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