dir.c revision 1.237 1 /* $NetBSD: dir.c,v 1.237 2020/11/29 15:14:32 rillig Exp $ */
2
3 /*
4 * Copyright (c) 1988, 1989, 1990 The Regents of the University of California.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Adam de Boor.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 */
34
35 /*
36 * Copyright (c) 1988, 1989 by Adam de Boor
37 * Copyright (c) 1989 by Berkeley Softworks
38 * All rights reserved.
39 *
40 * This code is derived from software contributed to Berkeley by
41 * Adam de Boor.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. All advertising materials mentioning features or use of this software
52 * must display the following acknowledgement:
53 * This product includes software developed by the University of
54 * California, Berkeley and its contributors.
55 * 4. Neither the name of the University nor the names of its contributors
56 * may be used to endorse or promote products derived from this software
57 * without specific prior written permission.
58 *
59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69 * SUCH DAMAGE.
70 */
71
72 /* Directory searching using wildcards and/or normal names.
73 * Used both for source wildcarding in the makefile and for finding
74 * implicit sources.
75 *
76 * The interface for this module is:
77 * Dir_Init Initialize the module.
78 *
79 * Dir_InitCur Set the cur CachedDir.
80 *
81 * Dir_InitDot Set the dot CachedDir.
82 *
83 * Dir_End Clean up the module.
84 *
85 * Dir_SetPATH Set ${.PATH} to reflect state of dirSearchPath.
86 *
87 * Dir_HasWildcards
88 * Returns TRUE if the name given it needs to
89 * be wildcard-expanded.
90 *
91 * Dir_Expand Given a pattern and a path, return a Lst of names
92 * which match the pattern on the search path.
93 *
94 * Dir_FindFile Searches for a file on a given search path.
95 * If it exists, the entire path is returned.
96 * Otherwise NULL is returned.
97 *
98 * Dir_FindHereOrAbove
99 * Search for a path in the current directory and
100 * then all the directories above it in turn until
101 * the path is found or we reach the root ("/").
102 *
103 * Dir_UpdateMTime
104 * Update the modification time and path of a node with
105 * data from the file corresponding to the node.
106 *
107 * Dir_AddDir Add a directory to a search path.
108 *
109 * SearchPath_ToFlags
110 * Given a search path and a command flag, create
111 * a string with each of the directories in the path
112 * preceded by the command flag and all of them
113 * separated by a space.
114 *
115 * Dir_Destroy Destroy an element of a search path. Frees up all
116 * things that can be freed for the element as long
117 * as the element is no longer referenced by any other
118 * search path.
119 *
120 * SearchPath_Clear
121 * Resets a search path to the empty list.
122 *
123 * For debugging:
124 * Dir_PrintDirectories
125 * Print stats about the directory cache.
126 */
127
128 #include <sys/types.h>
129 #include <sys/stat.h>
130
131 #include <dirent.h>
132 #include <errno.h>
133
134 #include "make.h"
135 #include "dir.h"
136 #include "job.h"
137
138 /* "@(#)dir.c 8.2 (Berkeley) 1/2/94" */
139 MAKE_RCSID("$NetBSD: dir.c,v 1.237 2020/11/29 15:14:32 rillig Exp $");
140
141 #define DIR_DEBUG0(text) DEBUG0(DIR, text)
142 #define DIR_DEBUG1(fmt, arg1) DEBUG1(DIR, fmt, arg1)
143 #define DIR_DEBUG2(fmt, arg1, arg2) DEBUG2(DIR, fmt, arg1, arg2)
144
145 /* A search path is a list of CachedDir structures. A CachedDir has in it the
146 * name of the directory and the names of all the files in the directory.
147 * This is used to cut down on the number of system calls necessary to find
148 * implicit dependents and their like. Since these searches are made before
149 * any actions are taken, we need not worry about the directory changing due
150 * to creation commands. If this hampers the style of some makefiles, they
151 * must be changed.
152 *
153 * All previously-read directories are kept in openDirs, which is checked
154 * first before a directory is opened.
155 *
156 * The need for the caching of whole directories is brought about by the
157 * multi-level transformation code in suff.c, which tends to search for far
158 * more files than regular make does. In the initial implementation, the
159 * amount of time spent performing "stat" calls was truly astronomical.
160 * The problem with caching at the start is, of course, that pmake doesn't
161 * then detect changes to these directories during the course of the make.
162 * Three possibilities suggest themselves:
163 *
164 * 1) just use stat to test for a file's existence. As mentioned above,
165 * this is very inefficient due to the number of checks engendered by
166 * the multi-level transformation code.
167 *
168 * 2) use readdir() and company to search the directories, keeping them
169 * open between checks. I have tried this and while it didn't slow down
170 * the process too much, it could severely affect the amount of
171 * parallelism available as each directory open would take another file
172 * descriptor out of play for handling I/O for another job. Given that
173 * it is only recently (as of 1993 or earlier) that UNIX OS's have taken
174 * to allowing more than 20 or 32 file descriptors for a process, this
175 * doesn't seem acceptable to me.
176 *
177 * 3) record the mtime of the directory in the CachedDir structure and
178 * verify the directory hasn't changed since the contents were cached.
179 * This will catch the creation or deletion of files, but not the
180 * updating of files. However, since it is the creation and deletion
181 * that is the problem, this could be a good thing to do. Unfortunately,
182 * if the directory (say ".") were fairly large and changed fairly
183 * frequently, the constant reloading could seriously degrade
184 * performance. It might be good in such cases to keep track of the
185 * number of reloadings and if the number goes over a (small) limit,
186 * resort to using stat in its place.
187 *
188 * An additional thing to consider is that pmake is used primarily to create
189 * C programs and until recently (as of 1993 or earlier) pcc-based compilers
190 * refused to allow you to specify where the resulting object file should be
191 * placed. This forced all objects to be created in the current directory.
192 * This isn't meant as a full excuse, just an explanation of some of the
193 * reasons for the caching used here.
194 *
195 * One more note: the location of a target's file is only performed on the
196 * downward traversal of the graph and then only for terminal nodes in the
197 * graph. This could be construed as wrong in some cases, but prevents
198 * inadvertent modification of files when the "installed" directory for a
199 * file is provided in the search path.
200 *
201 * Another data structure maintained by this module is an mtime cache used
202 * when the searching of cached directories fails to find a file. In the past,
203 * Dir_FindFile would simply perform an access() call in such a case to
204 * determine if the file could be found using just the name given. When this
205 * hit, however, all that was gained was the knowledge that the file existed.
206 * Given that an access() is essentially a stat() without the copyout() call,
207 * and that the same filesystem overhead would have to be incurred in
208 * Dir_MTime, it made sense to replace the access() with a stat() and record
209 * the mtime in a cache for when Dir_UpdateMTime was actually called.
210 */
211
212
213 /* A cache for the filenames in a directory. */
214 struct CachedDir {
215 /*
216 * Name of directory, either absolute or relative to the current
217 * directory. The name is not normalized in any way, that is, "."
218 * and "./." are different.
219 *
220 * Not sure what happens when .CURDIR is assigned a new value; see
221 * Parse_DoVar.
222 */
223 char *name;
224
225 /*
226 * The number of SearchPaths that refer to this directory.
227 * Plus the number of global variables that refer to this directory.
228 * References from openDirs do not count though.
229 *
230 * TODO: Check the reference counting; see Dir_Expand, partPath.
231 */
232 int refCount;
233
234 /* The number of times a file in this directory has been found. */
235 int hits;
236
237 /* The names of the files in the directory. */
238 HashSet files;
239 };
240
241 typedef List CachedDirList;
242 typedef ListNode CachedDirListNode;
243
244 typedef ListNode SearchPathNode;
245
246 /* A list of cached directories, with fast lookup by directory name. */
247 typedef struct OpenDirs {
248 CachedDirList list;
249 HashTable /* of CachedDirListNode */ table;
250 } OpenDirs;
251
252 typedef enum CachedStatsFlags {
253 CST_NONE = 0,
254 CST_LSTAT = 1 << 0, /* call lstat(2) instead of stat(2) */
255 CST_UPDATE = 1 << 1 /* ignore existing cached entry */
256 } CachedStatsFlags;
257
258
259 SearchPath dirSearchPath = LST_INIT; /* main search path */
260
261 static OpenDirs openDirs; /* all cached directories */
262
263 /*
264 * Variables for gathering statistics on the efficiency of the caching
265 * mechanism.
266 */
267 static int hits; /* Found in directory cache */
268 static int misses; /* Sad, but not evil misses */
269 static int nearmisses; /* Found under search path */
270 static int bigmisses; /* Sought by itself */
271
272 static CachedDir *dot; /* contents of current directory */
273 static CachedDir *cur; /* contents of current directory, if not dot */
274 static CachedDir *dotLast; /* a fake path entry indicating we need to
275 * look for . last */
276
277 /* Results of doing a last-resort stat in Dir_FindFile -- if we have to go to
278 * the system to find the file, we might as well have its mtime on record.
279 *
280 * XXX: If this is done way early, there's a chance other rules will have
281 * already updated the file, in which case we'll update it again. Generally,
282 * there won't be two rules to update a single file, so this should be ok,
283 * but... */
284 static HashTable mtimes;
285
286 static HashTable lmtimes; /* same as mtimes but for lstat */
287
288
289 static void CachedDir_Destroy(CachedDir *);
290
291
292 static CachedDir *
293 CachedDir_New(const char *name)
294 {
295 CachedDir *dir = bmake_malloc(sizeof *dir);
296
297 dir->name = bmake_strdup(name);
298 dir->refCount = 0;
299 dir->hits = 0;
300 HashSet_Init(&dir->files);
301
302 return dir;
303 }
304
305 static CachedDir *
306 CachedDir_Ref(CachedDir *dir)
307 {
308 dir->refCount++;
309 DEBUG2(DIR, "CachedDir refCount++ to %d for \"%s\"\n",
310 dir->refCount, dir->name);
311 return dir;
312 }
313
314 static CachedDir *
315 CachedDir_Unref(CachedDir *dir)
316 {
317 dir->refCount--;
318 DEBUG2(DIR, "CachedDir refCount-- to %d for \"%s\"\n",
319 dir->refCount, dir->name);
320 return dir;
321 }
322
323 static void
324 OpenDirs_Init(OpenDirs *odirs)
325 {
326 Lst_Init(&odirs->list);
327 HashTable_Init(&odirs->table);
328 }
329
330 #ifdef CLEANUP
331 static void
332 OpenDirs_Done(OpenDirs *odirs)
333 {
334 CachedDirListNode *ln = odirs->list.first;
335 DIR_DEBUG1("OpenDirs_Done: %u entries to remove\n",
336 odirs->table.numEntries);
337 while (ln != NULL) {
338 CachedDirListNode *next = ln->next;
339 CachedDir *dir = ln->datum;
340 DIR_DEBUG2("OpenDirs_Done: refCount %d for \"%s\"\n",
341 dir->refCount, dir->name);
342 CachedDir_Destroy(dir); /* removes the dir from odirs->list */
343 ln = next;
344 }
345 Lst_Done(&odirs->list);
346 HashTable_Done(&odirs->table);
347 }
348 #endif
349
350 static CachedDir *
351 OpenDirs_Find(OpenDirs *odirs, const char *name)
352 {
353 CachedDirListNode *ln = HashTable_FindValue(&odirs->table, name);
354 return ln != NULL ? ln->datum : NULL;
355 }
356
357 static void
358 OpenDirs_Add(OpenDirs *odirs, CachedDir *cdir)
359 {
360 if (HashTable_FindEntry(&odirs->table, cdir->name) != NULL)
361 return;
362 Lst_Append(&odirs->list, cdir);
363 HashTable_Set(&odirs->table, cdir->name, odirs->list.last);
364 }
365
366 static void
367 OpenDirs_Remove(OpenDirs *odirs, const char *name)
368 {
369 HashEntry *he = HashTable_FindEntry(&odirs->table, name);
370 CachedDirListNode *ln;
371 if (he == NULL)
372 return;
373 ln = HashEntry_Get(he);
374 HashTable_DeleteEntry(&odirs->table, he);
375 Lst_Remove(&odirs->list, ln);
376 }
377
378 /* Returns 0 and the result of stat(2) or lstat(2) in *out_cst,
379 * or -1 on error. */
380 static int
381 cached_stats(const char *pathname, struct cached_stat *out_cst,
382 CachedStatsFlags flags)
383 {
384 HashTable *tbl = flags & CST_LSTAT ? &lmtimes : &mtimes;
385 struct stat sys_st;
386 struct cached_stat *cst;
387 int rc;
388
389 if (pathname == NULL || pathname[0] == '\0')
390 return -1; /* This can happen in meta mode. */
391
392 cst = HashTable_FindValue(tbl, pathname);
393 if (cst != NULL && !(flags & CST_UPDATE)) {
394 *out_cst = *cst;
395 DIR_DEBUG2("Using cached time %s for %s\n",
396 Targ_FmtTime(cst->cst_mtime), pathname);
397 return 0;
398 }
399
400 rc = (flags & CST_LSTAT ? lstat : stat)(pathname, &sys_st);
401 if (rc == -1)
402 return -1; /* don't cache negative lookups */
403
404 if (sys_st.st_mtime == 0)
405 sys_st.st_mtime = 1; /* avoid confusion with missing file */
406
407 if (cst == NULL) {
408 cst = bmake_malloc(sizeof *cst);
409 HashTable_Set(tbl, pathname, cst);
410 }
411
412 cst->cst_mtime = sys_st.st_mtime;
413 cst->cst_mode = sys_st.st_mode;
414
415 *out_cst = *cst;
416 DIR_DEBUG2(" Caching %s for %s\n",
417 Targ_FmtTime(sys_st.st_mtime), pathname);
418
419 return 0;
420 }
421
422 int
423 cached_stat(const char *pathname, struct cached_stat *cst)
424 {
425 return cached_stats(pathname, cst, CST_NONE);
426 }
427
428 int
429 cached_lstat(const char *pathname, struct cached_stat *cst)
430 {
431 return cached_stats(pathname, cst, CST_LSTAT);
432 }
433
434 /* Initialize the directories module. */
435 void
436 Dir_Init(void)
437 {
438 OpenDirs_Init(&openDirs);
439 HashTable_Init(&mtimes);
440 HashTable_Init(&lmtimes);
441 }
442
443 void
444 Dir_InitDir(const char *cdname)
445 {
446 Dir_InitCur(cdname);
447
448 dotLast = CachedDir_Ref(CachedDir_New(".DOTLAST"));
449 }
450
451 /*
452 * Called by Dir_InitDir and whenever .CURDIR is assigned to.
453 */
454 void
455 Dir_InitCur(const char *cdname)
456 {
457 CachedDir *dir;
458
459 if (cdname == NULL)
460 return;
461
462 /*
463 * Our build directory is not the same as our source directory.
464 * Keep this one around too.
465 */
466 dir = Dir_AddDir(NULL, cdname);
467 if (dir == NULL)
468 return;
469
470 if (cur != NULL && cur != dir) {
471 CachedDir_Unref(cur); /* XXX: why unref twice? */
472 CachedDir_Destroy(cur);
473 }
474 cur = CachedDir_Ref(dir);
475 }
476
477 /* (Re)initialize "dot" (current/object directory) path hash.
478 * Some directories may be cached. */
479 void
480 Dir_InitDot(void)
481 {
482 if (dot != NULL) {
483 /* Remove old entry from openDirs, but do not destroy. */
484 /* XXX: Why not destroy? It's reference-counted after all. */
485 OpenDirs_Remove(&openDirs, dot->name);
486 }
487
488 /* XXX: Before assigning to the global variable, refCount++. */
489 dot = Dir_AddDir(NULL, ".");
490
491 if (dot == NULL) {
492 Error("Cannot open `.' (%s)", strerror(errno));
493 exit(1);
494 }
495
496 /*
497 * We always need to have dot around, so we increment its reference
498 * count to make sure it's not destroyed.
499 */
500 /*
501 * XXX: This is just the normal reference counting. Why is the above
502 * comment so long? And why doesn't the normal reference counting
503 * suffice? This sounds like someone misunderstood reference counting
504 * here.
505 */
506 CachedDir_Ref(dot);
507 Dir_SetPATH(); /* initialize */
508 }
509
510 /* Clean up the directories module. */
511 void
512 Dir_End(void)
513 {
514 #ifdef CLEANUP
515 if (cur != NULL) {
516 CachedDir_Unref(cur); /* XXX: why unref twice? */
517 CachedDir_Destroy(cur);
518 }
519 CachedDir_Unref(dotLast); /* XXX: why unref twice? */
520 CachedDir_Destroy(dotLast);
521 CachedDir_Unref(dot); /* XXX: why unref twice? */
522 CachedDir_Destroy(dot);
523 SearchPath_Clear(&dirSearchPath);
524 OpenDirs_Done(&openDirs);
525 HashTable_Done(&mtimes);
526 #endif
527 }
528
529 /*
530 * We want ${.PATH} to indicate the order in which we will actually
531 * search, so we rebuild it after any .PATH: target.
532 * This is the simplest way to deal with the effect of .DOTLAST.
533 */
534 void
535 Dir_SetPATH(void)
536 {
537 CachedDirListNode *ln;
538 Boolean hasLastDot = FALSE; /* true if we should search dot last */
539
540 Var_Delete(".PATH", VAR_GLOBAL);
541
542 if ((ln = dirSearchPath.first) != NULL) {
543 CachedDir *dir = ln->datum;
544 if (dir == dotLast) {
545 hasLastDot = TRUE;
546 Var_Append(".PATH", dotLast->name, VAR_GLOBAL);
547 }
548 }
549
550 if (!hasLastDot) {
551 if (dot != NULL)
552 Var_Append(".PATH", dot->name, VAR_GLOBAL);
553 if (cur != NULL)
554 Var_Append(".PATH", cur->name, VAR_GLOBAL);
555 }
556
557 for (ln = dirSearchPath.first; ln != NULL; ln = ln->next) {
558 CachedDir *dir = ln->datum;
559 if (dir == dotLast)
560 continue;
561 if (dir == dot && hasLastDot)
562 continue;
563 Var_Append(".PATH", dir->name, VAR_GLOBAL);
564 }
565
566 if (hasLastDot) {
567 if (dot != NULL)
568 Var_Append(".PATH", dot->name, VAR_GLOBAL);
569 if (cur != NULL)
570 Var_Append(".PATH", cur->name, VAR_GLOBAL);
571 }
572 }
573
574 /* See if the given name has any wildcard characters in it and all braces and
575 * brackets are properly balanced.
576 *
577 * XXX: This code is not 100% correct ([^]] fails etc.). I really don't think
578 * that make(1) should be expanding patterns, because then you have to set a
579 * mechanism for escaping the expansion!
580 *
581 * Return TRUE if the word should be expanded, FALSE otherwise.
582 */
583 Boolean
584 Dir_HasWildcards(const char *name)
585 {
586 const char *p;
587 Boolean wild = FALSE;
588 int braces = 0, brackets = 0;
589
590 for (p = name; *p != '\0'; p++) {
591 switch (*p) {
592 case '{':
593 braces++;
594 wild = TRUE;
595 break;
596 case '}':
597 braces--;
598 break;
599 case '[':
600 brackets++;
601 wild = TRUE;
602 break;
603 case ']':
604 brackets--;
605 break;
606 case '?':
607 case '*':
608 wild = TRUE;
609 break;
610 default:
611 break;
612 }
613 }
614 return wild && brackets == 0 && braces == 0;
615 }
616
617 /* See if any files match the pattern and add their names to the 'expansions'
618 * list if they do.
619 *
620 * This is incomplete -- wildcards are only expanded in the final path
621 * component, but not in directories like src/lib*c/file*.c, but it
622 * will do for now (now being 1993 until at least 2020). To expand these,
623 * use the ':sh' variable modifier such as in ${:!echo src/lib*c/file*.c!}.
624 *
625 * Input:
626 * pattern Pattern to look for
627 * dir Directory to search
628 * expansion Place to store the results
629 */
630 static void
631 DirMatchFiles(const char *pattern, CachedDir *dir, StringList *expansions)
632 {
633 const char *dirName = dir->name;
634 Boolean isDot = dirName[0] == '.' && dirName[1] == '\0';
635 HashIter hi;
636
637 /*
638 * XXX: Iterating over all hash entries is inefficient. If the
639 * pattern is a plain string without any wildcards, a direct lookup
640 * is faster.
641 */
642
643 HashIter_InitSet(&hi, &dir->files);
644 while (HashIter_Next(&hi) != NULL) {
645 const char *base = hi.entry->key;
646
647 if (!Str_Match(base, pattern))
648 continue;
649
650 /*
651 * Follow the UNIX convention that dot files are only found
652 * if the pattern begins with a dot. The pattern '.*' does
653 * not match '.' or '..' since these are not included in the
654 * directory cache.
655 *
656 * This means that the pattern '[a-z.]*' does not find
657 * '.file', which is consistent with bash, NetBSD sh and csh.
658 */
659 if (base[0] == '.' && pattern[0] != '.')
660 continue;
661
662 {
663 char *fullName = isDot
664 ? bmake_strdup(base)
665 : str_concat3(dirName, "/", base);
666 Lst_Append(expansions, fullName);
667 }
668 }
669 }
670
671 /* Find the next closing brace in the string, taking nested braces into
672 * account. */
673 static const char *
674 closing_brace(const char *p)
675 {
676 int nest = 0;
677 while (*p != '\0') {
678 if (*p == '}' && nest == 0)
679 break;
680 if (*p == '{')
681 nest++;
682 if (*p == '}')
683 nest--;
684 p++;
685 }
686 return p;
687 }
688
689 /* Find the next closing brace or comma in the string, taking nested braces
690 * into account. */
691 static const char *
692 separator_comma(const char *p)
693 {
694 int nest = 0;
695 while (*p != '\0') {
696 if ((*p == '}' || *p == ',') && nest == 0)
697 break;
698 if (*p == '{')
699 nest++;
700 if (*p == '}')
701 nest--;
702 p++;
703 }
704 return p;
705 }
706
707 static Boolean
708 contains_wildcard(const char *p)
709 {
710 for (; *p != '\0'; p++) {
711 switch (*p) {
712 case '*':
713 case '?':
714 case '{':
715 case '[':
716 return TRUE;
717 }
718 }
719 return FALSE;
720 }
721
722 static char *
723 concat3(const char *a, size_t a_len, const char *b, size_t b_len,
724 const char *c, size_t c_len)
725 {
726 size_t s_len = a_len + b_len + c_len;
727 char *s = bmake_malloc(s_len + 1);
728 memcpy(s, a, a_len);
729 memcpy(s + a_len, b, b_len);
730 memcpy(s + a_len + b_len, c, c_len);
731 s[s_len] = '\0';
732 return s;
733 }
734
735 /* Expand curly braces like the C shell. Brace expansion by itself is purely
736 * textual, the expansions are not looked up in the file system. But if an
737 * expanded word contains wildcard characters, it is expanded further,
738 * matching only the actually existing files.
739 *
740 * Example: "{a{b,c}}" expands to "ab" and "ac".
741 * Example: "{a}" expands to "a".
742 * Example: "{a,*.c}" expands to "a" and all "*.c" files that exist.
743 *
744 * Input:
745 * word Entire word to expand
746 * brace First curly brace in it
747 * path Search path to use
748 * expansions Place to store the expansions
749 */
750 static void
751 DirExpandCurly(const char *word, const char *brace, SearchPath *path,
752 StringList *expansions)
753 {
754 const char *prefix, *middle, *piece, *middle_end, *suffix;
755 size_t prefix_len, suffix_len;
756
757 /* Split the word into prefix '{' middle '}' suffix. */
758
759 middle = brace + 1;
760 middle_end = closing_brace(middle);
761 if (*middle_end == '\0') {
762 Error("Unterminated {} clause \"%s\"", middle);
763 return;
764 }
765
766 prefix = word;
767 prefix_len = (size_t)(brace - prefix);
768 suffix = middle_end + 1;
769 suffix_len = strlen(suffix);
770
771 /* Split the middle into pieces, separated by commas. */
772
773 piece = middle;
774 while (piece < middle_end + 1) {
775 const char *piece_end = separator_comma(piece);
776 size_t piece_len = (size_t)(piece_end - piece);
777
778 char *file = concat3(prefix, prefix_len, piece, piece_len,
779 suffix, suffix_len);
780
781 if (contains_wildcard(file)) {
782 Dir_Expand(file, path, expansions);
783 free(file);
784 } else {
785 Lst_Append(expansions, file);
786 }
787
788 /* skip over the comma or closing brace */
789 piece = piece_end + 1;
790 }
791 }
792
793
794 /* Expand the word in each of the directories from the path. */
795 static void
796 DirExpandPath(const char *word, SearchPath *path, StringList *expansions)
797 {
798 SearchPathNode *ln;
799 for (ln = path->first; ln != NULL; ln = ln->next) {
800 CachedDir *dir = ln->datum;
801 DirMatchFiles(word, dir, expansions);
802 }
803 }
804
805 static void
806 PrintExpansions(StringList *expansions)
807 {
808 const char *sep = "";
809 StringListNode *ln;
810 for (ln = expansions->first; ln != NULL; ln = ln->next) {
811 const char *word = ln->datum;
812 debug_printf("%s%s", sep, word);
813 sep = " ";
814 }
815 debug_printf("\n");
816 }
817
818 /* Expand the given word into a list of words by globbing it, looking in the
819 * directories on the given search path.
820 *
821 * Input:
822 * word the word to expand
823 * path the directories in which to find the files
824 * expansions the list on which to place the results
825 */
826 void
827 Dir_Expand(const char *word, SearchPath *path, StringList *expansions)
828 {
829 const char *cp;
830
831 assert(path != NULL);
832 assert(expansions != NULL);
833
834 DIR_DEBUG1("Expanding \"%s\"... ", word);
835
836 cp = strchr(word, '{');
837 if (cp != NULL) {
838 DirExpandCurly(word, cp, path, expansions);
839 goto done;
840 }
841
842 /* At this point, the word does not contain '{'. */
843
844 cp = strchr(word, '/');
845 if (cp == NULL) {
846 /* The word has no directory component. */
847 /* First the files in dot. */
848 DirMatchFiles(word, dot, expansions);
849
850 /* Then the files in every other directory on the path. */
851 DirExpandPath(word, path, expansions);
852 goto done;
853 }
854
855 /* At this point, the word has a directory component. */
856
857 /* Find the first wildcard in the word. */
858 for (cp = word; *cp != '\0'; cp++)
859 if (*cp == '?' || *cp == '[' || *cp == '*')
860 break;
861
862 if (*cp == '\0') {
863 /*
864 * No directory component and no wildcard at all -- this
865 * should never happen as in such a simple case there is no
866 * need to expand anything.
867 */
868 DirExpandPath(word, path, expansions);
869 goto done;
870 }
871
872 /* Back up to the start of the component containing the wildcard. */
873 /* XXX: This handles '///' and '/' differently. */
874 while (cp > word && *cp != '/')
875 cp--;
876
877 if (cp == word) {
878 /* The first component contains the wildcard. */
879 /* Start the search from the local directory */
880 DirExpandPath(word, path, expansions);
881 goto done;
882 }
883
884 {
885 char *prefix = bmake_strsedup(word, cp + 1);
886 /*
887 * The wildcard isn't in the first component.
888 * Find all the components up to the one with the wildcard.
889 */
890 /*
891 * XXX: Check the "the directory is added to the path" part.
892 * It is probably surprising that the directory before a
893 * wildcard gets added to the path.
894 */
895 /*
896 * XXX: Only the first match of the prefix in the path is
897 * taken, any others are ignored. The expectation may be
898 * that the pattern is expanded in the whole path.
899 */
900 char *dirpath = Dir_FindFile(prefix, path);
901 free(prefix);
902
903 /*
904 * dirpath is null if can't find the leading component
905 * XXX: Dir_FindFile won't find internal components.
906 * i.e. if the path contains ../Etc/Object and we're
907 * looking for Etc, it won't be found. Ah well.
908 * Probably not important.
909 * XXX: Check whether the above comment is still true.
910 */
911 if (dirpath != NULL) {
912 SearchPath *partPath;
913
914 char *end = &dirpath[strlen(dirpath) - 1];
915 /* XXX: What about multiple trailing slashes? */
916 if (*end == '/')
917 *end = '\0';
918
919 partPath = SearchPath_New();
920 (void)Dir_AddDir(partPath, dirpath);
921 DirExpandPath(cp + 1, partPath, expansions);
922 Lst_Free(partPath);
923 /* XXX: Should the dirs in partPath be freed here?
924 * It's not obvious whether to free them or not. */
925 }
926 }
927
928 done:
929 if (DEBUG(DIR))
930 PrintExpansions(expansions);
931 }
932
933 /* Find if the file with the given name exists in the given path.
934 * Return the freshly allocated path to the file, or NULL. */
935 static char *
936 DirLookup(CachedDir *dir, const char *base)
937 {
938 char *file; /* the current filename to check */
939
940 DIR_DEBUG1(" %s ...\n", dir->name);
941
942 if (!HashSet_Contains(&dir->files, base))
943 return NULL;
944
945 file = str_concat3(dir->name, "/", base);
946 DIR_DEBUG1(" returning %s\n", file);
947 dir->hits++;
948 hits++;
949 return file;
950 }
951
952
953 /* Find if the file with the given name exists in the given directory.
954 * Return the freshly allocated path to the file, or NULL. */
955 static char *
956 DirLookupSubdir(CachedDir *dir, const char *name)
957 {
958 struct cached_stat cst;
959 char *file = dir == dot ? bmake_strdup(name)
960 : str_concat3(dir->name, "/", name);
961
962 DIR_DEBUG1("checking %s ...\n", file);
963
964 if (cached_stat(file, &cst) == 0) {
965 nearmisses++;
966 return file;
967 }
968 free(file);
969 return NULL;
970 }
971
972 /* Find if the file with the given name exists in the given path.
973 * Return the freshly allocated path to the file, the empty string, or NULL.
974 * Returning the empty string means that the search should be terminated.
975 */
976 static char *
977 DirLookupAbs(CachedDir *dir, const char *name, const char *cp)
978 {
979 const char *dnp; /* pointer into dir->name */
980 const char *np; /* pointer into name */
981
982 DIR_DEBUG1(" %s ...\n", dir->name);
983
984 /*
985 * If the file has a leading path component and that component
986 * exactly matches the entire name of the current search
987 * directory, we can attempt another cache lookup. And if we don't
988 * have a hit, we can safely assume the file does not exist at all.
989 */
990 for (dnp = dir->name, np = name;
991 *dnp != '\0' && *dnp == *np; dnp++, np++)
992 continue;
993 if (*dnp != '\0' || np != cp - 1)
994 return NULL;
995
996 if (!HashSet_Contains(&dir->files, cp)) {
997 DIR_DEBUG0(" must be here but isn't -- returning\n");
998 return bmake_strdup(""); /* to terminate the search */
999 }
1000
1001 dir->hits++;
1002 hits++;
1003 DIR_DEBUG1(" returning %s\n", name);
1004 return bmake_strdup(name);
1005 }
1006
1007 /* Find the file given on "." or curdir.
1008 * Return the freshly allocated path to the file, or NULL. */
1009 static char *
1010 DirFindDot(const char *name, const char *base)
1011 {
1012
1013 if (HashSet_Contains(&dot->files, base)) {
1014 DIR_DEBUG0(" in '.'\n");
1015 hits++;
1016 dot->hits++;
1017 return bmake_strdup(name);
1018 }
1019
1020 if (cur != NULL && HashSet_Contains(&cur->files, base)) {
1021 DIR_DEBUG1(" in ${.CURDIR} = %s\n", cur->name);
1022 hits++;
1023 cur->hits++;
1024 return str_concat3(cur->name, "/", base);
1025 }
1026
1027 return NULL;
1028 }
1029
1030 /* Find the file with the given name along the given search path.
1031 *
1032 * If the file is found in a directory that is not on the path
1033 * already (either 'name' is absolute or it is a relative path
1034 * [ dir1/.../dirn/file ] which exists below one of the directories
1035 * already on the search path), its directory is added to the end
1036 * of the path, on the assumption that there will be more files in
1037 * that directory later on. Sometimes this is true. Sometimes not.
1038 *
1039 * Input:
1040 * name the file to find
1041 * path the directories to search, or NULL
1042 *
1043 * Results:
1044 * The freshly allocated path to the file, or NULL.
1045 */
1046 char *
1047 Dir_FindFile(const char *name, SearchPath *path)
1048 {
1049 char *file; /* the current filename to check */
1050 const char *base; /* Terminal name of file */
1051 Boolean hasLastDot = FALSE; /* true if we should search dot last */
1052 Boolean hasSlash; /* true if 'name' contains a / */
1053 struct cached_stat cst; /* Buffer for stat, if necessary */
1054 const char *trailing_dot = ".";
1055
1056 /*
1057 * Find the final component of the name and note whether it has a
1058 * slash in it (the name, I mean)
1059 */
1060 base = strrchr(name, '/');
1061 if (base != NULL) {
1062 hasSlash = TRUE;
1063 base++;
1064 } else {
1065 hasSlash = FALSE;
1066 base = name;
1067 }
1068
1069 DIR_DEBUG1("Searching for %s ...", name);
1070
1071 if (path == NULL) {
1072 DIR_DEBUG0("couldn't open path, file not found\n");
1073 misses++;
1074 return NULL;
1075 }
1076
1077 if (path->first != NULL) {
1078 CachedDir *dir = path->first->datum;
1079 if (dir == dotLast) {
1080 hasLastDot = TRUE;
1081 DIR_DEBUG0("[dot last]...");
1082 }
1083 }
1084 DIR_DEBUG0("\n");
1085
1086 /*
1087 * If there's no leading directory components or if the leading
1088 * directory component is exactly `./', consult the cached contents
1089 * of each of the directories on the search path.
1090 */
1091 if (!hasSlash || (base - name == 2 && *name == '.')) {
1092 SearchPathNode *ln;
1093
1094 /*
1095 * We look through all the directories on the path seeking one
1096 * which contains the final component of the given name. If
1097 * such a beast is found, we concatenate the directory name
1098 * and the final component and return the resulting string.
1099 * If we don't find any such thing, we go on to phase two.
1100 *
1101 * No matter what, we always look for the file in the current
1102 * directory before anywhere else (unless we found the magic
1103 * DOTLAST path, in which case we search it last) and we *do
1104 * not* add the ./ to it if it exists.
1105 * This is so there are no conflicts between what the user
1106 * specifies (fish.c) and what pmake finds (./fish.c).
1107 */
1108 if (!hasLastDot && (file = DirFindDot(name, base)) != NULL)
1109 return file;
1110
1111 for (ln = path->first; ln != NULL; ln = ln->next) {
1112 CachedDir *dir = ln->datum;
1113 if (dir == dotLast)
1114 continue;
1115 if ((file = DirLookup(dir, base)) != NULL)
1116 return file;
1117 }
1118
1119 if (hasLastDot && (file = DirFindDot(name, base)) != NULL)
1120 return file;
1121 }
1122
1123 /*
1124 * We didn't find the file on any directory in the search path.
1125 * If the name doesn't contain a slash, that means it doesn't exist.
1126 * If it *does* contain a slash, however, there is still hope: it
1127 * could be in a subdirectory of one of the members of the search
1128 * path. (eg. /usr/include and sys/types.h. The above search would
1129 * fail to turn up types.h in /usr/include, but it *is* in
1130 * /usr/include/sys/types.h).
1131 * [ This no longer applies: If we find such a beast, we assume there
1132 * will be more (what else can we assume?) and add all but the last
1133 * component of the resulting name onto the search path (at the
1134 * end).]
1135 * This phase is only performed if the file is *not* absolute.
1136 */
1137 if (!hasSlash) {
1138 DIR_DEBUG0(" failed.\n");
1139 misses++;
1140 return NULL;
1141 }
1142
1143 if (*base == '\0') {
1144 /* we were given a trailing "/" */
1145 base = trailing_dot;
1146 }
1147
1148 if (name[0] != '/') {
1149 SearchPathNode *ln;
1150 Boolean checkedDot = FALSE;
1151
1152 DIR_DEBUG0(" Trying subdirectories...\n");
1153
1154 if (!hasLastDot) {
1155 if (dot != NULL) {
1156 checkedDot = TRUE;
1157 if ((file = DirLookupSubdir(dot, name)) != NULL)
1158 return file;
1159 }
1160 if (cur && (file = DirLookupSubdir(cur, name)) != NULL)
1161 return file;
1162 }
1163
1164 for (ln = path->first; ln != NULL; ln = ln->next) {
1165 CachedDir *dir = ln->datum;
1166 if (dir == dotLast)
1167 continue;
1168 if (dir == dot) {
1169 if (checkedDot)
1170 continue;
1171 checkedDot = TRUE;
1172 }
1173 if ((file = DirLookupSubdir(dir, name)) != NULL)
1174 return file;
1175 }
1176
1177 if (hasLastDot) {
1178 if (dot && !checkedDot) {
1179 checkedDot = TRUE;
1180 if ((file = DirLookupSubdir(dot, name)) != NULL)
1181 return file;
1182 }
1183 if (cur && (file = DirLookupSubdir(cur, name)) != NULL)
1184 return file;
1185 }
1186
1187 if (checkedDot) {
1188 /*
1189 * Already checked by the given name, since . was in
1190 * the path, so no point in proceeding.
1191 */
1192 DIR_DEBUG0(" Checked . already, returning NULL\n");
1193 return NULL;
1194 }
1195
1196 } else { /* name[0] == '/' */
1197 SearchPathNode *ln;
1198
1199 /*
1200 * For absolute names, compare directory path prefix against
1201 * the the directory path of each member on the search path
1202 * for an exact match. If we have an exact match on any member
1203 * of the search path, use the cached contents of that member
1204 * to lookup the final file component. If that lookup fails we
1205 * can safely assume that the file does not exist at all.
1206 * This is signified by DirLookupAbs() returning an empty
1207 * string.
1208 */
1209 DIR_DEBUG0(" Trying exact path matches...\n");
1210
1211 if (!hasLastDot && cur &&
1212 ((file = DirLookupAbs(cur, name, base)) != NULL)) {
1213 if (file[0] == '\0') {
1214 free(file);
1215 return NULL;
1216 }
1217 return file;
1218 }
1219
1220 for (ln = path->first; ln != NULL; ln = ln->next) {
1221 CachedDir *dir = ln->datum;
1222 if (dir == dotLast)
1223 continue;
1224 if ((file = DirLookupAbs(dir, name, base)) != NULL) {
1225 if (file[0] == '\0') {
1226 free(file);
1227 return NULL;
1228 }
1229 return file;
1230 }
1231 }
1232
1233 if (hasLastDot && cur &&
1234 ((file = DirLookupAbs(cur, name, base)) != NULL)) {
1235 if (file[0] == '\0') {
1236 free(file);
1237 return NULL;
1238 }
1239 return file;
1240 }
1241 }
1242
1243 /*
1244 * Didn't find it that way, either. Sigh. Phase 3. Add its directory
1245 * onto the search path in any case, just in case, then look for the
1246 * thing in the hash table. If we find it, grand. We return a new
1247 * copy of the name. Otherwise we sadly return a NULL pointer. Sigh.
1248 * Note that if the directory holding the file doesn't exist, this
1249 * will do an extra search of the final directory on the path. Unless
1250 * something weird happens, this search won't succeed and life will
1251 * be groovy.
1252 *
1253 * Sigh. We cannot add the directory onto the search path because
1254 * of this amusing case:
1255 * $(INSTALLDIR)/$(FILE): $(FILE)
1256 *
1257 * $(FILE) exists in $(INSTALLDIR) but not in the current one.
1258 * When searching for $(FILE), we will find it in $(INSTALLDIR)
1259 * b/c we added it here. This is not good...
1260 */
1261 #if 0
1262 {
1263 CachedDir *dir;
1264 char *prefix;
1265
1266 if (base == trailing_dot) {
1267 base = strrchr(name, '/');
1268 base++;
1269 }
1270 prefix = bmake_strsedup(name, base - 1);
1271 (void)Dir_AddDir(path, prefix);
1272 free(prefix);
1273
1274 bigmisses++;
1275 if (path->last == NULL)
1276 return NULL;
1277
1278 dir = path->last->datum;
1279 if (HashSet_Contains(&dir->files, base))
1280 return bmake_strdup(name);
1281 return NULL;
1282 }
1283 #else
1284 DIR_DEBUG1(" Looking for \"%s\" ...\n", name);
1285
1286 bigmisses++;
1287 if (cached_stat(name, &cst) == 0) {
1288 return bmake_strdup(name);
1289 }
1290
1291 DIR_DEBUG0(" failed. Returning NULL\n");
1292 return NULL;
1293 #endif
1294 }
1295
1296
1297 /* Search for a path starting at a given directory and then working our way
1298 * up towards the root.
1299 *
1300 * Input:
1301 * here starting directory
1302 * search_path the relative path we are looking for
1303 *
1304 * Results:
1305 * The found path, or NULL.
1306 */
1307 char *
1308 Dir_FindHereOrAbove(const char *here, const char *search_path)
1309 {
1310 struct cached_stat cst;
1311 char *dirbase, *dirbase_end;
1312 char *try, *try_end;
1313
1314 /* copy out our starting point */
1315 dirbase = bmake_strdup(here);
1316 dirbase_end = dirbase + strlen(dirbase);
1317
1318 /* loop until we determine a result */
1319 for (;;) {
1320
1321 /* try and stat(2) it ... */
1322 try = str_concat3(dirbase, "/", search_path);
1323 if (cached_stat(try, &cst) != -1) {
1324 /*
1325 * success! if we found a file, chop off
1326 * the filename so we return a directory.
1327 */
1328 if ((cst.cst_mode & S_IFMT) != S_IFDIR) {
1329 try_end = try + strlen(try);
1330 while (try_end > try && *try_end != '/')
1331 try_end--;
1332 if (try_end > try)
1333 *try_end = '\0'; /* chop! */
1334 }
1335
1336 free(dirbase);
1337 return try;
1338 }
1339 free(try);
1340
1341 /*
1342 * nope, we didn't find it. if we used up dirbase we've
1343 * reached the root and failed.
1344 */
1345 if (dirbase_end == dirbase)
1346 break; /* failed! */
1347
1348 /*
1349 * truncate dirbase from the end to move up a dir
1350 */
1351 while (dirbase_end > dirbase && *dirbase_end != '/')
1352 dirbase_end--;
1353 *dirbase_end = '\0'; /* chop! */
1354 }
1355
1356 free(dirbase);
1357 return NULL;
1358 }
1359
1360 /*
1361 * This is an implied source, and it may have moved,
1362 * see if we can find it via the current .PATH
1363 */
1364 static char *
1365 ResolveMovedDepends(GNode *gn)
1366 {
1367 char *fullName;
1368
1369 char *base = strrchr(gn->name, '/');
1370 if (base == NULL)
1371 return NULL;
1372 base++;
1373
1374 fullName = Dir_FindFile(base, Suff_FindPath(gn));
1375 if (fullName == NULL)
1376 return NULL;
1377
1378 /*
1379 * Put the found file in gn->path so that we give that to the compiler.
1380 */
1381 /*
1382 * XXX: Better just reset gn->path to NULL; updating it is already done
1383 * by Dir_UpdateMTime.
1384 */
1385 gn->path = bmake_strdup(fullName);
1386 if (!Job_RunTarget(".STALE", gn->fname))
1387 fprintf(stdout, /* XXX: Why stdout? */
1388 "%s: %s, %d: ignoring stale %s for %s, found %s\n",
1389 progname, gn->fname, gn->lineno,
1390 makeDependfile, gn->name, fullName);
1391
1392 return fullName;
1393 }
1394
1395 static char *
1396 ResolveFullName(GNode *gn)
1397 {
1398 char *fullName;
1399
1400 fullName = gn->path;
1401 if (fullName == NULL && !(gn->type & OP_NOPATH)) {
1402
1403 fullName = Dir_FindFile(gn->name, Suff_FindPath(gn));
1404
1405 if (fullName == NULL && gn->flags & FROM_DEPEND &&
1406 !Lst_IsEmpty(&gn->implicitParents))
1407 fullName = ResolveMovedDepends(gn);
1408
1409 DIR_DEBUG2("Found '%s' as '%s'\n",
1410 gn->name, fullName ? fullName : "(not found)");
1411 }
1412
1413 if (fullName == NULL)
1414 fullName = bmake_strdup(gn->name);
1415
1416 /* XXX: Is every piece of memory freed as it should? */
1417
1418 return fullName;
1419 }
1420
1421 /* Search gn along dirSearchPath and store its modification time in gn->mtime.
1422 * If no file is found, store 0 instead.
1423 *
1424 * The found file is stored in gn->path, unless the node already had a path. */
1425 void
1426 Dir_UpdateMTime(GNode *gn, Boolean recheck)
1427 {
1428 char *fullName;
1429 struct cached_stat cst;
1430
1431 if (gn->type & OP_ARCHV) {
1432 Arch_UpdateMTime(gn);
1433 return;
1434 }
1435
1436 if (gn->type & OP_PHONY) {
1437 gn->mtime = 0;
1438 return;
1439 }
1440
1441 fullName = ResolveFullName(gn);
1442
1443 if (cached_stats(fullName, &cst, recheck ? CST_UPDATE : CST_NONE) < 0) {
1444 if (gn->type & OP_MEMBER) {
1445 if (fullName != gn->path)
1446 free(fullName);
1447 Arch_UpdateMemberMTime(gn);
1448 return;
1449 }
1450
1451 cst.cst_mtime = 0;
1452 }
1453
1454 if (fullName != NULL && gn->path == NULL)
1455 gn->path = fullName;
1456 /* XXX: else free(fullName)? */
1457
1458 gn->mtime = cst.cst_mtime;
1459 }
1460
1461 /*
1462 * Read the directory and add it to the cache in openDirs, even if it is
1463 * already there. If a path is given, add the directory to that path as
1464 * well.
1465 *
1466 * XXX: Why is it added to openDirs unconditionally?
1467 */
1468 static CachedDir *
1469 CacheNewDir(const char *name, SearchPath *path)
1470 {
1471 CachedDir *dir = NULL;
1472 DIR *d;
1473 struct dirent *dp;
1474
1475 if ((d = opendir(name)) == NULL) {
1476 DIR_DEBUG1("Caching %s ... not found\n", name);
1477 return dir;
1478 }
1479
1480 DIR_DEBUG1("Caching %s ...\n", name);
1481
1482 dir = CachedDir_New(name);
1483 CachedDir_Ref(dir); /* XXX: why here already? */
1484
1485 while ((dp = readdir(d)) != NULL) {
1486
1487 #if defined(sun) && defined(d_ino) /* d_ino is a sunos4 #define for d_fileno */
1488 /*
1489 * The sun directory library doesn't check for a 0 inode
1490 * (0-inode slots just take up space), so we have to do
1491 * it ourselves.
1492 */
1493 if (dp->d_fileno == 0)
1494 continue;
1495 #endif /* sun && d_ino */
1496
1497 (void)HashSet_Add(&dir->files, dp->d_name);
1498 }
1499 (void)closedir(d);
1500
1501 OpenDirs_Add(&openDirs, dir);
1502 if (path != NULL)
1503 Lst_Append(path, dir);
1504
1505 DIR_DEBUG1("Caching %s done\n", name);
1506 return dir;
1507 }
1508
1509 /* Read the list of filenames in the directory and store the result
1510 * in openDirs.
1511 *
1512 * If a path is given, append the directory to that path.
1513 *
1514 * Input:
1515 * path The path to which the directory should be
1516 * added, or NULL to only add the directory to openDirs
1517 * name The name of the directory to add.
1518 * The name is not normalized in any way.
1519 */
1520 CachedDir *
1521 Dir_AddDir(SearchPath *path, const char *name)
1522 {
1523
1524 if (path != NULL && strcmp(name, ".DOTLAST") == 0) {
1525 SearchPathNode *ln;
1526
1527 /* XXX: Linear search gets slow with thousands of entries. */
1528 for (ln = path->first; ln != NULL; ln = ln->next) {
1529 CachedDir *pathDir = ln->datum;
1530 if (strcmp(pathDir->name, name) == 0)
1531 return pathDir;
1532 }
1533
1534 CachedDir_Ref(dotLast);
1535 Lst_Prepend(path, dotLast);
1536 }
1537
1538 if (path != NULL) {
1539 CachedDir *dir = OpenDirs_Find(&openDirs, name);
1540 if (dir != NULL) {
1541 if (Lst_FindDatum(path, dir) == NULL)
1542 Lst_Append(path, CachedDir_Ref(dir));
1543 return dir;
1544 }
1545 }
1546
1547 return CacheNewDir(name, path);
1548 }
1549
1550 /* Return a copy of dirSearchPath, incrementing the reference counts for
1551 * the contained directories. */
1552 SearchPath *
1553 Dir_CopyDirSearchPath(void)
1554 {
1555 SearchPath *path = SearchPath_New();
1556 SearchPathNode *ln;
1557 for (ln = dirSearchPath.first; ln != NULL; ln = ln->next) {
1558 CachedDir *dir = ln->datum;
1559 Lst_Append(path, CachedDir_Ref(dir));
1560 }
1561 return path;
1562 }
1563
1564 /*-
1565 *-----------------------------------------------------------------------
1566 * SearchPath_ToFlags --
1567 * Make a string by taking all the directories in the given search
1568 * path and preceding them by the given flag. Used by the suffix
1569 * module to create variables for compilers based on suffix search
1570 * paths.
1571 *
1572 * Input:
1573 * flag flag which should precede each directory
1574 * path list of directories
1575 *
1576 * Results:
1577 * The string mentioned above. Note that there is no space between
1578 * the given flag and each directory. The empty string is returned if
1579 * Things don't go well.
1580 *
1581 * Side Effects:
1582 * None
1583 *-----------------------------------------------------------------------
1584 */
1585 char *
1586 SearchPath_ToFlags(const char *flag, SearchPath *path)
1587 {
1588 Buffer buf;
1589 SearchPathNode *ln;
1590
1591 Buf_Init(&buf);
1592
1593 if (path != NULL) {
1594 for (ln = path->first; ln != NULL; ln = ln->next) {
1595 CachedDir *dir = ln->datum;
1596 Buf_AddStr(&buf, " ");
1597 Buf_AddStr(&buf, flag);
1598 Buf_AddStr(&buf, dir->name);
1599 }
1600 }
1601
1602 return Buf_Destroy(&buf, FALSE);
1603 }
1604
1605 /* Nuke a directory descriptor, if it is no longer used. */
1606 static void
1607 CachedDir_Destroy(CachedDir *dir)
1608 {
1609 CachedDir_Unref(dir);
1610
1611 if (dir->refCount == 0) {
1612 OpenDirs_Remove(&openDirs, dir->name);
1613
1614 HashSet_Done(&dir->files);
1615 free(dir->name);
1616 free(dir);
1617 }
1618 }
1619
1620 /* Free the search path and all directories mentioned in it. */
1621 void
1622 SearchPath_Free(SearchPath *path)
1623 {
1624 SearchPathNode *ln;
1625
1626 for (ln = path->first; ln != NULL; ln = ln->next) {
1627 CachedDir *dir = ln->datum;
1628 CachedDir_Destroy(dir);
1629 }
1630 Lst_Free(path);
1631 }
1632
1633 /* Clear out all elements from the given search path.
1634 * The path is set to the empty list but is not destroyed. */
1635 void
1636 SearchPath_Clear(SearchPath *path)
1637 {
1638 while (!Lst_IsEmpty(path)) {
1639 CachedDir *dir = Lst_Dequeue(path);
1640 CachedDir_Destroy(dir);
1641 }
1642 }
1643
1644
1645 /* Concatenate two paths, adding the second to the end of the first,
1646 * skipping duplicates. */
1647 void
1648 SearchPath_AddAll(SearchPath *dst, SearchPath *src)
1649 {
1650 SearchPathNode *ln;
1651
1652 for (ln = src->first; ln != NULL; ln = ln->next) {
1653 CachedDir *dir = ln->datum;
1654 if (Lst_FindDatum(dst, dir) == NULL)
1655 Lst_Append(dst, CachedDir_Ref(dir));
1656 }
1657 }
1658
1659 static int
1660 percentage(int num, int den)
1661 {
1662 return den != 0 ? num * 100 / den : 0;
1663 }
1664
1665 /********** DEBUG INFO **********/
1666 void
1667 Dir_PrintDirectories(void)
1668 {
1669 CachedDirListNode *ln;
1670
1671 debug_printf("#*** Directory Cache:\n");
1672 debug_printf(
1673 "# Stats: %d hits %d misses %d near misses %d losers (%d%%)\n",
1674 hits, misses, nearmisses, bigmisses,
1675 percentage(hits, hits + bigmisses + nearmisses));
1676 debug_printf("# %-20s referenced\thits\n", "directory");
1677
1678 for (ln = openDirs.list.first; ln != NULL; ln = ln->next) {
1679 CachedDir *dir = ln->datum;
1680 debug_printf("# %-20s %10d\t%4d\n",
1681 dir->name, dir->refCount, dir->hits);
1682 }
1683 }
1684
1685 void
1686 SearchPath_Print(SearchPath *path)
1687 {
1688 SearchPathNode *node;
1689 for (node = path->first; node != NULL; node = node->next) {
1690 const CachedDir *dir = node->datum;
1691 debug_printf("%s ", dir->name);
1692 }
1693 }
1694