commit.con commit commit: use generation number in remove_redundant() (04bc8d1)
   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
  17static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
  18
  19int save_commit_buffer = 1;
  20
  21const char *commit_type = "commit";
  22
  23struct commit *lookup_commit_reference_gently(const struct object_id *oid,
  24                                              int quiet)
  25{
  26        struct object *obj = deref_tag(parse_object(oid), NULL, 0);
  27
  28        if (!obj)
  29                return NULL;
  30        return object_as_type(obj, OBJ_COMMIT, quiet);
  31}
  32
  33struct commit *lookup_commit_reference(const struct object_id *oid)
  34{
  35        return lookup_commit_reference_gently(oid, 0);
  36}
  37
  38struct commit *lookup_commit_or_die(const struct object_id *oid, const char *ref_name)
  39{
  40        struct commit *c = lookup_commit_reference(oid);
  41        if (!c)
  42                die(_("could not parse %s"), ref_name);
  43        if (oidcmp(oid, &c->object.oid)) {
  44                warning(_("%s %s is not a commit!"),
  45                        ref_name, oid_to_hex(oid));
  46        }
  47        return c;
  48}
  49
  50struct commit *lookup_commit(const struct object_id *oid)
  51{
  52        struct object *obj = lookup_object(oid->hash);
  53        if (!obj)
  54                return create_object(oid->hash, alloc_commit_node());
  55        return object_as_type(obj, OBJ_COMMIT, 0);
  56}
  57
  58struct commit *lookup_commit_reference_by_name(const char *name)
  59{
  60        struct object_id oid;
  61        struct commit *commit;
  62
  63        if (get_oid_committish(name, &oid))
  64                return NULL;
  65        commit = lookup_commit_reference(&oid);
  66        if (parse_commit(commit))
  67                return NULL;
  68        return commit;
  69}
  70
  71static timestamp_t parse_commit_date(const char *buf, const char *tail)
  72{
  73        const char *dateptr;
  74
  75        if (buf + 6 >= tail)
  76                return 0;
  77        if (memcmp(buf, "author", 6))
  78                return 0;
  79        while (buf < tail && *buf++ != '\n')
  80                /* nada */;
  81        if (buf + 9 >= tail)
  82                return 0;
  83        if (memcmp(buf, "committer", 9))
  84                return 0;
  85        while (buf < tail && *buf++ != '>')
  86                /* nada */;
  87        if (buf >= tail)
  88                return 0;
  89        dateptr = buf;
  90        while (buf < tail && *buf++ != '\n')
  91                /* nada */;
  92        if (buf >= tail)
  93                return 0;
  94        /* dateptr < buf && buf[-1] == '\n', so parsing will stop at buf-1 */
  95        return parse_timestamp(dateptr, NULL, 10);
  96}
  97
  98static struct commit_graft **commit_graft;
  99static int commit_graft_alloc, commit_graft_nr;
 100
 101static const unsigned char *commit_graft_sha1_access(size_t index, void *table)
 102{
 103        struct commit_graft **commit_graft_table = table;
 104        return commit_graft_table[index]->oid.hash;
 105}
 106
 107static int commit_graft_pos(const unsigned char *sha1)
 108{
 109        return sha1_pos(sha1, commit_graft, commit_graft_nr,
 110                        commit_graft_sha1_access);
 111}
 112
 113int register_commit_graft(struct commit_graft *graft, int ignore_dups)
 114{
 115        int pos = commit_graft_pos(graft->oid.hash);
 116
 117        if (0 <= pos) {
 118                if (ignore_dups)
 119                        free(graft);
 120                else {
 121                        free(commit_graft[pos]);
 122                        commit_graft[pos] = graft;
 123                }
 124                return 1;
 125        }
 126        pos = -pos - 1;
 127        ALLOC_GROW(commit_graft, commit_graft_nr + 1, commit_graft_alloc);
 128        commit_graft_nr++;
 129        if (pos < commit_graft_nr)
 130                MOVE_ARRAY(commit_graft + pos + 1, commit_graft + pos,
 131                           commit_graft_nr - pos - 1);
 132        commit_graft[pos] = graft;
 133        return 0;
 134}
 135
 136struct commit_graft *read_graft_line(struct strbuf *line)
 137{
 138        /* The format is just "Commit Parent1 Parent2 ...\n" */
 139        int i, phase;
 140        const char *tail = NULL;
 141        struct commit_graft *graft = NULL;
 142        struct object_id dummy_oid, *oid;
 143
 144        strbuf_rtrim(line);
 145        if (!line->len || line->buf[0] == '#')
 146                return NULL;
 147        /*
 148         * phase 0 verifies line, counts hashes in line and allocates graft
 149         * phase 1 fills graft
 150         */
 151        for (phase = 0; phase < 2; phase++) {
 152                oid = graft ? &graft->oid : &dummy_oid;
 153                if (parse_oid_hex(line->buf, oid, &tail))
 154                        goto bad_graft_data;
 155                for (i = 0; *tail != '\0'; i++) {
 156                        oid = graft ? &graft->parent[i] : &dummy_oid;
 157                        if (!isspace(*tail++) || parse_oid_hex(tail, oid, &tail))
 158                                goto bad_graft_data;
 159                }
 160                if (!graft) {
 161                        graft = xmalloc(st_add(sizeof(*graft),
 162                                               st_mult(sizeof(struct object_id), i)));
 163                        graft->nr_parent = i;
 164                }
 165        }
 166        return graft;
 167
 168bad_graft_data:
 169        error("bad graft data: %s", line->buf);
 170        assert(!graft);
 171        return NULL;
 172}
 173
 174static int read_graft_file(const char *graft_file)
 175{
 176        FILE *fp = fopen_or_warn(graft_file, "r");
 177        struct strbuf buf = STRBUF_INIT;
 178        if (!fp)
 179                return -1;
 180        while (!strbuf_getwholeline(&buf, fp, '\n')) {
 181                /* The format is just "Commit Parent1 Parent2 ...\n" */
 182                struct commit_graft *graft = read_graft_line(&buf);
 183                if (!graft)
 184                        continue;
 185                if (register_commit_graft(graft, 1))
 186                        error("duplicate graft data: %s", buf.buf);
 187        }
 188        fclose(fp);
 189        strbuf_release(&buf);
 190        return 0;
 191}
 192
 193static void prepare_commit_graft(void)
 194{
 195        static int commit_graft_prepared;
 196        char *graft_file;
 197
 198        if (commit_graft_prepared)
 199                return;
 200        graft_file = get_graft_file();
 201        read_graft_file(graft_file);
 202        /* make sure shallows are read */
 203        is_repository_shallow();
 204        commit_graft_prepared = 1;
 205}
 206
 207struct commit_graft *lookup_commit_graft(const struct object_id *oid)
 208{
 209        int pos;
 210        prepare_commit_graft();
 211        pos = commit_graft_pos(oid->hash);
 212        if (pos < 0)
 213                return NULL;
 214        return commit_graft[pos];
 215}
 216
 217int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
 218{
 219        int i, ret;
 220        for (i = ret = 0; i < commit_graft_nr && !ret; i++)
 221                ret = fn(commit_graft[i], cb_data);
 222        return ret;
 223}
 224
 225int unregister_shallow(const struct object_id *oid)
 226{
 227        int pos = commit_graft_pos(oid->hash);
 228        if (pos < 0)
 229                return -1;
 230        if (pos + 1 < commit_graft_nr)
 231                MOVE_ARRAY(commit_graft + pos, commit_graft + pos + 1,
 232                           commit_graft_nr - pos - 1);
 233        commit_graft_nr--;
 234        return 0;
 235}
 236
 237struct commit_buffer {
 238        void *buffer;
 239        unsigned long size;
 240};
 241define_commit_slab(buffer_slab, struct commit_buffer);
 242static struct buffer_slab buffer_slab = COMMIT_SLAB_INIT(1, buffer_slab);
 243
 244void set_commit_buffer(struct commit *commit, void *buffer, unsigned long size)
 245{
 246        struct commit_buffer *v = buffer_slab_at(&buffer_slab, commit);
 247        v->buffer = buffer;
 248        v->size = size;
 249}
 250
 251const void *get_cached_commit_buffer(const struct commit *commit, unsigned long *sizep)
 252{
 253        struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
 254        if (!v) {
 255                if (sizep)
 256                        *sizep = 0;
 257                return NULL;
 258        }
 259        if (sizep)
 260                *sizep = v->size;
 261        return v->buffer;
 262}
 263
 264const void *get_commit_buffer(const struct commit *commit, unsigned long *sizep)
 265{
 266        const void *ret = get_cached_commit_buffer(commit, sizep);
 267        if (!ret) {
 268                enum object_type type;
 269                unsigned long size;
 270                ret = read_sha1_file(commit->object.oid.hash, &type, &size);
 271                if (!ret)
 272                        die("cannot read commit object %s",
 273                            oid_to_hex(&commit->object.oid));
 274                if (type != OBJ_COMMIT)
 275                        die("expected commit for %s, got %s",
 276                            oid_to_hex(&commit->object.oid), type_name(type));
 277                if (sizep)
 278                        *sizep = size;
 279        }
 280        return ret;
 281}
 282
 283void unuse_commit_buffer(const struct commit *commit, const void *buffer)
 284{
 285        struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
 286        if (!(v && v->buffer == buffer))
 287                free((void *)buffer);
 288}
 289
 290void free_commit_buffer(struct commit *commit)
 291{
 292        struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
 293        if (v) {
 294                FREE_AND_NULL(v->buffer);
 295                v->size = 0;
 296        }
 297}
 298
 299struct tree *get_commit_tree(const struct commit *commit)
 300{
 301        if (commit->maybe_tree || !commit->object.parsed)
 302                return commit->maybe_tree;
 303
 304        if (commit->graph_pos == COMMIT_NOT_FROM_GRAPH)
 305                BUG("commit has NULL tree, but was not loaded from commit-graph");
 306
 307        return get_commit_tree_in_graph(commit);
 308}
 309
 310struct object_id *get_commit_tree_oid(const struct commit *commit)
 311{
 312        return &get_commit_tree(commit)->object.oid;
 313}
 314
 315const void *detach_commit_buffer(struct commit *commit, unsigned long *sizep)
 316{
 317        struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
 318        void *ret;
 319
 320        if (!v) {
 321                if (sizep)
 322                        *sizep = 0;
 323                return NULL;
 324        }
 325        ret = v->buffer;
 326        if (sizep)
 327                *sizep = v->size;
 328
 329        v->buffer = NULL;
 330        v->size = 0;
 331        return ret;
 332}
 333
 334int parse_commit_buffer(struct commit *item, const void *buffer, unsigned long size, int check_graph)
 335{
 336        const char *tail = buffer;
 337        const char *bufptr = buffer;
 338        struct object_id parent;
 339        struct commit_list **pptr;
 340        struct commit_graft *graft;
 341        const int tree_entry_len = GIT_SHA1_HEXSZ + 5;
 342        const int parent_entry_len = GIT_SHA1_HEXSZ + 7;
 343
 344        if (item->object.parsed)
 345                return 0;
 346        item->object.parsed = 1;
 347        tail += size;
 348        if (tail <= bufptr + tree_entry_len + 1 || memcmp(bufptr, "tree ", 5) ||
 349                        bufptr[tree_entry_len] != '\n')
 350                return error("bogus commit object %s", oid_to_hex(&item->object.oid));
 351        if (get_sha1_hex(bufptr + 5, parent.hash) < 0)
 352                return error("bad tree pointer in commit %s",
 353                             oid_to_hex(&item->object.oid));
 354        item->maybe_tree = lookup_tree(&parent);
 355        bufptr += tree_entry_len + 1; /* "tree " + "hex sha1" + "\n" */
 356        pptr = &item->parents;
 357
 358        graft = lookup_commit_graft(&item->object.oid);
 359        while (bufptr + parent_entry_len < tail && !memcmp(bufptr, "parent ", 7)) {
 360                struct commit *new_parent;
 361
 362                if (tail <= bufptr + parent_entry_len + 1 ||
 363                    get_sha1_hex(bufptr + 7, parent.hash) ||
 364                    bufptr[parent_entry_len] != '\n')
 365                        return error("bad parents in commit %s", oid_to_hex(&item->object.oid));
 366                bufptr += parent_entry_len + 1;
 367                /*
 368                 * The clone is shallow if nr_parent < 0, and we must
 369                 * not traverse its real parents even when we unhide them.
 370                 */
 371                if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
 372                        continue;
 373                new_parent = lookup_commit(&parent);
 374                if (new_parent)
 375                        pptr = &commit_list_insert(new_parent, pptr)->next;
 376        }
 377        if (graft) {
 378                int i;
 379                struct commit *new_parent;
 380                for (i = 0; i < graft->nr_parent; i++) {
 381                        new_parent = lookup_commit(&graft->parent[i]);
 382                        if (!new_parent)
 383                                continue;
 384                        pptr = &commit_list_insert(new_parent, pptr)->next;
 385                }
 386        }
 387        item->date = parse_commit_date(bufptr, tail);
 388
 389        if (check_graph)
 390                load_commit_graph_info(item);
 391
 392        return 0;
 393}
 394
 395int parse_commit_gently(struct commit *item, int quiet_on_missing)
 396{
 397        enum object_type type;
 398        void *buffer;
 399        unsigned long size;
 400        int ret;
 401
 402        if (!item)
 403                return -1;
 404        if (item->object.parsed)
 405                return 0;
 406        if (parse_commit_in_graph(item))
 407                return 0;
 408        buffer = read_sha1_file(item->object.oid.hash, &type, &size);
 409        if (!buffer)
 410                return quiet_on_missing ? -1 :
 411                        error("Could not read %s",
 412                             oid_to_hex(&item->object.oid));
 413        if (type != OBJ_COMMIT) {
 414                free(buffer);
 415                return error("Object %s not a commit",
 416                             oid_to_hex(&item->object.oid));
 417        }
 418        ret = parse_commit_buffer(item, buffer, size, 0);
 419        if (save_commit_buffer && !ret) {
 420                set_commit_buffer(item, buffer, size);
 421                return 0;
 422        }
 423        free(buffer);
 424        return ret;
 425}
 426
 427void parse_commit_or_die(struct commit *item)
 428{
 429        if (parse_commit(item))
 430                die("unable to parse commit %s",
 431                    item ? oid_to_hex(&item->object.oid) : "(null)");
 432}
 433
 434int find_commit_subject(const char *commit_buffer, const char **subject)
 435{
 436        const char *eol;
 437        const char *p = commit_buffer;
 438
 439        while (*p && (*p != '\n' || p[1] != '\n'))
 440                p++;
 441        if (*p) {
 442                p = skip_blank_lines(p + 2);
 443                eol = strchrnul(p, '\n');
 444        } else
 445                eol = p;
 446
 447        *subject = p;
 448
 449        return eol - p;
 450}
 451
 452struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
 453{
 454        struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
 455        new_list->item = item;
 456        new_list->next = *list_p;
 457        *list_p = new_list;
 458        return new_list;
 459}
 460
 461unsigned commit_list_count(const struct commit_list *l)
 462{
 463        unsigned c = 0;
 464        for (; l; l = l->next )
 465                c++;
 466        return c;
 467}
 468
 469struct commit_list *copy_commit_list(struct commit_list *list)
 470{
 471        struct commit_list *head = NULL;
 472        struct commit_list **pp = &head;
 473        while (list) {
 474                pp = commit_list_append(list->item, pp);
 475                list = list->next;
 476        }
 477        return head;
 478}
 479
 480void free_commit_list(struct commit_list *list)
 481{
 482        while (list)
 483                pop_commit(&list);
 484}
 485
 486struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
 487{
 488        struct commit_list **pp = list;
 489        struct commit_list *p;
 490        while ((p = *pp) != NULL) {
 491                if (p->item->date < item->date) {
 492                        break;
 493                }
 494                pp = &p->next;
 495        }
 496        return commit_list_insert(item, pp);
 497}
 498
 499static int commit_list_compare_by_date(const void *a, const void *b)
 500{
 501        timestamp_t a_date = ((const struct commit_list *)a)->item->date;
 502        timestamp_t b_date = ((const struct commit_list *)b)->item->date;
 503        if (a_date < b_date)
 504                return 1;
 505        if (a_date > b_date)
 506                return -1;
 507        return 0;
 508}
 509
 510static void *commit_list_get_next(const void *a)
 511{
 512        return ((const struct commit_list *)a)->next;
 513}
 514
 515static void commit_list_set_next(void *a, void *next)
 516{
 517        ((struct commit_list *)a)->next = next;
 518}
 519
 520void commit_list_sort_by_date(struct commit_list **list)
 521{
 522        *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
 523                                commit_list_compare_by_date);
 524}
 525
 526struct commit *pop_most_recent_commit(struct commit_list **list,
 527                                      unsigned int mark)
 528{
 529        struct commit *ret = pop_commit(list);
 530        struct commit_list *parents = ret->parents;
 531
 532        while (parents) {
 533                struct commit *commit = parents->item;
 534                if (!parse_commit(commit) && !(commit->object.flags & mark)) {
 535                        commit->object.flags |= mark;
 536                        commit_list_insert_by_date(commit, list);
 537                }
 538                parents = parents->next;
 539        }
 540        return ret;
 541}
 542
 543static void clear_commit_marks_1(struct commit_list **plist,
 544                                 struct commit *commit, unsigned int mark)
 545{
 546        while (commit) {
 547                struct commit_list *parents;
 548
 549                if (!(mark & commit->object.flags))
 550                        return;
 551
 552                commit->object.flags &= ~mark;
 553
 554                parents = commit->parents;
 555                if (!parents)
 556                        return;
 557
 558                while ((parents = parents->next))
 559                        commit_list_insert(parents->item, plist);
 560
 561                commit = commit->parents->item;
 562        }
 563}
 564
 565void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
 566{
 567        struct commit_list *list = NULL;
 568
 569        while (nr--) {
 570                clear_commit_marks_1(&list, *commit, mark);
 571                commit++;
 572        }
 573        while (list)
 574                clear_commit_marks_1(&list, pop_commit(&list), mark);
 575}
 576
 577void clear_commit_marks(struct commit *commit, unsigned int mark)
 578{
 579        clear_commit_marks_many(1, &commit, mark);
 580}
 581
 582struct commit *pop_commit(struct commit_list **stack)
 583{
 584        struct commit_list *top = *stack;
 585        struct commit *item = top ? top->item : NULL;
 586
 587        if (top) {
 588                *stack = top->next;
 589                free(top);
 590        }
 591        return item;
 592}
 593
 594/*
 595 * Topological sort support
 596 */
 597
 598/* count number of children that have not been emitted */
 599define_commit_slab(indegree_slab, int);
 600
 601/* record author-date for each commit object */
 602define_commit_slab(author_date_slab, unsigned long);
 603
 604static void record_author_date(struct author_date_slab *author_date,
 605                               struct commit *commit)
 606{
 607        const char *buffer = get_commit_buffer(commit, NULL);
 608        struct ident_split ident;
 609        const char *ident_line;
 610        size_t ident_len;
 611        char *date_end;
 612        timestamp_t date;
 613
 614        ident_line = find_commit_header(buffer, "author", &ident_len);
 615        if (!ident_line)
 616                goto fail_exit; /* no author line */
 617        if (split_ident_line(&ident, ident_line, ident_len) ||
 618            !ident.date_begin || !ident.date_end)
 619                goto fail_exit; /* malformed "author" line */
 620
 621        date = parse_timestamp(ident.date_begin, &date_end, 10);
 622        if (date_end != ident.date_end)
 623                goto fail_exit; /* malformed date */
 624        *(author_date_slab_at(author_date, commit)) = date;
 625
 626fail_exit:
 627        unuse_commit_buffer(commit, buffer);
 628}
 629
 630static int compare_commits_by_author_date(const void *a_, const void *b_,
 631                                          void *cb_data)
 632{
 633        const struct commit *a = a_, *b = b_;
 634        struct author_date_slab *author_date = cb_data;
 635        timestamp_t a_date = *(author_date_slab_at(author_date, a));
 636        timestamp_t b_date = *(author_date_slab_at(author_date, b));
 637
 638        /* newer commits with larger date first */
 639        if (a_date < b_date)
 640                return 1;
 641        else if (a_date > b_date)
 642                return -1;
 643        return 0;
 644}
 645
 646int compare_commits_by_gen_then_commit_date(const void *a_, const void *b_, void *unused)
 647{
 648        const struct commit *a = a_, *b = b_;
 649
 650        /* newer commits first */
 651        if (a->generation < b->generation)
 652                return 1;
 653        else if (a->generation > b->generation)
 654                return -1;
 655
 656        /* use date as a heuristic when generations are equal */
 657        if (a->date < b->date)
 658                return 1;
 659        else if (a->date > b->date)
 660                return -1;
 661        return 0;
 662}
 663
 664int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
 665{
 666        const struct commit *a = a_, *b = b_;
 667        /* newer commits with larger date first */
 668        if (a->date < b->date)
 669                return 1;
 670        else if (a->date > b->date)
 671                return -1;
 672        return 0;
 673}
 674
 675/*
 676 * Performs an in-place topological sort on the list supplied.
 677 */
 678void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
 679{
 680        struct commit_list *next, *orig = *list;
 681        struct commit_list **pptr;
 682        struct indegree_slab indegree;
 683        struct prio_queue queue;
 684        struct commit *commit;
 685        struct author_date_slab author_date;
 686
 687        if (!orig)
 688                return;
 689        *list = NULL;
 690
 691        init_indegree_slab(&indegree);
 692        memset(&queue, '\0', sizeof(queue));
 693
 694        switch (sort_order) {
 695        default: /* REV_SORT_IN_GRAPH_ORDER */
 696                queue.compare = NULL;
 697                break;
 698        case REV_SORT_BY_COMMIT_DATE:
 699                queue.compare = compare_commits_by_commit_date;
 700                break;
 701        case REV_SORT_BY_AUTHOR_DATE:
 702                init_author_date_slab(&author_date);
 703                queue.compare = compare_commits_by_author_date;
 704                queue.cb_data = &author_date;
 705                break;
 706        }
 707
 708        /* Mark them and clear the indegree */
 709        for (next = orig; next; next = next->next) {
 710                struct commit *commit = next->item;
 711                *(indegree_slab_at(&indegree, commit)) = 1;
 712                /* also record the author dates, if needed */
 713                if (sort_order == REV_SORT_BY_AUTHOR_DATE)
 714                        record_author_date(&author_date, commit);
 715        }
 716
 717        /* update the indegree */
 718        for (next = orig; next; next = next->next) {
 719                struct commit_list *parents = next->item->parents;
 720                while (parents) {
 721                        struct commit *parent = parents->item;
 722                        int *pi = indegree_slab_at(&indegree, parent);
 723
 724                        if (*pi)
 725                                (*pi)++;
 726                        parents = parents->next;
 727                }
 728        }
 729
 730        /*
 731         * find the tips
 732         *
 733         * tips are nodes not reachable from any other node in the list
 734         *
 735         * the tips serve as a starting set for the work queue.
 736         */
 737        for (next = orig; next; next = next->next) {
 738                struct commit *commit = next->item;
 739
 740                if (*(indegree_slab_at(&indegree, commit)) == 1)
 741                        prio_queue_put(&queue, commit);
 742        }
 743
 744        /*
 745         * This is unfortunate; the initial tips need to be shown
 746         * in the order given from the revision traversal machinery.
 747         */
 748        if (sort_order == REV_SORT_IN_GRAPH_ORDER)
 749                prio_queue_reverse(&queue);
 750
 751        /* We no longer need the commit list */
 752        free_commit_list(orig);
 753
 754        pptr = list;
 755        *list = NULL;
 756        while ((commit = prio_queue_get(&queue)) != NULL) {
 757                struct commit_list *parents;
 758
 759                for (parents = commit->parents; parents ; parents = parents->next) {
 760                        struct commit *parent = parents->item;
 761                        int *pi = indegree_slab_at(&indegree, parent);
 762
 763                        if (!*pi)
 764                                continue;
 765
 766                        /*
 767                         * parents are only enqueued for emission
 768                         * when all their children have been emitted thereby
 769                         * guaranteeing topological order.
 770                         */
 771                        if (--(*pi) == 1)
 772                                prio_queue_put(&queue, parent);
 773                }
 774                /*
 775                 * all children of commit have already been
 776                 * emitted. we can emit it now.
 777                 */
 778                *(indegree_slab_at(&indegree, commit)) = 0;
 779
 780                pptr = &commit_list_insert(commit, pptr)->next;
 781        }
 782
 783        clear_indegree_slab(&indegree);
 784        clear_prio_queue(&queue);
 785        if (sort_order == REV_SORT_BY_AUTHOR_DATE)
 786                clear_author_date_slab(&author_date);
 787}
 788
 789/* merge-base stuff */
 790
 791/* Remember to update object flag allocation in object.h */
 792#define PARENT1         (1u<<16)
 793#define PARENT2         (1u<<17)
 794#define STALE           (1u<<18)
 795#define RESULT          (1u<<19)
 796
 797static const unsigned all_flags = (PARENT1 | PARENT2 | STALE | RESULT);
 798
 799static int queue_has_nonstale(struct prio_queue *queue)
 800{
 801        int i;
 802        for (i = 0; i < queue->nr; i++) {
 803                struct commit *commit = queue->array[i].data;
 804                if (!(commit->object.flags & STALE))
 805                        return 1;
 806        }
 807        return 0;
 808}
 809
 810/* all input commits in one and twos[] must have been parsed! */
 811static struct commit_list *paint_down_to_common(struct commit *one, int n,
 812                                                struct commit **twos,
 813                                                int min_generation)
 814{
 815        struct prio_queue queue = { compare_commits_by_gen_then_commit_date };
 816        struct commit_list *result = NULL;
 817        int i;
 818        uint32_t last_gen = GENERATION_NUMBER_INFINITY;
 819
 820        one->object.flags |= PARENT1;
 821        if (!n) {
 822                commit_list_append(one, &result);
 823                return result;
 824        }
 825        prio_queue_put(&queue, one);
 826
 827        for (i = 0; i < n; i++) {
 828                twos[i]->object.flags |= PARENT2;
 829                prio_queue_put(&queue, twos[i]);
 830        }
 831
 832        while (queue_has_nonstale(&queue)) {
 833                struct commit *commit = prio_queue_get(&queue);
 834                struct commit_list *parents;
 835                int flags;
 836
 837                if (commit->generation > last_gen)
 838                        BUG("bad generation skip %8x > %8x at %s",
 839                            commit->generation, last_gen,
 840                            oid_to_hex(&commit->object.oid));
 841                last_gen = commit->generation;
 842
 843                if (commit->generation < min_generation)
 844                        break;
 845
 846                flags = commit->object.flags & (PARENT1 | PARENT2 | STALE);
 847                if (flags == (PARENT1 | PARENT2)) {
 848                        if (!(commit->object.flags & RESULT)) {
 849                                commit->object.flags |= RESULT;
 850                                commit_list_insert_by_date(commit, &result);
 851                        }
 852                        /* Mark parents of a found merge stale */
 853                        flags |= STALE;
 854                }
 855                parents = commit->parents;
 856                while (parents) {
 857                        struct commit *p = parents->item;
 858                        parents = parents->next;
 859                        if ((p->object.flags & flags) == flags)
 860                                continue;
 861                        if (parse_commit(p))
 862                                return NULL;
 863                        p->object.flags |= flags;
 864                        prio_queue_put(&queue, p);
 865                }
 866        }
 867
 868        clear_prio_queue(&queue);
 869        return result;
 870}
 871
 872static struct commit_list *merge_bases_many(struct commit *one, int n, struct commit **twos)
 873{
 874        struct commit_list *list = NULL;
 875        struct commit_list *result = NULL;
 876        int i;
 877
 878        for (i = 0; i < n; i++) {
 879                if (one == twos[i])
 880                        /*
 881                         * We do not mark this even with RESULT so we do not
 882                         * have to clean it up.
 883                         */
 884                        return commit_list_insert(one, &result);
 885        }
 886
 887        if (parse_commit(one))
 888                return NULL;
 889        for (i = 0; i < n; i++) {
 890                if (parse_commit(twos[i]))
 891                        return NULL;
 892        }
 893
 894        list = paint_down_to_common(one, n, twos, 0);
 895
 896        while (list) {
 897                struct commit *commit = pop_commit(&list);
 898                if (!(commit->object.flags & STALE))
 899                        commit_list_insert_by_date(commit, &result);
 900        }
 901        return result;
 902}
 903
 904struct commit_list *get_octopus_merge_bases(struct commit_list *in)
 905{
 906        struct commit_list *i, *j, *k, *ret = NULL;
 907
 908        if (!in)
 909                return ret;
 910
 911        commit_list_insert(in->item, &ret);
 912
 913        for (i = in->next; i; i = i->next) {
 914                struct commit_list *new_commits = NULL, *end = NULL;
 915
 916                for (j = ret; j; j = j->next) {
 917                        struct commit_list *bases;
 918                        bases = get_merge_bases(i->item, j->item);
 919                        if (!new_commits)
 920                                new_commits = bases;
 921                        else
 922                                end->next = bases;
 923                        for (k = bases; k; k = k->next)
 924                                end = k;
 925                }
 926                ret = new_commits;
 927        }
 928        return ret;
 929}
 930
 931static int remove_redundant(struct commit **array, int cnt)
 932{
 933        /*
 934         * Some commit in the array may be an ancestor of
 935         * another commit.  Move such commit to the end of
 936         * the array, and return the number of commits that
 937         * are independent from each other.
 938         */
 939        struct commit **work;
 940        unsigned char *redundant;
 941        int *filled_index;
 942        int i, j, filled;
 943
 944        work = xcalloc(cnt, sizeof(*work));
 945        redundant = xcalloc(cnt, 1);
 946        ALLOC_ARRAY(filled_index, cnt - 1);
 947
 948        for (i = 0; i < cnt; i++)
 949                parse_commit(array[i]);
 950        for (i = 0; i < cnt; i++) {
 951                struct commit_list *common;
 952                uint32_t min_generation = array[i]->generation;
 953
 954                if (redundant[i])
 955                        continue;
 956                for (j = filled = 0; j < cnt; j++) {
 957                        if (i == j || redundant[j])
 958                                continue;
 959                        filled_index[filled] = j;
 960                        work[filled++] = array[j];
 961
 962                        if (array[j]->generation < min_generation)
 963                                min_generation = array[j]->generation;
 964                }
 965                common = paint_down_to_common(array[i], filled, work,
 966                                              min_generation);
 967                if (array[i]->object.flags & PARENT2)
 968                        redundant[i] = 1;
 969                for (j = 0; j < filled; j++)
 970                        if (work[j]->object.flags & PARENT1)
 971                                redundant[filled_index[j]] = 1;
 972                clear_commit_marks(array[i], all_flags);
 973                clear_commit_marks_many(filled, work, all_flags);
 974                free_commit_list(common);
 975        }
 976
 977        /* Now collect the result */
 978        COPY_ARRAY(work, array, cnt);
 979        for (i = filled = 0; i < cnt; i++)
 980                if (!redundant[i])
 981                        array[filled++] = work[i];
 982        for (j = filled, i = 0; i < cnt; i++)
 983                if (redundant[i])
 984                        array[j++] = work[i];
 985        free(work);
 986        free(redundant);
 987        free(filled_index);
 988        return filled;
 989}
 990
 991static struct commit_list *get_merge_bases_many_0(struct commit *one,
 992                                                  int n,
 993                                                  struct commit **twos,
 994                                                  int cleanup)
 995{
 996        struct commit_list *list;
 997        struct commit **rslt;
 998        struct commit_list *result;
 999        int cnt, i;
1000
1001        result = merge_bases_many(one, n, twos);
1002        for (i = 0; i < n; i++) {
1003                if (one == twos[i])
1004                        return result;
1005        }
1006        if (!result || !result->next) {
1007                if (cleanup) {
1008                        clear_commit_marks(one, all_flags);
1009                        clear_commit_marks_many(n, twos, all_flags);
1010                }
1011                return result;
1012        }
1013
1014        /* There are more than one */
1015        cnt = commit_list_count(result);
1016        rslt = xcalloc(cnt, sizeof(*rslt));
1017        for (list = result, i = 0; list; list = list->next)
1018                rslt[i++] = list->item;
1019        free_commit_list(result);
1020
1021        clear_commit_marks(one, all_flags);
1022        clear_commit_marks_many(n, twos, all_flags);
1023
1024        cnt = remove_redundant(rslt, cnt);
1025        result = NULL;
1026        for (i = 0; i < cnt; i++)
1027                commit_list_insert_by_date(rslt[i], &result);
1028        free(rslt);
1029        return result;
1030}
1031
1032struct commit_list *get_merge_bases_many(struct commit *one,
1033                                         int n,
1034                                         struct commit **twos)
1035{
1036        return get_merge_bases_many_0(one, n, twos, 1);
1037}
1038
1039struct commit_list *get_merge_bases_many_dirty(struct commit *one,
1040                                               int n,
1041                                               struct commit **twos)
1042{
1043        return get_merge_bases_many_0(one, n, twos, 0);
1044}
1045
1046struct commit_list *get_merge_bases(struct commit *one, struct commit *two)
1047{
1048        return get_merge_bases_many_0(one, 1, &two, 1);
1049}
1050
1051/*
1052 * Is "commit" a descendant of one of the elements on the "with_commit" list?
1053 */
1054int is_descendant_of(struct commit *commit, struct commit_list *with_commit)
1055{
1056        if (!with_commit)
1057                return 1;
1058        while (with_commit) {
1059                struct commit *other;
1060
1061                other = with_commit->item;
1062                with_commit = with_commit->next;
1063                if (in_merge_bases(other, commit))
1064                        return 1;
1065        }
1066        return 0;
1067}
1068
1069/*
1070 * Is "commit" an ancestor of one of the "references"?
1071 */
1072int in_merge_bases_many(struct commit *commit, int nr_reference, struct commit **reference)
1073{
1074        struct commit_list *bases;
1075        int ret = 0, i;
1076        uint32_t min_generation = GENERATION_NUMBER_INFINITY;
1077
1078        if (parse_commit(commit))
1079                return ret;
1080        for (i = 0; i < nr_reference; i++) {
1081                if (parse_commit(reference[i]))
1082                        return ret;
1083                if (reference[i]->generation < min_generation)
1084                        min_generation = reference[i]->generation;
1085        }
1086
1087        if (commit->generation > min_generation)
1088                return ret;
1089
1090        bases = paint_down_to_common(commit, nr_reference, reference, commit->generation);
1091        if (commit->object.flags & PARENT2)
1092                ret = 1;
1093        clear_commit_marks(commit, all_flags);
1094        clear_commit_marks_many(nr_reference, reference, all_flags);
1095        free_commit_list(bases);
1096        return ret;
1097}
1098
1099/*
1100 * Is "commit" an ancestor of (i.e. reachable from) the "reference"?
1101 */
1102int in_merge_bases(struct commit *commit, struct commit *reference)
1103{
1104        return in_merge_bases_many(commit, 1, &reference);
1105}
1106
1107struct commit_list *reduce_heads(struct commit_list *heads)
1108{
1109        struct commit_list *p;
1110        struct commit_list *result = NULL, **tail = &result;
1111        struct commit **array;
1112        int num_head, i;
1113
1114        if (!heads)
1115                return NULL;
1116
1117        /* Uniquify */
1118        for (p = heads; p; p = p->next)
1119                p->item->object.flags &= ~STALE;
1120        for (p = heads, num_head = 0; p; p = p->next) {
1121                if (p->item->object.flags & STALE)
1122                        continue;
1123                p->item->object.flags |= STALE;
1124                num_head++;
1125        }
1126        array = xcalloc(num_head, sizeof(*array));
1127        for (p = heads, i = 0; p; p = p->next) {
1128                if (p->item->object.flags & STALE) {
1129                        array[i++] = p->item;
1130                        p->item->object.flags &= ~STALE;
1131                }
1132        }
1133        num_head = remove_redundant(array, num_head);
1134        for (i = 0; i < num_head; i++)
1135                tail = &commit_list_insert(array[i], tail)->next;
1136        free(array);
1137        return result;
1138}
1139
1140void reduce_heads_replace(struct commit_list **heads)
1141{
1142        struct commit_list *result = reduce_heads(*heads);
1143        free_commit_list(*heads);
1144        *heads = result;
1145}
1146
1147static const char gpg_sig_header[] = "gpgsig";
1148static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
1149
1150static int do_sign_commit(struct strbuf *buf, const char *keyid)
1151{
1152        struct strbuf sig = STRBUF_INIT;
1153        int inspos, copypos;
1154        const char *eoh;
1155
1156        /* find the end of the header */
1157        eoh = strstr(buf->buf, "\n\n");
1158        if (!eoh)
1159                inspos = buf->len;
1160        else
1161                inspos = eoh - buf->buf + 1;
1162
1163        if (!keyid || !*keyid)
1164                keyid = get_signing_key();
1165        if (sign_buffer(buf, &sig, keyid)) {
1166                strbuf_release(&sig);
1167                return -1;
1168        }
1169
1170        for (copypos = 0; sig.buf[copypos]; ) {
1171                const char *bol = sig.buf + copypos;
1172                const char *eol = strchrnul(bol, '\n');
1173                int len = (eol - bol) + !!*eol;
1174
1175                if (!copypos) {
1176                        strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
1177                        inspos += gpg_sig_header_len;
1178                }
1179                strbuf_insert(buf, inspos++, " ", 1);
1180                strbuf_insert(buf, inspos, bol, len);
1181                inspos += len;
1182                copypos += len;
1183        }
1184        strbuf_release(&sig);
1185        return 0;
1186}
1187
1188int parse_signed_commit(const struct commit *commit,
1189                        struct strbuf *payload, struct strbuf *signature)
1190{
1191
1192        unsigned long size;
1193        const char *buffer = get_commit_buffer(commit, &size);
1194        int in_signature, saw_signature = -1;
1195        const char *line, *tail;
1196
1197        line = buffer;
1198        tail = buffer + size;
1199        in_signature = 0;
1200        saw_signature = 0;
1201        while (line < tail) {
1202                const char *sig = NULL;
1203                const char *next = memchr(line, '\n', tail - line);
1204
1205                next = next ? next + 1 : tail;
1206                if (in_signature && line[0] == ' ')
1207                        sig = line + 1;
1208                else if (starts_with(line, gpg_sig_header) &&
1209                         line[gpg_sig_header_len] == ' ')
1210                        sig = line + gpg_sig_header_len + 1;
1211                if (sig) {
1212                        strbuf_add(signature, sig, next - sig);
1213                        saw_signature = 1;
1214                        in_signature = 1;
1215                } else {
1216                        if (*line == '\n')
1217                                /* dump the whole remainder of the buffer */
1218                                next = tail;
1219                        strbuf_add(payload, line, next - line);
1220                        in_signature = 0;
1221                }
1222                line = next;
1223        }
1224        unuse_commit_buffer(commit, buffer);
1225        return saw_signature;
1226}
1227
1228int remove_signature(struct strbuf *buf)
1229{
1230        const char *line = buf->buf;
1231        const char *tail = buf->buf + buf->len;
1232        int in_signature = 0;
1233        const char *sig_start = NULL;
1234        const char *sig_end = NULL;
1235
1236        while (line < tail) {
1237                const char *next = memchr(line, '\n', tail - line);
1238                next = next ? next + 1 : tail;
1239
1240                if (in_signature && line[0] == ' ')
1241                        sig_end = next;
1242                else if (starts_with(line, gpg_sig_header) &&
1243                         line[gpg_sig_header_len] == ' ') {
1244                        sig_start = line;
1245                        sig_end = next;
1246                        in_signature = 1;
1247                } else {
1248                        if (*line == '\n')
1249                                /* dump the whole remainder of the buffer */
1250                                next = tail;
1251                        in_signature = 0;
1252                }
1253                line = next;
1254        }
1255
1256        if (sig_start)
1257                strbuf_remove(buf, sig_start - buf->buf, sig_end - sig_start);
1258
1259        return sig_start != NULL;
1260}
1261
1262static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1263{
1264        struct merge_remote_desc *desc;
1265        struct commit_extra_header *mergetag;
1266        char *buf;
1267        unsigned long size, len;
1268        enum object_type type;
1269
1270        desc = merge_remote_util(parent);
1271        if (!desc || !desc->obj)
1272                return;
1273        buf = read_sha1_file(desc->obj->oid.hash, &type, &size);
1274        if (!buf || type != OBJ_TAG)
1275                goto free_return;
1276        len = parse_signature(buf, size);
1277        if (size == len)
1278                goto free_return;
1279        /*
1280         * We could verify this signature and either omit the tag when
1281         * it does not validate, but the integrator may not have the
1282         * public key of the signer of the tag he is merging, while a
1283         * later auditor may have it while auditing, so let's not run
1284         * verify-signed-buffer here for now...
1285         *
1286         * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1287         *      warn("warning: signed tag unverified.");
1288         */
1289        mergetag = xcalloc(1, sizeof(*mergetag));
1290        mergetag->key = xstrdup("mergetag");
1291        mergetag->value = buf;
1292        mergetag->len = size;
1293
1294        **tail = mergetag;
1295        *tail = &mergetag->next;
1296        return;
1297
1298free_return:
1299        free(buf);
1300}
1301
1302int check_commit_signature(const struct commit *commit, struct signature_check *sigc)
1303{
1304        struct strbuf payload = STRBUF_INIT;
1305        struct strbuf signature = STRBUF_INIT;
1306        int ret = 1;
1307
1308        sigc->result = 'N';
1309
1310        if (parse_signed_commit(commit, &payload, &signature) <= 0)
1311                goto out;
1312        ret = check_signature(payload.buf, payload.len, signature.buf,
1313                signature.len, sigc);
1314
1315 out:
1316        strbuf_release(&payload);
1317        strbuf_release(&signature);
1318
1319        return ret;
1320}
1321
1322
1323
1324void append_merge_tag_headers(struct commit_list *parents,
1325                              struct commit_extra_header ***tail)
1326{
1327        while (parents) {
1328                struct commit *parent = parents->item;
1329                handle_signed_tag(parent, tail);
1330                parents = parents->next;
1331        }
1332}
1333
1334static void add_extra_header(struct strbuf *buffer,
1335                             struct commit_extra_header *extra)
1336{
1337        strbuf_addstr(buffer, extra->key);
1338        if (extra->len)
1339                strbuf_add_lines(buffer, " ", extra->value, extra->len);
1340        else
1341                strbuf_addch(buffer, '\n');
1342}
1343
1344struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1345                                                      const char **exclude)
1346{
1347        struct commit_extra_header *extra = NULL;
1348        unsigned long size;
1349        const char *buffer = get_commit_buffer(commit, &size);
1350        extra = read_commit_extra_header_lines(buffer, size, exclude);
1351        unuse_commit_buffer(commit, buffer);
1352        return extra;
1353}
1354
1355void for_each_mergetag(each_mergetag_fn fn, struct commit *commit, void *data)
1356{
1357        struct commit_extra_header *extra, *to_free;
1358
1359        to_free = read_commit_extra_headers(commit, NULL);
1360        for (extra = to_free; extra; extra = extra->next) {
1361                if (strcmp(extra->key, "mergetag"))
1362                        continue; /* not a merge tag */
1363                fn(commit, extra, data);
1364        }
1365        free_commit_extra_headers(to_free);
1366}
1367
1368static inline int standard_header_field(const char *field, size_t len)
1369{
1370        return ((len == 4 && !memcmp(field, "tree", 4)) ||
1371                (len == 6 && !memcmp(field, "parent", 6)) ||
1372                (len == 6 && !memcmp(field, "author", 6)) ||
1373                (len == 9 && !memcmp(field, "committer", 9)) ||
1374                (len == 8 && !memcmp(field, "encoding", 8)));
1375}
1376
1377static int excluded_header_field(const char *field, size_t len, const char **exclude)
1378{
1379        if (!exclude)
1380                return 0;
1381
1382        while (*exclude) {
1383                size_t xlen = strlen(*exclude);
1384                if (len == xlen && !memcmp(field, *exclude, xlen))
1385                        return 1;
1386                exclude++;
1387        }
1388        return 0;
1389}
1390
1391static struct commit_extra_header *read_commit_extra_header_lines(
1392        const char *buffer, size_t size,
1393        const char **exclude)
1394{
1395        struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1396        const char *line, *next, *eof, *eob;
1397        struct strbuf buf = STRBUF_INIT;
1398
1399        for (line = buffer, eob = line + size;
1400             line < eob && *line != '\n';
1401             line = next) {
1402                next = memchr(line, '\n', eob - line);
1403                next = next ? next + 1 : eob;
1404                if (*line == ' ') {
1405                        /* continuation */
1406                        if (it)
1407                                strbuf_add(&buf, line + 1, next - (line + 1));
1408                        continue;
1409                }
1410                if (it)
1411                        it->value = strbuf_detach(&buf, &it->len);
1412                strbuf_reset(&buf);
1413                it = NULL;
1414
1415                eof = memchr(line, ' ', next - line);
1416                if (!eof)
1417                        eof = next;
1418                else if (standard_header_field(line, eof - line) ||
1419                         excluded_header_field(line, eof - line, exclude))
1420                        continue;
1421
1422                it = xcalloc(1, sizeof(*it));
1423                it->key = xmemdupz(line, eof-line);
1424                *tail = it;
1425                tail = &it->next;
1426                if (eof + 1 < next)
1427                        strbuf_add(&buf, eof + 1, next - (eof + 1));
1428        }
1429        if (it)
1430                it->value = strbuf_detach(&buf, &it->len);
1431        return extra;
1432}
1433
1434void free_commit_extra_headers(struct commit_extra_header *extra)
1435{
1436        while (extra) {
1437                struct commit_extra_header *next = extra->next;
1438                free(extra->key);
1439                free(extra->value);
1440                free(extra);
1441                extra = next;
1442        }
1443}
1444
1445int commit_tree(const char *msg, size_t msg_len, const struct object_id *tree,
1446                struct commit_list *parents, struct object_id *ret,
1447                const char *author, const char *sign_commit)
1448{
1449        struct commit_extra_header *extra = NULL, **tail = &extra;
1450        int result;
1451
1452        append_merge_tag_headers(parents, &tail);
1453        result = commit_tree_extended(msg, msg_len, tree, parents, ret,
1454                                      author, sign_commit, extra);
1455        free_commit_extra_headers(extra);
1456        return result;
1457}
1458
1459static int find_invalid_utf8(const char *buf, int len)
1460{
1461        int offset = 0;
1462        static const unsigned int max_codepoint[] = {
1463                0x7f, 0x7ff, 0xffff, 0x10ffff
1464        };
1465
1466        while (len) {
1467                unsigned char c = *buf++;
1468                int bytes, bad_offset;
1469                unsigned int codepoint;
1470                unsigned int min_val, max_val;
1471
1472                len--;
1473                offset++;
1474
1475                /* Simple US-ASCII? No worries. */
1476                if (c < 0x80)
1477                        continue;
1478
1479                bad_offset = offset-1;
1480
1481                /*
1482                 * Count how many more high bits set: that's how
1483                 * many more bytes this sequence should have.
1484                 */
1485                bytes = 0;
1486                while (c & 0x40) {
1487                        c <<= 1;
1488                        bytes++;
1489                }
1490
1491                /*
1492                 * Must be between 1 and 3 more bytes.  Longer sequences result in
1493                 * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1494                 */
1495                if (bytes < 1 || 3 < bytes)
1496                        return bad_offset;
1497
1498                /* Do we *have* that many bytes? */
1499                if (len < bytes)
1500                        return bad_offset;
1501
1502                /*
1503                 * Place the encoded bits at the bottom of the value and compute the
1504                 * valid range.
1505                 */
1506                codepoint = (c & 0x7f) >> bytes;
1507                min_val = max_codepoint[bytes-1] + 1;
1508                max_val = max_codepoint[bytes];
1509
1510                offset += bytes;
1511                len -= bytes;
1512
1513                /* And verify that they are good continuation bytes */
1514                do {
1515                        codepoint <<= 6;
1516                        codepoint |= *buf & 0x3f;
1517                        if ((*buf++ & 0xc0) != 0x80)
1518                                return bad_offset;
1519                } while (--bytes);
1520
1521                /* Reject codepoints that are out of range for the sequence length. */
1522                if (codepoint < min_val || codepoint > max_val)
1523                        return bad_offset;
1524                /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1525                if ((codepoint & 0x1ff800) == 0xd800)
1526                        return bad_offset;
1527                /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1528                if ((codepoint & 0xfffe) == 0xfffe)
1529                        return bad_offset;
1530                /* So are anything in the range U+FDD0..U+FDEF. */
1531                if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1532                        return bad_offset;
1533        }
1534        return -1;
1535}
1536
1537/*
1538 * This verifies that the buffer is in proper utf8 format.
1539 *
1540 * If it isn't, it assumes any non-utf8 characters are Latin1,
1541 * and does the conversion.
1542 */
1543static int verify_utf8(struct strbuf *buf)
1544{
1545        int ok = 1;
1546        long pos = 0;
1547
1548        for (;;) {
1549                int bad;
1550                unsigned char c;
1551                unsigned char replace[2];
1552
1553                bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1554                if (bad < 0)
1555                        return ok;
1556                pos += bad;
1557                ok = 0;
1558                c = buf->buf[pos];
1559                strbuf_remove(buf, pos, 1);
1560
1561                /* We know 'c' must be in the range 128-255 */
1562                replace[0] = 0xc0 + (c >> 6);
1563                replace[1] = 0x80 + (c & 0x3f);
1564                strbuf_insert(buf, pos, replace, 2);
1565                pos += 2;
1566        }
1567}
1568
1569static const char commit_utf8_warn[] =
1570N_("Warning: commit message did not conform to UTF-8.\n"
1571   "You may want to amend it after fixing the message, or set the config\n"
1572   "variable i18n.commitencoding to the encoding your project uses.\n");
1573
1574int commit_tree_extended(const char *msg, size_t msg_len,
1575                         const struct object_id *tree,
1576                         struct commit_list *parents, struct object_id *ret,
1577                         const char *author, const char *sign_commit,
1578                         struct commit_extra_header *extra)
1579{
1580        int result;
1581        int encoding_is_utf8;
1582        struct strbuf buffer;
1583
1584        assert_sha1_type(tree->hash, OBJ_TREE);
1585
1586        if (memchr(msg, '\0', msg_len))
1587                return error("a NUL byte in commit log message not allowed.");
1588
1589        /* Not having i18n.commitencoding is the same as having utf-8 */
1590        encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1591
1592        strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1593        strbuf_addf(&buffer, "tree %s\n", oid_to_hex(tree));
1594
1595        /*
1596         * NOTE! This ordering means that the same exact tree merged with a
1597         * different order of parents will be a _different_ changeset even
1598         * if everything else stays the same.
1599         */
1600        while (parents) {
1601                struct commit *parent = pop_commit(&parents);
1602                strbuf_addf(&buffer, "parent %s\n",
1603                            oid_to_hex(&parent->object.oid));
1604        }
1605
1606        /* Person/date information */
1607        if (!author)
1608                author = git_author_info(IDENT_STRICT);
1609        strbuf_addf(&buffer, "author %s\n", author);
1610        strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1611        if (!encoding_is_utf8)
1612                strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1613
1614        while (extra) {
1615                add_extra_header(&buffer, extra);
1616                extra = extra->next;
1617        }
1618        strbuf_addch(&buffer, '\n');
1619
1620        /* And add the comment */
1621        strbuf_add(&buffer, msg, msg_len);
1622
1623        /* And check the encoding */
1624        if (encoding_is_utf8 && !verify_utf8(&buffer))
1625                fprintf(stderr, _(commit_utf8_warn));
1626
1627        if (sign_commit && do_sign_commit(&buffer, sign_commit)) {
1628                result = -1;
1629                goto out;
1630        }
1631
1632        result = write_object_file(buffer.buf, buffer.len, commit_type, ret);
1633out:
1634        strbuf_release(&buffer);
1635        return result;
1636}
1637
1638void set_merge_remote_desc(struct commit *commit,
1639                           const char *name, struct object *obj)
1640{
1641        struct merge_remote_desc *desc;
1642        FLEX_ALLOC_STR(desc, name, name);
1643        desc->obj = obj;
1644        commit->util = desc;
1645}
1646
1647struct commit *get_merge_parent(const char *name)
1648{
1649        struct object *obj;
1650        struct commit *commit;
1651        struct object_id oid;
1652        if (get_oid(name, &oid))
1653                return NULL;
1654        obj = parse_object(&oid);
1655        commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1656        if (commit && !commit->util)
1657                set_merge_remote_desc(commit, name, obj);
1658        return commit;
1659}
1660
1661/*
1662 * Append a commit to the end of the commit_list.
1663 *
1664 * next starts by pointing to the variable that holds the head of an
1665 * empty commit_list, and is updated to point to the "next" field of
1666 * the last item on the list as new commits are appended.
1667 *
1668 * Usage example:
1669 *
1670 *     struct commit_list *list;
1671 *     struct commit_list **next = &list;
1672 *
1673 *     next = commit_list_append(c1, next);
1674 *     next = commit_list_append(c2, next);
1675 *     assert(commit_list_count(list) == 2);
1676 *     return list;
1677 */
1678struct commit_list **commit_list_append(struct commit *commit,
1679                                        struct commit_list **next)
1680{
1681        struct commit_list *new_commit = xmalloc(sizeof(struct commit_list));
1682        new_commit->item = commit;
1683        *next = new_commit;
1684        new_commit->next = NULL;
1685        return &new_commit->next;
1686}
1687
1688const char *find_commit_header(const char *msg, const char *key, size_t *out_len)
1689{
1690        int key_len = strlen(key);
1691        const char *line = msg;
1692
1693        while (line) {
1694                const char *eol = strchrnul(line, '\n');
1695
1696                if (line == eol)
1697                        return NULL;
1698
1699                if (eol - line > key_len &&
1700                    !strncmp(line, key, key_len) &&
1701                    line[key_len] == ' ') {
1702                        *out_len = eol - line - key_len - 1;
1703                        return line + key_len + 1;
1704                }
1705                line = *eol ? eol + 1 : NULL;
1706        }
1707        return NULL;
1708}
1709
1710/*
1711 * Inspect the given string and determine the true "end" of the log message, in
1712 * order to find where to put a new Signed-off-by: line.  Ignored are
1713 * trailing comment lines and blank lines.  To support "git commit -s
1714 * --amend" on an existing commit, we also ignore "Conflicts:".  To
1715 * support "git commit -v", we truncate at cut lines.
1716 *
1717 * Returns the number of bytes from the tail to ignore, to be fed as
1718 * the second parameter to append_signoff().
1719 */
1720int ignore_non_trailer(const char *buf, size_t len)
1721{
1722        int boc = 0;
1723        int bol = 0;
1724        int in_old_conflicts_block = 0;
1725        size_t cutoff = wt_status_locate_end(buf, len);
1726
1727        while (bol < cutoff) {
1728                const char *next_line = memchr(buf + bol, '\n', len - bol);
1729
1730                if (!next_line)
1731                        next_line = buf + len;
1732                else
1733                        next_line++;
1734
1735                if (buf[bol] == comment_line_char || buf[bol] == '\n') {
1736                        /* is this the first of the run of comments? */
1737                        if (!boc)
1738                                boc = bol;
1739                        /* otherwise, it is just continuing */
1740                } else if (starts_with(buf + bol, "Conflicts:\n")) {
1741                        in_old_conflicts_block = 1;
1742                        if (!boc)
1743                                boc = bol;
1744                } else if (in_old_conflicts_block && buf[bol] == '\t') {
1745                        ; /* a pathname in the conflicts block */
1746                } else if (boc) {
1747                        /* the previous was not trailing comment */
1748                        boc = 0;
1749                        in_old_conflicts_block = 0;
1750                }
1751                bol = next_line - buf;
1752        }
1753        return boc ? len - boc : len - cutoff;
1754}