dir.c revision 1.236 1 /* $NetBSD: dir.c,v 1.236 2020/11/29 14:29:19 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 1.236 2020/11/29 14:29:19 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 /* XXX: Reference counting is wrong here.
471 * If this function is called repeatedly with the same directory name,
472 * its reference count increases each time even though the number of
473 * actual references stays the same. */
474
475 CachedDir_Ref(dir); /* XXX: This can be expressed clearer. */
476 if (cur != NULL && cur != dir) {
477 /*
478 * We've been here before, clean up.
479 */
480 CachedDir_Unref(cur); /* XXX: why unref twice? */
481 CachedDir_Destroy(cur);
482 }
483 cur = dir;
484 }
485
486 /* (Re)initialize "dot" (current/object directory) path hash.
487 * Some directories may be cached. */
488 void
489 Dir_InitDot(void)
490 {
491 if (dot != NULL) {
492 /* Remove old entry from openDirs, but do not destroy. */
493 /* XXX: Why not destroy? It's reference-counted after all. */
494 OpenDirs_Remove(&openDirs, dot->name);
495 }
496
497 /* XXX: Before assigning to the global variable, refCount++. */
498 dot = Dir_AddDir(NULL, ".");
499
500 if (dot == NULL) {
501 Error("Cannot open `.' (%s)", strerror(errno));
502 exit(1);
503 }
504
505 /*
506 * We always need to have dot around, so we increment its reference
507 * count to make sure it's not destroyed.
508 */
509 /*
510 * XXX: This is just the normal reference counting. Why is the above
511 * comment so long? And why doesn't the normal reference counting
512 * suffice? This sounds like someone misunderstood reference counting
513 * here.
514 */
515 CachedDir_Ref(dot);
516 Dir_SetPATH(); /* initialize */
517 }
518
519 /* Clean up the directories module. */
520 void
521 Dir_End(void)
522 {
523 #ifdef CLEANUP
524 if (cur != NULL) {
525 CachedDir_Unref(cur); /* XXX: why unref twice? */
526 CachedDir_Destroy(cur);
527 }
528 CachedDir_Unref(dotLast); /* XXX: why unref twice? */
529 CachedDir_Destroy(dotLast);
530 CachedDir_Unref(dot); /* XXX: why unref twice? */
531 CachedDir_Destroy(dot);
532 SearchPath_Clear(&dirSearchPath);
533 OpenDirs_Done(&openDirs);
534 HashTable_Done(&mtimes);
535 #endif
536 }
537
538 /*
539 * We want ${.PATH} to indicate the order in which we will actually
540 * search, so we rebuild it after any .PATH: target.
541 * This is the simplest way to deal with the effect of .DOTLAST.
542 */
543 void
544 Dir_SetPATH(void)
545 {
546 CachedDirListNode *ln;
547 Boolean hasLastDot = FALSE; /* true if we should search dot last */
548
549 Var_Delete(".PATH", VAR_GLOBAL);
550
551 if ((ln = dirSearchPath.first) != NULL) {
552 CachedDir *dir = ln->datum;
553 if (dir == dotLast) {
554 hasLastDot = TRUE;
555 Var_Append(".PATH", dotLast->name, VAR_GLOBAL);
556 }
557 }
558
559 if (!hasLastDot) {
560 if (dot != NULL)
561 Var_Append(".PATH", dot->name, VAR_GLOBAL);
562 if (cur != NULL)
563 Var_Append(".PATH", cur->name, VAR_GLOBAL);
564 }
565
566 for (ln = dirSearchPath.first; ln != NULL; ln = ln->next) {
567 CachedDir *dir = ln->datum;
568 if (dir == dotLast)
569 continue;
570 if (dir == dot && hasLastDot)
571 continue;
572 Var_Append(".PATH", dir->name, VAR_GLOBAL);
573 }
574
575 if (hasLastDot) {
576 if (dot != NULL)
577 Var_Append(".PATH", dot->name, VAR_GLOBAL);
578 if (cur != NULL)
579 Var_Append(".PATH", cur->name, VAR_GLOBAL);
580 }
581 }
582
583 /* See if the given name has any wildcard characters in it and all braces and
584 * brackets are properly balanced.
585 *
586 * XXX: This code is not 100% correct ([^]] fails etc.). I really don't think
587 * that make(1) should be expanding patterns, because then you have to set a
588 * mechanism for escaping the expansion!
589 *
590 * Return TRUE if the word should be expanded, FALSE otherwise.
591 */
592 Boolean
593 Dir_HasWildcards(const char *name)
594 {
595 const char *p;
596 Boolean wild = FALSE;
597 int braces = 0, brackets = 0;
598
599 for (p = name; *p != '\0'; p++) {
600 switch (*p) {
601 case '{':
602 braces++;
603 wild = TRUE;
604 break;
605 case '}':
606 braces--;
607 break;
608 case '[':
609 brackets++;
610 wild = TRUE;
611 break;
612 case ']':
613 brackets--;
614 break;
615 case '?':
616 case '*':
617 wild = TRUE;
618 break;
619 default:
620 break;
621 }
622 }
623 return wild && brackets == 0 && braces == 0;
624 }
625
626 /* See if any files match the pattern and add their names to the 'expansions'
627 * list if they do.
628 *
629 * This is incomplete -- wildcards are only expanded in the final path
630 * component, but not in directories like src/lib*c/file*.c, but it
631 * will do for now (now being 1993 until at least 2020). To expand these,
632 * use the ':sh' variable modifier such as in ${:!echo src/lib*c/file*.c!}.
633 *
634 * Input:
635 * pattern Pattern to look for
636 * dir Directory to search
637 * expansion Place to store the results
638 */
639 static void
640 DirMatchFiles(const char *pattern, CachedDir *dir, StringList *expansions)
641 {
642 const char *dirName = dir->name;
643 Boolean isDot = dirName[0] == '.' && dirName[1] == '\0';
644 HashIter hi;
645
646 /*
647 * XXX: Iterating over all hash entries is inefficient. If the
648 * pattern is a plain string without any wildcards, a direct lookup
649 * is faster.
650 */
651
652 HashIter_InitSet(&hi, &dir->files);
653 while (HashIter_Next(&hi) != NULL) {
654 const char *base = hi.entry->key;
655
656 if (!Str_Match(base, pattern))
657 continue;
658
659 /*
660 * Follow the UNIX convention that dot files are only found
661 * if the pattern begins with a dot. The pattern '.*' does
662 * not match '.' or '..' since these are not included in the
663 * directory cache.
664 *
665 * This means that the pattern '[a-z.]*' does not find
666 * '.file', which is consistent with bash, NetBSD sh and csh.
667 */
668 if (base[0] == '.' && pattern[0] != '.')
669 continue;
670
671 {
672 char *fullName = isDot
673 ? bmake_strdup(base)
674 : str_concat3(dirName, "/", base);
675 Lst_Append(expansions, fullName);
676 }
677 }
678 }
679
680 /* Find the next closing brace in the string, taking nested braces into
681 * account. */
682 static const char *
683 closing_brace(const char *p)
684 {
685 int nest = 0;
686 while (*p != '\0') {
687 if (*p == '}' && nest == 0)
688 break;
689 if (*p == '{')
690 nest++;
691 if (*p == '}')
692 nest--;
693 p++;
694 }
695 return p;
696 }
697
698 /* Find the next closing brace or comma in the string, taking nested braces
699 * into account. */
700 static const char *
701 separator_comma(const char *p)
702 {
703 int nest = 0;
704 while (*p != '\0') {
705 if ((*p == '}' || *p == ',') && nest == 0)
706 break;
707 if (*p == '{')
708 nest++;
709 if (*p == '}')
710 nest--;
711 p++;
712 }
713 return p;
714 }
715
716 static Boolean
717 contains_wildcard(const char *p)
718 {
719 for (; *p != '\0'; p++) {
720 switch (*p) {
721 case '*':
722 case '?':
723 case '{':
724 case '[':
725 return TRUE;
726 }
727 }
728 return FALSE;
729 }
730
731 static char *
732 concat3(const char *a, size_t a_len, const char *b, size_t b_len,
733 const char *c, size_t c_len)
734 {
735 size_t s_len = a_len + b_len + c_len;
736 char *s = bmake_malloc(s_len + 1);
737 memcpy(s, a, a_len);
738 memcpy(s + a_len, b, b_len);
739 memcpy(s + a_len + b_len, c, c_len);
740 s[s_len] = '\0';
741 return s;
742 }
743
744 /* Expand curly braces like the C shell. Brace expansion by itself is purely
745 * textual, the expansions are not looked up in the file system. But if an
746 * expanded word contains wildcard characters, it is expanded further,
747 * matching only the actually existing files.
748 *
749 * Example: "{a{b,c}}" expands to "ab" and "ac".
750 * Example: "{a}" expands to "a".
751 * Example: "{a,*.c}" expands to "a" and all "*.c" files that exist.
752 *
753 * Input:
754 * word Entire word to expand
755 * brace First curly brace in it
756 * path Search path to use
757 * expansions Place to store the expansions
758 */
759 static void
760 DirExpandCurly(const char *word, const char *brace, SearchPath *path,
761 StringList *expansions)
762 {
763 const char *prefix, *middle, *piece, *middle_end, *suffix;
764 size_t prefix_len, suffix_len;
765
766 /* Split the word into prefix '{' middle '}' suffix. */
767
768 middle = brace + 1;
769 middle_end = closing_brace(middle);
770 if (*middle_end == '\0') {
771 Error("Unterminated {} clause \"%s\"", middle);
772 return;
773 }
774
775 prefix = word;
776 prefix_len = (size_t)(brace - prefix);
777 suffix = middle_end + 1;
778 suffix_len = strlen(suffix);
779
780 /* Split the middle into pieces, separated by commas. */
781
782 piece = middle;
783 while (piece < middle_end + 1) {
784 const char *piece_end = separator_comma(piece);
785 size_t piece_len = (size_t)(piece_end - piece);
786
787 char *file = concat3(prefix, prefix_len, piece, piece_len,
788 suffix, suffix_len);
789
790 if (contains_wildcard(file)) {
791 Dir_Expand(file, path, expansions);
792 free(file);
793 } else {
794 Lst_Append(expansions, file);
795 }
796
797 /* skip over the comma or closing brace */
798 piece = piece_end + 1;
799 }
800 }
801
802
803 /* Expand the word in each of the directories from the path. */
804 static void
805 DirExpandPath(const char *word, SearchPath *path, StringList *expansions)
806 {
807 SearchPathNode *ln;
808 for (ln = path->first; ln != NULL; ln = ln->next) {
809 CachedDir *dir = ln->datum;
810 DirMatchFiles(word, dir, expansions);
811 }
812 }
813
814 static void
815 PrintExpansions(StringList *expansions)
816 {
817 const char *sep = "";
818 StringListNode *ln;
819 for (ln = expansions->first; ln != NULL; ln = ln->next) {
820 const char *word = ln->datum;
821 debug_printf("%s%s", sep, word);
822 sep = " ";
823 }
824 debug_printf("\n");
825 }
826
827 /* Expand the given word into a list of words by globbing it, looking in the
828 * directories on the given search path.
829 *
830 * Input:
831 * word the word to expand
832 * path the directories in which to find the files
833 * expansions the list on which to place the results
834 */
835 void
836 Dir_Expand(const char *word, SearchPath *path, StringList *expansions)
837 {
838 const char *cp;
839
840 assert(path != NULL);
841 assert(expansions != NULL);
842
843 DIR_DEBUG1("Expanding \"%s\"... ", word);
844
845 cp = strchr(word, '{');
846 if (cp != NULL) {
847 DirExpandCurly(word, cp, path, expansions);
848 goto done;
849 }
850
851 /* At this point, the word does not contain '{'. */
852
853 cp = strchr(word, '/');
854 if (cp == NULL) {
855 /* The word has no directory component. */
856 /* First the files in dot. */
857 DirMatchFiles(word, dot, expansions);
858
859 /* Then the files in every other directory on the path. */
860 DirExpandPath(word, path, expansions);
861 goto done;
862 }
863
864 /* At this point, the word has a directory component. */
865
866 /* Find the first wildcard in the word. */
867 for (cp = word; *cp != '\0'; cp++)
868 if (*cp == '?' || *cp == '[' || *cp == '*')
869 break;
870
871 if (*cp == '\0') {
872 /*
873 * No directory component and no wildcard at all -- this
874 * should never happen as in such a simple case there is no
875 * need to expand anything.
876 */
877 DirExpandPath(word, path, expansions);
878 goto done;
879 }
880
881 /* Back up to the start of the component containing the wildcard. */
882 /* XXX: This handles '///' and '/' differently. */
883 while (cp > word && *cp != '/')
884 cp--;
885
886 if (cp == word) {
887 /* The first component contains the wildcard. */
888 /* Start the search from the local directory */
889 DirExpandPath(word, path, expansions);
890 goto done;
891 }
892
893 {
894 char *prefix = bmake_strsedup(word, cp + 1);
895 /*
896 * The wildcard isn't in the first component.
897 * Find all the components up to the one with the wildcard.
898 */
899 /*
900 * XXX: Check the "the directory is added to the path" part.
901 * It is probably surprising that the directory before a
902 * wildcard gets added to the path.
903 */
904 /*
905 * XXX: Only the first match of the prefix in the path is
906 * taken, any others are ignored. The expectation may be
907 * that the pattern is expanded in the whole path.
908 */
909 char *dirpath = Dir_FindFile(prefix, path);
910 free(prefix);
911
912 /*
913 * dirpath is null if can't find the leading component
914 * XXX: Dir_FindFile won't find internal components.
915 * i.e. if the path contains ../Etc/Object and we're
916 * looking for Etc, it won't be found. Ah well.
917 * Probably not important.
918 * XXX: Check whether the above comment is still true.
919 */
920 if (dirpath != NULL) {
921 SearchPath *partPath;
922
923 char *end = &dirpath[strlen(dirpath) - 1];
924 /* XXX: What about multiple trailing slashes? */
925 if (*end == '/')
926 *end = '\0';
927
928 partPath = SearchPath_New();
929 (void)Dir_AddDir(partPath, dirpath);
930 DirExpandPath(cp + 1, partPath, expansions);
931 Lst_Free(partPath);
932 /* XXX: Should the dirs in partPath be freed here?
933 * It's not obvious whether to free them or not. */
934 }
935 }
936
937 done:
938 if (DEBUG(DIR))
939 PrintExpansions(expansions);
940 }
941
942 /* Find if the file with the given name exists in the given path.
943 * Return the freshly allocated path to the file, or NULL. */
944 static char *
945 DirLookup(CachedDir *dir, const char *base)
946 {
947 char *file; /* the current filename to check */
948
949 DIR_DEBUG1(" %s ...\n", dir->name);
950
951 if (!HashSet_Contains(&dir->files, base))
952 return NULL;
953
954 file = str_concat3(dir->name, "/", base);
955 DIR_DEBUG1(" returning %s\n", file);
956 dir->hits++;
957 hits++;
958 return file;
959 }
960
961
962 /* Find if the file with the given name exists in the given directory.
963 * Return the freshly allocated path to the file, or NULL. */
964 static char *
965 DirLookupSubdir(CachedDir *dir, const char *name)
966 {
967 struct cached_stat cst;
968 char *file = dir == dot ? bmake_strdup(name)
969 : str_concat3(dir->name, "/", name);
970
971 DIR_DEBUG1("checking %s ...\n", file);
972
973 if (cached_stat(file, &cst) == 0) {
974 nearmisses++;
975 return file;
976 }
977 free(file);
978 return NULL;
979 }
980
981 /* Find if the file with the given name exists in the given path.
982 * Return the freshly allocated path to the file, the empty string, or NULL.
983 * Returning the empty string means that the search should be terminated.
984 */
985 static char *
986 DirLookupAbs(CachedDir *dir, const char *name, const char *cp)
987 {
988 const char *dnp; /* pointer into dir->name */
989 const char *np; /* pointer into name */
990
991 DIR_DEBUG1(" %s ...\n", dir->name);
992
993 /*
994 * If the file has a leading path component and that component
995 * exactly matches the entire name of the current search
996 * directory, we can attempt another cache lookup. And if we don't
997 * have a hit, we can safely assume the file does not exist at all.
998 */
999 for (dnp = dir->name, np = name;
1000 *dnp != '\0' && *dnp == *np; dnp++, np++)
1001 continue;
1002 if (*dnp != '\0' || np != cp - 1)
1003 return NULL;
1004
1005 if (!HashSet_Contains(&dir->files, cp)) {
1006 DIR_DEBUG0(" must be here but isn't -- returning\n");
1007 return bmake_strdup(""); /* to terminate the search */
1008 }
1009
1010 dir->hits++;
1011 hits++;
1012 DIR_DEBUG1(" returning %s\n", name);
1013 return bmake_strdup(name);
1014 }
1015
1016 /* Find the file given on "." or curdir.
1017 * Return the freshly allocated path to the file, or NULL. */
1018 static char *
1019 DirFindDot(const char *name, const char *base)
1020 {
1021
1022 if (HashSet_Contains(&dot->files, base)) {
1023 DIR_DEBUG0(" in '.'\n");
1024 hits++;
1025 dot->hits++;
1026 return bmake_strdup(name);
1027 }
1028
1029 if (cur != NULL && HashSet_Contains(&cur->files, base)) {
1030 DIR_DEBUG1(" in ${.CURDIR} = %s\n", cur->name);
1031 hits++;
1032 cur->hits++;
1033 return str_concat3(cur->name, "/", base);
1034 }
1035
1036 return NULL;
1037 }
1038
1039 /* Find the file with the given name along the given search path.
1040 *
1041 * If the file is found in a directory that is not on the path
1042 * already (either 'name' is absolute or it is a relative path
1043 * [ dir1/.../dirn/file ] which exists below one of the directories
1044 * already on the search path), its directory is added to the end
1045 * of the path, on the assumption that there will be more files in
1046 * that directory later on. Sometimes this is true. Sometimes not.
1047 *
1048 * Input:
1049 * name the file to find
1050 * path the directories to search, or NULL
1051 *
1052 * Results:
1053 * The freshly allocated path to the file, or NULL.
1054 */
1055 char *
1056 Dir_FindFile(const char *name, SearchPath *path)
1057 {
1058 char *file; /* the current filename to check */
1059 const char *base; /* Terminal name of file */
1060 Boolean hasLastDot = FALSE; /* true if we should search dot last */
1061 Boolean hasSlash; /* true if 'name' contains a / */
1062 struct cached_stat cst; /* Buffer for stat, if necessary */
1063 const char *trailing_dot = ".";
1064
1065 /*
1066 * Find the final component of the name and note whether it has a
1067 * slash in it (the name, I mean)
1068 */
1069 base = strrchr(name, '/');
1070 if (base != NULL) {
1071 hasSlash = TRUE;
1072 base++;
1073 } else {
1074 hasSlash = FALSE;
1075 base = name;
1076 }
1077
1078 DIR_DEBUG1("Searching for %s ...", name);
1079
1080 if (path == NULL) {
1081 DIR_DEBUG0("couldn't open path, file not found\n");
1082 misses++;
1083 return NULL;
1084 }
1085
1086 if (path->first != NULL) {
1087 CachedDir *dir = path->first->datum;
1088 if (dir == dotLast) {
1089 hasLastDot = TRUE;
1090 DIR_DEBUG0("[dot last]...");
1091 }
1092 }
1093 DIR_DEBUG0("\n");
1094
1095 /*
1096 * If there's no leading directory components or if the leading
1097 * directory component is exactly `./', consult the cached contents
1098 * of each of the directories on the search path.
1099 */
1100 if (!hasSlash || (base - name == 2 && *name == '.')) {
1101 SearchPathNode *ln;
1102
1103 /*
1104 * We look through all the directories on the path seeking one
1105 * which contains the final component of the given name. If
1106 * such a beast is found, we concatenate the directory name
1107 * and the final component and return the resulting string.
1108 * If we don't find any such thing, we go on to phase two.
1109 *
1110 * No matter what, we always look for the file in the current
1111 * directory before anywhere else (unless we found the magic
1112 * DOTLAST path, in which case we search it last) and we *do
1113 * not* add the ./ to it if it exists.
1114 * This is so there are no conflicts between what the user
1115 * specifies (fish.c) and what pmake finds (./fish.c).
1116 */
1117 if (!hasLastDot && (file = DirFindDot(name, base)) != NULL)
1118 return file;
1119
1120 for (ln = path->first; ln != NULL; ln = ln->next) {
1121 CachedDir *dir = ln->datum;
1122 if (dir == dotLast)
1123 continue;
1124 if ((file = DirLookup(dir, base)) != NULL)
1125 return file;
1126 }
1127
1128 if (hasLastDot && (file = DirFindDot(name, base)) != NULL)
1129 return file;
1130 }
1131
1132 /*
1133 * We didn't find the file on any directory in the search path.
1134 * If the name doesn't contain a slash, that means it doesn't exist.
1135 * If it *does* contain a slash, however, there is still hope: it
1136 * could be in a subdirectory of one of the members of the search
1137 * path. (eg. /usr/include and sys/types.h. The above search would
1138 * fail to turn up types.h in /usr/include, but it *is* in
1139 * /usr/include/sys/types.h).
1140 * [ This no longer applies: If we find such a beast, we assume there
1141 * will be more (what else can we assume?) and add all but the last
1142 * component of the resulting name onto the search path (at the
1143 * end).]
1144 * This phase is only performed if the file is *not* absolute.
1145 */
1146 if (!hasSlash) {
1147 DIR_DEBUG0(" failed.\n");
1148 misses++;
1149 return NULL;
1150 }
1151
1152 if (*base == '\0') {
1153 /* we were given a trailing "/" */
1154 base = trailing_dot;
1155 }
1156
1157 if (name[0] != '/') {
1158 SearchPathNode *ln;
1159 Boolean checkedDot = FALSE;
1160
1161 DIR_DEBUG0(" Trying subdirectories...\n");
1162
1163 if (!hasLastDot) {
1164 if (dot != NULL) {
1165 checkedDot = TRUE;
1166 if ((file = DirLookupSubdir(dot, name)) != NULL)
1167 return file;
1168 }
1169 if (cur && (file = DirLookupSubdir(cur, name)) != NULL)
1170 return file;
1171 }
1172
1173 for (ln = path->first; ln != NULL; ln = ln->next) {
1174 CachedDir *dir = ln->datum;
1175 if (dir == dotLast)
1176 continue;
1177 if (dir == dot) {
1178 if (checkedDot)
1179 continue;
1180 checkedDot = TRUE;
1181 }
1182 if ((file = DirLookupSubdir(dir, name)) != NULL)
1183 return file;
1184 }
1185
1186 if (hasLastDot) {
1187 if (dot && !checkedDot) {
1188 checkedDot = TRUE;
1189 if ((file = DirLookupSubdir(dot, name)) != NULL)
1190 return file;
1191 }
1192 if (cur && (file = DirLookupSubdir(cur, name)) != NULL)
1193 return file;
1194 }
1195
1196 if (checkedDot) {
1197 /*
1198 * Already checked by the given name, since . was in
1199 * the path, so no point in proceeding.
1200 */
1201 DIR_DEBUG0(" Checked . already, returning NULL\n");
1202 return NULL;
1203 }
1204
1205 } else { /* name[0] == '/' */
1206 SearchPathNode *ln;
1207
1208 /*
1209 * For absolute names, compare directory path prefix against
1210 * the the directory path of each member on the search path
1211 * for an exact match. If we have an exact match on any member
1212 * of the search path, use the cached contents of that member
1213 * to lookup the final file component. If that lookup fails we
1214 * can safely assume that the file does not exist at all.
1215 * This is signified by DirLookupAbs() returning an empty
1216 * string.
1217 */
1218 DIR_DEBUG0(" Trying exact path matches...\n");
1219
1220 if (!hasLastDot && cur &&
1221 ((file = DirLookupAbs(cur, name, base)) != NULL)) {
1222 if (file[0] == '\0') {
1223 free(file);
1224 return NULL;
1225 }
1226 return file;
1227 }
1228
1229 for (ln = path->first; ln != NULL; ln = ln->next) {
1230 CachedDir *dir = ln->datum;
1231 if (dir == dotLast)
1232 continue;
1233 if ((file = DirLookupAbs(dir, name, base)) != NULL) {
1234 if (file[0] == '\0') {
1235 free(file);
1236 return NULL;
1237 }
1238 return file;
1239 }
1240 }
1241
1242 if (hasLastDot && cur &&
1243 ((file = DirLookupAbs(cur, name, base)) != NULL)) {
1244 if (file[0] == '\0') {
1245 free(file);
1246 return NULL;
1247 }
1248 return file;
1249 }
1250 }
1251
1252 /*
1253 * Didn't find it that way, either. Sigh. Phase 3. Add its directory
1254 * onto the search path in any case, just in case, then look for the
1255 * thing in the hash table. If we find it, grand. We return a new
1256 * copy of the name. Otherwise we sadly return a NULL pointer. Sigh.
1257 * Note that if the directory holding the file doesn't exist, this
1258 * will do an extra search of the final directory on the path. Unless
1259 * something weird happens, this search won't succeed and life will
1260 * be groovy.
1261 *
1262 * Sigh. We cannot add the directory onto the search path because
1263 * of this amusing case:
1264 * $(INSTALLDIR)/$(FILE): $(FILE)
1265 *
1266 * $(FILE) exists in $(INSTALLDIR) but not in the current one.
1267 * When searching for $(FILE), we will find it in $(INSTALLDIR)
1268 * b/c we added it here. This is not good...
1269 */
1270 #if 0
1271 {
1272 CachedDir *dir;
1273 char *prefix;
1274
1275 if (base == trailing_dot) {
1276 base = strrchr(name, '/');
1277 base++;
1278 }
1279 prefix = bmake_strsedup(name, base - 1);
1280 (void)Dir_AddDir(path, prefix);
1281 free(prefix);
1282
1283 bigmisses++;
1284 if (path->last == NULL)
1285 return NULL;
1286
1287 dir = path->last->datum;
1288 if (HashSet_Contains(&dir->files, base))
1289 return bmake_strdup(name);
1290 return NULL;
1291 }
1292 #else
1293 DIR_DEBUG1(" Looking for \"%s\" ...\n", name);
1294
1295 bigmisses++;
1296 if (cached_stat(name, &cst) == 0) {
1297 return bmake_strdup(name);
1298 }
1299
1300 DIR_DEBUG0(" failed. Returning NULL\n");
1301 return NULL;
1302 #endif
1303 }
1304
1305
1306 /* Search for a path starting at a given directory and then working our way
1307 * up towards the root.
1308 *
1309 * Input:
1310 * here starting directory
1311 * search_path the relative path we are looking for
1312 *
1313 * Results:
1314 * The found path, or NULL.
1315 */
1316 char *
1317 Dir_FindHereOrAbove(const char *here, const char *search_path)
1318 {
1319 struct cached_stat cst;
1320 char *dirbase, *dirbase_end;
1321 char *try, *try_end;
1322
1323 /* copy out our starting point */
1324 dirbase = bmake_strdup(here);
1325 dirbase_end = dirbase + strlen(dirbase);
1326
1327 /* loop until we determine a result */
1328 for (;;) {
1329
1330 /* try and stat(2) it ... */
1331 try = str_concat3(dirbase, "/", search_path);
1332 if (cached_stat(try, &cst) != -1) {
1333 /*
1334 * success! if we found a file, chop off
1335 * the filename so we return a directory.
1336 */
1337 if ((cst.cst_mode & S_IFMT) != S_IFDIR) {
1338 try_end = try + strlen(try);
1339 while (try_end > try && *try_end != '/')
1340 try_end--;
1341 if (try_end > try)
1342 *try_end = '\0'; /* chop! */
1343 }
1344
1345 free(dirbase);
1346 return try;
1347 }
1348 free(try);
1349
1350 /*
1351 * nope, we didn't find it. if we used up dirbase we've
1352 * reached the root and failed.
1353 */
1354 if (dirbase_end == dirbase)
1355 break; /* failed! */
1356
1357 /*
1358 * truncate dirbase from the end to move up a dir
1359 */
1360 while (dirbase_end > dirbase && *dirbase_end != '/')
1361 dirbase_end--;
1362 *dirbase_end = '\0'; /* chop! */
1363 }
1364
1365 free(dirbase);
1366 return NULL;
1367 }
1368
1369 /*
1370 * This is an implied source, and it may have moved,
1371 * see if we can find it via the current .PATH
1372 */
1373 static char *
1374 ResolveMovedDepends(GNode *gn)
1375 {
1376 char *fullName;
1377
1378 char *base = strrchr(gn->name, '/');
1379 if (base == NULL)
1380 return NULL;
1381 base++;
1382
1383 fullName = Dir_FindFile(base, Suff_FindPath(gn));
1384 if (fullName == NULL)
1385 return NULL;
1386
1387 /*
1388 * Put the found file in gn->path so that we give that to the compiler.
1389 */
1390 /*
1391 * XXX: Better just reset gn->path to NULL; updating it is already done
1392 * by Dir_UpdateMTime.
1393 */
1394 gn->path = bmake_strdup(fullName);
1395 if (!Job_RunTarget(".STALE", gn->fname))
1396 fprintf(stdout, /* XXX: Why stdout? */
1397 "%s: %s, %d: ignoring stale %s for %s, found %s\n",
1398 progname, gn->fname, gn->lineno,
1399 makeDependfile, gn->name, fullName);
1400
1401 return fullName;
1402 }
1403
1404 static char *
1405 ResolveFullName(GNode *gn)
1406 {
1407 char *fullName;
1408
1409 fullName = gn->path;
1410 if (fullName == NULL && !(gn->type & OP_NOPATH)) {
1411
1412 fullName = Dir_FindFile(gn->name, Suff_FindPath(gn));
1413
1414 if (fullName == NULL && gn->flags & FROM_DEPEND &&
1415 !Lst_IsEmpty(&gn->implicitParents))
1416 fullName = ResolveMovedDepends(gn);
1417
1418 DIR_DEBUG2("Found '%s' as '%s'\n",
1419 gn->name, fullName ? fullName : "(not found)");
1420 }
1421
1422 if (fullName == NULL)
1423 fullName = bmake_strdup(gn->name);
1424
1425 /* XXX: Is every piece of memory freed as it should? */
1426
1427 return fullName;
1428 }
1429
1430 /* Search gn along dirSearchPath and store its modification time in gn->mtime.
1431 * If no file is found, store 0 instead.
1432 *
1433 * The found file is stored in gn->path, unless the node already had a path. */
1434 void
1435 Dir_UpdateMTime(GNode *gn, Boolean recheck)
1436 {
1437 char *fullName;
1438 struct cached_stat cst;
1439
1440 if (gn->type & OP_ARCHV) {
1441 Arch_UpdateMTime(gn);
1442 return;
1443 }
1444
1445 if (gn->type & OP_PHONY) {
1446 gn->mtime = 0;
1447 return;
1448 }
1449
1450 fullName = ResolveFullName(gn);
1451
1452 if (cached_stats(fullName, &cst, recheck ? CST_UPDATE : CST_NONE) < 0) {
1453 if (gn->type & OP_MEMBER) {
1454 if (fullName != gn->path)
1455 free(fullName);
1456 Arch_UpdateMemberMTime(gn);
1457 return;
1458 }
1459
1460 cst.cst_mtime = 0;
1461 }
1462
1463 if (fullName != NULL && gn->path == NULL)
1464 gn->path = fullName;
1465 /* XXX: else free(fullName)? */
1466
1467 gn->mtime = cst.cst_mtime;
1468 }
1469
1470 /*
1471 * Read the directory and add it to the cache in openDirs, even if it is
1472 * already there. If a path is given, add the directory to that path as
1473 * well.
1474 *
1475 * XXX: Why is it added to openDirs unconditionally?
1476 */
1477 static CachedDir *
1478 CacheNewDir(const char *name, SearchPath *path)
1479 {
1480 CachedDir *dir = NULL;
1481 DIR *d;
1482 struct dirent *dp;
1483
1484 if ((d = opendir(name)) == NULL) {
1485 DIR_DEBUG1("Caching %s ... not found\n", name);
1486 return dir;
1487 }
1488
1489 DIR_DEBUG1("Caching %s ...\n", name);
1490
1491 dir = CachedDir_New(name);
1492 CachedDir_Ref(dir); /* XXX: why here already? */
1493
1494 while ((dp = readdir(d)) != NULL) {
1495
1496 #if defined(sun) && defined(d_ino) /* d_ino is a sunos4 #define for d_fileno */
1497 /*
1498 * The sun directory library doesn't check for a 0 inode
1499 * (0-inode slots just take up space), so we have to do
1500 * it ourselves.
1501 */
1502 if (dp->d_fileno == 0)
1503 continue;
1504 #endif /* sun && d_ino */
1505
1506 (void)HashSet_Add(&dir->files, dp->d_name);
1507 }
1508 (void)closedir(d);
1509
1510 OpenDirs_Add(&openDirs, dir);
1511 if (path != NULL)
1512 Lst_Append(path, dir);
1513
1514 DIR_DEBUG1("Caching %s done\n", name);
1515 return dir;
1516 }
1517
1518 /* Read the list of filenames in the directory and store the result
1519 * in openDirs.
1520 *
1521 * If a path is given, append the directory to that path.
1522 *
1523 * Input:
1524 * path The path to which the directory should be
1525 * added, or NULL to only add the directory to openDirs
1526 * name The name of the directory to add.
1527 * The name is not normalized in any way.
1528 */
1529 CachedDir *
1530 Dir_AddDir(SearchPath *path, const char *name)
1531 {
1532
1533 if (path != NULL && strcmp(name, ".DOTLAST") == 0) {
1534 SearchPathNode *ln;
1535
1536 /* XXX: Linear search gets slow with thousands of entries. */
1537 for (ln = path->first; ln != NULL; ln = ln->next) {
1538 CachedDir *pathDir = ln->datum;
1539 if (strcmp(pathDir->name, name) == 0)
1540 return pathDir;
1541 }
1542
1543 CachedDir_Ref(dotLast);
1544 Lst_Prepend(path, dotLast);
1545 }
1546
1547 if (path != NULL) {
1548 CachedDir *dir = OpenDirs_Find(&openDirs, name);
1549 if (dir != NULL) {
1550 if (Lst_FindDatum(path, dir) == NULL)
1551 Lst_Append(path, CachedDir_Ref(dir));
1552 return dir;
1553 }
1554 }
1555
1556 return CacheNewDir(name, path);
1557 }
1558
1559 /* Return a copy of dirSearchPath, incrementing the reference counts for
1560 * the contained directories. */
1561 SearchPath *
1562 Dir_CopyDirSearchPath(void)
1563 {
1564 SearchPath *path = SearchPath_New();
1565 SearchPathNode *ln;
1566 for (ln = dirSearchPath.first; ln != NULL; ln = ln->next) {
1567 CachedDir *dir = ln->datum;
1568 Lst_Append(path, CachedDir_Ref(dir));
1569 }
1570 return path;
1571 }
1572
1573 /*-
1574 *-----------------------------------------------------------------------
1575 * SearchPath_ToFlags --
1576 * Make a string by taking all the directories in the given search
1577 * path and preceding them by the given flag. Used by the suffix
1578 * module to create variables for compilers based on suffix search
1579 * paths.
1580 *
1581 * Input:
1582 * flag flag which should precede each directory
1583 * path list of directories
1584 *
1585 * Results:
1586 * The string mentioned above. Note that there is no space between
1587 * the given flag and each directory. The empty string is returned if
1588 * Things don't go well.
1589 *
1590 * Side Effects:
1591 * None
1592 *-----------------------------------------------------------------------
1593 */
1594 char *
1595 SearchPath_ToFlags(const char *flag, SearchPath *path)
1596 {
1597 Buffer buf;
1598 SearchPathNode *ln;
1599
1600 Buf_Init(&buf);
1601
1602 if (path != NULL) {
1603 for (ln = path->first; ln != NULL; ln = ln->next) {
1604 CachedDir *dir = ln->datum;
1605 Buf_AddStr(&buf, " ");
1606 Buf_AddStr(&buf, flag);
1607 Buf_AddStr(&buf, dir->name);
1608 }
1609 }
1610
1611 return Buf_Destroy(&buf, FALSE);
1612 }
1613
1614 /* Nuke a directory descriptor, if it is no longer used. */
1615 static void
1616 CachedDir_Destroy(CachedDir *dir)
1617 {
1618 CachedDir_Unref(dir);
1619
1620 if (dir->refCount == 0) {
1621 OpenDirs_Remove(&openDirs, dir->name);
1622
1623 HashSet_Done(&dir->files);
1624 free(dir->name);
1625 free(dir);
1626 }
1627 }
1628
1629 /* Free the search path and all directories mentioned in it. */
1630 void
1631 SearchPath_Free(SearchPath *path)
1632 {
1633 SearchPathNode *ln;
1634
1635 for (ln = path->first; ln != NULL; ln = ln->next) {
1636 CachedDir *dir = ln->datum;
1637 CachedDir_Destroy(dir);
1638 }
1639 Lst_Free(path);
1640 }
1641
1642 /* Clear out all elements from the given search path.
1643 * The path is set to the empty list but is not destroyed. */
1644 void
1645 SearchPath_Clear(SearchPath *path)
1646 {
1647 while (!Lst_IsEmpty(path)) {
1648 CachedDir *dir = Lst_Dequeue(path);
1649 CachedDir_Destroy(dir);
1650 }
1651 }
1652
1653
1654 /* Concatenate two paths, adding the second to the end of the first,
1655 * skipping duplicates. */
1656 void
1657 SearchPath_AddAll(SearchPath *dst, SearchPath *src)
1658 {
1659 SearchPathNode *ln;
1660
1661 for (ln = src->first; ln != NULL; ln = ln->next) {
1662 CachedDir *dir = ln->datum;
1663 if (Lst_FindDatum(dst, dir) == NULL)
1664 Lst_Append(dst, CachedDir_Ref(dir));
1665 }
1666 }
1667
1668 static int
1669 percentage(int num, int den)
1670 {
1671 return den != 0 ? num * 100 / den : 0;
1672 }
1673
1674 /********** DEBUG INFO **********/
1675 void
1676 Dir_PrintDirectories(void)
1677 {
1678 CachedDirListNode *ln;
1679
1680 debug_printf("#*** Directory Cache:\n");
1681 debug_printf(
1682 "# Stats: %d hits %d misses %d near misses %d losers (%d%%)\n",
1683 hits, misses, nearmisses, bigmisses,
1684 percentage(hits, hits + bigmisses + nearmisses));
1685 debug_printf("# %-20s referenced\thits\n", "directory");
1686
1687 for (ln = openDirs.list.first; ln != NULL; ln = ln->next) {
1688 CachedDir *dir = ln->datum;
1689 debug_printf("# %-20s %10d\t%4d\n",
1690 dir->name, dir->refCount, dir->hits);
1691 }
1692 }
1693
1694 void
1695 SearchPath_Print(SearchPath *path)
1696 {
1697 SearchPathNode *node;
1698 for (node = path->first; node != NULL; node = node->next) {
1699 const CachedDir *dir = node->datum;
1700 debug_printf("%s ", dir->name);
1701 }
1702 }
1703