revision.con commit limit_list: avoid quadratic behavior from still_interesting (b6e8a3b)
   1#include "cache.h"
   2#include "tag.h"
   3#include "blob.h"
   4#include "tree.h"
   5#include "commit.h"
   6#include "diff.h"
   7#include "refs.h"
   8#include "revision.h"
   9#include "graph.h"
  10#include "grep.h"
  11#include "reflog-walk.h"
  12#include "patch-ids.h"
  13#include "decorate.h"
  14#include "log-tree.h"
  15#include "string-list.h"
  16#include "line-log.h"
  17#include "mailmap.h"
  18#include "commit-slab.h"
  19#include "dir.h"
  20
  21volatile show_early_output_fn_t show_early_output;
  22
  23char *path_name(const struct name_path *path, const char *name)
  24{
  25        const struct name_path *p;
  26        char *n, *m;
  27        int nlen = strlen(name);
  28        int len = nlen + 1;
  29
  30        for (p = path; p; p = p->up) {
  31                if (p->elem_len)
  32                        len += p->elem_len + 1;
  33        }
  34        n = xmalloc(len);
  35        m = n + len - (nlen + 1);
  36        strcpy(m, name);
  37        for (p = path; p; p = p->up) {
  38                if (p->elem_len) {
  39                        m -= p->elem_len + 1;
  40                        memcpy(m, p->elem, p->elem_len);
  41                        m[p->elem_len] = '/';
  42                }
  43        }
  44        return n;
  45}
  46
  47static int show_path_component_truncated(FILE *out, const char *name, int len)
  48{
  49        int cnt;
  50        for (cnt = 0; cnt < len; cnt++) {
  51                int ch = name[cnt];
  52                if (!ch || ch == '\n')
  53                        return -1;
  54                fputc(ch, out);
  55        }
  56        return len;
  57}
  58
  59static int show_path_truncated(FILE *out, const struct name_path *path)
  60{
  61        int emitted, ours;
  62
  63        if (!path)
  64                return 0;
  65        emitted = show_path_truncated(out, path->up);
  66        if (emitted < 0)
  67                return emitted;
  68        if (emitted)
  69                fputc('/', out);
  70        ours = show_path_component_truncated(out, path->elem, path->elem_len);
  71        if (ours < 0)
  72                return ours;
  73        return ours || emitted;
  74}
  75
  76void show_object_with_name(FILE *out, struct object *obj,
  77                           const struct name_path *path, const char *component)
  78{
  79        struct name_path leaf;
  80        leaf.up = (struct name_path *)path;
  81        leaf.elem = component;
  82        leaf.elem_len = strlen(component);
  83
  84        fprintf(out, "%s ", sha1_to_hex(obj->sha1));
  85        show_path_truncated(out, &leaf);
  86        fputc('\n', out);
  87}
  88
  89void add_object(struct object *obj,
  90                struct object_array *p,
  91                struct name_path *path,
  92                const char *name)
  93{
  94        char *pn = path_name(path, name);
  95        add_object_array(obj, pn, p);
  96        free(pn);
  97}
  98
  99static void mark_blob_uninteresting(struct blob *blob)
 100{
 101        if (!blob)
 102                return;
 103        if (blob->object.flags & UNINTERESTING)
 104                return;
 105        blob->object.flags |= UNINTERESTING;
 106}
 107
 108static void mark_tree_contents_uninteresting(struct tree *tree)
 109{
 110        struct tree_desc desc;
 111        struct name_entry entry;
 112        struct object *obj = &tree->object;
 113
 114        if (!has_sha1_file(obj->sha1))
 115                return;
 116        if (parse_tree(tree) < 0)
 117                die("bad tree %s", sha1_to_hex(obj->sha1));
 118
 119        init_tree_desc(&desc, tree->buffer, tree->size);
 120        while (tree_entry(&desc, &entry)) {
 121                switch (object_type(entry.mode)) {
 122                case OBJ_TREE:
 123                        mark_tree_uninteresting(lookup_tree(entry.sha1));
 124                        break;
 125                case OBJ_BLOB:
 126                        mark_blob_uninteresting(lookup_blob(entry.sha1));
 127                        break;
 128                default:
 129                        /* Subproject commit - not in this repository */
 130                        break;
 131                }
 132        }
 133
 134        /*
 135         * We don't care about the tree any more
 136         * after it has been marked uninteresting.
 137         */
 138        free_tree_buffer(tree);
 139}
 140
 141void mark_tree_uninteresting(struct tree *tree)
 142{
 143        struct object *obj = &tree->object;
 144
 145        if (!tree)
 146                return;
 147        if (obj->flags & UNINTERESTING)
 148                return;
 149        obj->flags |= UNINTERESTING;
 150        mark_tree_contents_uninteresting(tree);
 151}
 152
 153void mark_parents_uninteresting(struct commit *commit)
 154{
 155        struct commit_list *parents = NULL, *l;
 156
 157        for (l = commit->parents; l; l = l->next)
 158                commit_list_insert(l->item, &parents);
 159
 160        while (parents) {
 161                struct commit *commit = parents->item;
 162                l = parents;
 163                parents = parents->next;
 164                free(l);
 165
 166                while (commit) {
 167                        /*
 168                         * A missing commit is ok iff its parent is marked
 169                         * uninteresting.
 170                         *
 171                         * We just mark such a thing parsed, so that when
 172                         * it is popped next time around, we won't be trying
 173                         * to parse it and get an error.
 174                         */
 175                        if (!has_sha1_file(commit->object.sha1))
 176                                commit->object.parsed = 1;
 177
 178                        if (commit->object.flags & UNINTERESTING)
 179                                break;
 180
 181                        commit->object.flags |= UNINTERESTING;
 182
 183                        /*
 184                         * Normally we haven't parsed the parent
 185                         * yet, so we won't have a parent of a parent
 186                         * here. However, it may turn out that we've
 187                         * reached this commit some other way (where it
 188                         * wasn't uninteresting), in which case we need
 189                         * to mark its parents recursively too..
 190                         */
 191                        if (!commit->parents)
 192                                break;
 193
 194                        for (l = commit->parents->next; l; l = l->next)
 195                                commit_list_insert(l->item, &parents);
 196                        commit = commit->parents->item;
 197                }
 198        }
 199}
 200
 201static void add_pending_object_with_mode(struct rev_info *revs,
 202                                         struct object *obj,
 203                                         const char *name, unsigned mode)
 204{
 205        if (!obj)
 206                return;
 207        if (revs->no_walk && (obj->flags & UNINTERESTING))
 208                revs->no_walk = 0;
 209        if (revs->reflog_info && obj->type == OBJ_COMMIT) {
 210                struct strbuf buf = STRBUF_INIT;
 211                int len = interpret_branch_name(name, 0, &buf);
 212                int st;
 213
 214                if (0 < len && name[len] && buf.len)
 215                        strbuf_addstr(&buf, name + len);
 216                st = add_reflog_for_walk(revs->reflog_info,
 217                                         (struct commit *)obj,
 218                                         buf.buf[0] ? buf.buf: name);
 219                strbuf_release(&buf);
 220                if (st)
 221                        return;
 222        }
 223        add_object_array_with_mode(obj, name, &revs->pending, mode);
 224}
 225
 226void add_pending_object(struct rev_info *revs,
 227                        struct object *obj, const char *name)
 228{
 229        add_pending_object_with_mode(revs, obj, name, S_IFINVALID);
 230}
 231
 232void add_head_to_pending(struct rev_info *revs)
 233{
 234        unsigned char sha1[20];
 235        struct object *obj;
 236        if (get_sha1("HEAD", sha1))
 237                return;
 238        obj = parse_object(sha1);
 239        if (!obj)
 240                return;
 241        add_pending_object(revs, obj, "HEAD");
 242}
 243
 244static struct object *get_reference(struct rev_info *revs, const char *name,
 245                                    const unsigned char *sha1,
 246                                    unsigned int flags)
 247{
 248        struct object *object;
 249
 250        object = parse_object(sha1);
 251        if (!object) {
 252                if (revs->ignore_missing)
 253                        return object;
 254                die("bad object %s", name);
 255        }
 256        object->flags |= flags;
 257        return object;
 258}
 259
 260void add_pending_sha1(struct rev_info *revs, const char *name,
 261                      const unsigned char *sha1, unsigned int flags)
 262{
 263        struct object *object = get_reference(revs, name, sha1, flags);
 264        add_pending_object(revs, object, name);
 265}
 266
 267static struct commit *handle_commit(struct rev_info *revs,
 268                                    struct object *object, const char *name)
 269{
 270        unsigned long flags = object->flags;
 271
 272        /*
 273         * Tag object? Look what it points to..
 274         */
 275        while (object->type == OBJ_TAG) {
 276                struct tag *tag = (struct tag *) object;
 277                if (revs->tag_objects && !(flags & UNINTERESTING))
 278                        add_pending_object(revs, object, tag->tag);
 279                if (!tag->tagged)
 280                        die("bad tag");
 281                object = parse_object(tag->tagged->sha1);
 282                if (!object) {
 283                        if (flags & UNINTERESTING)
 284                                return NULL;
 285                        die("bad object %s", sha1_to_hex(tag->tagged->sha1));
 286                }
 287                object->flags |= flags;
 288        }
 289
 290        /*
 291         * Commit object? Just return it, we'll do all the complex
 292         * reachability crud.
 293         */
 294        if (object->type == OBJ_COMMIT) {
 295                struct commit *commit = (struct commit *)object;
 296                if (parse_commit(commit) < 0)
 297                        die("unable to parse commit %s", name);
 298                if (flags & UNINTERESTING) {
 299                        mark_parents_uninteresting(commit);
 300                        revs->limited = 1;
 301                }
 302                if (revs->show_source && !commit->util)
 303                        commit->util = (void *) name;
 304                return commit;
 305        }
 306
 307        /*
 308         * Tree object? Either mark it uninteresting, or add it
 309         * to the list of objects to look at later..
 310         */
 311        if (object->type == OBJ_TREE) {
 312                struct tree *tree = (struct tree *)object;
 313                if (!revs->tree_objects)
 314                        return NULL;
 315                if (flags & UNINTERESTING) {
 316                        mark_tree_contents_uninteresting(tree);
 317                        return NULL;
 318                }
 319                add_pending_object(revs, object, "");
 320                return NULL;
 321        }
 322
 323        /*
 324         * Blob object? You know the drill by now..
 325         */
 326        if (object->type == OBJ_BLOB) {
 327                if (!revs->blob_objects)
 328                        return NULL;
 329                if (flags & UNINTERESTING)
 330                        return NULL;
 331                add_pending_object(revs, object, "");
 332                return NULL;
 333        }
 334        die("%s is unknown object", name);
 335}
 336
 337static int everybody_uninteresting(struct commit_list *orig,
 338                                   struct commit **interesting_cache)
 339{
 340        struct commit_list *list = orig;
 341
 342        if (*interesting_cache) {
 343                struct commit *commit = *interesting_cache;
 344                if (!(commit->object.flags & UNINTERESTING))
 345                        return 0;
 346        }
 347
 348        while (list) {
 349                struct commit *commit = list->item;
 350                list = list->next;
 351                if (commit->object.flags & UNINTERESTING)
 352                        continue;
 353                if (interesting_cache)
 354                        *interesting_cache = commit;
 355                return 0;
 356        }
 357        return 1;
 358}
 359
 360/*
 361 * A definition of "relevant" commit that we can use to simplify limited graphs
 362 * by eliminating side branches.
 363 *
 364 * A "relevant" commit is one that is !UNINTERESTING (ie we are including it
 365 * in our list), or that is a specified BOTTOM commit. Then after computing
 366 * a limited list, during processing we can generally ignore boundary merges
 367 * coming from outside the graph, (ie from irrelevant parents), and treat
 368 * those merges as if they were single-parent. TREESAME is defined to consider
 369 * only relevant parents, if any. If we are TREESAME to our on-graph parents,
 370 * we don't care if we were !TREESAME to non-graph parents.
 371 *
 372 * Treating bottom commits as relevant ensures that a limited graph's
 373 * connection to the actual bottom commit is not viewed as a side branch, but
 374 * treated as part of the graph. For example:
 375 *
 376 *   ....Z...A---X---o---o---B
 377 *        .     /
 378 *         W---Y
 379 *
 380 * When computing "A..B", the A-X connection is at least as important as
 381 * Y-X, despite A being flagged UNINTERESTING.
 382 *
 383 * And when computing --ancestry-path "A..B", the A-X connection is more
 384 * important than Y-X, despite both A and Y being flagged UNINTERESTING.
 385 */
 386static inline int relevant_commit(struct commit *commit)
 387{
 388        return (commit->object.flags & (UNINTERESTING | BOTTOM)) != UNINTERESTING;
 389}
 390
 391/*
 392 * Return a single relevant commit from a parent list. If we are a TREESAME
 393 * commit, and this selects one of our parents, then we can safely simplify to
 394 * that parent.
 395 */
 396static struct commit *one_relevant_parent(const struct rev_info *revs,
 397                                          struct commit_list *orig)
 398{
 399        struct commit_list *list = orig;
 400        struct commit *relevant = NULL;
 401
 402        if (!orig)
 403                return NULL;
 404
 405        /*
 406         * For 1-parent commits, or if first-parent-only, then return that
 407         * first parent (even if not "relevant" by the above definition).
 408         * TREESAME will have been set purely on that parent.
 409         */
 410        if (revs->first_parent_only || !orig->next)
 411                return orig->item;
 412
 413        /*
 414         * For multi-parent commits, identify a sole relevant parent, if any.
 415         * If we have only one relevant parent, then TREESAME will be set purely
 416         * with regard to that parent, and we can simplify accordingly.
 417         *
 418         * If we have more than one relevant parent, or no relevant parents
 419         * (and multiple irrelevant ones), then we can't select a parent here
 420         * and return NULL.
 421         */
 422        while (list) {
 423                struct commit *commit = list->item;
 424                list = list->next;
 425                if (relevant_commit(commit)) {
 426                        if (relevant)
 427                                return NULL;
 428                        relevant = commit;
 429                }
 430        }
 431        return relevant;
 432}
 433
 434/*
 435 * The goal is to get REV_TREE_NEW as the result only if the
 436 * diff consists of all '+' (and no other changes), REV_TREE_OLD
 437 * if the whole diff is removal of old data, and otherwise
 438 * REV_TREE_DIFFERENT (of course if the trees are the same we
 439 * want REV_TREE_SAME).
 440 * That means that once we get to REV_TREE_DIFFERENT, we do not
 441 * have to look any further.
 442 */
 443static int tree_difference = REV_TREE_SAME;
 444
 445static void file_add_remove(struct diff_options *options,
 446                    int addremove, unsigned mode,
 447                    const unsigned char *sha1,
 448                    int sha1_valid,
 449                    const char *fullpath, unsigned dirty_submodule)
 450{
 451        int diff = addremove == '+' ? REV_TREE_NEW : REV_TREE_OLD;
 452
 453        tree_difference |= diff;
 454        if (tree_difference == REV_TREE_DIFFERENT)
 455                DIFF_OPT_SET(options, HAS_CHANGES);
 456}
 457
 458static void file_change(struct diff_options *options,
 459                 unsigned old_mode, unsigned new_mode,
 460                 const unsigned char *old_sha1,
 461                 const unsigned char *new_sha1,
 462                 int old_sha1_valid, int new_sha1_valid,
 463                 const char *fullpath,
 464                 unsigned old_dirty_submodule, unsigned new_dirty_submodule)
 465{
 466        tree_difference = REV_TREE_DIFFERENT;
 467        DIFF_OPT_SET(options, HAS_CHANGES);
 468}
 469
 470static int rev_compare_tree(struct rev_info *revs,
 471                            struct commit *parent, struct commit *commit)
 472{
 473        struct tree *t1 = parent->tree;
 474        struct tree *t2 = commit->tree;
 475
 476        if (!t1)
 477                return REV_TREE_NEW;
 478        if (!t2)
 479                return REV_TREE_OLD;
 480
 481        if (revs->simplify_by_decoration) {
 482                /*
 483                 * If we are simplifying by decoration, then the commit
 484                 * is worth showing if it has a tag pointing at it.
 485                 */
 486                if (lookup_decoration(&name_decoration, &commit->object))
 487                        return REV_TREE_DIFFERENT;
 488                /*
 489                 * A commit that is not pointed by a tag is uninteresting
 490                 * if we are not limited by path.  This means that you will
 491                 * see the usual "commits that touch the paths" plus any
 492                 * tagged commit by specifying both --simplify-by-decoration
 493                 * and pathspec.
 494                 */
 495                if (!revs->prune_data.nr)
 496                        return REV_TREE_SAME;
 497        }
 498
 499        tree_difference = REV_TREE_SAME;
 500        DIFF_OPT_CLR(&revs->pruning, HAS_CHANGES);
 501        if (diff_tree_sha1(t1->object.sha1, t2->object.sha1, "",
 502                           &revs->pruning) < 0)
 503                return REV_TREE_DIFFERENT;
 504        return tree_difference;
 505}
 506
 507static int rev_same_tree_as_empty(struct rev_info *revs, struct commit *commit)
 508{
 509        int retval;
 510        struct tree *t1 = commit->tree;
 511
 512        if (!t1)
 513                return 0;
 514
 515        tree_difference = REV_TREE_SAME;
 516        DIFF_OPT_CLR(&revs->pruning, HAS_CHANGES);
 517        retval = diff_tree_sha1(NULL, t1->object.sha1, "", &revs->pruning);
 518
 519        return retval >= 0 && (tree_difference == REV_TREE_SAME);
 520}
 521
 522struct treesame_state {
 523        unsigned int nparents;
 524        unsigned char treesame[FLEX_ARRAY];
 525};
 526
 527static struct treesame_state *initialise_treesame(struct rev_info *revs, struct commit *commit)
 528{
 529        unsigned n = commit_list_count(commit->parents);
 530        struct treesame_state *st = xcalloc(1, sizeof(*st) + n);
 531        st->nparents = n;
 532        add_decoration(&revs->treesame, &commit->object, st);
 533        return st;
 534}
 535
 536/*
 537 * Must be called immediately after removing the nth_parent from a commit's
 538 * parent list, if we are maintaining the per-parent treesame[] decoration.
 539 * This does not recalculate the master TREESAME flag - update_treesame()
 540 * should be called to update it after a sequence of treesame[] modifications
 541 * that may have affected it.
 542 */
 543static int compact_treesame(struct rev_info *revs, struct commit *commit, unsigned nth_parent)
 544{
 545        struct treesame_state *st;
 546        int old_same;
 547
 548        if (!commit->parents) {
 549                /*
 550                 * Have just removed the only parent from a non-merge.
 551                 * Different handling, as we lack decoration.
 552                 */
 553                if (nth_parent != 0)
 554                        die("compact_treesame %u", nth_parent);
 555                old_same = !!(commit->object.flags & TREESAME);
 556                if (rev_same_tree_as_empty(revs, commit))
 557                        commit->object.flags |= TREESAME;
 558                else
 559                        commit->object.flags &= ~TREESAME;
 560                return old_same;
 561        }
 562
 563        st = lookup_decoration(&revs->treesame, &commit->object);
 564        if (!st || nth_parent >= st->nparents)
 565                die("compact_treesame %u", nth_parent);
 566
 567        old_same = st->treesame[nth_parent];
 568        memmove(st->treesame + nth_parent,
 569                st->treesame + nth_parent + 1,
 570                st->nparents - nth_parent - 1);
 571
 572        /*
 573         * If we've just become a non-merge commit, update TREESAME
 574         * immediately, and remove the no-longer-needed decoration.
 575         * If still a merge, defer update until update_treesame().
 576         */
 577        if (--st->nparents == 1) {
 578                if (commit->parents->next)
 579                        die("compact_treesame parents mismatch");
 580                if (st->treesame[0] && revs->dense)
 581                        commit->object.flags |= TREESAME;
 582                else
 583                        commit->object.flags &= ~TREESAME;
 584                free(add_decoration(&revs->treesame, &commit->object, NULL));
 585        }
 586
 587        return old_same;
 588}
 589
 590static unsigned update_treesame(struct rev_info *revs, struct commit *commit)
 591{
 592        if (commit->parents && commit->parents->next) {
 593                unsigned n;
 594                struct treesame_state *st;
 595                struct commit_list *p;
 596                unsigned relevant_parents;
 597                unsigned relevant_change, irrelevant_change;
 598
 599                st = lookup_decoration(&revs->treesame, &commit->object);
 600                if (!st)
 601                        die("update_treesame %s", sha1_to_hex(commit->object.sha1));
 602                relevant_parents = 0;
 603                relevant_change = irrelevant_change = 0;
 604                for (p = commit->parents, n = 0; p; n++, p = p->next) {
 605                        if (relevant_commit(p->item)) {
 606                                relevant_change |= !st->treesame[n];
 607                                relevant_parents++;
 608                        } else
 609                                irrelevant_change |= !st->treesame[n];
 610                }
 611                if (relevant_parents ? relevant_change : irrelevant_change)
 612                        commit->object.flags &= ~TREESAME;
 613                else
 614                        commit->object.flags |= TREESAME;
 615        }
 616
 617        return commit->object.flags & TREESAME;
 618}
 619
 620static inline int limiting_can_increase_treesame(const struct rev_info *revs)
 621{
 622        /*
 623         * TREESAME is irrelevant unless prune && dense;
 624         * if simplify_history is set, we can't have a mixture of TREESAME and
 625         *    !TREESAME INTERESTING parents (and we don't have treesame[]
 626         *    decoration anyway);
 627         * if first_parent_only is set, then the TREESAME flag is locked
 628         *    against the first parent (and again we lack treesame[] decoration).
 629         */
 630        return revs->prune && revs->dense &&
 631               !revs->simplify_history &&
 632               !revs->first_parent_only;
 633}
 634
 635static void try_to_simplify_commit(struct rev_info *revs, struct commit *commit)
 636{
 637        struct commit_list **pp, *parent;
 638        struct treesame_state *ts = NULL;
 639        int relevant_change = 0, irrelevant_change = 0;
 640        int relevant_parents, nth_parent;
 641
 642        /*
 643         * If we don't do pruning, everything is interesting
 644         */
 645        if (!revs->prune)
 646                return;
 647
 648        if (!commit->tree)
 649                return;
 650
 651        if (!commit->parents) {
 652                if (rev_same_tree_as_empty(revs, commit))
 653                        commit->object.flags |= TREESAME;
 654                return;
 655        }
 656
 657        /*
 658         * Normal non-merge commit? If we don't want to make the
 659         * history dense, we consider it always to be a change..
 660         */
 661        if (!revs->dense && !commit->parents->next)
 662                return;
 663
 664        for (pp = &commit->parents, nth_parent = 0, relevant_parents = 0;
 665             (parent = *pp) != NULL;
 666             pp = &parent->next, nth_parent++) {
 667                struct commit *p = parent->item;
 668                if (relevant_commit(p))
 669                        relevant_parents++;
 670
 671                if (nth_parent == 1) {
 672                        /*
 673                         * This our second loop iteration - so we now know
 674                         * we're dealing with a merge.
 675                         *
 676                         * Do not compare with later parents when we care only about
 677                         * the first parent chain, in order to avoid derailing the
 678                         * traversal to follow a side branch that brought everything
 679                         * in the path we are limited to by the pathspec.
 680                         */
 681                        if (revs->first_parent_only)
 682                                break;
 683                        /*
 684                         * If this will remain a potentially-simplifiable
 685                         * merge, remember per-parent treesame if needed.
 686                         * Initialise the array with the comparison from our
 687                         * first iteration.
 688                         */
 689                        if (revs->treesame.name &&
 690                            !revs->simplify_history &&
 691                            !(commit->object.flags & UNINTERESTING)) {
 692                                ts = initialise_treesame(revs, commit);
 693                                if (!(irrelevant_change || relevant_change))
 694                                        ts->treesame[0] = 1;
 695                        }
 696                }
 697                if (parse_commit(p) < 0)
 698                        die("cannot simplify commit %s (because of %s)",
 699                            sha1_to_hex(commit->object.sha1),
 700                            sha1_to_hex(p->object.sha1));
 701                switch (rev_compare_tree(revs, p, commit)) {
 702                case REV_TREE_SAME:
 703                        if (!revs->simplify_history || !relevant_commit(p)) {
 704                                /* Even if a merge with an uninteresting
 705                                 * side branch brought the entire change
 706                                 * we are interested in, we do not want
 707                                 * to lose the other branches of this
 708                                 * merge, so we just keep going.
 709                                 */
 710                                if (ts)
 711                                        ts->treesame[nth_parent] = 1;
 712                                continue;
 713                        }
 714                        parent->next = NULL;
 715                        commit->parents = parent;
 716                        commit->object.flags |= TREESAME;
 717                        return;
 718
 719                case REV_TREE_NEW:
 720                        if (revs->remove_empty_trees &&
 721                            rev_same_tree_as_empty(revs, p)) {
 722                                /* We are adding all the specified
 723                                 * paths from this parent, so the
 724                                 * history beyond this parent is not
 725                                 * interesting.  Remove its parents
 726                                 * (they are grandparents for us).
 727                                 * IOW, we pretend this parent is a
 728                                 * "root" commit.
 729                                 */
 730                                if (parse_commit(p) < 0)
 731                                        die("cannot simplify commit %s (invalid %s)",
 732                                            sha1_to_hex(commit->object.sha1),
 733                                            sha1_to_hex(p->object.sha1));
 734                                p->parents = NULL;
 735                        }
 736                /* fallthrough */
 737                case REV_TREE_OLD:
 738                case REV_TREE_DIFFERENT:
 739                        if (relevant_commit(p))
 740                                relevant_change = 1;
 741                        else
 742                                irrelevant_change = 1;
 743                        continue;
 744                }
 745                die("bad tree compare for commit %s", sha1_to_hex(commit->object.sha1));
 746        }
 747
 748        /*
 749         * TREESAME is straightforward for single-parent commits. For merge
 750         * commits, it is most useful to define it so that "irrelevant"
 751         * parents cannot make us !TREESAME - if we have any relevant
 752         * parents, then we only consider TREESAMEness with respect to them,
 753         * allowing irrelevant merges from uninteresting branches to be
 754         * simplified away. Only if we have only irrelevant parents do we
 755         * base TREESAME on them. Note that this logic is replicated in
 756         * update_treesame, which should be kept in sync.
 757         */
 758        if (relevant_parents ? !relevant_change : !irrelevant_change)
 759                commit->object.flags |= TREESAME;
 760}
 761
 762static void commit_list_insert_by_date_cached(struct commit *p, struct commit_list **head,
 763                    struct commit_list *cached_base, struct commit_list **cache)
 764{
 765        struct commit_list *new_entry;
 766
 767        if (cached_base && p->date < cached_base->item->date)
 768                new_entry = commit_list_insert_by_date(p, &cached_base->next);
 769        else
 770                new_entry = commit_list_insert_by_date(p, head);
 771
 772        if (cache && (!*cache || p->date < (*cache)->item->date))
 773                *cache = new_entry;
 774}
 775
 776static int add_parents_to_list(struct rev_info *revs, struct commit *commit,
 777                    struct commit_list **list, struct commit_list **cache_ptr)
 778{
 779        struct commit_list *parent = commit->parents;
 780        unsigned left_flag;
 781        struct commit_list *cached_base = cache_ptr ? *cache_ptr : NULL;
 782
 783        if (commit->object.flags & ADDED)
 784                return 0;
 785        commit->object.flags |= ADDED;
 786
 787        if (revs->include_check &&
 788            !revs->include_check(commit, revs->include_check_data))
 789                return 0;
 790
 791        /*
 792         * If the commit is uninteresting, don't try to
 793         * prune parents - we want the maximal uninteresting
 794         * set.
 795         *
 796         * Normally we haven't parsed the parent
 797         * yet, so we won't have a parent of a parent
 798         * here. However, it may turn out that we've
 799         * reached this commit some other way (where it
 800         * wasn't uninteresting), in which case we need
 801         * to mark its parents recursively too..
 802         */
 803        if (commit->object.flags & UNINTERESTING) {
 804                while (parent) {
 805                        struct commit *p = parent->item;
 806                        parent = parent->next;
 807                        if (p)
 808                                p->object.flags |= UNINTERESTING;
 809                        if (parse_commit(p) < 0)
 810                                continue;
 811                        if (p->parents)
 812                                mark_parents_uninteresting(p);
 813                        if (p->object.flags & SEEN)
 814                                continue;
 815                        p->object.flags |= SEEN;
 816                        commit_list_insert_by_date_cached(p, list, cached_base, cache_ptr);
 817                }
 818                return 0;
 819        }
 820
 821        /*
 822         * Ok, the commit wasn't uninteresting. Try to
 823         * simplify the commit history and find the parent
 824         * that has no differences in the path set if one exists.
 825         */
 826        try_to_simplify_commit(revs, commit);
 827
 828        if (revs->no_walk)
 829                return 0;
 830
 831        left_flag = (commit->object.flags & SYMMETRIC_LEFT);
 832
 833        for (parent = commit->parents; parent; parent = parent->next) {
 834                struct commit *p = parent->item;
 835
 836                if (parse_commit(p) < 0)
 837                        return -1;
 838                if (revs->show_source && !p->util)
 839                        p->util = commit->util;
 840                p->object.flags |= left_flag;
 841                if (!(p->object.flags & SEEN)) {
 842                        p->object.flags |= SEEN;
 843                        commit_list_insert_by_date_cached(p, list, cached_base, cache_ptr);
 844                }
 845                if (revs->first_parent_only)
 846                        break;
 847        }
 848        return 0;
 849}
 850
 851static void cherry_pick_list(struct commit_list *list, struct rev_info *revs)
 852{
 853        struct commit_list *p;
 854        int left_count = 0, right_count = 0;
 855        int left_first;
 856        struct patch_ids ids;
 857        unsigned cherry_flag;
 858
 859        /* First count the commits on the left and on the right */
 860        for (p = list; p; p = p->next) {
 861                struct commit *commit = p->item;
 862                unsigned flags = commit->object.flags;
 863                if (flags & BOUNDARY)
 864                        ;
 865                else if (flags & SYMMETRIC_LEFT)
 866                        left_count++;
 867                else
 868                        right_count++;
 869        }
 870
 871        if (!left_count || !right_count)
 872                return;
 873
 874        left_first = left_count < right_count;
 875        init_patch_ids(&ids);
 876        ids.diffopts.pathspec = revs->diffopt.pathspec;
 877
 878        /* Compute patch-ids for one side */
 879        for (p = list; p; p = p->next) {
 880                struct commit *commit = p->item;
 881                unsigned flags = commit->object.flags;
 882
 883                if (flags & BOUNDARY)
 884                        continue;
 885                /*
 886                 * If we have fewer left, left_first is set and we omit
 887                 * commits on the right branch in this loop.  If we have
 888                 * fewer right, we skip the left ones.
 889                 */
 890                if (left_first != !!(flags & SYMMETRIC_LEFT))
 891                        continue;
 892                commit->util = add_commit_patch_id(commit, &ids);
 893        }
 894
 895        /* either cherry_mark or cherry_pick are true */
 896        cherry_flag = revs->cherry_mark ? PATCHSAME : SHOWN;
 897
 898        /* Check the other side */
 899        for (p = list; p; p = p->next) {
 900                struct commit *commit = p->item;
 901                struct patch_id *id;
 902                unsigned flags = commit->object.flags;
 903
 904                if (flags & BOUNDARY)
 905                        continue;
 906                /*
 907                 * If we have fewer left, left_first is set and we omit
 908                 * commits on the left branch in this loop.
 909                 */
 910                if (left_first == !!(flags & SYMMETRIC_LEFT))
 911                        continue;
 912
 913                /*
 914                 * Have we seen the same patch id?
 915                 */
 916                id = has_commit_patch_id(commit, &ids);
 917                if (!id)
 918                        continue;
 919                id->seen = 1;
 920                commit->object.flags |= cherry_flag;
 921        }
 922
 923        /* Now check the original side for seen ones */
 924        for (p = list; p; p = p->next) {
 925                struct commit *commit = p->item;
 926                struct patch_id *ent;
 927
 928                ent = commit->util;
 929                if (!ent)
 930                        continue;
 931                if (ent->seen)
 932                        commit->object.flags |= cherry_flag;
 933                commit->util = NULL;
 934        }
 935
 936        free_patch_ids(&ids);
 937}
 938
 939/* How many extra uninteresting commits we want to see.. */
 940#define SLOP 5
 941
 942static int still_interesting(struct commit_list *src, unsigned long date, int slop,
 943                             struct commit **interesting_cache)
 944{
 945        /*
 946         * No source list at all? We're definitely done..
 947         */
 948        if (!src)
 949                return 0;
 950
 951        /*
 952         * Does the destination list contain entries with a date
 953         * before the source list? Definitely _not_ done.
 954         */
 955        if (date <= src->item->date)
 956                return SLOP;
 957
 958        /*
 959         * Does the source list still have interesting commits in
 960         * it? Definitely not done..
 961         */
 962        if (!everybody_uninteresting(src, interesting_cache))
 963                return SLOP;
 964
 965        /* Ok, we're closing in.. */
 966        return slop-1;
 967}
 968
 969/*
 970 * "rev-list --ancestry-path A..B" computes commits that are ancestors
 971 * of B but not ancestors of A but further limits the result to those
 972 * that are descendants of A.  This takes the list of bottom commits and
 973 * the result of "A..B" without --ancestry-path, and limits the latter
 974 * further to the ones that can reach one of the commits in "bottom".
 975 */
 976static void limit_to_ancestry(struct commit_list *bottom, struct commit_list *list)
 977{
 978        struct commit_list *p;
 979        struct commit_list *rlist = NULL;
 980        int made_progress;
 981
 982        /*
 983         * Reverse the list so that it will be likely that we would
 984         * process parents before children.
 985         */
 986        for (p = list; p; p = p->next)
 987                commit_list_insert(p->item, &rlist);
 988
 989        for (p = bottom; p; p = p->next)
 990                p->item->object.flags |= TMP_MARK;
 991
 992        /*
 993         * Mark the ones that can reach bottom commits in "list",
 994         * in a bottom-up fashion.
 995         */
 996        do {
 997                made_progress = 0;
 998                for (p = rlist; p; p = p->next) {
 999                        struct commit *c = p->item;
1000                        struct commit_list *parents;
1001                        if (c->object.flags & (TMP_MARK | UNINTERESTING))
1002                                continue;
1003                        for (parents = c->parents;
1004                             parents;
1005                             parents = parents->next) {
1006                                if (!(parents->item->object.flags & TMP_MARK))
1007                                        continue;
1008                                c->object.flags |= TMP_MARK;
1009                                made_progress = 1;
1010                                break;
1011                        }
1012                }
1013        } while (made_progress);
1014
1015        /*
1016         * NEEDSWORK: decide if we want to remove parents that are
1017         * not marked with TMP_MARK from commit->parents for commits
1018         * in the resulting list.  We may not want to do that, though.
1019         */
1020
1021        /*
1022         * The ones that are not marked with TMP_MARK are uninteresting
1023         */
1024        for (p = list; p; p = p->next) {
1025                struct commit *c = p->item;
1026                if (c->object.flags & TMP_MARK)
1027                        continue;
1028                c->object.flags |= UNINTERESTING;
1029        }
1030
1031        /* We are done with the TMP_MARK */
1032        for (p = list; p; p = p->next)
1033                p->item->object.flags &= ~TMP_MARK;
1034        for (p = bottom; p; p = p->next)
1035                p->item->object.flags &= ~TMP_MARK;
1036        free_commit_list(rlist);
1037}
1038
1039/*
1040 * Before walking the history, keep the set of "negative" refs the
1041 * caller has asked to exclude.
1042 *
1043 * This is used to compute "rev-list --ancestry-path A..B", as we need
1044 * to filter the result of "A..B" further to the ones that can actually
1045 * reach A.
1046 */
1047static struct commit_list *collect_bottom_commits(struct commit_list *list)
1048{
1049        struct commit_list *elem, *bottom = NULL;
1050        for (elem = list; elem; elem = elem->next)
1051                if (elem->item->object.flags & BOTTOM)
1052                        commit_list_insert(elem->item, &bottom);
1053        return bottom;
1054}
1055
1056/* Assumes either left_only or right_only is set */
1057static void limit_left_right(struct commit_list *list, struct rev_info *revs)
1058{
1059        struct commit_list *p;
1060
1061        for (p = list; p; p = p->next) {
1062                struct commit *commit = p->item;
1063
1064                if (revs->right_only) {
1065                        if (commit->object.flags & SYMMETRIC_LEFT)
1066                                commit->object.flags |= SHOWN;
1067                } else  /* revs->left_only is set */
1068                        if (!(commit->object.flags & SYMMETRIC_LEFT))
1069                                commit->object.flags |= SHOWN;
1070        }
1071}
1072
1073static int limit_list(struct rev_info *revs)
1074{
1075        int slop = SLOP;
1076        unsigned long date = ~0ul;
1077        struct commit_list *list = revs->commits;
1078        struct commit_list *newlist = NULL;
1079        struct commit_list **p = &newlist;
1080        struct commit_list *bottom = NULL;
1081        struct commit *interesting_cache = NULL;
1082
1083        if (revs->ancestry_path) {
1084                bottom = collect_bottom_commits(list);
1085                if (!bottom)
1086                        die("--ancestry-path given but there are no bottom commits");
1087        }
1088
1089        while (list) {
1090                struct commit_list *entry = list;
1091                struct commit *commit = list->item;
1092                struct object *obj = &commit->object;
1093                show_early_output_fn_t show;
1094
1095                list = list->next;
1096                free(entry);
1097
1098                if (commit == interesting_cache)
1099                        interesting_cache = NULL;
1100
1101                if (revs->max_age != -1 && (commit->date < revs->max_age))
1102                        obj->flags |= UNINTERESTING;
1103                if (add_parents_to_list(revs, commit, &list, NULL) < 0)
1104                        return -1;
1105                if (obj->flags & UNINTERESTING) {
1106                        mark_parents_uninteresting(commit);
1107                        if (revs->show_all)
1108                                p = &commit_list_insert(commit, p)->next;
1109                        slop = still_interesting(list, date, slop, &interesting_cache);
1110                        if (slop)
1111                                continue;
1112                        /* If showing all, add the whole pending list to the end */
1113                        if (revs->show_all)
1114                                *p = list;
1115                        break;
1116                }
1117                if (revs->min_age != -1 && (commit->date > revs->min_age))
1118                        continue;
1119                date = commit->date;
1120                p = &commit_list_insert(commit, p)->next;
1121
1122                show = show_early_output;
1123                if (!show)
1124                        continue;
1125
1126                show(revs, newlist);
1127                show_early_output = NULL;
1128        }
1129        if (revs->cherry_pick || revs->cherry_mark)
1130                cherry_pick_list(newlist, revs);
1131
1132        if (revs->left_only || revs->right_only)
1133                limit_left_right(newlist, revs);
1134
1135        if (bottom) {
1136                limit_to_ancestry(bottom, newlist);
1137                free_commit_list(bottom);
1138        }
1139
1140        /*
1141         * Check if any commits have become TREESAME by some of their parents
1142         * becoming UNINTERESTING.
1143         */
1144        if (limiting_can_increase_treesame(revs))
1145                for (list = newlist; list; list = list->next) {
1146                        struct commit *c = list->item;
1147                        if (c->object.flags & (UNINTERESTING | TREESAME))
1148                                continue;
1149                        update_treesame(revs, c);
1150                }
1151
1152        revs->commits = newlist;
1153        return 0;
1154}
1155
1156/*
1157 * Add an entry to refs->cmdline with the specified information.
1158 * *name is copied.
1159 */
1160static void add_rev_cmdline(struct rev_info *revs,
1161                            struct object *item,
1162                            const char *name,
1163                            int whence,
1164                            unsigned flags)
1165{
1166        struct rev_cmdline_info *info = &revs->cmdline;
1167        int nr = info->nr;
1168
1169        ALLOC_GROW(info->rev, nr + 1, info->alloc);
1170        info->rev[nr].item = item;
1171        info->rev[nr].name = xstrdup(name);
1172        info->rev[nr].whence = whence;
1173        info->rev[nr].flags = flags;
1174        info->nr++;
1175}
1176
1177static void add_rev_cmdline_list(struct rev_info *revs,
1178                                 struct commit_list *commit_list,
1179                                 int whence,
1180                                 unsigned flags)
1181{
1182        while (commit_list) {
1183                struct object *object = &commit_list->item->object;
1184                add_rev_cmdline(revs, object, sha1_to_hex(object->sha1),
1185                                whence, flags);
1186                commit_list = commit_list->next;
1187        }
1188}
1189
1190struct all_refs_cb {
1191        int all_flags;
1192        int warned_bad_reflog;
1193        struct rev_info *all_revs;
1194        const char *name_for_errormsg;
1195};
1196
1197int ref_excluded(struct string_list *ref_excludes, const char *path)
1198{
1199        struct string_list_item *item;
1200
1201        if (!ref_excludes)
1202                return 0;
1203        for_each_string_list_item(item, ref_excludes) {
1204                if (!wildmatch(item->string, path, 0, NULL))
1205                        return 1;
1206        }
1207        return 0;
1208}
1209
1210static int handle_one_ref(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1211{
1212        struct all_refs_cb *cb = cb_data;
1213        struct object *object;
1214
1215        if (ref_excluded(cb->all_revs->ref_excludes, path))
1216            return 0;
1217
1218        object = get_reference(cb->all_revs, path, sha1, cb->all_flags);
1219        add_rev_cmdline(cb->all_revs, object, path, REV_CMD_REF, cb->all_flags);
1220        add_pending_sha1(cb->all_revs, path, sha1, cb->all_flags);
1221        return 0;
1222}
1223
1224static void init_all_refs_cb(struct all_refs_cb *cb, struct rev_info *revs,
1225        unsigned flags)
1226{
1227        cb->all_revs = revs;
1228        cb->all_flags = flags;
1229}
1230
1231void clear_ref_exclusion(struct string_list **ref_excludes_p)
1232{
1233        if (*ref_excludes_p) {
1234                string_list_clear(*ref_excludes_p, 0);
1235                free(*ref_excludes_p);
1236        }
1237        *ref_excludes_p = NULL;
1238}
1239
1240void add_ref_exclusion(struct string_list **ref_excludes_p, const char *exclude)
1241{
1242        if (!*ref_excludes_p) {
1243                *ref_excludes_p = xcalloc(1, sizeof(**ref_excludes_p));
1244                (*ref_excludes_p)->strdup_strings = 1;
1245        }
1246        string_list_append(*ref_excludes_p, exclude);
1247}
1248
1249static void handle_refs(const char *submodule, struct rev_info *revs, unsigned flags,
1250                int (*for_each)(const char *, each_ref_fn, void *))
1251{
1252        struct all_refs_cb cb;
1253        init_all_refs_cb(&cb, revs, flags);
1254        for_each(submodule, handle_one_ref, &cb);
1255}
1256
1257static void handle_one_reflog_commit(unsigned char *sha1, void *cb_data)
1258{
1259        struct all_refs_cb *cb = cb_data;
1260        if (!is_null_sha1(sha1)) {
1261                struct object *o = parse_object(sha1);
1262                if (o) {
1263                        o->flags |= cb->all_flags;
1264                        /* ??? CMDLINEFLAGS ??? */
1265                        add_pending_object(cb->all_revs, o, "");
1266                }
1267                else if (!cb->warned_bad_reflog) {
1268                        warning("reflog of '%s' references pruned commits",
1269                                cb->name_for_errormsg);
1270                        cb->warned_bad_reflog = 1;
1271                }
1272        }
1273}
1274
1275static int handle_one_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
1276                const char *email, unsigned long timestamp, int tz,
1277                const char *message, void *cb_data)
1278{
1279        handle_one_reflog_commit(osha1, cb_data);
1280        handle_one_reflog_commit(nsha1, cb_data);
1281        return 0;
1282}
1283
1284static int handle_one_reflog(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1285{
1286        struct all_refs_cb *cb = cb_data;
1287        cb->warned_bad_reflog = 0;
1288        cb->name_for_errormsg = path;
1289        for_each_reflog_ent(path, handle_one_reflog_ent, cb_data);
1290        return 0;
1291}
1292
1293static void handle_reflog(struct rev_info *revs, unsigned flags)
1294{
1295        struct all_refs_cb cb;
1296        cb.all_revs = revs;
1297        cb.all_flags = flags;
1298        for_each_reflog(handle_one_reflog, &cb);
1299}
1300
1301static int add_parents_only(struct rev_info *revs, const char *arg_, int flags)
1302{
1303        unsigned char sha1[20];
1304        struct object *it;
1305        struct commit *commit;
1306        struct commit_list *parents;
1307        const char *arg = arg_;
1308
1309        if (*arg == '^') {
1310                flags ^= UNINTERESTING | BOTTOM;
1311                arg++;
1312        }
1313        if (get_sha1_committish(arg, sha1))
1314                return 0;
1315        while (1) {
1316                it = get_reference(revs, arg, sha1, 0);
1317                if (!it && revs->ignore_missing)
1318                        return 0;
1319                if (it->type != OBJ_TAG)
1320                        break;
1321                if (!((struct tag*)it)->tagged)
1322                        return 0;
1323                hashcpy(sha1, ((struct tag*)it)->tagged->sha1);
1324        }
1325        if (it->type != OBJ_COMMIT)
1326                return 0;
1327        commit = (struct commit *)it;
1328        for (parents = commit->parents; parents; parents = parents->next) {
1329                it = &parents->item->object;
1330                it->flags |= flags;
1331                add_rev_cmdline(revs, it, arg_, REV_CMD_PARENTS_ONLY, flags);
1332                add_pending_object(revs, it, arg);
1333        }
1334        return 1;
1335}
1336
1337void init_revisions(struct rev_info *revs, const char *prefix)
1338{
1339        memset(revs, 0, sizeof(*revs));
1340
1341        revs->abbrev = DEFAULT_ABBREV;
1342        revs->ignore_merges = 1;
1343        revs->simplify_history = 1;
1344        DIFF_OPT_SET(&revs->pruning, RECURSIVE);
1345        DIFF_OPT_SET(&revs->pruning, QUICK);
1346        revs->pruning.add_remove = file_add_remove;
1347        revs->pruning.change = file_change;
1348        revs->sort_order = REV_SORT_IN_GRAPH_ORDER;
1349        revs->dense = 1;
1350        revs->prefix = prefix;
1351        revs->max_age = -1;
1352        revs->min_age = -1;
1353        revs->skip_count = -1;
1354        revs->max_count = -1;
1355        revs->max_parents = -1;
1356
1357        revs->commit_format = CMIT_FMT_DEFAULT;
1358
1359        init_grep_defaults();
1360        grep_init(&revs->grep_filter, prefix);
1361        revs->grep_filter.status_only = 1;
1362        revs->grep_filter.regflags = REG_NEWLINE;
1363
1364        diff_setup(&revs->diffopt);
1365        if (prefix && !revs->diffopt.prefix) {
1366                revs->diffopt.prefix = prefix;
1367                revs->diffopt.prefix_length = strlen(prefix);
1368        }
1369
1370        revs->notes_opt.use_default_notes = -1;
1371}
1372
1373static void add_pending_commit_list(struct rev_info *revs,
1374                                    struct commit_list *commit_list,
1375                                    unsigned int flags)
1376{
1377        while (commit_list) {
1378                struct object *object = &commit_list->item->object;
1379                object->flags |= flags;
1380                add_pending_object(revs, object, sha1_to_hex(object->sha1));
1381                commit_list = commit_list->next;
1382        }
1383}
1384
1385static void prepare_show_merge(struct rev_info *revs)
1386{
1387        struct commit_list *bases;
1388        struct commit *head, *other;
1389        unsigned char sha1[20];
1390        const char **prune = NULL;
1391        int i, prune_num = 1; /* counting terminating NULL */
1392
1393        if (get_sha1("HEAD", sha1))
1394                die("--merge without HEAD?");
1395        head = lookup_commit_or_die(sha1, "HEAD");
1396        if (get_sha1("MERGE_HEAD", sha1))
1397                die("--merge without MERGE_HEAD?");
1398        other = lookup_commit_or_die(sha1, "MERGE_HEAD");
1399        add_pending_object(revs, &head->object, "HEAD");
1400        add_pending_object(revs, &other->object, "MERGE_HEAD");
1401        bases = get_merge_bases(head, other, 1);
1402        add_rev_cmdline_list(revs, bases, REV_CMD_MERGE_BASE, UNINTERESTING | BOTTOM);
1403        add_pending_commit_list(revs, bases, UNINTERESTING | BOTTOM);
1404        free_commit_list(bases);
1405        head->object.flags |= SYMMETRIC_LEFT;
1406
1407        if (!active_nr)
1408                read_cache();
1409        for (i = 0; i < active_nr; i++) {
1410                const struct cache_entry *ce = active_cache[i];
1411                if (!ce_stage(ce))
1412                        continue;
1413                if (ce_path_match(ce, &revs->prune_data, NULL)) {
1414                        prune_num++;
1415                        prune = xrealloc(prune, sizeof(*prune) * prune_num);
1416                        prune[prune_num-2] = ce->name;
1417                        prune[prune_num-1] = NULL;
1418                }
1419                while ((i+1 < active_nr) &&
1420                       ce_same_name(ce, active_cache[i+1]))
1421                        i++;
1422        }
1423        free_pathspec(&revs->prune_data);
1424        parse_pathspec(&revs->prune_data, PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
1425                       PATHSPEC_PREFER_FULL | PATHSPEC_LITERAL_PATH, "", prune);
1426        revs->limited = 1;
1427}
1428
1429int handle_revision_arg(const char *arg_, struct rev_info *revs, int flags, unsigned revarg_opt)
1430{
1431        struct object_context oc;
1432        char *dotdot;
1433        struct object *object;
1434        unsigned char sha1[20];
1435        int local_flags;
1436        const char *arg = arg_;
1437        int cant_be_filename = revarg_opt & REVARG_CANNOT_BE_FILENAME;
1438        unsigned get_sha1_flags = 0;
1439
1440        flags = flags & UNINTERESTING ? flags | BOTTOM : flags & ~BOTTOM;
1441
1442        dotdot = strstr(arg, "..");
1443        if (dotdot) {
1444                unsigned char from_sha1[20];
1445                const char *next = dotdot + 2;
1446                const char *this = arg;
1447                int symmetric = *next == '.';
1448                unsigned int flags_exclude = flags ^ (UNINTERESTING | BOTTOM);
1449                static const char head_by_default[] = "HEAD";
1450                unsigned int a_flags;
1451
1452                *dotdot = 0;
1453                next += symmetric;
1454
1455                if (!*next)
1456                        next = head_by_default;
1457                if (dotdot == arg)
1458                        this = head_by_default;
1459                if (this == head_by_default && next == head_by_default &&
1460                    !symmetric) {
1461                        /*
1462                         * Just ".."?  That is not a range but the
1463                         * pathspec for the parent directory.
1464                         */
1465                        if (!cant_be_filename) {
1466                                *dotdot = '.';
1467                                return -1;
1468                        }
1469                }
1470                if (!get_sha1_committish(this, from_sha1) &&
1471                    !get_sha1_committish(next, sha1)) {
1472                        struct object *a_obj, *b_obj;
1473
1474                        if (!cant_be_filename) {
1475                                *dotdot = '.';
1476                                verify_non_filename(revs->prefix, arg);
1477                        }
1478
1479                        a_obj = parse_object(from_sha1);
1480                        b_obj = parse_object(sha1);
1481                        if (!a_obj || !b_obj) {
1482                        missing:
1483                                if (revs->ignore_missing)
1484                                        return 0;
1485                                die(symmetric
1486                                    ? "Invalid symmetric difference expression %s"
1487                                    : "Invalid revision range %s", arg);
1488                        }
1489
1490                        if (!symmetric) {
1491                                /* just A..B */
1492                                a_flags = flags_exclude;
1493                        } else {
1494                                /* A...B -- find merge bases between the two */
1495                                struct commit *a, *b;
1496                                struct commit_list *exclude;
1497
1498                                a = (a_obj->type == OBJ_COMMIT
1499                                     ? (struct commit *)a_obj
1500                                     : lookup_commit_reference(a_obj->sha1));
1501                                b = (b_obj->type == OBJ_COMMIT
1502                                     ? (struct commit *)b_obj
1503                                     : lookup_commit_reference(b_obj->sha1));
1504                                if (!a || !b)
1505                                        goto missing;
1506                                exclude = get_merge_bases(a, b, 1);
1507                                add_rev_cmdline_list(revs, exclude,
1508                                                     REV_CMD_MERGE_BASE,
1509                                                     flags_exclude);
1510                                add_pending_commit_list(revs, exclude,
1511                                                        flags_exclude);
1512                                free_commit_list(exclude);
1513
1514                                a_flags = flags | SYMMETRIC_LEFT;
1515                        }
1516
1517                        a_obj->flags |= a_flags;
1518                        b_obj->flags |= flags;
1519                        add_rev_cmdline(revs, a_obj, this,
1520                                        REV_CMD_LEFT, a_flags);
1521                        add_rev_cmdline(revs, b_obj, next,
1522                                        REV_CMD_RIGHT, flags);
1523                        add_pending_object(revs, a_obj, this);
1524                        add_pending_object(revs, b_obj, next);
1525                        return 0;
1526                }
1527                *dotdot = '.';
1528        }
1529        dotdot = strstr(arg, "^@");
1530        if (dotdot && !dotdot[2]) {
1531                *dotdot = 0;
1532                if (add_parents_only(revs, arg, flags))
1533                        return 0;
1534                *dotdot = '^';
1535        }
1536        dotdot = strstr(arg, "^!");
1537        if (dotdot && !dotdot[2]) {
1538                *dotdot = 0;
1539                if (!add_parents_only(revs, arg, flags ^ (UNINTERESTING | BOTTOM)))
1540                        *dotdot = '^';
1541        }
1542
1543        local_flags = 0;
1544        if (*arg == '^') {
1545                local_flags = UNINTERESTING | BOTTOM;
1546                arg++;
1547        }
1548
1549        if (revarg_opt & REVARG_COMMITTISH)
1550                get_sha1_flags = GET_SHA1_COMMITTISH;
1551
1552        if (get_sha1_with_context(arg, get_sha1_flags, sha1, &oc))
1553                return revs->ignore_missing ? 0 : -1;
1554        if (!cant_be_filename)
1555                verify_non_filename(revs->prefix, arg);
1556        object = get_reference(revs, arg, sha1, flags ^ local_flags);
1557        add_rev_cmdline(revs, object, arg_, REV_CMD_REV, flags ^ local_flags);
1558        add_pending_object_with_mode(revs, object, arg, oc.mode);
1559        return 0;
1560}
1561
1562struct cmdline_pathspec {
1563        int alloc;
1564        int nr;
1565        const char **path;
1566};
1567
1568static void append_prune_data(struct cmdline_pathspec *prune, const char **av)
1569{
1570        while (*av) {
1571                ALLOC_GROW(prune->path, prune->nr + 1, prune->alloc);
1572                prune->path[prune->nr++] = *(av++);
1573        }
1574}
1575
1576static void read_pathspec_from_stdin(struct rev_info *revs, struct strbuf *sb,
1577                                     struct cmdline_pathspec *prune)
1578{
1579        while (strbuf_getwholeline(sb, stdin, '\n') != EOF) {
1580                int len = sb->len;
1581                if (len && sb->buf[len - 1] == '\n')
1582                        sb->buf[--len] = '\0';
1583                ALLOC_GROW(prune->path, prune->nr + 1, prune->alloc);
1584                prune->path[prune->nr++] = xstrdup(sb->buf);
1585        }
1586}
1587
1588static void read_revisions_from_stdin(struct rev_info *revs,
1589                                      struct cmdline_pathspec *prune)
1590{
1591        struct strbuf sb;
1592        int seen_dashdash = 0;
1593        int save_warning;
1594
1595        save_warning = warn_on_object_refname_ambiguity;
1596        warn_on_object_refname_ambiguity = 0;
1597
1598        strbuf_init(&sb, 1000);
1599        while (strbuf_getwholeline(&sb, stdin, '\n') != EOF) {
1600                int len = sb.len;
1601                if (len && sb.buf[len - 1] == '\n')
1602                        sb.buf[--len] = '\0';
1603                if (!len)
1604                        break;
1605                if (sb.buf[0] == '-') {
1606                        if (len == 2 && sb.buf[1] == '-') {
1607                                seen_dashdash = 1;
1608                                break;
1609                        }
1610                        die("options not supported in --stdin mode");
1611                }
1612                if (handle_revision_arg(sb.buf, revs, 0,
1613                                        REVARG_CANNOT_BE_FILENAME))
1614                        die("bad revision '%s'", sb.buf);
1615        }
1616        if (seen_dashdash)
1617                read_pathspec_from_stdin(revs, &sb, prune);
1618
1619        strbuf_release(&sb);
1620        warn_on_object_refname_ambiguity = save_warning;
1621}
1622
1623static void add_grep(struct rev_info *revs, const char *ptn, enum grep_pat_token what)
1624{
1625        append_grep_pattern(&revs->grep_filter, ptn, "command line", 0, what);
1626}
1627
1628static void add_header_grep(struct rev_info *revs, enum grep_header_field field, const char *pattern)
1629{
1630        append_header_grep_pattern(&revs->grep_filter, field, pattern);
1631}
1632
1633static void add_message_grep(struct rev_info *revs, const char *pattern)
1634{
1635        add_grep(revs, pattern, GREP_PATTERN_BODY);
1636}
1637
1638static int handle_revision_opt(struct rev_info *revs, int argc, const char **argv,
1639                               int *unkc, const char **unkv)
1640{
1641        const char *arg = argv[0];
1642        const char *optarg;
1643        int argcount;
1644
1645        /* pseudo revision arguments */
1646        if (!strcmp(arg, "--all") || !strcmp(arg, "--branches") ||
1647            !strcmp(arg, "--tags") || !strcmp(arg, "--remotes") ||
1648            !strcmp(arg, "--reflog") || !strcmp(arg, "--not") ||
1649            !strcmp(arg, "--no-walk") || !strcmp(arg, "--do-walk") ||
1650            !strcmp(arg, "--bisect") || starts_with(arg, "--glob=") ||
1651            starts_with(arg, "--exclude=") ||
1652            starts_with(arg, "--branches=") || starts_with(arg, "--tags=") ||
1653            starts_with(arg, "--remotes=") || starts_with(arg, "--no-walk="))
1654        {
1655                unkv[(*unkc)++] = arg;
1656                return 1;
1657        }
1658
1659        if ((argcount = parse_long_opt("max-count", argv, &optarg))) {
1660                revs->max_count = atoi(optarg);
1661                revs->no_walk = 0;
1662                return argcount;
1663        } else if ((argcount = parse_long_opt("skip", argv, &optarg))) {
1664                revs->skip_count = atoi(optarg);
1665                return argcount;
1666        } else if ((*arg == '-') && isdigit(arg[1])) {
1667                /* accept -<digit>, like traditional "head" */
1668                if (strtol_i(arg + 1, 10, &revs->max_count) < 0 ||
1669                    revs->max_count < 0)
1670                        die("'%s': not a non-negative integer", arg + 1);
1671                revs->no_walk = 0;
1672        } else if (!strcmp(arg, "-n")) {
1673                if (argc <= 1)
1674                        return error("-n requires an argument");
1675                revs->max_count = atoi(argv[1]);
1676                revs->no_walk = 0;
1677                return 2;
1678        } else if (starts_with(arg, "-n")) {
1679                revs->max_count = atoi(arg + 2);
1680                revs->no_walk = 0;
1681        } else if ((argcount = parse_long_opt("max-age", argv, &optarg))) {
1682                revs->max_age = atoi(optarg);
1683                return argcount;
1684        } else if ((argcount = parse_long_opt("since", argv, &optarg))) {
1685                revs->max_age = approxidate(optarg);
1686                return argcount;
1687        } else if ((argcount = parse_long_opt("after", argv, &optarg))) {
1688                revs->max_age = approxidate(optarg);
1689                return argcount;
1690        } else if ((argcount = parse_long_opt("min-age", argv, &optarg))) {
1691                revs->min_age = atoi(optarg);
1692                return argcount;
1693        } else if ((argcount = parse_long_opt("before", argv, &optarg))) {
1694                revs->min_age = approxidate(optarg);
1695                return argcount;
1696        } else if ((argcount = parse_long_opt("until", argv, &optarg))) {
1697                revs->min_age = approxidate(optarg);
1698                return argcount;
1699        } else if (!strcmp(arg, "--first-parent")) {
1700                revs->first_parent_only = 1;
1701        } else if (!strcmp(arg, "--ancestry-path")) {
1702                revs->ancestry_path = 1;
1703                revs->simplify_history = 0;
1704                revs->limited = 1;
1705        } else if (!strcmp(arg, "-g") || !strcmp(arg, "--walk-reflogs")) {
1706                init_reflog_walk(&revs->reflog_info);
1707        } else if (!strcmp(arg, "--default")) {
1708                if (argc <= 1)
1709                        return error("bad --default argument");
1710                revs->def = argv[1];
1711                return 2;
1712        } else if (!strcmp(arg, "--merge")) {
1713                revs->show_merge = 1;
1714        } else if (!strcmp(arg, "--topo-order")) {
1715                revs->sort_order = REV_SORT_IN_GRAPH_ORDER;
1716                revs->topo_order = 1;
1717        } else if (!strcmp(arg, "--simplify-merges")) {
1718                revs->simplify_merges = 1;
1719                revs->topo_order = 1;
1720                revs->rewrite_parents = 1;
1721                revs->simplify_history = 0;
1722                revs->limited = 1;
1723        } else if (!strcmp(arg, "--simplify-by-decoration")) {
1724                revs->simplify_merges = 1;
1725                revs->topo_order = 1;
1726                revs->rewrite_parents = 1;
1727                revs->simplify_history = 0;
1728                revs->simplify_by_decoration = 1;
1729                revs->limited = 1;
1730                revs->prune = 1;
1731                load_ref_decorations(DECORATE_SHORT_REFS);
1732        } else if (!strcmp(arg, "--date-order")) {
1733                revs->sort_order = REV_SORT_BY_COMMIT_DATE;
1734                revs->topo_order = 1;
1735        } else if (!strcmp(arg, "--author-date-order")) {
1736                revs->sort_order = REV_SORT_BY_AUTHOR_DATE;
1737                revs->topo_order = 1;
1738        } else if (starts_with(arg, "--early-output")) {
1739                int count = 100;
1740                switch (arg[14]) {
1741                case '=':
1742                        count = atoi(arg+15);
1743                        /* Fallthrough */
1744                case 0:
1745                        revs->topo_order = 1;
1746                       revs->early_output = count;
1747                }
1748        } else if (!strcmp(arg, "--parents")) {
1749                revs->rewrite_parents = 1;
1750                revs->print_parents = 1;
1751        } else if (!strcmp(arg, "--dense")) {
1752                revs->dense = 1;
1753        } else if (!strcmp(arg, "--sparse")) {
1754                revs->dense = 0;
1755        } else if (!strcmp(arg, "--show-all")) {
1756                revs->show_all = 1;
1757        } else if (!strcmp(arg, "--remove-empty")) {
1758                revs->remove_empty_trees = 1;
1759        } else if (!strcmp(arg, "--merges")) {
1760                revs->min_parents = 2;
1761        } else if (!strcmp(arg, "--no-merges")) {
1762                revs->max_parents = 1;
1763        } else if (starts_with(arg, "--min-parents=")) {
1764                revs->min_parents = atoi(arg+14);
1765        } else if (starts_with(arg, "--no-min-parents")) {
1766                revs->min_parents = 0;
1767        } else if (starts_with(arg, "--max-parents=")) {
1768                revs->max_parents = atoi(arg+14);
1769        } else if (starts_with(arg, "--no-max-parents")) {
1770                revs->max_parents = -1;
1771        } else if (!strcmp(arg, "--boundary")) {
1772                revs->boundary = 1;
1773        } else if (!strcmp(arg, "--left-right")) {
1774                revs->left_right = 1;
1775        } else if (!strcmp(arg, "--left-only")) {
1776                if (revs->right_only)
1777                        die("--left-only is incompatible with --right-only"
1778                            " or --cherry");
1779                revs->left_only = 1;
1780        } else if (!strcmp(arg, "--right-only")) {
1781                if (revs->left_only)
1782                        die("--right-only is incompatible with --left-only");
1783                revs->right_only = 1;
1784        } else if (!strcmp(arg, "--cherry")) {
1785                if (revs->left_only)
1786                        die("--cherry is incompatible with --left-only");
1787                revs->cherry_mark = 1;
1788                revs->right_only = 1;
1789                revs->max_parents = 1;
1790                revs->limited = 1;
1791        } else if (!strcmp(arg, "--count")) {
1792                revs->count = 1;
1793        } else if (!strcmp(arg, "--cherry-mark")) {
1794                if (revs->cherry_pick)
1795                        die("--cherry-mark is incompatible with --cherry-pick");
1796                revs->cherry_mark = 1;
1797                revs->limited = 1; /* needs limit_list() */
1798        } else if (!strcmp(arg, "--cherry-pick")) {
1799                if (revs->cherry_mark)
1800                        die("--cherry-pick is incompatible with --cherry-mark");
1801                revs->cherry_pick = 1;
1802                revs->limited = 1;
1803        } else if (!strcmp(arg, "--objects")) {
1804                revs->tag_objects = 1;
1805                revs->tree_objects = 1;
1806                revs->blob_objects = 1;
1807        } else if (!strcmp(arg, "--objects-edge")) {
1808                revs->tag_objects = 1;
1809                revs->tree_objects = 1;
1810                revs->blob_objects = 1;
1811                revs->edge_hint = 1;
1812        } else if (!strcmp(arg, "--verify-objects")) {
1813                revs->tag_objects = 1;
1814                revs->tree_objects = 1;
1815                revs->blob_objects = 1;
1816                revs->verify_objects = 1;
1817        } else if (!strcmp(arg, "--unpacked")) {
1818                revs->unpacked = 1;
1819        } else if (starts_with(arg, "--unpacked=")) {
1820                die("--unpacked=<packfile> no longer supported.");
1821        } else if (!strcmp(arg, "-r")) {
1822                revs->diff = 1;
1823                DIFF_OPT_SET(&revs->diffopt, RECURSIVE);
1824        } else if (!strcmp(arg, "-t")) {
1825                revs->diff = 1;
1826                DIFF_OPT_SET(&revs->diffopt, RECURSIVE);
1827                DIFF_OPT_SET(&revs->diffopt, TREE_IN_RECURSIVE);
1828        } else if (!strcmp(arg, "-m")) {
1829                revs->ignore_merges = 0;
1830        } else if (!strcmp(arg, "-c")) {
1831                revs->diff = 1;
1832                revs->dense_combined_merges = 0;
1833                revs->combine_merges = 1;
1834        } else if (!strcmp(arg, "--cc")) {
1835                revs->diff = 1;
1836                revs->dense_combined_merges = 1;
1837                revs->combine_merges = 1;
1838        } else if (!strcmp(arg, "-v")) {
1839                revs->verbose_header = 1;
1840        } else if (!strcmp(arg, "--pretty")) {
1841                revs->verbose_header = 1;
1842                revs->pretty_given = 1;
1843                get_commit_format(arg+8, revs);
1844        } else if (starts_with(arg, "--pretty=") || starts_with(arg, "--format=")) {
1845                /*
1846                 * Detached form ("--pretty X" as opposed to "--pretty=X")
1847                 * not allowed, since the argument is optional.
1848                 */
1849                revs->verbose_header = 1;
1850                revs->pretty_given = 1;
1851                get_commit_format(arg+9, revs);
1852        } else if (!strcmp(arg, "--show-notes") || !strcmp(arg, "--notes")) {
1853                revs->show_notes = 1;
1854                revs->show_notes_given = 1;
1855                revs->notes_opt.use_default_notes = 1;
1856        } else if (!strcmp(arg, "--show-signature")) {
1857                revs->show_signature = 1;
1858        } else if (!strcmp(arg, "--show-linear-break") ||
1859                   starts_with(arg, "--show-linear-break=")) {
1860                if (starts_with(arg, "--show-linear-break="))
1861                        revs->break_bar = xstrdup(arg + 20);
1862                else
1863                        revs->break_bar = "                    ..........";
1864                revs->track_linear = 1;
1865                revs->track_first_time = 1;
1866        } else if (starts_with(arg, "--show-notes=") ||
1867                   starts_with(arg, "--notes=")) {
1868                struct strbuf buf = STRBUF_INIT;
1869                revs->show_notes = 1;
1870                revs->show_notes_given = 1;
1871                if (starts_with(arg, "--show-notes")) {
1872                        if (revs->notes_opt.use_default_notes < 0)
1873                                revs->notes_opt.use_default_notes = 1;
1874                        strbuf_addstr(&buf, arg+13);
1875                }
1876                else
1877                        strbuf_addstr(&buf, arg+8);
1878                expand_notes_ref(&buf);
1879                string_list_append(&revs->notes_opt.extra_notes_refs,
1880                                   strbuf_detach(&buf, NULL));
1881        } else if (!strcmp(arg, "--no-notes")) {
1882                revs->show_notes = 0;
1883                revs->show_notes_given = 1;
1884                revs->notes_opt.use_default_notes = -1;
1885                /* we have been strdup'ing ourselves, so trick
1886                 * string_list into free()ing strings */
1887                revs->notes_opt.extra_notes_refs.strdup_strings = 1;
1888                string_list_clear(&revs->notes_opt.extra_notes_refs, 0);
1889                revs->notes_opt.extra_notes_refs.strdup_strings = 0;
1890        } else if (!strcmp(arg, "--standard-notes")) {
1891                revs->show_notes_given = 1;
1892                revs->notes_opt.use_default_notes = 1;
1893        } else if (!strcmp(arg, "--no-standard-notes")) {
1894                revs->notes_opt.use_default_notes = 0;
1895        } else if (!strcmp(arg, "--oneline")) {
1896                revs->verbose_header = 1;
1897                get_commit_format("oneline", revs);
1898                revs->pretty_given = 1;
1899                revs->abbrev_commit = 1;
1900        } else if (!strcmp(arg, "--graph")) {
1901                revs->topo_order = 1;
1902                revs->rewrite_parents = 1;
1903                revs->graph = graph_init(revs);
1904        } else if (!strcmp(arg, "--root")) {
1905                revs->show_root_diff = 1;
1906        } else if (!strcmp(arg, "--no-commit-id")) {
1907                revs->no_commit_id = 1;
1908        } else if (!strcmp(arg, "--always")) {
1909                revs->always_show_header = 1;
1910        } else if (!strcmp(arg, "--no-abbrev")) {
1911                revs->abbrev = 0;
1912        } else if (!strcmp(arg, "--abbrev")) {
1913                revs->abbrev = DEFAULT_ABBREV;
1914        } else if (starts_with(arg, "--abbrev=")) {
1915                revs->abbrev = strtoul(arg + 9, NULL, 10);
1916                if (revs->abbrev < MINIMUM_ABBREV)
1917                        revs->abbrev = MINIMUM_ABBREV;
1918                else if (revs->abbrev > 40)
1919                        revs->abbrev = 40;
1920        } else if (!strcmp(arg, "--abbrev-commit")) {
1921                revs->abbrev_commit = 1;
1922                revs->abbrev_commit_given = 1;
1923        } else if (!strcmp(arg, "--no-abbrev-commit")) {
1924                revs->abbrev_commit = 0;
1925        } else if (!strcmp(arg, "--full-diff")) {
1926                revs->diff = 1;
1927                revs->full_diff = 1;
1928        } else if (!strcmp(arg, "--full-history")) {
1929                revs->simplify_history = 0;
1930        } else if (!strcmp(arg, "--relative-date")) {
1931                revs->date_mode = DATE_RELATIVE;
1932                revs->date_mode_explicit = 1;
1933        } else if ((argcount = parse_long_opt("date", argv, &optarg))) {
1934                revs->date_mode = parse_date_format(optarg);
1935                revs->date_mode_explicit = 1;
1936                return argcount;
1937        } else if (!strcmp(arg, "--log-size")) {
1938                revs->show_log_size = 1;
1939        }
1940        /*
1941         * Grepping the commit log
1942         */
1943        else if ((argcount = parse_long_opt("author", argv, &optarg))) {
1944                add_header_grep(revs, GREP_HEADER_AUTHOR, optarg);
1945                return argcount;
1946        } else if ((argcount = parse_long_opt("committer", argv, &optarg))) {
1947                add_header_grep(revs, GREP_HEADER_COMMITTER, optarg);
1948                return argcount;
1949        } else if ((argcount = parse_long_opt("grep-reflog", argv, &optarg))) {
1950                add_header_grep(revs, GREP_HEADER_REFLOG, optarg);
1951                return argcount;
1952        } else if ((argcount = parse_long_opt("grep", argv, &optarg))) {
1953                add_message_grep(revs, optarg);
1954                return argcount;
1955        } else if (!strcmp(arg, "--grep-debug")) {
1956                revs->grep_filter.debug = 1;
1957        } else if (!strcmp(arg, "--basic-regexp")) {
1958                grep_set_pattern_type_option(GREP_PATTERN_TYPE_BRE, &revs->grep_filter);
1959        } else if (!strcmp(arg, "--extended-regexp") || !strcmp(arg, "-E")) {
1960                grep_set_pattern_type_option(GREP_PATTERN_TYPE_ERE, &revs->grep_filter);
1961        } else if (!strcmp(arg, "--regexp-ignore-case") || !strcmp(arg, "-i")) {
1962                revs->grep_filter.regflags |= REG_ICASE;
1963                DIFF_OPT_SET(&revs->diffopt, PICKAXE_IGNORE_CASE);
1964        } else if (!strcmp(arg, "--fixed-strings") || !strcmp(arg, "-F")) {
1965                grep_set_pattern_type_option(GREP_PATTERN_TYPE_FIXED, &revs->grep_filter);
1966        } else if (!strcmp(arg, "--perl-regexp")) {
1967                grep_set_pattern_type_option(GREP_PATTERN_TYPE_PCRE, &revs->grep_filter);
1968        } else if (!strcmp(arg, "--all-match")) {
1969                revs->grep_filter.all_match = 1;
1970        } else if ((argcount = parse_long_opt("encoding", argv, &optarg))) {
1971                if (strcmp(optarg, "none"))
1972                        git_log_output_encoding = xstrdup(optarg);
1973                else
1974                        git_log_output_encoding = "";
1975                return argcount;
1976        } else if (!strcmp(arg, "--reverse")) {
1977                revs->reverse ^= 1;
1978        } else if (!strcmp(arg, "--children")) {
1979                revs->children.name = "children";
1980                revs->limited = 1;
1981        } else if (!strcmp(arg, "--ignore-missing")) {
1982                revs->ignore_missing = 1;
1983        } else {
1984                int opts = diff_opt_parse(&revs->diffopt, argv, argc);
1985                if (!opts)
1986                        unkv[(*unkc)++] = arg;
1987                return opts;
1988        }
1989        if (revs->graph && revs->track_linear)
1990                die("--show-linear-break and --graph are incompatible");
1991
1992        return 1;
1993}
1994
1995void parse_revision_opt(struct rev_info *revs, struct parse_opt_ctx_t *ctx,
1996                        const struct option *options,
1997                        const char * const usagestr[])
1998{
1999        int n = handle_revision_opt(revs, ctx->argc, ctx->argv,
2000                                    &ctx->cpidx, ctx->out);
2001        if (n <= 0) {
2002                error("unknown option `%s'", ctx->argv[0]);
2003                usage_with_options(usagestr, options);
2004        }
2005        ctx->argv += n;
2006        ctx->argc -= n;
2007}
2008
2009static int for_each_bad_bisect_ref(const char *submodule, each_ref_fn fn, void *cb_data)
2010{
2011        return for_each_ref_in_submodule(submodule, "refs/bisect/bad", fn, cb_data);
2012}
2013
2014static int for_each_good_bisect_ref(const char *submodule, each_ref_fn fn, void *cb_data)
2015{
2016        return for_each_ref_in_submodule(submodule, "refs/bisect/good", fn, cb_data);
2017}
2018
2019static int handle_revision_pseudo_opt(const char *submodule,
2020                                struct rev_info *revs,
2021                                int argc, const char **argv, int *flags)
2022{
2023        const char *arg = argv[0];
2024        const char *optarg;
2025        int argcount;
2026
2027        /*
2028         * NOTE!
2029         *
2030         * Commands like "git shortlog" will not accept the options below
2031         * unless parse_revision_opt queues them (as opposed to erroring
2032         * out).
2033         *
2034         * When implementing your new pseudo-option, remember to
2035         * register it in the list at the top of handle_revision_opt.
2036         */
2037        if (!strcmp(arg, "--all")) {
2038                handle_refs(submodule, revs, *flags, for_each_ref_submodule);
2039                handle_refs(submodule, revs, *flags, head_ref_submodule);
2040                clear_ref_exclusion(&revs->ref_excludes);
2041        } else if (!strcmp(arg, "--branches")) {
2042                handle_refs(submodule, revs, *flags, for_each_branch_ref_submodule);
2043                clear_ref_exclusion(&revs->ref_excludes);
2044        } else if (!strcmp(arg, "--bisect")) {
2045                handle_refs(submodule, revs, *flags, for_each_bad_bisect_ref);
2046                handle_refs(submodule, revs, *flags ^ (UNINTERESTING | BOTTOM), for_each_good_bisect_ref);
2047                revs->bisect = 1;
2048        } else if (!strcmp(arg, "--tags")) {
2049                handle_refs(submodule, revs, *flags, for_each_tag_ref_submodule);
2050                clear_ref_exclusion(&revs->ref_excludes);
2051        } else if (!strcmp(arg, "--remotes")) {
2052                handle_refs(submodule, revs, *flags, for_each_remote_ref_submodule);
2053                clear_ref_exclusion(&revs->ref_excludes);
2054        } else if ((argcount = parse_long_opt("glob", argv, &optarg))) {
2055                struct all_refs_cb cb;
2056                init_all_refs_cb(&cb, revs, *flags);
2057                for_each_glob_ref(handle_one_ref, optarg, &cb);
2058                clear_ref_exclusion(&revs->ref_excludes);
2059                return argcount;
2060        } else if ((argcount = parse_long_opt("exclude", argv, &optarg))) {
2061                add_ref_exclusion(&revs->ref_excludes, optarg);
2062                return argcount;
2063        } else if (starts_with(arg, "--branches=")) {
2064                struct all_refs_cb cb;
2065                init_all_refs_cb(&cb, revs, *flags);
2066                for_each_glob_ref_in(handle_one_ref, arg + 11, "refs/heads/", &cb);
2067                clear_ref_exclusion(&revs->ref_excludes);
2068        } else if (starts_with(arg, "--tags=")) {
2069                struct all_refs_cb cb;
2070                init_all_refs_cb(&cb, revs, *flags);
2071                for_each_glob_ref_in(handle_one_ref, arg + 7, "refs/tags/", &cb);
2072                clear_ref_exclusion(&revs->ref_excludes);
2073        } else if (starts_with(arg, "--remotes=")) {
2074                struct all_refs_cb cb;
2075                init_all_refs_cb(&cb, revs, *flags);
2076                for_each_glob_ref_in(handle_one_ref, arg + 10, "refs/remotes/", &cb);
2077                clear_ref_exclusion(&revs->ref_excludes);
2078        } else if (!strcmp(arg, "--reflog")) {
2079                handle_reflog(revs, *flags);
2080        } else if (!strcmp(arg, "--not")) {
2081                *flags ^= UNINTERESTING | BOTTOM;
2082        } else if (!strcmp(arg, "--no-walk")) {
2083                revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2084        } else if (starts_with(arg, "--no-walk=")) {
2085                /*
2086                 * Detached form ("--no-walk X" as opposed to "--no-walk=X")
2087                 * not allowed, since the argument is optional.
2088                 */
2089                if (!strcmp(arg + 10, "sorted"))
2090                        revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2091                else if (!strcmp(arg + 10, "unsorted"))
2092                        revs->no_walk = REVISION_WALK_NO_WALK_UNSORTED;
2093                else
2094                        return error("invalid argument to --no-walk");
2095        } else if (!strcmp(arg, "--do-walk")) {
2096                revs->no_walk = 0;
2097        } else {
2098                return 0;
2099        }
2100
2101        return 1;
2102}
2103
2104/*
2105 * Parse revision information, filling in the "rev_info" structure,
2106 * and removing the used arguments from the argument list.
2107 *
2108 * Returns the number of arguments left that weren't recognized
2109 * (which are also moved to the head of the argument list)
2110 */
2111int setup_revisions(int argc, const char **argv, struct rev_info *revs, struct setup_revision_opt *opt)
2112{
2113        int i, flags, left, seen_dashdash, read_from_stdin, got_rev_arg = 0, revarg_opt;
2114        struct cmdline_pathspec prune_data;
2115        const char *submodule = NULL;
2116
2117        memset(&prune_data, 0, sizeof(prune_data));
2118        if (opt)
2119                submodule = opt->submodule;
2120
2121        /* First, search for "--" */
2122        if (opt && opt->assume_dashdash) {
2123                seen_dashdash = 1;
2124        } else {
2125                seen_dashdash = 0;
2126                for (i = 1; i < argc; i++) {
2127                        const char *arg = argv[i];
2128                        if (strcmp(arg, "--"))
2129                                continue;
2130                        argv[i] = NULL;
2131                        argc = i;
2132                        if (argv[i + 1])
2133                                append_prune_data(&prune_data, argv + i + 1);
2134                        seen_dashdash = 1;
2135                        break;
2136                }
2137        }
2138
2139        /* Second, deal with arguments and options */
2140        flags = 0;
2141        revarg_opt = opt ? opt->revarg_opt : 0;
2142        if (seen_dashdash)
2143                revarg_opt |= REVARG_CANNOT_BE_FILENAME;
2144        read_from_stdin = 0;
2145        for (left = i = 1; i < argc; i++) {
2146                const char *arg = argv[i];
2147                if (*arg == '-') {
2148                        int opts;
2149
2150                        opts = handle_revision_pseudo_opt(submodule,
2151                                                revs, argc - i, argv + i,
2152                                                &flags);
2153                        if (opts > 0) {
2154                                i += opts - 1;
2155                                continue;
2156                        }
2157
2158                        if (!strcmp(arg, "--stdin")) {
2159                                if (revs->disable_stdin) {
2160                                        argv[left++] = arg;
2161                                        continue;
2162                                }
2163                                if (read_from_stdin++)
2164                                        die("--stdin given twice?");
2165                                read_revisions_from_stdin(revs, &prune_data);
2166                                continue;
2167                        }
2168
2169                        opts = handle_revision_opt(revs, argc - i, argv + i, &left, argv);
2170                        if (opts > 0) {
2171                                i += opts - 1;
2172                                continue;
2173                        }
2174                        if (opts < 0)
2175                                exit(128);
2176                        continue;
2177                }
2178
2179
2180                if (handle_revision_arg(arg, revs, flags, revarg_opt)) {
2181                        int j;
2182                        if (seen_dashdash || *arg == '^')
2183                                die("bad revision '%s'", arg);
2184
2185                        /* If we didn't have a "--":
2186                         * (1) all filenames must exist;
2187                         * (2) all rev-args must not be interpretable
2188                         *     as a valid filename.
2189                         * but the latter we have checked in the main loop.
2190                         */
2191                        for (j = i; j < argc; j++)
2192                                verify_filename(revs->prefix, argv[j], j == i);
2193
2194                        append_prune_data(&prune_data, argv + i);
2195                        break;
2196                }
2197                else
2198                        got_rev_arg = 1;
2199        }
2200
2201        if (prune_data.nr) {
2202                /*
2203                 * If we need to introduce the magic "a lone ':' means no
2204                 * pathspec whatsoever", here is the place to do so.
2205                 *
2206                 * if (prune_data.nr == 1 && !strcmp(prune_data[0], ":")) {
2207                 *      prune_data.nr = 0;
2208                 *      prune_data.alloc = 0;
2209                 *      free(prune_data.path);
2210                 *      prune_data.path = NULL;
2211                 * } else {
2212                 *      terminate prune_data.alloc with NULL and
2213                 *      call init_pathspec() to set revs->prune_data here.
2214                 * }
2215                 */
2216                ALLOC_GROW(prune_data.path, prune_data.nr + 1, prune_data.alloc);
2217                prune_data.path[prune_data.nr++] = NULL;
2218                parse_pathspec(&revs->prune_data, 0, 0,
2219                               revs->prefix, prune_data.path);
2220        }
2221
2222        if (revs->def == NULL)
2223                revs->def = opt ? opt->def : NULL;
2224        if (opt && opt->tweak)
2225                opt->tweak(revs, opt);
2226        if (revs->show_merge)
2227                prepare_show_merge(revs);
2228        if (revs->def && !revs->pending.nr && !got_rev_arg) {
2229                unsigned char sha1[20];
2230                struct object *object;
2231                struct object_context oc;
2232                if (get_sha1_with_context(revs->def, 0, sha1, &oc))
2233                        die("bad default revision '%s'", revs->def);
2234                object = get_reference(revs, revs->def, sha1, 0);
2235                add_pending_object_with_mode(revs, object, revs->def, oc.mode);
2236        }
2237
2238        /* Did the user ask for any diff output? Run the diff! */
2239        if (revs->diffopt.output_format & ~DIFF_FORMAT_NO_OUTPUT)
2240                revs->diff = 1;
2241
2242        /* Pickaxe, diff-filter and rename following need diffs */
2243        if (revs->diffopt.pickaxe ||
2244            revs->diffopt.filter ||
2245            DIFF_OPT_TST(&revs->diffopt, FOLLOW_RENAMES))
2246                revs->diff = 1;
2247
2248        if (revs->topo_order)
2249                revs->limited = 1;
2250
2251        if (revs->prune_data.nr) {
2252                copy_pathspec(&revs->pruning.pathspec, &revs->prune_data);
2253                /* Can't prune commits with rename following: the paths change.. */
2254                if (!DIFF_OPT_TST(&revs->diffopt, FOLLOW_RENAMES))
2255                        revs->prune = 1;
2256                if (!revs->full_diff)
2257                        copy_pathspec(&revs->diffopt.pathspec,
2258                                      &revs->prune_data);
2259        }
2260        if (revs->combine_merges)
2261                revs->ignore_merges = 0;
2262        revs->diffopt.abbrev = revs->abbrev;
2263
2264        if (revs->line_level_traverse) {
2265                revs->limited = 1;
2266                revs->topo_order = 1;
2267        }
2268
2269        diff_setup_done(&revs->diffopt);
2270
2271        grep_commit_pattern_type(GREP_PATTERN_TYPE_UNSPECIFIED,
2272                                 &revs->grep_filter);
2273        compile_grep_patterns(&revs->grep_filter);
2274
2275        if (revs->reverse && revs->reflog_info)
2276                die("cannot combine --reverse with --walk-reflogs");
2277        if (revs->rewrite_parents && revs->children.name)
2278                die("cannot combine --parents and --children");
2279
2280        /*
2281         * Limitations on the graph functionality
2282         */
2283        if (revs->reverse && revs->graph)
2284                die("cannot combine --reverse with --graph");
2285
2286        if (revs->reflog_info && revs->graph)
2287                die("cannot combine --walk-reflogs with --graph");
2288        if (!revs->reflog_info && revs->grep_filter.use_reflog_filter)
2289                die("cannot use --grep-reflog without --walk-reflogs");
2290
2291        return left;
2292}
2293
2294static void add_child(struct rev_info *revs, struct commit *parent, struct commit *child)
2295{
2296        struct commit_list *l = xcalloc(1, sizeof(*l));
2297
2298        l->item = child;
2299        l->next = add_decoration(&revs->children, &parent->object, l);
2300}
2301
2302static int remove_duplicate_parents(struct rev_info *revs, struct commit *commit)
2303{
2304        struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
2305        struct commit_list **pp, *p;
2306        int surviving_parents;
2307
2308        /* Examine existing parents while marking ones we have seen... */
2309        pp = &commit->parents;
2310        surviving_parents = 0;
2311        while ((p = *pp) != NULL) {
2312                struct commit *parent = p->item;
2313                if (parent->object.flags & TMP_MARK) {
2314                        *pp = p->next;
2315                        if (ts)
2316                                compact_treesame(revs, commit, surviving_parents);
2317                        continue;
2318                }
2319                parent->object.flags |= TMP_MARK;
2320                surviving_parents++;
2321                pp = &p->next;
2322        }
2323        /* clear the temporary mark */
2324        for (p = commit->parents; p; p = p->next) {
2325                p->item->object.flags &= ~TMP_MARK;
2326        }
2327        /* no update_treesame() - removing duplicates can't affect TREESAME */
2328        return surviving_parents;
2329}
2330
2331struct merge_simplify_state {
2332        struct commit *simplified;
2333};
2334
2335static struct merge_simplify_state *locate_simplify_state(struct rev_info *revs, struct commit *commit)
2336{
2337        struct merge_simplify_state *st;
2338
2339        st = lookup_decoration(&revs->merge_simplification, &commit->object);
2340        if (!st) {
2341                st = xcalloc(1, sizeof(*st));
2342                add_decoration(&revs->merge_simplification, &commit->object, st);
2343        }
2344        return st;
2345}
2346
2347static int mark_redundant_parents(struct rev_info *revs, struct commit *commit)
2348{
2349        struct commit_list *h = reduce_heads(commit->parents);
2350        int i = 0, marked = 0;
2351        struct commit_list *po, *pn;
2352
2353        /* Want these for sanity-checking only */
2354        int orig_cnt = commit_list_count(commit->parents);
2355        int cnt = commit_list_count(h);
2356
2357        /*
2358         * Not ready to remove items yet, just mark them for now, based
2359         * on the output of reduce_heads(). reduce_heads outputs the reduced
2360         * set in its original order, so this isn't too hard.
2361         */
2362        po = commit->parents;
2363        pn = h;
2364        while (po) {
2365                if (pn && po->item == pn->item) {
2366                        pn = pn->next;
2367                        i++;
2368                } else {
2369                        po->item->object.flags |= TMP_MARK;
2370                        marked++;
2371                }
2372                po=po->next;
2373        }
2374
2375        if (i != cnt || cnt+marked != orig_cnt)
2376                die("mark_redundant_parents %d %d %d %d", orig_cnt, cnt, i, marked);
2377
2378        free_commit_list(h);
2379
2380        return marked;
2381}
2382
2383static int mark_treesame_root_parents(struct rev_info *revs, struct commit *commit)
2384{
2385        struct commit_list *p;
2386        int marked = 0;
2387
2388        for (p = commit->parents; p; p = p->next) {
2389                struct commit *parent = p->item;
2390                if (!parent->parents && (parent->object.flags & TREESAME)) {
2391                        parent->object.flags |= TMP_MARK;
2392                        marked++;
2393                }
2394        }
2395
2396        return marked;
2397}
2398
2399/*
2400 * Awkward naming - this means one parent we are TREESAME to.
2401 * cf mark_treesame_root_parents: root parents that are TREESAME (to an
2402 * empty tree). Better name suggestions?
2403 */
2404static int leave_one_treesame_to_parent(struct rev_info *revs, struct commit *commit)
2405{
2406        struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
2407        struct commit *unmarked = NULL, *marked = NULL;
2408        struct commit_list *p;
2409        unsigned n;
2410
2411        for (p = commit->parents, n = 0; p; p = p->next, n++) {
2412                if (ts->treesame[n]) {
2413                        if (p->item->object.flags & TMP_MARK) {
2414                                if (!marked)
2415                                        marked = p->item;
2416                        } else {
2417                                if (!unmarked) {
2418                                        unmarked = p->item;
2419                                        break;
2420                                }
2421                        }
2422                }
2423        }
2424
2425        /*
2426         * If we are TREESAME to a marked-for-deletion parent, but not to any
2427         * unmarked parents, unmark the first TREESAME parent. This is the
2428         * parent that the default simplify_history==1 scan would have followed,
2429         * and it doesn't make sense to omit that path when asking for a
2430         * simplified full history. Retaining it improves the chances of
2431         * understanding odd missed merges that took an old version of a file.
2432         *
2433         * Example:
2434         *
2435         *   I--------*X       A modified the file, but mainline merge X used
2436         *    \       /        "-s ours", so took the version from I. X is
2437         *     `-*A--'         TREESAME to I and !TREESAME to A.
2438         *
2439         * Default log from X would produce "I". Without this check,
2440         * --full-history --simplify-merges would produce "I-A-X", showing
2441         * the merge commit X and that it changed A, but not making clear that
2442         * it had just taken the I version. With this check, the topology above
2443         * is retained.
2444         *
2445         * Note that it is possible that the simplification chooses a different
2446         * TREESAME parent from the default, in which case this test doesn't
2447         * activate, and we _do_ drop the default parent. Example:
2448         *
2449         *   I------X         A modified the file, but it was reverted in B,
2450         *    \    /          meaning mainline merge X is TREESAME to both
2451         *    *A-*B           parents.
2452         *
2453         * Default log would produce "I" by following the first parent;
2454         * --full-history --simplify-merges will produce "I-A-B". But this is a
2455         * reasonable result - it presents a logical full history leading from
2456         * I to X, and X is not an important merge.
2457         */
2458        if (!unmarked && marked) {
2459                marked->object.flags &= ~TMP_MARK;
2460                return 1;
2461        }
2462
2463        return 0;
2464}
2465
2466static int remove_marked_parents(struct rev_info *revs, struct commit *commit)
2467{
2468        struct commit_list **pp, *p;
2469        int nth_parent, removed = 0;
2470
2471        pp = &commit->parents;
2472        nth_parent = 0;
2473        while ((p = *pp) != NULL) {
2474                struct commit *parent = p->item;
2475                if (parent->object.flags & TMP_MARK) {
2476                        parent->object.flags &= ~TMP_MARK;
2477                        *pp = p->next;
2478                        free(p);
2479                        removed++;
2480                        compact_treesame(revs, commit, nth_parent);
2481                        continue;
2482                }
2483                pp = &p->next;
2484                nth_parent++;
2485        }
2486
2487        /* Removing parents can only increase TREESAMEness */
2488        if (removed && !(commit->object.flags & TREESAME))
2489                update_treesame(revs, commit);
2490
2491        return nth_parent;
2492}
2493
2494static struct commit_list **simplify_one(struct rev_info *revs, struct commit *commit, struct commit_list **tail)
2495{
2496        struct commit_list *p;
2497        struct commit *parent;
2498        struct merge_simplify_state *st, *pst;
2499        int cnt;
2500
2501        st = locate_simplify_state(revs, commit);
2502
2503        /*
2504         * Have we handled this one?
2505         */
2506        if (st->simplified)
2507                return tail;
2508
2509        /*
2510         * An UNINTERESTING commit simplifies to itself, so does a
2511         * root commit.  We do not rewrite parents of such commit
2512         * anyway.
2513         */
2514        if ((commit->object.flags & UNINTERESTING) || !commit->parents) {
2515                st->simplified = commit;
2516                return tail;
2517        }
2518
2519        /*
2520         * Do we know what commit all of our parents that matter
2521         * should be rewritten to?  Otherwise we are not ready to
2522         * rewrite this one yet.
2523         */
2524        for (cnt = 0, p = commit->parents; p; p = p->next) {
2525                pst = locate_simplify_state(revs, p->item);
2526                if (!pst->simplified) {
2527                        tail = &commit_list_insert(p->item, tail)->next;
2528                        cnt++;
2529                }
2530                if (revs->first_parent_only)
2531                        break;
2532        }
2533        if (cnt) {
2534                tail = &commit_list_insert(commit, tail)->next;
2535                return tail;
2536        }
2537
2538        /*
2539         * Rewrite our list of parents. Note that this cannot
2540         * affect our TREESAME flags in any way - a commit is
2541         * always TREESAME to its simplification.
2542         */
2543        for (p = commit->parents; p; p = p->next) {
2544                pst = locate_simplify_state(revs, p->item);
2545                p->item = pst->simplified;
2546                if (revs->first_parent_only)
2547                        break;
2548        }
2549
2550        if (revs->first_parent_only)
2551                cnt = 1;
2552        else
2553                cnt = remove_duplicate_parents(revs, commit);
2554
2555        /*
2556         * It is possible that we are a merge and one side branch
2557         * does not have any commit that touches the given paths;
2558         * in such a case, the immediate parent from that branch
2559         * will be rewritten to be the merge base.
2560         *
2561         *      o----X          X: the commit we are looking at;
2562         *     /    /           o: a commit that touches the paths;
2563         * ---o----'
2564         *
2565         * Further, a merge of an independent branch that doesn't
2566         * touch the path will reduce to a treesame root parent:
2567         *
2568         *  ----o----X          X: the commit we are looking at;
2569         *          /           o: a commit that touches the paths;
2570         *         r            r: a root commit not touching the paths
2571         *
2572         * Detect and simplify both cases.
2573         */
2574        if (1 < cnt) {
2575                int marked = mark_redundant_parents(revs, commit);
2576                marked += mark_treesame_root_parents(revs, commit);
2577                if (marked)
2578                        marked -= leave_one_treesame_to_parent(revs, commit);
2579                if (marked)
2580                        cnt = remove_marked_parents(revs, commit);
2581        }
2582
2583        /*
2584         * A commit simplifies to itself if it is a root, if it is
2585         * UNINTERESTING, if it touches the given paths, or if it is a
2586         * merge and its parents don't simplify to one relevant commit
2587         * (the first two cases are already handled at the beginning of
2588         * this function).
2589         *
2590         * Otherwise, it simplifies to what its sole relevant parent
2591         * simplifies to.
2592         */
2593        if (!cnt ||
2594            (commit->object.flags & UNINTERESTING) ||
2595            !(commit->object.flags & TREESAME) ||
2596            (parent = one_relevant_parent(revs, commit->parents)) == NULL)
2597                st->simplified = commit;
2598        else {
2599                pst = locate_simplify_state(revs, parent);
2600                st->simplified = pst->simplified;
2601        }
2602        return tail;
2603}
2604
2605static void simplify_merges(struct rev_info *revs)
2606{
2607        struct commit_list *list, *next;
2608        struct commit_list *yet_to_do, **tail;
2609        struct commit *commit;
2610
2611        if (!revs->prune)
2612                return;
2613
2614        /* feed the list reversed */
2615        yet_to_do = NULL;
2616        for (list = revs->commits; list; list = next) {
2617                commit = list->item;
2618                next = list->next;
2619                /*
2620                 * Do not free(list) here yet; the original list
2621                 * is used later in this function.
2622                 */
2623                commit_list_insert(commit, &yet_to_do);
2624        }
2625        while (yet_to_do) {
2626                list = yet_to_do;
2627                yet_to_do = NULL;
2628                tail = &yet_to_do;
2629                while (list) {
2630                        commit = list->item;
2631                        next = list->next;
2632                        free(list);
2633                        list = next;
2634                        tail = simplify_one(revs, commit, tail);
2635                }
2636        }
2637
2638        /* clean up the result, removing the simplified ones */
2639        list = revs->commits;
2640        revs->commits = NULL;
2641        tail = &revs->commits;
2642        while (list) {
2643                struct merge_simplify_state *st;
2644
2645                commit = list->item;
2646                next = list->next;
2647                free(list);
2648                list = next;
2649                st = locate_simplify_state(revs, commit);
2650                if (st->simplified == commit)
2651                        tail = &commit_list_insert(commit, tail)->next;
2652        }
2653}
2654
2655static void set_children(struct rev_info *revs)
2656{
2657        struct commit_list *l;
2658        for (l = revs->commits; l; l = l->next) {
2659                struct commit *commit = l->item;
2660                struct commit_list *p;
2661
2662                for (p = commit->parents; p; p = p->next)
2663                        add_child(revs, p->item, commit);
2664        }
2665}
2666
2667void reset_revision_walk(void)
2668{
2669        clear_object_flags(SEEN | ADDED | SHOWN);
2670}
2671
2672int prepare_revision_walk(struct rev_info *revs)
2673{
2674        int nr = revs->pending.nr;
2675        struct object_array_entry *e, *list;
2676        struct commit_list **next = &revs->commits;
2677
2678        e = list = revs->pending.objects;
2679        revs->pending.nr = 0;
2680        revs->pending.alloc = 0;
2681        revs->pending.objects = NULL;
2682        while (--nr >= 0) {
2683                struct commit *commit = handle_commit(revs, e->item, e->name);
2684                if (commit) {
2685                        if (!(commit->object.flags & SEEN)) {
2686                                commit->object.flags |= SEEN;
2687                                next = commit_list_append(commit, next);
2688                        }
2689                }
2690                e++;
2691        }
2692        if (!revs->leak_pending)
2693                free(list);
2694
2695        /* Signal whether we need per-parent treesame decoration */
2696        if (revs->simplify_merges ||
2697            (revs->limited && limiting_can_increase_treesame(revs)))
2698                revs->treesame.name = "treesame";
2699
2700        if (revs->no_walk != REVISION_WALK_NO_WALK_UNSORTED)
2701                commit_list_sort_by_date(&revs->commits);
2702        if (revs->no_walk)
2703                return 0;
2704        if (revs->limited)
2705                if (limit_list(revs) < 0)
2706                        return -1;
2707        if (revs->topo_order)
2708                sort_in_topological_order(&revs->commits, revs->sort_order);
2709        if (revs->line_level_traverse)
2710                line_log_filter(revs);
2711        if (revs->simplify_merges)
2712                simplify_merges(revs);
2713        if (revs->children.name)
2714                set_children(revs);
2715        return 0;
2716}
2717
2718static enum rewrite_result rewrite_one(struct rev_info *revs, struct commit **pp)
2719{
2720        struct commit_list *cache = NULL;
2721
2722        for (;;) {
2723                struct commit *p = *pp;
2724                if (!revs->limited)
2725                        if (add_parents_to_list(revs, p, &revs->commits, &cache) < 0)
2726                                return rewrite_one_error;
2727                if (p->object.flags & UNINTERESTING)
2728                        return rewrite_one_ok;
2729                if (!(p->object.flags & TREESAME))
2730                        return rewrite_one_ok;
2731                if (!p->parents)
2732                        return rewrite_one_noparents;
2733                if ((p = one_relevant_parent(revs, p->parents)) == NULL)
2734                        return rewrite_one_ok;
2735                *pp = p;
2736        }
2737}
2738
2739int rewrite_parents(struct rev_info *revs, struct commit *commit,
2740        rewrite_parent_fn_t rewrite_parent)
2741{
2742        struct commit_list **pp = &commit->parents;
2743        while (*pp) {
2744                struct commit_list *parent = *pp;
2745                switch (rewrite_parent(revs, &parent->item)) {
2746                case rewrite_one_ok:
2747                        break;
2748                case rewrite_one_noparents:
2749                        *pp = parent->next;
2750                        continue;
2751                case rewrite_one_error:
2752                        return -1;
2753                }
2754                pp = &parent->next;
2755        }
2756        remove_duplicate_parents(revs, commit);
2757        return 0;
2758}
2759
2760static int commit_rewrite_person(struct strbuf *buf, const char *what, struct string_list *mailmap)
2761{
2762        char *person, *endp;
2763        size_t len, namelen, maillen;
2764        const char *name;
2765        const char *mail;
2766        struct ident_split ident;
2767
2768        person = strstr(buf->buf, what);
2769        if (!person)
2770                return 0;
2771
2772        person += strlen(what);
2773        endp = strchr(person, '\n');
2774        if (!endp)
2775                return 0;
2776
2777        len = endp - person;
2778
2779        if (split_ident_line(&ident, person, len))
2780                return 0;
2781
2782        mail = ident.mail_begin;
2783        maillen = ident.mail_end - ident.mail_begin;
2784        name = ident.name_begin;
2785        namelen = ident.name_end - ident.name_begin;
2786
2787        if (map_user(mailmap, &mail, &maillen, &name, &namelen)) {
2788                struct strbuf namemail = STRBUF_INIT;
2789
2790                strbuf_addf(&namemail, "%.*s <%.*s>",
2791                            (int)namelen, name, (int)maillen, mail);
2792
2793                strbuf_splice(buf, ident.name_begin - buf->buf,
2794                              ident.mail_end - ident.name_begin + 1,
2795                              namemail.buf, namemail.len);
2796
2797                strbuf_release(&namemail);
2798
2799                return 1;
2800        }
2801
2802        return 0;
2803}
2804
2805static int commit_match(struct commit *commit, struct rev_info *opt)
2806{
2807        int retval;
2808        const char *encoding;
2809        const char *message;
2810        struct strbuf buf = STRBUF_INIT;
2811
2812        if (!opt->grep_filter.pattern_list && !opt->grep_filter.header_list)
2813                return 1;
2814
2815        /* Prepend "fake" headers as needed */
2816        if (opt->grep_filter.use_reflog_filter) {
2817                strbuf_addstr(&buf, "reflog ");
2818                get_reflog_message(&buf, opt->reflog_info);
2819                strbuf_addch(&buf, '\n');
2820        }
2821
2822        /*
2823         * We grep in the user's output encoding, under the assumption that it
2824         * is the encoding they are most likely to write their grep pattern
2825         * for. In addition, it means we will match the "notes" encoding below,
2826         * so we will not end up with a buffer that has two different encodings
2827         * in it.
2828         */
2829        encoding = get_log_output_encoding();
2830        message = logmsg_reencode(commit, NULL, encoding);
2831
2832        /* Copy the commit to temporary if we are using "fake" headers */
2833        if (buf.len)
2834                strbuf_addstr(&buf, message);
2835
2836        if (opt->grep_filter.header_list && opt->mailmap) {
2837                if (!buf.len)
2838                        strbuf_addstr(&buf, message);
2839
2840                commit_rewrite_person(&buf, "\nauthor ", opt->mailmap);
2841                commit_rewrite_person(&buf, "\ncommitter ", opt->mailmap);
2842        }
2843
2844        /* Append "fake" message parts as needed */
2845        if (opt->show_notes) {
2846                if (!buf.len)
2847                        strbuf_addstr(&buf, message);
2848                format_display_notes(commit->object.sha1, &buf, encoding, 1);
2849        }
2850
2851        /*
2852         * Find either in the original commit message, or in the temporary.
2853         * Note that we cast away the constness of "message" here. It is
2854         * const because it may come from the cached commit buffer. That's OK,
2855         * because we know that it is modifiable heap memory, and that while
2856         * grep_buffer may modify it for speed, it will restore any
2857         * changes before returning.
2858         */
2859        if (buf.len)
2860                retval = grep_buffer(&opt->grep_filter, buf.buf, buf.len);
2861        else
2862                retval = grep_buffer(&opt->grep_filter,
2863                                     (char *)message, strlen(message));
2864        strbuf_release(&buf);
2865        unuse_commit_buffer(commit, message);
2866        return retval;
2867}
2868
2869static inline int want_ancestry(const struct rev_info *revs)
2870{
2871        return (revs->rewrite_parents || revs->children.name);
2872}
2873
2874enum commit_action get_commit_action(struct rev_info *revs, struct commit *commit)
2875{
2876        if (commit->object.flags & SHOWN)
2877                return commit_ignore;
2878        if (revs->unpacked && has_sha1_pack(commit->object.sha1))
2879                return commit_ignore;
2880        if (revs->show_all)
2881                return commit_show;
2882        if (commit->object.flags & UNINTERESTING)
2883                return commit_ignore;
2884        if (revs->min_age != -1 && (commit->date > revs->min_age))
2885                return commit_ignore;
2886        if (revs->min_parents || (revs->max_parents >= 0)) {
2887                int n = commit_list_count(commit->parents);
2888                if ((n < revs->min_parents) ||
2889                    ((revs->max_parents >= 0) && (n > revs->max_parents)))
2890                        return commit_ignore;
2891        }
2892        if (!commit_match(commit, revs))
2893                return commit_ignore;
2894        if (revs->prune && revs->dense) {
2895                /* Commit without changes? */
2896                if (commit->object.flags & TREESAME) {
2897                        int n;
2898                        struct commit_list *p;
2899                        /* drop merges unless we want parenthood */
2900                        if (!want_ancestry(revs))
2901                                return commit_ignore;
2902                        /*
2903                         * If we want ancestry, then need to keep any merges
2904                         * between relevant commits to tie together topology.
2905                         * For consistency with TREESAME and simplification
2906                         * use "relevant" here rather than just INTERESTING,
2907                         * to treat bottom commit(s) as part of the topology.
2908                         */
2909                        for (n = 0, p = commit->parents; p; p = p->next)
2910                                if (relevant_commit(p->item))
2911                                        if (++n >= 2)
2912                                                return commit_show;
2913                        return commit_ignore;
2914                }
2915        }
2916        return commit_show;
2917}
2918
2919enum commit_action simplify_commit(struct rev_info *revs, struct commit *commit)
2920{
2921        enum commit_action action = get_commit_action(revs, commit);
2922
2923        if (action == commit_show &&
2924            !revs->show_all &&
2925            revs->prune && revs->dense && want_ancestry(revs)) {
2926                /*
2927                 * --full-diff on simplified parents is no good: it
2928                 * will show spurious changes from the commits that
2929                 * were elided.  So we save the parents on the side
2930                 * when --full-diff is in effect.
2931                 */
2932                if (revs->full_diff)
2933                        save_parents(revs, commit);
2934                if (rewrite_parents(revs, commit, rewrite_one) < 0)
2935                        return commit_error;
2936        }
2937        return action;
2938}
2939
2940static void track_linear(struct rev_info *revs, struct commit *commit)
2941{
2942        if (revs->track_first_time) {
2943                revs->linear = 1;
2944                revs->track_first_time = 0;
2945        } else {
2946                struct commit_list *p;
2947                for (p = revs->previous_parents; p; p = p->next)
2948                        if (p->item == NULL || /* first commit */
2949                            !hashcmp(p->item->object.sha1, commit->object.sha1))
2950                                break;
2951                revs->linear = p != NULL;
2952        }
2953        if (revs->reverse) {
2954                if (revs->linear)
2955                        commit->object.flags |= TRACK_LINEAR;
2956        }
2957        free_commit_list(revs->previous_parents);
2958        revs->previous_parents = copy_commit_list(commit->parents);
2959}
2960
2961static struct commit *get_revision_1(struct rev_info *revs)
2962{
2963        if (!revs->commits)
2964                return NULL;
2965
2966        do {
2967                struct commit_list *entry = revs->commits;
2968                struct commit *commit = entry->item;
2969
2970                revs->commits = entry->next;
2971                free(entry);
2972
2973                if (revs->reflog_info) {
2974                        save_parents(revs, commit);
2975                        fake_reflog_parent(revs->reflog_info, commit);
2976                        commit->object.flags &= ~(ADDED | SEEN | SHOWN);
2977                }
2978
2979                /*
2980                 * If we haven't done the list limiting, we need to look at
2981                 * the parents here. We also need to do the date-based limiting
2982                 * that we'd otherwise have done in limit_list().
2983                 */
2984                if (!revs->limited) {
2985                        if (revs->max_age != -1 &&
2986                            (commit->date < revs->max_age))
2987                                continue;
2988                        if (add_parents_to_list(revs, commit, &revs->commits, NULL) < 0) {
2989                                if (!revs->ignore_missing_links)
2990                                        die("Failed to traverse parents of commit %s",
2991                                                sha1_to_hex(commit->object.sha1));
2992                        }
2993                }
2994
2995                switch (simplify_commit(revs, commit)) {
2996                case commit_ignore:
2997                        continue;
2998                case commit_error:
2999                        die("Failed to simplify parents of commit %s",
3000                            sha1_to_hex(commit->object.sha1));
3001                default:
3002                        if (revs->track_linear)
3003                                track_linear(revs, commit);
3004                        return commit;
3005                }
3006        } while (revs->commits);
3007        return NULL;
3008}
3009
3010/*
3011 * Return true for entries that have not yet been shown.  (This is an
3012 * object_array_each_func_t.)
3013 */
3014static int entry_unshown(struct object_array_entry *entry, void *cb_data_unused)
3015{
3016        return !(entry->item->flags & SHOWN);
3017}
3018
3019/*
3020 * If array is on the verge of a realloc, garbage-collect any entries
3021 * that have already been shown to try to free up some space.
3022 */
3023static void gc_boundary(struct object_array *array)
3024{
3025        if (array->nr == array->alloc)
3026                object_array_filter(array, entry_unshown, NULL);
3027}
3028
3029static void create_boundary_commit_list(struct rev_info *revs)
3030{
3031        unsigned i;
3032        struct commit *c;
3033        struct object_array *array = &revs->boundary_commits;
3034        struct object_array_entry *objects = array->objects;
3035
3036        /*
3037         * If revs->commits is non-NULL at this point, an error occurred in
3038         * get_revision_1().  Ignore the error and continue printing the
3039         * boundary commits anyway.  (This is what the code has always
3040         * done.)
3041         */
3042        if (revs->commits) {
3043                free_commit_list(revs->commits);
3044                revs->commits = NULL;
3045        }
3046
3047        /*
3048         * Put all of the actual boundary commits from revs->boundary_commits
3049         * into revs->commits
3050         */
3051        for (i = 0; i < array->nr; i++) {
3052                c = (struct commit *)(objects[i].item);
3053                if (!c)
3054                        continue;
3055                if (!(c->object.flags & CHILD_SHOWN))
3056                        continue;
3057                if (c->object.flags & (SHOWN | BOUNDARY))
3058                        continue;
3059                c->object.flags |= BOUNDARY;
3060                commit_list_insert(c, &revs->commits);
3061        }
3062
3063        /*
3064         * If revs->topo_order is set, sort the boundary commits
3065         * in topological order
3066         */
3067        sort_in_topological_order(&revs->commits, revs->sort_order);
3068}
3069
3070static struct commit *get_revision_internal(struct rev_info *revs)
3071{
3072        struct commit *c = NULL;
3073        struct commit_list *l;
3074
3075        if (revs->boundary == 2) {
3076                /*
3077                 * All of the normal commits have already been returned,
3078                 * and we are now returning boundary commits.
3079                 * create_boundary_commit_list() has populated
3080                 * revs->commits with the remaining commits to return.
3081                 */
3082                c = pop_commit(&revs->commits);
3083                if (c)
3084                        c->object.flags |= SHOWN;
3085                return c;
3086        }
3087
3088        /*
3089         * If our max_count counter has reached zero, then we are done. We
3090         * don't simply return NULL because we still might need to show
3091         * boundary commits. But we want to avoid calling get_revision_1, which
3092         * might do a considerable amount of work finding the next commit only
3093         * for us to throw it away.
3094         *
3095         * If it is non-zero, then either we don't have a max_count at all
3096         * (-1), or it is still counting, in which case we decrement.
3097         */
3098        if (revs->max_count) {
3099                c = get_revision_1(revs);
3100                if (c) {
3101                        while (revs->skip_count > 0) {
3102                                revs->skip_count--;
3103                                c = get_revision_1(revs);
3104                                if (!c)
3105                                        break;
3106                        }
3107                }
3108
3109                if (revs->max_count > 0)
3110                        revs->max_count--;
3111        }
3112
3113        if (c)
3114                c->object.flags |= SHOWN;
3115
3116        if (!revs->boundary)
3117                return c;
3118
3119        if (!c) {
3120                /*
3121                 * get_revision_1() runs out the commits, and
3122                 * we are done computing the boundaries.
3123                 * switch to boundary commits output mode.
3124                 */
3125                revs->boundary = 2;
3126
3127                /*
3128                 * Update revs->commits to contain the list of
3129                 * boundary commits.
3130                 */
3131                create_boundary_commit_list(revs);
3132
3133                return get_revision_internal(revs);
3134        }
3135
3136        /*
3137         * boundary commits are the commits that are parents of the
3138         * ones we got from get_revision_1() but they themselves are
3139         * not returned from get_revision_1().  Before returning
3140         * 'c', we need to mark its parents that they could be boundaries.
3141         */
3142
3143        for (l = c->parents; l; l = l->next) {
3144                struct object *p;
3145                p = &(l->item->object);
3146                if (p->flags & (CHILD_SHOWN | SHOWN))
3147                        continue;
3148                p->flags |= CHILD_SHOWN;
3149                gc_boundary(&revs->boundary_commits);
3150                add_object_array(p, NULL, &revs->boundary_commits);
3151        }
3152
3153        return c;
3154}
3155
3156struct commit *get_revision(struct rev_info *revs)
3157{
3158        struct commit *c;
3159        struct commit_list *reversed;
3160
3161        if (revs->reverse) {
3162                reversed = NULL;
3163                while ((c = get_revision_internal(revs)))
3164                        commit_list_insert(c, &reversed);
3165                revs->commits = reversed;
3166                revs->reverse = 0;
3167                revs->reverse_output_stage = 1;
3168        }
3169
3170        if (revs->reverse_output_stage) {
3171                c = pop_commit(&revs->commits);
3172                if (revs->track_linear)
3173                        revs->linear = !!(c && c->object.flags & TRACK_LINEAR);
3174                return c;
3175        }
3176
3177        c = get_revision_internal(revs);
3178        if (c && revs->graph)
3179                graph_update(revs->graph, c);
3180        if (!c) {
3181                free_saved_parents(revs);
3182                if (revs->previous_parents) {
3183                        free_commit_list(revs->previous_parents);
3184                        revs->previous_parents = NULL;
3185                }
3186        }
3187        return c;
3188}
3189
3190char *get_revision_mark(const struct rev_info *revs, const struct commit *commit)
3191{
3192        if (commit->object.flags & BOUNDARY)
3193                return "-";
3194        else if (commit->object.flags & UNINTERESTING)
3195                return "^";
3196        else if (commit->object.flags & PATCHSAME)
3197                return "=";
3198        else if (!revs || revs->left_right) {
3199                if (commit->object.flags & SYMMETRIC_LEFT)
3200                        return "<";
3201                else
3202                        return ">";
3203        } else if (revs->graph)
3204                return "*";
3205        else if (revs->cherry_mark)
3206                return "+";
3207        return "";
3208}
3209
3210void put_revision_mark(const struct rev_info *revs, const struct commit *commit)
3211{
3212        char *mark = get_revision_mark(revs, commit);
3213        if (!strlen(mark))
3214                return;
3215        fputs(mark, stdout);
3216        putchar(' ');
3217}
3218
3219define_commit_slab(saved_parents, struct commit_list *);
3220
3221#define EMPTY_PARENT_LIST ((struct commit_list *)-1)
3222
3223void save_parents(struct rev_info *revs, struct commit *commit)
3224{
3225        struct commit_list **pp;
3226
3227        if (!revs->saved_parents_slab) {
3228                revs->saved_parents_slab = xmalloc(sizeof(struct saved_parents));
3229                init_saved_parents(revs->saved_parents_slab);
3230        }
3231
3232        pp = saved_parents_at(revs->saved_parents_slab, commit);
3233
3234        /*
3235         * When walking with reflogs, we may visit the same commit
3236         * several times: once for each appearance in the reflog.
3237         *
3238         * In this case, save_parents() will be called multiple times.
3239         * We want to keep only the first set of parents.  We need to
3240         * store a sentinel value for an empty (i.e., NULL) parent
3241         * list to distinguish it from a not-yet-saved list, however.
3242         */
3243        if (*pp)
3244                return;
3245        if (commit->parents)
3246                *pp = copy_commit_list(commit->parents);
3247        else
3248                *pp = EMPTY_PARENT_LIST;
3249}
3250
3251struct commit_list *get_saved_parents(struct rev_info *revs, const struct commit *commit)
3252{
3253        struct commit_list *parents;
3254
3255        if (!revs->saved_parents_slab)
3256                return commit->parents;
3257
3258        parents = *saved_parents_at(revs->saved_parents_slab, commit);
3259        if (parents == EMPTY_PARENT_LIST)
3260                return NULL;
3261        return parents;
3262}
3263
3264void free_saved_parents(struct rev_info *revs)
3265{
3266        if (revs->saved_parents_slab)
3267                clear_saved_parents(revs->saved_parents_slab);
3268}