commit.con commit Merge branch 'ds/commit-graph-lockfile-fix' (a856e7d)
   1#include "cache.h"
   2#include "tag.h"
   3#include "commit.h"
   4#include "commit-graph.h"
   5#include "pkt-line.h"
   6#include "utf8.h"
   7#include "diff.h"
   8#include "revision.h"
   9#include "notes.h"
  10#include "gpg-interface.h"
  11#include "mergesort.h"
  12#include "commit-slab.h"
  13#include "prio-queue.h"
  14#include "sha1-lookup.h"
  15#include "wt-status.h"
  16#include "advice.h"
  17
  18static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
  19
  20int save_commit_buffer = 1;
  21
  22const char *commit_type = "commit";
  23
  24struct commit *lookup_commit_reference_gently(const struct object_id *oid,
  25                                              int quiet)
  26{
  27        struct object *obj = deref_tag(parse_object(oid), NULL, 0);
  28
  29        if (!obj)
  30                return NULL;
  31        return object_as_type(obj, OBJ_COMMIT, quiet);
  32}
  33
  34struct commit *lookup_commit_reference(const struct object_id *oid)
  35{
  36        return lookup_commit_reference_gently(oid, 0);
  37}
  38
  39struct commit *lookup_commit_or_die(const struct object_id *oid, const char *ref_name)
  40{
  41        struct commit *c = lookup_commit_reference(oid);
  42        if (!c)
  43                die(_("could not parse %s"), ref_name);
  44        if (oidcmp(oid, &c->object.oid)) {
  45                warning(_("%s %s is not a commit!"),
  46                        ref_name, oid_to_hex(oid));
  47        }
  48        return c;
  49}
  50
  51struct commit *lookup_commit(const struct object_id *oid)
  52{
  53        struct object *obj = lookup_object(oid->hash);
  54        if (!obj)
  55                return create_object(oid->hash, alloc_commit_node());
  56        return object_as_type(obj, OBJ_COMMIT, 0);
  57}
  58
  59struct commit *lookup_commit_reference_by_name(const char *name)
  60{
  61        struct object_id oid;
  62        struct commit *commit;
  63
  64        if (get_oid_committish(name, &oid))
  65                return NULL;
  66        commit = lookup_commit_reference(&oid);
  67        if (parse_commit(commit))
  68                return NULL;
  69        return commit;
  70}
  71
  72static timestamp_t parse_commit_date(const char *buf, const char *tail)
  73{
  74        const char *dateptr;
  75
  76        if (buf + 6 >= tail)
  77                return 0;
  78        if (memcmp(buf, "author", 6))
  79                return 0;
  80        while (buf < tail && *buf++ != '\n')
  81                /* nada */;
  82        if (buf + 9 >= tail)
  83                return 0;
  84        if (memcmp(buf, "committer", 9))
  85                return 0;
  86        while (buf < tail && *buf++ != '>')
  87                /* nada */;
  88        if (buf >= tail)
  89                return 0;
  90        dateptr = buf;
  91        while (buf < tail && *buf++ != '\n')
  92                /* nada */;
  93        if (buf >= tail)
  94                return 0;
  95        /* dateptr < buf && buf[-1] == '\n', so parsing will stop at buf-1 */
  96        return parse_timestamp(dateptr, NULL, 10);
  97}
  98
  99static struct commit_graft **commit_graft;
 100static int commit_graft_alloc, commit_graft_nr;
 101
 102static const unsigned char *commit_graft_sha1_access(size_t index, void *table)
 103{
 104        struct commit_graft **commit_graft_table = table;
 105        return commit_graft_table[index]->oid.hash;
 106}
 107
 108static int commit_graft_pos(const unsigned char *sha1)
 109{
 110        return sha1_pos(sha1, commit_graft, commit_graft_nr,
 111                        commit_graft_sha1_access);
 112}
 113
 114int register_commit_graft(struct commit_graft *graft, int ignore_dups)
 115{
 116        int pos = commit_graft_pos(graft->oid.hash);
 117
 118        if (0 <= pos) {
 119                if (ignore_dups)
 120                        free(graft);
 121                else {
 122                        free(commit_graft[pos]);
 123                        commit_graft[pos] = graft;
 124                }
 125                return 1;
 126        }
 127        pos = -pos - 1;
 128        ALLOC_GROW(commit_graft, commit_graft_nr + 1, commit_graft_alloc);
 129        commit_graft_nr++;
 130        if (pos < commit_graft_nr)
 131                MOVE_ARRAY(commit_graft + pos + 1, commit_graft + pos,
 132                           commit_graft_nr - pos - 1);
 133        commit_graft[pos] = graft;
 134        return 0;
 135}
 136
 137struct commit_graft *read_graft_line(struct strbuf *line)
 138{
 139        /* The format is just "Commit Parent1 Parent2 ...\n" */
 140        int i, phase;
 141        const char *tail = NULL;
 142        struct commit_graft *graft = NULL;
 143        struct object_id dummy_oid, *oid;
 144
 145        strbuf_rtrim(line);
 146        if (!line->len || line->buf[0] == '#')
 147                return NULL;
 148        /*
 149         * phase 0 verifies line, counts hashes in line and allocates graft
 150         * phase 1 fills graft
 151         */
 152        for (phase = 0; phase < 2; phase++) {
 153                oid = graft ? &graft->oid : &dummy_oid;
 154                if (parse_oid_hex(line->buf, oid, &tail))
 155                        goto bad_graft_data;
 156                for (i = 0; *tail != '\0'; i++) {
 157                        oid = graft ? &graft->parent[i] : &dummy_oid;
 158                        if (!isspace(*tail++) || parse_oid_hex(tail, oid, &tail))
 159                                goto bad_graft_data;
 160                }
 161                if (!graft) {
 162                        graft = xmalloc(st_add(sizeof(*graft),
 163                                               st_mult(sizeof(struct object_id), i)));
 164                        graft->nr_parent = i;
 165                }
 166        }
 167        return graft;
 168
 169bad_graft_data:
 170        error("bad graft data: %s", line->buf);
 171        assert(!graft);
 172        return NULL;
 173}
 174
 175static int read_graft_file(const char *graft_file)
 176{
 177        FILE *fp = fopen_or_warn(graft_file, "r");
 178        struct strbuf buf = STRBUF_INIT;
 179        if (!fp)
 180                return -1;
 181        if (advice_graft_file_deprecated)
 182                advise(_("Support for <GIT_DIR>/info/grafts is deprecated\n"
 183                         "and will be removed in a future Git version.\n"
 184                         "\n"
 185                         "Please use \"git replace --convert-graft-file\"\n"
 186                         "to convert the grafts into replace refs.\n"
 187                         "\n"
 188                         "Turn this message off by running\n"
 189                         "\"git config advice.graftFileDeprecated false\""));
 190        while (!strbuf_getwholeline(&buf, fp, '\n')) {
 191                /* The format is just "Commit Parent1 Parent2 ...\n" */
 192                struct commit_graft *graft = read_graft_line(&buf);
 193                if (!graft)
 194                        continue;
 195                if (register_commit_graft(graft, 1))
 196                        error("duplicate graft data: %s", buf.buf);
 197        }
 198        fclose(fp);
 199        strbuf_release(&buf);
 200        return 0;
 201}
 202
 203static void prepare_commit_graft(void)
 204{
 205        static int commit_graft_prepared;
 206        char *graft_file;
 207
 208        if (commit_graft_prepared)
 209                return;
 210        if (!startup_info->have_repository)
 211                return;
 212
 213        graft_file = get_graft_file();
 214        read_graft_file(graft_file);
 215        /* make sure shallows are read */
 216        is_repository_shallow();
 217        commit_graft_prepared = 1;
 218}
 219
 220struct commit_graft *lookup_commit_graft(const struct object_id *oid)
 221{
 222        int pos;
 223        prepare_commit_graft();
 224        pos = commit_graft_pos(oid->hash);
 225        if (pos < 0)
 226                return NULL;
 227        return commit_graft[pos];
 228}
 229
 230int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
 231{
 232        int i, ret;
 233        for (i = ret = 0; i < commit_graft_nr && !ret; i++)
 234                ret = fn(commit_graft[i], cb_data);
 235        return ret;
 236}
 237
 238int unregister_shallow(const struct object_id *oid)
 239{
 240        int pos = commit_graft_pos(oid->hash);
 241        if (pos < 0)
 242                return -1;
 243        if (pos + 1 < commit_graft_nr)
 244                MOVE_ARRAY(commit_graft + pos, commit_graft + pos + 1,
 245                           commit_graft_nr - pos - 1);
 246        commit_graft_nr--;
 247        return 0;
 248}
 249
 250struct commit_buffer {
 251        void *buffer;
 252        unsigned long size;
 253};
 254define_commit_slab(buffer_slab, struct commit_buffer);
 255static struct buffer_slab buffer_slab = COMMIT_SLAB_INIT(1, buffer_slab);
 256
 257void set_commit_buffer(struct commit *commit, void *buffer, unsigned long size)
 258{
 259        struct commit_buffer *v = buffer_slab_at(&buffer_slab, commit);
 260        v->buffer = buffer;
 261        v->size = size;
 262}
 263
 264const void *get_cached_commit_buffer(const struct commit *commit, unsigned long *sizep)
 265{
 266        struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
 267        if (!v) {
 268                if (sizep)
 269                        *sizep = 0;
 270                return NULL;
 271        }
 272        if (sizep)
 273                *sizep = v->size;
 274        return v->buffer;
 275}
 276
 277const void *get_commit_buffer(const struct commit *commit, unsigned long *sizep)
 278{
 279        const void *ret = get_cached_commit_buffer(commit, sizep);
 280        if (!ret) {
 281                enum object_type type;
 282                unsigned long size;
 283                ret = read_object_file(&commit->object.oid, &type, &size);
 284                if (!ret)
 285                        die("cannot read commit object %s",
 286                            oid_to_hex(&commit->object.oid));
 287                if (type != OBJ_COMMIT)
 288                        die("expected commit for %s, got %s",
 289                            oid_to_hex(&commit->object.oid), type_name(type));
 290                if (sizep)
 291                        *sizep = size;
 292        }
 293        return ret;
 294}
 295
 296void unuse_commit_buffer(const struct commit *commit, const void *buffer)
 297{
 298        struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
 299        if (!(v && v->buffer == buffer))
 300                free((void *)buffer);
 301}
 302
 303void free_commit_buffer(struct commit *commit)
 304{
 305        struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
 306        if (v) {
 307                FREE_AND_NULL(v->buffer);
 308                v->size = 0;
 309        }
 310}
 311
 312struct tree *get_commit_tree(const struct commit *commit)
 313{
 314        if (commit->maybe_tree || !commit->object.parsed)
 315                return commit->maybe_tree;
 316
 317        if (commit->graph_pos == COMMIT_NOT_FROM_GRAPH)
 318                BUG("commit has NULL tree, but was not loaded from commit-graph");
 319
 320        return get_commit_tree_in_graph(commit);
 321}
 322
 323struct object_id *get_commit_tree_oid(const struct commit *commit)
 324{
 325        return &get_commit_tree(commit)->object.oid;
 326}
 327
 328const void *detach_commit_buffer(struct commit *commit, unsigned long *sizep)
 329{
 330        struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
 331        void *ret;
 332
 333        if (!v) {
 334                if (sizep)
 335                        *sizep = 0;
 336                return NULL;
 337        }
 338        ret = v->buffer;
 339        if (sizep)
 340                *sizep = v->size;
 341
 342        v->buffer = NULL;
 343        v->size = 0;
 344        return ret;
 345}
 346
 347int parse_commit_buffer(struct commit *item, const void *buffer, unsigned long size, int check_graph)
 348{
 349        const char *tail = buffer;
 350        const char *bufptr = buffer;
 351        struct object_id parent;
 352        struct commit_list **pptr;
 353        struct commit_graft *graft;
 354        const int tree_entry_len = GIT_SHA1_HEXSZ + 5;
 355        const int parent_entry_len = GIT_SHA1_HEXSZ + 7;
 356
 357        if (item->object.parsed)
 358                return 0;
 359        item->object.parsed = 1;
 360        tail += size;
 361        if (tail <= bufptr + tree_entry_len + 1 || memcmp(bufptr, "tree ", 5) ||
 362                        bufptr[tree_entry_len] != '\n')
 363                return error("bogus commit object %s", oid_to_hex(&item->object.oid));
 364        if (get_oid_hex(bufptr + 5, &parent) < 0)
 365                return error("bad tree pointer in commit %s",
 366                             oid_to_hex(&item->object.oid));
 367        item->maybe_tree = lookup_tree(&parent);
 368        bufptr += tree_entry_len + 1; /* "tree " + "hex sha1" + "\n" */
 369        pptr = &item->parents;
 370
 371        graft = lookup_commit_graft(&item->object.oid);
 372        while (bufptr + parent_entry_len < tail && !memcmp(bufptr, "parent ", 7)) {
 373                struct commit *new_parent;
 374
 375                if (tail <= bufptr + parent_entry_len + 1 ||
 376                    get_oid_hex(bufptr + 7, &parent) ||
 377                    bufptr[parent_entry_len] != '\n')
 378                        return error("bad parents in commit %s", oid_to_hex(&item->object.oid));
 379                bufptr += parent_entry_len + 1;
 380                /*
 381                 * The clone is shallow if nr_parent < 0, and we must
 382                 * not traverse its real parents even when we unhide them.
 383                 */
 384                if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
 385                        continue;
 386                new_parent = lookup_commit(&parent);
 387                if (new_parent)
 388                        pptr = &commit_list_insert(new_parent, pptr)->next;
 389        }
 390        if (graft) {
 391                int i;
 392                struct commit *new_parent;
 393                for (i = 0; i < graft->nr_parent; i++) {
 394                        new_parent = lookup_commit(&graft->parent[i]);
 395                        if (!new_parent)
 396                                continue;
 397                        pptr = &commit_list_insert(new_parent, pptr)->next;
 398                }
 399        }
 400        item->date = parse_commit_date(bufptr, tail);
 401
 402        if (check_graph)
 403                load_commit_graph_info(item);
 404
 405        return 0;
 406}
 407
 408int parse_commit_gently(struct commit *item, int quiet_on_missing)
 409{
 410        enum object_type type;
 411        void *buffer;
 412        unsigned long size;
 413        int ret;
 414
 415        if (!item)
 416                return -1;
 417        if (item->object.parsed)
 418                return 0;
 419        if (parse_commit_in_graph(item))
 420                return 0;
 421        buffer = read_object_file(&item->object.oid, &type, &size);
 422        if (!buffer)
 423                return quiet_on_missing ? -1 :
 424                        error("Could not read %s",
 425                             oid_to_hex(&item->object.oid));
 426        if (type != OBJ_COMMIT) {
 427                free(buffer);
 428                return error("Object %s not a commit",
 429                             oid_to_hex(&item->object.oid));
 430        }
 431        ret = parse_commit_buffer(item, buffer, size, 0);
 432        if (save_commit_buffer && !ret) {
 433                set_commit_buffer(item, buffer, size);
 434                return 0;
 435        }
 436        free(buffer);
 437        return ret;
 438}
 439
 440void parse_commit_or_die(struct commit *item)
 441{
 442        if (parse_commit(item))
 443                die("unable to parse commit %s",
 444                    item ? oid_to_hex(&item->object.oid) : "(null)");
 445}
 446
 447int find_commit_subject(const char *commit_buffer, const char **subject)
 448{
 449        const char *eol;
 450        const char *p = commit_buffer;
 451
 452        while (*p && (*p != '\n' || p[1] != '\n'))
 453                p++;
 454        if (*p) {
 455                p = skip_blank_lines(p + 2);
 456                eol = strchrnul(p, '\n');
 457        } else
 458                eol = p;
 459
 460        *subject = p;
 461
 462        return eol - p;
 463}
 464
 465struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
 466{
 467        struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
 468        new_list->item = item;
 469        new_list->next = *list_p;
 470        *list_p = new_list;
 471        return new_list;
 472}
 473
 474unsigned commit_list_count(const struct commit_list *l)
 475{
 476        unsigned c = 0;
 477        for (; l; l = l->next )
 478                c++;
 479        return c;
 480}
 481
 482struct commit_list *copy_commit_list(struct commit_list *list)
 483{
 484        struct commit_list *head = NULL;
 485        struct commit_list **pp = &head;
 486        while (list) {
 487                pp = commit_list_append(list->item, pp);
 488                list = list->next;
 489        }
 490        return head;
 491}
 492
 493void free_commit_list(struct commit_list *list)
 494{
 495        while (list)
 496                pop_commit(&list);
 497}
 498
 499struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
 500{
 501        struct commit_list **pp = list;
 502        struct commit_list *p;
 503        while ((p = *pp) != NULL) {
 504                if (p->item->date < item->date) {
 505                        break;
 506                }
 507                pp = &p->next;
 508        }
 509        return commit_list_insert(item, pp);
 510}
 511
 512static int commit_list_compare_by_date(const void *a, const void *b)
 513{
 514        timestamp_t a_date = ((const struct commit_list *)a)->item->date;
 515        timestamp_t b_date = ((const struct commit_list *)b)->item->date;
 516        if (a_date < b_date)
 517                return 1;
 518        if (a_date > b_date)
 519                return -1;
 520        return 0;
 521}
 522
 523static void *commit_list_get_next(const void *a)
 524{
 525        return ((const struct commit_list *)a)->next;
 526}
 527
 528static void commit_list_set_next(void *a, void *next)
 529{
 530        ((struct commit_list *)a)->next = next;
 531}
 532
 533void commit_list_sort_by_date(struct commit_list **list)
 534{
 535        *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
 536                                commit_list_compare_by_date);
 537}
 538
 539struct commit *pop_most_recent_commit(struct commit_list **list,
 540                                      unsigned int mark)
 541{
 542        struct commit *ret = pop_commit(list);
 543        struct commit_list *parents = ret->parents;
 544
 545        while (parents) {
 546                struct commit *commit = parents->item;
 547                if (!parse_commit(commit) && !(commit->object.flags & mark)) {
 548                        commit->object.flags |= mark;
 549                        commit_list_insert_by_date(commit, list);
 550                }
 551                parents = parents->next;
 552        }
 553        return ret;
 554}
 555
 556static void clear_commit_marks_1(struct commit_list **plist,
 557                                 struct commit *commit, unsigned int mark)
 558{
 559        while (commit) {
 560                struct commit_list *parents;
 561
 562                if (!(mark & commit->object.flags))
 563                        return;
 564
 565                commit->object.flags &= ~mark;
 566
 567                parents = commit->parents;
 568                if (!parents)
 569                        return;
 570
 571                while ((parents = parents->next))
 572                        commit_list_insert(parents->item, plist);
 573
 574                commit = commit->parents->item;
 575        }
 576}
 577
 578void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
 579{
 580        struct commit_list *list = NULL;
 581
 582        while (nr--) {
 583                clear_commit_marks_1(&list, *commit, mark);
 584                commit++;
 585        }
 586        while (list)
 587                clear_commit_marks_1(&list, pop_commit(&list), mark);
 588}
 589
 590void clear_commit_marks(struct commit *commit, unsigned int mark)
 591{
 592        clear_commit_marks_many(1, &commit, mark);
 593}
 594
 595struct commit *pop_commit(struct commit_list **stack)
 596{
 597        struct commit_list *top = *stack;
 598        struct commit *item = top ? top->item : NULL;
 599
 600        if (top) {
 601                *stack = top->next;
 602                free(top);
 603        }
 604        return item;
 605}
 606
 607/*
 608 * Topological sort support
 609 */
 610
 611/* count number of children that have not been emitted */
 612define_commit_slab(indegree_slab, int);
 613
 614/* record author-date for each commit object */
 615define_commit_slab(author_date_slab, unsigned long);
 616
 617static void record_author_date(struct author_date_slab *author_date,
 618                               struct commit *commit)
 619{
 620        const char *buffer = get_commit_buffer(commit, NULL);
 621        struct ident_split ident;
 622        const char *ident_line;
 623        size_t ident_len;
 624        char *date_end;
 625        timestamp_t date;
 626
 627        ident_line = find_commit_header(buffer, "author", &ident_len);
 628        if (!ident_line)
 629                goto fail_exit; /* no author line */
 630        if (split_ident_line(&ident, ident_line, ident_len) ||
 631            !ident.date_begin || !ident.date_end)
 632                goto fail_exit; /* malformed "author" line */
 633
 634        date = parse_timestamp(ident.date_begin, &date_end, 10);
 635        if (date_end != ident.date_end)
 636                goto fail_exit; /* malformed date */
 637        *(author_date_slab_at(author_date, commit)) = date;
 638
 639fail_exit:
 640        unuse_commit_buffer(commit, buffer);
 641}
 642
 643static int compare_commits_by_author_date(const void *a_, const void *b_,
 644                                          void *cb_data)
 645{
 646        const struct commit *a = a_, *b = b_;
 647        struct author_date_slab *author_date = cb_data;
 648        timestamp_t a_date = *(author_date_slab_at(author_date, a));
 649        timestamp_t b_date = *(author_date_slab_at(author_date, b));
 650
 651        /* newer commits with larger date first */
 652        if (a_date < b_date)
 653                return 1;
 654        else if (a_date > b_date)
 655                return -1;
 656        return 0;
 657}
 658
 659int compare_commits_by_gen_then_commit_date(const void *a_, const void *b_, void *unused)
 660{
 661        const struct commit *a = a_, *b = b_;
 662
 663        /* newer commits first */
 664        if (a->generation < b->generation)
 665                return 1;
 666        else if (a->generation > b->generation)
 667                return -1;
 668
 669        /* use date as a heuristic when generations are equal */
 670        if (a->date < b->date)
 671                return 1;
 672        else if (a->date > b->date)
 673                return -1;
 674        return 0;
 675}
 676
 677int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
 678{
 679        const struct commit *a = a_, *b = b_;
 680        /* newer commits with larger date first */
 681        if (a->date < b->date)
 682                return 1;
 683        else if (a->date > b->date)
 684                return -1;
 685        return 0;
 686}
 687
 688/*
 689 * Performs an in-place topological sort on the list supplied.
 690 */
 691void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
 692{
 693        struct commit_list *next, *orig = *list;
 694        struct commit_list **pptr;
 695        struct indegree_slab indegree;
 696        struct prio_queue queue;
 697        struct commit *commit;
 698        struct author_date_slab author_date;
 699
 700        if (!orig)
 701                return;
 702        *list = NULL;
 703
 704        init_indegree_slab(&indegree);
 705        memset(&queue, '\0', sizeof(queue));
 706
 707        switch (sort_order) {
 708        default: /* REV_SORT_IN_GRAPH_ORDER */
 709                queue.compare = NULL;
 710                break;
 711        case REV_SORT_BY_COMMIT_DATE:
 712                queue.compare = compare_commits_by_commit_date;
 713                break;
 714        case REV_SORT_BY_AUTHOR_DATE:
 715                init_author_date_slab(&author_date);
 716                queue.compare = compare_commits_by_author_date;
 717                queue.cb_data = &author_date;
 718                break;
 719        }
 720
 721        /* Mark them and clear the indegree */
 722        for (next = orig; next; next = next->next) {
 723                struct commit *commit = next->item;
 724                *(indegree_slab_at(&indegree, commit)) = 1;
 725                /* also record the author dates, if needed */
 726                if (sort_order == REV_SORT_BY_AUTHOR_DATE)
 727                        record_author_date(&author_date, commit);
 728        }
 729
 730        /* update the indegree */
 731        for (next = orig; next; next = next->next) {
 732                struct commit_list *parents = next->item->parents;
 733                while (parents) {
 734                        struct commit *parent = parents->item;
 735                        int *pi = indegree_slab_at(&indegree, parent);
 736
 737                        if (*pi)
 738                                (*pi)++;
 739                        parents = parents->next;
 740                }
 741        }
 742
 743        /*
 744         * find the tips
 745         *
 746         * tips are nodes not reachable from any other node in the list
 747         *
 748         * the tips serve as a starting set for the work queue.
 749         */
 750        for (next = orig; next; next = next->next) {
 751                struct commit *commit = next->item;
 752
 753                if (*(indegree_slab_at(&indegree, commit)) == 1)
 754                        prio_queue_put(&queue, commit);
 755        }
 756
 757        /*
 758         * This is unfortunate; the initial tips need to be shown
 759         * in the order given from the revision traversal machinery.
 760         */
 761        if (sort_order == REV_SORT_IN_GRAPH_ORDER)
 762                prio_queue_reverse(&queue);
 763
 764        /* We no longer need the commit list */
 765        free_commit_list(orig);
 766
 767        pptr = list;
 768        *list = NULL;
 769        while ((commit = prio_queue_get(&queue)) != NULL) {
 770                struct commit_list *parents;
 771
 772                for (parents = commit->parents; parents ; parents = parents->next) {
 773                        struct commit *parent = parents->item;
 774                        int *pi = indegree_slab_at(&indegree, parent);
 775
 776                        if (!*pi)
 777                                continue;
 778
 779                        /*
 780                         * parents are only enqueued for emission
 781                         * when all their children have been emitted thereby
 782                         * guaranteeing topological order.
 783                         */
 784                        if (--(*pi) == 1)
 785                                prio_queue_put(&queue, parent);
 786                }
 787                /*
 788                 * all children of commit have already been
 789                 * emitted. we can emit it now.
 790                 */
 791                *(indegree_slab_at(&indegree, commit)) = 0;
 792
 793                pptr = &commit_list_insert(commit, pptr)->next;
 794        }
 795
 796        clear_indegree_slab(&indegree);
 797        clear_prio_queue(&queue);
 798        if (sort_order == REV_SORT_BY_AUTHOR_DATE)
 799                clear_author_date_slab(&author_date);
 800}
 801
 802/* merge-base stuff */
 803
 804/* Remember to update object flag allocation in object.h */
 805#define PARENT1         (1u<<16)
 806#define PARENT2         (1u<<17)
 807#define STALE           (1u<<18)
 808#define RESULT          (1u<<19)
 809
 810static const unsigned all_flags = (PARENT1 | PARENT2 | STALE | RESULT);
 811
 812static int queue_has_nonstale(struct prio_queue *queue)
 813{
 814        int i;
 815        for (i = 0; i < queue->nr; i++) {
 816                struct commit *commit = queue->array[i].data;
 817                if (!(commit->object.flags & STALE))
 818                        return 1;
 819        }
 820        return 0;
 821}
 822
 823/* all input commits in one and twos[] must have been parsed! */
 824static struct commit_list *paint_down_to_common(struct commit *one, int n,
 825                                                struct commit **twos,
 826                                                int min_generation)
 827{
 828        struct prio_queue queue = { compare_commits_by_gen_then_commit_date };
 829        struct commit_list *result = NULL;
 830        int i;
 831        uint32_t last_gen = GENERATION_NUMBER_INFINITY;
 832
 833        one->object.flags |= PARENT1;
 834        if (!n) {
 835                commit_list_append(one, &result);
 836                return result;
 837        }
 838        prio_queue_put(&queue, one);
 839
 840        for (i = 0; i < n; i++) {
 841                twos[i]->object.flags |= PARENT2;
 842                prio_queue_put(&queue, twos[i]);
 843        }
 844
 845        while (queue_has_nonstale(&queue)) {
 846                struct commit *commit = prio_queue_get(&queue);
 847                struct commit_list *parents;
 848                int flags;
 849
 850                if (commit->generation > last_gen)
 851                        BUG("bad generation skip %8x > %8x at %s",
 852                            commit->generation, last_gen,
 853                            oid_to_hex(&commit->object.oid));
 854                last_gen = commit->generation;
 855
 856                if (commit->generation < min_generation)
 857                        break;
 858
 859                flags = commit->object.flags & (PARENT1 | PARENT2 | STALE);
 860                if (flags == (PARENT1 | PARENT2)) {
 861                        if (!(commit->object.flags & RESULT)) {
 862                                commit->object.flags |= RESULT;
 863                                commit_list_insert_by_date(commit, &result);
 864                        }
 865                        /* Mark parents of a found merge stale */
 866                        flags |= STALE;
 867                }
 868                parents = commit->parents;
 869                while (parents) {
 870                        struct commit *p = parents->item;
 871                        parents = parents->next;
 872                        if ((p->object.flags & flags) == flags)
 873                                continue;
 874                        if (parse_commit(p))
 875                                return NULL;
 876                        p->object.flags |= flags;
 877                        prio_queue_put(&queue, p);
 878                }
 879        }
 880
 881        clear_prio_queue(&queue);
 882        return result;
 883}
 884
 885static struct commit_list *merge_bases_many(struct commit *one, int n, struct commit **twos)
 886{
 887        struct commit_list *list = NULL;
 888        struct commit_list *result = NULL;
 889        int i;
 890
 891        for (i = 0; i < n; i++) {
 892                if (one == twos[i])
 893                        /*
 894                         * We do not mark this even with RESULT so we do not
 895                         * have to clean it up.
 896                         */
 897                        return commit_list_insert(one, &result);
 898        }
 899
 900        if (parse_commit(one))
 901                return NULL;
 902        for (i = 0; i < n; i++) {
 903                if (parse_commit(twos[i]))
 904                        return NULL;
 905        }
 906
 907        list = paint_down_to_common(one, n, twos, 0);
 908
 909        while (list) {
 910                struct commit *commit = pop_commit(&list);
 911                if (!(commit->object.flags & STALE))
 912                        commit_list_insert_by_date(commit, &result);
 913        }
 914        return result;
 915}
 916
 917struct commit_list *get_octopus_merge_bases(struct commit_list *in)
 918{
 919        struct commit_list *i, *j, *k, *ret = NULL;
 920
 921        if (!in)
 922                return ret;
 923
 924        commit_list_insert(in->item, &ret);
 925
 926        for (i = in->next; i; i = i->next) {
 927                struct commit_list *new_commits = NULL, *end = NULL;
 928
 929                for (j = ret; j; j = j->next) {
 930                        struct commit_list *bases;
 931                        bases = get_merge_bases(i->item, j->item);
 932                        if (!new_commits)
 933                                new_commits = bases;
 934                        else
 935                                end->next = bases;
 936                        for (k = bases; k; k = k->next)
 937                                end = k;
 938                }
 939                ret = new_commits;
 940        }
 941        return ret;
 942}
 943
 944static int remove_redundant(struct commit **array, int cnt)
 945{
 946        /*
 947         * Some commit in the array may be an ancestor of
 948         * another commit.  Move such commit to the end of
 949         * the array, and return the number of commits that
 950         * are independent from each other.
 951         */
 952        struct commit **work;
 953        unsigned char *redundant;
 954        int *filled_index;
 955        int i, j, filled;
 956
 957        work = xcalloc(cnt, sizeof(*work));
 958        redundant = xcalloc(cnt, 1);
 959        ALLOC_ARRAY(filled_index, cnt - 1);
 960
 961        for (i = 0; i < cnt; i++)
 962                parse_commit(array[i]);
 963        for (i = 0; i < cnt; i++) {
 964                struct commit_list *common;
 965                uint32_t min_generation = array[i]->generation;
 966
 967                if (redundant[i])
 968                        continue;
 969                for (j = filled = 0; j < cnt; j++) {
 970                        if (i == j || redundant[j])
 971                                continue;
 972                        filled_index[filled] = j;
 973                        work[filled++] = array[j];
 974
 975                        if (array[j]->generation < min_generation)
 976                                min_generation = array[j]->generation;
 977                }
 978                common = paint_down_to_common(array[i], filled, work,
 979                                              min_generation);
 980                if (array[i]->object.flags & PARENT2)
 981                        redundant[i] = 1;
 982                for (j = 0; j < filled; j++)
 983                        if (work[j]->object.flags & PARENT1)
 984                                redundant[filled_index[j]] = 1;
 985                clear_commit_marks(array[i], all_flags);
 986                clear_commit_marks_many(filled, work, all_flags);
 987                free_commit_list(common);
 988        }
 989
 990        /* Now collect the result */
 991        COPY_ARRAY(work, array, cnt);
 992        for (i = filled = 0; i < cnt; i++)
 993                if (!redundant[i])
 994                        array[filled++] = work[i];
 995        for (j = filled, i = 0; i < cnt; i++)
 996                if (redundant[i])
 997                        array[j++] = work[i];
 998        free(work);
 999        free(redundant);
1000        free(filled_index);
1001        return filled;
1002}
1003
1004static struct commit_list *get_merge_bases_many_0(struct commit *one,
1005                                                  int n,
1006                                                  struct commit **twos,
1007                                                  int cleanup)
1008{
1009        struct commit_list *list;
1010        struct commit **rslt;
1011        struct commit_list *result;
1012        int cnt, i;
1013
1014        result = merge_bases_many(one, n, twos);
1015        for (i = 0; i < n; i++) {
1016                if (one == twos[i])
1017                        return result;
1018        }
1019        if (!result || !result->next) {
1020                if (cleanup) {
1021                        clear_commit_marks(one, all_flags);
1022                        clear_commit_marks_many(n, twos, all_flags);
1023                }
1024                return result;
1025        }
1026
1027        /* There are more than one */
1028        cnt = commit_list_count(result);
1029        rslt = xcalloc(cnt, sizeof(*rslt));
1030        for (list = result, i = 0; list; list = list->next)
1031                rslt[i++] = list->item;
1032        free_commit_list(result);
1033
1034        clear_commit_marks(one, all_flags);
1035        clear_commit_marks_many(n, twos, all_flags);
1036
1037        cnt = remove_redundant(rslt, cnt);
1038        result = NULL;
1039        for (i = 0; i < cnt; i++)
1040                commit_list_insert_by_date(rslt[i], &result);
1041        free(rslt);
1042        return result;
1043}
1044
1045struct commit_list *get_merge_bases_many(struct commit *one,
1046                                         int n,
1047                                         struct commit **twos)
1048{
1049        return get_merge_bases_many_0(one, n, twos, 1);
1050}
1051
1052struct commit_list *get_merge_bases_many_dirty(struct commit *one,
1053                                               int n,
1054                                               struct commit **twos)
1055{
1056        return get_merge_bases_many_0(one, n, twos, 0);
1057}
1058
1059struct commit_list *get_merge_bases(struct commit *one, struct commit *two)
1060{
1061        return get_merge_bases_many_0(one, 1, &two, 1);
1062}
1063
1064/*
1065 * Is "commit" a descendant of one of the elements on the "with_commit" list?
1066 */
1067int is_descendant_of(struct commit *commit, struct commit_list *with_commit)
1068{
1069        if (!with_commit)
1070                return 1;
1071        while (with_commit) {
1072                struct commit *other;
1073
1074                other = with_commit->item;
1075                with_commit = with_commit->next;
1076                if (in_merge_bases(other, commit))
1077                        return 1;
1078        }
1079        return 0;
1080}
1081
1082/*
1083 * Is "commit" an ancestor of one of the "references"?
1084 */
1085int in_merge_bases_many(struct commit *commit, int nr_reference, struct commit **reference)
1086{
1087        struct commit_list *bases;
1088        int ret = 0, i;
1089        uint32_t min_generation = GENERATION_NUMBER_INFINITY;
1090
1091        if (parse_commit(commit))
1092                return ret;
1093        for (i = 0; i < nr_reference; i++) {
1094                if (parse_commit(reference[i]))
1095                        return ret;
1096                if (reference[i]->generation < min_generation)
1097                        min_generation = reference[i]->generation;
1098        }
1099
1100        if (commit->generation > min_generation)
1101                return ret;
1102
1103        bases = paint_down_to_common(commit, nr_reference, reference, commit->generation);
1104        if (commit->object.flags & PARENT2)
1105                ret = 1;
1106        clear_commit_marks(commit, all_flags);
1107        clear_commit_marks_many(nr_reference, reference, all_flags);
1108        free_commit_list(bases);
1109        return ret;
1110}
1111
1112/*
1113 * Is "commit" an ancestor of (i.e. reachable from) the "reference"?
1114 */
1115int in_merge_bases(struct commit *commit, struct commit *reference)
1116{
1117        return in_merge_bases_many(commit, 1, &reference);
1118}
1119
1120struct commit_list *reduce_heads(struct commit_list *heads)
1121{
1122        struct commit_list *p;
1123        struct commit_list *result = NULL, **tail = &result;
1124        struct commit **array;
1125        int num_head, i;
1126
1127        if (!heads)
1128                return NULL;
1129
1130        /* Uniquify */
1131        for (p = heads; p; p = p->next)
1132                p->item->object.flags &= ~STALE;
1133        for (p = heads, num_head = 0; p; p = p->next) {
1134                if (p->item->object.flags & STALE)
1135                        continue;
1136                p->item->object.flags |= STALE;
1137                num_head++;
1138        }
1139        array = xcalloc(num_head, sizeof(*array));
1140        for (p = heads, i = 0; p; p = p->next) {
1141                if (p->item->object.flags & STALE) {
1142                        array[i++] = p->item;
1143                        p->item->object.flags &= ~STALE;
1144                }
1145        }
1146        num_head = remove_redundant(array, num_head);
1147        for (i = 0; i < num_head; i++)
1148                tail = &commit_list_insert(array[i], tail)->next;
1149        free(array);
1150        return result;
1151}
1152
1153void reduce_heads_replace(struct commit_list **heads)
1154{
1155        struct commit_list *result = reduce_heads(*heads);
1156        free_commit_list(*heads);
1157        *heads = result;
1158}
1159
1160static const char gpg_sig_header[] = "gpgsig";
1161static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
1162
1163static int do_sign_commit(struct strbuf *buf, const char *keyid)
1164{
1165        struct strbuf sig = STRBUF_INIT;
1166        int inspos, copypos;
1167        const char *eoh;
1168
1169        /* find the end of the header */
1170        eoh = strstr(buf->buf, "\n\n");
1171        if (!eoh)
1172                inspos = buf->len;
1173        else
1174                inspos = eoh - buf->buf + 1;
1175
1176        if (!keyid || !*keyid)
1177                keyid = get_signing_key();
1178        if (sign_buffer(buf, &sig, keyid)) {
1179                strbuf_release(&sig);
1180                return -1;
1181        }
1182
1183        for (copypos = 0; sig.buf[copypos]; ) {
1184                const char *bol = sig.buf + copypos;
1185                const char *eol = strchrnul(bol, '\n');
1186                int len = (eol - bol) + !!*eol;
1187
1188                if (!copypos) {
1189                        strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
1190                        inspos += gpg_sig_header_len;
1191                }
1192                strbuf_insert(buf, inspos++, " ", 1);
1193                strbuf_insert(buf, inspos, bol, len);
1194                inspos += len;
1195                copypos += len;
1196        }
1197        strbuf_release(&sig);
1198        return 0;
1199}
1200
1201int parse_signed_commit(const struct commit *commit,
1202                        struct strbuf *payload, struct strbuf *signature)
1203{
1204
1205        unsigned long size;
1206        const char *buffer = get_commit_buffer(commit, &size);
1207        int in_signature, saw_signature = -1;
1208        const char *line, *tail;
1209
1210        line = buffer;
1211        tail = buffer + size;
1212        in_signature = 0;
1213        saw_signature = 0;
1214        while (line < tail) {
1215                const char *sig = NULL;
1216                const char *next = memchr(line, '\n', tail - line);
1217
1218                next = next ? next + 1 : tail;
1219                if (in_signature && line[0] == ' ')
1220                        sig = line + 1;
1221                else if (starts_with(line, gpg_sig_header) &&
1222                         line[gpg_sig_header_len] == ' ')
1223                        sig = line + gpg_sig_header_len + 1;
1224                if (sig) {
1225                        strbuf_add(signature, sig, next - sig);
1226                        saw_signature = 1;
1227                        in_signature = 1;
1228                } else {
1229                        if (*line == '\n')
1230                                /* dump the whole remainder of the buffer */
1231                                next = tail;
1232                        strbuf_add(payload, line, next - line);
1233                        in_signature = 0;
1234                }
1235                line = next;
1236        }
1237        unuse_commit_buffer(commit, buffer);
1238        return saw_signature;
1239}
1240
1241int remove_signature(struct strbuf *buf)
1242{
1243        const char *line = buf->buf;
1244        const char *tail = buf->buf + buf->len;
1245        int in_signature = 0;
1246        const char *sig_start = NULL;
1247        const char *sig_end = NULL;
1248
1249        while (line < tail) {
1250                const char *next = memchr(line, '\n', tail - line);
1251                next = next ? next + 1 : tail;
1252
1253                if (in_signature && line[0] == ' ')
1254                        sig_end = next;
1255                else if (starts_with(line, gpg_sig_header) &&
1256                         line[gpg_sig_header_len] == ' ') {
1257                        sig_start = line;
1258                        sig_end = next;
1259                        in_signature = 1;
1260                } else {
1261                        if (*line == '\n')
1262                                /* dump the whole remainder of the buffer */
1263                                next = tail;
1264                        in_signature = 0;
1265                }
1266                line = next;
1267        }
1268
1269        if (sig_start)
1270                strbuf_remove(buf, sig_start - buf->buf, sig_end - sig_start);
1271
1272        return sig_start != NULL;
1273}
1274
1275static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1276{
1277        struct merge_remote_desc *desc;
1278        struct commit_extra_header *mergetag;
1279        char *buf;
1280        unsigned long size, len;
1281        enum object_type type;
1282
1283        desc = merge_remote_util(parent);
1284        if (!desc || !desc->obj)
1285                return;
1286        buf = read_object_file(&desc->obj->oid, &type, &size);
1287        if (!buf || type != OBJ_TAG)
1288                goto free_return;
1289        len = parse_signature(buf, size);
1290        if (size == len)
1291                goto free_return;
1292        /*
1293         * We could verify this signature and either omit the tag when
1294         * it does not validate, but the integrator may not have the
1295         * public key of the signer of the tag he is merging, while a
1296         * later auditor may have it while auditing, so let's not run
1297         * verify-signed-buffer here for now...
1298         *
1299         * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1300         *      warn("warning: signed tag unverified.");
1301         */
1302        mergetag = xcalloc(1, sizeof(*mergetag));
1303        mergetag->key = xstrdup("mergetag");
1304        mergetag->value = buf;
1305        mergetag->len = size;
1306
1307        **tail = mergetag;
1308        *tail = &mergetag->next;
1309        return;
1310
1311free_return:
1312        free(buf);
1313}
1314
1315int check_commit_signature(const struct commit *commit, struct signature_check *sigc)
1316{
1317        struct strbuf payload = STRBUF_INIT;
1318        struct strbuf signature = STRBUF_INIT;
1319        int ret = 1;
1320
1321        sigc->result = 'N';
1322
1323        if (parse_signed_commit(commit, &payload, &signature) <= 0)
1324                goto out;
1325        ret = check_signature(payload.buf, payload.len, signature.buf,
1326                signature.len, sigc);
1327
1328 out:
1329        strbuf_release(&payload);
1330        strbuf_release(&signature);
1331
1332        return ret;
1333}
1334
1335
1336
1337void append_merge_tag_headers(struct commit_list *parents,
1338                              struct commit_extra_header ***tail)
1339{
1340        while (parents) {
1341                struct commit *parent = parents->item;
1342                handle_signed_tag(parent, tail);
1343                parents = parents->next;
1344        }
1345}
1346
1347static void add_extra_header(struct strbuf *buffer,
1348                             struct commit_extra_header *extra)
1349{
1350        strbuf_addstr(buffer, extra->key);
1351        if (extra->len)
1352                strbuf_add_lines(buffer, " ", extra->value, extra->len);
1353        else
1354                strbuf_addch(buffer, '\n');
1355}
1356
1357struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1358                                                      const char **exclude)
1359{
1360        struct commit_extra_header *extra = NULL;
1361        unsigned long size;
1362        const char *buffer = get_commit_buffer(commit, &size);
1363        extra = read_commit_extra_header_lines(buffer, size, exclude);
1364        unuse_commit_buffer(commit, buffer);
1365        return extra;
1366}
1367
1368int for_each_mergetag(each_mergetag_fn fn, struct commit *commit, void *data)
1369{
1370        struct commit_extra_header *extra, *to_free;
1371        int res = 0;
1372
1373        to_free = read_commit_extra_headers(commit, NULL);
1374        for (extra = to_free; !res && extra; extra = extra->next) {
1375                if (strcmp(extra->key, "mergetag"))
1376                        continue; /* not a merge tag */
1377                res = fn(commit, extra, data);
1378        }
1379        free_commit_extra_headers(to_free);
1380        return res;
1381}
1382
1383static inline int standard_header_field(const char *field, size_t len)
1384{
1385        return ((len == 4 && !memcmp(field, "tree", 4)) ||
1386                (len == 6 && !memcmp(field, "parent", 6)) ||
1387                (len == 6 && !memcmp(field, "author", 6)) ||
1388                (len == 9 && !memcmp(field, "committer", 9)) ||
1389                (len == 8 && !memcmp(field, "encoding", 8)));
1390}
1391
1392static int excluded_header_field(const char *field, size_t len, const char **exclude)
1393{
1394        if (!exclude)
1395                return 0;
1396
1397        while (*exclude) {
1398                size_t xlen = strlen(*exclude);
1399                if (len == xlen && !memcmp(field, *exclude, xlen))
1400                        return 1;
1401                exclude++;
1402        }
1403        return 0;
1404}
1405
1406static struct commit_extra_header *read_commit_extra_header_lines(
1407        const char *buffer, size_t size,
1408        const char **exclude)
1409{
1410        struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1411        const char *line, *next, *eof, *eob;
1412        struct strbuf buf = STRBUF_INIT;
1413
1414        for (line = buffer, eob = line + size;
1415             line < eob && *line != '\n';
1416             line = next) {
1417                next = memchr(line, '\n', eob - line);
1418                next = next ? next + 1 : eob;
1419                if (*line == ' ') {
1420                        /* continuation */
1421                        if (it)
1422                                strbuf_add(&buf, line + 1, next - (line + 1));
1423                        continue;
1424                }
1425                if (it)
1426                        it->value = strbuf_detach(&buf, &it->len);
1427                strbuf_reset(&buf);
1428                it = NULL;
1429
1430                eof = memchr(line, ' ', next - line);
1431                if (!eof)
1432                        eof = next;
1433                else if (standard_header_field(line, eof - line) ||
1434                         excluded_header_field(line, eof - line, exclude))
1435                        continue;
1436
1437                it = xcalloc(1, sizeof(*it));
1438                it->key = xmemdupz(line, eof-line);
1439                *tail = it;
1440                tail = &it->next;
1441                if (eof + 1 < next)
1442                        strbuf_add(&buf, eof + 1, next - (eof + 1));
1443        }
1444        if (it)
1445                it->value = strbuf_detach(&buf, &it->len);
1446        return extra;
1447}
1448
1449void free_commit_extra_headers(struct commit_extra_header *extra)
1450{
1451        while (extra) {
1452                struct commit_extra_header *next = extra->next;
1453                free(extra->key);
1454                free(extra->value);
1455                free(extra);
1456                extra = next;
1457        }
1458}
1459
1460int commit_tree(const char *msg, size_t msg_len, const struct object_id *tree,
1461                struct commit_list *parents, struct object_id *ret,
1462                const char *author, const char *sign_commit)
1463{
1464        struct commit_extra_header *extra = NULL, **tail = &extra;
1465        int result;
1466
1467        append_merge_tag_headers(parents, &tail);
1468        result = commit_tree_extended(msg, msg_len, tree, parents, ret,
1469                                      author, sign_commit, extra);
1470        free_commit_extra_headers(extra);
1471        return result;
1472}
1473
1474static int find_invalid_utf8(const char *buf, int len)
1475{
1476        int offset = 0;
1477        static const unsigned int max_codepoint[] = {
1478                0x7f, 0x7ff, 0xffff, 0x10ffff
1479        };
1480
1481        while (len) {
1482                unsigned char c = *buf++;
1483                int bytes, bad_offset;
1484                unsigned int codepoint;
1485                unsigned int min_val, max_val;
1486
1487                len--;
1488                offset++;
1489
1490                /* Simple US-ASCII? No worries. */
1491                if (c < 0x80)
1492                        continue;
1493
1494                bad_offset = offset-1;
1495
1496                /*
1497                 * Count how many more high bits set: that's how
1498                 * many more bytes this sequence should have.
1499                 */
1500                bytes = 0;
1501                while (c & 0x40) {
1502                        c <<= 1;
1503                        bytes++;
1504                }
1505
1506                /*
1507                 * Must be between 1 and 3 more bytes.  Longer sequences result in
1508                 * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1509                 */
1510                if (bytes < 1 || 3 < bytes)
1511                        return bad_offset;
1512
1513                /* Do we *have* that many bytes? */
1514                if (len < bytes)
1515                        return bad_offset;
1516
1517                /*
1518                 * Place the encoded bits at the bottom of the value and compute the
1519                 * valid range.
1520                 */
1521                codepoint = (c & 0x7f) >> bytes;
1522                min_val = max_codepoint[bytes-1] + 1;
1523                max_val = max_codepoint[bytes];
1524
1525                offset += bytes;
1526                len -= bytes;
1527
1528                /* And verify that they are good continuation bytes */
1529                do {
1530                        codepoint <<= 6;
1531                        codepoint |= *buf & 0x3f;
1532                        if ((*buf++ & 0xc0) != 0x80)
1533                                return bad_offset;
1534                } while (--bytes);
1535
1536                /* Reject codepoints that are out of range for the sequence length. */
1537                if (codepoint < min_val || codepoint > max_val)
1538                        return bad_offset;
1539                /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1540                if ((codepoint & 0x1ff800) == 0xd800)
1541                        return bad_offset;
1542                /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1543                if ((codepoint & 0xfffe) == 0xfffe)
1544                        return bad_offset;
1545                /* So are anything in the range U+FDD0..U+FDEF. */
1546                if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1547                        return bad_offset;
1548        }
1549        return -1;
1550}
1551
1552/*
1553 * This verifies that the buffer is in proper utf8 format.
1554 *
1555 * If it isn't, it assumes any non-utf8 characters are Latin1,
1556 * and does the conversion.
1557 */
1558static int verify_utf8(struct strbuf *buf)
1559{
1560        int ok = 1;
1561        long pos = 0;
1562
1563        for (;;) {
1564                int bad;
1565                unsigned char c;
1566                unsigned char replace[2];
1567
1568                bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1569                if (bad < 0)
1570                        return ok;
1571                pos += bad;
1572                ok = 0;
1573                c = buf->buf[pos];
1574                strbuf_remove(buf, pos, 1);
1575
1576                /* We know 'c' must be in the range 128-255 */
1577                replace[0] = 0xc0 + (c >> 6);
1578                replace[1] = 0x80 + (c & 0x3f);
1579                strbuf_insert(buf, pos, replace, 2);
1580                pos += 2;
1581        }
1582}
1583
1584static const char commit_utf8_warn[] =
1585N_("Warning: commit message did not conform to UTF-8.\n"
1586   "You may want to amend it after fixing the message, or set the config\n"
1587   "variable i18n.commitencoding to the encoding your project uses.\n");
1588
1589int commit_tree_extended(const char *msg, size_t msg_len,
1590                         const struct object_id *tree,
1591                         struct commit_list *parents, struct object_id *ret,
1592                         const char *author, const char *sign_commit,
1593                         struct commit_extra_header *extra)
1594{
1595        int result;
1596        int encoding_is_utf8;
1597        struct strbuf buffer;
1598
1599        assert_oid_type(tree, OBJ_TREE);
1600
1601        if (memchr(msg, '\0', msg_len))
1602                return error("a NUL byte in commit log message not allowed.");
1603
1604        /* Not having i18n.commitencoding is the same as having utf-8 */
1605        encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1606
1607        strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1608        strbuf_addf(&buffer, "tree %s\n", oid_to_hex(tree));
1609
1610        /*
1611         * NOTE! This ordering means that the same exact tree merged with a
1612         * different order of parents will be a _different_ changeset even
1613         * if everything else stays the same.
1614         */
1615        while (parents) {
1616                struct commit *parent = pop_commit(&parents);
1617                strbuf_addf(&buffer, "parent %s\n",
1618                            oid_to_hex(&parent->object.oid));
1619        }
1620
1621        /* Person/date information */
1622        if (!author)
1623                author = git_author_info(IDENT_STRICT);
1624        strbuf_addf(&buffer, "author %s\n", author);
1625        strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1626        if (!encoding_is_utf8)
1627                strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1628
1629        while (extra) {
1630                add_extra_header(&buffer, extra);
1631                extra = extra->next;
1632        }
1633        strbuf_addch(&buffer, '\n');
1634
1635        /* And add the comment */
1636        strbuf_add(&buffer, msg, msg_len);
1637
1638        /* And check the encoding */
1639        if (encoding_is_utf8 && !verify_utf8(&buffer))
1640                fprintf(stderr, _(commit_utf8_warn));
1641
1642        if (sign_commit && do_sign_commit(&buffer, sign_commit)) {
1643                result = -1;
1644                goto out;
1645        }
1646
1647        result = write_object_file(buffer.buf, buffer.len, commit_type, ret);
1648out:
1649        strbuf_release(&buffer);
1650        return result;
1651}
1652
1653define_commit_slab(merge_desc_slab, struct merge_remote_desc *);
1654static struct merge_desc_slab merge_desc_slab = COMMIT_SLAB_INIT(1, merge_desc_slab);
1655
1656struct merge_remote_desc *merge_remote_util(struct commit *commit)
1657{
1658        return *merge_desc_slab_at(&merge_desc_slab, commit);
1659}
1660
1661void set_merge_remote_desc(struct commit *commit,
1662                           const char *name, struct object *obj)
1663{
1664        struct merge_remote_desc *desc;
1665        FLEX_ALLOC_STR(desc, name, name);
1666        desc->obj = obj;
1667        *merge_desc_slab_at(&merge_desc_slab, commit) = desc;
1668}
1669
1670struct commit *get_merge_parent(const char *name)
1671{
1672        struct object *obj;
1673        struct commit *commit;
1674        struct object_id oid;
1675        if (get_oid(name, &oid))
1676                return NULL;
1677        obj = parse_object(&oid);
1678        commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1679        if (commit && !merge_remote_util(commit))
1680                set_merge_remote_desc(commit, name, obj);
1681        return commit;
1682}
1683
1684/*
1685 * Append a commit to the end of the commit_list.
1686 *
1687 * next starts by pointing to the variable that holds the head of an
1688 * empty commit_list, and is updated to point to the "next" field of
1689 * the last item on the list as new commits are appended.
1690 *
1691 * Usage example:
1692 *
1693 *     struct commit_list *list;
1694 *     struct commit_list **next = &list;
1695 *
1696 *     next = commit_list_append(c1, next);
1697 *     next = commit_list_append(c2, next);
1698 *     assert(commit_list_count(list) == 2);
1699 *     return list;
1700 */
1701struct commit_list **commit_list_append(struct commit *commit,
1702                                        struct commit_list **next)
1703{
1704        struct commit_list *new_commit = xmalloc(sizeof(struct commit_list));
1705        new_commit->item = commit;
1706        *next = new_commit;
1707        new_commit->next = NULL;
1708        return &new_commit->next;
1709}
1710
1711const char *find_commit_header(const char *msg, const char *key, size_t *out_len)
1712{
1713        int key_len = strlen(key);
1714        const char *line = msg;
1715
1716        while (line) {
1717                const char *eol = strchrnul(line, '\n');
1718
1719                if (line == eol)
1720                        return NULL;
1721
1722                if (eol - line > key_len &&
1723                    !strncmp(line, key, key_len) &&
1724                    line[key_len] == ' ') {
1725                        *out_len = eol - line - key_len - 1;
1726                        return line + key_len + 1;
1727                }
1728                line = *eol ? eol + 1 : NULL;
1729        }
1730        return NULL;
1731}
1732
1733/*
1734 * Inspect the given string and determine the true "end" of the log message, in
1735 * order to find where to put a new Signed-off-by: line.  Ignored are
1736 * trailing comment lines and blank lines.  To support "git commit -s
1737 * --amend" on an existing commit, we also ignore "Conflicts:".  To
1738 * support "git commit -v", we truncate at cut lines.
1739 *
1740 * Returns the number of bytes from the tail to ignore, to be fed as
1741 * the second parameter to append_signoff().
1742 */
1743int ignore_non_trailer(const char *buf, size_t len)
1744{
1745        int boc = 0;
1746        int bol = 0;
1747        int in_old_conflicts_block = 0;
1748        size_t cutoff = wt_status_locate_end(buf, len);
1749
1750        while (bol < cutoff) {
1751                const char *next_line = memchr(buf + bol, '\n', len - bol);
1752
1753                if (!next_line)
1754                        next_line = buf + len;
1755                else
1756                        next_line++;
1757
1758                if (buf[bol] == comment_line_char || buf[bol] == '\n') {
1759                        /* is this the first of the run of comments? */
1760                        if (!boc)
1761                                boc = bol;
1762                        /* otherwise, it is just continuing */
1763                } else if (starts_with(buf + bol, "Conflicts:\n")) {
1764                        in_old_conflicts_block = 1;
1765                        if (!boc)
1766                                boc = bol;
1767                } else if (in_old_conflicts_block && buf[bol] == '\t') {
1768                        ; /* a pathname in the conflicts block */
1769                } else if (boc) {
1770                        /* the previous was not trailing comment */
1771                        boc = 0;
1772                        in_old_conflicts_block = 0;
1773                }
1774                bol = next_line - buf;
1775        }
1776        return boc ? len - boc : len - cutoff;
1777}