commit.con commit commit-slab: avoid large realloc (66eb375)
   1#include "cache.h"
   2#include "tag.h"
   3#include "commit.h"
   4#include "pkt-line.h"
   5#include "utf8.h"
   6#include "diff.h"
   7#include "revision.h"
   8#include "notes.h"
   9#include "gpg-interface.h"
  10#include "mergesort.h"
  11
  12static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
  13
  14int save_commit_buffer = 1;
  15
  16const char *commit_type = "commit";
  17static int commit_count;
  18
  19static struct commit *check_commit(struct object *obj,
  20                                   const unsigned char *sha1,
  21                                   int quiet)
  22{
  23        if (obj->type != OBJ_COMMIT) {
  24                if (!quiet)
  25                        error("Object %s is a %s, not a commit",
  26                              sha1_to_hex(sha1), typename(obj->type));
  27                return NULL;
  28        }
  29        return (struct commit *) obj;
  30}
  31
  32struct commit *lookup_commit_reference_gently(const unsigned char *sha1,
  33                                              int quiet)
  34{
  35        struct object *obj = deref_tag(parse_object(sha1), NULL, 0);
  36
  37        if (!obj)
  38                return NULL;
  39        return check_commit(obj, sha1, quiet);
  40}
  41
  42struct commit *lookup_commit_reference(const unsigned char *sha1)
  43{
  44        return lookup_commit_reference_gently(sha1, 0);
  45}
  46
  47struct commit *lookup_commit_or_die(const unsigned char *sha1, const char *ref_name)
  48{
  49        struct commit *c = lookup_commit_reference(sha1);
  50        if (!c)
  51                die(_("could not parse %s"), ref_name);
  52        if (hashcmp(sha1, c->object.sha1)) {
  53                warning(_("%s %s is not a commit!"),
  54                        ref_name, sha1_to_hex(sha1));
  55        }
  56        return c;
  57}
  58
  59struct commit *lookup_commit(const unsigned char *sha1)
  60{
  61        struct object *obj = lookup_object(sha1);
  62        if (!obj) {
  63                struct commit *c = alloc_commit_node();
  64                c->index = commit_count++;
  65                return create_object(sha1, OBJ_COMMIT, c);
  66        }
  67        if (!obj->type)
  68                obj->type = OBJ_COMMIT;
  69        return check_commit(obj, sha1, 0);
  70}
  71
  72struct commit *lookup_commit_reference_by_name(const char *name)
  73{
  74        unsigned char sha1[20];
  75        struct commit *commit;
  76
  77        if (get_sha1_committish(name, sha1))
  78                return NULL;
  79        commit = lookup_commit_reference(sha1);
  80        if (!commit || parse_commit(commit))
  81                return NULL;
  82        return commit;
  83}
  84
  85static unsigned long parse_commit_date(const char *buf, const char *tail)
  86{
  87        const char *dateptr;
  88
  89        if (buf + 6 >= tail)
  90                return 0;
  91        if (memcmp(buf, "author", 6))
  92                return 0;
  93        while (buf < tail && *buf++ != '\n')
  94                /* nada */;
  95        if (buf + 9 >= tail)
  96                return 0;
  97        if (memcmp(buf, "committer", 9))
  98                return 0;
  99        while (buf < tail && *buf++ != '>')
 100                /* nada */;
 101        if (buf >= tail)
 102                return 0;
 103        dateptr = buf;
 104        while (buf < tail && *buf++ != '\n')
 105                /* nada */;
 106        if (buf >= tail)
 107                return 0;
 108        /* dateptr < buf && buf[-1] == '\n', so strtoul will stop at buf-1 */
 109        return strtoul(dateptr, NULL, 10);
 110}
 111
 112static struct commit_graft **commit_graft;
 113static int commit_graft_alloc, commit_graft_nr;
 114
 115static int commit_graft_pos(const unsigned char *sha1)
 116{
 117        int lo, hi;
 118        lo = 0;
 119        hi = commit_graft_nr;
 120        while (lo < hi) {
 121                int mi = (lo + hi) / 2;
 122                struct commit_graft *graft = commit_graft[mi];
 123                int cmp = hashcmp(sha1, graft->sha1);
 124                if (!cmp)
 125                        return mi;
 126                if (cmp < 0)
 127                        hi = mi;
 128                else
 129                        lo = mi + 1;
 130        }
 131        return -lo - 1;
 132}
 133
 134int register_commit_graft(struct commit_graft *graft, int ignore_dups)
 135{
 136        int pos = commit_graft_pos(graft->sha1);
 137
 138        if (0 <= pos) {
 139                if (ignore_dups)
 140                        free(graft);
 141                else {
 142                        free(commit_graft[pos]);
 143                        commit_graft[pos] = graft;
 144                }
 145                return 1;
 146        }
 147        pos = -pos - 1;
 148        if (commit_graft_alloc <= ++commit_graft_nr) {
 149                commit_graft_alloc = alloc_nr(commit_graft_alloc);
 150                commit_graft = xrealloc(commit_graft,
 151                                        sizeof(*commit_graft) *
 152                                        commit_graft_alloc);
 153        }
 154        if (pos < commit_graft_nr)
 155                memmove(commit_graft + pos + 1,
 156                        commit_graft + pos,
 157                        (commit_graft_nr - pos - 1) *
 158                        sizeof(*commit_graft));
 159        commit_graft[pos] = graft;
 160        return 0;
 161}
 162
 163struct commit_graft *read_graft_line(char *buf, int len)
 164{
 165        /* The format is just "Commit Parent1 Parent2 ...\n" */
 166        int i;
 167        struct commit_graft *graft = NULL;
 168
 169        while (len && isspace(buf[len-1]))
 170                buf[--len] = '\0';
 171        if (buf[0] == '#' || buf[0] == '\0')
 172                return NULL;
 173        if ((len + 1) % 41)
 174                goto bad_graft_data;
 175        i = (len + 1) / 41 - 1;
 176        graft = xmalloc(sizeof(*graft) + 20 * i);
 177        graft->nr_parent = i;
 178        if (get_sha1_hex(buf, graft->sha1))
 179                goto bad_graft_data;
 180        for (i = 40; i < len; i += 41) {
 181                if (buf[i] != ' ')
 182                        goto bad_graft_data;
 183                if (get_sha1_hex(buf + i + 1, graft->parent[i/41]))
 184                        goto bad_graft_data;
 185        }
 186        return graft;
 187
 188bad_graft_data:
 189        error("bad graft data: %s", buf);
 190        free(graft);
 191        return NULL;
 192}
 193
 194static int read_graft_file(const char *graft_file)
 195{
 196        FILE *fp = fopen(graft_file, "r");
 197        char buf[1024];
 198        if (!fp)
 199                return -1;
 200        while (fgets(buf, sizeof(buf), fp)) {
 201                /* The format is just "Commit Parent1 Parent2 ...\n" */
 202                int len = strlen(buf);
 203                struct commit_graft *graft = read_graft_line(buf, len);
 204                if (!graft)
 205                        continue;
 206                if (register_commit_graft(graft, 1))
 207                        error("duplicate graft data: %s", buf);
 208        }
 209        fclose(fp);
 210        return 0;
 211}
 212
 213static void prepare_commit_graft(void)
 214{
 215        static int commit_graft_prepared;
 216        char *graft_file;
 217
 218        if (commit_graft_prepared)
 219                return;
 220        graft_file = get_graft_file();
 221        read_graft_file(graft_file);
 222        /* make sure shallows are read */
 223        is_repository_shallow();
 224        commit_graft_prepared = 1;
 225}
 226
 227struct commit_graft *lookup_commit_graft(const unsigned char *sha1)
 228{
 229        int pos;
 230        prepare_commit_graft();
 231        pos = commit_graft_pos(sha1);
 232        if (pos < 0)
 233                return NULL;
 234        return commit_graft[pos];
 235}
 236
 237int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
 238{
 239        int i, ret;
 240        for (i = ret = 0; i < commit_graft_nr && !ret; i++)
 241                ret = fn(commit_graft[i], cb_data);
 242        return ret;
 243}
 244
 245int unregister_shallow(const unsigned char *sha1)
 246{
 247        int pos = commit_graft_pos(sha1);
 248        if (pos < 0)
 249                return -1;
 250        if (pos + 1 < commit_graft_nr)
 251                memmove(commit_graft + pos, commit_graft + pos + 1,
 252                                sizeof(struct commit_graft *)
 253                                * (commit_graft_nr - pos - 1));
 254        commit_graft_nr--;
 255        return 0;
 256}
 257
 258int parse_commit_buffer(struct commit *item, const void *buffer, unsigned long size)
 259{
 260        const char *tail = buffer;
 261        const char *bufptr = buffer;
 262        unsigned char parent[20];
 263        struct commit_list **pptr;
 264        struct commit_graft *graft;
 265
 266        if (item->object.parsed)
 267                return 0;
 268        item->object.parsed = 1;
 269        tail += size;
 270        if (tail <= bufptr + 46 || memcmp(bufptr, "tree ", 5) || bufptr[45] != '\n')
 271                return error("bogus commit object %s", sha1_to_hex(item->object.sha1));
 272        if (get_sha1_hex(bufptr + 5, parent) < 0)
 273                return error("bad tree pointer in commit %s",
 274                             sha1_to_hex(item->object.sha1));
 275        item->tree = lookup_tree(parent);
 276        bufptr += 46; /* "tree " + "hex sha1" + "\n" */
 277        pptr = &item->parents;
 278
 279        graft = lookup_commit_graft(item->object.sha1);
 280        while (bufptr + 48 < tail && !memcmp(bufptr, "parent ", 7)) {
 281                struct commit *new_parent;
 282
 283                if (tail <= bufptr + 48 ||
 284                    get_sha1_hex(bufptr + 7, parent) ||
 285                    bufptr[47] != '\n')
 286                        return error("bad parents in commit %s", sha1_to_hex(item->object.sha1));
 287                bufptr += 48;
 288                /*
 289                 * The clone is shallow if nr_parent < 0, and we must
 290                 * not traverse its real parents even when we unhide them.
 291                 */
 292                if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
 293                        continue;
 294                new_parent = lookup_commit(parent);
 295                if (new_parent)
 296                        pptr = &commit_list_insert(new_parent, pptr)->next;
 297        }
 298        if (graft) {
 299                int i;
 300                struct commit *new_parent;
 301                for (i = 0; i < graft->nr_parent; i++) {
 302                        new_parent = lookup_commit(graft->parent[i]);
 303                        if (!new_parent)
 304                                continue;
 305                        pptr = &commit_list_insert(new_parent, pptr)->next;
 306                }
 307        }
 308        item->date = parse_commit_date(bufptr, tail);
 309
 310        return 0;
 311}
 312
 313int parse_commit(struct commit *item)
 314{
 315        enum object_type type;
 316        void *buffer;
 317        unsigned long size;
 318        int ret;
 319
 320        if (!item)
 321                return -1;
 322        if (item->object.parsed)
 323                return 0;
 324        buffer = read_sha1_file(item->object.sha1, &type, &size);
 325        if (!buffer)
 326                return error("Could not read %s",
 327                             sha1_to_hex(item->object.sha1));
 328        if (type != OBJ_COMMIT) {
 329                free(buffer);
 330                return error("Object %s not a commit",
 331                             sha1_to_hex(item->object.sha1));
 332        }
 333        ret = parse_commit_buffer(item, buffer, size);
 334        if (save_commit_buffer && !ret) {
 335                item->buffer = buffer;
 336                return 0;
 337        }
 338        free(buffer);
 339        return ret;
 340}
 341
 342int find_commit_subject(const char *commit_buffer, const char **subject)
 343{
 344        const char *eol;
 345        const char *p = commit_buffer;
 346
 347        while (*p && (*p != '\n' || p[1] != '\n'))
 348                p++;
 349        if (*p) {
 350                p += 2;
 351                for (eol = p; *eol && *eol != '\n'; eol++)
 352                        ; /* do nothing */
 353        } else
 354                eol = p;
 355
 356        *subject = p;
 357
 358        return eol - p;
 359}
 360
 361struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
 362{
 363        struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
 364        new_list->item = item;
 365        new_list->next = *list_p;
 366        *list_p = new_list;
 367        return new_list;
 368}
 369
 370unsigned commit_list_count(const struct commit_list *l)
 371{
 372        unsigned c = 0;
 373        for (; l; l = l->next )
 374                c++;
 375        return c;
 376}
 377
 378void free_commit_list(struct commit_list *list)
 379{
 380        while (list) {
 381                struct commit_list *temp = list;
 382                list = temp->next;
 383                free(temp);
 384        }
 385}
 386
 387struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
 388{
 389        struct commit_list **pp = list;
 390        struct commit_list *p;
 391        while ((p = *pp) != NULL) {
 392                if (p->item->date < item->date) {
 393                        break;
 394                }
 395                pp = &p->next;
 396        }
 397        return commit_list_insert(item, pp);
 398}
 399
 400static int commit_list_compare_by_date(const void *a, const void *b)
 401{
 402        unsigned long a_date = ((const struct commit_list *)a)->item->date;
 403        unsigned long b_date = ((const struct commit_list *)b)->item->date;
 404        if (a_date < b_date)
 405                return 1;
 406        if (a_date > b_date)
 407                return -1;
 408        return 0;
 409}
 410
 411static void *commit_list_get_next(const void *a)
 412{
 413        return ((const struct commit_list *)a)->next;
 414}
 415
 416static void commit_list_set_next(void *a, void *next)
 417{
 418        ((struct commit_list *)a)->next = next;
 419}
 420
 421void commit_list_sort_by_date(struct commit_list **list)
 422{
 423        *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
 424                                commit_list_compare_by_date);
 425}
 426
 427struct commit *pop_most_recent_commit(struct commit_list **list,
 428                                      unsigned int mark)
 429{
 430        struct commit *ret = (*list)->item;
 431        struct commit_list *parents = ret->parents;
 432        struct commit_list *old = *list;
 433
 434        *list = (*list)->next;
 435        free(old);
 436
 437        while (parents) {
 438                struct commit *commit = parents->item;
 439                if (!parse_commit(commit) && !(commit->object.flags & mark)) {
 440                        commit->object.flags |= mark;
 441                        commit_list_insert_by_date(commit, list);
 442                }
 443                parents = parents->next;
 444        }
 445        return ret;
 446}
 447
 448static void clear_commit_marks_1(struct commit_list **plist,
 449                                 struct commit *commit, unsigned int mark)
 450{
 451        while (commit) {
 452                struct commit_list *parents;
 453
 454                if (!(mark & commit->object.flags))
 455                        return;
 456
 457                commit->object.flags &= ~mark;
 458
 459                parents = commit->parents;
 460                if (!parents)
 461                        return;
 462
 463                while ((parents = parents->next))
 464                        commit_list_insert(parents->item, plist);
 465
 466                commit = commit->parents->item;
 467        }
 468}
 469
 470void clear_commit_marks(struct commit *commit, unsigned int mark)
 471{
 472        struct commit_list *list = NULL;
 473        commit_list_insert(commit, &list);
 474        while (list)
 475                clear_commit_marks_1(&list, pop_commit(&list), mark);
 476}
 477
 478void clear_commit_marks_for_object_array(struct object_array *a, unsigned mark)
 479{
 480        struct object *object;
 481        struct commit *commit;
 482        unsigned int i;
 483
 484        for (i = 0; i < a->nr; i++) {
 485                object = a->objects[i].item;
 486                commit = lookup_commit_reference_gently(object->sha1, 1);
 487                if (commit)
 488                        clear_commit_marks(commit, mark);
 489        }
 490}
 491
 492struct commit *pop_commit(struct commit_list **stack)
 493{
 494        struct commit_list *top = *stack;
 495        struct commit *item = top ? top->item : NULL;
 496
 497        if (top) {
 498                *stack = top->next;
 499                free(top);
 500        }
 501        return item;
 502}
 503
 504struct commit_slab_piece {
 505        int buf;
 506};
 507
 508struct commit_slab {
 509        int piece_size;
 510        int piece_count;
 511        struct commit_slab_piece **piece;
 512};
 513
 514static void slab_init(struct commit_slab *s)
 515{
 516        /* allocate ~512kB at once, allowing for malloc overhead */
 517        int size = (512*1024-32) / sizeof(struct commit_slab_piece);
 518
 519        s->piece_size = size;
 520        s->piece_count = 0;
 521        s->piece = NULL;
 522}
 523
 524static void slab_clear(struct commit_slab *s)
 525{
 526        int i;
 527
 528        for (i = 0; i < s->piece_count; i++)
 529                free(s->piece[i]);
 530        s->piece_count = 0;
 531        free(s->piece);
 532        s->piece = NULL;
 533}
 534
 535static inline struct commit_slab_piece *slab_at(struct commit_slab *s,
 536                                                const struct commit *c)
 537{
 538        int nth_piece, nth_slot;
 539
 540        nth_piece = c->index / s->piece_size;
 541        nth_slot = c->index % s->piece_size;
 542
 543        if (s->piece_count <= nth_piece) {
 544                int i;
 545
 546                s->piece = xrealloc(s->piece, (nth_piece + 1) * sizeof(s->piece));
 547                for (i = s->piece_count; i <= nth_piece; i++)
 548                        s->piece[i] = NULL;
 549                s->piece_count = nth_piece + 1;
 550        }
 551        if (!s->piece[nth_piece])
 552                s->piece[nth_piece] = xcalloc(s->piece_size, sizeof(**s->piece));
 553        return &s->piece[nth_piece][nth_slot];
 554}
 555
 556/*
 557 * Performs an in-place topological sort on the list supplied.
 558 */
 559void sort_in_topological_order(struct commit_list ** list, int lifo)
 560{
 561        struct commit_list *next, *orig = *list;
 562        struct commit_list *work, **insert;
 563        struct commit_list **pptr;
 564        struct commit_slab indegree;
 565
 566        if (!orig)
 567                return;
 568        *list = NULL;
 569
 570        slab_init(&indegree);
 571
 572        /* Mark them and clear the indegree */
 573        for (next = orig; next; next = next->next) {
 574                struct commit *commit = next->item;
 575                slab_at(&indegree, commit)->buf = 1;
 576        }
 577
 578        /* update the indegree */
 579        for (next = orig; next; next = next->next) {
 580                struct commit_list * parents = next->item->parents;
 581                while (parents) {
 582                        struct commit *parent = parents->item;
 583                        int *pi = &slab_at(&indegree, parent)->buf;
 584
 585                        if (*pi)
 586                                (*pi)++;
 587                        parents = parents->next;
 588                }
 589        }
 590
 591        /*
 592         * find the tips
 593         *
 594         * tips are nodes not reachable from any other node in the list
 595         *
 596         * the tips serve as a starting set for the work queue.
 597         */
 598        work = NULL;
 599        insert = &work;
 600        for (next = orig; next; next = next->next) {
 601                struct commit *commit = next->item;
 602
 603                if (slab_at(&indegree, commit)->buf == 1)
 604                        insert = &commit_list_insert(commit, insert)->next;
 605        }
 606
 607        /* process the list in topological order */
 608        if (!lifo)
 609                commit_list_sort_by_date(&work);
 610
 611        pptr = list;
 612        *list = NULL;
 613        while (work) {
 614                struct commit *commit;
 615                struct commit_list *parents, *work_item;
 616
 617                work_item = work;
 618                work = work_item->next;
 619                work_item->next = NULL;
 620
 621                commit = work_item->item;
 622                for (parents = commit->parents; parents ; parents = parents->next) {
 623                        struct commit *parent = parents->item;
 624                        int *pi = &slab_at(&indegree, parent)->buf;
 625
 626                        if (!*pi)
 627                                continue;
 628
 629                        /*
 630                         * parents are only enqueued for emission
 631                         * when all their children have been emitted thereby
 632                         * guaranteeing topological order.
 633                         */
 634                        if (--(*pi) == 1) {
 635                                if (!lifo)
 636                                        commit_list_insert_by_date(parent, &work);
 637                                else
 638                                        commit_list_insert(parent, &work);
 639                        }
 640                }
 641                /*
 642                 * work_item is a commit all of whose children
 643                 * have already been emitted. we can emit it now.
 644                 */
 645                slab_at(&indegree, commit)->buf = 0;
 646                *pptr = work_item;
 647                pptr = &work_item->next;
 648        }
 649
 650        slab_clear(&indegree);
 651}
 652
 653/* merge-base stuff */
 654
 655/* bits #0..15 in revision.h */
 656#define PARENT1         (1u<<16)
 657#define PARENT2         (1u<<17)
 658#define STALE           (1u<<18)
 659#define RESULT          (1u<<19)
 660
 661static const unsigned all_flags = (PARENT1 | PARENT2 | STALE | RESULT);
 662
 663static struct commit *interesting(struct commit_list *list)
 664{
 665        while (list) {
 666                struct commit *commit = list->item;
 667                list = list->next;
 668                if (commit->object.flags & STALE)
 669                        continue;
 670                return commit;
 671        }
 672        return NULL;
 673}
 674
 675/* all input commits in one and twos[] must have been parsed! */
 676static struct commit_list *paint_down_to_common(struct commit *one, int n, struct commit **twos)
 677{
 678        struct commit_list *list = NULL;
 679        struct commit_list *result = NULL;
 680        int i;
 681
 682        one->object.flags |= PARENT1;
 683        commit_list_insert_by_date(one, &list);
 684        if (!n)
 685                return list;
 686        for (i = 0; i < n; i++) {
 687                twos[i]->object.flags |= PARENT2;
 688                commit_list_insert_by_date(twos[i], &list);
 689        }
 690
 691        while (interesting(list)) {
 692                struct commit *commit;
 693                struct commit_list *parents;
 694                struct commit_list *next;
 695                int flags;
 696
 697                commit = list->item;
 698                next = list->next;
 699                free(list);
 700                list = next;
 701
 702                flags = commit->object.flags & (PARENT1 | PARENT2 | STALE);
 703                if (flags == (PARENT1 | PARENT2)) {
 704                        if (!(commit->object.flags & RESULT)) {
 705                                commit->object.flags |= RESULT;
 706                                commit_list_insert_by_date(commit, &result);
 707                        }
 708                        /* Mark parents of a found merge stale */
 709                        flags |= STALE;
 710                }
 711                parents = commit->parents;
 712                while (parents) {
 713                        struct commit *p = parents->item;
 714                        parents = parents->next;
 715                        if ((p->object.flags & flags) == flags)
 716                                continue;
 717                        if (parse_commit(p))
 718                                return NULL;
 719                        p->object.flags |= flags;
 720                        commit_list_insert_by_date(p, &list);
 721                }
 722        }
 723
 724        free_commit_list(list);
 725        return result;
 726}
 727
 728static struct commit_list *merge_bases_many(struct commit *one, int n, struct commit **twos)
 729{
 730        struct commit_list *list = NULL;
 731        struct commit_list *result = NULL;
 732        int i;
 733
 734        for (i = 0; i < n; i++) {
 735                if (one == twos[i])
 736                        /*
 737                         * We do not mark this even with RESULT so we do not
 738                         * have to clean it up.
 739                         */
 740                        return commit_list_insert(one, &result);
 741        }
 742
 743        if (parse_commit(one))
 744                return NULL;
 745        for (i = 0; i < n; i++) {
 746                if (parse_commit(twos[i]))
 747                        return NULL;
 748        }
 749
 750        list = paint_down_to_common(one, n, twos);
 751
 752        while (list) {
 753                struct commit_list *next = list->next;
 754                if (!(list->item->object.flags & STALE))
 755                        commit_list_insert_by_date(list->item, &result);
 756                free(list);
 757                list = next;
 758        }
 759        return result;
 760}
 761
 762struct commit_list *get_octopus_merge_bases(struct commit_list *in)
 763{
 764        struct commit_list *i, *j, *k, *ret = NULL;
 765        struct commit_list **pptr = &ret;
 766
 767        for (i = in; i; i = i->next) {
 768                if (!ret)
 769                        pptr = &commit_list_insert(i->item, pptr)->next;
 770                else {
 771                        struct commit_list *new = NULL, *end = NULL;
 772
 773                        for (j = ret; j; j = j->next) {
 774                                struct commit_list *bases;
 775                                bases = get_merge_bases(i->item, j->item, 1);
 776                                if (!new)
 777                                        new = bases;
 778                                else
 779                                        end->next = bases;
 780                                for (k = bases; k; k = k->next)
 781                                        end = k;
 782                        }
 783                        ret = new;
 784                }
 785        }
 786        return ret;
 787}
 788
 789static int remove_redundant(struct commit **array, int cnt)
 790{
 791        /*
 792         * Some commit in the array may be an ancestor of
 793         * another commit.  Move such commit to the end of
 794         * the array, and return the number of commits that
 795         * are independent from each other.
 796         */
 797        struct commit **work;
 798        unsigned char *redundant;
 799        int *filled_index;
 800        int i, j, filled;
 801
 802        work = xcalloc(cnt, sizeof(*work));
 803        redundant = xcalloc(cnt, 1);
 804        filled_index = xmalloc(sizeof(*filled_index) * (cnt - 1));
 805
 806        for (i = 0; i < cnt; i++)
 807                parse_commit(array[i]);
 808        for (i = 0; i < cnt; i++) {
 809                struct commit_list *common;
 810
 811                if (redundant[i])
 812                        continue;
 813                for (j = filled = 0; j < cnt; j++) {
 814                        if (i == j || redundant[j])
 815                                continue;
 816                        filled_index[filled] = j;
 817                        work[filled++] = array[j];
 818                }
 819                common = paint_down_to_common(array[i], filled, work);
 820                if (array[i]->object.flags & PARENT2)
 821                        redundant[i] = 1;
 822                for (j = 0; j < filled; j++)
 823                        if (work[j]->object.flags & PARENT1)
 824                                redundant[filled_index[j]] = 1;
 825                clear_commit_marks(array[i], all_flags);
 826                for (j = 0; j < filled; j++)
 827                        clear_commit_marks(work[j], all_flags);
 828                free_commit_list(common);
 829        }
 830
 831        /* Now collect the result */
 832        memcpy(work, array, sizeof(*array) * cnt);
 833        for (i = filled = 0; i < cnt; i++)
 834                if (!redundant[i])
 835                        array[filled++] = work[i];
 836        for (j = filled, i = 0; i < cnt; i++)
 837                if (redundant[i])
 838                        array[j++] = work[i];
 839        free(work);
 840        free(redundant);
 841        free(filled_index);
 842        return filled;
 843}
 844
 845struct commit_list *get_merge_bases_many(struct commit *one,
 846                                         int n,
 847                                         struct commit **twos,
 848                                         int cleanup)
 849{
 850        struct commit_list *list;
 851        struct commit **rslt;
 852        struct commit_list *result;
 853        int cnt, i;
 854
 855        result = merge_bases_many(one, n, twos);
 856        for (i = 0; i < n; i++) {
 857                if (one == twos[i])
 858                        return result;
 859        }
 860        if (!result || !result->next) {
 861                if (cleanup) {
 862                        clear_commit_marks(one, all_flags);
 863                        for (i = 0; i < n; i++)
 864                                clear_commit_marks(twos[i], all_flags);
 865                }
 866                return result;
 867        }
 868
 869        /* There are more than one */
 870        cnt = 0;
 871        list = result;
 872        while (list) {
 873                list = list->next;
 874                cnt++;
 875        }
 876        rslt = xcalloc(cnt, sizeof(*rslt));
 877        for (list = result, i = 0; list; list = list->next)
 878                rslt[i++] = list->item;
 879        free_commit_list(result);
 880
 881        clear_commit_marks(one, all_flags);
 882        for (i = 0; i < n; i++)
 883                clear_commit_marks(twos[i], all_flags);
 884
 885        cnt = remove_redundant(rslt, cnt);
 886        result = NULL;
 887        for (i = 0; i < cnt; i++)
 888                commit_list_insert_by_date(rslt[i], &result);
 889        free(rslt);
 890        return result;
 891}
 892
 893struct commit_list *get_merge_bases(struct commit *one, struct commit *two,
 894                                    int cleanup)
 895{
 896        return get_merge_bases_many(one, 1, &two, cleanup);
 897}
 898
 899/*
 900 * Is "commit" a descendant of one of the elements on the "with_commit" list?
 901 */
 902int is_descendant_of(struct commit *commit, struct commit_list *with_commit)
 903{
 904        if (!with_commit)
 905                return 1;
 906        while (with_commit) {
 907                struct commit *other;
 908
 909                other = with_commit->item;
 910                with_commit = with_commit->next;
 911                if (in_merge_bases(other, commit))
 912                        return 1;
 913        }
 914        return 0;
 915}
 916
 917/*
 918 * Is "commit" an ancestor of (i.e. reachable from) the "reference"?
 919 */
 920int in_merge_bases(struct commit *commit, struct commit *reference)
 921{
 922        struct commit_list *bases;
 923        int ret = 0;
 924
 925        if (parse_commit(commit) || parse_commit(reference))
 926                return ret;
 927
 928        bases = paint_down_to_common(commit, 1, &reference);
 929        if (commit->object.flags & PARENT2)
 930                ret = 1;
 931        clear_commit_marks(commit, all_flags);
 932        clear_commit_marks(reference, all_flags);
 933        free_commit_list(bases);
 934        return ret;
 935}
 936
 937struct commit_list *reduce_heads(struct commit_list *heads)
 938{
 939        struct commit_list *p;
 940        struct commit_list *result = NULL, **tail = &result;
 941        struct commit **array;
 942        int num_head, i;
 943
 944        if (!heads)
 945                return NULL;
 946
 947        /* Uniquify */
 948        for (p = heads; p; p = p->next)
 949                p->item->object.flags &= ~STALE;
 950        for (p = heads, num_head = 0; p; p = p->next) {
 951                if (p->item->object.flags & STALE)
 952                        continue;
 953                p->item->object.flags |= STALE;
 954                num_head++;
 955        }
 956        array = xcalloc(sizeof(*array), num_head);
 957        for (p = heads, i = 0; p; p = p->next) {
 958                if (p->item->object.flags & STALE) {
 959                        array[i++] = p->item;
 960                        p->item->object.flags &= ~STALE;
 961                }
 962        }
 963        num_head = remove_redundant(array, num_head);
 964        for (i = 0; i < num_head; i++)
 965                tail = &commit_list_insert(array[i], tail)->next;
 966        return result;
 967}
 968
 969static const char gpg_sig_header[] = "gpgsig";
 970static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
 971
 972static int do_sign_commit(struct strbuf *buf, const char *keyid)
 973{
 974        struct strbuf sig = STRBUF_INIT;
 975        int inspos, copypos;
 976
 977        /* find the end of the header */
 978        inspos = strstr(buf->buf, "\n\n") - buf->buf + 1;
 979
 980        if (!keyid || !*keyid)
 981                keyid = get_signing_key();
 982        if (sign_buffer(buf, &sig, keyid)) {
 983                strbuf_release(&sig);
 984                return -1;
 985        }
 986
 987        for (copypos = 0; sig.buf[copypos]; ) {
 988                const char *bol = sig.buf + copypos;
 989                const char *eol = strchrnul(bol, '\n');
 990                int len = (eol - bol) + !!*eol;
 991
 992                if (!copypos) {
 993                        strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
 994                        inspos += gpg_sig_header_len;
 995                }
 996                strbuf_insert(buf, inspos++, " ", 1);
 997                strbuf_insert(buf, inspos, bol, len);
 998                inspos += len;
 999                copypos += len;
1000        }
1001        strbuf_release(&sig);
1002        return 0;
1003}
1004
1005int parse_signed_commit(const unsigned char *sha1,
1006                        struct strbuf *payload, struct strbuf *signature)
1007{
1008        unsigned long size;
1009        enum object_type type;
1010        char *buffer = read_sha1_file(sha1, &type, &size);
1011        int in_signature, saw_signature = -1;
1012        char *line, *tail;
1013
1014        if (!buffer || type != OBJ_COMMIT)
1015                goto cleanup;
1016
1017        line = buffer;
1018        tail = buffer + size;
1019        in_signature = 0;
1020        saw_signature = 0;
1021        while (line < tail) {
1022                const char *sig = NULL;
1023                char *next = memchr(line, '\n', tail - line);
1024
1025                next = next ? next + 1 : tail;
1026                if (in_signature && line[0] == ' ')
1027                        sig = line + 1;
1028                else if (!prefixcmp(line, gpg_sig_header) &&
1029                         line[gpg_sig_header_len] == ' ')
1030                        sig = line + gpg_sig_header_len + 1;
1031                if (sig) {
1032                        strbuf_add(signature, sig, next - sig);
1033                        saw_signature = 1;
1034                        in_signature = 1;
1035                } else {
1036                        if (*line == '\n')
1037                                /* dump the whole remainder of the buffer */
1038                                next = tail;
1039                        strbuf_add(payload, line, next - line);
1040                        in_signature = 0;
1041                }
1042                line = next;
1043        }
1044 cleanup:
1045        free(buffer);
1046        return saw_signature;
1047}
1048
1049static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1050{
1051        struct merge_remote_desc *desc;
1052        struct commit_extra_header *mergetag;
1053        char *buf;
1054        unsigned long size, len;
1055        enum object_type type;
1056
1057        desc = merge_remote_util(parent);
1058        if (!desc || !desc->obj)
1059                return;
1060        buf = read_sha1_file(desc->obj->sha1, &type, &size);
1061        if (!buf || type != OBJ_TAG)
1062                goto free_return;
1063        len = parse_signature(buf, size);
1064        if (size == len)
1065                goto free_return;
1066        /*
1067         * We could verify this signature and either omit the tag when
1068         * it does not validate, but the integrator may not have the
1069         * public key of the signer of the tag he is merging, while a
1070         * later auditor may have it while auditing, so let's not run
1071         * verify-signed-buffer here for now...
1072         *
1073         * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1074         *      warn("warning: signed tag unverified.");
1075         */
1076        mergetag = xcalloc(1, sizeof(*mergetag));
1077        mergetag->key = xstrdup("mergetag");
1078        mergetag->value = buf;
1079        mergetag->len = size;
1080
1081        **tail = mergetag;
1082        *tail = &mergetag->next;
1083        return;
1084
1085free_return:
1086        free(buf);
1087}
1088
1089void append_merge_tag_headers(struct commit_list *parents,
1090                              struct commit_extra_header ***tail)
1091{
1092        while (parents) {
1093                struct commit *parent = parents->item;
1094                handle_signed_tag(parent, tail);
1095                parents = parents->next;
1096        }
1097}
1098
1099static void add_extra_header(struct strbuf *buffer,
1100                             struct commit_extra_header *extra)
1101{
1102        strbuf_addstr(buffer, extra->key);
1103        if (extra->len)
1104                strbuf_add_lines(buffer, " ", extra->value, extra->len);
1105        else
1106                strbuf_addch(buffer, '\n');
1107}
1108
1109struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1110                                                      const char **exclude)
1111{
1112        struct commit_extra_header *extra = NULL;
1113        unsigned long size;
1114        enum object_type type;
1115        char *buffer = read_sha1_file(commit->object.sha1, &type, &size);
1116        if (buffer && type == OBJ_COMMIT)
1117                extra = read_commit_extra_header_lines(buffer, size, exclude);
1118        free(buffer);
1119        return extra;
1120}
1121
1122static inline int standard_header_field(const char *field, size_t len)
1123{
1124        return ((len == 4 && !memcmp(field, "tree ", 5)) ||
1125                (len == 6 && !memcmp(field, "parent ", 7)) ||
1126                (len == 6 && !memcmp(field, "author ", 7)) ||
1127                (len == 9 && !memcmp(field, "committer ", 10)) ||
1128                (len == 8 && !memcmp(field, "encoding ", 9)));
1129}
1130
1131static int excluded_header_field(const char *field, size_t len, const char **exclude)
1132{
1133        if (!exclude)
1134                return 0;
1135
1136        while (*exclude) {
1137                size_t xlen = strlen(*exclude);
1138                if (len == xlen &&
1139                    !memcmp(field, *exclude, xlen) && field[xlen] == ' ')
1140                        return 1;
1141                exclude++;
1142        }
1143        return 0;
1144}
1145
1146static struct commit_extra_header *read_commit_extra_header_lines(
1147        const char *buffer, size_t size,
1148        const char **exclude)
1149{
1150        struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1151        const char *line, *next, *eof, *eob;
1152        struct strbuf buf = STRBUF_INIT;
1153
1154        for (line = buffer, eob = line + size;
1155             line < eob && *line != '\n';
1156             line = next) {
1157                next = memchr(line, '\n', eob - line);
1158                next = next ? next + 1 : eob;
1159                if (*line == ' ') {
1160                        /* continuation */
1161                        if (it)
1162                                strbuf_add(&buf, line + 1, next - (line + 1));
1163                        continue;
1164                }
1165                if (it)
1166                        it->value = strbuf_detach(&buf, &it->len);
1167                strbuf_reset(&buf);
1168                it = NULL;
1169
1170                eof = strchr(line, ' ');
1171                if (next <= eof)
1172                        eof = next;
1173
1174                if (standard_header_field(line, eof - line) ||
1175                    excluded_header_field(line, eof - line, exclude))
1176                        continue;
1177
1178                it = xcalloc(1, sizeof(*it));
1179                it->key = xmemdupz(line, eof-line);
1180                *tail = it;
1181                tail = &it->next;
1182                if (eof + 1 < next)
1183                        strbuf_add(&buf, eof + 1, next - (eof + 1));
1184        }
1185        if (it)
1186                it->value = strbuf_detach(&buf, &it->len);
1187        return extra;
1188}
1189
1190void free_commit_extra_headers(struct commit_extra_header *extra)
1191{
1192        while (extra) {
1193                struct commit_extra_header *next = extra->next;
1194                free(extra->key);
1195                free(extra->value);
1196                free(extra);
1197                extra = next;
1198        }
1199}
1200
1201int commit_tree(const struct strbuf *msg, unsigned char *tree,
1202                struct commit_list *parents, unsigned char *ret,
1203                const char *author, const char *sign_commit)
1204{
1205        struct commit_extra_header *extra = NULL, **tail = &extra;
1206        int result;
1207
1208        append_merge_tag_headers(parents, &tail);
1209        result = commit_tree_extended(msg, tree, parents, ret,
1210                                      author, sign_commit, extra);
1211        free_commit_extra_headers(extra);
1212        return result;
1213}
1214
1215static int find_invalid_utf8(const char *buf, int len)
1216{
1217        int offset = 0;
1218
1219        while (len) {
1220                unsigned char c = *buf++;
1221                int bytes, bad_offset;
1222
1223                len--;
1224                offset++;
1225
1226                /* Simple US-ASCII? No worries. */
1227                if (c < 0x80)
1228                        continue;
1229
1230                bad_offset = offset-1;
1231
1232                /*
1233                 * Count how many more high bits set: that's how
1234                 * many more bytes this sequence should have.
1235                 */
1236                bytes = 0;
1237                while (c & 0x40) {
1238                        c <<= 1;
1239                        bytes++;
1240                }
1241
1242                /* Must be between 1 and 5 more bytes */
1243                if (bytes < 1 || bytes > 5)
1244                        return bad_offset;
1245
1246                /* Do we *have* that many bytes? */
1247                if (len < bytes)
1248                        return bad_offset;
1249
1250                offset += bytes;
1251                len -= bytes;
1252
1253                /* And verify that they are good continuation bytes */
1254                do {
1255                        if ((*buf++ & 0xc0) != 0x80)
1256                                return bad_offset;
1257                } while (--bytes);
1258
1259                /* We could/should check the value and length here too */
1260        }
1261        return -1;
1262}
1263
1264/*
1265 * This verifies that the buffer is in proper utf8 format.
1266 *
1267 * If it isn't, it assumes any non-utf8 characters are Latin1,
1268 * and does the conversion.
1269 *
1270 * Fixme: we should probably also disallow overlong forms and
1271 * invalid characters. But we don't do that currently.
1272 */
1273static int verify_utf8(struct strbuf *buf)
1274{
1275        int ok = 1;
1276        long pos = 0;
1277
1278        for (;;) {
1279                int bad;
1280                unsigned char c;
1281                unsigned char replace[2];
1282
1283                bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1284                if (bad < 0)
1285                        return ok;
1286                pos += bad;
1287                ok = 0;
1288                c = buf->buf[pos];
1289                strbuf_remove(buf, pos, 1);
1290
1291                /* We know 'c' must be in the range 128-255 */
1292                replace[0] = 0xc0 + (c >> 6);
1293                replace[1] = 0x80 + (c & 0x3f);
1294                strbuf_insert(buf, pos, replace, 2);
1295                pos += 2;
1296        }
1297}
1298
1299static const char commit_utf8_warn[] =
1300"Warning: commit message did not conform to UTF-8.\n"
1301"You may want to amend it after fixing the message, or set the config\n"
1302"variable i18n.commitencoding to the encoding your project uses.\n";
1303
1304int commit_tree_extended(const struct strbuf *msg, unsigned char *tree,
1305                         struct commit_list *parents, unsigned char *ret,
1306                         const char *author, const char *sign_commit,
1307                         struct commit_extra_header *extra)
1308{
1309        int result;
1310        int encoding_is_utf8;
1311        struct strbuf buffer;
1312
1313        assert_sha1_type(tree, OBJ_TREE);
1314
1315        if (memchr(msg->buf, '\0', msg->len))
1316                return error("a NUL byte in commit log message not allowed.");
1317
1318        /* Not having i18n.commitencoding is the same as having utf-8 */
1319        encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1320
1321        strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1322        strbuf_addf(&buffer, "tree %s\n", sha1_to_hex(tree));
1323
1324        /*
1325         * NOTE! This ordering means that the same exact tree merged with a
1326         * different order of parents will be a _different_ changeset even
1327         * if everything else stays the same.
1328         */
1329        while (parents) {
1330                struct commit_list *next = parents->next;
1331                struct commit *parent = parents->item;
1332
1333                strbuf_addf(&buffer, "parent %s\n",
1334                            sha1_to_hex(parent->object.sha1));
1335                free(parents);
1336                parents = next;
1337        }
1338
1339        /* Person/date information */
1340        if (!author)
1341                author = git_author_info(IDENT_STRICT);
1342        strbuf_addf(&buffer, "author %s\n", author);
1343        strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1344        if (!encoding_is_utf8)
1345                strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1346
1347        while (extra) {
1348                add_extra_header(&buffer, extra);
1349                extra = extra->next;
1350        }
1351        strbuf_addch(&buffer, '\n');
1352
1353        /* And add the comment */
1354        strbuf_addbuf(&buffer, msg);
1355
1356        /* And check the encoding */
1357        if (encoding_is_utf8 && !verify_utf8(&buffer))
1358                fprintf(stderr, commit_utf8_warn);
1359
1360        if (sign_commit && do_sign_commit(&buffer, sign_commit))
1361                return -1;
1362
1363        result = write_sha1_file(buffer.buf, buffer.len, commit_type, ret);
1364        strbuf_release(&buffer);
1365        return result;
1366}
1367
1368struct commit *get_merge_parent(const char *name)
1369{
1370        struct object *obj;
1371        struct commit *commit;
1372        unsigned char sha1[20];
1373        if (get_sha1(name, sha1))
1374                return NULL;
1375        obj = parse_object(sha1);
1376        commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1377        if (commit && !commit->util) {
1378                struct merge_remote_desc *desc;
1379                desc = xmalloc(sizeof(*desc));
1380                desc->obj = obj;
1381                desc->name = strdup(name);
1382                commit->util = desc;
1383        }
1384        return commit;
1385}
1386
1387/*
1388 * Append a commit to the end of the commit_list.
1389 *
1390 * next starts by pointing to the variable that holds the head of an
1391 * empty commit_list, and is updated to point to the "next" field of
1392 * the last item on the list as new commits are appended.
1393 *
1394 * Usage example:
1395 *
1396 *     struct commit_list *list;
1397 *     struct commit_list **next = &list;
1398 *
1399 *     next = commit_list_append(c1, next);
1400 *     next = commit_list_append(c2, next);
1401 *     assert(commit_list_count(list) == 2);
1402 *     return list;
1403 */
1404struct commit_list **commit_list_append(struct commit *commit,
1405                                        struct commit_list **next)
1406{
1407        struct commit_list *new = xmalloc(sizeof(struct commit_list));
1408        new->item = commit;
1409        *next = new;
1410        new->next = NULL;
1411        return &new->next;
1412}
1413
1414void print_commit_list(struct commit_list *list,
1415                       const char *format_cur,
1416                       const char *format_last)
1417{
1418        for ( ; list; list = list->next) {
1419                const char *format = list->next ? format_cur : format_last;
1420                printf(format, sha1_to_hex(list->item->object.sha1));
1421        }
1422}