]> asedeno.scripts.mit.edu Git - git.git/blob - fast-import.c
Merge branch 'master' of git://repo.or.cz/git/fastimport
[git.git] / fast-import.c
1 /*
2 Format of STDIN stream:
3
4   stream ::= cmd*;
5
6   cmd ::= new_blob
7         | new_commit
8         | new_tag
9         | reset_branch
10         | checkpoint
11         ;
12
13   new_blob ::= 'blob' lf
14         mark?
15     file_content;
16   file_content ::= data;
17
18   new_commit ::= 'commit' sp ref_str lf
19     mark?
20     ('author' sp name '<' email '>' when lf)?
21     'committer' sp name '<' email '>' when lf
22     commit_msg
23     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
24     ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
25     file_change*
26     lf;
27   commit_msg ::= data;
28
29   file_change ::= file_clr | file_del | file_obm | file_inm;
30   file_clr ::= 'deleteall' lf;
31   file_del ::= 'D' sp path_str lf;
32   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
33   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
34     data;
35
36   new_tag ::= 'tag' sp tag_str lf
37     'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
38         'tagger' sp name '<' email '>' when lf
39     tag_msg;
40   tag_msg ::= data;
41
42   reset_branch ::= 'reset' sp ref_str lf
43     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
44     lf;
45
46   checkpoint ::= 'checkpoint' lf
47     lf;
48
49      # note: the first idnum in a stream should be 1 and subsequent
50      # idnums should not have gaps between values as this will cause
51      # the stream parser to reserve space for the gapped values.  An
52          # idnum can be updated in the future to a new object by issuing
53      # a new mark directive with the old idnum.
54          #
55   mark ::= 'mark' sp idnum lf;
56   data ::= (delimited_data | exact_data)
57     lf;
58
59     # note: delim may be any string but must not contain lf.
60     # data_line may contain any data but must not be exactly
61     # delim.
62   delimited_data ::= 'data' sp '<<' delim lf
63     (data_line lf)*
64         delim lf;
65
66      # note: declen indicates the length of binary_data in bytes.
67      # declen does not include the lf preceeding the binary data.
68      #
69   exact_data ::= 'data' sp declen lf
70     binary_data;
71
72      # note: quoted strings are C-style quoting supporting \c for
73      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
74          # is the signed byte value in octal.  Note that the only
75      # characters which must actually be escaped to protect the
76      # stream formatting is: \, " and LF.  Otherwise these values
77          # are UTF8.
78      #
79   ref_str     ::= ref;
80   sha1exp_str ::= sha1exp;
81   tag_str     ::= tag;
82   path_str    ::= path    | '"' quoted(path)    '"' ;
83   mode        ::= '100644' | '644'
84                 | '100755' | '755'
85                 | '120000'
86                 ;
87
88   declen ::= # unsigned 32 bit value, ascii base10 notation;
89   bigint ::= # unsigned integer value, ascii base10 notation;
90   binary_data ::= # file content, not interpreted;
91
92   when         ::= raw_when | rfc2822_when;
93   raw_when     ::= ts sp tz;
94   rfc2822_when ::= # Valid RFC 2822 date and time;
95
96   sp ::= # ASCII space character;
97   lf ::= # ASCII newline (LF) character;
98
99      # note: a colon (':') must precede the numerical value assigned to
100          # an idnum.  This is to distinguish it from a ref or tag name as
101      # GIT does not permit ':' in ref or tag strings.
102          #
103   idnum   ::= ':' bigint;
104   path    ::= # GIT style file path, e.g. "a/b/c";
105   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
106   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
107   sha1exp ::= # Any valid GIT SHA1 expression;
108   hexsha1 ::= # SHA1 in hexadecimal format;
109
110      # note: name and email are UTF8 strings, however name must not
111          # contain '<' or lf and email must not contain any of the
112      # following: '<', '>', lf.
113          #
114   name  ::= # valid GIT author/committer name;
115   email ::= # valid GIT author/committer email;
116   ts    ::= # time since the epoch in seconds, ascii base10 notation;
117   tz    ::= # GIT style timezone;
118 */
119
120 #include "builtin.h"
121 #include "cache.h"
122 #include "object.h"
123 #include "blob.h"
124 #include "tree.h"
125 #include "commit.h"
126 #include "delta.h"
127 #include "pack.h"
128 #include "refs.h"
129 #include "csum-file.h"
130 #include "strbuf.h"
131 #include "quote.h"
132
133 #define PACK_ID_BITS 16
134 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
135
136 struct object_entry
137 {
138         struct object_entry *next;
139         uint32_t offset;
140         unsigned type : TYPE_BITS;
141         unsigned pack_id : PACK_ID_BITS;
142         unsigned char sha1[20];
143 };
144
145 struct object_entry_pool
146 {
147         struct object_entry_pool *next_pool;
148         struct object_entry *next_free;
149         struct object_entry *end;
150         struct object_entry entries[FLEX_ARRAY]; /* more */
151 };
152
153 struct mark_set
154 {
155         union {
156                 struct object_entry *marked[1024];
157                 struct mark_set *sets[1024];
158         } data;
159         unsigned int shift;
160 };
161
162 struct last_object
163 {
164         void *data;
165         unsigned long len;
166         uint32_t offset;
167         unsigned int depth;
168         unsigned no_free:1;
169 };
170
171 struct mem_pool
172 {
173         struct mem_pool *next_pool;
174         char *next_free;
175         char *end;
176         char space[FLEX_ARRAY]; /* more */
177 };
178
179 struct atom_str
180 {
181         struct atom_str *next_atom;
182         unsigned short str_len;
183         char str_dat[FLEX_ARRAY]; /* more */
184 };
185
186 struct tree_content;
187 struct tree_entry
188 {
189         struct tree_content *tree;
190         struct atom_str* name;
191         struct tree_entry_ms
192         {
193                 uint16_t mode;
194                 unsigned char sha1[20];
195         } versions[2];
196 };
197
198 struct tree_content
199 {
200         unsigned int entry_capacity; /* must match avail_tree_content */
201         unsigned int entry_count;
202         unsigned int delta_depth;
203         struct tree_entry *entries[FLEX_ARRAY]; /* more */
204 };
205
206 struct avail_tree_content
207 {
208         unsigned int entry_capacity; /* must match tree_content */
209         struct avail_tree_content *next_avail;
210 };
211
212 struct branch
213 {
214         struct branch *table_next_branch;
215         struct branch *active_next_branch;
216         const char *name;
217         struct tree_entry branch_tree;
218         uintmax_t last_commit;
219         unsigned int pack_id;
220         unsigned char sha1[20];
221 };
222
223 struct tag
224 {
225         struct tag *next_tag;
226         const char *name;
227         unsigned int pack_id;
228         unsigned char sha1[20];
229 };
230
231 struct dbuf
232 {
233         void *buffer;
234         size_t capacity;
235 };
236
237 struct hash_list
238 {
239         struct hash_list *next;
240         unsigned char sha1[20];
241 };
242
243 typedef enum {
244         WHENSPEC_RAW = 1,
245         WHENSPEC_RFC2822,
246         WHENSPEC_NOW,
247 } whenspec_type;
248
249 /* Configured limits on output */
250 static unsigned long max_depth = 10;
251 static unsigned long max_packsize = (1LL << 32) - 1;
252 static int force_update;
253
254 /* Stats and misc. counters */
255 static uintmax_t alloc_count;
256 static uintmax_t marks_set_count;
257 static uintmax_t object_count_by_type[1 << TYPE_BITS];
258 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
259 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
260 static unsigned long object_count;
261 static unsigned long branch_count;
262 static unsigned long branch_load_count;
263 static int failure;
264
265 /* Memory pools */
266 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
267 static size_t total_allocd;
268 static struct mem_pool *mem_pool;
269
270 /* Atom management */
271 static unsigned int atom_table_sz = 4451;
272 static unsigned int atom_cnt;
273 static struct atom_str **atom_table;
274
275 /* The .pack file being generated */
276 static unsigned int pack_id;
277 static struct packed_git *pack_data;
278 static struct packed_git **all_packs;
279 static unsigned long pack_size;
280
281 /* Table of objects we've written. */
282 static unsigned int object_entry_alloc = 5000;
283 static struct object_entry_pool *blocks;
284 static struct object_entry *object_table[1 << 16];
285 static struct mark_set *marks;
286 static const char* mark_file;
287
288 /* Our last blob */
289 static struct last_object last_blob;
290
291 /* Tree management */
292 static unsigned int tree_entry_alloc = 1000;
293 static void *avail_tree_entry;
294 static unsigned int avail_tree_table_sz = 100;
295 static struct avail_tree_content **avail_tree_table;
296 static struct dbuf old_tree;
297 static struct dbuf new_tree;
298
299 /* Branch data */
300 static unsigned long max_active_branches = 5;
301 static unsigned long cur_active_branches;
302 static unsigned long branch_table_sz = 1039;
303 static struct branch **branch_table;
304 static struct branch *active_branches;
305
306 /* Tag data */
307 static struct tag *first_tag;
308 static struct tag *last_tag;
309
310 /* Input stream parsing */
311 static whenspec_type whenspec = WHENSPEC_RAW;
312 static struct strbuf command_buf;
313 static uintmax_t next_mark;
314 static struct dbuf new_data;
315
316
317 static void alloc_objects(unsigned int cnt)
318 {
319         struct object_entry_pool *b;
320
321         b = xmalloc(sizeof(struct object_entry_pool)
322                 + cnt * sizeof(struct object_entry));
323         b->next_pool = blocks;
324         b->next_free = b->entries;
325         b->end = b->entries + cnt;
326         blocks = b;
327         alloc_count += cnt;
328 }
329
330 static struct object_entry *new_object(unsigned char *sha1)
331 {
332         struct object_entry *e;
333
334         if (blocks->next_free == blocks->end)
335                 alloc_objects(object_entry_alloc);
336
337         e = blocks->next_free++;
338         hashcpy(e->sha1, sha1);
339         return e;
340 }
341
342 static struct object_entry *find_object(unsigned char *sha1)
343 {
344         unsigned int h = sha1[0] << 8 | sha1[1];
345         struct object_entry *e;
346         for (e = object_table[h]; e; e = e->next)
347                 if (!hashcmp(sha1, e->sha1))
348                         return e;
349         return NULL;
350 }
351
352 static struct object_entry *insert_object(unsigned char *sha1)
353 {
354         unsigned int h = sha1[0] << 8 | sha1[1];
355         struct object_entry *e = object_table[h];
356         struct object_entry *p = NULL;
357
358         while (e) {
359                 if (!hashcmp(sha1, e->sha1))
360                         return e;
361                 p = e;
362                 e = e->next;
363         }
364
365         e = new_object(sha1);
366         e->next = NULL;
367         e->offset = 0;
368         if (p)
369                 p->next = e;
370         else
371                 object_table[h] = e;
372         return e;
373 }
374
375 static unsigned int hc_str(const char *s, size_t len)
376 {
377         unsigned int r = 0;
378         while (len-- > 0)
379                 r = r * 31 + *s++;
380         return r;
381 }
382
383 static void *pool_alloc(size_t len)
384 {
385         struct mem_pool *p;
386         void *r;
387
388         for (p = mem_pool; p; p = p->next_pool)
389                 if ((p->end - p->next_free >= len))
390                         break;
391
392         if (!p) {
393                 if (len >= (mem_pool_alloc/2)) {
394                         total_allocd += len;
395                         return xmalloc(len);
396                 }
397                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
398                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
399                 p->next_pool = mem_pool;
400                 p->next_free = p->space;
401                 p->end = p->next_free + mem_pool_alloc;
402                 mem_pool = p;
403         }
404
405         r = p->next_free;
406         /* round out to a pointer alignment */
407         if (len & (sizeof(void*) - 1))
408                 len += sizeof(void*) - (len & (sizeof(void*) - 1));
409         p->next_free += len;
410         return r;
411 }
412
413 static void *pool_calloc(size_t count, size_t size)
414 {
415         size_t len = count * size;
416         void *r = pool_alloc(len);
417         memset(r, 0, len);
418         return r;
419 }
420
421 static char *pool_strdup(const char *s)
422 {
423         char *r = pool_alloc(strlen(s) + 1);
424         strcpy(r, s);
425         return r;
426 }
427
428 static void size_dbuf(struct dbuf *b, size_t maxlen)
429 {
430         if (b->buffer) {
431                 if (b->capacity >= maxlen)
432                         return;
433                 free(b->buffer);
434         }
435         b->capacity = ((maxlen / 1024) + 1) * 1024;
436         b->buffer = xmalloc(b->capacity);
437 }
438
439 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
440 {
441         struct mark_set *s = marks;
442         while ((idnum >> s->shift) >= 1024) {
443                 s = pool_calloc(1, sizeof(struct mark_set));
444                 s->shift = marks->shift + 10;
445                 s->data.sets[0] = marks;
446                 marks = s;
447         }
448         while (s->shift) {
449                 uintmax_t i = idnum >> s->shift;
450                 idnum -= i << s->shift;
451                 if (!s->data.sets[i]) {
452                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
453                         s->data.sets[i]->shift = s->shift - 10;
454                 }
455                 s = s->data.sets[i];
456         }
457         if (!s->data.marked[idnum])
458                 marks_set_count++;
459         s->data.marked[idnum] = oe;
460 }
461
462 static struct object_entry *find_mark(uintmax_t idnum)
463 {
464         uintmax_t orig_idnum = idnum;
465         struct mark_set *s = marks;
466         struct object_entry *oe = NULL;
467         if ((idnum >> s->shift) < 1024) {
468                 while (s && s->shift) {
469                         uintmax_t i = idnum >> s->shift;
470                         idnum -= i << s->shift;
471                         s = s->data.sets[i];
472                 }
473                 if (s)
474                         oe = s->data.marked[idnum];
475         }
476         if (!oe)
477                 die("mark :%ju not declared", orig_idnum);
478         return oe;
479 }
480
481 static struct atom_str *to_atom(const char *s, unsigned short len)
482 {
483         unsigned int hc = hc_str(s, len) % atom_table_sz;
484         struct atom_str *c;
485
486         for (c = atom_table[hc]; c; c = c->next_atom)
487                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
488                         return c;
489
490         c = pool_alloc(sizeof(struct atom_str) + len + 1);
491         c->str_len = len;
492         strncpy(c->str_dat, s, len);
493         c->str_dat[len] = 0;
494         c->next_atom = atom_table[hc];
495         atom_table[hc] = c;
496         atom_cnt++;
497         return c;
498 }
499
500 static struct branch *lookup_branch(const char *name)
501 {
502         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
503         struct branch *b;
504
505         for (b = branch_table[hc]; b; b = b->table_next_branch)
506                 if (!strcmp(name, b->name))
507                         return b;
508         return NULL;
509 }
510
511 static struct branch *new_branch(const char *name)
512 {
513         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
514         struct branch* b = lookup_branch(name);
515
516         if (b)
517                 die("Invalid attempt to create duplicate branch: %s", name);
518         if (check_ref_format(name))
519                 die("Branch name doesn't conform to GIT standards: %s", name);
520
521         b = pool_calloc(1, sizeof(struct branch));
522         b->name = pool_strdup(name);
523         b->table_next_branch = branch_table[hc];
524         b->branch_tree.versions[0].mode = S_IFDIR;
525         b->branch_tree.versions[1].mode = S_IFDIR;
526         b->pack_id = MAX_PACK_ID;
527         branch_table[hc] = b;
528         branch_count++;
529         return b;
530 }
531
532 static unsigned int hc_entries(unsigned int cnt)
533 {
534         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
535         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
536 }
537
538 static struct tree_content *new_tree_content(unsigned int cnt)
539 {
540         struct avail_tree_content *f, *l = NULL;
541         struct tree_content *t;
542         unsigned int hc = hc_entries(cnt);
543
544         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
545                 if (f->entry_capacity >= cnt)
546                         break;
547
548         if (f) {
549                 if (l)
550                         l->next_avail = f->next_avail;
551                 else
552                         avail_tree_table[hc] = f->next_avail;
553         } else {
554                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
555                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
556                 f->entry_capacity = cnt;
557         }
558
559         t = (struct tree_content*)f;
560         t->entry_count = 0;
561         t->delta_depth = 0;
562         return t;
563 }
564
565 static void release_tree_entry(struct tree_entry *e);
566 static void release_tree_content(struct tree_content *t)
567 {
568         struct avail_tree_content *f = (struct avail_tree_content*)t;
569         unsigned int hc = hc_entries(f->entry_capacity);
570         f->next_avail = avail_tree_table[hc];
571         avail_tree_table[hc] = f;
572 }
573
574 static void release_tree_content_recursive(struct tree_content *t)
575 {
576         unsigned int i;
577         for (i = 0; i < t->entry_count; i++)
578                 release_tree_entry(t->entries[i]);
579         release_tree_content(t);
580 }
581
582 static struct tree_content *grow_tree_content(
583         struct tree_content *t,
584         int amt)
585 {
586         struct tree_content *r = new_tree_content(t->entry_count + amt);
587         r->entry_count = t->entry_count;
588         r->delta_depth = t->delta_depth;
589         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
590         release_tree_content(t);
591         return r;
592 }
593
594 static struct tree_entry *new_tree_entry(void)
595 {
596         struct tree_entry *e;
597
598         if (!avail_tree_entry) {
599                 unsigned int n = tree_entry_alloc;
600                 total_allocd += n * sizeof(struct tree_entry);
601                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
602                 while (n-- > 1) {
603                         *((void**)e) = e + 1;
604                         e++;
605                 }
606                 *((void**)e) = NULL;
607         }
608
609         e = avail_tree_entry;
610         avail_tree_entry = *((void**)e);
611         return e;
612 }
613
614 static void release_tree_entry(struct tree_entry *e)
615 {
616         if (e->tree)
617                 release_tree_content_recursive(e->tree);
618         *((void**)e) = avail_tree_entry;
619         avail_tree_entry = e;
620 }
621
622 static void start_packfile(void)
623 {
624         static char tmpfile[PATH_MAX];
625         struct packed_git *p;
626         struct pack_header hdr;
627         int pack_fd;
628
629         snprintf(tmpfile, sizeof(tmpfile),
630                 "%s/pack_XXXXXX", get_object_directory());
631         pack_fd = mkstemp(tmpfile);
632         if (pack_fd < 0)
633                 die("Can't create %s: %s", tmpfile, strerror(errno));
634         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
635         strcpy(p->pack_name, tmpfile);
636         p->pack_fd = pack_fd;
637
638         hdr.hdr_signature = htonl(PACK_SIGNATURE);
639         hdr.hdr_version = htonl(2);
640         hdr.hdr_entries = 0;
641         write_or_die(p->pack_fd, &hdr, sizeof(hdr));
642
643         pack_data = p;
644         pack_size = sizeof(hdr);
645         object_count = 0;
646
647         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
648         all_packs[pack_id] = p;
649 }
650
651 static void fixup_header_footer(void)
652 {
653         static const int buf_sz = 128 * 1024;
654         int pack_fd = pack_data->pack_fd;
655         SHA_CTX c;
656         struct pack_header hdr;
657         char *buf;
658
659         if (lseek(pack_fd, 0, SEEK_SET) != 0)
660                 die("Failed seeking to start: %s", strerror(errno));
661         if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
662                 die("Unable to reread header of %s", pack_data->pack_name);
663         if (lseek(pack_fd, 0, SEEK_SET) != 0)
664                 die("Failed seeking to start: %s", strerror(errno));
665         hdr.hdr_entries = htonl(object_count);
666         write_or_die(pack_fd, &hdr, sizeof(hdr));
667
668         SHA1_Init(&c);
669         SHA1_Update(&c, &hdr, sizeof(hdr));
670
671         buf = xmalloc(buf_sz);
672         for (;;) {
673                 size_t n = xread(pack_fd, buf, buf_sz);
674                 if (!n)
675                         break;
676                 if (n < 0)
677                         die("Failed to checksum %s", pack_data->pack_name);
678                 SHA1_Update(&c, buf, n);
679         }
680         free(buf);
681
682         SHA1_Final(pack_data->sha1, &c);
683         write_or_die(pack_fd, pack_data->sha1, sizeof(pack_data->sha1));
684         close(pack_fd);
685 }
686
687 static int oecmp (const void *a_, const void *b_)
688 {
689         struct object_entry *a = *((struct object_entry**)a_);
690         struct object_entry *b = *((struct object_entry**)b_);
691         return hashcmp(a->sha1, b->sha1);
692 }
693
694 static char *create_index(void)
695 {
696         static char tmpfile[PATH_MAX];
697         SHA_CTX ctx;
698         struct sha1file *f;
699         struct object_entry **idx, **c, **last, *e;
700         struct object_entry_pool *o;
701         uint32_t array[256];
702         int i, idx_fd;
703
704         /* Build the sorted table of object IDs. */
705         idx = xmalloc(object_count * sizeof(struct object_entry*));
706         c = idx;
707         for (o = blocks; o; o = o->next_pool)
708                 for (e = o->next_free; e-- != o->entries;)
709                         if (pack_id == e->pack_id)
710                                 *c++ = e;
711         last = idx + object_count;
712         if (c != last)
713                 die("internal consistency error creating the index");
714         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
715
716         /* Generate the fan-out array. */
717         c = idx;
718         for (i = 0; i < 256; i++) {
719                 struct object_entry **next = c;;
720                 while (next < last) {
721                         if ((*next)->sha1[0] != i)
722                                 break;
723                         next++;
724                 }
725                 array[i] = htonl(next - idx);
726                 c = next;
727         }
728
729         snprintf(tmpfile, sizeof(tmpfile),
730                 "%s/index_XXXXXX", get_object_directory());
731         idx_fd = mkstemp(tmpfile);
732         if (idx_fd < 0)
733                 die("Can't create %s: %s", tmpfile, strerror(errno));
734         f = sha1fd(idx_fd, tmpfile);
735         sha1write(f, array, 256 * sizeof(int));
736         SHA1_Init(&ctx);
737         for (c = idx; c != last; c++) {
738                 uint32_t offset = htonl((*c)->offset);
739                 sha1write(f, &offset, 4);
740                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
741                 SHA1_Update(&ctx, (*c)->sha1, 20);
742         }
743         sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
744         sha1close(f, NULL, 1);
745         free(idx);
746         SHA1_Final(pack_data->sha1, &ctx);
747         return tmpfile;
748 }
749
750 static char *keep_pack(char *curr_index_name)
751 {
752         static char name[PATH_MAX];
753         static char *keep_msg = "fast-import";
754         int keep_fd;
755
756         chmod(pack_data->pack_name, 0444);
757         chmod(curr_index_name, 0444);
758
759         snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
760                  get_object_directory(), sha1_to_hex(pack_data->sha1));
761         keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
762         if (keep_fd < 0)
763                 die("cannot create keep file");
764         write(keep_fd, keep_msg, strlen(keep_msg));
765         close(keep_fd);
766
767         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
768                  get_object_directory(), sha1_to_hex(pack_data->sha1));
769         if (move_temp_to_file(pack_data->pack_name, name))
770                 die("cannot store pack file");
771
772         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
773                  get_object_directory(), sha1_to_hex(pack_data->sha1));
774         if (move_temp_to_file(curr_index_name, name))
775                 die("cannot store index file");
776         return name;
777 }
778
779 static void unkeep_all_packs(void)
780 {
781         static char name[PATH_MAX];
782         int k;
783
784         for (k = 0; k < pack_id; k++) {
785                 struct packed_git *p = all_packs[k];
786                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
787                          get_object_directory(), sha1_to_hex(p->sha1));
788                 unlink(name);
789         }
790 }
791
792 static void end_packfile(void)
793 {
794         struct packed_git *old_p = pack_data, *new_p;
795
796         if (object_count) {
797                 char *idx_name;
798                 int i;
799                 struct branch *b;
800                 struct tag *t;
801
802                 fixup_header_footer();
803                 idx_name = keep_pack(create_index());
804
805                 /* Register the packfile with core git's machinary. */
806                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
807                 if (!new_p)
808                         die("core git rejected index %s", idx_name);
809                 new_p->windows = old_p->windows;
810                 all_packs[pack_id] = new_p;
811                 install_packed_git(new_p);
812
813                 /* Print the boundary */
814                 fprintf(stdout, "%s:", new_p->pack_name);
815                 for (i = 0; i < branch_table_sz; i++) {
816                         for (b = branch_table[i]; b; b = b->table_next_branch) {
817                                 if (b->pack_id == pack_id)
818                                         fprintf(stdout, " %s", sha1_to_hex(b->sha1));
819                         }
820                 }
821                 for (t = first_tag; t; t = t->next_tag) {
822                         if (t->pack_id == pack_id)
823                                 fprintf(stdout, " %s", sha1_to_hex(t->sha1));
824                 }
825                 fputc('\n', stdout);
826
827                 pack_id++;
828         }
829         else
830                 unlink(old_p->pack_name);
831         free(old_p);
832
833         /* We can't carry a delta across packfiles. */
834         free(last_blob.data);
835         last_blob.data = NULL;
836         last_blob.len = 0;
837         last_blob.offset = 0;
838         last_blob.depth = 0;
839 }
840
841 static void cycle_packfile(void)
842 {
843         end_packfile();
844         start_packfile();
845 }
846
847 static size_t encode_header(
848         enum object_type type,
849         size_t size,
850         unsigned char *hdr)
851 {
852         int n = 1;
853         unsigned char c;
854
855         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
856                 die("bad type %d", type);
857
858         c = (type << 4) | (size & 15);
859         size >>= 4;
860         while (size) {
861                 *hdr++ = c | 0x80;
862                 c = size & 0x7f;
863                 size >>= 7;
864                 n++;
865         }
866         *hdr = c;
867         return n;
868 }
869
870 static int store_object(
871         enum object_type type,
872         void *dat,
873         size_t datlen,
874         struct last_object *last,
875         unsigned char *sha1out,
876         uintmax_t mark)
877 {
878         void *out, *delta;
879         struct object_entry *e;
880         unsigned char hdr[96];
881         unsigned char sha1[20];
882         unsigned long hdrlen, deltalen;
883         SHA_CTX c;
884         z_stream s;
885
886         hdrlen = sprintf((char*)hdr,"%s %lu",type_names[type],datlen) + 1;
887         SHA1_Init(&c);
888         SHA1_Update(&c, hdr, hdrlen);
889         SHA1_Update(&c, dat, datlen);
890         SHA1_Final(sha1, &c);
891         if (sha1out)
892                 hashcpy(sha1out, sha1);
893
894         e = insert_object(sha1);
895         if (mark)
896                 insert_mark(mark, e);
897         if (e->offset) {
898                 duplicate_count_by_type[type]++;
899                 return 1;
900         }
901
902         if (last && last->data && last->depth < max_depth) {
903                 delta = diff_delta(last->data, last->len,
904                         dat, datlen,
905                         &deltalen, 0);
906                 if (delta && deltalen >= datlen) {
907                         free(delta);
908                         delta = NULL;
909                 }
910         } else
911                 delta = NULL;
912
913         memset(&s, 0, sizeof(s));
914         deflateInit(&s, zlib_compression_level);
915         if (delta) {
916                 s.next_in = delta;
917                 s.avail_in = deltalen;
918         } else {
919                 s.next_in = dat;
920                 s.avail_in = datlen;
921         }
922         s.avail_out = deflateBound(&s, s.avail_in);
923         s.next_out = out = xmalloc(s.avail_out);
924         while (deflate(&s, Z_FINISH) == Z_OK)
925                 /* nothing */;
926         deflateEnd(&s);
927
928         /* Determine if we should auto-checkpoint. */
929         if ((pack_size + 60 + s.total_out) > max_packsize
930                 || (pack_size + 60 + s.total_out) < pack_size) {
931
932                 /* This new object needs to *not* have the current pack_id. */
933                 e->pack_id = pack_id + 1;
934                 cycle_packfile();
935
936                 /* We cannot carry a delta into the new pack. */
937                 if (delta) {
938                         free(delta);
939                         delta = NULL;
940
941                         memset(&s, 0, sizeof(s));
942                         deflateInit(&s, zlib_compression_level);
943                         s.next_in = dat;
944                         s.avail_in = datlen;
945                         s.avail_out = deflateBound(&s, s.avail_in);
946                         s.next_out = out = xrealloc(out, s.avail_out);
947                         while (deflate(&s, Z_FINISH) == Z_OK)
948                                 /* nothing */;
949                         deflateEnd(&s);
950                 }
951         }
952
953         e->type = type;
954         e->pack_id = pack_id;
955         e->offset = pack_size;
956         object_count++;
957         object_count_by_type[type]++;
958
959         if (delta) {
960                 unsigned long ofs = e->offset - last->offset;
961                 unsigned pos = sizeof(hdr) - 1;
962
963                 delta_count_by_type[type]++;
964                 last->depth++;
965
966                 hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
967                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
968                 pack_size += hdrlen;
969
970                 hdr[pos] = ofs & 127;
971                 while (ofs >>= 7)
972                         hdr[--pos] = 128 | (--ofs & 127);
973                 write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
974                 pack_size += sizeof(hdr) - pos;
975         } else {
976                 if (last)
977                         last->depth = 0;
978                 hdrlen = encode_header(type, datlen, hdr);
979                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
980                 pack_size += hdrlen;
981         }
982
983         write_or_die(pack_data->pack_fd, out, s.total_out);
984         pack_size += s.total_out;
985
986         free(out);
987         free(delta);
988         if (last) {
989                 if (!last->no_free)
990                         free(last->data);
991                 last->data = dat;
992                 last->offset = e->offset;
993                 last->len = datlen;
994         }
995         return 0;
996 }
997
998 static void *gfi_unpack_entry(
999         struct object_entry *oe,
1000         unsigned long *sizep)
1001 {
1002         static char type[20];
1003         struct packed_git *p = all_packs[oe->pack_id];
1004         if (p == pack_data)
1005                 p->pack_size = pack_size + 20;
1006         return unpack_entry(p, oe->offset, type, sizep);
1007 }
1008
1009 static const char *get_mode(const char *str, uint16_t *modep)
1010 {
1011         unsigned char c;
1012         uint16_t mode = 0;
1013
1014         while ((c = *str++) != ' ') {
1015                 if (c < '0' || c > '7')
1016                         return NULL;
1017                 mode = (mode << 3) + (c - '0');
1018         }
1019         *modep = mode;
1020         return str;
1021 }
1022
1023 static void load_tree(struct tree_entry *root)
1024 {
1025         unsigned char* sha1 = root->versions[1].sha1;
1026         struct object_entry *myoe;
1027         struct tree_content *t;
1028         unsigned long size;
1029         char *buf;
1030         const char *c;
1031
1032         root->tree = t = new_tree_content(8);
1033         if (is_null_sha1(sha1))
1034                 return;
1035
1036         myoe = find_object(sha1);
1037         if (myoe) {
1038                 if (myoe->type != OBJ_TREE)
1039                         die("Not a tree: %s", sha1_to_hex(sha1));
1040                 t->delta_depth = 0;
1041                 buf = gfi_unpack_entry(myoe, &size);
1042         } else {
1043                 char type[20];
1044                 buf = read_sha1_file(sha1, type, &size);
1045                 if (!buf || strcmp(type, tree_type))
1046                         die("Can't load tree %s", sha1_to_hex(sha1));
1047         }
1048
1049         c = buf;
1050         while (c != (buf + size)) {
1051                 struct tree_entry *e = new_tree_entry();
1052
1053                 if (t->entry_count == t->entry_capacity)
1054                         root->tree = t = grow_tree_content(t, 8);
1055                 t->entries[t->entry_count++] = e;
1056
1057                 e->tree = NULL;
1058                 c = get_mode(c, &e->versions[1].mode);
1059                 if (!c)
1060                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1061                 e->versions[0].mode = e->versions[1].mode;
1062                 e->name = to_atom(c, (unsigned short)strlen(c));
1063                 c += e->name->str_len + 1;
1064                 hashcpy(e->versions[0].sha1, (unsigned char*)c);
1065                 hashcpy(e->versions[1].sha1, (unsigned char*)c);
1066                 c += 20;
1067         }
1068         free(buf);
1069 }
1070
1071 static int tecmp0 (const void *_a, const void *_b)
1072 {
1073         struct tree_entry *a = *((struct tree_entry**)_a);
1074         struct tree_entry *b = *((struct tree_entry**)_b);
1075         return base_name_compare(
1076                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1077                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1078 }
1079
1080 static int tecmp1 (const void *_a, const void *_b)
1081 {
1082         struct tree_entry *a = *((struct tree_entry**)_a);
1083         struct tree_entry *b = *((struct tree_entry**)_b);
1084         return base_name_compare(
1085                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1086                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1087 }
1088
1089 static void mktree(struct tree_content *t,
1090         int v,
1091         unsigned long *szp,
1092         struct dbuf *b)
1093 {
1094         size_t maxlen = 0;
1095         unsigned int i;
1096         char *c;
1097
1098         if (!v)
1099                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1100         else
1101                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1102
1103         for (i = 0; i < t->entry_count; i++) {
1104                 if (t->entries[i]->versions[v].mode)
1105                         maxlen += t->entries[i]->name->str_len + 34;
1106         }
1107
1108         size_dbuf(b, maxlen);
1109         c = b->buffer;
1110         for (i = 0; i < t->entry_count; i++) {
1111                 struct tree_entry *e = t->entries[i];
1112                 if (!e->versions[v].mode)
1113                         continue;
1114                 c += sprintf(c, "%o", (unsigned int)e->versions[v].mode);
1115                 *c++ = ' ';
1116                 strcpy(c, e->name->str_dat);
1117                 c += e->name->str_len + 1;
1118                 hashcpy((unsigned char*)c, e->versions[v].sha1);
1119                 c += 20;
1120         }
1121         *szp = c - (char*)b->buffer;
1122 }
1123
1124 static void store_tree(struct tree_entry *root)
1125 {
1126         struct tree_content *t = root->tree;
1127         unsigned int i, j, del;
1128         unsigned long new_len;
1129         struct last_object lo;
1130         struct object_entry *le;
1131
1132         if (!is_null_sha1(root->versions[1].sha1))
1133                 return;
1134
1135         for (i = 0; i < t->entry_count; i++) {
1136                 if (t->entries[i]->tree)
1137                         store_tree(t->entries[i]);
1138         }
1139
1140         le = find_object(root->versions[0].sha1);
1141         if (!S_ISDIR(root->versions[0].mode)
1142                 || !le
1143                 || le->pack_id != pack_id) {
1144                 lo.data = NULL;
1145                 lo.depth = 0;
1146         } else {
1147                 mktree(t, 0, &lo.len, &old_tree);
1148                 lo.data = old_tree.buffer;
1149                 lo.offset = le->offset;
1150                 lo.depth = t->delta_depth;
1151                 lo.no_free = 1;
1152         }
1153
1154         mktree(t, 1, &new_len, &new_tree);
1155         store_object(OBJ_TREE, new_tree.buffer, new_len,
1156                 &lo, root->versions[1].sha1, 0);
1157
1158         t->delta_depth = lo.depth;
1159         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1160                 struct tree_entry *e = t->entries[i];
1161                 if (e->versions[1].mode) {
1162                         e->versions[0].mode = e->versions[1].mode;
1163                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1164                         t->entries[j++] = e;
1165                 } else {
1166                         release_tree_entry(e);
1167                         del++;
1168                 }
1169         }
1170         t->entry_count -= del;
1171 }
1172
1173 static int tree_content_set(
1174         struct tree_entry *root,
1175         const char *p,
1176         const unsigned char *sha1,
1177         const uint16_t mode)
1178 {
1179         struct tree_content *t = root->tree;
1180         const char *slash1;
1181         unsigned int i, n;
1182         struct tree_entry *e;
1183
1184         slash1 = strchr(p, '/');
1185         if (slash1)
1186                 n = slash1 - p;
1187         else
1188                 n = strlen(p);
1189
1190         for (i = 0; i < t->entry_count; i++) {
1191                 e = t->entries[i];
1192                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1193                         if (!slash1) {
1194                                 if (e->versions[1].mode == mode
1195                                                 && !hashcmp(e->versions[1].sha1, sha1))
1196                                         return 0;
1197                                 e->versions[1].mode = mode;
1198                                 hashcpy(e->versions[1].sha1, sha1);
1199                                 if (e->tree) {
1200                                         release_tree_content_recursive(e->tree);
1201                                         e->tree = NULL;
1202                                 }
1203                                 hashclr(root->versions[1].sha1);
1204                                 return 1;
1205                         }
1206                         if (!S_ISDIR(e->versions[1].mode)) {
1207                                 e->tree = new_tree_content(8);
1208                                 e->versions[1].mode = S_IFDIR;
1209                         }
1210                         if (!e->tree)
1211                                 load_tree(e);
1212                         if (tree_content_set(e, slash1 + 1, sha1, mode)) {
1213                                 hashclr(root->versions[1].sha1);
1214                                 return 1;
1215                         }
1216                         return 0;
1217                 }
1218         }
1219
1220         if (t->entry_count == t->entry_capacity)
1221                 root->tree = t = grow_tree_content(t, 8);
1222         e = new_tree_entry();
1223         e->name = to_atom(p, (unsigned short)n);
1224         e->versions[0].mode = 0;
1225         hashclr(e->versions[0].sha1);
1226         t->entries[t->entry_count++] = e;
1227         if (slash1) {
1228                 e->tree = new_tree_content(8);
1229                 e->versions[1].mode = S_IFDIR;
1230                 tree_content_set(e, slash1 + 1, sha1, mode);
1231         } else {
1232                 e->tree = NULL;
1233                 e->versions[1].mode = mode;
1234                 hashcpy(e->versions[1].sha1, sha1);
1235         }
1236         hashclr(root->versions[1].sha1);
1237         return 1;
1238 }
1239
1240 static int tree_content_remove(struct tree_entry *root, const char *p)
1241 {
1242         struct tree_content *t = root->tree;
1243         const char *slash1;
1244         unsigned int i, n;
1245         struct tree_entry *e;
1246
1247         slash1 = strchr(p, '/');
1248         if (slash1)
1249                 n = slash1 - p;
1250         else
1251                 n = strlen(p);
1252
1253         for (i = 0; i < t->entry_count; i++) {
1254                 e = t->entries[i];
1255                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1256                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1257                                 goto del_entry;
1258                         if (!e->tree)
1259                                 load_tree(e);
1260                         if (tree_content_remove(e, slash1 + 1)) {
1261                                 for (n = 0; n < e->tree->entry_count; n++) {
1262                                         if (e->tree->entries[n]->versions[1].mode) {
1263                                                 hashclr(root->versions[1].sha1);
1264                                                 return 1;
1265                                         }
1266                                 }
1267                                 goto del_entry;
1268                         }
1269                         return 0;
1270                 }
1271         }
1272         return 0;
1273
1274 del_entry:
1275         if (e->tree) {
1276                 release_tree_content_recursive(e->tree);
1277                 e->tree = NULL;
1278         }
1279         e->versions[1].mode = 0;
1280         hashclr(e->versions[1].sha1);
1281         hashclr(root->versions[1].sha1);
1282         return 1;
1283 }
1284
1285 static int update_branch(struct branch *b)
1286 {
1287         static const char *msg = "fast-import";
1288         struct ref_lock *lock;
1289         unsigned char old_sha1[20];
1290
1291         if (read_ref(b->name, old_sha1))
1292                 hashclr(old_sha1);
1293         lock = lock_any_ref_for_update(b->name, old_sha1);
1294         if (!lock)
1295                 return error("Unable to lock %s", b->name);
1296         if (!force_update && !is_null_sha1(old_sha1)) {
1297                 struct commit *old_cmit, *new_cmit;
1298
1299                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1300                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1301                 if (!old_cmit || !new_cmit) {
1302                         unlock_ref(lock);
1303                         return error("Branch %s is missing commits.", b->name);
1304                 }
1305
1306                 if (!in_merge_bases(old_cmit, new_cmit)) {
1307                         unlock_ref(lock);
1308                         warn("Not updating %s"
1309                                 " (new tip %s does not contain %s)",
1310                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1311                         return -1;
1312                 }
1313         }
1314         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1315                 return error("Unable to update %s", b->name);
1316         return 0;
1317 }
1318
1319 static void dump_branches(void)
1320 {
1321         unsigned int i;
1322         struct branch *b;
1323
1324         for (i = 0; i < branch_table_sz; i++) {
1325                 for (b = branch_table[i]; b; b = b->table_next_branch)
1326                         failure |= update_branch(b);
1327         }
1328 }
1329
1330 static void dump_tags(void)
1331 {
1332         static const char *msg = "fast-import";
1333         struct tag *t;
1334         struct ref_lock *lock;
1335         char ref_name[PATH_MAX];
1336
1337         for (t = first_tag; t; t = t->next_tag) {
1338                 sprintf(ref_name, "tags/%s", t->name);
1339                 lock = lock_ref_sha1(ref_name, NULL);
1340                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1341                         failure |= error("Unable to update %s", ref_name);
1342         }
1343 }
1344
1345 static void dump_marks_helper(FILE *f,
1346         uintmax_t base,
1347         struct mark_set *m)
1348 {
1349         uintmax_t k;
1350         if (m->shift) {
1351                 for (k = 0; k < 1024; k++) {
1352                         if (m->data.sets[k])
1353                                 dump_marks_helper(f, (base + k) << m->shift,
1354                                         m->data.sets[k]);
1355                 }
1356         } else {
1357                 for (k = 0; k < 1024; k++) {
1358                         if (m->data.marked[k])
1359                                 fprintf(f, ":%ju %s\n", base + k,
1360                                         sha1_to_hex(m->data.marked[k]->sha1));
1361                 }
1362         }
1363 }
1364
1365 static void dump_marks(void)
1366 {
1367         if (mark_file)
1368         {
1369                 FILE *f = fopen(mark_file, "w");
1370                 if (f) {
1371                         dump_marks_helper(f, 0, marks);
1372                         fclose(f);
1373                 } else
1374                         failure |= error("Unable to write marks file %s: %s",
1375                                 mark_file, strerror(errno));
1376         }
1377 }
1378
1379 static void read_next_command(void)
1380 {
1381         read_line(&command_buf, stdin, '\n');
1382 }
1383
1384 static void cmd_mark(void)
1385 {
1386         if (!strncmp("mark :", command_buf.buf, 6)) {
1387                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1388                 read_next_command();
1389         }
1390         else
1391                 next_mark = 0;
1392 }
1393
1394 static void *cmd_data (size_t *size)
1395 {
1396         size_t length;
1397         char *buffer;
1398
1399         if (strncmp("data ", command_buf.buf, 5))
1400                 die("Expected 'data n' command, found: %s", command_buf.buf);
1401
1402         if (!strncmp("<<", command_buf.buf + 5, 2)) {
1403                 char *term = xstrdup(command_buf.buf + 5 + 2);
1404                 size_t sz = 8192, term_len = command_buf.len - 5 - 2;
1405                 length = 0;
1406                 buffer = xmalloc(sz);
1407                 for (;;) {
1408                         read_next_command();
1409                         if (command_buf.eof)
1410                                 die("EOF in data (terminator '%s' not found)", term);
1411                         if (term_len == command_buf.len
1412                                 && !strcmp(term, command_buf.buf))
1413                                 break;
1414                         if (sz < (length + command_buf.len)) {
1415                                 sz = sz * 3 / 2 + 16;
1416                                 if (sz < (length + command_buf.len))
1417                                         sz = length + command_buf.len;
1418                                 buffer = xrealloc(buffer, sz);
1419                         }
1420                         memcpy(buffer + length,
1421                                 command_buf.buf,
1422                                 command_buf.len - 1);
1423                         length += command_buf.len - 1;
1424                         buffer[length++] = '\n';
1425                 }
1426                 free(term);
1427         }
1428         else {
1429                 size_t n = 0;
1430                 length = strtoul(command_buf.buf + 5, NULL, 10);
1431                 buffer = xmalloc(length);
1432                 while (n < length) {
1433                         size_t s = fread(buffer + n, 1, length - n, stdin);
1434                         if (!s && feof(stdin))
1435                                 die("EOF in data (%lu bytes remaining)", length - n);
1436                         n += s;
1437                 }
1438         }
1439
1440         if (fgetc(stdin) != '\n')
1441                 die("An lf did not trail the binary data as expected.");
1442
1443         *size = length;
1444         return buffer;
1445 }
1446
1447 static int validate_raw_date(const char *src, char *result, int maxlen)
1448 {
1449         const char *orig_src = src;
1450         char *endp, sign;
1451
1452         strtoul(src, &endp, 10);
1453         if (endp == src || *endp != ' ')
1454                 return -1;
1455
1456         src = endp + 1;
1457         if (*src != '-' && *src != '+')
1458                 return -1;
1459         sign = *src;
1460
1461         strtoul(src + 1, &endp, 10);
1462         if (endp == src || *endp || (endp - orig_src) >= maxlen)
1463                 return -1;
1464
1465         strcpy(result, orig_src);
1466         return 0;
1467 }
1468
1469 static char *parse_ident(const char *buf)
1470 {
1471         const char *gt;
1472         size_t name_len;
1473         char *ident;
1474
1475         gt = strrchr(buf, '>');
1476         if (!gt)
1477                 die("Missing > in ident string: %s", buf);
1478         gt++;
1479         if (*gt != ' ')
1480                 die("Missing space after > in ident string: %s", buf);
1481         gt++;
1482         name_len = gt - buf;
1483         ident = xmalloc(name_len + 24);
1484         strncpy(ident, buf, name_len);
1485
1486         switch (whenspec) {
1487         case WHENSPEC_RAW:
1488                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1489                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1490                 break;
1491         case WHENSPEC_RFC2822:
1492                 if (parse_date(gt, ident + name_len, 24) < 0)
1493                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1494                 break;
1495         case WHENSPEC_NOW:
1496                 if (strcmp("now", gt))
1497                         die("Date in ident must be 'now': %s", buf);
1498                 datestamp(ident + name_len, 24);
1499                 break;
1500         }
1501
1502         return ident;
1503 }
1504
1505 static void cmd_new_blob(void)
1506 {
1507         size_t l;
1508         void *d;
1509
1510         read_next_command();
1511         cmd_mark();
1512         d = cmd_data(&l);
1513
1514         if (store_object(OBJ_BLOB, d, l, &last_blob, NULL, next_mark))
1515                 free(d);
1516 }
1517
1518 static void unload_one_branch(void)
1519 {
1520         while (cur_active_branches
1521                 && cur_active_branches >= max_active_branches) {
1522                 unsigned long min_commit = ULONG_MAX;
1523                 struct branch *e, *l = NULL, *p = NULL;
1524
1525                 for (e = active_branches; e; e = e->active_next_branch) {
1526                         if (e->last_commit < min_commit) {
1527                                 p = l;
1528                                 min_commit = e->last_commit;
1529                         }
1530                         l = e;
1531                 }
1532
1533                 if (p) {
1534                         e = p->active_next_branch;
1535                         p->active_next_branch = e->active_next_branch;
1536                 } else {
1537                         e = active_branches;
1538                         active_branches = e->active_next_branch;
1539                 }
1540                 e->active_next_branch = NULL;
1541                 if (e->branch_tree.tree) {
1542                         release_tree_content_recursive(e->branch_tree.tree);
1543                         e->branch_tree.tree = NULL;
1544                 }
1545                 cur_active_branches--;
1546         }
1547 }
1548
1549 static void load_branch(struct branch *b)
1550 {
1551         load_tree(&b->branch_tree);
1552         b->active_next_branch = active_branches;
1553         active_branches = b;
1554         cur_active_branches++;
1555         branch_load_count++;
1556 }
1557
1558 static void file_change_m(struct branch *b)
1559 {
1560         const char *p = command_buf.buf + 2;
1561         char *p_uq;
1562         const char *endp;
1563         struct object_entry *oe = oe;
1564         unsigned char sha1[20];
1565         uint16_t mode, inline_data = 0;
1566         char type[20];
1567
1568         p = get_mode(p, &mode);
1569         if (!p)
1570                 die("Corrupt mode: %s", command_buf.buf);
1571         switch (mode) {
1572         case S_IFREG | 0644:
1573         case S_IFREG | 0755:
1574         case S_IFLNK:
1575         case 0644:
1576         case 0755:
1577                 /* ok */
1578                 break;
1579         default:
1580                 die("Corrupt mode: %s", command_buf.buf);
1581         }
1582
1583         if (*p == ':') {
1584                 char *x;
1585                 oe = find_mark(strtoumax(p + 1, &x, 10));
1586                 hashcpy(sha1, oe->sha1);
1587                 p = x;
1588         } else if (!strncmp("inline", p, 6)) {
1589                 inline_data = 1;
1590                 p += 6;
1591         } else {
1592                 if (get_sha1_hex(p, sha1))
1593                         die("Invalid SHA1: %s", command_buf.buf);
1594                 oe = find_object(sha1);
1595                 p += 40;
1596         }
1597         if (*p++ != ' ')
1598                 die("Missing space after SHA1: %s", command_buf.buf);
1599
1600         p_uq = unquote_c_style(p, &endp);
1601         if (p_uq) {
1602                 if (*endp)
1603                         die("Garbage after path in: %s", command_buf.buf);
1604                 p = p_uq;
1605         }
1606
1607         if (inline_data) {
1608                 size_t l;
1609                 void *d;
1610                 if (!p_uq)
1611                         p = p_uq = xstrdup(p);
1612                 read_next_command();
1613                 d = cmd_data(&l);
1614                 if (store_object(OBJ_BLOB, d, l, &last_blob, sha1, 0))
1615                         free(d);
1616         } else if (oe) {
1617                 if (oe->type != OBJ_BLOB)
1618                         die("Not a blob (actually a %s): %s",
1619                                 command_buf.buf, type_names[oe->type]);
1620         } else {
1621                 if (sha1_object_info(sha1, type, NULL))
1622                         die("Blob not found: %s", command_buf.buf);
1623                 if (strcmp(blob_type, type))
1624                         die("Not a blob (actually a %s): %s",
1625                                 command_buf.buf, type);
1626         }
1627
1628         tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode);
1629         free(p_uq);
1630 }
1631
1632 static void file_change_d(struct branch *b)
1633 {
1634         const char *p = command_buf.buf + 2;
1635         char *p_uq;
1636         const char *endp;
1637
1638         p_uq = unquote_c_style(p, &endp);
1639         if (p_uq) {
1640                 if (*endp)
1641                         die("Garbage after path in: %s", command_buf.buf);
1642                 p = p_uq;
1643         }
1644         tree_content_remove(&b->branch_tree, p);
1645         free(p_uq);
1646 }
1647
1648 static void file_change_deleteall(struct branch *b)
1649 {
1650         release_tree_content_recursive(b->branch_tree.tree);
1651         hashclr(b->branch_tree.versions[0].sha1);
1652         hashclr(b->branch_tree.versions[1].sha1);
1653         load_tree(&b->branch_tree);
1654 }
1655
1656 static void cmd_from(struct branch *b)
1657 {
1658         const char *from;
1659         struct branch *s;
1660
1661         if (strncmp("from ", command_buf.buf, 5))
1662                 return;
1663
1664         if (b->last_commit)
1665                 die("Can't reinitailize branch %s", b->name);
1666
1667         from = strchr(command_buf.buf, ' ') + 1;
1668         s = lookup_branch(from);
1669         if (b == s)
1670                 die("Can't create a branch from itself: %s", b->name);
1671         else if (s) {
1672                 unsigned char *t = s->branch_tree.versions[1].sha1;
1673                 hashcpy(b->sha1, s->sha1);
1674                 hashcpy(b->branch_tree.versions[0].sha1, t);
1675                 hashcpy(b->branch_tree.versions[1].sha1, t);
1676         } else if (*from == ':') {
1677                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
1678                 struct object_entry *oe = find_mark(idnum);
1679                 unsigned long size;
1680                 char *buf;
1681                 if (oe->type != OBJ_COMMIT)
1682                         die("Mark :%ju not a commit", idnum);
1683                 hashcpy(b->sha1, oe->sha1);
1684                 buf = gfi_unpack_entry(oe, &size);
1685                 if (!buf || size < 46)
1686                         die("Not a valid commit: %s", from);
1687                 if (memcmp("tree ", buf, 5)
1688                         || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1689                         die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1690                 free(buf);
1691                 hashcpy(b->branch_tree.versions[0].sha1,
1692                         b->branch_tree.versions[1].sha1);
1693         } else if (!get_sha1(from, b->sha1)) {
1694                 if (is_null_sha1(b->sha1)) {
1695                         hashclr(b->branch_tree.versions[0].sha1);
1696                         hashclr(b->branch_tree.versions[1].sha1);
1697                 } else {
1698                         unsigned long size;
1699                         char *buf;
1700
1701                         buf = read_object_with_reference(b->sha1,
1702                                 type_names[OBJ_COMMIT], &size, b->sha1);
1703                         if (!buf || size < 46)
1704                                 die("Not a valid commit: %s", from);
1705                         if (memcmp("tree ", buf, 5)
1706                                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1707                                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1708                         free(buf);
1709                         hashcpy(b->branch_tree.versions[0].sha1,
1710                                 b->branch_tree.versions[1].sha1);
1711                 }
1712         } else
1713                 die("Invalid ref name or SHA1 expression: %s", from);
1714
1715         read_next_command();
1716 }
1717
1718 static struct hash_list *cmd_merge(unsigned int *count)
1719 {
1720         struct hash_list *list = NULL, *n, *e = e;
1721         const char *from;
1722         struct branch *s;
1723
1724         *count = 0;
1725         while (!strncmp("merge ", command_buf.buf, 6)) {
1726                 from = strchr(command_buf.buf, ' ') + 1;
1727                 n = xmalloc(sizeof(*n));
1728                 s = lookup_branch(from);
1729                 if (s)
1730                         hashcpy(n->sha1, s->sha1);
1731                 else if (*from == ':') {
1732                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
1733                         struct object_entry *oe = find_mark(idnum);
1734                         if (oe->type != OBJ_COMMIT)
1735                                 die("Mark :%ju not a commit", idnum);
1736                         hashcpy(n->sha1, oe->sha1);
1737                 } else if (get_sha1(from, n->sha1))
1738                         die("Invalid ref name or SHA1 expression: %s", from);
1739
1740                 n->next = NULL;
1741                 if (list)
1742                         e->next = n;
1743                 else
1744                         list = n;
1745                 e = n;
1746                 (*count)++;
1747                 read_next_command();
1748         }
1749         return list;
1750 }
1751
1752 static void cmd_new_commit(void)
1753 {
1754         struct branch *b;
1755         void *msg;
1756         size_t msglen;
1757         char *sp;
1758         char *author = NULL;
1759         char *committer = NULL;
1760         struct hash_list *merge_list = NULL;
1761         unsigned int merge_count;
1762
1763         /* Obtain the branch name from the rest of our command */
1764         sp = strchr(command_buf.buf, ' ') + 1;
1765         b = lookup_branch(sp);
1766         if (!b)
1767                 b = new_branch(sp);
1768
1769         read_next_command();
1770         cmd_mark();
1771         if (!strncmp("author ", command_buf.buf, 7)) {
1772                 author = parse_ident(command_buf.buf + 7);
1773                 read_next_command();
1774         }
1775         if (!strncmp("committer ", command_buf.buf, 10)) {
1776                 committer = parse_ident(command_buf.buf + 10);
1777                 read_next_command();
1778         }
1779         if (!committer)
1780                 die("Expected committer but didn't get one");
1781         msg = cmd_data(&msglen);
1782         read_next_command();
1783         cmd_from(b);
1784         merge_list = cmd_merge(&merge_count);
1785
1786         /* ensure the branch is active/loaded */
1787         if (!b->branch_tree.tree || !max_active_branches) {
1788                 unload_one_branch();
1789                 load_branch(b);
1790         }
1791
1792         /* file_change* */
1793         for (;;) {
1794                 if (1 == command_buf.len)
1795                         break;
1796                 else if (!strncmp("M ", command_buf.buf, 2))
1797                         file_change_m(b);
1798                 else if (!strncmp("D ", command_buf.buf, 2))
1799                         file_change_d(b);
1800                 else if (!strcmp("deleteall", command_buf.buf))
1801                         file_change_deleteall(b);
1802                 else
1803                         die("Unsupported file_change: %s", command_buf.buf);
1804                 read_next_command();
1805         }
1806
1807         /* build the tree and the commit */
1808         store_tree(&b->branch_tree);
1809         hashcpy(b->branch_tree.versions[0].sha1,
1810                 b->branch_tree.versions[1].sha1);
1811         size_dbuf(&new_data, 114 + msglen
1812                 + merge_count * 49
1813                 + (author
1814                         ? strlen(author) + strlen(committer)
1815                         : 2 * strlen(committer)));
1816         sp = new_data.buffer;
1817         sp += sprintf(sp, "tree %s\n",
1818                 sha1_to_hex(b->branch_tree.versions[1].sha1));
1819         if (!is_null_sha1(b->sha1))
1820                 sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
1821         while (merge_list) {
1822                 struct hash_list *next = merge_list->next;
1823                 sp += sprintf(sp, "parent %s\n", sha1_to_hex(merge_list->sha1));
1824                 free(merge_list);
1825                 merge_list = next;
1826         }
1827         sp += sprintf(sp, "author %s\n", author ? author : committer);
1828         sp += sprintf(sp, "committer %s\n", committer);
1829         *sp++ = '\n';
1830         memcpy(sp, msg, msglen);
1831         sp += msglen;
1832         free(author);
1833         free(committer);
1834         free(msg);
1835
1836         if (!store_object(OBJ_COMMIT,
1837                 new_data.buffer, sp - (char*)new_data.buffer,
1838                 NULL, b->sha1, next_mark))
1839                 b->pack_id = pack_id;
1840         b->last_commit = object_count_by_type[OBJ_COMMIT];
1841 }
1842
1843 static void cmd_new_tag(void)
1844 {
1845         char *sp;
1846         const char *from;
1847         char *tagger;
1848         struct branch *s;
1849         void *msg;
1850         size_t msglen;
1851         struct tag *t;
1852         uintmax_t from_mark = 0;
1853         unsigned char sha1[20];
1854
1855         /* Obtain the new tag name from the rest of our command */
1856         sp = strchr(command_buf.buf, ' ') + 1;
1857         t = pool_alloc(sizeof(struct tag));
1858         t->next_tag = NULL;
1859         t->name = pool_strdup(sp);
1860         if (last_tag)
1861                 last_tag->next_tag = t;
1862         else
1863                 first_tag = t;
1864         last_tag = t;
1865         read_next_command();
1866
1867         /* from ... */
1868         if (strncmp("from ", command_buf.buf, 5))
1869                 die("Expected from command, got %s", command_buf.buf);
1870         from = strchr(command_buf.buf, ' ') + 1;
1871         s = lookup_branch(from);
1872         if (s) {
1873                 hashcpy(sha1, s->sha1);
1874         } else if (*from == ':') {
1875                 struct object_entry *oe;
1876                 from_mark = strtoumax(from + 1, NULL, 10);
1877                 oe = find_mark(from_mark);
1878                 if (oe->type != OBJ_COMMIT)
1879                         die("Mark :%ju not a commit", from_mark);
1880                 hashcpy(sha1, oe->sha1);
1881         } else if (!get_sha1(from, sha1)) {
1882                 unsigned long size;
1883                 char *buf;
1884
1885                 buf = read_object_with_reference(sha1,
1886                         type_names[OBJ_COMMIT], &size, sha1);
1887                 if (!buf || size < 46)
1888                         die("Not a valid commit: %s", from);
1889                 free(buf);
1890         } else
1891                 die("Invalid ref name or SHA1 expression: %s", from);
1892         read_next_command();
1893
1894         /* tagger ... */
1895         if (strncmp("tagger ", command_buf.buf, 7))
1896                 die("Expected tagger command, got %s", command_buf.buf);
1897         tagger = parse_ident(command_buf.buf + 7);
1898
1899         /* tag payload/message */
1900         read_next_command();
1901         msg = cmd_data(&msglen);
1902
1903         /* build the tag object */
1904         size_dbuf(&new_data, 67+strlen(t->name)+strlen(tagger)+msglen);
1905         sp = new_data.buffer;
1906         sp += sprintf(sp, "object %s\n", sha1_to_hex(sha1));
1907         sp += sprintf(sp, "type %s\n", type_names[OBJ_COMMIT]);
1908         sp += sprintf(sp, "tag %s\n", t->name);
1909         sp += sprintf(sp, "tagger %s\n", tagger);
1910         *sp++ = '\n';
1911         memcpy(sp, msg, msglen);
1912         sp += msglen;
1913         free(tagger);
1914         free(msg);
1915
1916         if (store_object(OBJ_TAG, new_data.buffer,
1917                 sp - (char*)new_data.buffer,
1918                 NULL, t->sha1, 0))
1919                 t->pack_id = MAX_PACK_ID;
1920         else
1921                 t->pack_id = pack_id;
1922 }
1923
1924 static void cmd_reset_branch(void)
1925 {
1926         struct branch *b;
1927         char *sp;
1928
1929         /* Obtain the branch name from the rest of our command */
1930         sp = strchr(command_buf.buf, ' ') + 1;
1931         b = lookup_branch(sp);
1932         if (b) {
1933                 b->last_commit = 0;
1934                 if (b->branch_tree.tree) {
1935                         release_tree_content_recursive(b->branch_tree.tree);
1936                         b->branch_tree.tree = NULL;
1937                 }
1938         }
1939         else
1940                 b = new_branch(sp);
1941         read_next_command();
1942         cmd_from(b);
1943 }
1944
1945 static void cmd_checkpoint(void)
1946 {
1947         if (object_count) {
1948                 cycle_packfile();
1949                 dump_branches();
1950                 dump_tags();
1951                 dump_marks();
1952         }
1953         read_next_command();
1954 }
1955
1956 static const char fast_import_usage[] =
1957 "git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
1958
1959 int main(int argc, const char **argv)
1960 {
1961         int i, show_stats = 1;
1962
1963         git_config(git_default_config);
1964
1965         for (i = 1; i < argc; i++) {
1966                 const char *a = argv[i];
1967
1968                 if (*a != '-' || !strcmp(a, "--"))
1969                         break;
1970                 else if (!strncmp(a, "--date-format=", 14)) {
1971                         const char *fmt = a + 14;
1972                         if (!strcmp(fmt, "raw"))
1973                                 whenspec = WHENSPEC_RAW;
1974                         else if (!strcmp(fmt, "rfc2822"))
1975                                 whenspec = WHENSPEC_RFC2822;
1976                         else if (!strcmp(fmt, "now"))
1977                                 whenspec = WHENSPEC_NOW;
1978                         else
1979                                 die("unknown --date-format argument %s", fmt);
1980                 }
1981                 else if (!strncmp(a, "--max-pack-size=", 16))
1982                         max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
1983                 else if (!strncmp(a, "--depth=", 8))
1984                         max_depth = strtoul(a + 8, NULL, 0);
1985                 else if (!strncmp(a, "--active-branches=", 18))
1986                         max_active_branches = strtoul(a + 18, NULL, 0);
1987                 else if (!strncmp(a, "--export-marks=", 15))
1988                         mark_file = a + 15;
1989                 else if (!strcmp(a, "--force"))
1990                         force_update = 1;
1991                 else if (!strcmp(a, "--quiet"))
1992                         show_stats = 0;
1993                 else if (!strcmp(a, "--stats"))
1994                         show_stats = 1;
1995                 else
1996                         die("unknown option %s", a);
1997         }
1998         if (i != argc)
1999                 usage(fast_import_usage);
2000
2001         alloc_objects(object_entry_alloc);
2002         strbuf_init(&command_buf);
2003
2004         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2005         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2006         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2007         marks = pool_calloc(1, sizeof(struct mark_set));
2008
2009         start_packfile();
2010         for (;;) {
2011                 read_next_command();
2012                 if (command_buf.eof)
2013                         break;
2014                 else if (!strcmp("blob", command_buf.buf))
2015                         cmd_new_blob();
2016                 else if (!strncmp("commit ", command_buf.buf, 7))
2017                         cmd_new_commit();
2018                 else if (!strncmp("tag ", command_buf.buf, 4))
2019                         cmd_new_tag();
2020                 else if (!strncmp("reset ", command_buf.buf, 6))
2021                         cmd_reset_branch();
2022                 else if (!strcmp("checkpoint", command_buf.buf))
2023                         cmd_checkpoint();
2024                 else
2025                         die("Unsupported command: %s", command_buf.buf);
2026         }
2027         end_packfile();
2028
2029         dump_branches();
2030         dump_tags();
2031         unkeep_all_packs();
2032         dump_marks();
2033
2034         if (show_stats) {
2035                 uintmax_t total_count = 0, duplicate_count = 0;
2036                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2037                         total_count += object_count_by_type[i];
2038                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2039                         duplicate_count += duplicate_count_by_type[i];
2040
2041                 fprintf(stderr, "%s statistics:\n", argv[0]);
2042                 fprintf(stderr, "---------------------------------------------------------------------\n");
2043                 fprintf(stderr, "Alloc'd objects: %10ju\n", alloc_count);
2044                 fprintf(stderr, "Total objects:   %10ju (%10ju duplicates                  )\n", total_count, duplicate_count);
2045                 fprintf(stderr, "      blobs  :   %10ju (%10ju duplicates %10ju deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
2046                 fprintf(stderr, "      trees  :   %10ju (%10ju duplicates %10ju deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
2047                 fprintf(stderr, "      commits:   %10ju (%10ju duplicates %10ju deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
2048                 fprintf(stderr, "      tags   :   %10ju (%10ju duplicates %10ju deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
2049                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2050                 fprintf(stderr, "      marks:     %10ju (%10ju unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2051                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2052                 fprintf(stderr, "Memory total:    %10ju KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2053                 fprintf(stderr, "       pools:    %10lu KiB\n", total_allocd/1024);
2054                 fprintf(stderr, "     objects:    %10ju KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2055                 fprintf(stderr, "---------------------------------------------------------------------\n");
2056                 pack_report();
2057                 fprintf(stderr, "---------------------------------------------------------------------\n");
2058                 fprintf(stderr, "\n");
2059         }
2060
2061         return failure ? 1 : 0;
2062 }