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