builtin/fast-export.c

   1 /*
   2  * "git fast-export" builtin command
   3  *
   4  * Copyright (C) 2007 Johannes E. Schindelin
   5  */
   6 #include "builtin.h"
   7 #include "cache.h"
   8 #include "config.h"
   9 #include "refs.h"
  10 #include "refspec.h"
  11 #include "object-store.h"
  12 #include "commit.h"
  13 #include "object.h"
  14 #include "tag.h"
  15 #include "diff.h"
  16 #include "diffcore.h"
  17 #include "log-tree.h"
  18 #include "revision.h"
  19 #include "decorate.h"
  20 #include "string-list.h"
  21 #include "utf8.h"
  22 #include "parse-options.h"
  23 #include "quote.h"
  24 #include "remote.h"
  25 #include "blob.h"
  26 #include "commit-slab.h"
  27
  28 static const char *fast_export_usage[] = {
  29         N_("git fast-export [rev-list-opts]"),
  30         NULL
  31 };
  32
  33 static int progress;
  34 static enum { SIGNED_TAG_ABORT, VERBATIM, WARN, WARN_STRIP, STRIP } signed_tag_mode = SIGNED_TAG_ABORT;
  35 static enum { TAG_FILTERING_ABORT, DROP, REWRITE } tag_of_filtered_mode = TAG_FILTERING_ABORT;
  36 static enum { REENCODE_ABORT, REENCODE_YES, REENCODE_NO } reencode_mode = REENCODE_ABORT;
  37 static int fake_missing_tagger;
  38 static int use_done_feature;
  39 static int no_data;
  40 static int full_tree;
  41 static int reference_excluded_commits;
  42 static int show_original_ids;
  43 static struct string_list extra_refs = STRING_LIST_INIT_NODUP;
  44 static struct string_list tag_refs = STRING_LIST_INIT_NODUP;
  45 static struct refspec refspecs = REFSPEC_INIT_FETCH;
  46 static int anonymize;
  47 static struct revision_sources revision_sources;
  48
  49 static int parse_opt_signed_tag_mode(const struct option *opt,
  50                                      const char *arg, int unset)
  51 {
  52         if (unset || !strcmp(arg, "abort"))
  53                 signed_tag_mode = SIGNED_TAG_ABORT;
  54         else if (!strcmp(arg, "verbatim") || !strcmp(arg, "ignore"))
  55                 signed_tag_mode = VERBATIM;
  56         else if (!strcmp(arg, "warn"))
  57                 signed_tag_mode = WARN;
  58         else if (!strcmp(arg, "warn-strip"))
  59                 signed_tag_mode = WARN_STRIP;
  60         else if (!strcmp(arg, "strip"))
  61                 signed_tag_mode = STRIP;
  62         else
  63                 return error("Unknown signed-tags mode: %s", arg);
  64         return 0;
  65 }
  66
  67 static int parse_opt_tag_of_filtered_mode(const struct option *opt,
  68                                           const char *arg, int unset)
  69 {
  70         if (unset || !strcmp(arg, "abort"))
  71                 tag_of_filtered_mode = TAG_FILTERING_ABORT;
  72         else if (!strcmp(arg, "drop"))
  73                 tag_of_filtered_mode = DROP;
  74         else if (!strcmp(arg, "rewrite"))
  75                 tag_of_filtered_mode = REWRITE;
  76         else
  77                 return error("Unknown tag-of-filtered mode: %s", arg);
  78         return 0;
  79 }
  80
  81 static int parse_opt_reencode_mode(const struct option *opt,
  82                                    const char *arg, int unset)
  83 {
  84         if (unset) {
  85                 reencode_mode = REENCODE_ABORT;
  86                 return 0;
  87         }
  88
  89         switch (git_parse_maybe_bool(arg)) {
  90         case 0:
  91                 reencode_mode = REENCODE_NO;
  92                 break;
  93         case 1:
  94                 reencode_mode = REENCODE_YES;
  95                 break;
  96         default:
  97                 if (!strcasecmp(arg, "abort"))
  98                         reencode_mode = REENCODE_ABORT;
  99                 else
 100                         return error("Unknown reencoding mode: %s", arg);
 101         }
 102
 103         return 0;
 104 }
 105
 106 static struct decoration idnums;
 107 static uint32_t last_idnum;
 108
 109 static int has_unshown_parent(struct commit *commit)
 110 {
 111         struct commit_list *parent;
 112
 113         for (parent = commit->parents; parent; parent = parent->next)
 114                 if (!(parent->item->object.flags & SHOWN) &&
 115                     !(parent->item->object.flags & UNINTERESTING))
 116                         return 1;
 117         return 0;
 118 }
 119
 120 struct anonymized_entry {
 121         struct hashmap_entry hash;
 122         const char *orig;
 123         size_t orig_len;
 124         const char *anon;
 125         size_t anon_len;
 126 };
 127
 128 static int anonymized_entry_cmp(const void *unused_cmp_data,
 129                                 const struct hashmap_entry *eptr,
 130                                 const struct hashmap_entry *entry_or_key,
 131                                 const void *unused_keydata)
 132 {
 133         const struct anonymized_entry *a, *b;
 134
 135         a = container_of(eptr, const struct anonymized_entry, hash);
 136         b = container_of(entry_or_key, const struct anonymized_entry, hash);
 137
 138         return a->orig_len != b->orig_len ||
 139                 memcmp(a->orig, b->orig, a->orig_len);
 140 }
 141
 142 /*
 143  * Basically keep a cache of X->Y so that we can repeatedly replace
 144  * the same anonymized string with another. The actual generation
 145  * is farmed out to the generate function.
 146  */
 147 static const void *anonymize_mem(struct hashmap *map,
 148                                  void *(*generate)(const void *, size_t *),
 149                                  const void *orig, size_t *len)
 150 {
 151         struct anonymized_entry key, *ret;
 152
 153         if (!map->cmpfn)
 154                 hashmap_init(map, anonymized_entry_cmp, NULL, 0);
 155
 156         hashmap_entry_init(&key.hash, memhash(orig, *len));
 157         key.orig = orig;
 158         key.orig_len = *len;
 159         ret = hashmap_get_entry(map, &key, NULL, struct anonymized_entry, hash);
 160
 161         if (!ret) {
 162                 ret = xmalloc(sizeof(*ret));
 163                 hashmap_entry_init(&ret->hash, key.hash.hash);
 164                 ret->orig = xstrdup(orig);
 165                 ret->orig_len = *len;
 166                 ret->anon = generate(orig, len);
 167                 ret->anon_len = *len;
 168                 hashmap_put(map, &ret->hash);
 169         }
 170
 171         *len = ret->anon_len;
 172         return ret->anon;
 173 }
 174
 175 /*
 176  * We anonymize each component of a path individually,
 177  * so that paths a/b and a/c will share a common root.
 178  * The paths are cached via anonymize_mem so that repeated
 179  * lookups for "a" will yield the same value.
 180  */
 181 static void anonymize_path(struct strbuf *out, const char *path,
 182                            struct hashmap *map,
 183                            void *(*generate)(const void *, size_t *))
 184 {
 185         while (*path) {
 186                 const char *end_of_component = strchrnul(path, '/');
 187                 size_t len = end_of_component - path;
 188                 const char *c = anonymize_mem(map, generate, path, &len);
 189                 strbuf_add(out, c, len);
 190                 path = end_of_component;
 191                 if (*path)
 192                         strbuf_addch(out, *path++);
 193         }
 194 }
 195
 196 static inline void *mark_to_ptr(uint32_t mark)
 197 {
 198         return (void *)(uintptr_t)mark;
 199 }
 200
 201 static inline uint32_t ptr_to_mark(void * mark)
 202 {
 203         return (uint32_t)(uintptr_t)mark;
 204 }
 205
 206 static inline void mark_object(struct object *object, uint32_t mark)
 207 {
 208         add_decoration(&idnums, object, mark_to_ptr(mark));
 209 }
 210
 211 static inline void mark_next_object(struct object *object)
 212 {
 213         mark_object(object, ++last_idnum);
 214 }
 215
 216 static int get_object_mark(struct object *object)
 217 {
 218         void *decoration = lookup_decoration(&idnums, object);
 219         if (!decoration)
 220                 return 0;
 221         return ptr_to_mark(decoration);
 222 }
 223
 224 static struct commit *rewrite_commit(struct commit *p)
 225 {
 226         for (;;) {
 227                 if (p->parents && p->parents->next)
 228                         break;
 229                 if (p->object.flags & UNINTERESTING)
 230                         break;
 231                 if (!(p->object.flags & TREESAME))
 232                         break;
 233                 if (!p->parents)
 234                         return NULL;
 235                 p = p->parents->item;
 236         }
 237         return p;
 238 }
 239
 240 static void show_progress(void)
 241 {
 242         static int counter = 0;
 243         if (!progress)
 244                 return;
 245         if ((++counter % progress) == 0)
 246                 printf("progress %d objects\n", counter);
 247 }
 248
 249 /*
 250  * Ideally we would want some transformation of the blob data here
 251  * that is unreversible, but would still be the same size and have
 252  * the same data relationship to other blobs (so that we get the same
 253  * delta and packing behavior as the original). But the first and last
 254  * requirements there are probably mutually exclusive, so let's take
 255  * the easy way out for now, and just generate arbitrary content.
 256  *
 257  * There's no need to cache this result with anonymize_mem, since
 258  * we already handle blob content caching with marks.
 259  */
 260 static char *anonymize_blob(unsigned long *size)
 261 {
 262         static int counter;
 263         struct strbuf out = STRBUF_INIT;
 264         strbuf_addf(&out, "anonymous blob %d", counter++);
 265         *size = out.len;
 266         return strbuf_detach(&out, NULL);
 267 }
 268
 269 static void export_blob(const struct object_id *oid)
 270 {
 271         unsigned long size;
 272         enum object_type type;
 273         char *buf;
 274         struct object *object;
 275         int eaten;
 276
 277         if (no_data)
 278                 return;
 279
 280         if (is_null_oid(oid))
 281                 return;
 282
 283         object = lookup_object(the_repository, oid);
 284         if (object && object->flags & SHOWN)
 285                 return;
 286
 287         if (anonymize) {
 288                 buf = anonymize_blob(&size);
 289                 object = (struct object *)lookup_blob(the_repository, oid);
 290                 eaten = 0;
 291         } else {
 292                 buf = read_object_file(oid, &type, &size);
 293                 if (!buf)
 294                         die("could not read blob %s", oid_to_hex(oid));
 295                 if (check_object_signature(oid, buf, size, type_name(type)) < 0)
 296                         die("oid mismatch in blob %s", oid_to_hex(oid));
 297                 object = parse_object_buffer(the_repository, oid, type,
 298                                              size, buf, &eaten);
 299         }
 300
 301         if (!object)
 302                 die("Could not read blob %s", oid_to_hex(oid));
 303
 304         mark_next_object(object);
 305
 306         printf("blob\nmark :%"PRIu32"\n", last_idnum);
 307         if (show_original_ids)
 308                 printf("original-oid %s\n", oid_to_hex(oid));
 309         printf("data %"PRIuMAX"\n", (uintmax_t)size);
 310         if (size && fwrite(buf, size, 1, stdout) != 1)
 311                 die_errno("could not write blob '%s'", oid_to_hex(oid));
 312         printf("\n");
 313
 314         show_progress();
 315
 316         object->flags |= SHOWN;
 317         if (!eaten)
 318                 free(buf);
 319 }
 320
 321 static int depth_first(const void *a_, const void *b_)
 322 {
 323         const struct diff_filepair *a = *((const struct diff_filepair **)a_);
 324         const struct diff_filepair *b = *((const struct diff_filepair **)b_);
 325         const char *name_a, *name_b;
 326         int len_a, len_b, len;
 327         int cmp;
 328
 329         name_a = a->one ? a->one->path : a->two->path;
 330         name_b = b->one ? b->one->path : b->two->path;
 331
 332         len_a = strlen(name_a);
 333         len_b = strlen(name_b);
 334         len = (len_a < len_b) ? len_a : len_b;
 335
 336         /* strcmp will sort 'd' before 'd/e', we want 'd/e' before 'd' */
 337         cmp = memcmp(name_a, name_b, len);
 338         if (cmp)
 339                 return cmp;
 340         cmp = len_b - len_a;
 341         if (cmp)
 342                 return cmp;
 343         /*
 344          * Move 'R'ename entries last so that all references of the file
 345          * appear in the output before it is renamed (e.g., when a file
 346          * was copied and renamed in the same commit).
 347          */
 348         return (a->status == 'R') - (b->status == 'R');
 349 }
 350
 351 static void print_path_1(const char *path)
 352 {
 353         int need_quote = quote_c_style(path, NULL, NULL, 0);
 354         if (need_quote)
 355                 quote_c_style(path, NULL, stdout, 0);
 356         else if (strchr(path, ' '))
 357                 printf("\"%s\"", path);
 358         else
 359                 printf("%s", path);
 360 }
 361
 362 static void *anonymize_path_component(const void *path, size_t *len)
 363 {
 364         static int counter;
 365         struct strbuf out = STRBUF_INIT;
 366         strbuf_addf(&out, "path%d", counter++);
 367         return strbuf_detach(&out, len);
 368 }
 369
 370 static void print_path(const char *path)
 371 {
 372         if (!anonymize)
 373                 print_path_1(path);
 374         else {
 375                 static struct hashmap paths;
 376                 static struct strbuf anon = STRBUF_INIT;
 377
 378                 anonymize_path(&anon, path, &paths, anonymize_path_component);
 379                 print_path_1(anon.buf);
 380                 strbuf_reset(&anon);
 381         }
 382 }
 383
 384 static void *generate_fake_oid(const void *old, size_t *len)
 385 {
 386         static uint32_t counter = 1; /* avoid null oid */
 387         const unsigned hashsz = the_hash_algo->rawsz;
 388         unsigned char *out = xcalloc(hashsz, 1);
 389         put_be32(out + hashsz - 4, counter++);
 390         return out;
 391 }
 392
 393 static const struct object_id *anonymize_oid(const struct object_id *oid)
 394 {
 395         static struct hashmap objs;
 396         size_t len = the_hash_algo->rawsz;
 397         return anonymize_mem(&objs, generate_fake_oid, oid, &len);
 398 }
 399
 400 static void show_filemodify(struct diff_queue_struct *q,
 401                             struct diff_options *options, void *data)
 402 {
 403         int i;
 404         struct string_list *changed = data;
 405
 406         /*
 407          * Handle files below a directory first, in case they are all deleted
 408          * and the directory changes to a file or symlink.
 409          */
 410         QSORT(q->queue, q->nr, depth_first);
 411
 412         for (i = 0; i < q->nr; i++) {
 413                 struct diff_filespec *ospec = q->queue[i]->one;
 414                 struct diff_filespec *spec = q->queue[i]->two;
 415
 416                 switch (q->queue[i]->status) {
 417                 case DIFF_STATUS_DELETED:
 418                         printf("D ");
 419                         print_path(spec->path);
 420                         string_list_insert(changed, spec->path);
 421                         putchar('\n');
 422                         break;
 423
 424                 case DIFF_STATUS_COPIED:
 425                 case DIFF_STATUS_RENAMED:
 426                         /*
 427                          * If a change in the file corresponding to ospec->path
 428                          * has been observed, we cannot trust its contents
 429                          * because the diff is calculated based on the prior
 430                          * contents, not the current contents.  So, declare a
 431                          * copy or rename only if there was no change observed.
 432                          */
 433                         if (!string_list_has_string(changed, ospec->path)) {
 434                                 printf("%c ", q->queue[i]->status);
 435                                 print_path(ospec->path);
 436                                 putchar(' ');
 437                                 print_path(spec->path);
 438                                 string_list_insert(changed, spec->path);
 439                                 putchar('\n');
 440
 441                                 if (oideq(&ospec->oid, &spec->oid) &&
 442                                     ospec->mode == spec->mode)
 443                                         break;
 444                         }
 445                         /* fallthrough */
 446
 447                 case DIFF_STATUS_TYPE_CHANGED:
 448                 case DIFF_STATUS_MODIFIED:
 449                 case DIFF_STATUS_ADDED:
 450                         /*
 451                          * Links refer to objects in another repositories;
 452                          * output the SHA-1 verbatim.
 453                          */
 454                         if (no_data || S_ISGITLINK(spec->mode))
 455                                 printf("M %06o %s ", spec->mode,
 456                                        oid_to_hex(anonymize ?
 457                                                   anonymize_oid(&spec->oid) :
 458                                                   &spec->oid));
 459                         else {
 460                                 struct object *object = lookup_object(the_repository,
 461                                                                       &spec->oid);
 462                                 printf("M %06o :%d ", spec->mode,
 463                                        get_object_mark(object));
 464                         }
 465                         print_path(spec->path);
 466                         string_list_insert(changed, spec->path);
 467                         putchar('\n');
 468                         break;
 469
 470                 default:
 471                         die("Unexpected comparison status '%c' for %s, %s",
 472                                 q->queue[i]->status,
 473                                 ospec->path ? ospec->path : "none",
 474                                 spec->path ? spec->path : "none");
 475                 }
 476         }
 477 }
 478
 479 static const char *find_encoding(const char *begin, const char *end)
 480 {
 481         const char *needle = "\nencoding ";
 482         char *bol, *eol;
 483
 484         bol = memmem(begin, end ? end - begin : strlen(begin),
 485                      needle, strlen(needle));
 486         if (!bol)
 487                 return NULL;
 488         bol += strlen(needle);
 489         eol = strchrnul(bol, '\n');
 490         *eol = '\0';
 491         return bol;
 492 }
 493
 494 static void *anonymize_ref_component(const void *old, size_t *len)
 495 {
 496         static int counter;
 497         struct strbuf out = STRBUF_INIT;
 498         strbuf_addf(&out, "ref%d", counter++);
 499         return strbuf_detach(&out, len);
 500 }
 501
 502 static const char *anonymize_refname(const char *refname)
 503 {
 504         /*
 505          * If any of these prefixes is found, we will leave it intact
 506          * so that tags remain tags and so forth.
 507          */
 508         static const char *prefixes[] = {
 509                 "refs/heads/",
 510                 "refs/tags/",
 511                 "refs/remotes/",
 512                 "refs/"
 513         };
 514         static struct hashmap refs;
 515         static struct strbuf anon = STRBUF_INIT;
 516         int i;
 517
 518         /*
 519          * We also leave "master" as a special case, since it does not reveal
 520          * anything interesting.
 521          */
 522         if (!strcmp(refname, "refs/heads/master"))
 523                 return refname;
 524
 525         strbuf_reset(&anon);
 526         for (i = 0; i < ARRAY_SIZE(prefixes); i++) {
 527                 if (skip_prefix(refname, prefixes[i], &refname)) {
 528                         strbuf_addstr(&anon, prefixes[i]);
 529                         break;
 530                 }
 531         }
 532
 533         anonymize_path(&anon, refname, &refs, anonymize_ref_component);
 534         return anon.buf;
 535 }
 536
 537 /*
 538  * We do not even bother to cache commit messages, as they are unlikely
 539  * to be repeated verbatim, and it is not that interesting when they are.
 540  */
 541 static char *anonymize_commit_message(const char *old)
 542 {
 543         static int counter;
 544         return xstrfmt("subject %d\n\nbody\n", counter++);
 545 }
 546
 547 static struct hashmap idents;
 548 static void *anonymize_ident(const void *old, size_t *len)
 549 {
 550         static int counter;
 551         struct strbuf out = STRBUF_INIT;
 552         strbuf_addf(&out, "User %d <user%d@example.com>", counter, counter);
 553         counter++;
 554         return strbuf_detach(&out, len);
 555 }
 556
 557 /*
 558  * Our strategy here is to anonymize the names and email addresses,
 559  * but keep timestamps intact, as they influence things like traversal
 560  * order (and by themselves should not be too revealing).
 561  */
 562 static void anonymize_ident_line(const char **beg, const char **end)
 563 {
 564         static struct strbuf buffers[] = { STRBUF_INIT, STRBUF_INIT };
 565         static unsigned which_buffer;
 566
 567         struct strbuf *out;
 568         struct ident_split split;
 569         const char *end_of_header;
 570
 571         out = &buffers[which_buffer++];
 572         which_buffer %= ARRAY_SIZE(buffers);
 573         strbuf_reset(out);
 574
 575         /* skip "committer", "author", "tagger", etc */
 576         end_of_header = strchr(*beg, ' ');
 577         if (!end_of_header)
 578                 BUG("malformed line fed to anonymize_ident_line: %.*s",
 579                     (int)(*end - *beg), *beg);
 580         end_of_header++;
 581         strbuf_add(out, *beg, end_of_header - *beg);
 582
 583         if (!split_ident_line(&split, end_of_header, *end - end_of_header) &&
 584             split.date_begin) {
 585                 const char *ident;
 586                 size_t len;
 587
 588                 len = split.mail_end - split.name_begin;
 589                 ident = anonymize_mem(&idents, anonymize_ident,
 590                                       split.name_begin, &len);
 591                 strbuf_add(out, ident, len);
 592                 strbuf_addch(out, ' ');
 593                 strbuf_add(out, split.date_begin, split.tz_end - split.date_begin);
 594         } else {
 595                 strbuf_addstr(out, "Malformed Ident <malformed@example.com> 0 -0000");
 596         }
 597
 598         *beg = out->buf;
 599         *end = out->buf + out->len;
 600 }
 601
 602 static void handle_commit(struct commit *commit, struct rev_info *rev,
 603                           struct string_list *paths_of_changed_objects)
 604 {
 605         int saved_output_format = rev->diffopt.output_format;
 606         const char *commit_buffer;
 607         const char *author, *author_end, *committer, *committer_end;
 608         const char *encoding, *message;
 609         char *reencoded = NULL;
 610         struct commit_list *p;
 611         const char *refname;
 612         int i;
 613
 614         rev->diffopt.output_format = DIFF_FORMAT_CALLBACK;
 615
 616         parse_commit_or_die(commit);
 617         commit_buffer = get_commit_buffer(commit, NULL);
 618         author = strstr(commit_buffer, "\nauthor ");
 619         if (!author)
 620                 die("could not find author in commit %s",
 621                     oid_to_hex(&commit->object.oid));
 622         author++;
 623         author_end = strchrnul(author, '\n');
 624         committer = strstr(author_end, "\ncommitter ");
 625         if (!committer)
 626                 die("could not find committer in commit %s",
 627                     oid_to_hex(&commit->object.oid));
 628         committer++;
 629         committer_end = strchrnul(committer, '\n');
 630         message = strstr(committer_end, "\n\n");
 631         encoding = find_encoding(committer_end, message);
 632         if (message)
 633                 message += 2;
 634
 635         if (commit->parents &&
 636             (get_object_mark(&commit->parents->item->object) != 0 ||
 637              reference_excluded_commits) &&
 638             !full_tree) {
 639                 parse_commit_or_die(commit->parents->item);
 640                 diff_tree_oid(get_commit_tree_oid(commit->parents->item),
 641                               get_commit_tree_oid(commit), "", &rev->diffopt);
 642         }
 643         else
 644                 diff_root_tree_oid(get_commit_tree_oid(commit),
 645                                    "", &rev->diffopt);
 646
 647         /* Export the referenced blobs, and remember the marks. */
 648         for (i = 0; i < diff_queued_diff.nr; i++)
 649                 if (!S_ISGITLINK(diff_queued_diff.queue[i]->two->mode))
 650                         export_blob(&diff_queued_diff.queue[i]->two->oid);
 651
 652         refname = *revision_sources_at(&revision_sources, commit);
 653         /*
 654          * FIXME: string_list_remove() below for each ref is overall
 655          * O(N^2).  Compared to a history walk and diffing trees, this is
 656          * just lost in the noise in practice.  However, theoretically a
 657          * repo may have enough refs for this to become slow.
 658          */
 659         string_list_remove(&extra_refs, refname, 0);
 660         if (anonymize) {
 661                 refname = anonymize_refname(refname);
 662                 anonymize_ident_line(&committer, &committer_end);
 663                 anonymize_ident_line(&author, &author_end);
 664         }
 665
 666         mark_next_object(&commit->object);
 667         if (anonymize) {
 668                 reencoded = anonymize_commit_message(message);
 669         } else if (encoding) {
 670                 switch(reencode_mode) {
 671                 case REENCODE_YES:
 672                         reencoded = reencode_string(message, "UTF-8", encoding);
 673                         break;
 674                 case REENCODE_NO:
 675                         break;
 676                 case REENCODE_ABORT:
 677                         die("Encountered commit-specific encoding %s in commit "
 678                             "%s; use --reencode=[yes|no] to handle it",
 679                             encoding, oid_to_hex(&commit->object.oid));
 680                 }
 681         }
 682         if (!commit->parents)
 683                 printf("reset %s\n", refname);
 684         printf("commit %s\nmark :%"PRIu32"\n", refname, last_idnum);
 685         if (show_original_ids)
 686                 printf("original-oid %s\n", oid_to_hex(&commit->object.oid));
 687         printf("%.*s\n%.*s\n",
 688                (int)(author_end - author), author,
 689                (int)(committer_end - committer), committer);
 690         if (!reencoded && encoding)
 691                 printf("encoding %s\n", encoding);
 692         printf("data %u\n%s",
 693                (unsigned)(reencoded
 694                           ? strlen(reencoded) : message
 695                           ? strlen(message) : 0),
 696                reencoded ? reencoded : message ? message : "");
 697         free(reencoded);
 698         unuse_commit_buffer(commit, commit_buffer);
 699
 700         for (i = 0, p = commit->parents; p; p = p->next) {
 701                 struct object *obj = &p->item->object;
 702                 int mark = get_object_mark(obj);
 703
 704                 if (!mark && !reference_excluded_commits)
 705                         continue;
 706                 if (i == 0)
 707                         printf("from ");
 708                 else
 709                         printf("merge ");
 710                 if (mark)
 711                         printf(":%d\n", mark);
 712                 else
 713                         printf("%s\n", oid_to_hex(anonymize ?
 714                                                   anonymize_oid(&obj->oid) :
 715                                                   &obj->oid));
 716                 i++;
 717         }
 718
 719         if (full_tree)
 720                 printf("deleteall\n");
 721         log_tree_diff_flush(rev);
 722         string_list_clear(paths_of_changed_objects, 0);
 723         rev->diffopt.output_format = saved_output_format;
 724
 725         printf("\n");
 726
 727         show_progress();
 728 }
 729
 730 static void *anonymize_tag(const void *old, size_t *len)
 731 {
 732         static int counter;
 733         struct strbuf out = STRBUF_INIT;
 734         strbuf_addf(&out, "tag message %d", counter++);
 735         return strbuf_detach(&out, len);
 736 }
 737
 738 static void handle_tail(struct object_array *commits, struct rev_info *revs,
 739                         struct string_list *paths_of_changed_objects)
 740 {
 741         struct commit *commit;
 742         while (commits->nr) {
 743                 commit = (struct commit *)object_array_pop(commits);
 744                 if (has_unshown_parent(commit)) {
 745                         /* Queue again, to be handled later */
 746                         add_object_array(&commit->object, NULL, commits);
 747                         return;
 748                 }
 749                 handle_commit(commit, revs, paths_of_changed_objects);
 750         }
 751 }
 752
 753 static void handle_tag(const char *name, struct tag *tag)
 754 {
 755         unsigned long size;
 756         enum object_type type;
 757         char *buf;
 758         const char *tagger, *tagger_end, *message;
 759         size_t message_size = 0;
 760         struct object *tagged;
 761         int tagged_mark;
 762         struct commit *p;
 763
 764         /* Trees have no identifier in fast-export output, thus we have no way
 765          * to output tags of trees, tags of tags of trees, etc.  Simply omit
 766          * such tags.
 767          */
 768         tagged = tag->tagged;
 769         while (tagged->type == OBJ_TAG) {
 770                 tagged = ((struct tag *)tagged)->tagged;
 771         }
 772         if (tagged->type == OBJ_TREE) {
 773                 warning("Omitting tag %s,\nsince tags of trees (or tags of tags of trees, etc.) are not supported.",
 774                         oid_to_hex(&tag->object.oid));
 775                 return;
 776         }
 777
 778         buf = read_object_file(&tag->object.oid, &type, &size);
 779         if (!buf)
 780                 die("could not read tag %s", oid_to_hex(&tag->object.oid));
 781         message = memmem(buf, size, "\n\n", 2);
 782         if (message) {
 783                 message += 2;
 784                 message_size = strlen(message);
 785         }
 786         tagger = memmem(buf, message ? message - buf : size, "\ntagger ", 8);
 787         if (!tagger) {
 788                 if (fake_missing_tagger)
 789                         tagger = "tagger Unspecified Tagger "
 790                                 "<unspecified-tagger> 0 +0000";
 791                 else
 792                         tagger = "";
 793                 tagger_end = tagger + strlen(tagger);
 794         } else {
 795                 tagger++;
 796                 tagger_end = strchrnul(tagger, '\n');
 797                 if (anonymize)
 798                         anonymize_ident_line(&tagger, &tagger_end);
 799         }
 800
 801         if (anonymize) {
 802                 name = anonymize_refname(name);
 803                 if (message) {
 804                         static struct hashmap tags;
 805                         message = anonymize_mem(&tags, anonymize_tag,
 806                                                 message, &message_size);
 807                 }
 808         }
 809
 810         /* handle signed tags */
 811         if (message) {
 812                 const char *signature = strstr(message,
 813                                                "\n-----BEGIN PGP SIGNATURE-----\n");
 814                 if (signature)
 815                         switch(signed_tag_mode) {
 816                         case SIGNED_TAG_ABORT:
 817                                 die("encountered signed tag %s; use "
 818                                     "--signed-tags=<mode> to handle it",
 819                                     oid_to_hex(&tag->object.oid));
 820                         case WARN:
 821                                 warning("exporting signed tag %s",
 822                                         oid_to_hex(&tag->object.oid));
 823                                 /* fallthru */
 824                         case VERBATIM:
 825                                 break;
 826                         case WARN_STRIP:
 827                                 warning("stripping signature from tag %s",
 828                                         oid_to_hex(&tag->object.oid));
 829                                 /* fallthru */
 830                         case STRIP:
 831                                 message_size = signature + 1 - message;
 832                                 break;
 833                         }
 834         }
 835
 836         /* handle tag->tagged having been filtered out due to paths specified */
 837         tagged = tag->tagged;
 838         tagged_mark = get_object_mark(tagged);
 839         if (!tagged_mark) {
 840                 switch(tag_of_filtered_mode) {
 841                 case TAG_FILTERING_ABORT:
 842                         die("tag %s tags unexported object; use "
 843                             "--tag-of-filtered-object=<mode> to handle it",
 844                             oid_to_hex(&tag->object.oid));
 845                 case DROP:
 846                         /* Ignore this tag altogether */
 847                         free(buf);
 848                         return;
 849                 case REWRITE:
 850                         if (tagged->type != OBJ_COMMIT) {
 851                                 die("tag %s tags unexported %s!",
 852                                     oid_to_hex(&tag->object.oid),
 853                                     type_name(tagged->type));
 854                         }
 855                         p = rewrite_commit((struct commit *)tagged);
 856                         if (!p) {
 857                                 printf("reset %s\nfrom %s\n\n",
 858                                        name, oid_to_hex(&null_oid));
 859                                 free(buf);
 860                                 return;
 861                         }
 862                         tagged_mark = get_object_mark(&p->object);
 863                 }
 864         }
 865
 866         if (starts_with(name, "refs/tags/"))
 867                 name += 10;
 868         printf("tag %s\nfrom :%d\n", name, tagged_mark);
 869         if (show_original_ids)
 870                 printf("original-oid %s\n", oid_to_hex(&tag->object.oid));
 871         printf("%.*s%sdata %d\n%.*s\n",
 872                (int)(tagger_end - tagger), tagger,
 873                tagger == tagger_end ? "" : "\n",
 874                (int)message_size, (int)message_size, message ? message : "");
 875         free(buf);
 876 }
 877
 878 static struct commit *get_commit(struct rev_cmdline_entry *e, char *full_name)
 879 {
 880         switch (e->item->type) {
 881         case OBJ_COMMIT:
 882                 return (struct commit *)e->item;
 883         case OBJ_TAG: {
 884                 struct tag *tag = (struct tag *)e->item;
 885
 886                 /* handle nested tags */
 887                 while (tag && tag->object.type == OBJ_TAG) {
 888                         parse_object(the_repository, &tag->object.oid);
 889                         string_list_append(&tag_refs, full_name)->util = tag;
 890                         tag = (struct tag *)tag->tagged;
 891                 }
 892                 if (!tag)
 893                         die("Tag %s points nowhere?", e->name);
 894                 return (struct commit *)tag;
 895                 break;
 896         }
 897         default:
 898                 return NULL;
 899         }
 900 }
 901
 902 static void get_tags_and_duplicates(struct rev_cmdline_info *info)
 903 {
 904         int i;
 905
 906         for (i = 0; i < info->nr; i++) {
 907                 struct rev_cmdline_entry *e = info->rev + i;
 908                 struct object_id oid;
 909                 struct commit *commit;
 910                 char *full_name;
 911
 912                 if (e->flags & UNINTERESTING)
 913                         continue;
 914
 915                 if (dwim_ref(e->name, strlen(e->name), &oid, &full_name) != 1)
 916                         continue;
 917
 918                 if (refspecs.nr) {
 919                         char *private;
 920                         private = apply_refspecs(&refspecs, full_name);
 921                         if (private) {
 922                                 free(full_name);
 923                                 full_name = private;
 924                         }
 925                 }
 926
 927                 commit = get_commit(e, full_name);
 928                 if (!commit) {
 929                         warning("%s: Unexpected object of type %s, skipping.",
 930                                 e->name,
 931                                 type_name(e->item->type));
 932                         continue;
 933                 }
 934
 935                 switch(commit->object.type) {
 936                 case OBJ_COMMIT:
 937                         break;
 938                 case OBJ_BLOB:
 939                         export_blob(&commit->object.oid);
 940                         continue;
 941                 default: /* OBJ_TAG (nested tags) is already handled */
 942                         warning("Tag points to object of unexpected type %s, skipping.",
 943                                 type_name(commit->object.type));
 944                         continue;
 945                 }
 946
 947                 /*
 948                  * Make sure this ref gets properly updated eventually, whether
 949                  * through a commit or manually at the end.
 950                  */
 951                 if (e->item->type != OBJ_TAG)
 952                         string_list_append(&extra_refs, full_name)->util = commit;
 953
 954                 if (!*revision_sources_at(&revision_sources, commit))
 955                         *revision_sources_at(&revision_sources, commit) = full_name;
 956         }
 957
 958         string_list_sort(&extra_refs);
 959         string_list_remove_duplicates(&extra_refs, 0);
 960 }
 961
 962 static void handle_tags_and_duplicates(struct string_list *extras)
 963 {
 964         struct commit *commit;
 965         int i;
 966
 967         for (i = extras->nr - 1; i >= 0; i--) {
 968                 const char *name = extras->items[i].string;
 969                 struct object *object = extras->items[i].util;
 970                 int mark;
 971
 972                 switch (object->type) {
 973                 case OBJ_TAG:
 974                         handle_tag(name, (struct tag *)object);
 975                         break;
 976                 case OBJ_COMMIT:
 977                         if (anonymize)
 978                                 name = anonymize_refname(name);
 979                         /* create refs pointing to already seen commits */
 980                         commit = rewrite_commit((struct commit *)object);
 981                         if (!commit) {
 982                                 /*
 983                                  * Neither this object nor any of its
 984                                  * ancestors touch any relevant paths, so
 985                                  * it has been filtered to nothing.  Delete
 986                                  * it.
 987                                  */
 988                                 printf("reset %s\nfrom %s\n\n",
 989                                        name, oid_to_hex(&null_oid));
 990                                 continue;
 991                         }
 992
 993                         mark = get_object_mark(&commit->object);
 994                         if (!mark) {
 995                                 /*
 996                                  * Getting here means we have a commit which
 997                                  * was excluded by a negative refspec (e.g.
 998                                  * fast-export ^master master).  If we are
 999                                  * referencing excluded commits, set the ref
1000                                  * to the exact commit.  Otherwise, the user
1001                                  * wants the branch exported but every commit
1002                                  * in its history to be deleted, which basically
1003                                  * just means deletion of the ref.
1004                                  */
1005                                 if (!reference_excluded_commits) {
1006                                         /* delete the ref */
1007                                         printf("reset %s\nfrom %s\n\n",
1008                                                name, oid_to_hex(&null_oid));
1009                                         continue;
1010                                 }
1011                                 /* set ref to commit using oid, not mark */
1012                                 printf("reset %s\nfrom %s\n\n", name,
1013                                        oid_to_hex(&commit->object.oid));
1014                                 continue;
1015                         }
1016
1017                         printf("reset %s\nfrom :%d\n\n", name, mark
1018                                );
1019                         show_progress();
1020                         break;
1021                 }
1022         }
1023 }
1024
1025 static void export_marks(char *file)
1026 {
1027         unsigned int i;
1028         uint32_t mark;
1029         struct decoration_entry *deco = idnums.entries;
1030         FILE *f;
1031         int e = 0;
1032
1033         f = fopen_for_writing(file);
1034         if (!f)
1035                 die_errno("Unable to open marks file %s for writing.", file);
1036
1037         for (i = 0; i < idnums.size; i++) {
1038                 if (deco->base && deco->base->type == 1) {
1039                         mark = ptr_to_mark(deco->decoration);
1040                         if (fprintf(f, ":%"PRIu32" %s\n", mark,
1041                                 oid_to_hex(&deco->base->oid)) < 0) {
1042                             e = 1;
1043                             break;
1044                         }
1045                 }
1046                 deco++;
1047         }
1048
1049         e |= ferror(f);
1050         e |= fclose(f);
1051         if (e)
1052                 error("Unable to write marks file %s.", file);
1053 }
1054
1055 static void import_marks(char *input_file)
1056 {
1057         char line[512];
1058         FILE *f = xfopen(input_file, "r");
1059
1060         while (fgets(line, sizeof(line), f)) {
1061                 uint32_t mark;
1062                 char *line_end, *mark_end;
1063                 struct object_id oid;
1064                 struct object *object;
1065                 struct commit *commit;
1066                 enum object_type type;
1067
1068                 line_end = strchr(line, '\n');
1069                 if (line[0] != ':' || !line_end)
1070                         die("corrupt mark line: %s", line);
1071                 *line_end = '\0';
1072
1073                 mark = strtoumax(line + 1, &mark_end, 10);
1074                 if (!mark || mark_end == line + 1
1075                         || *mark_end != ' ' || get_oid_hex(mark_end + 1, &oid))
1076                         die("corrupt mark line: %s", line);
1077
1078                 if (last_idnum < mark)
1079                         last_idnum = mark;
1080
1081                 type = oid_object_info(the_repository, &oid, NULL);
1082                 if (type < 0)
1083                         die("object not found: %s", oid_to_hex(&oid));
1084
1085                 if (type != OBJ_COMMIT)
1086                         /* only commits */
1087                         continue;
1088
1089                 commit = lookup_commit(the_repository, &oid);
1090                 if (!commit)
1091                         die("not a commit? can't happen: %s", oid_to_hex(&oid));
1092
1093                 object = &commit->object;
1094
1095                 if (object->flags & SHOWN)
1096                         error("Object %s already has a mark", oid_to_hex(&oid));
1097
1098                 mark_object(object, mark);
1099
1100                 object->flags |= SHOWN;
1101         }
1102         fclose(f);
1103 }
1104
1105 static void handle_deletes(void)
1106 {
1107         int i;
1108         for (i = 0; i < refspecs.nr; i++) {
1109                 struct refspec_item *refspec = &refspecs.items[i];
1110                 if (*refspec->src)
1111                         continue;
1112
1113                 printf("reset %s\nfrom %s\n\n",
1114                                 refspec->dst, oid_to_hex(&null_oid));
1115         }
1116 }
1117
1118 int cmd_fast_export(int argc, const char **argv, const char *prefix)
1119 {
1120         struct rev_info revs;
1121         struct object_array commits = OBJECT_ARRAY_INIT;
1122         struct commit *commit;
1123         char *export_filename = NULL, *import_filename = NULL;
1124         uint32_t lastimportid;
1125         struct string_list refspecs_list = STRING_LIST_INIT_NODUP;
1126         struct string_list paths_of_changed_objects = STRING_LIST_INIT_DUP;
1127         struct option options[] = {
1128                 OPT_INTEGER(0, "progress", &progress,
1129                             N_("show progress after <n> objects")),
1130                 OPT_CALLBACK(0, "signed-tags", &signed_tag_mode, N_("mode"),
1131                              N_("select handling of signed tags"),
1132                              parse_opt_signed_tag_mode),
1133                 OPT_CALLBACK(0, "tag-of-filtered-object", &tag_of_filtered_mode, N_("mode"),
1134                              N_("select handling of tags that tag filtered objects"),
1135                              parse_opt_tag_of_filtered_mode),
1136                 OPT_CALLBACK(0, "reencode", &reencode_mode, N_("mode"),
1137                              N_("select handling of commit messages in an alternate encoding"),
1138                              parse_opt_reencode_mode),
1139                 OPT_STRING(0, "export-marks", &export_filename, N_("file"),
1140                              N_("Dump marks to this file")),
1141                 OPT_STRING(0, "import-marks", &import_filename, N_("file"),
1142                              N_("Import marks from this file")),
1143                 OPT_BOOL(0, "fake-missing-tagger", &fake_missing_tagger,
1144                          N_("Fake a tagger when tags lack one")),
1145                 OPT_BOOL(0, "full-tree", &full_tree,
1146                          N_("Output full tree for each commit")),
1147                 OPT_BOOL(0, "use-done-feature", &use_done_feature,
1148                              N_("Use the done feature to terminate the stream")),
1149                 OPT_BOOL(0, "no-data", &no_data, N_("Skip output of blob data")),
1150                 OPT_STRING_LIST(0, "refspec", &refspecs_list, N_("refspec"),
1151                              N_("Apply refspec to exported refs")),
1152                 OPT_BOOL(0, "anonymize", &anonymize, N_("anonymize output")),
1153                 OPT_BOOL(0, "reference-excluded-parents",
1154                          &reference_excluded_commits, N_("Reference parents which are not in fast-export stream by object id")),
1155                 OPT_BOOL(0, "show-original-ids", &show_original_ids,
1156                             N_("Show original object ids of blobs/commits")),
1157
1158                 OPT_END()
1159         };
1160
1161         if (argc == 1)
1162                 usage_with_options (fast_export_usage, options);
1163
1164         /* we handle encodings */
1165         git_config(git_default_config, NULL);
1166
1167         repo_init_revisions(the_repository, &revs, prefix);
1168         init_revision_sources(&revision_sources);
1169         revs.topo_order = 1;
1170         revs.sources = &revision_sources;
1171         revs.rewrite_parents = 1;
1172         argc = parse_options(argc, argv, prefix, options, fast_export_usage,
1173                         PARSE_OPT_KEEP_ARGV0 | PARSE_OPT_KEEP_UNKNOWN);
1174         argc = setup_revisions(argc, argv, &revs, NULL);
1175         if (argc > 1)
1176                 usage_with_options (fast_export_usage, options);
1177
1178         if (refspecs_list.nr) {
1179                 int i;
1180
1181                 for (i = 0; i < refspecs_list.nr; i++)
1182                         refspec_append(&refspecs, refspecs_list.items[i].string);
1183
1184                 string_list_clear(&refspecs_list, 1);
1185         }
1186
1187         if (use_done_feature)
1188                 printf("feature done\n");
1189
1190         if (import_filename)
1191                 import_marks(import_filename);
1192         lastimportid = last_idnum;
1193
1194         if (import_filename && revs.prune_data.nr)
1195                 full_tree = 1;
1196
1197         get_tags_and_duplicates(&revs.cmdline);
1198
1199         if (prepare_revision_walk(&revs))
1200                 die("revision walk setup failed");
1201         revs.diffopt.format_callback = show_filemodify;
1202         revs.diffopt.format_callback_data = &paths_of_changed_objects;
1203         revs.diffopt.flags.recursive = 1;
1204         while ((commit = get_revision(&revs))) {
1205                 if (has_unshown_parent(commit)) {
1206                         add_object_array(&commit->object, NULL, &commits);
1207                 }
1208                 else {
1209                         handle_commit(commit, &revs, &paths_of_changed_objects);
1210                         handle_tail(&commits, &revs, &paths_of_changed_objects);
1211                 }
1212         }
1213
1214         handle_tags_and_duplicates(&extra_refs);
1215         handle_tags_and_duplicates(&tag_refs);
1216         handle_deletes();
1217
1218         if (export_filename && lastimportid != last_idnum)
1219                 export_marks(export_filename);
1220
1221         if (use_done_feature)
1222                 printf("done\n");
1223
1224         refspec_clear(&refspecs);
1225
1226         return 0;
1227 }