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