pathspec: allow querying for attributes
[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 "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
19 const char git_attr__true[] = "(builtin)true";
20 const char git_attr__false[] = "\0(builtin)false";
21 static 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
31 struct git_attr {
32 int attr_nr; /* unique attribute number */
33 char name[FLEX_ARRAY]; /* attribute name */
34 };
35
36 const char *git_attr_name(const struct git_attr *attr)
37 {
38 return attr->name;
39 }
40
41 struct attr_hashmap {
42 struct hashmap map;
43 #ifndef NO_PTHREADS
44 pthread_mutex_t mutex;
45 #endif
46 };
47
48 static inline void hashmap_lock(struct attr_hashmap *map)
49 {
50 #ifndef NO_PTHREADS
51 pthread_mutex_lock(&map->mutex);
52 #endif
53 }
54
55 static 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 */
67 static struct attr_hashmap g_attr_hashmap;
68
69 /* The container for objects stored in "struct attr_hashmap" */
70 struct 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 */
78 static 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 */
86 static 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 */
95 static 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'. */
113 static 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
131 struct 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 */
147 static 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
189 static 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
208 static 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 */
223 static const 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
247 const struct 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? */
253 struct attr_state {
254 const struct git_attr *attr;
255 const char *setto;
256 };
257
258 struct 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 */
278 struct match_attr {
279 union {
280 struct pattern pat;
281 const struct git_attr *attr;
282 } u;
283 char is_macro;
284 unsigned num_attr;
285 struct attr_state state[FLEX_ARRAY];
286 };
287
288 static 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 */
297 static 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
342 static 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
424 fail_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 struct attr_stack {
449 struct attr_stack *prev;
450 char *origin;
451 size_t originlen;
452 unsigned num_matches;
453 unsigned alloc;
454 struct match_attr **attrs;
455 };
456
457 static void attr_stack_free(struct attr_stack *e)
458 {
459 int i;
460 free(e->origin);
461 for (i = 0; i < e->num_matches; i++) {
462 struct match_attr *a = e->attrs[i];
463 int j;
464 for (j = 0; j < a->num_attr; j++) {
465 const char *setto = a->state[j].setto;
466 if (setto == ATTR__TRUE ||
467 setto == ATTR__FALSE ||
468 setto == ATTR__UNSET ||
469 setto == ATTR__UNKNOWN)
470 ;
471 else
472 free((char *) setto);
473 }
474 free(a);
475 }
476 free(e->attrs);
477 free(e);
478 }
479
480 static void drop_attr_stack(struct attr_stack **stack)
481 {
482 while (*stack) {
483 struct attr_stack *elem = *stack;
484 *stack = elem->prev;
485 attr_stack_free(elem);
486 }
487 }
488
489 /* List of all attr_check structs; access should be surrounded by mutex */
490 static struct check_vector {
491 size_t nr;
492 size_t alloc;
493 struct attr_check **checks;
494 #ifndef NO_PTHREADS
495 pthread_mutex_t mutex;
496 #endif
497 } check_vector;
498
499 static inline void vector_lock(void)
500 {
501 #ifndef NO_PTHREADS
502 pthread_mutex_lock(&check_vector.mutex);
503 #endif
504 }
505
506 static inline void vector_unlock(void)
507 {
508 #ifndef NO_PTHREADS
509 pthread_mutex_unlock(&check_vector.mutex);
510 #endif
511 }
512
513 static void check_vector_add(struct attr_check *c)
514 {
515 vector_lock();
516
517 ALLOC_GROW(check_vector.checks,
518 check_vector.nr + 1,
519 check_vector.alloc);
520 check_vector.checks[check_vector.nr++] = c;
521
522 vector_unlock();
523 }
524
525 static void check_vector_remove(struct attr_check *check)
526 {
527 int i;
528
529 vector_lock();
530
531 /* Find entry */
532 for (i = 0; i < check_vector.nr; i++)
533 if (check_vector.checks[i] == check)
534 break;
535
536 if (i >= check_vector.nr)
537 die("BUG: no entry found");
538
539 /* shift entries over */
540 for (; i < check_vector.nr - 1; i++)
541 check_vector.checks[i] = check_vector.checks[i + 1];
542
543 check_vector.nr--;
544
545 vector_unlock();
546 }
547
548 /* Iterate through all attr_check instances and drop their stacks */
549 static void drop_all_attr_stacks(void)
550 {
551 int i;
552
553 vector_lock();
554
555 for (i = 0; i < check_vector.nr; i++) {
556 drop_attr_stack(&check_vector.checks[i]->stack);
557 }
558
559 vector_unlock();
560 }
561
562 struct attr_check *attr_check_alloc(void)
563 {
564 struct attr_check *c = xcalloc(1, sizeof(struct attr_check));
565
566 /* save pointer to the check struct */
567 check_vector_add(c);
568
569 return c;
570 }
571
572 struct attr_check *attr_check_initl(const char *one, ...)
573 {
574 struct attr_check *check;
575 int cnt;
576 va_list params;
577 const char *param;
578
579 va_start(params, one);
580 for (cnt = 1; (param = va_arg(params, const char *)) != NULL; cnt++)
581 ;
582 va_end(params);
583
584 check = attr_check_alloc();
585 check->nr = cnt;
586 check->alloc = cnt;
587 check->items = xcalloc(cnt, sizeof(struct attr_check_item));
588
589 check->items[0].attr = git_attr(one);
590 va_start(params, one);
591 for (cnt = 1; cnt < check->nr; cnt++) {
592 const struct git_attr *attr;
593 param = va_arg(params, const char *);
594 if (!param)
595 die("BUG: counted %d != ended at %d",
596 check->nr, cnt);
597 attr = git_attr(param);
598 if (!attr)
599 die("BUG: %s: not a valid attribute name", param);
600 check->items[cnt].attr = attr;
601 }
602 va_end(params);
603 return check;
604 }
605
606 struct attr_check *attr_check_dup(const struct attr_check *check)
607 {
608 struct attr_check *ret;
609
610 if (!check)
611 return NULL;
612
613 ret = attr_check_alloc();
614
615 ret->nr = check->nr;
616 ret->alloc = check->alloc;
617 ALLOC_ARRAY(ret->items, ret->nr);
618 COPY_ARRAY(ret->items, check->items, ret->nr);
619
620 return ret;
621 }
622
623 struct attr_check_item *attr_check_append(struct attr_check *check,
624 const struct git_attr *attr)
625 {
626 struct attr_check_item *item;
627
628 ALLOC_GROW(check->items, check->nr + 1, check->alloc);
629 item = &check->items[check->nr++];
630 item->attr = attr;
631 return item;
632 }
633
634 void attr_check_reset(struct attr_check *check)
635 {
636 check->nr = 0;
637 }
638
639 void attr_check_clear(struct attr_check *check)
640 {
641 free(check->items);
642 check->items = NULL;
643 check->alloc = 0;
644 check->nr = 0;
645
646 free(check->all_attrs);
647 check->all_attrs = NULL;
648 check->all_attrs_nr = 0;
649
650 drop_attr_stack(&check->stack);
651 }
652
653 void attr_check_free(struct attr_check *check)
654 {
655 if (check) {
656 /* Remove check from the check vector */
657 check_vector_remove(check);
658
659 attr_check_clear(check);
660 free(check);
661 }
662 }
663
664 static const char *builtin_attr[] = {
665 "[attr]binary -diff -merge -text",
666 NULL,
667 };
668
669 static void handle_attr_line(struct attr_stack *res,
670 const char *line,
671 const char *src,
672 int lineno,
673 int macro_ok)
674 {
675 struct match_attr *a;
676
677 a = parse_attr_line(line, src, lineno, macro_ok);
678 if (!a)
679 return;
680 ALLOC_GROW(res->attrs, res->num_matches + 1, res->alloc);
681 res->attrs[res->num_matches++] = a;
682 }
683
684 static struct attr_stack *read_attr_from_array(const char **list)
685 {
686 struct attr_stack *res;
687 const char *line;
688 int lineno = 0;
689
690 res = xcalloc(1, sizeof(*res));
691 while ((line = *(list++)) != NULL)
692 handle_attr_line(res, line, "[builtin]", ++lineno, 1);
693 return res;
694 }
695
696 /*
697 * Callers into the attribute system assume there is a single, system-wide
698 * global state where attributes are read from and when the state is flipped by
699 * calling git_attr_set_direction(), the stack frames that have been
700 * constructed need to be discarded so so that subsequent calls into the
701 * attribute system will lazily read from the right place. Since changing
702 * direction causes a global paradigm shift, it should not ever be called while
703 * another thread could potentially be calling into the attribute system.
704 */
705 static enum git_attr_direction direction;
706 static struct index_state *use_index;
707
708 void git_attr_set_direction(enum git_attr_direction new_direction,
709 struct index_state *istate)
710 {
711 if (is_bare_repository() && new_direction != GIT_ATTR_INDEX)
712 die("BUG: non-INDEX attr direction in a bare repo");
713
714 if (new_direction != direction)
715 drop_all_attr_stacks();
716
717 direction = new_direction;
718 use_index = istate;
719 }
720
721 static struct attr_stack *read_attr_from_file(const char *path, int macro_ok)
722 {
723 FILE *fp = fopen(path, "r");
724 struct attr_stack *res;
725 char buf[2048];
726 int lineno = 0;
727
728 if (!fp) {
729 if (errno != ENOENT && errno != ENOTDIR)
730 warn_on_inaccessible(path);
731 return NULL;
732 }
733 res = xcalloc(1, sizeof(*res));
734 while (fgets(buf, sizeof(buf), fp)) {
735 char *bufp = buf;
736 if (!lineno)
737 skip_utf8_bom(&bufp, strlen(bufp));
738 handle_attr_line(res, bufp, path, ++lineno, macro_ok);
739 }
740 fclose(fp);
741 return res;
742 }
743
744 static struct attr_stack *read_attr_from_index(const char *path, int macro_ok)
745 {
746 struct attr_stack *res;
747 char *buf, *sp;
748 int lineno = 0;
749
750 buf = read_blob_data_from_index(use_index ? use_index : &the_index, path, NULL);
751 if (!buf)
752 return NULL;
753
754 res = xcalloc(1, sizeof(*res));
755 for (sp = buf; *sp; ) {
756 char *ep;
757 int more;
758
759 ep = strchrnul(sp, '\n');
760 more = (*ep == '\n');
761 *ep = '\0';
762 handle_attr_line(res, sp, path, ++lineno, macro_ok);
763 sp = ep + more;
764 }
765 free(buf);
766 return res;
767 }
768
769 static struct attr_stack *read_attr(const char *path, int macro_ok)
770 {
771 struct attr_stack *res = NULL;
772
773 if (direction == GIT_ATTR_INDEX) {
774 res = read_attr_from_index(path, macro_ok);
775 } else if (!is_bare_repository()) {
776 if (direction == GIT_ATTR_CHECKOUT) {
777 res = read_attr_from_index(path, macro_ok);
778 if (!res)
779 res = read_attr_from_file(path, macro_ok);
780 } else if (direction == GIT_ATTR_CHECKIN) {
781 res = read_attr_from_file(path, macro_ok);
782 if (!res)
783 /*
784 * There is no checked out .gitattributes file
785 * there, but we might have it in the index.
786 * We allow operation in a sparsely checked out
787 * work tree, so read from it.
788 */
789 res = read_attr_from_index(path, macro_ok);
790 }
791 }
792
793 if (!res)
794 res = xcalloc(1, sizeof(*res));
795 return res;
796 }
797
798 #if DEBUG_ATTR
799 static void debug_info(const char *what, struct attr_stack *elem)
800 {
801 fprintf(stderr, "%s: %s\n", what, elem->origin ? elem->origin : "()");
802 }
803 static void debug_set(const char *what, const char *match, struct git_attr *attr, const void *v)
804 {
805 const char *value = v;
806
807 if (ATTR_TRUE(value))
808 value = "set";
809 else if (ATTR_FALSE(value))
810 value = "unset";
811 else if (ATTR_UNSET(value))
812 value = "unspecified";
813
814 fprintf(stderr, "%s: %s => %s (%s)\n",
815 what, attr->name, (char *) value, match);
816 }
817 #define debug_push(a) debug_info("push", (a))
818 #define debug_pop(a) debug_info("pop", (a))
819 #else
820 #define debug_push(a) do { ; } while (0)
821 #define debug_pop(a) do { ; } while (0)
822 #define debug_set(a,b,c,d) do { ; } while (0)
823 #endif /* DEBUG_ATTR */
824
825 static const char *git_etc_gitattributes(void)
826 {
827 static const char *system_wide;
828 if (!system_wide)
829 system_wide = system_path(ETC_GITATTRIBUTES);
830 return system_wide;
831 }
832
833 static const char *get_home_gitattributes(void)
834 {
835 if (!git_attributes_file)
836 git_attributes_file = xdg_config_home("attributes");
837
838 return git_attributes_file;
839 }
840
841 static int git_attr_system(void)
842 {
843 return !git_env_bool("GIT_ATTR_NOSYSTEM", 0);
844 }
845
846 static GIT_PATH_FUNC(git_path_info_attributes, INFOATTRIBUTES_FILE)
847
848 static void push_stack(struct attr_stack **attr_stack_p,
849 struct attr_stack *elem, char *origin, size_t originlen)
850 {
851 if (elem) {
852 elem->origin = origin;
853 if (origin)
854 elem->originlen = originlen;
855 elem->prev = *attr_stack_p;
856 *attr_stack_p = elem;
857 }
858 }
859
860 static void bootstrap_attr_stack(struct attr_stack **stack)
861 {
862 struct attr_stack *e;
863
864 if (*stack)
865 return;
866
867 /* builtin frame */
868 e = read_attr_from_array(builtin_attr);
869 push_stack(stack, e, NULL, 0);
870
871 /* system-wide frame */
872 if (git_attr_system()) {
873 e = read_attr_from_file(git_etc_gitattributes(), 1);
874 push_stack(stack, e, NULL, 0);
875 }
876
877 /* home directory */
878 if (get_home_gitattributes()) {
879 e = read_attr_from_file(get_home_gitattributes(), 1);
880 push_stack(stack, e, NULL, 0);
881 }
882
883 /* root directory */
884 e = read_attr(GITATTRIBUTES_FILE, 1);
885 push_stack(stack, e, xstrdup(""), 0);
886
887 /* info frame */
888 if (startup_info->have_repository)
889 e = read_attr_from_file(git_path_info_attributes(), 1);
890 else
891 e = NULL;
892 if (!e)
893 e = xcalloc(1, sizeof(struct attr_stack));
894 push_stack(stack, e, NULL, 0);
895 }
896
897 static void prepare_attr_stack(const char *path, int dirlen,
898 struct attr_stack **stack)
899 {
900 struct attr_stack *info;
901 struct strbuf pathbuf = STRBUF_INIT;
902
903 /*
904 * At the bottom of the attribute stack is the built-in
905 * set of attribute definitions, followed by the contents
906 * of $(prefix)/etc/gitattributes and a file specified by
907 * core.attributesfile. Then, contents from
908 * .gitattribute files from directories closer to the
909 * root to the ones in deeper directories are pushed
910 * to the stack. Finally, at the very top of the stack
911 * we always keep the contents of $GIT_DIR/info/attributes.
912 *
913 * When checking, we use entries from near the top of the
914 * stack, preferring $GIT_DIR/info/attributes, then
915 * .gitattributes in deeper directories to shallower ones,
916 * and finally use the built-in set as the default.
917 */
918 bootstrap_attr_stack(stack);
919
920 /*
921 * Pop the "info" one that is always at the top of the stack.
922 */
923 info = *stack;
924 *stack = info->prev;
925
926 /*
927 * Pop the ones from directories that are not the prefix of
928 * the path we are checking. Break out of the loop when we see
929 * the root one (whose origin is an empty string "") or the builtin
930 * one (whose origin is NULL) without popping it.
931 */
932 while ((*stack)->origin) {
933 int namelen = (*stack)->originlen;
934 struct attr_stack *elem;
935
936 elem = *stack;
937 if (namelen <= dirlen &&
938 !strncmp(elem->origin, path, namelen) &&
939 (!namelen || path[namelen] == '/'))
940 break;
941
942 debug_pop(elem);
943 *stack = elem->prev;
944 attr_stack_free(elem);
945 }
946
947 /*
948 * bootstrap_attr_stack() should have added, and the
949 * above loop should have stopped before popping, the
950 * root element whose attr_stack->origin is set to an
951 * empty string.
952 */
953 assert((*stack)->origin);
954
955 strbuf_addstr(&pathbuf, (*stack)->origin);
956 /* Build up to the directory 'path' is in */
957 while (pathbuf.len < dirlen) {
958 size_t len = pathbuf.len;
959 struct attr_stack *next;
960 char *origin;
961
962 /* Skip path-separator */
963 if (len < dirlen && is_dir_sep(path[len]))
964 len++;
965 /* Find the end of the next component */
966 while (len < dirlen && !is_dir_sep(path[len]))
967 len++;
968
969 if (pathbuf.len > 0)
970 strbuf_addch(&pathbuf, '/');
971 strbuf_add(&pathbuf, path + pathbuf.len, (len - pathbuf.len));
972 strbuf_addf(&pathbuf, "/%s", GITATTRIBUTES_FILE);
973
974 next = read_attr(pathbuf.buf, 0);
975
976 /* reset the pathbuf to not include "/.gitattributes" */
977 strbuf_setlen(&pathbuf, len);
978
979 origin = xstrdup(pathbuf.buf);
980 push_stack(stack, next, origin, len);
981 }
982
983 /*
984 * Finally push the "info" one at the top of the stack.
985 */
986 push_stack(stack, info, NULL, 0);
987
988 strbuf_release(&pathbuf);
989 }
990
991 static int path_matches(const char *pathname, int pathlen,
992 int basename_offset,
993 const struct pattern *pat,
994 const char *base, int baselen)
995 {
996 const char *pattern = pat->pattern;
997 int prefix = pat->nowildcardlen;
998 int isdir = (pathlen && pathname[pathlen - 1] == '/');
999
1000 if ((pat->flags & EXC_FLAG_MUSTBEDIR) && !isdir)
1001 return 0;
1002
1003 if (pat->flags & EXC_FLAG_NODIR) {
1004 return match_basename(pathname + basename_offset,
1005 pathlen - basename_offset - isdir,
1006 pattern, prefix,
1007 pat->patternlen, pat->flags);
1008 }
1009 return match_pathname(pathname, pathlen - isdir,
1010 base, baselen,
1011 pattern, prefix, pat->patternlen, pat->flags);
1012 }
1013
1014 static int macroexpand_one(struct all_attrs_item *all_attrs, int nr, int rem);
1015
1016 static int fill_one(const char *what, struct all_attrs_item *all_attrs,
1017 const struct match_attr *a, int rem)
1018 {
1019 int i;
1020
1021 for (i = a->num_attr - 1; rem > 0 && i >= 0; i--) {
1022 const struct git_attr *attr = a->state[i].attr;
1023 const char **n = &(all_attrs[attr->attr_nr].value);
1024 const char *v = a->state[i].setto;
1025
1026 if (*n == ATTR__UNKNOWN) {
1027 debug_set(what,
1028 a->is_macro ? a->u.attr->name : a->u.pat.pattern,
1029 attr, v);
1030 *n = v;
1031 rem--;
1032 rem = macroexpand_one(all_attrs, attr->attr_nr, rem);
1033 }
1034 }
1035 return rem;
1036 }
1037
1038 static int fill(const char *path, int pathlen, int basename_offset,
1039 const struct attr_stack *stack,
1040 struct all_attrs_item *all_attrs, int rem)
1041 {
1042 for (; rem > 0 && stack; stack = stack->prev) {
1043 int i;
1044 const char *base = stack->origin ? stack->origin : "";
1045
1046 for (i = stack->num_matches - 1; 0 < rem && 0 <= i; i--) {
1047 const struct match_attr *a = stack->attrs[i];
1048 if (a->is_macro)
1049 continue;
1050 if (path_matches(path, pathlen, basename_offset,
1051 &a->u.pat, base, stack->originlen))
1052 rem = fill_one("fill", all_attrs, a, rem);
1053 }
1054 }
1055
1056 return rem;
1057 }
1058
1059 static int macroexpand_one(struct all_attrs_item *all_attrs, int nr, int rem)
1060 {
1061 const struct all_attrs_item *item = &all_attrs[nr];
1062
1063 if (item->macro && item->value == ATTR__TRUE)
1064 return fill_one("expand", all_attrs, item->macro, rem);
1065 else
1066 return rem;
1067 }
1068
1069 /*
1070 * Marks the attributes which are macros based on the attribute stack.
1071 * This prevents having to search through the attribute stack each time
1072 * a macro needs to be expanded during the fill stage.
1073 */
1074 static void determine_macros(struct all_attrs_item *all_attrs,
1075 const struct attr_stack *stack)
1076 {
1077 for (; stack; stack = stack->prev) {
1078 int i;
1079 for (i = stack->num_matches - 1; i >= 0; i--) {
1080 const struct match_attr *ma = stack->attrs[i];
1081 if (ma->is_macro) {
1082 int n = ma->u.attr->attr_nr;
1083 if (!all_attrs[n].macro) {
1084 all_attrs[n].macro = ma;
1085 }
1086 }
1087 }
1088 }
1089 }
1090
1091 /*
1092 * Collect attributes for path into the array pointed to by check->all_attrs.
1093 * If check->check_nr is non-zero, only attributes in check[] are collected.
1094 * Otherwise all attributes are collected.
1095 */
1096 static void collect_some_attrs(const char *path, struct attr_check *check)
1097 {
1098 int i, pathlen, rem, dirlen;
1099 const char *cp, *last_slash = NULL;
1100 int basename_offset;
1101
1102 for (cp = path; *cp; cp++) {
1103 if (*cp == '/' && cp[1])
1104 last_slash = cp;
1105 }
1106 pathlen = cp - path;
1107 if (last_slash) {
1108 basename_offset = last_slash + 1 - path;
1109 dirlen = last_slash - path;
1110 } else {
1111 basename_offset = 0;
1112 dirlen = 0;
1113 }
1114
1115 prepare_attr_stack(path, dirlen, &check->stack);
1116 all_attrs_init(&g_attr_hashmap, check);
1117 determine_macros(check->all_attrs, check->stack);
1118
1119 if (check->nr) {
1120 rem = 0;
1121 for (i = 0; i < check->nr; i++) {
1122 int n = check->items[i].attr->attr_nr;
1123 struct all_attrs_item *item = &check->all_attrs[n];
1124 if (item->macro) {
1125 item->value = ATTR__UNSET;
1126 rem++;
1127 }
1128 }
1129 if (rem == check->nr)
1130 return;
1131 }
1132
1133 rem = check->all_attrs_nr;
1134 fill(path, pathlen, basename_offset, check->stack, check->all_attrs, rem);
1135 }
1136
1137 int git_check_attr(const char *path, struct attr_check *check)
1138 {
1139 int i;
1140
1141 collect_some_attrs(path, check);
1142
1143 for (i = 0; i < check->nr; i++) {
1144 size_t n = check->items[i].attr->attr_nr;
1145 const char *value = check->all_attrs[n].value;
1146 if (value == ATTR__UNKNOWN)
1147 value = ATTR__UNSET;
1148 check->items[i].value = value;
1149 }
1150
1151 return 0;
1152 }
1153
1154 void git_all_attrs(const char *path, struct attr_check *check)
1155 {
1156 int i;
1157
1158 attr_check_reset(check);
1159 collect_some_attrs(path, check);
1160
1161 for (i = 0; i < check->all_attrs_nr; i++) {
1162 const char *name = check->all_attrs[i].attr->name;
1163 const char *value = check->all_attrs[i].value;
1164 struct attr_check_item *item;
1165 if (value == ATTR__UNSET || value == ATTR__UNKNOWN)
1166 continue;
1167 item = attr_check_append(check, git_attr(name));
1168 item->value = value;
1169 }
1170 }
1171
1172 void attr_start(void)
1173 {
1174 #ifndef NO_PTHREADS
1175 pthread_mutex_init(&g_attr_hashmap.mutex, NULL);
1176 pthread_mutex_init(&check_vector.mutex, NULL);
1177 #endif
1178 }