]>
Commit | Line | Data |
---|---|---|
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 void decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size) | |
26 | { | |
27 | const char *path; | |
28 | unsigned int mode, len; | |
29 | ||
30 | if (size < 24 || buf[size - 21]) | |
31 | die("corrupt tree file"); | |
32 | ||
33 | path = get_mode(buf, &mode); | |
34 | if (!path || !*path) | |
35 | die("corrupt tree file"); | |
36 | len = strlen(path) + 1; | |
37 | ||
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); | |
42 | } | |
43 | ||
44 | void init_tree_desc(struct tree_desc *desc, const void *buffer, unsigned long size) | |
45 | { | |
46 | desc->buffer = buffer; | |
47 | desc->size = size; | |
48 | if (size) | |
49 | decode_tree_entry(desc, buffer, size); | |
50 | } | |
51 | ||
52 | void *fill_tree_descriptor(struct tree_desc *desc, const unsigned char *sha1) | |
53 | { | |
54 | unsigned long size = 0; | |
55 | void *buf = NULL; | |
56 | ||
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)); | |
61 | } | |
62 | init_tree_desc(desc, buf, size); | |
63 | return buf; | |
64 | } | |
65 | ||
66 | static void entry_clear(struct name_entry *a) | |
67 | { | |
68 | memset(a, 0, sizeof(*a)); | |
69 | } | |
70 | ||
71 | static void entry_extract(struct tree_desc *t, struct name_entry *a) | |
72 | { | |
73 | *a = t->entry; | |
74 | } | |
75 | ||
76 | void update_tree_entry(struct tree_desc *desc) | |
77 | { | |
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; | |
82 | ||
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); | |
91 | } | |
92 | ||
93 | int tree_entry(struct tree_desc *desc, struct name_entry *entry) | |
94 | { | |
95 | if (!desc->size) | |
96 | return 0; | |
97 | ||
98 | *entry = desc->entry; | |
99 | update_tree_entry(desc); | |
100 | return 1; | |
101 | } | |
102 | ||
103 | void setup_traverse_info(struct traverse_info *info, const char *base) | |
104 | { | |
105 | int pathlen = strlen(base); | |
106 | static struct traverse_info dummy; | |
107 | ||
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; | |
116 | } | |
117 | ||
118 | char *make_traverse_path(char *path, const struct traverse_info *info, const struct name_entry *n) | |
119 | { | |
120 | int len = tree_entry_len(n); | |
121 | int pathlen = info->pathlen; | |
122 | ||
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; | |
133 | } | |
134 | return path; | |
135 | } | |
136 | ||
137 | struct tree_desc_skip { | |
138 | struct tree_desc_skip *prev; | |
139 | const void *ptr; | |
140 | }; | |
141 | ||
142 | struct tree_desc_x { | |
143 | struct tree_desc d; | |
144 | struct tree_desc_skip *skip; | |
145 | }; | |
146 | ||
147 | static int check_entry_match(const char *a, int a_len, const char *b, int b_len) | |
148 | { | |
149 | /* | |
150 | * The caller wants to pick *a* from a tree or nothing. | |
151 | * We are looking at *b* in a tree. | |
152 | * | |
153 | * (0) If a and b are the same name, we are trivially happy. | |
154 | * | |
155 | * There are three possibilities where *a* could be hiding | |
156 | * behind *b*. | |
157 | * | |
158 | * (1) *a* == "t", *b* == "ab" i.e. *b* sorts earlier than *a* no | |
159 | * matter what. | |
160 | * (2) *a* == "t", *b* == "t-2" and "t" is a subtree in the tree; | |
161 | * (3) *a* == "t-2", *b* == "t" and "t-2" is a blob in the tree. | |
162 | * | |
163 | * Otherwise we know *a* won't appear in the tree without | |
164 | * scanning further. | |
165 | */ | |
166 | ||
167 | int cmp = name_compare(a, a_len, b, b_len); | |
168 | ||
169 | /* Most common case first -- reading sync'd trees */ | |
170 | if (!cmp) | |
171 | return cmp; | |
172 | ||
173 | if (0 < cmp) { | |
174 | /* a comes after b; it does not matter if it is case (3) | |
175 | if (b_len < a_len && !memcmp(a, b, b_len) && a[b_len] < '/') | |
176 | return 1; | |
177 | */ | |
178 | return 1; /* keep looking */ | |
179 | } | |
180 | ||
181 | /* b comes after a; are we looking at case (2)? */ | |
182 | if (a_len < b_len && !memcmp(a, b, a_len) && b[a_len] < '/') | |
183 | return 1; /* keep looking */ | |
184 | ||
185 | return -1; /* a cannot appear in the tree */ | |
186 | } | |
187 | ||
188 | /* | |
189 | * From the extended tree_desc, extract the first name entry, while | |
190 | * paying attention to the candidate "first" name. Most importantly, | |
191 | * when looking for an entry, if there are entries that sorts earlier | |
192 | * in the tree object representation than that name, skip them and | |
193 | * process the named entry first. We will remember that we haven't | |
194 | * processed the first entry yet, and in the later call skip the | |
195 | * entry we processed early when update_extended_entry() is called. | |
196 | * | |
197 | * E.g. if the underlying tree object has these entries: | |
198 | * | |
199 | * blob "t-1" | |
200 | * blob "t-2" | |
201 | * tree "t" | |
202 | * blob "t=1" | |
203 | * | |
204 | * and the "first" asks for "t", remember that we still need to | |
205 | * process "t-1" and "t-2" but extract "t". After processing the | |
206 | * entry "t" from this call, the caller will let us know by calling | |
207 | * update_extended_entry() that we can remember "t" has been processed | |
208 | * already. | |
209 | */ | |
210 | ||
211 | static void extended_entry_extract(struct tree_desc_x *t, | |
212 | struct name_entry *a, | |
213 | const char *first, | |
214 | int first_len) | |
215 | { | |
216 | const char *path; | |
217 | int len; | |
218 | struct tree_desc probe; | |
219 | struct tree_desc_skip *skip; | |
220 | ||
221 | /* | |
222 | * Extract the first entry from the tree_desc, but skip the | |
223 | * ones that we already returned in earlier rounds. | |
224 | */ | |
225 | while (1) { | |
226 | if (!t->d.size) { | |
227 | entry_clear(a); | |
228 | break; /* not found */ | |
229 | } | |
230 | entry_extract(&t->d, a); | |
231 | for (skip = t->skip; skip; skip = skip->prev) | |
232 | if (a->path == skip->ptr) | |
233 | break; /* found */ | |
234 | if (!skip) | |
235 | break; | |
236 | /* We have processed this entry already. */ | |
237 | update_tree_entry(&t->d); | |
238 | } | |
239 | ||
240 | if (!first || !a->path) | |
241 | return; | |
242 | ||
243 | /* | |
244 | * The caller wants "first" from this tree, or nothing. | |
245 | */ | |
246 | path = a->path; | |
247 | len = tree_entry_len(a); | |
248 | switch (check_entry_match(first, first_len, path, len)) { | |
249 | case -1: | |
250 | entry_clear(a); | |
251 | case 0: | |
252 | return; | |
253 | default: | |
254 | break; | |
255 | } | |
256 | ||
257 | /* | |
258 | * We need to look-ahead -- we suspect that a subtree whose | |
259 | * name is "first" may be hiding behind the current entry "path". | |
260 | */ | |
261 | probe = t->d; | |
262 | while (probe.size) { | |
263 | entry_extract(&probe, a); | |
264 | path = a->path; | |
265 | len = tree_entry_len(a); | |
266 | switch (check_entry_match(first, first_len, path, len)) { | |
267 | case -1: | |
268 | entry_clear(a); | |
269 | case 0: | |
270 | return; | |
271 | default: | |
272 | update_tree_entry(&probe); | |
273 | break; | |
274 | } | |
275 | /* keep looking */ | |
276 | } | |
277 | entry_clear(a); | |
278 | } | |
279 | ||
280 | static void update_extended_entry(struct tree_desc_x *t, struct name_entry *a) | |
281 | { | |
282 | if (t->d.entry.path == a->path) { | |
283 | update_tree_entry(&t->d); | |
284 | } else { | |
285 | /* we have returned this entry early */ | |
286 | struct tree_desc_skip *skip = xmalloc(sizeof(*skip)); | |
287 | skip->ptr = a->path; | |
288 | skip->prev = t->skip; | |
289 | t->skip = skip; | |
290 | } | |
291 | } | |
292 | ||
293 | static void free_extended_entry(struct tree_desc_x *t) | |
294 | { | |
295 | struct tree_desc_skip *p, *s; | |
296 | ||
297 | for (s = t->skip; s; s = p) { | |
298 | p = s->prev; | |
299 | free(s); | |
300 | } | |
301 | } | |
302 | ||
303 | static inline int prune_traversal(struct name_entry *e, | |
304 | struct traverse_info *info, | |
305 | struct strbuf *base, | |
306 | int still_interesting) | |
307 | { | |
308 | if (!info->pathspec || still_interesting == 2) | |
309 | return 2; | |
310 | if (still_interesting < 0) | |
311 | return still_interesting; | |
312 | return tree_entry_interesting(e, base, 0, info->pathspec); | |
313 | } | |
314 | ||
315 | int traverse_trees(int n, struct tree_desc *t, struct traverse_info *info) | |
316 | { | |
317 | int error = 0; | |
318 | struct name_entry *entry = xmalloc(n*sizeof(*entry)); | |
319 | int i; | |
320 | struct tree_desc_x *tx = xcalloc(n, sizeof(*tx)); | |
321 | struct strbuf base = STRBUF_INIT; | |
322 | int interesting = 1; | |
323 | ||
324 | for (i = 0; i < n; i++) | |
325 | tx[i].d = t[i]; | |
326 | ||
327 | if (info->prev) { | |
328 | strbuf_grow(&base, info->pathlen); | |
329 | make_traverse_path(base.buf, info->prev, &info->name); | |
330 | base.buf[info->pathlen-1] = '/'; | |
331 | strbuf_setlen(&base, info->pathlen); | |
332 | } | |
333 | for (;;) { | |
334 | int trees_used; | |
335 | unsigned long mask, dirmask; | |
336 | const char *first = NULL; | |
337 | int first_len = 0; | |
338 | struct name_entry *e = NULL; | |
339 | int len; | |
340 | ||
341 | for (i = 0; i < n; i++) { | |
342 | e = entry + i; | |
343 | extended_entry_extract(tx + i, e, NULL, 0); | |
344 | } | |
345 | ||
346 | /* | |
347 | * A tree may have "t-2" at the current location even | |
348 | * though it may have "t" that is a subtree behind it, | |
349 | * and another tree may return "t". We want to grab | |
350 | * all "t" from all trees to match in such a case. | |
351 | */ | |
352 | for (i = 0; i < n; i++) { | |
353 | e = entry + i; | |
354 | if (!e->path) | |
355 | continue; | |
356 | len = tree_entry_len(e); | |
357 | if (!first) { | |
358 | first = e->path; | |
359 | first_len = len; | |
360 | continue; | |
361 | } | |
362 | if (name_compare(e->path, len, first, first_len) < 0) { | |
363 | first = e->path; | |
364 | first_len = len; | |
365 | } | |
366 | } | |
367 | ||
368 | if (first) { | |
369 | for (i = 0; i < n; i++) { | |
370 | e = entry + i; | |
371 | extended_entry_extract(tx + i, e, first, first_len); | |
372 | /* Cull the ones that are not the earliest */ | |
373 | if (!e->path) | |
374 | continue; | |
375 | len = tree_entry_len(e); | |
376 | if (name_compare(e->path, len, first, first_len)) | |
377 | entry_clear(e); | |
378 | } | |
379 | } | |
380 | ||
381 | /* Now we have in entry[i] the earliest name from the trees */ | |
382 | mask = 0; | |
383 | dirmask = 0; | |
384 | for (i = 0; i < n; i++) { | |
385 | if (!entry[i].path) | |
386 | continue; | |
387 | mask |= 1ul << i; | |
388 | if (S_ISDIR(entry[i].mode)) | |
389 | dirmask |= 1ul << i; | |
390 | e = &entry[i]; | |
391 | } | |
392 | if (!mask) | |
393 | break; | |
394 | interesting = prune_traversal(e, info, &base, interesting); | |
395 | if (interesting < 0) | |
396 | break; | |
397 | if (interesting) { | |
398 | trees_used = info->fn(n, mask, dirmask, entry, info); | |
399 | if (trees_used < 0) { | |
400 | error = trees_used; | |
401 | if (!info->show_all_errors) | |
402 | break; | |
403 | } | |
404 | mask &= trees_used; | |
405 | } | |
406 | for (i = 0; i < n; i++) | |
407 | if (mask & (1ul << i)) | |
408 | update_extended_entry(tx + i, entry + i); | |
409 | } | |
410 | free(entry); | |
411 | for (i = 0; i < n; i++) | |
412 | free_extended_entry(tx + i); | |
413 | free(tx); | |
414 | strbuf_release(&base); | |
415 | return error; | |
416 | } | |
417 | ||
418 | static int find_tree_entry(struct tree_desc *t, const char *name, unsigned char *result, unsigned *mode) | |
419 | { | |
420 | int namelen = strlen(name); | |
421 | while (t->size) { | |
422 | const char *entry; | |
423 | const unsigned char *sha1; | |
424 | int entrylen, cmp; | |
425 | ||
426 | sha1 = tree_entry_extract(t, &entry, mode); | |
427 | entrylen = tree_entry_len(&t->entry); | |
428 | update_tree_entry(t); | |
429 | if (entrylen > namelen) | |
430 | continue; | |
431 | cmp = memcmp(name, entry, entrylen); | |
432 | if (cmp > 0) | |
433 | continue; | |
434 | if (cmp < 0) | |
435 | break; | |
436 | if (entrylen == namelen) { | |
437 | hashcpy(result, sha1); | |
438 | return 0; | |
439 | } | |
440 | if (name[entrylen] != '/') | |
441 | continue; | |
442 | if (!S_ISDIR(*mode)) | |
443 | break; | |
444 | if (++entrylen == namelen) { | |
445 | hashcpy(result, sha1); | |
446 | return 0; | |
447 | } | |
448 | return get_tree_entry(sha1, name + entrylen, result, mode); | |
449 | } | |
450 | return -1; | |
451 | } | |
452 | ||
453 | int get_tree_entry(const unsigned char *tree_sha1, const char *name, unsigned char *sha1, unsigned *mode) | |
454 | { | |
455 | int retval; | |
456 | void *tree; | |
457 | unsigned long size; | |
458 | unsigned char root[20]; | |
459 | ||
460 | tree = read_object_with_reference(tree_sha1, tree_type, &size, root); | |
461 | if (!tree) | |
462 | return -1; | |
463 | ||
464 | if (name[0] == '\0') { | |
465 | hashcpy(sha1, root); | |
466 | free(tree); | |
467 | return 0; | |
468 | } | |
469 | ||
470 | if (!size) { | |
471 | retval = -1; | |
472 | } else { | |
473 | struct tree_desc t; | |
474 | init_tree_desc(&t, tree, size); | |
475 | retval = find_tree_entry(&t, name, sha1, mode); | |
476 | } | |
477 | free(tree); | |
478 | return retval; | |
479 | } | |
480 | ||
481 | static int match_entry(const struct pathspec_item *item, | |
482 | const struct name_entry *entry, int pathlen, | |
483 | const char *match, int matchlen, | |
484 | enum interesting *never_interesting) | |
485 | { | |
486 | int m = -1; /* signals that we haven't called strncmp() */ | |
487 | ||
488 | if (item->magic & PATHSPEC_ICASE) | |
489 | /* | |
490 | * "Never interesting" trick requires exact | |
491 | * matching. We could do something clever with inexact | |
492 | * matching, but it's trickier (and not to forget that | |
493 | * strcasecmp is locale-dependent, at least in | |
494 | * glibc). Just disable it for now. It can't be worse | |
495 | * than the wildcard's codepath of '[Tt][Hi][Is][Ss]' | |
496 | * pattern. | |
497 | */ | |
498 | *never_interesting = entry_not_interesting; | |
499 | else if (*never_interesting != entry_not_interesting) { | |
500 | /* | |
501 | * We have not seen any match that sorts later | |
502 | * than the current path. | |
503 | */ | |
504 | ||
505 | /* | |
506 | * Does match sort strictly earlier than path | |
507 | * with their common parts? | |
508 | */ | |
509 | m = strncmp(match, entry->path, | |
510 | (matchlen < pathlen) ? matchlen : pathlen); | |
511 | if (m < 0) | |
512 | return 0; | |
513 | ||
514 | /* | |
515 | * If we come here even once, that means there is at | |
516 | * least one pathspec that would sort equal to or | |
517 | * later than the path we are currently looking at. | |
518 | * In other words, if we have never reached this point | |
519 | * after iterating all pathspecs, it means all | |
520 | * pathspecs are either outside of base, or inside the | |
521 | * base but sorts strictly earlier than the current | |
522 | * one. In either case, they will never match the | |
523 | * subsequent entries. In such a case, we initialized | |
524 | * the variable to -1 and that is what will be | |
525 | * returned, allowing the caller to terminate early. | |
526 | */ | |
527 | *never_interesting = entry_not_interesting; | |
528 | } | |
529 | ||
530 | if (pathlen > matchlen) | |
531 | return 0; | |
532 | ||
533 | if (matchlen > pathlen) { | |
534 | if (match[pathlen] != '/') | |
535 | return 0; | |
536 | if (!S_ISDIR(entry->mode) && !S_ISGITLINK(entry->mode)) | |
537 | return 0; | |
538 | } | |
539 | ||
540 | if (m == -1) | |
541 | /* | |
542 | * we cheated and did not do strncmp(), so we do | |
543 | * that here. | |
544 | */ | |
545 | m = ps_strncmp(item, match, entry->path, pathlen); | |
546 | ||
547 | /* | |
548 | * If common part matched earlier then it is a hit, | |
549 | * because we rejected the case where path is not a | |
550 | * leading directory and is shorter than match. | |
551 | */ | |
552 | if (!m) | |
553 | /* | |
554 | * match_entry does not check if the prefix part is | |
555 | * matched case-sensitively. If the entry is a | |
556 | * directory and part of prefix, it'll be rematched | |
557 | * eventually by basecmp with special treatment for | |
558 | * the prefix. | |
559 | */ | |
560 | return 1; | |
561 | ||
562 | return 0; | |
563 | } | |
564 | ||
565 | /* :(icase)-aware string compare */ | |
566 | static int basecmp(const struct pathspec_item *item, | |
567 | const char *base, const char *match, int len) | |
568 | { | |
569 | if (item->magic & PATHSPEC_ICASE) { | |
570 | int ret, n = len > item->prefix ? item->prefix : len; | |
571 | ret = strncmp(base, match, n); | |
572 | if (ret) | |
573 | return ret; | |
574 | base += n; | |
575 | match += n; | |
576 | len -= n; | |
577 | } | |
578 | return ps_strncmp(item, base, match, len); | |
579 | } | |
580 | ||
581 | static int match_dir_prefix(const struct pathspec_item *item, | |
582 | const char *base, | |
583 | const char *match, int matchlen) | |
584 | { | |
585 | if (basecmp(item, base, match, matchlen)) | |
586 | return 0; | |
587 | ||
588 | /* | |
589 | * If the base is a subdirectory of a path which | |
590 | * was specified, all of them are interesting. | |
591 | */ | |
592 | if (!matchlen || | |
593 | base[matchlen] == '/' || | |
594 | match[matchlen - 1] == '/') | |
595 | return 1; | |
596 | ||
597 | /* Just a random prefix match */ | |
598 | return 0; | |
599 | } | |
600 | ||
601 | /* | |
602 | * Perform matching on the leading non-wildcard part of | |
603 | * pathspec. item->nowildcard_len must be greater than zero. Return | |
604 | * non-zero if base is matched. | |
605 | */ | |
606 | static int match_wildcard_base(const struct pathspec_item *item, | |
607 | const char *base, int baselen, | |
608 | int *matched) | |
609 | { | |
610 | const char *match = item->match; | |
611 | /* the wildcard part is not considered in this function */ | |
612 | int matchlen = item->nowildcard_len; | |
613 | ||
614 | if (baselen) { | |
615 | int dirlen; | |
616 | /* | |
617 | * Return early if base is longer than the | |
618 | * non-wildcard part but it does not match. | |
619 | */ | |
620 | if (baselen >= matchlen) { | |
621 | *matched = matchlen; | |
622 | return !basecmp(item, base, match, matchlen); | |
623 | } | |
624 | ||
625 | dirlen = matchlen; | |
626 | while (dirlen && match[dirlen - 1] != '/') | |
627 | dirlen--; | |
628 | ||
629 | /* | |
630 | * Return early if base is shorter than the | |
631 | * non-wildcard part but it does not match. Note that | |
632 | * base ends with '/' so we are sure it really matches | |
633 | * directory | |
634 | */ | |
635 | if (basecmp(item, base, match, baselen)) | |
636 | return 0; | |
637 | *matched = baselen; | |
638 | } else | |
639 | *matched = 0; | |
640 | /* | |
641 | * we could have checked entry against the non-wildcard part | |
642 | * that is not in base and does similar never_interesting | |
643 | * optimization as in match_entry. For now just be happy with | |
644 | * base comparison. | |
645 | */ | |
646 | return entry_interesting; | |
647 | } | |
648 | ||
649 | /* | |
650 | * Is a tree entry interesting given the pathspec we have? | |
651 | * | |
652 | * Pre-condition: either baselen == base_offset (i.e. empty path) | |
653 | * or base[baselen-1] == '/' (i.e. with trailing slash). | |
654 | */ | |
655 | static enum interesting do_match(const struct name_entry *entry, | |
656 | struct strbuf *base, int base_offset, | |
657 | const struct pathspec *ps, | |
658 | int exclude) | |
659 | { | |
660 | int i; | |
661 | int pathlen, baselen = base->len - base_offset; | |
662 | enum interesting never_interesting = ps->has_wildcard ? | |
663 | entry_not_interesting : all_entries_not_interesting; | |
664 | ||
665 | GUARD_PATHSPEC(ps, | |
666 | PATHSPEC_FROMTOP | | |
667 | PATHSPEC_MAXDEPTH | | |
668 | PATHSPEC_LITERAL | | |
669 | PATHSPEC_GLOB | | |
670 | PATHSPEC_ICASE | | |
671 | PATHSPEC_EXCLUDE); | |
672 | ||
673 | if (!ps->nr) { | |
674 | if (!ps->recursive || | |
675 | !(ps->magic & PATHSPEC_MAXDEPTH) || | |
676 | ps->max_depth == -1) | |
677 | return all_entries_interesting; | |
678 | return within_depth(base->buf + base_offset, baselen, | |
679 | !!S_ISDIR(entry->mode), | |
680 | ps->max_depth) ? | |
681 | entry_interesting : entry_not_interesting; | |
682 | } | |
683 | ||
684 | pathlen = tree_entry_len(entry); | |
685 | ||
686 | for (i = ps->nr - 1; i >= 0; i--) { | |
687 | const struct pathspec_item *item = ps->items+i; | |
688 | const char *match = item->match; | |
689 | const char *base_str = base->buf + base_offset; | |
690 | int matchlen = item->len, matched = 0; | |
691 | ||
692 | if ((!exclude && item->magic & PATHSPEC_EXCLUDE) || | |
693 | ( exclude && !(item->magic & PATHSPEC_EXCLUDE))) | |
694 | continue; | |
695 | ||
696 | if (baselen >= matchlen) { | |
697 | /* If it doesn't match, move along... */ | |
698 | if (!match_dir_prefix(item, base_str, match, matchlen)) | |
699 | goto match_wildcards; | |
700 | ||
701 | if (!ps->recursive || | |
702 | !(ps->magic & PATHSPEC_MAXDEPTH) || | |
703 | ps->max_depth == -1) | |
704 | return all_entries_interesting; | |
705 | ||
706 | return within_depth(base_str + matchlen + 1, | |
707 | baselen - matchlen - 1, | |
708 | !!S_ISDIR(entry->mode), | |
709 | ps->max_depth) ? | |
710 | entry_interesting : entry_not_interesting; | |
711 | } | |
712 | ||
713 | /* Either there must be no base, or the base must match. */ | |
714 | if (baselen == 0 || !basecmp(item, base_str, match, baselen)) { | |
715 | if (match_entry(item, entry, pathlen, | |
716 | match + baselen, matchlen - baselen, | |
717 | &never_interesting)) | |
718 | return entry_interesting; | |
719 | ||
720 | if (item->nowildcard_len < item->len) { | |
721 | if (!git_fnmatch(item, match + baselen, entry->path, | |
722 | item->nowildcard_len - baselen)) | |
723 | return entry_interesting; | |
724 | ||
725 | /* | |
726 | * Match all directories. We'll try to | |
727 | * match files later on. | |
728 | */ | |
729 | if (ps->recursive && S_ISDIR(entry->mode)) | |
730 | return entry_interesting; | |
731 | } | |
732 | ||
733 | continue; | |
734 | } | |
735 | ||
736 | match_wildcards: | |
737 | if (item->nowildcard_len == item->len) | |
738 | continue; | |
739 | ||
740 | if (item->nowildcard_len && | |
741 | !match_wildcard_base(item, base_str, baselen, &matched)) | |
742 | continue; | |
743 | ||
744 | /* | |
745 | * Concatenate base and entry->path into one and do | |
746 | * fnmatch() on it. | |
747 | * | |
748 | * While we could avoid concatenation in certain cases | |
749 | * [1], which saves a memcpy and potentially a | |
750 | * realloc, it turns out not worth it. Measurement on | |
751 | * linux-2.6 does not show any clear improvements, | |
752 | * partly because of the nowildcard_len optimization | |
753 | * in git_fnmatch(). Avoid micro-optimizations here. | |
754 | * | |
755 | * [1] if match_wildcard_base() says the base | |
756 | * directory is already matched, we only need to match | |
757 | * the rest, which is shorter so _in theory_ faster. | |
758 | */ | |
759 | ||
760 | strbuf_add(base, entry->path, pathlen); | |
761 | ||
762 | if (!git_fnmatch(item, match, base->buf + base_offset, | |
763 | item->nowildcard_len)) { | |
764 | strbuf_setlen(base, base_offset + baselen); | |
765 | return entry_interesting; | |
766 | } | |
767 | strbuf_setlen(base, base_offset + baselen); | |
768 | ||
769 | /* | |
770 | * Match all directories. We'll try to match files | |
771 | * later on. | |
772 | * max_depth is ignored but we may consider support it | |
773 | * in future, see | |
774 | * http://thread.gmane.org/gmane.comp.version-control.git/163757/focus=163840 | |
775 | */ | |
776 | if (ps->recursive && S_ISDIR(entry->mode)) | |
777 | return entry_interesting; | |
778 | } | |
779 | return never_interesting; /* No matches */ | |
780 | } | |
781 | ||
782 | /* | |
783 | * Is a tree entry interesting given the pathspec we have? | |
784 | * | |
785 | * Pre-condition: either baselen == base_offset (i.e. empty path) | |
786 | * or base[baselen-1] == '/' (i.e. with trailing slash). | |
787 | */ | |
788 | enum interesting tree_entry_interesting(const struct name_entry *entry, | |
789 | struct strbuf *base, int base_offset, | |
790 | const struct pathspec *ps) | |
791 | { | |
792 | enum interesting positive, negative; | |
793 | positive = do_match(entry, base, base_offset, ps, 0); | |
794 | ||
795 | /* | |
796 | * case | entry | positive | negative | result | |
797 | * -----+-------+----------+----------+------- | |
798 | * 1 | file | -1 | -1..2 | -1 | |
799 | * 2 | file | 0 | -1..2 | 0 | |
800 | * 3 | file | 1 | -1 | 1 | |
801 | * 4 | file | 1 | 0 | 1 | |
802 | * 5 | file | 1 | 1 | 0 | |
803 | * 6 | file | 1 | 2 | 0 | |
804 | * 7 | file | 2 | -1 | 2 | |
805 | * 8 | file | 2 | 0 | 2 | |
806 | * 9 | file | 2 | 1 | 0 | |
807 | * 10 | file | 2 | 2 | -1 | |
808 | * -----+-------+----------+----------+------- | |
809 | * 11 | dir | -1 | -1..2 | -1 | |
810 | * 12 | dir | 0 | -1..2 | 0 | |
811 | * 13 | dir | 1 | -1 | 1 | |
812 | * 14 | dir | 1 | 0 | 1 | |
813 | * 15 | dir | 1 | 1 | 1 (*) | |
814 | * 16 | dir | 1 | 2 | 0 | |
815 | * 17 | dir | 2 | -1 | 2 | |
816 | * 18 | dir | 2 | 0 | 2 | |
817 | * 19 | dir | 2 | 1 | 1 (*) | |
818 | * 20 | dir | 2 | 2 | -1 | |
819 | * | |
820 | * (*) An exclude pattern interested in a directory does not | |
821 | * necessarily mean it will exclude all of the directory. In | |
822 | * wildcard case, it can't decide until looking at individual | |
823 | * files inside. So don't write such directories off yet. | |
824 | */ | |
825 | ||
826 | if (!(ps->magic & PATHSPEC_EXCLUDE) || | |
827 | positive <= entry_not_interesting) /* #1, #2, #11, #12 */ | |
828 | return positive; | |
829 | ||
830 | negative = do_match(entry, base, base_offset, ps, 1); | |
831 | ||
832 | /* #3, #4, #7, #8, #13, #14, #17, #18 */ | |
833 | if (negative <= entry_not_interesting) | |
834 | return positive; | |
835 | ||
836 | /* #15, #19 */ | |
837 | if (S_ISDIR(entry->mode) && | |
838 | positive >= entry_interesting && | |
839 | negative == entry_interesting) | |
840 | return entry_interesting; | |
841 | ||
842 | if ((positive == entry_interesting && | |
843 | negative >= entry_interesting) || /* #5, #6, #16 */ | |
844 | (positive == all_entries_interesting && | |
845 | negative == entry_interesting)) /* #9 */ | |
846 | return entry_not_interesting; | |
847 | ||
848 | return all_entries_not_interesting; /* #10, #20 */ | |
849 | } |