commit.c

   1 #include "cache.h"
   2 #include "tag.h"
   3 #include "commit.h"
   4 #include "commit-graph.h"
   5 #include "environment.h"
   6 #include "gettext.h"
   7 #include "hex.h"
   8 #include "repository.h"
   9 #include "object-store.h"
  10 #include "pkt-line.h"
  11 #include "utf8.h"
  12 #include "diff.h"
  13 #include "revision.h"
  14 #include "notes.h"
  15 #include "alloc.h"
  16 #include "gpg-interface.h"
  17 #include "mergesort.h"
  18 #include "commit-slab.h"
  19 #include "prio-queue.h"
  20 #include "hash-lookup.h"
  21 #include "wt-status.h"
  22 #include "advice.h"
  23 #include "refs.h"
  24 #include "commit-reach.h"
  25 #include "run-command.h"
  26 #include "setup.h"
  27 #include "shallow.h"
  28 #include "hook.h"
  29
  30 static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
  31
  32 int save_commit_buffer = 1;
  33 int no_graft_file_deprecated_advice;
  34
  35 const char *commit_type = "commit";
  36
  37 struct commit *lookup_commit_reference_gently(struct repository *r,
  38                 const struct object_id *oid, int quiet)
  39 {
  40         struct object *obj = deref_tag(r,
  41                                        parse_object(r, oid),
  42                                        NULL, 0);
  43
  44         if (!obj)
  45                 return NULL;
  46         return object_as_type(obj, OBJ_COMMIT, quiet);
  47 }
  48
  49 struct commit *lookup_commit_reference(struct repository *r, const struct object_id *oid)
  50 {
  51         return lookup_commit_reference_gently(r, oid, 0);
  52 }
  53
  54 struct commit *lookup_commit_or_die(const struct object_id *oid, const char *ref_name)
  55 {
  56         struct commit *c = lookup_commit_reference(the_repository, oid);
  57         if (!c)
  58                 die(_("could not parse %s"), ref_name);
  59         if (!oideq(oid, &c->object.oid)) {
  60                 warning(_("%s %s is not a commit!"),
  61                         ref_name, oid_to_hex(oid));
  62         }
  63         return c;
  64 }
  65
  66 struct commit *lookup_commit_object(struct repository *r,
  67                                     const struct object_id *oid)
  68 {
  69         struct object *obj = parse_object(r, oid);
  70         return obj ? object_as_type(obj, OBJ_COMMIT, 0) : NULL;
  71
  72 }
  73
  74 struct commit *lookup_commit(struct repository *r, const struct object_id *oid)
  75 {
  76         struct object *obj = lookup_object(r, oid);
  77         if (!obj)
  78                 return create_object(r, oid, alloc_commit_node(r));
  79         return object_as_type(obj, OBJ_COMMIT, 0);
  80 }
  81
  82 struct commit *lookup_commit_reference_by_name(const char *name)
  83 {
  84         struct object_id oid;
  85         struct commit *commit;
  86
  87         if (repo_get_oid_committish(the_repository, name, &oid))
  88                 return NULL;
  89         commit = lookup_commit_reference(the_repository, &oid);
  90         if (repo_parse_commit(the_repository, commit))
  91                 return NULL;
  92         return commit;
  93 }
  94
  95 static timestamp_t parse_commit_date(const char *buf, const char *tail)
  96 {
  97         const char *dateptr;
  98
  99         if (buf + 6 >= tail)
 100                 return 0;
 101         if (memcmp(buf, "author", 6))
 102                 return 0;
 103         while (buf < tail && *buf++ != '\n')
 104                 /* nada */;
 105         if (buf + 9 >= tail)
 106                 return 0;
 107         if (memcmp(buf, "committer", 9))
 108                 return 0;
 109         while (buf < tail && *buf++ != '>')
 110                 /* nada */;
 111         if (buf >= tail)
 112                 return 0;
 113         dateptr = buf;
 114         while (buf < tail && *buf++ != '\n')
 115                 /* nada */;
 116         if (buf >= tail)
 117                 return 0;
 118         /* dateptr < buf && buf[-1] == '\n', so parsing will stop at buf-1 */
 119         return parse_timestamp(dateptr, NULL, 10);
 120 }
 121
 122 static const struct object_id *commit_graft_oid_access(size_t index, const void *table)
 123 {
 124         const struct commit_graft * const *commit_graft_table = table;
 125         return &commit_graft_table[index]->oid;
 126 }
 127
 128 int commit_graft_pos(struct repository *r, const struct object_id *oid)
 129 {
 130         return oid_pos(oid, r->parsed_objects->grafts,
 131                        r->parsed_objects->grafts_nr,
 132                        commit_graft_oid_access);
 133 }
 134
 135 static void unparse_commit(struct repository *r, const struct object_id *oid)
 136 {
 137         struct commit *c = lookup_commit(r, oid);
 138
 139         if (!c->object.parsed)
 140                 return;
 141         free_commit_list(c->parents);
 142         c->parents = NULL;
 143         c->object.parsed = 0;
 144 }
 145
 146 int register_commit_graft(struct repository *r, struct commit_graft *graft,
 147                           int ignore_dups)
 148 {
 149         int pos = commit_graft_pos(r, &graft->oid);
 150
 151         if (0 <= pos) {
 152                 if (ignore_dups)
 153                         free(graft);
 154                 else {
 155                         free(r->parsed_objects->grafts[pos]);
 156                         r->parsed_objects->grafts[pos] = graft;
 157                 }
 158                 return 1;
 159         }
 160         pos = -pos - 1;
 161         ALLOC_GROW(r->parsed_objects->grafts,
 162                    r->parsed_objects->grafts_nr + 1,
 163                    r->parsed_objects->grafts_alloc);
 164         r->parsed_objects->grafts_nr++;
 165         if (pos < r->parsed_objects->grafts_nr)
 166                 memmove(r->parsed_objects->grafts + pos + 1,
 167                         r->parsed_objects->grafts + pos,
 168                         (r->parsed_objects->grafts_nr - pos - 1) *
 169                         sizeof(*r->parsed_objects->grafts));
 170         r->parsed_objects->grafts[pos] = graft;
 171         unparse_commit(r, &graft->oid);
 172         return 0;
 173 }
 174
 175 struct commit_graft *read_graft_line(struct strbuf *line)
 176 {
 177         /* The format is just "Commit Parent1 Parent2 ...\n" */
 178         int i, phase;
 179         const char *tail = NULL;
 180         struct commit_graft *graft = NULL;
 181         struct object_id dummy_oid, *oid;
 182
 183         strbuf_rtrim(line);
 184         if (!line->len || line->buf[0] == '#')
 185                 return NULL;
 186         /*
 187          * phase 0 verifies line, counts hashes in line and allocates graft
 188          * phase 1 fills graft
 189          */
 190         for (phase = 0; phase < 2; phase++) {
 191                 oid = graft ? &graft->oid : &dummy_oid;
 192                 if (parse_oid_hex(line->buf, oid, &tail))
 193                         goto bad_graft_data;
 194                 for (i = 0; *tail != '\0'; i++) {
 195                         oid = graft ? &graft->parent[i] : &dummy_oid;
 196                         if (!isspace(*tail++) || parse_oid_hex(tail, oid, &tail))
 197                                 goto bad_graft_data;
 198                 }
 199                 if (!graft) {
 200                         graft = xmalloc(st_add(sizeof(*graft),
 201                                                st_mult(sizeof(struct object_id), i)));
 202                         graft->nr_parent = i;
 203                 }
 204         }
 205         return graft;
 206
 207 bad_graft_data:
 208         error("bad graft data: %s", line->buf);
 209         assert(!graft);
 210         return NULL;
 211 }
 212
 213 static int read_graft_file(struct repository *r, const char *graft_file)
 214 {
 215         FILE *fp = fopen_or_warn(graft_file, "r");
 216         struct strbuf buf = STRBUF_INIT;
 217         if (!fp)
 218                 return -1;
 219         if (!no_graft_file_deprecated_advice &&
 220             advice_enabled(ADVICE_GRAFT_FILE_DEPRECATED))
 221                 advise(_("Support for <GIT_DIR>/info/grafts is deprecated\n"
 222                          "and will be removed in a future Git version.\n"
 223                          "\n"
 224                          "Please use \"git replace --convert-graft-file\"\n"
 225                          "to convert the grafts into replace refs.\n"
 226                          "\n"
 227                          "Turn this message off by running\n"
 228                          "\"git config advice.graftFileDeprecated false\""));
 229         while (!strbuf_getwholeline(&buf, fp, '\n')) {
 230                 /* The format is just "Commit Parent1 Parent2 ...\n" */
 231                 struct commit_graft *graft = read_graft_line(&buf);
 232                 if (!graft)
 233                         continue;
 234                 if (register_commit_graft(r, graft, 1))
 235                         error("duplicate graft data: %s", buf.buf);
 236         }
 237         fclose(fp);
 238         strbuf_release(&buf);
 239         return 0;
 240 }
 241
 242 void prepare_commit_graft(struct repository *r)
 243 {
 244         char *graft_file;
 245
 246         if (r->parsed_objects->commit_graft_prepared)
 247                 return;
 248         if (!startup_info->have_repository)
 249                 return;
 250
 251         graft_file = get_graft_file(r);
 252         read_graft_file(r, graft_file);
 253         /* make sure shallows are read */
 254         is_repository_shallow(r);
 255         r->parsed_objects->commit_graft_prepared = 1;
 256 }
 257
 258 struct commit_graft *lookup_commit_graft(struct repository *r, const struct object_id *oid)
 259 {
 260         int pos;
 261         prepare_commit_graft(r);
 262         pos = commit_graft_pos(r, oid);
 263         if (pos < 0)
 264                 return NULL;
 265         return r->parsed_objects->grafts[pos];
 266 }
 267
 268 int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
 269 {
 270         int i, ret;
 271         for (i = ret = 0; i < the_repository->parsed_objects->grafts_nr && !ret; i++)
 272                 ret = fn(the_repository->parsed_objects->grafts[i], cb_data);
 273         return ret;
 274 }
 275
 276 void reset_commit_grafts(struct repository *r)
 277 {
 278         int i;
 279
 280         for (i = 0; i < r->parsed_objects->grafts_nr; i++) {
 281                 unparse_commit(r, &r->parsed_objects->grafts[i]->oid);
 282                 free(r->parsed_objects->grafts[i]);
 283         }
 284         r->parsed_objects->grafts_nr = 0;
 285         r->parsed_objects->commit_graft_prepared = 0;
 286 }
 287
 288 struct commit_buffer {
 289         void *buffer;
 290         unsigned long size;
 291 };
 292 define_commit_slab(buffer_slab, struct commit_buffer);
 293
 294 struct buffer_slab *allocate_commit_buffer_slab(void)
 295 {
 296         struct buffer_slab *bs = xmalloc(sizeof(*bs));
 297         init_buffer_slab(bs);
 298         return bs;
 299 }
 300
 301 void free_commit_buffer_slab(struct buffer_slab *bs)
 302 {
 303         clear_buffer_slab(bs);
 304         free(bs);
 305 }
 306
 307 void set_commit_buffer(struct repository *r, struct commit *commit, void *buffer, unsigned long size)
 308 {
 309         struct commit_buffer *v = buffer_slab_at(
 310                 r->parsed_objects->buffer_slab, commit);
 311         v->buffer = buffer;
 312         v->size = size;
 313 }
 314
 315 const void *get_cached_commit_buffer(struct repository *r, const struct commit *commit, unsigned long *sizep)
 316 {
 317         struct commit_buffer *v = buffer_slab_peek(
 318                 r->parsed_objects->buffer_slab, commit);
 319         if (!v) {
 320                 if (sizep)
 321                         *sizep = 0;
 322                 return NULL;
 323         }
 324         if (sizep)
 325                 *sizep = v->size;
 326         return v->buffer;
 327 }
 328
 329 const void *repo_get_commit_buffer(struct repository *r,
 330                                    const struct commit *commit,
 331                                    unsigned long *sizep)
 332 {
 333         const void *ret = get_cached_commit_buffer(r, commit, sizep);
 334         if (!ret) {
 335                 enum object_type type;
 336                 unsigned long size;
 337                 ret = repo_read_object_file(r, &commit->object.oid, &type, &size);
 338                 if (!ret)
 339                         die("cannot read commit object %s",
 340                             oid_to_hex(&commit->object.oid));
 341                 if (type != OBJ_COMMIT)
 342                         die("expected commit for %s, got %s",
 343                             oid_to_hex(&commit->object.oid), type_name(type));
 344                 if (sizep)
 345                         *sizep = size;
 346         }
 347         return ret;
 348 }
 349
 350 void repo_unuse_commit_buffer(struct repository *r,
 351                               const struct commit *commit,
 352                               const void *buffer)
 353 {
 354         struct commit_buffer *v = buffer_slab_peek(
 355                 r->parsed_objects->buffer_slab, commit);
 356         if (!(v && v->buffer == buffer))
 357                 free((void *)buffer);
 358 }
 359
 360 void free_commit_buffer(struct parsed_object_pool *pool, struct commit *commit)
 361 {
 362         struct commit_buffer *v = buffer_slab_peek(
 363                 pool->buffer_slab, commit);
 364         if (v) {
 365                 FREE_AND_NULL(v->buffer);
 366                 v->size = 0;
 367         }
 368 }
 369
 370 static inline void set_commit_tree(struct commit *c, struct tree *t)
 371 {
 372         c->maybe_tree = t;
 373 }
 374
 375 struct tree *repo_get_commit_tree(struct repository *r,
 376                                   const struct commit *commit)
 377 {
 378         if (commit->maybe_tree || !commit->object.parsed)
 379                 return commit->maybe_tree;
 380
 381         if (commit_graph_position(commit) != COMMIT_NOT_FROM_GRAPH)
 382                 return get_commit_tree_in_graph(r, commit);
 383
 384         return NULL;
 385 }
 386
 387 struct object_id *get_commit_tree_oid(const struct commit *commit)
 388 {
 389         struct tree *tree = repo_get_commit_tree(the_repository, commit);
 390         return tree ? &tree->object.oid : NULL;
 391 }
 392
 393 void release_commit_memory(struct parsed_object_pool *pool, struct commit *c)
 394 {
 395         set_commit_tree(c, NULL);
 396         free_commit_buffer(pool, c);
 397         c->index = 0;
 398         free_commit_list(c->parents);
 399
 400         c->object.parsed = 0;
 401 }
 402
 403 const void *detach_commit_buffer(struct commit *commit, unsigned long *sizep)
 404 {
 405         struct commit_buffer *v = buffer_slab_peek(
 406                 the_repository->parsed_objects->buffer_slab, commit);
 407         void *ret;
 408
 409         if (!v) {
 410                 if (sizep)
 411                         *sizep = 0;
 412                 return NULL;
 413         }
 414         ret = v->buffer;
 415         if (sizep)
 416                 *sizep = v->size;
 417
 418         v->buffer = NULL;
 419         v->size = 0;
 420         return ret;
 421 }
 422
 423 int parse_commit_buffer(struct repository *r, struct commit *item, const void *buffer, unsigned long size, int check_graph)
 424 {
 425         const char *tail = buffer;
 426         const char *bufptr = buffer;
 427         struct object_id parent;
 428         struct commit_list **pptr;
 429         struct commit_graft *graft;
 430         const int tree_entry_len = the_hash_algo->hexsz + 5;
 431         const int parent_entry_len = the_hash_algo->hexsz + 7;
 432         struct tree *tree;
 433
 434         if (item->object.parsed)
 435                 return 0;
 436         /*
 437          * Presumably this is leftover from an earlier failed parse;
 438          * clear it out in preparation for us re-parsing (we'll hit the
 439          * same error, but that's good, since it lets our caller know
 440          * the result cannot be trusted.
 441          */
 442         free_commit_list(item->parents);
 443         item->parents = NULL;
 444
 445         tail += size;
 446         if (tail <= bufptr + tree_entry_len + 1 || memcmp(bufptr, "tree ", 5) ||
 447                         bufptr[tree_entry_len] != '\n')
 448                 return error("bogus commit object %s", oid_to_hex(&item->object.oid));
 449         if (get_oid_hex(bufptr + 5, &parent) < 0)
 450                 return error("bad tree pointer in commit %s",
 451                              oid_to_hex(&item->object.oid));
 452         tree = lookup_tree(r, &parent);
 453         if (!tree)
 454                 return error("bad tree pointer %s in commit %s",
 455                              oid_to_hex(&parent),
 456                              oid_to_hex(&item->object.oid));
 457         set_commit_tree(item, tree);
 458         bufptr += tree_entry_len + 1; /* "tree " + "hex sha1" + "\n" */
 459         pptr = &item->parents;
 460
 461         graft = lookup_commit_graft(r, &item->object.oid);
 462         if (graft)
 463                 r->parsed_objects->substituted_parent = 1;
 464         while (bufptr + parent_entry_len < tail && !memcmp(bufptr, "parent ", 7)) {
 465                 struct commit *new_parent;
 466
 467                 if (tail <= bufptr + parent_entry_len + 1 ||
 468                     get_oid_hex(bufptr + 7, &parent) ||
 469                     bufptr[parent_entry_len] != '\n')
 470                         return error("bad parents in commit %s", oid_to_hex(&item->object.oid));
 471                 bufptr += parent_entry_len + 1;
 472                 /*
 473                  * The clone is shallow if nr_parent < 0, and we must
 474                  * not traverse its real parents even when we unhide them.
 475                  */
 476                 if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
 477                         continue;
 478                 new_parent = lookup_commit(r, &parent);
 479                 if (!new_parent)
 480                         return error("bad parent %s in commit %s",
 481                                      oid_to_hex(&parent),
 482                                      oid_to_hex(&item->object.oid));
 483                 pptr = &commit_list_insert(new_parent, pptr)->next;
 484         }
 485         if (graft) {
 486                 int i;
 487                 struct commit *new_parent;
 488                 for (i = 0; i < graft->nr_parent; i++) {
 489                         new_parent = lookup_commit(r,
 490                                                    &graft->parent[i]);
 491                         if (!new_parent)
 492                                 return error("bad graft parent %s in commit %s",
 493                                              oid_to_hex(&graft->parent[i]),
 494                                              oid_to_hex(&item->object.oid));
 495                         pptr = &commit_list_insert(new_parent, pptr)->next;
 496                 }
 497         }
 498         item->date = parse_commit_date(bufptr, tail);
 499
 500         if (check_graph)
 501                 load_commit_graph_info(r, item);
 502
 503         item->object.parsed = 1;
 504         return 0;
 505 }
 506
 507 int repo_parse_commit_internal(struct repository *r,
 508                                struct commit *item,
 509                                int quiet_on_missing,
 510                                int use_commit_graph)
 511 {
 512         enum object_type type;
 513         void *buffer;
 514         unsigned long size;
 515         struct object_info oi = {
 516                 .typep = &type,
 517                 .sizep = &size,
 518                 .contentp = &buffer,
 519         };
 520         /*
 521          * Git does not support partial clones that exclude commits, so set
 522          * OBJECT_INFO_SKIP_FETCH_OBJECT to fail fast when an object is missing.
 523          */
 524         int flags = OBJECT_INFO_LOOKUP_REPLACE | OBJECT_INFO_SKIP_FETCH_OBJECT |
 525                 OBJECT_INFO_DIE_IF_CORRUPT;
 526         int ret;
 527
 528         if (!item)
 529                 return -1;
 530         if (item->object.parsed)
 531                 return 0;
 532         if (use_commit_graph && parse_commit_in_graph(r, item))
 533                 return 0;
 534
 535         if (oid_object_info_extended(r, &item->object.oid, &oi, flags) < 0)
 536                 return quiet_on_missing ? -1 :
 537                         error("Could not read %s",
 538                              oid_to_hex(&item->object.oid));
 539         if (type != OBJ_COMMIT) {
 540                 free(buffer);
 541                 return error("Object %s not a commit",
 542                              oid_to_hex(&item->object.oid));
 543         }
 544
 545         ret = parse_commit_buffer(r, item, buffer, size, 0);
 546         if (save_commit_buffer && !ret) {
 547                 set_commit_buffer(r, item, buffer, size);
 548                 return 0;
 549         }
 550         free(buffer);
 551         return ret;
 552 }
 553
 554 int repo_parse_commit_gently(struct repository *r,
 555                              struct commit *item, int quiet_on_missing)
 556 {
 557         return repo_parse_commit_internal(r, item, quiet_on_missing, 1);
 558 }
 559
 560 void parse_commit_or_die(struct commit *item)
 561 {
 562         if (repo_parse_commit(the_repository, item))
 563                 die("unable to parse commit %s",
 564                     item ? oid_to_hex(&item->object.oid) : "(null)");
 565 }
 566
 567 int find_commit_subject(const char *commit_buffer, const char **subject)
 568 {
 569         const char *eol;
 570         const char *p = commit_buffer;
 571
 572         while (*p && (*p != '\n' || p[1] != '\n'))
 573                 p++;
 574         if (*p) {
 575                 p = skip_blank_lines(p + 2);
 576                 eol = strchrnul(p, '\n');
 577         } else
 578                 eol = p;
 579
 580         *subject = p;
 581
 582         return eol - p;
 583 }
 584
 585 size_t commit_subject_length(const char *body)
 586 {
 587         const char *p = body;
 588         while (*p) {
 589                 const char *next = skip_blank_lines(p);
 590                 if (next != p)
 591                         break;
 592                 p = strchrnul(p, '\n');
 593                 if (*p)
 594                         p++;
 595         }
 596         return p - body;
 597 }
 598
 599 struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
 600 {
 601         struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
 602         new_list->item = item;
 603         new_list->next = *list_p;
 604         *list_p = new_list;
 605         return new_list;
 606 }
 607
 608 int commit_list_contains(struct commit *item, struct commit_list *list)
 609 {
 610         while (list) {
 611                 if (list->item == item)
 612                         return 1;
 613                 list = list->next;
 614         }
 615
 616         return 0;
 617 }
 618
 619 unsigned commit_list_count(const struct commit_list *l)
 620 {
 621         unsigned c = 0;
 622         for (; l; l = l->next )
 623                 c++;
 624         return c;
 625 }
 626
 627 struct commit_list *copy_commit_list(struct commit_list *list)
 628 {
 629         struct commit_list *head = NULL;
 630         struct commit_list **pp = &head;
 631         while (list) {
 632                 pp = commit_list_append(list->item, pp);
 633                 list = list->next;
 634         }
 635         return head;
 636 }
 637
 638 struct commit_list *reverse_commit_list(struct commit_list *list)
 639 {
 640         struct commit_list *next = NULL, *current, *backup;
 641         for (current = list; current; current = backup) {
 642                 backup = current->next;
 643                 current->next = next;
 644                 next = current;
 645         }
 646         return next;
 647 }
 648
 649 void free_commit_list(struct commit_list *list)
 650 {
 651         while (list)
 652                 pop_commit(&list);
 653 }
 654
 655 struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
 656 {
 657         struct commit_list **pp = list;
 658         struct commit_list *p;
 659         while ((p = *pp) != NULL) {
 660                 if (p->item->date < item->date) {
 661                         break;
 662                 }
 663                 pp = &p->next;
 664         }
 665         return commit_list_insert(item, pp);
 666 }
 667
 668 static int commit_list_compare_by_date(const struct commit_list *a,
 669                                        const struct commit_list *b)
 670 {
 671         timestamp_t a_date = a->item->date;
 672         timestamp_t b_date = b->item->date;
 673         if (a_date < b_date)
 674                 return 1;
 675         if (a_date > b_date)
 676                 return -1;
 677         return 0;
 678 }
 679
 680 DEFINE_LIST_SORT(static, commit_list_sort, struct commit_list, next);
 681
 682 void commit_list_sort_by_date(struct commit_list **list)
 683 {
 684         commit_list_sort(list, commit_list_compare_by_date);
 685 }
 686
 687 struct commit *pop_most_recent_commit(struct commit_list **list,
 688                                       unsigned int mark)
 689 {
 690         struct commit *ret = pop_commit(list);
 691         struct commit_list *parents = ret->parents;
 692
 693         while (parents) {
 694                 struct commit *commit = parents->item;
 695                 if (!repo_parse_commit(the_repository, commit) && !(commit->object.flags & mark)) {
 696                         commit->object.flags |= mark;
 697                         commit_list_insert_by_date(commit, list);
 698                 }
 699                 parents = parents->next;
 700         }
 701         return ret;
 702 }
 703
 704 static void clear_commit_marks_1(struct commit_list **plist,
 705                                  struct commit *commit, unsigned int mark)
 706 {
 707         while (commit) {
 708                 struct commit_list *parents;
 709
 710                 if (!(mark & commit->object.flags))
 711                         return;
 712
 713                 commit->object.flags &= ~mark;
 714
 715                 parents = commit->parents;
 716                 if (!parents)
 717                         return;
 718
 719                 while ((parents = parents->next)) {
 720                         if (parents->item->object.flags & mark)
 721                                 commit_list_insert(parents->item, plist);
 722                 }
 723
 724                 commit = commit->parents->item;
 725         }
 726 }
 727
 728 void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
 729 {
 730         struct commit_list *list = NULL;
 731
 732         while (nr--) {
 733                 clear_commit_marks_1(&list, *commit, mark);
 734                 commit++;
 735         }
 736         while (list)
 737                 clear_commit_marks_1(&list, pop_commit(&list), mark);
 738 }
 739
 740 void clear_commit_marks(struct commit *commit, unsigned int mark)
 741 {
 742         clear_commit_marks_many(1, &commit, mark);
 743 }
 744
 745 struct commit *pop_commit(struct commit_list **stack)
 746 {
 747         struct commit_list *top = *stack;
 748         struct commit *item = top ? top->item : NULL;
 749
 750         if (top) {
 751                 *stack = top->next;
 752                 free(top);
 753         }
 754         return item;
 755 }
 756
 757 /*
 758  * Topological sort support
 759  */
 760
 761 /* count number of children that have not been emitted */
 762 define_commit_slab(indegree_slab, int);
 763
 764 define_commit_slab(author_date_slab, timestamp_t);
 765
 766 void record_author_date(struct author_date_slab *author_date,
 767                         struct commit *commit)
 768 {
 769         const char *buffer = repo_get_commit_buffer(the_repository, commit,
 770                                                     NULL);
 771         struct ident_split ident;
 772         const char *ident_line;
 773         size_t ident_len;
 774         char *date_end;
 775         timestamp_t date;
 776
 777         ident_line = find_commit_header(buffer, "author", &ident_len);
 778         if (!ident_line)
 779                 goto fail_exit; /* no author line */
 780         if (split_ident_line(&ident, ident_line, ident_len) ||
 781             !ident.date_begin || !ident.date_end)
 782                 goto fail_exit; /* malformed "author" line */
 783
 784         date = parse_timestamp(ident.date_begin, &date_end, 10);
 785         if (date_end != ident.date_end)
 786                 goto fail_exit; /* malformed date */
 787         *(author_date_slab_at(author_date, commit)) = date;
 788
 789 fail_exit:
 790         repo_unuse_commit_buffer(the_repository, commit, buffer);
 791 }
 792
 793 int compare_commits_by_author_date(const void *a_, const void *b_,
 794                                    void *cb_data)
 795 {
 796         const struct commit *a = a_, *b = b_;
 797         struct author_date_slab *author_date = cb_data;
 798         timestamp_t a_date = *(author_date_slab_at(author_date, a));
 799         timestamp_t b_date = *(author_date_slab_at(author_date, b));
 800
 801         /* newer commits with larger date first */
 802         if (a_date < b_date)
 803                 return 1;
 804         else if (a_date > b_date)
 805                 return -1;
 806         return 0;
 807 }
 808
 809 int compare_commits_by_gen_then_commit_date(const void *a_, const void *b_,
 810                                             void *unused UNUSED)
 811 {
 812         const struct commit *a = a_, *b = b_;
 813         const timestamp_t generation_a = commit_graph_generation(a),
 814                           generation_b = commit_graph_generation(b);
 815
 816         /* newer commits first */
 817         if (generation_a < generation_b)
 818                 return 1;
 819         else if (generation_a > generation_b)
 820                 return -1;
 821
 822         /* use date as a heuristic when generations are equal */
 823         if (a->date < b->date)
 824                 return 1;
 825         else if (a->date > b->date)
 826                 return -1;
 827         return 0;
 828 }
 829
 830 int compare_commits_by_commit_date(const void *a_, const void *b_,
 831                                    void *unused UNUSED)
 832 {
 833         const struct commit *a = a_, *b = b_;
 834         /* newer commits with larger date first */
 835         if (a->date < b->date)
 836                 return 1;
 837         else if (a->date > b->date)
 838                 return -1;
 839         return 0;
 840 }
 841
 842 /*
 843  * Performs an in-place topological sort on the list supplied.
 844  */
 845 void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
 846 {
 847         struct commit_list *next, *orig = *list;
 848         struct commit_list **pptr;
 849         struct indegree_slab indegree;
 850         struct prio_queue queue;
 851         struct commit *commit;
 852         struct author_date_slab author_date;
 853
 854         if (!orig)
 855                 return;
 856         *list = NULL;
 857
 858         init_indegree_slab(&indegree);
 859         memset(&queue, '\0', sizeof(queue));
 860
 861         switch (sort_order) {
 862         default: /* REV_SORT_IN_GRAPH_ORDER */
 863                 queue.compare = NULL;
 864                 break;
 865         case REV_SORT_BY_COMMIT_DATE:
 866                 queue.compare = compare_commits_by_commit_date;
 867                 break;
 868         case REV_SORT_BY_AUTHOR_DATE:
 869                 init_author_date_slab(&author_date);
 870                 queue.compare = compare_commits_by_author_date;
 871                 queue.cb_data = &author_date;
 872                 break;
 873         }
 874
 875         /* Mark them and clear the indegree */
 876         for (next = orig; next; next = next->next) {
 877                 struct commit *commit = next->item;
 878                 *(indegree_slab_at(&indegree, commit)) = 1;
 879                 /* also record the author dates, if needed */
 880                 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
 881                         record_author_date(&author_date, commit);
 882         }
 883
 884         /* update the indegree */
 885         for (next = orig; next; next = next->next) {
 886                 struct commit_list *parents = next->item->parents;
 887                 while (parents) {
 888                         struct commit *parent = parents->item;
 889                         int *pi = indegree_slab_at(&indegree, parent);
 890
 891                         if (*pi)
 892                                 (*pi)++;
 893                         parents = parents->next;
 894                 }
 895         }
 896
 897         /*
 898          * find the tips
 899          *
 900          * tips are nodes not reachable from any other node in the list
 901          *
 902          * the tips serve as a starting set for the work queue.
 903          */
 904         for (next = orig; next; next = next->next) {
 905                 struct commit *commit = next->item;
 906
 907                 if (*(indegree_slab_at(&indegree, commit)) == 1)
 908                         prio_queue_put(&queue, commit);
 909         }
 910
 911         /*
 912          * This is unfortunate; the initial tips need to be shown
 913          * in the order given from the revision traversal machinery.
 914          */
 915         if (sort_order == REV_SORT_IN_GRAPH_ORDER)
 916                 prio_queue_reverse(&queue);
 917
 918         /* We no longer need the commit list */
 919         free_commit_list(orig);
 920
 921         pptr = list;
 922         *list = NULL;
 923         while ((commit = prio_queue_get(&queue)) != NULL) {
 924                 struct commit_list *parents;
 925
 926                 for (parents = commit->parents; parents ; parents = parents->next) {
 927                         struct commit *parent = parents->item;
 928                         int *pi = indegree_slab_at(&indegree, parent);
 929
 930                         if (!*pi)
 931                                 continue;
 932
 933                         /*
 934                          * parents are only enqueued for emission
 935                          * when all their children have been emitted thereby
 936                          * guaranteeing topological order.
 937                          */
 938                         if (--(*pi) == 1)
 939                                 prio_queue_put(&queue, parent);
 940                 }
 941                 /*
 942                  * all children of commit have already been
 943                  * emitted. we can emit it now.
 944                  */
 945                 *(indegree_slab_at(&indegree, commit)) = 0;
 946
 947                 pptr = &commit_list_insert(commit, pptr)->next;
 948         }
 949
 950         clear_indegree_slab(&indegree);
 951         clear_prio_queue(&queue);
 952         if (sort_order == REV_SORT_BY_AUTHOR_DATE)
 953                 clear_author_date_slab(&author_date);
 954 }
 955
 956 struct rev_collect {
 957         struct commit **commit;
 958         int nr;
 959         int alloc;
 960         unsigned int initial : 1;
 961 };
 962
 963 static void add_one_commit(struct object_id *oid, struct rev_collect *revs)
 964 {
 965         struct commit *commit;
 966
 967         if (is_null_oid(oid))
 968                 return;
 969
 970         commit = lookup_commit(the_repository, oid);
 971         if (!commit ||
 972             (commit->object.flags & TMP_MARK) ||
 973             repo_parse_commit(the_repository, commit))
 974                 return;
 975
 976         ALLOC_GROW(revs->commit, revs->nr + 1, revs->alloc);
 977         revs->commit[revs->nr++] = commit;
 978         commit->object.flags |= TMP_MARK;
 979 }
 980
 981 static int collect_one_reflog_ent(struct object_id *ooid, struct object_id *noid,
 982                                   const char *ident UNUSED,
 983                                   timestamp_t timestamp UNUSED, int tz UNUSED,
 984                                   const char *message UNUSED, void *cbdata)
 985 {
 986         struct rev_collect *revs = cbdata;
 987
 988         if (revs->initial) {
 989                 revs->initial = 0;
 990                 add_one_commit(ooid, revs);
 991         }
 992         add_one_commit(noid, revs);
 993         return 0;
 994 }
 995
 996 struct commit *get_fork_point(const char *refname, struct commit *commit)
 997 {
 998         struct object_id oid;
 999         struct rev_collect revs;
1000         struct commit_list *bases;
1001         int i;
1002         struct commit *ret = NULL;
1003         char *full_refname;
1004
1005         switch (repo_dwim_ref(the_repository, refname, strlen(refname), &oid,
1006                               &full_refname, 0)) {
1007         case 0:
1008                 die("No such ref: '%s'", refname);
1009         case 1:
1010                 break; /* good */
1011         default:
1012                 die("Ambiguous refname: '%s'", refname);
1013         }
1014
1015         memset(&revs, 0, sizeof(revs));
1016         revs.initial = 1;
1017         for_each_reflog_ent(full_refname, collect_one_reflog_ent, &revs);
1018
1019         if (!revs.nr)
1020                 add_one_commit(&oid, &revs);
1021
1022         for (i = 0; i < revs.nr; i++)
1023                 revs.commit[i]->object.flags &= ~TMP_MARK;
1024
1025         bases = repo_get_merge_bases_many(the_repository, commit, revs.nr,
1026                                           revs.commit);
1027
1028         /*
1029          * There should be one and only one merge base, when we found
1030          * a common ancestor among reflog entries.
1031          */
1032         if (!bases || bases->next)
1033                 goto cleanup_return;
1034
1035         /* And the found one must be one of the reflog entries */
1036         for (i = 0; i < revs.nr; i++)
1037                 if (&bases->item->object == &revs.commit[i]->object)
1038                         break; /* found */
1039         if (revs.nr <= i)
1040                 goto cleanup_return;
1041
1042         ret = bases->item;
1043
1044 cleanup_return:
1045         free(revs.commit);
1046         free_commit_list(bases);
1047         free(full_refname);
1048         return ret;
1049 }
1050
1051 /*
1052  * Indexed by hash algorithm identifier.
1053  */
1054 static const char *gpg_sig_headers[] = {
1055         NULL,
1056         "gpgsig",
1057         "gpgsig-sha256",
1058 };
1059
1060 int sign_with_header(struct strbuf *buf, const char *keyid)
1061 {
1062         struct strbuf sig = STRBUF_INIT;
1063         int inspos, copypos;
1064         const char *eoh;
1065         const char *gpg_sig_header = gpg_sig_headers[hash_algo_by_ptr(the_hash_algo)];
1066         int gpg_sig_header_len = strlen(gpg_sig_header);
1067
1068         /* find the end of the header */
1069         eoh = strstr(buf->buf, "\n\n");
1070         if (!eoh)
1071                 inspos = buf->len;
1072         else
1073                 inspos = eoh - buf->buf + 1;
1074
1075         if (!keyid || !*keyid)
1076                 keyid = get_signing_key();
1077         if (sign_buffer(buf, &sig, keyid)) {
1078                 strbuf_release(&sig);
1079                 return -1;
1080         }
1081
1082         for (copypos = 0; sig.buf[copypos]; ) {
1083                 const char *bol = sig.buf + copypos;
1084                 const char *eol = strchrnul(bol, '\n');
1085                 int len = (eol - bol) + !!*eol;
1086
1087                 if (!copypos) {
1088                         strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
1089                         inspos += gpg_sig_header_len;
1090                 }
1091                 strbuf_insertstr(buf, inspos++, " ");
1092                 strbuf_insert(buf, inspos, bol, len);
1093                 inspos += len;
1094                 copypos += len;
1095         }
1096         strbuf_release(&sig);
1097         return 0;
1098 }
1099
1100
1101
1102 int parse_signed_commit(const struct commit *commit,
1103                         struct strbuf *payload, struct strbuf *signature,
1104                         const struct git_hash_algo *algop)
1105 {
1106         unsigned long size;
1107         const char *buffer = repo_get_commit_buffer(the_repository, commit,
1108                                                     &size);
1109         int ret = parse_buffer_signed_by_header(buffer, size, payload, signature, algop);
1110
1111         repo_unuse_commit_buffer(the_repository, commit, buffer);
1112         return ret;
1113 }
1114
1115 int parse_buffer_signed_by_header(const char *buffer,
1116                                   unsigned long size,
1117                                   struct strbuf *payload,
1118                                   struct strbuf *signature,
1119                                   const struct git_hash_algo *algop)
1120 {
1121         int in_signature = 0, saw_signature = 0, other_signature = 0;
1122         const char *line, *tail, *p;
1123         const char *gpg_sig_header = gpg_sig_headers[hash_algo_by_ptr(algop)];
1124
1125         line = buffer;
1126         tail = buffer + size;
1127         while (line < tail) {
1128                 const char *sig = NULL;
1129                 const char *next = memchr(line, '\n', tail - line);
1130
1131                 next = next ? next + 1 : tail;
1132                 if (in_signature && line[0] == ' ')
1133                         sig = line + 1;
1134                 else if (skip_prefix(line, gpg_sig_header, &p) &&
1135                          *p == ' ') {
1136                         sig = line + strlen(gpg_sig_header) + 1;
1137                         other_signature = 0;
1138                 }
1139                 else if (starts_with(line, "gpgsig"))
1140                         other_signature = 1;
1141                 else if (other_signature && line[0] != ' ')
1142                         other_signature = 0;
1143                 if (sig) {
1144                         strbuf_add(signature, sig, next - sig);
1145                         saw_signature = 1;
1146                         in_signature = 1;
1147                 } else {
1148                         if (*line == '\n')
1149                                 /* dump the whole remainder of the buffer */
1150                                 next = tail;
1151                         if (!other_signature)
1152                                 strbuf_add(payload, line, next - line);
1153                         in_signature = 0;
1154                 }
1155                 line = next;
1156         }
1157         return saw_signature;
1158 }
1159
1160 int remove_signature(struct strbuf *buf)
1161 {
1162         const char *line = buf->buf;
1163         const char *tail = buf->buf + buf->len;
1164         int in_signature = 0;
1165         struct sigbuf {
1166                 const char *start;
1167                 const char *end;
1168         } sigs[2], *sigp = &sigs[0];
1169         int i;
1170         const char *orig_buf = buf->buf;
1171
1172         memset(sigs, 0, sizeof(sigs));
1173
1174         while (line < tail) {
1175                 const char *next = memchr(line, '\n', tail - line);
1176                 next = next ? next + 1 : tail;
1177
1178                 if (in_signature && line[0] == ' ')
1179                         sigp->end = next;
1180                 else if (starts_with(line, "gpgsig")) {
1181                         int i;
1182                         for (i = 1; i < GIT_HASH_NALGOS; i++) {
1183                                 const char *p;
1184                                 if (skip_prefix(line, gpg_sig_headers[i], &p) &&
1185                                     *p == ' ') {
1186                                         sigp->start = line;
1187                                         sigp->end = next;
1188                                         in_signature = 1;
1189                                 }
1190                         }
1191                 } else {
1192                         if (*line == '\n')
1193                                 /* dump the whole remainder of the buffer */
1194                                 next = tail;
1195                         if (in_signature && sigp - sigs != ARRAY_SIZE(sigs))
1196                                 sigp++;
1197                         in_signature = 0;
1198                 }
1199                 line = next;
1200         }
1201
1202         for (i = ARRAY_SIZE(sigs) - 1; i >= 0; i--)
1203                 if (sigs[i].start)
1204                         strbuf_remove(buf, sigs[i].start - orig_buf, sigs[i].end - sigs[i].start);
1205
1206         return sigs[0].start != NULL;
1207 }
1208
1209 static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1210 {
1211         struct merge_remote_desc *desc;
1212         struct commit_extra_header *mergetag;
1213         char *buf;
1214         unsigned long size;
1215         enum object_type type;
1216         struct strbuf payload = STRBUF_INIT;
1217         struct strbuf signature = STRBUF_INIT;
1218
1219         desc = merge_remote_util(parent);
1220         if (!desc || !desc->obj)
1221                 return;
1222         buf = repo_read_object_file(the_repository, &desc->obj->oid, &type,
1223                                     &size);
1224         if (!buf || type != OBJ_TAG)
1225                 goto free_return;
1226         if (!parse_signature(buf, size, &payload, &signature))
1227                 goto free_return;
1228         /*
1229          * We could verify this signature and either omit the tag when
1230          * it does not validate, but the integrator may not have the
1231          * public key of the signer of the tag being merged, while a
1232          * later auditor may have it while auditing, so let's not run
1233          * verify-signed-buffer here for now...
1234          *
1235          * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1236          *      warn("warning: signed tag unverified.");
1237          */
1238         CALLOC_ARRAY(mergetag, 1);
1239         mergetag->key = xstrdup("mergetag");
1240         mergetag->value = buf;
1241         mergetag->len = size;
1242
1243         **tail = mergetag;
1244         *tail = &mergetag->next;
1245         strbuf_release(&payload);
1246         strbuf_release(&signature);
1247         return;
1248
1249 free_return:
1250         free(buf);
1251 }
1252
1253 int check_commit_signature(const struct commit *commit, struct signature_check *sigc)
1254 {
1255         struct strbuf payload = STRBUF_INIT;
1256         struct strbuf signature = STRBUF_INIT;
1257         int ret = 1;
1258
1259         sigc->result = 'N';
1260
1261         if (parse_signed_commit(commit, &payload, &signature, the_hash_algo) <= 0)
1262                 goto out;
1263
1264         sigc->payload_type = SIGNATURE_PAYLOAD_COMMIT;
1265         sigc->payload = strbuf_detach(&payload, &sigc->payload_len);
1266         ret = check_signature(sigc, signature.buf, signature.len);
1267
1268  out:
1269         strbuf_release(&payload);
1270         strbuf_release(&signature);
1271
1272         return ret;
1273 }
1274
1275 void verify_merge_signature(struct commit *commit, int verbosity,
1276                             int check_trust)
1277 {
1278         char hex[GIT_MAX_HEXSZ + 1];
1279         struct signature_check signature_check;
1280         int ret;
1281         memset(&signature_check, 0, sizeof(signature_check));
1282
1283         ret = check_commit_signature(commit, &signature_check);
1284
1285         repo_find_unique_abbrev_r(the_repository, hex, &commit->object.oid,
1286                                   DEFAULT_ABBREV);
1287         switch (signature_check.result) {
1288         case 'G':
1289                 if (ret || (check_trust && signature_check.trust_level < TRUST_MARGINAL))
1290                         die(_("Commit %s has an untrusted GPG signature, "
1291                               "allegedly by %s."), hex, signature_check.signer);
1292                 break;
1293         case 'B':
1294                 die(_("Commit %s has a bad GPG signature "
1295                       "allegedly by %s."), hex, signature_check.signer);
1296         default: /* 'N' */
1297                 die(_("Commit %s does not have a GPG signature."), hex);
1298         }
1299         if (verbosity >= 0 && signature_check.result == 'G')
1300                 printf(_("Commit %s has a good GPG signature by %s\n"),
1301                        hex, signature_check.signer);
1302
1303         signature_check_clear(&signature_check);
1304 }
1305
1306 void append_merge_tag_headers(struct commit_list *parents,
1307                               struct commit_extra_header ***tail)
1308 {
1309         while (parents) {
1310                 struct commit *parent = parents->item;
1311                 handle_signed_tag(parent, tail);
1312                 parents = parents->next;
1313         }
1314 }
1315
1316 static void add_extra_header(struct strbuf *buffer,
1317                              struct commit_extra_header *extra)
1318 {
1319         strbuf_addstr(buffer, extra->key);
1320         if (extra->len)
1321                 strbuf_add_lines(buffer, " ", extra->value, extra->len);
1322         else
1323                 strbuf_addch(buffer, '\n');
1324 }
1325
1326 struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1327                                                       const char **exclude)
1328 {
1329         struct commit_extra_header *extra = NULL;
1330         unsigned long size;
1331         const char *buffer = repo_get_commit_buffer(the_repository, commit,
1332                                                     &size);
1333         extra = read_commit_extra_header_lines(buffer, size, exclude);
1334         repo_unuse_commit_buffer(the_repository, commit, buffer);
1335         return extra;
1336 }
1337
1338 int for_each_mergetag(each_mergetag_fn fn, struct commit *commit, void *data)
1339 {
1340         struct commit_extra_header *extra, *to_free;
1341         int res = 0;
1342
1343         to_free = read_commit_extra_headers(commit, NULL);
1344         for (extra = to_free; !res && extra; extra = extra->next) {
1345                 if (strcmp(extra->key, "mergetag"))
1346                         continue; /* not a merge tag */
1347                 res = fn(commit, extra, data);
1348         }
1349         free_commit_extra_headers(to_free);
1350         return res;
1351 }
1352
1353 static inline int standard_header_field(const char *field, size_t len)
1354 {
1355         return ((len == 4 && !memcmp(field, "tree", 4)) ||
1356                 (len == 6 && !memcmp(field, "parent", 6)) ||
1357                 (len == 6 && !memcmp(field, "author", 6)) ||
1358                 (len == 9 && !memcmp(field, "committer", 9)) ||
1359                 (len == 8 && !memcmp(field, "encoding", 8)));
1360 }
1361
1362 static int excluded_header_field(const char *field, size_t len, const char **exclude)
1363 {
1364         if (!exclude)
1365                 return 0;
1366
1367         while (*exclude) {
1368                 size_t xlen = strlen(*exclude);
1369                 if (len == xlen && !memcmp(field, *exclude, xlen))
1370                         return 1;
1371                 exclude++;
1372         }
1373         return 0;
1374 }
1375
1376 static struct commit_extra_header *read_commit_extra_header_lines(
1377         const char *buffer, size_t size,
1378         const char **exclude)
1379 {
1380         struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1381         const char *line, *next, *eof, *eob;
1382         struct strbuf buf = STRBUF_INIT;
1383
1384         for (line = buffer, eob = line + size;
1385              line < eob && *line != '\n';
1386              line = next) {
1387                 next = memchr(line, '\n', eob - line);
1388                 next = next ? next + 1 : eob;
1389                 if (*line == ' ') {
1390                         /* continuation */
1391                         if (it)
1392                                 strbuf_add(&buf, line + 1, next - (line + 1));
1393                         continue;
1394                 }
1395                 if (it)
1396                         it->value = strbuf_detach(&buf, &it->len);
1397                 strbuf_reset(&buf);
1398                 it = NULL;
1399
1400                 eof = memchr(line, ' ', next - line);
1401                 if (!eof)
1402                         eof = next;
1403                 else if (standard_header_field(line, eof - line) ||
1404                          excluded_header_field(line, eof - line, exclude))
1405                         continue;
1406
1407                 CALLOC_ARRAY(it, 1);
1408                 it->key = xmemdupz(line, eof-line);
1409                 *tail = it;
1410                 tail = &it->next;
1411                 if (eof + 1 < next)
1412                         strbuf_add(&buf, eof + 1, next - (eof + 1));
1413         }
1414         if (it)
1415                 it->value = strbuf_detach(&buf, &it->len);
1416         return extra;
1417 }
1418
1419 void free_commit_extra_headers(struct commit_extra_header *extra)
1420 {
1421         while (extra) {
1422                 struct commit_extra_header *next = extra->next;
1423                 free(extra->key);
1424                 free(extra->value);
1425                 free(extra);
1426                 extra = next;
1427         }
1428 }
1429
1430 int commit_tree(const char *msg, size_t msg_len, const struct object_id *tree,
1431                 struct commit_list *parents, struct object_id *ret,
1432                 const char *author, const char *sign_commit)
1433 {
1434         struct commit_extra_header *extra = NULL, **tail = &extra;
1435         int result;
1436
1437         append_merge_tag_headers(parents, &tail);
1438         result = commit_tree_extended(msg, msg_len, tree, parents, ret, author,
1439                                       NULL, sign_commit, extra);
1440         free_commit_extra_headers(extra);
1441         return result;
1442 }
1443
1444 static int find_invalid_utf8(const char *buf, int len)
1445 {
1446         int offset = 0;
1447         static const unsigned int max_codepoint[] = {
1448                 0x7f, 0x7ff, 0xffff, 0x10ffff
1449         };
1450
1451         while (len) {
1452                 unsigned char c = *buf++;
1453                 int bytes, bad_offset;
1454                 unsigned int codepoint;
1455                 unsigned int min_val, max_val;
1456
1457                 len--;
1458                 offset++;
1459
1460                 /* Simple US-ASCII? No worries. */
1461                 if (c < 0x80)
1462                         continue;
1463
1464                 bad_offset = offset-1;
1465
1466                 /*
1467                  * Count how many more high bits set: that's how
1468                  * many more bytes this sequence should have.
1469                  */
1470                 bytes = 0;
1471                 while (c & 0x40) {
1472                         c <<= 1;
1473                         bytes++;
1474                 }
1475
1476                 /*
1477                  * Must be between 1 and 3 more bytes.  Longer sequences result in
1478                  * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1479                  */
1480                 if (bytes < 1 || 3 < bytes)
1481                         return bad_offset;
1482
1483                 /* Do we *have* that many bytes? */
1484                 if (len < bytes)
1485                         return bad_offset;
1486
1487                 /*
1488                  * Place the encoded bits at the bottom of the value and compute the
1489                  * valid range.
1490                  */
1491                 codepoint = (c & 0x7f) >> bytes;
1492                 min_val = max_codepoint[bytes-1] + 1;
1493                 max_val = max_codepoint[bytes];
1494
1495                 offset += bytes;
1496                 len -= bytes;
1497
1498                 /* And verify that they are good continuation bytes */
1499                 do {
1500                         codepoint <<= 6;
1501                         codepoint |= *buf & 0x3f;
1502                         if ((*buf++ & 0xc0) != 0x80)
1503                                 return bad_offset;
1504                 } while (--bytes);
1505
1506                 /* Reject codepoints that are out of range for the sequence length. */
1507                 if (codepoint < min_val || codepoint > max_val)
1508                         return bad_offset;
1509                 /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1510                 if ((codepoint & 0x1ff800) == 0xd800)
1511                         return bad_offset;
1512                 /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1513                 if ((codepoint & 0xfffe) == 0xfffe)
1514                         return bad_offset;
1515                 /* So are anything in the range U+FDD0..U+FDEF. */
1516                 if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1517                         return bad_offset;
1518         }
1519         return -1;
1520 }
1521
1522 /*
1523  * This verifies that the buffer is in proper utf8 format.
1524  *
1525  * If it isn't, it assumes any non-utf8 characters are Latin1,
1526  * and does the conversion.
1527  */
1528 static int verify_utf8(struct strbuf *buf)
1529 {
1530         int ok = 1;
1531         long pos = 0;
1532
1533         for (;;) {
1534                 int bad;
1535                 unsigned char c;
1536                 unsigned char replace[2];
1537
1538                 bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1539                 if (bad < 0)
1540                         return ok;
1541                 pos += bad;
1542                 ok = 0;
1543                 c = buf->buf[pos];
1544                 strbuf_remove(buf, pos, 1);
1545
1546                 /* We know 'c' must be in the range 128-255 */
1547                 replace[0] = 0xc0 + (c >> 6);
1548                 replace[1] = 0x80 + (c & 0x3f);
1549                 strbuf_insert(buf, pos, replace, 2);
1550                 pos += 2;
1551         }
1552 }
1553
1554 static const char commit_utf8_warn[] =
1555 N_("Warning: commit message did not conform to UTF-8.\n"
1556    "You may want to amend it after fixing the message, or set the config\n"
1557    "variable i18n.commitEncoding to the encoding your project uses.\n");
1558
1559 int commit_tree_extended(const char *msg, size_t msg_len,
1560                          const struct object_id *tree,
1561                          struct commit_list *parents, struct object_id *ret,
1562                          const char *author, const char *committer,
1563                          const char *sign_commit,
1564                          struct commit_extra_header *extra)
1565 {
1566         int result;
1567         int encoding_is_utf8;
1568         struct strbuf buffer;
1569
1570         assert_oid_type(tree, OBJ_TREE);
1571
1572         if (memchr(msg, '\0', msg_len))
1573                 return error("a NUL byte in commit log message not allowed.");
1574
1575         /* Not having i18n.commitencoding is the same as having utf-8 */
1576         encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1577
1578         strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1579         strbuf_addf(&buffer, "tree %s\n", oid_to_hex(tree));
1580
1581         /*
1582          * NOTE! This ordering means that the same exact tree merged with a
1583          * different order of parents will be a _different_ changeset even
1584          * if everything else stays the same.
1585          */
1586         while (parents) {
1587                 struct commit *parent = pop_commit(&parents);
1588                 strbuf_addf(&buffer, "parent %s\n",
1589                             oid_to_hex(&parent->object.oid));
1590         }
1591
1592         /* Person/date information */
1593         if (!author)
1594                 author = git_author_info(IDENT_STRICT);
1595         strbuf_addf(&buffer, "author %s\n", author);
1596         if (!committer)
1597                 committer = git_committer_info(IDENT_STRICT);
1598         strbuf_addf(&buffer, "committer %s\n", committer);
1599         if (!encoding_is_utf8)
1600                 strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1601
1602         while (extra) {
1603                 add_extra_header(&buffer, extra);
1604                 extra = extra->next;
1605         }
1606         strbuf_addch(&buffer, '\n');
1607
1608         /* And add the comment */
1609         strbuf_add(&buffer, msg, msg_len);
1610
1611         /* And check the encoding */
1612         if (encoding_is_utf8 && !verify_utf8(&buffer))
1613                 fprintf(stderr, _(commit_utf8_warn));
1614
1615         if (sign_commit && sign_with_header(&buffer, sign_commit)) {
1616                 result = -1;
1617                 goto out;
1618         }
1619
1620         result = write_object_file(buffer.buf, buffer.len, OBJ_COMMIT, ret);
1621 out:
1622         strbuf_release(&buffer);
1623         return result;
1624 }
1625
1626 define_commit_slab(merge_desc_slab, struct merge_remote_desc *);
1627 static struct merge_desc_slab merge_desc_slab = COMMIT_SLAB_INIT(1, merge_desc_slab);
1628
1629 struct merge_remote_desc *merge_remote_util(struct commit *commit)
1630 {
1631         return *merge_desc_slab_at(&merge_desc_slab, commit);
1632 }
1633
1634 void set_merge_remote_desc(struct commit *commit,
1635                            const char *name, struct object *obj)
1636 {
1637         struct merge_remote_desc *desc;
1638         FLEX_ALLOC_STR(desc, name, name);
1639         desc->obj = obj;
1640         *merge_desc_slab_at(&merge_desc_slab, commit) = desc;
1641 }
1642
1643 struct commit *get_merge_parent(const char *name)
1644 {
1645         struct object *obj;
1646         struct commit *commit;
1647         struct object_id oid;
1648         if (repo_get_oid(the_repository, name, &oid))
1649                 return NULL;
1650         obj = parse_object(the_repository, &oid);
1651         commit = (struct commit *)repo_peel_to_type(the_repository, name, 0,
1652                                                     obj, OBJ_COMMIT);
1653         if (commit && !merge_remote_util(commit))
1654                 set_merge_remote_desc(commit, name, obj);
1655         return commit;
1656 }
1657
1658 /*
1659  * Append a commit to the end of the commit_list.
1660  *
1661  * next starts by pointing to the variable that holds the head of an
1662  * empty commit_list, and is updated to point to the "next" field of
1663  * the last item on the list as new commits are appended.
1664  *
1665  * Usage example:
1666  *
1667  *     struct commit_list *list;
1668  *     struct commit_list **next = &list;
1669  *
1670  *     next = commit_list_append(c1, next);
1671  *     next = commit_list_append(c2, next);
1672  *     assert(commit_list_count(list) == 2);
1673  *     return list;
1674  */
1675 struct commit_list **commit_list_append(struct commit *commit,
1676                                         struct commit_list **next)
1677 {
1678         struct commit_list *new_commit = xmalloc(sizeof(struct commit_list));
1679         new_commit->item = commit;
1680         *next = new_commit;
1681         new_commit->next = NULL;
1682         return &new_commit->next;
1683 }
1684
1685 const char *find_header_mem(const char *msg, size_t len,
1686                         const char *key, size_t *out_len)
1687 {
1688         int key_len = strlen(key);
1689         const char *line = msg;
1690
1691         /*
1692          * NEEDSWORK: It's possible for strchrnul() to scan beyond the range
1693          * given by len. However, current callers are safe because they compute
1694          * len by scanning a NUL-terminated block of memory starting at msg.
1695          * Nonetheless, it would be better to ensure the function does not look
1696          * at msg beyond the len provided by the caller.
1697          */
1698         while (line && line < msg + len) {
1699                 const char *eol = strchrnul(line, '\n');
1700
1701                 if (line == eol)
1702                         return NULL;
1703
1704                 if (eol - line > key_len &&
1705                     !strncmp(line, key, key_len) &&
1706                     line[key_len] == ' ') {
1707                         *out_len = eol - line - key_len - 1;
1708                         return line + key_len + 1;
1709                 }
1710                 line = *eol ? eol + 1 : NULL;
1711         }
1712         return NULL;
1713 }
1714
1715 const char *find_commit_header(const char *msg, const char *key, size_t *out_len)
1716 {
1717         return find_header_mem(msg, strlen(msg), key, out_len);
1718 }
1719 /*
1720  * Inspect the given string and determine the true "end" of the log message, in
1721  * order to find where to put a new Signed-off-by trailer.  Ignored are
1722  * trailing comment lines and blank lines.  To support "git commit -s
1723  * --amend" on an existing commit, we also ignore "Conflicts:".  To
1724  * support "git commit -v", we truncate at cut lines.
1725  *
1726  * Returns the number of bytes from the tail to ignore, to be fed as
1727  * the second parameter to append_signoff().
1728  */
1729 size_t ignore_non_trailer(const char *buf, size_t len)
1730 {
1731         size_t boc = 0;
1732         size_t bol = 0;
1733         int in_old_conflicts_block = 0;
1734         size_t cutoff = wt_status_locate_end(buf, len);
1735
1736         while (bol < cutoff) {
1737                 const char *next_line = memchr(buf + bol, '\n', len - bol);
1738
1739                 if (!next_line)
1740                         next_line = buf + len;
1741                 else
1742                         next_line++;
1743
1744                 if (buf[bol] == comment_line_char || buf[bol] == '\n') {
1745                         /* is this the first of the run of comments? */
1746                         if (!boc)
1747                                 boc = bol;
1748                         /* otherwise, it is just continuing */
1749                 } else if (starts_with(buf + bol, "Conflicts:\n")) {
1750                         in_old_conflicts_block = 1;
1751                         if (!boc)
1752                                 boc = bol;
1753                 } else if (in_old_conflicts_block && buf[bol] == '\t') {
1754                         ; /* a pathname in the conflicts block */
1755                 } else if (boc) {
1756                         /* the previous was not trailing comment */
1757                         boc = 0;
1758                         in_old_conflicts_block = 0;
1759                 }
1760                 bol = next_line - buf;
1761         }
1762         return boc ? len - boc : len - cutoff;
1763 }
1764
1765 int run_commit_hook(int editor_is_used, const char *index_file,
1766                     int *invoked_hook, const char *name, ...)
1767 {
1768         struct run_hooks_opt opt = RUN_HOOKS_OPT_INIT;
1769         va_list args;
1770         const char *arg;
1771
1772         strvec_pushf(&opt.env, "GIT_INDEX_FILE=%s", index_file);
1773
1774         /*
1775          * Let the hook know that no editor will be launched.
1776          */
1777         if (!editor_is_used)
1778                 strvec_push(&opt.env, "GIT_EDITOR=:");
1779
1780         va_start(args, name);
1781         while ((arg = va_arg(args, const char *)))
1782                 strvec_push(&opt.args, arg);
1783         va_end(args);
1784
1785         opt.invoked_hook = invoked_hook;
1786         return run_hooks_opt(name, &opt);
1787 }