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