2f49151736095cf2e2a76ea91d72743df3778a5d
[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
155 hashmap_lock(map);
156
157 if (map->map.size < check->all_attrs_nr)
158 die("BUG: interned attributes shouldn't be deleted");
159
160 /*
161 * If the number of attributes in the global dictionary has increased
162 * (or this attr_check instance doesn't have an initialized all_attrs
163 * field), reallocate the provided attr_check instance's all_attrs
164 * field and fill each entry with its corresponding git_attr.
165 */
166 if (map->map.size != check->all_attrs_nr) {
167 struct attr_hash_entry *e;
168 struct hashmap_iter iter;
169 hashmap_iter_init(&map->map, &iter);
170
171 REALLOC_ARRAY(check->all_attrs, map->map.size);
172 check->all_attrs_nr = map->map.size;
173
174 while ((e = hashmap_iter_next(&iter))) {
175 const struct git_attr *a = e->value;
176 check->all_attrs[a->attr_nr].attr = a;
177 }
178 }
179
180 hashmap_unlock(map);
181
182 /*
183 * Re-initialize every entry in check->all_attrs.
184 * This re-initialization can live outside of the locked region since
185 * the attribute dictionary is no longer being accessed.
186 */
187 for (i = 0; i < check->all_attrs_nr; i++) {
188 check->all_attrs[i].value = ATTR__UNKNOWN;
189 check->all_attrs[i].macro = NULL;
190 }
191 }
192
193 static int attr_name_valid(const char *name, size_t namelen)
194 {
195 /*
196 * Attribute name cannot begin with '-' and must consist of
197 * characters from [-A-Za-z0-9_.].
198 */
199 if (namelen <= 0 || *name == '-')
200 return 0;
201 while (namelen--) {
202 char ch = *name++;
203 if (! (ch == '-' || ch == '.' || ch == '_' ||
204 ('0' <= ch && ch <= '9') ||
205 ('a' <= ch && ch <= 'z') ||
206 ('A' <= ch && ch <= 'Z')) )
207 return 0;
208 }
209 return 1;
210 }
211
212 static void report_invalid_attr(const char *name, size_t len,
213 const char *src, int lineno)
214 {
215 struct strbuf err = STRBUF_INIT;
216 strbuf_addf(&err, _("%.*s is not a valid attribute name"),
217 (int) len, name);
218 fprintf(stderr, "%s: %s:%d\n", err.buf, src, lineno);
219 strbuf_release(&err);
220 }
221
222 /*
223 * Given a 'name', lookup and return the corresponding attribute in the global
224 * dictionary. If no entry is found, create a new attribute and store it in
225 * the dictionary.
226 */
227 static const struct git_attr *git_attr_internal(const char *name, int namelen)
228 {
229 struct git_attr *a;
230
231 if (!attr_name_valid(name, namelen))
232 return NULL;
233
234 hashmap_lock(&g_attr_hashmap);
235
236 a = attr_hashmap_get(&g_attr_hashmap, name, namelen);
237
238 if (!a) {
239 FLEX_ALLOC_MEM(a, name, name, namelen);
240 a->attr_nr = g_attr_hashmap.map.size;
241
242 attr_hashmap_add(&g_attr_hashmap, a->name, namelen, a);
243 assert(a->attr_nr == (g_attr_hashmap.map.size - 1));
244 }
245
246 hashmap_unlock(&g_attr_hashmap);
247
248 return a;
249 }
250
251 const struct git_attr *git_attr(const char *name)
252 {
253 return git_attr_internal(name, strlen(name));
254 }
255
256 /* What does a matched pattern decide? */
257 struct attr_state {
258 const struct git_attr *attr;
259 const char *setto;
260 };
261
262 struct pattern {
263 const char *pattern;
264 int patternlen;
265 int nowildcardlen;
266 unsigned flags; /* EXC_FLAG_* */
267 };
268
269 /*
270 * One rule, as from a .gitattributes file.
271 *
272 * If is_macro is true, then u.attr is a pointer to the git_attr being
273 * defined.
274 *
275 * If is_macro is false, then u.pat is the filename pattern to which the
276 * rule applies.
277 *
278 * In either case, num_attr is the number of attributes affected by
279 * this rule, and state is an array listing them. The attributes are
280 * listed as they appear in the file (macros unexpanded).
281 */
282 struct match_attr {
283 union {
284 struct pattern pat;
285 const struct git_attr *attr;
286 } u;
287 char is_macro;
288 unsigned num_attr;
289 struct attr_state state[FLEX_ARRAY];
290 };
291
292 static const char blank[] = " \t\r\n";
293
294 /*
295 * Parse a whitespace-delimited attribute state (i.e., "attr",
296 * "-attr", "!attr", or "attr=value") from the string starting at src.
297 * If e is not NULL, write the results to *e. Return a pointer to the
298 * remainder of the string (with leading whitespace removed), or NULL
299 * if there was an error.
300 */
301 static const char *parse_attr(const char *src, int lineno, const char *cp,
302 struct attr_state *e)
303 {
304 const char *ep, *equals;
305 int len;
306
307 ep = cp + strcspn(cp, blank);
308 equals = strchr(cp, '=');
309 if (equals && ep < equals)
310 equals = NULL;
311 if (equals)
312 len = equals - cp;
313 else
314 len = ep - cp;
315 if (!e) {
316 if (*cp == '-' || *cp == '!') {
317 cp++;
318 len--;
319 }
320 if (!attr_name_valid(cp, len)) {
321 report_invalid_attr(cp, len, src, lineno);
322 return NULL;
323 }
324 } else {
325 /*
326 * As this function is always called twice, once with
327 * e == NULL in the first pass and then e != NULL in
328 * the second pass, no need for attr_name_valid()
329 * check here.
330 */
331 if (*cp == '-' || *cp == '!') {
332 e->setto = (*cp == '-') ? ATTR__FALSE : ATTR__UNSET;
333 cp++;
334 len--;
335 }
336 else if (!equals)
337 e->setto = ATTR__TRUE;
338 else {
339 e->setto = xmemdupz(equals + 1, ep - equals - 1);
340 }
341 e->attr = git_attr_internal(cp, len);
342 }
343 return ep + strspn(ep, blank);
344 }
345
346 static struct match_attr *parse_attr_line(const char *line, const char *src,
347 int lineno, int macro_ok)
348 {
349 int namelen;
350 int num_attr, i;
351 const char *cp, *name, *states;
352 struct match_attr *res = NULL;
353 int is_macro;
354 struct strbuf pattern = STRBUF_INIT;
355
356 cp = line + strspn(line, blank);
357 if (!*cp || *cp == '#')
358 return NULL;
359 name = cp;
360
361 if (*cp == '"' && !unquote_c_style(&pattern, name, &states)) {
362 name = pattern.buf;
363 namelen = pattern.len;
364 } else {
365 namelen = strcspn(name, blank);
366 states = name + namelen;
367 }
368
369 if (strlen(ATTRIBUTE_MACRO_PREFIX) < namelen &&
370 starts_with(name, ATTRIBUTE_MACRO_PREFIX)) {
371 if (!macro_ok) {
372 fprintf(stderr, "%s not allowed: %s:%d\n",
373 name, src, lineno);
374 goto fail_return;
375 }
376 is_macro = 1;
377 name += strlen(ATTRIBUTE_MACRO_PREFIX);
378 name += strspn(name, blank);
379 namelen = strcspn(name, blank);
380 if (!attr_name_valid(name, namelen)) {
381 report_invalid_attr(name, namelen, src, lineno);
382 goto fail_return;
383 }
384 }
385 else
386 is_macro = 0;
387
388 states += strspn(states, blank);
389
390 /* First pass to count the attr_states */
391 for (cp = states, num_attr = 0; *cp; num_attr++) {
392 cp = parse_attr(src, lineno, cp, NULL);
393 if (!cp)
394 goto fail_return;
395 }
396
397 res = xcalloc(1,
398 sizeof(*res) +
399 sizeof(struct attr_state) * num_attr +
400 (is_macro ? 0 : namelen + 1));
401 if (is_macro) {
402 res->u.attr = git_attr_internal(name, namelen);
403 } else {
404 char *p = (char *)&(res->state[num_attr]);
405 memcpy(p, name, namelen);
406 res->u.pat.pattern = p;
407 parse_exclude_pattern(&res->u.pat.pattern,
408 &res->u.pat.patternlen,
409 &res->u.pat.flags,
410 &res->u.pat.nowildcardlen);
411 if (res->u.pat.flags & EXC_FLAG_NEGATIVE) {
412 warning(_("Negative patterns are ignored in git attributes\n"
413 "Use '\\!' for literal leading exclamation."));
414 goto fail_return;
415 }
416 }
417 res->is_macro = is_macro;
418 res->num_attr = num_attr;
419
420 /* Second pass to fill the attr_states */
421 for (cp = states, i = 0; *cp; i++) {
422 cp = parse_attr(src, lineno, cp, &(res->state[i]));
423 }
424
425 strbuf_release(&pattern);
426 return res;
427
428 fail_return:
429 strbuf_release(&pattern);
430 free(res);
431 return NULL;
432 }
433
434 /*
435 * Like info/exclude and .gitignore, the attribute information can
436 * come from many places.
437 *
438 * (1) .gitattribute file of the same directory;
439 * (2) .gitattribute file of the parent directory if (1) does not have
440 * any match; this goes recursively upwards, just like .gitignore.
441 * (3) $GIT_DIR/info/attributes, which overrides both of the above.
442 *
443 * In the same file, later entries override the earlier match, so in the
444 * global list, we would have entries from info/attributes the earliest
445 * (reading the file from top to bottom), .gitattribute of the root
446 * directory (again, reading the file from top to bottom) down to the
447 * current directory, and then scan the list backwards to find the first match.
448 * This is exactly the same as what is_excluded() does in dir.c to deal with
449 * .gitignore file and info/excludes file as a fallback.
450 */
451
452 struct attr_stack {
453 struct attr_stack *prev;
454 char *origin;
455 size_t originlen;
456 unsigned num_matches;
457 unsigned alloc;
458 struct match_attr **attrs;
459 };
460
461 static void attr_stack_free(struct attr_stack *e)
462 {
463 int i;
464 free(e->origin);
465 for (i = 0; i < e->num_matches; i++) {
466 struct match_attr *a = e->attrs[i];
467 int j;
468 for (j = 0; j < a->num_attr; j++) {
469 const char *setto = a->state[j].setto;
470 if (setto == ATTR__TRUE ||
471 setto == ATTR__FALSE ||
472 setto == ATTR__UNSET ||
473 setto == ATTR__UNKNOWN)
474 ;
475 else
476 free((char *) setto);
477 }
478 free(a);
479 }
480 free(e->attrs);
481 free(e);
482 }
483
484 static void drop_attr_stack(struct attr_stack **stack)
485 {
486 while (*stack) {
487 struct attr_stack *elem = *stack;
488 *stack = elem->prev;
489 attr_stack_free(elem);
490 }
491 }
492
493 /* List of all attr_check structs; access should be surrounded by mutex */
494 static struct check_vector {
495 size_t nr;
496 size_t alloc;
497 struct attr_check **checks;
498 #ifndef NO_PTHREADS
499 pthread_mutex_t mutex;
500 #endif
501 } check_vector;
502
503 static inline void vector_lock(void)
504 {
505 #ifndef NO_PTHREADS
506 pthread_mutex_lock(&check_vector.mutex);
507 #endif
508 }
509
510 static inline void vector_unlock(void)
511 {
512 #ifndef NO_PTHREADS
513 pthread_mutex_unlock(&check_vector.mutex);
514 #endif
515 }
516
517 static void check_vector_add(struct attr_check *c)
518 {
519 vector_lock();
520
521 ALLOC_GROW(check_vector.checks,
522 check_vector.nr + 1,
523 check_vector.alloc);
524 check_vector.checks[check_vector.nr++] = c;
525
526 vector_unlock();
527 }
528
529 static void check_vector_remove(struct attr_check *check)
530 {
531 int i;
532
533 vector_lock();
534
535 /* Find entry */
536 for (i = 0; i < check_vector.nr; i++)
537 if (check_vector.checks[i] == check)
538 break;
539
540 if (i >= check_vector.nr)
541 die("BUG: no entry found");
542
543 /* shift entries over */
544 for (; i < check_vector.nr - 1; i++)
545 check_vector.checks[i] = check_vector.checks[i + 1];
546
547 check_vector.nr--;
548
549 vector_unlock();
550 }
551
552 /* Iterate through all attr_check instances and drop their stacks */
553 static void drop_all_attr_stacks(void)
554 {
555 int i;
556
557 vector_lock();
558
559 for (i = 0; i < check_vector.nr; i++) {
560 drop_attr_stack(&check_vector.checks[i]->stack);
561 }
562
563 vector_unlock();
564 }
565
566 struct attr_check *attr_check_alloc(void)
567 {
568 struct attr_check *c = xcalloc(1, sizeof(struct attr_check));
569
570 /* save pointer to the check struct */
571 check_vector_add(c);
572
573 return c;
574 }
575
576 struct attr_check *attr_check_initl(const char *one, ...)
577 {
578 struct attr_check *check;
579 int cnt;
580 va_list params;
581 const char *param;
582
583 va_start(params, one);
584 for (cnt = 1; (param = va_arg(params, const char *)) != NULL; cnt++)
585 ;
586 va_end(params);
587
588 check = attr_check_alloc();
589 check->nr = cnt;
590 check->alloc = cnt;
591 check->items = xcalloc(cnt, sizeof(struct attr_check_item));
592
593 check->items[0].attr = git_attr(one);
594 va_start(params, one);
595 for (cnt = 1; cnt < check->nr; cnt++) {
596 const struct git_attr *attr;
597 param = va_arg(params, const char *);
598 if (!param)
599 die("BUG: counted %d != ended at %d",
600 check->nr, cnt);
601 attr = git_attr(param);
602 if (!attr)
603 die("BUG: %s: not a valid attribute name", param);
604 check->items[cnt].attr = attr;
605 }
606 va_end(params);
607 return check;
608 }
609
610 struct attr_check *attr_check_dup(const struct attr_check *check)
611 {
612 struct attr_check *ret;
613
614 if (!check)
615 return NULL;
616
617 ret = attr_check_alloc();
618
619 ret->nr = check->nr;
620 ret->alloc = check->alloc;
621 ALLOC_ARRAY(ret->items, ret->nr);
622 COPY_ARRAY(ret->items, check->items, ret->nr);
623
624 return ret;
625 }
626
627 struct attr_check_item *attr_check_append(struct attr_check *check,
628 const struct git_attr *attr)
629 {
630 struct attr_check_item *item;
631
632 ALLOC_GROW(check->items, check->nr + 1, check->alloc);
633 item = &check->items[check->nr++];
634 item->attr = attr;
635 return item;
636 }
637
638 void attr_check_reset(struct attr_check *check)
639 {
640 check->nr = 0;
641 }
642
643 void attr_check_clear(struct attr_check *check)
644 {
645 FREE_AND_NULL(check->items);
646 check->alloc = 0;
647 check->nr = 0;
648
649 FREE_AND_NULL(check->all_attrs);
650 check->all_attrs_nr = 0;
651
652 drop_attr_stack(&check->stack);
653 }
654
655 void attr_check_free(struct attr_check *check)
656 {
657 if (check) {
658 /* Remove check from the check vector */
659 check_vector_remove(check);
660
661 attr_check_clear(check);
662 free(check);
663 }
664 }
665
666 static const char *builtin_attr[] = {
667 "[attr]binary -diff -merge -text",
668 NULL,
669 };
670
671 static void handle_attr_line(struct attr_stack *res,
672 const char *line,
673 const char *src,
674 int lineno,
675 int macro_ok)
676 {
677 struct match_attr *a;
678
679 a = parse_attr_line(line, src, lineno, macro_ok);
680 if (!a)
681 return;
682 ALLOC_GROW(res->attrs, res->num_matches + 1, res->alloc);
683 res->attrs[res->num_matches++] = a;
684 }
685
686 static struct attr_stack *read_attr_from_array(const char **list)
687 {
688 struct attr_stack *res;
689 const char *line;
690 int lineno = 0;
691
692 res = xcalloc(1, sizeof(*res));
693 while ((line = *(list++)) != NULL)
694 handle_attr_line(res, line, "[builtin]", ++lineno, 1);
695 return res;
696 }
697
698 /*
699 * Callers into the attribute system assume there is a single, system-wide
700 * global state where attributes are read from and when the state is flipped by
701 * calling git_attr_set_direction(), the stack frames that have been
702 * constructed need to be discarded so so that subsequent calls into the
703 * attribute system will lazily read from the right place. Since changing
704 * direction causes a global paradigm shift, it should not ever be called while
705 * another thread could potentially be calling into the attribute system.
706 */
707 static enum git_attr_direction direction;
708 static struct index_state *use_index;
709
710 void git_attr_set_direction(enum git_attr_direction new_direction,
711 struct index_state *istate)
712 {
713 if (is_bare_repository() && new_direction != GIT_ATTR_INDEX)
714 die("BUG: non-INDEX attr direction in a bare repo");
715
716 if (new_direction != direction)
717 drop_all_attr_stacks();
718
719 direction = new_direction;
720 use_index = istate;
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 char *path, int macro_ok)
744 {
745 struct attr_stack *res;
746 char *buf, *sp;
747 int lineno = 0;
748
749 buf = read_blob_data_from_index(use_index ? use_index : &the_index, path, NULL);
750 if (!buf)
751 return NULL;
752
753 res = xcalloc(1, sizeof(*res));
754 for (sp = buf; *sp; ) {
755 char *ep;
756 int more;
757
758 ep = strchrnul(sp, '\n');
759 more = (*ep == '\n');
760 *ep = '\0';
761 handle_attr_line(res, sp, path, ++lineno, macro_ok);
762 sp = ep + more;
763 }
764 free(buf);
765 return res;
766 }
767
768 static struct attr_stack *read_attr(const char *path, int macro_ok)
769 {
770 struct attr_stack *res = NULL;
771
772 if (direction == GIT_ATTR_INDEX) {
773 res = read_attr_from_index(path, macro_ok);
774 } else if (!is_bare_repository()) {
775 if (direction == GIT_ATTR_CHECKOUT) {
776 res = read_attr_from_index(path, macro_ok);
777 if (!res)
778 res = read_attr_from_file(path, macro_ok);
779 } else if (direction == GIT_ATTR_CHECKIN) {
780 res = read_attr_from_file(path, macro_ok);
781 if (!res)
782 /*
783 * There is no checked out .gitattributes file
784 * there, but we might have it in the index.
785 * We allow operation in a sparsely checked out
786 * work tree, so read from it.
787 */
788 res = read_attr_from_index(path, macro_ok);
789 }
790 }
791
792 if (!res)
793 res = xcalloc(1, sizeof(*res));
794 return res;
795 }
796
797 #if DEBUG_ATTR
798 static void debug_info(const char *what, struct attr_stack *elem)
799 {
800 fprintf(stderr, "%s: %s\n", what, elem->origin ? elem->origin : "()");
801 }
802 static void debug_set(const char *what, const char *match, struct git_attr *attr, const void *v)
803 {
804 const char *value = v;
805
806 if (ATTR_TRUE(value))
807 value = "set";
808 else if (ATTR_FALSE(value))
809 value = "unset";
810 else if (ATTR_UNSET(value))
811 value = "unspecified";
812
813 fprintf(stderr, "%s: %s => %s (%s)\n",
814 what, attr->name, (char *) value, match);
815 }
816 #define debug_push(a) debug_info("push", (a))
817 #define debug_pop(a) debug_info("pop", (a))
818 #else
819 #define debug_push(a) do { ; } while (0)
820 #define debug_pop(a) do { ; } while (0)
821 #define debug_set(a,b,c,d) do { ; } while (0)
822 #endif /* DEBUG_ATTR */
823
824 static const char *git_etc_gitattributes(void)
825 {
826 static const char *system_wide;
827 if (!system_wide)
828 system_wide = system_path(ETC_GITATTRIBUTES);
829 return system_wide;
830 }
831
832 static const char *get_home_gitattributes(void)
833 {
834 if (!git_attributes_file)
835 git_attributes_file = xdg_config_home("attributes");
836
837 return git_attributes_file;
838 }
839
840 static int git_attr_system(void)
841 {
842 return !git_env_bool("GIT_ATTR_NOSYSTEM", 0);
843 }
844
845 static GIT_PATH_FUNC(git_path_info_attributes, INFOATTRIBUTES_FILE)
846
847 static void push_stack(struct attr_stack **attr_stack_p,
848 struct attr_stack *elem, char *origin, size_t originlen)
849 {
850 if (elem) {
851 elem->origin = origin;
852 if (origin)
853 elem->originlen = originlen;
854 elem->prev = *attr_stack_p;
855 *attr_stack_p = elem;
856 }
857 }
858
859 static void bootstrap_attr_stack(struct attr_stack **stack)
860 {
861 struct attr_stack *e;
862
863 if (*stack)
864 return;
865
866 /* builtin frame */
867 e = read_attr_from_array(builtin_attr);
868 push_stack(stack, e, NULL, 0);
869
870 /* system-wide frame */
871 if (git_attr_system()) {
872 e = read_attr_from_file(git_etc_gitattributes(), 1);
873 push_stack(stack, e, NULL, 0);
874 }
875
876 /* home directory */
877 if (get_home_gitattributes()) {
878 e = read_attr_from_file(get_home_gitattributes(), 1);
879 push_stack(stack, e, NULL, 0);
880 }
881
882 /* root directory */
883 e = read_attr(GITATTRIBUTES_FILE, 1);
884 push_stack(stack, e, xstrdup(""), 0);
885
886 /* info frame */
887 if (startup_info->have_repository)
888 e = read_attr_from_file(git_path_info_attributes(), 1);
889 else
890 e = NULL;
891 if (!e)
892 e = xcalloc(1, sizeof(struct attr_stack));
893 push_stack(stack, e, NULL, 0);
894 }
895
896 static void prepare_attr_stack(const char *path, int dirlen,
897 struct attr_stack **stack)
898 {
899 struct attr_stack *info;
900 struct strbuf pathbuf = STRBUF_INIT;
901
902 /*
903 * At the bottom of the attribute stack is the built-in
904 * set of attribute definitions, followed by the contents
905 * of $(prefix)/etc/gitattributes and a file specified by
906 * core.attributesfile. Then, contents from
907 * .gitattribute files from directories closer to the
908 * root to the ones in deeper directories are pushed
909 * to the stack. Finally, at the very top of the stack
910 * we always keep the contents of $GIT_DIR/info/attributes.
911 *
912 * When checking, we use entries from near the top of the
913 * stack, preferring $GIT_DIR/info/attributes, then
914 * .gitattributes in deeper directories to shallower ones,
915 * and finally use the built-in set as the default.
916 */
917 bootstrap_attr_stack(stack);
918
919 /*
920 * Pop the "info" one that is always at the top of the stack.
921 */
922 info = *stack;
923 *stack = info->prev;
924
925 /*
926 * Pop the ones from directories that are not the prefix of
927 * the path we are checking. Break out of the loop when we see
928 * the root one (whose origin is an empty string "") or the builtin
929 * one (whose origin is NULL) without popping it.
930 */
931 while ((*stack)->origin) {
932 int namelen = (*stack)->originlen;
933 struct attr_stack *elem;
934
935 elem = *stack;
936 if (namelen <= dirlen &&
937 !strncmp(elem->origin, path, namelen) &&
938 (!namelen || path[namelen] == '/'))
939 break;
940
941 debug_pop(elem);
942 *stack = elem->prev;
943 attr_stack_free(elem);
944 }
945
946 /*
947 * bootstrap_attr_stack() should have added, and the
948 * above loop should have stopped before popping, the
949 * root element whose attr_stack->origin is set to an
950 * empty string.
951 */
952 assert((*stack)->origin);
953
954 strbuf_addstr(&pathbuf, (*stack)->origin);
955 /* Build up to the directory 'path' is in */
956 while (pathbuf.len < dirlen) {
957 size_t len = pathbuf.len;
958 struct attr_stack *next;
959 char *origin;
960
961 /* Skip path-separator */
962 if (len < dirlen && is_dir_sep(path[len]))
963 len++;
964 /* Find the end of the next component */
965 while (len < dirlen && !is_dir_sep(path[len]))
966 len++;
967
968 if (pathbuf.len > 0)
969 strbuf_addch(&pathbuf, '/');
970 strbuf_add(&pathbuf, path + pathbuf.len, (len - pathbuf.len));
971 strbuf_addf(&pathbuf, "/%s", GITATTRIBUTES_FILE);
972
973 next = read_attr(pathbuf.buf, 0);
974
975 /* reset the pathbuf to not include "/.gitattributes" */
976 strbuf_setlen(&pathbuf, len);
977
978 origin = xstrdup(pathbuf.buf);
979 push_stack(stack, next, origin, len);
980 }
981
982 /*
983 * Finally push the "info" one at the top of the stack.
984 */
985 push_stack(stack, info, NULL, 0);
986
987 strbuf_release(&pathbuf);
988 }
989
990 static int path_matches(const char *pathname, int pathlen,
991 int basename_offset,
992 const struct pattern *pat,
993 const char *base, int baselen)
994 {
995 const char *pattern = pat->pattern;
996 int prefix = pat->nowildcardlen;
997 int isdir = (pathlen && pathname[pathlen - 1] == '/');
998
999 if ((pat->flags & EXC_FLAG_MUSTBEDIR) && !isdir)
1000 return 0;
1001
1002 if (pat->flags & EXC_FLAG_NODIR) {
1003 return match_basename(pathname + basename_offset,
1004 pathlen - basename_offset - isdir,
1005 pattern, prefix,
1006 pat->patternlen, pat->flags);
1007 }
1008 return match_pathname(pathname, pathlen - isdir,
1009 base, baselen,
1010 pattern, prefix, pat->patternlen, pat->flags);
1011 }
1012
1013 static int macroexpand_one(struct all_attrs_item *all_attrs, int nr, int rem);
1014
1015 static int fill_one(const char *what, struct all_attrs_item *all_attrs,
1016 const struct match_attr *a, int rem)
1017 {
1018 int i;
1019
1020 for (i = a->num_attr - 1; rem > 0 && i >= 0; i--) {
1021 const struct git_attr *attr = a->state[i].attr;
1022 const char **n = &(all_attrs[attr->attr_nr].value);
1023 const char *v = a->state[i].setto;
1024
1025 if (*n == ATTR__UNKNOWN) {
1026 debug_set(what,
1027 a->is_macro ? a->u.attr->name : a->u.pat.pattern,
1028 attr, v);
1029 *n = v;
1030 rem--;
1031 rem = macroexpand_one(all_attrs, attr->attr_nr, rem);
1032 }
1033 }
1034 return rem;
1035 }
1036
1037 static int fill(const char *path, int pathlen, int basename_offset,
1038 const struct attr_stack *stack,
1039 struct all_attrs_item *all_attrs, int rem)
1040 {
1041 for (; rem > 0 && stack; stack = stack->prev) {
1042 int i;
1043 const char *base = stack->origin ? stack->origin : "";
1044
1045 for (i = stack->num_matches - 1; 0 < rem && 0 <= i; i--) {
1046 const struct match_attr *a = stack->attrs[i];
1047 if (a->is_macro)
1048 continue;
1049 if (path_matches(path, pathlen, basename_offset,
1050 &a->u.pat, base, stack->originlen))
1051 rem = fill_one("fill", all_attrs, a, rem);
1052 }
1053 }
1054
1055 return rem;
1056 }
1057
1058 static int macroexpand_one(struct all_attrs_item *all_attrs, int nr, int rem)
1059 {
1060 const struct all_attrs_item *item = &all_attrs[nr];
1061
1062 if (item->macro && item->value == ATTR__TRUE)
1063 return fill_one("expand", all_attrs, item->macro, rem);
1064 else
1065 return rem;
1066 }
1067
1068 /*
1069 * Marks the attributes which are macros based on the attribute stack.
1070 * This prevents having to search through the attribute stack each time
1071 * a macro needs to be expanded during the fill stage.
1072 */
1073 static void determine_macros(struct all_attrs_item *all_attrs,
1074 const struct attr_stack *stack)
1075 {
1076 for (; stack; stack = stack->prev) {
1077 int i;
1078 for (i = stack->num_matches - 1; i >= 0; i--) {
1079 const struct match_attr *ma = stack->attrs[i];
1080 if (ma->is_macro) {
1081 int n = ma->u.attr->attr_nr;
1082 if (!all_attrs[n].macro) {
1083 all_attrs[n].macro = ma;
1084 }
1085 }
1086 }
1087 }
1088 }
1089
1090 /*
1091 * Collect attributes for path into the array pointed to by check->all_attrs.
1092 * If check->check_nr is non-zero, only attributes in check[] are collected.
1093 * Otherwise all attributes are collected.
1094 */
1095 static void collect_some_attrs(const char *path, struct attr_check *check)
1096 {
1097 int i, pathlen, rem, dirlen;
1098 const char *cp, *last_slash = NULL;
1099 int basename_offset;
1100
1101 for (cp = path; *cp; cp++) {
1102 if (*cp == '/' && cp[1])
1103 last_slash = cp;
1104 }
1105 pathlen = cp - path;
1106 if (last_slash) {
1107 basename_offset = last_slash + 1 - path;
1108 dirlen = last_slash - path;
1109 } else {
1110 basename_offset = 0;
1111 dirlen = 0;
1112 }
1113
1114 prepare_attr_stack(path, dirlen, &check->stack);
1115 all_attrs_init(&g_attr_hashmap, check);
1116 determine_macros(check->all_attrs, check->stack);
1117
1118 if (check->nr) {
1119 rem = 0;
1120 for (i = 0; i < check->nr; i++) {
1121 int n = check->items[i].attr->attr_nr;
1122 struct all_attrs_item *item = &check->all_attrs[n];
1123 if (item->macro) {
1124 item->value = ATTR__UNSET;
1125 rem++;
1126 }
1127 }
1128 if (rem == check->nr)
1129 return;
1130 }
1131
1132 rem = check->all_attrs_nr;
1133 fill(path, pathlen, basename_offset, check->stack, check->all_attrs, rem);
1134 }
1135
1136 int git_check_attr(const char *path, struct attr_check *check)
1137 {
1138 int i;
1139
1140 collect_some_attrs(path, check);
1141
1142 for (i = 0; i < check->nr; i++) {
1143 size_t n = check->items[i].attr->attr_nr;
1144 const char *value = check->all_attrs[n].value;
1145 if (value == ATTR__UNKNOWN)
1146 value = ATTR__UNSET;
1147 check->items[i].value = value;
1148 }
1149
1150 return 0;
1151 }
1152
1153 void git_all_attrs(const char *path, struct attr_check *check)
1154 {
1155 int i;
1156
1157 attr_check_reset(check);
1158 collect_some_attrs(path, check);
1159
1160 for (i = 0; i < check->all_attrs_nr; i++) {
1161 const char *name = check->all_attrs[i].attr->name;
1162 const char *value = check->all_attrs[i].value;
1163 struct attr_check_item *item;
1164 if (value == ATTR__UNSET || value == ATTR__UNKNOWN)
1165 continue;
1166 item = attr_check_append(check, git_attr(name));
1167 item->value = value;
1168 }
1169 }
1170
1171 void attr_start(void)
1172 {
1173 #ifndef NO_PTHREADS
1174 pthread_mutex_init(&g_attr_hashmap.mutex, NULL);
1175 pthread_mutex_init(&check_vector.mutex, NULL);
1176 #endif
1177 }