dir.c revision 1.240 1 /* $NetBSD: dir.c,v 1.240 2020/11/29 16:37:10 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 1.240 2020/11/29 16:37:10 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 /* A fake path entry indicating we need to look for '.' last. */
275 static CachedDir *dotLast = NULL;
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 OpenDirs_Remove(OpenDirs *, const char *);
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 /* Free a cached directory with reference count 0. */
315 static void
316 CachedDir_Free0(CachedDir *dir)
317 {
318
319 OpenDirs_Remove(&openDirs, dir->name);
320
321 free(dir->name);
322 HashSet_Done(&dir->files);
323 free(dir);
324 }
325
326 static void
327 CachedDir_Unref(CachedDir *dir)
328 {
329 dir->refCount--;
330 DEBUG2(DIR, "CachedDir refCount-- to %d for \"%s\"\n",
331 dir->refCount, dir->name);
332 }
333
334 /* Nuke a directory descriptor, if it is no longer used. */
335 static void
336 CachedDir_Destroy(CachedDir *dir)
337 {
338 CachedDir_Unref(dir);
339
340 if (dir->refCount == 0)
341 CachedDir_Free0(dir);
342 }
343
344 /* Update the value of the CachedDir variable, updating the reference counts. */
345 static void
346 CachedDir_Assign(CachedDir **var, CachedDir *dir)
347 {
348 CachedDir *prev;
349
350 prev = *var;
351 *var = dir;
352 if (dir != NULL)
353 CachedDir_Ref(dir);
354 if (prev != NULL)
355 CachedDir_Destroy(prev);
356 }
357
358 static void
359 OpenDirs_Init(OpenDirs *odirs)
360 {
361 Lst_Init(&odirs->list);
362 HashTable_Init(&odirs->table);
363 }
364
365 #ifdef CLEANUP
366 static void
367 OpenDirs_Done(OpenDirs *odirs)
368 {
369 CachedDirListNode *ln = odirs->list.first;
370 DIR_DEBUG1("OpenDirs_Done: %u entries to remove\n",
371 odirs->table.numEntries);
372 while (ln != NULL) {
373 CachedDirListNode *next = ln->next;
374 CachedDir *dir = ln->datum;
375 DIR_DEBUG2("OpenDirs_Done: refCount %d for \"%s\"\n",
376 dir->refCount, dir->name);
377 CachedDir_Destroy(dir); /* removes the dir from odirs->list */
378 ln = next;
379 }
380 Lst_Done(&odirs->list);
381 HashTable_Done(&odirs->table);
382 }
383 #endif
384
385 static CachedDir *
386 OpenDirs_Find(OpenDirs *odirs, const char *name)
387 {
388 CachedDirListNode *ln = HashTable_FindValue(&odirs->table, name);
389 return ln != NULL ? ln->datum : NULL;
390 }
391
392 static void
393 OpenDirs_Add(OpenDirs *odirs, CachedDir *cdir)
394 {
395 if (HashTable_FindEntry(&odirs->table, cdir->name) != NULL)
396 return;
397 Lst_Append(&odirs->list, cdir);
398 HashTable_Set(&odirs->table, cdir->name, odirs->list.last);
399 }
400
401 static void
402 OpenDirs_Remove(OpenDirs *odirs, const char *name)
403 {
404 HashEntry *he = HashTable_FindEntry(&odirs->table, name);
405 CachedDirListNode *ln;
406 if (he == NULL)
407 return;
408 ln = HashEntry_Get(he);
409 HashTable_DeleteEntry(&odirs->table, he);
410 Lst_Remove(&odirs->list, ln);
411 }
412
413 /* Returns 0 and the result of stat(2) or lstat(2) in *out_cst,
414 * or -1 on error. */
415 static int
416 cached_stats(const char *pathname, struct cached_stat *out_cst,
417 CachedStatsFlags flags)
418 {
419 HashTable *tbl = flags & CST_LSTAT ? &lmtimes : &mtimes;
420 struct stat sys_st;
421 struct cached_stat *cst;
422 int rc;
423
424 if (pathname == NULL || pathname[0] == '\0')
425 return -1; /* This can happen in meta mode. */
426
427 cst = HashTable_FindValue(tbl, pathname);
428 if (cst != NULL && !(flags & CST_UPDATE)) {
429 *out_cst = *cst;
430 DIR_DEBUG2("Using cached time %s for %s\n",
431 Targ_FmtTime(cst->cst_mtime), pathname);
432 return 0;
433 }
434
435 rc = (flags & CST_LSTAT ? lstat : stat)(pathname, &sys_st);
436 if (rc == -1)
437 return -1; /* don't cache negative lookups */
438
439 if (sys_st.st_mtime == 0)
440 sys_st.st_mtime = 1; /* avoid confusion with missing file */
441
442 if (cst == NULL) {
443 cst = bmake_malloc(sizeof *cst);
444 HashTable_Set(tbl, pathname, cst);
445 }
446
447 cst->cst_mtime = sys_st.st_mtime;
448 cst->cst_mode = sys_st.st_mode;
449
450 *out_cst = *cst;
451 DIR_DEBUG2(" Caching %s for %s\n",
452 Targ_FmtTime(sys_st.st_mtime), pathname);
453
454 return 0;
455 }
456
457 int
458 cached_stat(const char *pathname, struct cached_stat *cst)
459 {
460 return cached_stats(pathname, cst, CST_NONE);
461 }
462
463 int
464 cached_lstat(const char *pathname, struct cached_stat *cst)
465 {
466 return cached_stats(pathname, cst, CST_LSTAT);
467 }
468
469 /* Initialize the directories module. */
470 void
471 Dir_Init(void)
472 {
473 OpenDirs_Init(&openDirs);
474 HashTable_Init(&mtimes);
475 HashTable_Init(&lmtimes);
476 }
477
478 void
479 Dir_InitDir(const char *cdname)
480 {
481 Dir_InitCur(cdname);
482
483 CachedDir_Assign(&dotLast, CachedDir_New(".DOTLAST"));
484 }
485
486 /*
487 * Called by Dir_InitDir and whenever .CURDIR is assigned to.
488 */
489 void
490 Dir_InitCur(const char *cdname)
491 {
492 CachedDir *dir;
493
494 if (cdname == NULL)
495 return;
496
497 /*
498 * Our build directory is not the same as our source directory.
499 * Keep this one around too.
500 */
501 dir = Dir_AddDir(NULL, cdname);
502 if (dir == NULL)
503 return;
504
505 if (cur != NULL && cur != dir) {
506 CachedDir_Unref(cur); /* XXX: why unref twice? */
507 CachedDir_Destroy(cur);
508 }
509 cur = CachedDir_Ref(dir);
510 }
511
512 /* (Re)initialize "dot" (current/object directory) path hash.
513 * Some directories may be cached. */
514 void
515 Dir_InitDot(void)
516 {
517 if (dot != NULL) {
518 /* Remove old entry from openDirs, but do not destroy. */
519 /* XXX: Why not destroy? It's reference-counted after all. */
520 OpenDirs_Remove(&openDirs, dot->name);
521 }
522
523 /* XXX: Before assigning to the global variable, refCount++. */
524 dot = Dir_AddDir(NULL, ".");
525
526 if (dot == NULL) {
527 Error("Cannot open `.' (%s)", strerror(errno));
528 exit(1);
529 }
530
531 /*
532 * We always need to have dot around, so we increment its reference
533 * count to make sure it's not destroyed.
534 */
535 /*
536 * XXX: This is just the normal reference counting. Why is the above
537 * comment so long? And why doesn't the normal reference counting
538 * suffice? This sounds like someone misunderstood reference counting
539 * here.
540 */
541 CachedDir_Ref(dot);
542 Dir_SetPATH(); /* initialize */
543 }
544
545 /* Clean up the directories module. */
546 void
547 Dir_End(void)
548 {
549 #ifdef CLEANUP
550 if (cur != NULL) {
551 CachedDir_Unref(cur); /* XXX: why unref twice? */
552 CachedDir_Destroy(cur);
553 }
554 CachedDir_Assign(&dotLast, NULL);
555 CachedDir_Unref(dot); /* XXX: why unref twice? */
556 CachedDir_Destroy(dot);
557 SearchPath_Clear(&dirSearchPath);
558 OpenDirs_Done(&openDirs);
559 HashTable_Done(&mtimes);
560 #endif
561 }
562
563 /*
564 * We want ${.PATH} to indicate the order in which we will actually
565 * search, so we rebuild it after any .PATH: target.
566 * This is the simplest way to deal with the effect of .DOTLAST.
567 */
568 void
569 Dir_SetPATH(void)
570 {
571 CachedDirListNode *ln;
572 Boolean hasLastDot = FALSE; /* true if we should search dot last */
573
574 Var_Delete(".PATH", VAR_GLOBAL);
575
576 if ((ln = dirSearchPath.first) != NULL) {
577 CachedDir *dir = ln->datum;
578 if (dir == dotLast) {
579 hasLastDot = TRUE;
580 Var_Append(".PATH", dotLast->name, VAR_GLOBAL);
581 }
582 }
583
584 if (!hasLastDot) {
585 if (dot != NULL)
586 Var_Append(".PATH", dot->name, VAR_GLOBAL);
587 if (cur != NULL)
588 Var_Append(".PATH", cur->name, VAR_GLOBAL);
589 }
590
591 for (ln = dirSearchPath.first; ln != NULL; ln = ln->next) {
592 CachedDir *dir = ln->datum;
593 if (dir == dotLast)
594 continue;
595 if (dir == dot && hasLastDot)
596 continue;
597 Var_Append(".PATH", dir->name, VAR_GLOBAL);
598 }
599
600 if (hasLastDot) {
601 if (dot != NULL)
602 Var_Append(".PATH", dot->name, VAR_GLOBAL);
603 if (cur != NULL)
604 Var_Append(".PATH", cur->name, VAR_GLOBAL);
605 }
606 }
607
608 /* See if the given name has any wildcard characters in it and all braces and
609 * brackets are properly balanced.
610 *
611 * XXX: This code is not 100% correct ([^]] fails etc.). I really don't think
612 * that make(1) should be expanding patterns, because then you have to set a
613 * mechanism for escaping the expansion!
614 *
615 * Return TRUE if the word should be expanded, FALSE otherwise.
616 */
617 Boolean
618 Dir_HasWildcards(const char *name)
619 {
620 const char *p;
621 Boolean wild = FALSE;
622 int braces = 0, brackets = 0;
623
624 for (p = name; *p != '\0'; p++) {
625 switch (*p) {
626 case '{':
627 braces++;
628 wild = TRUE;
629 break;
630 case '}':
631 braces--;
632 break;
633 case '[':
634 brackets++;
635 wild = TRUE;
636 break;
637 case ']':
638 brackets--;
639 break;
640 case '?':
641 case '*':
642 wild = TRUE;
643 break;
644 default:
645 break;
646 }
647 }
648 return wild && brackets == 0 && braces == 0;
649 }
650
651 /* See if any files match the pattern and add their names to the 'expansions'
652 * list if they do.
653 *
654 * This is incomplete -- wildcards are only expanded in the final path
655 * component, but not in directories like src/lib*c/file*.c, but it
656 * will do for now (now being 1993 until at least 2020). To expand these,
657 * use the ':sh' variable modifier such as in ${:!echo src/lib*c/file*.c!}.
658 *
659 * Input:
660 * pattern Pattern to look for
661 * dir Directory to search
662 * expansion Place to store the results
663 */
664 static void
665 DirMatchFiles(const char *pattern, CachedDir *dir, StringList *expansions)
666 {
667 const char *dirName = dir->name;
668 Boolean isDot = dirName[0] == '.' && dirName[1] == '\0';
669 HashIter hi;
670
671 /*
672 * XXX: Iterating over all hash entries is inefficient. If the
673 * pattern is a plain string without any wildcards, a direct lookup
674 * is faster.
675 */
676
677 HashIter_InitSet(&hi, &dir->files);
678 while (HashIter_Next(&hi) != NULL) {
679 const char *base = hi.entry->key;
680
681 if (!Str_Match(base, pattern))
682 continue;
683
684 /*
685 * Follow the UNIX convention that dot files are only found
686 * if the pattern begins with a dot. The pattern '.*' does
687 * not match '.' or '..' since these are not included in the
688 * directory cache.
689 *
690 * This means that the pattern '[a-z.]*' does not find
691 * '.file', which is consistent with bash, NetBSD sh and csh.
692 */
693 if (base[0] == '.' && pattern[0] != '.')
694 continue;
695
696 {
697 char *fullName = isDot
698 ? bmake_strdup(base)
699 : str_concat3(dirName, "/", base);
700 Lst_Append(expansions, fullName);
701 }
702 }
703 }
704
705 /* Find the next closing brace in the string, taking nested braces into
706 * account. */
707 static const char *
708 closing_brace(const char *p)
709 {
710 int nest = 0;
711 while (*p != '\0') {
712 if (*p == '}' && nest == 0)
713 break;
714 if (*p == '{')
715 nest++;
716 if (*p == '}')
717 nest--;
718 p++;
719 }
720 return p;
721 }
722
723 /* Find the next closing brace or comma in the string, taking nested braces
724 * into account. */
725 static const char *
726 separator_comma(const char *p)
727 {
728 int nest = 0;
729 while (*p != '\0') {
730 if ((*p == '}' || *p == ',') && nest == 0)
731 break;
732 if (*p == '{')
733 nest++;
734 if (*p == '}')
735 nest--;
736 p++;
737 }
738 return p;
739 }
740
741 static Boolean
742 contains_wildcard(const char *p)
743 {
744 for (; *p != '\0'; p++) {
745 switch (*p) {
746 case '*':
747 case '?':
748 case '{':
749 case '[':
750 return TRUE;
751 }
752 }
753 return FALSE;
754 }
755
756 static char *
757 concat3(const char *a, size_t a_len, const char *b, size_t b_len,
758 const char *c, size_t c_len)
759 {
760 size_t s_len = a_len + b_len + c_len;
761 char *s = bmake_malloc(s_len + 1);
762 memcpy(s, a, a_len);
763 memcpy(s + a_len, b, b_len);
764 memcpy(s + a_len + b_len, c, c_len);
765 s[s_len] = '\0';
766 return s;
767 }
768
769 /* Expand curly braces like the C shell. Brace expansion by itself is purely
770 * textual, the expansions are not looked up in the file system. But if an
771 * expanded word contains wildcard characters, it is expanded further,
772 * matching only the actually existing files.
773 *
774 * Example: "{a{b,c}}" expands to "ab" and "ac".
775 * Example: "{a}" expands to "a".
776 * Example: "{a,*.c}" expands to "a" and all "*.c" files that exist.
777 *
778 * Input:
779 * word Entire word to expand
780 * brace First curly brace in it
781 * path Search path to use
782 * expansions Place to store the expansions
783 */
784 static void
785 DirExpandCurly(const char *word, const char *brace, SearchPath *path,
786 StringList *expansions)
787 {
788 const char *prefix, *middle, *piece, *middle_end, *suffix;
789 size_t prefix_len, suffix_len;
790
791 /* Split the word into prefix '{' middle '}' suffix. */
792
793 middle = brace + 1;
794 middle_end = closing_brace(middle);
795 if (*middle_end == '\0') {
796 Error("Unterminated {} clause \"%s\"", middle);
797 return;
798 }
799
800 prefix = word;
801 prefix_len = (size_t)(brace - prefix);
802 suffix = middle_end + 1;
803 suffix_len = strlen(suffix);
804
805 /* Split the middle into pieces, separated by commas. */
806
807 piece = middle;
808 while (piece < middle_end + 1) {
809 const char *piece_end = separator_comma(piece);
810 size_t piece_len = (size_t)(piece_end - piece);
811
812 char *file = concat3(prefix, prefix_len, piece, piece_len,
813 suffix, suffix_len);
814
815 if (contains_wildcard(file)) {
816 Dir_Expand(file, path, expansions);
817 free(file);
818 } else {
819 Lst_Append(expansions, file);
820 }
821
822 /* skip over the comma or closing brace */
823 piece = piece_end + 1;
824 }
825 }
826
827
828 /* Expand the word in each of the directories from the path. */
829 static void
830 DirExpandPath(const char *word, SearchPath *path, StringList *expansions)
831 {
832 SearchPathNode *ln;
833 for (ln = path->first; ln != NULL; ln = ln->next) {
834 CachedDir *dir = ln->datum;
835 DirMatchFiles(word, dir, expansions);
836 }
837 }
838
839 static void
840 PrintExpansions(StringList *expansions)
841 {
842 const char *sep = "";
843 StringListNode *ln;
844 for (ln = expansions->first; ln != NULL; ln = ln->next) {
845 const char *word = ln->datum;
846 debug_printf("%s%s", sep, word);
847 sep = " ";
848 }
849 debug_printf("\n");
850 }
851
852 /* Expand the given word into a list of words by globbing it, looking in the
853 * directories on the given search path.
854 *
855 * Input:
856 * word the word to expand
857 * path the directories in which to find the files
858 * expansions the list on which to place the results
859 */
860 void
861 Dir_Expand(const char *word, SearchPath *path, StringList *expansions)
862 {
863 const char *cp;
864
865 assert(path != NULL);
866 assert(expansions != NULL);
867
868 DIR_DEBUG1("Expanding \"%s\"... ", word);
869
870 cp = strchr(word, '{');
871 if (cp != NULL) {
872 DirExpandCurly(word, cp, path, expansions);
873 goto done;
874 }
875
876 /* At this point, the word does not contain '{'. */
877
878 cp = strchr(word, '/');
879 if (cp == NULL) {
880 /* The word has no directory component. */
881 /* First the files in dot. */
882 DirMatchFiles(word, dot, expansions);
883
884 /* Then the files in every other directory on the path. */
885 DirExpandPath(word, path, expansions);
886 goto done;
887 }
888
889 /* At this point, the word has a directory component. */
890
891 /* Find the first wildcard in the word. */
892 for (cp = word; *cp != '\0'; cp++)
893 if (*cp == '?' || *cp == '[' || *cp == '*')
894 break;
895
896 if (*cp == '\0') {
897 /*
898 * No directory component and no wildcard at all -- this
899 * should never happen as in such a simple case there is no
900 * need to expand anything.
901 */
902 DirExpandPath(word, path, expansions);
903 goto done;
904 }
905
906 /* Back up to the start of the component containing the wildcard. */
907 /* XXX: This handles '///' and '/' differently. */
908 while (cp > word && *cp != '/')
909 cp--;
910
911 if (cp == word) {
912 /* The first component contains the wildcard. */
913 /* Start the search from the local directory */
914 DirExpandPath(word, path, expansions);
915 goto done;
916 }
917
918 {
919 char *prefix = bmake_strsedup(word, cp + 1);
920 /*
921 * The wildcard isn't in the first component.
922 * Find all the components up to the one with the wildcard.
923 */
924 /*
925 * XXX: Check the "the directory is added to the path" part.
926 * It is probably surprising that the directory before a
927 * wildcard gets added to the path.
928 */
929 /*
930 * XXX: Only the first match of the prefix in the path is
931 * taken, any others are ignored. The expectation may be
932 * that the pattern is expanded in the whole path.
933 */
934 char *dirpath = Dir_FindFile(prefix, path);
935 free(prefix);
936
937 /*
938 * dirpath is null if can't find the leading component
939 * XXX: Dir_FindFile won't find internal components.
940 * i.e. if the path contains ../Etc/Object and we're
941 * looking for Etc, it won't be found. Ah well.
942 * Probably not important.
943 * XXX: Check whether the above comment is still true.
944 */
945 if (dirpath != NULL) {
946 SearchPath *partPath;
947
948 char *end = &dirpath[strlen(dirpath) - 1];
949 /* XXX: What about multiple trailing slashes? */
950 if (*end == '/')
951 *end = '\0';
952
953 partPath = SearchPath_New();
954 (void)Dir_AddDir(partPath, dirpath);
955 DirExpandPath(cp + 1, partPath, expansions);
956 Lst_Free(partPath);
957 /* XXX: Should the dirs in partPath be freed here?
958 * It's not obvious whether to free them or not. */
959 }
960 }
961
962 done:
963 if (DEBUG(DIR))
964 PrintExpansions(expansions);
965 }
966
967 /* Find if the file with the given name exists in the given path.
968 * Return the freshly allocated path to the file, or NULL. */
969 static char *
970 DirLookup(CachedDir *dir, const char *base)
971 {
972 char *file; /* the current filename to check */
973
974 DIR_DEBUG1(" %s ...\n", dir->name);
975
976 if (!HashSet_Contains(&dir->files, base))
977 return NULL;
978
979 file = str_concat3(dir->name, "/", base);
980 DIR_DEBUG1(" returning %s\n", file);
981 dir->hits++;
982 hits++;
983 return file;
984 }
985
986
987 /* Find if the file with the given name exists in the given directory.
988 * Return the freshly allocated path to the file, or NULL. */
989 static char *
990 DirLookupSubdir(CachedDir *dir, const char *name)
991 {
992 struct cached_stat cst;
993 char *file = dir == dot ? bmake_strdup(name)
994 : str_concat3(dir->name, "/", name);
995
996 DIR_DEBUG1("checking %s ...\n", file);
997
998 if (cached_stat(file, &cst) == 0) {
999 nearmisses++;
1000 return file;
1001 }
1002 free(file);
1003 return NULL;
1004 }
1005
1006 /* Find if the file with the given name exists in the given path.
1007 * Return the freshly allocated path to the file, the empty string, or NULL.
1008 * Returning the empty string means that the search should be terminated.
1009 */
1010 static char *
1011 DirLookupAbs(CachedDir *dir, const char *name, const char *cp)
1012 {
1013 const char *dnp; /* pointer into dir->name */
1014 const char *np; /* pointer into name */
1015
1016 DIR_DEBUG1(" %s ...\n", dir->name);
1017
1018 /*
1019 * If the file has a leading path component and that component
1020 * exactly matches the entire name of the current search
1021 * directory, we can attempt another cache lookup. And if we don't
1022 * have a hit, we can safely assume the file does not exist at all.
1023 */
1024 for (dnp = dir->name, np = name;
1025 *dnp != '\0' && *dnp == *np; dnp++, np++)
1026 continue;
1027 if (*dnp != '\0' || np != cp - 1)
1028 return NULL;
1029
1030 if (!HashSet_Contains(&dir->files, cp)) {
1031 DIR_DEBUG0(" must be here but isn't -- returning\n");
1032 return bmake_strdup(""); /* to terminate the search */
1033 }
1034
1035 dir->hits++;
1036 hits++;
1037 DIR_DEBUG1(" returning %s\n", name);
1038 return bmake_strdup(name);
1039 }
1040
1041 /* Find the file given on "." or curdir.
1042 * Return the freshly allocated path to the file, or NULL. */
1043 static char *
1044 DirFindDot(const char *name, const char *base)
1045 {
1046
1047 if (HashSet_Contains(&dot->files, base)) {
1048 DIR_DEBUG0(" in '.'\n");
1049 hits++;
1050 dot->hits++;
1051 return bmake_strdup(name);
1052 }
1053
1054 if (cur != NULL && HashSet_Contains(&cur->files, base)) {
1055 DIR_DEBUG1(" in ${.CURDIR} = %s\n", cur->name);
1056 hits++;
1057 cur->hits++;
1058 return str_concat3(cur->name, "/", base);
1059 }
1060
1061 return NULL;
1062 }
1063
1064 /* Find the file with the given name along the given search path.
1065 *
1066 * If the file is found in a directory that is not on the path
1067 * already (either 'name' is absolute or it is a relative path
1068 * [ dir1/.../dirn/file ] which exists below one of the directories
1069 * already on the search path), its directory is added to the end
1070 * of the path, on the assumption that there will be more files in
1071 * that directory later on. Sometimes this is true. Sometimes not.
1072 *
1073 * Input:
1074 * name the file to find
1075 * path the directories to search, or NULL
1076 *
1077 * Results:
1078 * The freshly allocated path to the file, or NULL.
1079 */
1080 char *
1081 Dir_FindFile(const char *name, SearchPath *path)
1082 {
1083 char *file; /* the current filename to check */
1084 const char *base; /* Terminal name of file */
1085 Boolean hasLastDot = FALSE; /* true if we should search dot last */
1086 Boolean hasSlash; /* true if 'name' contains a / */
1087 struct cached_stat cst; /* Buffer for stat, if necessary */
1088 const char *trailing_dot = ".";
1089
1090 /*
1091 * Find the final component of the name and note whether it has a
1092 * slash in it (the name, I mean)
1093 */
1094 base = strrchr(name, '/');
1095 if (base != NULL) {
1096 hasSlash = TRUE;
1097 base++;
1098 } else {
1099 hasSlash = FALSE;
1100 base = name;
1101 }
1102
1103 DIR_DEBUG1("Searching for %s ...", name);
1104
1105 if (path == NULL) {
1106 DIR_DEBUG0("couldn't open path, file not found\n");
1107 misses++;
1108 return NULL;
1109 }
1110
1111 if (path->first != NULL) {
1112 CachedDir *dir = path->first->datum;
1113 if (dir == dotLast) {
1114 hasLastDot = TRUE;
1115 DIR_DEBUG0("[dot last]...");
1116 }
1117 }
1118 DIR_DEBUG0("\n");
1119
1120 /*
1121 * If there's no leading directory components or if the leading
1122 * directory component is exactly `./', consult the cached contents
1123 * of each of the directories on the search path.
1124 */
1125 if (!hasSlash || (base - name == 2 && *name == '.')) {
1126 SearchPathNode *ln;
1127
1128 /*
1129 * We look through all the directories on the path seeking one
1130 * which contains the final component of the given name. If
1131 * such a beast is found, we concatenate the directory name
1132 * and the final component and return the resulting string.
1133 * If we don't find any such thing, we go on to phase two.
1134 *
1135 * No matter what, we always look for the file in the current
1136 * directory before anywhere else (unless we found the magic
1137 * DOTLAST path, in which case we search it last) and we *do
1138 * not* add the ./ to it if it exists.
1139 * This is so there are no conflicts between what the user
1140 * specifies (fish.c) and what pmake finds (./fish.c).
1141 */
1142 if (!hasLastDot && (file = DirFindDot(name, base)) != NULL)
1143 return file;
1144
1145 for (ln = path->first; ln != NULL; ln = ln->next) {
1146 CachedDir *dir = ln->datum;
1147 if (dir == dotLast)
1148 continue;
1149 if ((file = DirLookup(dir, base)) != NULL)
1150 return file;
1151 }
1152
1153 if (hasLastDot && (file = DirFindDot(name, base)) != NULL)
1154 return file;
1155 }
1156
1157 /*
1158 * We didn't find the file on any directory in the search path.
1159 * If the name doesn't contain a slash, that means it doesn't exist.
1160 * If it *does* contain a slash, however, there is still hope: it
1161 * could be in a subdirectory of one of the members of the search
1162 * path. (eg. /usr/include and sys/types.h. The above search would
1163 * fail to turn up types.h in /usr/include, but it *is* in
1164 * /usr/include/sys/types.h).
1165 * [ This no longer applies: If we find such a beast, we assume there
1166 * will be more (what else can we assume?) and add all but the last
1167 * component of the resulting name onto the search path (at the
1168 * end).]
1169 * This phase is only performed if the file is *not* absolute.
1170 */
1171 if (!hasSlash) {
1172 DIR_DEBUG0(" failed.\n");
1173 misses++;
1174 return NULL;
1175 }
1176
1177 if (*base == '\0') {
1178 /* we were given a trailing "/" */
1179 base = trailing_dot;
1180 }
1181
1182 if (name[0] != '/') {
1183 SearchPathNode *ln;
1184 Boolean checkedDot = FALSE;
1185
1186 DIR_DEBUG0(" Trying subdirectories...\n");
1187
1188 if (!hasLastDot) {
1189 if (dot != NULL) {
1190 checkedDot = TRUE;
1191 if ((file = DirLookupSubdir(dot, name)) != NULL)
1192 return file;
1193 }
1194 if (cur && (file = DirLookupSubdir(cur, name)) != NULL)
1195 return file;
1196 }
1197
1198 for (ln = path->first; ln != NULL; ln = ln->next) {
1199 CachedDir *dir = ln->datum;
1200 if (dir == dotLast)
1201 continue;
1202 if (dir == dot) {
1203 if (checkedDot)
1204 continue;
1205 checkedDot = TRUE;
1206 }
1207 if ((file = DirLookupSubdir(dir, name)) != NULL)
1208 return file;
1209 }
1210
1211 if (hasLastDot) {
1212 if (dot && !checkedDot) {
1213 checkedDot = TRUE;
1214 if ((file = DirLookupSubdir(dot, name)) != NULL)
1215 return file;
1216 }
1217 if (cur && (file = DirLookupSubdir(cur, name)) != NULL)
1218 return file;
1219 }
1220
1221 if (checkedDot) {
1222 /*
1223 * Already checked by the given name, since . was in
1224 * the path, so no point in proceeding.
1225 */
1226 DIR_DEBUG0(" Checked . already, returning NULL\n");
1227 return NULL;
1228 }
1229
1230 } else { /* name[0] == '/' */
1231 SearchPathNode *ln;
1232
1233 /*
1234 * For absolute names, compare directory path prefix against
1235 * the the directory path of each member on the search path
1236 * for an exact match. If we have an exact match on any member
1237 * of the search path, use the cached contents of that member
1238 * to lookup the final file component. If that lookup fails we
1239 * can safely assume that the file does not exist at all.
1240 * This is signified by DirLookupAbs() returning an empty
1241 * string.
1242 */
1243 DIR_DEBUG0(" Trying exact path matches...\n");
1244
1245 if (!hasLastDot && cur &&
1246 ((file = DirLookupAbs(cur, name, base)) != NULL)) {
1247 if (file[0] == '\0') {
1248 free(file);
1249 return NULL;
1250 }
1251 return file;
1252 }
1253
1254 for (ln = path->first; ln != NULL; ln = ln->next) {
1255 CachedDir *dir = ln->datum;
1256 if (dir == dotLast)
1257 continue;
1258 if ((file = DirLookupAbs(dir, name, base)) != NULL) {
1259 if (file[0] == '\0') {
1260 free(file);
1261 return NULL;
1262 }
1263 return file;
1264 }
1265 }
1266
1267 if (hasLastDot && cur &&
1268 ((file = DirLookupAbs(cur, name, base)) != NULL)) {
1269 if (file[0] == '\0') {
1270 free(file);
1271 return NULL;
1272 }
1273 return file;
1274 }
1275 }
1276
1277 /*
1278 * Didn't find it that way, either. Sigh. Phase 3. Add its directory
1279 * onto the search path in any case, just in case, then look for the
1280 * thing in the hash table. If we find it, grand. We return a new
1281 * copy of the name. Otherwise we sadly return a NULL pointer. Sigh.
1282 * Note that if the directory holding the file doesn't exist, this
1283 * will do an extra search of the final directory on the path. Unless
1284 * something weird happens, this search won't succeed and life will
1285 * be groovy.
1286 *
1287 * Sigh. We cannot add the directory onto the search path because
1288 * of this amusing case:
1289 * $(INSTALLDIR)/$(FILE): $(FILE)
1290 *
1291 * $(FILE) exists in $(INSTALLDIR) but not in the current one.
1292 * When searching for $(FILE), we will find it in $(INSTALLDIR)
1293 * b/c we added it here. This is not good...
1294 */
1295 #if 0
1296 {
1297 CachedDir *dir;
1298 char *prefix;
1299
1300 if (base == trailing_dot) {
1301 base = strrchr(name, '/');
1302 base++;
1303 }
1304 prefix = bmake_strsedup(name, base - 1);
1305 (void)Dir_AddDir(path, prefix);
1306 free(prefix);
1307
1308 bigmisses++;
1309 if (path->last == NULL)
1310 return NULL;
1311
1312 dir = path->last->datum;
1313 if (HashSet_Contains(&dir->files, base))
1314 return bmake_strdup(name);
1315 return NULL;
1316 }
1317 #else
1318 DIR_DEBUG1(" Looking for \"%s\" ...\n", name);
1319
1320 bigmisses++;
1321 if (cached_stat(name, &cst) == 0) {
1322 return bmake_strdup(name);
1323 }
1324
1325 DIR_DEBUG0(" failed. Returning NULL\n");
1326 return NULL;
1327 #endif
1328 }
1329
1330
1331 /* Search for a path starting at a given directory and then working our way
1332 * up towards the root.
1333 *
1334 * Input:
1335 * here starting directory
1336 * search_path the relative path we are looking for
1337 *
1338 * Results:
1339 * The found path, or NULL.
1340 */
1341 char *
1342 Dir_FindHereOrAbove(const char *here, const char *search_path)
1343 {
1344 struct cached_stat cst;
1345 char *dirbase, *dirbase_end;
1346 char *try, *try_end;
1347
1348 /* copy out our starting point */
1349 dirbase = bmake_strdup(here);
1350 dirbase_end = dirbase + strlen(dirbase);
1351
1352 /* loop until we determine a result */
1353 for (;;) {
1354
1355 /* try and stat(2) it ... */
1356 try = str_concat3(dirbase, "/", search_path);
1357 if (cached_stat(try, &cst) != -1) {
1358 /*
1359 * success! if we found a file, chop off
1360 * the filename so we return a directory.
1361 */
1362 if ((cst.cst_mode & S_IFMT) != S_IFDIR) {
1363 try_end = try + strlen(try);
1364 while (try_end > try && *try_end != '/')
1365 try_end--;
1366 if (try_end > try)
1367 *try_end = '\0'; /* chop! */
1368 }
1369
1370 free(dirbase);
1371 return try;
1372 }
1373 free(try);
1374
1375 /*
1376 * nope, we didn't find it. if we used up dirbase we've
1377 * reached the root and failed.
1378 */
1379 if (dirbase_end == dirbase)
1380 break; /* failed! */
1381
1382 /*
1383 * truncate dirbase from the end to move up a dir
1384 */
1385 while (dirbase_end > dirbase && *dirbase_end != '/')
1386 dirbase_end--;
1387 *dirbase_end = '\0'; /* chop! */
1388 }
1389
1390 free(dirbase);
1391 return NULL;
1392 }
1393
1394 /*
1395 * This is an implied source, and it may have moved,
1396 * see if we can find it via the current .PATH
1397 */
1398 static char *
1399 ResolveMovedDepends(GNode *gn)
1400 {
1401 char *fullName;
1402
1403 char *base = strrchr(gn->name, '/');
1404 if (base == NULL)
1405 return NULL;
1406 base++;
1407
1408 fullName = Dir_FindFile(base, Suff_FindPath(gn));
1409 if (fullName == NULL)
1410 return NULL;
1411
1412 /*
1413 * Put the found file in gn->path so that we give that to the compiler.
1414 */
1415 /*
1416 * XXX: Better just reset gn->path to NULL; updating it is already done
1417 * by Dir_UpdateMTime.
1418 */
1419 gn->path = bmake_strdup(fullName);
1420 if (!Job_RunTarget(".STALE", gn->fname))
1421 fprintf(stdout, /* XXX: Why stdout? */
1422 "%s: %s, %d: ignoring stale %s for %s, found %s\n",
1423 progname, gn->fname, gn->lineno,
1424 makeDependfile, gn->name, fullName);
1425
1426 return fullName;
1427 }
1428
1429 static char *
1430 ResolveFullName(GNode *gn)
1431 {
1432 char *fullName;
1433
1434 fullName = gn->path;
1435 if (fullName == NULL && !(gn->type & OP_NOPATH)) {
1436
1437 fullName = Dir_FindFile(gn->name, Suff_FindPath(gn));
1438
1439 if (fullName == NULL && gn->flags & FROM_DEPEND &&
1440 !Lst_IsEmpty(&gn->implicitParents))
1441 fullName = ResolveMovedDepends(gn);
1442
1443 DIR_DEBUG2("Found '%s' as '%s'\n",
1444 gn->name, fullName ? fullName : "(not found)");
1445 }
1446
1447 if (fullName == NULL)
1448 fullName = bmake_strdup(gn->name);
1449
1450 /* XXX: Is every piece of memory freed as it should? */
1451
1452 return fullName;
1453 }
1454
1455 /* Search gn along dirSearchPath and store its modification time in gn->mtime.
1456 * If no file is found, store 0 instead.
1457 *
1458 * The found file is stored in gn->path, unless the node already had a path. */
1459 void
1460 Dir_UpdateMTime(GNode *gn, Boolean recheck)
1461 {
1462 char *fullName;
1463 struct cached_stat cst;
1464
1465 if (gn->type & OP_ARCHV) {
1466 Arch_UpdateMTime(gn);
1467 return;
1468 }
1469
1470 if (gn->type & OP_PHONY) {
1471 gn->mtime = 0;
1472 return;
1473 }
1474
1475 fullName = ResolveFullName(gn);
1476
1477 if (cached_stats(fullName, &cst, recheck ? CST_UPDATE : CST_NONE) < 0) {
1478 if (gn->type & OP_MEMBER) {
1479 if (fullName != gn->path)
1480 free(fullName);
1481 Arch_UpdateMemberMTime(gn);
1482 return;
1483 }
1484
1485 cst.cst_mtime = 0;
1486 }
1487
1488 if (fullName != NULL && gn->path == NULL)
1489 gn->path = fullName;
1490 /* XXX: else free(fullName)? */
1491
1492 gn->mtime = cst.cst_mtime;
1493 }
1494
1495 /*
1496 * Read the directory and add it to the cache in openDirs, even if it is
1497 * already there. If a path is given, add the directory to that path as
1498 * well.
1499 *
1500 * XXX: Why is it added to openDirs unconditionally?
1501 */
1502 static CachedDir *
1503 CacheNewDir(const char *name, SearchPath *path)
1504 {
1505 CachedDir *dir = NULL;
1506 DIR *d;
1507 struct dirent *dp;
1508
1509 if ((d = opendir(name)) == NULL) {
1510 DIR_DEBUG1("Caching %s ... not found\n", name);
1511 return dir;
1512 }
1513
1514 DIR_DEBUG1("Caching %s ...\n", name);
1515
1516 dir = CachedDir_New(name);
1517 CachedDir_Ref(dir); /* XXX: why here already? */
1518
1519 while ((dp = readdir(d)) != NULL) {
1520
1521 #if defined(sun) && defined(d_ino) /* d_ino is a sunos4 #define for d_fileno */
1522 /*
1523 * The sun directory library doesn't check for a 0 inode
1524 * (0-inode slots just take up space), so we have to do
1525 * it ourselves.
1526 */
1527 if (dp->d_fileno == 0)
1528 continue;
1529 #endif /* sun && d_ino */
1530
1531 (void)HashSet_Add(&dir->files, dp->d_name);
1532 }
1533 (void)closedir(d);
1534
1535 OpenDirs_Add(&openDirs, dir);
1536 if (path != NULL)
1537 Lst_Append(path, dir);
1538
1539 DIR_DEBUG1("Caching %s done\n", name);
1540 return dir;
1541 }
1542
1543 /* Read the list of filenames in the directory and store the result
1544 * in openDirs.
1545 *
1546 * If a path is given, append the directory to that path.
1547 *
1548 * Input:
1549 * path The path to which the directory should be
1550 * added, or NULL to only add the directory to openDirs
1551 * name The name of the directory to add.
1552 * The name is not normalized in any way.
1553 */
1554 CachedDir *
1555 Dir_AddDir(SearchPath *path, const char *name)
1556 {
1557
1558 if (path != NULL && strcmp(name, ".DOTLAST") == 0) {
1559 SearchPathNode *ln;
1560
1561 /* XXX: Linear search gets slow with thousands of entries. */
1562 for (ln = path->first; ln != NULL; ln = ln->next) {
1563 CachedDir *pathDir = ln->datum;
1564 if (strcmp(pathDir->name, name) == 0)
1565 return pathDir;
1566 }
1567
1568 CachedDir_Ref(dotLast);
1569 Lst_Prepend(path, dotLast);
1570 }
1571
1572 if (path != NULL) {
1573 CachedDir *dir = OpenDirs_Find(&openDirs, name);
1574 if (dir != NULL) {
1575 if (Lst_FindDatum(path, dir) == NULL)
1576 Lst_Append(path, CachedDir_Ref(dir));
1577 return dir;
1578 }
1579 }
1580
1581 return CacheNewDir(name, path);
1582 }
1583
1584 /* Return a copy of dirSearchPath, incrementing the reference counts for
1585 * the contained directories. */
1586 SearchPath *
1587 Dir_CopyDirSearchPath(void)
1588 {
1589 SearchPath *path = SearchPath_New();
1590 SearchPathNode *ln;
1591 for (ln = dirSearchPath.first; ln != NULL; ln = ln->next) {
1592 CachedDir *dir = ln->datum;
1593 Lst_Append(path, CachedDir_Ref(dir));
1594 }
1595 return path;
1596 }
1597
1598 /*-
1599 *-----------------------------------------------------------------------
1600 * SearchPath_ToFlags --
1601 * Make a string by taking all the directories in the given search
1602 * path and preceding them by the given flag. Used by the suffix
1603 * module to create variables for compilers based on suffix search
1604 * paths.
1605 *
1606 * Input:
1607 * flag flag which should precede each directory
1608 * path list of directories
1609 *
1610 * Results:
1611 * The string mentioned above. Note that there is no space between
1612 * the given flag and each directory. The empty string is returned if
1613 * Things don't go well.
1614 *
1615 * Side Effects:
1616 * None
1617 *-----------------------------------------------------------------------
1618 */
1619 char *
1620 SearchPath_ToFlags(const char *flag, SearchPath *path)
1621 {
1622 Buffer buf;
1623 SearchPathNode *ln;
1624
1625 Buf_Init(&buf);
1626
1627 if (path != NULL) {
1628 for (ln = path->first; ln != NULL; ln = ln->next) {
1629 CachedDir *dir = ln->datum;
1630 Buf_AddStr(&buf, " ");
1631 Buf_AddStr(&buf, flag);
1632 Buf_AddStr(&buf, dir->name);
1633 }
1634 }
1635
1636 return Buf_Destroy(&buf, FALSE);
1637 }
1638
1639 /* Free the search path and all directories mentioned in it. */
1640 void
1641 SearchPath_Free(SearchPath *path)
1642 {
1643 SearchPathNode *ln;
1644
1645 for (ln = path->first; ln != NULL; ln = ln->next) {
1646 CachedDir *dir = ln->datum;
1647 CachedDir_Destroy(dir);
1648 }
1649 Lst_Free(path);
1650 }
1651
1652 /* Clear out all elements from the given search path.
1653 * The path is set to the empty list but is not destroyed. */
1654 void
1655 SearchPath_Clear(SearchPath *path)
1656 {
1657 while (!Lst_IsEmpty(path)) {
1658 CachedDir *dir = Lst_Dequeue(path);
1659 CachedDir_Destroy(dir);
1660 }
1661 }
1662
1663
1664 /* Concatenate two paths, adding the second to the end of the first,
1665 * skipping duplicates. */
1666 void
1667 SearchPath_AddAll(SearchPath *dst, SearchPath *src)
1668 {
1669 SearchPathNode *ln;
1670
1671 for (ln = src->first; ln != NULL; ln = ln->next) {
1672 CachedDir *dir = ln->datum;
1673 if (Lst_FindDatum(dst, dir) == NULL)
1674 Lst_Append(dst, CachedDir_Ref(dir));
1675 }
1676 }
1677
1678 static int
1679 percentage(int num, int den)
1680 {
1681 return den != 0 ? num * 100 / den : 0;
1682 }
1683
1684 /********** DEBUG INFO **********/
1685 void
1686 Dir_PrintDirectories(void)
1687 {
1688 CachedDirListNode *ln;
1689
1690 debug_printf("#*** Directory Cache:\n");
1691 debug_printf(
1692 "# Stats: %d hits %d misses %d near misses %d losers (%d%%)\n",
1693 hits, misses, nearmisses, bigmisses,
1694 percentage(hits, hits + bigmisses + nearmisses));
1695 debug_printf("# %-20s referenced\thits\n", "directory");
1696
1697 for (ln = openDirs.list.first; ln != NULL; ln = ln->next) {
1698 CachedDir *dir = ln->datum;
1699 debug_printf("# %-20s %10d\t%4d\n",
1700 dir->name, dir->refCount, dir->hits);
1701 }
1702 }
1703
1704 void
1705 SearchPath_Print(SearchPath *path)
1706 {
1707 SearchPathNode *node;
1708 for (node = path->first; node != NULL; node = node->next) {
1709 const CachedDir *dir = node->datum;
1710 debug_printf("%s ", dir->name);
1711 }
1712 }
1713