tree-walk.con commit Merge branch 'jk/vreport-sanitize' (daf75f2)
   1#include "cache.h"
   2#include "tree-walk.h"
   3#include "unpack-trees.h"
   4#include "dir.h"
   5#include "tree.h"
   6#include "pathspec.h"
   7
   8static const char *get_mode(const char *str, unsigned int *modep)
   9{
  10        unsigned char c;
  11        unsigned int mode = 0;
  12
  13        if (*str == ' ')
  14                return NULL;
  15
  16        while ((c = *str++) != ' ') {
  17                if (c < '0' || c > '7')
  18                        return NULL;
  19                mode = (mode << 3) + (c - '0');
  20        }
  21        *modep = mode;
  22        return str;
  23}
  24
  25static int decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size, struct strbuf *err)
  26{
  27        const char *path;
  28        unsigned int mode, len;
  29
  30        if (size < 23 || buf[size - 21]) {
  31                strbuf_addstr(err, _("too-short tree object"));
  32                return -1;
  33        }
  34
  35        path = get_mode(buf, &mode);
  36        if (!path) {
  37                strbuf_addstr(err, _("malformed mode in tree entry"));
  38                return -1;
  39        }
  40        if (!*path) {
  41                strbuf_addstr(err, _("empty filename in tree entry"));
  42                return -1;
  43        }
  44        len = strlen(path) + 1;
  45
  46        /* Initialize the descriptor entry */
  47        desc->entry.path = path;
  48        desc->entry.mode = canon_mode(mode);
  49        desc->entry.oid  = (const struct object_id *)(path + len);
  50
  51        return 0;
  52}
  53
  54static int init_tree_desc_internal(struct tree_desc *desc, const void *buffer, unsigned long size, struct strbuf *err)
  55{
  56        desc->buffer = buffer;
  57        desc->size = size;
  58        if (size)
  59                return decode_tree_entry(desc, buffer, size, err);
  60        return 0;
  61}
  62
  63void init_tree_desc(struct tree_desc *desc, const void *buffer, unsigned long size)
  64{
  65        struct strbuf err = STRBUF_INIT;
  66        if (init_tree_desc_internal(desc, buffer, size, &err))
  67                die("%s", err.buf);
  68        strbuf_release(&err);
  69}
  70
  71int init_tree_desc_gently(struct tree_desc *desc, const void *buffer, unsigned long size)
  72{
  73        struct strbuf err = STRBUF_INIT;
  74        int result = init_tree_desc_internal(desc, buffer, size, &err);
  75        if (result)
  76                error("%s", err.buf);
  77        strbuf_release(&err);
  78        return result;
  79}
  80
  81void *fill_tree_descriptor(struct tree_desc *desc, const unsigned char *sha1)
  82{
  83        unsigned long size = 0;
  84        void *buf = NULL;
  85
  86        if (sha1) {
  87                buf = read_object_with_reference(sha1, tree_type, &size, NULL);
  88                if (!buf)
  89                        die("unable to read tree %s", sha1_to_hex(sha1));
  90        }
  91        init_tree_desc(desc, buf, size);
  92        return buf;
  93}
  94
  95static void entry_clear(struct name_entry *a)
  96{
  97        memset(a, 0, sizeof(*a));
  98}
  99
 100static void entry_extract(struct tree_desc *t, struct name_entry *a)
 101{
 102        *a = t->entry;
 103}
 104
 105static int update_tree_entry_internal(struct tree_desc *desc, struct strbuf *err)
 106{
 107        const void *buf = desc->buffer;
 108        const unsigned char *end = desc->entry.oid->hash + 20;
 109        unsigned long size = desc->size;
 110        unsigned long len = end - (const unsigned char *)buf;
 111
 112        if (size < len)
 113                die(_("too-short tree file"));
 114        buf = end;
 115        size -= len;
 116        desc->buffer = buf;
 117        desc->size = size;
 118        if (size)
 119                return decode_tree_entry(desc, buf, size, err);
 120        return 0;
 121}
 122
 123void update_tree_entry(struct tree_desc *desc)
 124{
 125        struct strbuf err = STRBUF_INIT;
 126        if (update_tree_entry_internal(desc, &err))
 127                die("%s", err.buf);
 128        strbuf_release(&err);
 129}
 130
 131int update_tree_entry_gently(struct tree_desc *desc)
 132{
 133        struct strbuf err = STRBUF_INIT;
 134        if (update_tree_entry_internal(desc, &err)) {
 135                error("%s", err.buf);
 136                strbuf_release(&err);
 137                /* Stop processing this tree after error */
 138                desc->size = 0;
 139                return -1;
 140        }
 141        strbuf_release(&err);
 142        return 0;
 143}
 144
 145int tree_entry(struct tree_desc *desc, struct name_entry *entry)
 146{
 147        if (!desc->size)
 148                return 0;
 149
 150        *entry = desc->entry;
 151        update_tree_entry(desc);
 152        return 1;
 153}
 154
 155int tree_entry_gently(struct tree_desc *desc, struct name_entry *entry)
 156{
 157        if (!desc->size)
 158                return 0;
 159
 160        *entry = desc->entry;
 161        if (update_tree_entry_gently(desc))
 162                return 0;
 163        return 1;
 164}
 165
 166void setup_traverse_info(struct traverse_info *info, const char *base)
 167{
 168        int pathlen = strlen(base);
 169        static struct traverse_info dummy;
 170
 171        memset(info, 0, sizeof(*info));
 172        if (pathlen && base[pathlen-1] == '/')
 173                pathlen--;
 174        info->pathlen = pathlen ? pathlen + 1 : 0;
 175        info->name.path = base;
 176        info->name.oid = (void *)(base + pathlen + 1);
 177        if (pathlen)
 178                info->prev = &dummy;
 179}
 180
 181char *make_traverse_path(char *path, const struct traverse_info *info, const struct name_entry *n)
 182{
 183        int len = tree_entry_len(n);
 184        int pathlen = info->pathlen;
 185
 186        path[pathlen + len] = 0;
 187        for (;;) {
 188                memcpy(path + pathlen, n->path, len);
 189                if (!pathlen)
 190                        break;
 191                path[--pathlen] = '/';
 192                n = &info->name;
 193                len = tree_entry_len(n);
 194                info = info->prev;
 195                pathlen -= len;
 196        }
 197        return path;
 198}
 199
 200struct tree_desc_skip {
 201        struct tree_desc_skip *prev;
 202        const void *ptr;
 203};
 204
 205struct tree_desc_x {
 206        struct tree_desc d;
 207        struct tree_desc_skip *skip;
 208};
 209
 210static int check_entry_match(const char *a, int a_len, const char *b, int b_len)
 211{
 212        /*
 213         * The caller wants to pick *a* from a tree or nothing.
 214         * We are looking at *b* in a tree.
 215         *
 216         * (0) If a and b are the same name, we are trivially happy.
 217         *
 218         * There are three possibilities where *a* could be hiding
 219         * behind *b*.
 220         *
 221         * (1) *a* == "t",   *b* == "ab"  i.e. *b* sorts earlier than *a* no
 222         *                                matter what.
 223         * (2) *a* == "t",   *b* == "t-2" and "t" is a subtree in the tree;
 224         * (3) *a* == "t-2", *b* == "t"   and "t-2" is a blob in the tree.
 225         *
 226         * Otherwise we know *a* won't appear in the tree without
 227         * scanning further.
 228         */
 229
 230        int cmp = name_compare(a, a_len, b, b_len);
 231
 232        /* Most common case first -- reading sync'd trees */
 233        if (!cmp)
 234                return cmp;
 235
 236        if (0 < cmp) {
 237                /* a comes after b; it does not matter if it is case (3)
 238                if (b_len < a_len && !memcmp(a, b, b_len) && a[b_len] < '/')
 239                        return 1;
 240                */
 241                return 1; /* keep looking */
 242        }
 243
 244        /* b comes after a; are we looking at case (2)? */
 245        if (a_len < b_len && !memcmp(a, b, a_len) && b[a_len] < '/')
 246                return 1; /* keep looking */
 247
 248        return -1; /* a cannot appear in the tree */
 249}
 250
 251/*
 252 * From the extended tree_desc, extract the first name entry, while
 253 * paying attention to the candidate "first" name.  Most importantly,
 254 * when looking for an entry, if there are entries that sorts earlier
 255 * in the tree object representation than that name, skip them and
 256 * process the named entry first.  We will remember that we haven't
 257 * processed the first entry yet, and in the later call skip the
 258 * entry we processed early when update_extended_entry() is called.
 259 *
 260 * E.g. if the underlying tree object has these entries:
 261 *
 262 *    blob    "t-1"
 263 *    blob    "t-2"
 264 *    tree    "t"
 265 *    blob    "t=1"
 266 *
 267 * and the "first" asks for "t", remember that we still need to
 268 * process "t-1" and "t-2" but extract "t".  After processing the
 269 * entry "t" from this call, the caller will let us know by calling
 270 * update_extended_entry() that we can remember "t" has been processed
 271 * already.
 272 */
 273
 274static void extended_entry_extract(struct tree_desc_x *t,
 275                                   struct name_entry *a,
 276                                   const char *first,
 277                                   int first_len)
 278{
 279        const char *path;
 280        int len;
 281        struct tree_desc probe;
 282        struct tree_desc_skip *skip;
 283
 284        /*
 285         * Extract the first entry from the tree_desc, but skip the
 286         * ones that we already returned in earlier rounds.
 287         */
 288        while (1) {
 289                if (!t->d.size) {
 290                        entry_clear(a);
 291                        break; /* not found */
 292                }
 293                entry_extract(&t->d, a);
 294                for (skip = t->skip; skip; skip = skip->prev)
 295                        if (a->path == skip->ptr)
 296                                break; /* found */
 297                if (!skip)
 298                        break;
 299                /* We have processed this entry already. */
 300                update_tree_entry(&t->d);
 301        }
 302
 303        if (!first || !a->path)
 304                return;
 305
 306        /*
 307         * The caller wants "first" from this tree, or nothing.
 308         */
 309        path = a->path;
 310        len = tree_entry_len(a);
 311        switch (check_entry_match(first, first_len, path, len)) {
 312        case -1:
 313                entry_clear(a);
 314        case 0:
 315                return;
 316        default:
 317                break;
 318        }
 319
 320        /*
 321         * We need to look-ahead -- we suspect that a subtree whose
 322         * name is "first" may be hiding behind the current entry "path".
 323         */
 324        probe = t->d;
 325        while (probe.size) {
 326                entry_extract(&probe, a);
 327                path = a->path;
 328                len = tree_entry_len(a);
 329                switch (check_entry_match(first, first_len, path, len)) {
 330                case -1:
 331                        entry_clear(a);
 332                case 0:
 333                        return;
 334                default:
 335                        update_tree_entry(&probe);
 336                        break;
 337                }
 338                /* keep looking */
 339        }
 340        entry_clear(a);
 341}
 342
 343static void update_extended_entry(struct tree_desc_x *t, struct name_entry *a)
 344{
 345        if (t->d.entry.path == a->path) {
 346                update_tree_entry(&t->d);
 347        } else {
 348                /* we have returned this entry early */
 349                struct tree_desc_skip *skip = xmalloc(sizeof(*skip));
 350                skip->ptr = a->path;
 351                skip->prev = t->skip;
 352                t->skip = skip;
 353        }
 354}
 355
 356static void free_extended_entry(struct tree_desc_x *t)
 357{
 358        struct tree_desc_skip *p, *s;
 359
 360        for (s = t->skip; s; s = p) {
 361                p = s->prev;
 362                free(s);
 363        }
 364}
 365
 366static inline int prune_traversal(struct name_entry *e,
 367                                  struct traverse_info *info,
 368                                  struct strbuf *base,
 369                                  int still_interesting)
 370{
 371        if (!info->pathspec || still_interesting == 2)
 372                return 2;
 373        if (still_interesting < 0)
 374                return still_interesting;
 375        return tree_entry_interesting(e, base, 0, info->pathspec);
 376}
 377
 378int traverse_trees(int n, struct tree_desc *t, struct traverse_info *info)
 379{
 380        int error = 0;
 381        struct name_entry *entry = xmalloc(n*sizeof(*entry));
 382        int i;
 383        struct tree_desc_x *tx = xcalloc(n, sizeof(*tx));
 384        struct strbuf base = STRBUF_INIT;
 385        int interesting = 1;
 386        char *traverse_path;
 387
 388        for (i = 0; i < n; i++)
 389                tx[i].d = t[i];
 390
 391        if (info->prev) {
 392                strbuf_grow(&base, info->pathlen);
 393                make_traverse_path(base.buf, info->prev, &info->name);
 394                base.buf[info->pathlen-1] = '/';
 395                strbuf_setlen(&base, info->pathlen);
 396                traverse_path = xstrndup(base.buf, info->pathlen);
 397        } else {
 398                traverse_path = xstrndup(info->name.path, info->pathlen);
 399        }
 400        info->traverse_path = traverse_path;
 401        for (;;) {
 402                int trees_used;
 403                unsigned long mask, dirmask;
 404                const char *first = NULL;
 405                int first_len = 0;
 406                struct name_entry *e = NULL;
 407                int len;
 408
 409                for (i = 0; i < n; i++) {
 410                        e = entry + i;
 411                        extended_entry_extract(tx + i, e, NULL, 0);
 412                }
 413
 414                /*
 415                 * A tree may have "t-2" at the current location even
 416                 * though it may have "t" that is a subtree behind it,
 417                 * and another tree may return "t".  We want to grab
 418                 * all "t" from all trees to match in such a case.
 419                 */
 420                for (i = 0; i < n; i++) {
 421                        e = entry + i;
 422                        if (!e->path)
 423                                continue;
 424                        len = tree_entry_len(e);
 425                        if (!first) {
 426                                first = e->path;
 427                                first_len = len;
 428                                continue;
 429                        }
 430                        if (name_compare(e->path, len, first, first_len) < 0) {
 431                                first = e->path;
 432                                first_len = len;
 433                        }
 434                }
 435
 436                if (first) {
 437                        for (i = 0; i < n; i++) {
 438                                e = entry + i;
 439                                extended_entry_extract(tx + i, e, first, first_len);
 440                                /* Cull the ones that are not the earliest */
 441                                if (!e->path)
 442                                        continue;
 443                                len = tree_entry_len(e);
 444                                if (name_compare(e->path, len, first, first_len))
 445                                        entry_clear(e);
 446                        }
 447                }
 448
 449                /* Now we have in entry[i] the earliest name from the trees */
 450                mask = 0;
 451                dirmask = 0;
 452                for (i = 0; i < n; i++) {
 453                        if (!entry[i].path)
 454                                continue;
 455                        mask |= 1ul << i;
 456                        if (S_ISDIR(entry[i].mode))
 457                                dirmask |= 1ul << i;
 458                        e = &entry[i];
 459                }
 460                if (!mask)
 461                        break;
 462                interesting = prune_traversal(e, info, &base, interesting);
 463                if (interesting < 0)
 464                        break;
 465                if (interesting) {
 466                        trees_used = info->fn(n, mask, dirmask, entry, info);
 467                        if (trees_used < 0) {
 468                                error = trees_used;
 469                                if (!info->show_all_errors)
 470                                        break;
 471                        }
 472                        mask &= trees_used;
 473                }
 474                for (i = 0; i < n; i++)
 475                        if (mask & (1ul << i))
 476                                update_extended_entry(tx + i, entry + i);
 477        }
 478        free(entry);
 479        for (i = 0; i < n; i++)
 480                free_extended_entry(tx + i);
 481        free(tx);
 482        free(traverse_path);
 483        info->traverse_path = NULL;
 484        strbuf_release(&base);
 485        return error;
 486}
 487
 488struct dir_state {
 489        void *tree;
 490        unsigned long size;
 491        unsigned char sha1[20];
 492};
 493
 494static int find_tree_entry(struct tree_desc *t, const char *name, unsigned char *result, unsigned *mode)
 495{
 496        int namelen = strlen(name);
 497        while (t->size) {
 498                const char *entry;
 499                const struct object_id *oid;
 500                int entrylen, cmp;
 501
 502                oid = tree_entry_extract(t, &entry, mode);
 503                entrylen = tree_entry_len(&t->entry);
 504                update_tree_entry(t);
 505                if (entrylen > namelen)
 506                        continue;
 507                cmp = memcmp(name, entry, entrylen);
 508                if (cmp > 0)
 509                        continue;
 510                if (cmp < 0)
 511                        break;
 512                if (entrylen == namelen) {
 513                        hashcpy(result, oid->hash);
 514                        return 0;
 515                }
 516                if (name[entrylen] != '/')
 517                        continue;
 518                if (!S_ISDIR(*mode))
 519                        break;
 520                if (++entrylen == namelen) {
 521                        hashcpy(result, oid->hash);
 522                        return 0;
 523                }
 524                return get_tree_entry(oid->hash, name + entrylen, result, mode);
 525        }
 526        return -1;
 527}
 528
 529int get_tree_entry(const unsigned char *tree_sha1, const char *name, unsigned char *sha1, unsigned *mode)
 530{
 531        int retval;
 532        void *tree;
 533        unsigned long size;
 534        unsigned char root[20];
 535
 536        tree = read_object_with_reference(tree_sha1, tree_type, &size, root);
 537        if (!tree)
 538                return -1;
 539
 540        if (name[0] == '\0') {
 541                hashcpy(sha1, root);
 542                free(tree);
 543                return 0;
 544        }
 545
 546        if (!size) {
 547                retval = -1;
 548        } else {
 549                struct tree_desc t;
 550                init_tree_desc(&t, tree, size);
 551                retval = find_tree_entry(&t, name, sha1, mode);
 552        }
 553        free(tree);
 554        return retval;
 555}
 556
 557/*
 558 * This is Linux's built-in max for the number of symlinks to follow.
 559 * That limit, of course, does not affect git, but it's a reasonable
 560 * choice.
 561 */
 562#define GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS 40
 563
 564/**
 565 * Find a tree entry by following symlinks in tree_sha (which is
 566 * assumed to be the root of the repository).  In the event that a
 567 * symlink points outside the repository (e.g. a link to /foo or a
 568 * root-level link to ../foo), the portion of the link which is
 569 * outside the repository will be returned in result_path, and *mode
 570 * will be set to 0.  It is assumed that result_path is uninitialized.
 571 * If there are no symlinks, or the end result of the symlink chain
 572 * points to an object inside the repository, result will be filled in
 573 * with the sha1 of the found object, and *mode will hold the mode of
 574 * the object.
 575 *
 576 * See the code for enum follow_symlink_result for a description of
 577 * the return values.
 578 */
 579enum follow_symlinks_result get_tree_entry_follow_symlinks(unsigned char *tree_sha1, const char *name, unsigned char *result, struct strbuf *result_path, unsigned *mode)
 580{
 581        int retval = MISSING_OBJECT;
 582        struct dir_state *parents = NULL;
 583        size_t parents_alloc = 0;
 584        ssize_t parents_nr = 0;
 585        unsigned char current_tree_sha1[20];
 586        struct strbuf namebuf = STRBUF_INIT;
 587        struct tree_desc t;
 588        int follows_remaining = GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS;
 589        int i;
 590
 591        init_tree_desc(&t, NULL, 0UL);
 592        strbuf_init(result_path, 0);
 593        strbuf_addstr(&namebuf, name);
 594        hashcpy(current_tree_sha1, tree_sha1);
 595
 596        while (1) {
 597                int find_result;
 598                char *first_slash;
 599                char *remainder = NULL;
 600
 601                if (!t.buffer) {
 602                        void *tree;
 603                        unsigned char root[20];
 604                        unsigned long size;
 605                        tree = read_object_with_reference(current_tree_sha1,
 606                                                          tree_type, &size,
 607                                                          root);
 608                        if (!tree)
 609                                goto done;
 610
 611                        ALLOC_GROW(parents, parents_nr + 1, parents_alloc);
 612                        parents[parents_nr].tree = tree;
 613                        parents[parents_nr].size = size;
 614                        hashcpy(parents[parents_nr].sha1, root);
 615                        parents_nr++;
 616
 617                        if (namebuf.buf[0] == '\0') {
 618                                hashcpy(result, root);
 619                                retval = FOUND;
 620                                goto done;
 621                        }
 622
 623                        if (!size)
 624                                goto done;
 625
 626                        /* descend */
 627                        init_tree_desc(&t, tree, size);
 628                }
 629
 630                /* Handle symlinks to e.g. a//b by removing leading slashes */
 631                while (namebuf.buf[0] == '/') {
 632                        strbuf_remove(&namebuf, 0, 1);
 633                }
 634
 635                /* Split namebuf into a first component and a remainder */
 636                if ((first_slash = strchr(namebuf.buf, '/'))) {
 637                        *first_slash = 0;
 638                        remainder = first_slash + 1;
 639                }
 640
 641                if (!strcmp(namebuf.buf, "..")) {
 642                        struct dir_state *parent;
 643                        /*
 644                         * We could end up with .. in the namebuf if it
 645                         * appears in a symlink.
 646                         */
 647
 648                        if (parents_nr == 1) {
 649                                if (remainder)
 650                                        *first_slash = '/';
 651                                strbuf_add(result_path, namebuf.buf,
 652                                           namebuf.len);
 653                                *mode = 0;
 654                                retval = FOUND;
 655                                goto done;
 656                        }
 657                        parent = &parents[parents_nr - 1];
 658                        free(parent->tree);
 659                        parents_nr--;
 660                        parent = &parents[parents_nr - 1];
 661                        init_tree_desc(&t, parent->tree, parent->size);
 662                        strbuf_remove(&namebuf, 0, remainder ? 3 : 2);
 663                        continue;
 664                }
 665
 666                /* We could end up here via a symlink to dir/.. */
 667                if (namebuf.buf[0] == '\0') {
 668                        hashcpy(result, parents[parents_nr - 1].sha1);
 669                        retval = FOUND;
 670                        goto done;
 671                }
 672
 673                /* Look up the first (or only) path component in the tree. */
 674                find_result = find_tree_entry(&t, namebuf.buf,
 675                                              current_tree_sha1, mode);
 676                if (find_result) {
 677                        goto done;
 678                }
 679
 680                if (S_ISDIR(*mode)) {
 681                        if (!remainder) {
 682                                hashcpy(result, current_tree_sha1);
 683                                retval = FOUND;
 684                                goto done;
 685                        }
 686                        /* Descend the tree */
 687                        t.buffer = NULL;
 688                        strbuf_remove(&namebuf, 0,
 689                                      1 + first_slash - namebuf.buf);
 690                } else if (S_ISREG(*mode)) {
 691                        if (!remainder) {
 692                                hashcpy(result, current_tree_sha1);
 693                                retval = FOUND;
 694                        } else {
 695                                retval = NOT_DIR;
 696                        }
 697                        goto done;
 698                } else if (S_ISLNK(*mode)) {
 699                        /* Follow a symlink */
 700                        unsigned long link_len;
 701                        size_t len;
 702                        char *contents, *contents_start;
 703                        struct dir_state *parent;
 704                        enum object_type type;
 705
 706                        if (follows_remaining-- == 0) {
 707                                /* Too many symlinks followed */
 708                                retval = SYMLINK_LOOP;
 709                                goto done;
 710                        }
 711
 712                        /*
 713                         * At this point, we have followed at a least
 714                         * one symlink, so on error we need to report this.
 715                         */
 716                        retval = DANGLING_SYMLINK;
 717
 718                        contents = read_sha1_file(current_tree_sha1, &type,
 719                                                  &link_len);
 720
 721                        if (!contents)
 722                                goto done;
 723
 724                        if (contents[0] == '/') {
 725                                strbuf_addstr(result_path, contents);
 726                                free(contents);
 727                                *mode = 0;
 728                                retval = FOUND;
 729                                goto done;
 730                        }
 731
 732                        if (remainder)
 733                                len = first_slash - namebuf.buf;
 734                        else
 735                                len = namebuf.len;
 736
 737                        contents_start = contents;
 738
 739                        parent = &parents[parents_nr - 1];
 740                        init_tree_desc(&t, parent->tree, parent->size);
 741                        strbuf_splice(&namebuf, 0, len,
 742                                      contents_start, link_len);
 743                        if (remainder)
 744                                namebuf.buf[link_len] = '/';
 745                        free(contents);
 746                }
 747        }
 748done:
 749        for (i = 0; i < parents_nr; i++)
 750                free(parents[i].tree);
 751        free(parents);
 752
 753        strbuf_release(&namebuf);
 754        return retval;
 755}
 756
 757static int match_entry(const struct pathspec_item *item,
 758                       const struct name_entry *entry, int pathlen,
 759                       const char *match, int matchlen,
 760                       enum interesting *never_interesting)
 761{
 762        int m = -1; /* signals that we haven't called strncmp() */
 763
 764        if (item->magic & PATHSPEC_ICASE)
 765                /*
 766                 * "Never interesting" trick requires exact
 767                 * matching. We could do something clever with inexact
 768                 * matching, but it's trickier (and not to forget that
 769                 * strcasecmp is locale-dependent, at least in
 770                 * glibc). Just disable it for now. It can't be worse
 771                 * than the wildcard's codepath of '[Tt][Hi][Is][Ss]'
 772                 * pattern.
 773                 */
 774                *never_interesting = entry_not_interesting;
 775        else if (*never_interesting != entry_not_interesting) {
 776                /*
 777                 * We have not seen any match that sorts later
 778                 * than the current path.
 779                 */
 780
 781                /*
 782                 * Does match sort strictly earlier than path
 783                 * with their common parts?
 784                 */
 785                m = strncmp(match, entry->path,
 786                            (matchlen < pathlen) ? matchlen : pathlen);
 787                if (m < 0)
 788                        return 0;
 789
 790                /*
 791                 * If we come here even once, that means there is at
 792                 * least one pathspec that would sort equal to or
 793                 * later than the path we are currently looking at.
 794                 * In other words, if we have never reached this point
 795                 * after iterating all pathspecs, it means all
 796                 * pathspecs are either outside of base, or inside the
 797                 * base but sorts strictly earlier than the current
 798                 * one.  In either case, they will never match the
 799                 * subsequent entries.  In such a case, we initialized
 800                 * the variable to -1 and that is what will be
 801                 * returned, allowing the caller to terminate early.
 802                 */
 803                *never_interesting = entry_not_interesting;
 804        }
 805
 806        if (pathlen > matchlen)
 807                return 0;
 808
 809        if (matchlen > pathlen) {
 810                if (match[pathlen] != '/')
 811                        return 0;
 812                if (!S_ISDIR(entry->mode) && !S_ISGITLINK(entry->mode))
 813                        return 0;
 814        }
 815
 816        if (m == -1)
 817                /*
 818                 * we cheated and did not do strncmp(), so we do
 819                 * that here.
 820                 */
 821                m = ps_strncmp(item, match, entry->path, pathlen);
 822
 823        /*
 824         * If common part matched earlier then it is a hit,
 825         * because we rejected the case where path is not a
 826         * leading directory and is shorter than match.
 827         */
 828        if (!m)
 829                /*
 830                 * match_entry does not check if the prefix part is
 831                 * matched case-sensitively. If the entry is a
 832                 * directory and part of prefix, it'll be rematched
 833                 * eventually by basecmp with special treatment for
 834                 * the prefix.
 835                 */
 836                return 1;
 837
 838        return 0;
 839}
 840
 841/* :(icase)-aware string compare */
 842static int basecmp(const struct pathspec_item *item,
 843                   const char *base, const char *match, int len)
 844{
 845        if (item->magic & PATHSPEC_ICASE) {
 846                int ret, n = len > item->prefix ? item->prefix : len;
 847                ret = strncmp(base, match, n);
 848                if (ret)
 849                        return ret;
 850                base += n;
 851                match += n;
 852                len -= n;
 853        }
 854        return ps_strncmp(item, base, match, len);
 855}
 856
 857static int match_dir_prefix(const struct pathspec_item *item,
 858                            const char *base,
 859                            const char *match, int matchlen)
 860{
 861        if (basecmp(item, base, match, matchlen))
 862                return 0;
 863
 864        /*
 865         * If the base is a subdirectory of a path which
 866         * was specified, all of them are interesting.
 867         */
 868        if (!matchlen ||
 869            base[matchlen] == '/' ||
 870            match[matchlen - 1] == '/')
 871                return 1;
 872
 873        /* Just a random prefix match */
 874        return 0;
 875}
 876
 877/*
 878 * Perform matching on the leading non-wildcard part of
 879 * pathspec. item->nowildcard_len must be greater than zero. Return
 880 * non-zero if base is matched.
 881 */
 882static int match_wildcard_base(const struct pathspec_item *item,
 883                               const char *base, int baselen,
 884                               int *matched)
 885{
 886        const char *match = item->match;
 887        /* the wildcard part is not considered in this function */
 888        int matchlen = item->nowildcard_len;
 889
 890        if (baselen) {
 891                int dirlen;
 892                /*
 893                 * Return early if base is longer than the
 894                 * non-wildcard part but it does not match.
 895                 */
 896                if (baselen >= matchlen) {
 897                        *matched = matchlen;
 898                        return !basecmp(item, base, match, matchlen);
 899                }
 900
 901                dirlen = matchlen;
 902                while (dirlen && match[dirlen - 1] != '/')
 903                        dirlen--;
 904
 905                /*
 906                 * Return early if base is shorter than the
 907                 * non-wildcard part but it does not match. Note that
 908                 * base ends with '/' so we are sure it really matches
 909                 * directory
 910                 */
 911                if (basecmp(item, base, match, baselen))
 912                        return 0;
 913                *matched = baselen;
 914        } else
 915                *matched = 0;
 916        /*
 917         * we could have checked entry against the non-wildcard part
 918         * that is not in base and does similar never_interesting
 919         * optimization as in match_entry. For now just be happy with
 920         * base comparison.
 921         */
 922        return entry_interesting;
 923}
 924
 925/*
 926 * Is a tree entry interesting given the pathspec we have?
 927 *
 928 * Pre-condition: either baselen == base_offset (i.e. empty path)
 929 * or base[baselen-1] == '/' (i.e. with trailing slash).
 930 */
 931static enum interesting do_match(const struct name_entry *entry,
 932                                 struct strbuf *base, int base_offset,
 933                                 const struct pathspec *ps,
 934                                 int exclude)
 935{
 936        int i;
 937        int pathlen, baselen = base->len - base_offset;
 938        enum interesting never_interesting = ps->has_wildcard ?
 939                entry_not_interesting : all_entries_not_interesting;
 940
 941        GUARD_PATHSPEC(ps,
 942                       PATHSPEC_FROMTOP |
 943                       PATHSPEC_MAXDEPTH |
 944                       PATHSPEC_LITERAL |
 945                       PATHSPEC_GLOB |
 946                       PATHSPEC_ICASE |
 947                       PATHSPEC_EXCLUDE);
 948
 949        if (!ps->nr) {
 950                if (!ps->recursive ||
 951                    !(ps->magic & PATHSPEC_MAXDEPTH) ||
 952                    ps->max_depth == -1)
 953                        return all_entries_interesting;
 954                return within_depth(base->buf + base_offset, baselen,
 955                                    !!S_ISDIR(entry->mode),
 956                                    ps->max_depth) ?
 957                        entry_interesting : entry_not_interesting;
 958        }
 959
 960        pathlen = tree_entry_len(entry);
 961
 962        for (i = ps->nr - 1; i >= 0; i--) {
 963                const struct pathspec_item *item = ps->items+i;
 964                const char *match = item->match;
 965                const char *base_str = base->buf + base_offset;
 966                int matchlen = item->len, matched = 0;
 967
 968                if ((!exclude &&   item->magic & PATHSPEC_EXCLUDE) ||
 969                    ( exclude && !(item->magic & PATHSPEC_EXCLUDE)))
 970                        continue;
 971
 972                if (baselen >= matchlen) {
 973                        /* If it doesn't match, move along... */
 974                        if (!match_dir_prefix(item, base_str, match, matchlen))
 975                                goto match_wildcards;
 976
 977                        if (!ps->recursive ||
 978                            !(ps->magic & PATHSPEC_MAXDEPTH) ||
 979                            ps->max_depth == -1)
 980                                return all_entries_interesting;
 981
 982                        return within_depth(base_str + matchlen + 1,
 983                                            baselen - matchlen - 1,
 984                                            !!S_ISDIR(entry->mode),
 985                                            ps->max_depth) ?
 986                                entry_interesting : entry_not_interesting;
 987                }
 988
 989                /* Either there must be no base, or the base must match. */
 990                if (baselen == 0 || !basecmp(item, base_str, match, baselen)) {
 991                        if (match_entry(item, entry, pathlen,
 992                                        match + baselen, matchlen - baselen,
 993                                        &never_interesting))
 994                                return entry_interesting;
 995
 996                        if (item->nowildcard_len < item->len) {
 997                                if (!git_fnmatch(item, match + baselen, entry->path,
 998                                                 item->nowildcard_len - baselen))
 999                                        return entry_interesting;
1000
1001                                /*
1002                                 * Match all directories. We'll try to
1003                                 * match files later on.
1004                                 */
1005                                if (ps->recursive && S_ISDIR(entry->mode))
1006                                        return entry_interesting;
1007
1008                                /*
1009                                 * When matching against submodules with
1010                                 * wildcard characters, ensure that the entry
1011                                 * at least matches up to the first wild
1012                                 * character.  More accurate matching can then
1013                                 * be performed in the submodule itself.
1014                                 */
1015                                if (ps->recursive && S_ISGITLINK(entry->mode) &&
1016                                    !ps_strncmp(item, match + baselen,
1017                                                entry->path,
1018                                                item->nowildcard_len - baselen))
1019                                        return entry_interesting;
1020                        }
1021
1022                        continue;
1023                }
1024
1025match_wildcards:
1026                if (item->nowildcard_len == item->len)
1027                        continue;
1028
1029                if (item->nowildcard_len &&
1030                    !match_wildcard_base(item, base_str, baselen, &matched))
1031                        continue;
1032
1033                /*
1034                 * Concatenate base and entry->path into one and do
1035                 * fnmatch() on it.
1036                 *
1037                 * While we could avoid concatenation in certain cases
1038                 * [1], which saves a memcpy and potentially a
1039                 * realloc, it turns out not worth it. Measurement on
1040                 * linux-2.6 does not show any clear improvements,
1041                 * partly because of the nowildcard_len optimization
1042                 * in git_fnmatch(). Avoid micro-optimizations here.
1043                 *
1044                 * [1] if match_wildcard_base() says the base
1045                 * directory is already matched, we only need to match
1046                 * the rest, which is shorter so _in theory_ faster.
1047                 */
1048
1049                strbuf_add(base, entry->path, pathlen);
1050
1051                if (!git_fnmatch(item, match, base->buf + base_offset,
1052                                 item->nowildcard_len)) {
1053                        strbuf_setlen(base, base_offset + baselen);
1054                        return entry_interesting;
1055                }
1056
1057                /*
1058                 * When matching against submodules with
1059                 * wildcard characters, ensure that the entry
1060                 * at least matches up to the first wild
1061                 * character.  More accurate matching can then
1062                 * be performed in the submodule itself.
1063                 */
1064                if (ps->recursive && S_ISGITLINK(entry->mode) &&
1065                    !ps_strncmp(item, match, base->buf + base_offset,
1066                                item->nowildcard_len)) {
1067                        strbuf_setlen(base, base_offset + baselen);
1068                        return entry_interesting;
1069                }
1070
1071                strbuf_setlen(base, base_offset + baselen);
1072
1073                /*
1074                 * Match all directories. We'll try to match files
1075                 * later on.
1076                 * max_depth is ignored but we may consider support it
1077                 * in future, see
1078                 * http://thread.gmane.org/gmane.comp.version-control.git/163757/focus=163840
1079                 */
1080                if (ps->recursive && S_ISDIR(entry->mode))
1081                        return entry_interesting;
1082        }
1083        return never_interesting; /* No matches */
1084}
1085
1086/*
1087 * Is a tree entry interesting given the pathspec we have?
1088 *
1089 * Pre-condition: either baselen == base_offset (i.e. empty path)
1090 * or base[baselen-1] == '/' (i.e. with trailing slash).
1091 */
1092enum interesting tree_entry_interesting(const struct name_entry *entry,
1093                                        struct strbuf *base, int base_offset,
1094                                        const struct pathspec *ps)
1095{
1096        enum interesting positive, negative;
1097        positive = do_match(entry, base, base_offset, ps, 0);
1098
1099        /*
1100         * case | entry | positive | negative | result
1101         * -----+-------+----------+----------+-------
1102         *   1  |  file |   -1     |  -1..2   |  -1
1103         *   2  |  file |    0     |  -1..2   |   0
1104         *   3  |  file |    1     |   -1     |   1
1105         *   4  |  file |    1     |    0     |   1
1106         *   5  |  file |    1     |    1     |   0
1107         *   6  |  file |    1     |    2     |   0
1108         *   7  |  file |    2     |   -1     |   2
1109         *   8  |  file |    2     |    0     |   2
1110         *   9  |  file |    2     |    1     |   0
1111         *  10  |  file |    2     |    2     |  -1
1112         * -----+-------+----------+----------+-------
1113         *  11  |  dir  |   -1     |  -1..2   |  -1
1114         *  12  |  dir  |    0     |  -1..2   |   0
1115         *  13  |  dir  |    1     |   -1     |   1
1116         *  14  |  dir  |    1     |    0     |   1
1117         *  15  |  dir  |    1     |    1     |   1 (*)
1118         *  16  |  dir  |    1     |    2     |   0
1119         *  17  |  dir  |    2     |   -1     |   2
1120         *  18  |  dir  |    2     |    0     |   2
1121         *  19  |  dir  |    2     |    1     |   1 (*)
1122         *  20  |  dir  |    2     |    2     |  -1
1123         *
1124         * (*) An exclude pattern interested in a directory does not
1125         * necessarily mean it will exclude all of the directory. In
1126         * wildcard case, it can't decide until looking at individual
1127         * files inside. So don't write such directories off yet.
1128         */
1129
1130        if (!(ps->magic & PATHSPEC_EXCLUDE) ||
1131            positive <= entry_not_interesting) /* #1, #2, #11, #12 */
1132                return positive;
1133
1134        negative = do_match(entry, base, base_offset, ps, 1);
1135
1136        /* #3, #4, #7, #8, #13, #14, #17, #18 */
1137        if (negative <= entry_not_interesting)
1138                return positive;
1139
1140        /* #15, #19 */
1141        if (S_ISDIR(entry->mode) &&
1142            positive >= entry_interesting &&
1143            negative == entry_interesting)
1144                return entry_interesting;
1145
1146        if ((positive == entry_interesting &&
1147             negative >= entry_interesting) || /* #5, #6, #16 */
1148            (positive == all_entries_interesting &&
1149             negative == entry_interesting)) /* #9 */
1150                return entry_not_interesting;
1151
1152        return all_entries_not_interesting; /* #10, #20 */
1153}