Merge branch 'sb/submodule-rm-absorb'
[git] / tree-walk.c
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 unsigned char *sha1)
82 {
83         unsigned long size = 0;
84         void *buf = NULL;
85
86         if (sha1) {
87                 buf = read_object_with_reference(sha1, tree_type, &size, NULL);
88                 if (!buf)
89                         die("unable to read tree %s", sha1_to_hex(sha1));
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 + 20;
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         unsigned char sha1[20];
492 };
493
494 static int find_tree_entry(struct tree_desc *t, const char *name, unsigned char *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                         hashcpy(result, oid->hash);
514                         return 0;
515                 }
516                 if (name[entrylen] != '/')
517                         continue;
518                 if (!S_ISDIR(*mode))
519                         break;
520                 if (++entrylen == namelen) {
521                         hashcpy(result, oid->hash);
522                         return 0;
523                 }
524                 return get_tree_entry(oid->hash, name + entrylen, result, mode);
525         }
526         return -1;
527 }
528
529 int get_tree_entry(const unsigned char *tree_sha1, const char *name, unsigned char *sha1, unsigned *mode)
530 {
531         int retval;
532         void *tree;
533         unsigned long size;
534         unsigned char root[20];
535
536         tree = read_object_with_reference(tree_sha1, tree_type, &size, root);
537         if (!tree)
538                 return -1;
539
540         if (name[0] == '\0') {
541                 hashcpy(sha1, 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, sha1, 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(unsigned char *tree_sha1, const char *name, unsigned char *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         ssize_t parents_nr = 0;
585         unsigned char current_tree_sha1[20];
586         struct strbuf namebuf = STRBUF_INIT;
587         struct tree_desc t;
588         int follows_remaining = GET_TREE_ENTRY_FOLLOW_SYMLINKS_MAX_LINKS;
589         int i;
590
591         init_tree_desc(&t, NULL, 0UL);
592         strbuf_addstr(&namebuf, name);
593         hashcpy(current_tree_sha1, tree_sha1);
594
595         while (1) {
596                 int find_result;
597                 char *first_slash;
598                 char *remainder = NULL;
599
600                 if (!t.buffer) {
601                         void *tree;
602                         unsigned char root[20];
603                         unsigned long size;
604                         tree = read_object_with_reference(current_tree_sha1,
605                                                           tree_type, &size,
606                                                           root);
607                         if (!tree)
608                                 goto done;
609
610                         ALLOC_GROW(parents, parents_nr + 1, parents_alloc);
611                         parents[parents_nr].tree = tree;
612                         parents[parents_nr].size = size;
613                         hashcpy(parents[parents_nr].sha1, root);
614                         parents_nr++;
615
616                         if (namebuf.buf[0] == '\0') {
617                                 hashcpy(result, root);
618                                 retval = FOUND;
619                                 goto done;
620                         }
621
622                         if (!size)
623                                 goto done;
624
625                         /* descend */
626                         init_tree_desc(&t, tree, size);
627                 }
628
629                 /* Handle symlinks to e.g. a//b by removing leading slashes */
630                 while (namebuf.buf[0] == '/') {
631                         strbuf_remove(&namebuf, 0, 1);
632                 }
633
634                 /* Split namebuf into a first component and a remainder */
635                 if ((first_slash = strchr(namebuf.buf, '/'))) {
636                         *first_slash = 0;
637                         remainder = first_slash + 1;
638                 }
639
640                 if (!strcmp(namebuf.buf, "..")) {
641                         struct dir_state *parent;
642                         /*
643                          * We could end up with .. in the namebuf if it
644                          * appears in a symlink.
645                          */
646
647                         if (parents_nr == 1) {
648                                 if (remainder)
649                                         *first_slash = '/';
650                                 strbuf_add(result_path, namebuf.buf,
651                                            namebuf.len);
652                                 *mode = 0;
653                                 retval = FOUND;
654                                 goto done;
655                         }
656                         parent = &parents[parents_nr - 1];
657                         free(parent->tree);
658                         parents_nr--;
659                         parent = &parents[parents_nr - 1];
660                         init_tree_desc(&t, parent->tree, parent->size);
661                         strbuf_remove(&namebuf, 0, remainder ? 3 : 2);
662                         continue;
663                 }
664
665                 /* We could end up here via a symlink to dir/.. */
666                 if (namebuf.buf[0] == '\0') {
667                         hashcpy(result, parents[parents_nr - 1].sha1);
668                         retval = FOUND;
669                         goto done;
670                 }
671
672                 /* Look up the first (or only) path component in the tree. */
673                 find_result = find_tree_entry(&t, namebuf.buf,
674                                               current_tree_sha1, mode);
675                 if (find_result) {
676                         goto done;
677                 }
678
679                 if (S_ISDIR(*mode)) {
680                         if (!remainder) {
681                                 hashcpy(result, current_tree_sha1);
682                                 retval = FOUND;
683                                 goto done;
684                         }
685                         /* Descend the tree */
686                         t.buffer = NULL;
687                         strbuf_remove(&namebuf, 0,
688                                       1 + first_slash - namebuf.buf);
689                 } else if (S_ISREG(*mode)) {
690                         if (!remainder) {
691                                 hashcpy(result, current_tree_sha1);
692                                 retval = FOUND;
693                         } else {
694                                 retval = NOT_DIR;
695                         }
696                         goto done;
697                 } else if (S_ISLNK(*mode)) {
698                         /* Follow a symlink */
699                         unsigned long link_len;
700                         size_t len;
701                         char *contents, *contents_start;
702                         struct dir_state *parent;
703                         enum object_type type;
704
705                         if (follows_remaining-- == 0) {
706                                 /* Too many symlinks followed */
707                                 retval = SYMLINK_LOOP;
708                                 goto done;
709                         }
710
711                         /*
712                          * At this point, we have followed at a least
713                          * one symlink, so on error we need to report this.
714                          */
715                         retval = DANGLING_SYMLINK;
716
717                         contents = read_sha1_file(current_tree_sha1, &type,
718                                                   &link_len);
719
720                         if (!contents)
721                                 goto done;
722
723                         if (contents[0] == '/') {
724                                 strbuf_addstr(result_path, contents);
725                                 free(contents);
726                                 *mode = 0;
727                                 retval = FOUND;
728                                 goto done;
729                         }
730
731                         if (remainder)
732                                 len = first_slash - namebuf.buf;
733                         else
734                                 len = namebuf.len;
735
736                         contents_start = contents;
737
738                         parent = &parents[parents_nr - 1];
739                         init_tree_desc(&t, parent->tree, parent->size);
740                         strbuf_splice(&namebuf, 0, len,
741                                       contents_start, link_len);
742                         if (remainder)
743                                 namebuf.buf[link_len] = '/';
744                         free(contents);
745                 }
746         }
747 done:
748         for (i = 0; i < parents_nr; i++)
749                 free(parents[i].tree);
750         free(parents);
751
752         strbuf_release(&namebuf);
753         return retval;
754 }
755
756 static int match_entry(const struct pathspec_item *item,
757                        const struct name_entry *entry, int pathlen,
758                        const char *match, int matchlen,
759                        enum interesting *never_interesting)
760 {
761         int m = -1; /* signals that we haven't called strncmp() */
762
763         if (item->magic & PATHSPEC_ICASE)
764                 /*
765                  * "Never interesting" trick requires exact
766                  * matching. We could do something clever with inexact
767                  * matching, but it's trickier (and not to forget that
768                  * strcasecmp is locale-dependent, at least in
769                  * glibc). Just disable it for now. It can't be worse
770                  * than the wildcard's codepath of '[Tt][Hi][Is][Ss]'
771                  * pattern.
772                  */
773                 *never_interesting = entry_not_interesting;
774         else if (*never_interesting != entry_not_interesting) {
775                 /*
776                  * We have not seen any match that sorts later
777                  * than the current path.
778                  */
779
780                 /*
781                  * Does match sort strictly earlier than path
782                  * with their common parts?
783                  */
784                 m = strncmp(match, entry->path,
785                             (matchlen < pathlen) ? matchlen : pathlen);
786                 if (m < 0)
787                         return 0;
788
789                 /*
790                  * If we come here even once, that means there is at
791                  * least one pathspec that would sort equal to or
792                  * later than the path we are currently looking at.
793                  * In other words, if we have never reached this point
794                  * after iterating all pathspecs, it means all
795                  * pathspecs are either outside of base, or inside the
796                  * base but sorts strictly earlier than the current
797                  * one.  In either case, they will never match the
798                  * subsequent entries.  In such a case, we initialized
799                  * the variable to -1 and that is what will be
800                  * returned, allowing the caller to terminate early.
801                  */
802                 *never_interesting = entry_not_interesting;
803         }
804
805         if (pathlen > matchlen)
806                 return 0;
807
808         if (matchlen > pathlen) {
809                 if (match[pathlen] != '/')
810                         return 0;
811                 if (!S_ISDIR(entry->mode) && !S_ISGITLINK(entry->mode))
812                         return 0;
813         }
814
815         if (m == -1)
816                 /*
817                  * we cheated and did not do strncmp(), so we do
818                  * that here.
819                  */
820                 m = ps_strncmp(item, match, entry->path, pathlen);
821
822         /*
823          * If common part matched earlier then it is a hit,
824          * because we rejected the case where path is not a
825          * leading directory and is shorter than match.
826          */
827         if (!m)
828                 /*
829                  * match_entry does not check if the prefix part is
830                  * matched case-sensitively. If the entry is a
831                  * directory and part of prefix, it'll be rematched
832                  * eventually by basecmp with special treatment for
833                  * the prefix.
834                  */
835                 return 1;
836
837         return 0;
838 }
839
840 /* :(icase)-aware string compare */
841 static int basecmp(const struct pathspec_item *item,
842                    const char *base, const char *match, int len)
843 {
844         if (item->magic & PATHSPEC_ICASE) {
845                 int ret, n = len > item->prefix ? item->prefix : len;
846                 ret = strncmp(base, match, n);
847                 if (ret)
848                         return ret;
849                 base += n;
850                 match += n;
851                 len -= n;
852         }
853         return ps_strncmp(item, base, match, len);
854 }
855
856 static int match_dir_prefix(const struct pathspec_item *item,
857                             const char *base,
858                             const char *match, int matchlen)
859 {
860         if (basecmp(item, base, match, matchlen))
861                 return 0;
862
863         /*
864          * If the base is a subdirectory of a path which
865          * was specified, all of them are interesting.
866          */
867         if (!matchlen ||
868             base[matchlen] == '/' ||
869             match[matchlen - 1] == '/')
870                 return 1;
871
872         /* Just a random prefix match */
873         return 0;
874 }
875
876 /*
877  * Perform matching on the leading non-wildcard part of
878  * pathspec. item->nowildcard_len must be greater than zero. Return
879  * non-zero if base is matched.
880  */
881 static int match_wildcard_base(const struct pathspec_item *item,
882                                const char *base, int baselen,
883                                int *matched)
884 {
885         const char *match = item->match;
886         /* the wildcard part is not considered in this function */
887         int matchlen = item->nowildcard_len;
888
889         if (baselen) {
890                 int dirlen;
891                 /*
892                  * Return early if base is longer than the
893                  * non-wildcard part but it does not match.
894                  */
895                 if (baselen >= matchlen) {
896                         *matched = matchlen;
897                         return !basecmp(item, base, match, matchlen);
898                 }
899
900                 dirlen = matchlen;
901                 while (dirlen && match[dirlen - 1] != '/')
902                         dirlen--;
903
904                 /*
905                  * Return early if base is shorter than the
906                  * non-wildcard part but it does not match. Note that
907                  * base ends with '/' so we are sure it really matches
908                  * directory
909                  */
910                 if (basecmp(item, base, match, baselen))
911                         return 0;
912                 *matched = baselen;
913         } else
914                 *matched = 0;
915         /*
916          * we could have checked entry against the non-wildcard part
917          * that is not in base and does similar never_interesting
918          * optimization as in match_entry. For now just be happy with
919          * base comparison.
920          */
921         return entry_interesting;
922 }
923
924 /*
925  * Is a tree entry interesting given the pathspec we have?
926  *
927  * Pre-condition: either baselen == base_offset (i.e. empty path)
928  * or base[baselen-1] == '/' (i.e. with trailing slash).
929  */
930 static enum interesting do_match(const struct name_entry *entry,
931                                  struct strbuf *base, int base_offset,
932                                  const struct pathspec *ps,
933                                  int exclude)
934 {
935         int i;
936         int pathlen, baselen = base->len - base_offset;
937         enum interesting never_interesting = ps->has_wildcard ?
938                 entry_not_interesting : all_entries_not_interesting;
939
940         GUARD_PATHSPEC(ps,
941                        PATHSPEC_FROMTOP |
942                        PATHSPEC_MAXDEPTH |
943                        PATHSPEC_LITERAL |
944                        PATHSPEC_GLOB |
945                        PATHSPEC_ICASE |
946                        PATHSPEC_EXCLUDE);
947
948         if (!ps->nr) {
949                 if (!ps->recursive ||
950                     !(ps->magic & PATHSPEC_MAXDEPTH) ||
951                     ps->max_depth == -1)
952                         return all_entries_interesting;
953                 return within_depth(base->buf + base_offset, baselen,
954                                     !!S_ISDIR(entry->mode),
955                                     ps->max_depth) ?
956                         entry_interesting : entry_not_interesting;
957         }
958
959         pathlen = tree_entry_len(entry);
960
961         for (i = ps->nr - 1; i >= 0; i--) {
962                 const struct pathspec_item *item = ps->items+i;
963                 const char *match = item->match;
964                 const char *base_str = base->buf + base_offset;
965                 int matchlen = item->len, matched = 0;
966
967                 if ((!exclude &&   item->magic & PATHSPEC_EXCLUDE) ||
968                     ( exclude && !(item->magic & PATHSPEC_EXCLUDE)))
969                         continue;
970
971                 if (baselen >= matchlen) {
972                         /* If it doesn't match, move along... */
973                         if (!match_dir_prefix(item, base_str, match, matchlen))
974                                 goto match_wildcards;
975
976                         if (!ps->recursive ||
977                             !(ps->magic & PATHSPEC_MAXDEPTH) ||
978                             ps->max_depth == -1)
979                                 return all_entries_interesting;
980
981                         return within_depth(base_str + matchlen + 1,
982                                             baselen - matchlen - 1,
983                                             !!S_ISDIR(entry->mode),
984                                             ps->max_depth) ?
985                                 entry_interesting : entry_not_interesting;
986                 }
987
988                 /* Either there must be no base, or the base must match. */
989                 if (baselen == 0 || !basecmp(item, base_str, match, baselen)) {
990                         if (match_entry(item, entry, pathlen,
991                                         match + baselen, matchlen - baselen,
992                                         &never_interesting))
993                                 return entry_interesting;
994
995                         if (item->nowildcard_len < item->len) {
996                                 if (!git_fnmatch(item, match + baselen, entry->path,
997                                                  item->nowildcard_len - baselen))
998                                         return entry_interesting;
999
1000                                 /*
1001                                  * Match all directories. We'll try to
1002                                  * match files later on.
1003                                  */
1004                                 if (ps->recursive && S_ISDIR(entry->mode))
1005                                         return entry_interesting;
1006
1007                                 /*
1008                                  * When matching against submodules with
1009                                  * wildcard characters, ensure that the entry
1010                                  * at least matches up to the first wild
1011                                  * character.  More accurate matching can then
1012                                  * be performed in the submodule itself.
1013                                  */
1014                                 if (ps->recursive && 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->recursive && 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 }