bisect--helper: retire `--check-expected-revs` subcommand
[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 hashmap_entry ent;
43         uint32_t type : TYPE_BITS,
44                 pack_id : PACK_ID_BITS,
45                 depth : DEPTH_BITS;
46 };
47
48 static int object_entry_hashcmp(const void *map_data,
49                                 const struct hashmap_entry *eptr,
50                                 const struct hashmap_entry *entry_or_key,
51                                 const void *keydata)
52 {
53         const struct object_id *oid = keydata;
54         const struct object_entry *e1, *e2;
55
56         e1 = container_of(eptr, const struct object_entry, ent);
57         if (oid)
58                 return oidcmp(&e1->idx.oid, oid);
59
60         e2 = container_of(entry_or_key, const struct object_entry, ent);
61         return oidcmp(&e1->idx.oid, &e2->idx.oid);
62 }
63
64 struct object_entry_pool {
65         struct object_entry_pool *next_pool;
66         struct object_entry *next_free;
67         struct object_entry *end;
68         struct object_entry entries[FLEX_ARRAY]; /* more */
69 };
70
71 struct mark_set {
72         union {
73                 struct object_id *oids[1024];
74                 struct object_entry *marked[1024];
75                 struct mark_set *sets[1024];
76         } data;
77         unsigned int shift;
78 };
79
80 struct last_object {
81         struct strbuf data;
82         off_t offset;
83         unsigned int depth;
84         unsigned no_swap : 1;
85 };
86
87 struct atom_str {
88         struct atom_str *next_atom;
89         unsigned short str_len;
90         char str_dat[FLEX_ARRAY]; /* more */
91 };
92
93 struct tree_content;
94 struct tree_entry {
95         struct tree_content *tree;
96         struct atom_str *name;
97         struct tree_entry_ms {
98                 uint16_t mode;
99                 struct object_id oid;
100         } versions[2];
101 };
102
103 struct tree_content {
104         unsigned int entry_capacity; /* must match avail_tree_content */
105         unsigned int entry_count;
106         unsigned int delta_depth;
107         struct tree_entry *entries[FLEX_ARRAY]; /* more */
108 };
109
110 struct avail_tree_content {
111         unsigned int entry_capacity; /* must match tree_content */
112         struct avail_tree_content *next_avail;
113 };
114
115 struct branch {
116         struct branch *table_next_branch;
117         struct branch *active_next_branch;
118         const char *name;
119         struct tree_entry branch_tree;
120         uintmax_t last_commit;
121         uintmax_t num_notes;
122         unsigned active : 1;
123         unsigned delete : 1;
124         unsigned pack_id : PACK_ID_BITS;
125         struct object_id oid;
126 };
127
128 struct tag {
129         struct tag *next_tag;
130         const char *name;
131         unsigned int pack_id;
132         struct object_id oid;
133 };
134
135 struct hash_list {
136         struct hash_list *next;
137         struct object_id oid;
138 };
139
140 typedef enum {
141         WHENSPEC_RAW = 1,
142         WHENSPEC_RAW_PERMISSIVE,
143         WHENSPEC_RFC2822,
144         WHENSPEC_NOW
145 } whenspec_type;
146
147 struct recent_command {
148         struct recent_command *prev;
149         struct recent_command *next;
150         char *buf;
151 };
152
153 typedef void (*mark_set_inserter_t)(struct mark_set *s, struct object_id *oid, uintmax_t mark);
154 typedef void (*each_mark_fn_t)(uintmax_t mark, void *obj, void *cbp);
155
156 /* Configured limits on output */
157 static unsigned long max_depth = 50;
158 static off_t max_packsize;
159 static int unpack_limit = 100;
160 static int force_update;
161
162 /* Stats and misc. counters */
163 static uintmax_t alloc_count;
164 static uintmax_t marks_set_count;
165 static uintmax_t object_count_by_type[1 << TYPE_BITS];
166 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
167 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
168 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
169 static unsigned long object_count;
170 static unsigned long branch_count;
171 static unsigned long branch_load_count;
172 static int failure;
173 static FILE *pack_edges;
174 static unsigned int show_stats = 1;
175 static int global_argc;
176 static const char **global_argv;
177
178 /* Memory pools */
179 static struct mem_pool fi_mem_pool =  {NULL, 2*1024*1024 -
180                                        sizeof(struct mp_block), 0 };
181
182 /* Atom management */
183 static unsigned int atom_table_sz = 4451;
184 static unsigned int atom_cnt;
185 static struct atom_str **atom_table;
186
187 /* The .pack file being generated */
188 static struct pack_idx_option pack_idx_opts;
189 static unsigned int pack_id;
190 static struct hashfile *pack_file;
191 static struct packed_git *pack_data;
192 static struct packed_git **all_packs;
193 static off_t pack_size;
194
195 /* Table of objects we've written. */
196 static unsigned int object_entry_alloc = 5000;
197 static struct object_entry_pool *blocks;
198 static struct hashmap object_table;
199 static struct mark_set *marks;
200 static const char *export_marks_file;
201 static const char *import_marks_file;
202 static int import_marks_file_from_stream;
203 static int import_marks_file_ignore_missing;
204 static int import_marks_file_done;
205 static int relative_marks_paths;
206
207 /* Our last blob */
208 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
209
210 /* Tree management */
211 static unsigned int tree_entry_alloc = 1000;
212 static void *avail_tree_entry;
213 static unsigned int avail_tree_table_sz = 100;
214 static struct avail_tree_content **avail_tree_table;
215 static size_t tree_entry_allocd;
216 static struct strbuf old_tree = STRBUF_INIT;
217 static struct strbuf new_tree = STRBUF_INIT;
218
219 /* Branch data */
220 static unsigned long max_active_branches = 5;
221 static unsigned long cur_active_branches;
222 static unsigned long branch_table_sz = 1039;
223 static struct branch **branch_table;
224 static struct branch *active_branches;
225
226 /* Tag data */
227 static struct tag *first_tag;
228 static struct tag *last_tag;
229
230 /* Input stream parsing */
231 static whenspec_type whenspec = WHENSPEC_RAW;
232 static struct strbuf command_buf = STRBUF_INIT;
233 static int unread_command_buf;
234 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
235 static struct recent_command *cmd_tail = &cmd_hist;
236 static struct recent_command *rc_free;
237 static unsigned int cmd_save = 100;
238 static uintmax_t next_mark;
239 static struct strbuf new_data = STRBUF_INIT;
240 static int seen_data_command;
241 static int require_explicit_termination;
242 static int allow_unsafe_features;
243
244 /* Signal handling */
245 static volatile sig_atomic_t checkpoint_requested;
246
247 /* Submodule marks */
248 static struct string_list sub_marks_from = STRING_LIST_INIT_DUP;
249 static struct string_list sub_marks_to = STRING_LIST_INIT_DUP;
250 static kh_oid_map_t *sub_oid_map;
251
252 /* Where to write output of cat-blob commands */
253 static int cat_blob_fd = STDOUT_FILENO;
254
255 static void parse_argv(void);
256 static void parse_get_mark(const char *p);
257 static void parse_cat_blob(const char *p);
258 static void parse_ls(const char *p, struct branch *b);
259
260 static void for_each_mark(struct mark_set *m, uintmax_t base, each_mark_fn_t callback, void *p)
261 {
262         uintmax_t k;
263         if (m->shift) {
264                 for (k = 0; k < 1024; k++) {
265                         if (m->data.sets[k])
266                                 for_each_mark(m->data.sets[k], base + (k << m->shift), callback, p);
267                 }
268         } else {
269                 for (k = 0; k < 1024; k++) {
270                         if (m->data.marked[k])
271                                 callback(base + k, m->data.marked[k], p);
272                 }
273         }
274 }
275
276 static void dump_marks_fn(uintmax_t mark, void *object, void *cbp) {
277         struct object_entry *e = object;
278         FILE *f = cbp;
279
280         fprintf(f, ":%" PRIuMAX " %s\n", mark, oid_to_hex(&e->idx.oid));
281 }
282
283 static void write_branch_report(FILE *rpt, struct branch *b)
284 {
285         fprintf(rpt, "%s:\n", b->name);
286
287         fprintf(rpt, "  status      :");
288         if (b->active)
289                 fputs(" active", rpt);
290         if (b->branch_tree.tree)
291                 fputs(" loaded", rpt);
292         if (is_null_oid(&b->branch_tree.versions[1].oid))
293                 fputs(" dirty", rpt);
294         fputc('\n', rpt);
295
296         fprintf(rpt, "  tip commit  : %s\n", oid_to_hex(&b->oid));
297         fprintf(rpt, "  old tree    : %s\n",
298                 oid_to_hex(&b->branch_tree.versions[0].oid));
299         fprintf(rpt, "  cur tree    : %s\n",
300                 oid_to_hex(&b->branch_tree.versions[1].oid));
301         fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
302
303         fputs("  last pack   : ", rpt);
304         if (b->pack_id < MAX_PACK_ID)
305                 fprintf(rpt, "%u", b->pack_id);
306         fputc('\n', rpt);
307
308         fputc('\n', rpt);
309 }
310
311 static void write_crash_report(const char *err)
312 {
313         char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
314         FILE *rpt = fopen(loc, "w");
315         struct branch *b;
316         unsigned long lu;
317         struct recent_command *rc;
318
319         if (!rpt) {
320                 error_errno("can't write crash report %s", loc);
321                 free(loc);
322                 return;
323         }
324
325         fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
326
327         fprintf(rpt, "fast-import crash report:\n");
328         fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
329         fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
330         fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
331         fputc('\n', rpt);
332
333         fputs("fatal: ", rpt);
334         fputs(err, rpt);
335         fputc('\n', rpt);
336
337         fputc('\n', rpt);
338         fputs("Most Recent Commands Before Crash\n", rpt);
339         fputs("---------------------------------\n", rpt);
340         for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
341                 if (rc->next == &cmd_hist)
342                         fputs("* ", rpt);
343                 else
344                         fputs("  ", rpt);
345                 fputs(rc->buf, rpt);
346                 fputc('\n', rpt);
347         }
348
349         fputc('\n', rpt);
350         fputs("Active Branch LRU\n", rpt);
351         fputs("-----------------\n", rpt);
352         fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
353                 cur_active_branches,
354                 max_active_branches);
355         fputc('\n', rpt);
356         fputs("  pos  clock name\n", rpt);
357         fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
358         for (b = active_branches, lu = 0; b; b = b->active_next_branch)
359                 fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
360                         ++lu, b->last_commit, b->name);
361
362         fputc('\n', rpt);
363         fputs("Inactive Branches\n", rpt);
364         fputs("-----------------\n", rpt);
365         for (lu = 0; lu < branch_table_sz; lu++) {
366                 for (b = branch_table[lu]; b; b = b->table_next_branch)
367                         write_branch_report(rpt, b);
368         }
369
370         if (first_tag) {
371                 struct tag *tg;
372                 fputc('\n', rpt);
373                 fputs("Annotated Tags\n", rpt);
374                 fputs("--------------\n", rpt);
375                 for (tg = first_tag; tg; tg = tg->next_tag) {
376                         fputs(oid_to_hex(&tg->oid), rpt);
377                         fputc(' ', rpt);
378                         fputs(tg->name, rpt);
379                         fputc('\n', rpt);
380                 }
381         }
382
383         fputc('\n', rpt);
384         fputs("Marks\n", rpt);
385         fputs("-----\n", rpt);
386         if (export_marks_file)
387                 fprintf(rpt, "  exported to %s\n", export_marks_file);
388         else
389                 for_each_mark(marks, 0, dump_marks_fn, rpt);
390
391         fputc('\n', rpt);
392         fputs("-------------------\n", rpt);
393         fputs("END OF CRASH REPORT\n", rpt);
394         fclose(rpt);
395         free(loc);
396 }
397
398 static void end_packfile(void);
399 static void unkeep_all_packs(void);
400 static void dump_marks(void);
401
402 static NORETURN void die_nicely(const char *err, va_list params)
403 {
404         static int zombie;
405         char message[2 * PATH_MAX];
406
407         vsnprintf(message, sizeof(message), err, params);
408         fputs("fatal: ", stderr);
409         fputs(message, stderr);
410         fputc('\n', stderr);
411
412         if (!zombie) {
413                 zombie = 1;
414                 write_crash_report(message);
415                 end_packfile();
416                 unkeep_all_packs();
417                 dump_marks();
418         }
419         exit(128);
420 }
421
422 #ifndef SIGUSR1 /* Windows, for example */
423
424 static void set_checkpoint_signal(void)
425 {
426 }
427
428 #else
429
430 static void checkpoint_signal(int signo)
431 {
432         checkpoint_requested = 1;
433 }
434
435 static void set_checkpoint_signal(void)
436 {
437         struct sigaction sa;
438
439         memset(&sa, 0, sizeof(sa));
440         sa.sa_handler = checkpoint_signal;
441         sigemptyset(&sa.sa_mask);
442         sa.sa_flags = SA_RESTART;
443         sigaction(SIGUSR1, &sa, NULL);
444 }
445
446 #endif
447
448 static void alloc_objects(unsigned int cnt)
449 {
450         struct object_entry_pool *b;
451
452         b = xmalloc(sizeof(struct object_entry_pool)
453                 + cnt * sizeof(struct object_entry));
454         b->next_pool = blocks;
455         b->next_free = b->entries;
456         b->end = b->entries + cnt;
457         blocks = b;
458         alloc_count += cnt;
459 }
460
461 static struct object_entry *new_object(struct object_id *oid)
462 {
463         struct object_entry *e;
464
465         if (blocks->next_free == blocks->end)
466                 alloc_objects(object_entry_alloc);
467
468         e = blocks->next_free++;
469         oidcpy(&e->idx.oid, oid);
470         return e;
471 }
472
473 static struct object_entry *find_object(struct object_id *oid)
474 {
475         return hashmap_get_entry_from_hash(&object_table, oidhash(oid), oid,
476                                            struct object_entry, ent);
477 }
478
479 static struct object_entry *insert_object(struct object_id *oid)
480 {
481         struct object_entry *e;
482         unsigned int hash = oidhash(oid);
483
484         e = hashmap_get_entry_from_hash(&object_table, hash, oid,
485                                         struct object_entry, ent);
486         if (!e) {
487                 e = new_object(oid);
488                 e->idx.offset = 0;
489                 hashmap_entry_init(&e->ent, hash);
490                 hashmap_add(&object_table, &e->ent);
491         }
492
493         return e;
494 }
495
496 static void invalidate_pack_id(unsigned int id)
497 {
498         unsigned long lu;
499         struct tag *t;
500         struct hashmap_iter iter;
501         struct object_entry *e;
502
503         hashmap_for_each_entry(&object_table, &iter, e, ent) {
504                 if (e->pack_id == id)
505                         e->pack_id = MAX_PACK_ID;
506         }
507
508         for (lu = 0; lu < branch_table_sz; lu++) {
509                 struct branch *b;
510
511                 for (b = branch_table[lu]; b; b = b->table_next_branch)
512                         if (b->pack_id == id)
513                                 b->pack_id = MAX_PACK_ID;
514         }
515
516         for (t = first_tag; t; t = t->next_tag)
517                 if (t->pack_id == id)
518                         t->pack_id = MAX_PACK_ID;
519 }
520
521 static unsigned int hc_str(const char *s, size_t len)
522 {
523         unsigned int r = 0;
524         while (len-- > 0)
525                 r = r * 31 + *s++;
526         return r;
527 }
528
529 static char *pool_strdup(const char *s)
530 {
531         size_t len = strlen(s) + 1;
532         char *r = mem_pool_alloc(&fi_mem_pool, len);
533         memcpy(r, s, len);
534         return r;
535 }
536
537 static void insert_mark(struct mark_set *s, uintmax_t idnum, struct object_entry *oe)
538 {
539         while ((idnum >> s->shift) >= 1024) {
540                 s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
541                 s->shift = marks->shift + 10;
542                 s->data.sets[0] = marks;
543                 marks = s;
544         }
545         while (s->shift) {
546                 uintmax_t i = idnum >> s->shift;
547                 idnum -= i << s->shift;
548                 if (!s->data.sets[i]) {
549                         s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
550                         s->data.sets[i]->shift = s->shift - 10;
551                 }
552                 s = s->data.sets[i];
553         }
554         if (!s->data.marked[idnum])
555                 marks_set_count++;
556         s->data.marked[idnum] = oe;
557 }
558
559 static void *find_mark(struct mark_set *s, uintmax_t idnum)
560 {
561         uintmax_t orig_idnum = idnum;
562         struct object_entry *oe = NULL;
563         if ((idnum >> s->shift) < 1024) {
564                 while (s && s->shift) {
565                         uintmax_t i = idnum >> s->shift;
566                         idnum -= i << s->shift;
567                         s = s->data.sets[i];
568                 }
569                 if (s)
570                         oe = s->data.marked[idnum];
571         }
572         if (!oe)
573                 die("mark :%" PRIuMAX " not declared", orig_idnum);
574         return oe;
575 }
576
577 static struct atom_str *to_atom(const char *s, unsigned short len)
578 {
579         unsigned int hc = hc_str(s, len) % atom_table_sz;
580         struct atom_str *c;
581
582         for (c = atom_table[hc]; c; c = c->next_atom)
583                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
584                         return c;
585
586         c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
587         c->str_len = len;
588         memcpy(c->str_dat, s, len);
589         c->str_dat[len] = 0;
590         c->next_atom = atom_table[hc];
591         atom_table[hc] = c;
592         atom_cnt++;
593         return c;
594 }
595
596 static struct branch *lookup_branch(const char *name)
597 {
598         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
599         struct branch *b;
600
601         for (b = branch_table[hc]; b; b = b->table_next_branch)
602                 if (!strcmp(name, b->name))
603                         return b;
604         return NULL;
605 }
606
607 static struct branch *new_branch(const char *name)
608 {
609         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
610         struct branch *b = lookup_branch(name);
611
612         if (b)
613                 die("Invalid attempt to create duplicate branch: %s", name);
614         if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
615                 die("Branch name doesn't conform to GIT standards: %s", name);
616
617         b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
618         b->name = pool_strdup(name);
619         b->table_next_branch = branch_table[hc];
620         b->branch_tree.versions[0].mode = S_IFDIR;
621         b->branch_tree.versions[1].mode = S_IFDIR;
622         b->num_notes = 0;
623         b->active = 0;
624         b->pack_id = MAX_PACK_ID;
625         branch_table[hc] = b;
626         branch_count++;
627         return b;
628 }
629
630 static unsigned int hc_entries(unsigned int cnt)
631 {
632         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
633         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
634 }
635
636 static struct tree_content *new_tree_content(unsigned int cnt)
637 {
638         struct avail_tree_content *f, *l = NULL;
639         struct tree_content *t;
640         unsigned int hc = hc_entries(cnt);
641
642         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
643                 if (f->entry_capacity >= cnt)
644                         break;
645
646         if (f) {
647                 if (l)
648                         l->next_avail = f->next_avail;
649                 else
650                         avail_tree_table[hc] = f->next_avail;
651         } else {
652                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
653                 f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
654                 f->entry_capacity = cnt;
655         }
656
657         t = (struct tree_content*)f;
658         t->entry_count = 0;
659         t->delta_depth = 0;
660         return t;
661 }
662
663 static void release_tree_entry(struct tree_entry *e);
664 static void release_tree_content(struct tree_content *t)
665 {
666         struct avail_tree_content *f = (struct avail_tree_content*)t;
667         unsigned int hc = hc_entries(f->entry_capacity);
668         f->next_avail = avail_tree_table[hc];
669         avail_tree_table[hc] = f;
670 }
671
672 static void release_tree_content_recursive(struct tree_content *t)
673 {
674         unsigned int i;
675         for (i = 0; i < t->entry_count; i++)
676                 release_tree_entry(t->entries[i]);
677         release_tree_content(t);
678 }
679
680 static struct tree_content *grow_tree_content(
681         struct tree_content *t,
682         int amt)
683 {
684         struct tree_content *r = new_tree_content(t->entry_count + amt);
685         r->entry_count = t->entry_count;
686         r->delta_depth = t->delta_depth;
687         COPY_ARRAY(r->entries, t->entries, t->entry_count);
688         release_tree_content(t);
689         return r;
690 }
691
692 static struct tree_entry *new_tree_entry(void)
693 {
694         struct tree_entry *e;
695
696         if (!avail_tree_entry) {
697                 unsigned int n = tree_entry_alloc;
698                 tree_entry_allocd += n * sizeof(struct tree_entry);
699                 ALLOC_ARRAY(e, n);
700                 avail_tree_entry = e;
701                 while (n-- > 1) {
702                         *((void**)e) = e + 1;
703                         e++;
704                 }
705                 *((void**)e) = NULL;
706         }
707
708         e = avail_tree_entry;
709         avail_tree_entry = *((void**)e);
710         return e;
711 }
712
713 static void release_tree_entry(struct tree_entry *e)
714 {
715         if (e->tree)
716                 release_tree_content_recursive(e->tree);
717         *((void**)e) = avail_tree_entry;
718         avail_tree_entry = e;
719 }
720
721 static struct tree_content *dup_tree_content(struct tree_content *s)
722 {
723         struct tree_content *d;
724         struct tree_entry *a, *b;
725         unsigned int i;
726
727         if (!s)
728                 return NULL;
729         d = new_tree_content(s->entry_count);
730         for (i = 0; i < s->entry_count; i++) {
731                 a = s->entries[i];
732                 b = new_tree_entry();
733                 memcpy(b, a, sizeof(*a));
734                 if (a->tree && is_null_oid(&b->versions[1].oid))
735                         b->tree = dup_tree_content(a->tree);
736                 else
737                         b->tree = NULL;
738                 d->entries[i] = b;
739         }
740         d->entry_count = s->entry_count;
741         d->delta_depth = s->delta_depth;
742
743         return d;
744 }
745
746 static void start_packfile(void)
747 {
748         struct strbuf tmp_file = STRBUF_INIT;
749         struct packed_git *p;
750         struct pack_header hdr;
751         int pack_fd;
752
753         pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
754         FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
755         strbuf_release(&tmp_file);
756
757         p->pack_fd = pack_fd;
758         p->do_not_close = 1;
759         pack_file = hashfd(pack_fd, p->pack_name);
760
761         hdr.hdr_signature = htonl(PACK_SIGNATURE);
762         hdr.hdr_version = htonl(2);
763         hdr.hdr_entries = 0;
764         hashwrite(pack_file, &hdr, sizeof(hdr));
765
766         pack_data = p;
767         pack_size = sizeof(hdr);
768         object_count = 0;
769
770         REALLOC_ARRAY(all_packs, pack_id + 1);
771         all_packs[pack_id] = p;
772 }
773
774 static const char *create_index(void)
775 {
776         const char *tmpfile;
777         struct pack_idx_entry **idx, **c, **last;
778         struct object_entry *e;
779         struct object_entry_pool *o;
780
781         /* Build the table of object IDs. */
782         ALLOC_ARRAY(idx, object_count);
783         c = idx;
784         for (o = blocks; o; o = o->next_pool)
785                 for (e = o->next_free; e-- != o->entries;)
786                         if (pack_id == e->pack_id)
787                                 *c++ = &e->idx;
788         last = idx + object_count;
789         if (c != last)
790                 die("internal consistency error creating the index");
791
792         tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts,
793                                  pack_data->hash);
794         free(idx);
795         return tmpfile;
796 }
797
798 static char *keep_pack(const char *curr_index_name)
799 {
800         static const char *keep_msg = "fast-import";
801         struct strbuf name = STRBUF_INIT;
802         int keep_fd;
803
804         odb_pack_name(&name, pack_data->hash, "keep");
805         keep_fd = odb_pack_keep(name.buf);
806         if (keep_fd < 0)
807                 die_errno("cannot create keep file");
808         write_or_die(keep_fd, keep_msg, strlen(keep_msg));
809         if (close(keep_fd))
810                 die_errno("failed to write keep file");
811
812         odb_pack_name(&name, pack_data->hash, "pack");
813         if (finalize_object_file(pack_data->pack_name, name.buf))
814                 die("cannot store pack file");
815
816         odb_pack_name(&name, pack_data->hash, "idx");
817         if (finalize_object_file(curr_index_name, name.buf))
818                 die("cannot store index file");
819         free((void *)curr_index_name);
820         return strbuf_detach(&name, NULL);
821 }
822
823 static void unkeep_all_packs(void)
824 {
825         struct strbuf name = STRBUF_INIT;
826         int k;
827
828         for (k = 0; k < pack_id; k++) {
829                 struct packed_git *p = all_packs[k];
830                 odb_pack_name(&name, p->hash, "keep");
831                 unlink_or_warn(name.buf);
832         }
833         strbuf_release(&name);
834 }
835
836 static int loosen_small_pack(const struct packed_git *p)
837 {
838         struct child_process unpack = CHILD_PROCESS_INIT;
839
840         if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
841                 die_errno("Failed seeking to start of '%s'", p->pack_name);
842
843         unpack.in = p->pack_fd;
844         unpack.git_cmd = 1;
845         unpack.stdout_to_stderr = 1;
846         strvec_push(&unpack.args, "unpack-objects");
847         if (!show_stats)
848                 strvec_push(&unpack.args, "-q");
849
850         return run_command(&unpack);
851 }
852
853 static void end_packfile(void)
854 {
855         static int running;
856
857         if (running || !pack_data)
858                 return;
859
860         running = 1;
861         clear_delta_base_cache();
862         if (object_count) {
863                 struct packed_git *new_p;
864                 struct object_id cur_pack_oid;
865                 char *idx_name;
866                 int i;
867                 struct branch *b;
868                 struct tag *t;
869
870                 close_pack_windows(pack_data);
871                 finalize_hashfile(pack_file, cur_pack_oid.hash, 0);
872                 fixup_pack_header_footer(pack_data->pack_fd, pack_data->hash,
873                                          pack_data->pack_name, object_count,
874                                          cur_pack_oid.hash, pack_size);
875
876                 if (object_count <= unpack_limit) {
877                         if (!loosen_small_pack(pack_data)) {
878                                 invalidate_pack_id(pack_id);
879                                 goto discard_pack;
880                         }
881                 }
882
883                 close(pack_data->pack_fd);
884                 idx_name = keep_pack(create_index());
885
886                 /* Register the packfile with core git's machinery. */
887                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
888                 if (!new_p)
889                         die("core git rejected index %s", idx_name);
890                 all_packs[pack_id] = new_p;
891                 install_packed_git(the_repository, new_p);
892                 free(idx_name);
893
894                 /* Print the boundary */
895                 if (pack_edges) {
896                         fprintf(pack_edges, "%s:", new_p->pack_name);
897                         for (i = 0; i < branch_table_sz; i++) {
898                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
899                                         if (b->pack_id == pack_id)
900                                                 fprintf(pack_edges, " %s",
901                                                         oid_to_hex(&b->oid));
902                                 }
903                         }
904                         for (t = first_tag; t; t = t->next_tag) {
905                                 if (t->pack_id == pack_id)
906                                         fprintf(pack_edges, " %s",
907                                                 oid_to_hex(&t->oid));
908                         }
909                         fputc('\n', pack_edges);
910                         fflush(pack_edges);
911                 }
912
913                 pack_id++;
914         }
915         else {
916 discard_pack:
917                 close(pack_data->pack_fd);
918                 unlink_or_warn(pack_data->pack_name);
919         }
920         FREE_AND_NULL(pack_data);
921         running = 0;
922
923         /* We can't carry a delta across packfiles. */
924         strbuf_release(&last_blob.data);
925         last_blob.offset = 0;
926         last_blob.depth = 0;
927 }
928
929 static void cycle_packfile(void)
930 {
931         end_packfile();
932         start_packfile();
933 }
934
935 static int store_object(
936         enum object_type type,
937         struct strbuf *dat,
938         struct last_object *last,
939         struct object_id *oidout,
940         uintmax_t mark)
941 {
942         void *out, *delta;
943         struct object_entry *e;
944         unsigned char hdr[96];
945         struct object_id oid;
946         unsigned long hdrlen, deltalen;
947         git_hash_ctx c;
948         git_zstream s;
949
950         hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
951                            type_name(type), (unsigned long)dat->len) + 1;
952         the_hash_algo->init_fn(&c);
953         the_hash_algo->update_fn(&c, hdr, hdrlen);
954         the_hash_algo->update_fn(&c, dat->buf, dat->len);
955         the_hash_algo->final_fn(oid.hash, &c);
956         if (oidout)
957                 oidcpy(oidout, &oid);
958
959         e = insert_object(&oid);
960         if (mark)
961                 insert_mark(marks, mark, e);
962         if (e->idx.offset) {
963                 duplicate_count_by_type[type]++;
964                 return 1;
965         } else if (find_sha1_pack(oid.hash,
966                                   get_all_packs(the_repository))) {
967                 e->type = type;
968                 e->pack_id = MAX_PACK_ID;
969                 e->idx.offset = 1; /* just not zero! */
970                 duplicate_count_by_type[type]++;
971                 return 1;
972         }
973
974         if (last && last->data.len && last->data.buf && last->depth < max_depth
975                 && dat->len > the_hash_algo->rawsz) {
976
977                 delta_count_attempts_by_type[type]++;
978                 delta = diff_delta(last->data.buf, last->data.len,
979                         dat->buf, dat->len,
980                         &deltalen, dat->len - the_hash_algo->rawsz);
981         } else
982                 delta = NULL;
983
984         git_deflate_init(&s, pack_compression_level);
985         if (delta) {
986                 s.next_in = delta;
987                 s.avail_in = deltalen;
988         } else {
989                 s.next_in = (void *)dat->buf;
990                 s.avail_in = dat->len;
991         }
992         s.avail_out = git_deflate_bound(&s, s.avail_in);
993         s.next_out = out = xmalloc(s.avail_out);
994         while (git_deflate(&s, Z_FINISH) == Z_OK)
995                 ; /* nothing */
996         git_deflate_end(&s);
997
998         /* Determine if we should auto-checkpoint. */
999         if ((max_packsize
1000                 && (pack_size + PACK_SIZE_THRESHOLD + s.total_out) > max_packsize)
1001                 || (pack_size + PACK_SIZE_THRESHOLD + s.total_out) < pack_size) {
1002
1003                 /* This new object needs to *not* have the current pack_id. */
1004                 e->pack_id = pack_id + 1;
1005                 cycle_packfile();
1006
1007                 /* We cannot carry a delta into the new pack. */
1008                 if (delta) {
1009                         FREE_AND_NULL(delta);
1010
1011                         git_deflate_init(&s, pack_compression_level);
1012                         s.next_in = (void *)dat->buf;
1013                         s.avail_in = dat->len;
1014                         s.avail_out = git_deflate_bound(&s, s.avail_in);
1015                         s.next_out = out = xrealloc(out, s.avail_out);
1016                         while (git_deflate(&s, Z_FINISH) == Z_OK)
1017                                 ; /* nothing */
1018                         git_deflate_end(&s);
1019                 }
1020         }
1021
1022         e->type = type;
1023         e->pack_id = pack_id;
1024         e->idx.offset = pack_size;
1025         object_count++;
1026         object_count_by_type[type]++;
1027
1028         crc32_begin(pack_file);
1029
1030         if (delta) {
1031                 off_t ofs = e->idx.offset - last->offset;
1032                 unsigned pos = sizeof(hdr) - 1;
1033
1034                 delta_count_by_type[type]++;
1035                 e->depth = last->depth + 1;
1036
1037                 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1038                                                       OBJ_OFS_DELTA, deltalen);
1039                 hashwrite(pack_file, hdr, hdrlen);
1040                 pack_size += hdrlen;
1041
1042                 hdr[pos] = ofs & 127;
1043                 while (ofs >>= 7)
1044                         hdr[--pos] = 128 | (--ofs & 127);
1045                 hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1046                 pack_size += sizeof(hdr) - pos;
1047         } else {
1048                 e->depth = 0;
1049                 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1050                                                       type, dat->len);
1051                 hashwrite(pack_file, hdr, hdrlen);
1052                 pack_size += hdrlen;
1053         }
1054
1055         hashwrite(pack_file, out, s.total_out);
1056         pack_size += s.total_out;
1057
1058         e->idx.crc32 = crc32_end(pack_file);
1059
1060         free(out);
1061         free(delta);
1062         if (last) {
1063                 if (last->no_swap) {
1064                         last->data = *dat;
1065                 } else {
1066                         strbuf_swap(&last->data, dat);
1067                 }
1068                 last->offset = e->idx.offset;
1069                 last->depth = e->depth;
1070         }
1071         return 0;
1072 }
1073
1074 static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1075 {
1076         if (hashfile_truncate(pack_file, checkpoint))
1077                 die_errno("cannot truncate pack to skip duplicate");
1078         pack_size = checkpoint->offset;
1079 }
1080
1081 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1082 {
1083         size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1084         unsigned char *in_buf = xmalloc(in_sz);
1085         unsigned char *out_buf = xmalloc(out_sz);
1086         struct object_entry *e;
1087         struct object_id oid;
1088         unsigned long hdrlen;
1089         off_t offset;
1090         git_hash_ctx c;
1091         git_zstream s;
1092         struct hashfile_checkpoint checkpoint;
1093         int status = Z_OK;
1094
1095         /* Determine if we should auto-checkpoint. */
1096         if ((max_packsize
1097                 && (pack_size + PACK_SIZE_THRESHOLD + len) > max_packsize)
1098                 || (pack_size + PACK_SIZE_THRESHOLD + len) < pack_size)
1099                 cycle_packfile();
1100
1101         hashfile_checkpoint(pack_file, &checkpoint);
1102         offset = checkpoint.offset;
1103
1104         hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1105
1106         the_hash_algo->init_fn(&c);
1107         the_hash_algo->update_fn(&c, out_buf, hdrlen);
1108
1109         crc32_begin(pack_file);
1110
1111         git_deflate_init(&s, pack_compression_level);
1112
1113         hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1114
1115         s.next_out = out_buf + hdrlen;
1116         s.avail_out = out_sz - hdrlen;
1117
1118         while (status != Z_STREAM_END) {
1119                 if (0 < len && !s.avail_in) {
1120                         size_t cnt = in_sz < len ? in_sz : (size_t)len;
1121                         size_t n = fread(in_buf, 1, cnt, stdin);
1122                         if (!n && feof(stdin))
1123                                 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1124
1125                         the_hash_algo->update_fn(&c, in_buf, n);
1126                         s.next_in = in_buf;
1127                         s.avail_in = n;
1128                         len -= n;
1129                 }
1130
1131                 status = git_deflate(&s, len ? 0 : Z_FINISH);
1132
1133                 if (!s.avail_out || status == Z_STREAM_END) {
1134                         size_t n = s.next_out - out_buf;
1135                         hashwrite(pack_file, out_buf, n);
1136                         pack_size += n;
1137                         s.next_out = out_buf;
1138                         s.avail_out = out_sz;
1139                 }
1140
1141                 switch (status) {
1142                 case Z_OK:
1143                 case Z_BUF_ERROR:
1144                 case Z_STREAM_END:
1145                         continue;
1146                 default:
1147                         die("unexpected deflate failure: %d", status);
1148                 }
1149         }
1150         git_deflate_end(&s);
1151         the_hash_algo->final_fn(oid.hash, &c);
1152
1153         if (oidout)
1154                 oidcpy(oidout, &oid);
1155
1156         e = insert_object(&oid);
1157
1158         if (mark)
1159                 insert_mark(marks, mark, e);
1160
1161         if (e->idx.offset) {
1162                 duplicate_count_by_type[OBJ_BLOB]++;
1163                 truncate_pack(&checkpoint);
1164
1165         } else if (find_sha1_pack(oid.hash,
1166                                   get_all_packs(the_repository))) {
1167                 e->type = OBJ_BLOB;
1168                 e->pack_id = MAX_PACK_ID;
1169                 e->idx.offset = 1; /* just not zero! */
1170                 duplicate_count_by_type[OBJ_BLOB]++;
1171                 truncate_pack(&checkpoint);
1172
1173         } else {
1174                 e->depth = 0;
1175                 e->type = OBJ_BLOB;
1176                 e->pack_id = pack_id;
1177                 e->idx.offset = offset;
1178                 e->idx.crc32 = crc32_end(pack_file);
1179                 object_count++;
1180                 object_count_by_type[OBJ_BLOB]++;
1181         }
1182
1183         free(in_buf);
1184         free(out_buf);
1185 }
1186
1187 /* All calls must be guarded by find_object() or find_mark() to
1188  * ensure the 'struct object_entry' passed was written by this
1189  * process instance.  We unpack the entry by the offset, avoiding
1190  * the need for the corresponding .idx file.  This unpacking rule
1191  * works because we only use OBJ_REF_DELTA within the packfiles
1192  * created by fast-import.
1193  *
1194  * oe must not be NULL.  Such an oe usually comes from giving
1195  * an unknown SHA-1 to find_object() or an undefined mark to
1196  * find_mark().  Callers must test for this condition and use
1197  * the standard read_sha1_file() when it happens.
1198  *
1199  * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1200  * find_mark(), where the mark was reloaded from an existing marks
1201  * file and is referencing an object that this fast-import process
1202  * instance did not write out to a packfile.  Callers must test for
1203  * this condition and use read_sha1_file() instead.
1204  */
1205 static void *gfi_unpack_entry(
1206         struct object_entry *oe,
1207         unsigned long *sizep)
1208 {
1209         enum object_type type;
1210         struct packed_git *p = all_packs[oe->pack_id];
1211         if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1212                 /* The object is stored in the packfile we are writing to
1213                  * and we have modified it since the last time we scanned
1214                  * back to read a previously written object.  If an old
1215                  * window covered [p->pack_size, p->pack_size + rawsz) its
1216                  * data is stale and is not valid.  Closing all windows
1217                  * and updating the packfile length ensures we can read
1218                  * the newly written data.
1219                  */
1220                 close_pack_windows(p);
1221                 hashflush(pack_file);
1222
1223                 /* We have to offer rawsz bytes additional on the end of
1224                  * the packfile as the core unpacker code assumes the
1225                  * footer is present at the file end and must promise
1226                  * at least rawsz bytes within any window it maps.  But
1227                  * we don't actually create the footer here.
1228                  */
1229                 p->pack_size = pack_size + the_hash_algo->rawsz;
1230         }
1231         return unpack_entry(the_repository, p, oe->idx.offset, &type, sizep);
1232 }
1233
1234 static const char *get_mode(const char *str, uint16_t *modep)
1235 {
1236         unsigned char c;
1237         uint16_t mode = 0;
1238
1239         while ((c = *str++) != ' ') {
1240                 if (c < '0' || c > '7')
1241                         return NULL;
1242                 mode = (mode << 3) + (c - '0');
1243         }
1244         *modep = mode;
1245         return str;
1246 }
1247
1248 static void load_tree(struct tree_entry *root)
1249 {
1250         struct object_id *oid = &root->versions[1].oid;
1251         struct object_entry *myoe;
1252         struct tree_content *t;
1253         unsigned long size;
1254         char *buf;
1255         const char *c;
1256
1257         root->tree = t = new_tree_content(8);
1258         if (is_null_oid(oid))
1259                 return;
1260
1261         myoe = find_object(oid);
1262         if (myoe && myoe->pack_id != MAX_PACK_ID) {
1263                 if (myoe->type != OBJ_TREE)
1264                         die("Not a tree: %s", oid_to_hex(oid));
1265                 t->delta_depth = myoe->depth;
1266                 buf = gfi_unpack_entry(myoe, &size);
1267                 if (!buf)
1268                         die("Can't load tree %s", oid_to_hex(oid));
1269         } else {
1270                 enum object_type type;
1271                 buf = read_object_file(oid, &type, &size);
1272                 if (!buf || type != OBJ_TREE)
1273                         die("Can't load tree %s", oid_to_hex(oid));
1274         }
1275
1276         c = buf;
1277         while (c != (buf + size)) {
1278                 struct tree_entry *e = new_tree_entry();
1279
1280                 if (t->entry_count == t->entry_capacity)
1281                         root->tree = t = grow_tree_content(t, t->entry_count);
1282                 t->entries[t->entry_count++] = e;
1283
1284                 e->tree = NULL;
1285                 c = get_mode(c, &e->versions[1].mode);
1286                 if (!c)
1287                         die("Corrupt mode in %s", oid_to_hex(oid));
1288                 e->versions[0].mode = e->versions[1].mode;
1289                 e->name = to_atom(c, strlen(c));
1290                 c += e->name->str_len + 1;
1291                 hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1292                 hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1293                 c += the_hash_algo->rawsz;
1294         }
1295         free(buf);
1296 }
1297
1298 static int tecmp0 (const void *_a, const void *_b)
1299 {
1300         struct tree_entry *a = *((struct tree_entry**)_a);
1301         struct tree_entry *b = *((struct tree_entry**)_b);
1302         return base_name_compare(
1303                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1304                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1305 }
1306
1307 static int tecmp1 (const void *_a, const void *_b)
1308 {
1309         struct tree_entry *a = *((struct tree_entry**)_a);
1310         struct tree_entry *b = *((struct tree_entry**)_b);
1311         return base_name_compare(
1312                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1313                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1314 }
1315
1316 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1317 {
1318         size_t maxlen = 0;
1319         unsigned int i;
1320
1321         if (!v)
1322                 QSORT(t->entries, t->entry_count, tecmp0);
1323         else
1324                 QSORT(t->entries, t->entry_count, tecmp1);
1325
1326         for (i = 0; i < t->entry_count; i++) {
1327                 if (t->entries[i]->versions[v].mode)
1328                         maxlen += t->entries[i]->name->str_len + 34;
1329         }
1330
1331         strbuf_reset(b);
1332         strbuf_grow(b, maxlen);
1333         for (i = 0; i < t->entry_count; i++) {
1334                 struct tree_entry *e = t->entries[i];
1335                 if (!e->versions[v].mode)
1336                         continue;
1337                 strbuf_addf(b, "%o %s%c",
1338                         (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1339                         e->name->str_dat, '\0');
1340                 strbuf_add(b, e->versions[v].oid.hash, the_hash_algo->rawsz);
1341         }
1342 }
1343
1344 static void store_tree(struct tree_entry *root)
1345 {
1346         struct tree_content *t;
1347         unsigned int i, j, del;
1348         struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1349         struct object_entry *le = NULL;
1350
1351         if (!is_null_oid(&root->versions[1].oid))
1352                 return;
1353
1354         if (!root->tree)
1355                 load_tree(root);
1356         t = root->tree;
1357
1358         for (i = 0; i < t->entry_count; i++) {
1359                 if (t->entries[i]->tree)
1360                         store_tree(t->entries[i]);
1361         }
1362
1363         if (!(root->versions[0].mode & NO_DELTA))
1364                 le = find_object(&root->versions[0].oid);
1365         if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1366                 mktree(t, 0, &old_tree);
1367                 lo.data = old_tree;
1368                 lo.offset = le->idx.offset;
1369                 lo.depth = t->delta_depth;
1370         }
1371
1372         mktree(t, 1, &new_tree);
1373         store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1374
1375         t->delta_depth = lo.depth;
1376         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1377                 struct tree_entry *e = t->entries[i];
1378                 if (e->versions[1].mode) {
1379                         e->versions[0].mode = e->versions[1].mode;
1380                         oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1381                         t->entries[j++] = e;
1382                 } else {
1383                         release_tree_entry(e);
1384                         del++;
1385                 }
1386         }
1387         t->entry_count -= del;
1388 }
1389
1390 static void tree_content_replace(
1391         struct tree_entry *root,
1392         const struct object_id *oid,
1393         const uint16_t mode,
1394         struct tree_content *newtree)
1395 {
1396         if (!S_ISDIR(mode))
1397                 die("Root cannot be a non-directory");
1398         oidclr(&root->versions[0].oid);
1399         oidcpy(&root->versions[1].oid, oid);
1400         if (root->tree)
1401                 release_tree_content_recursive(root->tree);
1402         root->tree = newtree;
1403 }
1404
1405 static int tree_content_set(
1406         struct tree_entry *root,
1407         const char *p,
1408         const struct object_id *oid,
1409         const uint16_t mode,
1410         struct tree_content *subtree)
1411 {
1412         struct tree_content *t;
1413         const char *slash1;
1414         unsigned int i, n;
1415         struct tree_entry *e;
1416
1417         slash1 = strchrnul(p, '/');
1418         n = slash1 - p;
1419         if (!n)
1420                 die("Empty path component found in input");
1421         if (!*slash1 && !S_ISDIR(mode) && subtree)
1422                 die("Non-directories cannot have subtrees");
1423
1424         if (!root->tree)
1425                 load_tree(root);
1426         t = root->tree;
1427         for (i = 0; i < t->entry_count; i++) {
1428                 e = t->entries[i];
1429                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1430                         if (!*slash1) {
1431                                 if (!S_ISDIR(mode)
1432                                                 && e->versions[1].mode == mode
1433                                                 && oideq(&e->versions[1].oid, oid))
1434                                         return 0;
1435                                 e->versions[1].mode = mode;
1436                                 oidcpy(&e->versions[1].oid, oid);
1437                                 if (e->tree)
1438                                         release_tree_content_recursive(e->tree);
1439                                 e->tree = subtree;
1440
1441                                 /*
1442                                  * We need to leave e->versions[0].sha1 alone
1443                                  * to avoid modifying the preimage tree used
1444                                  * when writing out the parent directory.
1445                                  * But after replacing the subdir with a
1446                                  * completely different one, it's not a good
1447                                  * delta base any more, and besides, we've
1448                                  * thrown away the tree entries needed to
1449                                  * make a delta against it.
1450                                  *
1451                                  * So let's just explicitly disable deltas
1452                                  * for the subtree.
1453                                  */
1454                                 if (S_ISDIR(e->versions[0].mode))
1455                                         e->versions[0].mode |= NO_DELTA;
1456
1457                                 oidclr(&root->versions[1].oid);
1458                                 return 1;
1459                         }
1460                         if (!S_ISDIR(e->versions[1].mode)) {
1461                                 e->tree = new_tree_content(8);
1462                                 e->versions[1].mode = S_IFDIR;
1463                         }
1464                         if (!e->tree)
1465                                 load_tree(e);
1466                         if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1467                                 oidclr(&root->versions[1].oid);
1468                                 return 1;
1469                         }
1470                         return 0;
1471                 }
1472         }
1473
1474         if (t->entry_count == t->entry_capacity)
1475                 root->tree = t = grow_tree_content(t, t->entry_count);
1476         e = new_tree_entry();
1477         e->name = to_atom(p, n);
1478         e->versions[0].mode = 0;
1479         oidclr(&e->versions[0].oid);
1480         t->entries[t->entry_count++] = e;
1481         if (*slash1) {
1482                 e->tree = new_tree_content(8);
1483                 e->versions[1].mode = S_IFDIR;
1484                 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1485         } else {
1486                 e->tree = subtree;
1487                 e->versions[1].mode = mode;
1488                 oidcpy(&e->versions[1].oid, oid);
1489         }
1490         oidclr(&root->versions[1].oid);
1491         return 1;
1492 }
1493
1494 static int tree_content_remove(
1495         struct tree_entry *root,
1496         const char *p,
1497         struct tree_entry *backup_leaf,
1498         int allow_root)
1499 {
1500         struct tree_content *t;
1501         const char *slash1;
1502         unsigned int i, n;
1503         struct tree_entry *e;
1504
1505         slash1 = strchrnul(p, '/');
1506         n = slash1 - p;
1507
1508         if (!root->tree)
1509                 load_tree(root);
1510
1511         if (!*p && allow_root) {
1512                 e = root;
1513                 goto del_entry;
1514         }
1515
1516         t = root->tree;
1517         for (i = 0; i < t->entry_count; i++) {
1518                 e = t->entries[i];
1519                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1520                         if (*slash1 && !S_ISDIR(e->versions[1].mode))
1521                                 /*
1522                                  * If p names a file in some subdirectory, and a
1523                                  * file or symlink matching the name of the
1524                                  * parent directory of p exists, then p cannot
1525                                  * exist and need not be deleted.
1526                                  */
1527                                 return 1;
1528                         if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1529                                 goto del_entry;
1530                         if (!e->tree)
1531                                 load_tree(e);
1532                         if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1533                                 for (n = 0; n < e->tree->entry_count; n++) {
1534                                         if (e->tree->entries[n]->versions[1].mode) {
1535                                                 oidclr(&root->versions[1].oid);
1536                                                 return 1;
1537                                         }
1538                                 }
1539                                 backup_leaf = NULL;
1540                                 goto del_entry;
1541                         }
1542                         return 0;
1543                 }
1544         }
1545         return 0;
1546
1547 del_entry:
1548         if (backup_leaf)
1549                 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1550         else if (e->tree)
1551                 release_tree_content_recursive(e->tree);
1552         e->tree = NULL;
1553         e->versions[1].mode = 0;
1554         oidclr(&e->versions[1].oid);
1555         oidclr(&root->versions[1].oid);
1556         return 1;
1557 }
1558
1559 static int tree_content_get(
1560         struct tree_entry *root,
1561         const char *p,
1562         struct tree_entry *leaf,
1563         int allow_root)
1564 {
1565         struct tree_content *t;
1566         const char *slash1;
1567         unsigned int i, n;
1568         struct tree_entry *e;
1569
1570         slash1 = strchrnul(p, '/');
1571         n = slash1 - p;
1572         if (!n && !allow_root)
1573                 die("Empty path component found in input");
1574
1575         if (!root->tree)
1576                 load_tree(root);
1577
1578         if (!n) {
1579                 e = root;
1580                 goto found_entry;
1581         }
1582
1583         t = root->tree;
1584         for (i = 0; i < t->entry_count; i++) {
1585                 e = t->entries[i];
1586                 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1587                         if (!*slash1)
1588                                 goto found_entry;
1589                         if (!S_ISDIR(e->versions[1].mode))
1590                                 return 0;
1591                         if (!e->tree)
1592                                 load_tree(e);
1593                         return tree_content_get(e, slash1 + 1, leaf, 0);
1594                 }
1595         }
1596         return 0;
1597
1598 found_entry:
1599         memcpy(leaf, e, sizeof(*leaf));
1600         if (e->tree && is_null_oid(&e->versions[1].oid))
1601                 leaf->tree = dup_tree_content(e->tree);
1602         else
1603                 leaf->tree = NULL;
1604         return 1;
1605 }
1606
1607 static int update_branch(struct branch *b)
1608 {
1609         static const char *msg = "fast-import";
1610         struct ref_transaction *transaction;
1611         struct object_id old_oid;
1612         struct strbuf err = STRBUF_INIT;
1613
1614         if (is_null_oid(&b->oid)) {
1615                 if (b->delete)
1616                         delete_ref(NULL, b->name, NULL, 0);
1617                 return 0;
1618         }
1619         if (read_ref(b->name, &old_oid))
1620                 oidclr(&old_oid);
1621         if (!force_update && !is_null_oid(&old_oid)) {
1622                 struct commit *old_cmit, *new_cmit;
1623
1624                 old_cmit = lookup_commit_reference_gently(the_repository,
1625                                                           &old_oid, 0);
1626                 new_cmit = lookup_commit_reference_gently(the_repository,
1627                                                           &b->oid, 0);
1628                 if (!old_cmit || !new_cmit)
1629                         return error("Branch %s is missing commits.", b->name);
1630
1631                 if (!in_merge_bases(old_cmit, new_cmit)) {
1632                         warning("Not updating %s"
1633                                 " (new tip %s does not contain %s)",
1634                                 b->name, oid_to_hex(&b->oid),
1635                                 oid_to_hex(&old_oid));
1636                         return -1;
1637                 }
1638         }
1639         transaction = ref_transaction_begin(&err);
1640         if (!transaction ||
1641             ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1642                                    0, msg, &err) ||
1643             ref_transaction_commit(transaction, &err)) {
1644                 ref_transaction_free(transaction);
1645                 error("%s", err.buf);
1646                 strbuf_release(&err);
1647                 return -1;
1648         }
1649         ref_transaction_free(transaction);
1650         strbuf_release(&err);
1651         return 0;
1652 }
1653
1654 static void dump_branches(void)
1655 {
1656         unsigned int i;
1657         struct branch *b;
1658
1659         for (i = 0; i < branch_table_sz; i++) {
1660                 for (b = branch_table[i]; b; b = b->table_next_branch)
1661                         failure |= update_branch(b);
1662         }
1663 }
1664
1665 static void dump_tags(void)
1666 {
1667         static const char *msg = "fast-import";
1668         struct tag *t;
1669         struct strbuf ref_name = STRBUF_INIT;
1670         struct strbuf err = STRBUF_INIT;
1671         struct ref_transaction *transaction;
1672
1673         transaction = ref_transaction_begin(&err);
1674         if (!transaction) {
1675                 failure |= error("%s", err.buf);
1676                 goto cleanup;
1677         }
1678         for (t = first_tag; t; t = t->next_tag) {
1679                 strbuf_reset(&ref_name);
1680                 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1681
1682                 if (ref_transaction_update(transaction, ref_name.buf,
1683                                            &t->oid, NULL, 0, msg, &err)) {
1684                         failure |= error("%s", err.buf);
1685                         goto cleanup;
1686                 }
1687         }
1688         if (ref_transaction_commit(transaction, &err))
1689                 failure |= error("%s", err.buf);
1690
1691  cleanup:
1692         ref_transaction_free(transaction);
1693         strbuf_release(&ref_name);
1694         strbuf_release(&err);
1695 }
1696
1697 static void dump_marks(void)
1698 {
1699         struct lock_file mark_lock = LOCK_INIT;
1700         FILE *f;
1701
1702         if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1703                 return;
1704
1705         if (safe_create_leading_directories_const(export_marks_file)) {
1706                 failure |= error_errno("unable to create leading directories of %s",
1707                                        export_marks_file);
1708                 return;
1709         }
1710
1711         if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1712                 failure |= error_errno("Unable to write marks file %s",
1713                                        export_marks_file);
1714                 return;
1715         }
1716
1717         f = fdopen_lock_file(&mark_lock, "w");
1718         if (!f) {
1719                 int saved_errno = errno;
1720                 rollback_lock_file(&mark_lock);
1721                 failure |= error("Unable to write marks file %s: %s",
1722                         export_marks_file, strerror(saved_errno));
1723                 return;
1724         }
1725
1726         for_each_mark(marks, 0, dump_marks_fn, f);
1727         if (commit_lock_file(&mark_lock)) {
1728                 failure |= error_errno("Unable to write file %s",
1729                                        export_marks_file);
1730                 return;
1731         }
1732 }
1733
1734 static void insert_object_entry(struct mark_set *s, struct object_id *oid, uintmax_t mark)
1735 {
1736         struct object_entry *e;
1737         e = find_object(oid);
1738         if (!e) {
1739                 enum object_type type = oid_object_info(the_repository,
1740                                                         oid, NULL);
1741                 if (type < 0)
1742                         die("object not found: %s", oid_to_hex(oid));
1743                 e = insert_object(oid);
1744                 e->type = type;
1745                 e->pack_id = MAX_PACK_ID;
1746                 e->idx.offset = 1; /* just not zero! */
1747         }
1748         insert_mark(s, mark, e);
1749 }
1750
1751 static void insert_oid_entry(struct mark_set *s, struct object_id *oid, uintmax_t mark)
1752 {
1753         insert_mark(s, mark, xmemdupz(oid, sizeof(*oid)));
1754 }
1755
1756 static void read_mark_file(struct mark_set *s, FILE *f, mark_set_inserter_t inserter)
1757 {
1758         char line[512];
1759         while (fgets(line, sizeof(line), f)) {
1760                 uintmax_t mark;
1761                 char *end;
1762                 struct object_id oid;
1763
1764                 /* Ensure SHA-1 objects are padded with zeros. */
1765                 memset(oid.hash, 0, sizeof(oid.hash));
1766
1767                 end = strchr(line, '\n');
1768                 if (line[0] != ':' || !end)
1769                         die("corrupt mark line: %s", line);
1770                 *end = 0;
1771                 mark = strtoumax(line + 1, &end, 10);
1772                 if (!mark || end == line + 1
1773                         || *end != ' '
1774                         || get_oid_hex_any(end + 1, &oid) == GIT_HASH_UNKNOWN)
1775                         die("corrupt mark line: %s", line);
1776                 inserter(s, &oid, mark);
1777         }
1778 }
1779
1780 static void read_marks(void)
1781 {
1782         FILE *f = fopen(import_marks_file, "r");
1783         if (f)
1784                 ;
1785         else if (import_marks_file_ignore_missing && errno == ENOENT)
1786                 goto done; /* Marks file does not exist */
1787         else
1788                 die_errno("cannot read '%s'", import_marks_file);
1789         read_mark_file(marks, f, insert_object_entry);
1790         fclose(f);
1791 done:
1792         import_marks_file_done = 1;
1793 }
1794
1795
1796 static int read_next_command(void)
1797 {
1798         static int stdin_eof = 0;
1799
1800         if (stdin_eof) {
1801                 unread_command_buf = 0;
1802                 return EOF;
1803         }
1804
1805         for (;;) {
1806                 if (unread_command_buf) {
1807                         unread_command_buf = 0;
1808                 } else {
1809                         struct recent_command *rc;
1810
1811                         stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1812                         if (stdin_eof)
1813                                 return EOF;
1814
1815                         if (!seen_data_command
1816                                 && !starts_with(command_buf.buf, "feature ")
1817                                 && !starts_with(command_buf.buf, "option ")) {
1818                                 parse_argv();
1819                         }
1820
1821                         rc = rc_free;
1822                         if (rc)
1823                                 rc_free = rc->next;
1824                         else {
1825                                 rc = cmd_hist.next;
1826                                 cmd_hist.next = rc->next;
1827                                 cmd_hist.next->prev = &cmd_hist;
1828                                 free(rc->buf);
1829                         }
1830
1831                         rc->buf = xstrdup(command_buf.buf);
1832                         rc->prev = cmd_tail;
1833                         rc->next = cmd_hist.prev;
1834                         rc->prev->next = rc;
1835                         cmd_tail = rc;
1836                 }
1837                 if (command_buf.buf[0] == '#')
1838                         continue;
1839                 return 0;
1840         }
1841 }
1842
1843 static void skip_optional_lf(void)
1844 {
1845         int term_char = fgetc(stdin);
1846         if (term_char != '\n' && term_char != EOF)
1847                 ungetc(term_char, stdin);
1848 }
1849
1850 static void parse_mark(void)
1851 {
1852         const char *v;
1853         if (skip_prefix(command_buf.buf, "mark :", &v)) {
1854                 next_mark = strtoumax(v, NULL, 10);
1855                 read_next_command();
1856         }
1857         else
1858                 next_mark = 0;
1859 }
1860
1861 static void parse_original_identifier(void)
1862 {
1863         const char *v;
1864         if (skip_prefix(command_buf.buf, "original-oid ", &v))
1865                 read_next_command();
1866 }
1867
1868 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1869 {
1870         const char *data;
1871         strbuf_reset(sb);
1872
1873         if (!skip_prefix(command_buf.buf, "data ", &data))
1874                 die("Expected 'data n' command, found: %s", command_buf.buf);
1875
1876         if (skip_prefix(data, "<<", &data)) {
1877                 char *term = xstrdup(data);
1878                 size_t term_len = command_buf.len - (data - command_buf.buf);
1879
1880                 for (;;) {
1881                         if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1882                                 die("EOF in data (terminator '%s' not found)", term);
1883                         if (term_len == command_buf.len
1884                                 && !strcmp(term, command_buf.buf))
1885                                 break;
1886                         strbuf_addbuf(sb, &command_buf);
1887                         strbuf_addch(sb, '\n');
1888                 }
1889                 free(term);
1890         }
1891         else {
1892                 uintmax_t len = strtoumax(data, NULL, 10);
1893                 size_t n = 0, length = (size_t)len;
1894
1895                 if (limit && limit < len) {
1896                         *len_res = len;
1897                         return 0;
1898                 }
1899                 if (length < len)
1900                         die("data is too large to use in this context");
1901
1902                 while (n < length) {
1903                         size_t s = strbuf_fread(sb, length - n, stdin);
1904                         if (!s && feof(stdin))
1905                                 die("EOF in data (%lu bytes remaining)",
1906                                         (unsigned long)(length - n));
1907                         n += s;
1908                 }
1909         }
1910
1911         skip_optional_lf();
1912         return 1;
1913 }
1914
1915 static int validate_raw_date(const char *src, struct strbuf *result, int strict)
1916 {
1917         const char *orig_src = src;
1918         char *endp;
1919         unsigned long num;
1920
1921         errno = 0;
1922
1923         num = strtoul(src, &endp, 10);
1924         /*
1925          * NEEDSWORK: perhaps check for reasonable values? For example, we
1926          *            could error on values representing times more than a
1927          *            day in the future.
1928          */
1929         if (errno || endp == src || *endp != ' ')
1930                 return -1;
1931
1932         src = endp + 1;
1933         if (*src != '-' && *src != '+')
1934                 return -1;
1935
1936         num = strtoul(src + 1, &endp, 10);
1937         /*
1938          * NEEDSWORK: check for brokenness other than num > 1400, such as
1939          *            (num % 100) >= 60, or ((num % 100) % 15) != 0 ?
1940          */
1941         if (errno || endp == src + 1 || *endp || /* did not parse */
1942             (strict && (1400 < num))             /* parsed a broken timezone */
1943            )
1944                 return -1;
1945
1946         strbuf_addstr(result, orig_src);
1947         return 0;
1948 }
1949
1950 static char *parse_ident(const char *buf)
1951 {
1952         const char *ltgt;
1953         size_t name_len;
1954         struct strbuf ident = STRBUF_INIT;
1955
1956         /* ensure there is a space delimiter even if there is no name */
1957         if (*buf == '<')
1958                 --buf;
1959
1960         ltgt = buf + strcspn(buf, "<>");
1961         if (*ltgt != '<')
1962                 die("Missing < in ident string: %s", buf);
1963         if (ltgt != buf && ltgt[-1] != ' ')
1964                 die("Missing space before < in ident string: %s", buf);
1965         ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
1966         if (*ltgt != '>')
1967                 die("Missing > in ident string: %s", buf);
1968         ltgt++;
1969         if (*ltgt != ' ')
1970                 die("Missing space after > in ident string: %s", buf);
1971         ltgt++;
1972         name_len = ltgt - buf;
1973         strbuf_add(&ident, buf, name_len);
1974
1975         switch (whenspec) {
1976         case WHENSPEC_RAW:
1977                 if (validate_raw_date(ltgt, &ident, 1) < 0)
1978                         die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1979                 break;
1980         case WHENSPEC_RAW_PERMISSIVE:
1981                 if (validate_raw_date(ltgt, &ident, 0) < 0)
1982                         die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1983                 break;
1984         case WHENSPEC_RFC2822:
1985                 if (parse_date(ltgt, &ident) < 0)
1986                         die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
1987                 break;
1988         case WHENSPEC_NOW:
1989                 if (strcmp("now", ltgt))
1990                         die("Date in ident must be 'now': %s", buf);
1991                 datestamp(&ident);
1992                 break;
1993         }
1994
1995         return strbuf_detach(&ident, NULL);
1996 }
1997
1998 static void parse_and_store_blob(
1999         struct last_object *last,
2000         struct object_id *oidout,
2001         uintmax_t mark)
2002 {
2003         static struct strbuf buf = STRBUF_INIT;
2004         uintmax_t len;
2005
2006         if (parse_data(&buf, big_file_threshold, &len))
2007                 store_object(OBJ_BLOB, &buf, last, oidout, mark);
2008         else {
2009                 if (last) {
2010                         strbuf_release(&last->data);
2011                         last->offset = 0;
2012                         last->depth = 0;
2013                 }
2014                 stream_blob(len, oidout, mark);
2015                 skip_optional_lf();
2016         }
2017 }
2018
2019 static void parse_new_blob(void)
2020 {
2021         read_next_command();
2022         parse_mark();
2023         parse_original_identifier();
2024         parse_and_store_blob(&last_blob, NULL, next_mark);
2025 }
2026
2027 static void unload_one_branch(void)
2028 {
2029         while (cur_active_branches
2030                 && cur_active_branches >= max_active_branches) {
2031                 uintmax_t min_commit = ULONG_MAX;
2032                 struct branch *e, *l = NULL, *p = NULL;
2033
2034                 for (e = active_branches; e; e = e->active_next_branch) {
2035                         if (e->last_commit < min_commit) {
2036                                 p = l;
2037                                 min_commit = e->last_commit;
2038                         }
2039                         l = e;
2040                 }
2041
2042                 if (p) {
2043                         e = p->active_next_branch;
2044                         p->active_next_branch = e->active_next_branch;
2045                 } else {
2046                         e = active_branches;
2047                         active_branches = e->active_next_branch;
2048                 }
2049                 e->active = 0;
2050                 e->active_next_branch = NULL;
2051                 if (e->branch_tree.tree) {
2052                         release_tree_content_recursive(e->branch_tree.tree);
2053                         e->branch_tree.tree = NULL;
2054                 }
2055                 cur_active_branches--;
2056         }
2057 }
2058
2059 static void load_branch(struct branch *b)
2060 {
2061         load_tree(&b->branch_tree);
2062         if (!b->active) {
2063                 b->active = 1;
2064                 b->active_next_branch = active_branches;
2065                 active_branches = b;
2066                 cur_active_branches++;
2067                 branch_load_count++;
2068         }
2069 }
2070
2071 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2072 {
2073         unsigned char fanout = 0;
2074         while ((num_notes >>= 8))
2075                 fanout++;
2076         return fanout;
2077 }
2078
2079 static void construct_path_with_fanout(const char *hex_sha1,
2080                 unsigned char fanout, char *path)
2081 {
2082         unsigned int i = 0, j = 0;
2083         if (fanout >= the_hash_algo->rawsz)
2084                 die("Too large fanout (%u)", fanout);
2085         while (fanout) {
2086                 path[i++] = hex_sha1[j++];
2087                 path[i++] = hex_sha1[j++];
2088                 path[i++] = '/';
2089                 fanout--;
2090         }
2091         memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2092         path[i + the_hash_algo->hexsz - j] = '\0';
2093 }
2094
2095 static uintmax_t do_change_note_fanout(
2096                 struct tree_entry *orig_root, struct tree_entry *root,
2097                 char *hex_oid, unsigned int hex_oid_len,
2098                 char *fullpath, unsigned int fullpath_len,
2099                 unsigned char fanout)
2100 {
2101         struct tree_content *t;
2102         struct tree_entry *e, leaf;
2103         unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2104         uintmax_t num_notes = 0;
2105         struct object_id oid;
2106         /* hex oid + '/' between each pair of hex digits + NUL */
2107         char realpath[GIT_MAX_HEXSZ + ((GIT_MAX_HEXSZ / 2) - 1) + 1];
2108         const unsigned hexsz = the_hash_algo->hexsz;
2109
2110         if (!root->tree)
2111                 load_tree(root);
2112         t = root->tree;
2113
2114         for (i = 0; t && i < t->entry_count; i++) {
2115                 e = t->entries[i];
2116                 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2117                 tmp_fullpath_len = fullpath_len;
2118
2119                 /*
2120                  * We're interested in EITHER existing note entries (entries
2121                  * with exactly 40 hex chars in path, not including directory
2122                  * separators), OR directory entries that may contain note
2123                  * entries (with < 40 hex chars in path).
2124                  * Also, each path component in a note entry must be a multiple
2125                  * of 2 chars.
2126                  */
2127                 if (!e->versions[1].mode ||
2128                     tmp_hex_oid_len > hexsz ||
2129                     e->name->str_len % 2)
2130                         continue;
2131
2132                 /* This _may_ be a note entry, or a subdir containing notes */
2133                 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2134                        e->name->str_len);
2135                 if (tmp_fullpath_len)
2136                         fullpath[tmp_fullpath_len++] = '/';
2137                 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2138                        e->name->str_len);
2139                 tmp_fullpath_len += e->name->str_len;
2140                 fullpath[tmp_fullpath_len] = '\0';
2141
2142                 if (tmp_hex_oid_len == hexsz && !get_oid_hex(hex_oid, &oid)) {
2143                         /* This is a note entry */
2144                         if (fanout == 0xff) {
2145                                 /* Counting mode, no rename */
2146                                 num_notes++;
2147                                 continue;
2148                         }
2149                         construct_path_with_fanout(hex_oid, fanout, realpath);
2150                         if (!strcmp(fullpath, realpath)) {
2151                                 /* Note entry is in correct location */
2152                                 num_notes++;
2153                                 continue;
2154                         }
2155
2156                         /* Rename fullpath to realpath */
2157                         if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2158                                 die("Failed to remove path %s", fullpath);
2159                         tree_content_set(orig_root, realpath,
2160                                 &leaf.versions[1].oid,
2161                                 leaf.versions[1].mode,
2162                                 leaf.tree);
2163                 } else if (S_ISDIR(e->versions[1].mode)) {
2164                         /* This is a subdir that may contain note entries */
2165                         num_notes += do_change_note_fanout(orig_root, e,
2166                                 hex_oid, tmp_hex_oid_len,
2167                                 fullpath, tmp_fullpath_len, fanout);
2168                 }
2169
2170                 /* The above may have reallocated the current tree_content */
2171                 t = root->tree;
2172         }
2173         return num_notes;
2174 }
2175
2176 static uintmax_t change_note_fanout(struct tree_entry *root,
2177                 unsigned char fanout)
2178 {
2179         /*
2180          * The size of path is due to one slash between every two hex digits,
2181          * plus the terminating NUL.  Note that there is no slash at the end, so
2182          * the number of slashes is one less than half the number of hex
2183          * characters.
2184          */
2185         char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2186         return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2187 }
2188
2189 static int parse_mapped_oid_hex(const char *hex, struct object_id *oid, const char **end)
2190 {
2191         int algo;
2192         khiter_t it;
2193
2194         /* Make SHA-1 object IDs have all-zero padding. */
2195         memset(oid->hash, 0, sizeof(oid->hash));
2196
2197         algo = parse_oid_hex_any(hex, oid, end);
2198         if (algo == GIT_HASH_UNKNOWN)
2199                 return -1;
2200
2201         it = kh_get_oid_map(sub_oid_map, *oid);
2202         /* No such object? */
2203         if (it == kh_end(sub_oid_map)) {
2204                 /* If we're using the same algorithm, pass it through. */
2205                 if (hash_algos[algo].format_id == the_hash_algo->format_id)
2206                         return 0;
2207                 return -1;
2208         }
2209         oidcpy(oid, kh_value(sub_oid_map, it));
2210         return 0;
2211 }
2212
2213 /*
2214  * Given a pointer into a string, parse a mark reference:
2215  *
2216  *   idnum ::= ':' bigint;
2217  *
2218  * Return the first character after the value in *endptr.
2219  *
2220  * Complain if the following character is not what is expected,
2221  * either a space or end of the string.
2222  */
2223 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2224 {
2225         uintmax_t mark;
2226
2227         assert(*p == ':');
2228         p++;
2229         mark = strtoumax(p, endptr, 10);
2230         if (*endptr == p)
2231                 die("No value after ':' in mark: %s", command_buf.buf);
2232         return mark;
2233 }
2234
2235 /*
2236  * Parse the mark reference, and complain if this is not the end of
2237  * the string.
2238  */
2239 static uintmax_t parse_mark_ref_eol(const char *p)
2240 {
2241         char *end;
2242         uintmax_t mark;
2243
2244         mark = parse_mark_ref(p, &end);
2245         if (*end != '\0')
2246                 die("Garbage after mark: %s", command_buf.buf);
2247         return mark;
2248 }
2249
2250 /*
2251  * Parse the mark reference, demanding a trailing space.  Return a
2252  * pointer to the space.
2253  */
2254 static uintmax_t parse_mark_ref_space(const char **p)
2255 {
2256         uintmax_t mark;
2257         char *end;
2258
2259         mark = parse_mark_ref(*p, &end);
2260         if (*end++ != ' ')
2261                 die("Missing space after mark: %s", command_buf.buf);
2262         *p = end;
2263         return mark;
2264 }
2265
2266 static void file_change_m(const char *p, struct branch *b)
2267 {
2268         static struct strbuf uq = STRBUF_INIT;
2269         const char *endp;
2270         struct object_entry *oe;
2271         struct object_id oid;
2272         uint16_t mode, inline_data = 0;
2273
2274         p = get_mode(p, &mode);
2275         if (!p)
2276                 die("Corrupt mode: %s", command_buf.buf);
2277         switch (mode) {
2278         case 0644:
2279         case 0755:
2280                 mode |= S_IFREG;
2281         case S_IFREG | 0644:
2282         case S_IFREG | 0755:
2283         case S_IFLNK:
2284         case S_IFDIR:
2285         case S_IFGITLINK:
2286                 /* ok */
2287                 break;
2288         default:
2289                 die("Corrupt mode: %s", command_buf.buf);
2290         }
2291
2292         if (*p == ':') {
2293                 oe = find_mark(marks, parse_mark_ref_space(&p));
2294                 oidcpy(&oid, &oe->idx.oid);
2295         } else if (skip_prefix(p, "inline ", &p)) {
2296                 inline_data = 1;
2297                 oe = NULL; /* not used with inline_data, but makes gcc happy */
2298         } else {
2299                 if (parse_mapped_oid_hex(p, &oid, &p))
2300                         die("Invalid dataref: %s", command_buf.buf);
2301                 oe = find_object(&oid);
2302                 if (*p++ != ' ')
2303                         die("Missing space after SHA1: %s", command_buf.buf);
2304         }
2305
2306         strbuf_reset(&uq);
2307         if (!unquote_c_style(&uq, p, &endp)) {
2308                 if (*endp)
2309                         die("Garbage after path in: %s", command_buf.buf);
2310                 p = uq.buf;
2311         }
2312
2313         /* Git does not track empty, non-toplevel directories. */
2314         if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2315                 tree_content_remove(&b->branch_tree, p, NULL, 0);
2316                 return;
2317         }
2318
2319         if (S_ISGITLINK(mode)) {
2320                 if (inline_data)
2321                         die("Git links cannot be specified 'inline': %s",
2322                                 command_buf.buf);
2323                 else if (oe) {
2324                         if (oe->type != OBJ_COMMIT)
2325                                 die("Not a commit (actually a %s): %s",
2326                                         type_name(oe->type), command_buf.buf);
2327                 }
2328                 /*
2329                  * Accept the sha1 without checking; it expected to be in
2330                  * another repository.
2331                  */
2332         } else if (inline_data) {
2333                 if (S_ISDIR(mode))
2334                         die("Directories cannot be specified 'inline': %s",
2335                                 command_buf.buf);
2336                 if (p != uq.buf) {
2337                         strbuf_addstr(&uq, p);
2338                         p = uq.buf;
2339                 }
2340                 while (read_next_command() != EOF) {
2341                         const char *v;
2342                         if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2343                                 parse_cat_blob(v);
2344                         else {
2345                                 parse_and_store_blob(&last_blob, &oid, 0);
2346                                 break;
2347                         }
2348                 }
2349         } else {
2350                 enum object_type expected = S_ISDIR(mode) ?
2351                                                 OBJ_TREE: OBJ_BLOB;
2352                 enum object_type type = oe ? oe->type :
2353                                         oid_object_info(the_repository, &oid,
2354                                                         NULL);
2355                 if (type < 0)
2356                         die("%s not found: %s",
2357                                         S_ISDIR(mode) ?  "Tree" : "Blob",
2358                                         command_buf.buf);
2359                 if (type != expected)
2360                         die("Not a %s (actually a %s): %s",
2361                                 type_name(expected), type_name(type),
2362                                 command_buf.buf);
2363         }
2364
2365         if (!*p) {
2366                 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2367                 return;
2368         }
2369         tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2370 }
2371
2372 static void file_change_d(const char *p, struct branch *b)
2373 {
2374         static struct strbuf uq = STRBUF_INIT;
2375         const char *endp;
2376
2377         strbuf_reset(&uq);
2378         if (!unquote_c_style(&uq, p, &endp)) {
2379                 if (*endp)
2380                         die("Garbage after path in: %s", command_buf.buf);
2381                 p = uq.buf;
2382         }
2383         tree_content_remove(&b->branch_tree, p, NULL, 1);
2384 }
2385
2386 static void file_change_cr(const char *s, struct branch *b, int rename)
2387 {
2388         const char *d;
2389         static struct strbuf s_uq = STRBUF_INIT;
2390         static struct strbuf d_uq = STRBUF_INIT;
2391         const char *endp;
2392         struct tree_entry leaf;
2393
2394         strbuf_reset(&s_uq);
2395         if (!unquote_c_style(&s_uq, s, &endp)) {
2396                 if (*endp != ' ')
2397                         die("Missing space after source: %s", command_buf.buf);
2398         } else {
2399                 endp = strchr(s, ' ');
2400                 if (!endp)
2401                         die("Missing space after source: %s", command_buf.buf);
2402                 strbuf_add(&s_uq, s, endp - s);
2403         }
2404         s = s_uq.buf;
2405
2406         endp++;
2407         if (!*endp)
2408                 die("Missing dest: %s", command_buf.buf);
2409
2410         d = endp;
2411         strbuf_reset(&d_uq);
2412         if (!unquote_c_style(&d_uq, d, &endp)) {
2413                 if (*endp)
2414                         die("Garbage after dest in: %s", command_buf.buf);
2415                 d = d_uq.buf;
2416         }
2417
2418         memset(&leaf, 0, sizeof(leaf));
2419         if (rename)
2420                 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2421         else
2422                 tree_content_get(&b->branch_tree, s, &leaf, 1);
2423         if (!leaf.versions[1].mode)
2424                 die("Path %s not in branch", s);
2425         if (!*d) {      /* C "path/to/subdir" "" */
2426                 tree_content_replace(&b->branch_tree,
2427                         &leaf.versions[1].oid,
2428                         leaf.versions[1].mode,
2429                         leaf.tree);
2430                 return;
2431         }
2432         tree_content_set(&b->branch_tree, d,
2433                 &leaf.versions[1].oid,
2434                 leaf.versions[1].mode,
2435                 leaf.tree);
2436 }
2437
2438 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2439 {
2440         static struct strbuf uq = STRBUF_INIT;
2441         struct object_entry *oe;
2442         struct branch *s;
2443         struct object_id oid, commit_oid;
2444         char path[GIT_MAX_RAWSZ * 3];
2445         uint16_t inline_data = 0;
2446         unsigned char new_fanout;
2447
2448         /*
2449          * When loading a branch, we don't traverse its tree to count the real
2450          * number of notes (too expensive to do this for all non-note refs).
2451          * This means that recently loaded notes refs might incorrectly have
2452          * b->num_notes == 0, and consequently, old_fanout might be wrong.
2453          *
2454          * Fix this by traversing the tree and counting the number of notes
2455          * when b->num_notes == 0. If the notes tree is truly empty, the
2456          * calculation should not take long.
2457          */
2458         if (b->num_notes == 0 && *old_fanout == 0) {
2459                 /* Invoke change_note_fanout() in "counting mode". */
2460                 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2461                 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2462         }
2463
2464         /* Now parse the notemodify command. */
2465         /* <dataref> or 'inline' */
2466         if (*p == ':') {
2467                 oe = find_mark(marks, parse_mark_ref_space(&p));
2468                 oidcpy(&oid, &oe->idx.oid);
2469         } else if (skip_prefix(p, "inline ", &p)) {
2470                 inline_data = 1;
2471                 oe = NULL; /* not used with inline_data, but makes gcc happy */
2472         } else {
2473                 if (parse_mapped_oid_hex(p, &oid, &p))
2474                         die("Invalid dataref: %s", command_buf.buf);
2475                 oe = find_object(&oid);
2476                 if (*p++ != ' ')
2477                         die("Missing space after SHA1: %s", command_buf.buf);
2478         }
2479
2480         /* <commit-ish> */
2481         s = lookup_branch(p);
2482         if (s) {
2483                 if (is_null_oid(&s->oid))
2484                         die("Can't add a note on empty branch.");
2485                 oidcpy(&commit_oid, &s->oid);
2486         } else if (*p == ':') {
2487                 uintmax_t commit_mark = parse_mark_ref_eol(p);
2488                 struct object_entry *commit_oe = find_mark(marks, commit_mark);
2489                 if (commit_oe->type != OBJ_COMMIT)
2490                         die("Mark :%" PRIuMAX " not a commit", commit_mark);
2491                 oidcpy(&commit_oid, &commit_oe->idx.oid);
2492         } else if (!get_oid(p, &commit_oid)) {
2493                 unsigned long size;
2494                 char *buf = read_object_with_reference(the_repository,
2495                                                        &commit_oid,
2496                                                        commit_type, &size,
2497                                                        &commit_oid);
2498                 if (!buf || size < the_hash_algo->hexsz + 6)
2499                         die("Not a valid commit: %s", p);
2500                 free(buf);
2501         } else
2502                 die("Invalid ref name or SHA1 expression: %s", p);
2503
2504         if (inline_data) {
2505                 if (p != uq.buf) {
2506                         strbuf_addstr(&uq, p);
2507                         p = uq.buf;
2508                 }
2509                 read_next_command();
2510                 parse_and_store_blob(&last_blob, &oid, 0);
2511         } else if (oe) {
2512                 if (oe->type != OBJ_BLOB)
2513                         die("Not a blob (actually a %s): %s",
2514                                 type_name(oe->type), command_buf.buf);
2515         } else if (!is_null_oid(&oid)) {
2516                 enum object_type type = oid_object_info(the_repository, &oid,
2517                                                         NULL);
2518                 if (type < 0)
2519                         die("Blob not found: %s", command_buf.buf);
2520                 if (type != OBJ_BLOB)
2521                         die("Not a blob (actually a %s): %s",
2522                             type_name(type), command_buf.buf);
2523         }
2524
2525         construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2526         if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2527                 b->num_notes--;
2528
2529         if (is_null_oid(&oid))
2530                 return; /* nothing to insert */
2531
2532         b->num_notes++;
2533         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2534         construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2535         tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2536 }
2537
2538 static void file_change_deleteall(struct branch *b)
2539 {
2540         release_tree_content_recursive(b->branch_tree.tree);
2541         oidclr(&b->branch_tree.versions[0].oid);
2542         oidclr(&b->branch_tree.versions[1].oid);
2543         load_tree(&b->branch_tree);
2544         b->num_notes = 0;
2545 }
2546
2547 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2548 {
2549         if (!buf || size < the_hash_algo->hexsz + 6)
2550                 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2551         if (memcmp("tree ", buf, 5)
2552                 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2553                 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2554         oidcpy(&b->branch_tree.versions[0].oid,
2555                &b->branch_tree.versions[1].oid);
2556 }
2557
2558 static void parse_from_existing(struct branch *b)
2559 {
2560         if (is_null_oid(&b->oid)) {
2561                 oidclr(&b->branch_tree.versions[0].oid);
2562                 oidclr(&b->branch_tree.versions[1].oid);
2563         } else {
2564                 unsigned long size;
2565                 char *buf;
2566
2567                 buf = read_object_with_reference(the_repository,
2568                                                  &b->oid, commit_type, &size,
2569                                                  &b->oid);
2570                 parse_from_commit(b, buf, size);
2571                 free(buf);
2572         }
2573 }
2574
2575 static int parse_objectish(struct branch *b, const char *objectish)
2576 {
2577         struct branch *s;
2578         struct object_id oid;
2579
2580         oidcpy(&oid, &b->branch_tree.versions[1].oid);
2581
2582         s = lookup_branch(objectish);
2583         if (b == s)
2584                 die("Can't create a branch from itself: %s", b->name);
2585         else if (s) {
2586                 struct object_id *t = &s->branch_tree.versions[1].oid;
2587                 oidcpy(&b->oid, &s->oid);
2588                 oidcpy(&b->branch_tree.versions[0].oid, t);
2589                 oidcpy(&b->branch_tree.versions[1].oid, t);
2590         } else if (*objectish == ':') {
2591                 uintmax_t idnum = parse_mark_ref_eol(objectish);
2592                 struct object_entry *oe = find_mark(marks, idnum);
2593                 if (oe->type != OBJ_COMMIT)
2594                         die("Mark :%" PRIuMAX " not a commit", idnum);
2595                 if (!oideq(&b->oid, &oe->idx.oid)) {
2596                         oidcpy(&b->oid, &oe->idx.oid);
2597                         if (oe->pack_id != MAX_PACK_ID) {
2598                                 unsigned long size;
2599                                 char *buf = gfi_unpack_entry(oe, &size);
2600                                 parse_from_commit(b, buf, size);
2601                                 free(buf);
2602                         } else
2603                                 parse_from_existing(b);
2604                 }
2605         } else if (!get_oid(objectish, &b->oid)) {
2606                 parse_from_existing(b);
2607                 if (is_null_oid(&b->oid))
2608                         b->delete = 1;
2609         }
2610         else
2611                 die("Invalid ref name or SHA1 expression: %s", objectish);
2612
2613         if (b->branch_tree.tree && !oideq(&oid, &b->branch_tree.versions[1].oid)) {
2614                 release_tree_content_recursive(b->branch_tree.tree);
2615                 b->branch_tree.tree = NULL;
2616         }
2617
2618         read_next_command();
2619         return 1;
2620 }
2621
2622 static int parse_from(struct branch *b)
2623 {
2624         const char *from;
2625
2626         if (!skip_prefix(command_buf.buf, "from ", &from))
2627                 return 0;
2628
2629         return parse_objectish(b, from);
2630 }
2631
2632 static int parse_objectish_with_prefix(struct branch *b, const char *prefix)
2633 {
2634         const char *base;
2635
2636         if (!skip_prefix(command_buf.buf, prefix, &base))
2637                 return 0;
2638
2639         return parse_objectish(b, base);
2640 }
2641
2642 static struct hash_list *parse_merge(unsigned int *count)
2643 {
2644         struct hash_list *list = NULL, **tail = &list, *n;
2645         const char *from;
2646         struct branch *s;
2647
2648         *count = 0;
2649         while (skip_prefix(command_buf.buf, "merge ", &from)) {
2650                 n = xmalloc(sizeof(*n));
2651                 s = lookup_branch(from);
2652                 if (s)
2653                         oidcpy(&n->oid, &s->oid);
2654                 else if (*from == ':') {
2655                         uintmax_t idnum = parse_mark_ref_eol(from);
2656                         struct object_entry *oe = find_mark(marks, idnum);
2657                         if (oe->type != OBJ_COMMIT)
2658                                 die("Mark :%" PRIuMAX " not a commit", idnum);
2659                         oidcpy(&n->oid, &oe->idx.oid);
2660                 } else if (!get_oid(from, &n->oid)) {
2661                         unsigned long size;
2662                         char *buf = read_object_with_reference(the_repository,
2663                                                                &n->oid,
2664                                                                commit_type,
2665                                                                &size, &n->oid);
2666                         if (!buf || size < the_hash_algo->hexsz + 6)
2667                                 die("Not a valid commit: %s", from);
2668                         free(buf);
2669                 } else
2670                         die("Invalid ref name or SHA1 expression: %s", from);
2671
2672                 n->next = NULL;
2673                 *tail = n;
2674                 tail = &n->next;
2675
2676                 (*count)++;
2677                 read_next_command();
2678         }
2679         return list;
2680 }
2681
2682 static void parse_new_commit(const char *arg)
2683 {
2684         static struct strbuf msg = STRBUF_INIT;
2685         struct branch *b;
2686         char *author = NULL;
2687         char *committer = NULL;
2688         char *encoding = NULL;
2689         struct hash_list *merge_list = NULL;
2690         unsigned int merge_count;
2691         unsigned char prev_fanout, new_fanout;
2692         const char *v;
2693
2694         b = lookup_branch(arg);
2695         if (!b)
2696                 b = new_branch(arg);
2697
2698         read_next_command();
2699         parse_mark();
2700         parse_original_identifier();
2701         if (skip_prefix(command_buf.buf, "author ", &v)) {
2702                 author = parse_ident(v);
2703                 read_next_command();
2704         }
2705         if (skip_prefix(command_buf.buf, "committer ", &v)) {
2706                 committer = parse_ident(v);
2707                 read_next_command();
2708         }
2709         if (!committer)
2710                 die("Expected committer but didn't get one");
2711         if (skip_prefix(command_buf.buf, "encoding ", &v)) {
2712                 encoding = xstrdup(v);
2713                 read_next_command();
2714         }
2715         parse_data(&msg, 0, NULL);
2716         read_next_command();
2717         parse_from(b);
2718         merge_list = parse_merge(&merge_count);
2719
2720         /* ensure the branch is active/loaded */
2721         if (!b->branch_tree.tree || !max_active_branches) {
2722                 unload_one_branch();
2723                 load_branch(b);
2724         }
2725
2726         prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2727
2728         /* file_change* */
2729         while (command_buf.len > 0) {
2730                 if (skip_prefix(command_buf.buf, "M ", &v))
2731                         file_change_m(v, b);
2732                 else if (skip_prefix(command_buf.buf, "D ", &v))
2733                         file_change_d(v, b);
2734                 else if (skip_prefix(command_buf.buf, "R ", &v))
2735                         file_change_cr(v, b, 1);
2736                 else if (skip_prefix(command_buf.buf, "C ", &v))
2737                         file_change_cr(v, b, 0);
2738                 else if (skip_prefix(command_buf.buf, "N ", &v))
2739                         note_change_n(v, b, &prev_fanout);
2740                 else if (!strcmp("deleteall", command_buf.buf))
2741                         file_change_deleteall(b);
2742                 else if (skip_prefix(command_buf.buf, "ls ", &v))
2743                         parse_ls(v, b);
2744                 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2745                         parse_cat_blob(v);
2746                 else {
2747                         unread_command_buf = 1;
2748                         break;
2749                 }
2750                 if (read_next_command() == EOF)
2751                         break;
2752         }
2753
2754         new_fanout = convert_num_notes_to_fanout(b->num_notes);
2755         if (new_fanout != prev_fanout)
2756                 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2757
2758         /* build the tree and the commit */
2759         store_tree(&b->branch_tree);
2760         oidcpy(&b->branch_tree.versions[0].oid,
2761                &b->branch_tree.versions[1].oid);
2762
2763         strbuf_reset(&new_data);
2764         strbuf_addf(&new_data, "tree %s\n",
2765                 oid_to_hex(&b->branch_tree.versions[1].oid));
2766         if (!is_null_oid(&b->oid))
2767                 strbuf_addf(&new_data, "parent %s\n",
2768                             oid_to_hex(&b->oid));
2769         while (merge_list) {
2770                 struct hash_list *next = merge_list->next;
2771                 strbuf_addf(&new_data, "parent %s\n",
2772                             oid_to_hex(&merge_list->oid));
2773                 free(merge_list);
2774                 merge_list = next;
2775         }
2776         strbuf_addf(&new_data,
2777                 "author %s\n"
2778                 "committer %s\n",
2779                 author ? author : committer, committer);
2780         if (encoding)
2781                 strbuf_addf(&new_data,
2782                         "encoding %s\n",
2783                         encoding);
2784         strbuf_addch(&new_data, '\n');
2785         strbuf_addbuf(&new_data, &msg);
2786         free(author);
2787         free(committer);
2788         free(encoding);
2789
2790         if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2791                 b->pack_id = pack_id;
2792         b->last_commit = object_count_by_type[OBJ_COMMIT];
2793 }
2794
2795 static void parse_new_tag(const char *arg)
2796 {
2797         static struct strbuf msg = STRBUF_INIT;
2798         const char *from;
2799         char *tagger;
2800         struct branch *s;
2801         struct tag *t;
2802         uintmax_t from_mark = 0;
2803         struct object_id oid;
2804         enum object_type type;
2805         const char *v;
2806
2807         t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2808         memset(t, 0, sizeof(struct tag));
2809         t->name = pool_strdup(arg);
2810         if (last_tag)
2811                 last_tag->next_tag = t;
2812         else
2813                 first_tag = t;
2814         last_tag = t;
2815         read_next_command();
2816         parse_mark();
2817
2818         /* from ... */
2819         if (!skip_prefix(command_buf.buf, "from ", &from))
2820                 die("Expected from command, got %s", command_buf.buf);
2821         s = lookup_branch(from);
2822         if (s) {
2823                 if (is_null_oid(&s->oid))
2824                         die("Can't tag an empty branch.");
2825                 oidcpy(&oid, &s->oid);
2826                 type = OBJ_COMMIT;
2827         } else if (*from == ':') {
2828                 struct object_entry *oe;
2829                 from_mark = parse_mark_ref_eol(from);
2830                 oe = find_mark(marks, from_mark);
2831                 type = oe->type;
2832                 oidcpy(&oid, &oe->idx.oid);
2833         } else if (!get_oid(from, &oid)) {
2834                 struct object_entry *oe = find_object(&oid);
2835                 if (!oe) {
2836                         type = oid_object_info(the_repository, &oid, NULL);
2837                         if (type < 0)
2838                                 die("Not a valid object: %s", from);
2839                 } else
2840                         type = oe->type;
2841         } else
2842                 die("Invalid ref name or SHA1 expression: %s", from);
2843         read_next_command();
2844
2845         /* original-oid ... */
2846         parse_original_identifier();
2847
2848         /* tagger ... */
2849         if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2850                 tagger = parse_ident(v);
2851                 read_next_command();
2852         } else
2853                 tagger = NULL;
2854
2855         /* tag payload/message */
2856         parse_data(&msg, 0, NULL);
2857
2858         /* build the tag object */
2859         strbuf_reset(&new_data);
2860
2861         strbuf_addf(&new_data,
2862                     "object %s\n"
2863                     "type %s\n"
2864                     "tag %s\n",
2865                     oid_to_hex(&oid), type_name(type), t->name);
2866         if (tagger)
2867                 strbuf_addf(&new_data,
2868                             "tagger %s\n", tagger);
2869         strbuf_addch(&new_data, '\n');
2870         strbuf_addbuf(&new_data, &msg);
2871         free(tagger);
2872
2873         if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, next_mark))
2874                 t->pack_id = MAX_PACK_ID;
2875         else
2876                 t->pack_id = pack_id;
2877 }
2878
2879 static void parse_reset_branch(const char *arg)
2880 {
2881         struct branch *b;
2882         const char *tag_name;
2883
2884         b = lookup_branch(arg);
2885         if (b) {
2886                 oidclr(&b->oid);
2887                 oidclr(&b->branch_tree.versions[0].oid);
2888                 oidclr(&b->branch_tree.versions[1].oid);
2889                 if (b->branch_tree.tree) {
2890                         release_tree_content_recursive(b->branch_tree.tree);
2891                         b->branch_tree.tree = NULL;
2892                 }
2893         }
2894         else
2895                 b = new_branch(arg);
2896         read_next_command();
2897         parse_from(b);
2898         if (b->delete && skip_prefix(b->name, "refs/tags/", &tag_name)) {
2899                 /*
2900                  * Elsewhere, we call dump_branches() before dump_tags(),
2901                  * and dump_branches() will handle ref deletions first, so
2902                  * in order to make sure the deletion actually takes effect,
2903                  * we need to remove the tag from our list of tags to update.
2904                  *
2905                  * NEEDSWORK: replace list of tags with hashmap for faster
2906                  * deletion?
2907                  */
2908                 struct tag *t, *prev = NULL;
2909                 for (t = first_tag; t; t = t->next_tag) {
2910                         if (!strcmp(t->name, tag_name))
2911                                 break;
2912                         prev = t;
2913                 }
2914                 if (t) {
2915                         if (prev)
2916                                 prev->next_tag = t->next_tag;
2917                         else
2918                                 first_tag = t->next_tag;
2919                         if (!t->next_tag)
2920                                 last_tag = prev;
2921                         /* There is no mem_pool_free(t) function to call. */
2922                 }
2923         }
2924         if (command_buf.len > 0)
2925                 unread_command_buf = 1;
2926 }
2927
2928 static void cat_blob_write(const char *buf, unsigned long size)
2929 {
2930         if (write_in_full(cat_blob_fd, buf, size) < 0)
2931                 die_errno("Write to frontend failed");
2932 }
2933
2934 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2935 {
2936         struct strbuf line = STRBUF_INIT;
2937         unsigned long size;
2938         enum object_type type = 0;
2939         char *buf;
2940
2941         if (!oe || oe->pack_id == MAX_PACK_ID) {
2942                 buf = read_object_file(oid, &type, &size);
2943         } else {
2944                 type = oe->type;
2945                 buf = gfi_unpack_entry(oe, &size);
2946         }
2947
2948         /*
2949          * Output based on batch_one_object() from cat-file.c.
2950          */
2951         if (type <= 0) {
2952                 strbuf_reset(&line);
2953                 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2954                 cat_blob_write(line.buf, line.len);
2955                 strbuf_release(&line);
2956                 free(buf);
2957                 return;
2958         }
2959         if (!buf)
2960                 die("Can't read object %s", oid_to_hex(oid));
2961         if (type != OBJ_BLOB)
2962                 die("Object %s is a %s but a blob was expected.",
2963                     oid_to_hex(oid), type_name(type));
2964         strbuf_reset(&line);
2965         strbuf_addf(&line, "%s %s %"PRIuMAX"\n", oid_to_hex(oid),
2966                     type_name(type), (uintmax_t)size);
2967         cat_blob_write(line.buf, line.len);
2968         strbuf_release(&line);
2969         cat_blob_write(buf, size);
2970         cat_blob_write("\n", 1);
2971         if (oe && oe->pack_id == pack_id) {
2972                 last_blob.offset = oe->idx.offset;
2973                 strbuf_attach(&last_blob.data, buf, size, size);
2974                 last_blob.depth = oe->depth;
2975         } else
2976                 free(buf);
2977 }
2978
2979 static void parse_get_mark(const char *p)
2980 {
2981         struct object_entry *oe;
2982         char output[GIT_MAX_HEXSZ + 2];
2983
2984         /* get-mark SP <object> LF */
2985         if (*p != ':')
2986                 die("Not a mark: %s", p);
2987
2988         oe = find_mark(marks, parse_mark_ref_eol(p));
2989         if (!oe)
2990                 die("Unknown mark: %s", command_buf.buf);
2991
2992         xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2993         cat_blob_write(output, the_hash_algo->hexsz + 1);
2994 }
2995
2996 static void parse_cat_blob(const char *p)
2997 {
2998         struct object_entry *oe;
2999         struct object_id oid;
3000
3001         /* cat-blob SP <object> LF */
3002         if (*p == ':') {
3003                 oe = find_mark(marks, parse_mark_ref_eol(p));
3004                 if (!oe)
3005                         die("Unknown mark: %s", command_buf.buf);
3006                 oidcpy(&oid, &oe->idx.oid);
3007         } else {
3008                 if (parse_mapped_oid_hex(p, &oid, &p))
3009                         die("Invalid dataref: %s", command_buf.buf);
3010                 if (*p)
3011                         die("Garbage after SHA1: %s", command_buf.buf);
3012                 oe = find_object(&oid);
3013         }
3014
3015         cat_blob(oe, &oid);
3016 }
3017
3018 static struct object_entry *dereference(struct object_entry *oe,
3019                                         struct object_id *oid)
3020 {
3021         unsigned long size;
3022         char *buf = NULL;
3023         const unsigned hexsz = the_hash_algo->hexsz;
3024
3025         if (!oe) {
3026                 enum object_type type = oid_object_info(the_repository, oid,
3027                                                         NULL);
3028                 if (type < 0)
3029                         die("object not found: %s", oid_to_hex(oid));
3030                 /* cache it! */
3031                 oe = insert_object(oid);
3032                 oe->type = type;
3033                 oe->pack_id = MAX_PACK_ID;
3034                 oe->idx.offset = 1;
3035         }
3036         switch (oe->type) {
3037         case OBJ_TREE:  /* easy case. */
3038                 return oe;
3039         case OBJ_COMMIT:
3040         case OBJ_TAG:
3041                 break;
3042         default:
3043                 die("Not a tree-ish: %s", command_buf.buf);
3044         }
3045
3046         if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
3047                 buf = gfi_unpack_entry(oe, &size);
3048         } else {
3049                 enum object_type unused;
3050                 buf = read_object_file(oid, &unused, &size);
3051         }
3052         if (!buf)
3053                 die("Can't load object %s", oid_to_hex(oid));
3054
3055         /* Peel one layer. */
3056         switch (oe->type) {
3057         case OBJ_TAG:
3058                 if (size < hexsz + strlen("object ") ||
3059                     get_oid_hex(buf + strlen("object "), oid))
3060                         die("Invalid SHA1 in tag: %s", command_buf.buf);
3061                 break;
3062         case OBJ_COMMIT:
3063                 if (size < hexsz + strlen("tree ") ||
3064                     get_oid_hex(buf + strlen("tree "), oid))
3065                         die("Invalid SHA1 in commit: %s", command_buf.buf);
3066         }
3067
3068         free(buf);
3069         return find_object(oid);
3070 }
3071
3072 static void insert_mapped_mark(uintmax_t mark, void *object, void *cbp)
3073 {
3074         struct object_id *fromoid = object;
3075         struct object_id *tooid = find_mark(cbp, mark);
3076         int ret;
3077         khiter_t it;
3078
3079         it = kh_put_oid_map(sub_oid_map, *fromoid, &ret);
3080         /* We've already seen this object. */
3081         if (ret == 0)
3082                 return;
3083         kh_value(sub_oid_map, it) = tooid;
3084 }
3085
3086 static void build_mark_map_one(struct mark_set *from, struct mark_set *to)
3087 {
3088         for_each_mark(from, 0, insert_mapped_mark, to);
3089 }
3090
3091 static void build_mark_map(struct string_list *from, struct string_list *to)
3092 {
3093         struct string_list_item *fromp, *top;
3094
3095         sub_oid_map = kh_init_oid_map();
3096
3097         for_each_string_list_item(fromp, from) {
3098                 top = string_list_lookup(to, fromp->string);
3099                 if (!fromp->util) {
3100                         die(_("Missing from marks for submodule '%s'"), fromp->string);
3101                 } else if (!top || !top->util) {
3102                         die(_("Missing to marks for submodule '%s'"), fromp->string);
3103                 }
3104                 build_mark_map_one(fromp->util, top->util);
3105         }
3106 }
3107
3108 static struct object_entry *parse_treeish_dataref(const char **p)
3109 {
3110         struct object_id oid;
3111         struct object_entry *e;
3112
3113         if (**p == ':') {       /* <mark> */
3114                 e = find_mark(marks, parse_mark_ref_space(p));
3115                 if (!e)
3116                         die("Unknown mark: %s", command_buf.buf);
3117                 oidcpy(&oid, &e->idx.oid);
3118         } else {        /* <sha1> */
3119                 if (parse_mapped_oid_hex(*p, &oid, p))
3120                         die("Invalid dataref: %s", command_buf.buf);
3121                 e = find_object(&oid);
3122                 if (*(*p)++ != ' ')
3123                         die("Missing space after tree-ish: %s", command_buf.buf);
3124         }
3125
3126         while (!e || e->type != OBJ_TREE)
3127                 e = dereference(e, &oid);
3128         return e;
3129 }
3130
3131 static void print_ls(int mode, const unsigned char *hash, const char *path)
3132 {
3133         static struct strbuf line = STRBUF_INIT;
3134
3135         /* See show_tree(). */
3136         const char *type =
3137                 S_ISGITLINK(mode) ? commit_type :
3138                 S_ISDIR(mode) ? tree_type :
3139                 blob_type;
3140
3141         if (!mode) {
3142                 /* missing SP path LF */
3143                 strbuf_reset(&line);
3144                 strbuf_addstr(&line, "missing ");
3145                 quote_c_style(path, &line, NULL, 0);
3146                 strbuf_addch(&line, '\n');
3147         } else {
3148                 /* mode SP type SP object_name TAB path LF */
3149                 strbuf_reset(&line);
3150                 strbuf_addf(&line, "%06o %s %s\t",
3151                                 mode & ~NO_DELTA, type, hash_to_hex(hash));
3152                 quote_c_style(path, &line, NULL, 0);
3153                 strbuf_addch(&line, '\n');
3154         }
3155         cat_blob_write(line.buf, line.len);
3156 }
3157
3158 static void parse_ls(const char *p, struct branch *b)
3159 {
3160         struct tree_entry *root = NULL;
3161         struct tree_entry leaf = {NULL};
3162
3163         /* ls SP (<tree-ish> SP)? <path> */
3164         if (*p == '"') {
3165                 if (!b)
3166                         die("Not in a commit: %s", command_buf.buf);
3167                 root = &b->branch_tree;
3168         } else {
3169                 struct object_entry *e = parse_treeish_dataref(&p);
3170                 root = new_tree_entry();
3171                 oidcpy(&root->versions[1].oid, &e->idx.oid);
3172                 if (!is_null_oid(&root->versions[1].oid))
3173                         root->versions[1].mode = S_IFDIR;
3174                 load_tree(root);
3175         }
3176         if (*p == '"') {
3177                 static struct strbuf uq = STRBUF_INIT;
3178                 const char *endp;
3179                 strbuf_reset(&uq);
3180                 if (unquote_c_style(&uq, p, &endp))
3181                         die("Invalid path: %s", command_buf.buf);
3182                 if (*endp)
3183                         die("Garbage after path in: %s", command_buf.buf);
3184                 p = uq.buf;
3185         }
3186         tree_content_get(root, p, &leaf, 1);
3187         /*
3188          * A directory in preparation would have a sha1 of zero
3189          * until it is saved.  Save, for simplicity.
3190          */
3191         if (S_ISDIR(leaf.versions[1].mode))
3192                 store_tree(&leaf);
3193
3194         print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3195         if (leaf.tree)
3196                 release_tree_content_recursive(leaf.tree);
3197         if (!b || root != &b->branch_tree)
3198                 release_tree_entry(root);
3199 }
3200
3201 static void checkpoint(void)
3202 {
3203         checkpoint_requested = 0;
3204         if (object_count) {
3205                 cycle_packfile();
3206         }
3207         dump_branches();
3208         dump_tags();
3209         dump_marks();
3210 }
3211
3212 static void parse_checkpoint(void)
3213 {
3214         checkpoint_requested = 1;
3215         skip_optional_lf();
3216 }
3217
3218 static void parse_progress(void)
3219 {
3220         fwrite(command_buf.buf, 1, command_buf.len, stdout);
3221         fputc('\n', stdout);
3222         fflush(stdout);
3223         skip_optional_lf();
3224 }
3225
3226 static void parse_alias(void)
3227 {
3228         struct object_entry *e;
3229         struct branch b;
3230
3231         skip_optional_lf();
3232         read_next_command();
3233
3234         /* mark ... */
3235         parse_mark();
3236         if (!next_mark)
3237                 die(_("Expected 'mark' command, got %s"), command_buf.buf);
3238
3239         /* to ... */
3240         memset(&b, 0, sizeof(b));
3241         if (!parse_objectish_with_prefix(&b, "to "))
3242                 die(_("Expected 'to' command, got %s"), command_buf.buf);
3243         e = find_object(&b.oid);
3244         assert(e);
3245         insert_mark(marks, next_mark, e);
3246 }
3247
3248 static char* make_fast_import_path(const char *path)
3249 {
3250         if (!relative_marks_paths || is_absolute_path(path))
3251                 return xstrdup(path);
3252         return git_pathdup("info/fast-import/%s", path);
3253 }
3254
3255 static void option_import_marks(const char *marks,
3256                                         int from_stream, int ignore_missing)
3257 {
3258         if (import_marks_file) {
3259                 if (from_stream)
3260                         die("Only one import-marks command allowed per stream");
3261
3262                 /* read previous mark file */
3263                 if(!import_marks_file_from_stream)
3264                         read_marks();
3265         }
3266
3267         import_marks_file = make_fast_import_path(marks);
3268         import_marks_file_from_stream = from_stream;
3269         import_marks_file_ignore_missing = ignore_missing;
3270 }
3271
3272 static void option_date_format(const char *fmt)
3273 {
3274         if (!strcmp(fmt, "raw"))
3275                 whenspec = WHENSPEC_RAW;
3276         else if (!strcmp(fmt, "raw-permissive"))
3277                 whenspec = WHENSPEC_RAW_PERMISSIVE;
3278         else if (!strcmp(fmt, "rfc2822"))
3279                 whenspec = WHENSPEC_RFC2822;
3280         else if (!strcmp(fmt, "now"))
3281                 whenspec = WHENSPEC_NOW;
3282         else
3283                 die("unknown --date-format argument %s", fmt);
3284 }
3285
3286 static unsigned long ulong_arg(const char *option, const char *arg)
3287 {
3288         char *endptr;
3289         unsigned long rv = strtoul(arg, &endptr, 0);
3290         if (strchr(arg, '-') || endptr == arg || *endptr)
3291                 die("%s: argument must be a non-negative integer", option);
3292         return rv;
3293 }
3294
3295 static void option_depth(const char *depth)
3296 {
3297         max_depth = ulong_arg("--depth", depth);
3298         if (max_depth > MAX_DEPTH)
3299                 die("--depth cannot exceed %u", MAX_DEPTH);
3300 }
3301
3302 static void option_active_branches(const char *branches)
3303 {
3304         max_active_branches = ulong_arg("--active-branches", branches);
3305 }
3306
3307 static void option_export_marks(const char *marks)
3308 {
3309         export_marks_file = make_fast_import_path(marks);
3310 }
3311
3312 static void option_cat_blob_fd(const char *fd)
3313 {
3314         unsigned long n = ulong_arg("--cat-blob-fd", fd);
3315         if (n > (unsigned long) INT_MAX)
3316                 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3317         cat_blob_fd = (int) n;
3318 }
3319
3320 static void option_export_pack_edges(const char *edges)
3321 {
3322         if (pack_edges)
3323                 fclose(pack_edges);
3324         pack_edges = xfopen(edges, "a");
3325 }
3326
3327 static void option_rewrite_submodules(const char *arg, struct string_list *list)
3328 {
3329         struct mark_set *ms;
3330         FILE *fp;
3331         char *s = xstrdup(arg);
3332         char *f = strchr(s, ':');
3333         if (!f)
3334                 die(_("Expected format name:filename for submodule rewrite option"));
3335         *f = '\0';
3336         f++;
3337         ms = xcalloc(1, sizeof(*ms));
3338         string_list_insert(list, s)->util = ms;
3339
3340         fp = fopen(f, "r");
3341         if (!fp)
3342                 die_errno("cannot read '%s'", f);
3343         read_mark_file(ms, fp, insert_oid_entry);
3344         fclose(fp);
3345 }
3346
3347 static int parse_one_option(const char *option)
3348 {
3349         if (skip_prefix(option, "max-pack-size=", &option)) {
3350                 unsigned long v;
3351                 if (!git_parse_ulong(option, &v))
3352                         return 0;
3353                 if (v < 8192) {
3354                         warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3355                         v *= 1024 * 1024;
3356                 } else if (v < 1024 * 1024) {
3357                         warning("minimum max-pack-size is 1 MiB");
3358                         v = 1024 * 1024;
3359                 }
3360                 max_packsize = v;
3361         } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3362                 unsigned long v;
3363                 if (!git_parse_ulong(option, &v))
3364                         return 0;
3365                 big_file_threshold = v;
3366         } else if (skip_prefix(option, "depth=", &option)) {
3367                 option_depth(option);
3368         } else if (skip_prefix(option, "active-branches=", &option)) {
3369                 option_active_branches(option);
3370         } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3371                 option_export_pack_edges(option);
3372         } else if (!strcmp(option, "quiet")) {
3373                 show_stats = 0;
3374         } else if (!strcmp(option, "stats")) {
3375                 show_stats = 1;
3376         } else if (!strcmp(option, "allow-unsafe-features")) {
3377                 ; /* already handled during early option parsing */
3378         } else {
3379                 return 0;
3380         }
3381
3382         return 1;
3383 }
3384
3385 static void check_unsafe_feature(const char *feature, int from_stream)
3386 {
3387         if (from_stream && !allow_unsafe_features)
3388                 die(_("feature '%s' forbidden in input without --allow-unsafe-features"),
3389                     feature);
3390 }
3391
3392 static int parse_one_feature(const char *feature, int from_stream)
3393 {
3394         const char *arg;
3395
3396         if (skip_prefix(feature, "date-format=", &arg)) {
3397                 option_date_format(arg);
3398         } else if (skip_prefix(feature, "import-marks=", &arg)) {
3399                 check_unsafe_feature("import-marks", from_stream);
3400                 option_import_marks(arg, from_stream, 0);
3401         } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3402                 check_unsafe_feature("import-marks-if-exists", from_stream);
3403                 option_import_marks(arg, from_stream, 1);
3404         } else if (skip_prefix(feature, "export-marks=", &arg)) {
3405                 check_unsafe_feature(feature, from_stream);
3406                 option_export_marks(arg);
3407         } else if (!strcmp(feature, "alias")) {
3408                 ; /* Don't die - this feature is supported */
3409         } else if (skip_prefix(feature, "rewrite-submodules-to=", &arg)) {
3410                 option_rewrite_submodules(arg, &sub_marks_to);
3411         } else if (skip_prefix(feature, "rewrite-submodules-from=", &arg)) {
3412                 option_rewrite_submodules(arg, &sub_marks_from);
3413         } else if (skip_prefix(feature, "rewrite-submodules-from=", &arg)) {
3414         } else if (!strcmp(feature, "get-mark")) {
3415                 ; /* Don't die - this feature is supported */
3416         } else if (!strcmp(feature, "cat-blob")) {
3417                 ; /* Don't die - this feature is supported */
3418         } else if (!strcmp(feature, "relative-marks")) {
3419                 relative_marks_paths = 1;
3420         } else if (!strcmp(feature, "no-relative-marks")) {
3421                 relative_marks_paths = 0;
3422         } else if (!strcmp(feature, "done")) {
3423                 require_explicit_termination = 1;
3424         } else if (!strcmp(feature, "force")) {
3425                 force_update = 1;
3426         } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3427                 ; /* do nothing; we have the feature */
3428         } else {
3429                 return 0;
3430         }
3431
3432         return 1;
3433 }
3434
3435 static void parse_feature(const char *feature)
3436 {
3437         if (seen_data_command)
3438                 die("Got feature command '%s' after data command", feature);
3439
3440         if (parse_one_feature(feature, 1))
3441                 return;
3442
3443         die("This version of fast-import does not support feature %s.", feature);
3444 }
3445
3446 static void parse_option(const char *option)
3447 {
3448         if (seen_data_command)
3449                 die("Got option command '%s' after data command", option);
3450
3451         if (parse_one_option(option))
3452                 return;
3453
3454         die("This version of fast-import does not support option: %s", option);
3455 }
3456
3457 static void git_pack_config(void)
3458 {
3459         int indexversion_value;
3460         int limit;
3461         unsigned long packsizelimit_value;
3462
3463         if (!git_config_get_ulong("pack.depth", &max_depth)) {
3464                 if (max_depth > MAX_DEPTH)
3465                         max_depth = MAX_DEPTH;
3466         }
3467         if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3468                 pack_idx_opts.version = indexversion_value;
3469                 if (pack_idx_opts.version > 2)
3470                         git_die_config("pack.indexversion",
3471                                         "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3472         }
3473         if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3474                 max_packsize = packsizelimit_value;
3475
3476         if (!git_config_get_int("fastimport.unpacklimit", &limit))
3477                 unpack_limit = limit;
3478         else if (!git_config_get_int("transfer.unpacklimit", &limit))
3479                 unpack_limit = limit;
3480
3481         git_config(git_default_config, NULL);
3482 }
3483
3484 static const char fast_import_usage[] =
3485 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3486
3487 static void parse_argv(void)
3488 {
3489         unsigned int i;
3490
3491         for (i = 1; i < global_argc; i++) {
3492                 const char *a = global_argv[i];
3493
3494                 if (*a != '-' || !strcmp(a, "--"))
3495                         break;
3496
3497                 if (!skip_prefix(a, "--", &a))
3498                         die("unknown option %s", a);
3499
3500                 if (parse_one_option(a))
3501                         continue;
3502
3503                 if (parse_one_feature(a, 0))
3504                         continue;
3505
3506                 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3507                         option_cat_blob_fd(a);
3508                         continue;
3509                 }
3510
3511                 die("unknown option --%s", a);
3512         }
3513         if (i != global_argc)
3514                 usage(fast_import_usage);
3515
3516         seen_data_command = 1;
3517         if (import_marks_file)
3518                 read_marks();
3519         build_mark_map(&sub_marks_from, &sub_marks_to);
3520 }
3521
3522 int cmd_main(int argc, const char **argv)
3523 {
3524         unsigned int i;
3525
3526         if (argc == 2 && !strcmp(argv[1], "-h"))
3527                 usage(fast_import_usage);
3528
3529         setup_git_directory();
3530         reset_pack_idx_option(&pack_idx_opts);
3531         git_pack_config();
3532
3533         alloc_objects(object_entry_alloc);
3534         strbuf_init(&command_buf, 0);
3535         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3536         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3537         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3538         marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3539
3540         hashmap_init(&object_table, object_entry_hashcmp, NULL, 0);
3541
3542         /*
3543          * We don't parse most options until after we've seen the set of
3544          * "feature" lines at the start of the stream (which allows the command
3545          * line to override stream data). But we must do an early parse of any
3546          * command-line options that impact how we interpret the feature lines.
3547          */
3548         for (i = 1; i < argc; i++) {
3549                 const char *arg = argv[i];
3550                 if (*arg != '-' || !strcmp(arg, "--"))
3551                         break;
3552                 if (!strcmp(arg, "--allow-unsafe-features"))
3553                         allow_unsafe_features = 1;
3554         }
3555
3556         global_argc = argc;
3557         global_argv = argv;
3558
3559         rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3560         for (i = 0; i < (cmd_save - 1); i++)
3561                 rc_free[i].next = &rc_free[i + 1];
3562         rc_free[cmd_save - 1].next = NULL;
3563
3564         start_packfile();
3565         set_die_routine(die_nicely);
3566         set_checkpoint_signal();
3567         while (read_next_command() != EOF) {
3568                 const char *v;
3569                 if (!strcmp("blob", command_buf.buf))
3570                         parse_new_blob();
3571                 else if (skip_prefix(command_buf.buf, "commit ", &v))
3572                         parse_new_commit(v);
3573                 else if (skip_prefix(command_buf.buf, "tag ", &v))
3574                         parse_new_tag(v);
3575                 else if (skip_prefix(command_buf.buf, "reset ", &v))
3576                         parse_reset_branch(v);
3577                 else if (skip_prefix(command_buf.buf, "ls ", &v))
3578                         parse_ls(v, NULL);
3579                 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
3580                         parse_cat_blob(v);
3581                 else if (skip_prefix(command_buf.buf, "get-mark ", &v))
3582                         parse_get_mark(v);
3583                 else if (!strcmp("checkpoint", command_buf.buf))
3584                         parse_checkpoint();
3585                 else if (!strcmp("done", command_buf.buf))
3586                         break;
3587                 else if (!strcmp("alias", command_buf.buf))
3588                         parse_alias();
3589                 else if (starts_with(command_buf.buf, "progress "))
3590                         parse_progress();
3591                 else if (skip_prefix(command_buf.buf, "feature ", &v))
3592                         parse_feature(v);
3593                 else if (skip_prefix(command_buf.buf, "option git ", &v))
3594                         parse_option(v);
3595                 else if (starts_with(command_buf.buf, "option "))
3596                         /* ignore non-git options*/;
3597                 else
3598                         die("Unsupported command: %s", command_buf.buf);
3599
3600                 if (checkpoint_requested)
3601                         checkpoint();
3602         }
3603
3604         /* argv hasn't been parsed yet, do so */
3605         if (!seen_data_command)
3606                 parse_argv();
3607
3608         if (require_explicit_termination && feof(stdin))
3609                 die("stream ends early");
3610
3611         end_packfile();
3612
3613         dump_branches();
3614         dump_tags();
3615         unkeep_all_packs();
3616         dump_marks();
3617
3618         if (pack_edges)
3619                 fclose(pack_edges);
3620
3621         if (show_stats) {
3622                 uintmax_t total_count = 0, duplicate_count = 0;
3623                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3624                         total_count += object_count_by_type[i];
3625                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3626                         duplicate_count += duplicate_count_by_type[i];
3627
3628                 fprintf(stderr, "%s statistics:\n", argv[0]);
3629                 fprintf(stderr, "---------------------------------------------------------------------\n");
3630                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3631                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3632                 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]);
3633                 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]);
3634                 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]);
3635                 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]);
3636                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3637                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3638                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3639                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3640                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3641                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3642                 fprintf(stderr, "---------------------------------------------------------------------\n");
3643                 pack_report();
3644                 fprintf(stderr, "---------------------------------------------------------------------\n");
3645                 fprintf(stderr, "\n");
3646         }
3647
3648         return failure ? 1 : 0;
3649 }