Merge branch 'ls/no-double-utf8-author-name' of ../git-gui into ls/git-gui-no-double...
[git/git.git] / tree-walk.c
1 #include "cache.h"
2 #include "tree-walk.h"
3 #include "unpack-trees.h"
4 #include "dir.h"
5 #include "tree.h"
6 #include "pathspec.h"
7
8 static const char *get_mode(const char *str, unsigned int *modep)
9 {
10 unsigned char c;
11 unsigned int mode = 0;
12
13 if (*str == ' ')
14 return NULL;
15
16 while ((c = *str++) != ' ') {
17 if (c < '0' || c > '7')
18 return NULL;
19 mode = (mode << 3) + (c - '0');
20 }
21 *modep = mode;
22 return str;
23 }
24
25 static int decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size, struct strbuf *err)
26 {
27 const char *path;
28 unsigned int mode, len;
29
30 if (size < 23 || buf[size - 21]) {
31 strbuf_addstr(err, _("too-short tree object"));
32 return -1;
33 }
34
35 path = get_mode(buf, &mode);
36 if (!path) {
37 strbuf_addstr(err, _("malformed mode in tree entry"));
38 return -1;
39 }
40 if (!*path) {
41 strbuf_addstr(err, _("empty filename in tree entry"));
42 return -1;
43 }
44 len = strlen(path) + 1;
45
46 /* Initialize the descriptor entry */
47 desc->entry.path = path;
48 desc->entry.mode = canon_mode(mode);
49 desc->entry.oid = (const struct object_id *)(path + len);
50
51 return 0;
52 }
53
54 static int init_tree_desc_internal(struct tree_desc *desc, const void *buffer, unsigned long size, struct strbuf *err)
55 {
56 desc->buffer = buffer;
57 desc->size = size;
58 if (size)
59 return decode_tree_entry(desc, buffer, size, err);
60 return 0;
61 }
62
63 void init_tree_desc(struct tree_desc *desc, const void *buffer, unsigned long size)
64 {
65 struct strbuf err = STRBUF_INIT;
66 if (init_tree_desc_internal(desc, buffer, size, &err))
67 die("%s", err.buf);
68 strbuf_release(&err);
69 }
70
71 int init_tree_desc_gently(struct tree_desc *desc, const void *buffer, unsigned long size)
72 {
73 struct strbuf err = STRBUF_INIT;
74 int result = init_tree_desc_internal(desc, buffer, size, &err);
75 if (result)
76 error("%s", err.buf);
77 strbuf_release(&err);
78 return result;
79 }
80
81 void *fill_tree_descriptor(struct tree_desc *desc, const struct object_id *oid)
82 {
83 unsigned long size = 0;
84 void *buf = NULL;
85
86 if (oid) {
87 buf = read_object_with_reference(oid->hash, tree_type, &size,
88 NULL);
89 if (!buf)
90 die("unable to read tree %s", oid_to_hex(oid));
91 }
92 init_tree_desc(desc, buf, size);
93 return buf;
94 }
95
96 static void entry_clear(struct name_entry *a)
97 {
98 memset(a, 0, sizeof(*a));
99 }
100
101 static void entry_extract(struct tree_desc *t, struct name_entry *a)
102 {
103 *a = t->entry;
104 }
105
106 static int update_tree_entry_internal(struct tree_desc *desc, struct strbuf *err)
107 {
108 const void *buf = desc->buffer;
109 const unsigned char *end = desc->entry.oid->hash + 20;
110 unsigned long size = desc->size;
111 unsigned long len = end - (const unsigned char *)buf;
112
113 if (size < len)
114 die(_("too-short tree file"));
115 buf = end;
116 size -= len;
117 desc->buffer = buf;
118 desc->size = size;
119 if (size)
120 return decode_tree_entry(desc, buf, size, err);
121 return 0;
122 }
123
124 void update_tree_entry(struct tree_desc *desc)
125 {
126 struct strbuf err = STRBUF_INIT;
127 if (update_tree_entry_internal(desc, &err))
128 die("%s", err.buf);
129 strbuf_release(&err);
130 }
131
132 int update_tree_entry_gently(struct tree_desc *desc)
133 {
134 struct strbuf err = STRBUF_INIT;
135 if (update_tree_entry_internal(desc, &err)) {
136 error("%s", err.buf);
137 strbuf_release(&err);
138 /* Stop processing this tree after error */
139 desc->size = 0;
140 return -1;
141 }
142 strbuf_release(&err);
143 return 0;
144 }
145
146 int tree_entry(struct tree_desc *desc, struct name_entry *entry)
147 {
148 if (!desc->size)
149 return 0;
150
151 *entry = desc->entry;
152 update_tree_entry(desc);
153 return 1;
154 }
155
156 int tree_entry_gently(struct tree_desc *desc, struct name_entry *entry)
157 {
158 if (!desc->size)
159 return 0;
160
161 *entry = desc->entry;
162 if (update_tree_entry_gently(desc))
163 return 0;
164 return 1;
165 }
166
167 void setup_traverse_info(struct traverse_info *info, const char *base)
168 {
169 int pathlen = strlen(base);
170 static struct traverse_info dummy;
171
172 memset(info, 0, sizeof(*info));
173 if (pathlen && base[pathlen-1] == '/')
174 pathlen--;
175 info->pathlen = pathlen ? pathlen + 1 : 0;
176 info->name.path = base;
177 info->name.oid = (void *)(base + pathlen + 1);
178 if (pathlen)
179 info->prev = &dummy;
180 }
181
182 char *make_traverse_path(char *path, const struct traverse_info *info, const struct name_entry *n)
183 {
184 int len = tree_entry_len(n);
185 int pathlen = info->pathlen;
186
187 path[pathlen + len] = 0;
188 for (;;) {
189 memcpy(path + pathlen, n->path, len);
190 if (!pathlen)
191 break;
192 path[--pathlen] = '/';
193 n = &info->name;
194 len = tree_entry_len(n);
195 info = info->prev;
196 pathlen -= len;
197 }
198 return path;
199 }
200
201 struct tree_desc_skip {
202 struct tree_desc_skip *prev;
203 const void *ptr;
204 };
205
206 struct tree_desc_x {
207 struct tree_desc d;
208 struct tree_desc_skip *skip;
209 };
210
211 static int check_entry_match(const char *a, int a_len, const char *b, int b_len)
212 {
213 /*
214 * The caller wants to pick *a* from a tree or nothing.
215 * We are looking at *b* in a tree.
216 *
217 * (0) If a and b are the same name, we are trivially happy.
218 *
219 * There are three possibilities where *a* could be hiding
220 * behind *b*.
221 *
222 * (1) *a* == "t", *b* == "ab" i.e. *b* sorts earlier than *a* no
223 * matter what.
224 * (2) *a* == "t", *b* == "t-2" and "t" is a subtree in the tree;
225 * (3) *a* == "t-2", *b* == "t" and "t-2" is a blob in the tree.
226 *
227 * Otherwise we know *a* won't appear in the tree without
228 * scanning further.
229 */
230
231 int cmp = name_compare(a, a_len, b, b_len);
232
233 /* Most common case first -- reading sync'd trees */
234 if (!cmp)
235 return cmp;
236
237 if (0 < cmp) {
238 /* a comes after b; it does not matter if it is case (3)
239 if (b_len < a_len && !memcmp(a, b, b_len) && a[b_len] < '/')
240 return 1;
241 */
242 return 1; /* keep looking */
243 }
244
245 /* b comes after a; are we looking at case (2)? */
246 if (a_len < b_len && !memcmp(a, b, a_len) && b[a_len] < '/')
247 return 1; /* keep looking */
248
249 return -1; /* a cannot appear in the tree */
250 }
251
252 /*
253 * From the extended tree_desc, extract the first name entry, while
254 * paying attention to the candidate "first" name. Most importantly,
255 * when looking for an entry, if there are entries that sorts earlier
256 * in the tree object representation than that name, skip them and
257 * process the named entry first. We will remember that we haven't
258 * processed the first entry yet, and in the later call skip the
259 * entry we processed early when update_extended_entry() is called.
260 *
261 * E.g. if the underlying tree object has these entries:
262 *
263 * blob "t-1"
264 * blob "t-2"
265 * tree "t"
266 * blob "t=1"
267 *
268 * and the "first" asks for "t", remember that we still need to
269 * process "t-1" and "t-2" but extract "t". After processing the
270 * entry "t" from this call, the caller will let us know by calling
271 * update_extended_entry() that we can remember "t" has been processed
272 * already.
273 */
274
275 static void extended_entry_extract(struct tree_desc_x *t,
276 struct name_entry *a,
277 const char *first,
278 int first_len)
279 {
280 const char *path;
281 int len;
282 struct tree_desc probe;
283 struct tree_desc_skip *skip;
284
285 /*
286 * Extract the first entry from the tree_desc, but skip the
287 * ones that we already returned in earlier rounds.
288 */
289 while (1) {
290 if (!t->d.size) {
291 entry_clear(a);
292 break; /* not found */
293 }
294 entry_extract(&t->d, a);
295 for (skip = t->skip; skip; skip = skip->prev)
296 if (a->path == skip->ptr)
297 break; /* found */
298 if (!skip)
299 break;
300 /* We have processed this entry already. */
301 update_tree_entry(&t->d);
302 }
303
304 if (!first || !a->path)
305 return;
306
307 /*
308 * The caller wants "first" from this tree, or nothing.
309 */
310 path = a->path;
311 len = tree_entry_len(a);
312 switch (check_entry_match(first, first_len, path, len)) {
313 case -1:
314 entry_clear(a);
315 case 0:
316 return;
317 default:
318 break;
319 }
320
321 /*
322 * We need to look-ahead -- we suspect that a subtree whose
323 * name is "first" may be hiding behind the current entry "path".
324 */
325 probe = t->d;
326 while (probe.size) {
327 entry_extract(&probe, a);
328 path = a->path;
329 len = tree_entry_len(a);
330 switch (check_entry_match(first, first_len, path, len)) {
331 case -1:
332 entry_clear(a);
333 case 0:
334 return;
335 default:
336 update_tree_entry(&probe);
337 break;
338 }
339 /* keep looking */
340 }
341 entry_clear(a);
342 }
343
344 static void update_extended_entry(struct tree_desc_x *t, struct name_entry *a)
345 {
346 if (t->d.entry.path == a->path) {
347 update_tree_entry(&t->d);
348 } else {
349 /* we have returned this entry early */
350 struct tree_desc_skip *skip = xmalloc(sizeof(*skip));
351 skip->ptr = a->path;
352 skip->prev = t->skip;
353 t->skip = skip;
354 }
355 }
356
357 static void free_extended_entry(struct tree_desc_x *t)
358 {
359 struct tree_desc_skip *p, *s;
360
361 for (s = t->skip; s; s = p) {
362 p = s->prev;
363 free(s);
364 }
365 }
366
367 static inline int prune_traversal(struct name_entry *e,
368 struct traverse_info *info,
369 struct strbuf *base,
370 int still_interesting)
371 {
372 if (!info->pathspec || still_interesting == 2)
373 return 2;
374 if (still_interesting < 0)
375 return still_interesting;
376 return tree_entry_interesting(e, base, 0, info->pathspec);
377 }
378
379 int traverse_trees(int n, struct tree_desc *t, struct traverse_info *info)
380 {
381 int error = 0;
382 struct name_entry *entry = xmalloc(n*sizeof(*entry));
383 int i;
384 struct tree_desc_x *tx = xcalloc(n, sizeof(*tx));
385 struct strbuf base = STRBUF_INIT;
386 int interesting = 1;
387 char *traverse_path;
388
389 for (i = 0; i < n; i++)
390 tx[i].d = t[i];
391
392 if (info->prev) {
393 strbuf_grow(&base, info->pathlen);
394 make_traverse_path(base.buf, info->prev, &info->name);
395 base.buf[info->pathlen-1] = '/';
396 strbuf_setlen(&base, info->pathlen);
397 traverse_path = xstrndup(base.buf, info->pathlen);
398 } else {
399 traverse_path = xstrndup(info->name.path, info->pathlen);
400 }
401 info->traverse_path = traverse_path;
402 for (;;) {
403 int trees_used;
404 unsigned long mask, dirmask;
405 const char *first = NULL;
406 int first_len = 0;
407 struct name_entry *e = NULL;
408 int len;
409
410 for (i = 0; i < n; i++) {
411 e = entry + i;
412 extended_entry_extract(tx + i, e, NULL, 0);
413 }
414
415 /*
416 * A tree may have "t-2" at the current location even
417 * though it may have "t" that is a subtree behind it,
418 * and another tree may return "t". We want to grab
419 * all "t" from all trees to match in such a case.
420 */
421 for (i = 0; i < n; i++) {
422 e = entry + i;
423 if (!e->path)
424 continue;
425 len = tree_entry_len(e);
426 if (!first) {
427 first = e->path;
428 first_len = len;
429 continue;
430 }
431 if (name_compare(e->path, len, first, first_len) < 0) {
432 first = e->path;
433 first_len = len;
434 }
435 }
436
437 if (first) {
438 for (i = 0; i < n; i++) {
439 e = entry + i;
440 extended_entry_extract(tx + i, e, first, first_len);
441 /* Cull the ones that are not the earliest */
442 if (!e->path)
443 continue;
444 len = tree_entry_len(e);
445 if (name_compare(e->path, len, first, first_len))
446 entry_clear(e);
447 }
448 }
449
450 /* Now we have in entry[i] the earliest name from the trees */
451 mask = 0;
452 dirmask = 0;
453 for (i = 0; i < n; i++) {
454 if (!entry[i].path)
455 continue;
456 mask |= 1ul << i;
457 if (S_ISDIR(entry[i].mode))
458 dirmask |= 1ul << i;
459 e = &entry[i];
460 }
461 if (!mask)
462 break;
463 interesting = prune_traversal(e, info, &base, interesting);
464 if (interesting < 0)
465 break;
466 if (interesting) {
467 trees_used = info->fn(n, mask, dirmask, entry, info);
468 if (trees_used < 0) {
469 error = trees_used;
470 if (!info->show_all_errors)
471 break;
472 }
473 mask &= trees_used;
474 }
475 for (i = 0; i < n; i++)
476 if (mask & (1ul << i))
477 update_extended_entry(tx + i, entry + i);
478 }
479 free(entry);
480 for (i = 0; i < n; i++)
481 free_extended_entry(tx + i);
482 free(tx);
483 free(traverse_path);
484 info->traverse_path = NULL;
485 strbuf_release(&base);
486 return error;
487 }
488
489 struct dir_state {
490 void *tree;
491 unsigned long size;
492 unsigned char sha1[20];
493 };
494
495 static int find_tree_entry(struct tree_desc *t, const char *name, unsigned char *result, unsigned *mode)
496 {
497 int namelen = strlen(name);
498 while (t->size) {
499 const char *entry;
500 const struct object_id *oid;
501 int entrylen, cmp;
502
503 oid = tree_entry_extract(t, &entry, mode);
504 entrylen = tree_entry_len(&t->entry);
505 update_tree_entry(t);
506 if (entrylen > namelen)
507 continue;
508 cmp = memcmp(name, entry, entrylen);
509 if (cmp > 0)
510 continue;
511 if (cmp < 0)
512 break;
513 if (entrylen == namelen) {
514 hashcpy(result, oid->hash);
515 return 0;
516 }
517 if (name[entrylen] != '/')
518 continue;
519 if (!S_ISDIR(*mode))
520 break;
521 if (++entrylen == namelen) {
522 hashcpy(result, oid->hash);
523 return 0;
524 }
525 return get_tree_entry(oid->hash, name + entrylen, result, mode);
526 }
527 return -1;
528 }
529
530 int get_tree_entry(const unsigned char *tree_sha1, const char *name, unsigned char *sha1, unsigned *mode)
531 {
532 int retval;
533 void *tree;
534 unsigned long size;
535 unsigned char root[20];
536
537 tree = read_object_with_reference(tree_sha1, tree_type, &size, root);
538 if (!tree)
539 return -1;
540
541 if (name[0] == '\0') {
542 hashcpy(sha1, root);
543 free(tree);
544 return 0;
545 }
546
547 if (!size) {
548 retval = -1;
549 } else {
550 struct tree_desc t;
551 init_tree_desc(&t, tree, size);
552 retval = find_tree_entry(&t, name, sha1, mode);
553 }
554 free(tree);
555 return retval;
556 }
557
558 /*
559 * This is Linux's built-in max for the number of symlinks to follow.
560 * That limit, of course, does not affect git, but it's a reasonable
561 * choice.
562 */
563 #define GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS 40
564
565 /**
566 * Find a tree entry by following symlinks in tree_sha (which is
567 * assumed to be the root of the repository). In the event that a
568 * symlink points outside the repository (e.g. a link to /foo or a
569 * root-level link to ../foo), the portion of the link which is
570 * outside the repository will be returned in result_path, and *mode
571 * will be set to 0. It is assumed that result_path is uninitialized.
572 * If there are no symlinks, or the end result of the symlink chain
573 * points to an object inside the repository, result will be filled in
574 * with the sha1 of the found object, and *mode will hold the mode of
575 * the object.
576 *
577 * See the code for enum follow_symlink_result for a description of
578 * the return values.
579 */
580 enum follow_symlinks_result get_tree_entry_follow_symlinks(unsigned char *tree_sha1, const char *name, unsigned char *result, struct strbuf *result_path, unsigned *mode)
581 {
582 int retval = MISSING_OBJECT;
583 struct dir_state *parents = NULL;
584 size_t parents_alloc = 0;
585 size_t i, parents_nr = 0;
586 unsigned char current_tree_sha1[20];
587 struct strbuf namebuf = STRBUF_INIT;
588 struct tree_desc t;
589 int follows_remaining = GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS;
590
591 init_tree_desc(&t, NULL, 0UL);
592 strbuf_addstr(&namebuf, name);
593 hashcpy(current_tree_sha1, tree_sha1);
594
595 while (1) {
596 int find_result;
597 char *first_slash;
598 char *remainder = NULL;
599
600 if (!t.buffer) {
601 void *tree;
602 unsigned char root[20];
603 unsigned long size;
604 tree = read_object_with_reference(current_tree_sha1,
605 tree_type, &size,
606 root);
607 if (!tree)
608 goto done;
609
610 ALLOC_GROW(parents, parents_nr + 1, parents_alloc);
611 parents[parents_nr].tree = tree;
612 parents[parents_nr].size = size;
613 hashcpy(parents[parents_nr].sha1, root);
614 parents_nr++;
615
616 if (namebuf.buf[0] == '\0') {
617 hashcpy(result, root);
618 retval = FOUND;
619 goto done;
620 }
621
622 if (!size)
623 goto done;
624
625 /* descend */
626 init_tree_desc(&t, tree, size);
627 }
628
629 /* Handle symlinks to e.g. a//b by removing leading slashes */
630 while (namebuf.buf[0] == '/') {
631 strbuf_remove(&namebuf, 0, 1);
632 }
633
634 /* Split namebuf into a first component and a remainder */
635 if ((first_slash = strchr(namebuf.buf, '/'))) {
636 *first_slash = 0;
637 remainder = first_slash + 1;
638 }
639
640 if (!strcmp(namebuf.buf, "..")) {
641 struct dir_state *parent;
642 /*
643 * We could end up with .. in the namebuf if it
644 * appears in a symlink.
645 */
646
647 if (parents_nr == 1) {
648 if (remainder)
649 *first_slash = '/';
650 strbuf_add(result_path, namebuf.buf,
651 namebuf.len);
652 *mode = 0;
653 retval = FOUND;
654 goto done;
655 }
656 parent = &parents[parents_nr - 1];
657 free(parent->tree);
658 parents_nr--;
659 parent = &parents[parents_nr - 1];
660 init_tree_desc(&t, parent->tree, parent->size);
661 strbuf_remove(&namebuf, 0, remainder ? 3 : 2);
662 continue;
663 }
664
665 /* We could end up here via a symlink to dir/.. */
666 if (namebuf.buf[0] == '\0') {
667 hashcpy(result, parents[parents_nr - 1].sha1);
668 retval = FOUND;
669 goto done;
670 }
671
672 /* Look up the first (or only) path component in the tree. */
673 find_result = find_tree_entry(&t, namebuf.buf,
674 current_tree_sha1, mode);
675 if (find_result) {
676 goto done;
677 }
678
679 if (S_ISDIR(*mode)) {
680 if (!remainder) {
681 hashcpy(result, current_tree_sha1);
682 retval = FOUND;
683 goto done;
684 }
685 /* Descend the tree */
686 t.buffer = NULL;
687 strbuf_remove(&namebuf, 0,
688 1 + first_slash - namebuf.buf);
689 } else if (S_ISREG(*mode)) {
690 if (!remainder) {
691 hashcpy(result, current_tree_sha1);
692 retval = FOUND;
693 } else {
694 retval = NOT_DIR;
695 }
696 goto done;
697 } else if (S_ISLNK(*mode)) {
698 /* Follow a symlink */
699 unsigned long link_len;
700 size_t len;
701 char *contents, *contents_start;
702 struct dir_state *parent;
703 enum object_type type;
704
705 if (follows_remaining-- == 0) {
706 /* Too many symlinks followed */
707 retval = SYMLINK_LOOP;
708 goto done;
709 }
710
711 /*
712 * At this point, we have followed at a least
713 * one symlink, so on error we need to report this.
714 */
715 retval = DANGLING_SYMLINK;
716
717 contents = read_sha1_file(current_tree_sha1, &type,
718 &link_len);
719
720 if (!contents)
721 goto done;
722
723 if (contents[0] == '/') {
724 strbuf_addstr(result_path, contents);
725 free(contents);
726 *mode = 0;
727 retval = FOUND;
728 goto done;
729 }
730
731 if (remainder)
732 len = first_slash - namebuf.buf;
733 else
734 len = namebuf.len;
735
736 contents_start = contents;
737
738 parent = &parents[parents_nr - 1];
739 init_tree_desc(&t, parent->tree, parent->size);
740 strbuf_splice(&namebuf, 0, len,
741 contents_start, link_len);
742 if (remainder)
743 namebuf.buf[link_len] = '/';
744 free(contents);
745 }
746 }
747 done:
748 for (i = 0; i < parents_nr; i++)
749 free(parents[i].tree);
750 free(parents);
751
752 strbuf_release(&namebuf);
753 return retval;
754 }
755
756 static int match_entry(const struct pathspec_item *item,
757 const struct name_entry *entry, int pathlen,
758 const char *match, int matchlen,
759 enum interesting *never_interesting)
760 {
761 int m = -1; /* signals that we haven't called strncmp() */
762
763 if (item->magic & PATHSPEC_ICASE)
764 /*
765 * "Never interesting" trick requires exact
766 * matching. We could do something clever with inexact
767 * matching, but it's trickier (and not to forget that
768 * strcasecmp is locale-dependent, at least in
769 * glibc). Just disable it for now. It can't be worse
770 * than the wildcard's codepath of '[Tt][Hi][Is][Ss]'
771 * pattern.
772 */
773 *never_interesting = entry_not_interesting;
774 else if (*never_interesting != entry_not_interesting) {
775 /*
776 * We have not seen any match that sorts later
777 * than the current path.
778 */
779
780 /*
781 * Does match sort strictly earlier than path
782 * with their common parts?
783 */
784 m = strncmp(match, entry->path,
785 (matchlen < pathlen) ? matchlen : pathlen);
786 if (m < 0)
787 return 0;
788
789 /*
790 * If we come here even once, that means there is at
791 * least one pathspec that would sort equal to or
792 * later than the path we are currently looking at.
793 * In other words, if we have never reached this point
794 * after iterating all pathspecs, it means all
795 * pathspecs are either outside of base, or inside the
796 * base but sorts strictly earlier than the current
797 * one. In either case, they will never match the
798 * subsequent entries. In such a case, we initialized
799 * the variable to -1 and that is what will be
800 * returned, allowing the caller to terminate early.
801 */
802 *never_interesting = entry_not_interesting;
803 }
804
805 if (pathlen > matchlen)
806 return 0;
807
808 if (matchlen > pathlen) {
809 if (match[pathlen] != '/')
810 return 0;
811 if (!S_ISDIR(entry->mode) && !S_ISGITLINK(entry->mode))
812 return 0;
813 }
814
815 if (m == -1)
816 /*
817 * we cheated and did not do strncmp(), so we do
818 * that here.
819 */
820 m = ps_strncmp(item, match, entry->path, pathlen);
821
822 /*
823 * If common part matched earlier then it is a hit,
824 * because we rejected the case where path is not a
825 * leading directory and is shorter than match.
826 */
827 if (!m)
828 /*
829 * match_entry does not check if the prefix part is
830 * matched case-sensitively. If the entry is a
831 * directory and part of prefix, it'll be rematched
832 * eventually by basecmp with special treatment for
833 * the prefix.
834 */
835 return 1;
836
837 return 0;
838 }
839
840 /* :(icase)-aware string compare */
841 static int basecmp(const struct pathspec_item *item,
842 const char *base, const char *match, int len)
843 {
844 if (item->magic & PATHSPEC_ICASE) {
845 int ret, n = len > item->prefix ? item->prefix : len;
846 ret = strncmp(base, match, n);
847 if (ret)
848 return ret;
849 base += n;
850 match += n;
851 len -= n;
852 }
853 return ps_strncmp(item, base, match, len);
854 }
855
856 static int match_dir_prefix(const struct pathspec_item *item,
857 const char *base,
858 const char *match, int matchlen)
859 {
860 if (basecmp(item, base, match, matchlen))
861 return 0;
862
863 /*
864 * If the base is a subdirectory of a path which
865 * was specified, all of them are interesting.
866 */
867 if (!matchlen ||
868 base[matchlen] == '/' ||
869 match[matchlen - 1] == '/')
870 return 1;
871
872 /* Just a random prefix match */
873 return 0;
874 }
875
876 /*
877 * Perform matching on the leading non-wildcard part of
878 * pathspec. item->nowildcard_len must be greater than zero. Return
879 * non-zero if base is matched.
880 */
881 static int match_wildcard_base(const struct pathspec_item *item,
882 const char *base, int baselen,
883 int *matched)
884 {
885 const char *match = item->match;
886 /* the wildcard part is not considered in this function */
887 int matchlen = item->nowildcard_len;
888
889 if (baselen) {
890 int dirlen;
891 /*
892 * Return early if base is longer than the
893 * non-wildcard part but it does not match.
894 */
895 if (baselen >= matchlen) {
896 *matched = matchlen;
897 return !basecmp(item, base, match, matchlen);
898 }
899
900 dirlen = matchlen;
901 while (dirlen && match[dirlen - 1] != '/')
902 dirlen--;
903
904 /*
905 * Return early if base is shorter than the
906 * non-wildcard part but it does not match. Note that
907 * base ends with '/' so we are sure it really matches
908 * directory
909 */
910 if (basecmp(item, base, match, baselen))
911 return 0;
912 *matched = baselen;
913 } else
914 *matched = 0;
915 /*
916 * we could have checked entry against the non-wildcard part
917 * that is not in base and does similar never_interesting
918 * optimization as in match_entry. For now just be happy with
919 * base comparison.
920 */
921 return entry_interesting;
922 }
923
924 /*
925 * Is a tree entry interesting given the pathspec we have?
926 *
927 * Pre-condition: either baselen == base_offset (i.e. empty path)
928 * or base[baselen-1] == '/' (i.e. with trailing slash).
929 */
930 static enum interesting do_match(const struct name_entry *entry,
931 struct strbuf *base, int base_offset,
932 const struct pathspec *ps,
933 int exclude)
934 {
935 int i;
936 int pathlen, baselen = base->len - base_offset;
937 enum interesting never_interesting = ps->has_wildcard ?
938 entry_not_interesting : all_entries_not_interesting;
939
940 GUARD_PATHSPEC(ps,
941 PATHSPEC_FROMTOP |
942 PATHSPEC_MAXDEPTH |
943 PATHSPEC_LITERAL |
944 PATHSPEC_GLOB |
945 PATHSPEC_ICASE |
946 PATHSPEC_EXCLUDE);
947
948 if (!ps->nr) {
949 if (!ps->recursive ||
950 !(ps->magic & PATHSPEC_MAXDEPTH) ||
951 ps->max_depth == -1)
952 return all_entries_interesting;
953 return within_depth(base->buf + base_offset, baselen,
954 !!S_ISDIR(entry->mode),
955 ps->max_depth) ?
956 entry_interesting : entry_not_interesting;
957 }
958
959 pathlen = tree_entry_len(entry);
960
961 for (i = ps->nr - 1; i >= 0; i--) {
962 const struct pathspec_item *item = ps->items+i;
963 const char *match = item->match;
964 const char *base_str = base->buf + base_offset;
965 int matchlen = item->len, matched = 0;
966
967 if ((!exclude && item->magic & PATHSPEC_EXCLUDE) ||
968 ( exclude && !(item->magic & PATHSPEC_EXCLUDE)))
969 continue;
970
971 if (baselen >= matchlen) {
972 /* If it doesn't match, move along... */
973 if (!match_dir_prefix(item, base_str, match, matchlen))
974 goto match_wildcards;
975
976 if (!ps->recursive ||
977 !(ps->magic & PATHSPEC_MAXDEPTH) ||
978 ps->max_depth == -1)
979 return all_entries_interesting;
980
981 return within_depth(base_str + matchlen + 1,
982 baselen - matchlen - 1,
983 !!S_ISDIR(entry->mode),
984 ps->max_depth) ?
985 entry_interesting : entry_not_interesting;
986 }
987
988 /* Either there must be no base, or the base must match. */
989 if (baselen == 0 || !basecmp(item, base_str, match, baselen)) {
990 if (match_entry(item, entry, pathlen,
991 match + baselen, matchlen - baselen,
992 &never_interesting))
993 return entry_interesting;
994
995 if (item->nowildcard_len < item->len) {
996 if (!git_fnmatch(item, match + baselen, entry->path,
997 item->nowildcard_len - baselen))
998 return entry_interesting;
999
1000 /*
1001 * Match all directories. We'll try to
1002 * match files later on.
1003 */
1004 if (ps->recursive && S_ISDIR(entry->mode))
1005 return entry_interesting;
1006
1007 /*
1008 * When matching against submodules with
1009 * wildcard characters, ensure that the entry
1010 * at least matches up to the first wild
1011 * character. More accurate matching can then
1012 * be performed in the submodule itself.
1013 */
1014 if (ps->recursive && S_ISGITLINK(entry->mode) &&
1015 !ps_strncmp(item, match + baselen,
1016 entry->path,
1017 item->nowildcard_len - baselen))
1018 return entry_interesting;
1019 }
1020
1021 continue;
1022 }
1023
1024 match_wildcards:
1025 if (item->nowildcard_len == item->len)
1026 continue;
1027
1028 if (item->nowildcard_len &&
1029 !match_wildcard_base(item, base_str, baselen, &matched))
1030 continue;
1031
1032 /*
1033 * Concatenate base and entry->path into one and do
1034 * fnmatch() on it.
1035 *
1036 * While we could avoid concatenation in certain cases
1037 * [1], which saves a memcpy and potentially a
1038 * realloc, it turns out not worth it. Measurement on
1039 * linux-2.6 does not show any clear improvements,
1040 * partly because of the nowildcard_len optimization
1041 * in git_fnmatch(). Avoid micro-optimizations here.
1042 *
1043 * [1] if match_wildcard_base() says the base
1044 * directory is already matched, we only need to match
1045 * the rest, which is shorter so _in theory_ faster.
1046 */
1047
1048 strbuf_add(base, entry->path, pathlen);
1049
1050 if (!git_fnmatch(item, match, base->buf + base_offset,
1051 item->nowildcard_len)) {
1052 strbuf_setlen(base, base_offset + baselen);
1053 return entry_interesting;
1054 }
1055
1056 /*
1057 * When matching against submodules with
1058 * wildcard characters, ensure that the entry
1059 * at least matches up to the first wild
1060 * character. More accurate matching can then
1061 * be performed in the submodule itself.
1062 */
1063 if (ps->recursive && S_ISGITLINK(entry->mode) &&
1064 !ps_strncmp(item, match, base->buf + base_offset,
1065 item->nowildcard_len)) {
1066 strbuf_setlen(base, base_offset + baselen);
1067 return entry_interesting;
1068 }
1069
1070 strbuf_setlen(base, base_offset + baselen);
1071
1072 /*
1073 * Match all directories. We'll try to match files
1074 * later on.
1075 * max_depth is ignored but we may consider support it
1076 * in future, see
1077 * https://public-inbox.org/git/7vmxo5l2g4.fsf@alter.siamese.dyndns.org/
1078 */
1079 if (ps->recursive && S_ISDIR(entry->mode))
1080 return entry_interesting;
1081 }
1082 return never_interesting; /* No matches */
1083 }
1084
1085 /*
1086 * Is a tree entry interesting given the pathspec we have?
1087 *
1088 * Pre-condition: either baselen == base_offset (i.e. empty path)
1089 * or base[baselen-1] == '/' (i.e. with trailing slash).
1090 */
1091 enum interesting tree_entry_interesting(const struct name_entry *entry,
1092 struct strbuf *base, int base_offset,
1093 const struct pathspec *ps)
1094 {
1095 enum interesting positive, negative;
1096 positive = do_match(entry, base, base_offset, ps, 0);
1097
1098 /*
1099 * case | entry | positive | negative | result
1100 * -----+-------+----------+----------+-------
1101 * 1 | file | -1 | -1..2 | -1
1102 * 2 | file | 0 | -1..2 | 0
1103 * 3 | file | 1 | -1 | 1
1104 * 4 | file | 1 | 0 | 1
1105 * 5 | file | 1 | 1 | 0
1106 * 6 | file | 1 | 2 | 0
1107 * 7 | file | 2 | -1 | 2
1108 * 8 | file | 2 | 0 | 2
1109 * 9 | file | 2 | 1 | 0
1110 * 10 | file | 2 | 2 | -1
1111 * -----+-------+----------+----------+-------
1112 * 11 | dir | -1 | -1..2 | -1
1113 * 12 | dir | 0 | -1..2 | 0
1114 * 13 | dir | 1 | -1 | 1
1115 * 14 | dir | 1 | 0 | 1
1116 * 15 | dir | 1 | 1 | 1 (*)
1117 * 16 | dir | 1 | 2 | 0
1118 * 17 | dir | 2 | -1 | 2
1119 * 18 | dir | 2 | 0 | 2
1120 * 19 | dir | 2 | 1 | 1 (*)
1121 * 20 | dir | 2 | 2 | -1
1122 *
1123 * (*) An exclude pattern interested in a directory does not
1124 * necessarily mean it will exclude all of the directory. In
1125 * wildcard case, it can't decide until looking at individual
1126 * files inside. So don't write such directories off yet.
1127 */
1128
1129 if (!(ps->magic & PATHSPEC_EXCLUDE) ||
1130 positive <= entry_not_interesting) /* #1, #2, #11, #12 */
1131 return positive;
1132
1133 negative = do_match(entry, base, base_offset, ps, 1);
1134
1135 /* #3, #4, #7, #8, #13, #14, #17, #18 */
1136 if (negative <= entry_not_interesting)
1137 return positive;
1138
1139 /* #15, #19 */
1140 if (S_ISDIR(entry->mode) &&
1141 positive >= entry_interesting &&
1142 negative == entry_interesting)
1143 return entry_interesting;
1144
1145 if ((positive == entry_interesting &&
1146 negative >= entry_interesting) || /* #5, #6, #16 */
1147 (positive == all_entries_interesting &&
1148 negative == entry_interesting)) /* #9 */
1149 return entry_not_interesting;
1150
1151 return all_entries_not_interesting; /* #10, #20 */
1152 }