mingw: cope with the Isilon network file system
[git] / fast-import.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "lockfile.h"
6 #include "object.h"
7 #include "blob.h"
8 #include "tree.h"
9 #include "commit.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "refs.h"
13 #include "csum-file.h"
14 #include "quote.h"
15 #include "dir.h"
16 #include "run-command.h"
17 #include "packfile.h"
18 #include "object-store.h"
19 #include "mem-pool.h"
20 #include "commit-reach.h"
21 #include "khash.h"
22
23 #define PACK_ID_BITS 16
24 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
25 #define DEPTH_BITS 13
26 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
27
28 /*
29  * We abuse the setuid bit on directories to mean "do not delta".
30  */
31 #define NO_DELTA S_ISUID
32
33 /*
34  * The amount of additional space required in order to write an object into the
35  * current pack. This is the hash lengths at the end of the pack, plus the
36  * length of one object ID.
37  */
38 #define PACK_SIZE_THRESHOLD (the_hash_algo->rawsz * 3)
39
40 struct object_entry {
41         struct pack_idx_entry idx;
42         struct object_entry *next;
43         uint32_t type : TYPE_BITS,
44                 pack_id : PACK_ID_BITS,
45                 depth : DEPTH_BITS;
46 };
47
48 struct object_entry_pool {
49         struct object_entry_pool *next_pool;
50         struct object_entry *next_free;
51         struct object_entry *end;
52         struct object_entry entries[FLEX_ARRAY]; /* more */
53 };
54
55 struct mark_set {
56         union {
57                 struct object_id *oids[1024];
58                 struct object_entry *marked[1024];
59                 struct mark_set *sets[1024];
60         } data;
61         unsigned int shift;
62 };
63
64 struct last_object {
65         struct strbuf data;
66         off_t offset;
67         unsigned int depth;
68         unsigned no_swap : 1;
69 };
70
71 struct atom_str {
72         struct atom_str *next_atom;
73         unsigned short str_len;
74         char str_dat[FLEX_ARRAY]; /* more */
75 };
76
77 struct tree_content;
78 struct tree_entry {
79         struct tree_content *tree;
80         struct atom_str *name;
81         struct tree_entry_ms {
82                 uint16_t mode;
83                 struct object_id oid;
84         } versions[2];
85 };
86
87 struct tree_content {
88         unsigned int entry_capacity; /* must match avail_tree_content */
89         unsigned int entry_count;
90         unsigned int delta_depth;
91         struct tree_entry *entries[FLEX_ARRAY]; /* more */
92 };
93
94 struct avail_tree_content {
95         unsigned int entry_capacity; /* must match tree_content */
96         struct avail_tree_content *next_avail;
97 };
98
99 struct branch {
100         struct branch *table_next_branch;
101         struct branch *active_next_branch;
102         const char *name;
103         struct tree_entry branch_tree;
104         uintmax_t last_commit;
105         uintmax_t num_notes;
106         unsigned active : 1;
107         unsigned delete : 1;
108         unsigned pack_id : PACK_ID_BITS;
109         struct object_id oid;
110 };
111
112 struct tag {
113         struct tag *next_tag;
114         const char *name;
115         unsigned int pack_id;
116         struct object_id oid;
117 };
118
119 struct hash_list {
120         struct hash_list *next;
121         struct object_id oid;
122 };
123
124 typedef enum {
125         WHENSPEC_RAW = 1,
126         WHENSPEC_RFC2822,
127         WHENSPEC_NOW
128 } whenspec_type;
129
130 struct recent_command {
131         struct recent_command *prev;
132         struct recent_command *next;
133         char *buf;
134 };
135
136 typedef void (*mark_set_inserter_t)(struct mark_set *s, struct object_id *oid, uintmax_t mark);
137 typedef void (*each_mark_fn_t)(uintmax_t mark, void *obj, void *cbp);
138
139 /* Configured limits on output */
140 static unsigned long max_depth = 50;
141 static off_t max_packsize;
142 static int unpack_limit = 100;
143 static int force_update;
144
145 /* Stats and misc. counters */
146 static uintmax_t alloc_count;
147 static uintmax_t marks_set_count;
148 static uintmax_t object_count_by_type[1 << TYPE_BITS];
149 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
150 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
151 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
152 static unsigned long object_count;
153 static unsigned long branch_count;
154 static unsigned long branch_load_count;
155 static int failure;
156 static FILE *pack_edges;
157 static unsigned int show_stats = 1;
158 static int global_argc;
159 static const char **global_argv;
160
161 /* Memory pools */
162 static struct mem_pool fi_mem_pool =  {NULL, 2*1024*1024 -
163                                        sizeof(struct mp_block), 0 };
164
165 /* Atom management */
166 static unsigned int atom_table_sz = 4451;
167 static unsigned int atom_cnt;
168 static struct atom_str **atom_table;
169
170 /* The .pack file being generated */
171 static struct pack_idx_option pack_idx_opts;
172 static unsigned int pack_id;
173 static struct hashfile *pack_file;
174 static struct packed_git *pack_data;
175 static struct packed_git **all_packs;
176 static off_t pack_size;
177
178 /* Table of objects we've written. */
179 static unsigned int object_entry_alloc = 5000;
180 static struct object_entry_pool *blocks;
181 static struct object_entry *object_table[1 << 16];
182 static struct mark_set *marks;
183 static const char *export_marks_file;
184 static const char *import_marks_file;
185 static int import_marks_file_from_stream;
186 static int import_marks_file_ignore_missing;
187 static int import_marks_file_done;
188 static int relative_marks_paths;
189
190 /* Our last blob */
191 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
192
193 /* Tree management */
194 static unsigned int tree_entry_alloc = 1000;
195 static void *avail_tree_entry;
196 static unsigned int avail_tree_table_sz = 100;
197 static struct avail_tree_content **avail_tree_table;
198 static size_t tree_entry_allocd;
199 static struct strbuf old_tree = STRBUF_INIT;
200 static struct strbuf new_tree = STRBUF_INIT;
201
202 /* Branch data */
203 static unsigned long max_active_branches = 5;
204 static unsigned long cur_active_branches;
205 static unsigned long branch_table_sz = 1039;
206 static struct branch **branch_table;
207 static struct branch *active_branches;
208
209 /* Tag data */
210 static struct tag *first_tag;
211 static struct tag *last_tag;
212
213 /* Input stream parsing */
214 static whenspec_type whenspec = WHENSPEC_RAW;
215 static struct strbuf command_buf = STRBUF_INIT;
216 static int unread_command_buf;
217 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
218 static struct recent_command *cmd_tail = &cmd_hist;
219 static struct recent_command *rc_free;
220 static unsigned int cmd_save = 100;
221 static uintmax_t next_mark;
222 static struct strbuf new_data = STRBUF_INIT;
223 static int seen_data_command;
224 static int require_explicit_termination;
225 static int allow_unsafe_features;
226
227 /* Signal handling */
228 static volatile sig_atomic_t checkpoint_requested;
229
230 /* Submodule marks */
231 static struct string_list sub_marks_from = STRING_LIST_INIT_DUP;
232 static struct string_list sub_marks_to = STRING_LIST_INIT_DUP;
233 static kh_oid_map_t *sub_oid_map;
234
235 /* Where to write output of cat-blob commands */
236 static int cat_blob_fd = STDOUT_FILENO;
237
238 static void parse_argv(void);
239 static void parse_get_mark(const char *p);
240 static void parse_cat_blob(const char *p);
241 static void parse_ls(const char *p, struct branch *b);
242
243 static void for_each_mark(struct mark_set *m, uintmax_t base, each_mark_fn_t callback, void *p)
244 {
245         uintmax_t k;
246         if (m->shift) {
247                 for (k = 0; k < 1024; k++) {
248                         if (m->data.sets[k])
249                                 for_each_mark(m->data.sets[k], base + (k << m->shift), callback, p);
250                 }
251         } else {
252                 for (k = 0; k < 1024; k++) {
253                         if (m->data.marked[k])
254                                 callback(base + k, m->data.marked[k], p);
255                 }
256         }
257 }
258
259 static void dump_marks_fn(uintmax_t mark, void *object, void *cbp) {
260         struct object_entry *e = object;
261         FILE *f = cbp;
262
263         fprintf(f, ":%" PRIuMAX " %s\n", mark, oid_to_hex(&e->idx.oid));
264 }
265
266 static void write_branch_report(FILE *rpt, struct branch *b)
267 {
268         fprintf(rpt, "%s:\n", b->name);
269
270         fprintf(rpt, "  status      :");
271         if (b->active)
272                 fputs(" active", rpt);
273         if (b->branch_tree.tree)
274                 fputs(" loaded", rpt);
275         if (is_null_oid(&b->branch_tree.versions[1].oid))
276                 fputs(" dirty", rpt);
277         fputc('\n', rpt);
278
279         fprintf(rpt, "  tip commit  : %s\n", oid_to_hex(&b->oid));
280         fprintf(rpt, "  old tree    : %s\n",
281                 oid_to_hex(&b->branch_tree.versions[0].oid));
282         fprintf(rpt, "  cur tree    : %s\n",
283                 oid_to_hex(&b->branch_tree.versions[1].oid));
284         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
285
286         fputs("  last pack   : ", rpt);
287         if (b->pack_id < MAX_PACK_ID)
288                 fprintf(rpt, "%u", b->pack_id);
289         fputc('\n', rpt);
290
291         fputc('\n', rpt);
292 }
293
294 static void write_crash_report(const char *err)
295 {
296         char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
297         FILE *rpt = fopen(loc, "w");
298         struct branch *b;
299         unsigned long lu;
300         struct recent_command *rc;
301
302         if (!rpt) {
303                 error_errno("can't write crash report %s", loc);
304                 free(loc);
305                 return;
306         }
307
308         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
309
310         fprintf(rpt, "fast-import crash report:\n");
311         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
312         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
313         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
314         fputc('\n', rpt);
315
316         fputs("fatal: ", rpt);
317         fputs(err, rpt);
318         fputc('\n', rpt);
319
320         fputc('\n', rpt);
321         fputs("Most Recent Commands Before Crash\n", rpt);
322         fputs("---------------------------------\n", rpt);
323         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
324                 if (rc->next == &cmd_hist)
325                         fputs("* ", rpt);
326                 else
327                         fputs("  ", rpt);
328                 fputs(rc->buf, rpt);
329                 fputc('\n', rpt);
330         }
331
332         fputc('\n', rpt);
333         fputs("Active Branch LRU\n", rpt);
334         fputs("-----------------\n", rpt);
335         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
336                 cur_active_branches,
337                 max_active_branches);
338         fputc('\n', rpt);
339         fputs("  pos  clock name\n", rpt);
340         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
341         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
342                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
343                         ++lu, b->last_commit, b->name);
344
345         fputc('\n', rpt);
346         fputs("Inactive Branches\n", rpt);
347         fputs("-----------------\n", rpt);
348         for (lu = 0; lu < branch_table_sz; lu++) {
349                 for (b = branch_table[lu]; b; b = b->table_next_branch)
350                         write_branch_report(rpt, b);
351         }
352
353         if (first_tag) {
354                 struct tag *tg;
355                 fputc('\n', rpt);
356                 fputs("Annotated Tags\n", rpt);
357                 fputs("--------------\n", rpt);
358                 for (tg = first_tag; tg; tg = tg->next_tag) {
359                         fputs(oid_to_hex(&tg->oid), rpt);
360                         fputc(' ', rpt);
361                         fputs(tg->name, rpt);
362                         fputc('\n', rpt);
363                 }
364         }
365
366         fputc('\n', rpt);
367         fputs("Marks\n", rpt);
368         fputs("-----\n", rpt);
369         if (export_marks_file)
370                 fprintf(rpt, "  exported to %s\n", export_marks_file);
371         else
372                 for_each_mark(marks, 0, dump_marks_fn, rpt);
373
374         fputc('\n', rpt);
375         fputs("-------------------\n", rpt);
376         fputs("END OF CRASH REPORT\n", rpt);
377         fclose(rpt);
378         free(loc);
379 }
380
381 static void end_packfile(void);
382 static void unkeep_all_packs(void);
383 static void dump_marks(void);
384
385 static NORETURN void die_nicely(const char *err, va_list params)
386 {
387         static int zombie;
388         char message[2 * PATH_MAX];
389
390         vsnprintf(message, sizeof(message), err, params);
391         fputs("fatal: ", stderr);
392         fputs(message, stderr);
393         fputc('\n', stderr);
394
395         if (!zombie) {
396                 zombie = 1;
397                 write_crash_report(message);
398                 end_packfile();
399                 unkeep_all_packs();
400                 dump_marks();
401         }
402         exit(128);
403 }
404
405 #ifndef SIGUSR1 /* Windows, for example */
406
407 static void set_checkpoint_signal(void)
408 {
409 }
410
411 #else
412
413 static void checkpoint_signal(int signo)
414 {
415         checkpoint_requested = 1;
416 }
417
418 static void set_checkpoint_signal(void)
419 {
420         struct sigaction sa;
421
422         memset(&sa, 0, sizeof(sa));
423         sa.sa_handler = checkpoint_signal;
424         sigemptyset(&sa.sa_mask);
425         sa.sa_flags = SA_RESTART;
426         sigaction(SIGUSR1, &sa, NULL);
427 }
428
429 #endif
430
431 static void alloc_objects(unsigned int cnt)
432 {
433         struct object_entry_pool *b;
434
435         b = xmalloc(sizeof(struct object_entry_pool)
436                 + cnt * sizeof(struct object_entry));
437         b->next_pool = blocks;
438         b->next_free = b->entries;
439         b->end = b->entries + cnt;
440         blocks = b;
441         alloc_count += cnt;
442 }
443
444 static struct object_entry *new_object(struct object_id *oid)
445 {
446         struct object_entry *e;
447
448         if (blocks->next_free == blocks->end)
449                 alloc_objects(object_entry_alloc);
450
451         e = blocks->next_free++;
452         oidcpy(&e->idx.oid, oid);
453         return e;
454 }
455
456 static struct object_entry *find_object(struct object_id *oid)
457 {
458         unsigned int h = oid->hash[0] << 8 | oid->hash[1];
459         struct object_entry *e;
460         for (e = object_table[h]; e; e = e->next)
461                 if (oideq(oid, &e->idx.oid))
462                         return e;
463         return NULL;
464 }
465
466 static struct object_entry *insert_object(struct object_id *oid)
467 {
468         unsigned int h = oid->hash[0] << 8 | oid->hash[1];
469         struct object_entry *e = object_table[h];
470
471         while (e) {
472                 if (oideq(oid, &e->idx.oid))
473                         return e;
474                 e = e->next;
475         }
476
477         e = new_object(oid);
478         e->next = object_table[h];
479         e->idx.offset = 0;
480         object_table[h] = e;
481         return e;
482 }
483
484 static void invalidate_pack_id(unsigned int id)
485 {
486         unsigned int h;
487         unsigned long lu;
488         struct tag *t;
489
490         for (h = 0; h < ARRAY_SIZE(object_table); h++) {
491                 struct object_entry *e;
492
493                 for (e = object_table[h]; e; e = e->next)
494                         if (e->pack_id == id)
495                                 e->pack_id = MAX_PACK_ID;
496         }
497
498         for (lu = 0; lu < branch_table_sz; lu++) {
499                 struct branch *b;
500
501                 for (b = branch_table[lu]; b; b = b->table_next_branch)
502                         if (b->pack_id == id)
503                                 b->pack_id = MAX_PACK_ID;
504         }
505
506         for (t = first_tag; t; t = t->next_tag)
507                 if (t->pack_id == id)
508                         t->pack_id = MAX_PACK_ID;
509 }
510
511 static unsigned int hc_str(const char *s, size_t len)
512 {
513         unsigned int r = 0;
514         while (len-- > 0)
515                 r = r * 31 + *s++;
516         return r;
517 }
518
519 static char *pool_strdup(const char *s)
520 {
521         size_t len = strlen(s) + 1;
522         char *r = mem_pool_alloc(&fi_mem_pool, len);
523         memcpy(r, s, len);
524         return r;
525 }
526
527 static void insert_mark(struct mark_set *s, uintmax_t idnum, struct object_entry *oe)
528 {
529         while ((idnum >> s->shift) >= 1024) {
530                 s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
531                 s->shift = marks->shift + 10;
532                 s->data.sets[0] = marks;
533                 marks = s;
534         }
535         while (s->shift) {
536                 uintmax_t i = idnum >> s->shift;
537                 idnum -= i << s->shift;
538                 if (!s->data.sets[i]) {
539                         s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
540                         s->data.sets[i]->shift = s->shift - 10;
541                 }
542                 s = s->data.sets[i];
543         }
544         if (!s->data.marked[idnum])
545                 marks_set_count++;
546         s->data.marked[idnum] = oe;
547 }
548
549 static void *find_mark(struct mark_set *s, uintmax_t idnum)
550 {
551         uintmax_t orig_idnum = idnum;
552         struct object_entry *oe = NULL;
553         if ((idnum >> s->shift) < 1024) {
554                 while (s && s->shift) {
555                         uintmax_t i = idnum >> s->shift;
556                         idnum -= i << s->shift;
557                         s = s->data.sets[i];
558                 }
559                 if (s)
560                         oe = s->data.marked[idnum];
561         }
562         if (!oe)
563                 die("mark :%" PRIuMAX " not declared", orig_idnum);
564         return oe;
565 }
566
567 static struct atom_str *to_atom(const char *s, unsigned short len)
568 {
569         unsigned int hc = hc_str(s, len) % atom_table_sz;
570         struct atom_str *c;
571
572         for (c = atom_table[hc]; c; c = c->next_atom)
573                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
574                         return c;
575
576         c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
577         c->str_len = len;
578         memcpy(c->str_dat, s, len);
579         c->str_dat[len] = 0;
580         c->next_atom = atom_table[hc];
581         atom_table[hc] = c;
582         atom_cnt++;
583         return c;
584 }
585
586 static struct branch *lookup_branch(const char *name)
587 {
588         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
589         struct branch *b;
590
591         for (b = branch_table[hc]; b; b = b->table_next_branch)
592                 if (!strcmp(name, b->name))
593                         return b;
594         return NULL;
595 }
596
597 static struct branch *new_branch(const char *name)
598 {
599         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
600         struct branch *b = lookup_branch(name);
601
602         if (b)
603                 die("Invalid attempt to create duplicate branch: %s", name);
604         if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
605                 die("Branch name doesn't conform to GIT standards: %s", name);
606
607         b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
608         b->name = pool_strdup(name);
609         b->table_next_branch = branch_table[hc];
610         b->branch_tree.versions[0].mode = S_IFDIR;
611         b->branch_tree.versions[1].mode = S_IFDIR;
612         b->num_notes = 0;
613         b->active = 0;
614         b->pack_id = MAX_PACK_ID;
615         branch_table[hc] = b;
616         branch_count++;
617         return b;
618 }
619
620 static unsigned int hc_entries(unsigned int cnt)
621 {
622         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
623         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
624 }
625
626 static struct tree_content *new_tree_content(unsigned int cnt)
627 {
628         struct avail_tree_content *f, *l = NULL;
629         struct tree_content *t;
630         unsigned int hc = hc_entries(cnt);
631
632         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
633                 if (f->entry_capacity >= cnt)
634                         break;
635
636         if (f) {
637                 if (l)
638                         l->next_avail = f->next_avail;
639                 else
640                         avail_tree_table[hc] = f->next_avail;
641         } else {
642                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
643                 f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
644                 f->entry_capacity = cnt;
645         }
646
647         t = (struct tree_content*)f;
648         t->entry_count = 0;
649         t->delta_depth = 0;
650         return t;
651 }
652
653 static void release_tree_entry(struct tree_entry *e);
654 static void release_tree_content(struct tree_content *t)
655 {
656         struct avail_tree_content *f = (struct avail_tree_content*)t;
657         unsigned int hc = hc_entries(f->entry_capacity);
658         f->next_avail = avail_tree_table[hc];
659         avail_tree_table[hc] = f;
660 }
661
662 static void release_tree_content_recursive(struct tree_content *t)
663 {
664         unsigned int i;
665         for (i = 0; i < t->entry_count; i++)
666                 release_tree_entry(t->entries[i]);
667         release_tree_content(t);
668 }
669
670 static struct tree_content *grow_tree_content(
671         struct tree_content *t,
672         int amt)
673 {
674         struct tree_content *r = new_tree_content(t->entry_count + amt);
675         r->entry_count = t->entry_count;
676         r->delta_depth = t->delta_depth;
677         COPY_ARRAY(r->entries, t->entries, t->entry_count);
678         release_tree_content(t);
679         return r;
680 }
681
682 static struct tree_entry *new_tree_entry(void)
683 {
684         struct tree_entry *e;
685
686         if (!avail_tree_entry) {
687                 unsigned int n = tree_entry_alloc;
688                 tree_entry_allocd += n * sizeof(struct tree_entry);
689                 ALLOC_ARRAY(e, n);
690                 avail_tree_entry = e;
691                 while (n-- > 1) {
692                         *((void**)e) = e + 1;
693                         e++;
694                 }
695                 *((void**)e) = NULL;
696         }
697
698         e = avail_tree_entry;
699         avail_tree_entry = *((void**)e);
700         return e;
701 }
702
703 static void release_tree_entry(struct tree_entry *e)
704 {
705         if (e->tree)
706                 release_tree_content_recursive(e->tree);
707         *((void**)e) = avail_tree_entry;
708         avail_tree_entry = e;
709 }
710
711 static struct tree_content *dup_tree_content(struct tree_content *s)
712 {
713         struct tree_content *d;
714         struct tree_entry *a, *b;
715         unsigned int i;
716
717         if (!s)
718                 return NULL;
719         d = new_tree_content(s->entry_count);
720         for (i = 0; i < s->entry_count; i++) {
721                 a = s->entries[i];
722                 b = new_tree_entry();
723                 memcpy(b, a, sizeof(*a));
724                 if (a->tree && is_null_oid(&b->versions[1].oid))
725                         b->tree = dup_tree_content(a->tree);
726                 else
727                         b->tree = NULL;
728                 d->entries[i] = b;
729         }
730         d->entry_count = s->entry_count;
731         d->delta_depth = s->delta_depth;
732
733         return d;
734 }
735
736 static void start_packfile(void)
737 {
738         struct strbuf tmp_file = STRBUF_INIT;
739         struct packed_git *p;
740         struct pack_header hdr;
741         int pack_fd;
742
743         pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
744         FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
745         strbuf_release(&tmp_file);
746
747         p->pack_fd = pack_fd;
748         p->do_not_close = 1;
749         pack_file = hashfd(pack_fd, p->pack_name);
750
751         hdr.hdr_signature = htonl(PACK_SIGNATURE);
752         hdr.hdr_version = htonl(2);
753         hdr.hdr_entries = 0;
754         hashwrite(pack_file, &hdr, sizeof(hdr));
755
756         pack_data = p;
757         pack_size = sizeof(hdr);
758         object_count = 0;
759
760         REALLOC_ARRAY(all_packs, pack_id + 1);
761         all_packs[pack_id] = p;
762 }
763
764 static const char *create_index(void)
765 {
766         const char *tmpfile;
767         struct pack_idx_entry **idx, **c, **last;
768         struct object_entry *e;
769         struct object_entry_pool *o;
770
771         /* Build the table of object IDs. */
772         ALLOC_ARRAY(idx, object_count);
773         c = idx;
774         for (o = blocks; o; o = o->next_pool)
775                 for (e = o->next_free; e-- != o->entries;)
776                         if (pack_id == e->pack_id)
777                                 *c++ = &e->idx;
778         last = idx + object_count;
779         if (c != last)
780                 die("internal consistency error creating the index");
781
782         tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts,
783                                  pack_data->hash);
784         free(idx);
785         return tmpfile;
786 }
787
788 static char *keep_pack(const char *curr_index_name)
789 {
790         static const char *keep_msg = "fast-import";
791         struct strbuf name = STRBUF_INIT;
792         int keep_fd;
793
794         odb_pack_name(&name, pack_data->hash, "keep");
795         keep_fd = odb_pack_keep(name.buf);
796         if (keep_fd < 0)
797                 die_errno("cannot create keep file");
798         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
799         if (close(keep_fd))
800                 die_errno("failed to write keep file");
801
802         odb_pack_name(&name, pack_data->hash, "pack");
803         if (finalize_object_file(pack_data->pack_name, name.buf))
804                 die("cannot store pack file");
805
806         odb_pack_name(&name, pack_data->hash, "idx");
807         if (finalize_object_file(curr_index_name, name.buf))
808                 die("cannot store index file");
809         free((void *)curr_index_name);
810         return strbuf_detach(&name, NULL);
811 }
812
813 static void unkeep_all_packs(void)
814 {
815         struct strbuf name = STRBUF_INIT;
816         int k;
817
818         for (k = 0; k < pack_id; k++) {
819                 struct packed_git *p = all_packs[k];
820                 odb_pack_name(&name, p->hash, "keep");
821                 unlink_or_warn(name.buf);
822         }
823         strbuf_release(&name);
824 }
825
826 static int loosen_small_pack(const struct packed_git *p)
827 {
828         struct child_process unpack = CHILD_PROCESS_INIT;
829
830         if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
831                 die_errno("Failed seeking to start of '%s'", p->pack_name);
832
833         unpack.in = p->pack_fd;
834         unpack.git_cmd = 1;
835         unpack.stdout_to_stderr = 1;
836         argv_array_push(&unpack.args, "unpack-objects");
837         if (!show_stats)
838                 argv_array_push(&unpack.args, "-q");
839
840         return run_command(&unpack);
841 }
842
843 static void end_packfile(void)
844 {
845         static int running;
846
847         if (running || !pack_data)
848                 return;
849
850         running = 1;
851         clear_delta_base_cache();
852         if (object_count) {
853                 struct packed_git *new_p;
854                 struct object_id cur_pack_oid;
855                 char *idx_name;
856                 int i;
857                 struct branch *b;
858                 struct tag *t;
859
860                 close_pack_windows(pack_data);
861                 finalize_hashfile(pack_file, cur_pack_oid.hash, 0);
862                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->hash,
863                                          pack_data->pack_name, object_count,
864                                          cur_pack_oid.hash, pack_size);
865
866                 if (object_count <= unpack_limit) {
867                         if (!loosen_small_pack(pack_data)) {
868                                 invalidate_pack_id(pack_id);
869                                 goto discard_pack;
870                         }
871                 }
872
873                 close(pack_data->pack_fd);
874                 idx_name = keep_pack(create_index());
875
876                 /* Register the packfile with core git's machinery. */
877                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
878                 if (!new_p)
879                         die("core git rejected index %s", idx_name);
880                 all_packs[pack_id] = new_p;
881                 install_packed_git(the_repository, new_p);
882                 free(idx_name);
883
884                 /* Print the boundary */
885                 if (pack_edges) {
886                         fprintf(pack_edges, "%s:", new_p->pack_name);
887                         for (i = 0; i < branch_table_sz; i++) {
888                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
889                                         if (b->pack_id == pack_id)
890                                                 fprintf(pack_edges, " %s",
891                                                         oid_to_hex(&b->oid));
892                                 }
893                         }
894                         for (t = first_tag; t; t = t->next_tag) {
895                                 if (t->pack_id == pack_id)
896                                         fprintf(pack_edges, " %s",
897                                                 oid_to_hex(&t->oid));
898                         }
899                         fputc('\n', pack_edges);
900                         fflush(pack_edges);
901                 }
902
903                 pack_id++;
904         }
905         else {
906 discard_pack:
907                 close(pack_data->pack_fd);
908                 unlink_or_warn(pack_data->pack_name);
909         }
910         FREE_AND_NULL(pack_data);
911         running = 0;
912
913         /* We can't carry a delta across packfiles. */
914         strbuf_release(&last_blob.data);
915         last_blob.offset = 0;
916         last_blob.depth = 0;
917 }
918
919 static void cycle_packfile(void)
920 {
921         end_packfile();
922         start_packfile();
923 }
924
925 static int store_object(
926         enum object_type type,
927         struct strbuf *dat,
928         struct last_object *last,
929         struct object_id *oidout,
930         uintmax_t mark)
931 {
932         void *out, *delta;
933         struct object_entry *e;
934         unsigned char hdr[96];
935         struct object_id oid;
936         unsigned long hdrlen, deltalen;
937         git_hash_ctx c;
938         git_zstream s;
939
940         hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
941                            type_name(type), (unsigned long)dat->len) + 1;
942         the_hash_algo->init_fn(&c);
943         the_hash_algo->update_fn(&c, hdr, hdrlen);
944         the_hash_algo->update_fn(&c, dat->buf, dat->len);
945         the_hash_algo->final_fn(oid.hash, &c);
946         if (oidout)
947                 oidcpy(oidout, &oid);
948
949         e = insert_object(&oid);
950         if (mark)
951                 insert_mark(marks, mark, e);
952         if (e->idx.offset) {
953                 duplicate_count_by_type[type]++;
954                 return 1;
955         } else if (find_sha1_pack(oid.hash,
956                                   get_all_packs(the_repository))) {
957                 e->type = type;
958                 e->pack_id = MAX_PACK_ID;
959                 e->idx.offset = 1; /* just not zero! */
960                 duplicate_count_by_type[type]++;
961                 return 1;
962         }
963
964         if (last && last->data.len && last->data.buf && last->depth < max_depth
965                 && dat->len > the_hash_algo->rawsz) {
966
967                 delta_count_attempts_by_type[type]++;
968                 delta = diff_delta(last->data.buf, last->data.len,
969                         dat->buf, dat->len,
970                         &deltalen, dat->len - the_hash_algo->rawsz);
971         } else
972                 delta = NULL;
973
974         git_deflate_init(&s, pack_compression_level);
975         if (delta) {
976                 s.next_in = delta;
977                 s.avail_in = deltalen;
978         } else {
979                 s.next_in = (void *)dat->buf;
980                 s.avail_in = dat->len;
981         }
982         s.avail_out = git_deflate_bound(&s, s.avail_in);
983         s.next_out = out = xmalloc(s.avail_out);
984         while (git_deflate(&s, Z_FINISH) == Z_OK)
985                 ; /* nothing */
986         git_deflate_end(&s);
987
988         /* Determine if we should auto-checkpoint. */
989         if ((max_packsize
990                 && (pack_size + PACK_SIZE_THRESHOLD + s.total_out) > max_packsize)
991                 || (pack_size + PACK_SIZE_THRESHOLD + s.total_out) < pack_size) {
992
993                 /* This new object needs to *not* have the current pack_id. */
994                 e->pack_id = pack_id + 1;
995                 cycle_packfile();
996
997                 /* We cannot carry a delta into the new pack. */
998                 if (delta) {
999                         FREE_AND_NULL(delta);
1000
1001                         git_deflate_init(&s, pack_compression_level);
1002                         s.next_in = (void *)dat->buf;
1003                         s.avail_in = dat->len;
1004                         s.avail_out = git_deflate_bound(&s, s.avail_in);
1005                         s.next_out = out = xrealloc(out, s.avail_out);
1006                         while (git_deflate(&s, Z_FINISH) == Z_OK)
1007                                 ; /* nothing */
1008                         git_deflate_end(&s);
1009                 }
1010         }
1011
1012         e->type = type;
1013         e->pack_id = pack_id;
1014         e->idx.offset = pack_size;
1015         object_count++;
1016         object_count_by_type[type]++;
1017
1018         crc32_begin(pack_file);
1019
1020         if (delta) {
1021                 off_t ofs = e->idx.offset - last->offset;
1022                 unsigned pos = sizeof(hdr) - 1;
1023
1024                 delta_count_by_type[type]++;
1025                 e->depth = last->depth + 1;
1026
1027                 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1028                                                       OBJ_OFS_DELTA, deltalen);
1029                 hashwrite(pack_file, hdr, hdrlen);
1030                 pack_size += hdrlen;
1031
1032                 hdr[pos] = ofs & 127;
1033                 while (ofs >>= 7)
1034                         hdr[--pos] = 128 | (--ofs & 127);
1035                 hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1036                 pack_size += sizeof(hdr) - pos;
1037         } else {
1038                 e->depth = 0;
1039                 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1040                                                       type, dat->len);
1041                 hashwrite(pack_file, hdr, hdrlen);
1042                 pack_size += hdrlen;
1043         }
1044
1045         hashwrite(pack_file, out, s.total_out);
1046         pack_size += s.total_out;
1047
1048         e->idx.crc32 = crc32_end(pack_file);
1049
1050         free(out);
1051         free(delta);
1052         if (last) {
1053                 if (last->no_swap) {
1054                         last->data = *dat;
1055                 } else {
1056                         strbuf_swap(&last->data, dat);
1057                 }
1058                 last->offset = e->idx.offset;
1059                 last->depth = e->depth;
1060         }
1061         return 0;
1062 }
1063
1064 static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1065 {
1066         if (hashfile_truncate(pack_file, checkpoint))
1067                 die_errno("cannot truncate pack to skip duplicate");
1068         pack_size = checkpoint->offset;
1069 }
1070
1071 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1072 {
1073         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1074         unsigned char *in_buf = xmalloc(in_sz);
1075         unsigned char *out_buf = xmalloc(out_sz);
1076         struct object_entry *e;
1077         struct object_id oid;
1078         unsigned long hdrlen;
1079         off_t offset;
1080         git_hash_ctx c;
1081         git_zstream s;
1082         struct hashfile_checkpoint checkpoint;
1083         int status = Z_OK;
1084
1085         /* Determine if we should auto-checkpoint. */
1086         if ((max_packsize
1087                 && (pack_size + PACK_SIZE_THRESHOLD + len) > max_packsize)
1088                 || (pack_size + PACK_SIZE_THRESHOLD + len) < pack_size)
1089                 cycle_packfile();
1090
1091         hashfile_checkpoint(pack_file, &checkpoint);
1092         offset = checkpoint.offset;
1093
1094         hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1095
1096         the_hash_algo->init_fn(&c);
1097         the_hash_algo->update_fn(&c, out_buf, hdrlen);
1098
1099         crc32_begin(pack_file);
1100
1101         git_deflate_init(&s, pack_compression_level);
1102
1103         hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1104
1105         s.next_out = out_buf + hdrlen;
1106         s.avail_out = out_sz - hdrlen;
1107
1108         while (status != Z_STREAM_END) {
1109                 if (0 < len && !s.avail_in) {
1110                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1111                         size_t n = fread(in_buf, 1, cnt, stdin);
1112                         if (!n && feof(stdin))
1113                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1114
1115                         the_hash_algo->update_fn(&c, in_buf, n);
1116                         s.next_in = in_buf;
1117                         s.avail_in = n;
1118                         len -= n;
1119                 }
1120
1121                 status = git_deflate(&s, len ? 0 : Z_FINISH);
1122
1123                 if (!s.avail_out || status == Z_STREAM_END) {
1124                         size_t n = s.next_out - out_buf;
1125                         hashwrite(pack_file, out_buf, n);
1126                         pack_size += n;
1127                         s.next_out = out_buf;
1128                         s.avail_out = out_sz;
1129                 }
1130
1131                 switch (status) {
1132                 case Z_OK:
1133                 case Z_BUF_ERROR:
1134                 case Z_STREAM_END:
1135                         continue;
1136                 default:
1137                         die("unexpected deflate failure: %d", status);
1138                 }
1139         }
1140         git_deflate_end(&s);
1141         the_hash_algo->final_fn(oid.hash, &c);
1142
1143         if (oidout)
1144                 oidcpy(oidout, &oid);
1145
1146         e = insert_object(&oid);
1147
1148         if (mark)
1149                 insert_mark(marks, mark, e);
1150
1151         if (e->idx.offset) {
1152                 duplicate_count_by_type[OBJ_BLOB]++;
1153                 truncate_pack(&checkpoint);
1154
1155         } else if (find_sha1_pack(oid.hash,
1156                                   get_all_packs(the_repository))) {
1157                 e->type = OBJ_BLOB;
1158                 e->pack_id = MAX_PACK_ID;
1159                 e->idx.offset = 1; /* just not zero! */
1160                 duplicate_count_by_type[OBJ_BLOB]++;
1161                 truncate_pack(&checkpoint);
1162
1163         } else {
1164                 e->depth = 0;
1165                 e->type = OBJ_BLOB;
1166                 e->pack_id = pack_id;
1167                 e->idx.offset = offset;
1168                 e->idx.crc32 = crc32_end(pack_file);
1169                 object_count++;
1170                 object_count_by_type[OBJ_BLOB]++;
1171         }
1172
1173         free(in_buf);
1174         free(out_buf);
1175 }
1176
1177 /* All calls must be guarded by find_object() or find_mark() to
1178  * ensure the 'struct object_entry' passed was written by this
1179  * process instance.  We unpack the entry by the offset, avoiding
1180  * the need for the corresponding .idx file.  This unpacking rule
1181  * works because we only use OBJ_REF_DELTA within the packfiles
1182  * created by fast-import.
1183  *
1184  * oe must not be NULL.  Such an oe usually comes from giving
1185  * an unknown SHA-1 to find_object() or an undefined mark to
1186  * find_mark().  Callers must test for this condition and use
1187  * the standard read_sha1_file() when it happens.
1188  *
1189  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1190  * find_mark(), where the mark was reloaded from an existing marks
1191  * file and is referencing an object that this fast-import process
1192  * instance did not write out to a packfile.  Callers must test for
1193  * this condition and use read_sha1_file() instead.
1194  */
1195 static void *gfi_unpack_entry(
1196         struct object_entry *oe,
1197         unsigned long *sizep)
1198 {
1199         enum object_type type;
1200         struct packed_git *p = all_packs[oe->pack_id];
1201         if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1202                 /* The object is stored in the packfile we are writing to
1203                  * and we have modified it since the last time we scanned
1204                  * back to read a previously written object.  If an old
1205                  * window covered [p->pack_size, p->pack_size + rawsz) its
1206                  * data is stale and is not valid.  Closing all windows
1207                  * and updating the packfile length ensures we can read
1208                  * the newly written data.
1209                  */
1210                 close_pack_windows(p);
1211                 hashflush(pack_file);
1212
1213                 /* We have to offer rawsz bytes additional on the end of
1214                  * the packfile as the core unpacker code assumes the
1215                  * footer is present at the file end and must promise
1216                  * at least rawsz bytes within any window it maps.  But
1217                  * we don't actually create the footer here.
1218                  */
1219                 p->pack_size = pack_size + the_hash_algo->rawsz;
1220         }
1221         return unpack_entry(the_repository, p, oe->idx.offset, &type, sizep);
1222 }
1223
1224 static const char *get_mode(const char *str, uint16_t *modep)
1225 {
1226         unsigned char c;
1227         uint16_t mode = 0;
1228
1229         while ((c = *str++) != ' ') {
1230                 if (c < '0' || c > '7')
1231                         return NULL;
1232                 mode = (mode << 3) + (c - '0');
1233         }
1234         *modep = mode;
1235         return str;
1236 }
1237
1238 static void load_tree(struct tree_entry *root)
1239 {
1240         struct object_id *oid = &root->versions[1].oid;
1241         struct object_entry *myoe;
1242         struct tree_content *t;
1243         unsigned long size;
1244         char *buf;
1245         const char *c;
1246
1247         root->tree = t = new_tree_content(8);
1248         if (is_null_oid(oid))
1249                 return;
1250
1251         myoe = find_object(oid);
1252         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1253                 if (myoe->type != OBJ_TREE)
1254                         die("Not a tree: %s", oid_to_hex(oid));
1255                 t->delta_depth = myoe->depth;
1256                 buf = gfi_unpack_entry(myoe, &size);
1257                 if (!buf)
1258                         die("Can't load tree %s", oid_to_hex(oid));
1259         } else {
1260                 enum object_type type;
1261                 buf = read_object_file(oid, &type, &size);
1262                 if (!buf || type != OBJ_TREE)
1263                         die("Can't load tree %s", oid_to_hex(oid));
1264         }
1265
1266         c = buf;
1267         while (c != (buf + size)) {
1268                 struct tree_entry *e = new_tree_entry();
1269
1270                 if (t->entry_count == t->entry_capacity)
1271                         root->tree = t = grow_tree_content(t, t->entry_count);
1272                 t->entries[t->entry_count++] = e;
1273
1274                 e->tree = NULL;
1275                 c = get_mode(c, &e->versions[1].mode);
1276                 if (!c)
1277                         die("Corrupt mode in %s", oid_to_hex(oid));
1278                 e->versions[0].mode = e->versions[1].mode;
1279                 e->name = to_atom(c, strlen(c));
1280                 c += e->name->str_len + 1;
1281                 hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1282                 hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1283                 c += the_hash_algo->rawsz;
1284         }
1285         free(buf);
1286 }
1287
1288 static int tecmp0 (const void *_a, const void *_b)
1289 {
1290         struct tree_entry *a = *((struct tree_entry**)_a);
1291         struct tree_entry *b = *((struct tree_entry**)_b);
1292         return base_name_compare(
1293                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1294                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1295 }
1296
1297 static int tecmp1 (const void *_a, const void *_b)
1298 {
1299         struct tree_entry *a = *((struct tree_entry**)_a);
1300         struct tree_entry *b = *((struct tree_entry**)_b);
1301         return base_name_compare(
1302                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1303                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1304 }
1305
1306 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1307 {
1308         size_t maxlen = 0;
1309         unsigned int i;
1310
1311         if (!v)
1312                 QSORT(t->entries, t->entry_count, tecmp0);
1313         else
1314                 QSORT(t->entries, t->entry_count, tecmp1);
1315
1316         for (i = 0; i < t->entry_count; i++) {
1317                 if (t->entries[i]->versions[v].mode)
1318                         maxlen += t->entries[i]->name->str_len + 34;
1319         }
1320
1321         strbuf_reset(b);
1322         strbuf_grow(b, maxlen);
1323         for (i = 0; i < t->entry_count; i++) {
1324                 struct tree_entry *e = t->entries[i];
1325                 if (!e->versions[v].mode)
1326                         continue;
1327                 strbuf_addf(b, "%o %s%c",
1328                         (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1329                         e->name->str_dat, '\0');
1330                 strbuf_add(b, e->versions[v].oid.hash, the_hash_algo->rawsz);
1331         }
1332 }
1333
1334 static void store_tree(struct tree_entry *root)
1335 {
1336         struct tree_content *t;
1337         unsigned int i, j, del;
1338         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1339         struct object_entry *le = NULL;
1340
1341         if (!is_null_oid(&root->versions[1].oid))
1342                 return;
1343
1344         if (!root->tree)
1345                 load_tree(root);
1346         t = root->tree;
1347
1348         for (i = 0; i < t->entry_count; i++) {
1349                 if (t->entries[i]->tree)
1350                         store_tree(t->entries[i]);
1351         }
1352
1353         if (!(root->versions[0].mode & NO_DELTA))
1354                 le = find_object(&root->versions[0].oid);
1355         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1356                 mktree(t, 0, &old_tree);
1357                 lo.data = old_tree;
1358                 lo.offset = le->idx.offset;
1359                 lo.depth = t->delta_depth;
1360         }
1361
1362         mktree(t, 1, &new_tree);
1363         store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1364
1365         t->delta_depth = lo.depth;
1366         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1367                 struct tree_entry *e = t->entries[i];
1368                 if (e->versions[1].mode) {
1369                         e->versions[0].mode = e->versions[1].mode;
1370                         oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1371                         t->entries[j++] = e;
1372                 } else {
1373                         release_tree_entry(e);
1374                         del++;
1375                 }
1376         }
1377         t->entry_count -= del;
1378 }
1379
1380 static void tree_content_replace(
1381         struct tree_entry *root,
1382         const struct object_id *oid,
1383         const uint16_t mode,
1384         struct tree_content *newtree)
1385 {
1386         if (!S_ISDIR(mode))
1387                 die("Root cannot be a non-directory");
1388         oidclr(&root->versions[0].oid);
1389         oidcpy(&root->versions[1].oid, oid);
1390         if (root->tree)
1391                 release_tree_content_recursive(root->tree);
1392         root->tree = newtree;
1393 }
1394
1395 static int tree_content_set(
1396         struct tree_entry *root,
1397         const char *p,
1398         const struct object_id *oid,
1399         const uint16_t mode,
1400         struct tree_content *subtree)
1401 {
1402         struct tree_content *t;
1403         const char *slash1;
1404         unsigned int i, n;
1405         struct tree_entry *e;
1406
1407         slash1 = strchrnul(p, '/');
1408         n = slash1 - p;
1409         if (!n)
1410                 die("Empty path component found in input");
1411         if (!*slash1 && !S_ISDIR(mode) && subtree)
1412                 die("Non-directories cannot have subtrees");
1413
1414         if (!root->tree)
1415                 load_tree(root);
1416         t = root->tree;
1417         for (i = 0; i < t->entry_count; i++) {
1418                 e = t->entries[i];
1419                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1420                         if (!*slash1) {
1421                                 if (!S_ISDIR(mode)
1422                                                 && e->versions[1].mode == mode
1423                                                 && oideq(&e->versions[1].oid, oid))
1424                                         return 0;
1425                                 e->versions[1].mode = mode;
1426                                 oidcpy(&e->versions[1].oid, oid);
1427                                 if (e->tree)
1428                                         release_tree_content_recursive(e->tree);
1429                                 e->tree = subtree;
1430
1431                                 /*
1432                                  * We need to leave e->versions[0].sha1 alone
1433                                  * to avoid modifying the preimage tree used
1434                                  * when writing out the parent directory.
1435                                  * But after replacing the subdir with a
1436                                  * completely different one, it's not a good
1437                                  * delta base any more, and besides, we've
1438                                  * thrown away the tree entries needed to
1439                                  * make a delta against it.
1440                                  *
1441                                  * So let's just explicitly disable deltas
1442                                  * for the subtree.
1443                                  */
1444                                 if (S_ISDIR(e->versions[0].mode))
1445                                         e->versions[0].mode |= NO_DELTA;
1446
1447                                 oidclr(&root->versions[1].oid);
1448                                 return 1;
1449                         }
1450                         if (!S_ISDIR(e->versions[1].mode)) {
1451                                 e->tree = new_tree_content(8);
1452                                 e->versions[1].mode = S_IFDIR;
1453                         }
1454                         if (!e->tree)
1455                                 load_tree(e);
1456                         if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1457                                 oidclr(&root->versions[1].oid);
1458                                 return 1;
1459                         }
1460                         return 0;
1461                 }
1462         }
1463
1464         if (t->entry_count == t->entry_capacity)
1465                 root->tree = t = grow_tree_content(t, t->entry_count);
1466         e = new_tree_entry();
1467         e->name = to_atom(p, n);
1468         e->versions[0].mode = 0;
1469         oidclr(&e->versions[0].oid);
1470         t->entries[t->entry_count++] = e;
1471         if (*slash1) {
1472                 e->tree = new_tree_content(8);
1473                 e->versions[1].mode = S_IFDIR;
1474                 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1475         } else {
1476                 e->tree = subtree;
1477                 e->versions[1].mode = mode;
1478                 oidcpy(&e->versions[1].oid, oid);
1479         }
1480         oidclr(&root->versions[1].oid);
1481         return 1;
1482 }
1483
1484 static int tree_content_remove(
1485         struct tree_entry *root,
1486         const char *p,
1487         struct tree_entry *backup_leaf,
1488         int allow_root)
1489 {
1490         struct tree_content *t;
1491         const char *slash1;
1492         unsigned int i, n;
1493         struct tree_entry *e;
1494
1495         slash1 = strchrnul(p, '/');
1496         n = slash1 - p;
1497
1498         if (!root->tree)
1499                 load_tree(root);
1500
1501         if (!*p && allow_root) {
1502                 e = root;
1503                 goto del_entry;
1504         }
1505
1506         t = root->tree;
1507         for (i = 0; i < t->entry_count; i++) {
1508                 e = t->entries[i];
1509                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1510                         if (*slash1 && !S_ISDIR(e->versions[1].mode))
1511                                 /*
1512                                  * If p names a file in some subdirectory, and a
1513                                  * file or symlink matching the name of the
1514                                  * parent directory of p exists, then p cannot
1515                                  * exist and need not be deleted.
1516                                  */
1517                                 return 1;
1518                         if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1519                                 goto del_entry;
1520                         if (!e->tree)
1521                                 load_tree(e);
1522                         if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1523                                 for (n = 0; n < e->tree->entry_count; n++) {
1524                                         if (e->tree->entries[n]->versions[1].mode) {
1525                                                 oidclr(&root->versions[1].oid);
1526                                                 return 1;
1527                                         }
1528                                 }
1529                                 backup_leaf = NULL;
1530                                 goto del_entry;
1531                         }
1532                         return 0;
1533                 }
1534         }
1535         return 0;
1536
1537 del_entry:
1538         if (backup_leaf)
1539                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1540         else if (e->tree)
1541                 release_tree_content_recursive(e->tree);
1542         e->tree = NULL;
1543         e->versions[1].mode = 0;
1544         oidclr(&e->versions[1].oid);
1545         oidclr(&root->versions[1].oid);
1546         return 1;
1547 }
1548
1549 static int tree_content_get(
1550         struct tree_entry *root,
1551         const char *p,
1552         struct tree_entry *leaf,
1553         int allow_root)
1554 {
1555         struct tree_content *t;
1556         const char *slash1;
1557         unsigned int i, n;
1558         struct tree_entry *e;
1559
1560         slash1 = strchrnul(p, '/');
1561         n = slash1 - p;
1562         if (!n && !allow_root)
1563                 die("Empty path component found in input");
1564
1565         if (!root->tree)
1566                 load_tree(root);
1567
1568         if (!n) {
1569                 e = root;
1570                 goto found_entry;
1571         }
1572
1573         t = root->tree;
1574         for (i = 0; i < t->entry_count; i++) {
1575                 e = t->entries[i];
1576                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1577                         if (!*slash1)
1578                                 goto found_entry;
1579                         if (!S_ISDIR(e->versions[1].mode))
1580                                 return 0;
1581                         if (!e->tree)
1582                                 load_tree(e);
1583                         return tree_content_get(e, slash1 + 1, leaf, 0);
1584                 }
1585         }
1586         return 0;
1587
1588 found_entry:
1589         memcpy(leaf, e, sizeof(*leaf));
1590         if (e->tree && is_null_oid(&e->versions[1].oid))
1591                 leaf->tree = dup_tree_content(e->tree);
1592         else
1593                 leaf->tree = NULL;
1594         return 1;
1595 }
1596
1597 static int update_branch(struct branch *b)
1598 {
1599         static const char *msg = "fast-import";
1600         struct ref_transaction *transaction;
1601         struct object_id old_oid;
1602         struct strbuf err = STRBUF_INIT;
1603
1604         if (is_null_oid(&b->oid)) {
1605                 if (b->delete)
1606                         delete_ref(NULL, b->name, NULL, 0);
1607                 return 0;
1608         }
1609         if (read_ref(b->name, &old_oid))
1610                 oidclr(&old_oid);
1611         if (!force_update && !is_null_oid(&old_oid)) {
1612                 struct commit *old_cmit, *new_cmit;
1613
1614                 old_cmit = lookup_commit_reference_gently(the_repository,
1615                                                           &old_oid, 0);
1616                 new_cmit = lookup_commit_reference_gently(the_repository,
1617                                                           &b->oid, 0);
1618                 if (!old_cmit || !new_cmit)
1619                         return error("Branch %s is missing commits.", b->name);
1620
1621                 if (!in_merge_bases(old_cmit, new_cmit)) {
1622                         warning("Not updating %s"
1623                                 " (new tip %s does not contain %s)",
1624                                 b->name, oid_to_hex(&b->oid),
1625                                 oid_to_hex(&old_oid));
1626                         return -1;
1627                 }
1628         }
1629         transaction = ref_transaction_begin(&err);
1630         if (!transaction ||
1631             ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1632                                    0, msg, &err) ||
1633             ref_transaction_commit(transaction, &err)) {
1634                 ref_transaction_free(transaction);
1635                 error("%s", err.buf);
1636                 strbuf_release(&err);
1637                 return -1;
1638         }
1639         ref_transaction_free(transaction);
1640         strbuf_release(&err);
1641         return 0;
1642 }
1643
1644 static void dump_branches(void)
1645 {
1646         unsigned int i;
1647         struct branch *b;
1648
1649         for (i = 0; i < branch_table_sz; i++) {
1650                 for (b = branch_table[i]; b; b = b->table_next_branch)
1651                         failure |= update_branch(b);
1652         }
1653 }
1654
1655 static void dump_tags(void)
1656 {
1657         static const char *msg = "fast-import";
1658         struct tag *t;
1659         struct strbuf ref_name = STRBUF_INIT;
1660         struct strbuf err = STRBUF_INIT;
1661         struct ref_transaction *transaction;
1662
1663         transaction = ref_transaction_begin(&err);
1664         if (!transaction) {
1665                 failure |= error("%s", err.buf);
1666                 goto cleanup;
1667         }
1668         for (t = first_tag; t; t = t->next_tag) {
1669                 strbuf_reset(&ref_name);
1670                 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1671
1672                 if (ref_transaction_update(transaction, ref_name.buf,
1673                                            &t->oid, NULL, 0, msg, &err)) {
1674                         failure |= error("%s", err.buf);
1675                         goto cleanup;
1676                 }
1677         }
1678         if (ref_transaction_commit(transaction, &err))
1679                 failure |= error("%s", err.buf);
1680
1681  cleanup:
1682         ref_transaction_free(transaction);
1683         strbuf_release(&ref_name);
1684         strbuf_release(&err);
1685 }
1686
1687 static void dump_marks(void)
1688 {
1689         struct lock_file mark_lock = LOCK_INIT;
1690         FILE *f;
1691
1692         if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1693                 return;
1694
1695         if (safe_create_leading_directories_const(export_marks_file)) {
1696                 failure |= error_errno("unable to create leading directories of %s",
1697                                        export_marks_file);
1698                 return;
1699         }
1700
1701         if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1702                 failure |= error_errno("Unable to write marks file %s",
1703                                        export_marks_file);
1704                 return;
1705         }
1706
1707         f = fdopen_lock_file(&mark_lock, "w");
1708         if (!f) {
1709                 int saved_errno = errno;
1710                 rollback_lock_file(&mark_lock);
1711                 failure |= error("Unable to write marks file %s: %s",
1712                         export_marks_file, strerror(saved_errno));
1713                 return;
1714         }
1715
1716         for_each_mark(marks, 0, dump_marks_fn, f);
1717         if (commit_lock_file(&mark_lock)) {
1718                 failure |= error_errno("Unable to write file %s",
1719                                        export_marks_file);
1720                 return;
1721         }
1722 }
1723
1724 static void insert_object_entry(struct mark_set *s, struct object_id *oid, uintmax_t mark)
1725 {
1726         struct object_entry *e;
1727         e = find_object(oid);
1728         if (!e) {
1729                 enum object_type type = oid_object_info(the_repository,
1730                                                         oid, NULL);
1731                 if (type < 0)
1732                         die("object not found: %s", oid_to_hex(oid));
1733                 e = insert_object(oid);
1734                 e->type = type;
1735                 e->pack_id = MAX_PACK_ID;
1736                 e->idx.offset = 1; /* just not zero! */
1737         }
1738         insert_mark(s, mark, e);
1739 }
1740
1741 static void insert_oid_entry(struct mark_set *s, struct object_id *oid, uintmax_t mark)
1742 {
1743         insert_mark(s, mark, xmemdupz(oid, sizeof(*oid)));
1744 }
1745
1746 static void read_mark_file(struct mark_set *s, FILE *f, mark_set_inserter_t inserter)
1747 {
1748         char line[512];
1749         while (fgets(line, sizeof(line), f)) {
1750                 uintmax_t mark;
1751                 char *end;
1752                 struct object_id oid;
1753
1754                 /* Ensure SHA-1 objects are padded with zeros. */
1755                 memset(oid.hash, 0, sizeof(oid.hash));
1756
1757                 end = strchr(line, '\n');
1758                 if (line[0] != ':' || !end)
1759                         die("corrupt mark line: %s", line);
1760                 *end = 0;
1761                 mark = strtoumax(line + 1, &end, 10);
1762                 if (!mark || end == line + 1
1763                         || *end != ' '
1764                         || get_oid_hex_any(end + 1, &oid) == GIT_HASH_UNKNOWN)
1765                         die("corrupt mark line: %s", line);
1766                 inserter(s, &oid, mark);
1767         }
1768 }
1769
1770 static void read_marks(void)
1771 {
1772         FILE *f = fopen(import_marks_file, "r");
1773         if (f)
1774                 ;
1775         else if (import_marks_file_ignore_missing && errno == ENOENT)
1776                 goto done; /* Marks file does not exist */
1777         else
1778                 die_errno("cannot read '%s'", import_marks_file);
1779         read_mark_file(marks, f, insert_object_entry);
1780         fclose(f);
1781 done:
1782         import_marks_file_done = 1;
1783 }
1784
1785
1786 static int read_next_command(void)
1787 {
1788         static int stdin_eof = 0;
1789
1790         if (stdin_eof) {
1791                 unread_command_buf = 0;
1792                 return EOF;
1793         }
1794
1795         for (;;) {
1796                 if (unread_command_buf) {
1797                         unread_command_buf = 0;
1798                 } else {
1799                         struct recent_command *rc;
1800
1801                         stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1802                         if (stdin_eof)
1803                                 return EOF;
1804
1805                         if (!seen_data_command
1806                                 && !starts_with(command_buf.buf, "feature ")
1807                                 && !starts_with(command_buf.buf, "option ")) {
1808                                 parse_argv();
1809                         }
1810
1811                         rc = rc_free;
1812                         if (rc)
1813                                 rc_free = rc->next;
1814                         else {
1815                                 rc = cmd_hist.next;
1816                                 cmd_hist.next = rc->next;
1817                                 cmd_hist.next->prev = &cmd_hist;
1818                                 free(rc->buf);
1819                         }
1820
1821                         rc->buf = xstrdup(command_buf.buf);
1822                         rc->prev = cmd_tail;
1823                         rc->next = cmd_hist.prev;
1824                         rc->prev->next = rc;
1825                         cmd_tail = rc;
1826                 }
1827                 if (command_buf.buf[0] == '#')
1828                         continue;
1829                 return 0;
1830         }
1831 }
1832
1833 static void skip_optional_lf(void)
1834 {
1835         int term_char = fgetc(stdin);
1836         if (term_char != '\n' && term_char != EOF)
1837                 ungetc(term_char, stdin);
1838 }
1839
1840 static void parse_mark(void)
1841 {
1842         const char *v;
1843         if (skip_prefix(command_buf.buf, "mark :", &v)) {
1844                 next_mark = strtoumax(v, NULL, 10);
1845                 read_next_command();
1846         }
1847         else
1848                 next_mark = 0;
1849 }
1850
1851 static void parse_original_identifier(void)
1852 {
1853         const char *v;
1854         if (skip_prefix(command_buf.buf, "original-oid ", &v))
1855                 read_next_command();
1856 }
1857
1858 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1859 {
1860         const char *data;
1861         strbuf_reset(sb);
1862
1863         if (!skip_prefix(command_buf.buf, "data ", &data))
1864                 die("Expected 'data n' command, found: %s", command_buf.buf);
1865
1866         if (skip_prefix(data, "<<", &data)) {
1867                 char *term = xstrdup(data);
1868                 size_t term_len = command_buf.len - (data - command_buf.buf);
1869
1870                 for (;;) {
1871                         if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1872                                 die("EOF in data (terminator '%s' not found)", term);
1873                         if (term_len == command_buf.len
1874                                 && !strcmp(term, command_buf.buf))
1875                                 break;
1876                         strbuf_addbuf(sb, &command_buf);
1877                         strbuf_addch(sb, '\n');
1878                 }
1879                 free(term);
1880         }
1881         else {
1882                 uintmax_t len = strtoumax(data, NULL, 10);
1883                 size_t n = 0, length = (size_t)len;
1884
1885                 if (limit && limit < len) {
1886                         *len_res = len;
1887                         return 0;
1888                 }
1889                 if (length < len)
1890                         die("data is too large to use in this context");
1891
1892                 while (n < length) {
1893                         size_t s = strbuf_fread(sb, length - n, stdin);
1894                         if (!s && feof(stdin))
1895                                 die("EOF in data (%lu bytes remaining)",
1896                                         (unsigned long)(length - n));
1897                         n += s;
1898                 }
1899         }
1900
1901         skip_optional_lf();
1902         return 1;
1903 }
1904
1905 static int validate_raw_date(const char *src, struct strbuf *result)
1906 {
1907         const char *orig_src = src;
1908         char *endp;
1909         unsigned long num;
1910
1911         errno = 0;
1912
1913         num = strtoul(src, &endp, 10);
1914         /* NEEDSWORK: perhaps check for reasonable values? */
1915         if (errno || endp == src || *endp != ' ')
1916                 return -1;
1917
1918         src = endp + 1;
1919         if (*src != '-' && *src != '+')
1920                 return -1;
1921
1922         num = strtoul(src + 1, &endp, 10);
1923         if (errno || endp == src + 1 || *endp || 1400 < num)
1924                 return -1;
1925
1926         strbuf_addstr(result, orig_src);
1927         return 0;
1928 }
1929
1930 static char *parse_ident(const char *buf)
1931 {
1932         const char *ltgt;
1933         size_t name_len;
1934         struct strbuf ident = STRBUF_INIT;
1935
1936         /* ensure there is a space delimiter even if there is no name */
1937         if (*buf == '<')
1938                 --buf;
1939
1940         ltgt = buf + strcspn(buf, "<>");
1941         if (*ltgt != '<')
1942                 die("Missing < in ident string: %s", buf);
1943         if (ltgt != buf && ltgt[-1] != ' ')
1944                 die("Missing space before < in ident string: %s", buf);
1945         ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
1946         if (*ltgt != '>')
1947                 die("Missing > in ident string: %s", buf);
1948         ltgt++;
1949         if (*ltgt != ' ')
1950                 die("Missing space after > in ident string: %s", buf);
1951         ltgt++;
1952         name_len = ltgt - buf;
1953         strbuf_add(&ident, buf, name_len);
1954
1955         switch (whenspec) {
1956         case WHENSPEC_RAW:
1957                 if (validate_raw_date(ltgt, &ident) < 0)
1958                         die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1959                 break;
1960         case WHENSPEC_RFC2822:
1961                 if (parse_date(ltgt, &ident) < 0)
1962                         die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
1963                 break;
1964         case WHENSPEC_NOW:
1965                 if (strcmp("now", ltgt))
1966                         die("Date in ident must be 'now': %s", buf);
1967                 datestamp(&ident);
1968                 break;
1969         }
1970
1971         return strbuf_detach(&ident, NULL);
1972 }
1973
1974 static void parse_and_store_blob(
1975         struct last_object *last,
1976         struct object_id *oidout,
1977         uintmax_t mark)
1978 {
1979         static struct strbuf buf = STRBUF_INIT;
1980         uintmax_t len;
1981
1982         if (parse_data(&buf, big_file_threshold, &len))
1983                 store_object(OBJ_BLOB, &buf, last, oidout, mark);
1984         else {
1985                 if (last) {
1986                         strbuf_release(&last->data);
1987                         last->offset = 0;
1988                         last->depth = 0;
1989                 }
1990                 stream_blob(len, oidout, mark);
1991                 skip_optional_lf();
1992         }
1993 }
1994
1995 static void parse_new_blob(void)
1996 {
1997         read_next_command();
1998         parse_mark();
1999         parse_original_identifier();
2000         parse_and_store_blob(&last_blob, NULL, next_mark);
2001 }
2002
2003 static void unload_one_branch(void)
2004 {
2005         while (cur_active_branches
2006                 && cur_active_branches >= max_active_branches) {
2007                 uintmax_t min_commit = ULONG_MAX;
2008                 struct branch *e, *l = NULL, *p = NULL;
2009
2010                 for (e = active_branches; e; e = e->active_next_branch) {
2011                         if (e->last_commit < min_commit) {
2012                                 p = l;
2013                                 min_commit = e->last_commit;
2014                         }
2015                         l = e;
2016                 }
2017
2018                 if (p) {
2019                         e = p->active_next_branch;
2020                         p->active_next_branch = e->active_next_branch;
2021                 } else {
2022                         e = active_branches;
2023                         active_branches = e->active_next_branch;
2024                 }
2025                 e->active = 0;
2026                 e->active_next_branch = NULL;
2027                 if (e->branch_tree.tree) {
2028                         release_tree_content_recursive(e->branch_tree.tree);
2029                         e->branch_tree.tree = NULL;
2030                 }
2031                 cur_active_branches--;
2032         }
2033 }
2034
2035 static void load_branch(struct branch *b)
2036 {
2037         load_tree(&b->branch_tree);
2038         if (!b->active) {
2039                 b->active = 1;
2040                 b->active_next_branch = active_branches;
2041                 active_branches = b;
2042                 cur_active_branches++;
2043                 branch_load_count++;
2044         }
2045 }
2046
2047 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2048 {
2049         unsigned char fanout = 0;
2050         while ((num_notes >>= 8))
2051                 fanout++;
2052         return fanout;
2053 }
2054
2055 static void construct_path_with_fanout(const char *hex_sha1,
2056                 unsigned char fanout, char *path)
2057 {
2058         unsigned int i = 0, j = 0;
2059         if (fanout >= the_hash_algo->rawsz)
2060                 die("Too large fanout (%u)", fanout);
2061         while (fanout) {
2062                 path[i++] = hex_sha1[j++];
2063                 path[i++] = hex_sha1[j++];
2064                 path[i++] = '/';
2065                 fanout--;
2066         }
2067         memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2068         path[i + the_hash_algo->hexsz - j] = '\0';
2069 }
2070
2071 static uintmax_t do_change_note_fanout(
2072                 struct tree_entry *orig_root, struct tree_entry *root,
2073                 char *hex_oid, unsigned int hex_oid_len,
2074                 char *fullpath, unsigned int fullpath_len,
2075                 unsigned char fanout)
2076 {
2077         struct tree_content *t;
2078         struct tree_entry *e, leaf;
2079         unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2080         uintmax_t num_notes = 0;
2081         struct object_id oid;
2082         /* hex oid + '/' between each pair of hex digits + NUL */
2083         char realpath[GIT_MAX_HEXSZ + ((GIT_MAX_HEXSZ / 2) - 1) + 1];
2084         const unsigned hexsz = the_hash_algo->hexsz;
2085
2086         if (!root->tree)
2087                 load_tree(root);
2088         t = root->tree;
2089
2090         for (i = 0; t && i < t->entry_count; i++) {
2091                 e = t->entries[i];
2092                 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2093                 tmp_fullpath_len = fullpath_len;
2094
2095                 /*
2096                  * We're interested in EITHER existing note entries (entries
2097                  * with exactly 40 hex chars in path, not including directory
2098                  * separators), OR directory entries that may contain note
2099                  * entries (with < 40 hex chars in path).
2100                  * Also, each path component in a note entry must be a multiple
2101                  * of 2 chars.
2102                  */
2103                 if (!e->versions[1].mode ||
2104                     tmp_hex_oid_len > hexsz ||
2105                     e->name->str_len % 2)
2106                         continue;
2107
2108                 /* This _may_ be a note entry, or a subdir containing notes */
2109                 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2110                        e->name->str_len);
2111                 if (tmp_fullpath_len)
2112                         fullpath[tmp_fullpath_len++] = '/';
2113                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2114                        e->name->str_len);
2115                 tmp_fullpath_len += e->name->str_len;
2116                 fullpath[tmp_fullpath_len] = '\0';
2117
2118                 if (tmp_hex_oid_len == hexsz && !get_oid_hex(hex_oid, &oid)) {
2119                         /* This is a note entry */
2120                         if (fanout == 0xff) {
2121                                 /* Counting mode, no rename */
2122                                 num_notes++;
2123                                 continue;
2124                         }
2125                         construct_path_with_fanout(hex_oid, fanout, realpath);
2126                         if (!strcmp(fullpath, realpath)) {
2127                                 /* Note entry is in correct location */
2128                                 num_notes++;
2129                                 continue;
2130                         }
2131
2132                         /* Rename fullpath to realpath */
2133                         if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2134                                 die("Failed to remove path %s", fullpath);
2135                         tree_content_set(orig_root, realpath,
2136                                 &leaf.versions[1].oid,
2137                                 leaf.versions[1].mode,
2138                                 leaf.tree);
2139                 } else if (S_ISDIR(e->versions[1].mode)) {
2140                         /* This is a subdir that may contain note entries */
2141                         num_notes += do_change_note_fanout(orig_root, e,
2142                                 hex_oid, tmp_hex_oid_len,
2143                                 fullpath, tmp_fullpath_len, fanout);
2144                 }
2145
2146                 /* The above may have reallocated the current tree_content */
2147                 t = root->tree;
2148         }
2149         return num_notes;
2150 }
2151
2152 static uintmax_t change_note_fanout(struct tree_entry *root,
2153                 unsigned char fanout)
2154 {
2155         /*
2156          * The size of path is due to one slash between every two hex digits,
2157          * plus the terminating NUL.  Note that there is no slash at the end, so
2158          * the number of slashes is one less than half the number of hex
2159          * characters.
2160          */
2161         char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2162         return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2163 }
2164
2165 static int parse_mapped_oid_hex(const char *hex, struct object_id *oid, const char **end)
2166 {
2167         int algo;
2168         khiter_t it;
2169
2170         /* Make SHA-1 object IDs have all-zero padding. */
2171         memset(oid->hash, 0, sizeof(oid->hash));
2172
2173         algo = parse_oid_hex_any(hex, oid, end);
2174         if (algo == GIT_HASH_UNKNOWN)
2175                 return -1;
2176
2177         it = kh_get_oid_map(sub_oid_map, *oid);
2178         /* No such object? */
2179         if (it == kh_end(sub_oid_map)) {
2180                 /* If we're using the same algorithm, pass it through. */
2181                 if (hash_algos[algo].format_id == the_hash_algo->format_id)
2182                         return 0;
2183                 return -1;
2184         }
2185         oidcpy(oid, kh_value(sub_oid_map, it));
2186         return 0;
2187 }
2188
2189 /*
2190  * Given a pointer into a string, parse a mark reference:
2191  *
2192  *   idnum ::= ':' bigint;
2193  *
2194  * Return the first character after the value in *endptr.
2195  *
2196  * Complain if the following character is not what is expected,
2197  * either a space or end of the string.
2198  */
2199 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2200 {
2201         uintmax_t mark;
2202
2203         assert(*p == ':');
2204         p++;
2205         mark = strtoumax(p, endptr, 10);
2206         if (*endptr == p)
2207                 die("No value after ':' in mark: %s", command_buf.buf);
2208         return mark;
2209 }
2210
2211 /*
2212  * Parse the mark reference, and complain if this is not the end of
2213  * the string.
2214  */
2215 static uintmax_t parse_mark_ref_eol(const char *p)
2216 {
2217         char *end;
2218         uintmax_t mark;
2219
2220         mark = parse_mark_ref(p, &end);
2221         if (*end != '\0')
2222                 die("Garbage after mark: %s", command_buf.buf);
2223         return mark;
2224 }
2225
2226 /*
2227  * Parse the mark reference, demanding a trailing space.  Return a
2228  * pointer to the space.
2229  */
2230 static uintmax_t parse_mark_ref_space(const char **p)
2231 {
2232         uintmax_t mark;
2233         char *end;
2234
2235         mark = parse_mark_ref(*p, &end);
2236         if (*end++ != ' ')
2237                 die("Missing space after mark: %s", command_buf.buf);
2238         *p = end;
2239         return mark;
2240 }
2241
2242 static void file_change_m(const char *p, struct branch *b)
2243 {
2244         static struct strbuf uq = STRBUF_INIT;
2245         const char *endp;
2246         struct object_entry *oe;
2247         struct object_id oid;
2248         uint16_t mode, inline_data = 0;
2249
2250         p = get_mode(p, &mode);
2251         if (!p)
2252                 die("Corrupt mode: %s", command_buf.buf);
2253         switch (mode) {
2254         case 0644:
2255         case 0755:
2256                 mode |= S_IFREG;
2257         case S_IFREG | 0644:
2258         case S_IFREG | 0755:
2259         case S_IFLNK:
2260         case S_IFDIR:
2261         case S_IFGITLINK:
2262                 /* ok */
2263                 break;
2264         default:
2265                 die("Corrupt mode: %s", command_buf.buf);
2266         }
2267
2268         if (*p == ':') {
2269                 oe = find_mark(marks, parse_mark_ref_space(&p));
2270                 oidcpy(&oid, &oe->idx.oid);
2271         } else if (skip_prefix(p, "inline ", &p)) {
2272                 inline_data = 1;
2273                 oe = NULL; /* not used with inline_data, but makes gcc happy */
2274         } else {
2275                 if (parse_mapped_oid_hex(p, &oid, &p))
2276                         die("Invalid dataref: %s", command_buf.buf);
2277                 oe = find_object(&oid);
2278                 if (*p++ != ' ')
2279                         die("Missing space after SHA1: %s", command_buf.buf);
2280         }
2281
2282         strbuf_reset(&uq);
2283         if (!unquote_c_style(&uq, p, &endp)) {
2284                 if (*endp)
2285                         die("Garbage after path in: %s", command_buf.buf);
2286                 p = uq.buf;
2287         }
2288
2289         /* Git does not track empty, non-toplevel directories. */
2290         if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2291                 tree_content_remove(&b->branch_tree, p, NULL, 0);
2292                 return;
2293         }
2294
2295         if (S_ISGITLINK(mode)) {
2296                 if (inline_data)
2297                         die("Git links cannot be specified 'inline': %s",
2298                                 command_buf.buf);
2299                 else if (oe) {
2300                         if (oe->type != OBJ_COMMIT)
2301                                 die("Not a commit (actually a %s): %s",
2302                                         type_name(oe->type), command_buf.buf);
2303                 }
2304                 /*
2305                  * Accept the sha1 without checking; it expected to be in
2306                  * another repository.
2307                  */
2308         } else if (inline_data) {
2309                 if (S_ISDIR(mode))
2310                         die("Directories cannot be specified 'inline': %s",
2311                                 command_buf.buf);
2312                 if (p != uq.buf) {
2313                         strbuf_addstr(&uq, p);
2314                         p = uq.buf;
2315                 }
2316                 while (read_next_command() != EOF) {
2317                         const char *v;
2318                         if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2319                                 parse_cat_blob(v);
2320                         else {
2321                                 parse_and_store_blob(&last_blob, &oid, 0);
2322                                 break;
2323                         }
2324                 }
2325         } else {
2326                 enum object_type expected = S_ISDIR(mode) ?
2327                                                 OBJ_TREE: OBJ_BLOB;
2328                 enum object_type type = oe ? oe->type :
2329                                         oid_object_info(the_repository, &oid,
2330                                                         NULL);
2331                 if (type < 0)
2332                         die("%s not found: %s",
2333                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2334                                         command_buf.buf);
2335                 if (type != expected)
2336                         die("Not a %s (actually a %s): %s",
2337                                 type_name(expected), type_name(type),
2338                                 command_buf.buf);
2339         }
2340
2341         if (!*p) {
2342                 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2343                 return;
2344         }
2345         tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2346 }
2347
2348 static void file_change_d(const char *p, struct branch *b)
2349 {
2350         static struct strbuf uq = STRBUF_INIT;
2351         const char *endp;
2352
2353         strbuf_reset(&uq);
2354         if (!unquote_c_style(&uq, p, &endp)) {
2355                 if (*endp)
2356                         die("Garbage after path in: %s", command_buf.buf);
2357                 p = uq.buf;
2358         }
2359         tree_content_remove(&b->branch_tree, p, NULL, 1);
2360 }
2361
2362 static void file_change_cr(const char *s, struct branch *b, int rename)
2363 {
2364         const char *d;
2365         static struct strbuf s_uq = STRBUF_INIT;
2366         static struct strbuf d_uq = STRBUF_INIT;
2367         const char *endp;
2368         struct tree_entry leaf;
2369
2370         strbuf_reset(&s_uq);
2371         if (!unquote_c_style(&s_uq, s, &endp)) {
2372                 if (*endp != ' ')
2373                         die("Missing space after source: %s", command_buf.buf);
2374         } else {
2375                 endp = strchr(s, ' ');
2376                 if (!endp)
2377                         die("Missing space after source: %s", command_buf.buf);
2378                 strbuf_add(&s_uq, s, endp - s);
2379         }
2380         s = s_uq.buf;
2381
2382         endp++;
2383         if (!*endp)
2384                 die("Missing dest: %s", command_buf.buf);
2385
2386         d = endp;
2387         strbuf_reset(&d_uq);
2388         if (!unquote_c_style(&d_uq, d, &endp)) {
2389                 if (*endp)
2390                         die("Garbage after dest in: %s", command_buf.buf);
2391                 d = d_uq.buf;
2392         }
2393
2394         memset(&leaf, 0, sizeof(leaf));
2395         if (rename)
2396                 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2397         else
2398                 tree_content_get(&b->branch_tree, s, &leaf, 1);
2399         if (!leaf.versions[1].mode)
2400                 die("Path %s not in branch", s);
2401         if (!*d) {      /* C "path/to/subdir" "" */
2402                 tree_content_replace(&b->branch_tree,
2403                         &leaf.versions[1].oid,
2404                         leaf.versions[1].mode,
2405                         leaf.tree);
2406                 return;
2407         }
2408         tree_content_set(&b->branch_tree, d,
2409                 &leaf.versions[1].oid,
2410                 leaf.versions[1].mode,
2411                 leaf.tree);
2412 }
2413
2414 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2415 {
2416         static struct strbuf uq = STRBUF_INIT;
2417         struct object_entry *oe;
2418         struct branch *s;
2419         struct object_id oid, commit_oid;
2420         char path[GIT_MAX_RAWSZ * 3];
2421         uint16_t inline_data = 0;
2422         unsigned char new_fanout;
2423
2424         /*
2425          * When loading a branch, we don't traverse its tree to count the real
2426          * number of notes (too expensive to do this for all non-note refs).
2427          * This means that recently loaded notes refs might incorrectly have
2428          * b->num_notes == 0, and consequently, old_fanout might be wrong.
2429          *
2430          * Fix this by traversing the tree and counting the number of notes
2431          * when b->num_notes == 0. If the notes tree is truly empty, the
2432          * calculation should not take long.
2433          */
2434         if (b->num_notes == 0 && *old_fanout == 0) {
2435                 /* Invoke change_note_fanout() in "counting mode". */
2436                 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2437                 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2438         }
2439
2440         /* Now parse the notemodify command. */
2441         /* <dataref> or 'inline' */
2442         if (*p == ':') {
2443                 oe = find_mark(marks, parse_mark_ref_space(&p));
2444                 oidcpy(&oid, &oe->idx.oid);
2445         } else if (skip_prefix(p, "inline ", &p)) {
2446                 inline_data = 1;
2447                 oe = NULL; /* not used with inline_data, but makes gcc happy */
2448         } else {
2449                 if (parse_mapped_oid_hex(p, &oid, &p))
2450                         die("Invalid dataref: %s", command_buf.buf);
2451                 oe = find_object(&oid);
2452                 if (*p++ != ' ')
2453                         die("Missing space after SHA1: %s", command_buf.buf);
2454         }
2455
2456         /* <commit-ish> */
2457         s = lookup_branch(p);
2458         if (s) {
2459                 if (is_null_oid(&s->oid))
2460                         die("Can't add a note on empty branch.");
2461                 oidcpy(&commit_oid, &s->oid);
2462         } else if (*p == ':') {
2463                 uintmax_t commit_mark = parse_mark_ref_eol(p);
2464                 struct object_entry *commit_oe = find_mark(marks, commit_mark);
2465                 if (commit_oe->type != OBJ_COMMIT)
2466                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2467                 oidcpy(&commit_oid, &commit_oe->idx.oid);
2468         } else if (!get_oid(p, &commit_oid)) {
2469                 unsigned long size;
2470                 char *buf = read_object_with_reference(the_repository,
2471                                                        &commit_oid,
2472                                                        commit_type, &size,
2473                                                        &commit_oid);
2474                 if (!buf || size < the_hash_algo->hexsz + 6)
2475                         die("Not a valid commit: %s", p);
2476                 free(buf);
2477         } else
2478                 die("Invalid ref name or SHA1 expression: %s", p);
2479
2480         if (inline_data) {
2481                 if (p != uq.buf) {
2482                         strbuf_addstr(&uq, p);
2483                         p = uq.buf;
2484                 }
2485                 read_next_command();
2486                 parse_and_store_blob(&last_blob, &oid, 0);
2487         } else if (oe) {
2488                 if (oe->type != OBJ_BLOB)
2489                         die("Not a blob (actually a %s): %s",
2490                                 type_name(oe->type), command_buf.buf);
2491         } else if (!is_null_oid(&oid)) {
2492                 enum object_type type = oid_object_info(the_repository, &oid,
2493                                                         NULL);
2494                 if (type < 0)
2495                         die("Blob not found: %s", command_buf.buf);
2496                 if (type != OBJ_BLOB)
2497                         die("Not a blob (actually a %s): %s",
2498                             type_name(type), command_buf.buf);
2499         }
2500
2501         construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2502         if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2503                 b->num_notes--;
2504
2505         if (is_null_oid(&oid))
2506                 return; /* nothing to insert */
2507
2508         b->num_notes++;
2509         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2510         construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2511         tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2512 }
2513
2514 static void file_change_deleteall(struct branch *b)
2515 {
2516         release_tree_content_recursive(b->branch_tree.tree);
2517         oidclr(&b->branch_tree.versions[0].oid);
2518         oidclr(&b->branch_tree.versions[1].oid);
2519         load_tree(&b->branch_tree);
2520         b->num_notes = 0;
2521 }
2522
2523 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2524 {
2525         if (!buf || size < the_hash_algo->hexsz + 6)
2526                 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2527         if (memcmp("tree ", buf, 5)
2528                 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2529                 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2530         oidcpy(&b->branch_tree.versions[0].oid,
2531                &b->branch_tree.versions[1].oid);
2532 }
2533
2534 static void parse_from_existing(struct branch *b)
2535 {
2536         if (is_null_oid(&b->oid)) {
2537                 oidclr(&b->branch_tree.versions[0].oid);
2538                 oidclr(&b->branch_tree.versions[1].oid);
2539         } else {
2540                 unsigned long size;
2541                 char *buf;
2542
2543                 buf = read_object_with_reference(the_repository,
2544                                                  &b->oid, commit_type, &size,
2545                                                  &b->oid);
2546                 parse_from_commit(b, buf, size);
2547                 free(buf);
2548         }
2549 }
2550
2551 static int parse_objectish(struct branch *b, const char *objectish)
2552 {
2553         struct branch *s;
2554         struct object_id oid;
2555
2556         oidcpy(&oid, &b->branch_tree.versions[1].oid);
2557
2558         s = lookup_branch(objectish);
2559         if (b == s)
2560                 die("Can't create a branch from itself: %s", b->name);
2561         else if (s) {
2562                 struct object_id *t = &s->branch_tree.versions[1].oid;
2563                 oidcpy(&b->oid, &s->oid);
2564                 oidcpy(&b->branch_tree.versions[0].oid, t);
2565                 oidcpy(&b->branch_tree.versions[1].oid, t);
2566         } else if (*objectish == ':') {
2567                 uintmax_t idnum = parse_mark_ref_eol(objectish);
2568                 struct object_entry *oe = find_mark(marks, idnum);
2569                 if (oe->type != OBJ_COMMIT)
2570                         die("Mark :%" PRIuMAX " not a commit", idnum);
2571                 if (!oideq(&b->oid, &oe->idx.oid)) {
2572                         oidcpy(&b->oid, &oe->idx.oid);
2573                         if (oe->pack_id != MAX_PACK_ID) {
2574                                 unsigned long size;
2575                                 char *buf = gfi_unpack_entry(oe, &size);
2576                                 parse_from_commit(b, buf, size);
2577                                 free(buf);
2578                         } else
2579                                 parse_from_existing(b);
2580                 }
2581         } else if (!get_oid(objectish, &b->oid)) {
2582                 parse_from_existing(b);
2583                 if (is_null_oid(&b->oid))
2584                         b->delete = 1;
2585         }
2586         else
2587                 die("Invalid ref name or SHA1 expression: %s", objectish);
2588
2589         if (b->branch_tree.tree && !oideq(&oid, &b->branch_tree.versions[1].oid)) {
2590                 release_tree_content_recursive(b->branch_tree.tree);
2591                 b->branch_tree.tree = NULL;
2592         }
2593
2594         read_next_command();
2595         return 1;
2596 }
2597
2598 static int parse_from(struct branch *b)
2599 {
2600         const char *from;
2601
2602         if (!skip_prefix(command_buf.buf, "from ", &from))
2603                 return 0;
2604
2605         return parse_objectish(b, from);
2606 }
2607
2608 static int parse_objectish_with_prefix(struct branch *b, const char *prefix)
2609 {
2610         const char *base;
2611
2612         if (!skip_prefix(command_buf.buf, prefix, &base))
2613                 return 0;
2614
2615         return parse_objectish(b, base);
2616 }
2617
2618 static struct hash_list *parse_merge(unsigned int *count)
2619 {
2620         struct hash_list *list = NULL, **tail = &list, *n;
2621         const char *from;
2622         struct branch *s;
2623
2624         *count = 0;
2625         while (skip_prefix(command_buf.buf, "merge ", &from)) {
2626                 n = xmalloc(sizeof(*n));
2627                 s = lookup_branch(from);
2628                 if (s)
2629                         oidcpy(&n->oid, &s->oid);
2630                 else if (*from == ':') {
2631                         uintmax_t idnum = parse_mark_ref_eol(from);
2632                         struct object_entry *oe = find_mark(marks, idnum);
2633                         if (oe->type != OBJ_COMMIT)
2634                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2635                         oidcpy(&n->oid, &oe->idx.oid);
2636                 } else if (!get_oid(from, &n->oid)) {
2637                         unsigned long size;
2638                         char *buf = read_object_with_reference(the_repository,
2639                                                                &n->oid,
2640                                                                commit_type,
2641                                                                &size, &n->oid);
2642                         if (!buf || size < the_hash_algo->hexsz + 6)
2643                                 die("Not a valid commit: %s", from);
2644                         free(buf);
2645                 } else
2646                         die("Invalid ref name or SHA1 expression: %s", from);
2647
2648                 n->next = NULL;
2649                 *tail = n;
2650                 tail = &n->next;
2651
2652                 (*count)++;
2653                 read_next_command();
2654         }
2655         return list;
2656 }
2657
2658 static void parse_new_commit(const char *arg)
2659 {
2660         static struct strbuf msg = STRBUF_INIT;
2661         struct branch *b;
2662         char *author = NULL;
2663         char *committer = NULL;
2664         char *encoding = NULL;
2665         struct hash_list *merge_list = NULL;
2666         unsigned int merge_count;
2667         unsigned char prev_fanout, new_fanout;
2668         const char *v;
2669
2670         b = lookup_branch(arg);
2671         if (!b)
2672                 b = new_branch(arg);
2673
2674         read_next_command();
2675         parse_mark();
2676         parse_original_identifier();
2677         if (skip_prefix(command_buf.buf, "author ", &v)) {
2678                 author = parse_ident(v);
2679                 read_next_command();
2680         }
2681         if (skip_prefix(command_buf.buf, "committer ", &v)) {
2682                 committer = parse_ident(v);
2683                 read_next_command();
2684         }
2685         if (!committer)
2686                 die("Expected committer but didn't get one");
2687         if (skip_prefix(command_buf.buf, "encoding ", &v)) {
2688                 encoding = xstrdup(v);
2689                 read_next_command();
2690         }
2691         parse_data(&msg, 0, NULL);
2692         read_next_command();
2693         parse_from(b);
2694         merge_list = parse_merge(&merge_count);
2695
2696         /* ensure the branch is active/loaded */
2697         if (!b->branch_tree.tree || !max_active_branches) {
2698                 unload_one_branch();
2699                 load_branch(b);
2700         }
2701
2702         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2703
2704         /* file_change* */
2705         while (command_buf.len > 0) {
2706                 if (skip_prefix(command_buf.buf, "M ", &v))
2707                         file_change_m(v, b);
2708                 else if (skip_prefix(command_buf.buf, "D ", &v))
2709                         file_change_d(v, b);
2710                 else if (skip_prefix(command_buf.buf, "R ", &v))
2711                         file_change_cr(v, b, 1);
2712                 else if (skip_prefix(command_buf.buf, "C ", &v))
2713                         file_change_cr(v, b, 0);
2714                 else if (skip_prefix(command_buf.buf, "N ", &v))
2715                         note_change_n(v, b, &prev_fanout);
2716                 else if (!strcmp("deleteall", command_buf.buf))
2717                         file_change_deleteall(b);
2718                 else if (skip_prefix(command_buf.buf, "ls ", &v))
2719                         parse_ls(v, b);
2720                 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2721                         parse_cat_blob(v);
2722                 else {
2723                         unread_command_buf = 1;
2724                         break;
2725                 }
2726                 if (read_next_command() == EOF)
2727                         break;
2728         }
2729
2730         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2731         if (new_fanout != prev_fanout)
2732                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2733
2734         /* build the tree and the commit */
2735         store_tree(&b->branch_tree);
2736         oidcpy(&b->branch_tree.versions[0].oid,
2737                &b->branch_tree.versions[1].oid);
2738
2739         strbuf_reset(&new_data);
2740         strbuf_addf(&new_data, "tree %s\n",
2741                 oid_to_hex(&b->branch_tree.versions[1].oid));
2742         if (!is_null_oid(&b->oid))
2743                 strbuf_addf(&new_data, "parent %s\n",
2744                             oid_to_hex(&b->oid));
2745         while (merge_list) {
2746                 struct hash_list *next = merge_list->next;
2747                 strbuf_addf(&new_data, "parent %s\n",
2748                             oid_to_hex(&merge_list->oid));
2749                 free(merge_list);
2750                 merge_list = next;
2751         }
2752         strbuf_addf(&new_data,
2753                 "author %s\n"
2754                 "committer %s\n",
2755                 author ? author : committer, committer);
2756         if (encoding)
2757                 strbuf_addf(&new_data,
2758                         "encoding %s\n",
2759                         encoding);
2760         strbuf_addch(&new_data, '\n');
2761         strbuf_addbuf(&new_data, &msg);
2762         free(author);
2763         free(committer);
2764         free(encoding);
2765
2766         if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2767                 b->pack_id = pack_id;
2768         b->last_commit = object_count_by_type[OBJ_COMMIT];
2769 }
2770
2771 static void parse_new_tag(const char *arg)
2772 {
2773         static struct strbuf msg = STRBUF_INIT;
2774         const char *from;
2775         char *tagger;
2776         struct branch *s;
2777         struct tag *t;
2778         uintmax_t from_mark = 0;
2779         struct object_id oid;
2780         enum object_type type;
2781         const char *v;
2782
2783         t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2784         memset(t, 0, sizeof(struct tag));
2785         t->name = pool_strdup(arg);
2786         if (last_tag)
2787                 last_tag->next_tag = t;
2788         else
2789                 first_tag = t;
2790         last_tag = t;
2791         read_next_command();
2792         parse_mark();
2793
2794         /* from ... */
2795         if (!skip_prefix(command_buf.buf, "from ", &from))
2796                 die("Expected from command, got %s", command_buf.buf);
2797         s = lookup_branch(from);
2798         if (s) {
2799                 if (is_null_oid(&s->oid))
2800                         die("Can't tag an empty branch.");
2801                 oidcpy(&oid, &s->oid);
2802                 type = OBJ_COMMIT;
2803         } else if (*from == ':') {
2804                 struct object_entry *oe;
2805                 from_mark = parse_mark_ref_eol(from);
2806                 oe = find_mark(marks, from_mark);
2807                 type = oe->type;
2808                 oidcpy(&oid, &oe->idx.oid);
2809         } else if (!get_oid(from, &oid)) {
2810                 struct object_entry *oe = find_object(&oid);
2811                 if (!oe) {
2812                         type = oid_object_info(the_repository, &oid, NULL);
2813                         if (type < 0)
2814                                 die("Not a valid object: %s", from);
2815                 } else
2816                         type = oe->type;
2817         } else
2818                 die("Invalid ref name or SHA1 expression: %s", from);
2819         read_next_command();
2820
2821         /* original-oid ... */
2822         parse_original_identifier();
2823
2824         /* tagger ... */
2825         if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2826                 tagger = parse_ident(v);
2827                 read_next_command();
2828         } else
2829                 tagger = NULL;
2830
2831         /* tag payload/message */
2832         parse_data(&msg, 0, NULL);
2833
2834         /* build the tag object */
2835         strbuf_reset(&new_data);
2836
2837         strbuf_addf(&new_data,
2838                     "object %s\n"
2839                     "type %s\n"
2840                     "tag %s\n",
2841                     oid_to_hex(&oid), type_name(type), t->name);
2842         if (tagger)
2843                 strbuf_addf(&new_data,
2844                             "tagger %s\n", tagger);
2845         strbuf_addch(&new_data, '\n');
2846         strbuf_addbuf(&new_data, &msg);
2847         free(tagger);
2848
2849         if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, next_mark))
2850                 t->pack_id = MAX_PACK_ID;
2851         else
2852                 t->pack_id = pack_id;
2853 }
2854
2855 static void parse_reset_branch(const char *arg)
2856 {
2857         struct branch *b;
2858         const char *tag_name;
2859
2860         b = lookup_branch(arg);
2861         if (b) {
2862                 oidclr(&b->oid);
2863                 oidclr(&b->branch_tree.versions[0].oid);
2864                 oidclr(&b->branch_tree.versions[1].oid);
2865                 if (b->branch_tree.tree) {
2866                         release_tree_content_recursive(b->branch_tree.tree);
2867                         b->branch_tree.tree = NULL;
2868                 }
2869         }
2870         else
2871                 b = new_branch(arg);
2872         read_next_command();
2873         parse_from(b);
2874         if (b->delete && skip_prefix(b->name, "refs/tags/", &tag_name)) {
2875                 /*
2876                  * Elsewhere, we call dump_branches() before dump_tags(),
2877                  * and dump_branches() will handle ref deletions first, so
2878                  * in order to make sure the deletion actually takes effect,
2879                  * we need to remove the tag from our list of tags to update.
2880                  *
2881                  * NEEDSWORK: replace list of tags with hashmap for faster
2882                  * deletion?
2883                  */
2884                 struct tag *t, *prev = NULL;
2885                 for (t = first_tag; t; t = t->next_tag) {
2886                         if (!strcmp(t->name, tag_name))
2887                                 break;
2888                         prev = t;
2889                 }
2890                 if (t) {
2891                         if (prev)
2892                                 prev->next_tag = t->next_tag;
2893                         else
2894                                 first_tag = t->next_tag;
2895                         if (!t->next_tag)
2896                                 last_tag = prev;
2897                         /* There is no mem_pool_free(t) function to call. */
2898                 }
2899         }
2900         if (command_buf.len > 0)
2901                 unread_command_buf = 1;
2902 }
2903
2904 static void cat_blob_write(const char *buf, unsigned long size)
2905 {
2906         if (write_in_full(cat_blob_fd, buf, size) < 0)
2907                 die_errno("Write to frontend failed");
2908 }
2909
2910 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2911 {
2912         struct strbuf line = STRBUF_INIT;
2913         unsigned long size;
2914         enum object_type type = 0;
2915         char *buf;
2916
2917         if (!oe || oe->pack_id == MAX_PACK_ID) {
2918                 buf = read_object_file(oid, &type, &size);
2919         } else {
2920                 type = oe->type;
2921                 buf = gfi_unpack_entry(oe, &size);
2922         }
2923
2924         /*
2925          * Output based on batch_one_object() from cat-file.c.
2926          */
2927         if (type <= 0) {
2928                 strbuf_reset(&line);
2929                 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2930                 cat_blob_write(line.buf, line.len);
2931                 strbuf_release(&line);
2932                 free(buf);
2933                 return;
2934         }
2935         if (!buf)
2936                 die("Can't read object %s", oid_to_hex(oid));
2937         if (type != OBJ_BLOB)
2938                 die("Object %s is a %s but a blob was expected.",
2939                     oid_to_hex(oid), type_name(type));
2940         strbuf_reset(&line);
2941         strbuf_addf(&line, "%s %s %"PRIuMAX"\n", oid_to_hex(oid),
2942                     type_name(type), (uintmax_t)size);
2943         cat_blob_write(line.buf, line.len);
2944         strbuf_release(&line);
2945         cat_blob_write(buf, size);
2946         cat_blob_write("\n", 1);
2947         if (oe && oe->pack_id == pack_id) {
2948                 last_blob.offset = oe->idx.offset;
2949                 strbuf_attach(&last_blob.data, buf, size, size);
2950                 last_blob.depth = oe->depth;
2951         } else
2952                 free(buf);
2953 }
2954
2955 static void parse_get_mark(const char *p)
2956 {
2957         struct object_entry *oe;
2958         char output[GIT_MAX_HEXSZ + 2];
2959
2960         /* get-mark SP <object> LF */
2961         if (*p != ':')
2962                 die("Not a mark: %s", p);
2963
2964         oe = find_mark(marks, parse_mark_ref_eol(p));
2965         if (!oe)
2966                 die("Unknown mark: %s", command_buf.buf);
2967
2968         xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2969         cat_blob_write(output, the_hash_algo->hexsz + 1);
2970 }
2971
2972 static void parse_cat_blob(const char *p)
2973 {
2974         struct object_entry *oe;
2975         struct object_id oid;
2976
2977         /* cat-blob SP <object> LF */
2978         if (*p == ':') {
2979                 oe = find_mark(marks, parse_mark_ref_eol(p));
2980                 if (!oe)
2981                         die("Unknown mark: %s", command_buf.buf);
2982                 oidcpy(&oid, &oe->idx.oid);
2983         } else {
2984                 if (parse_mapped_oid_hex(p, &oid, &p))
2985                         die("Invalid dataref: %s", command_buf.buf);
2986                 if (*p)
2987                         die("Garbage after SHA1: %s", command_buf.buf);
2988                 oe = find_object(&oid);
2989         }
2990
2991         cat_blob(oe, &oid);
2992 }
2993
2994 static struct object_entry *dereference(struct object_entry *oe,
2995                                         struct object_id *oid)
2996 {
2997         unsigned long size;
2998         char *buf = NULL;
2999         const unsigned hexsz = the_hash_algo->hexsz;
3000
3001         if (!oe) {
3002                 enum object_type type = oid_object_info(the_repository, oid,
3003                                                         NULL);
3004                 if (type < 0)
3005                         die("object not found: %s", oid_to_hex(oid));
3006                 /* cache it! */
3007                 oe = insert_object(oid);
3008                 oe->type = type;
3009                 oe->pack_id = MAX_PACK_ID;
3010                 oe->idx.offset = 1;
3011         }
3012         switch (oe->type) {
3013         case OBJ_TREE:  /* easy case. */
3014                 return oe;
3015         case OBJ_COMMIT:
3016         case OBJ_TAG:
3017                 break;
3018         default:
3019                 die("Not a tree-ish: %s", command_buf.buf);
3020         }
3021
3022         if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
3023                 buf = gfi_unpack_entry(oe, &size);
3024         } else {
3025                 enum object_type unused;
3026                 buf = read_object_file(oid, &unused, &size);
3027         }
3028         if (!buf)
3029                 die("Can't load object %s", oid_to_hex(oid));
3030
3031         /* Peel one layer. */
3032         switch (oe->type) {
3033         case OBJ_TAG:
3034                 if (size < hexsz + strlen("object ") ||
3035                     get_oid_hex(buf + strlen("object "), oid))
3036                         die("Invalid SHA1 in tag: %s", command_buf.buf);
3037                 break;
3038         case OBJ_COMMIT:
3039                 if (size < hexsz + strlen("tree ") ||
3040                     get_oid_hex(buf + strlen("tree "), oid))
3041                         die("Invalid SHA1 in commit: %s", command_buf.buf);
3042         }
3043
3044         free(buf);
3045         return find_object(oid);
3046 }
3047
3048 static void insert_mapped_mark(uintmax_t mark, void *object, void *cbp)
3049 {
3050         struct object_id *fromoid = object;
3051         struct object_id *tooid = find_mark(cbp, mark);
3052         int ret;
3053         khiter_t it;
3054
3055         it = kh_put_oid_map(sub_oid_map, *fromoid, &ret);
3056         /* We've already seen this object. */
3057         if (ret == 0)
3058                 return;
3059         kh_value(sub_oid_map, it) = tooid;
3060 }
3061
3062 static void build_mark_map_one(struct mark_set *from, struct mark_set *to)
3063 {
3064         for_each_mark(from, 0, insert_mapped_mark, to);
3065 }
3066
3067 static void build_mark_map(struct string_list *from, struct string_list *to)
3068 {
3069         struct string_list_item *fromp, *top;
3070
3071         sub_oid_map = kh_init_oid_map();
3072
3073         for_each_string_list_item(fromp, from) {
3074                 top = string_list_lookup(to, fromp->string);
3075                 if (!fromp->util) {
3076                         die(_("Missing from marks for submodule '%s'"), fromp->string);
3077                 } else if (!top || !top->util) {
3078                         die(_("Missing to marks for submodule '%s'"), fromp->string);
3079                 }
3080                 build_mark_map_one(fromp->util, top->util);
3081         }
3082 }
3083
3084 static struct object_entry *parse_treeish_dataref(const char **p)
3085 {
3086         struct object_id oid;
3087         struct object_entry *e;
3088
3089         if (**p == ':') {       /* <mark> */
3090                 e = find_mark(marks, parse_mark_ref_space(p));
3091                 if (!e)
3092                         die("Unknown mark: %s", command_buf.buf);
3093                 oidcpy(&oid, &e->idx.oid);
3094         } else {        /* <sha1> */
3095                 if (parse_mapped_oid_hex(*p, &oid, p))
3096                         die("Invalid dataref: %s", command_buf.buf);
3097                 e = find_object(&oid);
3098                 if (*(*p)++ != ' ')
3099                         die("Missing space after tree-ish: %s", command_buf.buf);
3100         }
3101
3102         while (!e || e->type != OBJ_TREE)
3103                 e = dereference(e, &oid);
3104         return e;
3105 }
3106
3107 static void print_ls(int mode, const unsigned char *hash, const char *path)
3108 {
3109         static struct strbuf line = STRBUF_INIT;
3110
3111         /* See show_tree(). */
3112         const char *type =
3113                 S_ISGITLINK(mode) ? commit_type :
3114                 S_ISDIR(mode) ? tree_type :
3115                 blob_type;
3116
3117         if (!mode) {
3118                 /* missing SP path LF */
3119                 strbuf_reset(&line);
3120                 strbuf_addstr(&line, "missing ");
3121                 quote_c_style(path, &line, NULL, 0);
3122                 strbuf_addch(&line, '\n');
3123         } else {
3124                 /* mode SP type SP object_name TAB path LF */
3125                 strbuf_reset(&line);
3126                 strbuf_addf(&line, "%06o %s %s\t",
3127                                 mode & ~NO_DELTA, type, hash_to_hex(hash));
3128                 quote_c_style(path, &line, NULL, 0);
3129                 strbuf_addch(&line, '\n');
3130         }
3131         cat_blob_write(line.buf, line.len);
3132 }
3133
3134 static void parse_ls(const char *p, struct branch *b)
3135 {
3136         struct tree_entry *root = NULL;
3137         struct tree_entry leaf = {NULL};
3138
3139         /* ls SP (<tree-ish> SP)? <path> */
3140         if (*p == '"') {
3141                 if (!b)
3142                         die("Not in a commit: %s", command_buf.buf);
3143                 root = &b->branch_tree;
3144         } else {
3145                 struct object_entry *e = parse_treeish_dataref(&p);
3146                 root = new_tree_entry();
3147                 oidcpy(&root->versions[1].oid, &e->idx.oid);
3148                 if (!is_null_oid(&root->versions[1].oid))
3149                         root->versions[1].mode = S_IFDIR;
3150                 load_tree(root);
3151         }
3152         if (*p == '"') {
3153                 static struct strbuf uq = STRBUF_INIT;
3154                 const char *endp;
3155                 strbuf_reset(&uq);
3156                 if (unquote_c_style(&uq, p, &endp))
3157                         die("Invalid path: %s", command_buf.buf);
3158                 if (*endp)
3159                         die("Garbage after path in: %s", command_buf.buf);
3160                 p = uq.buf;
3161         }
3162         tree_content_get(root, p, &leaf, 1);
3163         /*
3164          * A directory in preparation would have a sha1 of zero
3165          * until it is saved.  Save, for simplicity.
3166          */
3167         if (S_ISDIR(leaf.versions[1].mode))
3168                 store_tree(&leaf);
3169
3170         print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3171         if (leaf.tree)
3172                 release_tree_content_recursive(leaf.tree);
3173         if (!b || root != &b->branch_tree)
3174                 release_tree_entry(root);
3175 }
3176
3177 static void checkpoint(void)
3178 {
3179         checkpoint_requested = 0;
3180         if (object_count) {
3181                 cycle_packfile();
3182         }
3183         dump_branches();
3184         dump_tags();
3185         dump_marks();
3186 }
3187
3188 static void parse_checkpoint(void)
3189 {
3190         checkpoint_requested = 1;
3191         skip_optional_lf();
3192 }
3193
3194 static void parse_progress(void)
3195 {
3196         fwrite(command_buf.buf, 1, command_buf.len, stdout);
3197         fputc('\n', stdout);
3198         fflush(stdout);
3199         skip_optional_lf();
3200 }
3201
3202 static void parse_alias(void)
3203 {
3204         struct object_entry *e;
3205         struct branch b;
3206
3207         skip_optional_lf();
3208         read_next_command();
3209
3210         /* mark ... */
3211         parse_mark();
3212         if (!next_mark)
3213                 die(_("Expected 'mark' command, got %s"), command_buf.buf);
3214
3215         /* to ... */
3216         memset(&b, 0, sizeof(b));
3217         if (!parse_objectish_with_prefix(&b, "to "))
3218                 die(_("Expected 'to' command, got %s"), command_buf.buf);
3219         e = find_object(&b.oid);
3220         assert(e);
3221         insert_mark(marks, next_mark, e);
3222 }
3223
3224 static char* make_fast_import_path(const char *path)
3225 {
3226         if (!relative_marks_paths || is_absolute_path(path))
3227                 return xstrdup(path);
3228         return git_pathdup("info/fast-import/%s", path);
3229 }
3230
3231 static void option_import_marks(const char *marks,
3232                                         int from_stream, int ignore_missing)
3233 {
3234         if (import_marks_file) {
3235                 if (from_stream)
3236                         die("Only one import-marks command allowed per stream");
3237
3238                 /* read previous mark file */
3239                 if(!import_marks_file_from_stream)
3240                         read_marks();
3241         }
3242
3243         import_marks_file = make_fast_import_path(marks);
3244         import_marks_file_from_stream = from_stream;
3245         import_marks_file_ignore_missing = ignore_missing;
3246 }
3247
3248 static void option_date_format(const char *fmt)
3249 {
3250         if (!strcmp(fmt, "raw"))
3251                 whenspec = WHENSPEC_RAW;
3252         else if (!strcmp(fmt, "rfc2822"))
3253                 whenspec = WHENSPEC_RFC2822;
3254         else if (!strcmp(fmt, "now"))
3255                 whenspec = WHENSPEC_NOW;
3256         else
3257                 die("unknown --date-format argument %s", fmt);
3258 }
3259
3260 static unsigned long ulong_arg(const char *option, const char *arg)
3261 {
3262         char *endptr;
3263         unsigned long rv = strtoul(arg, &endptr, 0);
3264         if (strchr(arg, '-') || endptr == arg || *endptr)
3265                 die("%s: argument must be a non-negative integer", option);
3266         return rv;
3267 }
3268
3269 static void option_depth(const char *depth)
3270 {
3271         max_depth = ulong_arg("--depth", depth);
3272         if (max_depth > MAX_DEPTH)
3273                 die("--depth cannot exceed %u", MAX_DEPTH);
3274 }
3275
3276 static void option_active_branches(const char *branches)
3277 {
3278         max_active_branches = ulong_arg("--active-branches", branches);
3279 }
3280
3281 static void option_export_marks(const char *marks)
3282 {
3283         export_marks_file = make_fast_import_path(marks);
3284 }
3285
3286 static void option_cat_blob_fd(const char *fd)
3287 {
3288         unsigned long n = ulong_arg("--cat-blob-fd", fd);
3289         if (n > (unsigned long) INT_MAX)
3290                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3291         cat_blob_fd = (int) n;
3292 }
3293
3294 static void option_export_pack_edges(const char *edges)
3295 {
3296         if (pack_edges)
3297                 fclose(pack_edges);
3298         pack_edges = xfopen(edges, "a");
3299 }
3300
3301 static void option_rewrite_submodules(const char *arg, struct string_list *list)
3302 {
3303         struct mark_set *ms;
3304         FILE *fp;
3305         char *s = xstrdup(arg);
3306         char *f = strchr(s, ':');
3307         if (!f)
3308                 die(_("Expected format name:filename for submodule rewrite option"));
3309         *f = '\0';
3310         f++;
3311         ms = xcalloc(1, sizeof(*ms));
3312         string_list_insert(list, s)->util = ms;
3313
3314         fp = fopen(f, "r");
3315         if (!fp)
3316                 die_errno("cannot read '%s'", f);
3317         read_mark_file(ms, fp, insert_oid_entry);
3318         fclose(fp);
3319 }
3320
3321 static int parse_one_option(const char *option)
3322 {
3323         if (skip_prefix(option, "max-pack-size=", &option)) {
3324                 unsigned long v;
3325                 if (!git_parse_ulong(option, &v))
3326                         return 0;
3327                 if (v < 8192) {
3328                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3329                         v *= 1024 * 1024;
3330                 } else if (v < 1024 * 1024) {
3331                         warning("minimum max-pack-size is 1 MiB");
3332                         v = 1024 * 1024;
3333                 }
3334                 max_packsize = v;
3335         } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3336                 unsigned long v;
3337                 if (!git_parse_ulong(option, &v))
3338                         return 0;
3339                 big_file_threshold = v;
3340         } else if (skip_prefix(option, "depth=", &option)) {
3341                 option_depth(option);
3342         } else if (skip_prefix(option, "active-branches=", &option)) {
3343                 option_active_branches(option);
3344         } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3345                 option_export_pack_edges(option);
3346         } else if (!strcmp(option, "quiet")) {
3347                 show_stats = 0;
3348         } else if (!strcmp(option, "stats")) {
3349                 show_stats = 1;
3350         } else if (!strcmp(option, "allow-unsafe-features")) {
3351                 ; /* already handled during early option parsing */
3352         } else {
3353                 return 0;
3354         }
3355
3356         return 1;
3357 }
3358
3359 static void check_unsafe_feature(const char *feature, int from_stream)
3360 {
3361         if (from_stream && !allow_unsafe_features)
3362                 die(_("feature '%s' forbidden in input without --allow-unsafe-features"),
3363                     feature);
3364 }
3365
3366 static int parse_one_feature(const char *feature, int from_stream)
3367 {
3368         const char *arg;
3369
3370         if (skip_prefix(feature, "date-format=", &arg)) {
3371                 option_date_format(arg);
3372         } else if (skip_prefix(feature, "import-marks=", &arg)) {
3373                 check_unsafe_feature("import-marks", from_stream);
3374                 option_import_marks(arg, from_stream, 0);
3375         } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3376                 check_unsafe_feature("import-marks-if-exists", from_stream);
3377                 option_import_marks(arg, from_stream, 1);
3378         } else if (skip_prefix(feature, "export-marks=", &arg)) {
3379                 check_unsafe_feature(feature, from_stream);
3380                 option_export_marks(arg);
3381         } else if (!strcmp(feature, "alias")) {
3382                 ; /* Don't die - this feature is supported */
3383         } else if (skip_prefix(feature, "rewrite-submodules-to=", &arg)) {
3384                 option_rewrite_submodules(arg, &sub_marks_to);
3385         } else if (skip_prefix(feature, "rewrite-submodules-from=", &arg)) {
3386                 option_rewrite_submodules(arg, &sub_marks_from);
3387         } else if (skip_prefix(feature, "rewrite-submodules-from=", &arg)) {
3388         } else if (!strcmp(feature, "get-mark")) {
3389                 ; /* Don't die - this feature is supported */
3390         } else if (!strcmp(feature, "cat-blob")) {
3391                 ; /* Don't die - this feature is supported */
3392         } else if (!strcmp(feature, "relative-marks")) {
3393                 relative_marks_paths = 1;
3394         } else if (!strcmp(feature, "no-relative-marks")) {
3395                 relative_marks_paths = 0;
3396         } else if (!strcmp(feature, "done")) {
3397                 require_explicit_termination = 1;
3398         } else if (!strcmp(feature, "force")) {
3399                 force_update = 1;
3400         } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3401                 ; /* do nothing; we have the feature */
3402         } else {
3403                 return 0;
3404         }
3405
3406         return 1;
3407 }
3408
3409 static void parse_feature(const char *feature)
3410 {
3411         if (seen_data_command)
3412                 die("Got feature command '%s' after data command", feature);
3413
3414         if (parse_one_feature(feature, 1))
3415                 return;
3416
3417         die("This version of fast-import does not support feature %s.", feature);
3418 }
3419
3420 static void parse_option(const char *option)
3421 {
3422         if (seen_data_command)
3423                 die("Got option command '%s' after data command", option);
3424
3425         if (parse_one_option(option))
3426                 return;
3427
3428         die("This version of fast-import does not support option: %s", option);
3429 }
3430
3431 static void git_pack_config(void)
3432 {
3433         int indexversion_value;
3434         int limit;
3435         unsigned long packsizelimit_value;
3436
3437         if (!git_config_get_ulong("pack.depth", &max_depth)) {
3438                 if (max_depth > MAX_DEPTH)
3439                         max_depth = MAX_DEPTH;
3440         }
3441         if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3442                 pack_idx_opts.version = indexversion_value;
3443                 if (pack_idx_opts.version > 2)
3444                         git_die_config("pack.indexversion",
3445                                         "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3446         }
3447         if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3448                 max_packsize = packsizelimit_value;
3449
3450         if (!git_config_get_int("fastimport.unpacklimit", &limit))
3451                 unpack_limit = limit;
3452         else if (!git_config_get_int("transfer.unpacklimit", &limit))
3453                 unpack_limit = limit;
3454
3455         git_config(git_default_config, NULL);
3456 }
3457
3458 static const char fast_import_usage[] =
3459 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3460
3461 static void parse_argv(void)
3462 {
3463         unsigned int i;
3464
3465         for (i = 1; i < global_argc; i++) {
3466                 const char *a = global_argv[i];
3467
3468                 if (*a != '-' || !strcmp(a, "--"))
3469                         break;
3470
3471                 if (!skip_prefix(a, "--", &a))
3472                         die("unknown option %s", a);
3473
3474                 if (parse_one_option(a))
3475                         continue;
3476
3477                 if (parse_one_feature(a, 0))
3478                         continue;
3479
3480                 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3481                         option_cat_blob_fd(a);
3482                         continue;
3483                 }
3484
3485                 die("unknown option --%s", a);
3486         }
3487         if (i != global_argc)
3488                 usage(fast_import_usage);
3489
3490         seen_data_command = 1;
3491         if (import_marks_file)
3492                 read_marks();
3493         build_mark_map(&sub_marks_from, &sub_marks_to);
3494 }
3495
3496 int cmd_main(int argc, const char **argv)
3497 {
3498         unsigned int i;
3499
3500         if (argc == 2 && !strcmp(argv[1], "-h"))
3501                 usage(fast_import_usage);
3502
3503         setup_git_directory();
3504         reset_pack_idx_option(&pack_idx_opts);
3505         git_pack_config();
3506
3507         alloc_objects(object_entry_alloc);
3508         strbuf_init(&command_buf, 0);
3509         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3510         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3511         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3512         marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3513
3514         /*
3515          * We don't parse most options until after we've seen the set of
3516          * "feature" lines at the start of the stream (which allows the command
3517          * line to override stream data). But we must do an early parse of any
3518          * command-line options that impact how we interpret the feature lines.
3519          */
3520         for (i = 1; i < argc; i++) {
3521                 const char *arg = argv[i];
3522                 if (*arg != '-' || !strcmp(arg, "--"))
3523                         break;
3524                 if (!strcmp(arg, "--allow-unsafe-features"))
3525                         allow_unsafe_features = 1;
3526         }
3527
3528         global_argc = argc;
3529         global_argv = argv;
3530
3531         rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3532         for (i = 0; i < (cmd_save - 1); i++)
3533                 rc_free[i].next = &rc_free[i + 1];
3534         rc_free[cmd_save - 1].next = NULL;
3535
3536         start_packfile();
3537         set_die_routine(die_nicely);
3538         set_checkpoint_signal();
3539         while (read_next_command() != EOF) {
3540                 const char *v;
3541                 if (!strcmp("blob", command_buf.buf))
3542                         parse_new_blob();
3543                 else if (skip_prefix(command_buf.buf, "commit ", &v))
3544                         parse_new_commit(v);
3545                 else if (skip_prefix(command_buf.buf, "tag ", &v))
3546                         parse_new_tag(v);
3547                 else if (skip_prefix(command_buf.buf, "reset ", &v))
3548                         parse_reset_branch(v);
3549                 else if (skip_prefix(command_buf.buf, "ls ", &v))
3550                         parse_ls(v, NULL);
3551                 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
3552                         parse_cat_blob(v);
3553                 else if (skip_prefix(command_buf.buf, "get-mark ", &v))
3554                         parse_get_mark(v);
3555                 else if (!strcmp("checkpoint", command_buf.buf))
3556                         parse_checkpoint();
3557                 else if (!strcmp("done", command_buf.buf))
3558                         break;
3559                 else if (!strcmp("alias", command_buf.buf))
3560                         parse_alias();
3561                 else if (starts_with(command_buf.buf, "progress "))
3562                         parse_progress();
3563                 else if (skip_prefix(command_buf.buf, "feature ", &v))
3564                         parse_feature(v);
3565                 else if (skip_prefix(command_buf.buf, "option git ", &v))
3566                         parse_option(v);
3567                 else if (starts_with(command_buf.buf, "option "))
3568                         /* ignore non-git options*/;
3569                 else
3570                         die("Unsupported command: %s", command_buf.buf);
3571
3572                 if (checkpoint_requested)
3573                         checkpoint();
3574         }
3575
3576         /* argv hasn't been parsed yet, do so */
3577         if (!seen_data_command)
3578                 parse_argv();
3579
3580         if (require_explicit_termination && feof(stdin))
3581                 die("stream ends early");
3582
3583         end_packfile();
3584
3585         dump_branches();
3586         dump_tags();
3587         unkeep_all_packs();
3588         dump_marks();
3589
3590         if (pack_edges)
3591                 fclose(pack_edges);
3592
3593         if (show_stats) {
3594                 uintmax_t total_count = 0, duplicate_count = 0;
3595                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3596                         total_count += object_count_by_type[i];
3597                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3598                         duplicate_count += duplicate_count_by_type[i];
3599
3600                 fprintf(stderr, "%s statistics:\n", argv[0]);
3601                 fprintf(stderr, "---------------------------------------------------------------------\n");
3602                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3603                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3604                 fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB], delta_count_attempts_by_type[OBJ_BLOB]);
3605                 fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE], delta_count_attempts_by_type[OBJ_TREE]);
3606                 fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT], delta_count_attempts_by_type[OBJ_COMMIT]);
3607                 fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG], delta_count_attempts_by_type[OBJ_TAG]);
3608                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3609                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3610                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3611                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3612                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3613                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3614                 fprintf(stderr, "---------------------------------------------------------------------\n");
3615                 pack_report();
3616                 fprintf(stderr, "---------------------------------------------------------------------\n");
3617                 fprintf(stderr, "\n");
3618         }
3619
3620         return failure ? 1 : 0;
3621 }