commit.con commit fetch-pack: write shallow, then check connectivity (cf1e7c0)
   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)
 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        return 0;
 403}
 404
 405int parse_commit_gently(struct commit *item, int quiet_on_missing)
 406{
 407        enum object_type type;
 408        void *buffer;
 409        unsigned long size;
 410        int ret;
 411
 412        if (!item)
 413                return -1;
 414        if (item->object.parsed)
 415                return 0;
 416        if (parse_commit_in_graph(item))
 417                return 0;
 418        buffer = read_object_file(&item->object.oid, &type, &size);
 419        if (!buffer)
 420                return quiet_on_missing ? -1 :
 421                        error("Could not read %s",
 422                             oid_to_hex(&item->object.oid));
 423        if (type != OBJ_COMMIT) {
 424                free(buffer);
 425                return error("Object %s not a commit",
 426                             oid_to_hex(&item->object.oid));
 427        }
 428        ret = parse_commit_buffer(item, buffer, size);
 429        if (save_commit_buffer && !ret) {
 430                set_commit_buffer(item, buffer, size);
 431                return 0;
 432        }
 433        free(buffer);
 434        return ret;
 435}
 436
 437void parse_commit_or_die(struct commit *item)
 438{
 439        if (parse_commit(item))
 440                die("unable to parse commit %s",
 441                    item ? oid_to_hex(&item->object.oid) : "(null)");
 442}
 443
 444int find_commit_subject(const char *commit_buffer, const char **subject)
 445{
 446        const char *eol;
 447        const char *p = commit_buffer;
 448
 449        while (*p && (*p != '\n' || p[1] != '\n'))
 450                p++;
 451        if (*p) {
 452                p = skip_blank_lines(p + 2);
 453                eol = strchrnul(p, '\n');
 454        } else
 455                eol = p;
 456
 457        *subject = p;
 458
 459        return eol - p;
 460}
 461
 462struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
 463{
 464        struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
 465        new_list->item = item;
 466        new_list->next = *list_p;
 467        *list_p = new_list;
 468        return new_list;
 469}
 470
 471unsigned commit_list_count(const struct commit_list *l)
 472{
 473        unsigned c = 0;
 474        for (; l; l = l->next )
 475                c++;
 476        return c;
 477}
 478
 479struct commit_list *copy_commit_list(struct commit_list *list)
 480{
 481        struct commit_list *head = NULL;
 482        struct commit_list **pp = &head;
 483        while (list) {
 484                pp = commit_list_append(list->item, pp);
 485                list = list->next;
 486        }
 487        return head;
 488}
 489
 490void free_commit_list(struct commit_list *list)
 491{
 492        while (list)
 493                pop_commit(&list);
 494}
 495
 496struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
 497{
 498        struct commit_list **pp = list;
 499        struct commit_list *p;
 500        while ((p = *pp) != NULL) {
 501                if (p->item->date < item->date) {
 502                        break;
 503                }
 504                pp = &p->next;
 505        }
 506        return commit_list_insert(item, pp);
 507}
 508
 509static int commit_list_compare_by_date(const void *a, const void *b)
 510{
 511        timestamp_t a_date = ((const struct commit_list *)a)->item->date;
 512        timestamp_t b_date = ((const struct commit_list *)b)->item->date;
 513        if (a_date < b_date)
 514                return 1;
 515        if (a_date > b_date)
 516                return -1;
 517        return 0;
 518}
 519
 520static void *commit_list_get_next(const void *a)
 521{
 522        return ((const struct commit_list *)a)->next;
 523}
 524
 525static void commit_list_set_next(void *a, void *next)
 526{
 527        ((struct commit_list *)a)->next = next;
 528}
 529
 530void commit_list_sort_by_date(struct commit_list **list)
 531{
 532        *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
 533                                commit_list_compare_by_date);
 534}
 535
 536struct commit *pop_most_recent_commit(struct commit_list **list,
 537                                      unsigned int mark)
 538{
 539        struct commit *ret = pop_commit(list);
 540        struct commit_list *parents = ret->parents;
 541
 542        while (parents) {
 543                struct commit *commit = parents->item;
 544                if (!parse_commit(commit) && !(commit->object.flags & mark)) {
 545                        commit->object.flags |= mark;
 546                        commit_list_insert_by_date(commit, list);
 547                }
 548                parents = parents->next;
 549        }
 550        return ret;
 551}
 552
 553static void clear_commit_marks_1(struct commit_list **plist,
 554                                 struct commit *commit, unsigned int mark)
 555{
 556        while (commit) {
 557                struct commit_list *parents;
 558
 559                if (!(mark & commit->object.flags))
 560                        return;
 561
 562                commit->object.flags &= ~mark;
 563
 564                parents = commit->parents;
 565                if (!parents)
 566                        return;
 567
 568                while ((parents = parents->next))
 569                        commit_list_insert(parents->item, plist);
 570
 571                commit = commit->parents->item;
 572        }
 573}
 574
 575void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
 576{
 577        struct commit_list *list = NULL;
 578
 579        while (nr--) {
 580                clear_commit_marks_1(&list, *commit, mark);
 581                commit++;
 582        }
 583        while (list)
 584                clear_commit_marks_1(&list, pop_commit(&list), mark);
 585}
 586
 587void clear_commit_marks(struct commit *commit, unsigned int mark)
 588{
 589        clear_commit_marks_many(1, &commit, mark);
 590}
 591
 592struct commit *pop_commit(struct commit_list **stack)
 593{
 594        struct commit_list *top = *stack;
 595        struct commit *item = top ? top->item : NULL;
 596
 597        if (top) {
 598                *stack = top->next;
 599                free(top);
 600        }
 601        return item;
 602}
 603
 604/*
 605 * Topological sort support
 606 */
 607
 608/* count number of children that have not been emitted */
 609define_commit_slab(indegree_slab, int);
 610
 611/* record author-date for each commit object */
 612define_commit_slab(author_date_slab, unsigned long);
 613
 614static void record_author_date(struct author_date_slab *author_date,
 615                               struct commit *commit)
 616{
 617        const char *buffer = get_commit_buffer(commit, NULL);
 618        struct ident_split ident;
 619        const char *ident_line;
 620        size_t ident_len;
 621        char *date_end;
 622        timestamp_t date;
 623
 624        ident_line = find_commit_header(buffer, "author", &ident_len);
 625        if (!ident_line)
 626                goto fail_exit; /* no author line */
 627        if (split_ident_line(&ident, ident_line, ident_len) ||
 628            !ident.date_begin || !ident.date_end)
 629                goto fail_exit; /* malformed "author" line */
 630
 631        date = parse_timestamp(ident.date_begin, &date_end, 10);
 632        if (date_end != ident.date_end)
 633                goto fail_exit; /* malformed date */
 634        *(author_date_slab_at(author_date, commit)) = date;
 635
 636fail_exit:
 637        unuse_commit_buffer(commit, buffer);
 638}
 639
 640static int compare_commits_by_author_date(const void *a_, const void *b_,
 641                                          void *cb_data)
 642{
 643        const struct commit *a = a_, *b = b_;
 644        struct author_date_slab *author_date = cb_data;
 645        timestamp_t a_date = *(author_date_slab_at(author_date, a));
 646        timestamp_t b_date = *(author_date_slab_at(author_date, b));
 647
 648        /* newer commits with larger date first */
 649        if (a_date < b_date)
 650                return 1;
 651        else if (a_date > b_date)
 652                return -1;
 653        return 0;
 654}
 655
 656int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
 657{
 658        const struct commit *a = a_, *b = b_;
 659        /* newer commits with larger date first */
 660        if (a->date < b->date)
 661                return 1;
 662        else if (a->date > b->date)
 663                return -1;
 664        return 0;
 665}
 666
 667/*
 668 * Performs an in-place topological sort on the list supplied.
 669 */
 670void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
 671{
 672        struct commit_list *next, *orig = *list;
 673        struct commit_list **pptr;
 674        struct indegree_slab indegree;
 675        struct prio_queue queue;
 676        struct commit *commit;
 677        struct author_date_slab author_date;
 678
 679        if (!orig)
 680                return;
 681        *list = NULL;
 682
 683        init_indegree_slab(&indegree);
 684        memset(&queue, '\0', sizeof(queue));
 685
 686        switch (sort_order) {
 687        default: /* REV_SORT_IN_GRAPH_ORDER */
 688                queue.compare = NULL;
 689                break;
 690        case REV_SORT_BY_COMMIT_DATE:
 691                queue.compare = compare_commits_by_commit_date;
 692                break;
 693        case REV_SORT_BY_AUTHOR_DATE:
 694                init_author_date_slab(&author_date);
 695                queue.compare = compare_commits_by_author_date;
 696                queue.cb_data = &author_date;
 697                break;
 698        }
 699
 700        /* Mark them and clear the indegree */
 701        for (next = orig; next; next = next->next) {
 702                struct commit *commit = next->item;
 703                *(indegree_slab_at(&indegree, commit)) = 1;
 704                /* also record the author dates, if needed */
 705                if (sort_order == REV_SORT_BY_AUTHOR_DATE)
 706                        record_author_date(&author_date, commit);
 707        }
 708
 709        /* update the indegree */
 710        for (next = orig; next; next = next->next) {
 711                struct commit_list *parents = next->item->parents;
 712                while (parents) {
 713                        struct commit *parent = parents->item;
 714                        int *pi = indegree_slab_at(&indegree, parent);
 715
 716                        if (*pi)
 717                                (*pi)++;
 718                        parents = parents->next;
 719                }
 720        }
 721
 722        /*
 723         * find the tips
 724         *
 725         * tips are nodes not reachable from any other node in the list
 726         *
 727         * the tips serve as a starting set for the work queue.
 728         */
 729        for (next = orig; next; next = next->next) {
 730                struct commit *commit = next->item;
 731
 732                if (*(indegree_slab_at(&indegree, commit)) == 1)
 733                        prio_queue_put(&queue, commit);
 734        }
 735
 736        /*
 737         * This is unfortunate; the initial tips need to be shown
 738         * in the order given from the revision traversal machinery.
 739         */
 740        if (sort_order == REV_SORT_IN_GRAPH_ORDER)
 741                prio_queue_reverse(&queue);
 742
 743        /* We no longer need the commit list */
 744        free_commit_list(orig);
 745
 746        pptr = list;
 747        *list = NULL;
 748        while ((commit = prio_queue_get(&queue)) != NULL) {
 749                struct commit_list *parents;
 750
 751                for (parents = commit->parents; parents ; parents = parents->next) {
 752                        struct commit *parent = parents->item;
 753                        int *pi = indegree_slab_at(&indegree, parent);
 754
 755                        if (!*pi)
 756                                continue;
 757
 758                        /*
 759                         * parents are only enqueued for emission
 760                         * when all their children have been emitted thereby
 761                         * guaranteeing topological order.
 762                         */
 763                        if (--(*pi) == 1)
 764                                prio_queue_put(&queue, parent);
 765                }
 766                /*
 767                 * all children of commit have already been
 768                 * emitted. we can emit it now.
 769                 */
 770                *(indegree_slab_at(&indegree, commit)) = 0;
 771
 772                pptr = &commit_list_insert(commit, pptr)->next;
 773        }
 774
 775        clear_indegree_slab(&indegree);
 776        clear_prio_queue(&queue);
 777        if (sort_order == REV_SORT_BY_AUTHOR_DATE)
 778                clear_author_date_slab(&author_date);
 779}
 780
 781/* merge-base stuff */
 782
 783/* Remember to update object flag allocation in object.h */
 784#define PARENT1         (1u<<16)
 785#define PARENT2         (1u<<17)
 786#define STALE           (1u<<18)
 787#define RESULT          (1u<<19)
 788
 789static const unsigned all_flags = (PARENT1 | PARENT2 | STALE | RESULT);
 790
 791static int queue_has_nonstale(struct prio_queue *queue)
 792{
 793        int i;
 794        for (i = 0; i < queue->nr; i++) {
 795                struct commit *commit = queue->array[i].data;
 796                if (!(commit->object.flags & STALE))
 797                        return 1;
 798        }
 799        return 0;
 800}
 801
 802/* all input commits in one and twos[] must have been parsed! */
 803static struct commit_list *paint_down_to_common(struct commit *one, int n, struct commit **twos)
 804{
 805        struct prio_queue queue = { compare_commits_by_commit_date };
 806        struct commit_list *result = NULL;
 807        int i;
 808
 809        one->object.flags |= PARENT1;
 810        if (!n) {
 811                commit_list_append(one, &result);
 812                return result;
 813        }
 814        prio_queue_put(&queue, one);
 815
 816        for (i = 0; i < n; i++) {
 817                twos[i]->object.flags |= PARENT2;
 818                prio_queue_put(&queue, twos[i]);
 819        }
 820
 821        while (queue_has_nonstale(&queue)) {
 822                struct commit *commit = prio_queue_get(&queue);
 823                struct commit_list *parents;
 824                int flags;
 825
 826                flags = commit->object.flags & (PARENT1 | PARENT2 | STALE);
 827                if (flags == (PARENT1 | PARENT2)) {
 828                        if (!(commit->object.flags & RESULT)) {
 829                                commit->object.flags |= RESULT;
 830                                commit_list_insert_by_date(commit, &result);
 831                        }
 832                        /* Mark parents of a found merge stale */
 833                        flags |= STALE;
 834                }
 835                parents = commit->parents;
 836                while (parents) {
 837                        struct commit *p = parents->item;
 838                        parents = parents->next;
 839                        if ((p->object.flags & flags) == flags)
 840                                continue;
 841                        if (parse_commit(p))
 842                                return NULL;
 843                        p->object.flags |= flags;
 844                        prio_queue_put(&queue, p);
 845                }
 846        }
 847
 848        clear_prio_queue(&queue);
 849        return result;
 850}
 851
 852static struct commit_list *merge_bases_many(struct commit *one, int n, struct commit **twos)
 853{
 854        struct commit_list *list = NULL;
 855        struct commit_list *result = NULL;
 856        int i;
 857
 858        for (i = 0; i < n; i++) {
 859                if (one == twos[i])
 860                        /*
 861                         * We do not mark this even with RESULT so we do not
 862                         * have to clean it up.
 863                         */
 864                        return commit_list_insert(one, &result);
 865        }
 866
 867        if (parse_commit(one))
 868                return NULL;
 869        for (i = 0; i < n; i++) {
 870                if (parse_commit(twos[i]))
 871                        return NULL;
 872        }
 873
 874        list = paint_down_to_common(one, n, twos);
 875
 876        while (list) {
 877                struct commit *commit = pop_commit(&list);
 878                if (!(commit->object.flags & STALE))
 879                        commit_list_insert_by_date(commit, &result);
 880        }
 881        return result;
 882}
 883
 884struct commit_list *get_octopus_merge_bases(struct commit_list *in)
 885{
 886        struct commit_list *i, *j, *k, *ret = NULL;
 887
 888        if (!in)
 889                return ret;
 890
 891        commit_list_insert(in->item, &ret);
 892
 893        for (i = in->next; i; i = i->next) {
 894                struct commit_list *new_commits = NULL, *end = NULL;
 895
 896                for (j = ret; j; j = j->next) {
 897                        struct commit_list *bases;
 898                        bases = get_merge_bases(i->item, j->item);
 899                        if (!new_commits)
 900                                new_commits = bases;
 901                        else
 902                                end->next = bases;
 903                        for (k = bases; k; k = k->next)
 904                                end = k;
 905                }
 906                ret = new_commits;
 907        }
 908        return ret;
 909}
 910
 911static int remove_redundant(struct commit **array, int cnt)
 912{
 913        /*
 914         * Some commit in the array may be an ancestor of
 915         * another commit.  Move such commit to the end of
 916         * the array, and return the number of commits that
 917         * are independent from each other.
 918         */
 919        struct commit **work;
 920        unsigned char *redundant;
 921        int *filled_index;
 922        int i, j, filled;
 923
 924        work = xcalloc(cnt, sizeof(*work));
 925        redundant = xcalloc(cnt, 1);
 926        ALLOC_ARRAY(filled_index, cnt - 1);
 927
 928        for (i = 0; i < cnt; i++)
 929                parse_commit(array[i]);
 930        for (i = 0; i < cnt; i++) {
 931                struct commit_list *common;
 932
 933                if (redundant[i])
 934                        continue;
 935                for (j = filled = 0; j < cnt; j++) {
 936                        if (i == j || redundant[j])
 937                                continue;
 938                        filled_index[filled] = j;
 939                        work[filled++] = array[j];
 940                }
 941                common = paint_down_to_common(array[i], filled, work);
 942                if (array[i]->object.flags & PARENT2)
 943                        redundant[i] = 1;
 944                for (j = 0; j < filled; j++)
 945                        if (work[j]->object.flags & PARENT1)
 946                                redundant[filled_index[j]] = 1;
 947                clear_commit_marks(array[i], all_flags);
 948                clear_commit_marks_many(filled, work, all_flags);
 949                free_commit_list(common);
 950        }
 951
 952        /* Now collect the result */
 953        COPY_ARRAY(work, array, cnt);
 954        for (i = filled = 0; i < cnt; i++)
 955                if (!redundant[i])
 956                        array[filled++] = work[i];
 957        for (j = filled, i = 0; i < cnt; i++)
 958                if (redundant[i])
 959                        array[j++] = work[i];
 960        free(work);
 961        free(redundant);
 962        free(filled_index);
 963        return filled;
 964}
 965
 966static struct commit_list *get_merge_bases_many_0(struct commit *one,
 967                                                  int n,
 968                                                  struct commit **twos,
 969                                                  int cleanup)
 970{
 971        struct commit_list *list;
 972        struct commit **rslt;
 973        struct commit_list *result;
 974        int cnt, i;
 975
 976        result = merge_bases_many(one, n, twos);
 977        for (i = 0; i < n; i++) {
 978                if (one == twos[i])
 979                        return result;
 980        }
 981        if (!result || !result->next) {
 982                if (cleanup) {
 983                        clear_commit_marks(one, all_flags);
 984                        clear_commit_marks_many(n, twos, all_flags);
 985                }
 986                return result;
 987        }
 988
 989        /* There are more than one */
 990        cnt = commit_list_count(result);
 991        rslt = xcalloc(cnt, sizeof(*rslt));
 992        for (list = result, i = 0; list; list = list->next)
 993                rslt[i++] = list->item;
 994        free_commit_list(result);
 995
 996        clear_commit_marks(one, all_flags);
 997        clear_commit_marks_many(n, twos, all_flags);
 998
 999        cnt = remove_redundant(rslt, cnt);
1000        result = NULL;
1001        for (i = 0; i < cnt; i++)
1002                commit_list_insert_by_date(rslt[i], &result);
1003        free(rslt);
1004        return result;
1005}
1006
1007struct commit_list *get_merge_bases_many(struct commit *one,
1008                                         int n,
1009                                         struct commit **twos)
1010{
1011        return get_merge_bases_many_0(one, n, twos, 1);
1012}
1013
1014struct commit_list *get_merge_bases_many_dirty(struct commit *one,
1015                                               int n,
1016                                               struct commit **twos)
1017{
1018        return get_merge_bases_many_0(one, n, twos, 0);
1019}
1020
1021struct commit_list *get_merge_bases(struct commit *one, struct commit *two)
1022{
1023        return get_merge_bases_many_0(one, 1, &two, 1);
1024}
1025
1026/*
1027 * Is "commit" a descendant of one of the elements on the "with_commit" list?
1028 */
1029int is_descendant_of(struct commit *commit, struct commit_list *with_commit)
1030{
1031        if (!with_commit)
1032                return 1;
1033        while (with_commit) {
1034                struct commit *other;
1035
1036                other = with_commit->item;
1037                with_commit = with_commit->next;
1038                if (in_merge_bases(other, commit))
1039                        return 1;
1040        }
1041        return 0;
1042}
1043
1044/*
1045 * Is "commit" an ancestor of one of the "references"?
1046 */
1047int in_merge_bases_many(struct commit *commit, int nr_reference, struct commit **reference)
1048{
1049        struct commit_list *bases;
1050        int ret = 0, i;
1051
1052        if (parse_commit(commit))
1053                return ret;
1054        for (i = 0; i < nr_reference; i++)
1055                if (parse_commit(reference[i]))
1056                        return ret;
1057
1058        bases = paint_down_to_common(commit, nr_reference, reference);
1059        if (commit->object.flags & PARENT2)
1060                ret = 1;
1061        clear_commit_marks(commit, all_flags);
1062        clear_commit_marks_many(nr_reference, reference, all_flags);
1063        free_commit_list(bases);
1064        return ret;
1065}
1066
1067/*
1068 * Is "commit" an ancestor of (i.e. reachable from) the "reference"?
1069 */
1070int in_merge_bases(struct commit *commit, struct commit *reference)
1071{
1072        return in_merge_bases_many(commit, 1, &reference);
1073}
1074
1075struct commit_list *reduce_heads(struct commit_list *heads)
1076{
1077        struct commit_list *p;
1078        struct commit_list *result = NULL, **tail = &result;
1079        struct commit **array;
1080        int num_head, i;
1081
1082        if (!heads)
1083                return NULL;
1084
1085        /* Uniquify */
1086        for (p = heads; p; p = p->next)
1087                p->item->object.flags &= ~STALE;
1088        for (p = heads, num_head = 0; p; p = p->next) {
1089                if (p->item->object.flags & STALE)
1090                        continue;
1091                p->item->object.flags |= STALE;
1092                num_head++;
1093        }
1094        array = xcalloc(num_head, sizeof(*array));
1095        for (p = heads, i = 0; p; p = p->next) {
1096                if (p->item->object.flags & STALE) {
1097                        array[i++] = p->item;
1098                        p->item->object.flags &= ~STALE;
1099                }
1100        }
1101        num_head = remove_redundant(array, num_head);
1102        for (i = 0; i < num_head; i++)
1103                tail = &commit_list_insert(array[i], tail)->next;
1104        free(array);
1105        return result;
1106}
1107
1108void reduce_heads_replace(struct commit_list **heads)
1109{
1110        struct commit_list *result = reduce_heads(*heads);
1111        free_commit_list(*heads);
1112        *heads = result;
1113}
1114
1115static const char gpg_sig_header[] = "gpgsig";
1116static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
1117
1118static int do_sign_commit(struct strbuf *buf, const char *keyid)
1119{
1120        struct strbuf sig = STRBUF_INIT;
1121        int inspos, copypos;
1122        const char *eoh;
1123
1124        /* find the end of the header */
1125        eoh = strstr(buf->buf, "\n\n");
1126        if (!eoh)
1127                inspos = buf->len;
1128        else
1129                inspos = eoh - buf->buf + 1;
1130
1131        if (!keyid || !*keyid)
1132                keyid = get_signing_key();
1133        if (sign_buffer(buf, &sig, keyid)) {
1134                strbuf_release(&sig);
1135                return -1;
1136        }
1137
1138        for (copypos = 0; sig.buf[copypos]; ) {
1139                const char *bol = sig.buf + copypos;
1140                const char *eol = strchrnul(bol, '\n');
1141                int len = (eol - bol) + !!*eol;
1142
1143                if (!copypos) {
1144                        strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
1145                        inspos += gpg_sig_header_len;
1146                }
1147                strbuf_insert(buf, inspos++, " ", 1);
1148                strbuf_insert(buf, inspos, bol, len);
1149                inspos += len;
1150                copypos += len;
1151        }
1152        strbuf_release(&sig);
1153        return 0;
1154}
1155
1156int parse_signed_commit(const struct commit *commit,
1157                        struct strbuf *payload, struct strbuf *signature)
1158{
1159
1160        unsigned long size;
1161        const char *buffer = get_commit_buffer(commit, &size);
1162        int in_signature, saw_signature = -1;
1163        const char *line, *tail;
1164
1165        line = buffer;
1166        tail = buffer + size;
1167        in_signature = 0;
1168        saw_signature = 0;
1169        while (line < tail) {
1170                const char *sig = NULL;
1171                const char *next = memchr(line, '\n', tail - line);
1172
1173                next = next ? next + 1 : tail;
1174                if (in_signature && line[0] == ' ')
1175                        sig = line + 1;
1176                else if (starts_with(line, gpg_sig_header) &&
1177                         line[gpg_sig_header_len] == ' ')
1178                        sig = line + gpg_sig_header_len + 1;
1179                if (sig) {
1180                        strbuf_add(signature, sig, next - sig);
1181                        saw_signature = 1;
1182                        in_signature = 1;
1183                } else {
1184                        if (*line == '\n')
1185                                /* dump the whole remainder of the buffer */
1186                                next = tail;
1187                        strbuf_add(payload, line, next - line);
1188                        in_signature = 0;
1189                }
1190                line = next;
1191        }
1192        unuse_commit_buffer(commit, buffer);
1193        return saw_signature;
1194}
1195
1196int remove_signature(struct strbuf *buf)
1197{
1198        const char *line = buf->buf;
1199        const char *tail = buf->buf + buf->len;
1200        int in_signature = 0;
1201        const char *sig_start = NULL;
1202        const char *sig_end = NULL;
1203
1204        while (line < tail) {
1205                const char *next = memchr(line, '\n', tail - line);
1206                next = next ? next + 1 : tail;
1207
1208                if (in_signature && line[0] == ' ')
1209                        sig_end = next;
1210                else if (starts_with(line, gpg_sig_header) &&
1211                         line[gpg_sig_header_len] == ' ') {
1212                        sig_start = line;
1213                        sig_end = next;
1214                        in_signature = 1;
1215                } else {
1216                        if (*line == '\n')
1217                                /* dump the whole remainder of the buffer */
1218                                next = tail;
1219                        in_signature = 0;
1220                }
1221                line = next;
1222        }
1223
1224        if (sig_start)
1225                strbuf_remove(buf, sig_start - buf->buf, sig_end - sig_start);
1226
1227        return sig_start != NULL;
1228}
1229
1230static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1231{
1232        struct merge_remote_desc *desc;
1233        struct commit_extra_header *mergetag;
1234        char *buf;
1235        unsigned long size, len;
1236        enum object_type type;
1237
1238        desc = merge_remote_util(parent);
1239        if (!desc || !desc->obj)
1240                return;
1241        buf = read_object_file(&desc->obj->oid, &type, &size);
1242        if (!buf || type != OBJ_TAG)
1243                goto free_return;
1244        len = parse_signature(buf, size);
1245        if (size == len)
1246                goto free_return;
1247        /*
1248         * We could verify this signature and either omit the tag when
1249         * it does not validate, but the integrator may not have the
1250         * public key of the signer of the tag he is merging, while a
1251         * later auditor may have it while auditing, so let's not run
1252         * verify-signed-buffer here for now...
1253         *
1254         * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1255         *      warn("warning: signed tag unverified.");
1256         */
1257        mergetag = xcalloc(1, sizeof(*mergetag));
1258        mergetag->key = xstrdup("mergetag");
1259        mergetag->value = buf;
1260        mergetag->len = size;
1261
1262        **tail = mergetag;
1263        *tail = &mergetag->next;
1264        return;
1265
1266free_return:
1267        free(buf);
1268}
1269
1270int check_commit_signature(const struct commit *commit, struct signature_check *sigc)
1271{
1272        struct strbuf payload = STRBUF_INIT;
1273        struct strbuf signature = STRBUF_INIT;
1274        int ret = 1;
1275
1276        sigc->result = 'N';
1277
1278        if (parse_signed_commit(commit, &payload, &signature) <= 0)
1279                goto out;
1280        ret = check_signature(payload.buf, payload.len, signature.buf,
1281                signature.len, sigc);
1282
1283 out:
1284        strbuf_release(&payload);
1285        strbuf_release(&signature);
1286
1287        return ret;
1288}
1289
1290
1291
1292void append_merge_tag_headers(struct commit_list *parents,
1293                              struct commit_extra_header ***tail)
1294{
1295        while (parents) {
1296                struct commit *parent = parents->item;
1297                handle_signed_tag(parent, tail);
1298                parents = parents->next;
1299        }
1300}
1301
1302static void add_extra_header(struct strbuf *buffer,
1303                             struct commit_extra_header *extra)
1304{
1305        strbuf_addstr(buffer, extra->key);
1306        if (extra->len)
1307                strbuf_add_lines(buffer, " ", extra->value, extra->len);
1308        else
1309                strbuf_addch(buffer, '\n');
1310}
1311
1312struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1313                                                      const char **exclude)
1314{
1315        struct commit_extra_header *extra = NULL;
1316        unsigned long size;
1317        const char *buffer = get_commit_buffer(commit, &size);
1318        extra = read_commit_extra_header_lines(buffer, size, exclude);
1319        unuse_commit_buffer(commit, buffer);
1320        return extra;
1321}
1322
1323int for_each_mergetag(each_mergetag_fn fn, struct commit *commit, void *data)
1324{
1325        struct commit_extra_header *extra, *to_free;
1326        int res = 0;
1327
1328        to_free = read_commit_extra_headers(commit, NULL);
1329        for (extra = to_free; !res && extra; extra = extra->next) {
1330                if (strcmp(extra->key, "mergetag"))
1331                        continue; /* not a merge tag */
1332                res = fn(commit, extra, data);
1333        }
1334        free_commit_extra_headers(to_free);
1335        return res;
1336}
1337
1338static inline int standard_header_field(const char *field, size_t len)
1339{
1340        return ((len == 4 && !memcmp(field, "tree", 4)) ||
1341                (len == 6 && !memcmp(field, "parent", 6)) ||
1342                (len == 6 && !memcmp(field, "author", 6)) ||
1343                (len == 9 && !memcmp(field, "committer", 9)) ||
1344                (len == 8 && !memcmp(field, "encoding", 8)));
1345}
1346
1347static int excluded_header_field(const char *field, size_t len, const char **exclude)
1348{
1349        if (!exclude)
1350                return 0;
1351
1352        while (*exclude) {
1353                size_t xlen = strlen(*exclude);
1354                if (len == xlen && !memcmp(field, *exclude, xlen))
1355                        return 1;
1356                exclude++;
1357        }
1358        return 0;
1359}
1360
1361static struct commit_extra_header *read_commit_extra_header_lines(
1362        const char *buffer, size_t size,
1363        const char **exclude)
1364{
1365        struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1366        const char *line, *next, *eof, *eob;
1367        struct strbuf buf = STRBUF_INIT;
1368
1369        for (line = buffer, eob = line + size;
1370             line < eob && *line != '\n';
1371             line = next) {
1372                next = memchr(line, '\n', eob - line);
1373                next = next ? next + 1 : eob;
1374                if (*line == ' ') {
1375                        /* continuation */
1376                        if (it)
1377                                strbuf_add(&buf, line + 1, next - (line + 1));
1378                        continue;
1379                }
1380                if (it)
1381                        it->value = strbuf_detach(&buf, &it->len);
1382                strbuf_reset(&buf);
1383                it = NULL;
1384
1385                eof = memchr(line, ' ', next - line);
1386                if (!eof)
1387                        eof = next;
1388                else if (standard_header_field(line, eof - line) ||
1389                         excluded_header_field(line, eof - line, exclude))
1390                        continue;
1391
1392                it = xcalloc(1, sizeof(*it));
1393                it->key = xmemdupz(line, eof-line);
1394                *tail = it;
1395                tail = &it->next;
1396                if (eof + 1 < next)
1397                        strbuf_add(&buf, eof + 1, next - (eof + 1));
1398        }
1399        if (it)
1400                it->value = strbuf_detach(&buf, &it->len);
1401        return extra;
1402}
1403
1404void free_commit_extra_headers(struct commit_extra_header *extra)
1405{
1406        while (extra) {
1407                struct commit_extra_header *next = extra->next;
1408                free(extra->key);
1409                free(extra->value);
1410                free(extra);
1411                extra = next;
1412        }
1413}
1414
1415int commit_tree(const char *msg, size_t msg_len, const struct object_id *tree,
1416                struct commit_list *parents, struct object_id *ret,
1417                const char *author, const char *sign_commit)
1418{
1419        struct commit_extra_header *extra = NULL, **tail = &extra;
1420        int result;
1421
1422        append_merge_tag_headers(parents, &tail);
1423        result = commit_tree_extended(msg, msg_len, tree, parents, ret,
1424                                      author, sign_commit, extra);
1425        free_commit_extra_headers(extra);
1426        return result;
1427}
1428
1429static int find_invalid_utf8(const char *buf, int len)
1430{
1431        int offset = 0;
1432        static const unsigned int max_codepoint[] = {
1433                0x7f, 0x7ff, 0xffff, 0x10ffff
1434        };
1435
1436        while (len) {
1437                unsigned char c = *buf++;
1438                int bytes, bad_offset;
1439                unsigned int codepoint;
1440                unsigned int min_val, max_val;
1441
1442                len--;
1443                offset++;
1444
1445                /* Simple US-ASCII? No worries. */
1446                if (c < 0x80)
1447                        continue;
1448
1449                bad_offset = offset-1;
1450
1451                /*
1452                 * Count how many more high bits set: that's how
1453                 * many more bytes this sequence should have.
1454                 */
1455                bytes = 0;
1456                while (c & 0x40) {
1457                        c <<= 1;
1458                        bytes++;
1459                }
1460
1461                /*
1462                 * Must be between 1 and 3 more bytes.  Longer sequences result in
1463                 * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1464                 */
1465                if (bytes < 1 || 3 < bytes)
1466                        return bad_offset;
1467
1468                /* Do we *have* that many bytes? */
1469                if (len < bytes)
1470                        return bad_offset;
1471
1472                /*
1473                 * Place the encoded bits at the bottom of the value and compute the
1474                 * valid range.
1475                 */
1476                codepoint = (c & 0x7f) >> bytes;
1477                min_val = max_codepoint[bytes-1] + 1;
1478                max_val = max_codepoint[bytes];
1479
1480                offset += bytes;
1481                len -= bytes;
1482
1483                /* And verify that they are good continuation bytes */
1484                do {
1485                        codepoint <<= 6;
1486                        codepoint |= *buf & 0x3f;
1487                        if ((*buf++ & 0xc0) != 0x80)
1488                                return bad_offset;
1489                } while (--bytes);
1490
1491                /* Reject codepoints that are out of range for the sequence length. */
1492                if (codepoint < min_val || codepoint > max_val)
1493                        return bad_offset;
1494                /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1495                if ((codepoint & 0x1ff800) == 0xd800)
1496                        return bad_offset;
1497                /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1498                if ((codepoint & 0xfffe) == 0xfffe)
1499                        return bad_offset;
1500                /* So are anything in the range U+FDD0..U+FDEF. */
1501                if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1502                        return bad_offset;
1503        }
1504        return -1;
1505}
1506
1507/*
1508 * This verifies that the buffer is in proper utf8 format.
1509 *
1510 * If it isn't, it assumes any non-utf8 characters are Latin1,
1511 * and does the conversion.
1512 */
1513static int verify_utf8(struct strbuf *buf)
1514{
1515        int ok = 1;
1516        long pos = 0;
1517
1518        for (;;) {
1519                int bad;
1520                unsigned char c;
1521                unsigned char replace[2];
1522
1523                bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1524                if (bad < 0)
1525                        return ok;
1526                pos += bad;
1527                ok = 0;
1528                c = buf->buf[pos];
1529                strbuf_remove(buf, pos, 1);
1530
1531                /* We know 'c' must be in the range 128-255 */
1532                replace[0] = 0xc0 + (c >> 6);
1533                replace[1] = 0x80 + (c & 0x3f);
1534                strbuf_insert(buf, pos, replace, 2);
1535                pos += 2;
1536        }
1537}
1538
1539static const char commit_utf8_warn[] =
1540N_("Warning: commit message did not conform to UTF-8.\n"
1541   "You may want to amend it after fixing the message, or set the config\n"
1542   "variable i18n.commitencoding to the encoding your project uses.\n");
1543
1544int commit_tree_extended(const char *msg, size_t msg_len,
1545                         const struct object_id *tree,
1546                         struct commit_list *parents, struct object_id *ret,
1547                         const char *author, const char *sign_commit,
1548                         struct commit_extra_header *extra)
1549{
1550        int result;
1551        int encoding_is_utf8;
1552        struct strbuf buffer;
1553
1554        assert_oid_type(tree, OBJ_TREE);
1555
1556        if (memchr(msg, '\0', msg_len))
1557                return error("a NUL byte in commit log message not allowed.");
1558
1559        /* Not having i18n.commitencoding is the same as having utf-8 */
1560        encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1561
1562        strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1563        strbuf_addf(&buffer, "tree %s\n", oid_to_hex(tree));
1564
1565        /*
1566         * NOTE! This ordering means that the same exact tree merged with a
1567         * different order of parents will be a _different_ changeset even
1568         * if everything else stays the same.
1569         */
1570        while (parents) {
1571                struct commit *parent = pop_commit(&parents);
1572                strbuf_addf(&buffer, "parent %s\n",
1573                            oid_to_hex(&parent->object.oid));
1574        }
1575
1576        /* Person/date information */
1577        if (!author)
1578                author = git_author_info(IDENT_STRICT);
1579        strbuf_addf(&buffer, "author %s\n", author);
1580        strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1581        if (!encoding_is_utf8)
1582                strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1583
1584        while (extra) {
1585                add_extra_header(&buffer, extra);
1586                extra = extra->next;
1587        }
1588        strbuf_addch(&buffer, '\n');
1589
1590        /* And add the comment */
1591        strbuf_add(&buffer, msg, msg_len);
1592
1593        /* And check the encoding */
1594        if (encoding_is_utf8 && !verify_utf8(&buffer))
1595                fprintf(stderr, _(commit_utf8_warn));
1596
1597        if (sign_commit && do_sign_commit(&buffer, sign_commit)) {
1598                result = -1;
1599                goto out;
1600        }
1601
1602        result = write_object_file(buffer.buf, buffer.len, commit_type, ret);
1603out:
1604        strbuf_release(&buffer);
1605        return result;
1606}
1607
1608void set_merge_remote_desc(struct commit *commit,
1609                           const char *name, struct object *obj)
1610{
1611        struct merge_remote_desc *desc;
1612        FLEX_ALLOC_STR(desc, name, name);
1613        desc->obj = obj;
1614        commit->util = desc;
1615}
1616
1617struct commit *get_merge_parent(const char *name)
1618{
1619        struct object *obj;
1620        struct commit *commit;
1621        struct object_id oid;
1622        if (get_oid(name, &oid))
1623                return NULL;
1624        obj = parse_object(&oid);
1625        commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1626        if (commit && !commit->util)
1627                set_merge_remote_desc(commit, name, obj);
1628        return commit;
1629}
1630
1631/*
1632 * Append a commit to the end of the commit_list.
1633 *
1634 * next starts by pointing to the variable that holds the head of an
1635 * empty commit_list, and is updated to point to the "next" field of
1636 * the last item on the list as new commits are appended.
1637 *
1638 * Usage example:
1639 *
1640 *     struct commit_list *list;
1641 *     struct commit_list **next = &list;
1642 *
1643 *     next = commit_list_append(c1, next);
1644 *     next = commit_list_append(c2, next);
1645 *     assert(commit_list_count(list) == 2);
1646 *     return list;
1647 */
1648struct commit_list **commit_list_append(struct commit *commit,
1649                                        struct commit_list **next)
1650{
1651        struct commit_list *new_commit = xmalloc(sizeof(struct commit_list));
1652        new_commit->item = commit;
1653        *next = new_commit;
1654        new_commit->next = NULL;
1655        return &new_commit->next;
1656}
1657
1658const char *find_commit_header(const char *msg, const char *key, size_t *out_len)
1659{
1660        int key_len = strlen(key);
1661        const char *line = msg;
1662
1663        while (line) {
1664                const char *eol = strchrnul(line, '\n');
1665
1666                if (line == eol)
1667                        return NULL;
1668
1669                if (eol - line > key_len &&
1670                    !strncmp(line, key, key_len) &&
1671                    line[key_len] == ' ') {
1672                        *out_len = eol - line - key_len - 1;
1673                        return line + key_len + 1;
1674                }
1675                line = *eol ? eol + 1 : NULL;
1676        }
1677        return NULL;
1678}
1679
1680/*
1681 * Inspect the given string and determine the true "end" of the log message, in
1682 * order to find where to put a new Signed-off-by: line.  Ignored are
1683 * trailing comment lines and blank lines.  To support "git commit -s
1684 * --amend" on an existing commit, we also ignore "Conflicts:".  To
1685 * support "git commit -v", we truncate at cut lines.
1686 *
1687 * Returns the number of bytes from the tail to ignore, to be fed as
1688 * the second parameter to append_signoff().
1689 */
1690int ignore_non_trailer(const char *buf, size_t len)
1691{
1692        int boc = 0;
1693        int bol = 0;
1694        int in_old_conflicts_block = 0;
1695        size_t cutoff = wt_status_locate_end(buf, len);
1696
1697        while (bol < cutoff) {
1698                const char *next_line = memchr(buf + bol, '\n', len - bol);
1699
1700                if (!next_line)
1701                        next_line = buf + len;
1702                else
1703                        next_line++;
1704
1705                if (buf[bol] == comment_line_char || buf[bol] == '\n') {
1706                        /* is this the first of the run of comments? */
1707                        if (!boc)
1708                                boc = bol;
1709                        /* otherwise, it is just continuing */
1710                } else if (starts_with(buf + bol, "Conflicts:\n")) {
1711                        in_old_conflicts_block = 1;
1712                        if (!boc)
1713                                boc = bol;
1714                } else if (in_old_conflicts_block && buf[bol] == '\t') {
1715                        ; /* a pathname in the conflicts block */
1716                } else if (boc) {
1717                        /* the previous was not trailing comment */
1718                        boc = 0;
1719                        in_old_conflicts_block = 0;
1720                }
1721                bol = next_line - buf;
1722        }
1723        return boc ? len - boc : len - cutoff;
1724}