]>
Commit | Line | Data |
---|---|---|
1 | #include "cache.h" | |
2 | #include "tree-walk.h" | |
3 | #include "unpack-trees.h" | |
4 | #include "dir.h" | |
5 | #include "object-store.h" | |
6 | #include "tree.h" | |
7 | #include "pathspec.h" | |
8 | ||
9 | static const char *get_mode(const char *str, unsigned int *modep) | |
10 | { | |
11 | unsigned char c; | |
12 | unsigned int mode = 0; | |
13 | ||
14 | if (*str == ' ') | |
15 | return NULL; | |
16 | ||
17 | while ((c = *str++) != ' ') { | |
18 | if (c < '0' || c > '7') | |
19 | return NULL; | |
20 | mode = (mode << 3) + (c - '0'); | |
21 | } | |
22 | *modep = mode; | |
23 | return str; | |
24 | } | |
25 | ||
26 | static int decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size, struct strbuf *err) | |
27 | { | |
28 | const char *path; | |
29 | unsigned int mode, len; | |
30 | const unsigned hashsz = the_hash_algo->rawsz; | |
31 | ||
32 | if (size < hashsz + 3 || buf[size - (hashsz + 1)]) { | |
33 | strbuf_addstr(err, _("too-short tree object")); | |
34 | return -1; | |
35 | } | |
36 | ||
37 | path = get_mode(buf, &mode); | |
38 | if (!path) { | |
39 | strbuf_addstr(err, _("malformed mode in tree entry")); | |
40 | return -1; | |
41 | } | |
42 | if (!*path) { | |
43 | strbuf_addstr(err, _("empty filename in tree entry")); | |
44 | return -1; | |
45 | } | |
46 | len = strlen(path) + 1; | |
47 | ||
48 | /* Initialize the descriptor entry */ | |
49 | desc->entry.path = path; | |
50 | desc->entry.mode = canon_mode(mode); | |
51 | desc->entry.oid = (const struct object_id *)(path + len); | |
52 | ||
53 | return 0; | |
54 | } | |
55 | ||
56 | static int init_tree_desc_internal(struct tree_desc *desc, const void *buffer, unsigned long size, struct strbuf *err) | |
57 | { | |
58 | desc->buffer = buffer; | |
59 | desc->size = size; | |
60 | if (size) | |
61 | return decode_tree_entry(desc, buffer, size, err); | |
62 | return 0; | |
63 | } | |
64 | ||
65 | void init_tree_desc(struct tree_desc *desc, const void *buffer, unsigned long size) | |
66 | { | |
67 | struct strbuf err = STRBUF_INIT; | |
68 | if (init_tree_desc_internal(desc, buffer, size, &err)) | |
69 | die("%s", err.buf); | |
70 | strbuf_release(&err); | |
71 | } | |
72 | ||
73 | int init_tree_desc_gently(struct tree_desc *desc, const void *buffer, unsigned long size) | |
74 | { | |
75 | struct strbuf err = STRBUF_INIT; | |
76 | int result = init_tree_desc_internal(desc, buffer, size, &err); | |
77 | if (result) | |
78 | error("%s", err.buf); | |
79 | strbuf_release(&err); | |
80 | return result; | |
81 | } | |
82 | ||
83 | void *fill_tree_descriptor(struct tree_desc *desc, const struct object_id *oid) | |
84 | { | |
85 | unsigned long size = 0; | |
86 | void *buf = NULL; | |
87 | ||
88 | if (oid) { | |
89 | buf = read_object_with_reference(oid, tree_type, &size, NULL); | |
90 | if (!buf) | |
91 | die("unable to read tree %s", oid_to_hex(oid)); | |
92 | } | |
93 | init_tree_desc(desc, buf, size); | |
94 | return buf; | |
95 | } | |
96 | ||
97 | static void entry_clear(struct name_entry *a) | |
98 | { | |
99 | memset(a, 0, sizeof(*a)); | |
100 | } | |
101 | ||
102 | static void entry_extract(struct tree_desc *t, struct name_entry *a) | |
103 | { | |
104 | *a = t->entry; | |
105 | } | |
106 | ||
107 | static int update_tree_entry_internal(struct tree_desc *desc, struct strbuf *err) | |
108 | { | |
109 | const void *buf = desc->buffer; | |
110 | const unsigned char *end = desc->entry.oid->hash + the_hash_algo->rawsz; | |
111 | unsigned long size = desc->size; | |
112 | unsigned long len = end - (const unsigned char *)buf; | |
113 | ||
114 | if (size < len) | |
115 | die(_("too-short tree file")); | |
116 | buf = end; | |
117 | size -= len; | |
118 | desc->buffer = buf; | |
119 | desc->size = size; | |
120 | if (size) | |
121 | return decode_tree_entry(desc, buf, size, err); | |
122 | return 0; | |
123 | } | |
124 | ||
125 | void update_tree_entry(struct tree_desc *desc) | |
126 | { | |
127 | struct strbuf err = STRBUF_INIT; | |
128 | if (update_tree_entry_internal(desc, &err)) | |
129 | die("%s", err.buf); | |
130 | strbuf_release(&err); | |
131 | } | |
132 | ||
133 | int update_tree_entry_gently(struct tree_desc *desc) | |
134 | { | |
135 | struct strbuf err = STRBUF_INIT; | |
136 | if (update_tree_entry_internal(desc, &err)) { | |
137 | error("%s", err.buf); | |
138 | strbuf_release(&err); | |
139 | /* Stop processing this tree after error */ | |
140 | desc->size = 0; | |
141 | return -1; | |
142 | } | |
143 | strbuf_release(&err); | |
144 | return 0; | |
145 | } | |
146 | ||
147 | int tree_entry(struct tree_desc *desc, struct name_entry *entry) | |
148 | { | |
149 | if (!desc->size) | |
150 | return 0; | |
151 | ||
152 | *entry = desc->entry; | |
153 | update_tree_entry(desc); | |
154 | return 1; | |
155 | } | |
156 | ||
157 | int tree_entry_gently(struct tree_desc *desc, struct name_entry *entry) | |
158 | { | |
159 | if (!desc->size) | |
160 | return 0; | |
161 | ||
162 | *entry = desc->entry; | |
163 | if (update_tree_entry_gently(desc)) | |
164 | return 0; | |
165 | return 1; | |
166 | } | |
167 | ||
168 | void setup_traverse_info(struct traverse_info *info, const char *base) | |
169 | { | |
170 | int pathlen = strlen(base); | |
171 | static struct traverse_info dummy; | |
172 | ||
173 | memset(info, 0, sizeof(*info)); | |
174 | if (pathlen && base[pathlen-1] == '/') | |
175 | pathlen--; | |
176 | info->pathlen = pathlen ? pathlen + 1 : 0; | |
177 | info->name.path = base; | |
178 | info->name.oid = (void *)(base + pathlen + 1); | |
179 | if (pathlen) | |
180 | info->prev = &dummy; | |
181 | } | |
182 | ||
183 | char *make_traverse_path(char *path, const struct traverse_info *info, const struct name_entry *n) | |
184 | { | |
185 | int len = tree_entry_len(n); | |
186 | int pathlen = info->pathlen; | |
187 | ||
188 | path[pathlen + len] = 0; | |
189 | for (;;) { | |
190 | memcpy(path + pathlen, n->path, len); | |
191 | if (!pathlen) | |
192 | break; | |
193 | path[--pathlen] = '/'; | |
194 | n = &info->name; | |
195 | len = tree_entry_len(n); | |
196 | info = info->prev; | |
197 | pathlen -= len; | |
198 | } | |
199 | return path; | |
200 | } | |
201 | ||
202 | struct tree_desc_skip { | |
203 | struct tree_desc_skip *prev; | |
204 | const void *ptr; | |
205 | }; | |
206 | ||
207 | struct tree_desc_x { | |
208 | struct tree_desc d; | |
209 | struct tree_desc_skip *skip; | |
210 | }; | |
211 | ||
212 | static int check_entry_match(const char *a, int a_len, const char *b, int b_len) | |
213 | { | |
214 | /* | |
215 | * The caller wants to pick *a* from a tree or nothing. | |
216 | * We are looking at *b* in a tree. | |
217 | * | |
218 | * (0) If a and b are the same name, we are trivially happy. | |
219 | * | |
220 | * There are three possibilities where *a* could be hiding | |
221 | * behind *b*. | |
222 | * | |
223 | * (1) *a* == "t", *b* == "ab" i.e. *b* sorts earlier than *a* no | |
224 | * matter what. | |
225 | * (2) *a* == "t", *b* == "t-2" and "t" is a subtree in the tree; | |
226 | * (3) *a* == "t-2", *b* == "t" and "t-2" is a blob in the tree. | |
227 | * | |
228 | * Otherwise we know *a* won't appear in the tree without | |
229 | * scanning further. | |
230 | */ | |
231 | ||
232 | int cmp = name_compare(a, a_len, b, b_len); | |
233 | ||
234 | /* Most common case first -- reading sync'd trees */ | |
235 | if (!cmp) | |
236 | return cmp; | |
237 | ||
238 | if (0 < cmp) { | |
239 | /* a comes after b; it does not matter if it is case (3) | |
240 | if (b_len < a_len && !memcmp(a, b, b_len) && a[b_len] < '/') | |
241 | return 1; | |
242 | */ | |
243 | return 1; /* keep looking */ | |
244 | } | |
245 | ||
246 | /* b comes after a; are we looking at case (2)? */ | |
247 | if (a_len < b_len && !memcmp(a, b, a_len) && b[a_len] < '/') | |
248 | return 1; /* keep looking */ | |
249 | ||
250 | return -1; /* a cannot appear in the tree */ | |
251 | } | |
252 | ||
253 | /* | |
254 | * From the extended tree_desc, extract the first name entry, while | |
255 | * paying attention to the candidate "first" name. Most importantly, | |
256 | * when looking for an entry, if there are entries that sorts earlier | |
257 | * in the tree object representation than that name, skip them and | |
258 | * process the named entry first. We will remember that we haven't | |
259 | * processed the first entry yet, and in the later call skip the | |
260 | * entry we processed early when update_extended_entry() is called. | |
261 | * | |
262 | * E.g. if the underlying tree object has these entries: | |
263 | * | |
264 | * blob "t-1" | |
265 | * blob "t-2" | |
266 | * tree "t" | |
267 | * blob "t=1" | |
268 | * | |
269 | * and the "first" asks for "t", remember that we still need to | |
270 | * process "t-1" and "t-2" but extract "t". After processing the | |
271 | * entry "t" from this call, the caller will let us know by calling | |
272 | * update_extended_entry() that we can remember "t" has been processed | |
273 | * already. | |
274 | */ | |
275 | ||
276 | static void extended_entry_extract(struct tree_desc_x *t, | |
277 | struct name_entry *a, | |
278 | const char *first, | |
279 | int first_len) | |
280 | { | |
281 | const char *path; | |
282 | int len; | |
283 | struct tree_desc probe; | |
284 | struct tree_desc_skip *skip; | |
285 | ||
286 | /* | |
287 | * Extract the first entry from the tree_desc, but skip the | |
288 | * ones that we already returned in earlier rounds. | |
289 | */ | |
290 | while (1) { | |
291 | if (!t->d.size) { | |
292 | entry_clear(a); | |
293 | break; /* not found */ | |
294 | } | |
295 | entry_extract(&t->d, a); | |
296 | for (skip = t->skip; skip; skip = skip->prev) | |
297 | if (a->path == skip->ptr) | |
298 | break; /* found */ | |
299 | if (!skip) | |
300 | break; | |
301 | /* We have processed this entry already. */ | |
302 | update_tree_entry(&t->d); | |
303 | } | |
304 | ||
305 | if (!first || !a->path) | |
306 | return; | |
307 | ||
308 | /* | |
309 | * The caller wants "first" from this tree, or nothing. | |
310 | */ | |
311 | path = a->path; | |
312 | len = tree_entry_len(a); | |
313 | switch (check_entry_match(first, first_len, path, len)) { | |
314 | case -1: | |
315 | entry_clear(a); | |
316 | case 0: | |
317 | return; | |
318 | default: | |
319 | break; | |
320 | } | |
321 | ||
322 | /* | |
323 | * We need to look-ahead -- we suspect that a subtree whose | |
324 | * name is "first" may be hiding behind the current entry "path". | |
325 | */ | |
326 | probe = t->d; | |
327 | while (probe.size) { | |
328 | entry_extract(&probe, a); | |
329 | path = a->path; | |
330 | len = tree_entry_len(a); | |
331 | switch (check_entry_match(first, first_len, path, len)) { | |
332 | case -1: | |
333 | entry_clear(a); | |
334 | case 0: | |
335 | return; | |
336 | default: | |
337 | update_tree_entry(&probe); | |
338 | break; | |
339 | } | |
340 | /* keep looking */ | |
341 | } | |
342 | entry_clear(a); | |
343 | } | |
344 | ||
345 | static void update_extended_entry(struct tree_desc_x *t, struct name_entry *a) | |
346 | { | |
347 | if (t->d.entry.path == a->path) { | |
348 | update_tree_entry(&t->d); | |
349 | } else { | |
350 | /* we have returned this entry early */ | |
351 | struct tree_desc_skip *skip = xmalloc(sizeof(*skip)); | |
352 | skip->ptr = a->path; | |
353 | skip->prev = t->skip; | |
354 | t->skip = skip; | |
355 | } | |
356 | } | |
357 | ||
358 | static void free_extended_entry(struct tree_desc_x *t) | |
359 | { | |
360 | struct tree_desc_skip *p, *s; | |
361 | ||
362 | for (s = t->skip; s; s = p) { | |
363 | p = s->prev; | |
364 | free(s); | |
365 | } | |
366 | } | |
367 | ||
368 | static inline int prune_traversal(struct name_entry *e, | |
369 | struct traverse_info *info, | |
370 | struct strbuf *base, | |
371 | int still_interesting) | |
372 | { | |
373 | if (!info->pathspec || still_interesting == 2) | |
374 | return 2; | |
375 | if (still_interesting < 0) | |
376 | return still_interesting; | |
377 | return tree_entry_interesting(e, base, 0, info->pathspec); | |
378 | } | |
379 | ||
380 | int traverse_trees(int n, struct tree_desc *t, struct traverse_info *info) | |
381 | { | |
382 | int error = 0; | |
383 | struct name_entry *entry = xmalloc(n*sizeof(*entry)); | |
384 | int i; | |
385 | struct tree_desc_x *tx = xcalloc(n, sizeof(*tx)); | |
386 | struct strbuf base = STRBUF_INIT; | |
387 | int interesting = 1; | |
388 | char *traverse_path; | |
389 | ||
390 | for (i = 0; i < n; i++) | |
391 | tx[i].d = t[i]; | |
392 | ||
393 | if (info->prev) { | |
394 | strbuf_grow(&base, info->pathlen); | |
395 | make_traverse_path(base.buf, info->prev, &info->name); | |
396 | base.buf[info->pathlen-1] = '/'; | |
397 | strbuf_setlen(&base, info->pathlen); | |
398 | traverse_path = xstrndup(base.buf, info->pathlen); | |
399 | } else { | |
400 | traverse_path = xstrndup(info->name.path, info->pathlen); | |
401 | } | |
402 | info->traverse_path = traverse_path; | |
403 | for (;;) { | |
404 | int trees_used; | |
405 | unsigned long mask, dirmask; | |
406 | const char *first = NULL; | |
407 | int first_len = 0; | |
408 | struct name_entry *e = NULL; | |
409 | int len; | |
410 | ||
411 | for (i = 0; i < n; i++) { | |
412 | e = entry + i; | |
413 | extended_entry_extract(tx + i, e, NULL, 0); | |
414 | } | |
415 | ||
416 | /* | |
417 | * A tree may have "t-2" at the current location even | |
418 | * though it may have "t" that is a subtree behind it, | |
419 | * and another tree may return "t". We want to grab | |
420 | * all "t" from all trees to match in such a case. | |
421 | */ | |
422 | for (i = 0; i < n; i++) { | |
423 | e = entry + i; | |
424 | if (!e->path) | |
425 | continue; | |
426 | len = tree_entry_len(e); | |
427 | if (!first) { | |
428 | first = e->path; | |
429 | first_len = len; | |
430 | continue; | |
431 | } | |
432 | if (name_compare(e->path, len, first, first_len) < 0) { | |
433 | first = e->path; | |
434 | first_len = len; | |
435 | } | |
436 | } | |
437 | ||
438 | if (first) { | |
439 | for (i = 0; i < n; i++) { | |
440 | e = entry + i; | |
441 | extended_entry_extract(tx + i, e, first, first_len); | |
442 | /* Cull the ones that are not the earliest */ | |
443 | if (!e->path) | |
444 | continue; | |
445 | len = tree_entry_len(e); | |
446 | if (name_compare(e->path, len, first, first_len)) | |
447 | entry_clear(e); | |
448 | } | |
449 | } | |
450 | ||
451 | /* Now we have in entry[i] the earliest name from the trees */ | |
452 | mask = 0; | |
453 | dirmask = 0; | |
454 | for (i = 0; i < n; i++) { | |
455 | if (!entry[i].path) | |
456 | continue; | |
457 | mask |= 1ul << i; | |
458 | if (S_ISDIR(entry[i].mode)) | |
459 | dirmask |= 1ul << i; | |
460 | e = &entry[i]; | |
461 | } | |
462 | if (!mask) | |
463 | break; | |
464 | interesting = prune_traversal(e, info, &base, interesting); | |
465 | if (interesting < 0) | |
466 | break; | |
467 | if (interesting) { | |
468 | trees_used = info->fn(n, mask, dirmask, entry, info); | |
469 | if (trees_used < 0) { | |
470 | error = trees_used; | |
471 | if (!info->show_all_errors) | |
472 | break; | |
473 | } | |
474 | mask &= trees_used; | |
475 | } | |
476 | for (i = 0; i < n; i++) | |
477 | if (mask & (1ul << i)) | |
478 | update_extended_entry(tx + i, entry + i); | |
479 | } | |
480 | free(entry); | |
481 | for (i = 0; i < n; i++) | |
482 | free_extended_entry(tx + i); | |
483 | free(tx); | |
484 | free(traverse_path); | |
485 | info->traverse_path = NULL; | |
486 | strbuf_release(&base); | |
487 | return error; | |
488 | } | |
489 | ||
490 | struct dir_state { | |
491 | void *tree; | |
492 | unsigned long size; | |
493 | struct object_id oid; | |
494 | }; | |
495 | ||
496 | static int find_tree_entry(struct tree_desc *t, const char *name, struct object_id *result, unsigned *mode) | |
497 | { | |
498 | int namelen = strlen(name); | |
499 | while (t->size) { | |
500 | const char *entry; | |
501 | const struct object_id *oid; | |
502 | int entrylen, cmp; | |
503 | ||
504 | oid = tree_entry_extract(t, &entry, mode); | |
505 | entrylen = tree_entry_len(&t->entry); | |
506 | update_tree_entry(t); | |
507 | if (entrylen > namelen) | |
508 | continue; | |
509 | cmp = memcmp(name, entry, entrylen); | |
510 | if (cmp > 0) | |
511 | continue; | |
512 | if (cmp < 0) | |
513 | break; | |
514 | if (entrylen == namelen) { | |
515 | oidcpy(result, oid); | |
516 | return 0; | |
517 | } | |
518 | if (name[entrylen] != '/') | |
519 | continue; | |
520 | if (!S_ISDIR(*mode)) | |
521 | break; | |
522 | if (++entrylen == namelen) { | |
523 | oidcpy(result, oid); | |
524 | return 0; | |
525 | } | |
526 | return get_tree_entry(oid, name + entrylen, result, mode); | |
527 | } | |
528 | return -1; | |
529 | } | |
530 | ||
531 | int get_tree_entry(const struct object_id *tree_oid, const char *name, struct object_id *oid, unsigned *mode) | |
532 | { | |
533 | int retval; | |
534 | void *tree; | |
535 | unsigned long size; | |
536 | struct object_id root; | |
537 | ||
538 | tree = read_object_with_reference(tree_oid, tree_type, &size, &root); | |
539 | if (!tree) | |
540 | return -1; | |
541 | ||
542 | if (name[0] == '\0') { | |
543 | oidcpy(oid, &root); | |
544 | free(tree); | |
545 | return 0; | |
546 | } | |
547 | ||
548 | if (!size) { | |
549 | retval = -1; | |
550 | } else { | |
551 | struct tree_desc t; | |
552 | init_tree_desc(&t, tree, size); | |
553 | retval = find_tree_entry(&t, name, oid, mode); | |
554 | } | |
555 | free(tree); | |
556 | return retval; | |
557 | } | |
558 | ||
559 | /* | |
560 | * This is Linux's built-in max for the number of symlinks to follow. | |
561 | * That limit, of course, does not affect git, but it's a reasonable | |
562 | * choice. | |
563 | */ | |
564 | #define GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS 40 | |
565 | ||
566 | /** | |
567 | * Find a tree entry by following symlinks in tree_sha (which is | |
568 | * assumed to be the root of the repository). In the event that a | |
569 | * symlink points outside the repository (e.g. a link to /foo or a | |
570 | * root-level link to ../foo), the portion of the link which is | |
571 | * outside the repository will be returned in result_path, and *mode | |
572 | * will be set to 0. It is assumed that result_path is uninitialized. | |
573 | * If there are no symlinks, or the end result of the symlink chain | |
574 | * points to an object inside the repository, result will be filled in | |
575 | * with the sha1 of the found object, and *mode will hold the mode of | |
576 | * the object. | |
577 | * | |
578 | * See the code for enum follow_symlink_result for a description of | |
579 | * the return values. | |
580 | */ | |
581 | 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) | |
582 | { | |
583 | int retval = MISSING_OBJECT; | |
584 | struct dir_state *parents = NULL; | |
585 | size_t parents_alloc = 0; | |
586 | size_t i, parents_nr = 0; | |
587 | struct object_id current_tree_oid; | |
588 | struct strbuf namebuf = STRBUF_INIT; | |
589 | struct tree_desc t; | |
590 | int follows_remaining = GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS; | |
591 | ||
592 | init_tree_desc(&t, NULL, 0UL); | |
593 | strbuf_addstr(&namebuf, name); | |
594 | oidcpy(¤t_tree_oid, tree_oid); | |
595 | ||
596 | while (1) { | |
597 | int find_result; | |
598 | char *first_slash; | |
599 | char *remainder = NULL; | |
600 | ||
601 | if (!t.buffer) { | |
602 | void *tree; | |
603 | struct object_id root; | |
604 | unsigned long size; | |
605 | tree = read_object_with_reference(¤t_tree_oid, | |
606 | tree_type, &size, | |
607 | &root); | |
608 | if (!tree) | |
609 | goto done; | |
610 | ||
611 | ALLOC_GROW(parents, parents_nr + 1, parents_alloc); | |
612 | parents[parents_nr].tree = tree; | |
613 | parents[parents_nr].size = size; | |
614 | oidcpy(&parents[parents_nr].oid, &root); | |
615 | parents_nr++; | |
616 | ||
617 | if (namebuf.buf[0] == '\0') { | |
618 | oidcpy(result, &root); | |
619 | retval = FOUND; | |
620 | goto done; | |
621 | } | |
622 | ||
623 | if (!size) | |
624 | goto done; | |
625 | ||
626 | /* descend */ | |
627 | init_tree_desc(&t, tree, size); | |
628 | } | |
629 | ||
630 | /* Handle symlinks to e.g. a//b by removing leading slashes */ | |
631 | while (namebuf.buf[0] == '/') { | |
632 | strbuf_remove(&namebuf, 0, 1); | |
633 | } | |
634 | ||
635 | /* Split namebuf into a first component and a remainder */ | |
636 | if ((first_slash = strchr(namebuf.buf, '/'))) { | |
637 | *first_slash = 0; | |
638 | remainder = first_slash + 1; | |
639 | } | |
640 | ||
641 | if (!strcmp(namebuf.buf, "..")) { | |
642 | struct dir_state *parent; | |
643 | /* | |
644 | * We could end up with .. in the namebuf if it | |
645 | * appears in a symlink. | |
646 | */ | |
647 | ||
648 | if (parents_nr == 1) { | |
649 | if (remainder) | |
650 | *first_slash = '/'; | |
651 | strbuf_add(result_path, namebuf.buf, | |
652 | namebuf.len); | |
653 | *mode = 0; | |
654 | retval = FOUND; | |
655 | goto done; | |
656 | } | |
657 | parent = &parents[parents_nr - 1]; | |
658 | free(parent->tree); | |
659 | parents_nr--; | |
660 | parent = &parents[parents_nr - 1]; | |
661 | init_tree_desc(&t, parent->tree, parent->size); | |
662 | strbuf_remove(&namebuf, 0, remainder ? 3 : 2); | |
663 | continue; | |
664 | } | |
665 | ||
666 | /* We could end up here via a symlink to dir/.. */ | |
667 | if (namebuf.buf[0] == '\0') { | |
668 | oidcpy(result, &parents[parents_nr - 1].oid); | |
669 | retval = FOUND; | |
670 | goto done; | |
671 | } | |
672 | ||
673 | /* Look up the first (or only) path component in the tree. */ | |
674 | find_result = find_tree_entry(&t, namebuf.buf, | |
675 | ¤t_tree_oid, mode); | |
676 | if (find_result) { | |
677 | goto done; | |
678 | } | |
679 | ||
680 | if (S_ISDIR(*mode)) { | |
681 | if (!remainder) { | |
682 | oidcpy(result, ¤t_tree_oid); | |
683 | retval = FOUND; | |
684 | goto done; | |
685 | } | |
686 | /* Descend the tree */ | |
687 | t.buffer = NULL; | |
688 | strbuf_remove(&namebuf, 0, | |
689 | 1 + first_slash - namebuf.buf); | |
690 | } else if (S_ISREG(*mode)) { | |
691 | if (!remainder) { | |
692 | oidcpy(result, ¤t_tree_oid); | |
693 | retval = FOUND; | |
694 | } else { | |
695 | retval = NOT_DIR; | |
696 | } | |
697 | goto done; | |
698 | } else if (S_ISLNK(*mode)) { | |
699 | /* Follow a symlink */ | |
700 | unsigned long link_len; | |
701 | size_t len; | |
702 | char *contents, *contents_start; | |
703 | struct dir_state *parent; | |
704 | enum object_type type; | |
705 | ||
706 | if (follows_remaining-- == 0) { | |
707 | /* Too many symlinks followed */ | |
708 | retval = SYMLINK_LOOP; | |
709 | goto done; | |
710 | } | |
711 | ||
712 | /* | |
713 | * At this point, we have followed at a least | |
714 | * one symlink, so on error we need to report this. | |
715 | */ | |
716 | retval = DANGLING_SYMLINK; | |
717 | ||
718 | contents = read_object_file(¤t_tree_oid, &type, | |
719 | &link_len); | |
720 | ||
721 | if (!contents) | |
722 | goto done; | |
723 | ||
724 | if (contents[0] == '/') { | |
725 | strbuf_addstr(result_path, contents); | |
726 | free(contents); | |
727 | *mode = 0; | |
728 | retval = FOUND; | |
729 | goto done; | |
730 | } | |
731 | ||
732 | if (remainder) | |
733 | len = first_slash - namebuf.buf; | |
734 | else | |
735 | len = namebuf.len; | |
736 | ||
737 | contents_start = contents; | |
738 | ||
739 | parent = &parents[parents_nr - 1]; | |
740 | init_tree_desc(&t, parent->tree, parent->size); | |
741 | strbuf_splice(&namebuf, 0, len, | |
742 | contents_start, link_len); | |
743 | if (remainder) | |
744 | namebuf.buf[link_len] = '/'; | |
745 | free(contents); | |
746 | } | |
747 | } | |
748 | done: | |
749 | for (i = 0; i < parents_nr; i++) | |
750 | free(parents[i].tree); | |
751 | free(parents); | |
752 | ||
753 | strbuf_release(&namebuf); | |
754 | return retval; | |
755 | } | |
756 | ||
757 | static int match_entry(const struct pathspec_item *item, | |
758 | const struct name_entry *entry, int pathlen, | |
759 | const char *match, int matchlen, | |
760 | enum interesting *never_interesting) | |
761 | { | |
762 | int m = -1; /* signals that we haven't called strncmp() */ | |
763 | ||
764 | if (item->magic & PATHSPEC_ICASE) | |
765 | /* | |
766 | * "Never interesting" trick requires exact | |
767 | * matching. We could do something clever with inexact | |
768 | * matching, but it's trickier (and not to forget that | |
769 | * strcasecmp is locale-dependent, at least in | |
770 | * glibc). Just disable it for now. It can't be worse | |
771 | * than the wildcard's codepath of '[Tt][Hi][Is][Ss]' | |
772 | * pattern. | |
773 | */ | |
774 | *never_interesting = entry_not_interesting; | |
775 | else if (*never_interesting != entry_not_interesting) { | |
776 | /* | |
777 | * We have not seen any match that sorts later | |
778 | * than the current path. | |
779 | */ | |
780 | ||
781 | /* | |
782 | * Does match sort strictly earlier than path | |
783 | * with their common parts? | |
784 | */ | |
785 | m = strncmp(match, entry->path, | |
786 | (matchlen < pathlen) ? matchlen : pathlen); | |
787 | if (m < 0) | |
788 | return 0; | |
789 | ||
790 | /* | |
791 | * If we come here even once, that means there is at | |
792 | * least one pathspec that would sort equal to or | |
793 | * later than the path we are currently looking at. | |
794 | * In other words, if we have never reached this point | |
795 | * after iterating all pathspecs, it means all | |
796 | * pathspecs are either outside of base, or inside the | |
797 | * base but sorts strictly earlier than the current | |
798 | * one. In either case, they will never match the | |
799 | * subsequent entries. In such a case, we initialized | |
800 | * the variable to -1 and that is what will be | |
801 | * returned, allowing the caller to terminate early. | |
802 | */ | |
803 | *never_interesting = entry_not_interesting; | |
804 | } | |
805 | ||
806 | if (pathlen > matchlen) | |
807 | return 0; | |
808 | ||
809 | if (matchlen > pathlen) { | |
810 | if (match[pathlen] != '/') | |
811 | return 0; | |
812 | if (!S_ISDIR(entry->mode) && !S_ISGITLINK(entry->mode)) | |
813 | return 0; | |
814 | } | |
815 | ||
816 | if (m == -1) | |
817 | /* | |
818 | * we cheated and did not do strncmp(), so we do | |
819 | * that here. | |
820 | */ | |
821 | m = ps_strncmp(item, match, entry->path, pathlen); | |
822 | ||
823 | /* | |
824 | * If common part matched earlier then it is a hit, | |
825 | * because we rejected the case where path is not a | |
826 | * leading directory and is shorter than match. | |
827 | */ | |
828 | if (!m) | |
829 | /* | |
830 | * match_entry does not check if the prefix part is | |
831 | * matched case-sensitively. If the entry is a | |
832 | * directory and part of prefix, it'll be rematched | |
833 | * eventually by basecmp with special treatment for | |
834 | * the prefix. | |
835 | */ | |
836 | return 1; | |
837 | ||
838 | return 0; | |
839 | } | |
840 | ||
841 | /* :(icase)-aware string compare */ | |
842 | static int basecmp(const struct pathspec_item *item, | |
843 | const char *base, const char *match, int len) | |
844 | { | |
845 | if (item->magic & PATHSPEC_ICASE) { | |
846 | int ret, n = len > item->prefix ? item->prefix : len; | |
847 | ret = strncmp(base, match, n); | |
848 | if (ret) | |
849 | return ret; | |
850 | base += n; | |
851 | match += n; | |
852 | len -= n; | |
853 | } | |
854 | return ps_strncmp(item, base, match, len); | |
855 | } | |
856 | ||
857 | static int match_dir_prefix(const struct pathspec_item *item, | |
858 | const char *base, | |
859 | const char *match, int matchlen) | |
860 | { | |
861 | if (basecmp(item, base, match, matchlen)) | |
862 | return 0; | |
863 | ||
864 | /* | |
865 | * If the base is a subdirectory of a path which | |
866 | * was specified, all of them are interesting. | |
867 | */ | |
868 | if (!matchlen || | |
869 | base[matchlen] == '/' || | |
870 | match[matchlen - 1] == '/') | |
871 | return 1; | |
872 | ||
873 | /* Just a random prefix match */ | |
874 | return 0; | |
875 | } | |
876 | ||
877 | /* | |
878 | * Perform matching on the leading non-wildcard part of | |
879 | * pathspec. item->nowildcard_len must be greater than zero. Return | |
880 | * non-zero if base is matched. | |
881 | */ | |
882 | static int match_wildcard_base(const struct pathspec_item *item, | |
883 | const char *base, int baselen, | |
884 | int *matched) | |
885 | { | |
886 | const char *match = item->match; | |
887 | /* the wildcard part is not considered in this function */ | |
888 | int matchlen = item->nowildcard_len; | |
889 | ||
890 | if (baselen) { | |
891 | int dirlen; | |
892 | /* | |
893 | * Return early if base is longer than the | |
894 | * non-wildcard part but it does not match. | |
895 | */ | |
896 | if (baselen >= matchlen) { | |
897 | *matched = matchlen; | |
898 | return !basecmp(item, base, match, matchlen); | |
899 | } | |
900 | ||
901 | dirlen = matchlen; | |
902 | while (dirlen && match[dirlen - 1] != '/') | |
903 | dirlen--; | |
904 | ||
905 | /* | |
906 | * Return early if base is shorter than the | |
907 | * non-wildcard part but it does not match. Note that | |
908 | * base ends with '/' so we are sure it really matches | |
909 | * directory | |
910 | */ | |
911 | if (basecmp(item, base, match, baselen)) | |
912 | return 0; | |
913 | *matched = baselen; | |
914 | } else | |
915 | *matched = 0; | |
916 | /* | |
917 | * we could have checked entry against the non-wildcard part | |
918 | * that is not in base and does similar never_interesting | |
919 | * optimization as in match_entry. For now just be happy with | |
920 | * base comparison. | |
921 | */ | |
922 | return entry_interesting; | |
923 | } | |
924 | ||
925 | /* | |
926 | * Is a tree entry interesting given the pathspec we have? | |
927 | * | |
928 | * Pre-condition: either baselen == base_offset (i.e. empty path) | |
929 | * or base[baselen-1] == '/' (i.e. with trailing slash). | |
930 | */ | |
931 | static enum interesting do_match(const struct name_entry *entry, | |
932 | struct strbuf *base, int base_offset, | |
933 | const struct pathspec *ps, | |
934 | int exclude) | |
935 | { | |
936 | int i; | |
937 | int pathlen, baselen = base->len - base_offset; | |
938 | enum interesting never_interesting = ps->has_wildcard ? | |
939 | entry_not_interesting : all_entries_not_interesting; | |
940 | ||
941 | GUARD_PATHSPEC(ps, | |
942 | PATHSPEC_FROMTOP | | |
943 | PATHSPEC_MAXDEPTH | | |
944 | PATHSPEC_LITERAL | | |
945 | PATHSPEC_GLOB | | |
946 | PATHSPEC_ICASE | | |
947 | PATHSPEC_EXCLUDE); | |
948 | ||
949 | if (!ps->nr) { | |
950 | if (!ps->recursive || | |
951 | !(ps->magic & PATHSPEC_MAXDEPTH) || | |
952 | ps->max_depth == -1) | |
953 | return all_entries_interesting; | |
954 | return within_depth(base->buf + base_offset, baselen, | |
955 | !!S_ISDIR(entry->mode), | |
956 | ps->max_depth) ? | |
957 | entry_interesting : entry_not_interesting; | |
958 | } | |
959 | ||
960 | pathlen = tree_entry_len(entry); | |
961 | ||
962 | for (i = ps->nr - 1; i >= 0; i--) { | |
963 | const struct pathspec_item *item = ps->items+i; | |
964 | const char *match = item->match; | |
965 | const char *base_str = base->buf + base_offset; | |
966 | int matchlen = item->len, matched = 0; | |
967 | ||
968 | if ((!exclude && item->magic & PATHSPEC_EXCLUDE) || | |
969 | ( exclude && !(item->magic & PATHSPEC_EXCLUDE))) | |
970 | continue; | |
971 | ||
972 | if (baselen >= matchlen) { | |
973 | /* If it doesn't match, move along... */ | |
974 | if (!match_dir_prefix(item, base_str, match, matchlen)) | |
975 | goto match_wildcards; | |
976 | ||
977 | if (!ps->recursive || | |
978 | !(ps->magic & PATHSPEC_MAXDEPTH) || | |
979 | ps->max_depth == -1) | |
980 | return all_entries_interesting; | |
981 | ||
982 | return within_depth(base_str + matchlen + 1, | |
983 | baselen - matchlen - 1, | |
984 | !!S_ISDIR(entry->mode), | |
985 | ps->max_depth) ? | |
986 | entry_interesting : entry_not_interesting; | |
987 | } | |
988 | ||
989 | /* Either there must be no base, or the base must match. */ | |
990 | if (baselen == 0 || !basecmp(item, base_str, match, baselen)) { | |
991 | if (match_entry(item, entry, pathlen, | |
992 | match + baselen, matchlen - baselen, | |
993 | &never_interesting)) | |
994 | return entry_interesting; | |
995 | ||
996 | if (item->nowildcard_len < item->len) { | |
997 | if (!git_fnmatch(item, match + baselen, entry->path, | |
998 | item->nowildcard_len - baselen)) | |
999 | return entry_interesting; | |
1000 | ||
1001 | /* | |
1002 | * Match all directories. We'll try to | |
1003 | * match files later on. | |
1004 | */ | |
1005 | if (ps->recursive && S_ISDIR(entry->mode)) | |
1006 | return entry_interesting; | |
1007 | ||
1008 | /* | |
1009 | * When matching against submodules with | |
1010 | * wildcard characters, ensure that the entry | |
1011 | * at least matches up to the first wild | |
1012 | * character. More accurate matching can then | |
1013 | * be performed in the submodule itself. | |
1014 | */ | |
1015 | if (ps->recurse_submodules && | |
1016 | S_ISGITLINK(entry->mode) && | |
1017 | !ps_strncmp(item, match + baselen, | |
1018 | entry->path, | |
1019 | item->nowildcard_len - baselen)) | |
1020 | return entry_interesting; | |
1021 | } | |
1022 | ||
1023 | continue; | |
1024 | } | |
1025 | ||
1026 | match_wildcards: | |
1027 | if (item->nowildcard_len == item->len) | |
1028 | continue; | |
1029 | ||
1030 | if (item->nowildcard_len && | |
1031 | !match_wildcard_base(item, base_str, baselen, &matched)) | |
1032 | continue; | |
1033 | ||
1034 | /* | |
1035 | * Concatenate base and entry->path into one and do | |
1036 | * fnmatch() on it. | |
1037 | * | |
1038 | * While we could avoid concatenation in certain cases | |
1039 | * [1], which saves a memcpy and potentially a | |
1040 | * realloc, it turns out not worth it. Measurement on | |
1041 | * linux-2.6 does not show any clear improvements, | |
1042 | * partly because of the nowildcard_len optimization | |
1043 | * in git_fnmatch(). Avoid micro-optimizations here. | |
1044 | * | |
1045 | * [1] if match_wildcard_base() says the base | |
1046 | * directory is already matched, we only need to match | |
1047 | * the rest, which is shorter so _in theory_ faster. | |
1048 | */ | |
1049 | ||
1050 | strbuf_add(base, entry->path, pathlen); | |
1051 | ||
1052 | if (!git_fnmatch(item, match, base->buf + base_offset, | |
1053 | item->nowildcard_len)) { | |
1054 | strbuf_setlen(base, base_offset + baselen); | |
1055 | return entry_interesting; | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * When matching against submodules with | |
1060 | * wildcard characters, ensure that the entry | |
1061 | * at least matches up to the first wild | |
1062 | * character. More accurate matching can then | |
1063 | * be performed in the submodule itself. | |
1064 | */ | |
1065 | if (ps->recurse_submodules && S_ISGITLINK(entry->mode) && | |
1066 | !ps_strncmp(item, match, base->buf + base_offset, | |
1067 | item->nowildcard_len)) { | |
1068 | strbuf_setlen(base, base_offset + baselen); | |
1069 | return entry_interesting; | |
1070 | } | |
1071 | ||
1072 | strbuf_setlen(base, base_offset + baselen); | |
1073 | ||
1074 | /* | |
1075 | * Match all directories. We'll try to match files | |
1076 | * later on. | |
1077 | * max_depth is ignored but we may consider support it | |
1078 | * in future, see | |
1079 | * https://public-inbox.org/git/7vmxo5l2g4.fsf@alter.siamese.dyndns.org/ | |
1080 | */ | |
1081 | if (ps->recursive && S_ISDIR(entry->mode)) | |
1082 | return entry_interesting; | |
1083 | } | |
1084 | return never_interesting; /* No matches */ | |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * Is a tree entry interesting given the pathspec we have? | |
1089 | * | |
1090 | * Pre-condition: either baselen == base_offset (i.e. empty path) | |
1091 | * or base[baselen-1] == '/' (i.e. with trailing slash). | |
1092 | */ | |
1093 | enum interesting tree_entry_interesting(const struct name_entry *entry, | |
1094 | struct strbuf *base, int base_offset, | |
1095 | const struct pathspec *ps) | |
1096 | { | |
1097 | enum interesting positive, negative; | |
1098 | positive = do_match(entry, base, base_offset, ps, 0); | |
1099 | ||
1100 | /* | |
1101 | * case | entry | positive | negative | result | |
1102 | * -----+-------+----------+----------+------- | |
1103 | * 1 | file | -1 | -1..2 | -1 | |
1104 | * 2 | file | 0 | -1..2 | 0 | |
1105 | * 3 | file | 1 | -1 | 1 | |
1106 | * 4 | file | 1 | 0 | 1 | |
1107 | * 5 | file | 1 | 1 | 0 | |
1108 | * 6 | file | 1 | 2 | 0 | |
1109 | * 7 | file | 2 | -1 | 2 | |
1110 | * 8 | file | 2 | 0 | 2 | |
1111 | * 9 | file | 2 | 1 | 0 | |
1112 | * 10 | file | 2 | 2 | -1 | |
1113 | * -----+-------+----------+----------+------- | |
1114 | * 11 | dir | -1 | -1..2 | -1 | |
1115 | * 12 | dir | 0 | -1..2 | 0 | |
1116 | * 13 | dir | 1 | -1 | 1 | |
1117 | * 14 | dir | 1 | 0 | 1 | |
1118 | * 15 | dir | 1 | 1 | 1 (*) | |
1119 | * 16 | dir | 1 | 2 | 0 | |
1120 | * 17 | dir | 2 | -1 | 2 | |
1121 | * 18 | dir | 2 | 0 | 2 | |
1122 | * 19 | dir | 2 | 1 | 1 (*) | |
1123 | * 20 | dir | 2 | 2 | -1 | |
1124 | * | |
1125 | * (*) An exclude pattern interested in a directory does not | |
1126 | * necessarily mean it will exclude all of the directory. In | |
1127 | * wildcard case, it can't decide until looking at individual | |
1128 | * files inside. So don't write such directories off yet. | |
1129 | */ | |
1130 | ||
1131 | if (!(ps->magic & PATHSPEC_EXCLUDE) || | |
1132 | positive <= entry_not_interesting) /* #1, #2, #11, #12 */ | |
1133 | return positive; | |
1134 | ||
1135 | negative = do_match(entry, base, base_offset, ps, 1); | |
1136 | ||
1137 | /* #3, #4, #7, #8, #13, #14, #17, #18 */ | |
1138 | if (negative <= entry_not_interesting) | |
1139 | return positive; | |
1140 | ||
1141 | /* #15, #19 */ | |
1142 | if (S_ISDIR(entry->mode) && | |
1143 | positive >= entry_interesting && | |
1144 | negative == entry_interesting) | |
1145 | return entry_interesting; | |
1146 | ||
1147 | if ((positive == entry_interesting && | |
1148 | negative >= entry_interesting) || /* #5, #6, #16 */ | |
1149 | (positive == all_entries_interesting && | |
1150 | negative == entry_interesting)) /* #9 */ | |
1151 | return entry_not_interesting; | |
1152 | ||
1153 | return all_entries_not_interesting; /* #10, #20 */ | |
1154 | } |