sha1_file.c: Don't retain open fds on small packs
[git] / fast-import.c
1 /*
2 (See Documentation/git-fast-import.txt for maintained documentation.)
3 Format of STDIN stream:
4
5   stream ::= cmd*;
6
7   cmd ::= new_blob
8         | new_commit
9         | new_tag
10         | reset_branch
11         | checkpoint
12         | progress
13         ;
14
15   new_blob ::= 'blob' lf
16     mark?
17     file_content;
18   file_content ::= data;
19
20   new_commit ::= 'commit' sp ref_str lf
21     mark?
22     ('author' (sp name)? sp '<' email '>' sp when lf)?
23     'committer' (sp name)? sp '<' email '>' sp when lf
24     commit_msg
25     ('from' sp committish lf)?
26     ('merge' sp committish lf)*
27     file_change*
28     lf?;
29   commit_msg ::= data;
30
31   file_change ::= file_clr
32     | file_del
33     | file_rnm
34     | file_cpy
35     | file_obm
36     | file_inm;
37   file_clr ::= 'deleteall' lf;
38   file_del ::= 'D' sp path_str lf;
39   file_rnm ::= 'R' sp path_str sp path_str lf;
40   file_cpy ::= 'C' sp path_str sp path_str lf;
41   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
42   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
43     data;
44   note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
45   note_inm ::= 'N' sp 'inline' sp committish lf
46     data;
47
48   new_tag ::= 'tag' sp tag_str lf
49     'from' sp committish lf
50     ('tagger' (sp name)? sp '<' email '>' sp when lf)?
51     tag_msg;
52   tag_msg ::= data;
53
54   reset_branch ::= 'reset' sp ref_str lf
55     ('from' sp committish lf)?
56     lf?;
57
58   checkpoint ::= 'checkpoint' lf
59     lf?;
60
61   progress ::= 'progress' sp not_lf* lf
62     lf?;
63
64      # note: the first idnum in a stream should be 1 and subsequent
65      # idnums should not have gaps between values as this will cause
66      # the stream parser to reserve space for the gapped values.  An
67      # idnum can be updated in the future to a new object by issuing
68      # a new mark directive with the old idnum.
69      #
70   mark ::= 'mark' sp idnum lf;
71   data ::= (delimited_data | exact_data)
72     lf?;
73
74     # note: delim may be any string but must not contain lf.
75     # data_line may contain any data but must not be exactly
76     # delim.
77   delimited_data ::= 'data' sp '<<' delim lf
78     (data_line lf)*
79     delim lf;
80
81      # note: declen indicates the length of binary_data in bytes.
82      # declen does not include the lf preceding the binary data.
83      #
84   exact_data ::= 'data' sp declen lf
85     binary_data;
86
87      # note: quoted strings are C-style quoting supporting \c for
88      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
89      # is the signed byte value in octal.  Note that the only
90      # characters which must actually be escaped to protect the
91      # stream formatting is: \, " and LF.  Otherwise these values
92      # are UTF8.
93      #
94   committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
95   ref_str     ::= ref;
96   sha1exp_str ::= sha1exp;
97   tag_str     ::= tag;
98   path_str    ::= path    | '"' quoted(path)    '"' ;
99   mode        ::= '100644' | '644'
100                 | '100755' | '755'
101                 | '120000'
102                 ;
103
104   declen ::= # unsigned 32 bit value, ascii base10 notation;
105   bigint ::= # unsigned integer value, ascii base10 notation;
106   binary_data ::= # file content, not interpreted;
107
108   when         ::= raw_when | rfc2822_when;
109   raw_when     ::= ts sp tz;
110   rfc2822_when ::= # Valid RFC 2822 date and time;
111
112   sp ::= # ASCII space character;
113   lf ::= # ASCII newline (LF) character;
114
115      # note: a colon (':') must precede the numerical value assigned to
116      # an idnum.  This is to distinguish it from a ref or tag name as
117      # GIT does not permit ':' in ref or tag strings.
118      #
119   idnum   ::= ':' bigint;
120   path    ::= # GIT style file path, e.g. "a/b/c";
121   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
122   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
123   sha1exp ::= # Any valid GIT SHA1 expression;
124   hexsha1 ::= # SHA1 in hexadecimal format;
125
126      # note: name and email are UTF8 strings, however name must not
127      # contain '<' or lf and email must not contain any of the
128      # following: '<', '>', lf.
129      #
130   name  ::= # valid GIT author/committer name;
131   email ::= # valid GIT author/committer email;
132   ts    ::= # time since the epoch in seconds, ascii base10 notation;
133   tz    ::= # GIT style timezone;
134
135      # note: comments and cat requests may appear anywhere
136      # in the input, except within a data command.  Any form
137      # of the data command always escapes the related input
138      # from comment processing.
139      #
140      # In case it is not clear, the '#' that starts the comment
141      # must be the first character on that line (an lf
142      # preceded it).
143      #
144   cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
145
146   comment ::= '#' not_lf* lf;
147   not_lf  ::= # Any byte that is not ASCII newline (LF);
148 */
149
150 #include "builtin.h"
151 #include "cache.h"
152 #include "object.h"
153 #include "blob.h"
154 #include "tree.h"
155 #include "commit.h"
156 #include "delta.h"
157 #include "pack.h"
158 #include "refs.h"
159 #include "csum-file.h"
160 #include "quote.h"
161 #include "exec_cmd.h"
162 #include "dir.h"
163
164 #define PACK_ID_BITS 16
165 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
166 #define DEPTH_BITS 13
167 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
168
169 struct object_entry
170 {
171         struct pack_idx_entry idx;
172         struct object_entry *next;
173         uint32_t type : TYPE_BITS,
174                 pack_id : PACK_ID_BITS,
175                 depth : DEPTH_BITS;
176 };
177
178 struct object_entry_pool
179 {
180         struct object_entry_pool *next_pool;
181         struct object_entry *next_free;
182         struct object_entry *end;
183         struct object_entry entries[FLEX_ARRAY]; /* more */
184 };
185
186 struct mark_set
187 {
188         union {
189                 struct object_entry *marked[1024];
190                 struct mark_set *sets[1024];
191         } data;
192         unsigned int shift;
193 };
194
195 struct last_object
196 {
197         struct strbuf data;
198         off_t offset;
199         unsigned int depth;
200         unsigned no_swap : 1;
201 };
202
203 struct mem_pool
204 {
205         struct mem_pool *next_pool;
206         char *next_free;
207         char *end;
208         uintmax_t space[FLEX_ARRAY]; /* more */
209 };
210
211 struct atom_str
212 {
213         struct atom_str *next_atom;
214         unsigned short str_len;
215         char str_dat[FLEX_ARRAY]; /* more */
216 };
217
218 struct tree_content;
219 struct tree_entry
220 {
221         struct tree_content *tree;
222         struct atom_str *name;
223         struct tree_entry_ms
224         {
225                 uint16_t mode;
226                 unsigned char sha1[20];
227         } versions[2];
228 };
229
230 struct tree_content
231 {
232         unsigned int entry_capacity; /* must match avail_tree_content */
233         unsigned int entry_count;
234         unsigned int delta_depth;
235         struct tree_entry *entries[FLEX_ARRAY]; /* more */
236 };
237
238 struct avail_tree_content
239 {
240         unsigned int entry_capacity; /* must match tree_content */
241         struct avail_tree_content *next_avail;
242 };
243
244 struct branch
245 {
246         struct branch *table_next_branch;
247         struct branch *active_next_branch;
248         const char *name;
249         struct tree_entry branch_tree;
250         uintmax_t last_commit;
251         uintmax_t num_notes;
252         unsigned active : 1;
253         unsigned pack_id : PACK_ID_BITS;
254         unsigned char sha1[20];
255 };
256
257 struct tag
258 {
259         struct tag *next_tag;
260         const char *name;
261         unsigned int pack_id;
262         unsigned char sha1[20];
263 };
264
265 struct hash_list
266 {
267         struct hash_list *next;
268         unsigned char sha1[20];
269 };
270
271 typedef enum {
272         WHENSPEC_RAW = 1,
273         WHENSPEC_RFC2822,
274         WHENSPEC_NOW
275 } whenspec_type;
276
277 struct recent_command
278 {
279         struct recent_command *prev;
280         struct recent_command *next;
281         char *buf;
282 };
283
284 /* Configured limits on output */
285 static unsigned long max_depth = 10;
286 static off_t max_packsize;
287 static uintmax_t big_file_threshold = 512 * 1024 * 1024;
288 static int force_update;
289 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
290 static int pack_compression_seen;
291
292 /* Stats and misc. counters */
293 static uintmax_t alloc_count;
294 static uintmax_t marks_set_count;
295 static uintmax_t object_count_by_type[1 << TYPE_BITS];
296 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
297 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
298 static unsigned long object_count;
299 static unsigned long branch_count;
300 static unsigned long branch_load_count;
301 static int failure;
302 static FILE *pack_edges;
303 static unsigned int show_stats = 1;
304 static int global_argc;
305 static const char **global_argv;
306
307 /* Memory pools */
308 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
309 static size_t total_allocd;
310 static struct mem_pool *mem_pool;
311
312 /* Atom management */
313 static unsigned int atom_table_sz = 4451;
314 static unsigned int atom_cnt;
315 static struct atom_str **atom_table;
316
317 /* The .pack file being generated */
318 static unsigned int pack_id;
319 static struct sha1file *pack_file;
320 static struct packed_git *pack_data;
321 static struct packed_git **all_packs;
322 static off_t pack_size;
323
324 /* Table of objects we've written. */
325 static unsigned int object_entry_alloc = 5000;
326 static struct object_entry_pool *blocks;
327 static struct object_entry *object_table[1 << 16];
328 static struct mark_set *marks;
329 static const char *export_marks_file;
330 static const char *import_marks_file;
331 static int import_marks_file_from_stream;
332 static int relative_marks_paths;
333
334 /* Our last blob */
335 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
336
337 /* Tree management */
338 static unsigned int tree_entry_alloc = 1000;
339 static void *avail_tree_entry;
340 static unsigned int avail_tree_table_sz = 100;
341 static struct avail_tree_content **avail_tree_table;
342 static struct strbuf old_tree = STRBUF_INIT;
343 static struct strbuf new_tree = STRBUF_INIT;
344
345 /* Branch data */
346 static unsigned long max_active_branches = 5;
347 static unsigned long cur_active_branches;
348 static unsigned long branch_table_sz = 1039;
349 static struct branch **branch_table;
350 static struct branch *active_branches;
351
352 /* Tag data */
353 static struct tag *first_tag;
354 static struct tag *last_tag;
355
356 /* Input stream parsing */
357 static whenspec_type whenspec = WHENSPEC_RAW;
358 static struct strbuf command_buf = STRBUF_INIT;
359 static int unread_command_buf;
360 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
361 static struct recent_command *cmd_tail = &cmd_hist;
362 static struct recent_command *rc_free;
363 static unsigned int cmd_save = 100;
364 static uintmax_t next_mark;
365 static struct strbuf new_data = STRBUF_INIT;
366 static int seen_data_command;
367
368 /* Signal handling */
369 static volatile sig_atomic_t checkpoint_requested;
370
371 /* Where to write output of cat-blob commands */
372 static int cat_blob_fd = STDOUT_FILENO;
373
374 static void parse_argv(void);
375 static void parse_cat_blob(void);
376
377 static void write_branch_report(FILE *rpt, struct branch *b)
378 {
379         fprintf(rpt, "%s:\n", b->name);
380
381         fprintf(rpt, "  status      :");
382         if (b->active)
383                 fputs(" active", rpt);
384         if (b->branch_tree.tree)
385                 fputs(" loaded", rpt);
386         if (is_null_sha1(b->branch_tree.versions[1].sha1))
387                 fputs(" dirty", rpt);
388         fputc('\n', rpt);
389
390         fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
391         fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
392         fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
393         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
394
395         fputs("  last pack   : ", rpt);
396         if (b->pack_id < MAX_PACK_ID)
397                 fprintf(rpt, "%u", b->pack_id);
398         fputc('\n', rpt);
399
400         fputc('\n', rpt);
401 }
402
403 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
404
405 static void write_crash_report(const char *err)
406 {
407         char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
408         FILE *rpt = fopen(loc, "w");
409         struct branch *b;
410         unsigned long lu;
411         struct recent_command *rc;
412
413         if (!rpt) {
414                 error("can't write crash report %s: %s", loc, strerror(errno));
415                 return;
416         }
417
418         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
419
420         fprintf(rpt, "fast-import crash report:\n");
421         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
422         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
423         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
424         fputc('\n', rpt);
425
426         fputs("fatal: ", rpt);
427         fputs(err, rpt);
428         fputc('\n', rpt);
429
430         fputc('\n', rpt);
431         fputs("Most Recent Commands Before Crash\n", rpt);
432         fputs("---------------------------------\n", rpt);
433         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
434                 if (rc->next == &cmd_hist)
435                         fputs("* ", rpt);
436                 else
437                         fputs("  ", rpt);
438                 fputs(rc->buf, rpt);
439                 fputc('\n', rpt);
440         }
441
442         fputc('\n', rpt);
443         fputs("Active Branch LRU\n", rpt);
444         fputs("-----------------\n", rpt);
445         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
446                 cur_active_branches,
447                 max_active_branches);
448         fputc('\n', rpt);
449         fputs("  pos  clock name\n", rpt);
450         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
451         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
452                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
453                         ++lu, b->last_commit, b->name);
454
455         fputc('\n', rpt);
456         fputs("Inactive Branches\n", rpt);
457         fputs("-----------------\n", rpt);
458         for (lu = 0; lu < branch_table_sz; lu++) {
459                 for (b = branch_table[lu]; b; b = b->table_next_branch)
460                         write_branch_report(rpt, b);
461         }
462
463         if (first_tag) {
464                 struct tag *tg;
465                 fputc('\n', rpt);
466                 fputs("Annotated Tags\n", rpt);
467                 fputs("--------------\n", rpt);
468                 for (tg = first_tag; tg; tg = tg->next_tag) {
469                         fputs(sha1_to_hex(tg->sha1), rpt);
470                         fputc(' ', rpt);
471                         fputs(tg->name, rpt);
472                         fputc('\n', rpt);
473                 }
474         }
475
476         fputc('\n', rpt);
477         fputs("Marks\n", rpt);
478         fputs("-----\n", rpt);
479         if (export_marks_file)
480                 fprintf(rpt, "  exported to %s\n", export_marks_file);
481         else
482                 dump_marks_helper(rpt, 0, marks);
483
484         fputc('\n', rpt);
485         fputs("-------------------\n", rpt);
486         fputs("END OF CRASH REPORT\n", rpt);
487         fclose(rpt);
488 }
489
490 static void end_packfile(void);
491 static void unkeep_all_packs(void);
492 static void dump_marks(void);
493
494 static NORETURN void die_nicely(const char *err, va_list params)
495 {
496         static int zombie;
497         char message[2 * PATH_MAX];
498
499         vsnprintf(message, sizeof(message), err, params);
500         fputs("fatal: ", stderr);
501         fputs(message, stderr);
502         fputc('\n', stderr);
503
504         if (!zombie) {
505                 zombie = 1;
506                 write_crash_report(message);
507                 end_packfile();
508                 unkeep_all_packs();
509                 dump_marks();
510         }
511         exit(128);
512 }
513
514 #ifndef SIGUSR1 /* Windows, for example */
515
516 static void set_checkpoint_signal(void)
517 {
518 }
519
520 #else
521
522 static void checkpoint_signal(int signo)
523 {
524         checkpoint_requested = 1;
525 }
526
527 static void set_checkpoint_signal(void)
528 {
529         struct sigaction sa;
530
531         memset(&sa, 0, sizeof(sa));
532         sa.sa_handler = checkpoint_signal;
533         sigemptyset(&sa.sa_mask);
534         sa.sa_flags = SA_RESTART;
535         sigaction(SIGUSR1, &sa, NULL);
536 }
537
538 #endif
539
540 static void alloc_objects(unsigned int cnt)
541 {
542         struct object_entry_pool *b;
543
544         b = xmalloc(sizeof(struct object_entry_pool)
545                 + cnt * sizeof(struct object_entry));
546         b->next_pool = blocks;
547         b->next_free = b->entries;
548         b->end = b->entries + cnt;
549         blocks = b;
550         alloc_count += cnt;
551 }
552
553 static struct object_entry *new_object(unsigned char *sha1)
554 {
555         struct object_entry *e;
556
557         if (blocks->next_free == blocks->end)
558                 alloc_objects(object_entry_alloc);
559
560         e = blocks->next_free++;
561         hashcpy(e->idx.sha1, sha1);
562         return e;
563 }
564
565 static struct object_entry *find_object(unsigned char *sha1)
566 {
567         unsigned int h = sha1[0] << 8 | sha1[1];
568         struct object_entry *e;
569         for (e = object_table[h]; e; e = e->next)
570                 if (!hashcmp(sha1, e->idx.sha1))
571                         return e;
572         return NULL;
573 }
574
575 static struct object_entry *insert_object(unsigned char *sha1)
576 {
577         unsigned int h = sha1[0] << 8 | sha1[1];
578         struct object_entry *e = object_table[h];
579
580         while (e) {
581                 if (!hashcmp(sha1, e->idx.sha1))
582                         return e;
583                 e = e->next;
584         }
585
586         e = new_object(sha1);
587         e->next = object_table[h];
588         e->idx.offset = 0;
589         object_table[h] = e;
590         return e;
591 }
592
593 static unsigned int hc_str(const char *s, size_t len)
594 {
595         unsigned int r = 0;
596         while (len-- > 0)
597                 r = r * 31 + *s++;
598         return r;
599 }
600
601 static void *pool_alloc(size_t len)
602 {
603         struct mem_pool *p;
604         void *r;
605
606         /* round up to a 'uintmax_t' alignment */
607         if (len & (sizeof(uintmax_t) - 1))
608                 len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
609
610         for (p = mem_pool; p; p = p->next_pool)
611                 if ((p->end - p->next_free >= len))
612                         break;
613
614         if (!p) {
615                 if (len >= (mem_pool_alloc/2)) {
616                         total_allocd += len;
617                         return xmalloc(len);
618                 }
619                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
620                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
621                 p->next_pool = mem_pool;
622                 p->next_free = (char *) p->space;
623                 p->end = p->next_free + mem_pool_alloc;
624                 mem_pool = p;
625         }
626
627         r = p->next_free;
628         p->next_free += len;
629         return r;
630 }
631
632 static void *pool_calloc(size_t count, size_t size)
633 {
634         size_t len = count * size;
635         void *r = pool_alloc(len);
636         memset(r, 0, len);
637         return r;
638 }
639
640 static char *pool_strdup(const char *s)
641 {
642         char *r = pool_alloc(strlen(s) + 1);
643         strcpy(r, s);
644         return r;
645 }
646
647 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
648 {
649         struct mark_set *s = marks;
650         while ((idnum >> s->shift) >= 1024) {
651                 s = pool_calloc(1, sizeof(struct mark_set));
652                 s->shift = marks->shift + 10;
653                 s->data.sets[0] = marks;
654                 marks = s;
655         }
656         while (s->shift) {
657                 uintmax_t i = idnum >> s->shift;
658                 idnum -= i << s->shift;
659                 if (!s->data.sets[i]) {
660                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
661                         s->data.sets[i]->shift = s->shift - 10;
662                 }
663                 s = s->data.sets[i];
664         }
665         if (!s->data.marked[idnum])
666                 marks_set_count++;
667         s->data.marked[idnum] = oe;
668 }
669
670 static struct object_entry *find_mark(uintmax_t idnum)
671 {
672         uintmax_t orig_idnum = idnum;
673         struct mark_set *s = marks;
674         struct object_entry *oe = NULL;
675         if ((idnum >> s->shift) < 1024) {
676                 while (s && s->shift) {
677                         uintmax_t i = idnum >> s->shift;
678                         idnum -= i << s->shift;
679                         s = s->data.sets[i];
680                 }
681                 if (s)
682                         oe = s->data.marked[idnum];
683         }
684         if (!oe)
685                 die("mark :%" PRIuMAX " not declared", orig_idnum);
686         return oe;
687 }
688
689 static struct atom_str *to_atom(const char *s, unsigned short len)
690 {
691         unsigned int hc = hc_str(s, len) % atom_table_sz;
692         struct atom_str *c;
693
694         for (c = atom_table[hc]; c; c = c->next_atom)
695                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
696                         return c;
697
698         c = pool_alloc(sizeof(struct atom_str) + len + 1);
699         c->str_len = len;
700         strncpy(c->str_dat, s, len);
701         c->str_dat[len] = 0;
702         c->next_atom = atom_table[hc];
703         atom_table[hc] = c;
704         atom_cnt++;
705         return c;
706 }
707
708 static struct branch *lookup_branch(const char *name)
709 {
710         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
711         struct branch *b;
712
713         for (b = branch_table[hc]; b; b = b->table_next_branch)
714                 if (!strcmp(name, b->name))
715                         return b;
716         return NULL;
717 }
718
719 static struct branch *new_branch(const char *name)
720 {
721         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
722         struct branch *b = lookup_branch(name);
723
724         if (b)
725                 die("Invalid attempt to create duplicate branch: %s", name);
726         switch (check_ref_format(name)) {
727         case 0: break; /* its valid */
728         case CHECK_REF_FORMAT_ONELEVEL:
729                 break; /* valid, but too few '/', allow anyway */
730         default:
731                 die("Branch name doesn't conform to GIT standards: %s", name);
732         }
733
734         b = pool_calloc(1, sizeof(struct branch));
735         b->name = pool_strdup(name);
736         b->table_next_branch = branch_table[hc];
737         b->branch_tree.versions[0].mode = S_IFDIR;
738         b->branch_tree.versions[1].mode = S_IFDIR;
739         b->num_notes = 0;
740         b->active = 0;
741         b->pack_id = MAX_PACK_ID;
742         branch_table[hc] = b;
743         branch_count++;
744         return b;
745 }
746
747 static unsigned int hc_entries(unsigned int cnt)
748 {
749         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
750         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
751 }
752
753 static struct tree_content *new_tree_content(unsigned int cnt)
754 {
755         struct avail_tree_content *f, *l = NULL;
756         struct tree_content *t;
757         unsigned int hc = hc_entries(cnt);
758
759         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
760                 if (f->entry_capacity >= cnt)
761                         break;
762
763         if (f) {
764                 if (l)
765                         l->next_avail = f->next_avail;
766                 else
767                         avail_tree_table[hc] = f->next_avail;
768         } else {
769                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
770                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
771                 f->entry_capacity = cnt;
772         }
773
774         t = (struct tree_content*)f;
775         t->entry_count = 0;
776         t->delta_depth = 0;
777         return t;
778 }
779
780 static void release_tree_entry(struct tree_entry *e);
781 static void release_tree_content(struct tree_content *t)
782 {
783         struct avail_tree_content *f = (struct avail_tree_content*)t;
784         unsigned int hc = hc_entries(f->entry_capacity);
785         f->next_avail = avail_tree_table[hc];
786         avail_tree_table[hc] = f;
787 }
788
789 static void release_tree_content_recursive(struct tree_content *t)
790 {
791         unsigned int i;
792         for (i = 0; i < t->entry_count; i++)
793                 release_tree_entry(t->entries[i]);
794         release_tree_content(t);
795 }
796
797 static struct tree_content *grow_tree_content(
798         struct tree_content *t,
799         int amt)
800 {
801         struct tree_content *r = new_tree_content(t->entry_count + amt);
802         r->entry_count = t->entry_count;
803         r->delta_depth = t->delta_depth;
804         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
805         release_tree_content(t);
806         return r;
807 }
808
809 static struct tree_entry *new_tree_entry(void)
810 {
811         struct tree_entry *e;
812
813         if (!avail_tree_entry) {
814                 unsigned int n = tree_entry_alloc;
815                 total_allocd += n * sizeof(struct tree_entry);
816                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
817                 while (n-- > 1) {
818                         *((void**)e) = e + 1;
819                         e++;
820                 }
821                 *((void**)e) = NULL;
822         }
823
824         e = avail_tree_entry;
825         avail_tree_entry = *((void**)e);
826         return e;
827 }
828
829 static void release_tree_entry(struct tree_entry *e)
830 {
831         if (e->tree)
832                 release_tree_content_recursive(e->tree);
833         *((void**)e) = avail_tree_entry;
834         avail_tree_entry = e;
835 }
836
837 static struct tree_content *dup_tree_content(struct tree_content *s)
838 {
839         struct tree_content *d;
840         struct tree_entry *a, *b;
841         unsigned int i;
842
843         if (!s)
844                 return NULL;
845         d = new_tree_content(s->entry_count);
846         for (i = 0; i < s->entry_count; i++) {
847                 a = s->entries[i];
848                 b = new_tree_entry();
849                 memcpy(b, a, sizeof(*a));
850                 if (a->tree && is_null_sha1(b->versions[1].sha1))
851                         b->tree = dup_tree_content(a->tree);
852                 else
853                         b->tree = NULL;
854                 d->entries[i] = b;
855         }
856         d->entry_count = s->entry_count;
857         d->delta_depth = s->delta_depth;
858
859         return d;
860 }
861
862 static void start_packfile(void)
863 {
864         static char tmpfile[PATH_MAX];
865         struct packed_git *p;
866         struct pack_header hdr;
867         int pack_fd;
868
869         pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
870                               "pack/tmp_pack_XXXXXX");
871         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
872         strcpy(p->pack_name, tmpfile);
873         p->pack_fd = pack_fd;
874         p->do_not_close = 1;
875         pack_file = sha1fd(pack_fd, p->pack_name);
876
877         hdr.hdr_signature = htonl(PACK_SIGNATURE);
878         hdr.hdr_version = htonl(2);
879         hdr.hdr_entries = 0;
880         sha1write(pack_file, &hdr, sizeof(hdr));
881
882         pack_data = p;
883         pack_size = sizeof(hdr);
884         object_count = 0;
885
886         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
887         all_packs[pack_id] = p;
888 }
889
890 static const char *create_index(void)
891 {
892         const char *tmpfile;
893         struct pack_idx_entry **idx, **c, **last;
894         struct object_entry *e;
895         struct object_entry_pool *o;
896
897         /* Build the table of object IDs. */
898         idx = xmalloc(object_count * sizeof(*idx));
899         c = idx;
900         for (o = blocks; o; o = o->next_pool)
901                 for (e = o->next_free; e-- != o->entries;)
902                         if (pack_id == e->pack_id)
903                                 *c++ = &e->idx;
904         last = idx + object_count;
905         if (c != last)
906                 die("internal consistency error creating the index");
907
908         tmpfile = write_idx_file(NULL, idx, object_count, pack_data->sha1);
909         free(idx);
910         return tmpfile;
911 }
912
913 static char *keep_pack(const char *curr_index_name)
914 {
915         static char name[PATH_MAX];
916         static const char *keep_msg = "fast-import";
917         int keep_fd;
918
919         keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
920         if (keep_fd < 0)
921                 die_errno("cannot create keep file");
922         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
923         if (close(keep_fd))
924                 die_errno("failed to write keep file");
925
926         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
927                  get_object_directory(), sha1_to_hex(pack_data->sha1));
928         if (move_temp_to_file(pack_data->pack_name, name))
929                 die("cannot store pack file");
930
931         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
932                  get_object_directory(), sha1_to_hex(pack_data->sha1));
933         if (move_temp_to_file(curr_index_name, name))
934                 die("cannot store index file");
935         free((void *)curr_index_name);
936         return name;
937 }
938
939 static void unkeep_all_packs(void)
940 {
941         static char name[PATH_MAX];
942         int k;
943
944         for (k = 0; k < pack_id; k++) {
945                 struct packed_git *p = all_packs[k];
946                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
947                          get_object_directory(), sha1_to_hex(p->sha1));
948                 unlink_or_warn(name);
949         }
950 }
951
952 static void end_packfile(void)
953 {
954         struct packed_git *old_p = pack_data, *new_p;
955
956         clear_delta_base_cache();
957         if (object_count) {
958                 unsigned char cur_pack_sha1[20];
959                 char *idx_name;
960                 int i;
961                 struct branch *b;
962                 struct tag *t;
963
964                 close_pack_windows(pack_data);
965                 sha1close(pack_file, cur_pack_sha1, 0);
966                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
967                                     pack_data->pack_name, object_count,
968                                     cur_pack_sha1, pack_size);
969                 close(pack_data->pack_fd);
970                 idx_name = keep_pack(create_index());
971
972                 /* Register the packfile with core git's machinery. */
973                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
974                 if (!new_p)
975                         die("core git rejected index %s", idx_name);
976                 all_packs[pack_id] = new_p;
977                 install_packed_git(new_p);
978
979                 /* Print the boundary */
980                 if (pack_edges) {
981                         fprintf(pack_edges, "%s:", new_p->pack_name);
982                         for (i = 0; i < branch_table_sz; i++) {
983                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
984                                         if (b->pack_id == pack_id)
985                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
986                                 }
987                         }
988                         for (t = first_tag; t; t = t->next_tag) {
989                                 if (t->pack_id == pack_id)
990                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
991                         }
992                         fputc('\n', pack_edges);
993                         fflush(pack_edges);
994                 }
995
996                 pack_id++;
997         }
998         else {
999                 close(old_p->pack_fd);
1000                 unlink_or_warn(old_p->pack_name);
1001         }
1002         free(old_p);
1003
1004         /* We can't carry a delta across packfiles. */
1005         strbuf_release(&last_blob.data);
1006         last_blob.offset = 0;
1007         last_blob.depth = 0;
1008 }
1009
1010 static void cycle_packfile(void)
1011 {
1012         end_packfile();
1013         start_packfile();
1014 }
1015
1016 static int store_object(
1017         enum object_type type,
1018         struct strbuf *dat,
1019         struct last_object *last,
1020         unsigned char *sha1out,
1021         uintmax_t mark)
1022 {
1023         void *out, *delta;
1024         struct object_entry *e;
1025         unsigned char hdr[96];
1026         unsigned char sha1[20];
1027         unsigned long hdrlen, deltalen;
1028         git_SHA_CTX c;
1029         z_stream s;
1030
1031         hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1032                 (unsigned long)dat->len) + 1;
1033         git_SHA1_Init(&c);
1034         git_SHA1_Update(&c, hdr, hdrlen);
1035         git_SHA1_Update(&c, dat->buf, dat->len);
1036         git_SHA1_Final(sha1, &c);
1037         if (sha1out)
1038                 hashcpy(sha1out, sha1);
1039
1040         e = insert_object(sha1);
1041         if (mark)
1042                 insert_mark(mark, e);
1043         if (e->idx.offset) {
1044                 duplicate_count_by_type[type]++;
1045                 return 1;
1046         } else if (find_sha1_pack(sha1, packed_git)) {
1047                 e->type = type;
1048                 e->pack_id = MAX_PACK_ID;
1049                 e->idx.offset = 1; /* just not zero! */
1050                 duplicate_count_by_type[type]++;
1051                 return 1;
1052         }
1053
1054         if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1055                 delta = diff_delta(last->data.buf, last->data.len,
1056                         dat->buf, dat->len,
1057                         &deltalen, dat->len - 20);
1058         } else
1059                 delta = NULL;
1060
1061         memset(&s, 0, sizeof(s));
1062         deflateInit(&s, pack_compression_level);
1063         if (delta) {
1064                 s.next_in = delta;
1065                 s.avail_in = deltalen;
1066         } else {
1067                 s.next_in = (void *)dat->buf;
1068                 s.avail_in = dat->len;
1069         }
1070         s.avail_out = deflateBound(&s, s.avail_in);
1071         s.next_out = out = xmalloc(s.avail_out);
1072         while (deflate(&s, Z_FINISH) == Z_OK)
1073                 /* nothing */;
1074         deflateEnd(&s);
1075
1076         /* Determine if we should auto-checkpoint. */
1077         if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1078                 || (pack_size + 60 + s.total_out) < pack_size) {
1079
1080                 /* This new object needs to *not* have the current pack_id. */
1081                 e->pack_id = pack_id + 1;
1082                 cycle_packfile();
1083
1084                 /* We cannot carry a delta into the new pack. */
1085                 if (delta) {
1086                         free(delta);
1087                         delta = NULL;
1088
1089                         memset(&s, 0, sizeof(s));
1090                         deflateInit(&s, pack_compression_level);
1091                         s.next_in = (void *)dat->buf;
1092                         s.avail_in = dat->len;
1093                         s.avail_out = deflateBound(&s, s.avail_in);
1094                         s.next_out = out = xrealloc(out, s.avail_out);
1095                         while (deflate(&s, Z_FINISH) == Z_OK)
1096                                 /* nothing */;
1097                         deflateEnd(&s);
1098                 }
1099         }
1100
1101         e->type = type;
1102         e->pack_id = pack_id;
1103         e->idx.offset = pack_size;
1104         object_count++;
1105         object_count_by_type[type]++;
1106
1107         crc32_begin(pack_file);
1108
1109         if (delta) {
1110                 off_t ofs = e->idx.offset - last->offset;
1111                 unsigned pos = sizeof(hdr) - 1;
1112
1113                 delta_count_by_type[type]++;
1114                 e->depth = last->depth + 1;
1115
1116                 hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1117                 sha1write(pack_file, hdr, hdrlen);
1118                 pack_size += hdrlen;
1119
1120                 hdr[pos] = ofs & 127;
1121                 while (ofs >>= 7)
1122                         hdr[--pos] = 128 | (--ofs & 127);
1123                 sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1124                 pack_size += sizeof(hdr) - pos;
1125         } else {
1126                 e->depth = 0;
1127                 hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1128                 sha1write(pack_file, hdr, hdrlen);
1129                 pack_size += hdrlen;
1130         }
1131
1132         sha1write(pack_file, out, s.total_out);
1133         pack_size += s.total_out;
1134
1135         e->idx.crc32 = crc32_end(pack_file);
1136
1137         free(out);
1138         free(delta);
1139         if (last) {
1140                 if (last->no_swap) {
1141                         last->data = *dat;
1142                 } else {
1143                         strbuf_swap(&last->data, dat);
1144                 }
1145                 last->offset = e->idx.offset;
1146                 last->depth = e->depth;
1147         }
1148         return 0;
1149 }
1150
1151 static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1152 {
1153         if (ftruncate(pack_data->pack_fd, to)
1154          || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1155                 die_errno("cannot truncate pack to skip duplicate");
1156         pack_size = to;
1157
1158         /* yes this is a layering violation */
1159         pack_file->total = to;
1160         pack_file->offset = 0;
1161         pack_file->ctx = *ctx;
1162 }
1163
1164 static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1165 {
1166         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1167         unsigned char *in_buf = xmalloc(in_sz);
1168         unsigned char *out_buf = xmalloc(out_sz);
1169         struct object_entry *e;
1170         unsigned char sha1[20];
1171         unsigned long hdrlen;
1172         off_t offset;
1173         git_SHA_CTX c;
1174         git_SHA_CTX pack_file_ctx;
1175         z_stream s;
1176         int status = Z_OK;
1177
1178         /* Determine if we should auto-checkpoint. */
1179         if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1180                 || (pack_size + 60 + len) < pack_size)
1181                 cycle_packfile();
1182
1183         offset = pack_size;
1184
1185         /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1186         sha1flush(pack_file);
1187         pack_file_ctx = pack_file->ctx;
1188
1189         hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1190         if (out_sz <= hdrlen)
1191                 die("impossibly large object header");
1192
1193         git_SHA1_Init(&c);
1194         git_SHA1_Update(&c, out_buf, hdrlen);
1195
1196         crc32_begin(pack_file);
1197
1198         memset(&s, 0, sizeof(s));
1199         deflateInit(&s, pack_compression_level);
1200
1201         hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1202         if (out_sz <= hdrlen)
1203                 die("impossibly large object header");
1204
1205         s.next_out = out_buf + hdrlen;
1206         s.avail_out = out_sz - hdrlen;
1207
1208         while (status != Z_STREAM_END) {
1209                 if (0 < len && !s.avail_in) {
1210                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1211                         size_t n = fread(in_buf, 1, cnt, stdin);
1212                         if (!n && feof(stdin))
1213                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1214
1215                         git_SHA1_Update(&c, in_buf, n);
1216                         s.next_in = in_buf;
1217                         s.avail_in = n;
1218                         len -= n;
1219                 }
1220
1221                 status = deflate(&s, len ? 0 : Z_FINISH);
1222
1223                 if (!s.avail_out || status == Z_STREAM_END) {
1224                         size_t n = s.next_out - out_buf;
1225                         sha1write(pack_file, out_buf, n);
1226                         pack_size += n;
1227                         s.next_out = out_buf;
1228                         s.avail_out = out_sz;
1229                 }
1230
1231                 switch (status) {
1232                 case Z_OK:
1233                 case Z_BUF_ERROR:
1234                 case Z_STREAM_END:
1235                         continue;
1236                 default:
1237                         die("unexpected deflate failure: %d", status);
1238                 }
1239         }
1240         deflateEnd(&s);
1241         git_SHA1_Final(sha1, &c);
1242
1243         if (sha1out)
1244                 hashcpy(sha1out, sha1);
1245
1246         e = insert_object(sha1);
1247
1248         if (mark)
1249                 insert_mark(mark, e);
1250
1251         if (e->idx.offset) {
1252                 duplicate_count_by_type[OBJ_BLOB]++;
1253                 truncate_pack(offset, &pack_file_ctx);
1254
1255         } else if (find_sha1_pack(sha1, packed_git)) {
1256                 e->type = OBJ_BLOB;
1257                 e->pack_id = MAX_PACK_ID;
1258                 e->idx.offset = 1; /* just not zero! */
1259                 duplicate_count_by_type[OBJ_BLOB]++;
1260                 truncate_pack(offset, &pack_file_ctx);
1261
1262         } else {
1263                 e->depth = 0;
1264                 e->type = OBJ_BLOB;
1265                 e->pack_id = pack_id;
1266                 e->idx.offset = offset;
1267                 e->idx.crc32 = crc32_end(pack_file);
1268                 object_count++;
1269                 object_count_by_type[OBJ_BLOB]++;
1270         }
1271
1272         free(in_buf);
1273         free(out_buf);
1274 }
1275
1276 /* All calls must be guarded by find_object() or find_mark() to
1277  * ensure the 'struct object_entry' passed was written by this
1278  * process instance.  We unpack the entry by the offset, avoiding
1279  * the need for the corresponding .idx file.  This unpacking rule
1280  * works because we only use OBJ_REF_DELTA within the packfiles
1281  * created by fast-import.
1282  *
1283  * oe must not be NULL.  Such an oe usually comes from giving
1284  * an unknown SHA-1 to find_object() or an undefined mark to
1285  * find_mark().  Callers must test for this condition and use
1286  * the standard read_sha1_file() when it happens.
1287  *
1288  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1289  * find_mark(), where the mark was reloaded from an existing marks
1290  * file and is referencing an object that this fast-import process
1291  * instance did not write out to a packfile.  Callers must test for
1292  * this condition and use read_sha1_file() instead.
1293  */
1294 static void *gfi_unpack_entry(
1295         struct object_entry *oe,
1296         unsigned long *sizep)
1297 {
1298         enum object_type type;
1299         struct packed_git *p = all_packs[oe->pack_id];
1300         if (p == pack_data && p->pack_size < (pack_size + 20)) {
1301                 /* The object is stored in the packfile we are writing to
1302                  * and we have modified it since the last time we scanned
1303                  * back to read a previously written object.  If an old
1304                  * window covered [p->pack_size, p->pack_size + 20) its
1305                  * data is stale and is not valid.  Closing all windows
1306                  * and updating the packfile length ensures we can read
1307                  * the newly written data.
1308                  */
1309                 close_pack_windows(p);
1310                 sha1flush(pack_file);
1311
1312                 /* We have to offer 20 bytes additional on the end of
1313                  * the packfile as the core unpacker code assumes the
1314                  * footer is present at the file end and must promise
1315                  * at least 20 bytes within any window it maps.  But
1316                  * we don't actually create the footer here.
1317                  */
1318                 p->pack_size = pack_size + 20;
1319         }
1320         return unpack_entry(p, oe->idx.offset, &type, sizep);
1321 }
1322
1323 static const char *get_mode(const char *str, uint16_t *modep)
1324 {
1325         unsigned char c;
1326         uint16_t mode = 0;
1327
1328         while ((c = *str++) != ' ') {
1329                 if (c < '0' || c > '7')
1330                         return NULL;
1331                 mode = (mode << 3) + (c - '0');
1332         }
1333         *modep = mode;
1334         return str;
1335 }
1336
1337 static void load_tree(struct tree_entry *root)
1338 {
1339         unsigned char *sha1 = root->versions[1].sha1;
1340         struct object_entry *myoe;
1341         struct tree_content *t;
1342         unsigned long size;
1343         char *buf;
1344         const char *c;
1345
1346         root->tree = t = new_tree_content(8);
1347         if (is_null_sha1(sha1))
1348                 return;
1349
1350         myoe = find_object(sha1);
1351         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1352                 if (myoe->type != OBJ_TREE)
1353                         die("Not a tree: %s", sha1_to_hex(sha1));
1354                 t->delta_depth = myoe->depth;
1355                 buf = gfi_unpack_entry(myoe, &size);
1356                 if (!buf)
1357                         die("Can't load tree %s", sha1_to_hex(sha1));
1358         } else {
1359                 enum object_type type;
1360                 buf = read_sha1_file(sha1, &type, &size);
1361                 if (!buf || type != OBJ_TREE)
1362                         die("Can't load tree %s", sha1_to_hex(sha1));
1363         }
1364
1365         c = buf;
1366         while (c != (buf + size)) {
1367                 struct tree_entry *e = new_tree_entry();
1368
1369                 if (t->entry_count == t->entry_capacity)
1370                         root->tree = t = grow_tree_content(t, t->entry_count);
1371                 t->entries[t->entry_count++] = e;
1372
1373                 e->tree = NULL;
1374                 c = get_mode(c, &e->versions[1].mode);
1375                 if (!c)
1376                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1377                 e->versions[0].mode = e->versions[1].mode;
1378                 e->name = to_atom(c, strlen(c));
1379                 c += e->name->str_len + 1;
1380                 hashcpy(e->versions[0].sha1, (unsigned char *)c);
1381                 hashcpy(e->versions[1].sha1, (unsigned char *)c);
1382                 c += 20;
1383         }
1384         free(buf);
1385 }
1386
1387 static int tecmp0 (const void *_a, const void *_b)
1388 {
1389         struct tree_entry *a = *((struct tree_entry**)_a);
1390         struct tree_entry *b = *((struct tree_entry**)_b);
1391         return base_name_compare(
1392                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1393                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1394 }
1395
1396 static int tecmp1 (const void *_a, const void *_b)
1397 {
1398         struct tree_entry *a = *((struct tree_entry**)_a);
1399         struct tree_entry *b = *((struct tree_entry**)_b);
1400         return base_name_compare(
1401                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1402                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1403 }
1404
1405 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1406 {
1407         size_t maxlen = 0;
1408         unsigned int i;
1409
1410         if (!v)
1411                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1412         else
1413                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1414
1415         for (i = 0; i < t->entry_count; i++) {
1416                 if (t->entries[i]->versions[v].mode)
1417                         maxlen += t->entries[i]->name->str_len + 34;
1418         }
1419
1420         strbuf_reset(b);
1421         strbuf_grow(b, maxlen);
1422         for (i = 0; i < t->entry_count; i++) {
1423                 struct tree_entry *e = t->entries[i];
1424                 if (!e->versions[v].mode)
1425                         continue;
1426                 strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1427                                         e->name->str_dat, '\0');
1428                 strbuf_add(b, e->versions[v].sha1, 20);
1429         }
1430 }
1431
1432 static void store_tree(struct tree_entry *root)
1433 {
1434         struct tree_content *t = root->tree;
1435         unsigned int i, j, del;
1436         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1437         struct object_entry *le;
1438
1439         if (!is_null_sha1(root->versions[1].sha1))
1440                 return;
1441
1442         for (i = 0; i < t->entry_count; i++) {
1443                 if (t->entries[i]->tree)
1444                         store_tree(t->entries[i]);
1445         }
1446
1447         le = find_object(root->versions[0].sha1);
1448         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1449                 mktree(t, 0, &old_tree);
1450                 lo.data = old_tree;
1451                 lo.offset = le->idx.offset;
1452                 lo.depth = t->delta_depth;
1453         }
1454
1455         mktree(t, 1, &new_tree);
1456         store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1457
1458         t->delta_depth = lo.depth;
1459         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1460                 struct tree_entry *e = t->entries[i];
1461                 if (e->versions[1].mode) {
1462                         e->versions[0].mode = e->versions[1].mode;
1463                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1464                         t->entries[j++] = e;
1465                 } else {
1466                         release_tree_entry(e);
1467                         del++;
1468                 }
1469         }
1470         t->entry_count -= del;
1471 }
1472
1473 static void tree_content_replace(
1474         struct tree_entry *root,
1475         const unsigned char *sha1,
1476         const uint16_t mode,
1477         struct tree_content *newtree)
1478 {
1479         if (!S_ISDIR(mode))
1480                 die("Root cannot be a non-directory");
1481         hashcpy(root->versions[1].sha1, sha1);
1482         if (root->tree)
1483                 release_tree_content_recursive(root->tree);
1484         root->tree = newtree;
1485 }
1486
1487 static int tree_content_set(
1488         struct tree_entry *root,
1489         const char *p,
1490         const unsigned char *sha1,
1491         const uint16_t mode,
1492         struct tree_content *subtree)
1493 {
1494         struct tree_content *t;
1495         const char *slash1;
1496         unsigned int i, n;
1497         struct tree_entry *e;
1498
1499         slash1 = strchr(p, '/');
1500         if (slash1)
1501                 n = slash1 - p;
1502         else
1503                 n = strlen(p);
1504         if (!n)
1505                 die("Empty path component found in input");
1506         if (!slash1 && !S_ISDIR(mode) && subtree)
1507                 die("Non-directories cannot have subtrees");
1508
1509         if (!root->tree)
1510                 load_tree(root);
1511         t = root->tree;
1512         for (i = 0; i < t->entry_count; i++) {
1513                 e = t->entries[i];
1514                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1515                         if (!slash1) {
1516                                 if (!S_ISDIR(mode)
1517                                                 && e->versions[1].mode == mode
1518                                                 && !hashcmp(e->versions[1].sha1, sha1))
1519                                         return 0;
1520                                 e->versions[1].mode = mode;
1521                                 hashcpy(e->versions[1].sha1, sha1);
1522                                 if (e->tree)
1523                                         release_tree_content_recursive(e->tree);
1524                                 e->tree = subtree;
1525                                 hashclr(root->versions[1].sha1);
1526                                 return 1;
1527                         }
1528                         if (!S_ISDIR(e->versions[1].mode)) {
1529                                 e->tree = new_tree_content(8);
1530                                 e->versions[1].mode = S_IFDIR;
1531                         }
1532                         if (!e->tree)
1533                                 load_tree(e);
1534                         if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1535                                 hashclr(root->versions[1].sha1);
1536                                 return 1;
1537                         }
1538                         return 0;
1539                 }
1540         }
1541
1542         if (t->entry_count == t->entry_capacity)
1543                 root->tree = t = grow_tree_content(t, t->entry_count);
1544         e = new_tree_entry();
1545         e->name = to_atom(p, n);
1546         e->versions[0].mode = 0;
1547         hashclr(e->versions[0].sha1);
1548         t->entries[t->entry_count++] = e;
1549         if (slash1) {
1550                 e->tree = new_tree_content(8);
1551                 e->versions[1].mode = S_IFDIR;
1552                 tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1553         } else {
1554                 e->tree = subtree;
1555                 e->versions[1].mode = mode;
1556                 hashcpy(e->versions[1].sha1, sha1);
1557         }
1558         hashclr(root->versions[1].sha1);
1559         return 1;
1560 }
1561
1562 static int tree_content_remove(
1563         struct tree_entry *root,
1564         const char *p,
1565         struct tree_entry *backup_leaf)
1566 {
1567         struct tree_content *t;
1568         const char *slash1;
1569         unsigned int i, n;
1570         struct tree_entry *e;
1571
1572         slash1 = strchr(p, '/');
1573         if (slash1)
1574                 n = slash1 - p;
1575         else
1576                 n = strlen(p);
1577
1578         if (!root->tree)
1579                 load_tree(root);
1580         t = root->tree;
1581         for (i = 0; i < t->entry_count; i++) {
1582                 e = t->entries[i];
1583                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1584                         if (slash1 && !S_ISDIR(e->versions[1].mode))
1585                                 /*
1586                                  * If p names a file in some subdirectory, and a
1587                                  * file or symlink matching the name of the
1588                                  * parent directory of p exists, then p cannot
1589                                  * exist and need not be deleted.
1590                                  */
1591                                 return 1;
1592                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1593                                 goto del_entry;
1594                         if (!e->tree)
1595                                 load_tree(e);
1596                         if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1597                                 for (n = 0; n < e->tree->entry_count; n++) {
1598                                         if (e->tree->entries[n]->versions[1].mode) {
1599                                                 hashclr(root->versions[1].sha1);
1600                                                 return 1;
1601                                         }
1602                                 }
1603                                 backup_leaf = NULL;
1604                                 goto del_entry;
1605                         }
1606                         return 0;
1607                 }
1608         }
1609         return 0;
1610
1611 del_entry:
1612         if (backup_leaf)
1613                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1614         else if (e->tree)
1615                 release_tree_content_recursive(e->tree);
1616         e->tree = NULL;
1617         e->versions[1].mode = 0;
1618         hashclr(e->versions[1].sha1);
1619         hashclr(root->versions[1].sha1);
1620         return 1;
1621 }
1622
1623 static int tree_content_get(
1624         struct tree_entry *root,
1625         const char *p,
1626         struct tree_entry *leaf)
1627 {
1628         struct tree_content *t;
1629         const char *slash1;
1630         unsigned int i, n;
1631         struct tree_entry *e;
1632
1633         slash1 = strchr(p, '/');
1634         if (slash1)
1635                 n = slash1 - p;
1636         else
1637                 n = strlen(p);
1638
1639         if (!root->tree)
1640                 load_tree(root);
1641         t = root->tree;
1642         for (i = 0; i < t->entry_count; i++) {
1643                 e = t->entries[i];
1644                 if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1645                         if (!slash1) {
1646                                 memcpy(leaf, e, sizeof(*leaf));
1647                                 if (e->tree && is_null_sha1(e->versions[1].sha1))
1648                                         leaf->tree = dup_tree_content(e->tree);
1649                                 else
1650                                         leaf->tree = NULL;
1651                                 return 1;
1652                         }
1653                         if (!S_ISDIR(e->versions[1].mode))
1654                                 return 0;
1655                         if (!e->tree)
1656                                 load_tree(e);
1657                         return tree_content_get(e, slash1 + 1, leaf);
1658                 }
1659         }
1660         return 0;
1661 }
1662
1663 static int update_branch(struct branch *b)
1664 {
1665         static const char *msg = "fast-import";
1666         struct ref_lock *lock;
1667         unsigned char old_sha1[20];
1668
1669         if (is_null_sha1(b->sha1))
1670                 return 0;
1671         if (read_ref(b->name, old_sha1))
1672                 hashclr(old_sha1);
1673         lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1674         if (!lock)
1675                 return error("Unable to lock %s", b->name);
1676         if (!force_update && !is_null_sha1(old_sha1)) {
1677                 struct commit *old_cmit, *new_cmit;
1678
1679                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1680                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1681                 if (!old_cmit || !new_cmit) {
1682                         unlock_ref(lock);
1683                         return error("Branch %s is missing commits.", b->name);
1684                 }
1685
1686                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1687                         unlock_ref(lock);
1688                         warning("Not updating %s"
1689                                 " (new tip %s does not contain %s)",
1690                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1691                         return -1;
1692                 }
1693         }
1694         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1695                 return error("Unable to update %s", b->name);
1696         return 0;
1697 }
1698
1699 static void dump_branches(void)
1700 {
1701         unsigned int i;
1702         struct branch *b;
1703
1704         for (i = 0; i < branch_table_sz; i++) {
1705                 for (b = branch_table[i]; b; b = b->table_next_branch)
1706                         failure |= update_branch(b);
1707         }
1708 }
1709
1710 static void dump_tags(void)
1711 {
1712         static const char *msg = "fast-import";
1713         struct tag *t;
1714         struct ref_lock *lock;
1715         char ref_name[PATH_MAX];
1716
1717         for (t = first_tag; t; t = t->next_tag) {
1718                 sprintf(ref_name, "tags/%s", t->name);
1719                 lock = lock_ref_sha1(ref_name, NULL);
1720                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1721                         failure |= error("Unable to update %s", ref_name);
1722         }
1723 }
1724
1725 static void dump_marks_helper(FILE *f,
1726         uintmax_t base,
1727         struct mark_set *m)
1728 {
1729         uintmax_t k;
1730         if (m->shift) {
1731                 for (k = 0; k < 1024; k++) {
1732                         if (m->data.sets[k])
1733                                 dump_marks_helper(f, base + (k << m->shift),
1734                                         m->data.sets[k]);
1735                 }
1736         } else {
1737                 for (k = 0; k < 1024; k++) {
1738                         if (m->data.marked[k])
1739                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1740                                         sha1_to_hex(m->data.marked[k]->idx.sha1));
1741                 }
1742         }
1743 }
1744
1745 static void dump_marks(void)
1746 {
1747         static struct lock_file mark_lock;
1748         int mark_fd;
1749         FILE *f;
1750
1751         if (!export_marks_file)
1752                 return;
1753
1754         mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1755         if (mark_fd < 0) {
1756                 failure |= error("Unable to write marks file %s: %s",
1757                         export_marks_file, strerror(errno));
1758                 return;
1759         }
1760
1761         f = fdopen(mark_fd, "w");
1762         if (!f) {
1763                 int saved_errno = errno;
1764                 rollback_lock_file(&mark_lock);
1765                 failure |= error("Unable to write marks file %s: %s",
1766                         export_marks_file, strerror(saved_errno));
1767                 return;
1768         }
1769
1770         /*
1771          * Since the lock file was fdopen()'ed, it should not be close()'ed.
1772          * Assign -1 to the lock file descriptor so that commit_lock_file()
1773          * won't try to close() it.
1774          */
1775         mark_lock.fd = -1;
1776
1777         dump_marks_helper(f, 0, marks);
1778         if (ferror(f) || fclose(f)) {
1779                 int saved_errno = errno;
1780                 rollback_lock_file(&mark_lock);
1781                 failure |= error("Unable to write marks file %s: %s",
1782                         export_marks_file, strerror(saved_errno));
1783                 return;
1784         }
1785
1786         if (commit_lock_file(&mark_lock)) {
1787                 int saved_errno = errno;
1788                 rollback_lock_file(&mark_lock);
1789                 failure |= error("Unable to commit marks file %s: %s",
1790                         export_marks_file, strerror(saved_errno));
1791                 return;
1792         }
1793 }
1794
1795 static void read_marks(void)
1796 {
1797         char line[512];
1798         FILE *f = fopen(import_marks_file, "r");
1799         if (!f)
1800                 die_errno("cannot read '%s'", import_marks_file);
1801         while (fgets(line, sizeof(line), f)) {
1802                 uintmax_t mark;
1803                 char *end;
1804                 unsigned char sha1[20];
1805                 struct object_entry *e;
1806
1807                 end = strchr(line, '\n');
1808                 if (line[0] != ':' || !end)
1809                         die("corrupt mark line: %s", line);
1810                 *end = 0;
1811                 mark = strtoumax(line + 1, &end, 10);
1812                 if (!mark || end == line + 1
1813                         || *end != ' ' || get_sha1(end + 1, sha1))
1814                         die("corrupt mark line: %s", line);
1815                 e = find_object(sha1);
1816                 if (!e) {
1817                         enum object_type type = sha1_object_info(sha1, NULL);
1818                         if (type < 0)
1819                                 die("object not found: %s", sha1_to_hex(sha1));
1820                         e = insert_object(sha1);
1821                         e->type = type;
1822                         e->pack_id = MAX_PACK_ID;
1823                         e->idx.offset = 1; /* just not zero! */
1824                 }
1825                 insert_mark(mark, e);
1826         }
1827         fclose(f);
1828 }
1829
1830
1831 static int read_next_command(void)
1832 {
1833         static int stdin_eof = 0;
1834
1835         if (stdin_eof) {
1836                 unread_command_buf = 0;
1837                 return EOF;
1838         }
1839
1840         for (;;) {
1841                 if (unread_command_buf) {
1842                         unread_command_buf = 0;
1843                 } else {
1844                         struct recent_command *rc;
1845
1846                         strbuf_detach(&command_buf, NULL);
1847                         stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1848                         if (stdin_eof)
1849                                 return EOF;
1850
1851                         if (!seen_data_command
1852                                 && prefixcmp(command_buf.buf, "feature ")
1853                                 && prefixcmp(command_buf.buf, "option ")) {
1854                                 parse_argv();
1855                         }
1856
1857                         rc = rc_free;
1858                         if (rc)
1859                                 rc_free = rc->next;
1860                         else {
1861                                 rc = cmd_hist.next;
1862                                 cmd_hist.next = rc->next;
1863                                 cmd_hist.next->prev = &cmd_hist;
1864                                 free(rc->buf);
1865                         }
1866
1867                         rc->buf = command_buf.buf;
1868                         rc->prev = cmd_tail;
1869                         rc->next = cmd_hist.prev;
1870                         rc->prev->next = rc;
1871                         cmd_tail = rc;
1872                 }
1873                 if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1874                         parse_cat_blob();
1875                         continue;
1876                 }
1877                 if (command_buf.buf[0] == '#')
1878                         continue;
1879                 return 0;
1880         }
1881 }
1882
1883 static void skip_optional_lf(void)
1884 {
1885         int term_char = fgetc(stdin);
1886         if (term_char != '\n' && term_char != EOF)
1887                 ungetc(term_char, stdin);
1888 }
1889
1890 static void parse_mark(void)
1891 {
1892         if (!prefixcmp(command_buf.buf, "mark :")) {
1893                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1894                 read_next_command();
1895         }
1896         else
1897                 next_mark = 0;
1898 }
1899
1900 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1901 {
1902         strbuf_reset(sb);
1903
1904         if (prefixcmp(command_buf.buf, "data "))
1905                 die("Expected 'data n' command, found: %s", command_buf.buf);
1906
1907         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1908                 char *term = xstrdup(command_buf.buf + 5 + 2);
1909                 size_t term_len = command_buf.len - 5 - 2;
1910
1911                 strbuf_detach(&command_buf, NULL);
1912                 for (;;) {
1913                         if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1914                                 die("EOF in data (terminator '%s' not found)", term);
1915                         if (term_len == command_buf.len
1916                                 && !strcmp(term, command_buf.buf))
1917                                 break;
1918                         strbuf_addbuf(sb, &command_buf);
1919                         strbuf_addch(sb, '\n');
1920                 }
1921                 free(term);
1922         }
1923         else {
1924                 uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1925                 size_t n = 0, length = (size_t)len;
1926
1927                 if (limit && limit < len) {
1928                         *len_res = len;
1929                         return 0;
1930                 }
1931                 if (length < len)
1932                         die("data is too large to use in this context");
1933
1934                 while (n < length) {
1935                         size_t s = strbuf_fread(sb, length - n, stdin);
1936                         if (!s && feof(stdin))
1937                                 die("EOF in data (%lu bytes remaining)",
1938                                         (unsigned long)(length - n));
1939                         n += s;
1940                 }
1941         }
1942
1943         skip_optional_lf();
1944         return 1;
1945 }
1946
1947 static int validate_raw_date(const char *src, char *result, int maxlen)
1948 {
1949         const char *orig_src = src;
1950         char *endp;
1951         unsigned long num;
1952
1953         errno = 0;
1954
1955         num = strtoul(src, &endp, 10);
1956         /* NEEDSWORK: perhaps check for reasonable values? */
1957         if (errno || endp == src || *endp != ' ')
1958                 return -1;
1959
1960         src = endp + 1;
1961         if (*src != '-' && *src != '+')
1962                 return -1;
1963
1964         num = strtoul(src + 1, &endp, 10);
1965         if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1966             1400 < num)
1967                 return -1;
1968
1969         strcpy(result, orig_src);
1970         return 0;
1971 }
1972
1973 static char *parse_ident(const char *buf)
1974 {
1975         const char *gt;
1976         size_t name_len;
1977         char *ident;
1978
1979         gt = strrchr(buf, '>');
1980         if (!gt)
1981                 die("Missing > in ident string: %s", buf);
1982         gt++;
1983         if (*gt != ' ')
1984                 die("Missing space after > in ident string: %s", buf);
1985         gt++;
1986         name_len = gt - buf;
1987         ident = xmalloc(name_len + 24);
1988         strncpy(ident, buf, name_len);
1989
1990         switch (whenspec) {
1991         case WHENSPEC_RAW:
1992                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1993                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1994                 break;
1995         case WHENSPEC_RFC2822:
1996                 if (parse_date(gt, ident + name_len, 24) < 0)
1997                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1998                 break;
1999         case WHENSPEC_NOW:
2000                 if (strcmp("now", gt))
2001                         die("Date in ident must be 'now': %s", buf);
2002                 datestamp(ident + name_len, 24);
2003                 break;
2004         }
2005
2006         return ident;
2007 }
2008
2009 static void parse_and_store_blob(
2010         struct last_object *last,
2011         unsigned char *sha1out,
2012         uintmax_t mark)
2013 {
2014         static struct strbuf buf = STRBUF_INIT;
2015         uintmax_t len;
2016
2017         if (parse_data(&buf, big_file_threshold, &len))
2018                 store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2019         else {
2020                 if (last) {
2021                         strbuf_release(&last->data);
2022                         last->offset = 0;
2023                         last->depth = 0;
2024                 }
2025                 stream_blob(len, sha1out, mark);
2026                 skip_optional_lf();
2027         }
2028 }
2029
2030 static void parse_new_blob(void)
2031 {
2032         read_next_command();
2033         parse_mark();
2034         parse_and_store_blob(&last_blob, NULL, next_mark);
2035 }
2036
2037 static void unload_one_branch(void)
2038 {
2039         while (cur_active_branches
2040                 && cur_active_branches >= max_active_branches) {
2041                 uintmax_t min_commit = ULONG_MAX;
2042                 struct branch *e, *l = NULL, *p = NULL;
2043
2044                 for (e = active_branches; e; e = e->active_next_branch) {
2045                         if (e->last_commit < min_commit) {
2046                                 p = l;
2047                                 min_commit = e->last_commit;
2048                         }
2049                         l = e;
2050                 }
2051
2052                 if (p) {
2053                         e = p->active_next_branch;
2054                         p->active_next_branch = e->active_next_branch;
2055                 } else {
2056                         e = active_branches;
2057                         active_branches = e->active_next_branch;
2058                 }
2059                 e->active = 0;
2060                 e->active_next_branch = NULL;
2061                 if (e->branch_tree.tree) {
2062                         release_tree_content_recursive(e->branch_tree.tree);
2063                         e->branch_tree.tree = NULL;
2064                 }
2065                 cur_active_branches--;
2066         }
2067 }
2068
2069 static void load_branch(struct branch *b)
2070 {
2071         load_tree(&b->branch_tree);
2072         if (!b->active) {
2073                 b->active = 1;
2074                 b->active_next_branch = active_branches;
2075                 active_branches = b;
2076                 cur_active_branches++;
2077                 branch_load_count++;
2078         }
2079 }
2080
2081 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2082 {
2083         unsigned char fanout = 0;
2084         while ((num_notes >>= 8))
2085                 fanout++;
2086         return fanout;
2087 }
2088
2089 static void construct_path_with_fanout(const char *hex_sha1,
2090                 unsigned char fanout, char *path)
2091 {
2092         unsigned int i = 0, j = 0;
2093         if (fanout >= 20)
2094                 die("Too large fanout (%u)", fanout);
2095         while (fanout) {
2096                 path[i++] = hex_sha1[j++];
2097                 path[i++] = hex_sha1[j++];
2098                 path[i++] = '/';
2099                 fanout--;
2100         }
2101         memcpy(path + i, hex_sha1 + j, 40 - j);
2102         path[i + 40 - j] = '\0';
2103 }
2104
2105 static uintmax_t do_change_note_fanout(
2106                 struct tree_entry *orig_root, struct tree_entry *root,
2107                 char *hex_sha1, unsigned int hex_sha1_len,
2108                 char *fullpath, unsigned int fullpath_len,
2109                 unsigned char fanout)
2110 {
2111         struct tree_content *t = root->tree;
2112         struct tree_entry *e, leaf;
2113         unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2114         uintmax_t num_notes = 0;
2115         unsigned char sha1[20];
2116         char realpath[60];
2117
2118         for (i = 0; t && i < t->entry_count; i++) {
2119                 e = t->entries[i];
2120                 tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2121                 tmp_fullpath_len = fullpath_len;
2122
2123                 /*
2124                  * We're interested in EITHER existing note entries (entries
2125                  * with exactly 40 hex chars in path, not including directory
2126                  * separators), OR directory entries that may contain note
2127                  * entries (with < 40 hex chars in path).
2128                  * Also, each path component in a note entry must be a multiple
2129                  * of 2 chars.
2130                  */
2131                 if (!e->versions[1].mode ||
2132                     tmp_hex_sha1_len > 40 ||
2133                     e->name->str_len % 2)
2134                         continue;
2135
2136                 /* This _may_ be a note entry, or a subdir containing notes */
2137                 memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2138                        e->name->str_len);
2139                 if (tmp_fullpath_len)
2140                         fullpath[tmp_fullpath_len++] = '/';
2141                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2142                        e->name->str_len);
2143                 tmp_fullpath_len += e->name->str_len;
2144                 fullpath[tmp_fullpath_len] = '\0';
2145
2146                 if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2147                         /* This is a note entry */
2148                         construct_path_with_fanout(hex_sha1, fanout, realpath);
2149                         if (!strcmp(fullpath, realpath)) {
2150                                 /* Note entry is in correct location */
2151                                 num_notes++;
2152                                 continue;
2153                         }
2154
2155                         /* Rename fullpath to realpath */
2156                         if (!tree_content_remove(orig_root, fullpath, &leaf))
2157                                 die("Failed to remove path %s", fullpath);
2158                         tree_content_set(orig_root, realpath,
2159                                 leaf.versions[1].sha1,
2160                                 leaf.versions[1].mode,
2161                                 leaf.tree);
2162                 } else if (S_ISDIR(e->versions[1].mode)) {
2163                         /* This is a subdir that may contain note entries */
2164                         if (!e->tree)
2165                                 load_tree(e);
2166                         num_notes += do_change_note_fanout(orig_root, e,
2167                                 hex_sha1, tmp_hex_sha1_len,
2168                                 fullpath, tmp_fullpath_len, fanout);
2169                 }
2170
2171                 /* The above may have reallocated the current tree_content */
2172                 t = root->tree;
2173         }
2174         return num_notes;
2175 }
2176
2177 static uintmax_t change_note_fanout(struct tree_entry *root,
2178                 unsigned char fanout)
2179 {
2180         char hex_sha1[40], path[60];
2181         return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2182 }
2183
2184 static void file_change_m(struct branch *b)
2185 {
2186         const char *p = command_buf.buf + 2;
2187         static struct strbuf uq = STRBUF_INIT;
2188         const char *endp;
2189         struct object_entry *oe = oe;
2190         unsigned char sha1[20];
2191         uint16_t mode, inline_data = 0;
2192
2193         p = get_mode(p, &mode);
2194         if (!p)
2195                 die("Corrupt mode: %s", command_buf.buf);
2196         switch (mode) {
2197         case 0644:
2198         case 0755:
2199                 mode |= S_IFREG;
2200         case S_IFREG | 0644:
2201         case S_IFREG | 0755:
2202         case S_IFLNK:
2203         case S_IFDIR:
2204         case S_IFGITLINK:
2205                 /* ok */
2206                 break;
2207         default:
2208                 die("Corrupt mode: %s", command_buf.buf);
2209         }
2210
2211         if (*p == ':') {
2212                 char *x;
2213                 oe = find_mark(strtoumax(p + 1, &x, 10));
2214                 hashcpy(sha1, oe->idx.sha1);
2215                 p = x;
2216         } else if (!prefixcmp(p, "inline")) {
2217                 inline_data = 1;
2218                 p += 6;
2219         } else {
2220                 if (get_sha1_hex(p, sha1))
2221                         die("Invalid SHA1: %s", command_buf.buf);
2222                 oe = find_object(sha1);
2223                 p += 40;
2224         }
2225         if (*p++ != ' ')
2226                 die("Missing space after SHA1: %s", command_buf.buf);
2227
2228         strbuf_reset(&uq);
2229         if (!unquote_c_style(&uq, p, &endp)) {
2230                 if (*endp)
2231                         die("Garbage after path in: %s", command_buf.buf);
2232                 p = uq.buf;
2233         }
2234
2235         /* Git does not track empty, non-toplevel directories. */
2236         if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2237                 tree_content_remove(&b->branch_tree, p, NULL);
2238                 return;
2239         }
2240
2241         if (S_ISGITLINK(mode)) {
2242                 if (inline_data)
2243                         die("Git links cannot be specified 'inline': %s",
2244                                 command_buf.buf);
2245                 else if (oe) {
2246                         if (oe->type != OBJ_COMMIT)
2247                                 die("Not a commit (actually a %s): %s",
2248                                         typename(oe->type), command_buf.buf);
2249                 }
2250                 /*
2251                  * Accept the sha1 without checking; it expected to be in
2252                  * another repository.
2253                  */
2254         } else if (inline_data) {
2255                 if (S_ISDIR(mode))
2256                         die("Directories cannot be specified 'inline': %s",
2257                                 command_buf.buf);
2258                 if (p != uq.buf) {
2259                         strbuf_addstr(&uq, p);
2260                         p = uq.buf;
2261                 }
2262                 read_next_command();
2263                 parse_and_store_blob(&last_blob, sha1, 0);
2264         } else {
2265                 enum object_type expected = S_ISDIR(mode) ?
2266                                                 OBJ_TREE: OBJ_BLOB;
2267                 enum object_type type = oe ? oe->type :
2268                                         sha1_object_info(sha1, NULL);
2269                 if (type < 0)
2270                         die("%s not found: %s",
2271                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2272                                         command_buf.buf);
2273                 if (type != expected)
2274                         die("Not a %s (actually a %s): %s",
2275                                 typename(expected), typename(type),
2276                                 command_buf.buf);
2277         }
2278
2279         if (!*p) {
2280                 tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2281                 return;
2282         }
2283         tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2284 }
2285
2286 static void file_change_d(struct branch *b)
2287 {
2288         const char *p = command_buf.buf + 2;
2289         static struct strbuf uq = STRBUF_INIT;
2290         const char *endp;
2291
2292         strbuf_reset(&uq);
2293         if (!unquote_c_style(&uq, p, &endp)) {
2294                 if (*endp)
2295                         die("Garbage after path in: %s", command_buf.buf);
2296                 p = uq.buf;
2297         }
2298         tree_content_remove(&b->branch_tree, p, NULL);
2299 }
2300
2301 static void file_change_cr(struct branch *b, int rename)
2302 {
2303         const char *s, *d;
2304         static struct strbuf s_uq = STRBUF_INIT;
2305         static struct strbuf d_uq = STRBUF_INIT;
2306         const char *endp;
2307         struct tree_entry leaf;
2308
2309         s = command_buf.buf + 2;
2310         strbuf_reset(&s_uq);
2311         if (!unquote_c_style(&s_uq, s, &endp)) {
2312                 if (*endp != ' ')
2313                         die("Missing space after source: %s", command_buf.buf);
2314         } else {
2315                 endp = strchr(s, ' ');
2316                 if (!endp)
2317                         die("Missing space after source: %s", command_buf.buf);
2318                 strbuf_add(&s_uq, s, endp - s);
2319         }
2320         s = s_uq.buf;
2321
2322         endp++;
2323         if (!*endp)
2324                 die("Missing dest: %s", command_buf.buf);
2325
2326         d = endp;
2327         strbuf_reset(&d_uq);
2328         if (!unquote_c_style(&d_uq, d, &endp)) {
2329                 if (*endp)
2330                         die("Garbage after dest in: %s", command_buf.buf);
2331                 d = d_uq.buf;
2332         }
2333
2334         memset(&leaf, 0, sizeof(leaf));
2335         if (rename)
2336                 tree_content_remove(&b->branch_tree, s, &leaf);
2337         else
2338                 tree_content_get(&b->branch_tree, s, &leaf);
2339         if (!leaf.versions[1].mode)
2340                 die("Path %s not in branch", s);
2341         if (!*d) {      /* C "path/to/subdir" "" */
2342                 tree_content_replace(&b->branch_tree,
2343                         leaf.versions[1].sha1,
2344                         leaf.versions[1].mode,
2345                         leaf.tree);
2346                 return;
2347         }
2348         tree_content_set(&b->branch_tree, d,
2349                 leaf.versions[1].sha1,
2350                 leaf.versions[1].mode,
2351                 leaf.tree);
2352 }
2353
2354 static void note_change_n(struct branch *b, unsigned char old_fanout)
2355 {
2356         const char *p = command_buf.buf + 2;
2357         static struct strbuf uq = STRBUF_INIT;
2358         struct object_entry *oe = oe;
2359         struct branch *s;
2360         unsigned char sha1[20], commit_sha1[20];
2361         char path[60];
2362         uint16_t inline_data = 0;
2363         unsigned char new_fanout;
2364
2365         /* <dataref> or 'inline' */
2366         if (*p == ':') {
2367                 char *x;
2368                 oe = find_mark(strtoumax(p + 1, &x, 10));
2369                 hashcpy(sha1, oe->idx.sha1);
2370                 p = x;
2371         } else if (!prefixcmp(p, "inline")) {
2372                 inline_data = 1;
2373                 p += 6;
2374         } else {
2375                 if (get_sha1_hex(p, sha1))
2376                         die("Invalid SHA1: %s", command_buf.buf);
2377                 oe = find_object(sha1);
2378                 p += 40;
2379         }
2380         if (*p++ != ' ')
2381                 die("Missing space after SHA1: %s", command_buf.buf);
2382
2383         /* <committish> */
2384         s = lookup_branch(p);
2385         if (s) {
2386                 hashcpy(commit_sha1, s->sha1);
2387         } else if (*p == ':') {
2388                 uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2389                 struct object_entry *commit_oe = find_mark(commit_mark);
2390                 if (commit_oe->type != OBJ_COMMIT)
2391                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2392                 hashcpy(commit_sha1, commit_oe->idx.sha1);
2393         } else if (!get_sha1(p, commit_sha1)) {
2394                 unsigned long size;
2395                 char *buf = read_object_with_reference(commit_sha1,
2396                         commit_type, &size, commit_sha1);
2397                 if (!buf || size < 46)
2398                         die("Not a valid commit: %s", p);
2399                 free(buf);
2400         } else
2401                 die("Invalid ref name or SHA1 expression: %s", p);
2402
2403         if (inline_data) {
2404                 if (p != uq.buf) {
2405                         strbuf_addstr(&uq, p);
2406                         p = uq.buf;
2407                 }
2408                 read_next_command();
2409                 parse_and_store_blob(&last_blob, sha1, 0);
2410         } else if (oe) {
2411                 if (oe->type != OBJ_BLOB)
2412                         die("Not a blob (actually a %s): %s",
2413                                 typename(oe->type), command_buf.buf);
2414         } else if (!is_null_sha1(sha1)) {
2415                 enum object_type type = sha1_object_info(sha1, NULL);
2416                 if (type < 0)
2417                         die("Blob not found: %s", command_buf.buf);
2418                 if (type != OBJ_BLOB)
2419                         die("Not a blob (actually a %s): %s",
2420                             typename(type), command_buf.buf);
2421         }
2422
2423         construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
2424         if (tree_content_remove(&b->branch_tree, path, NULL))
2425                 b->num_notes--;
2426
2427         if (is_null_sha1(sha1))
2428                 return; /* nothing to insert */
2429
2430         b->num_notes++;
2431         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2432         construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2433         tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2434 }
2435
2436 static void file_change_deleteall(struct branch *b)
2437 {
2438         release_tree_content_recursive(b->branch_tree.tree);
2439         hashclr(b->branch_tree.versions[0].sha1);
2440         hashclr(b->branch_tree.versions[1].sha1);
2441         load_tree(&b->branch_tree);
2442         b->num_notes = 0;
2443 }
2444
2445 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2446 {
2447         if (!buf || size < 46)
2448                 die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2449         if (memcmp("tree ", buf, 5)
2450                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2451                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2452         hashcpy(b->branch_tree.versions[0].sha1,
2453                 b->branch_tree.versions[1].sha1);
2454 }
2455
2456 static void parse_from_existing(struct branch *b)
2457 {
2458         if (is_null_sha1(b->sha1)) {
2459                 hashclr(b->branch_tree.versions[0].sha1);
2460                 hashclr(b->branch_tree.versions[1].sha1);
2461         } else {
2462                 unsigned long size;
2463                 char *buf;
2464
2465                 buf = read_object_with_reference(b->sha1,
2466                         commit_type, &size, b->sha1);
2467                 parse_from_commit(b, buf, size);
2468                 free(buf);
2469         }
2470 }
2471
2472 static int parse_from(struct branch *b)
2473 {
2474         const char *from;
2475         struct branch *s;
2476
2477         if (prefixcmp(command_buf.buf, "from "))
2478                 return 0;
2479
2480         if (b->branch_tree.tree) {
2481                 release_tree_content_recursive(b->branch_tree.tree);
2482                 b->branch_tree.tree = NULL;
2483         }
2484
2485         from = strchr(command_buf.buf, ' ') + 1;
2486         s = lookup_branch(from);
2487         if (b == s)
2488                 die("Can't create a branch from itself: %s", b->name);
2489         else if (s) {
2490                 unsigned char *t = s->branch_tree.versions[1].sha1;
2491                 hashcpy(b->sha1, s->sha1);
2492                 hashcpy(b->branch_tree.versions[0].sha1, t);
2493                 hashcpy(b->branch_tree.versions[1].sha1, t);
2494         } else if (*from == ':') {
2495                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2496                 struct object_entry *oe = find_mark(idnum);
2497                 if (oe->type != OBJ_COMMIT)
2498                         die("Mark :%" PRIuMAX " not a commit", idnum);
2499                 hashcpy(b->sha1, oe->idx.sha1);
2500                 if (oe->pack_id != MAX_PACK_ID) {
2501                         unsigned long size;
2502                         char *buf = gfi_unpack_entry(oe, &size);
2503                         parse_from_commit(b, buf, size);
2504                         free(buf);
2505                 } else
2506                         parse_from_existing(b);
2507         } else if (!get_sha1(from, b->sha1))
2508                 parse_from_existing(b);
2509         else
2510                 die("Invalid ref name or SHA1 expression: %s", from);
2511
2512         read_next_command();
2513         return 1;
2514 }
2515
2516 static struct hash_list *parse_merge(unsigned int *count)
2517 {
2518         struct hash_list *list = NULL, *n, *e = e;
2519         const char *from;
2520         struct branch *s;
2521
2522         *count = 0;
2523         while (!prefixcmp(command_buf.buf, "merge ")) {
2524                 from = strchr(command_buf.buf, ' ') + 1;
2525                 n = xmalloc(sizeof(*n));
2526                 s = lookup_branch(from);
2527                 if (s)
2528                         hashcpy(n->sha1, s->sha1);
2529                 else if (*from == ':') {
2530                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2531                         struct object_entry *oe = find_mark(idnum);
2532                         if (oe->type != OBJ_COMMIT)
2533                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2534                         hashcpy(n->sha1, oe->idx.sha1);
2535                 } else if (!get_sha1(from, n->sha1)) {
2536                         unsigned long size;
2537                         char *buf = read_object_with_reference(n->sha1,
2538                                 commit_type, &size, n->sha1);
2539                         if (!buf || size < 46)
2540                                 die("Not a valid commit: %s", from);
2541                         free(buf);
2542                 } else
2543                         die("Invalid ref name or SHA1 expression: %s", from);
2544
2545                 n->next = NULL;
2546                 if (list)
2547                         e->next = n;
2548                 else
2549                         list = n;
2550                 e = n;
2551                 (*count)++;
2552                 read_next_command();
2553         }
2554         return list;
2555 }
2556
2557 static void parse_new_commit(void)
2558 {
2559         static struct strbuf msg = STRBUF_INIT;
2560         struct branch *b;
2561         char *sp;
2562         char *author = NULL;
2563         char *committer = NULL;
2564         struct hash_list *merge_list = NULL;
2565         unsigned int merge_count;
2566         unsigned char prev_fanout, new_fanout;
2567
2568         /* Obtain the branch name from the rest of our command */
2569         sp = strchr(command_buf.buf, ' ') + 1;
2570         b = lookup_branch(sp);
2571         if (!b)
2572                 b = new_branch(sp);
2573
2574         read_next_command();
2575         parse_mark();
2576         if (!prefixcmp(command_buf.buf, "author ")) {
2577                 author = parse_ident(command_buf.buf + 7);
2578                 read_next_command();
2579         }
2580         if (!prefixcmp(command_buf.buf, "committer ")) {
2581                 committer = parse_ident(command_buf.buf + 10);
2582                 read_next_command();
2583         }
2584         if (!committer)
2585                 die("Expected committer but didn't get one");
2586         parse_data(&msg, 0, NULL);
2587         read_next_command();
2588         parse_from(b);
2589         merge_list = parse_merge(&merge_count);
2590
2591         /* ensure the branch is active/loaded */
2592         if (!b->branch_tree.tree || !max_active_branches) {
2593                 unload_one_branch();
2594                 load_branch(b);
2595         }
2596
2597         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2598
2599         /* file_change* */
2600         while (command_buf.len > 0) {
2601                 if (!prefixcmp(command_buf.buf, "M "))
2602                         file_change_m(b);
2603                 else if (!prefixcmp(command_buf.buf, "D "))
2604                         file_change_d(b);
2605                 else if (!prefixcmp(command_buf.buf, "R "))
2606                         file_change_cr(b, 1);
2607                 else if (!prefixcmp(command_buf.buf, "C "))
2608                         file_change_cr(b, 0);
2609                 else if (!prefixcmp(command_buf.buf, "N "))
2610                         note_change_n(b, prev_fanout);
2611                 else if (!strcmp("deleteall", command_buf.buf))
2612                         file_change_deleteall(b);
2613                 else {
2614                         unread_command_buf = 1;
2615                         break;
2616                 }
2617                 if (read_next_command() == EOF)
2618                         break;
2619         }
2620
2621         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2622         if (new_fanout != prev_fanout)
2623                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2624
2625         /* build the tree and the commit */
2626         store_tree(&b->branch_tree);
2627         hashcpy(b->branch_tree.versions[0].sha1,
2628                 b->branch_tree.versions[1].sha1);
2629
2630         strbuf_reset(&new_data);
2631         strbuf_addf(&new_data, "tree %s\n",
2632                 sha1_to_hex(b->branch_tree.versions[1].sha1));
2633         if (!is_null_sha1(b->sha1))
2634                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2635         while (merge_list) {
2636                 struct hash_list *next = merge_list->next;
2637                 strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2638                 free(merge_list);
2639                 merge_list = next;
2640         }
2641         strbuf_addf(&new_data,
2642                 "author %s\n"
2643                 "committer %s\n"
2644                 "\n",
2645                 author ? author : committer, committer);
2646         strbuf_addbuf(&new_data, &msg);
2647         free(author);
2648         free(committer);
2649
2650         if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2651                 b->pack_id = pack_id;
2652         b->last_commit = object_count_by_type[OBJ_COMMIT];
2653 }
2654
2655 static void parse_new_tag(void)
2656 {
2657         static struct strbuf msg = STRBUF_INIT;
2658         char *sp;
2659         const char *from;
2660         char *tagger;
2661         struct branch *s;
2662         struct tag *t;
2663         uintmax_t from_mark = 0;
2664         unsigned char sha1[20];
2665         enum object_type type;
2666
2667         /* Obtain the new tag name from the rest of our command */
2668         sp = strchr(command_buf.buf, ' ') + 1;
2669         t = pool_alloc(sizeof(struct tag));
2670         t->next_tag = NULL;
2671         t->name = pool_strdup(sp);
2672         if (last_tag)
2673                 last_tag->next_tag = t;
2674         else
2675                 first_tag = t;
2676         last_tag = t;
2677         read_next_command();
2678
2679         /* from ... */
2680         if (prefixcmp(command_buf.buf, "from "))
2681                 die("Expected from command, got %s", command_buf.buf);
2682         from = strchr(command_buf.buf, ' ') + 1;
2683         s = lookup_branch(from);
2684         if (s) {
2685                 hashcpy(sha1, s->sha1);
2686                 type = OBJ_COMMIT;
2687         } else if (*from == ':') {
2688                 struct object_entry *oe;
2689                 from_mark = strtoumax(from + 1, NULL, 10);
2690                 oe = find_mark(from_mark);
2691                 type = oe->type;
2692                 hashcpy(sha1, oe->idx.sha1);
2693         } else if (!get_sha1(from, sha1)) {
2694                 unsigned long size;
2695                 char *buf;
2696
2697                 buf = read_sha1_file(sha1, &type, &size);
2698                 if (!buf || size < 46)
2699                         die("Not a valid commit: %s", from);
2700                 free(buf);
2701         } else
2702                 die("Invalid ref name or SHA1 expression: %s", from);
2703         read_next_command();
2704
2705         /* tagger ... */
2706         if (!prefixcmp(command_buf.buf, "tagger ")) {
2707                 tagger = parse_ident(command_buf.buf + 7);
2708                 read_next_command();
2709         } else
2710                 tagger = NULL;
2711
2712         /* tag payload/message */
2713         parse_data(&msg, 0, NULL);
2714
2715         /* build the tag object */
2716         strbuf_reset(&new_data);
2717
2718         strbuf_addf(&new_data,
2719                     "object %s\n"
2720                     "type %s\n"
2721                     "tag %s\n",
2722                     sha1_to_hex(sha1), typename(type), t->name);
2723         if (tagger)
2724                 strbuf_addf(&new_data,
2725                             "tagger %s\n", tagger);
2726         strbuf_addch(&new_data, '\n');
2727         strbuf_addbuf(&new_data, &msg);
2728         free(tagger);
2729
2730         if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2731                 t->pack_id = MAX_PACK_ID;
2732         else
2733                 t->pack_id = pack_id;
2734 }
2735
2736 static void parse_reset_branch(void)
2737 {
2738         struct branch *b;
2739         char *sp;
2740
2741         /* Obtain the branch name from the rest of our command */
2742         sp = strchr(command_buf.buf, ' ') + 1;
2743         b = lookup_branch(sp);
2744         if (b) {
2745                 hashclr(b->sha1);
2746                 hashclr(b->branch_tree.versions[0].sha1);
2747                 hashclr(b->branch_tree.versions[1].sha1);
2748                 if (b->branch_tree.tree) {
2749                         release_tree_content_recursive(b->branch_tree.tree);
2750                         b->branch_tree.tree = NULL;
2751                 }
2752         }
2753         else
2754                 b = new_branch(sp);
2755         read_next_command();
2756         parse_from(b);
2757         if (command_buf.len > 0)
2758                 unread_command_buf = 1;
2759 }
2760
2761 static void cat_blob_write(const char *buf, unsigned long size)
2762 {
2763         if (write_in_full(cat_blob_fd, buf, size) != size)
2764                 die_errno("Write to frontend failed");
2765 }
2766
2767 static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2768 {
2769         struct strbuf line = STRBUF_INIT;
2770         unsigned long size;
2771         enum object_type type = 0;
2772         char *buf;
2773
2774         if (!oe || oe->pack_id == MAX_PACK_ID) {
2775                 buf = read_sha1_file(sha1, &type, &size);
2776         } else {
2777                 type = oe->type;
2778                 buf = gfi_unpack_entry(oe, &size);
2779         }
2780
2781         /*
2782          * Output based on batch_one_object() from cat-file.c.
2783          */
2784         if (type <= 0) {
2785                 strbuf_reset(&line);
2786                 strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2787                 cat_blob_write(line.buf, line.len);
2788                 strbuf_release(&line);
2789                 free(buf);
2790                 return;
2791         }
2792         if (!buf)
2793                 die("Can't read object %s", sha1_to_hex(sha1));
2794         if (type != OBJ_BLOB)
2795                 die("Object %s is a %s but a blob was expected.",
2796                     sha1_to_hex(sha1), typename(type));
2797         strbuf_reset(&line);
2798         strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2799                                                 typename(type), size);
2800         cat_blob_write(line.buf, line.len);
2801         strbuf_release(&line);
2802         cat_blob_write(buf, size);
2803         cat_blob_write("\n", 1);
2804         free(buf);
2805 }
2806
2807 static void parse_cat_blob(void)
2808 {
2809         const char *p;
2810         struct object_entry *oe = oe;
2811         unsigned char sha1[20];
2812
2813         /* cat-blob SP <object> LF */
2814         p = command_buf.buf + strlen("cat-blob ");
2815         if (*p == ':') {
2816                 char *x;
2817                 oe = find_mark(strtoumax(p + 1, &x, 10));
2818                 if (x == p + 1)
2819                         die("Invalid mark: %s", command_buf.buf);
2820                 if (!oe)
2821                         die("Unknown mark: %s", command_buf.buf);
2822                 if (*x)
2823                         die("Garbage after mark: %s", command_buf.buf);
2824                 hashcpy(sha1, oe->idx.sha1);
2825         } else {
2826                 if (get_sha1_hex(p, sha1))
2827                         die("Invalid SHA1: %s", command_buf.buf);
2828                 if (p[40])
2829                         die("Garbage after SHA1: %s", command_buf.buf);
2830                 oe = find_object(sha1);
2831         }
2832
2833         cat_blob(oe, sha1);
2834 }
2835
2836 static void checkpoint(void)
2837 {
2838         checkpoint_requested = 0;
2839         if (object_count) {
2840                 cycle_packfile();
2841                 dump_branches();
2842                 dump_tags();
2843                 dump_marks();
2844         }
2845 }
2846
2847 static void parse_checkpoint(void)
2848 {
2849         checkpoint_requested = 1;
2850         skip_optional_lf();
2851 }
2852
2853 static void parse_progress(void)
2854 {
2855         fwrite(command_buf.buf, 1, command_buf.len, stdout);
2856         fputc('\n', stdout);
2857         fflush(stdout);
2858         skip_optional_lf();
2859 }
2860
2861 static char* make_fast_import_path(const char *path)
2862 {
2863         struct strbuf abs_path = STRBUF_INIT;
2864
2865         if (!relative_marks_paths || is_absolute_path(path))
2866                 return xstrdup(path);
2867         strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
2868         return strbuf_detach(&abs_path, NULL);
2869 }
2870
2871 static void option_import_marks(const char *marks, int from_stream)
2872 {
2873         if (import_marks_file) {
2874                 if (from_stream)
2875                         die("Only one import-marks command allowed per stream");
2876
2877                 /* read previous mark file */
2878                 if(!import_marks_file_from_stream)
2879                         read_marks();
2880         }
2881
2882         import_marks_file = make_fast_import_path(marks);
2883         safe_create_leading_directories_const(import_marks_file);
2884         import_marks_file_from_stream = from_stream;
2885 }
2886
2887 static void option_date_format(const char *fmt)
2888 {
2889         if (!strcmp(fmt, "raw"))
2890                 whenspec = WHENSPEC_RAW;
2891         else if (!strcmp(fmt, "rfc2822"))
2892                 whenspec = WHENSPEC_RFC2822;
2893         else if (!strcmp(fmt, "now"))
2894                 whenspec = WHENSPEC_NOW;
2895         else
2896                 die("unknown --date-format argument %s", fmt);
2897 }
2898
2899 static unsigned long ulong_arg(const char *option, const char *arg)
2900 {
2901         char *endptr;
2902         unsigned long rv = strtoul(arg, &endptr, 0);
2903         if (strchr(arg, '-') || endptr == arg || *endptr)
2904                 die("%s: argument must be a non-negative integer", option);
2905         return rv;
2906 }
2907
2908 static void option_depth(const char *depth)
2909 {
2910         max_depth = ulong_arg("--depth", depth);
2911         if (max_depth > MAX_DEPTH)
2912                 die("--depth cannot exceed %u", MAX_DEPTH);
2913 }
2914
2915 static void option_active_branches(const char *branches)
2916 {
2917         max_active_branches = ulong_arg("--active-branches", branches);
2918 }
2919
2920 static void option_export_marks(const char *marks)
2921 {
2922         export_marks_file = make_fast_import_path(marks);
2923         safe_create_leading_directories_const(export_marks_file);
2924 }
2925
2926 static void option_cat_blob_fd(const char *fd)
2927 {
2928         unsigned long n = ulong_arg("--cat-blob-fd", fd);
2929         if (n > (unsigned long) INT_MAX)
2930                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
2931         cat_blob_fd = (int) n;
2932 }
2933
2934 static void option_export_pack_edges(const char *edges)
2935 {
2936         if (pack_edges)
2937                 fclose(pack_edges);
2938         pack_edges = fopen(edges, "a");
2939         if (!pack_edges)
2940                 die_errno("Cannot open '%s'", edges);
2941 }
2942
2943 static int parse_one_option(const char *option)
2944 {
2945         if (!prefixcmp(option, "max-pack-size=")) {
2946                 unsigned long v;
2947                 if (!git_parse_ulong(option + 14, &v))
2948                         return 0;
2949                 if (v < 8192) {
2950                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
2951                         v *= 1024 * 1024;
2952                 } else if (v < 1024 * 1024) {
2953                         warning("minimum max-pack-size is 1 MiB");
2954                         v = 1024 * 1024;
2955                 }
2956                 max_packsize = v;
2957         } else if (!prefixcmp(option, "big-file-threshold=")) {
2958                 unsigned long v;
2959                 if (!git_parse_ulong(option + 19, &v))
2960                         return 0;
2961                 big_file_threshold = v;
2962         } else if (!prefixcmp(option, "depth=")) {
2963                 option_depth(option + 6);
2964         } else if (!prefixcmp(option, "active-branches=")) {
2965                 option_active_branches(option + 16);
2966         } else if (!prefixcmp(option, "export-pack-edges=")) {
2967                 option_export_pack_edges(option + 18);
2968         } else if (!prefixcmp(option, "quiet")) {
2969                 show_stats = 0;
2970         } else if (!prefixcmp(option, "stats")) {
2971                 show_stats = 1;
2972         } else {
2973                 return 0;
2974         }
2975
2976         return 1;
2977 }
2978
2979 static int parse_one_feature(const char *feature, int from_stream)
2980 {
2981         if (!prefixcmp(feature, "date-format=")) {
2982                 option_date_format(feature + 12);
2983         } else if (!prefixcmp(feature, "import-marks=")) {
2984                 option_import_marks(feature + 13, from_stream);
2985         } else if (!prefixcmp(feature, "export-marks=")) {
2986                 option_export_marks(feature + 13);
2987         } else if (!strcmp(feature, "cat-blob")) {
2988                 ; /* Don't die - this feature is supported */
2989         } else if (!prefixcmp(feature, "relative-marks")) {
2990                 relative_marks_paths = 1;
2991         } else if (!prefixcmp(feature, "no-relative-marks")) {
2992                 relative_marks_paths = 0;
2993         } else if (!prefixcmp(feature, "force")) {
2994                 force_update = 1;
2995         } else if (!strcmp(feature, "notes")) {
2996                 ; /* do nothing; we have the feature */
2997         } else {
2998                 return 0;
2999         }
3000
3001         return 1;
3002 }
3003
3004 static void parse_feature(void)
3005 {
3006         char *feature = command_buf.buf + 8;
3007
3008         if (seen_data_command)
3009                 die("Got feature command '%s' after data command", feature);
3010
3011         if (parse_one_feature(feature, 1))
3012                 return;
3013
3014         die("This version of fast-import does not support feature %s.", feature);
3015 }
3016
3017 static void parse_option(void)
3018 {
3019         char *option = command_buf.buf + 11;
3020
3021         if (seen_data_command)
3022                 die("Got option command '%s' after data command", option);
3023
3024         if (parse_one_option(option))
3025                 return;
3026
3027         die("This version of fast-import does not support option: %s", option);
3028 }
3029
3030 static int git_pack_config(const char *k, const char *v, void *cb)
3031 {
3032         if (!strcmp(k, "pack.depth")) {
3033                 max_depth = git_config_int(k, v);
3034                 if (max_depth > MAX_DEPTH)
3035                         max_depth = MAX_DEPTH;
3036                 return 0;
3037         }
3038         if (!strcmp(k, "pack.compression")) {
3039                 int level = git_config_int(k, v);
3040                 if (level == -1)
3041                         level = Z_DEFAULT_COMPRESSION;
3042                 else if (level < 0 || level > Z_BEST_COMPRESSION)
3043                         die("bad pack compression level %d", level);
3044                 pack_compression_level = level;
3045                 pack_compression_seen = 1;
3046                 return 0;
3047         }
3048         if (!strcmp(k, "pack.indexversion")) {
3049                 pack_idx_default_version = git_config_int(k, v);
3050                 if (pack_idx_default_version > 2)
3051                         die("bad pack.indexversion=%"PRIu32,
3052                             pack_idx_default_version);
3053                 return 0;
3054         }
3055         if (!strcmp(k, "pack.packsizelimit")) {
3056                 max_packsize = git_config_ulong(k, v);
3057                 return 0;
3058         }
3059         if (!strcmp(k, "core.bigfilethreshold")) {
3060                 long n = git_config_int(k, v);
3061                 big_file_threshold = 0 < n ? n : 0;
3062         }
3063         return git_default_config(k, v, cb);
3064 }
3065
3066 static const char fast_import_usage[] =
3067 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3068
3069 static void parse_argv(void)
3070 {
3071         unsigned int i;
3072
3073         for (i = 1; i < global_argc; i++) {
3074                 const char *a = global_argv[i];
3075
3076                 if (*a != '-' || !strcmp(a, "--"))
3077                         break;
3078
3079                 if (parse_one_option(a + 2))
3080                         continue;
3081
3082                 if (parse_one_feature(a + 2, 0))
3083                         continue;
3084
3085                 if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3086                         option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3087                         continue;
3088                 }
3089
3090                 die("unknown option %s", a);
3091         }
3092         if (i != global_argc)
3093                 usage(fast_import_usage);
3094
3095         seen_data_command = 1;
3096         if (import_marks_file)
3097                 read_marks();
3098 }
3099
3100 int main(int argc, const char **argv)
3101 {
3102         unsigned int i;
3103
3104         git_extract_argv0_path(argv[0]);
3105
3106         if (argc == 2 && !strcmp(argv[1], "-h"))
3107                 usage(fast_import_usage);
3108
3109         setup_git_directory();
3110         git_config(git_pack_config, NULL);
3111         if (!pack_compression_seen && core_compression_seen)
3112                 pack_compression_level = core_compression_level;
3113
3114         alloc_objects(object_entry_alloc);
3115         strbuf_init(&command_buf, 0);
3116         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3117         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3118         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3119         marks = pool_calloc(1, sizeof(struct mark_set));
3120
3121         global_argc = argc;
3122         global_argv = argv;
3123
3124         rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3125         for (i = 0; i < (cmd_save - 1); i++)
3126                 rc_free[i].next = &rc_free[i + 1];
3127         rc_free[cmd_save - 1].next = NULL;
3128
3129         prepare_packed_git();
3130         start_packfile();
3131         set_die_routine(die_nicely);
3132         set_checkpoint_signal();
3133         while (read_next_command() != EOF) {
3134                 if (!strcmp("blob", command_buf.buf))
3135                         parse_new_blob();
3136                 else if (!prefixcmp(command_buf.buf, "commit "))
3137                         parse_new_commit();
3138                 else if (!prefixcmp(command_buf.buf, "tag "))
3139                         parse_new_tag();
3140                 else if (!prefixcmp(command_buf.buf, "reset "))
3141                         parse_reset_branch();
3142                 else if (!strcmp("checkpoint", command_buf.buf))
3143                         parse_checkpoint();
3144                 else if (!prefixcmp(command_buf.buf, "progress "))
3145                         parse_progress();
3146                 else if (!prefixcmp(command_buf.buf, "feature "))
3147                         parse_feature();
3148                 else if (!prefixcmp(command_buf.buf, "option git "))
3149                         parse_option();
3150                 else if (!prefixcmp(command_buf.buf, "option "))
3151                         /* ignore non-git options*/;
3152                 else
3153                         die("Unsupported command: %s", command_buf.buf);
3154
3155                 if (checkpoint_requested)
3156                         checkpoint();
3157         }
3158
3159         /* argv hasn't been parsed yet, do so */
3160         if (!seen_data_command)
3161                 parse_argv();
3162
3163         end_packfile();
3164
3165         dump_branches();
3166         dump_tags();
3167         unkeep_all_packs();
3168         dump_marks();
3169
3170         if (pack_edges)
3171                 fclose(pack_edges);
3172
3173         if (show_stats) {
3174                 uintmax_t total_count = 0, duplicate_count = 0;
3175                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3176                         total_count += object_count_by_type[i];
3177                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3178                         duplicate_count += duplicate_count_by_type[i];
3179
3180                 fprintf(stderr, "%s statistics:\n", argv[0]);
3181                 fprintf(stderr, "---------------------------------------------------------------------\n");
3182                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3183                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3184                 fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
3185                 fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
3186                 fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
3187                 fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
3188                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3189                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3190                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3191                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3192                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3193                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3194                 fprintf(stderr, "---------------------------------------------------------------------\n");
3195                 pack_report();
3196                 fprintf(stderr, "---------------------------------------------------------------------\n");
3197                 fprintf(stderr, "\n");
3198         }
3199
3200         return failure ? 1 : 0;
3201 }