Allow symlink blobs in trees during fast-import.
[git] / fast-import.c
1 /*
2 Format of STDIN stream:
3
4   stream ::= cmd*;
5
6   cmd ::= new_blob
7         | new_branch
8         | new_commit
9         | new_tag
10         ;
11
12   new_blob ::= 'blob' lf
13         mark?
14     file_content;
15   file_content ::= data;
16
17   new_branch ::= 'branch' sp ref_str lf
18     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
19     lf;
20
21   new_commit ::= 'commit' sp ref_str lf
22         mark?
23         ('author' sp name '<' email '>' ts tz lf)?
24         'committer' sp name '<' email '>' ts tz lf
25         commit_msg
26     file_change*
27     lf;
28   commit_msg ::= data;
29
30   file_change ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf
31                 | 'D' sp path_str lf
32                 ;
33   mode ::= '644' | '755';
34
35   new_tag ::= 'tag' sp tag_str lf
36     'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
37         'tagger' sp name '<' email '>' ts tz lf
38     tag_msg;
39   tag_msg ::= data;
40
41      # note: the first idnum in a stream should be 1 and subsequent
42      # idnums should not have gaps between values as this will cause
43      # the stream parser to reserve space for the gapped values.  An
44          # idnum can be updated in the future to a new object by issuing
45      # a new mark directive with the old idnum.
46          #
47   mark ::= 'mark' sp idnum lf;
48
49      # note: declen indicates the length of binary_data in bytes.
50      # declen does not include the lf preceeding or trailing the
51      # binary data.
52      #
53   data ::= 'data' sp declen lf
54     binary_data
55         lf;
56
57      # note: quoted strings are C-style quoting supporting \c for
58      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
59          # is the signed byte value in octal.  Note that the only
60      # characters which must actually be escaped to protect the
61      # stream formatting is: \, " and LF.  Otherwise these values
62          # are UTF8.
63      #
64   ref_str     ::= ref     | '"' quoted(ref)     '"' ;
65   sha1exp_str ::= sha1exp | '"' quoted(sha1exp) '"' ;
66   tag_str     ::= tag     | '"' quoted(tag)     '"' ;
67   path_str    ::= path    | '"' quoted(path)    '"' ;
68
69   declen ::= # unsigned 32 bit value, ascii base10 notation;
70   binary_data ::= # file content, not interpreted;
71
72   sp ::= # ASCII space character;
73   lf ::= # ASCII newline (LF) character;
74
75      # note: a colon (':') must precede the numerical value assigned to
76          # an idnum.  This is to distinguish it from a ref or tag name as
77      # GIT does not permit ':' in ref or tag strings.
78          #
79   idnum   ::= ':' declen;
80   path    ::= # GIT style file path, e.g. "a/b/c";
81   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
82   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
83   sha1exp ::= # Any valid GIT SHA1 expression;
84   hexsha1 ::= # SHA1 in hexadecimal format;
85
86      # note: name and email are UTF8 strings, however name must not
87          # contain '<' or lf and email must not contain any of the
88      # following: '<', '>', lf.
89          #
90   name  ::= # valid GIT author/committer name;
91   email ::= # valid GIT author/committer email;
92   ts    ::= # time since the epoch in seconds, ascii base10 notation;
93   tz    ::= # GIT style timezone;
94 */
95
96 #include "builtin.h"
97 #include "cache.h"
98 #include "object.h"
99 #include "blob.h"
100 #include "tree.h"
101 #include "delta.h"
102 #include "pack.h"
103 #include "refs.h"
104 #include "csum-file.h"
105 #include "strbuf.h"
106 #include "quote.h"
107
108 struct object_entry
109 {
110         struct object_entry *next;
111         enum object_type type;
112         unsigned long offset;
113         unsigned char sha1[20];
114 };
115
116 struct object_entry_pool
117 {
118         struct object_entry_pool *next_pool;
119         struct object_entry *next_free;
120         struct object_entry *end;
121         struct object_entry entries[FLEX_ARRAY]; /* more */
122 };
123
124 struct last_object
125 {
126         void *data;
127         unsigned int len;
128         unsigned int depth;
129         unsigned char sha1[20];
130 };
131
132 struct mem_pool
133 {
134         struct mem_pool *next_pool;
135         char *next_free;
136         char *end;
137         char space[FLEX_ARRAY]; /* more */
138 };
139
140 struct atom_str
141 {
142         struct atom_str *next_atom;
143         int str_len;
144         char str_dat[FLEX_ARRAY]; /* more */
145 };
146
147 struct tree_content;
148 struct tree_entry
149 {
150         struct tree_content *tree;
151         struct atom_str* name;
152         unsigned int mode;
153         unsigned char sha1[20];
154 };
155
156 struct tree_content
157 {
158         unsigned int entry_capacity; /* must match avail_tree_content */
159         unsigned int entry_count;
160         struct tree_entry *entries[FLEX_ARRAY]; /* more */
161 };
162
163 struct avail_tree_content
164 {
165         unsigned int entry_capacity; /* must match tree_content */
166         struct avail_tree_content *next_avail;
167 };
168
169 struct branch
170 {
171         struct branch *table_next_branch;
172         struct branch *active_next_branch;
173         const char *name;
174         unsigned long last_commit;
175         struct tree_entry branch_tree;
176         unsigned char sha1[20];
177 };
178
179
180 /* Stats and misc. counters */
181 static int max_depth = 10;
182 static unsigned long alloc_count;
183 static unsigned long branch_count;
184 static unsigned long object_count;
185 static unsigned long duplicate_count;
186 static unsigned long object_count_by_type[9];
187 static unsigned long duplicate_count_by_type[9];
188
189 /* Memory pools */
190 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
191 static size_t total_allocd;
192 static struct mem_pool *mem_pool;
193
194 /* Atom management */
195 static unsigned int atom_table_sz = 4451;
196 static unsigned int atom_cnt;
197 static struct atom_str **atom_table;
198
199 /* The .pack file being generated */
200 static int pack_fd;
201 static unsigned long pack_offset;
202 static unsigned char pack_sha1[20];
203
204 /* Table of objects we've written. */
205 static unsigned int object_entry_alloc = 1000;
206 static struct object_entry_pool *blocks;
207 static struct object_entry *object_table[1 << 16];
208
209 /* Our last blob */
210 static struct last_object last_blob;
211
212 /* Tree management */
213 static unsigned int tree_entry_alloc = 1000;
214 static void *avail_tree_entry;
215 static unsigned int avail_tree_table_sz = 100;
216 static struct avail_tree_content **avail_tree_table;
217
218 /* Branch data */
219 static unsigned int max_active_branches = 5;
220 static unsigned int cur_active_branches;
221 static unsigned int branch_table_sz = 1039;
222 static struct branch **branch_table;
223 static struct branch *active_branches;
224
225 /* Input stream parsing */
226 static struct strbuf command_buf;
227 static unsigned long command_mark;
228
229
230 static void alloc_objects(int cnt)
231 {
232         struct object_entry_pool *b;
233
234         b = xmalloc(sizeof(struct object_entry_pool)
235                 + cnt * sizeof(struct object_entry));
236         b->next_pool = blocks;
237         b->next_free = b->entries;
238         b->end = b->entries + cnt;
239         blocks = b;
240         alloc_count += cnt;
241 }
242
243 static struct object_entry* new_object(unsigned char *sha1)
244 {
245         struct object_entry *e;
246
247         if (blocks->next_free == blocks->end)
248                 alloc_objects(object_entry_alloc);
249
250         e = blocks->next_free++;
251         memcpy(e->sha1, sha1, sizeof(e->sha1));
252         return e;
253 }
254
255 static struct object_entry* find_object(unsigned char *sha1)
256 {
257         unsigned int h = sha1[0] << 8 | sha1[1];
258         struct object_entry *e;
259         for (e = object_table[h]; e; e = e->next)
260                 if (!memcmp(sha1, e->sha1, sizeof(e->sha1)))
261                         return e;
262         return NULL;
263 }
264
265 static struct object_entry* insert_object(unsigned char *sha1)
266 {
267         unsigned int h = sha1[0] << 8 | sha1[1];
268         struct object_entry *e = object_table[h];
269         struct object_entry *p = NULL;
270
271         while (e) {
272                 if (!memcmp(sha1, e->sha1, sizeof(e->sha1)))
273                         return e;
274                 p = e;
275                 e = e->next;
276         }
277
278         e = new_object(sha1);
279         e->next = NULL;
280         e->offset = 0;
281         if (p)
282                 p->next = e;
283         else
284                 object_table[h] = e;
285         return e;
286 }
287
288 static unsigned int hc_str(const char *s, size_t len)
289 {
290         unsigned int r = 0;
291         while (len-- > 0)
292                 r = r * 31 + *s++;
293         return r;
294 }
295
296 static void* pool_alloc(size_t len)
297 {
298         struct mem_pool *p;
299         void *r;
300
301         for (p = mem_pool; p; p = p->next_pool)
302                 if ((p->end - p->next_free >= len))
303                         break;
304
305         if (!p) {
306                 if (len >= (mem_pool_alloc/2)) {
307                         total_allocd += len;
308                         return xmalloc(len);
309                 }
310                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
311                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
312                 p->next_pool = mem_pool;
313                 p->next_free = p->space;
314                 p->end = p->next_free + mem_pool_alloc;
315                 mem_pool = p;
316         }
317
318         r = p->next_free;
319         p->next_free += len;
320         return r;
321 }
322
323 static void* pool_calloc(size_t count, size_t size)
324 {
325         size_t len = count * size;
326         void *r = pool_alloc(len);
327         memset(r, 0, len);
328         return r;
329 }
330
331 static char* pool_strdup(const char *s)
332 {
333         char *r = pool_alloc(strlen(s) + 1);
334         strcpy(r, s);
335         return r;
336 }
337
338 static struct atom_str* to_atom(const char *s, size_t len)
339 {
340         unsigned int hc = hc_str(s, len) % atom_table_sz;
341         struct atom_str *c;
342
343         for (c = atom_table[hc]; c; c = c->next_atom)
344                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
345                         return c;
346
347         c = pool_alloc(sizeof(struct atom_str) + len + 1);
348         c->str_len = len;
349         strncpy(c->str_dat, s, len);
350         c->str_dat[len] = 0;
351         c->next_atom = atom_table[hc];
352         atom_table[hc] = c;
353         atom_cnt++;
354         return c;
355 }
356
357 static struct branch* lookup_branch(const char *name)
358 {
359         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
360         struct branch *b;
361
362         for (b = branch_table[hc]; b; b = b->table_next_branch)
363                 if (!strcmp(name, b->name))
364                         return b;
365         return NULL;
366 }
367
368 static struct branch* new_branch(const char *name)
369 {
370         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
371         struct branch* b = lookup_branch(name);
372
373         if (b)
374                 die("Invalid attempt to create duplicate branch: %s", name);
375         if (check_ref_format(name))
376                 die("Branch name doesn't conform to GIT standards: %s", name);
377
378         b = pool_calloc(1, sizeof(struct branch));
379         b->name = pool_strdup(name);
380         b->table_next_branch = branch_table[hc];
381         branch_table[hc] = b;
382         branch_count++;
383         return b;
384 }
385
386 static unsigned int hc_entries(unsigned int cnt)
387 {
388         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
389         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
390 }
391
392 static struct tree_content* new_tree_content(unsigned int cnt)
393 {
394         struct avail_tree_content *f, *l = NULL;
395         struct tree_content *t;
396         unsigned int hc = hc_entries(cnt);
397
398         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
399                 if (f->entry_capacity >= cnt)
400                         break;
401
402         if (f) {
403                 if (l)
404                         l->next_avail = f->next_avail;
405                 else
406                         avail_tree_table[hc] = f->next_avail;
407         } else {
408                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
409                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
410                 f->entry_capacity = cnt;
411         }
412
413         t = (struct tree_content*)f;
414         t->entry_count = 0;
415         return t;
416 }
417
418 static void release_tree_entry(struct tree_entry *e);
419 static void release_tree_content(struct tree_content *t)
420 {
421         struct avail_tree_content *f = (struct avail_tree_content*)t;
422         unsigned int hc = hc_entries(f->entry_capacity);
423         unsigned int i;
424         for (i = 0; i < t->entry_count; i++)
425                 release_tree_entry(t->entries[i]);
426         f->next_avail = avail_tree_table[hc];
427         avail_tree_table[hc] = f;
428 }
429
430 static struct tree_content* grow_tree_content(
431         struct tree_content *t,
432         int amt)
433 {
434         struct tree_content *r = new_tree_content(t->entry_count + amt);
435         r->entry_count = t->entry_count;
436         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
437         release_tree_content(t);
438         return r;
439 }
440
441 static struct tree_entry* new_tree_entry()
442 {
443         struct tree_entry *e;
444
445         if (!avail_tree_entry) {
446                 unsigned int n = tree_entry_alloc;
447                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
448                 while (n--) {
449                         *((void**)e) = e + 1;
450                         e++;
451                 }
452         }
453
454         e = avail_tree_entry;
455         avail_tree_entry = *((void**)e);
456         return e;
457 }
458
459 static void release_tree_entry(struct tree_entry *e)
460 {
461         if (e->tree)
462                 release_tree_content(e->tree);
463         *((void**)e) = avail_tree_entry;
464         avail_tree_entry = e;
465 }
466
467 static void yread(int fd, void *buffer, size_t length)
468 {
469         ssize_t ret = 0;
470         while (ret < length) {
471                 ssize_t size = xread(fd, (char *) buffer + ret, length - ret);
472                 if (!size)
473                         die("Read from descriptor %i: end of stream", fd);
474                 if (size < 0)
475                         die("Read from descriptor %i: %s", fd, strerror(errno));
476                 ret += size;
477         }
478 }
479
480 static void ywrite(int fd, void *buffer, size_t length)
481 {
482         ssize_t ret = 0;
483         while (ret < length) {
484                 ssize_t size = xwrite(fd, (char *) buffer + ret, length - ret);
485                 if (!size)
486                         die("Write to descriptor %i: end of file", fd);
487                 if (size < 0)
488                         die("Write to descriptor %i: %s", fd, strerror(errno));
489                 ret += size;
490         }
491 }
492
493 static size_t encode_header(
494         enum object_type type,
495         size_t size,
496         unsigned char *hdr)
497 {
498         int n = 1;
499         unsigned char c;
500
501         if (type < OBJ_COMMIT || type > OBJ_DELTA)
502                 die("bad type %d", type);
503
504         c = (type << 4) | (size & 15);
505         size >>= 4;
506         while (size) {
507                 *hdr++ = c | 0x80;
508                 c = size & 0x7f;
509                 size >>= 7;
510                 n++;
511         }
512         *hdr = c;
513         return n;
514 }
515
516 static int store_object(
517         enum object_type type,
518         void *dat,
519         size_t datlen,
520         struct last_object *last,
521         unsigned char *sha1out)
522 {
523         void *out, *delta;
524         struct object_entry *e;
525         unsigned char hdr[96];
526         unsigned char sha1[20];
527         unsigned long hdrlen, deltalen;
528         SHA_CTX c;
529         z_stream s;
530
531         hdrlen = sprintf((char*)hdr,"%s %lu",type_names[type],datlen) + 1;
532         SHA1_Init(&c);
533         SHA1_Update(&c, hdr, hdrlen);
534         SHA1_Update(&c, dat, datlen);
535         SHA1_Final(sha1, &c);
536         if (sha1out)
537                 memcpy(sha1out, sha1, sizeof(sha1));
538
539         e = insert_object(sha1);
540         if (e->offset) {
541                 duplicate_count++;
542                 duplicate_count_by_type[type]++;
543                 return 1;
544         }
545         e->type = type;
546         e->offset = pack_offset;
547         object_count++;
548         object_count_by_type[type]++;
549
550         if (last && last->data && last->depth < max_depth)
551                 delta = diff_delta(last->data, last->len,
552                         dat, datlen,
553                         &deltalen, 0);
554         else
555                 delta = 0;
556
557         memset(&s, 0, sizeof(s));
558         deflateInit(&s, zlib_compression_level);
559
560         if (delta) {
561                 last->depth++;
562                 s.next_in = delta;
563                 s.avail_in = deltalen;
564                 hdrlen = encode_header(OBJ_DELTA, deltalen, hdr);
565                 ywrite(pack_fd, hdr, hdrlen);
566                 ywrite(pack_fd, last->sha1, sizeof(sha1));
567                 pack_offset += hdrlen + sizeof(sha1);
568         } else {
569                 if (last)
570                         last->depth = 0;
571                 s.next_in = dat;
572                 s.avail_in = datlen;
573                 hdrlen = encode_header(type, datlen, hdr);
574                 ywrite(pack_fd, hdr, hdrlen);
575                 pack_offset += hdrlen;
576         }
577
578         s.avail_out = deflateBound(&s, s.avail_in);
579         s.next_out = out = xmalloc(s.avail_out);
580         while (deflate(&s, Z_FINISH) == Z_OK)
581                 /* nothing */;
582         deflateEnd(&s);
583
584         ywrite(pack_fd, out, s.total_out);
585         pack_offset += s.total_out;
586
587         free(out);
588         if (delta)
589                 free(delta);
590         if (last) {
591                 if (last->data)
592                         free(last->data);
593                 last->data = dat;
594                 last->len = datlen;
595                 memcpy(last->sha1, sha1, sizeof(sha1));
596         }
597         return 0;
598 }
599
600 static const char *get_mode(const char *str, unsigned int *modep)
601 {
602         unsigned char c;
603         unsigned int mode = 0;
604
605         while ((c = *str++) != ' ') {
606                 if (c < '0' || c > '7')
607                         return NULL;
608                 mode = (mode << 3) + (c - '0');
609         }
610         *modep = mode;
611         return str;
612 }
613
614 static void load_tree(struct tree_entry *root)
615 {
616         struct object_entry *myoe;
617         struct tree_content *t;
618         unsigned long size;
619         char *buf;
620         const char *c;
621         char type[20];
622
623         root->tree = t = new_tree_content(8);
624         if (!memcmp(root->sha1, null_sha1, 20))
625                 return;
626
627         myoe = find_object(root->sha1);
628         if (myoe) {
629                 die("FIXME");
630         } else {
631                 buf = read_sha1_file(root->sha1, type, &size);
632                 if (!buf || strcmp(type, tree_type))
633                         die("Can't load existing tree %s", sha1_to_hex(root->sha1));
634         }
635
636         c = buf;
637         while (c != (buf + size)) {
638                 struct tree_entry *e = new_tree_entry();
639
640                 if (t->entry_count == t->entry_capacity)
641                         root->tree = t = grow_tree_content(t, 8);
642                 t->entries[t->entry_count++] = e;
643
644                 e->tree = NULL;
645                 c = get_mode(c, &e->mode);
646                 if (!c)
647                         die("Corrupt mode in %s", sha1_to_hex(root->sha1));
648                 e->name = to_atom(c, strlen(c));
649                 c += e->name->str_len + 1;
650                 memcpy(e->sha1, c, sizeof(e->sha1));
651                 c += 20;
652         }
653         free(buf);
654 }
655
656 static int tecmp (const void *_a, const void *_b)
657 {
658         struct tree_entry *a = *((struct tree_entry**)_a);
659         struct tree_entry *b = *((struct tree_entry**)_b);
660         return base_name_compare(
661                 a->name->str_dat, a->name->str_len, a->mode,
662                 b->name->str_dat, b->name->str_len, b->mode);
663 }
664
665 static void store_tree(struct tree_entry *root)
666 {
667         struct tree_content *t = root->tree;
668         unsigned int i;
669         size_t maxlen;
670         char *buf, *c;
671
672         if (memcmp(root->sha1, null_sha1, 20))
673                 return;
674
675         maxlen = 0;
676         for (i = 0; i < t->entry_count; i++) {
677                 maxlen += t->entries[i]->name->str_len + 34;
678                 if (t->entries[i]->tree)
679                         store_tree(t->entries[i]);
680         }
681
682         qsort(t->entries, t->entry_count, sizeof(t->entries[0]), tecmp);
683         buf = c = xmalloc(maxlen);
684         for (i = 0; i < t->entry_count; i++) {
685                 struct tree_entry *e = t->entries[i];
686                 c += sprintf(c, "%o", e->mode);
687                 *c++ = ' ';
688                 strcpy(c, e->name->str_dat);
689                 c += e->name->str_len + 1;
690                 memcpy(c, e->sha1, 20);
691                 c += 20;
692         }
693         store_object(OBJ_TREE, buf, c - buf, NULL, root->sha1);
694         free(buf);
695 }
696
697 static int tree_content_set(
698         struct tree_entry *root,
699         const char *p,
700         const unsigned char *sha1,
701         const unsigned int mode)
702 {
703         struct tree_content *t = root->tree;
704         const char *slash1;
705         unsigned int i, n;
706         struct tree_entry *e;
707
708         slash1 = strchr(p, '/');
709         if (slash1)
710                 n = slash1 - p;
711         else
712                 n = strlen(p);
713
714         for (i = 0; i < t->entry_count; i++) {
715                 e = t->entries[i];
716                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
717                         if (!slash1) {
718                                 if (e->mode == mode && !memcmp(e->sha1, sha1, 20))
719                                         return 0;
720                                 e->mode = mode;
721                                 memcpy(e->sha1, sha1, 20);
722                                 if (e->tree) {
723                                         release_tree_content(e->tree);
724                                         e->tree = NULL;
725                                 }
726                                 memcpy(root->sha1, null_sha1, 20);
727                                 return 1;
728                         }
729                         if (!S_ISDIR(e->mode)) {
730                                 e->tree = new_tree_content(8);
731                                 e->mode = S_IFDIR;
732                         }
733                         if (!e->tree)
734                                 load_tree(e);
735                         if (tree_content_set(e, slash1 + 1, sha1, mode)) {
736                                 memcpy(root->sha1, null_sha1, 20);
737                                 return 1;
738                         }
739                         return 0;
740                 }
741         }
742
743         if (t->entry_count == t->entry_capacity)
744                 root->tree = t = grow_tree_content(t, 8);
745         e = new_tree_entry();
746         e->name = to_atom(p, n);
747         t->entries[t->entry_count++] = e;
748         if (slash1) {
749                 e->tree = new_tree_content(8);
750                 e->mode = S_IFDIR;
751                 tree_content_set(e, slash1 + 1, sha1, mode);
752         } else {
753                 e->tree = NULL;
754                 e->mode = mode;
755                 memcpy(e->sha1, sha1, 20);
756         }
757         memcpy(root->sha1, null_sha1, 20);
758         return 1;
759 }
760
761 static int tree_content_remove(struct tree_entry *root, const char *p)
762 {
763         struct tree_content *t = root->tree;
764         const char *slash1;
765         unsigned int i, n;
766         struct tree_entry *e;
767
768         slash1 = strchr(p, '/');
769         if (slash1)
770                 n = slash1 - p;
771         else
772                 n = strlen(p);
773
774         for (i = 0; i < t->entry_count; i++) {
775                 e = t->entries[i];
776                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
777                         if (!slash1 || !S_ISDIR(e->mode))
778                                 goto del_entry;
779                         if (!e->tree)
780                                 load_tree(e);
781                         if (tree_content_remove(e, slash1 + 1)) {
782                                 if (!e->tree->entry_count)
783                                         goto del_entry;
784                                 memcpy(root->sha1, null_sha1, 20);
785                                 return 1;
786                         }
787                         return 0;
788                 }
789         }
790         return 0;
791
792 del_entry:
793         for (i++; i < t->entry_count; i++)
794                 t->entries[i-1] = t->entries[i];
795         t->entry_count--;
796         release_tree_entry(e);
797         memcpy(root->sha1, null_sha1, 20);
798         return 1;
799 }
800
801 static void init_pack_header()
802 {
803         struct pack_header hdr;
804
805         hdr.hdr_signature = htonl(PACK_SIGNATURE);
806         hdr.hdr_version = htonl(2);
807         hdr.hdr_entries = 0;
808
809         ywrite(pack_fd, &hdr, sizeof(hdr));
810         pack_offset = sizeof(hdr);
811 }
812
813 static void fixup_header_footer()
814 {
815         SHA_CTX c;
816         char hdr[8];
817         unsigned long cnt;
818         char *buf;
819         size_t n;
820
821         if (lseek(pack_fd, 0, SEEK_SET) != 0)
822                 die("Failed seeking to start: %s", strerror(errno));
823
824         SHA1_Init(&c);
825         yread(pack_fd, hdr, 8);
826         SHA1_Update(&c, hdr, 8);
827
828         cnt = htonl(object_count);
829         SHA1_Update(&c, &cnt, 4);
830         ywrite(pack_fd, &cnt, 4);
831
832         buf = xmalloc(128 * 1024);
833         for (;;) {
834                 n = xread(pack_fd, buf, 128 * 1024);
835                 if (n <= 0)
836                         break;
837                 SHA1_Update(&c, buf, n);
838         }
839         free(buf);
840
841         SHA1_Final(pack_sha1, &c);
842         ywrite(pack_fd, pack_sha1, sizeof(pack_sha1));
843 }
844
845 static int oecmp (const void *_a, const void *_b)
846 {
847         struct object_entry *a = *((struct object_entry**)_a);
848         struct object_entry *b = *((struct object_entry**)_b);
849         return memcmp(a->sha1, b->sha1, sizeof(a->sha1));
850 }
851
852 static void write_index(const char *idx_name)
853 {
854         struct sha1file *f;
855         struct object_entry **idx, **c, **last;
856         struct object_entry *e;
857         struct object_entry_pool *o;
858         unsigned int array[256];
859         int i;
860
861         /* Build the sorted table of object IDs. */
862         idx = xmalloc(object_count * sizeof(struct object_entry*));
863         c = idx;
864         for (o = blocks; o; o = o->next_pool)
865                 for (e = o->entries; e != o->next_free; e++)
866                         *c++ = e;
867         last = idx + object_count;
868         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
869
870         /* Generate the fan-out array. */
871         c = idx;
872         for (i = 0; i < 256; i++) {
873                 struct object_entry **next = c;;
874                 while (next < last) {
875                         if ((*next)->sha1[0] != i)
876                                 break;
877                         next++;
878                 }
879                 array[i] = htonl(next - idx);
880                 c = next;
881         }
882
883         f = sha1create("%s", idx_name);
884         sha1write(f, array, 256 * sizeof(int));
885         for (c = idx; c != last; c++) {
886                 unsigned int offset = htonl((*c)->offset);
887                 sha1write(f, &offset, 4);
888                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
889         }
890         sha1write(f, pack_sha1, sizeof(pack_sha1));
891         sha1close(f, NULL, 1);
892         free(idx);
893 }
894
895 static void dump_branches()
896 {
897         static const char *msg = "fast-import";
898         unsigned int i;
899         struct branch *b;
900         struct ref_lock *lock;
901
902         for (i = 0; i < branch_table_sz; i++) {
903                 for (b = branch_table[i]; b; b = b->table_next_branch) {
904                         lock = lock_any_ref_for_update(b->name, NULL, 0);
905                         if (!lock || write_ref_sha1(lock, b->sha1, msg) < 0)
906                                 die("Can't write %s", b->name);
907                 }
908         }
909 }
910
911 static void read_next_command()
912 {
913         read_line(&command_buf, stdin, '\n');
914 }
915
916 static void cmd_mark()
917 {
918         if (!strncmp("mark :", command_buf.buf, 6)) {
919                 command_mark = strtoul(command_buf.buf + 6, NULL, 10);
920                 read_next_command();
921         }
922         else
923                 command_mark = 0;
924 }
925
926 static void* cmd_data (size_t *size)
927 {
928         size_t n = 0;
929         void *buffer;
930         size_t length;
931
932         if (strncmp("data ", command_buf.buf, 5))
933                 die("Expected 'data n' command, found: %s", command_buf.buf);
934
935         length = strtoul(command_buf.buf + 5, NULL, 10);
936         buffer = xmalloc(length);
937
938         while (n < length) {
939                 size_t s = fread((char*)buffer + n, 1, length - n, stdin);
940                 if (!s && feof(stdin))
941                         die("EOF in data (%lu bytes remaining)", length - n);
942                 n += s;
943         }
944
945         if (fgetc(stdin) != '\n')
946                 die("An lf did not trail the binary data as expected.");
947
948         *size = length;
949         return buffer;
950 }
951
952 static void cmd_new_blob()
953 {
954         size_t datlen;
955         void *dat;
956         unsigned char sha1[20];
957
958         read_next_command();
959         cmd_mark();
960         dat = cmd_data(&datlen);
961
962         if (store_object(OBJ_BLOB, dat, datlen, &last_blob, sha1))
963                 free(dat);
964 }
965
966 static void unload_one_branch()
967 {
968         while (cur_active_branches >= max_active_branches) {
969                 unsigned long min_commit = ULONG_MAX;
970                 struct branch *e, *l = NULL, *p = NULL;
971
972                 for (e = active_branches; e; e = e->active_next_branch) {
973                         if (e->last_commit < min_commit) {
974                                 p = l;
975                                 min_commit = e->last_commit;
976                         }
977                         l = e;
978                 }
979
980                 if (p) {
981                         e = p->active_next_branch;
982                         p->active_next_branch = e->active_next_branch;
983                 } else {
984                         e = active_branches;
985                         active_branches = e->active_next_branch;
986                 }
987                 e->active_next_branch = NULL;
988                 if (e->branch_tree.tree) {
989                         release_tree_content(e->branch_tree.tree);
990                         e->branch_tree.tree = NULL;
991                 }
992                 cur_active_branches--;
993         }
994 }
995
996 static void load_branch(struct branch *b)
997 {
998         load_tree(&b->branch_tree);
999         b->active_next_branch = active_branches;
1000         active_branches = b;
1001         cur_active_branches++;
1002 }
1003
1004 static void file_change_m(struct branch *b)
1005 {
1006         const char *p = command_buf.buf + 2;
1007         char *p_uq;
1008         const char *endp;
1009         struct object_entry *oe;
1010         unsigned char sha1[20];
1011         unsigned int mode;
1012         char type[20];
1013
1014         p = get_mode(p, &mode);
1015         if (!p)
1016                 die("Corrupt mode: %s", command_buf.buf);
1017         switch (mode) {
1018         case S_IFREG | 0644:
1019         case S_IFREG | 0755:
1020         case S_IFLNK:
1021         case 0644:
1022         case 0755:
1023                 /* ok */
1024                 break;
1025         default:
1026                 die("Corrupt mode: %s", command_buf.buf);
1027         }
1028
1029         if (get_sha1_hex(p, sha1))
1030                 die("Invalid SHA1: %s", command_buf.buf);
1031         p += 40;
1032         if (*p++ != ' ')
1033                 die("Missing space after SHA1: %s", command_buf.buf);
1034
1035         p_uq = unquote_c_style(p, &endp);
1036         if (p_uq) {
1037                 if (*endp)
1038                         die("Garbage after path in: %s", command_buf.buf);
1039                 p = p_uq;
1040         }
1041
1042         oe = find_object(sha1);
1043         if (oe) {
1044                 if (oe->type != OBJ_BLOB)
1045                         die("Not a blob (actually a %s): %s",
1046                                 command_buf.buf, type_names[oe->type]);
1047         } else {
1048                 if (sha1_object_info(sha1, type, NULL))
1049                         die("Blob not found: %s", command_buf.buf);
1050                 if (strcmp(blob_type, type))
1051                         die("Not a blob (actually a %s): %s",
1052                                 command_buf.buf, type);
1053         }
1054
1055         tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode);
1056
1057         if (p_uq)
1058                 free(p_uq);
1059 }
1060
1061 static void file_change_d(struct branch *b)
1062 {
1063         const char *p = command_buf.buf + 2;
1064         char *p_uq;
1065         const char *endp;
1066
1067         p_uq = unquote_c_style(p, &endp);
1068         if (p_uq) {
1069                 if (*endp)
1070                         die("Garbage after path in: %s", command_buf.buf);
1071                 p = p_uq;
1072         }
1073         tree_content_remove(&b->branch_tree, p);
1074         if (p_uq)
1075                 free(p_uq);
1076 }
1077
1078 static void cmd_new_commit()
1079 {
1080         struct branch *b;
1081         void *msg;
1082         size_t msglen;
1083         char *str_uq;
1084         const char *endp;
1085         char *sp;
1086         char *author = NULL;
1087         char *committer = NULL;
1088         char *body;
1089
1090         /* Obtain the branch name from the rest of our command */
1091         sp = strchr(command_buf.buf, ' ') + 1;
1092         str_uq = unquote_c_style(sp, &endp);
1093         if (str_uq) {
1094                 if (*endp)
1095                         die("Garbage after ref in: %s", command_buf.buf);
1096                 sp = str_uq;
1097         }
1098         b = lookup_branch(sp);
1099         if (!b)
1100                 die("Branch not declared: %s", sp);
1101         if (str_uq)
1102                 free(str_uq);
1103
1104         read_next_command();
1105         cmd_mark();
1106         if (!strncmp("author ", command_buf.buf, 7)) {
1107                 author = strdup(command_buf.buf);
1108                 read_next_command();
1109         }
1110         if (!strncmp("committer ", command_buf.buf, 10)) {
1111                 committer = strdup(command_buf.buf);
1112                 read_next_command();
1113         }
1114         if (!committer)
1115                 die("Expected committer but didn't get one");
1116         msg = cmd_data(&msglen);
1117
1118         /* ensure the branch is active/loaded */
1119         if (!b->branch_tree.tree) {
1120                 unload_one_branch();
1121                 load_branch(b);
1122         }
1123
1124         /* file_change* */
1125         for (;;) {
1126                 read_next_command();
1127                 if (1 == command_buf.len)
1128                         break;
1129                 else if (!strncmp("M ", command_buf.buf, 2))
1130                         file_change_m(b);
1131                 else if (!strncmp("D ", command_buf.buf, 2))
1132                         file_change_d(b);
1133                 else
1134                         die("Unsupported file_change: %s", command_buf.buf);
1135         }
1136
1137         /* build the tree and the commit */
1138         store_tree(&b->branch_tree);
1139         body = xmalloc(97 + msglen
1140                 + (author
1141                         ? strlen(author) + strlen(committer)
1142                         : 2 * strlen(committer)));
1143         sp = body;
1144         sp += sprintf(sp, "tree %s\n", sha1_to_hex(b->branch_tree.sha1));
1145         if (memcmp(b->sha1, null_sha1, 20))
1146                 sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
1147         if (author)
1148                 sp += sprintf(sp, "%s\n", author);
1149         else
1150                 sp += sprintf(sp, "author %s\n", committer + 10);
1151         sp += sprintf(sp, "%s\n\n", committer);
1152         memcpy(sp, msg, msglen);
1153         sp += msglen;
1154         if (author)
1155                 free(author);
1156         free(committer);
1157         free(msg);
1158
1159         store_object(OBJ_COMMIT, body, sp - body, NULL, b->sha1);
1160         free(body);
1161         b->last_commit = object_count_by_type[OBJ_COMMIT];
1162 }
1163
1164 static void cmd_new_branch()
1165 {
1166         struct branch *b;
1167         char *str_uq;
1168         const char *endp;
1169         char *sp;
1170
1171         /* Obtain the new branch name from the rest of our command */
1172         sp = strchr(command_buf.buf, ' ') + 1;
1173         str_uq = unquote_c_style(sp, &endp);
1174         if (str_uq) {
1175                 if (*endp)
1176                         die("Garbage after ref in: %s", command_buf.buf);
1177                 sp = str_uq;
1178         }
1179         b = new_branch(sp);
1180         if (str_uq)
1181                 free(str_uq);
1182         read_next_command();
1183
1184         /* from ... */
1185         if (!strncmp("from ", command_buf.buf, 5)) {
1186                 const char *from;
1187                 struct branch *s;
1188
1189                 from = strchr(command_buf.buf, ' ') + 1;
1190                 str_uq = unquote_c_style(from, &endp);
1191                 if (str_uq) {
1192                         if (*endp)
1193                                 die("Garbage after string in: %s", command_buf.buf);
1194                         from = str_uq;
1195                 }
1196
1197                 s = lookup_branch(from);
1198                 if (b == s)
1199                         die("Can't create a branch from itself: %s", b->name);
1200                 else if (s) {
1201                         memcpy(b->sha1, s->sha1, 20);
1202                         memcpy(b->branch_tree.sha1, s->branch_tree.sha1, 20);
1203                 } else if (!get_sha1(from, b->sha1)) {
1204                         if (!memcmp(b->sha1, null_sha1, 20))
1205                                 memcpy(b->branch_tree.sha1, null_sha1, 20);
1206                         else {
1207                                 unsigned long size;
1208                                 char *buf;
1209
1210                                 buf = read_object_with_reference(b->sha1,
1211                                         type_names[OBJ_COMMIT], &size, b->sha1);
1212                                 if (!buf || size < 46)
1213                                         die("Not a valid commit: %s", from);
1214                                 if (memcmp("tree ", buf, 5)
1215                                         || get_sha1_hex(buf + 5, b->branch_tree.sha1))
1216                                         die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1217                                 free(buf);
1218                         }
1219                 } else
1220                         die("Invalid ref name or SHA1 expression: %s", from);
1221
1222                 if (str_uq)
1223                         free(str_uq);
1224                 read_next_command();
1225         } else {
1226                 memcpy(b->sha1, null_sha1, 20);
1227                 memcpy(b->branch_tree.sha1, null_sha1, 20);
1228         }
1229
1230         if (command_buf.eof || command_buf.len > 1)
1231                 die("An lf did not terminate the branch command as expected.");
1232 }
1233
1234 int main(int argc, const char **argv)
1235 {
1236         const char *base_name = argv[1];
1237         int est_obj_cnt = atoi(argv[2]);
1238         char *pack_name;
1239         char *idx_name;
1240         struct stat sb;
1241
1242         setup_ident();
1243         git_config(git_default_config);
1244
1245         pack_name = xmalloc(strlen(base_name) + 6);
1246         sprintf(pack_name, "%s.pack", base_name);
1247         idx_name = xmalloc(strlen(base_name) + 5);
1248         sprintf(idx_name, "%s.idx", base_name);
1249
1250         pack_fd = open(pack_name, O_RDWR|O_CREAT|O_EXCL, 0666);
1251         if (pack_fd < 0)
1252                 die("Can't create %s: %s", pack_name, strerror(errno));
1253
1254         init_pack_header();
1255         alloc_objects(est_obj_cnt);
1256         strbuf_init(&command_buf);
1257
1258         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
1259         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
1260         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
1261
1262         for (;;) {
1263                 read_next_command();
1264                 if (command_buf.eof)
1265                         break;
1266                 else if (!strcmp("blob", command_buf.buf))
1267                         cmd_new_blob();
1268                 else if (!strncmp("branch ", command_buf.buf, 7))
1269                         cmd_new_branch();
1270                 else if (!strncmp("commit ", command_buf.buf, 7))
1271                         cmd_new_commit();
1272                 else
1273                         die("Unsupported command: %s", command_buf.buf);
1274         }
1275
1276         fixup_header_footer();
1277         close(pack_fd);
1278         write_index(idx_name);
1279         dump_branches();
1280
1281         fprintf(stderr, "%s statistics:\n", argv[0]);
1282         fprintf(stderr, "---------------------------------------------------\n");
1283         fprintf(stderr, "Alloc'd objects: %10lu (%10lu overflow  )\n", alloc_count, alloc_count - est_obj_cnt);
1284         fprintf(stderr, "Total objects:   %10lu (%10lu duplicates)\n", object_count, duplicate_count);
1285         fprintf(stderr, "      blobs  :   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB]);
1286         fprintf(stderr, "      trees  :   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE]);
1287         fprintf(stderr, "      commits:   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT]);
1288         fprintf(stderr, "      tags   :   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG]);
1289         fprintf(stderr, "Total branches:  %10lu\n", branch_count);
1290         fprintf(stderr, "Total atoms:     %10u\n", atom_cnt);
1291         fprintf(stderr, "Memory total:    %10lu KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
1292         fprintf(stderr, "       pools:    %10lu KiB\n", total_allocd/1024);
1293         fprintf(stderr, "     objects:    %10lu KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
1294         fprintf(stderr, "---------------------------------------------------\n");
1295
1296         stat(pack_name, &sb);
1297         fprintf(stderr, "Pack size:       %10lu KiB\n", (unsigned long)(sb.st_size/1024));
1298         stat(idx_name, &sb);
1299         fprintf(stderr, "Index size:      %10lu KiB\n", (unsigned long)(sb.st_size/1024));
1300
1301         fprintf(stderr, "\n");
1302
1303         return 0;
1304 }