The sixth batch
[git.git] / tree-walk.c
blobc20b62f49e4709253d71e18eea0a2472809d8e9d
1 #include "cache.h"
2 #include "tree-walk.h"
3 #include "unpack-trees.h"
4 #include "dir.h"
5 #include "object-store.h"
6 #include "tree.h"
7 #include "pathspec.h"
9 static const char *get_mode(const char *str, unsigned int *modep)
11 unsigned char c;
12 unsigned int mode = 0;
14 if (*str == ' ')
15 return NULL;
17 while ((c = *str++) != ' ') {
18 if (c < '0' || c > '7')
19 return NULL;
20 mode = (mode << 3) + (c - '0');
22 *modep = mode;
23 return str;
26 static int decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size, struct strbuf *err)
28 const char *path;
29 unsigned int mode, len;
30 const unsigned hashsz = the_hash_algo->rawsz;
32 if (size < hashsz + 3 || buf[size - (hashsz + 1)]) {
33 strbuf_addstr(err, _("too-short tree object"));
34 return -1;
37 path = get_mode(buf, &mode);
38 if (!path) {
39 strbuf_addstr(err, _("malformed mode in tree entry"));
40 return -1;
42 if (!*path) {
43 strbuf_addstr(err, _("empty filename in tree entry"));
44 return -1;
46 len = strlen(path) + 1;
48 /* Initialize the descriptor entry */
49 desc->entry.path = path;
50 desc->entry.mode = canon_mode(mode);
51 desc->entry.pathlen = len - 1;
52 hashcpy(desc->entry.oid.hash, (const unsigned char *)path + len);
54 return 0;
57 static int init_tree_desc_internal(struct tree_desc *desc, const void *buffer, unsigned long size, struct strbuf *err)
59 desc->buffer = buffer;
60 desc->size = size;
61 if (size)
62 return decode_tree_entry(desc, buffer, size, err);
63 return 0;
66 void init_tree_desc(struct tree_desc *desc, const void *buffer, unsigned long size)
68 struct strbuf err = STRBUF_INIT;
69 if (init_tree_desc_internal(desc, buffer, size, &err))
70 die("%s", err.buf);
71 strbuf_release(&err);
74 int init_tree_desc_gently(struct tree_desc *desc, const void *buffer, unsigned long size)
76 struct strbuf err = STRBUF_INIT;
77 int result = init_tree_desc_internal(desc, buffer, size, &err);
78 if (result)
79 error("%s", err.buf);
80 strbuf_release(&err);
81 return result;
84 void *fill_tree_descriptor(struct repository *r,
85 struct tree_desc *desc,
86 const struct object_id *oid)
88 unsigned long size = 0;
89 void *buf = NULL;
91 if (oid) {
92 buf = read_object_with_reference(r, oid, tree_type, &size, NULL);
93 if (!buf)
94 die("unable to read tree %s", oid_to_hex(oid));
96 init_tree_desc(desc, buf, size);
97 return buf;
100 static void entry_clear(struct name_entry *a)
102 memset(a, 0, sizeof(*a));
105 static void entry_extract(struct tree_desc *t, struct name_entry *a)
107 *a = t->entry;
110 static int update_tree_entry_internal(struct tree_desc *desc, struct strbuf *err)
112 const void *buf = desc->buffer;
113 const unsigned char *end = (const unsigned char *)desc->entry.path + desc->entry.pathlen + 1 + the_hash_algo->rawsz;
114 unsigned long size = desc->size;
115 unsigned long len = end - (const unsigned char *)buf;
117 if (size < len)
118 die(_("too-short tree file"));
119 buf = end;
120 size -= len;
121 desc->buffer = buf;
122 desc->size = size;
123 if (size)
124 return decode_tree_entry(desc, buf, size, err);
125 return 0;
128 void update_tree_entry(struct tree_desc *desc)
130 struct strbuf err = STRBUF_INIT;
131 if (update_tree_entry_internal(desc, &err))
132 die("%s", err.buf);
133 strbuf_release(&err);
136 int update_tree_entry_gently(struct tree_desc *desc)
138 struct strbuf err = STRBUF_INIT;
139 if (update_tree_entry_internal(desc, &err)) {
140 error("%s", err.buf);
141 strbuf_release(&err);
142 /* Stop processing this tree after error */
143 desc->size = 0;
144 return -1;
146 strbuf_release(&err);
147 return 0;
150 int tree_entry(struct tree_desc *desc, struct name_entry *entry)
152 if (!desc->size)
153 return 0;
155 *entry = desc->entry;
156 update_tree_entry(desc);
157 return 1;
160 int tree_entry_gently(struct tree_desc *desc, struct name_entry *entry)
162 if (!desc->size)
163 return 0;
165 *entry = desc->entry;
166 if (update_tree_entry_gently(desc))
167 return 0;
168 return 1;
171 void setup_traverse_info(struct traverse_info *info, const char *base)
173 int pathlen = strlen(base);
174 static struct traverse_info dummy;
176 memset(info, 0, sizeof(*info));
177 if (pathlen && base[pathlen-1] == '/')
178 pathlen--;
179 info->pathlen = pathlen ? pathlen + 1 : 0;
180 info->name.path = base;
181 info->name.pathlen = pathlen;
182 if (pathlen) {
183 hashcpy(info->name.oid.hash, (const unsigned char *)base + pathlen + 1);
184 info->prev = &dummy;
188 char *make_traverse_path(char *path, const struct traverse_info *info, const struct name_entry *n)
190 int len = tree_entry_len(n);
191 int pathlen = info->pathlen;
193 path[pathlen + len] = 0;
194 for (;;) {
195 memcpy(path + pathlen, n->path, len);
196 if (!pathlen)
197 break;
198 path[--pathlen] = '/';
199 n = &info->name;
200 len = tree_entry_len(n);
201 info = info->prev;
202 pathlen -= len;
204 return path;
207 struct tree_desc_skip {
208 struct tree_desc_skip *prev;
209 const void *ptr;
212 struct tree_desc_x {
213 struct tree_desc d;
214 struct tree_desc_skip *skip;
217 static int check_entry_match(const char *a, int a_len, const char *b, int b_len)
220 * The caller wants to pick *a* from a tree or nothing.
221 * We are looking at *b* in a tree.
223 * (0) If a and b are the same name, we are trivially happy.
225 * There are three possibilities where *a* could be hiding
226 * behind *b*.
228 * (1) *a* == "t", *b* == "ab" i.e. *b* sorts earlier than *a* no
229 * matter what.
230 * (2) *a* == "t", *b* == "t-2" and "t" is a subtree in the tree;
231 * (3) *a* == "t-2", *b* == "t" and "t-2" is a blob in the tree.
233 * Otherwise we know *a* won't appear in the tree without
234 * scanning further.
237 int cmp = name_compare(a, a_len, b, b_len);
239 /* Most common case first -- reading sync'd trees */
240 if (!cmp)
241 return cmp;
243 if (0 < cmp) {
244 /* a comes after b; it does not matter if it is case (3)
245 if (b_len < a_len && !memcmp(a, b, b_len) && a[b_len] < '/')
246 return 1;
248 return 1; /* keep looking */
251 /* b comes after a; are we looking at case (2)? */
252 if (a_len < b_len && !memcmp(a, b, a_len) && b[a_len] < '/')
253 return 1; /* keep looking */
255 return -1; /* a cannot appear in the tree */
259 * From the extended tree_desc, extract the first name entry, while
260 * paying attention to the candidate "first" name. Most importantly,
261 * when looking for an entry, if there are entries that sorts earlier
262 * in the tree object representation than that name, skip them and
263 * process the named entry first. We will remember that we haven't
264 * processed the first entry yet, and in the later call skip the
265 * entry we processed early when update_extended_entry() is called.
267 * E.g. if the underlying tree object has these entries:
269 * blob "t-1"
270 * blob "t-2"
271 * tree "t"
272 * blob "t=1"
274 * and the "first" asks for "t", remember that we still need to
275 * process "t-1" and "t-2" but extract "t". After processing the
276 * entry "t" from this call, the caller will let us know by calling
277 * update_extended_entry() that we can remember "t" has been processed
278 * already.
281 static void extended_entry_extract(struct tree_desc_x *t,
282 struct name_entry *a,
283 const char *first,
284 int first_len)
286 const char *path;
287 int len;
288 struct tree_desc probe;
289 struct tree_desc_skip *skip;
292 * Extract the first entry from the tree_desc, but skip the
293 * ones that we already returned in earlier rounds.
295 while (1) {
296 if (!t->d.size) {
297 entry_clear(a);
298 break; /* not found */
300 entry_extract(&t->d, a);
301 for (skip = t->skip; skip; skip = skip->prev)
302 if (a->path == skip->ptr)
303 break; /* found */
304 if (!skip)
305 break;
306 /* We have processed this entry already. */
307 update_tree_entry(&t->d);
310 if (!first || !a->path)
311 return;
314 * The caller wants "first" from this tree, or nothing.
316 path = a->path;
317 len = tree_entry_len(a);
318 switch (check_entry_match(first, first_len, path, len)) {
319 case -1:
320 entry_clear(a);
321 case 0:
322 return;
323 default:
324 break;
328 * We need to look-ahead -- we suspect that a subtree whose
329 * name is "first" may be hiding behind the current entry "path".
331 probe = t->d;
332 while (probe.size) {
333 entry_extract(&probe, a);
334 path = a->path;
335 len = tree_entry_len(a);
336 switch (check_entry_match(first, first_len, path, len)) {
337 case -1:
338 entry_clear(a);
339 case 0:
340 return;
341 default:
342 update_tree_entry(&probe);
343 break;
345 /* keep looking */
347 entry_clear(a);
350 static void update_extended_entry(struct tree_desc_x *t, struct name_entry *a)
352 if (t->d.entry.path == a->path) {
353 update_tree_entry(&t->d);
354 } else {
355 /* we have returned this entry early */
356 struct tree_desc_skip *skip = xmalloc(sizeof(*skip));
357 skip->ptr = a->path;
358 skip->prev = t->skip;
359 t->skip = skip;
363 static void free_extended_entry(struct tree_desc_x *t)
365 struct tree_desc_skip *p, *s;
367 for (s = t->skip; s; s = p) {
368 p = s->prev;
369 free(s);
373 static inline int prune_traversal(struct index_state *istate,
374 struct name_entry *e,
375 struct traverse_info *info,
376 struct strbuf *base,
377 int still_interesting)
379 if (!info->pathspec || still_interesting == 2)
380 return 2;
381 if (still_interesting < 0)
382 return still_interesting;
383 return tree_entry_interesting(istate, e, base,
384 0, info->pathspec);
387 int traverse_trees(struct index_state *istate,
388 int n, struct tree_desc *t,
389 struct traverse_info *info)
391 int error = 0;
392 struct name_entry *entry = xmalloc(n*sizeof(*entry));
393 int i;
394 struct tree_desc_x *tx = xcalloc(n, sizeof(*tx));
395 struct strbuf base = STRBUF_INIT;
396 int interesting = 1;
397 char *traverse_path;
399 for (i = 0; i < n; i++)
400 tx[i].d = t[i];
402 if (info->prev) {
403 strbuf_grow(&base, info->pathlen);
404 make_traverse_path(base.buf, info->prev, &info->name);
405 base.buf[info->pathlen-1] = '/';
406 strbuf_setlen(&base, info->pathlen);
407 traverse_path = xstrndup(base.buf, info->pathlen);
408 } else {
409 traverse_path = xstrndup(info->name.path, info->pathlen);
411 info->traverse_path = traverse_path;
412 for (;;) {
413 int trees_used;
414 unsigned long mask, dirmask;
415 const char *first = NULL;
416 int first_len = 0;
417 struct name_entry *e = NULL;
418 int len;
420 for (i = 0; i < n; i++) {
421 e = entry + i;
422 extended_entry_extract(tx + i, e, NULL, 0);
426 * A tree may have "t-2" at the current location even
427 * though it may have "t" that is a subtree behind it,
428 * and another tree may return "t". We want to grab
429 * all "t" from all trees to match in such a case.
431 for (i = 0; i < n; i++) {
432 e = entry + i;
433 if (!e->path)
434 continue;
435 len = tree_entry_len(e);
436 if (!first) {
437 first = e->path;
438 first_len = len;
439 continue;
441 if (name_compare(e->path, len, first, first_len) < 0) {
442 first = e->path;
443 first_len = len;
447 if (first) {
448 for (i = 0; i < n; i++) {
449 e = entry + i;
450 extended_entry_extract(tx + i, e, first, first_len);
451 /* Cull the ones that are not the earliest */
452 if (!e->path)
453 continue;
454 len = tree_entry_len(e);
455 if (name_compare(e->path, len, first, first_len))
456 entry_clear(e);
460 /* Now we have in entry[i] the earliest name from the trees */
461 mask = 0;
462 dirmask = 0;
463 for (i = 0; i < n; i++) {
464 if (!entry[i].path)
465 continue;
466 mask |= 1ul << i;
467 if (S_ISDIR(entry[i].mode))
468 dirmask |= 1ul << i;
469 e = &entry[i];
471 if (!mask)
472 break;
473 interesting = prune_traversal(istate, e, info, &base, interesting);
474 if (interesting < 0)
475 break;
476 if (interesting) {
477 trees_used = info->fn(n, mask, dirmask, entry, info);
478 if (trees_used < 0) {
479 error = trees_used;
480 if (!info->show_all_errors)
481 break;
483 mask &= trees_used;
485 for (i = 0; i < n; i++)
486 if (mask & (1ul << i))
487 update_extended_entry(tx + i, entry + i);
489 free(entry);
490 for (i = 0; i < n; i++)
491 free_extended_entry(tx + i);
492 free(tx);
493 free(traverse_path);
494 info->traverse_path = NULL;
495 strbuf_release(&base);
496 return error;
499 struct dir_state {
500 void *tree;
501 unsigned long size;
502 struct object_id oid;
505 static int find_tree_entry(struct repository *r, struct tree_desc *t,
506 const char *name, struct object_id *result,
507 unsigned short *mode)
509 int namelen = strlen(name);
510 while (t->size) {
511 const char *entry;
512 struct object_id oid;
513 int entrylen, cmp;
515 oidcpy(&oid, tree_entry_extract(t, &entry, mode));
516 entrylen = tree_entry_len(&t->entry);
517 update_tree_entry(t);
518 if (entrylen > namelen)
519 continue;
520 cmp = memcmp(name, entry, entrylen);
521 if (cmp > 0)
522 continue;
523 if (cmp < 0)
524 break;
525 if (entrylen == namelen) {
526 oidcpy(result, &oid);
527 return 0;
529 if (name[entrylen] != '/')
530 continue;
531 if (!S_ISDIR(*mode))
532 break;
533 if (++entrylen == namelen) {
534 oidcpy(result, &oid);
535 return 0;
537 return get_tree_entry(r, &oid, name + entrylen, result, mode);
539 return -1;
542 int get_tree_entry(struct repository *r,
543 const struct object_id *tree_oid,
544 const char *name,
545 struct object_id *oid,
546 unsigned short *mode)
548 int retval;
549 void *tree;
550 unsigned long size;
551 struct object_id root;
553 tree = read_object_with_reference(r, tree_oid, tree_type, &size, &root);
554 if (!tree)
555 return -1;
557 if (name[0] == '\0') {
558 oidcpy(oid, &root);
559 free(tree);
560 return 0;
563 if (!size) {
564 retval = -1;
565 } else {
566 struct tree_desc t;
567 init_tree_desc(&t, tree, size);
568 retval = find_tree_entry(r, &t, name, oid, mode);
570 free(tree);
571 return retval;
575 * This is Linux's built-in max for the number of symlinks to follow.
576 * That limit, of course, does not affect git, but it's a reasonable
577 * choice.
579 #define GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS 40
582 * Find a tree entry by following symlinks in tree_sha (which is
583 * assumed to be the root of the repository). In the event that a
584 * symlink points outside the repository (e.g. a link to /foo or a
585 * root-level link to ../foo), the portion of the link which is
586 * outside the repository will be returned in result_path, and *mode
587 * will be set to 0. It is assumed that result_path is uninitialized.
588 * If there are no symlinks, or the end result of the symlink chain
589 * points to an object inside the repository, result will be filled in
590 * with the sha1 of the found object, and *mode will hold the mode of
591 * the object.
593 * See the code for enum get_oid_result for a description of
594 * the return values.
596 enum get_oid_result get_tree_entry_follow_symlinks(struct repository *r,
597 struct object_id *tree_oid, const char *name,
598 struct object_id *result, struct strbuf *result_path,
599 unsigned short *mode)
601 int retval = MISSING_OBJECT;
602 struct dir_state *parents = NULL;
603 size_t parents_alloc = 0;
604 size_t i, parents_nr = 0;
605 struct object_id current_tree_oid;
606 struct strbuf namebuf = STRBUF_INIT;
607 struct tree_desc t;
608 int follows_remaining = GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS;
610 init_tree_desc(&t, NULL, 0UL);
611 strbuf_addstr(&namebuf, name);
612 oidcpy(&current_tree_oid, tree_oid);
614 while (1) {
615 int find_result;
616 char *first_slash;
617 char *remainder = NULL;
619 if (!t.buffer) {
620 void *tree;
621 struct object_id root;
622 unsigned long size;
623 tree = read_object_with_reference(r,
624 &current_tree_oid,
625 tree_type, &size,
626 &root);
627 if (!tree)
628 goto done;
630 ALLOC_GROW(parents, parents_nr + 1, parents_alloc);
631 parents[parents_nr].tree = tree;
632 parents[parents_nr].size = size;
633 oidcpy(&parents[parents_nr].oid, &root);
634 parents_nr++;
636 if (namebuf.buf[0] == '\0') {
637 oidcpy(result, &root);
638 retval = FOUND;
639 goto done;
642 if (!size)
643 goto done;
645 /* descend */
646 init_tree_desc(&t, tree, size);
649 /* Handle symlinks to e.g. a//b by removing leading slashes */
650 while (namebuf.buf[0] == '/') {
651 strbuf_remove(&namebuf, 0, 1);
654 /* Split namebuf into a first component and a remainder */
655 if ((first_slash = strchr(namebuf.buf, '/'))) {
656 *first_slash = 0;
657 remainder = first_slash + 1;
660 if (!strcmp(namebuf.buf, "..")) {
661 struct dir_state *parent;
663 * We could end up with .. in the namebuf if it
664 * appears in a symlink.
667 if (parents_nr == 1) {
668 if (remainder)
669 *first_slash = '/';
670 strbuf_add(result_path, namebuf.buf,
671 namebuf.len);
672 *mode = 0;
673 retval = FOUND;
674 goto done;
676 parent = &parents[parents_nr - 1];
677 free(parent->tree);
678 parents_nr--;
679 parent = &parents[parents_nr - 1];
680 init_tree_desc(&t, parent->tree, parent->size);
681 strbuf_remove(&namebuf, 0, remainder ? 3 : 2);
682 continue;
685 /* We could end up here via a symlink to dir/.. */
686 if (namebuf.buf[0] == '\0') {
687 oidcpy(result, &parents[parents_nr - 1].oid);
688 retval = FOUND;
689 goto done;
692 /* Look up the first (or only) path component in the tree. */
693 find_result = find_tree_entry(r, &t, namebuf.buf,
694 &current_tree_oid, mode);
695 if (find_result) {
696 goto done;
699 if (S_ISDIR(*mode)) {
700 if (!remainder) {
701 oidcpy(result, &current_tree_oid);
702 retval = FOUND;
703 goto done;
705 /* Descend the tree */
706 t.buffer = NULL;
707 strbuf_remove(&namebuf, 0,
708 1 + first_slash - namebuf.buf);
709 } else if (S_ISREG(*mode)) {
710 if (!remainder) {
711 oidcpy(result, &current_tree_oid);
712 retval = FOUND;
713 } else {
714 retval = NOT_DIR;
716 goto done;
717 } else if (S_ISLNK(*mode)) {
718 /* Follow a symlink */
719 unsigned long link_len;
720 size_t len;
721 char *contents, *contents_start;
722 struct dir_state *parent;
723 enum object_type type;
725 if (follows_remaining-- == 0) {
726 /* Too many symlinks followed */
727 retval = SYMLINK_LOOP;
728 goto done;
732 * At this point, we have followed at a least
733 * one symlink, so on error we need to report this.
735 retval = DANGLING_SYMLINK;
737 contents = repo_read_object_file(r,
738 &current_tree_oid, &type,
739 &link_len);
741 if (!contents)
742 goto done;
744 if (contents[0] == '/') {
745 strbuf_addstr(result_path, contents);
746 free(contents);
747 *mode = 0;
748 retval = FOUND;
749 goto done;
752 if (remainder)
753 len = first_slash - namebuf.buf;
754 else
755 len = namebuf.len;
757 contents_start = contents;
759 parent = &parents[parents_nr - 1];
760 init_tree_desc(&t, parent->tree, parent->size);
761 strbuf_splice(&namebuf, 0, len,
762 contents_start, link_len);
763 if (remainder)
764 namebuf.buf[link_len] = '/';
765 free(contents);
768 done:
769 for (i = 0; i < parents_nr; i++)
770 free(parents[i].tree);
771 free(parents);
773 strbuf_release(&namebuf);
774 return retval;
777 static int match_entry(const struct pathspec_item *item,
778 const struct name_entry *entry, int pathlen,
779 const char *match, int matchlen,
780 enum interesting *never_interesting)
782 int m = -1; /* signals that we haven't called strncmp() */
784 if (item->magic & PATHSPEC_ICASE)
786 * "Never interesting" trick requires exact
787 * matching. We could do something clever with inexact
788 * matching, but it's trickier (and not to forget that
789 * strcasecmp is locale-dependent, at least in
790 * glibc). Just disable it for now. It can't be worse
791 * than the wildcard's codepath of '[Tt][Hi][Is][Ss]'
792 * pattern.
794 *never_interesting = entry_not_interesting;
795 else if (*never_interesting != entry_not_interesting) {
797 * We have not seen any match that sorts later
798 * than the current path.
802 * Does match sort strictly earlier than path
803 * with their common parts?
805 m = strncmp(match, entry->path,
806 (matchlen < pathlen) ? matchlen : pathlen);
807 if (m < 0)
808 return 0;
811 * If we come here even once, that means there is at
812 * least one pathspec that would sort equal to or
813 * later than the path we are currently looking at.
814 * In other words, if we have never reached this point
815 * after iterating all pathspecs, it means all
816 * pathspecs are either outside of base, or inside the
817 * base but sorts strictly earlier than the current
818 * one. In either case, they will never match the
819 * subsequent entries. In such a case, we initialized
820 * the variable to -1 and that is what will be
821 * returned, allowing the caller to terminate early.
823 *never_interesting = entry_not_interesting;
826 if (pathlen > matchlen)
827 return 0;
829 if (matchlen > pathlen) {
830 if (match[pathlen] != '/')
831 return 0;
832 if (!S_ISDIR(entry->mode) && !S_ISGITLINK(entry->mode))
833 return 0;
836 if (m == -1)
838 * we cheated and did not do strncmp(), so we do
839 * that here.
841 m = ps_strncmp(item, match, entry->path, pathlen);
844 * If common part matched earlier then it is a hit,
845 * because we rejected the case where path is not a
846 * leading directory and is shorter than match.
848 if (!m)
850 * match_entry does not check if the prefix part is
851 * matched case-sensitively. If the entry is a
852 * directory and part of prefix, it'll be rematched
853 * eventually by basecmp with special treatment for
854 * the prefix.
856 return 1;
858 return 0;
861 /* :(icase)-aware string compare */
862 static int basecmp(const struct pathspec_item *item,
863 const char *base, const char *match, int len)
865 if (item->magic & PATHSPEC_ICASE) {
866 int ret, n = len > item->prefix ? item->prefix : len;
867 ret = strncmp(base, match, n);
868 if (ret)
869 return ret;
870 base += n;
871 match += n;
872 len -= n;
874 return ps_strncmp(item, base, match, len);
877 static int match_dir_prefix(const struct pathspec_item *item,
878 const char *base,
879 const char *match, int matchlen)
881 if (basecmp(item, base, match, matchlen))
882 return 0;
885 * If the base is a subdirectory of a path which
886 * was specified, all of them are interesting.
888 if (!matchlen ||
889 base[matchlen] == '/' ||
890 match[matchlen - 1] == '/')
891 return 1;
893 /* Just a random prefix match */
894 return 0;
898 * Perform matching on the leading non-wildcard part of
899 * pathspec. item->nowildcard_len must be greater than zero. Return
900 * non-zero if base is matched.
902 static int match_wildcard_base(const struct pathspec_item *item,
903 const char *base, int baselen,
904 int *matched)
906 const char *match = item->match;
907 /* the wildcard part is not considered in this function */
908 int matchlen = item->nowildcard_len;
910 if (baselen) {
911 int dirlen;
913 * Return early if base is longer than the
914 * non-wildcard part but it does not match.
916 if (baselen >= matchlen) {
917 *matched = matchlen;
918 return !basecmp(item, base, match, matchlen);
921 dirlen = matchlen;
922 while (dirlen && match[dirlen - 1] != '/')
923 dirlen--;
926 * Return early if base is shorter than the
927 * non-wildcard part but it does not match. Note that
928 * base ends with '/' so we are sure it really matches
929 * directory
931 if (basecmp(item, base, match, baselen))
932 return 0;
933 *matched = baselen;
934 } else
935 *matched = 0;
937 * we could have checked entry against the non-wildcard part
938 * that is not in base and does similar never_interesting
939 * optimization as in match_entry. For now just be happy with
940 * base comparison.
942 return entry_interesting;
946 * Is a tree entry interesting given the pathspec we have?
948 * Pre-condition: either baselen == base_offset (i.e. empty path)
949 * or base[baselen-1] == '/' (i.e. with trailing slash).
951 static enum interesting do_match(struct index_state *istate,
952 const struct name_entry *entry,
953 struct strbuf *base, int base_offset,
954 const struct pathspec *ps,
955 int exclude)
957 int i;
958 int pathlen, baselen = base->len - base_offset;
959 enum interesting never_interesting = ps->has_wildcard ?
960 entry_not_interesting : all_entries_not_interesting;
962 GUARD_PATHSPEC(ps,
963 PATHSPEC_FROMTOP |
964 PATHSPEC_MAXDEPTH |
965 PATHSPEC_LITERAL |
966 PATHSPEC_GLOB |
967 PATHSPEC_ICASE |
968 PATHSPEC_EXCLUDE |
969 PATHSPEC_ATTR);
971 if (!ps->nr) {
972 if (!ps->recursive ||
973 !(ps->magic & PATHSPEC_MAXDEPTH) ||
974 ps->max_depth == -1)
975 return all_entries_interesting;
976 return within_depth(base->buf + base_offset, baselen,
977 !!S_ISDIR(entry->mode),
978 ps->max_depth) ?
979 entry_interesting : entry_not_interesting;
982 pathlen = tree_entry_len(entry);
984 for (i = ps->nr - 1; i >= 0; i--) {
985 const struct pathspec_item *item = ps->items+i;
986 const char *match = item->match;
987 const char *base_str = base->buf + base_offset;
988 int matchlen = item->len, matched = 0;
990 if ((!exclude && item->magic & PATHSPEC_EXCLUDE) ||
991 ( exclude && !(item->magic & PATHSPEC_EXCLUDE)))
992 continue;
994 if (baselen >= matchlen) {
995 /* If it doesn't match, move along... */
996 if (!match_dir_prefix(item, base_str, match, matchlen))
997 goto match_wildcards;
999 if (!ps->recursive ||
1000 !(ps->magic & PATHSPEC_MAXDEPTH) ||
1001 ps->max_depth == -1) {
1002 if (!item->attr_match_nr)
1003 return all_entries_interesting;
1004 else
1005 goto interesting;
1008 if (within_depth(base_str + matchlen + 1,
1009 baselen - matchlen - 1,
1010 !!S_ISDIR(entry->mode),
1011 ps->max_depth))
1012 goto interesting;
1013 else
1014 return entry_not_interesting;
1017 /* Either there must be no base, or the base must match. */
1018 if (baselen == 0 || !basecmp(item, base_str, match, baselen)) {
1019 if (match_entry(item, entry, pathlen,
1020 match + baselen, matchlen - baselen,
1021 &never_interesting))
1022 goto interesting;
1024 if (item->nowildcard_len < item->len) {
1025 if (!git_fnmatch(item, match + baselen, entry->path,
1026 item->nowildcard_len - baselen))
1027 goto interesting;
1030 * Match all directories. We'll try to
1031 * match files later on.
1033 if (ps->recursive && S_ISDIR(entry->mode))
1034 return entry_interesting;
1037 * When matching against submodules with
1038 * wildcard characters, ensure that the entry
1039 * at least matches up to the first wild
1040 * character. More accurate matching can then
1041 * be performed in the submodule itself.
1043 if (ps->recurse_submodules &&
1044 S_ISGITLINK(entry->mode) &&
1045 !ps_strncmp(item, match + baselen,
1046 entry->path,
1047 item->nowildcard_len - baselen))
1048 goto interesting;
1051 continue;
1054 match_wildcards:
1055 if (item->nowildcard_len == item->len)
1056 continue;
1058 if (item->nowildcard_len &&
1059 !match_wildcard_base(item, base_str, baselen, &matched))
1060 continue;
1063 * Concatenate base and entry->path into one and do
1064 * fnmatch() on it.
1066 * While we could avoid concatenation in certain cases
1067 * [1], which saves a memcpy and potentially a
1068 * realloc, it turns out not worth it. Measurement on
1069 * linux-2.6 does not show any clear improvements,
1070 * partly because of the nowildcard_len optimization
1071 * in git_fnmatch(). Avoid micro-optimizations here.
1073 * [1] if match_wildcard_base() says the base
1074 * directory is already matched, we only need to match
1075 * the rest, which is shorter so _in theory_ faster.
1078 strbuf_add(base, entry->path, pathlen);
1080 if (!git_fnmatch(item, match, base->buf + base_offset,
1081 item->nowildcard_len)) {
1082 strbuf_setlen(base, base_offset + baselen);
1083 goto interesting;
1087 * When matching against submodules with
1088 * wildcard characters, ensure that the entry
1089 * at least matches up to the first wild
1090 * character. More accurate matching can then
1091 * be performed in the submodule itself.
1093 if (ps->recurse_submodules && S_ISGITLINK(entry->mode) &&
1094 !ps_strncmp(item, match, base->buf + base_offset,
1095 item->nowildcard_len)) {
1096 strbuf_setlen(base, base_offset + baselen);
1097 goto interesting;
1100 strbuf_setlen(base, base_offset + baselen);
1103 * Match all directories. We'll try to match files
1104 * later on.
1105 * max_depth is ignored but we may consider support it
1106 * in future, see
1107 * https://public-inbox.org/git/7vmxo5l2g4.fsf@alter.siamese.dyndns.org/
1109 if (ps->recursive && S_ISDIR(entry->mode))
1110 return entry_interesting;
1111 continue;
1112 interesting:
1113 if (item->attr_match_nr) {
1114 int ret;
1117 * Must not return all_entries_not_interesting
1118 * prematurely. We do not know if all entries do not
1119 * match some attributes with current attr API.
1121 never_interesting = entry_not_interesting;
1124 * Consider all directories interesting (because some
1125 * of those files inside may match some attributes
1126 * even though the parent dir does not)
1128 * FIXME: attributes _can_ match directories and we
1129 * can probably return all_entries_interesting or
1130 * all_entries_not_interesting here if matched.
1132 if (S_ISDIR(entry->mode))
1133 return entry_interesting;
1135 strbuf_add(base, entry->path, pathlen);
1136 ret = match_pathspec_attrs(istate, base->buf + base_offset,
1137 base->len - base_offset, item);
1138 strbuf_setlen(base, base_offset + baselen);
1139 if (!ret)
1140 continue;
1142 return entry_interesting;
1144 return never_interesting; /* No matches */
1148 * Is a tree entry interesting given the pathspec we have?
1150 * Pre-condition: either baselen == base_offset (i.e. empty path)
1151 * or base[baselen-1] == '/' (i.e. with trailing slash).
1153 enum interesting tree_entry_interesting(struct index_state *istate,
1154 const struct name_entry *entry,
1155 struct strbuf *base, int base_offset,
1156 const struct pathspec *ps)
1158 enum interesting positive, negative;
1159 positive = do_match(istate, entry, base, base_offset, ps, 0);
1162 * case | entry | positive | negative | result
1163 * -----+-------+----------+----------+-------
1164 * 1 | file | -1 | -1..2 | -1
1165 * 2 | file | 0 | -1..2 | 0
1166 * 3 | file | 1 | -1 | 1
1167 * 4 | file | 1 | 0 | 1
1168 * 5 | file | 1 | 1 | 0
1169 * 6 | file | 1 | 2 | 0
1170 * 7 | file | 2 | -1 | 2
1171 * 8 | file | 2 | 0 | 1
1172 * 9 | file | 2 | 1 | 0
1173 * 10 | file | 2 | 2 | -1
1174 * -----+-------+----------+----------+-------
1175 * 11 | dir | -1 | -1..2 | -1
1176 * 12 | dir | 0 | -1..2 | 0
1177 * 13 | dir | 1 | -1 | 1
1178 * 14 | dir | 1 | 0 | 1
1179 * 15 | dir | 1 | 1 | 1 (*)
1180 * 16 | dir | 1 | 2 | 0
1181 * 17 | dir | 2 | -1 | 2
1182 * 18 | dir | 2 | 0 | 1
1183 * 19 | dir | 2 | 1 | 1 (*)
1184 * 20 | dir | 2 | 2 | -1
1186 * (*) An exclude pattern interested in a directory does not
1187 * necessarily mean it will exclude all of the directory. In
1188 * wildcard case, it can't decide until looking at individual
1189 * files inside. So don't write such directories off yet.
1192 if (!(ps->magic & PATHSPEC_EXCLUDE) ||
1193 positive <= entry_not_interesting) /* #1, #2, #11, #12 */
1194 return positive;
1196 negative = do_match(istate, entry, base, base_offset, ps, 1);
1198 /* #8, #18 */
1199 if (positive == all_entries_interesting &&
1200 negative == entry_not_interesting)
1201 return entry_interesting;
1203 /* #3, #4, #7, #13, #14, #17 */
1204 if (negative <= entry_not_interesting)
1205 return positive;
1207 /* #15, #19 */
1208 if (S_ISDIR(entry->mode) &&
1209 positive >= entry_interesting &&
1210 negative == entry_interesting)
1211 return entry_interesting;
1213 if ((positive == entry_interesting &&
1214 negative >= entry_interesting) || /* #5, #6, #16 */
1215 (positive == all_entries_interesting &&
1216 negative == entry_interesting)) /* #9 */
1217 return entry_not_interesting;
1219 return all_entries_not_interesting; /* #10, #20 */