unpack-trees.con commit general improvements (43abf13)
   1#include "cache.h"
   2#include "argv-array.h"
   3#include "repository.h"
   4#include "config.h"
   5#include "dir.h"
   6#include "tree.h"
   7#include "tree-walk.h"
   8#include "cache-tree.h"
   9#include "unpack-trees.h"
  10#include "progress.h"
  11#include "refs.h"
  12#include "attr.h"
  13#include "split-index.h"
  14#include "dir.h"
  15#include "submodule.h"
  16#include "submodule-config.h"
  17#include "fsmonitor.h"
  18#include "object-store.h"
  19#include "promisor-remote.h"
  20
  21/*
  22 * Error messages expected by scripts out of plumbing commands such as
  23 * read-tree.  Non-scripted Porcelain is not required to use these messages
  24 * and in fact are encouraged to reword them to better suit their particular
  25 * situation better.  See how "git checkout" and "git merge" replaces
  26 * them using setup_unpack_trees_porcelain(), for example.
  27 */
  28static const char *unpack_plumbing_errors[NB_UNPACK_TREES_ERROR_TYPES] = {
  29        /* ERROR_WOULD_OVERWRITE */
  30        "Entry '%s' would be overwritten by merge. Cannot merge.",
  31
  32        /* ERROR_NOT_UPTODATE_FILE */
  33        "Entry '%s' not uptodate. Cannot merge.",
  34
  35        /* ERROR_NOT_UPTODATE_DIR */
  36        "Updating '%s' would lose untracked files in it",
  37
  38        /* ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN */
  39        "Untracked working tree file '%s' would be overwritten by merge.",
  40
  41        /* ERROR_WOULD_LOSE_UNTRACKED_REMOVED */
  42        "Untracked working tree file '%s' would be removed by merge.",
  43
  44        /* ERROR_BIND_OVERLAP */
  45        "Entry '%s' overlaps with '%s'.  Cannot bind.",
  46
  47        /* ERROR_SPARSE_NOT_UPTODATE_FILE */
  48        "Entry '%s' not uptodate. Cannot update sparse checkout.",
  49
  50        /* ERROR_WOULD_LOSE_ORPHANED_OVERWRITTEN */
  51        "Working tree file '%s' would be overwritten by sparse checkout update.",
  52
  53        /* ERROR_WOULD_LOSE_ORPHANED_REMOVED */
  54        "Working tree file '%s' would be removed by sparse checkout update.",
  55
  56        /* ERROR_WOULD_LOSE_SUBMODULE */
  57        "Submodule '%s' cannot checkout new HEAD.",
  58};
  59
  60#define ERRORMSG(o,type) \
  61        ( ((o) && (o)->msgs[(type)]) \
  62          ? ((o)->msgs[(type)])      \
  63          : (unpack_plumbing_errors[(type)]) )
  64
  65static const char *super_prefixed(const char *path)
  66{
  67        /*
  68         * It is necessary and sufficient to have two static buffers
  69         * here, as the return value of this function is fed to
  70         * error() using the unpack_*_errors[] templates we see above.
  71         */
  72        static struct strbuf buf[2] = {STRBUF_INIT, STRBUF_INIT};
  73        static int super_prefix_len = -1;
  74        static unsigned idx = ARRAY_SIZE(buf) - 1;
  75
  76        if (super_prefix_len < 0) {
  77                const char *super_prefix = get_super_prefix();
  78                if (!super_prefix) {
  79                        super_prefix_len = 0;
  80                } else {
  81                        int i;
  82                        for (i = 0; i < ARRAY_SIZE(buf); i++)
  83                                strbuf_addstr(&buf[i], super_prefix);
  84                        super_prefix_len = buf[0].len;
  85                }
  86        }
  87
  88        if (!super_prefix_len)
  89                return path;
  90
  91        if (++idx >= ARRAY_SIZE(buf))
  92                idx = 0;
  93
  94        strbuf_setlen(&buf[idx], super_prefix_len);
  95        strbuf_addstr(&buf[idx], path);
  96
  97        return buf[idx].buf;
  98}
  99
 100void setup_unpack_trees_porcelain(struct unpack_trees_options *opts,
 101                                  const char *cmd)
 102{
 103        int i;
 104        const char **msgs = opts->msgs;
 105        const char *msg;
 106
 107        argv_array_init(&opts->msgs_to_free);
 108
 109        if (!strcmp(cmd, "checkout"))
 110                msg = advice_commit_before_merge
 111                      ? _("Your local changes to the following files would be overwritten by checkout:\n%%s"
 112                          "Please commit your changes or stash them before you switch branches.")
 113                      : _("Your local changes to the following files would be overwritten by checkout:\n%%s");
 114        else if (!strcmp(cmd, "merge"))
 115                msg = advice_commit_before_merge
 116                      ? _("Your local changes to the following files would be overwritten by merge:\n%%s"
 117                          "Please commit your changes or stash them before you merge.")
 118                      : _("Your local changes to the following files would be overwritten by merge:\n%%s");
 119        else
 120                msg = advice_commit_before_merge
 121                      ? _("Your local changes to the following files would be overwritten by %s:\n%%s"
 122                          "Please commit your changes or stash them before you %s.")
 123                      : _("Your local changes to the following files would be overwritten by %s:\n%%s");
 124        msgs[ERROR_WOULD_OVERWRITE] = msgs[ERROR_NOT_UPTODATE_FILE] =
 125                argv_array_pushf(&opts->msgs_to_free, msg, cmd, cmd);
 126
 127        msgs[ERROR_NOT_UPTODATE_DIR] =
 128                _("Updating the following directories would lose untracked files in them:\n%s");
 129
 130        if (!strcmp(cmd, "checkout"))
 131                msg = advice_commit_before_merge
 132                      ? _("The following untracked working tree files would be removed by checkout:\n%%s"
 133                          "Please move or remove them before you switch branches.")
 134                      : _("The following untracked working tree files would be removed by checkout:\n%%s");
 135        else if (!strcmp(cmd, "merge"))
 136                msg = advice_commit_before_merge
 137                      ? _("The following untracked working tree files would be removed by merge:\n%%s"
 138                          "Please move or remove them before you merge.")
 139                      : _("The following untracked working tree files would be removed by merge:\n%%s");
 140        else
 141                msg = advice_commit_before_merge
 142                      ? _("The following untracked working tree files would be removed by %s:\n%%s"
 143                          "Please move or remove them before you %s.")
 144                      : _("The following untracked working tree files would be removed by %s:\n%%s");
 145        msgs[ERROR_WOULD_LOSE_UNTRACKED_REMOVED] =
 146                argv_array_pushf(&opts->msgs_to_free, msg, cmd, cmd);
 147
 148        if (!strcmp(cmd, "checkout"))
 149                msg = advice_commit_before_merge
 150                      ? _("The following untracked working tree files would be overwritten by checkout:\n%%s"
 151                          "Please move or remove them before you switch branches.")
 152                      : _("The following untracked working tree files would be overwritten by checkout:\n%%s");
 153        else if (!strcmp(cmd, "merge"))
 154                msg = advice_commit_before_merge
 155                      ? _("The following untracked working tree files would be overwritten by merge:\n%%s"
 156                          "Please move or remove them before you merge.")
 157                      : _("The following untracked working tree files would be overwritten by merge:\n%%s");
 158        else
 159                msg = advice_commit_before_merge
 160                      ? _("The following untracked working tree files would be overwritten by %s:\n%%s"
 161                          "Please move or remove them before you %s.")
 162                      : _("The following untracked working tree files would be overwritten by %s:\n%%s");
 163        msgs[ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN] =
 164                argv_array_pushf(&opts->msgs_to_free, msg, cmd, cmd);
 165
 166        /*
 167         * Special case: ERROR_BIND_OVERLAP refers to a pair of paths, we
 168         * cannot easily display it as a list.
 169         */
 170        msgs[ERROR_BIND_OVERLAP] = _("Entry '%s' overlaps with '%s'.  Cannot bind.");
 171
 172        msgs[ERROR_SPARSE_NOT_UPTODATE_FILE] =
 173                _("Cannot update sparse checkout: the following entries are not up to date:\n%s");
 174        msgs[ERROR_WOULD_LOSE_ORPHANED_OVERWRITTEN] =
 175                _("The following working tree files would be overwritten by sparse checkout update:\n%s");
 176        msgs[ERROR_WOULD_LOSE_ORPHANED_REMOVED] =
 177                _("The following working tree files would be removed by sparse checkout update:\n%s");
 178        msgs[ERROR_WOULD_LOSE_SUBMODULE] =
 179                _("Cannot update submodule:\n%s");
 180
 181        opts->show_all_errors = 1;
 182        /* rejected paths may not have a static buffer */
 183        for (i = 0; i < ARRAY_SIZE(opts->unpack_rejects); i++)
 184                opts->unpack_rejects[i].strdup_strings = 1;
 185}
 186
 187void clear_unpack_trees_porcelain(struct unpack_trees_options *opts)
 188{
 189        argv_array_clear(&opts->msgs_to_free);
 190        memset(opts->msgs, 0, sizeof(opts->msgs));
 191}
 192
 193static int do_add_entry(struct unpack_trees_options *o, struct cache_entry *ce,
 194                         unsigned int set, unsigned int clear)
 195{
 196        clear |= CE_HASHED;
 197
 198        if (set & CE_REMOVE)
 199                set |= CE_WT_REMOVE;
 200
 201        ce->ce_flags = (ce->ce_flags & ~clear) | set;
 202        return add_index_entry(&o->result, ce,
 203                               ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
 204}
 205
 206static void add_entry(struct unpack_trees_options *o,
 207                      const struct cache_entry *ce,
 208                      unsigned int set, unsigned int clear)
 209{
 210        do_add_entry(o, dup_cache_entry(ce, &o->result), set, clear);
 211}
 212
 213/*
 214 * add error messages on path <path>
 215 * corresponding to the type <e> with the message <msg>
 216 * indicating if it should be display in porcelain or not
 217 */
 218static int add_rejected_path(struct unpack_trees_options *o,
 219                             enum unpack_trees_error_types e,
 220                             const char *path)
 221{
 222        if (o->quiet)
 223                return -1;
 224
 225        if (!o->show_all_errors)
 226                return error(ERRORMSG(o, e), super_prefixed(path));
 227
 228        /*
 229         * Otherwise, insert in a list for future display by
 230         * display_error_msgs()
 231         */
 232        string_list_append(&o->unpack_rejects[e], path);
 233        return -1;
 234}
 235
 236/*
 237 * display all the error messages stored in a nice way
 238 */
 239static void display_error_msgs(struct unpack_trees_options *o)
 240{
 241        int e, i;
 242        int something_displayed = 0;
 243        for (e = 0; e < NB_UNPACK_TREES_ERROR_TYPES; e++) {
 244                struct string_list *rejects = &o->unpack_rejects[e];
 245                if (rejects->nr > 0) {
 246                        struct strbuf path = STRBUF_INIT;
 247                        something_displayed = 1;
 248                        for (i = 0; i < rejects->nr; i++)
 249                                strbuf_addf(&path, "\t%s\n", rejects->items[i].string);
 250                        error(ERRORMSG(o, e), super_prefixed(path.buf));
 251                        strbuf_release(&path);
 252                }
 253                string_list_clear(rejects, 0);
 254        }
 255        if (something_displayed)
 256                fprintf(stderr, _("Aborting\n"));
 257}
 258
 259static int check_submodule_move_head(const struct cache_entry *ce,
 260                                     const char *old_id,
 261                                     const char *new_id,
 262                                     struct unpack_trees_options *o)
 263{
 264        unsigned flags = SUBMODULE_MOVE_HEAD_DRY_RUN;
 265        const struct submodule *sub = submodule_from_ce(ce);
 266
 267        if (!sub)
 268                return 0;
 269
 270        if (o->reset)
 271                flags |= SUBMODULE_MOVE_HEAD_FORCE;
 272
 273        if (submodule_move_head(ce->name, old_id, new_id, flags))
 274                return add_rejected_path(o, ERROR_WOULD_LOSE_SUBMODULE, ce->name);
 275        return 0;
 276}
 277
 278/*
 279 * Preform the loading of the repository's gitmodules file.  This function is
 280 * used by 'check_update()' to perform loading of the gitmodules file in two
 281 * differnt situations:
 282 * (1) before removing entries from the working tree if the gitmodules file has
 283 *     been marked for removal.  This situation is specified by 'state' == NULL.
 284 * (2) before checking out entries to the working tree if the gitmodules file
 285 *     has been marked for update.  This situation is specified by 'state' != NULL.
 286 */
 287static void load_gitmodules_file(struct index_state *index,
 288                                 struct checkout *state)
 289{
 290        int pos = index_name_pos(index, GITMODULES_FILE, strlen(GITMODULES_FILE));
 291
 292        if (pos >= 0) {
 293                struct cache_entry *ce = index->cache[pos];
 294                if (!state && ce->ce_flags & CE_WT_REMOVE) {
 295                        repo_read_gitmodules(the_repository);
 296                } else if (state && (ce->ce_flags & CE_UPDATE)) {
 297                        submodule_free(the_repository);
 298                        checkout_entry(ce, state, NULL, NULL);
 299                        repo_read_gitmodules(the_repository);
 300                }
 301        }
 302}
 303
 304static struct progress *get_progress(struct unpack_trees_options *o)
 305{
 306        unsigned cnt = 0, total = 0;
 307        struct index_state *index = &o->result;
 308
 309        if (!o->update || !o->verbose_update)
 310                return NULL;
 311
 312        for (; cnt < index->cache_nr; cnt++) {
 313                const struct cache_entry *ce = index->cache[cnt];
 314                if (ce->ce_flags & (CE_UPDATE | CE_WT_REMOVE))
 315                        total++;
 316        }
 317
 318        return start_delayed_progress(_("Updating files"), total);
 319}
 320
 321static void setup_collided_checkout_detection(struct checkout *state,
 322                                              struct index_state *index)
 323{
 324        int i;
 325
 326        state->clone = 1;
 327        for (i = 0; i < index->cache_nr; i++)
 328                index->cache[i]->ce_flags &= ~CE_MATCHED;
 329}
 330
 331static void report_collided_checkout(struct index_state *index)
 332{
 333        struct string_list list = STRING_LIST_INIT_NODUP;
 334        int i;
 335
 336        for (i = 0; i < index->cache_nr; i++) {
 337                struct cache_entry *ce = index->cache[i];
 338
 339                if (!(ce->ce_flags & CE_MATCHED))
 340                        continue;
 341
 342                string_list_append(&list, ce->name);
 343                ce->ce_flags &= ~CE_MATCHED;
 344        }
 345
 346        list.cmp = fspathcmp;
 347        string_list_sort(&list);
 348
 349        if (list.nr) {
 350                warning(_("the following paths have collided (e.g. case-sensitive paths\n"
 351                          "on a case-insensitive filesystem) and only one from the same\n"
 352                          "colliding group is in the working tree:\n"));
 353
 354                for (i = 0; i < list.nr; i++)
 355                        fprintf(stderr, "  '%s'\n", list.items[i].string);
 356        }
 357
 358        string_list_clear(&list, 0);
 359}
 360
 361static int check_updates(struct unpack_trees_options *o)
 362{
 363        unsigned cnt = 0;
 364        int errs = 0;
 365        struct progress *progress;
 366        struct index_state *index = &o->result;
 367        struct checkout state = CHECKOUT_INIT;
 368        int i;
 369
 370        trace_performance_enter();
 371        state.force = 1;
 372        state.quiet = 1;
 373        state.refresh_cache = 1;
 374        state.istate = index;
 375
 376        if (o->clone)
 377                setup_collided_checkout_detection(&state, index);
 378
 379        progress = get_progress(o);
 380
 381        if (o->update)
 382                git_attr_set_direction(GIT_ATTR_CHECKOUT);
 383
 384        if (should_update_submodules() && o->update && !o->dry_run)
 385                load_gitmodules_file(index, NULL);
 386
 387        for (i = 0; i < index->cache_nr; i++) {
 388                const struct cache_entry *ce = index->cache[i];
 389
 390                if (ce->ce_flags & CE_WT_REMOVE) {
 391                        display_progress(progress, ++cnt);
 392                        if (o->update && !o->dry_run)
 393                                unlink_entry(ce);
 394                }
 395        }
 396        remove_marked_cache_entries(index, 0);
 397        remove_scheduled_dirs();
 398
 399        if (should_update_submodules() && o->update && !o->dry_run)
 400                load_gitmodules_file(index, &state);
 401
 402        enable_delayed_checkout(&state);
 403        if (has_promisor_remote() && o->update && !o->dry_run) {
 404                /*
 405                 * Prefetch the objects that are to be checked out in the loop
 406                 * below.
 407                 */
 408                struct oid_array to_fetch = OID_ARRAY_INIT;
 409                for (i = 0; i < index->cache_nr; i++) {
 410                        struct cache_entry *ce = index->cache[i];
 411
 412                        if (!(ce->ce_flags & CE_UPDATE) ||
 413                            S_ISGITLINK(ce->ce_mode))
 414                                continue;
 415                        if (!oid_object_info_extended(the_repository, &ce->oid,
 416                                                      NULL,
 417                                                      OBJECT_INFO_FOR_PREFETCH))
 418                                continue;
 419                        oid_array_append(&to_fetch, &ce->oid);
 420                }
 421                if (to_fetch.nr)
 422                        promisor_remote_get_direct(the_repository,
 423                                                   to_fetch.oid, to_fetch.nr);
 424                oid_array_clear(&to_fetch);
 425        }
 426        for (i = 0; i < index->cache_nr; i++) {
 427                struct cache_entry *ce = index->cache[i];
 428
 429                if (ce->ce_flags & CE_UPDATE) {
 430                        if (ce->ce_flags & CE_WT_REMOVE)
 431                                BUG("both update and delete flags are set on %s",
 432                                    ce->name);
 433                        display_progress(progress, ++cnt);
 434                        ce->ce_flags &= ~CE_UPDATE;
 435                        if (o->update && !o->dry_run) {
 436                                errs |= checkout_entry(ce, &state, NULL, NULL);
 437                        }
 438                }
 439        }
 440        stop_progress(&progress);
 441        errs |= finish_delayed_checkout(&state, NULL);
 442        if (o->update)
 443                git_attr_set_direction(GIT_ATTR_CHECKIN);
 444
 445        if (o->clone)
 446                report_collided_checkout(index);
 447
 448        trace_performance_leave("check_updates");
 449        return errs != 0;
 450}
 451
 452static int verify_uptodate_sparse(const struct cache_entry *ce,
 453                                  struct unpack_trees_options *o);
 454static int verify_absent_sparse(const struct cache_entry *ce,
 455                                enum unpack_trees_error_types,
 456                                struct unpack_trees_options *o);
 457
 458static int apply_sparse_checkout(struct index_state *istate,
 459                                 struct cache_entry *ce,
 460                                 struct unpack_trees_options *o)
 461{
 462        int was_skip_worktree = ce_skip_worktree(ce);
 463
 464        if (ce->ce_flags & CE_NEW_SKIP_WORKTREE)
 465                ce->ce_flags |= CE_SKIP_WORKTREE;
 466        else
 467                ce->ce_flags &= ~CE_SKIP_WORKTREE;
 468        if (was_skip_worktree != ce_skip_worktree(ce)) {
 469                ce->ce_flags |= CE_UPDATE_IN_BASE;
 470                mark_fsmonitor_invalid(istate, ce);
 471                istate->cache_changed |= CE_ENTRY_CHANGED;
 472        }
 473
 474        /*
 475         * if (!was_skip_worktree && !ce_skip_worktree()) {
 476         *      This is perfectly normal. Move on;
 477         * }
 478         */
 479
 480        /*
 481         * Merge strategies may set CE_UPDATE|CE_REMOVE outside checkout
 482         * area as a result of ce_skip_worktree() shortcuts in
 483         * verify_absent() and verify_uptodate().
 484         * Make sure they don't modify worktree if they are already
 485         * outside checkout area
 486         */
 487        if (was_skip_worktree && ce_skip_worktree(ce)) {
 488                ce->ce_flags &= ~CE_UPDATE;
 489
 490                /*
 491                 * By default, when CE_REMOVE is on, CE_WT_REMOVE is also
 492                 * on to get that file removed from both index and worktree.
 493                 * If that file is already outside worktree area, don't
 494                 * bother remove it.
 495                 */
 496                if (ce->ce_flags & CE_REMOVE)
 497                        ce->ce_flags &= ~CE_WT_REMOVE;
 498        }
 499
 500        if (!was_skip_worktree && ce_skip_worktree(ce)) {
 501                /*
 502                 * If CE_UPDATE is set, verify_uptodate() must be called already
 503                 * also stat info may have lost after merged_entry() so calling
 504                 * verify_uptodate() again may fail
 505                 */
 506                if (!(ce->ce_flags & CE_UPDATE) && verify_uptodate_sparse(ce, o))
 507                        return -1;
 508                ce->ce_flags |= CE_WT_REMOVE;
 509                ce->ce_flags &= ~CE_UPDATE;
 510        }
 511        if (was_skip_worktree && !ce_skip_worktree(ce)) {
 512                if (verify_absent_sparse(ce, ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o))
 513                        return -1;
 514                ce->ce_flags |= CE_UPDATE;
 515        }
 516        return 0;
 517}
 518
 519static inline int call_unpack_fn(const struct cache_entry * const *src,
 520                                 struct unpack_trees_options *o)
 521{
 522        int ret = o->fn(src, o);
 523        if (ret > 0)
 524                ret = 0;
 525        return ret;
 526}
 527
 528static void mark_ce_used(struct cache_entry *ce, struct unpack_trees_options *o)
 529{
 530        ce->ce_flags |= CE_UNPACKED;
 531
 532        if (o->cache_bottom < o->src_index->cache_nr &&
 533            o->src_index->cache[o->cache_bottom] == ce) {
 534                int bottom = o->cache_bottom;
 535                while (bottom < o->src_index->cache_nr &&
 536                       o->src_index->cache[bottom]->ce_flags & CE_UNPACKED)
 537                        bottom++;
 538                o->cache_bottom = bottom;
 539        }
 540}
 541
 542static void mark_all_ce_unused(struct index_state *index)
 543{
 544        int i;
 545        for (i = 0; i < index->cache_nr; i++)
 546                index->cache[i]->ce_flags &= ~(CE_UNPACKED | CE_ADDED | CE_NEW_SKIP_WORKTREE);
 547}
 548
 549static int locate_in_src_index(const struct cache_entry *ce,
 550                               struct unpack_trees_options *o)
 551{
 552        struct index_state *index = o->src_index;
 553        int len = ce_namelen(ce);
 554        int pos = index_name_pos(index, ce->name, len);
 555        if (pos < 0)
 556                pos = -1 - pos;
 557        return pos;
 558}
 559
 560/*
 561 * We call unpack_index_entry() with an unmerged cache entry
 562 * only in diff-index, and it wants a single callback.  Skip
 563 * the other unmerged entry with the same name.
 564 */
 565static void mark_ce_used_same_name(struct cache_entry *ce,
 566                                   struct unpack_trees_options *o)
 567{
 568        struct index_state *index = o->src_index;
 569        int len = ce_namelen(ce);
 570        int pos;
 571
 572        for (pos = locate_in_src_index(ce, o); pos < index->cache_nr; pos++) {
 573                struct cache_entry *next = index->cache[pos];
 574                if (len != ce_namelen(next) ||
 575                    memcmp(ce->name, next->name, len))
 576                        break;
 577                mark_ce_used(next, o);
 578        }
 579}
 580
 581static struct cache_entry *next_cache_entry(struct unpack_trees_options *o)
 582{
 583        const struct index_state *index = o->src_index;
 584        int pos = o->cache_bottom;
 585
 586        while (pos < index->cache_nr) {
 587                struct cache_entry *ce = index->cache[pos];
 588                if (!(ce->ce_flags & CE_UNPACKED))
 589                        return ce;
 590                pos++;
 591        }
 592        return NULL;
 593}
 594
 595static void add_same_unmerged(const struct cache_entry *ce,
 596                              struct unpack_trees_options *o)
 597{
 598        struct index_state *index = o->src_index;
 599        int len = ce_namelen(ce);
 600        int pos = index_name_pos(index, ce->name, len);
 601
 602        if (0 <= pos)
 603                die("programming error in a caller of mark_ce_used_same_name");
 604        for (pos = -pos - 1; pos < index->cache_nr; pos++) {
 605                struct cache_entry *next = index->cache[pos];
 606                if (len != ce_namelen(next) ||
 607                    memcmp(ce->name, next->name, len))
 608                        break;
 609                add_entry(o, next, 0, 0);
 610                mark_ce_used(next, o);
 611        }
 612}
 613
 614static int unpack_index_entry(struct cache_entry *ce,
 615                              struct unpack_trees_options *o)
 616{
 617        const struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
 618        int ret;
 619
 620        src[0] = ce;
 621
 622        mark_ce_used(ce, o);
 623        if (ce_stage(ce)) {
 624                if (o->skip_unmerged) {
 625                        add_entry(o, ce, 0, 0);
 626                        return 0;
 627                }
 628        }
 629        ret = call_unpack_fn(src, o);
 630        if (ce_stage(ce))
 631                mark_ce_used_same_name(ce, o);
 632        return ret;
 633}
 634
 635static int find_cache_pos(struct traverse_info *, const char *p, size_t len);
 636
 637static void restore_cache_bottom(struct traverse_info *info, int bottom)
 638{
 639        struct unpack_trees_options *o = info->data;
 640
 641        if (o->diff_index_cached)
 642                return;
 643        o->cache_bottom = bottom;
 644}
 645
 646static int switch_cache_bottom(struct traverse_info *info)
 647{
 648        struct unpack_trees_options *o = info->data;
 649        int ret, pos;
 650
 651        if (o->diff_index_cached)
 652                return 0;
 653        ret = o->cache_bottom;
 654        pos = find_cache_pos(info->prev, info->name, info->namelen);
 655
 656        if (pos < -1)
 657                o->cache_bottom = -2 - pos;
 658        else if (pos < 0)
 659                o->cache_bottom = o->src_index->cache_nr;
 660        return ret;
 661}
 662
 663static inline int are_same_oid(struct name_entry *name_j, struct name_entry *name_k)
 664{
 665        return !is_null_oid(&name_j->oid) && !is_null_oid(&name_k->oid) && oideq(&name_j->oid, &name_k->oid);
 666}
 667
 668static int all_trees_same_as_cache_tree(int n, unsigned long dirmask,
 669                                        struct name_entry *names,
 670                                        struct traverse_info *info)
 671{
 672        struct unpack_trees_options *o = info->data;
 673        int i;
 674
 675        if (!o->merge || dirmask != ((1 << n) - 1))
 676                return 0;
 677
 678        for (i = 1; i < n; i++)
 679                if (!are_same_oid(names, names + i))
 680                        return 0;
 681
 682        return cache_tree_matches_traversal(o->src_index->cache_tree, names, info);
 683}
 684
 685static int index_pos_by_traverse_info(struct name_entry *names,
 686                                      struct traverse_info *info)
 687{
 688        struct unpack_trees_options *o = info->data;
 689        struct strbuf name = STRBUF_INIT;
 690        int pos;
 691
 692        strbuf_make_traverse_path(&name, info, names->path, names->pathlen);
 693        strbuf_addch(&name, '/');
 694        pos = index_name_pos(o->src_index, name.buf, name.len);
 695        if (pos >= 0)
 696                BUG("This is a directory and should not exist in index");
 697        pos = -pos - 1;
 698        if (!starts_with(o->src_index->cache[pos]->name, name.buf) ||
 699            (pos > 0 && starts_with(o->src_index->cache[pos-1]->name, name.buf)))
 700                BUG("pos must point at the first entry in this directory");
 701        strbuf_release(&name);
 702        return pos;
 703}
 704
 705/*
 706 * Fast path if we detect that all trees are the same as cache-tree at this
 707 * path. We'll walk these trees in an iterative loop using cache-tree/index
 708 * instead of ODB since we already know what these trees contain.
 709 */
 710static int traverse_by_cache_tree(int pos, int nr_entries, int nr_names,
 711                                  struct traverse_info *info)
 712{
 713        struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
 714        struct unpack_trees_options *o = info->data;
 715        struct cache_entry *tree_ce = NULL;
 716        int ce_len = 0;
 717        int i, d;
 718
 719        if (!o->merge)
 720                BUG("We need cache-tree to do this optimization");
 721
 722        /*
 723         * Do what unpack_callback() and unpack_nondirectories() normally
 724         * do. But we walk all paths in an iterative loop instead.
 725         *
 726         * D/F conflicts and higher stage entries are not a concern
 727         * because cache-tree would be invalidated and we would never
 728         * get here in the first place.
 729         */
 730        for (i = 0; i < nr_entries; i++) {
 731                int new_ce_len, len, rc;
 732
 733                src[0] = o->src_index->cache[pos + i];
 734
 735                len = ce_namelen(src[0]);
 736                new_ce_len = cache_entry_size(len);
 737
 738                if (new_ce_len > ce_len) {
 739                        new_ce_len <<= 1;
 740                        tree_ce = xrealloc(tree_ce, new_ce_len);
 741                        memset(tree_ce, 0, new_ce_len);
 742                        ce_len = new_ce_len;
 743
 744                        tree_ce->ce_flags = create_ce_flags(0);
 745
 746                        for (d = 1; d <= nr_names; d++)
 747                                src[d] = tree_ce;
 748                }
 749
 750                tree_ce->ce_mode = src[0]->ce_mode;
 751                tree_ce->ce_namelen = len;
 752                oidcpy(&tree_ce->oid, &src[0]->oid);
 753                memcpy(tree_ce->name, src[0]->name, len + 1);
 754
 755                rc = call_unpack_fn((const struct cache_entry * const *)src, o);
 756                if (rc < 0) {
 757                        free(tree_ce);
 758                        return rc;
 759                }
 760
 761                mark_ce_used(src[0], o);
 762        }
 763        free(tree_ce);
 764        if (o->debug_unpack)
 765                printf("Unpacked %d entries from %s to %s using cache-tree\n",
 766                       nr_entries,
 767                       o->src_index->cache[pos]->name,
 768                       o->src_index->cache[pos + nr_entries - 1]->name);
 769        return 0;
 770}
 771
 772static int traverse_trees_recursive(int n, unsigned long dirmask,
 773                                    unsigned long df_conflicts,
 774                                    struct name_entry *names,
 775                                    struct traverse_info *info)
 776{
 777        struct unpack_trees_options *o = info->data;
 778        int i, ret, bottom;
 779        int nr_buf = 0;
 780        struct tree_desc t[MAX_UNPACK_TREES];
 781        void *buf[MAX_UNPACK_TREES];
 782        struct traverse_info newinfo;
 783        struct name_entry *p;
 784        int nr_entries;
 785
 786        nr_entries = all_trees_same_as_cache_tree(n, dirmask, names, info);
 787        if (nr_entries > 0) {
 788                int pos = index_pos_by_traverse_info(names, info);
 789
 790                if (!o->merge || df_conflicts)
 791                        BUG("Wrong condition to get here buddy");
 792
 793                /*
 794                 * All entries up to 'pos' must have been processed
 795                 * (i.e. marked CE_UNPACKED) at this point. But to be safe,
 796                 * save and restore cache_bottom anyway to not miss
 797                 * unprocessed entries before 'pos'.
 798                 */
 799                bottom = o->cache_bottom;
 800                ret = traverse_by_cache_tree(pos, nr_entries, n, info);
 801                o->cache_bottom = bottom;
 802                return ret;
 803        }
 804
 805        p = names;
 806        while (!p->mode)
 807                p++;
 808
 809        newinfo = *info;
 810        newinfo.prev = info;
 811        newinfo.pathspec = info->pathspec;
 812        newinfo.name = p->path;
 813        newinfo.namelen = p->pathlen;
 814        newinfo.mode = p->mode;
 815        newinfo.pathlen = st_add3(newinfo.pathlen, tree_entry_len(p), 1);
 816        newinfo.df_conflicts |= df_conflicts;
 817
 818        /*
 819         * Fetch the tree from the ODB for each peer directory in the
 820         * n commits.
 821         *
 822         * For 2- and 3-way traversals, we try to avoid hitting the
 823         * ODB twice for the same OID.  This should yield a nice speed
 824         * up in checkouts and merges when the commits are similar.
 825         *
 826         * We don't bother doing the full O(n^2) search for larger n,
 827         * because wider traversals don't happen that often and we
 828         * avoid the search setup.
 829         *
 830         * When 2 peer OIDs are the same, we just copy the tree
 831         * descriptor data.  This implicitly borrows the buffer
 832         * data from the earlier cell.
 833         */
 834        for (i = 0; i < n; i++, dirmask >>= 1) {
 835                if (i > 0 && are_same_oid(&names[i], &names[i - 1]))
 836                        t[i] = t[i - 1];
 837                else if (i > 1 && are_same_oid(&names[i], &names[i - 2]))
 838                        t[i] = t[i - 2];
 839                else {
 840                        const struct object_id *oid = NULL;
 841                        if (dirmask & 1)
 842                                oid = &names[i].oid;
 843                        buf[nr_buf++] = fill_tree_descriptor(the_repository, t + i, oid);
 844                }
 845        }
 846
 847        bottom = switch_cache_bottom(&newinfo);
 848        ret = traverse_trees(o->src_index, n, t, &newinfo);
 849        restore_cache_bottom(&newinfo, bottom);
 850
 851        for (i = 0; i < nr_buf; i++)
 852                free(buf[i]);
 853
 854        return ret;
 855}
 856
 857/*
 858 * Compare the traverse-path to the cache entry without actually
 859 * having to generate the textual representation of the traverse
 860 * path.
 861 *
 862 * NOTE! This *only* compares up to the size of the traverse path
 863 * itself - the caller needs to do the final check for the cache
 864 * entry having more data at the end!
 865 */
 866static int do_compare_entry_piecewise(const struct cache_entry *ce,
 867                                      const struct traverse_info *info,
 868                                      const char *name, size_t namelen,
 869                                      unsigned mode)
 870{
 871        int pathlen, ce_len;
 872        const char *ce_name;
 873
 874        if (info->prev) {
 875                int cmp = do_compare_entry_piecewise(ce, info->prev,
 876                                                     info->name, info->namelen,
 877                                                     info->mode);
 878                if (cmp)
 879                        return cmp;
 880        }
 881        pathlen = info->pathlen;
 882        ce_len = ce_namelen(ce);
 883
 884        /* If ce_len < pathlen then we must have previously hit "name == directory" entry */
 885        if (ce_len < pathlen)
 886                return -1;
 887
 888        ce_len -= pathlen;
 889        ce_name = ce->name + pathlen;
 890
 891        return df_name_compare(ce_name, ce_len, S_IFREG, name, namelen, mode);
 892}
 893
 894static int do_compare_entry(const struct cache_entry *ce,
 895                            const struct traverse_info *info,
 896                            const char *name, size_t namelen,
 897                            unsigned mode)
 898{
 899        int pathlen, ce_len;
 900        const char *ce_name;
 901        int cmp;
 902
 903        /*
 904         * If we have not precomputed the traverse path, it is quicker
 905         * to avoid doing so.  But if we have precomputed it,
 906         * it is quicker to use the precomputed version.
 907         */
 908        if (!info->traverse_path)
 909                return do_compare_entry_piecewise(ce, info, name, namelen, mode);
 910
 911        cmp = strncmp(ce->name, info->traverse_path, info->pathlen);
 912        if (cmp)
 913                return cmp;
 914
 915        pathlen = info->pathlen;
 916        ce_len = ce_namelen(ce);
 917
 918        if (ce_len < pathlen)
 919                return -1;
 920
 921        ce_len -= pathlen;
 922        ce_name = ce->name + pathlen;
 923
 924        return df_name_compare(ce_name, ce_len, S_IFREG, name, namelen, mode);
 925}
 926
 927static int compare_entry(const struct cache_entry *ce, const struct traverse_info *info, const struct name_entry *n)
 928{
 929        int cmp = do_compare_entry(ce, info, n->path, n->pathlen, n->mode);
 930        if (cmp)
 931                return cmp;
 932
 933        /*
 934         * Even if the beginning compared identically, the ce should
 935         * compare as bigger than a directory leading up to it!
 936         */
 937        return ce_namelen(ce) > traverse_path_len(info, tree_entry_len(n));
 938}
 939
 940static int ce_in_traverse_path(const struct cache_entry *ce,
 941                               const struct traverse_info *info)
 942{
 943        if (!info->prev)
 944                return 1;
 945        if (do_compare_entry(ce, info->prev,
 946                             info->name, info->namelen, info->mode))
 947                return 0;
 948        /*
 949         * If ce (blob) is the same name as the path (which is a tree
 950         * we will be descending into), it won't be inside it.
 951         */
 952        return (info->pathlen < ce_namelen(ce));
 953}
 954
 955static struct cache_entry *create_ce_entry(const struct traverse_info *info,
 956        const struct name_entry *n,
 957        int stage,
 958        struct index_state *istate,
 959        int is_transient)
 960{
 961        size_t len = traverse_path_len(info, tree_entry_len(n));
 962        struct cache_entry *ce =
 963                is_transient ?
 964                make_empty_transient_cache_entry(len) :
 965                make_empty_cache_entry(istate, len);
 966
 967        ce->ce_mode = create_ce_mode(n->mode);
 968        ce->ce_flags = create_ce_flags(stage);
 969        ce->ce_namelen = len;
 970        oidcpy(&ce->oid, &n->oid);
 971        /* len+1 because the cache_entry allocates space for NUL */
 972        make_traverse_path(ce->name, len + 1, info, n->path, n->pathlen);
 973
 974        return ce;
 975}
 976
 977/*
 978 * Note that traverse_by_cache_tree() duplicates some logic in this function
 979 * without actually calling it. If you change the logic here you may need to
 980 * check and change there as well.
 981 */
 982static int unpack_nondirectories(int n, unsigned long mask,
 983                                 unsigned long dirmask,
 984                                 struct cache_entry **src,
 985                                 const struct name_entry *names,
 986                                 const struct traverse_info *info)
 987{
 988        int i;
 989        struct unpack_trees_options *o = info->data;
 990        unsigned long conflicts = info->df_conflicts | dirmask;
 991
 992        /* Do we have *only* directories? Nothing to do */
 993        if (mask == dirmask && !src[0])
 994                return 0;
 995
 996        /*
 997         * Ok, we've filled in up to any potential index entry in src[0],
 998         * now do the rest.
 999         */
1000        for (i = 0; i < n; i++) {
1001                int stage;
1002                unsigned int bit = 1ul << i;
1003                if (conflicts & bit) {
1004                        src[i + o->merge] = o->df_conflict_entry;
1005                        continue;
1006                }
1007                if (!(mask & bit))
1008                        continue;
1009                if (!o->merge)
1010                        stage = 0;
1011                else if (i + 1 < o->head_idx)
1012                        stage = 1;
1013                else if (i + 1 > o->head_idx)
1014                        stage = 3;
1015                else
1016                        stage = 2;
1017
1018                /*
1019                 * If the merge bit is set, then the cache entries are
1020                 * discarded in the following block.  In this case,
1021                 * construct "transient" cache_entries, as they are
1022                 * not stored in the index.  otherwise construct the
1023                 * cache entry from the index aware logic.
1024                 */
1025                src[i + o->merge] = create_ce_entry(info, names + i, stage, &o->result, o->merge);
1026        }
1027
1028        if (o->merge) {
1029                int rc = call_unpack_fn((const struct cache_entry * const *)src,
1030                                        o);
1031                for (i = 0; i < n; i++) {
1032                        struct cache_entry *ce = src[i + o->merge];
1033                        if (ce != o->df_conflict_entry)
1034                                discard_cache_entry(ce);
1035                }
1036                return rc;
1037        }
1038
1039        for (i = 0; i < n; i++)
1040                if (src[i] && src[i] != o->df_conflict_entry)
1041                        if (do_add_entry(o, src[i], 0, 0))
1042                                return -1;
1043
1044        return 0;
1045}
1046
1047static int unpack_failed(struct unpack_trees_options *o, const char *message)
1048{
1049        discard_index(&o->result);
1050        if (!o->quiet && !o->exiting_early) {
1051                if (message)
1052                        return error("%s", message);
1053                return -1;
1054        }
1055        return -1;
1056}
1057
1058/*
1059 * The tree traversal is looking at name p.  If we have a matching entry,
1060 * return it.  If name p is a directory in the index, do not return
1061 * anything, as we will want to match it when the traversal descends into
1062 * the directory.
1063 */
1064static int find_cache_pos(struct traverse_info *info,
1065                          const char *p, size_t p_len)
1066{
1067        int pos;
1068        struct unpack_trees_options *o = info->data;
1069        struct index_state *index = o->src_index;
1070        int pfxlen = info->pathlen;
1071
1072        for (pos = o->cache_bottom; pos < index->cache_nr; pos++) {
1073                const struct cache_entry *ce = index->cache[pos];
1074                const char *ce_name, *ce_slash;
1075                int cmp, ce_len;
1076
1077                if (ce->ce_flags & CE_UNPACKED) {
1078                        /*
1079                         * cache_bottom entry is already unpacked, so
1080                         * we can never match it; don't check it
1081                         * again.
1082                         */
1083                        if (pos == o->cache_bottom)
1084                                ++o->cache_bottom;
1085                        continue;
1086                }
1087                if (!ce_in_traverse_path(ce, info)) {
1088                        /*
1089                         * Check if we can skip future cache checks
1090                         * (because we're already past all possible
1091                         * entries in the traverse path).
1092                         */
1093                        if (info->traverse_path) {
1094                                if (strncmp(ce->name, info->traverse_path,
1095                                            info->pathlen) > 0)
1096                                        break;
1097                        }
1098                        continue;
1099                }
1100                ce_name = ce->name + pfxlen;
1101                ce_slash = strchr(ce_name, '/');
1102                if (ce_slash)
1103                        ce_len = ce_slash - ce_name;
1104                else
1105                        ce_len = ce_namelen(ce) - pfxlen;
1106                cmp = name_compare(p, p_len, ce_name, ce_len);
1107                /*
1108                 * Exact match; if we have a directory we need to
1109                 * delay returning it.
1110                 */
1111                if (!cmp)
1112                        return ce_slash ? -2 - pos : pos;
1113                if (0 < cmp)
1114                        continue; /* keep looking */
1115                /*
1116                 * ce_name sorts after p->path; could it be that we
1117                 * have files under p->path directory in the index?
1118                 * E.g.  ce_name == "t-i", and p->path == "t"; we may
1119                 * have "t/a" in the index.
1120                 */
1121                if (p_len < ce_len && !memcmp(ce_name, p, p_len) &&
1122                    ce_name[p_len] < '/')
1123                        continue; /* keep looking */
1124                break;
1125        }
1126        return -1;
1127}
1128
1129static struct cache_entry *find_cache_entry(struct traverse_info *info,
1130                                            const struct name_entry *p)
1131{
1132        int pos = find_cache_pos(info, p->path, p->pathlen);
1133        struct unpack_trees_options *o = info->data;
1134
1135        if (0 <= pos)
1136                return o->src_index->cache[pos];
1137        else
1138                return NULL;
1139}
1140
1141static void debug_path(struct traverse_info *info)
1142{
1143        if (info->prev) {
1144                debug_path(info->prev);
1145                if (*info->prev->name)
1146                        putchar('/');
1147        }
1148        printf("%s", info->name);
1149}
1150
1151static void debug_name_entry(int i, struct name_entry *n)
1152{
1153        printf("ent#%d %06o %s\n", i,
1154               n->path ? n->mode : 0,
1155               n->path ? n->path : "(missing)");
1156}
1157
1158static void debug_unpack_callback(int n,
1159                                  unsigned long mask,
1160                                  unsigned long dirmask,
1161                                  struct name_entry *names,
1162                                  struct traverse_info *info)
1163{
1164        int i;
1165        printf("* unpack mask %lu, dirmask %lu, cnt %d ",
1166               mask, dirmask, n);
1167        debug_path(info);
1168        putchar('\n');
1169        for (i = 0; i < n; i++)
1170                debug_name_entry(i, names + i);
1171}
1172
1173/*
1174 * Note that traverse_by_cache_tree() duplicates some logic in this function
1175 * without actually calling it. If you change the logic here you may need to
1176 * check and change there as well.
1177 */
1178static int unpack_callback(int n, unsigned long mask, unsigned long dirmask, struct name_entry *names, struct traverse_info *info)
1179{
1180        struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
1181        struct unpack_trees_options *o = info->data;
1182        const struct name_entry *p = names;
1183
1184        /* Find first entry with a real name (we could use "mask" too) */
1185        while (!p->mode)
1186                p++;
1187
1188        if (o->debug_unpack)
1189                debug_unpack_callback(n, mask, dirmask, names, info);
1190
1191        /* Are we supposed to look at the index too? */
1192        if (o->merge) {
1193                while (1) {
1194                        int cmp;
1195                        struct cache_entry *ce;
1196
1197                        if (o->diff_index_cached)
1198                                ce = next_cache_entry(o);
1199                        else
1200                                ce = find_cache_entry(info, p);
1201
1202                        if (!ce)
1203                                break;
1204                        cmp = compare_entry(ce, info, p);
1205                        if (cmp < 0) {
1206                                if (unpack_index_entry(ce, o) < 0)
1207                                        return unpack_failed(o, NULL);
1208                                continue;
1209                        }
1210                        if (!cmp) {
1211                                if (ce_stage(ce)) {
1212                                        /*
1213                                         * If we skip unmerged index
1214                                         * entries, we'll skip this
1215                                         * entry *and* the tree
1216                                         * entries associated with it!
1217                                         */
1218                                        if (o->skip_unmerged) {
1219                                                add_same_unmerged(ce, o);
1220                                                return mask;
1221                                        }
1222                                }
1223                                src[0] = ce;
1224                        }
1225                        break;
1226                }
1227        }
1228
1229        if (unpack_nondirectories(n, mask, dirmask, src, names, info) < 0)
1230                return -1;
1231
1232        if (o->merge && src[0]) {
1233                if (ce_stage(src[0]))
1234                        mark_ce_used_same_name(src[0], o);
1235                else
1236                        mark_ce_used(src[0], o);
1237        }
1238
1239        /* Now handle any directories.. */
1240        if (dirmask) {
1241                /* special case: "diff-index --cached" looking at a tree */
1242                if (o->diff_index_cached &&
1243                    n == 1 && dirmask == 1 && S_ISDIR(names->mode)) {
1244                        int matches;
1245                        matches = cache_tree_matches_traversal(o->src_index->cache_tree,
1246                                                               names, info);
1247                        /*
1248                         * Everything under the name matches; skip the
1249                         * entire hierarchy.  diff_index_cached codepath
1250                         * special cases D/F conflicts in such a way that
1251                         * it does not do any look-ahead, so this is safe.
1252                         */
1253                        if (matches) {
1254                                o->cache_bottom += matches;
1255                                return mask;
1256                        }
1257                }
1258
1259                if (traverse_trees_recursive(n, dirmask, mask & ~dirmask,
1260                                             names, info) < 0)
1261                        return -1;
1262                return mask;
1263        }
1264
1265        return mask;
1266}
1267
1268static int clear_ce_flags_1(struct index_state *istate,
1269                            struct cache_entry **cache, int nr,
1270                            struct strbuf *prefix,
1271                            int select_mask, int clear_mask,
1272                            struct pattern_list *pl,
1273                            enum pattern_match_result default_match);
1274
1275/* Whole directory matching */
1276static int clear_ce_flags_dir(struct index_state *istate,
1277                              struct cache_entry **cache, int nr,
1278                              struct strbuf *prefix,
1279                              char *basename,
1280                              int select_mask, int clear_mask,
1281                              struct pattern_list *pl,
1282                              enum pattern_match_result default_match)
1283{
1284        struct cache_entry **cache_end;
1285        int dtype = DT_DIR;
1286        int rc;
1287        enum pattern_match_result ret;
1288        ret = path_matches_pattern_list(prefix->buf, prefix->len,
1289                                        basename, &dtype, pl, istate);
1290
1291        strbuf_addch(prefix, '/');
1292
1293        /* If undecided, use matching result of parent dir in defval */
1294        if (ret == UNDECIDED)
1295                ret = default_match;
1296
1297        for (cache_end = cache; cache_end != cache + nr; cache_end++) {
1298                struct cache_entry *ce = *cache_end;
1299                if (strncmp(ce->name, prefix->buf, prefix->len))
1300                        break;
1301        }
1302
1303        /*
1304         * TODO: check pl, if there are no patterns that may conflict
1305         * with ret (iow, we know in advance the incl/excl
1306         * decision for the entire directory), clear flag here without
1307         * calling clear_ce_flags_1(). That function will call
1308         * the expensive path_matches_pattern_list() on every entry.
1309         */
1310        rc = clear_ce_flags_1(istate, cache, cache_end - cache,
1311                              prefix,
1312                              select_mask, clear_mask,
1313                              pl, ret);
1314        strbuf_setlen(prefix, prefix->len - 1);
1315        return rc;
1316}
1317
1318/*
1319 * Traverse the index, find every entry that matches according to
1320 * o->pl. Do "ce_flags &= ~clear_mask" on those entries. Return the
1321 * number of traversed entries.
1322 *
1323 * If select_mask is non-zero, only entries whose ce_flags has on of
1324 * those bits enabled are traversed.
1325 *
1326 * cache        : pointer to an index entry
1327 * prefix_len   : an offset to its path
1328 *
1329 * The current path ("prefix") including the trailing '/' is
1330 *   cache[0]->name[0..(prefix_len-1)]
1331 * Top level path has prefix_len zero.
1332 */
1333static int clear_ce_flags_1(struct index_state *istate,
1334                            struct cache_entry **cache, int nr,
1335                            struct strbuf *prefix,
1336                            int select_mask, int clear_mask,
1337                            struct pattern_list *pl,
1338                            enum pattern_match_result default_match)
1339{
1340        struct cache_entry **cache_end = cache + nr;
1341
1342        /*
1343         * Process all entries that have the given prefix and meet
1344         * select_mask condition
1345         */
1346        while(cache != cache_end) {
1347                struct cache_entry *ce = *cache;
1348                const char *name, *slash;
1349                int len, dtype;
1350                enum pattern_match_result ret;
1351
1352                if (select_mask && !(ce->ce_flags & select_mask)) {
1353                        cache++;
1354                        continue;
1355                }
1356
1357                if (prefix->len && strncmp(ce->name, prefix->buf, prefix->len))
1358                        break;
1359
1360                name = ce->name + prefix->len;
1361                slash = strchr(name, '/');
1362
1363                /* If it's a directory, try whole directory match first */
1364                if (slash) {
1365                        int processed;
1366
1367                        len = slash - name;
1368                        strbuf_add(prefix, name, len);
1369
1370                        processed = clear_ce_flags_dir(istate, cache, cache_end - cache,
1371                                                       prefix,
1372                                                       prefix->buf + prefix->len - len,
1373                                                       select_mask, clear_mask,
1374                                                       pl, default_match);
1375
1376                        /* clear_c_f_dir eats a whole dir already? */
1377                        if (processed) {
1378                                cache += processed;
1379                                strbuf_setlen(prefix, prefix->len - len);
1380                                continue;
1381                        }
1382
1383                        strbuf_addch(prefix, '/');
1384                        cache += clear_ce_flags_1(istate, cache, cache_end - cache,
1385                                                  prefix,
1386                                                  select_mask, clear_mask, pl,
1387                                                  default_match);
1388                        strbuf_setlen(prefix, prefix->len - len - 1);
1389                        continue;
1390                }
1391
1392                /* Non-directory */
1393                dtype = ce_to_dtype(ce);
1394                ret = path_matches_pattern_list(ce->name,
1395                                                ce_namelen(ce),
1396                                                name, &dtype, pl, istate);
1397                if (ret == UNDECIDED)
1398                        ret = default_match;
1399                if (ret == MATCHED)
1400                        ce->ce_flags &= ~clear_mask;
1401                cache++;
1402        }
1403        return nr - (cache_end - cache);
1404}
1405
1406static int clear_ce_flags(struct index_state *istate,
1407                          int select_mask, int clear_mask,
1408                          struct pattern_list *pl)
1409{
1410        static struct strbuf prefix = STRBUF_INIT;
1411
1412        strbuf_reset(&prefix);
1413
1414        return clear_ce_flags_1(istate,
1415                                istate->cache,
1416                                istate->cache_nr,
1417                                &prefix,
1418                                select_mask, clear_mask,
1419                                pl, 0);
1420}
1421
1422/*
1423 * Set/Clear CE_NEW_SKIP_WORKTREE according to $GIT_DIR/info/sparse-checkout
1424 */
1425static void mark_new_skip_worktree(struct pattern_list *pl,
1426                                   struct index_state *istate,
1427                                   int select_flag, int skip_wt_flag)
1428{
1429        int i;
1430
1431        /*
1432         * 1. Pretend the narrowest worktree: only unmerged entries
1433         * are checked out
1434         */
1435        for (i = 0; i < istate->cache_nr; i++) {
1436                struct cache_entry *ce = istate->cache[i];
1437
1438                if (select_flag && !(ce->ce_flags & select_flag))
1439                        continue;
1440
1441                if (!ce_stage(ce) && !(ce->ce_flags & CE_CONFLICTED))
1442                        ce->ce_flags |= skip_wt_flag;
1443                else
1444                        ce->ce_flags &= ~skip_wt_flag;
1445        }
1446
1447        /*
1448         * 2. Widen worktree according to sparse-checkout file.
1449         * Matched entries will have skip_wt_flag cleared (i.e. "in")
1450         */
1451        clear_ce_flags(istate, select_flag, skip_wt_flag, pl);
1452}
1453
1454static int verify_absent(const struct cache_entry *,
1455                         enum unpack_trees_error_types,
1456                         struct unpack_trees_options *);
1457/*
1458 * N-way merge "len" trees.  Returns 0 on success, -1 on failure to manipulate the
1459 * resulting index, -2 on failure to reflect the changes to the work tree.
1460 *
1461 * CE_ADDED, CE_UNPACKED and CE_NEW_SKIP_WORKTREE are used internally
1462 */
1463int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options *o)
1464{
1465        int i, ret;
1466        static struct cache_entry *dfc;
1467        struct pattern_list pl;
1468
1469        if (len > MAX_UNPACK_TREES)
1470                die("unpack_trees takes at most %d trees", MAX_UNPACK_TREES);
1471
1472        trace_performance_enter();
1473        memset(&pl, 0, sizeof(pl));
1474        if (!core_apply_sparse_checkout || !o->update)
1475                o->skip_sparse_checkout = 1;
1476        if (!o->skip_sparse_checkout) {
1477                char *sparse = git_pathdup("info/sparse-checkout");
1478                if (add_patterns_from_file_to_list(sparse, "", 0, &pl, NULL) < 0)
1479                        o->skip_sparse_checkout = 1;
1480                else
1481                        o->pl = &pl;
1482                free(sparse);
1483        }
1484
1485        memset(&o->result, 0, sizeof(o->result));
1486        o->result.initialized = 1;
1487        o->result.timestamp.sec = o->src_index->timestamp.sec;
1488        o->result.timestamp.nsec = o->src_index->timestamp.nsec;
1489        o->result.version = o->src_index->version;
1490        if (!o->src_index->split_index) {
1491                o->result.split_index = NULL;
1492        } else if (o->src_index == o->dst_index) {
1493                /*
1494                 * o->dst_index (and thus o->src_index) will be discarded
1495                 * and overwritten with o->result at the end of this function,
1496                 * so just use src_index's split_index to avoid having to
1497                 * create a new one.
1498                 */
1499                o->result.split_index = o->src_index->split_index;
1500                o->result.split_index->refcount++;
1501        } else {
1502                o->result.split_index = init_split_index(&o->result);
1503        }
1504        oidcpy(&o->result.oid, &o->src_index->oid);
1505        o->merge_size = len;
1506        mark_all_ce_unused(o->src_index);
1507
1508        /*
1509         * Sparse checkout loop #1: set NEW_SKIP_WORKTREE on existing entries
1510         */
1511        if (!o->skip_sparse_checkout)
1512                mark_new_skip_worktree(o->pl, o->src_index, 0, CE_NEW_SKIP_WORKTREE);
1513
1514        if (!dfc)
1515                dfc = xcalloc(1, cache_entry_size(0));
1516        o->df_conflict_entry = dfc;
1517
1518        if (len) {
1519                const char *prefix = o->prefix ? o->prefix : "";
1520                struct traverse_info info;
1521
1522                setup_traverse_info(&info, prefix);
1523                info.fn = unpack_callback;
1524                info.data = o;
1525                info.show_all_errors = o->show_all_errors;
1526                info.pathspec = o->pathspec;
1527
1528                if (o->prefix) {
1529                        /*
1530                         * Unpack existing index entries that sort before the
1531                         * prefix the tree is spliced into.  Note that o->merge
1532                         * is always true in this case.
1533                         */
1534                        while (1) {
1535                                struct cache_entry *ce = next_cache_entry(o);
1536                                if (!ce)
1537                                        break;
1538                                if (ce_in_traverse_path(ce, &info))
1539                                        break;
1540                                if (unpack_index_entry(ce, o) < 0)
1541                                        goto return_failed;
1542                        }
1543                }
1544
1545                trace_performance_enter();
1546                ret = traverse_trees(o->src_index, len, t, &info);
1547                trace_performance_leave("traverse_trees");
1548                if (ret < 0)
1549                        goto return_failed;
1550        }
1551
1552        /* Any left-over entries in the index? */
1553        if (o->merge) {
1554                while (1) {
1555                        struct cache_entry *ce = next_cache_entry(o);
1556                        if (!ce)
1557                                break;
1558                        if (unpack_index_entry(ce, o) < 0)
1559                                goto return_failed;
1560                }
1561        }
1562        mark_all_ce_unused(o->src_index);
1563
1564        if (o->trivial_merges_only && o->nontrivial_merge) {
1565                ret = unpack_failed(o, "Merge requires file-level merging");
1566                goto done;
1567        }
1568
1569        if (!o->skip_sparse_checkout) {
1570                int empty_worktree = 1;
1571
1572                /*
1573                 * Sparse checkout loop #2: set NEW_SKIP_WORKTREE on entries not in loop #1
1574                 * If the will have NEW_SKIP_WORKTREE, also set CE_SKIP_WORKTREE
1575                 * so apply_sparse_checkout() won't attempt to remove it from worktree
1576                 */
1577                mark_new_skip_worktree(o->pl, &o->result, CE_ADDED, CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE);
1578
1579                ret = 0;
1580                for (i = 0; i < o->result.cache_nr; i++) {
1581                        struct cache_entry *ce = o->result.cache[i];
1582
1583                        /*
1584                         * Entries marked with CE_ADDED in merged_entry() do not have
1585                         * verify_absent() check (the check is effectively disabled
1586                         * because CE_NEW_SKIP_WORKTREE is set unconditionally).
1587                         *
1588                         * Do the real check now because we have had
1589                         * correct CE_NEW_SKIP_WORKTREE
1590                         */
1591                        if (ce->ce_flags & CE_ADDED &&
1592                            verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o)) {
1593                                if (!o->show_all_errors)
1594                                        goto return_failed;
1595                                ret = -1;
1596                        }
1597
1598                        if (apply_sparse_checkout(&o->result, ce, o)) {
1599                                if (!o->show_all_errors)
1600                                        goto return_failed;
1601                                ret = -1;
1602                        }
1603                        if (!ce_skip_worktree(ce))
1604                                empty_worktree = 0;
1605
1606                }
1607                if (ret < 0)
1608                        goto return_failed;
1609                /*
1610                 * Sparse checkout is meant to narrow down checkout area
1611                 * but it does not make sense to narrow down to empty working
1612                 * tree. This is usually a mistake in sparse checkout rules.
1613                 * Do not allow users to do that.
1614                 */
1615                if (o->result.cache_nr && empty_worktree) {
1616                        ret = unpack_failed(o, "Sparse checkout leaves no entry on working directory");
1617                        goto done;
1618                }
1619        }
1620
1621        ret = check_updates(o) ? (-2) : 0;
1622        if (o->dst_index) {
1623                move_index_extensions(&o->result, o->src_index);
1624                if (!ret) {
1625                        if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
1626                                cache_tree_verify(the_repository, &o->result);
1627                        if (!o->result.cache_tree)
1628                                o->result.cache_tree = cache_tree();
1629                        if (!cache_tree_fully_valid(o->result.cache_tree))
1630                                cache_tree_update(&o->result,
1631                                                  WRITE_TREE_SILENT |
1632                                                  WRITE_TREE_REPAIR);
1633                }
1634
1635                o->result.updated_workdir = 1;
1636                discard_index(o->dst_index);
1637                *o->dst_index = o->result;
1638        } else {
1639                discard_index(&o->result);
1640        }
1641        o->src_index = NULL;
1642
1643done:
1644        trace_performance_leave("unpack_trees");
1645        clear_pattern_list(&pl);
1646        return ret;
1647
1648return_failed:
1649        if (o->show_all_errors)
1650                display_error_msgs(o);
1651        mark_all_ce_unused(o->src_index);
1652        ret = unpack_failed(o, NULL);
1653        if (o->exiting_early)
1654                ret = 0;
1655        goto done;
1656}
1657
1658/* Here come the merge functions */
1659
1660static int reject_merge(const struct cache_entry *ce,
1661                        struct unpack_trees_options *o)
1662{
1663        return add_rejected_path(o, ERROR_WOULD_OVERWRITE, ce->name);
1664}
1665
1666static int same(const struct cache_entry *a, const struct cache_entry *b)
1667{
1668        if (!!a != !!b)
1669                return 0;
1670        if (!a && !b)
1671                return 1;
1672        if ((a->ce_flags | b->ce_flags) & CE_CONFLICTED)
1673                return 0;
1674        return a->ce_mode == b->ce_mode &&
1675               oideq(&a->oid, &b->oid);
1676}
1677
1678
1679/*
1680 * When a CE gets turned into an unmerged entry, we
1681 * want it to be up-to-date
1682 */
1683static int verify_uptodate_1(const struct cache_entry *ce,
1684                             struct unpack_trees_options *o,
1685                             enum unpack_trees_error_types error_type)
1686{
1687        struct stat st;
1688
1689        if (o->index_only)
1690                return 0;
1691
1692        /*
1693         * CE_VALID and CE_SKIP_WORKTREE cheat, we better check again
1694         * if this entry is truly up-to-date because this file may be
1695         * overwritten.
1696         */
1697        if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
1698                ; /* keep checking */
1699        else if (o->reset || ce_uptodate(ce))
1700                return 0;
1701
1702        if (!lstat(ce->name, &st)) {
1703                int flags = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE;
1704                unsigned changed = ie_match_stat(o->src_index, ce, &st, flags);
1705
1706                if (submodule_from_ce(ce)) {
1707                        int r = check_submodule_move_head(ce,
1708                                "HEAD", oid_to_hex(&ce->oid), o);
1709                        if (r)
1710                                return add_rejected_path(o, error_type, ce->name);
1711                        return 0;
1712                }
1713
1714                if (!changed)
1715                        return 0;
1716                /*
1717                 * Historic default policy was to allow submodule to be out
1718                 * of sync wrt the superproject index. If the submodule was
1719                 * not considered interesting above, we don't care here.
1720                 */
1721                if (S_ISGITLINK(ce->ce_mode))
1722                        return 0;
1723
1724                errno = 0;
1725        }
1726        if (errno == ENOENT)
1727                return 0;
1728        return add_rejected_path(o, error_type, ce->name);
1729}
1730
1731int verify_uptodate(const struct cache_entry *ce,
1732                    struct unpack_trees_options *o)
1733{
1734        if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
1735                return 0;
1736        return verify_uptodate_1(ce, o, ERROR_NOT_UPTODATE_FILE);
1737}
1738
1739static int verify_uptodate_sparse(const struct cache_entry *ce,
1740                                  struct unpack_trees_options *o)
1741{
1742        return verify_uptodate_1(ce, o, ERROR_SPARSE_NOT_UPTODATE_FILE);
1743}
1744
1745/*
1746 * TODO: We should actually invalidate o->result, not src_index [1].
1747 * But since cache tree and untracked cache both are not copied to
1748 * o->result until unpacking is complete, we invalidate them on
1749 * src_index instead with the assumption that they will be copied to
1750 * dst_index at the end.
1751 *
1752 * [1] src_index->cache_tree is also used in unpack_callback() so if
1753 * we invalidate o->result, we need to update it to use
1754 * o->result.cache_tree as well.
1755 */
1756static void invalidate_ce_path(const struct cache_entry *ce,
1757                               struct unpack_trees_options *o)
1758{
1759        if (!ce)
1760                return;
1761        cache_tree_invalidate_path(o->src_index, ce->name);
1762        untracked_cache_invalidate_path(o->src_index, ce->name, 1);
1763}
1764
1765/*
1766 * Check that checking out ce->sha1 in subdir ce->name is not
1767 * going to overwrite any working files.
1768 *
1769 * Currently, git does not checkout subprojects during a superproject
1770 * checkout, so it is not going to overwrite anything.
1771 */
1772static int verify_clean_submodule(const char *old_sha1,
1773                                  const struct cache_entry *ce,
1774                                  struct unpack_trees_options *o)
1775{
1776        if (!submodule_from_ce(ce))
1777                return 0;
1778
1779        return check_submodule_move_head(ce, old_sha1,
1780                                         oid_to_hex(&ce->oid), o);
1781}
1782
1783static int verify_clean_subdirectory(const struct cache_entry *ce,
1784                                     struct unpack_trees_options *o)
1785{
1786        /*
1787         * we are about to extract "ce->name"; we would not want to lose
1788         * anything in the existing directory there.
1789         */
1790        int namelen;
1791        int i;
1792        struct dir_struct d;
1793        char *pathbuf;
1794        int cnt = 0;
1795
1796        if (S_ISGITLINK(ce->ce_mode)) {
1797                struct object_id oid;
1798                int sub_head = resolve_gitlink_ref(ce->name, "HEAD", &oid);
1799                /*
1800                 * If we are not going to update the submodule, then
1801                 * we don't care.
1802                 */
1803                if (!sub_head && oideq(&oid, &ce->oid))
1804                        return 0;
1805                return verify_clean_submodule(sub_head ? NULL : oid_to_hex(&oid),
1806                                              ce, o);
1807        }
1808
1809        /*
1810         * First let's make sure we do not have a local modification
1811         * in that directory.
1812         */
1813        namelen = ce_namelen(ce);
1814        for (i = locate_in_src_index(ce, o);
1815             i < o->src_index->cache_nr;
1816             i++) {
1817                struct cache_entry *ce2 = o->src_index->cache[i];
1818                int len = ce_namelen(ce2);
1819                if (len < namelen ||
1820                    strncmp(ce->name, ce2->name, namelen) ||
1821                    ce2->name[namelen] != '/')
1822                        break;
1823                /*
1824                 * ce2->name is an entry in the subdirectory to be
1825                 * removed.
1826                 */
1827                if (!ce_stage(ce2)) {
1828                        if (verify_uptodate(ce2, o))
1829                                return -1;
1830                        add_entry(o, ce2, CE_REMOVE, 0);
1831                        invalidate_ce_path(ce, o);
1832                        mark_ce_used(ce2, o);
1833                }
1834                cnt++;
1835        }
1836
1837        /*
1838         * Then we need to make sure that we do not lose a locally
1839         * present file that is not ignored.
1840         */
1841        pathbuf = xstrfmt("%.*s/", namelen, ce->name);
1842
1843        memset(&d, 0, sizeof(d));
1844        if (o->dir)
1845                d.exclude_per_dir = o->dir->exclude_per_dir;
1846        i = read_directory(&d, o->src_index, pathbuf, namelen+1, NULL);
1847        if (i)
1848                return add_rejected_path(o, ERROR_NOT_UPTODATE_DIR, ce->name);
1849        free(pathbuf);
1850        return cnt;
1851}
1852
1853/*
1854 * This gets called when there was no index entry for the tree entry 'dst',
1855 * but we found a file in the working tree that 'lstat()' said was fine,
1856 * and we're on a case-insensitive filesystem.
1857 *
1858 * See if we can find a case-insensitive match in the index that also
1859 * matches the stat information, and assume it's that other file!
1860 */
1861static int icase_exists(struct unpack_trees_options *o, const char *name, int len, struct stat *st)
1862{
1863        const struct cache_entry *src;
1864
1865        src = index_file_exists(o->src_index, name, len, 1);
1866        return src && !ie_match_stat(o->src_index, src, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
1867}
1868
1869static int check_ok_to_remove(const char *name, int len, int dtype,
1870                              const struct cache_entry *ce, struct stat *st,
1871                              enum unpack_trees_error_types error_type,
1872                              struct unpack_trees_options *o)
1873{
1874        const struct cache_entry *result;
1875
1876        /*
1877         * It may be that the 'lstat()' succeeded even though
1878         * target 'ce' was absent, because there is an old
1879         * entry that is different only in case..
1880         *
1881         * Ignore that lstat() if it matches.
1882         */
1883        if (ignore_case && icase_exists(o, name, len, st))
1884                return 0;
1885
1886        if (o->dir &&
1887            is_excluded(o->dir, o->src_index, name, &dtype))
1888                /*
1889                 * ce->name is explicitly excluded, so it is Ok to
1890                 * overwrite it.
1891                 */
1892                return 0;
1893        if (S_ISDIR(st->st_mode)) {
1894                /*
1895                 * We are checking out path "foo" and
1896                 * found "foo/." in the working tree.
1897                 * This is tricky -- if we have modified
1898                 * files that are in "foo/" we would lose
1899                 * them.
1900                 */
1901                if (verify_clean_subdirectory(ce, o) < 0)
1902                        return -1;
1903                return 0;
1904        }
1905
1906        /*
1907         * The previous round may already have decided to
1908         * delete this path, which is in a subdirectory that
1909         * is being replaced with a blob.
1910         */
1911        result = index_file_exists(&o->result, name, len, 0);
1912        if (result) {
1913                if (result->ce_flags & CE_REMOVE)
1914                        return 0;
1915        }
1916
1917        return add_rejected_path(o, error_type, name);
1918}
1919
1920/*
1921 * We do not want to remove or overwrite a working tree file that
1922 * is not tracked, unless it is ignored.
1923 */
1924static int verify_absent_1(const struct cache_entry *ce,
1925                           enum unpack_trees_error_types error_type,
1926                           struct unpack_trees_options *o)
1927{
1928        int len;
1929        struct stat st;
1930
1931        if (o->index_only || o->reset || !o->update)
1932                return 0;
1933
1934        len = check_leading_path(ce->name, ce_namelen(ce));
1935        if (!len)
1936                return 0;
1937        else if (len > 0) {
1938                char *path;
1939                int ret;
1940
1941                path = xmemdupz(ce->name, len);
1942                if (lstat(path, &st))
1943                        ret = error_errno("cannot stat '%s'", path);
1944                else {
1945                        if (submodule_from_ce(ce))
1946                                ret = check_submodule_move_head(ce,
1947                                                                oid_to_hex(&ce->oid),
1948                                                                NULL, o);
1949                        else
1950                                ret = check_ok_to_remove(path, len, DT_UNKNOWN, NULL,
1951                                                         &st, error_type, o);
1952                }
1953                free(path);
1954                return ret;
1955        } else if (lstat(ce->name, &st)) {
1956                if (errno != ENOENT)
1957                        return error_errno("cannot stat '%s'", ce->name);
1958                return 0;
1959        } else {
1960                if (submodule_from_ce(ce))
1961                        return check_submodule_move_head(ce, oid_to_hex(&ce->oid),
1962                                                         NULL, o);
1963
1964                return check_ok_to_remove(ce->name, ce_namelen(ce),
1965                                          ce_to_dtype(ce), ce, &st,
1966                                          error_type, o);
1967        }
1968}
1969
1970static int verify_absent(const struct cache_entry *ce,
1971                         enum unpack_trees_error_types error_type,
1972                         struct unpack_trees_options *o)
1973{
1974        if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
1975                return 0;
1976        return verify_absent_1(ce, error_type, o);
1977}
1978
1979static int verify_absent_sparse(const struct cache_entry *ce,
1980                                enum unpack_trees_error_types error_type,
1981                                struct unpack_trees_options *o)
1982{
1983        enum unpack_trees_error_types orphaned_error = error_type;
1984        if (orphaned_error == ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN)
1985                orphaned_error = ERROR_WOULD_LOSE_ORPHANED_OVERWRITTEN;
1986
1987        return verify_absent_1(ce, orphaned_error, o);
1988}
1989
1990static int merged_entry(const struct cache_entry *ce,
1991                        const struct cache_entry *old,
1992                        struct unpack_trees_options *o)
1993{
1994        int update = CE_UPDATE;
1995        struct cache_entry *merge = dup_cache_entry(ce, &o->result);
1996
1997        if (!old) {
1998                /*
1999                 * New index entries. In sparse checkout, the following
2000                 * verify_absent() will be delayed until after
2001                 * traverse_trees() finishes in unpack_trees(), then:
2002                 *
2003                 *  - CE_NEW_SKIP_WORKTREE will be computed correctly
2004                 *  - verify_absent() be called again, this time with
2005                 *    correct CE_NEW_SKIP_WORKTREE
2006                 *
2007                 * verify_absent() call here does nothing in sparse
2008                 * checkout (i.e. o->skip_sparse_checkout == 0)
2009                 */
2010                update |= CE_ADDED;
2011                merge->ce_flags |= CE_NEW_SKIP_WORKTREE;
2012
2013                if (verify_absent(merge,
2014                                  ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o)) {
2015                        discard_cache_entry(merge);
2016                        return -1;
2017                }
2018                invalidate_ce_path(merge, o);
2019
2020                if (submodule_from_ce(ce)) {
2021                        int ret = check_submodule_move_head(ce, NULL,
2022                                                            oid_to_hex(&ce->oid),
2023                                                            o);
2024                        if (ret)
2025                                return ret;
2026                }
2027
2028        } else if (!(old->ce_flags & CE_CONFLICTED)) {
2029                /*
2030                 * See if we can re-use the old CE directly?
2031                 * That way we get the uptodate stat info.
2032                 *
2033                 * This also removes the UPDATE flag on a match; otherwise
2034                 * we will end up overwriting local changes in the work tree.
2035                 */
2036                if (same(old, merge)) {
2037                        copy_cache_entry(merge, old);
2038                        update = 0;
2039                } else {
2040                        if (verify_uptodate(old, o)) {
2041                                discard_cache_entry(merge);
2042                                return -1;
2043                        }
2044                        /* Migrate old flags over */
2045                        update |= old->ce_flags & (CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE);
2046                        invalidate_ce_path(old, o);
2047                }
2048
2049                if (submodule_from_ce(ce)) {
2050                        int ret = check_submodule_move_head(ce, oid_to_hex(&old->oid),
2051                                                            oid_to_hex(&ce->oid),
2052                                                            o);
2053                        if (ret)
2054                                return ret;
2055                }
2056        } else {
2057                /*
2058                 * Previously unmerged entry left as an existence
2059                 * marker by read_index_unmerged();
2060                 */
2061                invalidate_ce_path(old, o);
2062        }
2063
2064        do_add_entry(o, merge, update, CE_STAGEMASK);
2065        return 1;
2066}
2067
2068static int deleted_entry(const struct cache_entry *ce,
2069                         const struct cache_entry *old,
2070                         struct unpack_trees_options *o)
2071{
2072        /* Did it exist in the index? */
2073        if (!old) {
2074                if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2075                        return -1;
2076                return 0;
2077        }
2078        if (!(old->ce_flags & CE_CONFLICTED) && verify_uptodate(old, o))
2079                return -1;
2080        add_entry(o, ce, CE_REMOVE, 0);
2081        invalidate_ce_path(ce, o);
2082        return 1;
2083}
2084
2085static int keep_entry(const struct cache_entry *ce,
2086                      struct unpack_trees_options *o)
2087{
2088        add_entry(o, ce, 0, 0);
2089        if (ce_stage(ce))
2090                invalidate_ce_path(ce, o);
2091        return 1;
2092}
2093
2094#if DBRT_DEBUG
2095static void show_stage_entry(FILE *o,
2096                             const char *label, const struct cache_entry *ce)
2097{
2098        if (!ce)
2099                fprintf(o, "%s (missing)\n", label);
2100        else
2101                fprintf(o, "%s%06o %s %d\t%s\n",
2102                        label,
2103                        ce->ce_mode,
2104                        oid_to_hex(&ce->oid),
2105                        ce_stage(ce),
2106                        ce->name);
2107}
2108#endif
2109
2110int threeway_merge(const struct cache_entry * const *stages,
2111                   struct unpack_trees_options *o)
2112{
2113        const struct cache_entry *index;
2114        const struct cache_entry *head;
2115        const struct cache_entry *remote = stages[o->head_idx + 1];
2116        int count;
2117        int head_match = 0;
2118        int remote_match = 0;
2119
2120        int df_conflict_head = 0;
2121        int df_conflict_remote = 0;
2122
2123        int any_anc_missing = 0;
2124        int no_anc_exists = 1;
2125        int i;
2126
2127        for (i = 1; i < o->head_idx; i++) {
2128                if (!stages[i] || stages[i] == o->df_conflict_entry)
2129                        any_anc_missing = 1;
2130                else
2131                        no_anc_exists = 0;
2132        }
2133
2134        index = stages[0];
2135        head = stages[o->head_idx];
2136
2137        if (head == o->df_conflict_entry) {
2138                df_conflict_head = 1;
2139                head = NULL;
2140        }
2141
2142        if (remote == o->df_conflict_entry) {
2143                df_conflict_remote = 1;
2144                remote = NULL;
2145        }
2146
2147        /*
2148         * First, if there's a #16 situation, note that to prevent #13
2149         * and #14.
2150         */
2151        if (!same(remote, head)) {
2152                for (i = 1; i < o->head_idx; i++) {
2153                        if (same(stages[i], head)) {
2154                                head_match = i;
2155                        }
2156                        if (same(stages[i], remote)) {
2157                                remote_match = i;
2158                        }
2159                }
2160        }
2161
2162        /*
2163         * We start with cases where the index is allowed to match
2164         * something other than the head: #14(ALT) and #2ALT, where it
2165         * is permitted to match the result instead.
2166         */
2167        /* #14, #14ALT, #2ALT */
2168        if (remote && !df_conflict_head && head_match && !remote_match) {
2169                if (index && !same(index, remote) && !same(index, head))
2170                        return reject_merge(index, o);
2171                return merged_entry(remote, index, o);
2172        }
2173        /*
2174         * If we have an entry in the index cache, then we want to
2175         * make sure that it matches head.
2176         */
2177        if (index && !same(index, head))
2178                return reject_merge(index, o);
2179
2180        if (head) {
2181                /* #5ALT, #15 */
2182                if (same(head, remote))
2183                        return merged_entry(head, index, o);
2184                /* #13, #3ALT */
2185                if (!df_conflict_remote && remote_match && !head_match)
2186                        return merged_entry(head, index, o);
2187        }
2188
2189        /* #1 */
2190        if (!head && !remote && any_anc_missing)
2191                return 0;
2192
2193        /*
2194         * Under the "aggressive" rule, we resolve mostly trivial
2195         * cases that we historically had git-merge-one-file resolve.
2196         */
2197        if (o->aggressive) {
2198                int head_deleted = !head;
2199                int remote_deleted = !remote;
2200                const struct cache_entry *ce = NULL;
2201
2202                if (index)
2203                        ce = index;
2204                else if (head)
2205                        ce = head;
2206                else if (remote)
2207                        ce = remote;
2208                else {
2209                        for (i = 1; i < o->head_idx; i++) {
2210                                if (stages[i] && stages[i] != o->df_conflict_entry) {
2211                                        ce = stages[i];
2212                                        break;
2213                                }
2214                        }
2215                }
2216
2217                /*
2218                 * Deleted in both.
2219                 * Deleted in one and unchanged in the other.
2220                 */
2221                if ((head_deleted && remote_deleted) ||
2222                    (head_deleted && remote && remote_match) ||
2223                    (remote_deleted && head && head_match)) {
2224                        if (index)
2225                                return deleted_entry(index, index, o);
2226                        if (ce && !head_deleted) {
2227                                if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2228                                        return -1;
2229                        }
2230                        return 0;
2231                }
2232                /*
2233                 * Added in both, identically.
2234                 */
2235                if (no_anc_exists && head && remote && same(head, remote))
2236                        return merged_entry(head, index, o);
2237
2238        }
2239
2240        /* Below are "no merge" cases, which require that the index be
2241         * up-to-date to avoid the files getting overwritten with
2242         * conflict resolution files.
2243         */
2244        if (index) {
2245                if (verify_uptodate(index, o))
2246                        return -1;
2247        }
2248
2249        o->nontrivial_merge = 1;
2250
2251        /* #2, #3, #4, #6, #7, #9, #10, #11. */
2252        count = 0;
2253        if (!head_match || !remote_match) {
2254                for (i = 1; i < o->head_idx; i++) {
2255                        if (stages[i] && stages[i] != o->df_conflict_entry) {
2256                                keep_entry(stages[i], o);
2257                                count++;
2258                                break;
2259                        }
2260                }
2261        }
2262#if DBRT_DEBUG
2263        else {
2264                fprintf(stderr, "read-tree: warning #16 detected\n");
2265                show_stage_entry(stderr, "head   ", stages[head_match]);
2266                show_stage_entry(stderr, "remote ", stages[remote_match]);
2267        }
2268#endif
2269        if (head) { count += keep_entry(head, o); }
2270        if (remote) { count += keep_entry(remote, o); }
2271        return count;
2272}
2273
2274/*
2275 * Two-way merge.
2276 *
2277 * The rule is to "carry forward" what is in the index without losing
2278 * information across a "fast-forward", favoring a successful merge
2279 * over a merge failure when it makes sense.  For details of the
2280 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
2281 *
2282 */
2283int twoway_merge(const struct cache_entry * const *src,
2284                 struct unpack_trees_options *o)
2285{
2286        const struct cache_entry *current = src[0];
2287        const struct cache_entry *oldtree = src[1];
2288        const struct cache_entry *newtree = src[2];
2289
2290        if (o->merge_size != 2)
2291                return error("Cannot do a twoway merge of %d trees",
2292                             o->merge_size);
2293
2294        if (oldtree == o->df_conflict_entry)
2295                oldtree = NULL;
2296        if (newtree == o->df_conflict_entry)
2297                newtree = NULL;
2298
2299        if (current) {
2300                if (current->ce_flags & CE_CONFLICTED) {
2301                        if (same(oldtree, newtree) || o->reset) {
2302                                if (!newtree)
2303                                        return deleted_entry(current, current, o);
2304                                else
2305                                        return merged_entry(newtree, current, o);
2306                        }
2307                        return reject_merge(current, o);
2308                } else if ((!oldtree && !newtree) || /* 4 and 5 */
2309                         (!oldtree && newtree &&
2310                          same(current, newtree)) || /* 6 and 7 */
2311                         (oldtree && newtree &&
2312                          same(oldtree, newtree)) || /* 14 and 15 */
2313                         (oldtree && newtree &&
2314                          !same(oldtree, newtree) && /* 18 and 19 */
2315                          same(current, newtree))) {
2316                        return keep_entry(current, o);
2317                } else if (oldtree && !newtree && same(current, oldtree)) {
2318                        /* 10 or 11 */
2319                        return deleted_entry(oldtree, current, o);
2320                } else if (oldtree && newtree &&
2321                         same(current, oldtree) && !same(current, newtree)) {
2322                        /* 20 or 21 */
2323                        return merged_entry(newtree, current, o);
2324                } else
2325                        return reject_merge(current, o);
2326        }
2327        else if (newtree) {
2328                if (oldtree && !o->initial_checkout) {
2329                        /*
2330                         * deletion of the path was staged;
2331                         */
2332                        if (same(oldtree, newtree))
2333                                return 1;
2334                        return reject_merge(oldtree, o);
2335                }
2336                return merged_entry(newtree, current, o);
2337        }
2338        return deleted_entry(oldtree, current, o);
2339}
2340
2341/*
2342 * Bind merge.
2343 *
2344 * Keep the index entries at stage0, collapse stage1 but make sure
2345 * stage0 does not have anything there.
2346 */
2347int bind_merge(const struct cache_entry * const *src,
2348               struct unpack_trees_options *o)
2349{
2350        const struct cache_entry *old = src[0];
2351        const struct cache_entry *a = src[1];
2352
2353        if (o->merge_size != 1)
2354                return error("Cannot do a bind merge of %d trees",
2355                             o->merge_size);
2356        if (a && old)
2357                return o->quiet ? -1 :
2358                        error(ERRORMSG(o, ERROR_BIND_OVERLAP),
2359                              super_prefixed(a->name),
2360                              super_prefixed(old->name));
2361        if (!a)
2362                return keep_entry(old, o);
2363        else
2364                return merged_entry(a, NULL, o);
2365}
2366
2367/*
2368 * One-way merge.
2369 *
2370 * The rule is:
2371 * - take the stat information from stage0, take the data from stage1
2372 */
2373int oneway_merge(const struct cache_entry * const *src,
2374                 struct unpack_trees_options *o)
2375{
2376        const struct cache_entry *old = src[0];
2377        const struct cache_entry *a = src[1];
2378
2379        if (o->merge_size != 1)
2380                return error("Cannot do a oneway merge of %d trees",
2381                             o->merge_size);
2382
2383        if (!a || a == o->df_conflict_entry)
2384                return deleted_entry(old, old, o);
2385
2386        if (old && same(old, a)) {
2387                int update = 0;
2388                if (o->reset && o->update && !ce_uptodate(old) && !ce_skip_worktree(old)) {
2389                        struct stat st;
2390                        if (lstat(old->name, &st) ||
2391                            ie_match_stat(o->src_index, old, &st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE))
2392                                update |= CE_UPDATE;
2393                }
2394                if (o->update && S_ISGITLINK(old->ce_mode) &&
2395                    should_update_submodules() && !verify_uptodate(old, o))
2396                        update |= CE_UPDATE;
2397                add_entry(o, old, update, CE_STAGEMASK);
2398                return 0;
2399        }
2400        return merged_entry(a, old, o);
2401}