Merge branch 'mc/doc-submodule-sync-recurse'
[git.git] / tree-walk.c
blob4dc86c7fe5f66b858b6bd7834a76b1dbbb63a339
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"
8 static const char *get_mode(const char *str, unsigned int *modep)
10 unsigned char c;
11 unsigned int mode = 0;
13 if (*str == ' ')
14 return NULL;
16 while ((c = *str++) != ' ') {
17 if (c < '0' || c > '7')
18 return NULL;
19 mode = (mode << 3) + (c - '0');
21 *modep = mode;
22 return str;
25 static void decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size)
27 const char *path;
28 unsigned int mode, len;
30 if (size < 24 || buf[size - 21])
31 die("corrupt tree file");
33 path = get_mode(buf, &mode);
34 if (!path || !*path)
35 die("corrupt tree file");
36 len = strlen(path) + 1;
38 /* Initialize the descriptor entry */
39 desc->entry.path = path;
40 desc->entry.mode = canon_mode(mode);
41 desc->entry.sha1 = (const unsigned char *)(path + len);
44 void init_tree_desc(struct tree_desc *desc, const void *buffer, unsigned long size)
46 desc->buffer = buffer;
47 desc->size = size;
48 if (size)
49 decode_tree_entry(desc, buffer, size);
52 void *fill_tree_descriptor(struct tree_desc *desc, const unsigned char *sha1)
54 unsigned long size = 0;
55 void *buf = NULL;
57 if (sha1) {
58 buf = read_object_with_reference(sha1, tree_type, &size, NULL);
59 if (!buf)
60 die("unable to read tree %s", sha1_to_hex(sha1));
62 init_tree_desc(desc, buf, size);
63 return buf;
66 static void entry_clear(struct name_entry *a)
68 memset(a, 0, sizeof(*a));
71 static void entry_extract(struct tree_desc *t, struct name_entry *a)
73 *a = t->entry;
76 void update_tree_entry(struct tree_desc *desc)
78 const void *buf = desc->buffer;
79 const unsigned char *end = desc->entry.sha1 + 20;
80 unsigned long size = desc->size;
81 unsigned long len = end - (const unsigned char *)buf;
83 if (size < len)
84 die("corrupt tree file");
85 buf = end;
86 size -= len;
87 desc->buffer = buf;
88 desc->size = size;
89 if (size)
90 decode_tree_entry(desc, buf, size);
93 int tree_entry(struct tree_desc *desc, struct name_entry *entry)
95 if (!desc->size)
96 return 0;
98 *entry = desc->entry;
99 update_tree_entry(desc);
100 return 1;
103 void setup_traverse_info(struct traverse_info *info, const char *base)
105 int pathlen = strlen(base);
106 static struct traverse_info dummy;
108 memset(info, 0, sizeof(*info));
109 if (pathlen && base[pathlen-1] == '/')
110 pathlen--;
111 info->pathlen = pathlen ? pathlen + 1 : 0;
112 info->name.path = base;
113 info->name.sha1 = (void *)(base + pathlen + 1);
114 if (pathlen)
115 info->prev = &dummy;
118 char *make_traverse_path(char *path, const struct traverse_info *info, const struct name_entry *n)
120 int len = tree_entry_len(n);
121 int pathlen = info->pathlen;
123 path[pathlen + len] = 0;
124 for (;;) {
125 memcpy(path + pathlen, n->path, len);
126 if (!pathlen)
127 break;
128 path[--pathlen] = '/';
129 n = &info->name;
130 len = tree_entry_len(n);
131 info = info->prev;
132 pathlen -= len;
134 return path;
137 struct tree_desc_skip {
138 struct tree_desc_skip *prev;
139 const void *ptr;
142 struct tree_desc_x {
143 struct tree_desc d;
144 struct tree_desc_skip *skip;
147 static int name_compare(const char *a, int a_len,
148 const char *b, int b_len)
150 int len = (a_len < b_len) ? a_len : b_len;
151 int cmp = memcmp(a, b, len);
152 if (cmp)
153 return cmp;
154 return (a_len - b_len);
157 static int check_entry_match(const char *a, int a_len, const char *b, int b_len)
160 * The caller wants to pick *a* from a tree or nothing.
161 * We are looking at *b* in a tree.
163 * (0) If a and b are the same name, we are trivially happy.
165 * There are three possibilities where *a* could be hiding
166 * behind *b*.
168 * (1) *a* == "t", *b* == "ab" i.e. *b* sorts earlier than *a* no
169 * matter what.
170 * (2) *a* == "t", *b* == "t-2" and "t" is a subtree in the tree;
171 * (3) *a* == "t-2", *b* == "t" and "t-2" is a blob in the tree.
173 * Otherwise we know *a* won't appear in the tree without
174 * scanning further.
177 int cmp = name_compare(a, a_len, b, b_len);
179 /* Most common case first -- reading sync'd trees */
180 if (!cmp)
181 return cmp;
183 if (0 < cmp) {
184 /* a comes after b; it does not matter if it is case (3)
185 if (b_len < a_len && !memcmp(a, b, b_len) && a[b_len] < '/')
186 return 1;
188 return 1; /* keep looking */
191 /* b comes after a; are we looking at case (2)? */
192 if (a_len < b_len && !memcmp(a, b, a_len) && b[a_len] < '/')
193 return 1; /* keep looking */
195 return -1; /* a cannot appear in the tree */
199 * From the extended tree_desc, extract the first name entry, while
200 * paying attention to the candidate "first" name. Most importantly,
201 * when looking for an entry, if there are entries that sorts earlier
202 * in the tree object representation than that name, skip them and
203 * process the named entry first. We will remember that we haven't
204 * processed the first entry yet, and in the later call skip the
205 * entry we processed early when update_extended_entry() is called.
207 * E.g. if the underlying tree object has these entries:
209 * blob "t-1"
210 * blob "t-2"
211 * tree "t"
212 * blob "t=1"
214 * and the "first" asks for "t", remember that we still need to
215 * process "t-1" and "t-2" but extract "t". After processing the
216 * entry "t" from this call, the caller will let us know by calling
217 * update_extended_entry() that we can remember "t" has been processed
218 * already.
221 static void extended_entry_extract(struct tree_desc_x *t,
222 struct name_entry *a,
223 const char *first,
224 int first_len)
226 const char *path;
227 int len;
228 struct tree_desc probe;
229 struct tree_desc_skip *skip;
232 * Extract the first entry from the tree_desc, but skip the
233 * ones that we already returned in earlier rounds.
235 while (1) {
236 if (!t->d.size) {
237 entry_clear(a);
238 break; /* not found */
240 entry_extract(&t->d, a);
241 for (skip = t->skip; skip; skip = skip->prev)
242 if (a->path == skip->ptr)
243 break; /* found */
244 if (!skip)
245 break;
246 /* We have processed this entry already. */
247 update_tree_entry(&t->d);
250 if (!first || !a->path)
251 return;
254 * The caller wants "first" from this tree, or nothing.
256 path = a->path;
257 len = tree_entry_len(a);
258 switch (check_entry_match(first, first_len, path, len)) {
259 case -1:
260 entry_clear(a);
261 case 0:
262 return;
263 default:
264 break;
268 * We need to look-ahead -- we suspect that a subtree whose
269 * name is "first" may be hiding behind the current entry "path".
271 probe = t->d;
272 while (probe.size) {
273 entry_extract(&probe, a);
274 path = a->path;
275 len = tree_entry_len(a);
276 switch (check_entry_match(first, first_len, path, len)) {
277 case -1:
278 entry_clear(a);
279 case 0:
280 return;
281 default:
282 update_tree_entry(&probe);
283 break;
285 /* keep looking */
287 entry_clear(a);
290 static void update_extended_entry(struct tree_desc_x *t, struct name_entry *a)
292 if (t->d.entry.path == a->path) {
293 update_tree_entry(&t->d);
294 } else {
295 /* we have returned this entry early */
296 struct tree_desc_skip *skip = xmalloc(sizeof(*skip));
297 skip->ptr = a->path;
298 skip->prev = t->skip;
299 t->skip = skip;
303 static void free_extended_entry(struct tree_desc_x *t)
305 struct tree_desc_skip *p, *s;
307 for (s = t->skip; s; s = p) {
308 p = s->prev;
309 free(s);
313 static inline int prune_traversal(struct name_entry *e,
314 struct traverse_info *info,
315 struct strbuf *base,
316 int still_interesting)
318 if (!info->pathspec || still_interesting == 2)
319 return 2;
320 if (still_interesting < 0)
321 return still_interesting;
322 return tree_entry_interesting(e, base, 0, info->pathspec);
325 int traverse_trees(int n, struct tree_desc *t, struct traverse_info *info)
327 int error = 0;
328 struct name_entry *entry = xmalloc(n*sizeof(*entry));
329 int i;
330 struct tree_desc_x *tx = xcalloc(n, sizeof(*tx));
331 struct strbuf base = STRBUF_INIT;
332 int interesting = 1;
334 for (i = 0; i < n; i++)
335 tx[i].d = t[i];
337 if (info->prev) {
338 strbuf_grow(&base, info->pathlen);
339 make_traverse_path(base.buf, info->prev, &info->name);
340 base.buf[info->pathlen-1] = '/';
341 strbuf_setlen(&base, info->pathlen);
343 for (;;) {
344 int trees_used;
345 unsigned long mask, dirmask;
346 const char *first = NULL;
347 int first_len = 0;
348 struct name_entry *e = NULL;
349 int len;
351 for (i = 0; i < n; i++) {
352 e = entry + i;
353 extended_entry_extract(tx + i, e, NULL, 0);
357 * A tree may have "t-2" at the current location even
358 * though it may have "t" that is a subtree behind it,
359 * and another tree may return "t". We want to grab
360 * all "t" from all trees to match in such a case.
362 for (i = 0; i < n; i++) {
363 e = entry + i;
364 if (!e->path)
365 continue;
366 len = tree_entry_len(e);
367 if (!first) {
368 first = e->path;
369 first_len = len;
370 continue;
372 if (name_compare(e->path, len, first, first_len) < 0) {
373 first = e->path;
374 first_len = len;
378 if (first) {
379 for (i = 0; i < n; i++) {
380 e = entry + i;
381 extended_entry_extract(tx + i, e, first, first_len);
382 /* Cull the ones that are not the earliest */
383 if (!e->path)
384 continue;
385 len = tree_entry_len(e);
386 if (name_compare(e->path, len, first, first_len))
387 entry_clear(e);
391 /* Now we have in entry[i] the earliest name from the trees */
392 mask = 0;
393 dirmask = 0;
394 for (i = 0; i < n; i++) {
395 if (!entry[i].path)
396 continue;
397 mask |= 1ul << i;
398 if (S_ISDIR(entry[i].mode))
399 dirmask |= 1ul << i;
400 e = &entry[i];
402 if (!mask)
403 break;
404 interesting = prune_traversal(e, info, &base, interesting);
405 if (interesting < 0)
406 break;
407 if (interesting) {
408 trees_used = info->fn(n, mask, dirmask, entry, info);
409 if (trees_used < 0) {
410 error = trees_used;
411 if (!info->show_all_errors)
412 break;
414 mask &= trees_used;
416 for (i = 0; i < n; i++)
417 if (mask & (1ul << i))
418 update_extended_entry(tx + i, entry + i);
420 free(entry);
421 for (i = 0; i < n; i++)
422 free_extended_entry(tx + i);
423 free(tx);
424 strbuf_release(&base);
425 return error;
428 static int find_tree_entry(struct tree_desc *t, const char *name, unsigned char *result, unsigned *mode)
430 int namelen = strlen(name);
431 while (t->size) {
432 const char *entry;
433 const unsigned char *sha1;
434 int entrylen, cmp;
436 sha1 = tree_entry_extract(t, &entry, mode);
437 entrylen = tree_entry_len(&t->entry);
438 update_tree_entry(t);
439 if (entrylen > namelen)
440 continue;
441 cmp = memcmp(name, entry, entrylen);
442 if (cmp > 0)
443 continue;
444 if (cmp < 0)
445 break;
446 if (entrylen == namelen) {
447 hashcpy(result, sha1);
448 return 0;
450 if (name[entrylen] != '/')
451 continue;
452 if (!S_ISDIR(*mode))
453 break;
454 if (++entrylen == namelen) {
455 hashcpy(result, sha1);
456 return 0;
458 return get_tree_entry(sha1, name + entrylen, result, mode);
460 return -1;
463 int get_tree_entry(const unsigned char *tree_sha1, const char *name, unsigned char *sha1, unsigned *mode)
465 int retval;
466 void *tree;
467 unsigned long size;
468 unsigned char root[20];
470 tree = read_object_with_reference(tree_sha1, tree_type, &size, root);
471 if (!tree)
472 return -1;
474 if (name[0] == '\0') {
475 hashcpy(sha1, root);
476 free(tree);
477 return 0;
480 if (!size) {
481 retval = -1;
482 } else {
483 struct tree_desc t;
484 init_tree_desc(&t, tree, size);
485 retval = find_tree_entry(&t, name, sha1, mode);
487 free(tree);
488 return retval;
491 static int match_entry(const struct pathspec_item *item,
492 const struct name_entry *entry, int pathlen,
493 const char *match, int matchlen,
494 enum interesting *never_interesting)
496 int m = -1; /* signals that we haven't called strncmp() */
498 if (item->magic & PATHSPEC_ICASE)
500 * "Never interesting" trick requires exact
501 * matching. We could do something clever with inexact
502 * matching, but it's trickier (and not to forget that
503 * strcasecmp is locale-dependent, at least in
504 * glibc). Just disable it for now. It can't be worse
505 * than the wildcard's codepath of '[Tt][Hi][Is][Ss]'
506 * pattern.
508 *never_interesting = entry_not_interesting;
509 else if (*never_interesting != entry_not_interesting) {
511 * We have not seen any match that sorts later
512 * than the current path.
516 * Does match sort strictly earlier than path
517 * with their common parts?
519 m = strncmp(match, entry->path,
520 (matchlen < pathlen) ? matchlen : pathlen);
521 if (m < 0)
522 return 0;
525 * If we come here even once, that means there is at
526 * least one pathspec that would sort equal to or
527 * later than the path we are currently looking at.
528 * In other words, if we have never reached this point
529 * after iterating all pathspecs, it means all
530 * pathspecs are either outside of base, or inside the
531 * base but sorts strictly earlier than the current
532 * one. In either case, they will never match the
533 * subsequent entries. In such a case, we initialized
534 * the variable to -1 and that is what will be
535 * returned, allowing the caller to terminate early.
537 *never_interesting = entry_not_interesting;
540 if (pathlen > matchlen)
541 return 0;
543 if (matchlen > pathlen) {
544 if (match[pathlen] != '/')
545 return 0;
546 if (!S_ISDIR(entry->mode) && !S_ISGITLINK(entry->mode))
547 return 0;
550 if (m == -1)
552 * we cheated and did not do strncmp(), so we do
553 * that here.
555 m = ps_strncmp(item, match, entry->path, pathlen);
558 * If common part matched earlier then it is a hit,
559 * because we rejected the case where path is not a
560 * leading directory and is shorter than match.
562 if (!m)
564 * match_entry does not check if the prefix part is
565 * matched case-sensitively. If the entry is a
566 * directory and part of prefix, it'll be rematched
567 * eventually by basecmp with special treatment for
568 * the prefix.
570 return 1;
572 return 0;
575 /* :(icase)-aware string compare */
576 static int basecmp(const struct pathspec_item *item,
577 const char *base, const char *match, int len)
579 if (item->magic & PATHSPEC_ICASE) {
580 int ret, n = len > item->prefix ? item->prefix : len;
581 ret = strncmp(base, match, n);
582 if (ret)
583 return ret;
584 base += n;
585 match += n;
586 len -= n;
588 return ps_strncmp(item, base, match, len);
591 static int match_dir_prefix(const struct pathspec_item *item,
592 const char *base,
593 const char *match, int matchlen)
595 if (basecmp(item, base, match, matchlen))
596 return 0;
599 * If the base is a subdirectory of a path which
600 * was specified, all of them are interesting.
602 if (!matchlen ||
603 base[matchlen] == '/' ||
604 match[matchlen - 1] == '/')
605 return 1;
607 /* Just a random prefix match */
608 return 0;
612 * Perform matching on the leading non-wildcard part of
613 * pathspec. item->nowildcard_len must be greater than zero. Return
614 * non-zero if base is matched.
616 static int match_wildcard_base(const struct pathspec_item *item,
617 const char *base, int baselen,
618 int *matched)
620 const char *match = item->match;
621 /* the wildcard part is not considered in this function */
622 int matchlen = item->nowildcard_len;
624 if (baselen) {
625 int dirlen;
627 * Return early if base is longer than the
628 * non-wildcard part but it does not match.
630 if (baselen >= matchlen) {
631 *matched = matchlen;
632 return !basecmp(item, base, match, matchlen);
635 dirlen = matchlen;
636 while (dirlen && match[dirlen - 1] != '/')
637 dirlen--;
640 * Return early if base is shorter than the
641 * non-wildcard part but it does not match. Note that
642 * base ends with '/' so we are sure it really matches
643 * directory
645 if (basecmp(item, base, match, baselen))
646 return 0;
647 *matched = baselen;
648 } else
649 *matched = 0;
651 * we could have checked entry against the non-wildcard part
652 * that is not in base and does similar never_interesting
653 * optimization as in match_entry. For now just be happy with
654 * base comparison.
656 return entry_interesting;
660 * Is a tree entry interesting given the pathspec we have?
662 * Pre-condition: either baselen == base_offset (i.e. empty path)
663 * or base[baselen-1] == '/' (i.e. with trailing slash).
665 static enum interesting do_match(const struct name_entry *entry,
666 struct strbuf *base, int base_offset,
667 const struct pathspec *ps,
668 int exclude)
670 int i;
671 int pathlen, baselen = base->len - base_offset;
672 enum interesting never_interesting = ps->has_wildcard ?
673 entry_not_interesting : all_entries_not_interesting;
675 GUARD_PATHSPEC(ps,
676 PATHSPEC_FROMTOP |
677 PATHSPEC_MAXDEPTH |
678 PATHSPEC_LITERAL |
679 PATHSPEC_GLOB |
680 PATHSPEC_ICASE |
681 PATHSPEC_EXCLUDE);
683 if (!ps->nr) {
684 if (!ps->recursive ||
685 !(ps->magic & PATHSPEC_MAXDEPTH) ||
686 ps->max_depth == -1)
687 return all_entries_interesting;
688 return within_depth(base->buf + base_offset, baselen,
689 !!S_ISDIR(entry->mode),
690 ps->max_depth) ?
691 entry_interesting : entry_not_interesting;
694 pathlen = tree_entry_len(entry);
696 for (i = ps->nr - 1; i >= 0; i--) {
697 const struct pathspec_item *item = ps->items+i;
698 const char *match = item->match;
699 const char *base_str = base->buf + base_offset;
700 int matchlen = item->len, matched = 0;
702 if ((!exclude && item->magic & PATHSPEC_EXCLUDE) ||
703 ( exclude && !(item->magic & PATHSPEC_EXCLUDE)))
704 continue;
706 if (baselen >= matchlen) {
707 /* If it doesn't match, move along... */
708 if (!match_dir_prefix(item, base_str, match, matchlen))
709 goto match_wildcards;
711 if (!ps->recursive ||
712 !(ps->magic & PATHSPEC_MAXDEPTH) ||
713 ps->max_depth == -1)
714 return all_entries_interesting;
716 return within_depth(base_str + matchlen + 1,
717 baselen - matchlen - 1,
718 !!S_ISDIR(entry->mode),
719 ps->max_depth) ?
720 entry_interesting : entry_not_interesting;
723 /* Either there must be no base, or the base must match. */
724 if (baselen == 0 || !basecmp(item, base_str, match, baselen)) {
725 if (match_entry(item, entry, pathlen,
726 match + baselen, matchlen - baselen,
727 &never_interesting))
728 return entry_interesting;
730 if (item->nowildcard_len < item->len) {
731 if (!git_fnmatch(item, match + baselen, entry->path,
732 item->nowildcard_len - baselen))
733 return entry_interesting;
736 * Match all directories. We'll try to
737 * match files later on.
739 if (ps->recursive && S_ISDIR(entry->mode))
740 return entry_interesting;
743 continue;
746 match_wildcards:
747 if (item->nowildcard_len == item->len)
748 continue;
750 if (item->nowildcard_len &&
751 !match_wildcard_base(item, base_str, baselen, &matched))
752 continue;
755 * Concatenate base and entry->path into one and do
756 * fnmatch() on it.
758 * While we could avoid concatenation in certain cases
759 * [1], which saves a memcpy and potentially a
760 * realloc, it turns out not worth it. Measurement on
761 * linux-2.6 does not show any clear improvements,
762 * partly because of the nowildcard_len optimization
763 * in git_fnmatch(). Avoid micro-optimizations here.
765 * [1] if match_wildcard_base() says the base
766 * directory is already matched, we only need to match
767 * the rest, which is shorter so _in theory_ faster.
770 strbuf_add(base, entry->path, pathlen);
772 if (!git_fnmatch(item, match, base->buf + base_offset,
773 item->nowildcard_len)) {
774 strbuf_setlen(base, base_offset + baselen);
775 return entry_interesting;
777 strbuf_setlen(base, base_offset + baselen);
780 * Match all directories. We'll try to match files
781 * later on.
782 * max_depth is ignored but we may consider support it
783 * in future, see
784 * http://thread.gmane.org/gmane.comp.version-control.git/163757/focus=163840
786 if (ps->recursive && S_ISDIR(entry->mode))
787 return entry_interesting;
789 return never_interesting; /* No matches */
793 * Is a tree entry interesting given the pathspec we have?
795 * Pre-condition: either baselen == base_offset (i.e. empty path)
796 * or base[baselen-1] == '/' (i.e. with trailing slash).
798 enum interesting tree_entry_interesting(const struct name_entry *entry,
799 struct strbuf *base, int base_offset,
800 const struct pathspec *ps)
802 enum interesting positive, negative;
803 positive = do_match(entry, base, base_offset, ps, 0);
806 * case | entry | positive | negative | result
807 * -----+-------+----------+----------+-------
808 * 1 | file | -1 | -1..2 | -1
809 * 2 | file | 0 | -1..2 | 0
810 * 3 | file | 1 | -1 | 1
811 * 4 | file | 1 | 0 | 1
812 * 5 | file | 1 | 1 | 0
813 * 6 | file | 1 | 2 | 0
814 * 7 | file | 2 | -1 | 2
815 * 8 | file | 2 | 0 | 2
816 * 9 | file | 2 | 1 | 0
817 * 10 | file | 2 | 2 | -1
818 * -----+-------+----------+----------+-------
819 * 11 | dir | -1 | -1..2 | -1
820 * 12 | dir | 0 | -1..2 | 0
821 * 13 | dir | 1 | -1 | 1
822 * 14 | dir | 1 | 0 | 1
823 * 15 | dir | 1 | 1 | 1 (*)
824 * 16 | dir | 1 | 2 | 0
825 * 17 | dir | 2 | -1 | 2
826 * 18 | dir | 2 | 0 | 2
827 * 19 | dir | 2 | 1 | 1 (*)
828 * 20 | dir | 2 | 2 | -1
830 * (*) An exclude pattern interested in a directory does not
831 * necessarily mean it will exclude all of the directory. In
832 * wildcard case, it can't decide until looking at individual
833 * files inside. So don't write such directories off yet.
836 if (!(ps->magic & PATHSPEC_EXCLUDE) ||
837 positive <= entry_not_interesting) /* #1, #2, #11, #12 */
838 return positive;
840 negative = do_match(entry, base, base_offset, ps, 1);
842 /* #3, #4, #7, #8, #13, #14, #17, #18 */
843 if (negative <= entry_not_interesting)
844 return positive;
846 /* #15, #19 */
847 if (S_ISDIR(entry->mode) &&
848 positive >= entry_interesting &&
849 negative == entry_interesting)
850 return entry_interesting;
852 if ((positive == entry_interesting &&
853 negative >= entry_interesting) || /* #5, #6, #16 */
854 (positive == all_entries_interesting &&
855 negative == entry_interesting)) /* #9 */
856 return entry_not_interesting;
858 return all_entries_not_interesting; /* #10, #20 */