0c7722b2205107273cd5da5f70a6776c99006078
   1#include "cache.h"
   2#include "tree-walk.h"
   3#include "unpack-trees.h"
   4#include "dir.h"
   5#include "object-store.h"
   6#include "tree.h"
   7#include "pathspec.h"
   8
   9static const char *get_mode(const char *str, unsigned int *modep)
  10{
  11        unsigned char c;
  12        unsigned int mode = 0;
  13
  14        if (*str == ' ')
  15                return NULL;
  16
  17        while ((c = *str++) != ' ') {
  18                if (c < '0' || c > '7')
  19                        return NULL;
  20                mode = (mode << 3) + (c - '0');
  21        }
  22        *modep = mode;
  23        return str;
  24}
  25
  26static int decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size, struct strbuf *err)
  27{
  28        const char *path;
  29        unsigned int mode, len;
  30        const unsigned hashsz = the_hash_algo->rawsz;
  31
  32        if (size < hashsz + 3 || buf[size - (hashsz + 1)]) {
  33                strbuf_addstr(err, _("too-short tree object"));
  34                return -1;
  35        }
  36
  37        path = get_mode(buf, &mode);
  38        if (!path) {
  39                strbuf_addstr(err, _("malformed mode in tree entry"));
  40                return -1;
  41        }
  42        if (!*path) {
  43                strbuf_addstr(err, _("empty filename in tree entry"));
  44                return -1;
  45        }
  46        len = strlen(path) + 1;
  47
  48        /* Initialize the descriptor entry */
  49        desc->entry.path = path;
  50        desc->entry.mode = canon_mode(mode);
  51        desc->entry.pathlen = len - 1;
  52        hashcpy(desc->entry.oid.hash, (const unsigned char *)path + len);
  53
  54        return 0;
  55}
  56
  57static int init_tree_desc_internal(struct tree_desc *desc, const void *buffer, unsigned long size, struct strbuf *err)
  58{
  59        desc->buffer = buffer;
  60        desc->size = size;
  61        if (size)
  62                return decode_tree_entry(desc, buffer, size, err);
  63        return 0;
  64}
  65
  66void init_tree_desc(struct tree_desc *desc, const void *buffer, unsigned long size)
  67{
  68        struct strbuf err = STRBUF_INIT;
  69        if (init_tree_desc_internal(desc, buffer, size, &err))
  70                die("%s", err.buf);
  71        strbuf_release(&err);
  72}
  73
  74int init_tree_desc_gently(struct tree_desc *desc, const void *buffer, unsigned long size)
  75{
  76        struct strbuf err = STRBUF_INIT;
  77        int result = init_tree_desc_internal(desc, buffer, size, &err);
  78        if (result)
  79                error("%s", err.buf);
  80        strbuf_release(&err);
  81        return result;
  82}
  83
  84void *fill_tree_descriptor(struct tree_desc *desc, const struct object_id *oid)
  85{
  86        unsigned long size = 0;
  87        void *buf = NULL;
  88
  89        if (oid) {
  90                buf = read_object_with_reference(the_repository, oid, tree_type, &size, NULL);
  91                if (!buf)
  92                        die("unable to read tree %s", oid_to_hex(oid));
  93        }
  94        init_tree_desc(desc, buf, size);
  95        return buf;
  96}
  97
  98static void entry_clear(struct name_entry *a)
  99{
 100        memset(a, 0, sizeof(*a));
 101}
 102
 103static void entry_extract(struct tree_desc *t, struct name_entry *a)
 104{
 105        *a = t->entry;
 106}
 107
 108static int update_tree_entry_internal(struct tree_desc *desc, struct strbuf *err)
 109{
 110        const void *buf = desc->buffer;
 111        const unsigned char *end = (const unsigned char *)desc->entry.path + desc->entry.pathlen + 1 + the_hash_algo->rawsz;
 112        unsigned long size = desc->size;
 113        unsigned long len = end - (const unsigned char *)buf;
 114
 115        if (size < len)
 116                die(_("too-short tree file"));
 117        buf = end;
 118        size -= len;
 119        desc->buffer = buf;
 120        desc->size = size;
 121        if (size)
 122                return decode_tree_entry(desc, buf, size, err);
 123        return 0;
 124}
 125
 126void update_tree_entry(struct tree_desc *desc)
 127{
 128        struct strbuf err = STRBUF_INIT;
 129        if (update_tree_entry_internal(desc, &err))
 130                die("%s", err.buf);
 131        strbuf_release(&err);
 132}
 133
 134int update_tree_entry_gently(struct tree_desc *desc)
 135{
 136        struct strbuf err = STRBUF_INIT;
 137        if (update_tree_entry_internal(desc, &err)) {
 138                error("%s", err.buf);
 139                strbuf_release(&err);
 140                /* Stop processing this tree after error */
 141                desc->size = 0;
 142                return -1;
 143        }
 144        strbuf_release(&err);
 145        return 0;
 146}
 147
 148int tree_entry(struct tree_desc *desc, struct name_entry *entry)
 149{
 150        if (!desc->size)
 151                return 0;
 152
 153        *entry = desc->entry;
 154        update_tree_entry(desc);
 155        return 1;
 156}
 157
 158int tree_entry_gently(struct tree_desc *desc, struct name_entry *entry)
 159{
 160        if (!desc->size)
 161                return 0;
 162
 163        *entry = desc->entry;
 164        if (update_tree_entry_gently(desc))
 165                return 0;
 166        return 1;
 167}
 168
 169void setup_traverse_info(struct traverse_info *info, const char *base)
 170{
 171        int pathlen = strlen(base);
 172        static struct traverse_info dummy;
 173
 174        memset(info, 0, sizeof(*info));
 175        if (pathlen && base[pathlen-1] == '/')
 176                pathlen--;
 177        info->pathlen = pathlen ? pathlen + 1 : 0;
 178        info->name.path = base;
 179        info->name.pathlen = pathlen;
 180        if (pathlen) {
 181                hashcpy(info->name.oid.hash, (const unsigned char *)base + pathlen + 1);
 182                info->prev = &dummy;
 183        }
 184}
 185
 186char *make_traverse_path(char *path, const struct traverse_info *info, const struct name_entry *n)
 187{
 188        int len = tree_entry_len(n);
 189        int pathlen = info->pathlen;
 190
 191        path[pathlen + len] = 0;
 192        for (;;) {
 193                memcpy(path + pathlen, n->path, len);
 194                if (!pathlen)
 195                        break;
 196                path[--pathlen] = '/';
 197                n = &info->name;
 198                len = tree_entry_len(n);
 199                info = info->prev;
 200                pathlen -= len;
 201        }
 202        return path;
 203}
 204
 205struct tree_desc_skip {
 206        struct tree_desc_skip *prev;
 207        const void *ptr;
 208};
 209
 210struct tree_desc_x {
 211        struct tree_desc d;
 212        struct tree_desc_skip *skip;
 213};
 214
 215static int check_entry_match(const char *a, int a_len, const char *b, int b_len)
 216{
 217        /*
 218         * The caller wants to pick *a* from a tree or nothing.
 219         * We are looking at *b* in a tree.
 220         *
 221         * (0) If a and b are the same name, we are trivially happy.
 222         *
 223         * There are three possibilities where *a* could be hiding
 224         * behind *b*.
 225         *
 226         * (1) *a* == "t",   *b* == "ab"  i.e. *b* sorts earlier than *a* no
 227         *                                matter what.
 228         * (2) *a* == "t",   *b* == "t-2" and "t" is a subtree in the tree;
 229         * (3) *a* == "t-2", *b* == "t"   and "t-2" is a blob in the tree.
 230         *
 231         * Otherwise we know *a* won't appear in the tree without
 232         * scanning further.
 233         */
 234
 235        int cmp = name_compare(a, a_len, b, b_len);
 236
 237        /* Most common case first -- reading sync'd trees */
 238        if (!cmp)
 239                return cmp;
 240
 241        if (0 < cmp) {
 242                /* a comes after b; it does not matter if it is case (3)
 243                if (b_len < a_len && !memcmp(a, b, b_len) && a[b_len] < '/')
 244                        return 1;
 245                */
 246                return 1; /* keep looking */
 247        }
 248
 249        /* b comes after a; are we looking at case (2)? */
 250        if (a_len < b_len && !memcmp(a, b, a_len) && b[a_len] < '/')
 251                return 1; /* keep looking */
 252
 253        return -1; /* a cannot appear in the tree */
 254}
 255
 256/*
 257 * From the extended tree_desc, extract the first name entry, while
 258 * paying attention to the candidate "first" name.  Most importantly,
 259 * when looking for an entry, if there are entries that sorts earlier
 260 * in the tree object representation than that name, skip them and
 261 * process the named entry first.  We will remember that we haven't
 262 * processed the first entry yet, and in the later call skip the
 263 * entry we processed early when update_extended_entry() is called.
 264 *
 265 * E.g. if the underlying tree object has these entries:
 266 *
 267 *    blob    "t-1"
 268 *    blob    "t-2"
 269 *    tree    "t"
 270 *    blob    "t=1"
 271 *
 272 * and the "first" asks for "t", remember that we still need to
 273 * process "t-1" and "t-2" but extract "t".  After processing the
 274 * entry "t" from this call, the caller will let us know by calling
 275 * update_extended_entry() that we can remember "t" has been processed
 276 * already.
 277 */
 278
 279static void extended_entry_extract(struct tree_desc_x *t,
 280                                   struct name_entry *a,
 281                                   const char *first,
 282                                   int first_len)
 283{
 284        const char *path;
 285        int len;
 286        struct tree_desc probe;
 287        struct tree_desc_skip *skip;
 288
 289        /*
 290         * Extract the first entry from the tree_desc, but skip the
 291         * ones that we already returned in earlier rounds.
 292         */
 293        while (1) {
 294                if (!t->d.size) {
 295                        entry_clear(a);
 296                        break; /* not found */
 297                }
 298                entry_extract(&t->d, a);
 299                for (skip = t->skip; skip; skip = skip->prev)
 300                        if (a->path == skip->ptr)
 301                                break; /* found */
 302                if (!skip)
 303                        break;
 304                /* We have processed this entry already. */
 305                update_tree_entry(&t->d);
 306        }
 307
 308        if (!first || !a->path)
 309                return;
 310
 311        /*
 312         * The caller wants "first" from this tree, or nothing.
 313         */
 314        path = a->path;
 315        len = tree_entry_len(a);
 316        switch (check_entry_match(first, first_len, path, len)) {
 317        case -1:
 318                entry_clear(a);
 319        case 0:
 320                return;
 321        default:
 322                break;
 323        }
 324
 325        /*
 326         * We need to look-ahead -- we suspect that a subtree whose
 327         * name is "first" may be hiding behind the current entry "path".
 328         */
 329        probe = t->d;
 330        while (probe.size) {
 331                entry_extract(&probe, a);
 332                path = a->path;
 333                len = tree_entry_len(a);
 334                switch (check_entry_match(first, first_len, path, len)) {
 335                case -1:
 336                        entry_clear(a);
 337                case 0:
 338                        return;
 339                default:
 340                        update_tree_entry(&probe);
 341                        break;
 342                }
 343                /* keep looking */
 344        }
 345        entry_clear(a);
 346}
 347
 348static void update_extended_entry(struct tree_desc_x *t, struct name_entry *a)
 349{
 350        if (t->d.entry.path == a->path) {
 351                update_tree_entry(&t->d);
 352        } else {
 353                /* we have returned this entry early */
 354                struct tree_desc_skip *skip = xmalloc(sizeof(*skip));
 355                skip->ptr = a->path;
 356                skip->prev = t->skip;
 357                t->skip = skip;
 358        }
 359}
 360
 361static void free_extended_entry(struct tree_desc_x *t)
 362{
 363        struct tree_desc_skip *p, *s;
 364
 365        for (s = t->skip; s; s = p) {
 366                p = s->prev;
 367                free(s);
 368        }
 369}
 370
 371static inline int prune_traversal(struct index_state *istate,
 372                                  struct name_entry *e,
 373                                  struct traverse_info *info,
 374                                  struct strbuf *base,
 375                                  int still_interesting)
 376{
 377        if (!info->pathspec || still_interesting == 2)
 378                return 2;
 379        if (still_interesting < 0)
 380                return still_interesting;
 381        return tree_entry_interesting(istate, e, base,
 382                                      0, info->pathspec);
 383}
 384
 385int traverse_trees(struct index_state *istate,
 386                   int n, struct tree_desc *t,
 387                   struct traverse_info *info)
 388{
 389        int error = 0;
 390        struct name_entry *entry = xmalloc(n*sizeof(*entry));
 391        int i;
 392        struct tree_desc_x *tx = xcalloc(n, sizeof(*tx));
 393        struct strbuf base = STRBUF_INIT;
 394        int interesting = 1;
 395        char *traverse_path;
 396
 397        for (i = 0; i < n; i++)
 398                tx[i].d = t[i];
 399
 400        if (info->prev) {
 401                strbuf_grow(&base, info->pathlen);
 402                make_traverse_path(base.buf, info->prev, &info->name);
 403                base.buf[info->pathlen-1] = '/';
 404                strbuf_setlen(&base, info->pathlen);
 405                traverse_path = xstrndup(base.buf, info->pathlen);
 406        } else {
 407                traverse_path = xstrndup(info->name.path, info->pathlen);
 408        }
 409        info->traverse_path = traverse_path;
 410        for (;;) {
 411                int trees_used;
 412                unsigned long mask, dirmask;
 413                const char *first = NULL;
 414                int first_len = 0;
 415                struct name_entry *e = NULL;
 416                int len;
 417
 418                for (i = 0; i < n; i++) {
 419                        e = entry + i;
 420                        extended_entry_extract(tx + i, e, NULL, 0);
 421                }
 422
 423                /*
 424                 * A tree may have "t-2" at the current location even
 425                 * though it may have "t" that is a subtree behind it,
 426                 * and another tree may return "t".  We want to grab
 427                 * all "t" from all trees to match in such a case.
 428                 */
 429                for (i = 0; i < n; i++) {
 430                        e = entry + i;
 431                        if (!e->path)
 432                                continue;
 433                        len = tree_entry_len(e);
 434                        if (!first) {
 435                                first = e->path;
 436                                first_len = len;
 437                                continue;
 438                        }
 439                        if (name_compare(e->path, len, first, first_len) < 0) {
 440                                first = e->path;
 441                                first_len = len;
 442                        }
 443                }
 444
 445                if (first) {
 446                        for (i = 0; i < n; i++) {
 447                                e = entry + i;
 448                                extended_entry_extract(tx + i, e, first, first_len);
 449                                /* Cull the ones that are not the earliest */
 450                                if (!e->path)
 451                                        continue;
 452                                len = tree_entry_len(e);
 453                                if (name_compare(e->path, len, first, first_len))
 454                                        entry_clear(e);
 455                        }
 456                }
 457
 458                /* Now we have in entry[i] the earliest name from the trees */
 459                mask = 0;
 460                dirmask = 0;
 461                for (i = 0; i < n; i++) {
 462                        if (!entry[i].path)
 463                                continue;
 464                        mask |= 1ul << i;
 465                        if (S_ISDIR(entry[i].mode))
 466                                dirmask |= 1ul << i;
 467                        e = &entry[i];
 468                }
 469                if (!mask)
 470                        break;
 471                interesting = prune_traversal(istate, e, info, &base, interesting);
 472                if (interesting < 0)
 473                        break;
 474                if (interesting) {
 475                        trees_used = info->fn(n, mask, dirmask, entry, info);
 476                        if (trees_used < 0) {
 477                                error = trees_used;
 478                                if (!info->show_all_errors)
 479                                        break;
 480                        }
 481                        mask &= trees_used;
 482                }
 483                for (i = 0; i < n; i++)
 484                        if (mask & (1ul << i))
 485                                update_extended_entry(tx + i, entry + i);
 486        }
 487        free(entry);
 488        for (i = 0; i < n; i++)
 489                free_extended_entry(tx + i);
 490        free(tx);
 491        free(traverse_path);
 492        info->traverse_path = NULL;
 493        strbuf_release(&base);
 494        return error;
 495}
 496
 497struct dir_state {
 498        void *tree;
 499        unsigned long size;
 500        struct object_id oid;
 501};
 502
 503static int find_tree_entry(struct tree_desc *t, const char *name, struct object_id *result, unsigned short *mode)
 504{
 505        int namelen = strlen(name);
 506        while (t->size) {
 507                const char *entry;
 508                struct object_id oid;
 509                int entrylen, cmp;
 510
 511                oidcpy(&oid, tree_entry_extract(t, &entry, mode));
 512                entrylen = tree_entry_len(&t->entry);
 513                update_tree_entry(t);
 514                if (entrylen > namelen)
 515                        continue;
 516                cmp = memcmp(name, entry, entrylen);
 517                if (cmp > 0)
 518                        continue;
 519                if (cmp < 0)
 520                        break;
 521                if (entrylen == namelen) {
 522                        oidcpy(result, &oid);
 523                        return 0;
 524                }
 525                if (name[entrylen] != '/')
 526                        continue;
 527                if (!S_ISDIR(*mode))
 528                        break;
 529                if (++entrylen == namelen) {
 530                        oidcpy(result, &oid);
 531                        return 0;
 532                }
 533                return get_tree_entry(&oid, name + entrylen, result, mode);
 534        }
 535        return -1;
 536}
 537
 538int get_tree_entry(const struct object_id *tree_oid, const char *name, struct object_id *oid, unsigned short *mode)
 539{
 540        int retval;
 541        void *tree;
 542        unsigned long size;
 543        struct object_id root;
 544
 545        tree = read_object_with_reference(the_repository, tree_oid, tree_type, &size, &root);
 546        if (!tree)
 547                return -1;
 548
 549        if (name[0] == '\0') {
 550                oidcpy(oid, &root);
 551                free(tree);
 552                return 0;
 553        }
 554
 555        if (!size) {
 556                retval = -1;
 557        } else {
 558                struct tree_desc t;
 559                init_tree_desc(&t, tree, size);
 560                retval = find_tree_entry(&t, name, oid, mode);
 561        }
 562        free(tree);
 563        return retval;
 564}
 565
 566/*
 567 * This is Linux's built-in max for the number of symlinks to follow.
 568 * That limit, of course, does not affect git, but it's a reasonable
 569 * choice.
 570 */
 571#define GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS 40
 572
 573/**
 574 * Find a tree entry by following symlinks in tree_sha (which is
 575 * assumed to be the root of the repository).  In the event that a
 576 * symlink points outside the repository (e.g. a link to /foo or a
 577 * root-level link to ../foo), the portion of the link which is
 578 * outside the repository will be returned in result_path, and *mode
 579 * will be set to 0.  It is assumed that result_path is uninitialized.
 580 * If there are no symlinks, or the end result of the symlink chain
 581 * points to an object inside the repository, result will be filled in
 582 * with the sha1 of the found object, and *mode will hold the mode of
 583 * the object.
 584 *
 585 * See the code for enum get_oid_result for a description of
 586 * the return values.
 587 */
 588enum get_oid_result get_tree_entry_follow_symlinks(struct object_id *tree_oid, const char *name, struct object_id *result, struct strbuf *result_path, unsigned short *mode)
 589{
 590        int retval = MISSING_OBJECT;
 591        struct dir_state *parents = NULL;
 592        size_t parents_alloc = 0;
 593        size_t i, parents_nr = 0;
 594        struct object_id current_tree_oid;
 595        struct strbuf namebuf = STRBUF_INIT;
 596        struct tree_desc t;
 597        int follows_remaining = GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS;
 598
 599        init_tree_desc(&t, NULL, 0UL);
 600        strbuf_addstr(&namebuf, name);
 601        oidcpy(&current_tree_oid, tree_oid);
 602
 603        while (1) {
 604                int find_result;
 605                char *first_slash;
 606                char *remainder = NULL;
 607
 608                if (!t.buffer) {
 609                        void *tree;
 610                        struct object_id root;
 611                        unsigned long size;
 612                        tree = read_object_with_reference(the_repository,
 613                                                          &current_tree_oid,
 614                                                          tree_type, &size,
 615                                                          &root);
 616                        if (!tree)
 617                                goto done;
 618
 619                        ALLOC_GROW(parents, parents_nr + 1, parents_alloc);
 620                        parents[parents_nr].tree = tree;
 621                        parents[parents_nr].size = size;
 622                        oidcpy(&parents[parents_nr].oid, &root);
 623                        parents_nr++;
 624
 625                        if (namebuf.buf[0] == '\0') {
 626                                oidcpy(result, &root);
 627                                retval = FOUND;
 628                                goto done;
 629                        }
 630
 631                        if (!size)
 632                                goto done;
 633
 634                        /* descend */
 635                        init_tree_desc(&t, tree, size);
 636                }
 637
 638                /* Handle symlinks to e.g. a//b by removing leading slashes */
 639                while (namebuf.buf[0] == '/') {
 640                        strbuf_remove(&namebuf, 0, 1);
 641                }
 642
 643                /* Split namebuf into a first component and a remainder */
 644                if ((first_slash = strchr(namebuf.buf, '/'))) {
 645                        *first_slash = 0;
 646                        remainder = first_slash + 1;
 647                }
 648
 649                if (!strcmp(namebuf.buf, "..")) {
 650                        struct dir_state *parent;
 651                        /*
 652                         * We could end up with .. in the namebuf if it
 653                         * appears in a symlink.
 654                         */
 655
 656                        if (parents_nr == 1) {
 657                                if (remainder)
 658                                        *first_slash = '/';
 659                                strbuf_add(result_path, namebuf.buf,
 660                                           namebuf.len);
 661                                *mode = 0;
 662                                retval = FOUND;
 663                                goto done;
 664                        }
 665                        parent = &parents[parents_nr - 1];
 666                        free(parent->tree);
 667                        parents_nr--;
 668                        parent = &parents[parents_nr - 1];
 669                        init_tree_desc(&t, parent->tree, parent->size);
 670                        strbuf_remove(&namebuf, 0, remainder ? 3 : 2);
 671                        continue;
 672                }
 673
 674                /* We could end up here via a symlink to dir/.. */
 675                if (namebuf.buf[0] == '\0') {
 676                        oidcpy(result, &parents[parents_nr - 1].oid);
 677                        retval = FOUND;
 678                        goto done;
 679                }
 680
 681                /* Look up the first (or only) path component in the tree. */
 682                find_result = find_tree_entry(&t, namebuf.buf,
 683                                              &current_tree_oid, mode);
 684                if (find_result) {
 685                        goto done;
 686                }
 687
 688                if (S_ISDIR(*mode)) {
 689                        if (!remainder) {
 690                                oidcpy(result, &current_tree_oid);
 691                                retval = FOUND;
 692                                goto done;
 693                        }
 694                        /* Descend the tree */
 695                        t.buffer = NULL;
 696                        strbuf_remove(&namebuf, 0,
 697                                      1 + first_slash - namebuf.buf);
 698                } else if (S_ISREG(*mode)) {
 699                        if (!remainder) {
 700                                oidcpy(result, &current_tree_oid);
 701                                retval = FOUND;
 702                        } else {
 703                                retval = NOT_DIR;
 704                        }
 705                        goto done;
 706                } else if (S_ISLNK(*mode)) {
 707                        /* Follow a symlink */
 708                        unsigned long link_len;
 709                        size_t len;
 710                        char *contents, *contents_start;
 711                        struct dir_state *parent;
 712                        enum object_type type;
 713
 714                        if (follows_remaining-- == 0) {
 715                                /* Too many symlinks followed */
 716                                retval = SYMLINK_LOOP;
 717                                goto done;
 718                        }
 719
 720                        /*
 721                         * At this point, we have followed at a least
 722                         * one symlink, so on error we need to report this.
 723                         */
 724                        retval = DANGLING_SYMLINK;
 725
 726                        contents = read_object_file(&current_tree_oid, &type,
 727                                                    &link_len);
 728
 729                        if (!contents)
 730                                goto done;
 731
 732                        if (contents[0] == '/') {
 733                                strbuf_addstr(result_path, contents);
 734                                free(contents);
 735                                *mode = 0;
 736                                retval = FOUND;
 737                                goto done;
 738                        }
 739
 740                        if (remainder)
 741                                len = first_slash - namebuf.buf;
 742                        else
 743                                len = namebuf.len;
 744
 745                        contents_start = contents;
 746
 747                        parent = &parents[parents_nr - 1];
 748                        init_tree_desc(&t, parent->tree, parent->size);
 749                        strbuf_splice(&namebuf, 0, len,
 750                                      contents_start, link_len);
 751                        if (remainder)
 752                                namebuf.buf[link_len] = '/';
 753                        free(contents);
 754                }
 755        }
 756done:
 757        for (i = 0; i < parents_nr; i++)
 758                free(parents[i].tree);
 759        free(parents);
 760
 761        strbuf_release(&namebuf);
 762        return retval;
 763}
 764
 765static int match_entry(const struct pathspec_item *item,
 766                       const struct name_entry *entry, int pathlen,
 767                       const char *match, int matchlen,
 768                       enum interesting *never_interesting)
 769{
 770        int m = -1; /* signals that we haven't called strncmp() */
 771
 772        if (item->magic & PATHSPEC_ICASE)
 773                /*
 774                 * "Never interesting" trick requires exact
 775                 * matching. We could do something clever with inexact
 776                 * matching, but it's trickier (and not to forget that
 777                 * strcasecmp is locale-dependent, at least in
 778                 * glibc). Just disable it for now. It can't be worse
 779                 * than the wildcard's codepath of '[Tt][Hi][Is][Ss]'
 780                 * pattern.
 781                 */
 782                *never_interesting = entry_not_interesting;
 783        else if (*never_interesting != entry_not_interesting) {
 784                /*
 785                 * We have not seen any match that sorts later
 786                 * than the current path.
 787                 */
 788
 789                /*
 790                 * Does match sort strictly earlier than path
 791                 * with their common parts?
 792                 */
 793                m = strncmp(match, entry->path,
 794                            (matchlen < pathlen) ? matchlen : pathlen);
 795                if (m < 0)
 796                        return 0;
 797
 798                /*
 799                 * If we come here even once, that means there is at
 800                 * least one pathspec that would sort equal to or
 801                 * later than the path we are currently looking at.
 802                 * In other words, if we have never reached this point
 803                 * after iterating all pathspecs, it means all
 804                 * pathspecs are either outside of base, or inside the
 805                 * base but sorts strictly earlier than the current
 806                 * one.  In either case, they will never match the
 807                 * subsequent entries.  In such a case, we initialized
 808                 * the variable to -1 and that is what will be
 809                 * returned, allowing the caller to terminate early.
 810                 */
 811                *never_interesting = entry_not_interesting;
 812        }
 813
 814        if (pathlen > matchlen)
 815                return 0;
 816
 817        if (matchlen > pathlen) {
 818                if (match[pathlen] != '/')
 819                        return 0;
 820                if (!S_ISDIR(entry->mode) && !S_ISGITLINK(entry->mode))
 821                        return 0;
 822        }
 823
 824        if (m == -1)
 825                /*
 826                 * we cheated and did not do strncmp(), so we do
 827                 * that here.
 828                 */
 829                m = ps_strncmp(item, match, entry->path, pathlen);
 830
 831        /*
 832         * If common part matched earlier then it is a hit,
 833         * because we rejected the case where path is not a
 834         * leading directory and is shorter than match.
 835         */
 836        if (!m)
 837                /*
 838                 * match_entry does not check if the prefix part is
 839                 * matched case-sensitively. If the entry is a
 840                 * directory and part of prefix, it'll be rematched
 841                 * eventually by basecmp with special treatment for
 842                 * the prefix.
 843                 */
 844                return 1;
 845
 846        return 0;
 847}
 848
 849/* :(icase)-aware string compare */
 850static int basecmp(const struct pathspec_item *item,
 851                   const char *base, const char *match, int len)
 852{
 853        if (item->magic & PATHSPEC_ICASE) {
 854                int ret, n = len > item->prefix ? item->prefix : len;
 855                ret = strncmp(base, match, n);
 856                if (ret)
 857                        return ret;
 858                base += n;
 859                match += n;
 860                len -= n;
 861        }
 862        return ps_strncmp(item, base, match, len);
 863}
 864
 865static int match_dir_prefix(const struct pathspec_item *item,
 866                            const char *base,
 867                            const char *match, int matchlen)
 868{
 869        if (basecmp(item, base, match, matchlen))
 870                return 0;
 871
 872        /*
 873         * If the base is a subdirectory of a path which
 874         * was specified, all of them are interesting.
 875         */
 876        if (!matchlen ||
 877            base[matchlen] == '/' ||
 878            match[matchlen - 1] == '/')
 879                return 1;
 880
 881        /* Just a random prefix match */
 882        return 0;
 883}
 884
 885/*
 886 * Perform matching on the leading non-wildcard part of
 887 * pathspec. item->nowildcard_len must be greater than zero. Return
 888 * non-zero if base is matched.
 889 */
 890static int match_wildcard_base(const struct pathspec_item *item,
 891                               const char *base, int baselen,
 892                               int *matched)
 893{
 894        const char *match = item->match;
 895        /* the wildcard part is not considered in this function */
 896        int matchlen = item->nowildcard_len;
 897
 898        if (baselen) {
 899                int dirlen;
 900                /*
 901                 * Return early if base is longer than the
 902                 * non-wildcard part but it does not match.
 903                 */
 904                if (baselen >= matchlen) {
 905                        *matched = matchlen;
 906                        return !basecmp(item, base, match, matchlen);
 907                }
 908
 909                dirlen = matchlen;
 910                while (dirlen && match[dirlen - 1] != '/')
 911                        dirlen--;
 912
 913                /*
 914                 * Return early if base is shorter than the
 915                 * non-wildcard part but it does not match. Note that
 916                 * base ends with '/' so we are sure it really matches
 917                 * directory
 918                 */
 919                if (basecmp(item, base, match, baselen))
 920                        return 0;
 921                *matched = baselen;
 922        } else
 923                *matched = 0;
 924        /*
 925         * we could have checked entry against the non-wildcard part
 926         * that is not in base and does similar never_interesting
 927         * optimization as in match_entry. For now just be happy with
 928         * base comparison.
 929         */
 930        return entry_interesting;
 931}
 932
 933/*
 934 * Is a tree entry interesting given the pathspec we have?
 935 *
 936 * Pre-condition: either baselen == base_offset (i.e. empty path)
 937 * or base[baselen-1] == '/' (i.e. with trailing slash).
 938 */
 939static enum interesting do_match(struct index_state *istate,
 940                                 const struct name_entry *entry,
 941                                 struct strbuf *base, int base_offset,
 942                                 const struct pathspec *ps,
 943                                 int exclude)
 944{
 945        int i;
 946        int pathlen, baselen = base->len - base_offset;
 947        enum interesting never_interesting = ps->has_wildcard ?
 948                entry_not_interesting : all_entries_not_interesting;
 949
 950        GUARD_PATHSPEC(ps,
 951                       PATHSPEC_FROMTOP |
 952                       PATHSPEC_MAXDEPTH |
 953                       PATHSPEC_LITERAL |
 954                       PATHSPEC_GLOB |
 955                       PATHSPEC_ICASE |
 956                       PATHSPEC_EXCLUDE |
 957                       PATHSPEC_ATTR);
 958
 959        if (!ps->nr) {
 960                if (!ps->recursive ||
 961                    !(ps->magic & PATHSPEC_MAXDEPTH) ||
 962                    ps->max_depth == -1)
 963                        return all_entries_interesting;
 964                return within_depth(base->buf + base_offset, baselen,
 965                                    !!S_ISDIR(entry->mode),
 966                                    ps->max_depth) ?
 967                        entry_interesting : entry_not_interesting;
 968        }
 969
 970        pathlen = tree_entry_len(entry);
 971
 972        for (i = ps->nr - 1; i >= 0; i--) {
 973                const struct pathspec_item *item = ps->items+i;
 974                const char *match = item->match;
 975                const char *base_str = base->buf + base_offset;
 976                int matchlen = item->len, matched = 0;
 977
 978                if ((!exclude &&   item->magic & PATHSPEC_EXCLUDE) ||
 979                    ( exclude && !(item->magic & PATHSPEC_EXCLUDE)))
 980                        continue;
 981
 982                if (baselen >= matchlen) {
 983                        /* If it doesn't match, move along... */
 984                        if (!match_dir_prefix(item, base_str, match, matchlen))
 985                                goto match_wildcards;
 986
 987                        if (!ps->recursive ||
 988                            !(ps->magic & PATHSPEC_MAXDEPTH) ||
 989                            ps->max_depth == -1) {
 990                                if (!item->attr_match_nr)
 991                                        return all_entries_interesting;
 992                                else
 993                                        goto interesting;
 994                        }
 995
 996                        if (within_depth(base_str + matchlen + 1,
 997                                         baselen - matchlen - 1,
 998                                         !!S_ISDIR(entry->mode),
 999                                         ps->max_depth))
1000                                goto interesting;
1001                        else
1002                                return entry_not_interesting;
1003                }
1004
1005                /* Either there must be no base, or the base must match. */
1006                if (baselen == 0 || !basecmp(item, base_str, match, baselen)) {
1007                        if (match_entry(item, entry, pathlen,
1008                                        match + baselen, matchlen - baselen,
1009                                        &never_interesting))
1010                                goto interesting;
1011
1012                        if (item->nowildcard_len < item->len) {
1013                                if (!git_fnmatch(item, match + baselen, entry->path,
1014                                                 item->nowildcard_len - baselen))
1015                                        goto interesting;
1016
1017                                /*
1018                                 * Match all directories. We'll try to
1019                                 * match files later on.
1020                                 */
1021                                if (ps->recursive && S_ISDIR(entry->mode))
1022                                        return entry_interesting;
1023
1024                                /*
1025                                 * When matching against submodules with
1026                                 * wildcard characters, ensure that the entry
1027                                 * at least matches up to the first wild
1028                                 * character.  More accurate matching can then
1029                                 * be performed in the submodule itself.
1030                                 */
1031                                if (ps->recurse_submodules &&
1032                                    S_ISGITLINK(entry->mode) &&
1033                                    !ps_strncmp(item, match + baselen,
1034                                                entry->path,
1035                                                item->nowildcard_len - baselen))
1036                                        goto interesting;
1037                        }
1038
1039                        continue;
1040                }
1041
1042match_wildcards:
1043                if (item->nowildcard_len == item->len)
1044                        continue;
1045
1046                if (item->nowildcard_len &&
1047                    !match_wildcard_base(item, base_str, baselen, &matched))
1048                        continue;
1049
1050                /*
1051                 * Concatenate base and entry->path into one and do
1052                 * fnmatch() on it.
1053                 *
1054                 * While we could avoid concatenation in certain cases
1055                 * [1], which saves a memcpy and potentially a
1056                 * realloc, it turns out not worth it. Measurement on
1057                 * linux-2.6 does not show any clear improvements,
1058                 * partly because of the nowildcard_len optimization
1059                 * in git_fnmatch(). Avoid micro-optimizations here.
1060                 *
1061                 * [1] if match_wildcard_base() says the base
1062                 * directory is already matched, we only need to match
1063                 * the rest, which is shorter so _in theory_ faster.
1064                 */
1065
1066                strbuf_add(base, entry->path, pathlen);
1067
1068                if (!git_fnmatch(item, match, base->buf + base_offset,
1069                                 item->nowildcard_len)) {
1070                        strbuf_setlen(base, base_offset + baselen);
1071                        goto interesting;
1072                }
1073
1074                /*
1075                 * When matching against submodules with
1076                 * wildcard characters, ensure that the entry
1077                 * at least matches up to the first wild
1078                 * character.  More accurate matching can then
1079                 * be performed in the submodule itself.
1080                 */
1081                if (ps->recurse_submodules && S_ISGITLINK(entry->mode) &&
1082                    !ps_strncmp(item, match, base->buf + base_offset,
1083                                item->nowildcard_len)) {
1084                        strbuf_setlen(base, base_offset + baselen);
1085                        goto interesting;
1086                }
1087
1088                strbuf_setlen(base, base_offset + baselen);
1089
1090                /*
1091                 * Match all directories. We'll try to match files
1092                 * later on.
1093                 * max_depth is ignored but we may consider support it
1094                 * in future, see
1095                 * https://public-inbox.org/git/7vmxo5l2g4.fsf@alter.siamese.dyndns.org/
1096                 */
1097                if (ps->recursive && S_ISDIR(entry->mode))
1098                        return entry_interesting;
1099                continue;
1100interesting:
1101                if (item->attr_match_nr) {
1102                        int ret;
1103
1104                        /*
1105                         * Must not return all_entries_not_interesting
1106                         * prematurely. We do not know if all entries do not
1107                         * match some attributes with current attr API.
1108                         */
1109                        never_interesting = entry_not_interesting;
1110
1111                        /*
1112                         * Consider all directories interesting (because some
1113                         * of those files inside may match some attributes
1114                         * even though the parent dir does not)
1115                         *
1116                         * FIXME: attributes _can_ match directories and we
1117                         * can probably return all_entries_interesting or
1118                         * all_entries_not_interesting here if matched.
1119                         */
1120                        if (S_ISDIR(entry->mode))
1121                                return entry_interesting;
1122
1123                        strbuf_add(base, entry->path, pathlen);
1124                        ret = match_pathspec_attrs(istate, base->buf + base_offset,
1125                                                   base->len - base_offset, item);
1126                        strbuf_setlen(base, base_offset + baselen);
1127                        if (!ret)
1128                                continue;
1129                }
1130                return entry_interesting;
1131        }
1132        return never_interesting; /* No matches */
1133}
1134
1135/*
1136 * Is a tree entry interesting given the pathspec we have?
1137 *
1138 * Pre-condition: either baselen == base_offset (i.e. empty path)
1139 * or base[baselen-1] == '/' (i.e. with trailing slash).
1140 */
1141enum interesting tree_entry_interesting(struct index_state *istate,
1142                                        const struct name_entry *entry,
1143                                        struct strbuf *base, int base_offset,
1144                                        const struct pathspec *ps)
1145{
1146        enum interesting positive, negative;
1147        positive = do_match(istate, entry, base, base_offset, ps, 0);
1148
1149        /*
1150         * case | entry | positive | negative | result
1151         * -----+-------+----------+----------+-------
1152         *   1  |  file |   -1     |  -1..2   |  -1
1153         *   2  |  file |    0     |  -1..2   |   0
1154         *   3  |  file |    1     |   -1     |   1
1155         *   4  |  file |    1     |    0     |   1
1156         *   5  |  file |    1     |    1     |   0
1157         *   6  |  file |    1     |    2     |   0
1158         *   7  |  file |    2     |   -1     |   2
1159         *   8  |  file |    2     |    0     |   1
1160         *   9  |  file |    2     |    1     |   0
1161         *  10  |  file |    2     |    2     |  -1
1162         * -----+-------+----------+----------+-------
1163         *  11  |  dir  |   -1     |  -1..2   |  -1
1164         *  12  |  dir  |    0     |  -1..2   |   0
1165         *  13  |  dir  |    1     |   -1     |   1
1166         *  14  |  dir  |    1     |    0     |   1
1167         *  15  |  dir  |    1     |    1     |   1 (*)
1168         *  16  |  dir  |    1     |    2     |   0
1169         *  17  |  dir  |    2     |   -1     |   2
1170         *  18  |  dir  |    2     |    0     |   1
1171         *  19  |  dir  |    2     |    1     |   1 (*)
1172         *  20  |  dir  |    2     |    2     |  -1
1173         *
1174         * (*) An exclude pattern interested in a directory does not
1175         * necessarily mean it will exclude all of the directory. In
1176         * wildcard case, it can't decide until looking at individual
1177         * files inside. So don't write such directories off yet.
1178         */
1179
1180        if (!(ps->magic & PATHSPEC_EXCLUDE) ||
1181            positive <= entry_not_interesting) /* #1, #2, #11, #12 */
1182                return positive;
1183
1184        negative = do_match(istate, entry, base, base_offset, ps, 1);
1185
1186        /* #8, #18 */
1187        if (positive == all_entries_interesting &&
1188            negative == entry_not_interesting)
1189                return entry_interesting;
1190
1191        /* #3, #4, #7, #13, #14, #17 */
1192        if (negative <= entry_not_interesting)
1193                return positive;
1194
1195        /* #15, #19 */
1196        if (S_ISDIR(entry->mode) &&
1197            positive >= entry_interesting &&
1198            negative == entry_interesting)
1199                return entry_interesting;
1200
1201        if ((positive == entry_interesting &&
1202             negative >= entry_interesting) || /* #5, #6, #16 */
1203            (positive == all_entries_interesting &&
1204             negative == entry_interesting)) /* #9 */
1205                return entry_not_interesting;
1206
1207        return all_entries_not_interesting; /* #10, #20 */
1208}