309                 duplicate_count_by_type[OBJ_BLOB]++;
1310                 truncate_pack(&checkpoint);
1311
1312         } else {
1313                 e->depth = 0;
1314                 e->type = OBJ_BLOB;
1315                 e->pack_id = pack_id;
1316                 e->idx.offset = offset;
1317                 e->idx.crc32 = crc32_end(pack_file);
1318                 object_count++;
1319                 object_count_by_type[OBJ_BLOB]++;
1320         }
1321
1322         free(in_buf);
1323         free(out_buf);
1324 }
1325
1326 /* All calls must be guarded by find_object() or find_mark() to
1327  * ensure the 'struct object_entry' passed was written by this
1328  * process instance.  We unpack the entry by the offset, avoiding
1329  * the need for the corresponding .idx file.  This unpacking rule
1330  * works because we only use OBJ_REF_DELTA within the packfiles
1331  * created by fast-import.
1332  *
1333  * oe must not be NULL.  Such an oe usually comes from giving
1334  * an unknown SHA-1 to find_object() or an undefined mark to
1335  * find_mark().  Callers must test for this condition and use
1336  * the standard read_sha1_file() when it happens.
1337  *
1338  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1339  * find_mark(), where the mark was reloaded from an existing marks
1340  * file and is referencing an object that this fast-import process
1341  * instance did not write out to a packfile.  Callers must test for
1342  * this condition and use read_sha1_file() instead.
1343  */
1344 static void *gfi_unpack_entry(
1345         struct object_entry *oe,
1346         unsigned long *sizep)
1347 {
1348         enum object_type type;
1349         struct packed_git *p = all_packs[oe->pack_id];
1350         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1351                 /* The object is stored in the packfile we are writing to
1352                  * and we have modified it since the last time we scanned
1353                  * back to read a previously written object.  If an old
1354                  * window covered [p->pack_size, p->pack_size + 20) its
1355                  * data is stale and is not valid.  Closing all windows
1356                  * and updating the packfile length ensures we can read
1357                  * the newly written data.
1358                  */
1359                 close_pack_windows(p);
1360                 sha1flush(pack_file);
1361
1362                 /* We have to offer 20 bytes additional on the end of
1363                  * the packfile as the core unpacker code assumes the
1364                  * footer is present at the file end and must promise
1365                  * at least 20 bytes within any window it maps.  But
1366                  * we don't actually create the footer here.
1367                  */
1368                 p->pack_size = pack_size + 20;
1369         }
1370         return unpack_entry(p, oe->idx.offset, &type, sizep);
1371 }
1372
1373 static const char *get_mode(const char *str, uint16_t *modep)
1374 {
1375         unsigned char c;
1376         uint16_t mode = 0;
1377
1378         while ((c = *str++) != ' ') {
1379                 if (c < '0' || c > '7')
1380                         return NULL;
1381                 mode = (mode << 3) + (c - '0');
1382         }
1383         *modep = mode;
1384         return str;
1385 }
1386
1387 static void load_tree(struct tree_entry *root)
1388 {
1389         unsigned char *sha1 = root->versions[1].sha1;
1390         struct object_entry *myoe;
1391         struct tree_content *t;
1392         unsigned long size;
1393         char *buf;
1394         const char *c;
1395
1396         root->tree = t = new_tree_content(8);
1397         if (is_null_sha1(sha1))
1398                 return;
1399
1400         myoe = find_object(sha1);
1401         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1402                 if (myoe->type != OBJ_TREE)
1403                         die("Not a tree: %s", sha1_to_hex(sha1));
1404                 t->delta_depth = myoe->depth;
1405                 buf = gfi_unpack_entry(myoe, &size);
1406                 if (!buf)
1407                         die("Can't load tree %s", sha1_to_hex(sha1));
1408         } else {
1409                 enum object_type type;
1410                 buf = read_sha1_file(sha1, &type, &size);
1411                 if (!buf || type != OBJ_TREE)
1412                         die("Can't load tree %s", sha1_to_hex(sha1));
1413         }
1414
1415         c = buf;
1416         while (c != (buf + size)) {
1417                 struct tree_entry *e = new_tree_entry();
1418
1419                 if (t->entry_count == t->entry_capacity)
1420                         root->tree = t = grow_tree_content(t, t->entry_count);
1421                 t->entries[t->entry_count++] = e;
1422
1423                 e->tree = NULL;
1424                 c = get_mode(c, &e->versions[1].mode);
1425                 if (!c)
1426                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1427                 e->versions[0].mode = e->versions[1].mode;
1428                 e->name = to_atom(c, strlen(c));
1429                 c += e->name->str_len + 1;
1430                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1431                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1432                 c += 20;
1433         }
1434         free(buf);
1435 }
1436
1437 static int tecmp0 (const void *_a, const void *_b)
1438 {
1439         struct tree_entry *a = *((struct tree_entry**)_a);
1440         struct tree_entry *b = *((struct tree_entry**)_b);
1441         return base_name_compare(
1442                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1443                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1444 }
1445
1446 static int tecmp1 (const void *_a, const void *_b)
1447 {
1448         struct tree_entry *a = *((struct tree_entry**)_a);
1449         struct tree_entry *b = *((struct tree_entry**)_b);
1450         return base_name_compare(
1451                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1452                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1453 }
1454
1455 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1456 {
1457         size_t maxlen = 0;
1458         unsigned int i;
1459
1460         if (!v)
1461                 QSORT(t->entries, t->entry_count, tecmp0);
1462         else
1463                 QSORT(t->entries, t->entry_count, tecmp1);
1464
1465         for (i = 0; i < t->entry_count; i++) {
1466                 if (t->entries[i]->versions[v].mode)
1467                         maxlen += t->entries[i]->name->str_len + 34;
1468         }
1469
1470         strbuf_reset(b);
1471         strbuf_grow(b, maxlen);
1472         for (i = 0; i < t->entry_count; i++) {
1473                 struct tree_entry *e = t->entries[i];
1474                 if (!e->versions[v].mode)
1475                         continue;
1476                 strbuf_addf(b, "%o %s%c",
1477                         (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1478                         e->name->str_dat, '\0');
1479                 strbuf_add(b, e->versions[v].sha1, 20);
1480         }
1481 }
1482
1483 static void store_tree(struct tree_entry *root)
1484 {
1485         struct tree_content *t;
1486         unsigned int i, j, del;
1487         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1488         struct object_entry *le = NULL;
1489
1490         if (!is_null_sha1(root->versions[1].sha1))
1491                 return;
1492
1493         if (!root->tree)
1494                 load_tree(root);
1495         t = root->tree;
1496
1497         for (i = 0; i < t->entry_count; i++) {
1498                 if (t->entries[i]->tree)
1499                         store_tree(t->entries[i]);
1500         }
1501
1502         if (!(root->versions[0].mode & NO_DELTA))
1503                 le = find_object(root->versions[0].sha1);
1504         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1505                 mktree(t, 0, &old_tree);
1506                 lo.data = old_tree;
1507                 lo.offset = le->idx.offset;
1508                 lo.depth = t->delta_depth;
1509         }
1510
1511         mktree(t, 1, &new_tree);
1512         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1513
1514         t->delta_depth = lo.depth;
1515         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1516                 struct tree_entry *e = t->entries[i];
1517                 if (e->versions[1].mode) {
1518                         e->versions[0].mode = e->versions[1].mode;
1519                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1520                         t->entries[j++] = e;
1521                 } else {
1522                         release_tree_entry(e);
1523                         del++;
1524                 }
1525         }
1526         t->entry_count -= del;
1527 }
1528
1529 static void tree_content_replace(
1530         struct tree_entry *root,
1531         const unsigned char *sha1,
1532         const uint16_t mode,
1533         struct tree_content *newtree)
1534 {
1535         if (!S_ISDIR(mode))
1536                 die("Root cannot be a non-directory");
1537         hashclr(root->versions[0].sha1);
1538         hashcpy(root->versions[1].sha1, sha1);
1539         if (root->tree)
1540                 release_tree_content_recursive(root->tree);
1541         root->tree = newtree;
1542 }
1543
1544 static int tree_content_set(
1545         struct tree_entry *root,
1546         const char *p,
1547         const unsigned char *sha1,
1548         const uint16_t mode,
1549         struct tree_content *subtree)
1550 {
1551         struct tree_content *t;
1552         const char *slash1;
1553         unsigned int i, n;
1554         struct tree_entry *e;
1555
1556         slash1 = strchrnul(p, '/');
1557         n = slash1 - p;
1558         if (!n)
1559                 die("Empty path component found in input");
1560         if (!*slash1 && !S_ISDIR(mode) && subtree)
1561                 die("Non-directories cannot have subtrees");
1562
1563         if (!root->tree)
1564                 load_tree(root);
1565         t = root->tree;
1566         for (i = 0; i < t->entry_count; i++) {
1567                 e = t->entries[i];
1568                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1569                         if (!*slash1) {
1570                                 if (!S_ISDIR(mode)
1571                                                 && e->versions[1].mode == mode
1572                                                 && !hashcmp(e->versions[1].sha1, sha1))
1573                                         return 0;
1574                                 e->versions[1].mode = mode;
1575                                 hashcpy(e->versions[1].sha1, sha1);
1576                                 if (e->tree)
1577                                         release_tree_content_recursive(e->tree);
1578                                 e->tree = subtree;
1579
1580                                 /*
1581                                  * We need to leave e->versions[0].sha1 alone
1582                                  * to avoid modifying the preimage tree used
1583                                  * when writing out the parent directory.
1584                                  * But after replacing the subdir with a
1585                                  * completely different one, it's not a good
1586                                  * delta base any more, and besides, we've
1587                                  * thrown away the tree entries needed to
1588                                  * make a delta against it.
1589                                  *
1590                                  * So let's just explicitly disable deltas
1591                                  * for the subtree.
1592                                  */
1593                                 if (S_ISDIR(e->versions[0].mode))
1594                                         e->versions[0].mode |= NO_DELTA;
1595
1596                                 hashclr(root->versions[1].sha1);
1597                                 return 1;
1598                         }
1599                         if (!S_ISDIR(e->versions[1].mode)) {
1600                                 e->tree = new_tree_content(8);
1601                                 e->versions[1].mode = S_IFDIR;
1602                         }
1603                         if (!e->tree)
1604                                 load_tree(e);
1605                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1606                                 hashclr(root->versions[1].sha1);
1607                                 return 1;
1608                         }
1609                         return 0;
1610                 }
1611         }
1612
1613         if (t->entry_count == t->entry_capacity)
1614                 root->tree = t = grow_tree_content(t, t->entry_count);
1615         e = new_tree_entry();
1616         e->name = to_atom(p, n);
1617         e->versions[0].mode = 0;
1618         hashclr(e->versions[0].sha1);
1619         t->entries[t->entry_count++] = e;
1620         if (*slash1) {
1621                 e->tree = new_tree_content(8);
1622                 e->versions[1].mode = S_IFDIR;
1623                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1624         } else {
1625                 e->tree = subtree;
1626                 e->versions[1].mode = mode;
1627                 hashcpy(e->versions[1].sha1, sha1);
1628         }
1629         hashclr(root->versions[1].sha1);
1630         return 1;
1631 }
1632
1633 static int tree_content_remove(
1634         struct tree_entry *root,
1635         const char *p,
1636         struct tree_entry *backup_leaf,
1637         int allow_root)
1638 {
1639         struct tree_content *t;
1640         const char *slash1;
1641         unsigned int i, n;
1642         struct tree_entry *e;
1643
1644         slash1 = strchrnul(p, '/');
1645         n = slash1 - p;
1646
1647         if (!root->tree)
1648                 load_tree(root);
1649
1650         if (!*p && allow_root) {
1651                 e = root;
1652                 goto del_entry;
1653         }
1654
1655         t = root->tree;
1656         for (i = 0; i < t->entry_count; i++) {
1657                 e = t->entries[i];
1658                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1659                         if (*slash1 && !S_ISDIR(e->versions[1].mode))
1660                                 /*
1661                                  * If p names a file in some subdirectory, and a
1662                                  * file or symlink matching the name of the
1663                                  * parent directory of p exists, then p cannot
1664                                  * exist and need not be deleted.
1665                                  */
1666                                 return 1;
1667                         if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1668                                 goto del_entry;
1669                         if (!e->tree)
1670                                 load_tree(e);
1671                         if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1672                                 for (n = 0; n < e->tree->entry_count; n++) {
1673                                         if (e->tree->entries[n]->versions[1].mode) {
1674                                                 hashclr(root->versions[1].sha1);
1675                                                 return 1;
1676                                         }
1677                                 }
1678                                 backup_leaf = NULL;
1679                                 goto del_entry;
1680                         }
1681                         return 0;
1682                 }
1683         }
1684         return 0;
1685
1686 del_entry:
1687         if (backup_leaf)
1688                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1689         else if (e->tree)
1690                 release_tree_content_recursive(e->tree);
1691         e->tree = NULL;
1692         e->versions[1].mode = 0;
1693         hashclr(e->versions[1].sha1);
1694         hashclr(root->versions[1].sha1);
1695         return 1;
1696 }
1697
1698 static int tree_content_get(
1699         struct tree_entry *root,
1700         const char *p,
1701         struct tree_entry *leaf,
1702         int allow_root)
1703 {
1704         struct tree_content *t;
1705         const char *slash1;
1706         unsigned int i, n;
1707         struct tree_entry *e;
1708
1709         slash1 = strchrnul(p, '/');
1710         n = slash1 - p;
1711         if (!n && !allow_root)
1712                 die("Empty path component found in input");
1713
1714         if (!root->tree)
1715                 load_tree(root);
1716
1717         if (!n) {
1718                 e = root;
1719                 goto found_entry;
1720         }
1721
1722         t = root->tree;
1723         for (i = 0; i < t->entry_count; i++) {
1724                 e = t->entries[i];
1725                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1726                         if (!*slash1)
1727                                 goto found_entry;
1728                         if (!S_ISDIR(e->versions[1].mode))
1729                                 return 0;
1730                         if (!e->tree)
1731                                 load_tree(e);
1732                         return tree_content_get(e, slash1 + 1, leaf, 0);
1733                 }
1734         }
1735         return 0;
1736
1737 found_entry:
1738         memcpy(leaf, e, sizeof(*leaf));
1739         if (e->tree && is_null_sha1(e->versions[1].sha1))
1740                 leaf->tree = dup_tree_content(e->tree);
1741         else
1742                 leaf->tree = NULL;
1743         return 1;
1744 }
1745
1746 static int update_branch(struct branch *b)
1747 {
1748         static const char *msg = "fast-import";
1749         struct ref_transaction *transaction;
1750         unsigned char old_sha1[20];
1751         struct strbuf err = STRBUF_INIT;
1752
1753         if (is_null_sha1(b->sha1)) {
1754                 if (b->delete)
1755                         delete_ref(b->name, NULL, 0);
1756                 return 0;
1757         }
1758         if (read_ref(b->name, old_sha1))
1759                 hashclr(old_sha1);
1760         if (!force_update && !is_null_sha1(old_sha1)) {
1761                 struct commit *old_cmit, *new_cmit;
1762
1763                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1764                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1765                 if (!old_cmit || !new_cmit)
1766                         return error("Branch %s is missing commits.", b->name);
1767
1768                 if (!in_merge_bases(old_cmit, new_cmit)) {
1769                         warning("Not updating %s"
1770                                 " (new tip %s does not contain %s)",
1771                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1772                         return -1;
1773                 }
1774         }
1775         transaction = ref_transaction_begin(&err);
1776         if (!transaction ||
1777             ref_transaction_update(transaction, b->name, b->sha1, old_sha1,
1778                                    0, msg, &err) ||
1779             ref_transaction_commit(transaction, &err)) {
1780                 ref_transaction_free(transaction);
1781                 error("%s", err.buf);
1782                 strbuf_release(&err);
1783                 return -1;
1784         }
1785         ref_transaction_free(transaction);
1786         strbuf_release(&err);
1787         return 0;
1788 }
1789
1790 static void dump_branches(void)
1791 {
1792         unsigned int i;
1793         struct branch *b;
1794
1795         for (i = 0; i < branch_table_sz; i++) {
1796                 for (b = branch_table[i]; b; b = b->table_next_branch)
1797                         failure |= update_branch(b);
1798         }
1799 }
1800
1801 static void dump_tags(void)
1802 {
1803         static const char *msg = "fast-import";
1804         struct tag *t;
1805         struct strbuf ref_name = STRBUF_INIT;
1806         struct strbuf err = STRBUF_INIT;
1807         struct ref_transaction *transaction;
1808
1809         transaction = ref_transaction_begin(&err);
1810         if (!transaction) {
1811                 failure |= error("%s", err.buf);
1812                 goto cleanup;
1813         }
1814         for (t = first_tag; t; t = t->next_tag) {
1815                 strbuf_reset(&ref_name);
1816                 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1817
1818                 if (ref_transaction_update(transaction, ref_name.buf,
1819                                            t->sha1, NULL, 0, msg, &err)) {
1820                         failure |= error("%s", err.buf);
1821                         goto cleanup;
1822                 }
1823         }
1824         if (ref_transaction_commit(transaction, &err))
1825                 failure |= error("%s", err.buf);
1826
1827  cleanup:
1828         ref_transaction_free(transaction);
1829         strbuf_release(&ref_name);
1830         strbuf_release(&err);
1831 }
1832
1833 static void dump_marks_helper(FILE *f,
1834         uintmax_t base,
1835         struct mark_set *m)
1836 {
1837         uintmax_t k;
1838         if (m->shift) {
1839                 for (k = 0; k < 1024; k++) {
1840                         if (m->data.sets[k])
1841                                 dump_marks_helper(f, base + (k << m->shift),
1842                                         m->data.sets[k]);
1843                 }
1844         } else {
1845                 for (k = 0; k < 1024; k++) {
1846                         if (m->data.marked[k])
1847                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1848                                         sha1_to_hex(m->data.marked[k]->idx.sha1));
1849                 }
1850         }
1851 }
1852
1853 static void dump_marks(void)
1854 {
1855         static struct lock_file mark_lock;
1856         FILE *f;
1857
1858         if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1859                 return;
1860
1861         if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1862                 failure |= error_errno("Unable to write marks file %s",
1863                                        export_marks_file);
1864                 return;
1865         }
1866
1867         f = fdopen_lock_file(&mark_lock, "w");
1868         if (!f) {
1869                 int saved_errno = errno;
1870                 rollback_lock_file(&mark_lock);
1871                 failure |= error("Unable to write marks file %s: %s",
1872                         export_marks_file, strerror(saved_errno));
1873                 return;
1874         }
1875
1876         dump_marks_helper(f, 0, marks);
1877         if (commit_lock_file(&mark_lock)) {
1878                 failure |= error_errno("Unable to write file %s",
1879                                        export_marks_file);
1880                 return;
1881         }
1882 }
1883
1884 static void read_marks(void)
1885 {
1886         char line[512];
1887         FILE *f = fopen(import_marks_file, "r");
1888         if (f)
1889                 ;
1890         else if (import_marks_file_ignore_missing && errno == ENOENT)
1891                 goto done; /* Marks file does not exist */
1892         else
1893                 die_errno("cannot read '%s'", import_marks_file);
1894         while (fgets(line, sizeof(line), f)) {
1895                 uintmax_t mark;
1896                 char *end;
1897                 unsigned char sha1[20];
1898                 struct object_entry *e;
1899
1900                 end = strchr(line, '\n');
1901                 if (line[0] != ':' || !end)
1902                         die("corrupt mark line: %s", line);
1903                 *end = 0;
1904                 mark = strtoumax(line + 1, &end, 10);
1905                 if (!mark || end == line + 1
1906                         || *end != ' ' || get_sha1_hex(end + 1, sha1))
1907                         die("corrupt mark line: %s", line);
1908                 e = find_object(sha1);
1909                 if (!e) {
1910                         enum object_type type = sha1_object_info(sha1, NULL);
1911                         if (type < 0)
1912                                 die("object not found: %s", sha1_to_hex(sha1));
1913                         e = insert_object(sha1);
1914                         e->type = type;
1915                         e->pack_id = MAX_PACK_ID;
1916                         e->idx.offset = 1; /* just not zero! */
1917                 }
1918                 insert_mark(mark, e);
1919         }
1920         fclose(f);
1921 done:
1922         import_marks_file_done = 1;
1923 }
1924
1925
1926 static int read_next_command(void)
1927 {
1928         static int stdin_eof = 0;
1929
1930         if (stdin_eof) {
1931                 unread_command_buf = 0;
1932                 return EOF;
1933         }
1934
1935         for (;;) {
1936                 const char *p;
1937
1938                 if (unread_command_buf) {
1939                         unread_command_buf = 0;
1940                 } else {
1941                         struct recent_command *rc;
1942
1943                         strbuf_detach(&command_buf, NULL);
1944                         stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1945                         if (stdin_eof)
1946                                 return EOF;
1947
1948                         if (!seen_data_command
1949                                 && !starts_with(command_buf.buf, "feature ")
1950                                 && !starts_with(command_buf.buf, "option ")) {
1951                                 parse_argv();
1952                         }
1953
1954                         rc = rc_free;
1955                         if (rc)
1956                                 rc_free = rc->next;
1957                         else {
1958                                 rc = cmd_hist.next;
1959                                 cmd_hist.next = rc->next;
1960                                 cmd_hist.next->prev = &cmd_hist;
1961                                 free(rc->buf);
1962                         }
1963
1964                         rc->buf = command_buf.buf;
1965                         rc->prev = cmd_tail;
1966                         rc->next = cmd_hist.prev;
1967                         rc->prev->next = rc;
1968                         cmd_tail = rc;
1969                 }
1970                 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1971                         parse_get_mark(p);
1972                         continue;
1973                 }
1974                 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1975                         parse_cat_blob(p);
1976                         continue;
1977                 }
1978                 if (command_buf.buf[0] == '#')
1979                         continue;
1980                 return 0;
1981         }
1982 }
1983
1984 static void skip_optional_lf(void)
1985 {
1986         int term_char = fgetc(stdin);
1987         if (term_char != '\n' && term_char != EOF)
1988                 ungetc(term_char, stdin);
1989 }
1990
1991 static void parse_mark(void)
1992 {
1993         const char *v;
1994         if (skip_prefix(command_buf.buf, "mark :", &v)) {
1995                 next_mark = strtoumax(v, NULL, 10);
1996                 read_next_command();
1997         }
1998         else
1999                 next_mark = 0;
2000 }
2001
2002 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
2003 {
2004         const char *data;
2005         strbuf_reset(sb);
2006
2007         if (!skip_prefix(command_buf.buf, "data ", &data))
2008                 die("Expected 'data n' command, found: %s", command_buf.buf);
2009
2010         if (skip_prefix(data, "<<", &data)) {
2011                 char *term = xstrdup(data);
2012                 size_t term_len = command_buf.len - (data - command_buf.buf);
2013
2014                 strbuf_detach(&command_buf, NULL);
2015                 for (;;) {
2016                         if (strbuf_getline_lf(&command_buf, stdin) == EOF)
2017                                 die("EOF in data (terminator '%s' not found)", term);
2018                         if (term_len == command_buf.len
2019                                 && !strcmp(term, command_buf.buf))
2020                                 break;
2021                         strbuf_addbuf(sb, &command_buf);
2022                         strbuf_addch(sb, '\n');
2023                 }
2024                 free(term);
2025         }
2026         else {
2027                 uintmax_t len = strtoumax(data, NULL, 10);
2028                 size_t n = 0, length = (size_t)len;
2029
2030                 if (limit && limit < len) {
2031                         *len_res = len;
2032                         return 0;
2033                 }
2034                 if (length < len)
2035                         die("data is too large to use in this context");
2036
2037                 while (n < length) {
2038                         size_t s = strbuf_fread(sb, length - n, stdin);
2039                         if (!s && feof(stdin))
2040                                 die("EOF in data (%lu bytes remaining)",
2041                                         (unsigned long)(length - n));
2042                         n += s;
2043                 }
2044         }
2045
2046         skip_optional_lf();
2047         return 1;
2048 }
2049
2050 static int validate_raw_date(const char *src, struct strbuf *result)
2051 {
2052         const char *orig_src = src;
2053         char *endp;
2054         unsigned long num;
2055
2056         errno = 0;
2057
2058         num = strtoul(src, &endp, 10);
2059         /* NEEDSWORK: perhaps check for reasonable values? */
2060         if (errno || endp == src || *endp != ' ')
2061                 return -1;
2062
2063         src = endp + 1;
2064         if (*src != '-' && *src != '+')
2065                 return -1;
2066
2067         num = strtoul(src + 1, &endp, 10);
2068         if (errno || endp == src + 1 || *endp || 1400 < num)
2069                 return -1;
2070
2071         strbuf_addstr(result, orig_src);
2072         return 0;
2073 }
2074
2075 static char *parse_ident(const char *buf)
2076 {
2077         const char *ltgt;
2078         size_t name_len;
2079         struct strbuf ident = STRBUF_INIT;
2080
2081         /* ensure there is a space delimiter even if there is no name */
2082         if (*buf == '<')
2083                 --buf;
2084
2085         ltgt = buf + strcspn(buf, "<>");
2086         if (*ltgt != '<')
2087                 die("Missing < in ident string: %s", buf);
2088         if (ltgt != buf && ltgt[-1] != ' ')
2089                 die("Missing space before < in ident string: %s", buf);
2090         ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2091         if (*ltgt != '>')
2092                 die("Missing > in ident string: %s", buf);
2093         ltgt++;
2094         if (*ltgt != ' ')
2095                 die("Missing space after > in ident string: %s", buf);
2096         ltgt++;
2097         name_len = ltgt - buf;
2098         strbuf_add(&ident, buf, name_len);
2099
2100         switch (whenspec) {
2101         case WHENSPEC_RAW:
2102                 if (validate_raw_date(ltgt, &ident) < 0)
2103                         die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2104                 break;
2105         case WHENSPEC_RFC2822:
2106                 if (parse_date(ltgt, &ident) < 0)
2107                         die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2108                 break;
2109         case WHENSPEC_NOW:
2110                 if (strcmp("now", ltgt))
2111                         die("Date in ident must be 'now': %s", buf);
2112                 datestamp(&ident);
2113                 break;
2114         }
2115
2116         return strbuf_detach(&ident, NULL);
2117 }
2118
2119 static void parse_and_store_blob(
2120         struct last_object *last,
2121         unsigned char *sha1out,
2122         uintmax_t mark)
2123 {
2124         static struct strbuf buf = STRBUF_INIT;
2125         uintmax_t len;
2126
2127         if (parse_data(&buf, big_file_threshold, &len))
2128                 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2129         else {
2130                 if (last) {
2131                         strbuf_release(&last->data);
2132                         last->offset = 0;
2133                         last->depth = 0;
2134                 }
2135                 stream_blob(len, sha1out, mark);
2136                 skip_optional_lf();
2137         }
2138 }
2139
2140 static void parse_new_blob(void)
2141 {
2142         read_next_command();
2143         parse_mark();
2144         parse_and_store_blob(&last_blob, NULL, next_mark);
2145 }
2146
2147 static void unload_one_branch(void)
2148 {
2149         while (cur_active_branches
2150                 && cur_active_branches >= max_active_branches) {
2151                 uintmax_t min_commit = ULONG_MAX;
2152                 struct branch *e, *l = NULL, *p = NULL;
2153
2154                 for (e = active_branches; e; e = e->active_next_branch) {
2155                         if (e->last_commit < min_commit) {
2156                                 p = l;
2157                                 min_commit = e->last_commit;
2158                         }
2159                         l = e;
2160                 }
2161
2162                 if (p) {
2163                         e = p->active_next_branch;
2164                         p->active_next_branch = e->active_next_branch;
2165                 } else {
2166                         e = active_branches;
2167                         active_branches = e->active_next_branch;
2168                 }
2169                 e->active = 0;
2170                 e->active_next_branch = NULL;
2171                 if (e->branch_tree.tree) {
2172                         release_tree_content_recursive(e->branch_tree.tree);
2173                         e->branch_tree.tree = NULL;
2174                 }
2175                 cur_active_branches--;
2176         }
2177 }
2178
2179 static void load_branch(struct branch *b)
2180 {
2181         load_tree(&b->branch_tree);
2182         if (!b->active) {
2183                 b->active = 1;
2184                 b->active_next_branch = active_branches;
2185                 active_branches = b;
2186                 cur_active_branches++;
2187                 branch_load_count++;
2188         }
2189 }
2190
2191 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2192 {
2193         unsigned char fanout = 0;
2194         while ((num_notes >>= 8))
2195                 fanout++;
2196         return fanout;
2197 }
2198
2199 static void construct_path_with_fanout(const char *hex_sha1,
2200                 unsigned char fanout, char *path)
2201 {
2202         unsigned int i = 0, j = 0;
2203         if (fanout >= 20)
2204                 die("Too large fanout (%u)", fanout);
2205         while (fanout) {
2206                 path[i++] = hex_sha1[j++];
2207                 path[i++] = hex_sha1[j++];
2208                 path[i++] = '/';
2209                 fanout--;
2210         }
2211         memcpy(path + i, hex_sha1 + j, 40 - j);
2212         path[i + 40 - j] = '\0';
2213 }
2214
2215 static uintmax_t do_change_note_fanout(
2216                 struct tree_entry *orig_root, struct tree_entry *root,
2217                 char *hex_sha1, unsigned int hex_sha1_len,
2218                 char *fullpath, unsigned int fullpath_len,
2219                 unsigned char fanout)
2220 {
2221         struct tree_content *t;
2222         struct tree_entry *e, leaf;
2223         unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2224         uintmax_t num_notes = 0;
2225         unsigned char sha1[20];
2226         char realpath[60];
2227
2228         if (!root->tree)
2229                 load_tree(root);
2230         t = root->tree;
2231
2232         for (i = 0; t && i < t->entry_count; i++) {
2233                 e = t->entries[i];
2234                 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2235                 tmp_fullpath_len = fullpath_len;
2236
2237                 /*
2238                  * We're interested in EITHER existing note entries (entries
2239                  * with exactly 40 hex chars in path, not including directory
2240                  * separators), OR directory entries that may contain note
2241                  * entries (with < 40 hex chars in path).
2242                  * Also, each path component in a note entry must be a multiple
2243                  * of 2 chars.
2244                  */
2245                 if (!e->versions[1].mode ||
2246                     tmp_hex_sha1_len > 40 ||
2247                     e->name->str_len % 2)
2248                         continue;
2249
2250                 /* This _may_ be a note entry, or a subdir containing notes */
2251                 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2252                        e->name->str_len);
2253                 if (tmp_fullpath_len)
2254                         fullpath[tmp_fullpath_len++] = '/';
2255                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2256                        e->name->str_len);
2257                 tmp_fullpath_len += e->name->str_len;
2258                 fullpath[tmp_fullpath_len] = '\0';
2259
2260                 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2261                         /* This is a note entry */
2262                         if (fanout == 0xff) {
2263                                 /* Counting mode, no rename */
2264                                 num_notes++;
2265                                 continue;
2266                         }
2267                         construct_path_with_fanout(hex_sha1, fanout, realpath);
2268                         if (!strcmp(fullpath, realpath)) {
2269                                 /* Note entry is in correct location */
2270                                 num_notes++;
2271                                 continue;
2272                         }
2273
2274                         /* Rename fullpath to realpath */
2275                         if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2276                                 die("Failed to remove path %s", fullpath);
2277                         tree_content_set(orig_root, realpath,
2278                                 leaf.versions[1].sha1,
2279                                 leaf.versions[1].mode,
2280                                 leaf.tree);
2281                 } else if (S_ISDIR(e->versions[1].mode)) {
2282                         /* This is a subdir that may contain note entries */
2283                         num_notes += do_change_note_fanout(orig_root, e,
2284                                 hex_sha1, tmp_hex_sha1_len,
2285                                 fullpath, tmp_fullpath_len, fanout);
2286                 }
2287
2288                 /* The above may have reallocated the current tree_content */
2289                 t = root->tree;
2290         }
2291         return num_notes;
2292 }
2293
2294 static uintmax_t change_note_fanout(struct tree_entry *root,
2295                 unsigned char fanout)
2296 {
2297         char hex_sha1[40], path[60];
2298         return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2299 }
2300
2301 /*
2302  * Given a pointer into a string, parse a mark reference:
2303  *
2304  *   idnum ::= ':' bigint;
2305  *
2306  * Return the first character after the value in *endptr.
2307  *
2308  * Complain if the following character is not what is expected,
2309  * either a space or end of the string.
2310  */
2311 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2312 {
2313         uintmax_t mark;
2314
2315         assert(*p == ':');
2316         p++;
2317         mark = strtoumax(p, endptr, 10);
2318         if (*endptr == p)
2319                 die("No value after ':' in mark: %s", command_buf.buf);
2320         return mark;
2321 }
2322
2323 /*
2324  * Parse the mark reference, and complain if this is not the end of
2325  * the string.
2326  */
2327 static uintmax_t parse_mark_ref_eol(const char *p)
2328 {
2329         char *end;
2330         uintmax_t mark;
2331
2332         mark = parse_mark_ref(p, &end);
2333         if (*end != '\0')
2334                 die("Garbage after mark: %s", command_buf.buf);
2335         return mark;
2336 }
2337
2338 /*
2339  * Parse the mark reference, demanding a trailing space.  Return a
2340  * pointer to the space.
2341  */
2342 static uintmax_t parse_mark_ref_space(const char **p)
2343 {
2344         uintmax_t mark;
2345         char *end;
2346
2347         mark = parse_mark_ref(*p, &end);
2348         if (*end++ != ' ')
2349                 die("Missing space after mark: %s", command_buf.buf);
2350         *p = end;
2351         return mark;
2352 }
2353
2354 static void file_change_m(const char *p, struct branch *b)
2355 {
2356         static struct strbuf uq = STRBUF_INIT;
2357         const char *endp;
2358         struct object_entry *oe;
2359         unsigned char sha1[20];
2360         uint16_t mode, inline_data = 0;
2361
2362         p = get_mode(p, &mode);
2363         if (!p)
2364                 die("Corrupt mode: %s", command_buf.buf);
2365         switch (mode) {
2366         case 0644:
2367         case 0755:
2368                 mode |= S_IFREG;
2369         case S_IFREG | 0644:
2370         case S_IFREG | 0755:
2371         case S_IFLNK:
2372         case S_IFDIR:
2373         case S_IFGITLINK:
2374                 /* ok */
2375                 break;
2376         default:
2377                 die("Corrupt mode: %s", command_buf.buf);
2378         }
2379
2380         if (*p == ':') {
2381                 oe = find_mark(parse_mark_ref_space(&p));
2382                 hashcpy(sha1, oe->idx.sha1);
2383         } else if (skip_prefix(p, "inline ", &p)) {
2384                 inline_data = 1;
2385                 oe = NULL; /* not used with inline_data, but makes gcc happy */
2386         } else {
2387                 if (get_sha1_hex(p, sha1))
2388                         die("Invalid dataref: %s", command_buf.buf);
2389                 oe = find_object(sha1);
2390                 p += 40;
2391                 if (*p++ != ' ')
2392                         die("Missing space after SHA1: %s", command_buf.buf);
2393         }
2394
2395         strbuf_reset(&uq);
2396         if (!unquote_c_style(&uq, p, &endp)) {
2397                 if (*endp)
2398                         die("Garbage after path in: %s", command_buf.buf);
2399                 p = uq.buf;
2400         }
2401
2402         /* Git does not track empty, non-toplevel directories. */
2403         if (S_ISDIR(mode) && !hashcmp(sha1, EMPTY_TREE_SHA1_BIN) && *p) {
2404                 tree_content_remove(&b->branch_tree, p, NULL, 0);
2405                 return;
2406         }
2407
2408         if (S_ISGITLINK(mode)) {
2409                 if (inline_data)
2410                         die("Git links cannot be specified 'inline': %s",
2411                                 command_buf.buf);
2412                 else if (oe) {
2413                         if (oe->type != OBJ_COMMIT)
2414                                 die("Not a commit (actually a %s): %s",
2415                                         typename(oe->type), command_buf.buf);
2416                 }
2417                 /*
2418                  * Accept the sha1 without checking; it expected to be in
2419                  * another repository.
2420                  */
2421         } else if (inline_data) {
2422                 if (S_ISDIR(mode))
2423                         die("Directories cannot be specified 'inline': %s",
2424                                 command_buf.buf);
2425                 if (p != uq.buf) {
2426                         strbuf_addstr(&uq, p);
2427                         p = uq.buf;
2428                 }
2429                 read_next_command();
2430                 parse_and_store_blob(&last_blob, sha1, 0);
2431         } else {
2432                 enum object_type expected = S_ISDIR(mode) ?
2433                                                 OBJ_TREE: OBJ_BLOB;
2434                 enum object_type type = oe ? oe->type :
2435                                         sha1_object_info(sha1, NULL);
2436                 if (type < 0)
2437                         die("%s not found: %s",
2438                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2439                                         command_buf.buf);
2440                 if (type != expected)
2441                         die("Not a %s (actually a %s): %s",
2442                                 typename(expected), typename(type),
2443                                 command_buf.buf);
2444         }
2445
2446         if (!*p) {
2447                 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2448                 return;
2449         }
2450         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2451 }
2452
2453 static void file_change_d(const char *p, struct branch *b)
2454 {
2455         static struct strbuf uq = STRBUF_INIT;
2456         const char *endp;
2457
2458         strbuf_reset(&uq);
2459         if (!unquote_c_style(&uq, p, &endp)) {
2460                 if (*endp)
2461                         die("Garbage after path in: %s", command_buf.buf);
2462                 p = uq.buf;
2463         }
2464         tree_content_remove(&b->branch_tree, p, NULL, 1);
2465 }
2466
2467 static void file_change_cr(const char *s, struct branch *b, int rename)
2468 {
2469         const char *d;
2470         static struct strbuf s_uq = STRBUF_INIT;
2471         static struct strbuf d_uq = STRBUF_INIT;
2472         const char *endp;
2473         struct tree_entry leaf;
2474
2475         strbuf_reset(&s_uq);
2476         if (!unquote_c_style(&s_uq, s, &endp)) {
2477                 if (*endp != ' ')
2478                         die("Missing space after source: %s", command_buf.buf);
2479         } else {
2480                 endp = strchr(s, ' ');
2481                 if (!endp)
2482                         die("Missing space after source: %s", command_buf.buf);
2483                 strbuf_add(&s_uq, s, endp - s);
2484         }
2485         s = s_uq.buf;
2486
2487         endp++;
2488         if (!*endp)
2489                 die("Missing dest: %s", command_buf.buf);
2490
2491         d = endp;
2492         strbuf_reset(&d_uq);
2493         if (!unquote_c_style(&d_uq, d, &endp)) {
2494                 if (*endp)
2495                         die("Garbage after dest in: %s", command_buf.buf);
2496                 d = d_uq.buf;
2497         }
2498
2499         memset(&leaf, 0, sizeof(leaf));
2500         if (rename)
2501                 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2502         else
2503                 tree_content_get(&b->branch_tree, s, &leaf, 1);
2504         if (!leaf.versions[1].mode)
2505                 die("Path %s not in branch", s);
2506         if (!*d) {      /* C "path/to/subdir" "" */
2507                 tree_content_replace(&b->branch_tree,
2508                         leaf.versions[1].sha1,
2509                         leaf.versions[1].mode,
2510                         leaf.tree);
2511                 return;
2512         }
2513         tree_content_set(&b->branch_tree, d,
2514                 leaf.versions[1].sha1,
2515                 leaf.versions[1].mode,
2516                 leaf.tree);
2517 }
2518
2519 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2520 {
2521         static struct strbuf uq = STRBUF_INIT;
2522         struct object_entry *oe;
2523         struct branch *s;
2524         unsigned char sha1[20], commit_sha1[20];
2525         char path[60];
2526         uint16_t inline_data = 0;
2527         unsigned char new_fanout;
2528
2529         /*
2530          * When loading a branch, we don't traverse its tree to count the real
2531          * number of notes (too expensive to do this for all non-note refs).
2532          * This means that recently loaded notes refs might incorrectly have
2533          * b->num_notes == 0, and consequently, old_fanout might be wrong.
2534          *
2535          * Fix this by traversing the tree and counting the number of notes
2536          * when b->num_notes == 0. If the notes tree is truly empty, the
2537          * calculation should not take long.
2538          */
2539         if (b->num_notes == 0 && *old_fanout == 0) {
2540                 /* Invoke change_note_fanout() in "counting mode". */
2541                 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2542                 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2543         }
2544
2545         /* Now parse the notemodify command. */
2546         /* <dataref> or 'inline' */
2547         if (*p == ':') {
2548                 oe = find_mark(parse_mark_ref_space(&p));
2549                 hashcpy(sha1, oe->idx.sha1);
2550         } else if (skip_prefix(p, "inline ", &p)) {
2551                 inline_data = 1;
2552                 oe = NULL; /* not used with inline_data, but makes gcc happy */
2553         } else {
2554                 if (get_sha1_hex(p, sha1))
2555                         die("Invalid dataref: %s", command_buf.buf);
2556                 oe = find_object(sha1);
2557                 p += 40;
2558                 if (*p++ != ' ')
2559                         die("Missing space after SHA1: %s", command_buf.buf);
2560         }
2561
2562         /* <commit-ish> */
2563         s = lookup_branch(p);
2564         if (s) {
2565                 if (is_null_sha1(s->sha1))
2566                         die("Can't add a note on empty branch.");
2567                 hashcpy(commit_sha1, s->sha1);
2568         } else if (*p == ':') {
2569                 uintmax_t commit_mark = parse_mark_ref_eol(p);
2570                 struct object_entry *commit_oe = find_mark(commit_mark);
2571                 if (commit_oe->type != OBJ_COMMIT)
2572                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2573                 hashcpy(commit_sha1, commit_oe->idx.sha1);
2574         } else if (!get_sha1(p, commit_sha1)) {
2575                 unsigned long size;
2576                 char *buf = read_object_with_reference(commit_sha1,
2577                         commit_type, &size, commit_sha1);
2578                 if (!buf || size < 46)
2579                         die("Not a valid commit: %s", p);
2580                 free(buf);
2581         } else
2582                 die("Invalid ref name or SHA1 expression: %s", p);
2583
2584         if (inline_data) {
2585                 if (p != uq.buf) {
2586                         strbuf_addstr(&uq, p);
2587                         p = uq.buf;
2588                 }
2589                 read_next_command();
2590                 parse_and_store_blob(&last_blob, sha1, 0);
2591         } else if (oe) {
2592                 if (oe->type != OBJ_BLOB)
2593                         die("Not a blob (actually a %s): %s",
2594                                 typename(oe->type), command_buf.buf);
2595         } else if (!is_null_sha1(sha1)) {
2596                 enum object_type type = sha1_object_info(sha1, NULL);
2597                 if (type < 0)
2598                         die("Blob not found: %s", command_buf.buf);
2599                 if (type != OBJ_BLOB)
2600                         die("Not a blob (actually a %s): %s",
2601                             typename(type), command_buf.buf);
2602         }
2603
2604         construct_path_with_fanout(sha1_to_hex(commit_sha1), *old_fanout, path);
2605         if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2606                 b->num_notes--;
2607
2608         if (is_null_sha1(sha1))
2609                 return; /* nothing to insert */
2610
2611         b->num_notes++;
2612         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2613         construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2614         tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2615 }
2616
2617 static void file_change_deleteall(struct branch *b)
2618 {
2619         release_tree_content_recursive(b->branch_tree.tree);
2620         hashclr(b->branch_tree.versions[0].sha1);
2621         hashclr(b->branch_tree.versions[1].sha1);
2622         load_tree(&b->branch_tree);
2623         b->num_notes = 0;
2624 }
2625
2626 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2627 {
2628         if (!buf || size < 46)
2629                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2630         if (memcmp("tree ", buf, 5)
2631                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2632                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2633         hashcpy(b->branch_tree.versions[0].sha1,
2634                 b->branch_tree.versions[1].sha1);
2635 }
2636
2637 static void parse_from_existing(struct branch *b)
2638 {
2639         if (is_null_sha1(b->sha1)) {
2640                 hashclr(b->branch_tree.versions[0].sha1);
2641                 hashclr(b->branch_tree.versions[1].sha1);
2642         } else {
2643                 unsigned long size;
2644                 char *buf;
2645
2646                 buf = read_object_with_reference(b->sha1,
2647                         commit_type, &size, b->sha1);
2648                 parse_from_commit(b, buf, size);
2649                 free(buf);
2650         }
2651 }
2652
2653 static int parse_from(struct branch *b)
2654 {
2655         const char *from;
2656         struct branch *s;
2657         unsigned char sha1[20];
2658
2659         if (!skip_prefix(command_buf.buf, "from ", &from))
2660                 return 0;
2661
2662         hashcpy(sha1, b->branch_tree.versions[1].sha1);
2663
2664         s = lookup_branch(from);
2665         if (b == s)
2666                 die("Can't create a branch from itself: %s", b->name);
2667         else if (s) {
2668                 unsigned char *t = s->branch_tree.versions[1].sha1;
2669                 hashcpy(b->sha1, s->sha1);
2670                 hashcpy(b->branch_tree.versions[0].sha1, t);
2671                 hashcpy(b->branch_tree.versions[1].sha1, t);
2672         } else if (*from == ':') {
2673                 uintmax_t idnum = parse_mark_ref_eol(from);
2674                 struct object_entry *oe = find_mark(idnum);
2675                 if (oe->type != OBJ_COMMIT)
2676                         die("Mark :%" PRIuMAX " not a commit", idnum);
2677                 if (hashcmp(b->sha1, oe->idx.sha1)) {
2678                         hashcpy(b->sha1, oe->idx.sha1);
2679                         if (oe->pack_id != MAX_PACK_ID) {
2680                                 unsigned long size;
2681                                 char *buf = gfi_unpack_entry(oe, &size);
2682                                 parse_from_commit(b, buf, size);
2683                                 free(buf);
2684                         } else
2685                                 parse_from_existing(b);
2686                 }
2687         } else if (!get_sha1(from, b->sha1)) {
2688                 parse_from_existing(b);
2689                 if (is_null_sha1(b->sha1))
2690                         b->delete = 1;
2691         }
2692         else
2693                 die("Invalid ref name or SHA1 expression: %s", from);
2694
2695         if (b->branch_tree.tree && hashcmp(sha1, b->branch_tree.versions[1].sha1)) {
2696                 release_tree_content_recursive(b->branch_tree.tree);
2697                 b->branch_tree.tree = NULL;
2698         }
2699
2700         read_next_command();
2701         return 1;
2702 }
2703
2704 static struct hash_list *parse_merge(unsigned int *count)
2705 {
2706         struct hash_list *list = NULL, **tail = &list, *n;
2707         const char *from;
2708         struct branch *s;
2709
2710         *count = 0;
2711         while (skip_prefix(command_buf.buf, "merge ", &from)) {
2712                 n = xmalloc(sizeof(*n));
2713                 s = lookup_branch(from);
2714                 if (s)
2715                         hashcpy(n->sha1, s->sha1);
2716                 else if (*from == ':') {
2717                         uintmax_t idnum = parse_mark_ref_eol(from);
2718                         struct object_entry *oe = find_mark(idnum);
2719                         if (oe->type != OBJ_COMMIT)
2720                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2721                         hashcpy(n->sha1, oe->idx.sha1);
2722                 } else if (!get_sha1(from, n->sha1)) {
2723                         unsigned long size;
2724                         char *buf = read_object_with_reference(n->sha1,
2725                                 commit_type, &size, n->sha1);
2726                         if (!buf || size < 46)
2727                                 die("Not a valid commit: %s", from);
2728                         free(buf);
2729                 } else
2730                         die("Invalid ref name or SHA1 expression: %s", from);
2731
2732                 n->next = NULL;
2733                 *tail = n;
2734                 tail = &n->next;
2735
2736                 (*count)++;
2737                 read_next_command();
2738         }
2739         return list;
2740 }
2741
2742 static void parse_new_commit(const char *arg)
2743 {
2744         static struct strbuf msg = STRBUF_INIT;
2745         struct branch *b;
2746         char *author = NULL;
2747         char *committer = NULL;
2748         struct hash_list *merge_list = NULL;
2749         unsigned int merge_count;
2750         unsigned char prev_fanout, new_fanout;
2751         const char *v;
2752
2753         b = lookup_branch(arg);
2754         if (!b)
2755                 b = new_branch(arg);
2756
2757         read_next_command();
2758         parse_mark();
2759         if (skip_prefix(command_buf.buf, "author ", &v)) {
2760                 author = parse_ident(v);
2761                 read_next_command();
2762         }
2763         if (skip_prefix(command_buf.buf, "committer ", &v)) {
2764                 committer = parse_ident(v);
2765                 read_next_command();
2766         }
2767         if (!committer)
2768                 die("Expected committer but didn't get one");
2769         parse_data(&msg, 0, NULL);
2770         read_next_command();
2771         parse_from(b);
2772         merge_list = parse_merge(&merge_count);
2773
2774         /* ensure the branch is active/loaded */
2775         if (!b->branch_tree.tree || !max_active_branches) {
2776                 unload_one_branch();
2777                 load_branch(b);
2778         }
2779
2780         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2781
2782         /* file_change* */
2783         while (command_buf.len > 0) {
2784                 if (skip_prefix(command_buf.buf, "M ", &v))
2785                         file_change_m(v, b);
2786                 else if (skip_prefix(command_buf.buf, "D ", &v))
2787                         file_change_d(v, b);
2788                 else if (skip_prefix(command_buf.buf, "R ", &v))
2789                         file_change_cr(v, b, 1);
2790                 else if (skip_prefix(command_buf.buf, "C ", &v))
2791                         file_change_cr(v, b, 0);
2792                 else if (skip_prefix(command_buf.buf, "N ", &v))
2793                         note_change_n(v, b, &prev_fanout);
2794                 else if (!strcmp("deleteall", command_buf.buf))
2795                         file_change_deleteall(b);
2796                 else if (skip_prefix(command_buf.buf, "ls ", &v))
2797                         parse_ls(v, b);
2798                 else {
2799                         unread_command_buf = 1;
2800                         break;
2801                 }
2802                 if (read_next_command() == EOF)
2803                         break;
2804         }
2805
2806         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2807         if (new_fanout != prev_fanout)
2808                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2809
2810         /* build the tree and the commit */
2811         store_tree(&b->branch_tree);
2812         hashcpy(b->branch_tree.versions[0].sha1,
2813                 b->branch_tree.versions[1].sha1);
2814
2815         strbuf_reset(&new_data);
2816         strbuf_addf(&new_data, "tree %s\n",
2817                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2818         if (!is_null_sha1(b->sha1))
2819                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2820         while (merge_list) {
2821                 struct hash_list *next = merge_list->next;
2822                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2823                 free(merge_list);
2824                 merge_list = next;
2825         }
2826         strbuf_addf(&new_data,
2827                 "author %s\n"
2828                 "committer %s\n"
2829                 "\n",
2830                 author ? author : committer, committer);
2831         strbuf_addbuf(&new_data, &msg);
2832         free(author);
2833         free(committer);
2834
2835         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2836                 b->pack_id = pack_id;
2837         b->last_commit = object_count_by_type[OBJ_COMMIT];
2838 }
2839
2840 static void parse_new_tag(const char *arg)
2841 {
2842         static struct strbuf msg = STRBUF_INIT;
2843         const char *from;
2844         char *tagger;
2845         struct branch *s;
2846         struct tag *t;
2847         uintmax_t from_mark = 0;
2848         unsigned char sha1[20];
2849         enum object_type type;
2850         const char *v;
2851
2852         t = pool_alloc(sizeof(struct tag));
2853         memset(t, 0, sizeof(struct tag));
2854         t->name = pool_strdup(arg);
2855         if (last_tag)
2856                 last_tag->next_tag = t;
2857         else
2858                 first_tag = t;
2859         last_tag = t;
2860         read_next_command();
2861
2862         /* from ... */
2863         if (!skip_prefix(command_buf.buf, "from ", &from))
2864                 die("Expected from command, got %s", command_buf.buf);
2865         s = lookup_branch(from);
2866         if (s) {
2867                 if (is_null_sha1(s->sha1))
2868                         die("Can't tag an empty branch.");
2869                 hashcpy(sha1, s->sha1);
2870                 type = OBJ_COMMIT;
2871         } else if (*from == ':') {
2872                 struct object_entry *oe;
2873                 from_mark = parse_mark_ref_eol(from);
2874                 oe = find_mark(from_mark);
2875                 type = oe->type;
2876                 hashcpy(sha1, oe->idx.sha1);
2877         } else if (!get_sha1(from, sha1)) {
2878                 struct object_entry *oe = find_object(sha1);
2879                 if (!oe) {
2880                         type = sha1_object_info(sha1, NULL);
2881                         if (type < 0)
2882                                 die("Not a valid object: %s", from);
2883                 } else
2884                         type = oe->type;
2885         } else
2886                 die("Invalid ref name or SHA1 expression: %s", from);
2887         read_next_command();
2888
2889         /* tagger ... */
2890         if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2891                 tagger = parse_ident(v);
2892                 read_next_command();
2893         } else
2894                 tagger = NULL;
2895
2896         /* tag payload/message */
2897         parse_data(&msg, 0, NULL);
2898
2899         /* build the tag object */
2900         strbuf_reset(&new_data);
2901
2902         strbuf_addf(&new_data,
2903                     "object %s\n"
2904                     "type %s\n"
2905                     "tag %s\n",
2906                     sha1_to_hex(sha1), typename(type), t->name);
2907         if (tagger)
2908                 strbuf_addf(&new_data,
2909                             "tagger %s\n", tagger);
2910         strbuf_addch(&new_data, '\n');
2911         strbuf_addbuf(&new_data, &msg);
2912         free(tagger);
2913
2914         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2915                 t->pack_id = MAX_PACK_ID;
2916         else
2917                 t->pack_id = pack_id;
2918 }
2919
2920 static void parse_reset_branch(const char *arg)
2921 {
2922         struct branch *b;
2923
2924         b = lookup_branch(arg);
2925         if (b) {
2926                 hashclr(b->sha1);
2927                 hashclr(b->branch_tree.versions[0].sha1);
2928                 hashclr(b->branch_tree.versions[1].sha1);
2929                 if (b->branch_tree.tree) {
2930                         release_tree_content_recursive(b->branch_tree.tree);
2931                         b->branch_tree.tree = NULL;
2932                 }
2933         }
2934         else
2935                 b = new_branch(arg);
2936         read_next_command();
2937         parse_from(b);
2938         if (command_buf.len > 0)
2939                 unread_command_buf = 1;
2940 }
2941
2942 static void cat_blob_write(const char *buf, unsigned long size)
2943 {
2944         if (write_in_full(cat_blob_fd, buf, size) != size)
2945                 die_errno("Write to frontend failed");
2946 }
2947
2948 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2949 {
2950         struct strbuf line = STRBUF_INIT;
2951         unsigned long size;
2952         enum object_type type = 0;
2953         char *buf;
2954
2955         if (!oe || oe->pack_id == MAX_PACK_ID) {
2956                 buf = read_sha1_file(sha1, &type, &size);
2957         } else {
2958                 type = oe->type;
2959                 buf = gfi_unpack_entry(oe, &size);
2960         }
2961
2962         /*
2963          * Output based on batch_one_object() from cat-file.c.
2964          */
2965         if (type <= 0) {
2966                 strbuf_reset(&line);
2967                 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2968                 cat_blob_write(line.buf, line.len);
2969                 strbuf_release(&line);
2970                 free(buf);
2971                 return;
2972         }
2973         if (!buf)
2974                 die("Can't read object %s", sha1_to_hex(sha1));
2975         if (type != OBJ_BLOB)
2976                 die("Object %s is a %s but a blob was expected.",
2977                     sha1_to_hex(sha1), typename(type));
2978         strbuf_reset(&line);
2979         strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2980                                                 typename(type), size);
2981         cat_blob_write(line.buf, line.len);
2982         strbuf_release(&line);
2983         cat_blob_write(buf, size);
2984         cat_blob_write("\n", 1);
2985         if (oe && oe->pack_id == pack_id) {
2986                 last_blob.offset = oe->idx.offset;
2987                 strbuf_attach(&last_blob.data, buf, size, size);
2988                 last_blob.depth = oe->depth;
2989         } else
2990                 free(buf);
2991 }
2992
2993 static void parse_get_mark(const char *p)
2994 {
2995         struct object_entry *oe = oe;
2996         char output[42];
2997
2998         /* get-mark SP <object> LF */
2999         if (*p != ':')
3000                 die("Not a mark: %s", p);
3001
3002         oe = find_mark(parse_mark_ref_eol(p));
3003         if (!oe)
3004                 die("Unknown mark: %s", command_buf.buf);
3005
3006         snprintf(output, sizeof(output), "%s\n", sha1_to_hex(oe->idx.sha1));
3007         cat_blob_write(output, 41);
3008 }
3009
3010 static void parse_cat_blob(const char *p)
3011 {
3012         struct object_entry *oe = oe;
3013         unsigned char sha1[20];
3014
3015         /* cat-blob SP <object> LF */
3016         if (*p == ':') {
3017                 oe = find_mark(parse_mark_ref_eol(p));
3018                 if (!oe)
3019                         die("Unknown mark: %s", command_buf.buf);
3020                 hashcpy(sha1, oe->idx.sha1);
3021         } else {
3022                 if (get_sha1_hex(p, sha1))
3023                         die("Invalid dataref: %s", command_buf.buf);
3024                 if (p[40])
3025                         die("Garbage after SHA1: %s", command_buf.buf);
3026                 oe = find_object(sha1);
3027         }
3028
3029         cat_blob(oe, sha1);
3030 }
3031
3032 static struct object_entry *dereference(struct object_entry *oe,
3033                                         unsigned char sha1[20])
3034 {
3035         unsigned long size;
3036         char *buf = NULL;
3037         if (!oe) {
3038                 enum object_type type = sha1_object_info(sha1, NULL);
3039                 if (type < 0)
3040                         die("object not found: %s", sha1_to_hex(sha1));
3041                 /* cache it! */
3042                 oe = insert_object(sha1);
3043                 oe->type = type;
3044                 oe->pack_id = MAX_PACK_ID;
3045                 oe->idx.offset = 1;
3046         }
3047         switch (oe->type) {
3048         case OBJ_TREE:  /* easy case. */
3049                 return oe;
3050         case OBJ_COMMIT:
3051         case OBJ_TAG:
3052                 break;
3053         default:
3054                 die("Not a tree-ish: %s", command_buf.buf);
3055         }
3056
3057         if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
3058                 buf = gfi_unpack_entry(oe, &size);
3059         } else {
3060                 enum object_type unused;
3061                 buf = read_sha1_file(sha1, &unused, &size);
3062         }
3063         if (!buf)
3064                 die("Can't load object %s", sha1_to_hex(sha1));
3065
3066         /* Peel one layer. */
3067         switch (oe->type) {
3068         case OBJ_TAG:
3069                 if (size < 40 + strlen("object ") ||
3070                     get_sha1_hex(buf + strlen("object "), sha1))
3071                         die("Invalid SHA1 in tag: %s", command_buf.buf);
3072                 break;
3073         case OBJ_COMMIT:
3074                 if (size < 40 + strlen("tree ") ||
3075                     get_sha1_hex(buf + strlen("tree "), sha1))
3076                         die("Invalid SHA1 in commit: %s", command_buf.buf);
3077         }
3078
3079         free(buf);
3080         return find_object(sha1);
3081 }
3082
3083 static struct object_entry *parse_treeish_dataref(const char **p)
3084 {
3085         unsigned char sha1[20];
3086         struct object_entry *e;
3087
3088         if (**p == ':') {       /* <mark> */
3089                 e = find_mark(parse_mark_ref_space(p));
3090                 if (!e)
3091                         die("Unknown mark: %s", command_buf.buf);
3092                 hashcpy(sha1, e->idx.sha1);
3093         } else {        /* <sha1> */
3094                 if (get_sha1_hex(*p, sha1))
3095                         die("Invalid dataref: %s", command_buf.buf);
3096                 e = find_object(sha1);
3097                 *p += 40;
3098                 if (*(*p)++ != ' ')
3099                         die("Missing space after tree-ish: %s", command_buf.buf);
3100         }
3101
3102         while (!e || e->type != OBJ_TREE)
3103                 e = dereference(e, sha1);
3104         return e;
3105 }
3106
3107 static void print_ls(int mode, const unsigned char *sha1, const char *path)
3108 {
3109         static struct strbuf line = STRBUF_INIT;
3110
3111         /* See show_tree(). */
3112         const char *type =
3113                 S_ISGITLINK(mode) ? commit_type :
3114                 S_ISDIR(mode) ? tree_type :
3115                 blob_type;
3116
3117         if (!mode) {
3118                 /* missing SP path LF */
3119                 strbuf_reset(&line);
3120                 strbuf_addstr(&line, "missing ");
3121                 quote_c_style(path, &line, NULL, 0);
3122                 strbuf_addch(&line, '\n');
3123         } else {
3124                 /* mode SP type SP object_name TAB path LF */
3125                 strbuf_reset(&line);
3126                 strbuf_addf(&line, "%06o %s %s\t",
3127                                 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3128                 quote_c_style(path, &line, NULL, 0);
3129                 strbuf_addch(&line, '\n');
3130         }
3131         cat_blob_write(line.buf, line.len);
3132 }
3133
3134 static void parse_ls(const char *p, struct branch *b)
3135 {
3136         struct tree_entry *root = NULL;
3137         struct tree_entry leaf = {NULL};
3138
3139         /* ls SP (<tree-ish> SP)? <path> */
3140         if (*p == '"') {
3141                 if (!b)
3142                         die("Not in a commit: %s", command_buf.buf);
3143                 root = &b->branch_tree;
3144         } else {
3145                 struct object_entry *e = parse_treeish_dataref(&p);
3146                 root = new_tree_entry();
3147                 hashcpy(root->versions[1].sha1, e->idx.sha1);
3148                 if (!is_null_sha1(root->versions[1].sha1))
3149                         root->versions[1].mode = S_IFDIR;
3150                 load_tree(root);
3151         }
3152         if (*p == '"') {
3153                 static struct strbuf uq = STRBUF_INIT;
3154                 const char *endp;
3155                 strbuf_reset(&uq);
3156                 if (unquote_c_style(&uq, p, &endp))
3157                         die("Invalid path: %s", command_buf.buf);
3158                 if (*endp)
3159                         die("Garbage after path in: %s", command_buf.buf);
3160                 p = uq.buf;
3161         }
3162         tree_content_get(root, p, &leaf, 1);
3163         /*
3164          * A directory in preparation would have a sha1 of zero
3165          * until it is saved.  Save, for simplicity.
3166          */
3167         if (S_ISDIR(leaf.versions[1].mode))
3168                 store_tree(&leaf);
3169
3170         print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3171         if (leaf.tree)
3172                 release_tree_content_recursive(leaf.tree);
3173         if (!b || root != &b->branch_tree)
3174                 release_tree_entry(root);
3175 }
3176
3177 static void checkpoint(void)
3178 {
3179         checkpoint_requested = 0;
3180         if (object_count) {
3181                 cycle_packfile();
3182                 dump_branches();
3183                 dump_tags();
3184                 dump_marks();
3185         }
3186 }
3187
3188 static void parse_checkpoint(void)
3189 {
3190         checkpoint_requested = 1;
3191         skip_optional_lf();
3192 }
3193
3194 static void parse_progress(void)
3195 {
3196         fwrite(command_buf.buf, 1, command_buf.len, stdout);
3197         fputc('\n', stdout);
3198         fflush(stdout);
3199         skip_optional_lf();
3200 }
3201
3202 static char* make_fast_import_path(const char *path)
3203 {
3204         if (!relative_marks_paths || is_absolute_path(path))
3205                 return xstrdup(path);
3206         return xstrdup(git_path("info/fast-import/%s", path));
3207 }
3208
3209 static void option_import_marks(const char *marks,
3210                                         int from_stream, int ignore_missing)
3211 {
3212         if (import_marks_file) {
3213                 if (from_stream)
3214                         die("Only one import-marks command allowed per stream");
3215
3216                 /* read previous mark file */
3217                 if(!import_marks_file_from_stream)
3218                         read_marks();
3219         }
3220
3221         import_marks_file = make_fast_import_path(marks);
3222         safe_create_leading_directories_const(import_marks_file);
3223         import_marks_file_from_stream = from_stream;
3224         import_marks_file_ignore_missing = ignore_missing;
3225 }
3226
3227 static void option_date_format(const char *fmt)
3228 {
3229         if (!strcmp(fmt, "raw"))
3230                 whenspec = WHENSPEC_RAW;
3231         else if (!strcmp(fmt, "rfc2822"))
3232                 whenspec = WHENSPEC_RFC2822;
3233         else if (!strcmp(fmt, "now"))
3234                 whenspec = WHENSPEC_NOW;
3235         else
3236                 die("unknown --date-format argument %s", fmt);
3237 }
3238
3239 static unsigned long ulong_arg(const char *option, const char *arg)
3240 {
3241         char *endptr;
3242         unsigned long rv = strtoul(arg, &endptr, 0);
3243         if (strchr(arg, '-') || endptr == arg || *endptr)
3244                 die("%s: argument must be a non-negative integer", option);
3245         return rv;
3246 }
3247
3248 static void option_depth(const char *depth)
3249 {
3250         max_depth = ulong_arg("--depth", depth);
3251         if (max_depth > MAX_DEPTH)
3252                 die("--depth cannot exceed %u", MAX_DEPTH);
3253 }
3254
3255 static void option_active_branches(const char *branches)
3256 {
3257         max_active_branches = ulong_arg("--active-branches", branches);
3258 }
3259
3260 static void option_export_marks(const char *marks)
3261 {
3262         export_marks_file = make_fast_import_path(marks);
3263         safe_create_leading_directories_const(export_marks_file);
3264 }
3265
3266 static void option_cat_blob_fd(const char *fd)
3267 {
3268         unsigned long n = ulong_arg("--cat-blob-fd", fd);
3269         if (n > (unsigned long) INT_MAX)
3270                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3271         cat_blob_fd = (int) n;
3272 }
3273
3274 static void option_export_pack_edges(const char *edges)
3275 {
3276         if (pack_edges)
3277                 fclose(pack_edges);
3278         pack_edges = fopen(edges, "a");
3279         if (!pack_edges)
3280                 die_errno("Cannot open '%s'", edges);
3281 }
3282
3283 static int parse_one_option(const char *option)
3284 {
3285         if (skip_prefix(option, "max-pack-size=", &option)) {
3286                 unsigned long v;
3287                 if (!git_parse_ulong(option, &v))
3288                         return 0;
3289                 if (v < 8192) {
3290                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3291                         v *= 1024 * 1024;
3292                 } else if (v < 1024 * 1024) {
3293                         warning("minimum max-pack-size is 1 MiB");
3294                         v = 1024 * 1024;
3295                 }
3296                 max_packsize = v;
3297         } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3298                 unsigned long v;
3299                 if (!git_parse_ulong(option, &v))
3300                         return 0;
3301                 big_file_threshold = v;
3302         } else if (skip_prefix(option, "depth=", &option)) {
3303                 option_depth(option);
3304         } else if (skip_prefix(option, "active-branches=", &option)) {
3305                 option_active_branches(option);
3306         } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3307                 option_export_pack_edges(option);
3308         } else if (starts_with(option, "quiet")) {
3309                 show_stats = 0;
3310         } else if (starts_with(option, "stats")) {
3311                 show_stats = 1;
3312         } else {
3313                 return 0;
3314         }
3315
3316         return 1;
3317 }
3318
3319 static int parse_one_feature(const char *feature, int from_stream)
3320 {
3321         const char *arg;
3322
3323         if (skip_prefix(feature, "date-format=", &arg)) {
3324                 option_date_format(arg);
3325         } else if (skip_prefix(feature, "import-marks=", &arg)) {
3326                 option_import_marks(arg, from_stream, 0);
3327         } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3328                 option_import_marks(arg, from_stream, 1);
3329         } else if (skip_prefix(feature, "export-marks=", &arg)) {
3330                 option_export_marks(arg);
3331         } else if (!strcmp(feature, "get-mark")) {
3332                 ; /* Don't die - this feature is supported */
3333         } else if (!strcmp(feature, "cat-blob")) {
3334                 ; /* Don't die - this feature is supported */
3335         } else if (!strcmp(feature, "relative-marks")) {
3336                 relative_marks_paths = 1;
3337         } else if (!strcmp(feature, "no-relative-marks")) {
3338                 relative_marks_paths = 0;
3339         } else if (!strcmp(feature, "done")) {
3340                 require_explicit_termination = 1;
3341         } else if (!strcmp(feature, "force")) {
3342                 force_update = 1;
3343         } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3344                 ; /* do nothing; we have the feature */
3345         } else {
3346                 return 0;
3347         }
3348
3349         return 1;
3350 }
3351
3352 static void parse_feature(const char *feature)
3353 {
3354         if (seen_data_command)
3355                 die("Got feature command '%s' after data command", feature);
3356
3357         if (parse_one_feature(feature, 1))
3358                 return;
3359
3360         die("This version of fast-import does not support feature %s.", feature);
3361 }
3362
3363 static void parse_option(const char *option)
3364 {
3365         if (seen_data_command)
3366                 die("Got option command '%s' after data command", option);
3367
3368         if (parse_one_option(option))
3369                 return;
3370
3371         die("This version of fast-import does not support option: %s", option);
3372 }
3373
3374 static void git_pack_config(void)
3375 {
3376         int indexversion_value;
3377         int limit;
3378         unsigned long packsizelimit_value;
3379
3380         if (!git_config_get_ulong("pack.depth", &max_depth)) {
3381                 if (max_depth > MAX_DEPTH)
3382                         max_depth = MAX_DEPTH;
3383         }
3384         if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3385                 pack_idx_opts.version = indexversion_value;
3386                 if (pack_idx_opts.version > 2)
3387                         git_die_config("pack.indexversion",
3388                                         "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3389         }
3390         if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3391                 max_packsize = packsizelimit_value;
3392
3393         if (!git_config_get_int("fastimport.unpacklimit", &limit))
3394                 unpack_limit = limit;
3395         else if (!git_config_get_int("transfer.unpacklimit", &limit))
3396                 unpack_limit = limit;
3397
3398         git_config(git_default_config, NULL);
3399 }
3400
3401 static const char fast_import_usage[] =
3402 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3403
3404 static void parse_argv(void)
3405 {
3406         unsigned int i;
3407
3408         for (i = 1; i < global_argc; i++) {
3409                 const char *a = global_argv[i];
3410
3411                 if (*a != '-' || !strcmp(a, "--"))
3412                         break;
3413
3414                 if (!skip_prefix(a, "--", &a))
3415                         die("unknown option %s", a);
3416
3417                 if (parse_one_option(a))
3418                         continue;
3419
3420                 if (parse_one_feature(a, 0))
3421                         continue;
3422
3423                 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3424                         option_cat_blob_fd(a);
3425                         continue;
3426                 }
3427
3428                 die("unknown option --%s", a);
3429         }
3430         if (i != global_argc)
3431                 usage(fast_import_usage);
3432
3433         seen_data_command = 1;
3434         if (import_marks_file)
3435                 read_marks();
3436 }
3437
3438 int cmd_main(int argc, const char **argv)
3439 {
3440         unsigned int i;
3441
3442         if (argc == 2 && !strcmp(argv[1], "-h"))
3443                 usage(fast_import_usage);
3444
3445         setup_git_directory();
3446         reset_pack_idx_option(&pack_idx_opts);
3447         git_pack_config();
3448
3449         alloc_objects(object_entry_alloc);
3450         strbuf_init(&command_buf, 0);
3451         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3452         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3453         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3454         marks = pool_calloc(1, sizeof(struct mark_set));
3455
3456         global_argc = argc;
3457         global_argv = argv;
3458
3459         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3460         for (i = 0; i < (cmd_save - 1); i++)
3461                 rc_free[i].next = &rc_free[i + 1];
3462         rc_free[cmd_save - 1].next = NULL;
3463
3464         prepare_packed_git();
3465         start_packfile();
3466         set_die_routine(die_nicely);
3467         set_checkpoint_signal();
3468         while (read_next_command() != EOF) {
3469                 const char *v;
3470                 if (!strcmp("blob", command_buf.buf))
3471                         parse_new_blob();
3472                 else if (skip_prefix(command_buf.buf, "ls ", &v))
3473                         parse_ls(v, NULL);
3474                 else if (skip_prefix(command_buf.buf, "commit ", &v))
3475                         parse_new_commit(v);
3476                 else if (skip_prefix(command_buf.buf, "tag ", &v))
3477                         parse_new_tag(v);
3478                 else if (skip_prefix(command_buf.buf, "reset ", &v))
3479                         parse_reset_branch(v);
3480                 else if (!strcmp("checkpoint", command_buf.buf))
3481                         parse_checkpoint();
3482                 else if (!strcmp("done", command_buf.buf))
3483                         break;
3484                 else if (starts_with(command_buf.buf, "progress "))
3485                         parse_progress();
3486                 else if (skip_prefix(command_buf.buf, "feature ", &v))
3487                         parse_feature(v);
3488                 else if (skip_prefix(command_buf.buf, "option git ", &v))
3489                         parse_option(v);
3490                 else if (starts_with(command_buf.buf, "option "))
3491                         /* ignore non-git options*/;
3492                 else
3493                         die("Unsupported command: %s", command_buf.buf);
3494
3495                 if (checkpoint_requested)
3496                         checkpoint();
3497         }
3498
3499         /* argv hasn't been parsed yet, do so */
3500         if (!seen_data_command)
3501                 parse_argv();
3502
3503         if (require_explicit_termination && feof(stdin))
3504                 die("stream ends early");
3505
3506         end_packfile();
3507
3508         dump_branches();
3509         dump_tags();
3510         unkeep_all_packs();
3511         dump_marks();
3512
3513         if (pack_edges)
3514                 fclose(pack_edges);
3515
3516         if (show_stats) {
3517                 uintmax_t total_count = 0, duplicate_count = 0;
3518                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3519                         total_count += object_count_by_type[i];
3520                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3521                         duplicate_count += duplicate_count_by_type[i];
3522
3523                 fprintf(stderr, "%s statistics:\n", argv[0]);
3524                 fprintf(stderr, "---------------------------------------------------------------------\n");
3525                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3526                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3527                 fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB], delta_count_attempts_by_type[OBJ_BLOB]);
3528                 fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE], delta_count_attempts_by_type[OBJ_TREE]);
3529                 fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT], delta_count_attempts_by_type[OBJ_COMMIT]);
3530                 fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG], delta_count_attempts_by_type[OBJ_TAG]);
3531                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3532                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3533                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3534                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3535                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3536                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3537                 fprintf(stderr, "---------------------------------------------------------------------\n");
3538                 pack_report();
3539                 fprintf(stderr, "---------------------------------------------------------------------\n");
3540                 fprintf(stderr, "\n");
3541         }
3542
3543         return failure ? 1 : 0;
3544 }