Merge branch 'master' into next
[git] / fast-import.c
1 /*
2 (See Documentation/git-fast-import.txt for maintained documentation.)
3 Format of STDIN stream:
4
5   stream ::= cmd*;
6
7   cmd ::= new_blob
8         | new_commit
9         | new_tag
10         | reset_branch
11         | checkpoint
12         | progress
13         ;
14
15   new_blob ::= 'blob' lf
16     mark?
17     file_content;
18   file_content ::= data;
19
20   new_commit ::= 'commit' sp ref_str lf
21     mark?
22     ('author' (sp name)? sp '<' email '>' sp when lf)?
23     'committer' (sp name)? sp '<' email '>' sp when lf
24     commit_msg
25     ('from' sp committish lf)?
26     ('merge' sp committish lf)*
27     (file_change | ls)*
28     lf?;
29   commit_msg ::= data;
30
31   ls ::= 'ls' sp '"' quoted(path) '"' lf;
32
33   file_change ::= file_clr
34     | file_del
35     | file_rnm
36     | file_cpy
37     | file_obm
38     | file_inm;
39   file_clr ::= 'deleteall' lf;
40   file_del ::= 'D' sp path_str lf;
41   file_rnm ::= 'R' sp path_str sp path_str lf;
42   file_cpy ::= 'C' sp path_str sp path_str lf;
43   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
44   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
45     data;
46   note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
47   note_inm ::= 'N' sp 'inline' sp committish lf
48     data;
49
50   new_tag ::= 'tag' sp tag_str lf
51     'from' sp committish lf
52     ('tagger' (sp name)? sp '<' email '>' sp when lf)?
53     tag_msg;
54   tag_msg ::= data;
55
56   reset_branch ::= 'reset' sp ref_str lf
57     ('from' sp committish lf)?
58     lf?;
59
60   checkpoint ::= 'checkpoint' lf
61     lf?;
62
63   progress ::= 'progress' sp not_lf* lf
64     lf?;
65
66      # note: the first idnum in a stream should be 1 and subsequent
67      # idnums should not have gaps between values as this will cause
68      # the stream parser to reserve space for the gapped values.  An
69      # idnum can be updated in the future to a new object by issuing
70      # a new mark directive with the old idnum.
71      #
72   mark ::= 'mark' sp idnum lf;
73   data ::= (delimited_data | exact_data)
74     lf?;
75
76     # note: delim may be any string but must not contain lf.
77     # data_line may contain any data but must not be exactly
78     # delim.
79   delimited_data ::= 'data' sp '<<' delim lf
80     (data_line lf)*
81     delim lf;
82
83      # note: declen indicates the length of binary_data in bytes.
84      # declen does not include the lf preceding the binary data.
85      #
86   exact_data ::= 'data' sp declen lf
87     binary_data;
88
89      # note: quoted strings are C-style quoting supporting \c for
90      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
91      # is the signed byte value in octal.  Note that the only
92      # characters which must actually be escaped to protect the
93      # stream formatting is: \, " and LF.  Otherwise these values
94      # are UTF8.
95      #
96   committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
97   ref_str     ::= ref;
98   sha1exp_str ::= sha1exp;
99   tag_str     ::= tag;
100   path_str    ::= path    | '"' quoted(path)    '"' ;
101   mode        ::= '100644' | '644'
102                 | '100755' | '755'
103                 | '120000'
104                 ;
105
106   declen ::= # unsigned 32 bit value, ascii base10 notation;
107   bigint ::= # unsigned integer value, ascii base10 notation;
108   binary_data ::= # file content, not interpreted;
109
110   when         ::= raw_when | rfc2822_when;
111   raw_when     ::= ts sp tz;
112   rfc2822_when ::= # Valid RFC 2822 date and time;
113
114   sp ::= # ASCII space character;
115   lf ::= # ASCII newline (LF) character;
116
117      # note: a colon (':') must precede the numerical value assigned to
118      # an idnum.  This is to distinguish it from a ref or tag name as
119      # GIT does not permit ':' in ref or tag strings.
120      #
121   idnum   ::= ':' bigint;
122   path    ::= # GIT style file path, e.g. "a/b/c";
123   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
124   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
125   sha1exp ::= # Any valid GIT SHA1 expression;
126   hexsha1 ::= # SHA1 in hexadecimal format;
127
128      # note: name and email are UTF8 strings, however name must not
129      # contain '<' or lf and email must not contain any of the
130      # following: '<', '>', lf.
131      #
132   name  ::= # valid GIT author/committer name;
133   email ::= # valid GIT author/committer email;
134   ts    ::= # time since the epoch in seconds, ascii base10 notation;
135   tz    ::= # GIT style timezone;
136
137      # note: comments, ls and cat requests may appear anywhere
138      # in the input, except within a data command.  Any form
139      # of the data command always escapes the related input
140      # from comment processing.
141      #
142      # In case it is not clear, the '#' that starts the comment
143      # must be the first character on that line (an lf
144      # preceded it).
145      #
146
147   cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
148   ls_tree  ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
149
150   comment ::= '#' not_lf* lf;
151   not_lf  ::= # Any byte that is not ASCII newline (LF);
152 */
153
154 #include "builtin.h"
155 #include "cache.h"
156 #include "object.h"
157 #include "blob.h"
158 #include "tree.h"
159 #include "commit.h"
160 #include "delta.h"
161 #include "pack.h"
162 #include "refs.h"
163 #include "csum-file.h"
164 #include "quote.h"
165 #include "exec_cmd.h"
166 #include "dir.h"
167
168 #define PACK_ID_BITS 16
169 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
170 #define DEPTH_BITS 13
171 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
172
173 struct object_entry {
174         struct pack_idx_entry idx;
175         struct object_entry *next;
176         uint32_t type : TYPE_BITS,
177                 pack_id : PACK_ID_BITS,
178                 depth : DEPTH_BITS;
179 };
180
181 struct object_entry_pool {
182         struct object_entry_pool *next_pool;
183         struct object_entry *next_free;
184         struct object_entry *end;
185         struct object_entry entries[FLEX_ARRAY]; /* more */
186 };
187
188 struct mark_set {
189         union {
190                 struct object_entry *marked[1024];
191                 struct mark_set *sets[1024];
192         } data;
193         unsigned int shift;
194 };
195
196 struct last_object {
197         struct strbuf data;
198         off_t offset;
199         unsigned int depth;
200         unsigned no_swap : 1;
201 };
202
203 struct mem_pool {
204         struct mem_pool *next_pool;
205         char *next_free;
206         char *end;
207         uintmax_t space[FLEX_ARRAY]; /* more */
208 };
209
210 struct atom_str {
211         struct atom_str *next_atom;
212         unsigned short str_len;
213         char str_dat[FLEX_ARRAY]; /* more */
214 };
215
216 struct tree_content;
217 struct tree_entry {
218         struct tree_content *tree;
219         struct atom_str *name;
220         struct tree_entry_ms {
221                 uint16_t mode;
222                 unsigned char sha1[20];
223         } versions[2];
224 };
225
226 struct tree_content {
227         unsigned int entry_capacity; /* must match avail_tree_content */
228         unsigned int entry_count;
229         unsigned int delta_depth;
230         struct tree_entry *entries[FLEX_ARRAY]; /* more */
231 };
232
233 struct avail_tree_content {
234         unsigned int entry_capacity; /* must match tree_content */
235         struct avail_tree_content *next_avail;
236 };
237
238 struct branch {
239         struct branch *table_next_branch;
240         struct branch *active_next_branch;
241         const char *name;
242         struct tree_entry branch_tree;
243         uintmax_t last_commit;
244         uintmax_t num_notes;
245         unsigned active : 1;
246         unsigned pack_id : PACK_ID_BITS;
247         unsigned char sha1[20];
248 };
249
250 struct tag {
251         struct tag *next_tag;
252         const char *name;
253         unsigned int pack_id;
254         unsigned char sha1[20];
255 };
256
257 struct hash_list {
258         struct hash_list *next;
259         unsigned char sha1[20];
260 };
261
262 typedef enum {
263         WHENSPEC_RAW = 1,
264         WHENSPEC_RFC2822,
265         WHENSPEC_NOW
266 } whenspec_type;
267
268 struct recent_command {
269         struct recent_command *prev;
270         struct recent_command *next;
271         char *buf;
272 };
273
274 /* Configured limits on output */
275 static unsigned long max_depth = 10;
276 static off_t max_packsize;
277 static int force_update;
278 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
279 static int pack_compression_seen;
280
281 /* Stats and misc. counters */
282 static uintmax_t alloc_count;
283 static uintmax_t marks_set_count;
284 static uintmax_t object_count_by_type[1 << TYPE_BITS];
285 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
286 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
287 static unsigned long object_count;
288 static unsigned long branch_count;
289 static unsigned long branch_load_count;
290 static int failure;
291 static FILE *pack_edges;
292 static unsigned int show_stats = 1;
293 static int global_argc;
294 static const char **global_argv;
295
296 /* Memory pools */
297 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
298 static size_t total_allocd;
299 static struct mem_pool *mem_pool;
300
301 /* Atom management */
302 static unsigned int atom_table_sz = 4451;
303 static unsigned int atom_cnt;
304 static struct atom_str **atom_table;
305
306 /* The .pack file being generated */
307 static struct pack_idx_option pack_idx_opts;
308 static unsigned int pack_id;
309 static struct sha1file *pack_file;
310 static struct packed_git *pack_data;
311 static struct packed_git **all_packs;
312 static off_t pack_size;
313
314 /* Table of objects we've written. */
315 static unsigned int object_entry_alloc = 5000;
316 static struct object_entry_pool *blocks;
317 static struct object_entry *object_table[1 << 16];
318 static struct mark_set *marks;
319 static const char *export_marks_file;
320 static const char *import_marks_file;
321 static int import_marks_file_from_stream;
322 static int import_marks_file_ignore_missing;
323 static int relative_marks_paths;
324
325 /* Our last blob */
326 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
327
328 /* Tree management */
329 static unsigned int tree_entry_alloc = 1000;
330 static void *avail_tree_entry;
331 static unsigned int avail_tree_table_sz = 100;
332 static struct avail_tree_content **avail_tree_table;
333 static struct strbuf old_tree = STRBUF_INIT;
334 static struct strbuf new_tree = STRBUF_INIT;
335
336 /* Branch data */
337 static unsigned long max_active_branches = 5;
338 static unsigned long cur_active_branches;
339 static unsigned long branch_table_sz = 1039;
340 static struct branch **branch_table;
341 static struct branch *active_branches;
342
343 /* Tag data */
344 static struct tag *first_tag;
345 static struct tag *last_tag;
346
347 /* Input stream parsing */
348 static whenspec_type whenspec = WHENSPEC_RAW;
349 static struct strbuf command_buf = STRBUF_INIT;
350 static int unread_command_buf;
351 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
352 static struct recent_command *cmd_tail = &cmd_hist;
353 static struct recent_command *rc_free;
354 static unsigned int cmd_save = 100;
355 static uintmax_t next_mark;
356 static struct strbuf new_data = STRBUF_INIT;
357 static int seen_data_command;
358 static int require_explicit_termination;
359
360 /* Signal handling */
361 static volatile sig_atomic_t checkpoint_requested;
362
363 /* Where to write output of cat-blob commands */
364 static int cat_blob_fd = STDOUT_FILENO;
365
366 static void parse_argv(void);
367 static void parse_cat_blob(void);
368 static void parse_ls(struct branch *b);
369
370 static void write_branch_report(FILE *rpt, struct branch *b)
371 {
372         fprintf(rpt, "%s:\n", b->name);
373
374         fprintf(rpt, "  status      :");
375         if (b->active)
376                 fputs(" active", rpt);
377         if (b->branch_tree.tree)
378                 fputs(" loaded", rpt);
379         if (is_null_sha1(b->branch_tree.versions[1].sha1))
380                 fputs(" dirty", rpt);
381         fputc('\n', rpt);
382
383         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
384         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
385         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
386         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
387
388         fputs("  last pack   : ", rpt);
389         if (b->pack_id < MAX_PACK_ID)
390                 fprintf(rpt, "%u", b->pack_id);
391         fputc('\n', rpt);
392
393         fputc('\n', rpt);
394 }
395
396 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
397
398 static void write_crash_report(const char *err)
399 {
400         char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
401         FILE *rpt = fopen(loc, "w");
402         struct branch *b;
403         unsigned long lu;
404         struct recent_command *rc;
405
406         if (!rpt) {
407                 error("can't write crash report %s: %s", loc, strerror(errno));
408                 return;
409         }
410
411         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
412
413         fprintf(rpt, "fast-import crash report:\n");
414         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
415         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
416         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
417         fputc('\n', rpt);
418
419         fputs("fatal: ", rpt);
420         fputs(err, rpt);
421         fputc('\n', rpt);
422
423         fputc('\n', rpt);
424         fputs("Most Recent Commands Before Crash\n", rpt);
425         fputs("---------------------------------\n", rpt);
426         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
427                 if (rc->next == &cmd_hist)
428                         fputs("* ", rpt);
429                 else
430                         fputs("  ", rpt);
431                 fputs(rc->buf, rpt);
432                 fputc('\n', rpt);
433         }
434
435         fputc('\n', rpt);
436         fputs("Active Branch LRU\n", rpt);
437         fputs("-----------------\n", rpt);
438         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
439                 cur_active_branches,
440                 max_active_branches);
441         fputc('\n', rpt);
442         fputs("  pos  clock name\n", rpt);
443         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
444         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
445                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
446                         ++lu, b->last_commit, b->name);
447
448         fputc('\n', rpt);
449         fputs("Inactive Branches\n", rpt);
450         fputs("-----------------\n", rpt);
451         for (lu = 0; lu < branch_table_sz; lu++) {
452                 for (b = branch_table[lu]; b; b = b->table_next_branch)
453                         write_branch_report(rpt, b);
454         }
455
456         if (first_tag) {
457                 struct tag *tg;
458                 fputc('\n', rpt);
459                 fputs("Annotated Tags\n", rpt);
460                 fputs("--------------\n", rpt);
461                 for (tg = first_tag; tg; tg = tg->next_tag) {
462                         fputs(sha1_to_hex(tg->sha1), rpt);
463                         fputc(' ', rpt);
464                         fputs(tg->name, rpt);
465                         fputc('\n', rpt);
466                 }
467         }
468
469         fputc('\n', rpt);
470         fputs("Marks\n", rpt);
471         fputs("-----\n", rpt);
472         if (export_marks_file)
473                 fprintf(rpt, "  exported to %s\n", export_marks_file);
474         else
475                 dump_marks_helper(rpt, 0, marks);
476
477         fputc('\n', rpt);
478         fputs("-------------------\n", rpt);
479         fputs("END OF CRASH REPORT\n", rpt);
480         fclose(rpt);
481 }
482
483 static void end_packfile(void);
484 static void unkeep_all_packs(void);
485 static void dump_marks(void);
486
487 static NORETURN void die_nicely(const char *err, va_list params)
488 {
489         static int zombie;
490         char message[2 * PATH_MAX];
491
492         vsnprintf(message, sizeof(message), err, params);
493         fputs("fatal: ", stderr);
494         fputs(message, stderr);
495         fputc('\n', stderr);
496
497         if (!zombie) {
498                 zombie = 1;
499                 write_crash_report(message);
500                 end_packfile();
501                 unkeep_all_packs();
502                 dump_marks();
503         }
504         exit(128);
505 }
506
507 #ifndef SIGUSR1 /* Windows, for example */
508
509 static void set_checkpoint_signal(void)
510 {
511 }
512
513 #else
514
515 static void checkpoint_signal(int signo)
516 {
517         checkpoint_requested = 1;
518 }
519
520 static void set_checkpoint_signal(void)
521 {
522         struct sigaction sa;
523
524         memset(&sa, 0, sizeof(sa));
525         sa.sa_handler = checkpoint_signal;
526         sigemptyset(&sa.sa_mask);
527         sa.sa_flags = SA_RESTART;
528         sigaction(SIGUSR1, &sa, NULL);
529 }
530
531 #endif
532
533 static void alloc_objects(unsigned int cnt)
534 {
535         struct object_entry_pool *b;
536
537         b = xmalloc(sizeof(struct object_entry_pool)
538                 + cnt * sizeof(struct object_entry));
539         b->next_pool = blocks;
540         b->next_free = b->entries;
541         b->end = b->entries + cnt;
542         blocks = b;
543         alloc_count += cnt;
544 }
545
546 static struct object_entry *new_object(unsigned char *sha1)
547 {
548         struct object_entry *e;
549
550         if (blocks->next_free == blocks->end)
551                 alloc_objects(object_entry_alloc);
552
553         e = blocks->next_free++;
554         hashcpy(e->idx.sha1, sha1);
555         return e;
556 }
557
558 static struct object_entry *find_object(unsigned char *sha1)
559 {
560         unsigned int h = sha1[0] << 8 | sha1[1];
561         struct object_entry *e;
562         for (e = object_table[h]; e; e = e->next)
563                 if (!hashcmp(sha1, e->idx.sha1))
564                         return e;
565         return NULL;
566 }
567
568 static struct object_entry *insert_object(unsigned char *sha1)
569 {
570         unsigned int h = sha1[0] << 8 | sha1[1];
571         struct object_entry *e = object_table[h];
572
573         while (e) {
574                 if (!hashcmp(sha1, e->idx.sha1))
575                         return e;
576                 e = e->next;
577         }
578
579         e = new_object(sha1);
580         e->next = object_table[h];
581         e->idx.offset = 0;
582         object_table[h] = e;
583         return e;
584 }
585
586 static unsigned int hc_str(const char *s, size_t len)
587 {
588         unsigned int r = 0;
589         while (len-- > 0)
590                 r = r * 31 + *s++;
591         return r;
592 }
593
594 static void *pool_alloc(size_t len)
595 {
596         struct mem_pool *p;
597         void *r;
598
599         /* round up to a 'uintmax_t' alignment */
600         if (len & (sizeof(uintmax_t) - 1))
601                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
602
603         for (p = mem_pool; p; p = p->next_pool)
604                 if ((p->end - p->next_free >= len))
605                         break;
606
607         if (!p) {
608                 if (len >= (mem_pool_alloc/2)) {
609                         total_allocd += len;
610                         return xmalloc(len);
611                 }
612                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
613                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
614                 p->next_pool = mem_pool;
615                 p->next_free = (char *) p->space;
616                 p->end = p->next_free + mem_pool_alloc;
617                 mem_pool = p;
618         }
619
620         r = p->next_free;
621         p->next_free += len;
622         return r;
623 }
624
625 static void *pool_calloc(size_t count, size_t size)
626 {
627         size_t len = count * size;
628         void *r = pool_alloc(len);
629         memset(r, 0, len);
630         return r;
631 }
632
633 static char *pool_strdup(const char *s)
634 {
635         char *r = pool_alloc(strlen(s) + 1);
636         strcpy(r, s);
637         return r;
638 }
639
640 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
641 {
642         struct mark_set *s = marks;
643         while ((idnum >> s->shift) >= 1024) {
644                 s = pool_calloc(1, sizeof(struct mark_set));
645                 s->shift = marks->shift + 10;
646                 s->data.sets[0] = marks;
647                 marks = s;
648         }
649         while (s->shift) {
650                 uintmax_t i = idnum >> s->shift;
651                 idnum -= i << s->shift;
652                 if (!s->data.sets[i]) {
653                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
654                         s->data.sets[i]->shift = s->shift - 10;
655                 }
656                 s = s->data.sets[i];
657         }
658         if (!s->data.marked[idnum])
659                 marks_set_count++;
660         s->data.marked[idnum] = oe;
661 }
662
663 static struct object_entry *find_mark(uintmax_t idnum)
664 {
665         uintmax_t orig_idnum = idnum;
666         struct mark_set *s = marks;
667         struct object_entry *oe = NULL;
668         if ((idnum >> s->shift) < 1024) {
669                 while (s && s->shift) {
670                         uintmax_t i = idnum >> s->shift;
671                         idnum -= i << s->shift;
672                         s = s->data.sets[i];
673                 }
674                 if (s)
675                         oe = s->data.marked[idnum];
676         }
677         if (!oe)
678                 die("mark :%" PRIuMAX " not declared", orig_idnum);
679         return oe;
680 }
681
682 static struct atom_str *to_atom(const char *s, unsigned short len)
683 {
684         unsigned int hc = hc_str(s, len) % atom_table_sz;
685         struct atom_str *c;
686
687         for (c = atom_table[hc]; c; c = c->next_atom)
688                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
689                         return c;
690
691         c = pool_alloc(sizeof(struct atom_str) + len + 1);
692         c->str_len = len;
693         strncpy(c->str_dat, s, len);
694         c->str_dat[len] = 0;
695         c->next_atom = atom_table[hc];
696         atom_table[hc] = c;
697         atom_cnt++;
698         return c;
699 }
700
701 static struct branch *lookup_branch(const char *name)
702 {
703         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
704         struct branch *b;
705
706         for (b = branch_table[hc]; b; b = b->table_next_branch)
707                 if (!strcmp(name, b->name))
708                         return b;
709         return NULL;
710 }
711
712 static struct branch *new_branch(const char *name)
713 {
714         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
715         struct branch *b = lookup_branch(name);
716
717         if (b)
718                 die("Invalid attempt to create duplicate branch: %s", name);
719         switch (check_ref_format(name)) {
720         case 0: break; /* its valid */
721         case CHECK_REF_FORMAT_ONELEVEL:
722                 break; /* valid, but too few '/', allow anyway */
723         default:
724                 die("Branch name doesn't conform to GIT standards: %s", name);
725         }
726
727         b = pool_calloc(1, sizeof(struct branch));
728         b->name = pool_strdup(name);
729         b->table_next_branch = branch_table[hc];
730         b->branch_tree.versions[0].mode = S_IFDIR;
731         b->branch_tree.versions[1].mode = S_IFDIR;
732         b->num_notes = 0;
733         b->active = 0;
734         b->pack_id = MAX_PACK_ID;
735         branch_table[hc] = b;
736         branch_count++;
737         return b;
738 }
739
740 static unsigned int hc_entries(unsigned int cnt)
741 {
742         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
743         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
744 }
745
746 static struct tree_content *new_tree_content(unsigned int cnt)
747 {
748         struct avail_tree_content *f, *l = NULL;
749         struct tree_content *t;
750         unsigned int hc = hc_entries(cnt);
751
752         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
753                 if (f->entry_capacity >= cnt)
754                         break;
755
756         if (f) {
757                 if (l)
758                         l->next_avail = f->next_avail;
759                 else
760                         avail_tree_table[hc] = f->next_avail;
761         } else {
762                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
763                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
764                 f->entry_capacity = cnt;
765         }
766
767         t = (struct tree_content*)f;
768         t->entry_count = 0;
769         t->delta_depth = 0;
770         return t;
771 }
772
773 static void release_tree_entry(struct tree_entry *e);
774 static void release_tree_content(struct tree_content *t)
775 {
776         struct avail_tree_content *f = (struct avail_tree_content*)t;
777         unsigned int hc = hc_entries(f->entry_capacity);
778         f->next_avail = avail_tree_table[hc];
779         avail_tree_table[hc] = f;
780 }
781
782 static void release_tree_content_recursive(struct tree_content *t)
783 {
784         unsigned int i;
785         for (i = 0; i < t->entry_count; i++)
786                 release_tree_entry(t->entries[i]);
787         release_tree_content(t);
788 }
789
790 static struct tree_content *grow_tree_content(
791         struct tree_content *t,
792         int amt)
793 {
794         struct tree_content *r = new_tree_content(t->entry_count + amt);
795         r->entry_count = t->entry_count;
796         r->delta_depth = t->delta_depth;
797         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
798         release_tree_content(t);
799         return r;
800 }
801
802 static struct tree_entry *new_tree_entry(void)
803 {
804         struct tree_entry *e;
805
806         if (!avail_tree_entry) {
807                 unsigned int n = tree_entry_alloc;
808                 total_allocd += n * sizeof(struct tree_entry);
809                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
810                 while (n-- > 1) {
811                         *((void**)e) = e + 1;
812                         e++;
813                 }
814                 *((void**)e) = NULL;
815         }
816
817         e = avail_tree_entry;
818         avail_tree_entry = *((void**)e);
819         return e;
820 }
821
822 static void release_tree_entry(struct tree_entry *e)
823 {
824         if (e->tree)
825                 release_tree_content_recursive(e->tree);
826         *((void**)e) = avail_tree_entry;
827         avail_tree_entry = e;
828 }
829
830 static struct tree_content *dup_tree_content(struct tree_content *s)
831 {
832         struct tree_content *d;
833         struct tree_entry *a, *b;
834         unsigned int i;
835
836         if (!s)
837                 return NULL;
838         d = new_tree_content(s->entry_count);
839         for (i = 0; i < s->entry_count; i++) {
840                 a = s->entries[i];
841                 b = new_tree_entry();
842                 memcpy(b, a, sizeof(*a));
843                 if (a->tree && is_null_sha1(b->versions[1].sha1))
844                         b->tree = dup_tree_content(a->tree);
845                 else
846                         b->tree = NULL;
847                 d->entries[i] = b;
848         }
849         d->entry_count = s->entry_count;
850         d->delta_depth = s->delta_depth;
851
852         return d;
853 }
854
855 static void start_packfile(void)
856 {
857         static char tmpfile[PATH_MAX];
858         struct packed_git *p;
859         struct pack_header hdr;
860         int pack_fd;
861
862         pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
863                               "pack/tmp_pack_XXXXXX");
864         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
865         strcpy(p->pack_name, tmpfile);
866         p->pack_fd = pack_fd;
867         p->do_not_close = 1;
868         pack_file = sha1fd(pack_fd, p->pack_name);
869
870         hdr.hdr_signature = htonl(PACK_SIGNATURE);
871         hdr.hdr_version = htonl(2);
872         hdr.hdr_entries = 0;
873         sha1write(pack_file, &hdr, sizeof(hdr));
874
875         pack_data = p;
876         pack_size = sizeof(hdr);
877         object_count = 0;
878
879         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
880         all_packs[pack_id] = p;
881 }
882
883 static const char *create_index(void)
884 {
885         const char *tmpfile;
886         struct pack_idx_entry **idx, **c, **last;
887         struct object_entry *e;
888         struct object_entry_pool *o;
889
890         /* Build the table of object IDs. */
891         idx = xmalloc(object_count * sizeof(*idx));
892         c = idx;
893         for (o = blocks; o; o = o->next_pool)
894                 for (e = o->next_free; e-- != o->entries;)
895                         if (pack_id == e->pack_id)
896                                 *c++ = &e->idx;
897         last = idx + object_count;
898         if (c != last)
899                 die("internal consistency error creating the index");
900
901         tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
902         free(idx);
903         return tmpfile;
904 }
905
906 static char *keep_pack(const char *curr_index_name)
907 {
908         static char name[PATH_MAX];
909         static const char *keep_msg = "fast-import";
910         int keep_fd;
911
912         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
913         if (keep_fd < 0)
914                 die_errno("cannot create keep file");
915         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
916         if (close(keep_fd))
917                 die_errno("failed to write keep file");
918
919         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
920                  get_object_directory(), sha1_to_hex(pack_data->sha1));
921         if (move_temp_to_file(pack_data->pack_name, name))
922                 die("cannot store pack file");
923
924         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
925                  get_object_directory(), sha1_to_hex(pack_data->sha1));
926         if (move_temp_to_file(curr_index_name, name))
927                 die("cannot store index file");
928         free((void *)curr_index_name);
929         return name;
930 }
931
932 static void unkeep_all_packs(void)
933 {
934         static char name[PATH_MAX];
935         int k;
936
937         for (k = 0; k < pack_id; k++) {
938                 struct packed_git *p = all_packs[k];
939                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
940                          get_object_directory(), sha1_to_hex(p->sha1));
941                 unlink_or_warn(name);
942         }
943 }
944
945 static void end_packfile(void)
946 {
947         struct packed_git *old_p = pack_data, *new_p;
948
949         clear_delta_base_cache();
950         if (object_count) {
951                 unsigned char cur_pack_sha1[20];
952                 char *idx_name;
953                 int i;
954                 struct branch *b;
955                 struct tag *t;
956
957                 close_pack_windows(pack_data);
958                 sha1close(pack_file, cur_pack_sha1, 0);
959                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
960                                     pack_data->pack_name, object_count,
961                                     cur_pack_sha1, pack_size);
962                 close(pack_data->pack_fd);
963                 idx_name = keep_pack(create_index());
964
965                 /* Register the packfile with core git's machinery. */
966                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
967                 if (!new_p)
968                         die("core git rejected index %s", idx_name);
969                 all_packs[pack_id] = new_p;
970                 install_packed_git(new_p);
971
972                 /* Print the boundary */
973                 if (pack_edges) {
974                         fprintf(pack_edges, "%s:", new_p->pack_name);
975                         for (i = 0; i < branch_table_sz; i++) {
976                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
977                                         if (b->pack_id == pack_id)
978                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
979                                 }
980                         }
981                         for (t = first_tag; t; t = t->next_tag) {
982                                 if (t->pack_id == pack_id)
983                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
984                         }
985                         fputc('\n', pack_edges);
986                         fflush(pack_edges);
987                 }
988
989                 pack_id++;
990         }
991         else {
992                 close(old_p->pack_fd);
993                 unlink_or_warn(old_p->pack_name);
994         }
995         free(old_p);
996
997         /* We can't carry a delta across packfiles. */
998         strbuf_release(&last_blob.data);
999         last_blob.offset = 0;
1000         last_blob.depth = 0;
1001 }
1002
1003 static void cycle_packfile(void)
1004 {
1005         end_packfile();
1006         start_packfile();
1007 }
1008
1009 static int store_object(
1010         enum object_type type,
1011         struct strbuf *dat,
1012         struct last_object *last,
1013         unsigned char *sha1out,
1014         uintmax_t mark)
1015 {
1016         void *out, *delta;
1017         struct object_entry *e;
1018         unsigned char hdr[96];
1019         unsigned char sha1[20];
1020         unsigned long hdrlen, deltalen;
1021         git_SHA_CTX c;
1022         git_zstream s;
1023
1024         hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1025                 (unsigned long)dat->len) + 1;
1026         git_SHA1_Init(&c);
1027         git_SHA1_Update(&c, hdr, hdrlen);
1028         git_SHA1_Update(&c, dat->buf, dat->len);
1029         git_SHA1_Final(sha1, &c);
1030         if (sha1out)
1031                 hashcpy(sha1out, sha1);
1032
1033         e = insert_object(sha1);
1034         if (mark)
1035                 insert_mark(mark, e);
1036         if (e->idx.offset) {
1037                 duplicate_count_by_type[type]++;
1038                 return 1;
1039         } else if (find_sha1_pack(sha1, packed_git)) {
1040                 e->type = type;
1041                 e->pack_id = MAX_PACK_ID;
1042                 e->idx.offset = 1; /* just not zero! */
1043                 duplicate_count_by_type[type]++;
1044                 return 1;
1045         }
1046
1047         if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1048                 delta = diff_delta(last->data.buf, last->data.len,
1049                         dat->buf, dat->len,
1050                         &deltalen, dat->len - 20);
1051         } else
1052                 delta = NULL;
1053
1054         memset(&s, 0, sizeof(s));
1055         git_deflate_init(&s, pack_compression_level);
1056         if (delta) {
1057                 s.next_in = delta;
1058                 s.avail_in = deltalen;
1059         } else {
1060                 s.next_in = (void *)dat->buf;
1061                 s.avail_in = dat->len;
1062         }
1063         s.avail_out = git_deflate_bound(&s, s.avail_in);
1064         s.next_out = out = xmalloc(s.avail_out);
1065         while (git_deflate(&s, Z_FINISH) == Z_OK)
1066                 ; /* nothing */
1067         git_deflate_end(&s);
1068
1069         /* Determine if we should auto-checkpoint. */
1070         if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1071                 || (pack_size + 60 + s.total_out) < pack_size) {
1072
1073                 /* This new object needs to *not* have the current pack_id. */
1074                 e->pack_id = pack_id + 1;
1075                 cycle_packfile();
1076
1077                 /* We cannot carry a delta into the new pack. */
1078                 if (delta) {
1079                         free(delta);
1080                         delta = NULL;
1081
1082                         memset(&s, 0, sizeof(s));
1083                         git_deflate_init(&s, pack_compression_level);
1084                         s.next_in = (void *)dat->buf;
1085                         s.avail_in = dat->len;
1086                         s.avail_out = git_deflate_bound(&s, s.avail_in);
1087                         s.next_out = out = xrealloc(out, s.avail_out);
1088                         while (git_deflate(&s, Z_FINISH) == Z_OK)
1089                                 ; /* nothing */
1090                         git_deflate_end(&s);
1091                 }
1092         }
1093
1094         e->type = type;
1095         e->pack_id = pack_id;
1096         e->idx.offset = pack_size;
1097         object_count++;
1098         object_count_by_type[type]++;
1099
1100         crc32_begin(pack_file);
1101
1102         if (delta) {
1103                 off_t ofs = e->idx.offset - last->offset;
1104                 unsigned pos = sizeof(hdr) - 1;
1105
1106                 delta_count_by_type[type]++;
1107                 e->depth = last->depth + 1;
1108
1109                 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1110                 sha1write(pack_file, hdr, hdrlen);
1111                 pack_size += hdrlen;
1112
1113                 hdr[pos] = ofs & 127;
1114                 while (ofs >>= 7)
1115                         hdr[--pos] = 128 | (--ofs & 127);
1116                 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1117                 pack_size += sizeof(hdr) - pos;
1118         } else {
1119                 e->depth = 0;
1120                 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1121                 sha1write(pack_file, hdr, hdrlen);
1122                 pack_size += hdrlen;
1123         }
1124
1125         sha1write(pack_file, out, s.total_out);
1126         pack_size += s.total_out;
1127
1128         e->idx.crc32 = crc32_end(pack_file);
1129
1130         free(out);
1131         free(delta);
1132         if (last) {
1133                 if (last->no_swap) {
1134                         last->data = *dat;
1135                 } else {
1136                         strbuf_swap(&last->data, dat);
1137                 }
1138                 last->offset = e->idx.offset;
1139                 last->depth = e->depth;
1140         }
1141         return 0;
1142 }
1143
1144 static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1145 {
1146         if (ftruncate(pack_data->pack_fd, to)
1147          || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1148                 die_errno("cannot truncate pack to skip duplicate");
1149         pack_size = to;
1150
1151         /* yes this is a layering violation */
1152         pack_file->total = to;
1153         pack_file->offset = 0;
1154         pack_file->ctx = *ctx;
1155 }
1156
1157 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1158 {
1159         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1160         unsigned char *in_buf = xmalloc(in_sz);
1161         unsigned char *out_buf = xmalloc(out_sz);
1162         struct object_entry *e;
1163         unsigned char sha1[20];
1164         unsigned long hdrlen;
1165         off_t offset;
1166         git_SHA_CTX c;
1167         git_SHA_CTX pack_file_ctx;
1168         git_zstream s;
1169         int status = Z_OK;
1170
1171         /* Determine if we should auto-checkpoint. */
1172         if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1173                 || (pack_size + 60 + len) < pack_size)
1174                 cycle_packfile();
1175
1176         offset = pack_size;
1177
1178         /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1179         sha1flush(pack_file);
1180         pack_file_ctx = pack_file->ctx;
1181
1182         hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1183         if (out_sz <= hdrlen)
1184                 die("impossibly large object header");
1185
1186         git_SHA1_Init(&c);
1187         git_SHA1_Update(&c, out_buf, hdrlen);
1188
1189         crc32_begin(pack_file);
1190
1191         memset(&s, 0, sizeof(s));
1192         git_deflate_init(&s, pack_compression_level);
1193
1194         hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1195         if (out_sz <= hdrlen)
1196                 die("impossibly large object header");
1197
1198         s.next_out = out_buf + hdrlen;
1199         s.avail_out = out_sz - hdrlen;
1200
1201         while (status != Z_STREAM_END) {
1202                 if (0 < len && !s.avail_in) {
1203                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1204                         size_t n = fread(in_buf, 1, cnt, stdin);
1205                         if (!n && feof(stdin))
1206                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1207
1208                         git_SHA1_Update(&c, in_buf, n);
1209                         s.next_in = in_buf;
1210                         s.avail_in = n;
1211                         len -= n;
1212                 }
1213
1214                 status = git_deflate(&s, len ? 0 : Z_FINISH);
1215
1216                 if (!s.avail_out || status == Z_STREAM_END) {
1217                         size_t n = s.next_out - out_buf;
1218                         sha1write(pack_file, out_buf, n);
1219                         pack_size += n;
1220                         s.next_out = out_buf;
1221                         s.avail_out = out_sz;
1222                 }
1223
1224                 switch (status) {
1225                 case Z_OK:
1226                 case Z_BUF_ERROR:
1227                 case Z_STREAM_END:
1228                         continue;
1229                 default:
1230                         die("unexpected deflate failure: %d", status);
1231                 }
1232         }
1233         git_deflate_end(&s);
1234         git_SHA1_Final(sha1, &c);
1235
1236         if (sha1out)
1237                 hashcpy(sha1out, sha1);
1238
1239         e = insert_object(sha1);
1240
1241         if (mark)
1242                 insert_mark(mark, e);
1243
1244         if (e->idx.offset) {
1245                 duplicate_count_by_type[OBJ_BLOB]++;
1246                 truncate_pack(offset, &pack_file_ctx);
1247
1248         } else if (find_sha1_pack(sha1, packed_git)) {
1249                 e->type = OBJ_BLOB;
1250                 e->pack_id = MAX_PACK_ID;
1251                 e->idx.offset = 1; /* just not zero! */
1252                 duplicate_count_by_type[OBJ_BLOB]++;
1253                 truncate_pack(offset, &pack_file_ctx);
1254
1255         } else {
1256                 e->depth = 0;
1257                 e->type = OBJ_BLOB;
1258                 e->pack_id = pack_id;
1259                 e->idx.offset = offset;
1260                 e->idx.crc32 = crc32_end(pack_file);
1261                 object_count++;
1262                 object_count_by_type[OBJ_BLOB]++;
1263         }
1264
1265         free(in_buf);
1266         free(out_buf);
1267 }
1268
1269 /* All calls must be guarded by find_object() or find_mark() to
1270  * ensure the 'struct object_entry' passed was written by this
1271  * process instance.  We unpack the entry by the offset, avoiding
1272  * the need for the corresponding .idx file.  This unpacking rule
1273  * works because we only use OBJ_REF_DELTA within the packfiles
1274  * created by fast-import.
1275  *
1276  * oe must not be NULL.  Such an oe usually comes from giving
1277  * an unknown SHA-1 to find_object() or an undefined mark to
1278  * find_mark().  Callers must test for this condition and use
1279  * the standard read_sha1_file() when it happens.
1280  *
1281  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1282  * find_mark(), where the mark was reloaded from an existing marks
1283  * file and is referencing an object that this fast-import process
1284  * instance did not write out to a packfile.  Callers must test for
1285  * this condition and use read_sha1_file() instead.
1286  */
1287 static void *gfi_unpack_entry(
1288         struct object_entry *oe,
1289         unsigned long *sizep)
1290 {
1291         enum object_type type;
1292         struct packed_git *p = all_packs[oe->pack_id];
1293         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1294                 /* The object is stored in the packfile we are writing to
1295                  * and we have modified it since the last time we scanned
1296                  * back to read a previously written object.  If an old
1297                  * window covered [p->pack_size, p->pack_size + 20) its
1298                  * data is stale and is not valid.  Closing all windows
1299                  * and updating the packfile length ensures we can read
1300                  * the newly written data.
1301                  */
1302                 close_pack_windows(p);
1303                 sha1flush(pack_file);
1304
1305                 /* We have to offer 20 bytes additional on the end of
1306                  * the packfile as the core unpacker code assumes the
1307                  * footer is present at the file end and must promise
1308                  * at least 20 bytes within any window it maps.  But
1309                  * we don't actually create the footer here.
1310                  */
1311                 p->pack_size = pack_size + 20;
1312         }
1313         return unpack_entry(p, oe->idx.offset, &type, sizep);
1314 }
1315
1316 static const char *get_mode(const char *str, uint16_t *modep)
1317 {
1318         unsigned char c;
1319         uint16_t mode = 0;
1320
1321         while ((c = *str++) != ' ') {
1322                 if (c < '0' || c > '7')
1323                         return NULL;
1324                 mode = (mode << 3) + (c - '0');
1325         }
1326         *modep = mode;
1327         return str;
1328 }
1329
1330 static void load_tree(struct tree_entry *root)
1331 {
1332         unsigned char *sha1 = root->versions[1].sha1;
1333         struct object_entry *myoe;
1334         struct tree_content *t;
1335         unsigned long size;
1336         char *buf;
1337         const char *c;
1338
1339         root->tree = t = new_tree_content(8);
1340         if (is_null_sha1(sha1))
1341                 return;
1342
1343         myoe = find_object(sha1);
1344         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1345                 if (myoe->type != OBJ_TREE)
1346                         die("Not a tree: %s", sha1_to_hex(sha1));
1347                 t->delta_depth = myoe->depth;
1348                 buf = gfi_unpack_entry(myoe, &size);
1349                 if (!buf)
1350                         die("Can't load tree %s", sha1_to_hex(sha1));
1351         } else {
1352                 enum object_type type;
1353                 buf = read_sha1_file(sha1, &type, &size);
1354                 if (!buf || type != OBJ_TREE)
1355                         die("Can't load tree %s", sha1_to_hex(sha1));
1356         }
1357
1358         c = buf;
1359         while (c != (buf + size)) {
1360                 struct tree_entry *e = new_tree_entry();
1361
1362                 if (t->entry_count == t->entry_capacity)
1363                         root->tree = t = grow_tree_content(t, t->entry_count);
1364                 t->entries[t->entry_count++] = e;
1365
1366                 e->tree = NULL;
1367                 c = get_mode(c, &e->versions[1].mode);
1368                 if (!c)
1369                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1370                 e->versions[0].mode = e->versions[1].mode;
1371                 e->name = to_atom(c, strlen(c));
1372                 c += e->name->str_len + 1;
1373                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1374                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1375                 c += 20;
1376         }
1377         free(buf);
1378 }
1379
1380 static int tecmp0 (const void *_a, const void *_b)
1381 {
1382         struct tree_entry *a = *((struct tree_entry**)_a);
1383         struct tree_entry *b = *((struct tree_entry**)_b);
1384         return base_name_compare(
1385                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1386                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1387 }
1388
1389 static int tecmp1 (const void *_a, const void *_b)
1390 {
1391         struct tree_entry *a = *((struct tree_entry**)_a);
1392         struct tree_entry *b = *((struct tree_entry**)_b);
1393         return base_name_compare(
1394                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1395                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1396 }
1397
1398 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1399 {
1400         size_t maxlen = 0;
1401         unsigned int i;
1402
1403         if (!v)
1404                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1405         else
1406                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1407
1408         for (i = 0; i < t->entry_count; i++) {
1409                 if (t->entries[i]->versions[v].mode)
1410                         maxlen += t->entries[i]->name->str_len + 34;
1411         }
1412
1413         strbuf_reset(b);
1414         strbuf_grow(b, maxlen);
1415         for (i = 0; i < t->entry_count; i++) {
1416                 struct tree_entry *e = t->entries[i];
1417                 if (!e->versions[v].mode)
1418                         continue;
1419                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1420                                         e->name->str_dat, '\0');
1421                 strbuf_add(b, e->versions[v].sha1, 20);
1422         }
1423 }
1424
1425 static void store_tree(struct tree_entry *root)
1426 {
1427         struct tree_content *t = root->tree;
1428         unsigned int i, j, del;
1429         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1430         struct object_entry *le;
1431
1432         if (!is_null_sha1(root->versions[1].sha1))
1433                 return;
1434
1435         for (i = 0; i < t->entry_count; i++) {
1436                 if (t->entries[i]->tree)
1437                         store_tree(t->entries[i]);
1438         }
1439
1440         le = find_object(root->versions[0].sha1);
1441         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1442                 mktree(t, 0, &old_tree);
1443                 lo.data = old_tree;
1444                 lo.offset = le->idx.offset;
1445                 lo.depth = t->delta_depth;
1446         }
1447
1448         mktree(t, 1, &new_tree);
1449         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1450
1451         t->delta_depth = lo.depth;
1452         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1453                 struct tree_entry *e = t->entries[i];
1454                 if (e->versions[1].mode) {
1455                         e->versions[0].mode = e->versions[1].mode;
1456                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1457                         t->entries[j++] = e;
1458                 } else {
1459                         release_tree_entry(e);
1460                         del++;
1461                 }
1462         }
1463         t->entry_count -= del;
1464 }
1465
1466 static void tree_content_replace(
1467         struct tree_entry *root,
1468         const unsigned char *sha1,
1469         const uint16_t mode,
1470         struct tree_content *newtree)
1471 {
1472         if (!S_ISDIR(mode))
1473                 die("Root cannot be a non-directory");
1474         hashcpy(root->versions[1].sha1, sha1);
1475         if (root->tree)
1476                 release_tree_content_recursive(root->tree);
1477         root->tree = newtree;
1478 }
1479
1480 static int tree_content_set(
1481         struct tree_entry *root,
1482         const char *p,
1483         const unsigned char *sha1,
1484         const uint16_t mode,
1485         struct tree_content *subtree)
1486 {
1487         struct tree_content *t;
1488         const char *slash1;
1489         unsigned int i, n;
1490         struct tree_entry *e;
1491
1492         slash1 = strchr(p, '/');
1493         if (slash1)
1494                 n = slash1 - p;
1495         else
1496                 n = strlen(p);
1497         if (!n)
1498                 die("Empty path component found in input");
1499         if (!slash1 && !S_ISDIR(mode) && subtree)
1500                 die("Non-directories cannot have subtrees");
1501
1502         if (!root->tree)
1503                 load_tree(root);
1504         t = root->tree;
1505         for (i = 0; i < t->entry_count; i++) {
1506                 e = t->entries[i];
1507                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1508                         if (!slash1) {
1509                                 if (!S_ISDIR(mode)
1510                                                 && e->versions[1].mode == mode
1511                                                 && !hashcmp(e->versions[1].sha1, sha1))
1512                                         return 0;
1513                                 e->versions[1].mode = mode;
1514                                 hashcpy(e->versions[1].sha1, sha1);
1515                                 if (e->tree)
1516                                         release_tree_content_recursive(e->tree);
1517                                 e->tree = subtree;
1518                                 hashclr(root->versions[1].sha1);
1519                                 return 1;
1520                         }
1521                         if (!S_ISDIR(e->versions[1].mode)) {
1522                                 e->tree = new_tree_content(8);
1523                                 e->versions[1].mode = S_IFDIR;
1524                         }
1525                         if (!e->tree)
1526                                 load_tree(e);
1527                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1528                                 hashclr(root->versions[1].sha1);
1529                                 return 1;
1530                         }
1531                         return 0;
1532                 }
1533         }
1534
1535         if (t->entry_count == t->entry_capacity)
1536                 root->tree = t = grow_tree_content(t, t->entry_count);
1537         e = new_tree_entry();
1538         e->name = to_atom(p, n);
1539         e->versions[0].mode = 0;
1540         hashclr(e->versions[0].sha1);
1541         t->entries[t->entry_count++] = e;
1542         if (slash1) {
1543                 e->tree = new_tree_content(8);
1544                 e->versions[1].mode = S_IFDIR;
1545                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1546         } else {
1547                 e->tree = subtree;
1548                 e->versions[1].mode = mode;
1549                 hashcpy(e->versions[1].sha1, sha1);
1550         }
1551         hashclr(root->versions[1].sha1);
1552         return 1;
1553 }
1554
1555 static int tree_content_remove(
1556         struct tree_entry *root,
1557         const char *p,
1558         struct tree_entry *backup_leaf)
1559 {
1560         struct tree_content *t;
1561         const char *slash1;
1562         unsigned int i, n;
1563         struct tree_entry *e;
1564
1565         slash1 = strchr(p, '/');
1566         if (slash1)
1567                 n = slash1 - p;
1568         else
1569                 n = strlen(p);
1570
1571         if (!root->tree)
1572                 load_tree(root);
1573         t = root->tree;
1574         for (i = 0; i < t->entry_count; i++) {
1575                 e = t->entries[i];
1576                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1577                         if (slash1 && !S_ISDIR(e->versions[1].mode))
1578                                 /*
1579                                  * If p names a file in some subdirectory, and a
1580                                  * file or symlink matching the name of the
1581                                  * parent directory of p exists, then p cannot
1582                                  * exist and need not be deleted.
1583                                  */
1584                                 return 1;
1585                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1586                                 goto del_entry;
1587                         if (!e->tree)
1588                                 load_tree(e);
1589                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1590                                 for (n = 0; n < e->tree->entry_count; n++) {
1591                                         if (e->tree->entries[n]->versions[1].mode) {
1592                                                 hashclr(root->versions[1].sha1);
1593                                                 return 1;
1594                                         }
1595                                 }
1596                                 backup_leaf = NULL;
1597                                 goto del_entry;
1598                         }
1599                         return 0;
1600                 }
1601         }
1602         return 0;
1603
1604 del_entry:
1605         if (backup_leaf)
1606                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1607         else if (e->tree)
1608                 release_tree_content_recursive(e->tree);
1609         e->tree = NULL;
1610         e->versions[1].mode = 0;
1611         hashclr(e->versions[1].sha1);
1612         hashclr(root->versions[1].sha1);
1613         return 1;
1614 }
1615
1616 static int tree_content_get(
1617         struct tree_entry *root,
1618         const char *p,
1619         struct tree_entry *leaf)
1620 {
1621         struct tree_content *t;
1622         const char *slash1;
1623         unsigned int i, n;
1624         struct tree_entry *e;
1625
1626         slash1 = strchr(p, '/');
1627         if (slash1)
1628                 n = slash1 - p;
1629         else
1630                 n = strlen(p);
1631
1632         if (!root->tree)
1633                 load_tree(root);
1634         t = root->tree;
1635         for (i = 0; i < t->entry_count; i++) {
1636                 e = t->entries[i];
1637                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1638                         if (!slash1) {
1639                                 memcpy(leaf, e, sizeof(*leaf));
1640                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1641                                         leaf->tree = dup_tree_content(e->tree);
1642                                 else
1643                                         leaf->tree = NULL;
1644                                 return 1;
1645                         }
1646                         if (!S_ISDIR(e->versions[1].mode))
1647                                 return 0;
1648                         if (!e->tree)
1649                                 load_tree(e);
1650                         return tree_content_get(e, slash1 + 1, leaf);
1651                 }
1652         }
1653         return 0;
1654 }
1655
1656 static int update_branch(struct branch *b)
1657 {
1658         static const char *msg = "fast-import";
1659         struct ref_lock *lock;
1660         unsigned char old_sha1[20];
1661
1662         if (is_null_sha1(b->sha1))
1663                 return 0;
1664         if (read_ref(b->name, old_sha1))
1665                 hashclr(old_sha1);
1666         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1667         if (!lock)
1668                 return error("Unable to lock %s", b->name);
1669         if (!force_update && !is_null_sha1(old_sha1)) {
1670                 struct commit *old_cmit, *new_cmit;
1671
1672                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1673                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1674                 if (!old_cmit || !new_cmit) {
1675                         unlock_ref(lock);
1676                         return error("Branch %s is missing commits.", b->name);
1677                 }
1678
1679                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1680                         unlock_ref(lock);
1681                         warning("Not updating %s"
1682                                 " (new tip %s does not contain %s)",
1683                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1684                         return -1;
1685                 }
1686         }
1687         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1688                 return error("Unable to update %s", b->name);
1689         return 0;
1690 }
1691
1692 static void dump_branches(void)
1693 {
1694         unsigned int i;
1695         struct branch *b;
1696
1697         for (i = 0; i < branch_table_sz; i++) {
1698                 for (b = branch_table[i]; b; b = b->table_next_branch)
1699                         failure |= update_branch(b);
1700         }
1701 }
1702
1703 static void dump_tags(void)
1704 {
1705         static const char *msg = "fast-import";
1706         struct tag *t;
1707         struct ref_lock *lock;
1708         char ref_name[PATH_MAX];
1709
1710         for (t = first_tag; t; t = t->next_tag) {
1711                 sprintf(ref_name, "tags/%s", t->name);
1712                 lock = lock_ref_sha1(ref_name, NULL);
1713                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1714                         failure |= error("Unable to update %s", ref_name);
1715         }
1716 }
1717
1718 static void dump_marks_helper(FILE *f,
1719         uintmax_t base,
1720         struct mark_set *m)
1721 {
1722         uintmax_t k;
1723         if (m->shift) {
1724                 for (k = 0; k < 1024; k++) {
1725                         if (m->data.sets[k])
1726                                 dump_marks_helper(f, base + (k << m->shift),
1727                                         m->data.sets[k]);
1728                 }
1729         } else {
1730                 for (k = 0; k < 1024; k++) {
1731                         if (m->data.marked[k])
1732                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1733                                         sha1_to_hex(m->data.marked[k]->idx.sha1));
1734                 }
1735         }
1736 }
1737
1738 static void dump_marks(void)
1739 {
1740         static struct lock_file mark_lock;
1741         int mark_fd;
1742         FILE *f;
1743
1744         if (!export_marks_file)
1745                 return;
1746
1747         mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1748         if (mark_fd < 0) {
1749                 failure |= error("Unable to write marks file %s: %s",
1750                         export_marks_file, strerror(errno));
1751                 return;
1752         }
1753
1754         f = fdopen(mark_fd, "w");
1755         if (!f) {
1756                 int saved_errno = errno;
1757                 rollback_lock_file(&mark_lock);
1758                 failure |= error("Unable to write marks file %s: %s",
1759                         export_marks_file, strerror(saved_errno));
1760                 return;
1761         }
1762
1763         /*
1764          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1765          * Assign -1 to the lock file descriptor so that commit_lock_file()
1766          * won't try to close() it.
1767          */
1768         mark_lock.fd = -1;
1769
1770         dump_marks_helper(f, 0, marks);
1771         if (ferror(f) || fclose(f)) {
1772                 int saved_errno = errno;
1773                 rollback_lock_file(&mark_lock);
1774                 failure |= error("Unable to write marks file %s: %s",
1775                         export_marks_file, strerror(saved_errno));
1776                 return;
1777         }
1778
1779         if (commit_lock_file(&mark_lock)) {
1780                 int saved_errno = errno;
1781                 rollback_lock_file(&mark_lock);
1782                 failure |= error("Unable to commit marks file %s: %s",
1783                         export_marks_file, strerror(saved_errno));
1784                 return;
1785         }
1786 }
1787
1788 static void read_marks(void)
1789 {
1790         char line[512];
1791         FILE *f = fopen(import_marks_file, "r");
1792         if (f)
1793                 ;
1794         else if (import_marks_file_ignore_missing && errno == ENOENT)
1795                 return; /* Marks file does not exist */
1796         else
1797                 die_errno("cannot read '%s'", import_marks_file);
1798         while (fgets(line, sizeof(line), f)) {
1799                 uintmax_t mark;
1800                 char *end;
1801                 unsigned char sha1[20];
1802                 struct object_entry *e;
1803
1804                 end = strchr(line, '\n');
1805                 if (line[0] != ':' || !end)
1806                         die("corrupt mark line: %s", line);
1807                 *end = 0;
1808                 mark = strtoumax(line + 1, &end, 10);
1809                 if (!mark || end == line + 1
1810                         || *end != ' ' || get_sha1(end + 1, sha1))
1811                         die("corrupt mark line: %s", line);
1812                 e = find_object(sha1);
1813                 if (!e) {
1814                         enum object_type type = sha1_object_info(sha1, NULL);
1815                         if (type < 0)
1816                                 die("object not found: %s", sha1_to_hex(sha1));
1817                         e = insert_object(sha1);
1818                         e->type = type;
1819                         e->pack_id = MAX_PACK_ID;
1820                         e->idx.offset = 1; /* just not zero! */
1821                 }
1822                 insert_mark(mark, e);
1823         }
1824         fclose(f);
1825 }
1826
1827
1828 static int read_next_command(void)
1829 {
1830         static int stdin_eof = 0;
1831
1832         if (stdin_eof) {
1833                 unread_command_buf = 0;
1834                 return EOF;
1835         }
1836
1837         for (;;) {
1838                 if (unread_command_buf) {
1839                         unread_command_buf = 0;
1840                 } else {
1841                         struct recent_command *rc;
1842
1843                         strbuf_detach(&command_buf, NULL);
1844                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1845                         if (stdin_eof)
1846                                 return EOF;
1847
1848                         if (!seen_data_command
1849                                 && prefixcmp(command_buf.buf, "feature ")
1850                                 && prefixcmp(command_buf.buf, "option ")) {
1851                                 parse_argv();
1852                         }
1853
1854                         rc = rc_free;
1855                         if (rc)
1856                                 rc_free = rc->next;
1857                         else {
1858                                 rc = cmd_hist.next;
1859                                 cmd_hist.next = rc->next;
1860                                 cmd_hist.next->prev = &cmd_hist;
1861                                 free(rc->buf);
1862                         }
1863
1864                         rc->buf = command_buf.buf;
1865                         rc->prev = cmd_tail;
1866                         rc->next = cmd_hist.prev;
1867                         rc->prev->next = rc;
1868                         cmd_tail = rc;
1869                 }
1870                 if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1871                         parse_cat_blob();
1872                         continue;
1873                 }
1874                 if (command_buf.buf[0] == '#')
1875                         continue;
1876                 return 0;
1877         }
1878 }
1879
1880 static void skip_optional_lf(void)
1881 {
1882         int term_char = fgetc(stdin);
1883         if (term_char != '\n' && term_char != EOF)
1884                 ungetc(term_char, stdin);
1885 }
1886
1887 static void parse_mark(void)
1888 {
1889         if (!prefixcmp(command_buf.buf, "mark :")) {
1890                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1891                 read_next_command();
1892         }
1893         else
1894                 next_mark = 0;
1895 }
1896
1897 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1898 {
1899         strbuf_reset(sb);
1900
1901         if (prefixcmp(command_buf.buf, "data "))
1902                 die("Expected 'data n' command, found: %s", command_buf.buf);
1903
1904         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1905                 char *term = xstrdup(command_buf.buf + 5 + 2);
1906                 size_t term_len = command_buf.len - 5 - 2;
1907
1908                 strbuf_detach(&command_buf, NULL);
1909                 for (;;) {
1910                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1911                                 die("EOF in data (terminator '%s' not found)", term);
1912                         if (term_len == command_buf.len
1913                                 && !strcmp(term, command_buf.buf))
1914                                 break;
1915                         strbuf_addbuf(sb, &command_buf);
1916                         strbuf_addch(sb, '\n');
1917                 }
1918                 free(term);
1919         }
1920         else {
1921                 uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1922                 size_t n = 0, length = (size_t)len;
1923
1924                 if (limit && limit < len) {
1925                         *len_res = len;
1926                         return 0;
1927                 }
1928                 if (length < len)
1929                         die("data is too large to use in this context");
1930
1931                 while (n < length) {
1932                         size_t s = strbuf_fread(sb, length - n, stdin);
1933                         if (!s && feof(stdin))
1934                                 die("EOF in data (%lu bytes remaining)",
1935                                         (unsigned long)(length - n));
1936                         n += s;
1937                 }
1938         }
1939
1940         skip_optional_lf();
1941         return 1;
1942 }
1943
1944 static int validate_raw_date(const char *src, char *result, int maxlen)
1945 {
1946         const char *orig_src = src;
1947         char *endp;
1948         unsigned long num;
1949
1950         errno = 0;
1951
1952         num = strtoul(src, &endp, 10);
1953         /* NEEDSWORK: perhaps check for reasonable values? */
1954         if (errno || endp == src || *endp != ' ')
1955                 return -1;
1956
1957         src = endp + 1;
1958         if (*src != '-' && *src != '+')
1959                 return -1;
1960
1961         num = strtoul(src + 1, &endp, 10);
1962         if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1963             1400 < num)
1964                 return -1;
1965
1966         strcpy(result, orig_src);
1967         return 0;
1968 }
1969
1970 static char *parse_ident(const char *buf)
1971 {
1972         const char *gt;
1973         size_t name_len;
1974         char *ident;
1975
1976         gt = strrchr(buf, '>');
1977         if (!gt)
1978                 die("Missing > in ident string: %s", buf);
1979         gt++;
1980         if (*gt != ' ')
1981                 die("Missing space after > in ident string: %s", buf);
1982         gt++;
1983         name_len = gt - buf;
1984         ident = xmalloc(name_len + 24);
1985         strncpy(ident, buf, name_len);
1986
1987         switch (whenspec) {
1988         case WHENSPEC_RAW:
1989                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1990                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1991                 break;
1992         case WHENSPEC_RFC2822:
1993                 if (parse_date(gt, ident + name_len, 24) < 0)
1994                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1995                 break;
1996         case WHENSPEC_NOW:
1997                 if (strcmp("now", gt))
1998                         die("Date in ident must be 'now': %s", buf);
1999                 datestamp(ident + name_len, 24);
2000                 break;
2001         }
2002
2003         return ident;
2004 }
2005
2006 static void parse_and_store_blob(
2007         struct last_object *last,
2008         unsigned char *sha1out,
2009         uintmax_t mark)
2010 {
2011         static struct strbuf buf = STRBUF_INIT;
2012         uintmax_t len;
2013
2014         if (parse_data(&buf, big_file_threshold, &len))
2015                 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2016         else {
2017                 if (last) {
2018                         strbuf_release(&last->data);
2019                         last->offset = 0;
2020                         last->depth = 0;
2021                 }
2022                 stream_blob(len, sha1out, mark);
2023                 skip_optional_lf();
2024         }
2025 }
2026
2027 static void parse_new_blob(void)
2028 {
2029         read_next_command();
2030         parse_mark();
2031         parse_and_store_blob(&last_blob, NULL, next_mark);
2032 }
2033
2034 static void unload_one_branch(void)
2035 {
2036         while (cur_active_branches
2037                 && cur_active_branches >= max_active_branches) {
2038                 uintmax_t min_commit = ULONG_MAX;
2039                 struct branch *e, *l = NULL, *p = NULL;
2040
2041                 for (e = active_branches; e; e = e->active_next_branch) {
2042                         if (e->last_commit < min_commit) {
2043                                 p = l;
2044                                 min_commit = e->last_commit;
2045                         }
2046                         l = e;
2047                 }
2048
2049                 if (p) {
2050                         e = p->active_next_branch;
2051                         p->active_next_branch = e->active_next_branch;
2052                 } else {
2053                         e = active_branches;
2054                         active_branches = e->active_next_branch;
2055                 }
2056                 e->active = 0;
2057                 e->active_next_branch = NULL;
2058                 if (e->branch_tree.tree) {
2059                         release_tree_content_recursive(e->branch_tree.tree);
2060                         e->branch_tree.tree = NULL;
2061                 }
2062                 cur_active_branches--;
2063         }
2064 }
2065
2066 static void load_branch(struct branch *b)
2067 {
2068         load_tree(&b->branch_tree);
2069         if (!b->active) {
2070                 b->active = 1;
2071                 b->active_next_branch = active_branches;
2072                 active_branches = b;
2073                 cur_active_branches++;
2074                 branch_load_count++;
2075         }
2076 }
2077
2078 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2079 {
2080         unsigned char fanout = 0;
2081         while ((num_notes >>= 8))
2082                 fanout++;
2083         return fanout;
2084 }
2085
2086 static void construct_path_with_fanout(const char *hex_sha1,
2087                 unsigned char fanout, char *path)
2088 {
2089         unsigned int i = 0, j = 0;
2090         if (fanout >= 20)
2091                 die("Too large fanout (%u)", fanout);
2092         while (fanout) {
2093                 path[i++] = hex_sha1[j++];
2094                 path[i++] = hex_sha1[j++];
2095                 path[i++] = '/';
2096                 fanout--;
2097         }
2098         memcpy(path + i, hex_sha1 + j, 40 - j);
2099         path[i + 40 - j] = '\0';
2100 }
2101
2102 static uintmax_t do_change_note_fanout(
2103                 struct tree_entry *orig_root, struct tree_entry *root,
2104                 char *hex_sha1, unsigned int hex_sha1_len,
2105                 char *fullpath, unsigned int fullpath_len,
2106                 unsigned char fanout)
2107 {
2108         struct tree_content *t = root->tree;
2109         struct tree_entry *e, leaf;
2110         unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2111         uintmax_t num_notes = 0;
2112         unsigned char sha1[20];
2113         char realpath[60];
2114
2115         for (i = 0; t && i < t->entry_count; i++) {
2116                 e = t->entries[i];
2117                 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2118                 tmp_fullpath_len = fullpath_len;
2119
2120                 /*
2121                  * We're interested in EITHER existing note entries (entries
2122                  * with exactly 40 hex chars in path, not including directory
2123                  * separators), OR directory entries that may contain note
2124                  * entries (with < 40 hex chars in path).
2125                  * Also, each path component in a note entry must be a multiple
2126                  * of 2 chars.
2127                  */
2128                 if (!e->versions[1].mode ||
2129                     tmp_hex_sha1_len > 40 ||
2130                     e->name->str_len % 2)
2131                         continue;
2132
2133                 /* This _may_ be a note entry, or a subdir containing notes */
2134                 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2135                        e->name->str_len);
2136                 if (tmp_fullpath_len)
2137                         fullpath[tmp_fullpath_len++] = '/';
2138                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2139                        e->name->str_len);
2140                 tmp_fullpath_len += e->name->str_len;
2141                 fullpath[tmp_fullpath_len] = '\0';
2142
2143                 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2144                         /* This is a note entry */
2145                         construct_path_with_fanout(hex_sha1, fanout, realpath);
2146                         if (!strcmp(fullpath, realpath)) {
2147                                 /* Note entry is in correct location */
2148                                 num_notes++;
2149                                 continue;
2150                         }
2151
2152                         /* Rename fullpath to realpath */
2153                         if (!tree_content_remove(orig_root, fullpath, &leaf))
2154                                 die("Failed to remove path %s", fullpath);
2155                         tree_content_set(orig_root, realpath,
2156                                 leaf.versions[1].sha1,
2157                                 leaf.versions[1].mode,
2158                                 leaf.tree);
2159                 } else if (S_ISDIR(e->versions[1].mode)) {
2160                         /* This is a subdir that may contain note entries */
2161                         if (!e->tree)
2162                                 load_tree(e);
2163                         num_notes += do_change_note_fanout(orig_root, e,
2164                                 hex_sha1, tmp_hex_sha1_len,
2165                                 fullpath, tmp_fullpath_len, fanout);
2166                 }
2167
2168                 /* The above may have reallocated the current tree_content */
2169                 t = root->tree;
2170         }
2171         return num_notes;
2172 }
2173
2174 static uintmax_t change_note_fanout(struct tree_entry *root,
2175                 unsigned char fanout)
2176 {
2177         char hex_sha1[40], path[60];
2178         return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2179 }
2180
2181 static void file_change_m(struct branch *b)
2182 {
2183         const char *p = command_buf.buf + 2;
2184         static struct strbuf uq = STRBUF_INIT;
2185         const char *endp;
2186         struct object_entry *oe = oe;
2187         unsigned char sha1[20];
2188         uint16_t mode, inline_data = 0;
2189
2190         p = get_mode(p, &mode);
2191         if (!p)
2192                 die("Corrupt mode: %s", command_buf.buf);
2193         switch (mode) {
2194         case 0644:
2195         case 0755:
2196                 mode |= S_IFREG;
2197         case S_IFREG | 0644:
2198         case S_IFREG | 0755:
2199         case S_IFLNK:
2200         case S_IFDIR:
2201         case S_IFGITLINK:
2202                 /* ok */
2203                 break;
2204         default:
2205                 die("Corrupt mode: %s", command_buf.buf);
2206         }
2207
2208         if (*p == ':') {
2209                 char *x;
2210                 oe = find_mark(strtoumax(p + 1, &x, 10));
2211                 hashcpy(sha1, oe->idx.sha1);
2212                 p = x;
2213         } else if (!prefixcmp(p, "inline")) {
2214                 inline_data = 1;
2215                 p += 6;
2216         } else {
2217                 if (get_sha1_hex(p, sha1))
2218                         die("Invalid SHA1: %s", command_buf.buf);
2219                 oe = find_object(sha1);
2220                 p += 40;
2221         }
2222         if (*p++ != ' ')
2223                 die("Missing space after SHA1: %s", command_buf.buf);
2224
2225         strbuf_reset(&uq);
2226         if (!unquote_c_style(&uq, p, &endp)) {
2227                 if (*endp)
2228                         die("Garbage after path in: %s", command_buf.buf);
2229                 p = uq.buf;
2230         }
2231
2232         /* Git does not track empty, non-toplevel directories. */
2233         if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2234                 tree_content_remove(&b->branch_tree, p, NULL);
2235                 return;
2236         }
2237
2238         if (S_ISGITLINK(mode)) {
2239                 if (inline_data)
2240                         die("Git links cannot be specified 'inline': %s",
2241                                 command_buf.buf);
2242                 else if (oe) {
2243                         if (oe->type != OBJ_COMMIT)
2244                                 die("Not a commit (actually a %s): %s",
2245                                         typename(oe->type), command_buf.buf);
2246                 }
2247                 /*
2248                  * Accept the sha1 without checking; it expected to be in
2249                  * another repository.
2250                  */
2251         } else if (inline_data) {
2252                 if (S_ISDIR(mode))
2253                         die("Directories cannot be specified 'inline': %s",
2254                                 command_buf.buf);
2255                 if (p != uq.buf) {
2256                         strbuf_addstr(&uq, p);
2257                         p = uq.buf;
2258                 }
2259                 read_next_command();
2260                 parse_and_store_blob(&last_blob, sha1, 0);
2261         } else {
2262                 enum object_type expected = S_ISDIR(mode) ?
2263                                                 OBJ_TREE: OBJ_BLOB;
2264                 enum object_type type = oe ? oe->type :
2265                                         sha1_object_info(sha1, NULL);
2266                 if (type < 0)
2267                         die("%s not found: %s",
2268                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2269                                         command_buf.buf);
2270                 if (type != expected)
2271                         die("Not a %s (actually a %s): %s",
2272                                 typename(expected), typename(type),
2273                                 command_buf.buf);
2274         }
2275
2276         if (!*p) {
2277                 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2278                 return;
2279         }
2280         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2281 }
2282
2283 static void file_change_d(struct branch *b)
2284 {
2285         const char *p = command_buf.buf + 2;
2286         static struct strbuf uq = STRBUF_INIT;
2287         const char *endp;
2288
2289         strbuf_reset(&uq);
2290         if (!unquote_c_style(&uq, p, &endp)) {
2291                 if (*endp)
2292                         die("Garbage after path in: %s", command_buf.buf);
2293                 p = uq.buf;
2294         }
2295         tree_content_remove(&b->branch_tree, p, NULL);
2296 }
2297
2298 static void file_change_cr(struct branch *b, int rename)
2299 {
2300         const char *s, *d;
2301         static struct strbuf s_uq = STRBUF_INIT;
2302         static struct strbuf d_uq = STRBUF_INIT;
2303         const char *endp;
2304         struct tree_entry leaf;
2305
2306         s = command_buf.buf + 2;
2307         strbuf_reset(&s_uq);
2308         if (!unquote_c_style(&s_uq, s, &endp)) {
2309                 if (*endp != ' ')
2310                         die("Missing space after source: %s", command_buf.buf);
2311         } else {
2312                 endp = strchr(s, ' ');
2313                 if (!endp)
2314                         die("Missing space after source: %s", command_buf.buf);
2315                 strbuf_add(&s_uq, s, endp - s);
2316         }
2317         s = s_uq.buf;
2318
2319         endp++;
2320         if (!*endp)
2321                 die("Missing dest: %s", command_buf.buf);
2322
2323         d = endp;
2324         strbuf_reset(&d_uq);
2325         if (!unquote_c_style(&d_uq, d, &endp)) {
2326                 if (*endp)
2327                         die("Garbage after dest in: %s", command_buf.buf);
2328                 d = d_uq.buf;
2329         }
2330
2331         memset(&leaf, 0, sizeof(leaf));
2332         if (rename)
2333                 tree_content_remove(&b->branch_tree, s, &leaf);
2334         else
2335                 tree_content_get(&b->branch_tree, s, &leaf);
2336         if (!leaf.versions[1].mode)
2337                 die("Path %s not in branch", s);
2338         if (!*d) {      /* C "path/to/subdir" "" */
2339                 tree_content_replace(&b->branch_tree,
2340                         leaf.versions[1].sha1,
2341                         leaf.versions[1].mode,
2342                         leaf.tree);
2343                 return;
2344         }
2345         tree_content_set(&b->branch_tree, d,
2346                 leaf.versions[1].sha1,
2347                 leaf.versions[1].mode,
2348                 leaf.tree);
2349 }
2350
2351 static void note_change_n(struct branch *b, unsigned char old_fanout)
2352 {
2353         const char *p = command_buf.buf + 2;
2354         static struct strbuf uq = STRBUF_INIT;
2355         struct object_entry *oe = oe;
2356         struct branch *s;
2357         unsigned char sha1[20], commit_sha1[20];
2358         char path[60];
2359         uint16_t inline_data = 0;
2360         unsigned char new_fanout;
2361
2362         /* <dataref> or 'inline' */
2363         if (*p == ':') {
2364                 char *x;
2365                 oe = find_mark(strtoumax(p + 1, &x, 10));
2366                 hashcpy(sha1, oe->idx.sha1);
2367                 p = x;
2368         } else if (!prefixcmp(p, "inline")) {
2369                 inline_data = 1;
2370                 p += 6;
2371         } else {
2372                 if (get_sha1_hex(p, sha1))
2373                         die("Invalid SHA1: %s", command_buf.buf);
2374                 oe = find_object(sha1);
2375                 p += 40;
2376         }
2377         if (*p++ != ' ')
2378                 die("Missing space after SHA1: %s", command_buf.buf);
2379
2380         /* <committish> */
2381         s = lookup_branch(p);
2382         if (s) {
2383                 hashcpy(commit_sha1, s->sha1);
2384         } else if (*p == ':') {
2385                 uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2386                 struct object_entry *commit_oe = find_mark(commit_mark);
2387                 if (commit_oe->type != OBJ_COMMIT)
2388                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2389                 hashcpy(commit_sha1, commit_oe->idx.sha1);
2390         } else if (!get_sha1(p, commit_sha1)) {
2391                 unsigned long size;
2392                 char *buf = read_object_with_reference(commit_sha1,
2393                         commit_type, &size, commit_sha1);
2394                 if (!buf || size < 46)
2395                         die("Not a valid commit: %s", p);
2396                 free(buf);
2397         } else
2398                 die("Invalid ref name or SHA1 expression: %s", p);
2399
2400         if (inline_data) {
2401                 if (p != uq.buf) {
2402                         strbuf_addstr(&uq, p);
2403                         p = uq.buf;
2404                 }
2405                 read_next_command();
2406                 parse_and_store_blob(&last_blob, sha1, 0);
2407         } else if (oe) {
2408                 if (oe->type != OBJ_BLOB)
2409                         die("Not a blob (actually a %s): %s",
2410                                 typename(oe->type), command_buf.buf);
2411         } else if (!is_null_sha1(sha1)) {
2412                 enum object_type type = sha1_object_info(sha1, NULL);
2413                 if (type < 0)
2414                         die("Blob not found: %s", command_buf.buf);
2415                 if (type != OBJ_BLOB)
2416                         die("Not a blob (actually a %s): %s",
2417                             typename(type), command_buf.buf);
2418         }
2419
2420         construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
2421         if (tree_content_remove(&b->branch_tree, path, NULL))
2422                 b->num_notes--;
2423
2424         if (is_null_sha1(sha1))
2425                 return; /* nothing to insert */
2426
2427         b->num_notes++;
2428         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2429         construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2430         tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2431 }
2432
2433 static void file_change_deleteall(struct branch *b)
2434 {
2435         release_tree_content_recursive(b->branch_tree.tree);
2436         hashclr(b->branch_tree.versions[0].sha1);
2437         hashclr(b->branch_tree.versions[1].sha1);
2438         load_tree(&b->branch_tree);
2439         b->num_notes = 0;
2440 }
2441
2442 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2443 {
2444         if (!buf || size < 46)
2445                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2446         if (memcmp("tree ", buf, 5)
2447                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2448                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2449         hashcpy(b->branch_tree.versions[0].sha1,
2450                 b->branch_tree.versions[1].sha1);
2451 }
2452
2453 static void parse_from_existing(struct branch *b)
2454 {
2455         if (is_null_sha1(b->sha1)) {
2456                 hashclr(b->branch_tree.versions[0].sha1);
2457                 hashclr(b->branch_tree.versions[1].sha1);
2458         } else {
2459                 unsigned long size;
2460                 char *buf;
2461
2462                 buf = read_object_with_reference(b->sha1,
2463                         commit_type, &size, b->sha1);
2464                 parse_from_commit(b, buf, size);
2465                 free(buf);
2466         }
2467 }
2468
2469 static int parse_from(struct branch *b)
2470 {
2471         const char *from;
2472         struct branch *s;
2473
2474         if (prefixcmp(command_buf.buf, "from "))
2475                 return 0;
2476
2477         if (b->branch_tree.tree) {
2478                 release_tree_content_recursive(b->branch_tree.tree);
2479                 b->branch_tree.tree = NULL;
2480         }
2481
2482         from = strchr(command_buf.buf, ' ') + 1;
2483         s = lookup_branch(from);
2484         if (b == s)
2485                 die("Can't create a branch from itself: %s", b->name);
2486         else if (s) {
2487                 unsigned char *t = s->branch_tree.versions[1].sha1;
2488                 hashcpy(b->sha1, s->sha1);
2489                 hashcpy(b->branch_tree.versions[0].sha1, t);
2490                 hashcpy(b->branch_tree.versions[1].sha1, t);
2491         } else if (*from == ':') {
2492                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2493                 struct object_entry *oe = find_mark(idnum);
2494                 if (oe->type != OBJ_COMMIT)
2495                         die("Mark :%" PRIuMAX " not a commit", idnum);
2496                 hashcpy(b->sha1, oe->idx.sha1);
2497                 if (oe->pack_id != MAX_PACK_ID) {
2498                         unsigned long size;
2499                         char *buf = gfi_unpack_entry(oe, &size);
2500                         parse_from_commit(b, buf, size);
2501                         free(buf);
2502                 } else
2503                         parse_from_existing(b);
2504         } else if (!get_sha1(from, b->sha1))
2505                 parse_from_existing(b);
2506         else
2507                 die("Invalid ref name or SHA1 expression: %s", from);
2508
2509         read_next_command();
2510         return 1;
2511 }
2512
2513 static struct hash_list *parse_merge(unsigned int *count)
2514 {
2515         struct hash_list *list = NULL, *n, *e = e;
2516         const char *from;
2517         struct branch *s;
2518
2519         *count = 0;
2520         while (!prefixcmp(command_buf.buf, "merge ")) {
2521                 from = strchr(command_buf.buf, ' ') + 1;
2522                 n = xmalloc(sizeof(*n));
2523                 s = lookup_branch(from);
2524                 if (s)
2525                         hashcpy(n->sha1, s->sha1);
2526                 else if (*from == ':') {
2527                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2528                         struct object_entry *oe = find_mark(idnum);
2529                         if (oe->type != OBJ_COMMIT)
2530                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2531                         hashcpy(n->sha1, oe->idx.sha1);
2532                 } else if (!get_sha1(from, n->sha1)) {
2533                         unsigned long size;
2534                         char *buf = read_object_with_reference(n->sha1,
2535                                 commit_type, &size, n->sha1);
2536                         if (!buf || size < 46)
2537                                 die("Not a valid commit: %s", from);
2538                         free(buf);
2539                 } else
2540                         die("Invalid ref name or SHA1 expression: %s", from);
2541
2542                 n->next = NULL;
2543                 if (list)
2544                         e->next = n;
2545                 else
2546                         list = n;
2547                 e = n;
2548                 (*count)++;
2549                 read_next_command();
2550         }
2551         return list;
2552 }
2553
2554 static void parse_new_commit(void)
2555 {
2556         static struct strbuf msg = STRBUF_INIT;
2557         struct branch *b;
2558         char *sp;
2559         char *author = NULL;
2560         char *committer = NULL;
2561         struct hash_list *merge_list = NULL;
2562         unsigned int merge_count;
2563         unsigned char prev_fanout, new_fanout;
2564
2565         /* Obtain the branch name from the rest of our command */
2566         sp = strchr(command_buf.buf, ' ') + 1;
2567         b = lookup_branch(sp);
2568         if (!b)
2569                 b = new_branch(sp);
2570
2571         read_next_command();
2572         parse_mark();
2573         if (!prefixcmp(command_buf.buf, "author ")) {
2574                 author = parse_ident(command_buf.buf + 7);
2575                 read_next_command();
2576         }
2577         if (!prefixcmp(command_buf.buf, "committer ")) {
2578                 committer = parse_ident(command_buf.buf + 10);
2579                 read_next_command();
2580         }
2581         if (!committer)
2582                 die("Expected committer but didn't get one");
2583         parse_data(&msg, 0, NULL);
2584         read_next_command();
2585         parse_from(b);
2586         merge_list = parse_merge(&merge_count);
2587
2588         /* ensure the branch is active/loaded */
2589         if (!b->branch_tree.tree || !max_active_branches) {
2590                 unload_one_branch();
2591                 load_branch(b);
2592         }
2593
2594         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2595
2596         /* file_change* */
2597         while (command_buf.len > 0) {
2598                 if (!prefixcmp(command_buf.buf, "M "))
2599                         file_change_m(b);
2600                 else if (!prefixcmp(command_buf.buf, "D "))
2601                         file_change_d(b);
2602                 else if (!prefixcmp(command_buf.buf, "R "))
2603                         file_change_cr(b, 1);
2604                 else if (!prefixcmp(command_buf.buf, "C "))
2605                         file_change_cr(b, 0);
2606                 else if (!prefixcmp(command_buf.buf, "N "))
2607                         note_change_n(b, prev_fanout);
2608                 else if (!strcmp("deleteall", command_buf.buf))
2609                         file_change_deleteall(b);
2610                 else if (!prefixcmp(command_buf.buf, "ls "))
2611                         parse_ls(b);
2612                 else {
2613                         unread_command_buf = 1;
2614                         break;
2615                 }
2616                 if (read_next_command() == EOF)
2617                         break;
2618         }
2619
2620         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2621         if (new_fanout != prev_fanout)
2622                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2623
2624         /* build the tree and the commit */
2625         store_tree(&b->branch_tree);
2626         hashcpy(b->branch_tree.versions[0].sha1,
2627                 b->branch_tree.versions[1].sha1);
2628
2629         strbuf_reset(&new_data);
2630         strbuf_addf(&new_data, "tree %s\n",
2631                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2632         if (!is_null_sha1(b->sha1))
2633                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2634         while (merge_list) {
2635                 struct hash_list *next = merge_list->next;
2636                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2637                 free(merge_list);
2638                 merge_list = next;
2639         }
2640         strbuf_addf(&new_data,
2641                 "author %s\n"
2642                 "committer %s\n"
2643                 "\n",
2644                 author ? author : committer, committer);
2645         strbuf_addbuf(&new_data, &msg);
2646         free(author);
2647         free(committer);
2648
2649         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2650                 b->pack_id = pack_id;
2651         b->last_commit = object_count_by_type[OBJ_COMMIT];
2652 }
2653
2654 static void parse_new_tag(void)
2655 {
2656         static struct strbuf msg = STRBUF_INIT;
2657         char *sp;
2658         const char *from;
2659         char *tagger;
2660         struct branch *s;
2661         struct tag *t;
2662         uintmax_t from_mark = 0;
2663         unsigned char sha1[20];
2664         enum object_type type;
2665
2666         /* Obtain the new tag name from the rest of our command */
2667         sp = strchr(command_buf.buf, ' ') + 1;
2668         t = pool_alloc(sizeof(struct tag));
2669         t->next_tag = NULL;
2670         t->name = pool_strdup(sp);
2671         if (last_tag)
2672                 last_tag->next_tag = t;
2673         else
2674                 first_tag = t;
2675         last_tag = t;
2676         read_next_command();
2677
2678         /* from ... */
2679         if (prefixcmp(command_buf.buf, "from "))
2680                 die("Expected from command, got %s", command_buf.buf);
2681         from = strchr(command_buf.buf, ' ') + 1;
2682         s = lookup_branch(from);
2683         if (s) {
2684                 hashcpy(sha1, s->sha1);
2685                 type = OBJ_COMMIT;
2686         } else if (*from == ':') {
2687                 struct object_entry *oe;
2688                 from_mark = strtoumax(from + 1, NULL, 10);
2689                 oe = find_mark(from_mark);
2690                 type = oe->type;
2691                 hashcpy(sha1, oe->idx.sha1);
2692         } else if (!get_sha1(from, sha1)) {
2693                 unsigned long size;
2694                 char *buf;
2695
2696                 buf = read_sha1_file(sha1, &type, &size);
2697                 if (!buf || size < 46)
2698                         die("Not a valid commit: %s", from);
2699                 free(buf);
2700         } else
2701                 die("Invalid ref name or SHA1 expression: %s", from);
2702         read_next_command();
2703
2704         /* tagger ... */
2705         if (!prefixcmp(command_buf.buf, "tagger ")) {
2706                 tagger = parse_ident(command_buf.buf + 7);
2707                 read_next_command();
2708         } else
2709                 tagger = NULL;
2710
2711         /* tag payload/message */
2712         parse_data(&msg, 0, NULL);
2713
2714         /* build the tag object */
2715         strbuf_reset(&new_data);
2716
2717         strbuf_addf(&new_data,
2718                     "object %s\n"
2719                     "type %s\n"
2720                     "tag %s\n",
2721                     sha1_to_hex(sha1), typename(type), t->name);
2722         if (tagger)
2723                 strbuf_addf(&new_data,
2724                             "tagger %s\n", tagger);
2725         strbuf_addch(&new_data, '\n');
2726         strbuf_addbuf(&new_data, &msg);
2727         free(tagger);
2728
2729         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2730                 t->pack_id = MAX_PACK_ID;
2731         else
2732                 t->pack_id = pack_id;
2733 }
2734
2735 static void parse_reset_branch(void)
2736 {
2737         struct branch *b;
2738         char *sp;
2739
2740         /* Obtain the branch name from the rest of our command */
2741         sp = strchr(command_buf.buf, ' ') + 1;
2742         b = lookup_branch(sp);
2743         if (b) {
2744                 hashclr(b->sha1);
2745                 hashclr(b->branch_tree.versions[0].sha1);
2746                 hashclr(b->branch_tree.versions[1].sha1);
2747                 if (b->branch_tree.tree) {
2748                         release_tree_content_recursive(b->branch_tree.tree);
2749                         b->branch_tree.tree = NULL;
2750                 }
2751         }
2752         else
2753                 b = new_branch(sp);
2754         read_next_command();
2755         parse_from(b);
2756         if (command_buf.len > 0)
2757                 unread_command_buf = 1;
2758 }
2759
2760 static void cat_blob_write(const char *buf, unsigned long size)
2761 {
2762         if (write_in_full(cat_blob_fd, buf, size) != size)
2763                 die_errno("Write to frontend failed");
2764 }
2765
2766 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2767 {
2768         struct strbuf line = STRBUF_INIT;
2769         unsigned long size;
2770         enum object_type type = 0;
2771         char *buf;
2772
2773         if (!oe || oe->pack_id == MAX_PACK_ID) {
2774                 buf = read_sha1_file(sha1, &type, &size);
2775         } else {
2776                 type = oe->type;
2777                 buf = gfi_unpack_entry(oe, &size);
2778         }
2779
2780         /*
2781          * Output based on batch_one_object() from cat-file.c.
2782          */
2783         if (type <= 0) {
2784                 strbuf_reset(&line);
2785                 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2786                 cat_blob_write(line.buf, line.len);
2787                 strbuf_release(&line);
2788                 free(buf);
2789                 return;
2790         }
2791         if (!buf)
2792                 die("Can't read object %s", sha1_to_hex(sha1));
2793         if (type != OBJ_BLOB)
2794                 die("Object %s is a %s but a blob was expected.",
2795                     sha1_to_hex(sha1), typename(type));
2796         strbuf_reset(&line);
2797         strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2798                                                 typename(type), size);
2799         cat_blob_write(line.buf, line.len);
2800         strbuf_release(&line);
2801         cat_blob_write(buf, size);
2802         cat_blob_write("\n", 1);
2803         free(buf);
2804 }
2805
2806 static void parse_cat_blob(void)
2807 {
2808         const char *p;
2809         struct object_entry *oe = oe;
2810         unsigned char sha1[20];
2811
2812         /* cat-blob SP <object> LF */
2813         p = command_buf.buf + strlen("cat-blob ");
2814         if (*p == ':') {
2815                 char *x;
2816                 oe = find_mark(strtoumax(p + 1, &x, 10));
2817                 if (x == p + 1)
2818                         die("Invalid mark: %s", command_buf.buf);
2819                 if (!oe)
2820                         die("Unknown mark: %s", command_buf.buf);
2821                 if (*x)
2822                         die("Garbage after mark: %s", command_buf.buf);
2823                 hashcpy(sha1, oe->idx.sha1);
2824         } else {
2825                 if (get_sha1_hex(p, sha1))
2826                         die("Invalid SHA1: %s", command_buf.buf);
2827                 if (p[40])
2828                         die("Garbage after SHA1: %s", command_buf.buf);
2829                 oe = find_object(sha1);
2830         }
2831
2832         cat_blob(oe, sha1);
2833 }
2834
2835 static struct object_entry *dereference(struct object_entry *oe,
2836                                         unsigned char sha1[20])
2837 {
2838         unsigned long size;
2839         char *buf = NULL;
2840         if (!oe) {
2841                 enum object_type type = sha1_object_info(sha1, NULL);
2842                 if (type < 0)
2843                         die("object not found: %s", sha1_to_hex(sha1));
2844                 /* cache it! */
2845                 oe = insert_object(sha1);
2846                 oe->type = type;
2847                 oe->pack_id = MAX_PACK_ID;
2848                 oe->idx.offset = 1;
2849         }
2850         switch (oe->type) {
2851         case OBJ_TREE:  /* easy case. */
2852                 return oe;
2853         case OBJ_COMMIT:
2854         case OBJ_TAG:
2855                 break;
2856         default:
2857                 die("Not a treeish: %s", command_buf.buf);
2858         }
2859
2860         if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
2861                 buf = gfi_unpack_entry(oe, &size);
2862         } else {
2863                 enum object_type unused;
2864                 buf = read_sha1_file(sha1, &unused, &size);
2865         }
2866         if (!buf)
2867                 die("Can't load object %s", sha1_to_hex(sha1));
2868
2869         /* Peel one layer. */
2870         switch (oe->type) {
2871         case OBJ_TAG:
2872                 if (size < 40 + strlen("object ") ||
2873                     get_sha1_hex(buf + strlen("object "), sha1))
2874                         die("Invalid SHA1 in tag: %s", command_buf.buf);
2875                 break;
2876         case OBJ_COMMIT:
2877                 if (size < 40 + strlen("tree ") ||
2878                     get_sha1_hex(buf + strlen("tree "), sha1))
2879                         die("Invalid SHA1 in commit: %s", command_buf.buf);
2880         }
2881
2882         free(buf);
2883         return find_object(sha1);
2884 }
2885
2886 static struct object_entry *parse_treeish_dataref(const char **p)
2887 {
2888         unsigned char sha1[20];
2889         struct object_entry *e;
2890
2891         if (**p == ':') {       /* <mark> */
2892                 char *endptr;
2893                 e = find_mark(strtoumax(*p + 1, &endptr, 10));
2894                 if (endptr == *p + 1)
2895                         die("Invalid mark: %s", command_buf.buf);
2896                 if (!e)
2897                         die("Unknown mark: %s", command_buf.buf);
2898                 *p = endptr;
2899                 hashcpy(sha1, e->idx.sha1);
2900         } else {        /* <sha1> */
2901                 if (get_sha1_hex(*p, sha1))
2902                         die("Invalid SHA1: %s", command_buf.buf);
2903                 e = find_object(sha1);
2904                 *p += 40;
2905         }
2906
2907         while (!e || e->type != OBJ_TREE)
2908                 e = dereference(e, sha1);
2909         return e;
2910 }
2911
2912 static void print_ls(int mode, const unsigned char *sha1, const char *path)
2913 {
2914         static struct strbuf line = STRBUF_INIT;
2915
2916         /* See show_tree(). */
2917         const char *type =
2918                 S_ISGITLINK(mode) ? commit_type :
2919                 S_ISDIR(mode) ? tree_type :
2920                 blob_type;
2921
2922         if (!mode) {
2923                 /* missing SP path LF */
2924                 strbuf_reset(&line);
2925                 strbuf_addstr(&line, "missing ");
2926                 quote_c_style(path, &line, NULL, 0);
2927                 strbuf_addch(&line, '\n');
2928         } else {
2929                 /* mode SP type SP object_name TAB path LF */
2930                 strbuf_reset(&line);
2931                 strbuf_addf(&line, "%06o %s %s\t",
2932                                 mode, type, sha1_to_hex(sha1));
2933                 quote_c_style(path, &line, NULL, 0);
2934                 strbuf_addch(&line, '\n');
2935         }
2936         cat_blob_write(line.buf, line.len);
2937 }
2938
2939 static void parse_ls(struct branch *b)
2940 {
2941         const char *p;
2942         struct tree_entry *root = NULL;
2943         struct tree_entry leaf = {NULL};
2944
2945         /* ls SP (<treeish> SP)? <path> */
2946         p = command_buf.buf + strlen("ls ");
2947         if (*p == '"') {
2948                 if (!b)
2949                         die("Not in a commit: %s", command_buf.buf);
2950                 root = &b->branch_tree;
2951         } else {
2952                 struct object_entry *e = parse_treeish_dataref(&p);
2953                 root = new_tree_entry();
2954                 hashcpy(root->versions[1].sha1, e->idx.sha1);
2955                 load_tree(root);
2956                 if (*p++ != ' ')
2957                         die("Missing space after tree-ish: %s", command_buf.buf);
2958         }
2959         if (*p == '"') {
2960                 static struct strbuf uq = STRBUF_INIT;
2961                 const char *endp;
2962                 strbuf_reset(&uq);
2963                 if (unquote_c_style(&uq, p, &endp))
2964                         die("Invalid path: %s", command_buf.buf);
2965                 if (*endp)
2966                         die("Garbage after path in: %s", command_buf.buf);
2967                 p = uq.buf;
2968         }
2969         tree_content_get(root, p, &leaf);
2970         /*
2971          * A directory in preparation would have a sha1 of zero
2972          * until it is saved.  Save, for simplicity.
2973          */
2974         if (S_ISDIR(leaf.versions[1].mode))
2975                 store_tree(&leaf);
2976
2977         print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
2978         if (!b || root != &b->branch_tree)
2979                 release_tree_entry(root);
2980 }
2981
2982 static void checkpoint(void)
2983 {
2984         checkpoint_requested = 0;
2985         if (object_count) {
2986                 cycle_packfile();
2987                 dump_branches();
2988                 dump_tags();
2989                 dump_marks();
2990         }
2991 }
2992
2993 static void parse_checkpoint(void)
2994 {
2995         checkpoint_requested = 1;
2996         skip_optional_lf();
2997 }
2998
2999 static void parse_progress(void)
3000 {
3001         fwrite(command_buf.buf, 1, command_buf.len, stdout);
3002         fputc('\n', stdout);
3003         fflush(stdout);
3004         skip_optional_lf();
3005 }
3006
3007 static char* make_fast_import_path(const char *path)
3008 {
3009         struct strbuf abs_path = STRBUF_INIT;
3010
3011         if (!relative_marks_paths || is_absolute_path(path))
3012                 return xstrdup(path);
3013         strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
3014         return strbuf_detach(&abs_path, NULL);
3015 }
3016
3017 static void option_import_marks(const char *marks,
3018                                         int from_stream, int ignore_missing)
3019 {
3020         if (import_marks_file) {
3021                 if (from_stream)
3022                         die("Only one import-marks command allowed per stream");
3023
3024                 /* read previous mark file */
3025                 if(!import_marks_file_from_stream)
3026                         read_marks();
3027         }
3028
3029         import_marks_file = make_fast_import_path(marks);
3030         safe_create_leading_directories_const(import_marks_file);
3031         import_marks_file_from_stream = from_stream;
3032         import_marks_file_ignore_missing = ignore_missing;
3033 }
3034
3035 static void option_date_format(const char *fmt)
3036 {
3037         if (!strcmp(fmt, "raw"))
3038                 whenspec = WHENSPEC_RAW;
3039         else if (!strcmp(fmt, "rfc2822"))
3040                 whenspec = WHENSPEC_RFC2822;
3041         else if (!strcmp(fmt, "now"))
3042                 whenspec = WHENSPEC_NOW;
3043         else
3044                 die("unknown --date-format argument %s", fmt);
3045 }
3046
3047 static unsigned long ulong_arg(const char *option, const char *arg)
3048 {
3049         char *endptr;
3050         unsigned long rv = strtoul(arg, &endptr, 0);
3051         if (strchr(arg, '-') || endptr == arg || *endptr)
3052                 die("%s: argument must be a non-negative integer", option);
3053         return rv;
3054 }
3055
3056 static void option_depth(const char *depth)
3057 {
3058         max_depth = ulong_arg("--depth", depth);
3059         if (max_depth > MAX_DEPTH)
3060                 die("--depth cannot exceed %u", MAX_DEPTH);
3061 }
3062
3063 static void option_active_branches(const char *branches)
3064 {
3065         max_active_branches = ulong_arg("--active-branches", branches);
3066 }
3067
3068 static void option_export_marks(const char *marks)
3069 {
3070         export_marks_file = make_fast_import_path(marks);
3071         safe_create_leading_directories_const(export_marks_file);
3072 }
3073
3074 static void option_cat_blob_fd(const char *fd)
3075 {
3076         unsigned long n = ulong_arg("--cat-blob-fd", fd);
3077         if (n > (unsigned long) INT_MAX)
3078                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3079         cat_blob_fd = (int) n;
3080 }
3081
3082 static void option_export_pack_edges(const char *edges)
3083 {
3084         if (pack_edges)
3085                 fclose(pack_edges);
3086         pack_edges = fopen(edges, "a");
3087         if (!pack_edges)
3088                 die_errno("Cannot open '%s'", edges);
3089 }
3090
3091 static int parse_one_option(const char *option)
3092 {
3093         if (!prefixcmp(option, "max-pack-size=")) {
3094                 unsigned long v;
3095                 if (!git_parse_ulong(option + 14, &v))
3096                         return 0;
3097                 if (v < 8192) {
3098                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3099                         v *= 1024 * 1024;
3100                 } else if (v < 1024 * 1024) {
3101                         warning("minimum max-pack-size is 1 MiB");
3102                         v = 1024 * 1024;
3103                 }
3104                 max_packsize = v;
3105         } else if (!prefixcmp(option, "big-file-threshold=")) {
3106                 unsigned long v;
3107                 if (!git_parse_ulong(option + 19, &v))
3108                         return 0;
3109                 big_file_threshold = v;
3110         } else if (!prefixcmp(option, "depth=")) {
3111                 option_depth(option + 6);
3112         } else if (!prefixcmp(option, "active-branches=")) {
3113                 option_active_branches(option + 16);
3114         } else if (!prefixcmp(option, "export-pack-edges=")) {
3115                 option_export_pack_edges(option + 18);
3116         } else if (!prefixcmp(option, "quiet")) {
3117                 show_stats = 0;
3118         } else if (!prefixcmp(option, "stats")) {
3119                 show_stats = 1;
3120         } else {
3121                 return 0;
3122         }
3123
3124         return 1;
3125 }
3126
3127 static int parse_one_feature(const char *feature, int from_stream)
3128 {
3129         if (!prefixcmp(feature, "date-format=")) {
3130                 option_date_format(feature + 12);
3131         } else if (!prefixcmp(feature, "import-marks=")) {
3132                 option_import_marks(feature + 13, from_stream, 0);
3133         } else if (!prefixcmp(feature, "import-marks-if-exists=")) {
3134                 option_import_marks(feature + strlen("import-marks-if-exists="),
3135                                         from_stream, 1);
3136         } else if (!prefixcmp(feature, "export-marks=")) {
3137                 option_export_marks(feature + 13);
3138         } else if (!strcmp(feature, "cat-blob")) {
3139                 ; /* Don't die - this feature is supported */
3140         } else if (!strcmp(feature, "relative-marks")) {
3141                 relative_marks_paths = 1;
3142         } else if (!strcmp(feature, "no-relative-marks")) {
3143                 relative_marks_paths = 0;
3144         } else if (!strcmp(feature, "done")) {
3145                 require_explicit_termination = 1;
3146         } else if (!strcmp(feature, "force")) {
3147                 force_update = 1;
3148         } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3149                 ; /* do nothing; we have the feature */
3150         } else {
3151                 return 0;
3152         }
3153
3154         return 1;
3155 }
3156
3157 static void parse_feature(void)
3158 {
3159         char *feature = command_buf.buf + 8;
3160
3161         if (seen_data_command)
3162                 die("Got feature command '%s' after data command", feature);
3163
3164         if (parse_one_feature(feature, 1))
3165                 return;
3166
3167         die("This version of fast-import does not support feature %s.", feature);
3168 }
3169
3170 static void parse_option(void)
3171 {
3172         char *option = command_buf.buf + 11;
3173
3174         if (seen_data_command)
3175                 die("Got option command '%s' after data command", option);
3176
3177         if (parse_one_option(option))
3178                 return;
3179
3180         die("This version of fast-import does not support option: %s", option);
3181 }
3182
3183 static int git_pack_config(const char *k, const char *v, void *cb)
3184 {
3185         if (!strcmp(k, "pack.depth")) {
3186                 max_depth = git_config_int(k, v);
3187                 if (max_depth > MAX_DEPTH)
3188                         max_depth = MAX_DEPTH;
3189                 return 0;
3190         }
3191         if (!strcmp(k, "pack.compression")) {
3192                 int level = git_config_int(k, v);
3193                 if (level == -1)
3194                         level = Z_DEFAULT_COMPRESSION;
3195                 else if (level < 0 || level > Z_BEST_COMPRESSION)
3196                         die("bad pack compression level %d", level);
3197                 pack_compression_level = level;
3198                 pack_compression_seen = 1;
3199                 return 0;
3200         }
3201         if (!strcmp(k, "pack.indexversion")) {
3202                 pack_idx_opts.version = git_config_int(k, v);
3203                 if (pack_idx_opts.version > 2)
3204                         die("bad pack.indexversion=%"PRIu32,
3205                             pack_idx_opts.version);
3206                 return 0;
3207         }
3208         if (!strcmp(k, "pack.packsizelimit")) {
3209                 max_packsize = git_config_ulong(k, v);
3210                 return 0;
3211         }
3212         return git_default_config(k, v, cb);
3213 }
3214
3215 static const char fast_import_usage[] =
3216 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3217
3218 static void parse_argv(void)
3219 {
3220         unsigned int i;
3221
3222         for (i = 1; i < global_argc; i++) {
3223                 const char *a = global_argv[i];
3224
3225                 if (*a != '-' || !strcmp(a, "--"))
3226                         break;
3227
3228                 if (parse_one_option(a + 2))
3229                         continue;
3230
3231                 if (parse_one_feature(a + 2, 0))
3232                         continue;
3233
3234                 if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3235                         option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3236                         continue;
3237                 }
3238
3239                 die("unknown option %s", a);
3240         }
3241         if (i != global_argc)
3242                 usage(fast_import_usage);
3243
3244         seen_data_command = 1;
3245         if (import_marks_file)
3246                 read_marks();
3247 }
3248
3249 int main(int argc, const char **argv)
3250 {
3251         unsigned int i;
3252
3253         git_extract_argv0_path(argv[0]);
3254
3255         if (argc == 2 && !strcmp(argv[1], "-h"))
3256                 usage(fast_import_usage);
3257
3258         setup_git_directory();
3259         reset_pack_idx_option(&pack_idx_opts);
3260         git_config(git_pack_config, NULL);
3261         if (!pack_compression_seen && core_compression_seen)
3262                 pack_compression_level = core_compression_level;
3263
3264         alloc_objects(object_entry_alloc);
3265         strbuf_init(&command_buf, 0);
3266         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3267         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3268         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3269         marks = pool_calloc(1, sizeof(struct mark_set));
3270
3271         global_argc = argc;
3272         global_argv = argv;
3273
3274         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3275         for (i = 0; i < (cmd_save - 1); i++)
3276                 rc_free[i].next = &rc_free[i + 1];
3277         rc_free[cmd_save - 1].next = NULL;
3278
3279         prepare_packed_git();
3280         start_packfile();
3281         set_die_routine(die_nicely);
3282         set_checkpoint_signal();
3283         while (read_next_command() != EOF) {
3284                 if (!strcmp("blob", command_buf.buf))
3285                         parse_new_blob();
3286                 else if (!prefixcmp(command_buf.buf, "ls "))
3287                         parse_ls(NULL);
3288                 else if (!prefixcmp(command_buf.buf, "commit "))
3289                         parse_new_commit();
3290                 else if (!prefixcmp(command_buf.buf, "tag "))
3291                         parse_new_tag();
3292                 else if (!prefixcmp(command_buf.buf, "reset "))
3293                         parse_reset_branch();
3294                 else if (!strcmp("checkpoint", command_buf.buf))
3295                         parse_checkpoint();
3296                 else if (!strcmp("done", command_buf.buf))
3297                         break;
3298                 else if (!prefixcmp(command_buf.buf, "progress "))
3299                         parse_progress();
3300                 else if (!prefixcmp(command_buf.buf, "feature "))
3301                         parse_feature();
3302                 else if (!prefixcmp(command_buf.buf, "option git "))
3303                         parse_option();
3304                 else if (!prefixcmp(command_buf.buf, "option "))
3305                         /* ignore non-git options*/;
3306                 else
3307                         die("Unsupported command: %s", command_buf.buf);
3308
3309                 if (checkpoint_requested)
3310                         checkpoint();
3311         }
3312
3313         /* argv hasn't been parsed yet, do so */
3314         if (!seen_data_command)
3315                 parse_argv();
3316
3317         if (require_explicit_termination && feof(stdin))
3318                 die("stream ends early");
3319
3320         end_packfile();
3321
3322         dump_branches();
3323         dump_tags();
3324         unkeep_all_packs();
3325         dump_marks();
3326
3327         if (pack_edges)
3328                 fclose(pack_edges);
3329
3330         if (show_stats) {
3331                 uintmax_t total_count = 0, duplicate_count = 0;
3332                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3333                         total_count += object_count_by_type[i];
3334                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3335                         duplicate_count += duplicate_count_by_type[i];
3336
3337                 fprintf(stderr, "%s statistics:\n", argv[0]);
3338                 fprintf(stderr, "---------------------------------------------------------------------\n");
3339                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3340                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3341                 fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
3342                 fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
3343                 fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
3344                 fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
3345                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3346                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3347                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3348                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3349                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3350                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3351                 fprintf(stderr, "---------------------------------------------------------------------\n");
3352                 pack_report();
3353                 fprintf(stderr, "---------------------------------------------------------------------\n");
3354                 fprintf(stderr, "\n");
3355         }
3356
3357         return failure ? 1 : 0;
3358 }