attr.con commit attr: remove maybe-real, maybe-macro from git_attr (60a1272)
   1/*
   2 * Handle git attributes.  See gitattributes(5) for a description of
   3 * the file syntax, and Documentation/technical/api-gitattributes.txt
   4 * for a description of the API.
   5 *
   6 * One basic design decision here is that we are not going to support
   7 * an insanely large number of attributes.
   8 */
   9
  10#define NO_THE_INDEX_COMPATIBILITY_MACROS
  11#include "cache.h"
  12#include "exec_cmd.h"
  13#include "attr.h"
  14#include "dir.h"
  15#include "utf8.h"
  16#include "quote.h"
  17#include "thread-utils.h"
  18
  19const char git_attr__true[] = "(builtin)true";
  20const char git_attr__false[] = "\0(builtin)false";
  21static const char git_attr__unknown[] = "(builtin)unknown";
  22#define ATTR__TRUE git_attr__true
  23#define ATTR__FALSE git_attr__false
  24#define ATTR__UNSET NULL
  25#define ATTR__UNKNOWN git_attr__unknown
  26
  27#ifndef DEBUG_ATTR
  28#define DEBUG_ATTR 0
  29#endif
  30
  31struct git_attr {
  32        int attr_nr; /* unique attribute number */
  33        char name[FLEX_ARRAY]; /* attribute name */
  34};
  35
  36const char *git_attr_name(const struct git_attr *attr)
  37{
  38        return attr->name;
  39}
  40
  41struct attr_hashmap {
  42        struct hashmap map;
  43#ifndef NO_PTHREADS
  44        pthread_mutex_t mutex;
  45#endif
  46};
  47
  48static inline void hashmap_lock(struct attr_hashmap *map)
  49{
  50#ifndef NO_PTHREADS
  51        pthread_mutex_lock(&map->mutex);
  52#endif
  53}
  54
  55static inline void hashmap_unlock(struct attr_hashmap *map)
  56{
  57#ifndef NO_PTHREADS
  58        pthread_mutex_unlock(&map->mutex);
  59#endif
  60}
  61
  62/*
  63 * The global dictionary of all interned attributes.  This
  64 * is a singleton object which is shared between threads.
  65 * Access to this dictionary must be surrounded with a mutex.
  66 */
  67static struct attr_hashmap g_attr_hashmap;
  68
  69/* The container for objects stored in "struct attr_hashmap" */
  70struct attr_hash_entry {
  71        struct hashmap_entry ent; /* must be the first member! */
  72        const char *key; /* the key; memory should be owned by value */
  73        size_t keylen; /* length of the key */
  74        void *value; /* the stored value */
  75};
  76
  77/* attr_hashmap comparison function */
  78static int attr_hash_entry_cmp(const struct attr_hash_entry *a,
  79                               const struct attr_hash_entry *b,
  80                               void *unused)
  81{
  82        return (a->keylen != b->keylen) || strncmp(a->key, b->key, a->keylen);
  83}
  84
  85/* Initialize an 'attr_hashmap' object */
  86static void attr_hashmap_init(struct attr_hashmap *map)
  87{
  88        hashmap_init(&map->map, (hashmap_cmp_fn) attr_hash_entry_cmp, 0);
  89}
  90
  91/*
  92 * Retrieve the 'value' stored in a hashmap given the provided 'key'.
  93 * If there is no matching entry, return NULL.
  94 */
  95static void *attr_hashmap_get(struct attr_hashmap *map,
  96                              const char *key, size_t keylen)
  97{
  98        struct attr_hash_entry k;
  99        struct attr_hash_entry *e;
 100
 101        if (!map->map.tablesize)
 102                attr_hashmap_init(map);
 103
 104        hashmap_entry_init(&k, memhash(key, keylen));
 105        k.key = key;
 106        k.keylen = keylen;
 107        e = hashmap_get(&map->map, &k, NULL);
 108
 109        return e ? e->value : NULL;
 110}
 111
 112/* Add 'value' to a hashmap based on the provided 'key'. */
 113static void attr_hashmap_add(struct attr_hashmap *map,
 114                             const char *key, size_t keylen,
 115                             void *value)
 116{
 117        struct attr_hash_entry *e;
 118
 119        if (!map->map.tablesize)
 120                attr_hashmap_init(map);
 121
 122        e = xmalloc(sizeof(struct attr_hash_entry));
 123        hashmap_entry_init(e, memhash(key, keylen));
 124        e->key = key;
 125        e->keylen = keylen;
 126        e->value = value;
 127
 128        hashmap_add(&map->map, e);
 129}
 130
 131struct all_attrs_item {
 132        const struct git_attr *attr;
 133        const char *value;
 134        /*
 135         * If 'macro' is non-NULL, indicates that 'attr' is a macro based on
 136         * the current attribute stack and contains a pointer to the match_attr
 137         * definition of the macro
 138         */
 139        const struct match_attr *macro;
 140};
 141
 142/*
 143 * Reallocate and reinitialize the array of all attributes (which is used in
 144 * the attribute collection process) in 'check' based on the global dictionary
 145 * of attributes.
 146 */
 147static void all_attrs_init(struct attr_hashmap *map, struct attr_check *check)
 148{
 149        int i;
 150
 151        hashmap_lock(map);
 152
 153        if (map->map.size < check->all_attrs_nr)
 154                die("BUG: interned attributes shouldn't be deleted");
 155
 156        /*
 157         * If the number of attributes in the global dictionary has increased
 158         * (or this attr_check instance doesn't have an initialized all_attrs
 159         * field), reallocate the provided attr_check instance's all_attrs
 160         * field and fill each entry with its corresponding git_attr.
 161         */
 162        if (map->map.size != check->all_attrs_nr) {
 163                struct attr_hash_entry *e;
 164                struct hashmap_iter iter;
 165                hashmap_iter_init(&map->map, &iter);
 166
 167                REALLOC_ARRAY(check->all_attrs, map->map.size);
 168                check->all_attrs_nr = map->map.size;
 169
 170                while ((e = hashmap_iter_next(&iter))) {
 171                        const struct git_attr *a = e->value;
 172                        check->all_attrs[a->attr_nr].attr = a;
 173                }
 174        }
 175
 176        hashmap_unlock(map);
 177
 178        /*
 179         * Re-initialize every entry in check->all_attrs.
 180         * This re-initialization can live outside of the locked region since
 181         * the attribute dictionary is no longer being accessed.
 182         */
 183        for (i = 0; i < check->all_attrs_nr; i++) {
 184                check->all_attrs[i].value = ATTR__UNKNOWN;
 185                check->all_attrs[i].macro = NULL;
 186        }
 187}
 188
 189static int attr_name_valid(const char *name, size_t namelen)
 190{
 191        /*
 192         * Attribute name cannot begin with '-' and must consist of
 193         * characters from [-A-Za-z0-9_.].
 194         */
 195        if (namelen <= 0 || *name == '-')
 196                return 0;
 197        while (namelen--) {
 198                char ch = *name++;
 199                if (! (ch == '-' || ch == '.' || ch == '_' ||
 200                       ('0' <= ch && ch <= '9') ||
 201                       ('a' <= ch && ch <= 'z') ||
 202                       ('A' <= ch && ch <= 'Z')) )
 203                        return 0;
 204        }
 205        return 1;
 206}
 207
 208static void report_invalid_attr(const char *name, size_t len,
 209                                const char *src, int lineno)
 210{
 211        struct strbuf err = STRBUF_INIT;
 212        strbuf_addf(&err, _("%.*s is not a valid attribute name"),
 213                    (int) len, name);
 214        fprintf(stderr, "%s: %s:%d\n", err.buf, src, lineno);
 215        strbuf_release(&err);
 216}
 217
 218/*
 219 * Given a 'name', lookup and return the corresponding attribute in the global
 220 * dictionary.  If no entry is found, create a new attribute and store it in
 221 * the dictionary.
 222 */
 223static struct git_attr *git_attr_internal(const char *name, int namelen)
 224{
 225        struct git_attr *a;
 226
 227        if (!attr_name_valid(name, namelen))
 228                return NULL;
 229
 230        hashmap_lock(&g_attr_hashmap);
 231
 232        a = attr_hashmap_get(&g_attr_hashmap, name, namelen);
 233
 234        if (!a) {
 235                FLEX_ALLOC_MEM(a, name, name, namelen);
 236                a->attr_nr = g_attr_hashmap.map.size;
 237
 238                attr_hashmap_add(&g_attr_hashmap, a->name, namelen, a);
 239                assert(a->attr_nr == (g_attr_hashmap.map.size - 1));
 240        }
 241
 242        hashmap_unlock(&g_attr_hashmap);
 243
 244        return a;
 245}
 246
 247struct git_attr *git_attr(const char *name)
 248{
 249        return git_attr_internal(name, strlen(name));
 250}
 251
 252/* What does a matched pattern decide? */
 253struct attr_state {
 254        struct git_attr *attr;
 255        const char *setto;
 256};
 257
 258struct pattern {
 259        const char *pattern;
 260        int patternlen;
 261        int nowildcardlen;
 262        unsigned flags;         /* EXC_FLAG_* */
 263};
 264
 265/*
 266 * One rule, as from a .gitattributes file.
 267 *
 268 * If is_macro is true, then u.attr is a pointer to the git_attr being
 269 * defined.
 270 *
 271 * If is_macro is false, then u.pat is the filename pattern to which the
 272 * rule applies.
 273 *
 274 * In either case, num_attr is the number of attributes affected by
 275 * this rule, and state is an array listing them.  The attributes are
 276 * listed as they appear in the file (macros unexpanded).
 277 */
 278struct match_attr {
 279        union {
 280                struct pattern pat;
 281                struct git_attr *attr;
 282        } u;
 283        char is_macro;
 284        unsigned num_attr;
 285        struct attr_state state[FLEX_ARRAY];
 286};
 287
 288static const char blank[] = " \t\r\n";
 289
 290/*
 291 * Parse a whitespace-delimited attribute state (i.e., "attr",
 292 * "-attr", "!attr", or "attr=value") from the string starting at src.
 293 * If e is not NULL, write the results to *e.  Return a pointer to the
 294 * remainder of the string (with leading whitespace removed), or NULL
 295 * if there was an error.
 296 */
 297static const char *parse_attr(const char *src, int lineno, const char *cp,
 298                              struct attr_state *e)
 299{
 300        const char *ep, *equals;
 301        int len;
 302
 303        ep = cp + strcspn(cp, blank);
 304        equals = strchr(cp, '=');
 305        if (equals && ep < equals)
 306                equals = NULL;
 307        if (equals)
 308                len = equals - cp;
 309        else
 310                len = ep - cp;
 311        if (!e) {
 312                if (*cp == '-' || *cp == '!') {
 313                        cp++;
 314                        len--;
 315                }
 316                if (!attr_name_valid(cp, len)) {
 317                        report_invalid_attr(cp, len, src, lineno);
 318                        return NULL;
 319                }
 320        } else {
 321                /*
 322                 * As this function is always called twice, once with
 323                 * e == NULL in the first pass and then e != NULL in
 324                 * the second pass, no need for attr_name_valid()
 325                 * check here.
 326                 */
 327                if (*cp == '-' || *cp == '!') {
 328                        e->setto = (*cp == '-') ? ATTR__FALSE : ATTR__UNSET;
 329                        cp++;
 330                        len--;
 331                }
 332                else if (!equals)
 333                        e->setto = ATTR__TRUE;
 334                else {
 335                        e->setto = xmemdupz(equals + 1, ep - equals - 1);
 336                }
 337                e->attr = git_attr_internal(cp, len);
 338        }
 339        return ep + strspn(ep, blank);
 340}
 341
 342static struct match_attr *parse_attr_line(const char *line, const char *src,
 343                                          int lineno, int macro_ok)
 344{
 345        int namelen;
 346        int num_attr, i;
 347        const char *cp, *name, *states;
 348        struct match_attr *res = NULL;
 349        int is_macro;
 350        struct strbuf pattern = STRBUF_INIT;
 351
 352        cp = line + strspn(line, blank);
 353        if (!*cp || *cp == '#')
 354                return NULL;
 355        name = cp;
 356
 357        if (*cp == '"' && !unquote_c_style(&pattern, name, &states)) {
 358                name = pattern.buf;
 359                namelen = pattern.len;
 360        } else {
 361                namelen = strcspn(name, blank);
 362                states = name + namelen;
 363        }
 364
 365        if (strlen(ATTRIBUTE_MACRO_PREFIX) < namelen &&
 366            starts_with(name, ATTRIBUTE_MACRO_PREFIX)) {
 367                if (!macro_ok) {
 368                        fprintf(stderr, "%s not allowed: %s:%d\n",
 369                                name, src, lineno);
 370                        goto fail_return;
 371                }
 372                is_macro = 1;
 373                name += strlen(ATTRIBUTE_MACRO_PREFIX);
 374                name += strspn(name, blank);
 375                namelen = strcspn(name, blank);
 376                if (!attr_name_valid(name, namelen)) {
 377                        report_invalid_attr(name, namelen, src, lineno);
 378                        goto fail_return;
 379                }
 380        }
 381        else
 382                is_macro = 0;
 383
 384        states += strspn(states, blank);
 385
 386        /* First pass to count the attr_states */
 387        for (cp = states, num_attr = 0; *cp; num_attr++) {
 388                cp = parse_attr(src, lineno, cp, NULL);
 389                if (!cp)
 390                        goto fail_return;
 391        }
 392
 393        res = xcalloc(1,
 394                      sizeof(*res) +
 395                      sizeof(struct attr_state) * num_attr +
 396                      (is_macro ? 0 : namelen + 1));
 397        if (is_macro) {
 398                res->u.attr = git_attr_internal(name, namelen);
 399        } else {
 400                char *p = (char *)&(res->state[num_attr]);
 401                memcpy(p, name, namelen);
 402                res->u.pat.pattern = p;
 403                parse_exclude_pattern(&res->u.pat.pattern,
 404                                      &res->u.pat.patternlen,
 405                                      &res->u.pat.flags,
 406                                      &res->u.pat.nowildcardlen);
 407                if (res->u.pat.flags & EXC_FLAG_NEGATIVE) {
 408                        warning(_("Negative patterns are ignored in git attributes\n"
 409                                  "Use '\\!' for literal leading exclamation."));
 410                        goto fail_return;
 411                }
 412        }
 413        res->is_macro = is_macro;
 414        res->num_attr = num_attr;
 415
 416        /* Second pass to fill the attr_states */
 417        for (cp = states, i = 0; *cp; i++) {
 418                cp = parse_attr(src, lineno, cp, &(res->state[i]));
 419        }
 420
 421        strbuf_release(&pattern);
 422        return res;
 423
 424fail_return:
 425        strbuf_release(&pattern);
 426        free(res);
 427        return NULL;
 428}
 429
 430/*
 431 * Like info/exclude and .gitignore, the attribute information can
 432 * come from many places.
 433 *
 434 * (1) .gitattribute file of the same directory;
 435 * (2) .gitattribute file of the parent directory if (1) does not have
 436 *      any match; this goes recursively upwards, just like .gitignore.
 437 * (3) $GIT_DIR/info/attributes, which overrides both of the above.
 438 *
 439 * In the same file, later entries override the earlier match, so in the
 440 * global list, we would have entries from info/attributes the earliest
 441 * (reading the file from top to bottom), .gitattribute of the root
 442 * directory (again, reading the file from top to bottom) down to the
 443 * current directory, and then scan the list backwards to find the first match.
 444 * This is exactly the same as what is_excluded() does in dir.c to deal with
 445 * .gitignore file and info/excludes file as a fallback.
 446 */
 447
 448/* NEEDSWORK: This will become per git_attr_check */
 449static struct attr_stack {
 450        struct attr_stack *prev;
 451        char *origin;
 452        size_t originlen;
 453        unsigned num_matches;
 454        unsigned alloc;
 455        struct match_attr **attrs;
 456} *attr_stack;
 457
 458static void free_attr_elem(struct attr_stack *e)
 459{
 460        int i;
 461        free(e->origin);
 462        for (i = 0; i < e->num_matches; i++) {
 463                struct match_attr *a = e->attrs[i];
 464                int j;
 465                for (j = 0; j < a->num_attr; j++) {
 466                        const char *setto = a->state[j].setto;
 467                        if (setto == ATTR__TRUE ||
 468                            setto == ATTR__FALSE ||
 469                            setto == ATTR__UNSET ||
 470                            setto == ATTR__UNKNOWN)
 471                                ;
 472                        else
 473                                free((char *) setto);
 474                }
 475                free(a);
 476        }
 477        free(e->attrs);
 478        free(e);
 479}
 480
 481struct attr_check *attr_check_alloc(void)
 482{
 483        return xcalloc(1, sizeof(struct attr_check));
 484}
 485
 486struct attr_check *attr_check_initl(const char *one, ...)
 487{
 488        struct attr_check *check;
 489        int cnt;
 490        va_list params;
 491        const char *param;
 492
 493        va_start(params, one);
 494        for (cnt = 1; (param = va_arg(params, const char *)) != NULL; cnt++)
 495                ;
 496        va_end(params);
 497
 498        check = attr_check_alloc();
 499        check->nr = cnt;
 500        check->alloc = cnt;
 501        check->items = xcalloc(cnt, sizeof(struct attr_check_item));
 502
 503        check->items[0].attr = git_attr(one);
 504        va_start(params, one);
 505        for (cnt = 1; cnt < check->nr; cnt++) {
 506                const struct git_attr *attr;
 507                param = va_arg(params, const char *);
 508                if (!param)
 509                        die("BUG: counted %d != ended at %d",
 510                            check->nr, cnt);
 511                attr = git_attr(param);
 512                if (!attr)
 513                        die("BUG: %s: not a valid attribute name", param);
 514                check->items[cnt].attr = attr;
 515        }
 516        va_end(params);
 517        return check;
 518}
 519
 520struct attr_check_item *attr_check_append(struct attr_check *check,
 521                                          const struct git_attr *attr)
 522{
 523        struct attr_check_item *item;
 524
 525        ALLOC_GROW(check->items, check->nr + 1, check->alloc);
 526        item = &check->items[check->nr++];
 527        item->attr = attr;
 528        return item;
 529}
 530
 531void attr_check_reset(struct attr_check *check)
 532{
 533        check->nr = 0;
 534}
 535
 536void attr_check_clear(struct attr_check *check)
 537{
 538        free(check->items);
 539        check->items = NULL;
 540        check->alloc = 0;
 541        check->nr = 0;
 542
 543        free(check->all_attrs);
 544        check->all_attrs = NULL;
 545        check->all_attrs_nr = 0;
 546}
 547
 548void attr_check_free(struct attr_check *check)
 549{
 550        attr_check_clear(check);
 551        free(check);
 552}
 553
 554static const char *builtin_attr[] = {
 555        "[attr]binary -diff -merge -text",
 556        NULL,
 557};
 558
 559static void handle_attr_line(struct attr_stack *res,
 560                             const char *line,
 561                             const char *src,
 562                             int lineno,
 563                             int macro_ok)
 564{
 565        struct match_attr *a;
 566
 567        a = parse_attr_line(line, src, lineno, macro_ok);
 568        if (!a)
 569                return;
 570        ALLOC_GROW(res->attrs, res->num_matches + 1, res->alloc);
 571        res->attrs[res->num_matches++] = a;
 572}
 573
 574static struct attr_stack *read_attr_from_array(const char **list)
 575{
 576        struct attr_stack *res;
 577        const char *line;
 578        int lineno = 0;
 579
 580        res = xcalloc(1, sizeof(*res));
 581        while ((line = *(list++)) != NULL)
 582                handle_attr_line(res, line, "[builtin]", ++lineno, 1);
 583        return res;
 584}
 585
 586/*
 587 * NEEDSWORK: these two are tricky.  The callers assume there is a
 588 * single, system-wide global state "where we read attributes from?"
 589 * and when the state is flipped by calling git_attr_set_direction(),
 590 * attr_stack is discarded so that subsequent attr_check will lazily
 591 * read from the right place.  And they do not know or care who called
 592 * by them uses the attribute subsystem, hence have no knowledge of
 593 * existing git_attr_check instances or future ones that will be
 594 * created).
 595 *
 596 * Probably we need a thread_local that holds these two variables,
 597 * and a list of git_attr_check instances (which need to be maintained
 598 * by hooking into git_attr_check_alloc(), git_attr_check_initl(), and
 599 * git_attr_check_clear().  Then git_attr_set_direction() updates the
 600 * fields in that thread_local for these two variables, iterate over
 601 * all the active git_attr_check instances and discard the attr_stack
 602 * they hold.  Yuck, but it sounds doable.
 603 */
 604static enum git_attr_direction direction;
 605static struct index_state *use_index;
 606
 607static struct attr_stack *read_attr_from_file(const char *path, int macro_ok)
 608{
 609        FILE *fp = fopen(path, "r");
 610        struct attr_stack *res;
 611        char buf[2048];
 612        int lineno = 0;
 613
 614        if (!fp) {
 615                if (errno != ENOENT && errno != ENOTDIR)
 616                        warn_on_inaccessible(path);
 617                return NULL;
 618        }
 619        res = xcalloc(1, sizeof(*res));
 620        while (fgets(buf, sizeof(buf), fp)) {
 621                char *bufp = buf;
 622                if (!lineno)
 623                        skip_utf8_bom(&bufp, strlen(bufp));
 624                handle_attr_line(res, bufp, path, ++lineno, macro_ok);
 625        }
 626        fclose(fp);
 627        return res;
 628}
 629
 630static struct attr_stack *read_attr_from_index(const char *path, int macro_ok)
 631{
 632        struct attr_stack *res;
 633        char *buf, *sp;
 634        int lineno = 0;
 635
 636        buf = read_blob_data_from_index(use_index ? use_index : &the_index, path, NULL);
 637        if (!buf)
 638                return NULL;
 639
 640        res = xcalloc(1, sizeof(*res));
 641        for (sp = buf; *sp; ) {
 642                char *ep;
 643                int more;
 644
 645                ep = strchrnul(sp, '\n');
 646                more = (*ep == '\n');
 647                *ep = '\0';
 648                handle_attr_line(res, sp, path, ++lineno, macro_ok);
 649                sp = ep + more;
 650        }
 651        free(buf);
 652        return res;
 653}
 654
 655static struct attr_stack *read_attr(const char *path, int macro_ok)
 656{
 657        struct attr_stack *res;
 658
 659        if (direction == GIT_ATTR_CHECKOUT) {
 660                res = read_attr_from_index(path, macro_ok);
 661                if (!res)
 662                        res = read_attr_from_file(path, macro_ok);
 663        }
 664        else if (direction == GIT_ATTR_CHECKIN) {
 665                res = read_attr_from_file(path, macro_ok);
 666                if (!res)
 667                        /*
 668                         * There is no checked out .gitattributes file there, but
 669                         * we might have it in the index.  We allow operation in a
 670                         * sparsely checked out work tree, so read from it.
 671                         */
 672                        res = read_attr_from_index(path, macro_ok);
 673        }
 674        else
 675                res = read_attr_from_index(path, macro_ok);
 676        if (!res)
 677                res = xcalloc(1, sizeof(*res));
 678        return res;
 679}
 680
 681#if DEBUG_ATTR
 682static void debug_info(const char *what, struct attr_stack *elem)
 683{
 684        fprintf(stderr, "%s: %s\n", what, elem->origin ? elem->origin : "()");
 685}
 686static void debug_set(const char *what, const char *match, struct git_attr *attr, const void *v)
 687{
 688        const char *value = v;
 689
 690        if (ATTR_TRUE(value))
 691                value = "set";
 692        else if (ATTR_FALSE(value))
 693                value = "unset";
 694        else if (ATTR_UNSET(value))
 695                value = "unspecified";
 696
 697        fprintf(stderr, "%s: %s => %s (%s)\n",
 698                what, attr->name, (char *) value, match);
 699}
 700#define debug_push(a) debug_info("push", (a))
 701#define debug_pop(a) debug_info("pop", (a))
 702#else
 703#define debug_push(a) do { ; } while (0)
 704#define debug_pop(a) do { ; } while (0)
 705#define debug_set(a,b,c,d) do { ; } while (0)
 706#endif /* DEBUG_ATTR */
 707
 708static void drop_attr_stack(void)
 709{
 710        while (attr_stack) {
 711                struct attr_stack *elem = attr_stack;
 712                attr_stack = elem->prev;
 713                free_attr_elem(elem);
 714        }
 715}
 716
 717static const char *git_etc_gitattributes(void)
 718{
 719        static const char *system_wide;
 720        if (!system_wide)
 721                system_wide = system_path(ETC_GITATTRIBUTES);
 722        return system_wide;
 723}
 724
 725static int git_attr_system(void)
 726{
 727        return !git_env_bool("GIT_ATTR_NOSYSTEM", 0);
 728}
 729
 730static GIT_PATH_FUNC(git_path_info_attributes, INFOATTRIBUTES_FILE)
 731
 732static void push_stack(struct attr_stack **attr_stack_p,
 733                       struct attr_stack *elem, char *origin, size_t originlen)
 734{
 735        if (elem) {
 736                elem->origin = origin;
 737                if (origin)
 738                        elem->originlen = originlen;
 739                elem->prev = *attr_stack_p;
 740                *attr_stack_p = elem;
 741        }
 742}
 743
 744static void bootstrap_attr_stack(void)
 745{
 746        struct attr_stack *elem;
 747
 748        if (attr_stack)
 749                return;
 750
 751        push_stack(&attr_stack, read_attr_from_array(builtin_attr), NULL, 0);
 752
 753        if (git_attr_system())
 754                push_stack(&attr_stack,
 755                           read_attr_from_file(git_etc_gitattributes(), 1),
 756                           NULL, 0);
 757
 758        if (!git_attributes_file)
 759                git_attributes_file = xdg_config_home("attributes");
 760        if (git_attributes_file)
 761                push_stack(&attr_stack,
 762                           read_attr_from_file(git_attributes_file, 1),
 763                           NULL, 0);
 764
 765        if (!is_bare_repository() || direction == GIT_ATTR_INDEX) {
 766                elem = read_attr(GITATTRIBUTES_FILE, 1);
 767                push_stack(&attr_stack, elem, xstrdup(""), 0);
 768                debug_push(elem);
 769        }
 770
 771        if (startup_info->have_repository)
 772                elem = read_attr_from_file(git_path_info_attributes(), 1);
 773        else
 774                elem = NULL;
 775
 776        if (!elem)
 777                elem = xcalloc(1, sizeof(*elem));
 778        push_stack(&attr_stack, elem, NULL, 0);
 779}
 780
 781static void prepare_attr_stack(const char *path, int dirlen)
 782{
 783        struct attr_stack *elem, *info;
 784        const char *cp;
 785
 786        /*
 787         * At the bottom of the attribute stack is the built-in
 788         * set of attribute definitions, followed by the contents
 789         * of $(prefix)/etc/gitattributes and a file specified by
 790         * core.attributesfile.  Then, contents from
 791         * .gitattribute files from directories closer to the
 792         * root to the ones in deeper directories are pushed
 793         * to the stack.  Finally, at the very top of the stack
 794         * we always keep the contents of $GIT_DIR/info/attributes.
 795         *
 796         * When checking, we use entries from near the top of the
 797         * stack, preferring $GIT_DIR/info/attributes, then
 798         * .gitattributes in deeper directories to shallower ones,
 799         * and finally use the built-in set as the default.
 800         */
 801        bootstrap_attr_stack();
 802
 803        /*
 804         * Pop the "info" one that is always at the top of the stack.
 805         */
 806        info = attr_stack;
 807        attr_stack = info->prev;
 808
 809        /*
 810         * Pop the ones from directories that are not the prefix of
 811         * the path we are checking. Break out of the loop when we see
 812         * the root one (whose origin is an empty string "") or the builtin
 813         * one (whose origin is NULL) without popping it.
 814         */
 815        while (attr_stack->origin) {
 816                int namelen = strlen(attr_stack->origin);
 817
 818                elem = attr_stack;
 819                if (namelen <= dirlen &&
 820                    !strncmp(elem->origin, path, namelen) &&
 821                    (!namelen || path[namelen] == '/'))
 822                        break;
 823
 824                debug_pop(elem);
 825                attr_stack = elem->prev;
 826                free_attr_elem(elem);
 827        }
 828
 829        /*
 830         * Read from parent directories and push them down
 831         */
 832        if (!is_bare_repository() || direction == GIT_ATTR_INDEX) {
 833                /*
 834                 * bootstrap_attr_stack() should have added, and the
 835                 * above loop should have stopped before popping, the
 836                 * root element whose attr_stack->origin is set to an
 837                 * empty string.
 838                 */
 839                struct strbuf pathbuf = STRBUF_INIT;
 840
 841                assert(attr_stack->origin);
 842                while (1) {
 843                        size_t len = strlen(attr_stack->origin);
 844                        char *origin;
 845
 846                        if (dirlen <= len)
 847                                break;
 848                        cp = memchr(path + len + 1, '/', dirlen - len - 1);
 849                        if (!cp)
 850                                cp = path + dirlen;
 851                        strbuf_addf(&pathbuf,
 852                                    "%.*s/%s", (int)(cp - path), path,
 853                                    GITATTRIBUTES_FILE);
 854                        elem = read_attr(pathbuf.buf, 0);
 855                        strbuf_setlen(&pathbuf, cp - path);
 856                        origin = strbuf_detach(&pathbuf, &len);
 857                        push_stack(&attr_stack, elem, origin, len);
 858                        debug_push(elem);
 859                }
 860
 861                strbuf_release(&pathbuf);
 862        }
 863
 864        /*
 865         * Finally push the "info" one at the top of the stack.
 866         */
 867        push_stack(&attr_stack, info, NULL, 0);
 868}
 869
 870static int path_matches(const char *pathname, int pathlen,
 871                        int basename_offset,
 872                        const struct pattern *pat,
 873                        const char *base, int baselen)
 874{
 875        const char *pattern = pat->pattern;
 876        int prefix = pat->nowildcardlen;
 877        int isdir = (pathlen && pathname[pathlen - 1] == '/');
 878
 879        if ((pat->flags & EXC_FLAG_MUSTBEDIR) && !isdir)
 880                return 0;
 881
 882        if (pat->flags & EXC_FLAG_NODIR) {
 883                return match_basename(pathname + basename_offset,
 884                                      pathlen - basename_offset - isdir,
 885                                      pattern, prefix,
 886                                      pat->patternlen, pat->flags);
 887        }
 888        return match_pathname(pathname, pathlen - isdir,
 889                              base, baselen,
 890                              pattern, prefix, pat->patternlen, pat->flags);
 891}
 892
 893static int macroexpand_one(struct all_attrs_item *all_attrs, int nr, int rem);
 894
 895static int fill_one(const char *what, struct all_attrs_item *all_attrs,
 896                    const struct match_attr *a, int rem)
 897{
 898        int i;
 899
 900        for (i = a->num_attr - 1; rem > 0 && i >= 0; i--) {
 901                struct git_attr *attr = a->state[i].attr;
 902                const char **n = &(all_attrs[attr->attr_nr].value);
 903                const char *v = a->state[i].setto;
 904
 905                if (*n == ATTR__UNKNOWN) {
 906                        debug_set(what,
 907                                  a->is_macro ? a->u.attr->name : a->u.pat.pattern,
 908                                  attr, v);
 909                        *n = v;
 910                        rem--;
 911                        rem = macroexpand_one(all_attrs, attr->attr_nr, rem);
 912                }
 913        }
 914        return rem;
 915}
 916
 917static int fill(const char *path, int pathlen, int basename_offset,
 918                struct attr_stack *stk, struct all_attrs_item *all_attrs,
 919                int rem)
 920{
 921        int i;
 922        const char *base = stk->origin ? stk->origin : "";
 923
 924        for (i = stk->num_matches - 1; 0 < rem && 0 <= i; i--) {
 925                struct match_attr *a = stk->attrs[i];
 926                if (a->is_macro)
 927                        continue;
 928                if (path_matches(path, pathlen, basename_offset,
 929                                 &a->u.pat, base, stk->originlen))
 930                        rem = fill_one("fill", all_attrs, a, rem);
 931        }
 932        return rem;
 933}
 934
 935static int macroexpand_one(struct all_attrs_item *all_attrs, int nr, int rem)
 936{
 937        const struct all_attrs_item *item = &all_attrs[nr];
 938
 939        if (item->macro && item->value == ATTR__TRUE)
 940                return fill_one("expand", all_attrs, item->macro, rem);
 941        else
 942                return rem;
 943}
 944
 945/*
 946 * Marks the attributes which are macros based on the attribute stack.
 947 * This prevents having to search through the attribute stack each time
 948 * a macro needs to be expanded during the fill stage.
 949 */
 950static void determine_macros(struct all_attrs_item *all_attrs,
 951                             const struct attr_stack *stack)
 952{
 953        for (; stack; stack = stack->prev) {
 954                int i;
 955                for (i = stack->num_matches - 1; i >= 0; i--) {
 956                        const struct match_attr *ma = stack->attrs[i];
 957                        if (ma->is_macro) {
 958                                int n = ma->u.attr->attr_nr;
 959                                if (!all_attrs[n].macro) {
 960                                        all_attrs[n].macro = ma;
 961                                }
 962                        }
 963                }
 964        }
 965}
 966
 967/*
 968 * Collect attributes for path into the array pointed to by check->all_attrs.
 969 * If check->check_nr is non-zero, only attributes in check[] are collected.
 970 * Otherwise all attributes are collected.
 971 */
 972static void collect_some_attrs(const char *path, struct attr_check *check)
 973{
 974        struct attr_stack *stk;
 975        int i, pathlen, rem, dirlen;
 976        const char *cp, *last_slash = NULL;
 977        int basename_offset;
 978
 979        for (cp = path; *cp; cp++) {
 980                if (*cp == '/' && cp[1])
 981                        last_slash = cp;
 982        }
 983        pathlen = cp - path;
 984        if (last_slash) {
 985                basename_offset = last_slash + 1 - path;
 986                dirlen = last_slash - path;
 987        } else {
 988                basename_offset = 0;
 989                dirlen = 0;
 990        }
 991
 992        prepare_attr_stack(path, dirlen);
 993        all_attrs_init(&g_attr_hashmap, check);
 994        determine_macros(check->all_attrs, attr_stack);
 995
 996        if (check->nr) {
 997                rem = 0;
 998                for (i = 0; i < check->nr; i++) {
 999                        int n = check->items[i].attr->attr_nr;
1000                        struct all_attrs_item *item = &check->all_attrs[n];
1001                        if (item->macro) {
1002                                item->value = ATTR__UNSET;
1003                                rem++;
1004                        }
1005                }
1006                if (rem == check->nr)
1007                        return;
1008        }
1009
1010        rem = check->all_attrs_nr;
1011        for (stk = attr_stack; 0 < rem && stk; stk = stk->prev)
1012                rem = fill(path, pathlen, basename_offset, stk, check->all_attrs, rem);
1013}
1014
1015int git_check_attr(const char *path, struct attr_check *check)
1016{
1017        int i;
1018
1019        collect_some_attrs(path, check);
1020
1021        for (i = 0; i < check->nr; i++) {
1022                size_t n = check->items[i].attr->attr_nr;
1023                const char *value = check->all_attrs[n].value;
1024                if (value == ATTR__UNKNOWN)
1025                        value = ATTR__UNSET;
1026                check->items[i].value = value;
1027        }
1028
1029        return 0;
1030}
1031
1032void git_all_attrs(const char *path, struct attr_check *check)
1033{
1034        int i;
1035
1036        attr_check_reset(check);
1037        collect_some_attrs(path, check);
1038
1039        for (i = 0; i < check->all_attrs_nr; i++) {
1040                const char *name = check->all_attrs[i].attr->name;
1041                const char *value = check->all_attrs[i].value;
1042                struct attr_check_item *item;
1043                if (value == ATTR__UNSET || value == ATTR__UNKNOWN)
1044                        continue;
1045                item = attr_check_append(check, git_attr(name));
1046                item->value = value;
1047        }
1048}
1049
1050void git_attr_set_direction(enum git_attr_direction new, struct index_state *istate)
1051{
1052        enum git_attr_direction old = direction;
1053
1054        if (is_bare_repository() && new != GIT_ATTR_INDEX)
1055                die("BUG: non-INDEX attr direction in a bare repo");
1056
1057        direction = new;
1058        if (new != old)
1059                drop_attr_stack();
1060        use_index = istate;
1061}
1062
1063void attr_start(void)
1064{
1065#ifndef NO_PTHREADS
1066        pthread_mutex_init(&g_attr_hashmap.mutex, NULL);
1067#endif
1068}