Make git_check_attr() a void function
[git/git.git] / attr.c
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 "config.h"
13 #include "exec-cmd.h"
14 #include "attr.h"
15 #include "dir.h"
16 #include "utf8.h"
17 #include "quote.h"
18 #include "thread-utils.h"
19
20 const char git_attr__true[] = "(builtin)true";
21 const char git_attr__false[] = "\0(builtin)false";
22 static const char git_attr__unknown[] = "(builtin)unknown";
23 #define ATTR__TRUE git_attr__true
24 #define ATTR__FALSE git_attr__false
25 #define ATTR__UNSET NULL
26 #define ATTR__UNKNOWN git_attr__unknown
27
28 #ifndef DEBUG_ATTR
29 #define DEBUG_ATTR 0
30 #endif
31
32 struct git_attr {
33 int attr_nr; /* unique attribute number */
34 char name[FLEX_ARRAY]; /* attribute name */
35 };
36
37 const char *git_attr_name(const struct git_attr *attr)
38 {
39 return attr->name;
40 }
41
42 struct attr_hashmap {
43 struct hashmap map;
44 #ifndef NO_PTHREADS
45 pthread_mutex_t mutex;
46 #endif
47 };
48
49 static inline void hashmap_lock(struct attr_hashmap *map)
50 {
51 #ifndef NO_PTHREADS
52 pthread_mutex_lock(&map->mutex);
53 #endif
54 }
55
56 static inline void hashmap_unlock(struct attr_hashmap *map)
57 {
58 #ifndef NO_PTHREADS
59 pthread_mutex_unlock(&map->mutex);
60 #endif
61 }
62
63 /*
64 * The global dictionary of all interned attributes. This
65 * is a singleton object which is shared between threads.
66 * Access to this dictionary must be surrounded with a mutex.
67 */
68 static struct attr_hashmap g_attr_hashmap;
69
70 /* The container for objects stored in "struct attr_hashmap" */
71 struct attr_hash_entry {
72 struct hashmap_entry ent; /* must be the first member! */
73 const char *key; /* the key; memory should be owned by value */
74 size_t keylen; /* length of the key */
75 void *value; /* the stored value */
76 };
77
78 /* attr_hashmap comparison function */
79 static int attr_hash_entry_cmp(const void *unused_cmp_data,
80 const void *entry,
81 const void *entry_or_key,
82 const void *unused_keydata)
83 {
84 const struct attr_hash_entry *a = entry;
85 const struct attr_hash_entry *b = entry_or_key;
86 return (a->keylen != b->keylen) || strncmp(a->key, b->key, a->keylen);
87 }
88
89 /* Initialize an 'attr_hashmap' object */
90 static void attr_hashmap_init(struct attr_hashmap *map)
91 {
92 hashmap_init(&map->map, attr_hash_entry_cmp, NULL, 0);
93 }
94
95 /*
96 * Retrieve the 'value' stored in a hashmap given the provided 'key'.
97 * If there is no matching entry, return NULL.
98 */
99 static void *attr_hashmap_get(struct attr_hashmap *map,
100 const char *key, size_t keylen)
101 {
102 struct attr_hash_entry k;
103 struct attr_hash_entry *e;
104
105 if (!map->map.tablesize)
106 attr_hashmap_init(map);
107
108 hashmap_entry_init(&k, memhash(key, keylen));
109 k.key = key;
110 k.keylen = keylen;
111 e = hashmap_get(&map->map, &k, NULL);
112
113 return e ? e->value : NULL;
114 }
115
116 /* Add 'value' to a hashmap based on the provided 'key'. */
117 static void attr_hashmap_add(struct attr_hashmap *map,
118 const char *key, size_t keylen,
119 void *value)
120 {
121 struct attr_hash_entry *e;
122
123 if (!map->map.tablesize)
124 attr_hashmap_init(map);
125
126 e = xmalloc(sizeof(struct attr_hash_entry));
127 hashmap_entry_init(e, memhash(key, keylen));
128 e->key = key;
129 e->keylen = keylen;
130 e->value = value;
131
132 hashmap_add(&map->map, e);
133 }
134
135 struct all_attrs_item {
136 const struct git_attr *attr;
137 const char *value;
138 /*
139 * If 'macro' is non-NULL, indicates that 'attr' is a macro based on
140 * the current attribute stack and contains a pointer to the match_attr
141 * definition of the macro
142 */
143 const struct match_attr *macro;
144 };
145
146 /*
147 * Reallocate and reinitialize the array of all attributes (which is used in
148 * the attribute collection process) in 'check' based on the global dictionary
149 * of attributes.
150 */
151 static void all_attrs_init(struct attr_hashmap *map, struct attr_check *check)
152 {
153 int i;
154 unsigned int size;
155
156 hashmap_lock(map);
157
158 size = hashmap_get_size(&map->map);
159 if (size < check->all_attrs_nr)
160 BUG("interned attributes shouldn't be deleted");
161
162 /*
163 * If the number of attributes in the global dictionary has increased
164 * (or this attr_check instance doesn't have an initialized all_attrs
165 * field), reallocate the provided attr_check instance's all_attrs
166 * field and fill each entry with its corresponding git_attr.
167 */
168 if (size != check->all_attrs_nr) {
169 struct attr_hash_entry *e;
170 struct hashmap_iter iter;
171 hashmap_iter_init(&map->map, &iter);
172
173 REALLOC_ARRAY(check->all_attrs, size);
174 check->all_attrs_nr = size;
175
176 while ((e = hashmap_iter_next(&iter))) {
177 const struct git_attr *a = e->value;
178 check->all_attrs[a->attr_nr].attr = a;
179 }
180 }
181
182 hashmap_unlock(map);
183
184 /*
185 * Re-initialize every entry in check->all_attrs.
186 * This re-initialization can live outside of the locked region since
187 * the attribute dictionary is no longer being accessed.
188 */
189 for (i = 0; i < check->all_attrs_nr; i++) {
190 check->all_attrs[i].value = ATTR__UNKNOWN;
191 check->all_attrs[i].macro = NULL;
192 }
193 }
194
195 static int attr_name_valid(const char *name, size_t namelen)
196 {
197 /*
198 * Attribute name cannot begin with '-' and must consist of
199 * characters from [-A-Za-z0-9_.].
200 */
201 if (namelen <= 0 || *name == '-')
202 return 0;
203 while (namelen--) {
204 char ch = *name++;
205 if (! (ch == '-' || ch == '.' || ch == '_' ||
206 ('0' <= ch && ch <= '9') ||
207 ('a' <= ch && ch <= 'z') ||
208 ('A' <= ch && ch <= 'Z')) )
209 return 0;
210 }
211 return 1;
212 }
213
214 static void report_invalid_attr(const char *name, size_t len,
215 const char *src, int lineno)
216 {
217 struct strbuf err = STRBUF_INIT;
218 strbuf_addf(&err, _("%.*s is not a valid attribute name"),
219 (int) len, name);
220 fprintf(stderr, "%s: %s:%d\n", err.buf, src, lineno);
221 strbuf_release(&err);
222 }
223
224 /*
225 * Given a 'name', lookup and return the corresponding attribute in the global
226 * dictionary. If no entry is found, create a new attribute and store it in
227 * the dictionary.
228 */
229 static const struct git_attr *git_attr_internal(const char *name, int namelen)
230 {
231 struct git_attr *a;
232
233 if (!attr_name_valid(name, namelen))
234 return NULL;
235
236 hashmap_lock(&g_attr_hashmap);
237
238 a = attr_hashmap_get(&g_attr_hashmap, name, namelen);
239
240 if (!a) {
241 FLEX_ALLOC_MEM(a, name, name, namelen);
242 a->attr_nr = hashmap_get_size(&g_attr_hashmap.map);
243
244 attr_hashmap_add(&g_attr_hashmap, a->name, namelen, a);
245 assert(a->attr_nr ==
246 (hashmap_get_size(&g_attr_hashmap.map) - 1));
247 }
248
249 hashmap_unlock(&g_attr_hashmap);
250
251 return a;
252 }
253
254 const struct git_attr *git_attr(const char *name)
255 {
256 return git_attr_internal(name, strlen(name));
257 }
258
259 /* What does a matched pattern decide? */
260 struct attr_state {
261 const struct git_attr *attr;
262 const char *setto;
263 };
264
265 struct pattern {
266 const char *pattern;
267 int patternlen;
268 int nowildcardlen;
269 unsigned flags; /* EXC_FLAG_* */
270 };
271
272 /*
273 * One rule, as from a .gitattributes file.
274 *
275 * If is_macro is true, then u.attr is a pointer to the git_attr being
276 * defined.
277 *
278 * If is_macro is false, then u.pat is the filename pattern to which the
279 * rule applies.
280 *
281 * In either case, num_attr is the number of attributes affected by
282 * this rule, and state is an array listing them. The attributes are
283 * listed as they appear in the file (macros unexpanded).
284 */
285 struct match_attr {
286 union {
287 struct pattern pat;
288 const struct git_attr *attr;
289 } u;
290 char is_macro;
291 unsigned num_attr;
292 struct attr_state state[FLEX_ARRAY];
293 };
294
295 static const char blank[] = " \t\r\n";
296
297 /*
298 * Parse a whitespace-delimited attribute state (i.e., "attr",
299 * "-attr", "!attr", or "attr=value") from the string starting at src.
300 * If e is not NULL, write the results to *e. Return a pointer to the
301 * remainder of the string (with leading whitespace removed), or NULL
302 * if there was an error.
303 */
304 static const char *parse_attr(const char *src, int lineno, const char *cp,
305 struct attr_state *e)
306 {
307 const char *ep, *equals;
308 int len;
309
310 ep = cp + strcspn(cp, blank);
311 equals = strchr(cp, '=');
312 if (equals && ep < equals)
313 equals = NULL;
314 if (equals)
315 len = equals - cp;
316 else
317 len = ep - cp;
318 if (!e) {
319 if (*cp == '-' || *cp == '!') {
320 cp++;
321 len--;
322 }
323 if (!attr_name_valid(cp, len)) {
324 report_invalid_attr(cp, len, src, lineno);
325 return NULL;
326 }
327 } else {
328 /*
329 * As this function is always called twice, once with
330 * e == NULL in the first pass and then e != NULL in
331 * the second pass, no need for attr_name_valid()
332 * check here.
333 */
334 if (*cp == '-' || *cp == '!') {
335 e->setto = (*cp == '-') ? ATTR__FALSE : ATTR__UNSET;
336 cp++;
337 len--;
338 }
339 else if (!equals)
340 e->setto = ATTR__TRUE;
341 else {
342 e->setto = xmemdupz(equals + 1, ep - equals - 1);
343 }
344 e->attr = git_attr_internal(cp, len);
345 }
346 return ep + strspn(ep, blank);
347 }
348
349 static struct match_attr *parse_attr_line(const char *line, const char *src,
350 int lineno, int macro_ok)
351 {
352 int namelen;
353 int num_attr, i;
354 const char *cp, *name, *states;
355 struct match_attr *res = NULL;
356 int is_macro;
357 struct strbuf pattern = STRBUF_INIT;
358
359 cp = line + strspn(line, blank);
360 if (!*cp || *cp == '#')
361 return NULL;
362 name = cp;
363
364 if (*cp == '"' && !unquote_c_style(&pattern, name, &states)) {
365 name = pattern.buf;
366 namelen = pattern.len;
367 } else {
368 namelen = strcspn(name, blank);
369 states = name + namelen;
370 }
371
372 if (strlen(ATTRIBUTE_MACRO_PREFIX) < namelen &&
373 starts_with(name, ATTRIBUTE_MACRO_PREFIX)) {
374 if (!macro_ok) {
375 fprintf(stderr, "%s not allowed: %s:%d\n",
376 name, src, lineno);
377 goto fail_return;
378 }
379 is_macro = 1;
380 name += strlen(ATTRIBUTE_MACRO_PREFIX);
381 name += strspn(name, blank);
382 namelen = strcspn(name, blank);
383 if (!attr_name_valid(name, namelen)) {
384 report_invalid_attr(name, namelen, src, lineno);
385 goto fail_return;
386 }
387 }
388 else
389 is_macro = 0;
390
391 states += strspn(states, blank);
392
393 /* First pass to count the attr_states */
394 for (cp = states, num_attr = 0; *cp; num_attr++) {
395 cp = parse_attr(src, lineno, cp, NULL);
396 if (!cp)
397 goto fail_return;
398 }
399
400 res = xcalloc(1,
401 sizeof(*res) +
402 sizeof(struct attr_state) * num_attr +
403 (is_macro ? 0 : namelen + 1));
404 if (is_macro) {
405 res->u.attr = git_attr_internal(name, namelen);
406 } else {
407 char *p = (char *)&(res->state[num_attr]);
408 memcpy(p, name, namelen);
409 res->u.pat.pattern = p;
410 parse_exclude_pattern(&res->u.pat.pattern,
411 &res->u.pat.patternlen,
412 &res->u.pat.flags,
413 &res->u.pat.nowildcardlen);
414 if (res->u.pat.flags & EXC_FLAG_NEGATIVE) {
415 warning(_("Negative patterns are ignored in git attributes\n"
416 "Use '\\!' for literal leading exclamation."));
417 goto fail_return;
418 }
419 }
420 res->is_macro = is_macro;
421 res->num_attr = num_attr;
422
423 /* Second pass to fill the attr_states */
424 for (cp = states, i = 0; *cp; i++) {
425 cp = parse_attr(src, lineno, cp, &(res->state[i]));
426 }
427
428 strbuf_release(&pattern);
429 return res;
430
431 fail_return:
432 strbuf_release(&pattern);
433 free(res);
434 return NULL;
435 }
436
437 /*
438 * Like info/exclude and .gitignore, the attribute information can
439 * come from many places.
440 *
441 * (1) .gitattribute file of the same directory;
442 * (2) .gitattribute file of the parent directory if (1) does not have
443 * any match; this goes recursively upwards, just like .gitignore.
444 * (3) $GIT_DIR/info/attributes, which overrides both of the above.
445 *
446 * In the same file, later entries override the earlier match, so in the
447 * global list, we would have entries from info/attributes the earliest
448 * (reading the file from top to bottom), .gitattribute of the root
449 * directory (again, reading the file from top to bottom) down to the
450 * current directory, and then scan the list backwards to find the first match.
451 * This is exactly the same as what is_excluded() does in dir.c to deal with
452 * .gitignore file and info/excludes file as a fallback.
453 */
454
455 struct attr_stack {
456 struct attr_stack *prev;
457 char *origin;
458 size_t originlen;
459 unsigned num_matches;
460 unsigned alloc;
461 struct match_attr **attrs;
462 };
463
464 static void attr_stack_free(struct attr_stack *e)
465 {
466 int i;
467 free(e->origin);
468 for (i = 0; i < e->num_matches; i++) {
469 struct match_attr *a = e->attrs[i];
470 int j;
471 for (j = 0; j < a->num_attr; j++) {
472 const char *setto = a->state[j].setto;
473 if (setto == ATTR__TRUE ||
474 setto == ATTR__FALSE ||
475 setto == ATTR__UNSET ||
476 setto == ATTR__UNKNOWN)
477 ;
478 else
479 free((char *) setto);
480 }
481 free(a);
482 }
483 free(e->attrs);
484 free(e);
485 }
486
487 static void drop_attr_stack(struct attr_stack **stack)
488 {
489 while (*stack) {
490 struct attr_stack *elem = *stack;
491 *stack = elem->prev;
492 attr_stack_free(elem);
493 }
494 }
495
496 /* List of all attr_check structs; access should be surrounded by mutex */
497 static struct check_vector {
498 size_t nr;
499 size_t alloc;
500 struct attr_check **checks;
501 #ifndef NO_PTHREADS
502 pthread_mutex_t mutex;
503 #endif
504 } check_vector;
505
506 static inline void vector_lock(void)
507 {
508 #ifndef NO_PTHREADS
509 pthread_mutex_lock(&check_vector.mutex);
510 #endif
511 }
512
513 static inline void vector_unlock(void)
514 {
515 #ifndef NO_PTHREADS
516 pthread_mutex_unlock(&check_vector.mutex);
517 #endif
518 }
519
520 static void check_vector_add(struct attr_check *c)
521 {
522 vector_lock();
523
524 ALLOC_GROW(check_vector.checks,
525 check_vector.nr + 1,
526 check_vector.alloc);
527 check_vector.checks[check_vector.nr++] = c;
528
529 vector_unlock();
530 }
531
532 static void check_vector_remove(struct attr_check *check)
533 {
534 int i;
535
536 vector_lock();
537
538 /* Find entry */
539 for (i = 0; i < check_vector.nr; i++)
540 if (check_vector.checks[i] == check)
541 break;
542
543 if (i >= check_vector.nr)
544 BUG("no entry found");
545
546 /* shift entries over */
547 for (; i < check_vector.nr - 1; i++)
548 check_vector.checks[i] = check_vector.checks[i + 1];
549
550 check_vector.nr--;
551
552 vector_unlock();
553 }
554
555 /* Iterate through all attr_check instances and drop their stacks */
556 static void drop_all_attr_stacks(void)
557 {
558 int i;
559
560 vector_lock();
561
562 for (i = 0; i < check_vector.nr; i++) {
563 drop_attr_stack(&check_vector.checks[i]->stack);
564 }
565
566 vector_unlock();
567 }
568
569 struct attr_check *attr_check_alloc(void)
570 {
571 struct attr_check *c = xcalloc(1, sizeof(struct attr_check));
572
573 /* save pointer to the check struct */
574 check_vector_add(c);
575
576 return c;
577 }
578
579 struct attr_check *attr_check_initl(const char *one, ...)
580 {
581 struct attr_check *check;
582 int cnt;
583 va_list params;
584 const char *param;
585
586 va_start(params, one);
587 for (cnt = 1; (param = va_arg(params, const char *)) != NULL; cnt++)
588 ;
589 va_end(params);
590
591 check = attr_check_alloc();
592 check->nr = cnt;
593 check->alloc = cnt;
594 check->items = xcalloc(cnt, sizeof(struct attr_check_item));
595
596 check->items[0].attr = git_attr(one);
597 va_start(params, one);
598 for (cnt = 1; cnt < check->nr; cnt++) {
599 const struct git_attr *attr;
600 param = va_arg(params, const char *);
601 if (!param)
602 BUG("counted %d != ended at %d",
603 check->nr, cnt);
604 attr = git_attr(param);
605 if (!attr)
606 BUG("%s: not a valid attribute name", param);
607 check->items[cnt].attr = attr;
608 }
609 va_end(params);
610 return check;
611 }
612
613 struct attr_check *attr_check_dup(const struct attr_check *check)
614 {
615 struct attr_check *ret;
616
617 if (!check)
618 return NULL;
619
620 ret = attr_check_alloc();
621
622 ret->nr = check->nr;
623 ret->alloc = check->alloc;
624 ALLOC_ARRAY(ret->items, ret->nr);
625 COPY_ARRAY(ret->items, check->items, ret->nr);
626
627 return ret;
628 }
629
630 struct attr_check_item *attr_check_append(struct attr_check *check,
631 const struct git_attr *attr)
632 {
633 struct attr_check_item *item;
634
635 ALLOC_GROW(check->items, check->nr + 1, check->alloc);
636 item = &check->items[check->nr++];
637 item->attr = attr;
638 return item;
639 }
640
641 void attr_check_reset(struct attr_check *check)
642 {
643 check->nr = 0;
644 }
645
646 void attr_check_clear(struct attr_check *check)
647 {
648 FREE_AND_NULL(check->items);
649 check->alloc = 0;
650 check->nr = 0;
651
652 FREE_AND_NULL(check->all_attrs);
653 check->all_attrs_nr = 0;
654
655 drop_attr_stack(&check->stack);
656 }
657
658 void attr_check_free(struct attr_check *check)
659 {
660 if (check) {
661 /* Remove check from the check vector */
662 check_vector_remove(check);
663
664 attr_check_clear(check);
665 free(check);
666 }
667 }
668
669 static const char *builtin_attr[] = {
670 "[attr]binary -diff -merge -text",
671 NULL,
672 };
673
674 static void handle_attr_line(struct attr_stack *res,
675 const char *line,
676 const char *src,
677 int lineno,
678 int macro_ok)
679 {
680 struct match_attr *a;
681
682 a = parse_attr_line(line, src, lineno, macro_ok);
683 if (!a)
684 return;
685 ALLOC_GROW(res->attrs, res->num_matches + 1, res->alloc);
686 res->attrs[res->num_matches++] = a;
687 }
688
689 static struct attr_stack *read_attr_from_array(const char **list)
690 {
691 struct attr_stack *res;
692 const char *line;
693 int lineno = 0;
694
695 res = xcalloc(1, sizeof(*res));
696 while ((line = *(list++)) != NULL)
697 handle_attr_line(res, line, "[builtin]", ++lineno, 1);
698 return res;
699 }
700
701 /*
702 * Callers into the attribute system assume there is a single, system-wide
703 * global state where attributes are read from and when the state is flipped by
704 * calling git_attr_set_direction(), the stack frames that have been
705 * constructed need to be discarded so so that subsequent calls into the
706 * attribute system will lazily read from the right place. Since changing
707 * direction causes a global paradigm shift, it should not ever be called while
708 * another thread could potentially be calling into the attribute system.
709 */
710 static enum git_attr_direction direction;
711
712 void git_attr_set_direction(enum git_attr_direction new_direction)
713 {
714 if (is_bare_repository() && new_direction != GIT_ATTR_INDEX)
715 BUG("non-INDEX attr direction in a bare repo");
716
717 if (new_direction != direction)
718 drop_all_attr_stacks();
719
720 direction = new_direction;
721 }
722
723 static struct attr_stack *read_attr_from_file(const char *path, int macro_ok)
724 {
725 FILE *fp = fopen_or_warn(path, "r");
726 struct attr_stack *res;
727 char buf[2048];
728 int lineno = 0;
729
730 if (!fp)
731 return NULL;
732 res = xcalloc(1, sizeof(*res));
733 while (fgets(buf, sizeof(buf), fp)) {
734 char *bufp = buf;
735 if (!lineno)
736 skip_utf8_bom(&bufp, strlen(bufp));
737 handle_attr_line(res, bufp, path, ++lineno, macro_ok);
738 }
739 fclose(fp);
740 return res;
741 }
742
743 static struct attr_stack *read_attr_from_index(const struct index_state *istate,
744 const char *path,
745 int macro_ok)
746 {
747 struct attr_stack *res;
748 char *buf, *sp;
749 int lineno = 0;
750
751 if (!istate)
752 return NULL;
753
754 buf = read_blob_data_from_index(istate, path, NULL);
755 if (!buf)
756 return NULL;
757
758 res = xcalloc(1, sizeof(*res));
759 for (sp = buf; *sp; ) {
760 char *ep;
761 int more;
762
763 ep = strchrnul(sp, '\n');
764 more = (*ep == '\n');
765 *ep = '\0';
766 handle_attr_line(res, sp, path, ++lineno, macro_ok);
767 sp = ep + more;
768 }
769 free(buf);
770 return res;
771 }
772
773 static struct attr_stack *read_attr(const struct index_state *istate,
774 const char *path, int macro_ok)
775 {
776 struct attr_stack *res = NULL;
777
778 if (direction == GIT_ATTR_INDEX) {
779 res = read_attr_from_index(istate, path, macro_ok);
780 } else if (!is_bare_repository()) {
781 if (direction == GIT_ATTR_CHECKOUT) {
782 res = read_attr_from_index(istate, path, macro_ok);
783 if (!res)
784 res = read_attr_from_file(path, macro_ok);
785 } else if (direction == GIT_ATTR_CHECKIN) {
786 res = read_attr_from_file(path, macro_ok);
787 if (!res)
788 /*
789 * There is no checked out .gitattributes file
790 * there, but we might have it in the index.
791 * We allow operation in a sparsely checked out
792 * work tree, so read from it.
793 */
794 res = read_attr_from_index(istate, path, macro_ok);
795 }
796 }
797
798 if (!res)
799 res = xcalloc(1, sizeof(*res));
800 return res;
801 }
802
803 #if DEBUG_ATTR
804 static void debug_info(const char *what, struct attr_stack *elem)
805 {
806 fprintf(stderr, "%s: %s\n", what, elem->origin ? elem->origin : "()");
807 }
808 static void debug_set(const char *what, const char *match, struct git_attr *attr, const void *v)
809 {
810 const char *value = v;
811
812 if (ATTR_TRUE(value))
813 value = "set";
814 else if (ATTR_FALSE(value))
815 value = "unset";
816 else if (ATTR_UNSET(value))
817 value = "unspecified";
818
819 fprintf(stderr, "%s: %s => %s (%s)\n",
820 what, attr->name, (char *) value, match);
821 }
822 #define debug_push(a) debug_info("push", (a))
823 #define debug_pop(a) debug_info("pop", (a))
824 #else
825 #define debug_push(a) do { ; } while (0)
826 #define debug_pop(a) do { ; } while (0)
827 #define debug_set(a,b,c,d) do { ; } while (0)
828 #endif /* DEBUG_ATTR */
829
830 static const char *git_etc_gitattributes(void)
831 {
832 static const char *system_wide;
833 if (!system_wide)
834 system_wide = system_path(ETC_GITATTRIBUTES);
835 return system_wide;
836 }
837
838 static const char *get_home_gitattributes(void)
839 {
840 if (!git_attributes_file)
841 git_attributes_file = xdg_config_home("attributes");
842
843 return git_attributes_file;
844 }
845
846 static int git_attr_system(void)
847 {
848 return !git_env_bool("GIT_ATTR_NOSYSTEM", 0);
849 }
850
851 static GIT_PATH_FUNC(git_path_info_attributes, INFOATTRIBUTES_FILE)
852
853 static void push_stack(struct attr_stack **attr_stack_p,
854 struct attr_stack *elem, char *origin, size_t originlen)
855 {
856 if (elem) {
857 elem->origin = origin;
858 if (origin)
859 elem->originlen = originlen;
860 elem->prev = *attr_stack_p;
861 *attr_stack_p = elem;
862 }
863 }
864
865 static void bootstrap_attr_stack(const struct index_state *istate,
866 struct attr_stack **stack)
867 {
868 struct attr_stack *e;
869
870 if (*stack)
871 return;
872
873 /* builtin frame */
874 e = read_attr_from_array(builtin_attr);
875 push_stack(stack, e, NULL, 0);
876
877 /* system-wide frame */
878 if (git_attr_system()) {
879 e = read_attr_from_file(git_etc_gitattributes(), 1);
880 push_stack(stack, e, NULL, 0);
881 }
882
883 /* home directory */
884 if (get_home_gitattributes()) {
885 e = read_attr_from_file(get_home_gitattributes(), 1);
886 push_stack(stack, e, NULL, 0);
887 }
888
889 /* root directory */
890 e = read_attr(istate, GITATTRIBUTES_FILE, 1);
891 push_stack(stack, e, xstrdup(""), 0);
892
893 /* info frame */
894 if (startup_info->have_repository)
895 e = read_attr_from_file(git_path_info_attributes(), 1);
896 else
897 e = NULL;
898 if (!e)
899 e = xcalloc(1, sizeof(struct attr_stack));
900 push_stack(stack, e, NULL, 0);
901 }
902
903 static void prepare_attr_stack(const struct index_state *istate,
904 const char *path, int dirlen,
905 struct attr_stack **stack)
906 {
907 struct attr_stack *info;
908 struct strbuf pathbuf = STRBUF_INIT;
909
910 /*
911 * At the bottom of the attribute stack is the built-in
912 * set of attribute definitions, followed by the contents
913 * of $(prefix)/etc/gitattributes and a file specified by
914 * core.attributesfile. Then, contents from
915 * .gitattribute files from directories closer to the
916 * root to the ones in deeper directories are pushed
917 * to the stack. Finally, at the very top of the stack
918 * we always keep the contents of $GIT_DIR/info/attributes.
919 *
920 * When checking, we use entries from near the top of the
921 * stack, preferring $GIT_DIR/info/attributes, then
922 * .gitattributes in deeper directories to shallower ones,
923 * and finally use the built-in set as the default.
924 */
925 bootstrap_attr_stack(istate, stack);
926
927 /*
928 * Pop the "info" one that is always at the top of the stack.
929 */
930 info = *stack;
931 *stack = info->prev;
932
933 /*
934 * Pop the ones from directories that are not the prefix of
935 * the path we are checking. Break out of the loop when we see
936 * the root one (whose origin is an empty string "") or the builtin
937 * one (whose origin is NULL) without popping it.
938 */
939 while ((*stack)->origin) {
940 int namelen = (*stack)->originlen;
941 struct attr_stack *elem;
942
943 elem = *stack;
944 if (namelen <= dirlen &&
945 !strncmp(elem->origin, path, namelen) &&
946 (!namelen || path[namelen] == '/'))
947 break;
948
949 debug_pop(elem);
950 *stack = elem->prev;
951 attr_stack_free(elem);
952 }
953
954 /*
955 * bootstrap_attr_stack() should have added, and the
956 * above loop should have stopped before popping, the
957 * root element whose attr_stack->origin is set to an
958 * empty string.
959 */
960 assert((*stack)->origin);
961
962 strbuf_addstr(&pathbuf, (*stack)->origin);
963 /* Build up to the directory 'path' is in */
964 while (pathbuf.len < dirlen) {
965 size_t len = pathbuf.len;
966 struct attr_stack *next;
967 char *origin;
968
969 /* Skip path-separator */
970 if (len < dirlen && is_dir_sep(path[len]))
971 len++;
972 /* Find the end of the next component */
973 while (len < dirlen && !is_dir_sep(path[len]))
974 len++;
975
976 if (pathbuf.len > 0)
977 strbuf_addch(&pathbuf, '/');
978 strbuf_add(&pathbuf, path + pathbuf.len, (len - pathbuf.len));
979 strbuf_addf(&pathbuf, "/%s", GITATTRIBUTES_FILE);
980
981 next = read_attr(istate, pathbuf.buf, 0);
982
983 /* reset the pathbuf to not include "/.gitattributes" */
984 strbuf_setlen(&pathbuf, len);
985
986 origin = xstrdup(pathbuf.buf);
987 push_stack(stack, next, origin, len);
988 }
989
990 /*
991 * Finally push the "info" one at the top of the stack.
992 */
993 push_stack(stack, info, NULL, 0);
994
995 strbuf_release(&pathbuf);
996 }
997
998 static int path_matches(const char *pathname, int pathlen,
999 int basename_offset,
1000 const struct pattern *pat,
1001 const char *base, int baselen)
1002 {
1003 const char *pattern = pat->pattern;
1004 int prefix = pat->nowildcardlen;
1005 int isdir = (pathlen && pathname[pathlen - 1] == '/');
1006
1007 if ((pat->flags & EXC_FLAG_MUSTBEDIR) && !isdir)
1008 return 0;
1009
1010 if (pat->flags & EXC_FLAG_NODIR) {
1011 return match_basename(pathname + basename_offset,
1012 pathlen - basename_offset - isdir,
1013 pattern, prefix,
1014 pat->patternlen, pat->flags);
1015 }
1016 return match_pathname(pathname, pathlen - isdir,
1017 base, baselen,
1018 pattern, prefix, pat->patternlen, pat->flags);
1019 }
1020
1021 static int macroexpand_one(struct all_attrs_item *all_attrs, int nr, int rem);
1022
1023 static int fill_one(const char *what, struct all_attrs_item *all_attrs,
1024 const struct match_attr *a, int rem)
1025 {
1026 int i;
1027
1028 for (i = a->num_attr - 1; rem > 0 && i >= 0; i--) {
1029 const struct git_attr *attr = a->state[i].attr;
1030 const char **n = &(all_attrs[attr->attr_nr].value);
1031 const char *v = a->state[i].setto;
1032
1033 if (*n == ATTR__UNKNOWN) {
1034 debug_set(what,
1035 a->is_macro ? a->u.attr->name : a->u.pat.pattern,
1036 attr, v);
1037 *n = v;
1038 rem--;
1039 rem = macroexpand_one(all_attrs, attr->attr_nr, rem);
1040 }
1041 }
1042 return rem;
1043 }
1044
1045 static int fill(const char *path, int pathlen, int basename_offset,
1046 const struct attr_stack *stack,
1047 struct all_attrs_item *all_attrs, int rem)
1048 {
1049 for (; rem > 0 && stack; stack = stack->prev) {
1050 int i;
1051 const char *base = stack->origin ? stack->origin : "";
1052
1053 for (i = stack->num_matches - 1; 0 < rem && 0 <= i; i--) {
1054 const struct match_attr *a = stack->attrs[i];
1055 if (a->is_macro)
1056 continue;
1057 if (path_matches(path, pathlen, basename_offset,
1058 &a->u.pat, base, stack->originlen))
1059 rem = fill_one("fill", all_attrs, a, rem);
1060 }
1061 }
1062
1063 return rem;
1064 }
1065
1066 static int macroexpand_one(struct all_attrs_item *all_attrs, int nr, int rem)
1067 {
1068 const struct all_attrs_item *item = &all_attrs[nr];
1069
1070 if (item->macro && item->value == ATTR__TRUE)
1071 return fill_one("expand", all_attrs, item->macro, rem);
1072 else
1073 return rem;
1074 }
1075
1076 /*
1077 * Marks the attributes which are macros based on the attribute stack.
1078 * This prevents having to search through the attribute stack each time
1079 * a macro needs to be expanded during the fill stage.
1080 */
1081 static void determine_macros(struct all_attrs_item *all_attrs,
1082 const struct attr_stack *stack)
1083 {
1084 for (; stack; stack = stack->prev) {
1085 int i;
1086 for (i = stack->num_matches - 1; i >= 0; i--) {
1087 const struct match_attr *ma = stack->attrs[i];
1088 if (ma->is_macro) {
1089 int n = ma->u.attr->attr_nr;
1090 if (!all_attrs[n].macro) {
1091 all_attrs[n].macro = ma;
1092 }
1093 }
1094 }
1095 }
1096 }
1097
1098 /*
1099 * Collect attributes for path into the array pointed to by check->all_attrs.
1100 * If check->check_nr is non-zero, only attributes in check[] are collected.
1101 * Otherwise all attributes are collected.
1102 */
1103 static void collect_some_attrs(const struct index_state *istate,
1104 const char *path,
1105 struct attr_check *check)
1106 {
1107 int i, pathlen, rem, dirlen;
1108 const char *cp, *last_slash = NULL;
1109 int basename_offset;
1110
1111 for (cp = path; *cp; cp++) {
1112 if (*cp == '/' && cp[1])
1113 last_slash = cp;
1114 }
1115 pathlen = cp - path;
1116 if (last_slash) {
1117 basename_offset = last_slash + 1 - path;
1118 dirlen = last_slash - path;
1119 } else {
1120 basename_offset = 0;
1121 dirlen = 0;
1122 }
1123
1124 prepare_attr_stack(istate, path, dirlen, &check->stack);
1125 all_attrs_init(&g_attr_hashmap, check);
1126 determine_macros(check->all_attrs, check->stack);
1127
1128 if (check->nr) {
1129 rem = 0;
1130 for (i = 0; i < check->nr; i++) {
1131 int n = check->items[i].attr->attr_nr;
1132 struct all_attrs_item *item = &check->all_attrs[n];
1133 if (item->macro) {
1134 item->value = ATTR__UNSET;
1135 rem++;
1136 }
1137 }
1138 if (rem == check->nr)
1139 return;
1140 }
1141
1142 rem = check->all_attrs_nr;
1143 fill(path, pathlen, basename_offset, check->stack, check->all_attrs, rem);
1144 }
1145
1146 void git_check_attr(const struct index_state *istate,
1147 const char *path,
1148 struct attr_check *check)
1149 {
1150 int i;
1151
1152 collect_some_attrs(istate, path, check);
1153
1154 for (i = 0; i < check->nr; i++) {
1155 size_t n = check->items[i].attr->attr_nr;
1156 const char *value = check->all_attrs[n].value;
1157 if (value == ATTR__UNKNOWN)
1158 value = ATTR__UNSET;
1159 check->items[i].value = value;
1160 }
1161 }
1162
1163 void git_all_attrs(const struct index_state *istate,
1164 const char *path, struct attr_check *check)
1165 {
1166 int i;
1167
1168 attr_check_reset(check);
1169 collect_some_attrs(istate, path, check);
1170
1171 for (i = 0; i < check->all_attrs_nr; i++) {
1172 const char *name = check->all_attrs[i].attr->name;
1173 const char *value = check->all_attrs[i].value;
1174 struct attr_check_item *item;
1175 if (value == ATTR__UNSET || value == ATTR__UNKNOWN)
1176 continue;
1177 item = attr_check_append(check, git_attr(name));
1178 item->value = value;
1179 }
1180 }
1181
1182 void attr_start(void)
1183 {
1184 #ifndef NO_PTHREADS
1185 pthread_mutex_init(&g_attr_hashmap.mutex, NULL);
1186 pthread_mutex_init(&check_vector.mutex, NULL);
1187 #endif
1188 }