merge-ort/merge-recursive: do report errors in `merge_submodule()`
[git.git] / tree-walk.c
blobb517792ba23a2104f5f8207519e64cbc5d2daaa4
1 #include "git-compat-util.h"
2 #include "tree-walk.h"
3 #include "dir.h"
4 #include "gettext.h"
5 #include "hex.h"
6 #include "object-file.h"
7 #include "object-store-ll.h"
8 #include "trace2.h"
9 #include "tree.h"
10 #include "pathspec.h"
11 #include "json-writer.h"
12 #include "environment.h"
14 static const char *get_mode(const char *str, unsigned int *modep)
16 unsigned char c;
17 unsigned int mode = 0;
19 if (*str == ' ')
20 return NULL;
22 while ((c = *str++) != ' ') {
23 if (c < '0' || c > '7')
24 return NULL;
25 mode = (mode << 3) + (c - '0');
27 *modep = mode;
28 return str;
31 static int decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size, struct strbuf *err)
33 const char *path;
34 unsigned int mode, len;
35 const unsigned hashsz = the_hash_algo->rawsz;
37 if (size < hashsz + 3 || buf[size - (hashsz + 1)]) {
38 strbuf_addstr(err, _("too-short tree object"));
39 return -1;
42 path = get_mode(buf, &mode);
43 if (!path) {
44 strbuf_addstr(err, _("malformed mode in tree entry"));
45 return -1;
47 if (!*path) {
48 strbuf_addstr(err, _("empty filename in tree entry"));
49 return -1;
51 len = strlen(path) + 1;
53 /* Initialize the descriptor entry */
54 desc->entry.path = path;
55 desc->entry.mode = (desc->flags & TREE_DESC_RAW_MODES) ? mode : canon_mode(mode);
56 desc->entry.pathlen = len - 1;
57 oidread(&desc->entry.oid, (const unsigned char *)path + len);
59 return 0;
62 static int init_tree_desc_internal(struct tree_desc *desc, const void *buffer,
63 unsigned long size, struct strbuf *err,
64 enum tree_desc_flags flags)
66 desc->buffer = buffer;
67 desc->size = size;
68 desc->flags = flags;
69 if (size)
70 return decode_tree_entry(desc, buffer, size, err);
71 return 0;
74 void init_tree_desc(struct tree_desc *desc, const void *buffer, unsigned long size)
76 struct strbuf err = STRBUF_INIT;
77 if (init_tree_desc_internal(desc, buffer, size, &err, 0))
78 die("%s", err.buf);
79 strbuf_release(&err);
82 int init_tree_desc_gently(struct tree_desc *desc, const void *buffer, unsigned long size,
83 enum tree_desc_flags flags)
85 struct strbuf err = STRBUF_INIT;
86 int result = init_tree_desc_internal(desc, buffer, size, &err, flags);
87 if (result)
88 error("%s", err.buf);
89 strbuf_release(&err);
90 return result;
93 void *fill_tree_descriptor(struct repository *r,
94 struct tree_desc *desc,
95 const struct object_id *oid)
97 unsigned long size = 0;
98 void *buf = NULL;
100 if (oid) {
101 buf = read_object_with_reference(r, oid, OBJ_TREE, &size, NULL);
102 if (!buf)
103 die("unable to read tree %s", oid_to_hex(oid));
105 init_tree_desc(desc, buf, size);
106 return buf;
109 static void entry_clear(struct name_entry *a)
111 memset(a, 0, sizeof(*a));
114 static void entry_extract(struct tree_desc *t, struct name_entry *a)
116 *a = t->entry;
119 static int update_tree_entry_internal(struct tree_desc *desc, struct strbuf *err)
121 const void *buf = desc->buffer;
122 const unsigned char *end = (const unsigned char *)desc->entry.path + desc->entry.pathlen + 1 + the_hash_algo->rawsz;
123 unsigned long size = desc->size;
124 unsigned long len = end - (const unsigned char *)buf;
126 if (size < len)
127 die(_("too-short tree file"));
128 buf = end;
129 size -= len;
130 desc->buffer = buf;
131 desc->size = size;
132 if (size)
133 return decode_tree_entry(desc, buf, size, err);
134 return 0;
137 void update_tree_entry(struct tree_desc *desc)
139 struct strbuf err = STRBUF_INIT;
140 if (update_tree_entry_internal(desc, &err))
141 die("%s", err.buf);
142 strbuf_release(&err);
145 int update_tree_entry_gently(struct tree_desc *desc)
147 struct strbuf err = STRBUF_INIT;
148 if (update_tree_entry_internal(desc, &err)) {
149 error("%s", err.buf);
150 strbuf_release(&err);
151 /* Stop processing this tree after error */
152 desc->size = 0;
153 return -1;
155 strbuf_release(&err);
156 return 0;
159 int tree_entry(struct tree_desc *desc, struct name_entry *entry)
161 if (!desc->size)
162 return 0;
164 *entry = desc->entry;
165 update_tree_entry(desc);
166 return 1;
169 int tree_entry_gently(struct tree_desc *desc, struct name_entry *entry)
171 if (!desc->size)
172 return 0;
174 *entry = desc->entry;
175 if (update_tree_entry_gently(desc))
176 return 0;
177 return 1;
180 static int traverse_trees_atexit_registered;
181 static int traverse_trees_count;
182 static int traverse_trees_cur_depth;
183 static int traverse_trees_max_depth;
185 static void trace2_traverse_trees_statistics_atexit(void)
187 struct json_writer jw = JSON_WRITER_INIT;
189 jw_object_begin(&jw, 0);
190 jw_object_intmax(&jw, "traverse_trees_count", traverse_trees_count);
191 jw_object_intmax(&jw, "traverse_trees_max_depth", traverse_trees_max_depth);
192 jw_end(&jw);
194 trace2_data_json("traverse_trees", the_repository, "statistics", &jw);
196 jw_release(&jw);
199 void setup_traverse_info(struct traverse_info *info, const char *base)
201 size_t pathlen = strlen(base);
202 static struct traverse_info dummy;
204 memset(info, 0, sizeof(*info));
205 if (pathlen && base[pathlen-1] == '/')
206 pathlen--;
207 info->pathlen = pathlen ? pathlen + 1 : 0;
208 info->name = base;
209 info->namelen = pathlen;
210 if (pathlen)
211 info->prev = &dummy;
213 if (trace2_is_enabled() && !traverse_trees_atexit_registered) {
214 atexit(trace2_traverse_trees_statistics_atexit);
215 traverse_trees_atexit_registered = 1;
219 char *make_traverse_path(char *path, size_t pathlen,
220 const struct traverse_info *info,
221 const char *name, size_t namelen)
223 /* Always points to the end of the name we're about to add */
224 size_t pos = st_add(info->pathlen, namelen);
226 if (pos >= pathlen)
227 BUG("too small buffer passed to make_traverse_path");
229 path[pos] = 0;
230 for (;;) {
231 if (pos < namelen)
232 BUG("traverse_info pathlen does not match strings");
233 pos -= namelen;
234 memcpy(path + pos, name, namelen);
236 if (!pos)
237 break;
238 path[--pos] = '/';
240 if (!info)
241 BUG("traverse_info ran out of list items");
242 name = info->name;
243 namelen = info->namelen;
244 info = info->prev;
246 return path;
249 void strbuf_make_traverse_path(struct strbuf *out,
250 const struct traverse_info *info,
251 const char *name, size_t namelen)
253 size_t len = traverse_path_len(info, namelen);
255 strbuf_grow(out, len);
256 make_traverse_path(out->buf + out->len, out->alloc - out->len,
257 info, name, namelen);
258 strbuf_setlen(out, out->len + len);
261 struct tree_desc_skip {
262 struct tree_desc_skip *prev;
263 const void *ptr;
266 struct tree_desc_x {
267 struct tree_desc d;
268 struct tree_desc_skip *skip;
271 static int check_entry_match(const char *a, int a_len, const char *b, int b_len)
274 * The caller wants to pick *a* from a tree or nothing.
275 * We are looking at *b* in a tree.
277 * (0) If a and b are the same name, we are trivially happy.
279 * There are three possibilities where *a* could be hiding
280 * behind *b*.
282 * (1) *a* == "t", *b* == "ab" i.e. *b* sorts earlier than *a* no
283 * matter what.
284 * (2) *a* == "t", *b* == "t-2" and "t" is a subtree in the tree;
285 * (3) *a* == "t-2", *b* == "t" and "t-2" is a blob in the tree.
287 * Otherwise we know *a* won't appear in the tree without
288 * scanning further.
291 int cmp = name_compare(a, a_len, b, b_len);
293 /* Most common case first -- reading sync'd trees */
294 if (!cmp)
295 return cmp;
297 if (0 < cmp) {
298 /* a comes after b; it does not matter if it is case (3)
299 if (b_len < a_len && !memcmp(a, b, b_len) && a[b_len] < '/')
300 return 1;
302 return 1; /* keep looking */
305 /* b comes after a; are we looking at case (2)? */
306 if (a_len < b_len && !memcmp(a, b, a_len) && b[a_len] < '/')
307 return 1; /* keep looking */
309 return -1; /* a cannot appear in the tree */
313 * From the extended tree_desc, extract the first name entry, while
314 * paying attention to the candidate "first" name. Most importantly,
315 * when looking for an entry, if there are entries that sorts earlier
316 * in the tree object representation than that name, skip them and
317 * process the named entry first. We will remember that we haven't
318 * processed the first entry yet, and in the later call skip the
319 * entry we processed early when update_extended_entry() is called.
321 * E.g. if the underlying tree object has these entries:
323 * blob "t-1"
324 * blob "t-2"
325 * tree "t"
326 * blob "t=1"
328 * and the "first" asks for "t", remember that we still need to
329 * process "t-1" and "t-2" but extract "t". After processing the
330 * entry "t" from this call, the caller will let us know by calling
331 * update_extended_entry() that we can remember "t" has been processed
332 * already.
335 static void extended_entry_extract(struct tree_desc_x *t,
336 struct name_entry *a,
337 const char *first,
338 int first_len)
340 const char *path;
341 int len;
342 struct tree_desc probe;
343 struct tree_desc_skip *skip;
346 * Extract the first entry from the tree_desc, but skip the
347 * ones that we already returned in earlier rounds.
349 while (1) {
350 if (!t->d.size) {
351 entry_clear(a);
352 break; /* not found */
354 entry_extract(&t->d, a);
355 for (skip = t->skip; skip; skip = skip->prev)
356 if (a->path == skip->ptr)
357 break; /* found */
358 if (!skip)
359 break;
360 /* We have processed this entry already. */
361 update_tree_entry(&t->d);
364 if (!first || !a->path)
365 return;
368 * The caller wants "first" from this tree, or nothing.
370 path = a->path;
371 len = tree_entry_len(a);
372 switch (check_entry_match(first, first_len, path, len)) {
373 case -1:
374 entry_clear(a);
375 case 0:
376 return;
377 default:
378 break;
382 * We need to look-ahead -- we suspect that a subtree whose
383 * name is "first" may be hiding behind the current entry "path".
385 probe = t->d;
386 while (probe.size) {
387 entry_extract(&probe, a);
388 path = a->path;
389 len = tree_entry_len(a);
390 switch (check_entry_match(first, first_len, path, len)) {
391 case -1:
392 entry_clear(a);
393 case 0:
394 return;
395 default:
396 update_tree_entry(&probe);
397 break;
399 /* keep looking */
401 entry_clear(a);
404 static void update_extended_entry(struct tree_desc_x *t, struct name_entry *a)
406 if (t->d.entry.path == a->path) {
407 update_tree_entry(&t->d);
408 } else {
409 /* we have returned this entry early */
410 struct tree_desc_skip *skip = xmalloc(sizeof(*skip));
411 skip->ptr = a->path;
412 skip->prev = t->skip;
413 t->skip = skip;
417 static void free_extended_entry(struct tree_desc_x *t)
419 struct tree_desc_skip *p, *s;
421 for (s = t->skip; s; s = p) {
422 p = s->prev;
423 free(s);
427 static inline int prune_traversal(struct index_state *istate,
428 struct name_entry *e,
429 struct traverse_info *info,
430 struct strbuf *base,
431 int still_interesting)
433 if (!info->pathspec || still_interesting == 2)
434 return 2;
435 if (still_interesting < 0)
436 return still_interesting;
437 return tree_entry_interesting(istate, e, base,
438 info->pathspec);
441 int traverse_trees(struct index_state *istate,
442 int n, struct tree_desc *t,
443 struct traverse_info *info)
445 int ret = 0;
446 struct name_entry *entry;
447 int i;
448 struct tree_desc_x *tx;
449 struct strbuf base = STRBUF_INIT;
450 int interesting = 1;
451 char *traverse_path;
453 if (traverse_trees_cur_depth > max_allowed_tree_depth)
454 return error("exceeded maximum allowed tree depth");
456 traverse_trees_count++;
457 traverse_trees_cur_depth++;
459 if (traverse_trees_cur_depth > traverse_trees_max_depth)
460 traverse_trees_max_depth = traverse_trees_cur_depth;
462 ALLOC_ARRAY(entry, n);
463 ALLOC_ARRAY(tx, n);
465 for (i = 0; i < n; i++) {
466 tx[i].d = t[i];
467 tx[i].skip = NULL;
470 if (info->prev) {
471 strbuf_make_traverse_path(&base, info->prev,
472 info->name, info->namelen);
473 strbuf_addch(&base, '/');
474 traverse_path = xstrndup(base.buf, base.len);
475 } else {
476 traverse_path = xstrndup(info->name, info->pathlen);
478 info->traverse_path = traverse_path;
479 for (;;) {
480 int trees_used;
481 unsigned long mask, dirmask;
482 const char *first = NULL;
483 int first_len = 0;
484 struct name_entry *e = NULL;
485 int len;
487 for (i = 0; i < n; i++) {
488 e = entry + i;
489 extended_entry_extract(tx + i, e, NULL, 0);
493 * A tree may have "t-2" at the current location even
494 * though it may have "t" that is a subtree behind it,
495 * and another tree may return "t". We want to grab
496 * all "t" from all trees to match in such a case.
498 for (i = 0; i < n; i++) {
499 e = entry + i;
500 if (!e->path)
501 continue;
502 len = tree_entry_len(e);
503 if (!first) {
504 first = e->path;
505 first_len = len;
506 continue;
508 if (name_compare(e->path, len, first, first_len) < 0) {
509 first = e->path;
510 first_len = len;
514 if (first) {
515 for (i = 0; i < n; i++) {
516 e = entry + i;
517 extended_entry_extract(tx + i, e, first, first_len);
518 /* Cull the ones that are not the earliest */
519 if (!e->path)
520 continue;
521 len = tree_entry_len(e);
522 if (name_compare(e->path, len, first, first_len))
523 entry_clear(e);
527 /* Now we have in entry[i] the earliest name from the trees */
528 mask = 0;
529 dirmask = 0;
530 for (i = 0; i < n; i++) {
531 if (!entry[i].path)
532 continue;
533 mask |= 1ul << i;
534 if (S_ISDIR(entry[i].mode))
535 dirmask |= 1ul << i;
536 e = &entry[i];
538 if (!mask)
539 break;
540 interesting = prune_traversal(istate, e, info, &base, interesting);
541 if (interesting < 0)
542 break;
543 if (interesting) {
544 trees_used = info->fn(n, mask, dirmask, entry, info);
545 if (trees_used < 0) {
546 ret = trees_used;
547 if (!info->show_all_errors)
548 break;
550 mask &= trees_used;
552 for (i = 0; i < n; i++)
553 if (mask & (1ul << i))
554 update_extended_entry(tx + i, entry + i);
556 for (i = 0; i < n; i++)
557 free_extended_entry(tx + i);
558 free(tx);
559 free(entry);
560 free(traverse_path);
561 info->traverse_path = NULL;
562 strbuf_release(&base);
564 traverse_trees_cur_depth--;
565 return ret;
568 struct dir_state {
569 void *tree;
570 unsigned long size;
571 struct object_id oid;
574 static int find_tree_entry(struct repository *r, struct tree_desc *t,
575 const char *name, struct object_id *result,
576 unsigned short *mode)
578 int namelen = strlen(name);
579 while (t->size) {
580 const char *entry;
581 struct object_id oid;
582 int entrylen, cmp;
584 oidcpy(&oid, tree_entry_extract(t, &entry, mode));
585 entrylen = tree_entry_len(&t->entry);
586 update_tree_entry(t);
587 if (entrylen > namelen)
588 continue;
589 cmp = memcmp(name, entry, entrylen);
590 if (cmp > 0)
591 continue;
592 if (cmp < 0)
593 break;
594 if (entrylen == namelen) {
595 oidcpy(result, &oid);
596 return 0;
598 if (name[entrylen] != '/')
599 continue;
600 if (!S_ISDIR(*mode))
601 break;
602 if (++entrylen == namelen) {
603 oidcpy(result, &oid);
604 return 0;
606 return get_tree_entry(r, &oid, name + entrylen, result, mode);
608 return -1;
611 int get_tree_entry(struct repository *r,
612 const struct object_id *tree_oid,
613 const char *name,
614 struct object_id *oid,
615 unsigned short *mode)
617 int retval;
618 void *tree;
619 unsigned long size;
620 struct object_id root;
622 tree = read_object_with_reference(r, tree_oid, OBJ_TREE, &size, &root);
623 if (!tree)
624 return -1;
626 if (name[0] == '\0') {
627 oidcpy(oid, &root);
628 free(tree);
629 return 0;
632 if (!size) {
633 retval = -1;
634 } else {
635 struct tree_desc t;
636 init_tree_desc(&t, tree, size);
637 retval = find_tree_entry(r, &t, name, oid, mode);
639 free(tree);
640 return retval;
644 * This is Linux's built-in max for the number of symlinks to follow.
645 * That limit, of course, does not affect git, but it's a reasonable
646 * choice.
648 #define GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS 40
651 * Find a tree entry by following symlinks in tree_sha (which is
652 * assumed to be the root of the repository). In the event that a
653 * symlink points outside the repository (e.g. a link to /foo or a
654 * root-level link to ../foo), the portion of the link which is
655 * outside the repository will be returned in result_path, and *mode
656 * will be set to 0. It is assumed that result_path is uninitialized.
657 * If there are no symlinks, or the end result of the symlink chain
658 * points to an object inside the repository, result will be filled in
659 * with the sha1 of the found object, and *mode will hold the mode of
660 * the object.
662 * See the code for enum get_oid_result for a description of
663 * the return values.
665 enum get_oid_result get_tree_entry_follow_symlinks(struct repository *r,
666 struct object_id *tree_oid, const char *name,
667 struct object_id *result, struct strbuf *result_path,
668 unsigned short *mode)
670 int retval = MISSING_OBJECT;
671 struct dir_state *parents = NULL;
672 size_t parents_alloc = 0;
673 size_t i, parents_nr = 0;
674 struct object_id current_tree_oid;
675 struct strbuf namebuf = STRBUF_INIT;
676 struct tree_desc t;
677 int follows_remaining = GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS;
679 init_tree_desc(&t, NULL, 0UL);
680 strbuf_addstr(&namebuf, name);
681 oidcpy(&current_tree_oid, tree_oid);
683 while (1) {
684 int find_result;
685 char *first_slash;
686 char *remainder = NULL;
688 if (!t.buffer) {
689 void *tree;
690 struct object_id root;
691 unsigned long size;
692 tree = read_object_with_reference(r,
693 &current_tree_oid,
694 OBJ_TREE, &size,
695 &root);
696 if (!tree)
697 goto done;
699 ALLOC_GROW(parents, parents_nr + 1, parents_alloc);
700 parents[parents_nr].tree = tree;
701 parents[parents_nr].size = size;
702 oidcpy(&parents[parents_nr].oid, &root);
703 parents_nr++;
705 if (namebuf.buf[0] == '\0') {
706 oidcpy(result, &root);
707 retval = FOUND;
708 goto done;
711 if (!size)
712 goto done;
714 /* descend */
715 init_tree_desc(&t, tree, size);
718 /* Handle symlinks to e.g. a//b by removing leading slashes */
719 while (namebuf.buf[0] == '/') {
720 strbuf_remove(&namebuf, 0, 1);
723 /* Split namebuf into a first component and a remainder */
724 if ((first_slash = strchr(namebuf.buf, '/'))) {
725 *first_slash = 0;
726 remainder = first_slash + 1;
729 if (!strcmp(namebuf.buf, "..")) {
730 struct dir_state *parent;
732 * We could end up with .. in the namebuf if it
733 * appears in a symlink.
736 if (parents_nr == 1) {
737 if (remainder)
738 *first_slash = '/';
739 strbuf_add(result_path, namebuf.buf,
740 namebuf.len);
741 *mode = 0;
742 retval = FOUND;
743 goto done;
745 parent = &parents[parents_nr - 1];
746 free(parent->tree);
747 parents_nr--;
748 parent = &parents[parents_nr - 1];
749 init_tree_desc(&t, parent->tree, parent->size);
750 strbuf_remove(&namebuf, 0, remainder ? 3 : 2);
751 continue;
754 /* We could end up here via a symlink to dir/.. */
755 if (namebuf.buf[0] == '\0') {
756 oidcpy(result, &parents[parents_nr - 1].oid);
757 retval = FOUND;
758 goto done;
761 /* Look up the first (or only) path component in the tree. */
762 find_result = find_tree_entry(r, &t, namebuf.buf,
763 &current_tree_oid, mode);
764 if (find_result) {
765 goto done;
768 if (S_ISDIR(*mode)) {
769 if (!remainder) {
770 oidcpy(result, &current_tree_oid);
771 retval = FOUND;
772 goto done;
774 /* Descend the tree */
775 t.buffer = NULL;
776 strbuf_remove(&namebuf, 0,
777 1 + first_slash - namebuf.buf);
778 } else if (S_ISREG(*mode)) {
779 if (!remainder) {
780 oidcpy(result, &current_tree_oid);
781 retval = FOUND;
782 } else {
783 retval = NOT_DIR;
785 goto done;
786 } else if (S_ISLNK(*mode)) {
787 /* Follow a symlink */
788 unsigned long link_len;
789 size_t len;
790 char *contents, *contents_start;
791 struct dir_state *parent;
792 enum object_type type;
794 if (follows_remaining-- == 0) {
795 /* Too many symlinks followed */
796 retval = SYMLINK_LOOP;
797 goto done;
801 * At this point, we have followed at a least
802 * one symlink, so on error we need to report this.
804 retval = DANGLING_SYMLINK;
806 contents = repo_read_object_file(r,
807 &current_tree_oid, &type,
808 &link_len);
810 if (!contents)
811 goto done;
813 if (contents[0] == '/') {
814 strbuf_addstr(result_path, contents);
815 free(contents);
816 *mode = 0;
817 retval = FOUND;
818 goto done;
821 if (remainder)
822 len = first_slash - namebuf.buf;
823 else
824 len = namebuf.len;
826 contents_start = contents;
828 parent = &parents[parents_nr - 1];
829 init_tree_desc(&t, parent->tree, parent->size);
830 strbuf_splice(&namebuf, 0, len,
831 contents_start, link_len);
832 if (remainder)
833 namebuf.buf[link_len] = '/';
834 free(contents);
837 done:
838 for (i = 0; i < parents_nr; i++)
839 free(parents[i].tree);
840 free(parents);
842 strbuf_release(&namebuf);
843 return retval;
846 static int match_entry(const struct pathspec_item *item,
847 const struct name_entry *entry, int pathlen,
848 const char *match, int matchlen,
849 enum interesting *never_interesting)
851 int m = -1; /* signals that we haven't called strncmp() */
853 if (item->magic & PATHSPEC_ICASE)
855 * "Never interesting" trick requires exact
856 * matching. We could do something clever with inexact
857 * matching, but it's trickier (and not to forget that
858 * strcasecmp is locale-dependent, at least in
859 * glibc). Just disable it for now. It can't be worse
860 * than the wildcard's codepath of '[Tt][Hi][Is][Ss]'
861 * pattern.
863 *never_interesting = entry_not_interesting;
864 else if (*never_interesting != entry_not_interesting) {
866 * We have not seen any match that sorts later
867 * than the current path.
871 * Does match sort strictly earlier than path
872 * with their common parts?
874 m = strncmp(match, entry->path,
875 (matchlen < pathlen) ? matchlen : pathlen);
876 if (m < 0)
877 return 0;
880 * If we come here even once, that means there is at
881 * least one pathspec that would sort equal to or
882 * later than the path we are currently looking at.
883 * In other words, if we have never reached this point
884 * after iterating all pathspecs, it means all
885 * pathspecs are either outside of base, or inside the
886 * base but sorts strictly earlier than the current
887 * one. In either case, they will never match the
888 * subsequent entries. In such a case, we initialized
889 * the variable to -1 and that is what will be
890 * returned, allowing the caller to terminate early.
892 *never_interesting = entry_not_interesting;
895 if (pathlen > matchlen)
896 return 0;
898 if (matchlen > pathlen) {
899 if (match[pathlen] != '/')
900 return 0;
902 * Reject non-directories as partial pathnames, except
903 * when match is a submodule with a trailing slash and
904 * nothing else (to handle 'submod/' and 'submod'
905 * uniformly).
907 if (!S_ISDIR(entry->mode) &&
908 (!S_ISGITLINK(entry->mode) || matchlen > pathlen + 1))
909 return 0;
912 if (m == -1)
914 * we cheated and did not do strncmp(), so we do
915 * that here.
917 m = ps_strncmp(item, match, entry->path, pathlen);
920 * If common part matched earlier then it is a hit,
921 * because we rejected the case where path is not a
922 * leading directory and is shorter than match.
924 if (!m)
926 * match_entry does not check if the prefix part is
927 * matched case-sensitively. If the entry is a
928 * directory and part of prefix, it'll be rematched
929 * eventually by basecmp with special treatment for
930 * the prefix.
932 return 1;
934 return 0;
937 /* :(icase)-aware string compare */
938 static int basecmp(const struct pathspec_item *item,
939 const char *base, const char *match, int len)
941 if (item->magic & PATHSPEC_ICASE) {
942 int ret, n = len > item->prefix ? item->prefix : len;
943 ret = strncmp(base, match, n);
944 if (ret)
945 return ret;
946 base += n;
947 match += n;
948 len -= n;
950 return ps_strncmp(item, base, match, len);
953 static int match_dir_prefix(const struct pathspec_item *item,
954 const char *base,
955 const char *match, int matchlen)
957 if (basecmp(item, base, match, matchlen))
958 return 0;
961 * If the base is a subdirectory of a path which
962 * was specified, all of them are interesting.
964 if (!matchlen ||
965 base[matchlen] == '/' ||
966 match[matchlen - 1] == '/')
967 return 1;
969 /* Just a random prefix match */
970 return 0;
974 * Perform matching on the leading non-wildcard part of
975 * pathspec. item->nowildcard_len must be greater than zero. Return
976 * non-zero if base is matched.
978 static int match_wildcard_base(const struct pathspec_item *item,
979 const char *base, int baselen,
980 int *matched)
982 const char *match = item->match;
983 /* the wildcard part is not considered in this function */
984 int matchlen = item->nowildcard_len;
986 if (baselen) {
987 int dirlen;
989 * Return early if base is longer than the
990 * non-wildcard part but it does not match.
992 if (baselen >= matchlen) {
993 *matched = matchlen;
994 return !basecmp(item, base, match, matchlen);
997 dirlen = matchlen;
998 while (dirlen && match[dirlen - 1] != '/')
999 dirlen--;
1002 * Return early if base is shorter than the
1003 * non-wildcard part but it does not match. Note that
1004 * base ends with '/' so we are sure it really matches
1005 * directory
1007 if (basecmp(item, base, match, baselen))
1008 return 0;
1009 *matched = baselen;
1010 } else
1011 *matched = 0;
1013 * we could have checked entry against the non-wildcard part
1014 * that is not in base and does similar never_interesting
1015 * optimization as in match_entry. For now just be happy with
1016 * base comparison.
1018 return entry_interesting;
1022 * Is a tree entry interesting given the pathspec we have?
1024 * Pre-condition: either baselen == 0 (i.e. empty path)
1025 * or base[baselen-1] == '/' (i.e. with trailing slash).
1027 static enum interesting do_match(struct index_state *istate,
1028 const struct name_entry *entry,
1029 struct strbuf *base,
1030 const struct pathspec *ps,
1031 int exclude)
1033 int i;
1034 int pathlen, baselen = base->len;
1035 enum interesting never_interesting = ps->has_wildcard ?
1036 entry_not_interesting : all_entries_not_interesting;
1038 GUARD_PATHSPEC(ps,
1039 PATHSPEC_FROMTOP |
1040 PATHSPEC_MAXDEPTH |
1041 PATHSPEC_LITERAL |
1042 PATHSPEC_GLOB |
1043 PATHSPEC_ICASE |
1044 PATHSPEC_EXCLUDE |
1045 PATHSPEC_ATTR);
1047 if (!ps->nr) {
1048 if (!ps->recursive ||
1049 !(ps->magic & PATHSPEC_MAXDEPTH) ||
1050 ps->max_depth == -1)
1051 return all_entries_interesting;
1052 return within_depth(base->buf, baselen,
1053 !!S_ISDIR(entry->mode),
1054 ps->max_depth) ?
1055 entry_interesting : entry_not_interesting;
1058 pathlen = tree_entry_len(entry);
1060 for (i = ps->nr - 1; i >= 0; i--) {
1061 const struct pathspec_item *item = ps->items+i;
1062 const char *match = item->match;
1063 const char *base_str = base->buf;
1064 int matchlen = item->len, matched = 0;
1066 if ((!exclude && item->magic & PATHSPEC_EXCLUDE) ||
1067 ( exclude && !(item->magic & PATHSPEC_EXCLUDE)))
1068 continue;
1070 if (baselen >= matchlen) {
1071 /* If it doesn't match, move along... */
1072 if (!match_dir_prefix(item, base_str, match, matchlen))
1073 goto match_wildcards;
1075 if (!ps->recursive ||
1076 !(ps->magic & PATHSPEC_MAXDEPTH) ||
1077 ps->max_depth == -1) {
1078 if (!item->attr_match_nr)
1079 return all_entries_interesting;
1080 else
1081 goto interesting;
1084 if (within_depth(base_str + matchlen + 1,
1085 baselen - matchlen - 1,
1086 !!S_ISDIR(entry->mode),
1087 ps->max_depth))
1088 goto interesting;
1089 else
1090 return entry_not_interesting;
1093 /* Either there must be no base, or the base must match. */
1094 if (baselen == 0 || !basecmp(item, base_str, match, baselen)) {
1095 if (match_entry(item, entry, pathlen,
1096 match + baselen, matchlen - baselen,
1097 &never_interesting))
1098 goto interesting;
1100 if (item->nowildcard_len < item->len) {
1101 if (!git_fnmatch(item, match + baselen, entry->path,
1102 item->nowildcard_len - baselen))
1103 goto interesting;
1106 * Match all directories. We'll try to
1107 * match files later on.
1109 if (ps->recursive && S_ISDIR(entry->mode))
1110 return entry_interesting;
1113 * When matching against submodules with
1114 * wildcard characters, ensure that the entry
1115 * at least matches up to the first wild
1116 * character. More accurate matching can then
1117 * be performed in the submodule itself.
1119 if (ps->recurse_submodules &&
1120 S_ISGITLINK(entry->mode) &&
1121 !ps_strncmp(item, match + baselen,
1122 entry->path,
1123 item->nowildcard_len - baselen))
1124 goto interesting;
1127 continue;
1130 match_wildcards:
1131 if (item->nowildcard_len == item->len)
1132 continue;
1134 if (item->nowildcard_len &&
1135 !match_wildcard_base(item, base_str, baselen, &matched))
1136 continue;
1139 * Concatenate base and entry->path into one and do
1140 * fnmatch() on it.
1142 * While we could avoid concatenation in certain cases
1143 * [1], which saves a memcpy and potentially a
1144 * realloc, it turns out not worth it. Measurement on
1145 * linux-2.6 does not show any clear improvements,
1146 * partly because of the nowildcard_len optimization
1147 * in git_fnmatch(). Avoid micro-optimizations here.
1149 * [1] if match_wildcard_base() says the base
1150 * directory is already matched, we only need to match
1151 * the rest, which is shorter so _in theory_ faster.
1154 strbuf_add(base, entry->path, pathlen);
1156 if (!git_fnmatch(item, match, base->buf,
1157 item->nowildcard_len)) {
1158 strbuf_setlen(base, baselen);
1159 goto interesting;
1163 * When matching against submodules with
1164 * wildcard characters, ensure that the entry
1165 * at least matches up to the first wild
1166 * character. More accurate matching can then
1167 * be performed in the submodule itself.
1169 if (ps->recurse_submodules && S_ISGITLINK(entry->mode) &&
1170 !ps_strncmp(item, match, base->buf,
1171 item->nowildcard_len)) {
1172 strbuf_setlen(base, baselen);
1173 goto interesting;
1176 strbuf_setlen(base, baselen);
1179 * Match all directories. We'll try to match files
1180 * later on.
1181 * max_depth is ignored but we may consider support it
1182 * in future, see
1183 * https://lore.kernel.org/git/7vmxo5l2g4.fsf@alter.siamese.dyndns.org/
1185 if (ps->recursive && S_ISDIR(entry->mode))
1186 return entry_interesting;
1187 continue;
1188 interesting:
1189 if (item->attr_match_nr) {
1190 int ret;
1193 * Must not return all_entries_not_interesting
1194 * prematurely. We do not know if all entries do not
1195 * match some attributes with current attr API.
1197 never_interesting = entry_not_interesting;
1200 * Consider all directories interesting (because some
1201 * of those files inside may match some attributes
1202 * even though the parent dir does not)
1204 * FIXME: attributes _can_ match directories and we
1205 * can probably return all_entries_interesting or
1206 * all_entries_not_interesting here if matched.
1208 if (S_ISDIR(entry->mode))
1209 return entry_interesting;
1211 strbuf_add(base, entry->path, pathlen);
1212 ret = match_pathspec_attrs(istate, base->buf,
1213 base->len, item);
1214 strbuf_setlen(base, baselen);
1215 if (!ret)
1216 continue;
1218 return entry_interesting;
1220 return never_interesting; /* No matches */
1224 * Is a tree entry interesting given the pathspec we have?
1226 * Pre-condition: either baselen == 0 (i.e. empty path)
1227 * or base[baselen-1] == '/' (i.e. with trailing slash).
1229 enum interesting tree_entry_interesting(struct index_state *istate,
1230 const struct name_entry *entry,
1231 struct strbuf *base,
1232 const struct pathspec *ps)
1234 enum interesting positive, negative;
1235 positive = do_match(istate, entry, base, ps, 0);
1238 * case | entry | positive | negative | result
1239 * -----+-------+----------+----------+-------
1240 * 1 | file | -1 | -1..2 | -1
1241 * 2 | file | 0 | -1..2 | 0
1242 * 3 | file | 1 | -1 | 1
1243 * 4 | file | 1 | 0 | 1
1244 * 5 | file | 1 | 1 | 0
1245 * 6 | file | 1 | 2 | 0
1246 * 7 | file | 2 | -1 | 2
1247 * 8 | file | 2 | 0 | 1
1248 * 9 | file | 2 | 1 | 0
1249 * 10 | file | 2 | 2 | -1
1250 * -----+-------+----------+----------+-------
1251 * 11 | dir | -1 | -1..2 | -1
1252 * 12 | dir | 0 | -1..2 | 0
1253 * 13 | dir | 1 | -1 | 1
1254 * 14 | dir | 1 | 0 | 1
1255 * 15 | dir | 1 | 1 | 1 (*)
1256 * 16 | dir | 1 | 2 | 0
1257 * 17 | dir | 2 | -1 | 2
1258 * 18 | dir | 2 | 0 | 1
1259 * 19 | dir | 2 | 1 | 1 (*)
1260 * 20 | dir | 2 | 2 | -1
1262 * (*) An exclude pattern interested in a directory does not
1263 * necessarily mean it will exclude all of the directory. In
1264 * wildcard case, it can't decide until looking at individual
1265 * files inside. So don't write such directories off yet.
1268 if (!(ps->magic & PATHSPEC_EXCLUDE) ||
1269 positive <= entry_not_interesting) /* #1, #2, #11, #12 */
1270 return positive;
1272 negative = do_match(istate, entry, base, ps, 1);
1274 /* #8, #18 */
1275 if (positive == all_entries_interesting &&
1276 negative == entry_not_interesting)
1277 return entry_interesting;
1279 /* #3, #4, #7, #13, #14, #17 */
1280 if (negative <= entry_not_interesting)
1281 return positive;
1283 /* #15, #19 */
1284 if (S_ISDIR(entry->mode) &&
1285 positive >= entry_interesting &&
1286 negative == entry_interesting)
1287 return entry_interesting;
1289 if ((positive == entry_interesting &&
1290 negative >= entry_interesting) || /* #5, #6, #16 */
1291 (positive == all_entries_interesting &&
1292 negative == entry_interesting)) /* #9 */
1293 return entry_not_interesting;
1295 return all_entries_not_interesting; /* #10, #20 */