xref: /netbsd-src/usr.bin/make/dir.c (revision 154bfe8e089c1a0a4e9ed8414f08d3da90949162)
1 /*	$NetBSD: dir.c,v 1.137 2020/09/07 19:48:08 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 #ifndef MAKE_NATIVE
73 static char rcsid[] = "$NetBSD: dir.c,v 1.137 2020/09/07 19:48:08 rillig Exp $";
74 #else
75 #include <sys/cdefs.h>
76 #ifndef lint
77 #if 0
78 static char sccsid[] = "@(#)dir.c	8.2 (Berkeley) 1/2/94";
79 #else
80 __RCSID("$NetBSD: dir.c,v 1.137 2020/09/07 19:48:08 rillig Exp $");
81 #endif
82 #endif /* not lint */
83 #endif
84 
85 /*-
86  * dir.c --
87  *	Directory searching using wildcards and/or normal names...
88  *	Used both for source wildcarding in the Makefile and for finding
89  *	implicit sources.
90  *
91  * The interface for this module is:
92  *	Dir_Init  	    Initialize the module.
93  *
94  *	Dir_InitCur	    Set the cur Path.
95  *
96  *	Dir_InitDot	    Set the dot Path.
97  *
98  *	Dir_End  	    Cleanup the module.
99  *
100  *	Dir_SetPATH	    Set ${.PATH} to reflect state of dirSearchPath.
101  *
102  *	Dir_HasWildcards    Returns TRUE if the name given it needs to
103  *	    	  	    be wildcard-expanded.
104  *
105  *	Dir_Expand	    Given a pattern and a path, return a Lst of names
106  *	    	  	    which match the pattern on the search path.
107  *
108  *	Dir_FindFile	    Searches for a file on a given search path.
109  *	    	  	    If it exists, the entire path is returned.
110  *	    	  	    Otherwise NULL is returned.
111  *
112  *	Dir_FindHereOrAbove Search for a path in the current directory and
113  *			    then all the directories above it in turn until
114  *			    the path is found or we reach the root ("/").
115  *
116  *	Dir_MTime 	    Return the modification time of a node. The file
117  *	    	  	    is searched for along the default search path.
118  *	    	  	    The path and mtime fields of the node are filled
119  *	    	  	    in.
120  *
121  *	Dir_AddDir	    Add a directory to a search path.
122  *
123  *	Dir_MakeFlags	    Given a search path and a command flag, create
124  *	    	  	    a string with each of the directories in the path
125  *	    	  	    preceded by the command flag and all of them
126  *	    	  	    separated by a space.
127  *
128  *	Dir_Destroy	    Destroy an element of a search path. Frees up all
129  *	    	  	    things that can be freed for the element as long
130  *	    	  	    as the element is no longer referenced by any other
131  *	    	  	    search path.
132  *	Dir_ClearPath	    Resets a search path to the empty list.
133  *
134  * For debugging:
135  *	Dir_PrintDirectories	Print stats about the directory cache.
136  */
137 
138 #include <sys/types.h>
139 #include <sys/stat.h>
140 
141 #include <dirent.h>
142 #include <errno.h>
143 #include <stdio.h>
144 
145 #include "make.h"
146 #include "dir.h"
147 #include "job.h"
148 
149 
150 #define DIR_DEBUG0(fmt) \
151     if (!DEBUG(DIR)) (void) 0; else fprintf(debug_file, fmt)
152 
153 #define DIR_DEBUG1(fmt, arg1) \
154     if (!DEBUG(DIR)) (void) 0; else fprintf(debug_file, fmt, arg1)
155 
156 #define DIR_DEBUG2(fmt, arg1, arg2) \
157     if (!DEBUG(DIR)) (void) 0; else fprintf(debug_file, fmt, arg1, arg2)
158 
159 
160 /*
161  *	A search path consists of a Lst of Path structures. A Path structure
162  *	has in it the name of the directory and a hash table of all the files
163  *	in the directory. This is used to cut down on the number of system
164  *	calls necessary to find implicit dependents and their like. Since
165  *	these searches are made before any actions are taken, we need not
166  *	worry about the directory changing due to creation commands. If this
167  *	hampers the style of some makefiles, they must be changed.
168  *
169  *	A list of all previously-read directories is kept in the
170  *	openDirectories Lst. This list is checked first before a directory
171  *	is opened.
172  *
173  *	The need for the caching of whole directories is brought about by
174  *	the multi-level transformation code in suff.c, which tends to search
175  *	for far more files than regular make does. In the initial
176  *	implementation, the amount of time spent performing "stat" calls was
177  *	truly astronomical. The problem with hashing at the start is,
178  *	of course, that pmake doesn't then detect changes to these directories
179  *	during the course of the make. Three possibilities suggest themselves:
180  *
181  *	    1) just use stat to test for a file's existence. As mentioned
182  *	       above, this is very inefficient due to the number of checks
183  *	       engendered by the multi-level transformation code.
184  *	    2) use readdir() and company to search the directories, keeping
185  *	       them open between checks. I have tried this and while it
186  *	       didn't slow down the process too much, it could severely
187  *	       affect the amount of parallelism available as each directory
188  *	       open would take another file descriptor out of play for
189  *	       handling I/O for another job. Given that it is only recently
190  *	       that UNIX OS's have taken to allowing more than 20 or 32
191  *	       file descriptors for a process, this doesn't seem acceptable
192  *	       to me.
193  *	    3) record the mtime of the directory in the Path structure and
194  *	       verify the directory hasn't changed since the contents were
195  *	       hashed. This will catch the creation or deletion of files,
196  *	       but not the updating of files. However, since it is the
197  *	       creation and deletion that is the problem, this could be
198  *	       a good thing to do. Unfortunately, if the directory (say ".")
199  *	       were fairly large and changed fairly frequently, the constant
200  *	       rehashing could seriously degrade performance. It might be
201  *	       good in such cases to keep track of the number of rehashes
202  *	       and if the number goes over a (small) limit, resort to using
203  *	       stat in its place.
204  *
205  *	An additional thing to consider is that pmake is used primarily
206  *	to create C programs and until recently pcc-based compilers refused
207  *	to allow you to specify where the resulting object file should be
208  *	placed. This forced all objects to be created in the current
209  *	directory. This isn't meant as a full excuse, just an explanation of
210  *	some of the reasons for the caching used here.
211  *
212  *	One more note: the location of a target's file is only performed
213  *	on the downward traversal of the graph and then only for terminal
214  *	nodes in the graph. This could be construed as wrong in some cases,
215  *	but prevents inadvertent modification of files when the "installed"
216  *	directory for a file is provided in the search path.
217  *
218  *	Another data structure maintained by this module is an mtime
219  *	cache used when the searching of cached directories fails to find
220  *	a file. In the past, Dir_FindFile would simply perform an access()
221  *	call in such a case to determine if the file could be found using
222  *	just the name given. When this hit, however, all that was gained
223  *	was the knowledge that the file existed. Given that an access() is
224  *	essentially a stat() without the copyout() call, and that the same
225  *	filesystem overhead would have to be incurred in Dir_MTime, it made
226  *	sense to replace the access() with a stat() and record the mtime
227  *	in a cache for when Dir_MTime was actually called.
228  */
229 
230 Lst dirSearchPath;		/* main search path */
231 
232 static Lst openDirectories;	/* the list of all open directories */
233 
234 /*
235  * Variables for gathering statistics on the efficiency of the hashing
236  * mechanism.
237  */
238 static int hits;		/* Found in directory cache */
239 static int misses;		/* Sad, but not evil misses */
240 static int nearmisses;		/* Found under search path */
241 static int bigmisses;		/* Sought by itself */
242 
243 static Path *dot;		/* contents of current directory */
244 static Path *cur;		/* contents of current directory, if not dot */
245 static Path *dotLast;		/* a fake path entry indicating we need to
246 				 * look for . last */
247 
248 /* Results of doing a last-resort stat in Dir_FindFile -- if we have to go to
249  * the system to find the file, we might as well have its mtime on record.
250  *
251  * XXX: If this is done way early, there's a chance other rules will have
252  * already updated the file, in which case we'll update it again. Generally,
253  * there won't be two rules to update a single file, so this should be ok,
254  * but... */
255 static Hash_Table mtimes;
256 
257 static Hash_Table lmtimes;	/* same as mtimes but for lstat */
258 
259 static void DirExpandInt(const char *, Lst, Lst);
260 static int DirPrintWord(void *, void *);
261 static int DirPrintDir(void *, void *);
262 static char *DirLookup(Path *, const char *, const char *, Boolean);
263 static char *DirLookupSubdir(Path *, const char *);
264 static char *DirFindDot(Boolean, const char *, const char *);
265 static char *DirLookupAbs(Path *, const char *, const char *);
266 
267 
268 /*
269  * We use stat(2) a lot, cache the results.
270  * mtime and mode are all we care about.
271  */
272 struct cache_st {
273     time_t lmtime;		/* lstat */
274     time_t mtime;		/* stat */
275     mode_t mode;
276 };
277 
278 /* minimize changes below */
279 typedef enum {
280     CST_LSTAT = 0x01,		/* call lstat(2) instead of stat(2) */
281     CST_UPDATE = 0x02		/* ignore existing cached entry */
282 } CachedStatsFlags;
283 
284 /* Returns 0 and the result of stat(2) or lstat(2) in *mst, or -1 on error. */
285 static int
286 cached_stats(Hash_Table *htp, const char *pathname, struct make_stat *mst,
287 	     CachedStatsFlags flags)
288 {
289     Hash_Entry *entry;
290     struct stat sys_st;
291     struct cache_st *cst;
292     int rc;
293 
294     if (!pathname || !pathname[0])
295 	return -1;
296 
297     entry = Hash_FindEntry(htp, pathname);
298 
299     if (entry && !(flags & CST_UPDATE)) {
300 	cst = Hash_GetValue(entry);
301 
302 	mst->mst_mode = cst->mode;
303 	mst->mst_mtime = (flags & CST_LSTAT) ? cst->lmtime : cst->mtime;
304 	if (mst->mst_mtime) {
305 	    DIR_DEBUG2("Using cached time %s for %s\n",
306 		       Targ_FmtTime(mst->mst_mtime), pathname);
307 	    return 0;
308 	}
309     }
310 
311     rc = (flags & CST_LSTAT)
312 	 ? lstat(pathname, &sys_st)
313 	 : stat(pathname, &sys_st);
314     if (rc == -1)
315 	return -1;
316 
317     if (sys_st.st_mtime == 0)
318 	sys_st.st_mtime = 1;	/* avoid confusion with missing file */
319 
320     mst->mst_mode = sys_st.st_mode;
321     mst->mst_mtime = sys_st.st_mtime;
322 
323     if (entry == NULL)
324 	entry = Hash_CreateEntry(htp, pathname, NULL);
325     if (Hash_GetValue(entry) == NULL) {
326 	Hash_SetValue(entry, bmake_malloc(sizeof(*cst)));
327 	memset(Hash_GetValue(entry), 0, sizeof(*cst));
328     }
329     cst = Hash_GetValue(entry);
330     if (flags & CST_LSTAT) {
331 	cst->lmtime = sys_st.st_mtime;
332     } else {
333 	cst->mtime = sys_st.st_mtime;
334     }
335     cst->mode = sys_st.st_mode;
336     DIR_DEBUG2("   Caching %s for %s\n",
337 	       Targ_FmtTime(sys_st.st_mtime), pathname);
338 
339     return 0;
340 }
341 
342 int
343 cached_stat(const char *pathname, struct make_stat *st)
344 {
345     return cached_stats(&mtimes, pathname, st, 0);
346 }
347 
348 int
349 cached_lstat(const char *pathname, struct make_stat *st)
350 {
351     return cached_stats(&lmtimes, pathname, st, CST_LSTAT);
352 }
353 
354 /* Initialize things for this module. */
355 void
356 Dir_Init(void)
357 {
358     dirSearchPath = Lst_Init();
359     openDirectories = Lst_Init();
360     Hash_InitTable(&mtimes);
361     Hash_InitTable(&lmtimes);
362 }
363 
364 void
365 Dir_InitDir(const char *cdname)
366 {
367     Dir_InitCur(cdname);
368 
369     dotLast = bmake_malloc(sizeof(Path));
370     dotLast->refCount = 1;
371     dotLast->hits = 0;
372     dotLast->name = bmake_strdup(".DOTLAST");
373     Hash_InitTable(&dotLast->files);
374 }
375 
376 /*
377  * Called by Dir_InitDir and whenever .CURDIR is assigned to.
378  */
379 void
380 Dir_InitCur(const char *cdname)
381 {
382     Path *p;
383 
384     if (cdname != NULL) {
385 	/*
386 	 * Our build directory is not the same as our source directory.
387 	 * Keep this one around too.
388 	 */
389 	if ((p = Dir_AddDir(NULL, cdname))) {
390 	    p->refCount += 1;
391 	    if (cur && cur != p) {
392 		/*
393 		 * We've been here before, cleanup.
394 		 */
395 		cur->refCount -= 1;
396 		Dir_Destroy(cur);
397 	    }
398 	    cur = p;
399 	}
400     }
401 }
402 
403 /* (Re)initialize "dot" (current/object directory) path hash.
404  * Some directories may be opened. */
405 void
406 Dir_InitDot(void)
407 {
408     if (dot != NULL) {
409 	LstNode ln;
410 
411 	/* Remove old entry from openDirectories, but do not destroy. */
412 	ln = Lst_FindDatum(openDirectories, dot);
413 	Lst_Remove(openDirectories, ln);
414     }
415 
416     dot = Dir_AddDir(NULL, ".");
417 
418     if (dot == NULL) {
419 	Error("Cannot open `.' (%s)", strerror(errno));
420 	exit(1);
421     }
422 
423     /*
424      * We always need to have dot around, so we increment its reference count
425      * to make sure it's not destroyed.
426      */
427     dot->refCount += 1;
428     Dir_SetPATH();		/* initialize */
429 }
430 
431 /* Clean up things for this module. */
432 void
433 Dir_End(void)
434 {
435 #ifdef CLEANUP
436     if (cur) {
437 	cur->refCount -= 1;
438 	Dir_Destroy(cur);
439     }
440     dot->refCount -= 1;
441     dotLast->refCount -= 1;
442     Dir_Destroy(dotLast);
443     Dir_Destroy(dot);
444     Dir_ClearPath(dirSearchPath);
445     Lst_Free(dirSearchPath);
446     Dir_ClearPath(openDirectories);
447     Lst_Free(openDirectories);
448     Hash_DeleteTable(&mtimes);
449 #endif
450 }
451 
452 /*
453  * We want ${.PATH} to indicate the order in which we will actually
454  * search, so we rebuild it after any .PATH: target.
455  * This is the simplest way to deal with the effect of .DOTLAST.
456  */
457 void
458 Dir_SetPATH(void)
459 {
460     LstNode ln;			/* a list element */
461     Path *p;
462     Boolean hasLastDot = FALSE;	/* true if we should search dot last */
463 
464     Var_Delete(".PATH", VAR_GLOBAL);
465 
466     Lst_Open(dirSearchPath);
467     if ((ln = Lst_First(dirSearchPath)) != NULL) {
468 	p = LstNode_Datum(ln);
469 	if (p == dotLast) {
470 	    hasLastDot = TRUE;
471 	    Var_Append(".PATH", dotLast->name, VAR_GLOBAL);
472 	}
473     }
474 
475     if (!hasLastDot) {
476 	if (dot)
477 	    Var_Append(".PATH", dot->name, VAR_GLOBAL);
478 	if (cur)
479 	    Var_Append(".PATH", cur->name, VAR_GLOBAL);
480     }
481 
482     while ((ln = Lst_Next(dirSearchPath)) != NULL) {
483 	p = LstNode_Datum(ln);
484 	if (p == dotLast)
485 	    continue;
486 	if (p == dot && hasLastDot)
487 	    continue;
488 	Var_Append(".PATH", p->name, VAR_GLOBAL);
489     }
490 
491     if (hasLastDot) {
492 	if (dot)
493 	    Var_Append(".PATH", dot->name, VAR_GLOBAL);
494 	if (cur)
495 	    Var_Append(".PATH", cur->name, VAR_GLOBAL);
496     }
497     Lst_Close(dirSearchPath);
498 }
499 
500 /* See if the Path structure describes the same directory as the
501  * given one by comparing their names. Called from Dir_AddDir via
502  * Lst_Find when searching the list of open directories. */
503 static Boolean
504 DirFindName(const void *p, const void *desiredName)
505 {
506     return strcmp(((const Path *)p)->name, desiredName) == 0;
507 }
508 
509 /* See if the given name has any wildcard characters in it. Be careful not to
510  * expand unmatching brackets or braces.
511  *
512  * XXX: This code is not 100% correct ([^]] fails etc.). I really don't think
513  * that make(1) should be expanding patterns, because then you have to set a
514  * mechanism for escaping the expansion!
515  *
516  * Input:
517  *	name		name to check
518  *
519  * Results:
520  *	returns TRUE if the word should be expanded, FALSE otherwise
521  */
522 Boolean
523 Dir_HasWildcards(const char *name)
524 {
525     const char *cp;
526     Boolean wild = FALSE;
527     int braces = 0, brackets = 0;
528 
529     for (cp = name; *cp; cp++) {
530 	switch (*cp) {
531 	case '{':
532 	    braces++;
533 	    wild = TRUE;
534 	    break;
535 	case '}':
536 	    braces--;
537 	    break;
538 	case '[':
539 	    brackets++;
540 	    wild = TRUE;
541 	    break;
542 	case ']':
543 	    brackets--;
544 	    break;
545 	case '?':
546 	case '*':
547 	    wild = TRUE;
548 	    break;
549 	default:
550 	    break;
551 	}
552     }
553     return wild && brackets == 0 && braces == 0;
554 }
555 
556 /*-
557  *-----------------------------------------------------------------------
558  * DirMatchFiles --
559  * 	Given a pattern and a Path structure, see if any files
560  *	match the pattern and add their names to the 'expansions' list if
561  *	any do. This is incomplete -- it doesn't take care of patterns like
562  *	src / *src / *.c properly (just *.c on any of the directories), but it
563  *	will do for now.
564  *
565  * Input:
566  *	pattern		Pattern to look for
567  *	p		Directory to search
568  *	expansion	Place to store the results
569  *
570  * Side Effects:
571  *	File names are added to the expansions lst. The directory will be
572  *	fully hashed when this is done.
573  *-----------------------------------------------------------------------
574  */
575 static void
576 DirMatchFiles(const char *pattern, Path *p, Lst expansions)
577 {
578     Hash_Search search;		/* Index into the directory's table */
579     Hash_Entry *entry;		/* Current entry in the table */
580     Boolean isDot;		/* TRUE if the directory being searched is . */
581 
582     isDot = (*p->name == '.' && p->name[1] == '\0');
583 
584     for (entry = Hash_EnumFirst(&p->files, &search);
585 	 entry != NULL;
586 	 entry = Hash_EnumNext(&search))
587     {
588 	/*
589 	 * See if the file matches the given pattern. Note we follow the UNIX
590 	 * convention that dot files will only be found if the pattern
591 	 * begins with a dot (note also that as a side effect of the hashing
592 	 * scheme, .* won't match . or .. since they aren't hashed).
593 	 */
594 	if (Str_Match(entry->name, pattern) &&
595 	    ((entry->name[0] != '.') ||
596 	     (pattern[0] == '.')))
597 	{
598 	    Lst_Append(expansions,
599 			(isDot ? bmake_strdup(entry->name) :
600 			 str_concat3(p->name, "/", entry->name)));
601 	}
602     }
603 }
604 
605 /* Find the next closing brace in the string, taking nested braces into
606  * account. */
607 static const char *
608 closing_brace(const char *p)
609 {
610     int nest = 0;
611     while (*p != '\0') {
612 	if (*p == '}' && nest == 0)
613 	    break;
614 	if (*p == '{')
615 	    nest++;
616 	if (*p == '}')
617 	    nest--;
618 	p++;
619     }
620     return p;
621 }
622 
623 /* Find the next closing brace or comma in the string, taking nested braces
624  * into account. */
625 static const char *
626 separator_comma(const char *p)
627 {
628     int nest = 0;
629     while (*p != '\0') {
630 	if ((*p == '}' || *p == ',') && nest == 0)
631 	    break;
632 	if (*p == '{')
633 	    nest++;
634 	if (*p == '}')
635 	    nest--;
636 	p++;
637     }
638     return p;
639 }
640 
641 static Boolean
642 contains_wildcard(const char *p)
643 {
644     for (; *p != '\0'; p++) {
645 	switch (*p) {
646 	case '*':
647 	case '?':
648 	case '{':
649 	case '[':
650 	    return TRUE;
651 	}
652     }
653     return FALSE;
654 }
655 
656 static char *
657 concat3(const char *a, size_t a_len, const char *b, size_t b_len,
658 	const char *c, size_t c_len)
659 {
660     size_t s_len = a_len + b_len + c_len;
661     char *s = bmake_malloc(s_len + 1);
662     memcpy(s, a, a_len);
663     memcpy(s + a_len, b, b_len);
664     memcpy(s + a_len + b_len, c, c_len);
665     s[s_len] = '\0';
666     return s;
667 }
668 
669 /*-
670  *-----------------------------------------------------------------------
671  * DirExpandCurly --
672  *	Expand curly braces like the C shell. Does this recursively.
673  *	Note the special case: if after the piece of the curly brace is
674  *	done there are no wildcard characters in the result, the result is
675  *	placed on the list WITHOUT CHECKING FOR ITS EXISTENCE.
676  *
677  * Input:
678  *	word		Entire word to expand
679  *	brace		First curly brace in it
680  *	path		Search path to use
681  *	expansions	Place to store the expansions
682  *
683  * Results:
684  *	None.
685  *
686  * Side Effects:
687  *	The given list is filled with the expansions...
688  *
689  *-----------------------------------------------------------------------
690  */
691 static void
692 DirExpandCurly(const char *word, const char *brace, Lst path, Lst expansions)
693 {
694     const char *prefix, *middle, *piece, *middle_end, *suffix;
695     size_t prefix_len, suffix_len;
696 
697     /* Split the word into prefix '{' middle '}' suffix. */
698 
699     middle = brace + 1;
700     middle_end = closing_brace(middle);
701     if (*middle_end == '\0') {
702 	Error("Unterminated {} clause \"%s\"", middle);
703 	return;
704     }
705 
706     prefix = word;
707     prefix_len = (size_t)(brace - prefix);
708     suffix = middle_end + 1;
709     suffix_len = strlen(suffix);
710 
711     /* Split the middle into pieces, separated by commas. */
712 
713     piece = middle;
714     while (piece < middle_end + 1) {
715 	const char *piece_end = separator_comma(piece);
716 	size_t piece_len = (size_t)(piece_end - piece);
717 
718 	char *file = concat3(prefix, prefix_len, piece, piece_len,
719 			     suffix, suffix_len);
720 
721 	if (contains_wildcard(file)) {
722 	    Dir_Expand(file, path, expansions);
723 	    free(file);
724 	} else {
725 	    Lst_Append(expansions, file);
726 	}
727 
728 	piece = piece_end + 1;	/* skip over the comma or closing brace */
729     }
730 }
731 
732 
733 /*-
734  *-----------------------------------------------------------------------
735  * DirExpandInt --
736  *	Internal expand routine. Passes through the directories in the
737  *	path one by one, calling DirMatchFiles for each. NOTE: This still
738  *	doesn't handle patterns in directories...
739  *
740  * Input:
741  *	word		Word to expand
742  *	path		Path on which to look
743  *	expansions	Place to store the result
744  *
745  * Results:
746  *	None.
747  *
748  * Side Effects:
749  *	Things are added to the expansions list.
750  *
751  *-----------------------------------------------------------------------
752  */
753 static void
754 DirExpandInt(const char *word, Lst path, Lst expansions)
755 {
756     LstNode ln;			/* Current node */
757 
758     Lst_Open(path);
759     while ((ln = Lst_Next(path)) != NULL) {
760 	Path *p = LstNode_Datum(ln);
761 	DirMatchFiles(word, p, expansions);
762     }
763     Lst_Close(path);
764 }
765 
766 /* Print a word in the list of expansions.
767  * Callback for Dir_Expand when DEBUG(DIR), via Lst_ForEach. */
768 static int
769 DirPrintWord(void *word, void *dummy MAKE_ATTR_UNUSED)
770 {
771     fprintf(debug_file, "%s ", (char *)word);
772 
773     return 0;
774 }
775 
776 /*-
777  *-----------------------------------------------------------------------
778  * Dir_Expand  --
779  *	Expand the given word into a list of words by globbing it looking
780  *	in the directories on the given search path.
781  *
782  * Input:
783  *	word		the word to expand
784  *	path		the list of directories in which to find the
785  *			resulting files
786  *	expansions	the list on which to place the results
787  *
788  * Results:
789  *	A list of words consisting of the files which exist along the search
790  *	path matching the given pattern.
791  *
792  * Side Effects:
793  *	Directories may be opened. Who knows?
794  *	Undefined behavior if the word is really in read-only memory.
795  *-----------------------------------------------------------------------
796  */
797 void
798 Dir_Expand(const char *word, Lst path, Lst expansions)
799 {
800     const char *cp;
801 
802     assert(path != NULL);
803     assert(expansions != NULL);
804 
805     DIR_DEBUG1("Expanding \"%s\"... ", word);
806 
807     cp = strchr(word, '{');
808     if (cp) {
809 	DirExpandCurly(word, cp, path, expansions);
810     } else {
811 	cp = strchr(word, '/');
812 	if (cp) {
813 	    /*
814 	     * The thing has a directory component -- find the first wildcard
815 	     * in the string.
816 	     */
817 	    for (cp = word; *cp; cp++) {
818 		if (*cp == '?' || *cp == '[' || *cp == '*' || *cp == '{') {
819 		    break;
820 		}
821 	    }
822 	    if (*cp == '{') {
823 		/*
824 		 * This one will be fun.
825 		 */
826 		DirExpandCurly(word, cp, path, expansions);
827 		return;
828 	    } else if (*cp != '\0') {
829 		/*
830 		 * Back up to the start of the component
831 		 */
832 		while (cp > word && *cp != '/') {
833 		    cp--;
834 		}
835 		if (cp != word) {
836 		    char sc;
837 		    char *dirpath;
838 		    /*
839 		     * If the glob isn't in the first component, try and find
840 		     * all the components up to the one with a wildcard.
841 		     */
842 		    sc = cp[1];
843 		    ((char *)UNCONST(cp))[1] = '\0';
844 		    dirpath = Dir_FindFile(word, path);
845 		    ((char *)UNCONST(cp))[1] = sc;
846 		    /*
847 		     * dirpath is null if can't find the leading component
848 		     * XXX: Dir_FindFile won't find internal components.
849 		     * i.e. if the path contains ../Etc/Object and we're
850 		     * looking for Etc, it won't be found. Ah well.
851 		     * Probably not important.
852 		     */
853 		    if (dirpath != NULL) {
854 			char *dp = &dirpath[strlen(dirpath) - 1];
855 			if (*dp == '/')
856 			    *dp = '\0';
857 			path = Lst_Init();
858 			(void)Dir_AddDir(path, dirpath);
859 			DirExpandInt(cp + 1, path, expansions);
860 			Lst_Free(path);
861 		    }
862 		} else {
863 		    /*
864 		     * Start the search from the local directory
865 		     */
866 		    DirExpandInt(word, path, expansions);
867 		}
868 	    } else {
869 		/*
870 		 * Return the file -- this should never happen.
871 		 */
872 		DirExpandInt(word, path, expansions);
873 	    }
874 	} else {
875 	    /*
876 	     * First the files in dot
877 	     */
878 	    DirMatchFiles(word, dot, expansions);
879 
880 	    /*
881 	     * Then the files in every other directory on the path.
882 	     */
883 	    DirExpandInt(word, path, expansions);
884 	}
885     }
886     if (DEBUG(DIR)) {
887 	Lst_ForEach(expansions, DirPrintWord, NULL);
888 	fprintf(debug_file, "\n");
889     }
890 }
891 
892 /*-
893  *-----------------------------------------------------------------------
894  * DirLookup  --
895  *	Find if the file with the given name exists in the given path.
896  *
897  * Results:
898  *	The path to the file or NULL. This path is guaranteed to be in a
899  *	different part of memory than name and so may be safely free'd.
900  *
901  * Side Effects:
902  *	None.
903  *-----------------------------------------------------------------------
904  */
905 static char *
906 DirLookup(Path *p, const char *name MAKE_ATTR_UNUSED, const char *cp,
907 	  Boolean hasSlash MAKE_ATTR_UNUSED)
908 {
909     char *file;			/* the current filename to check */
910 
911     DIR_DEBUG1("   %s ...\n", p->name);
912 
913     if (Hash_FindEntry(&p->files, cp) == NULL)
914 	return NULL;
915 
916     file = str_concat3(p->name, "/", cp);
917     DIR_DEBUG1("   returning %s\n", file);
918     p->hits += 1;
919     hits += 1;
920     return file;
921 }
922 
923 
924 /*-
925  *-----------------------------------------------------------------------
926  * DirLookupSubdir  --
927  *	Find if the file with the given name exists in the given path.
928  *
929  * Results:
930  *	The path to the file or NULL. This path is guaranteed to be in a
931  *	different part of memory than name and so may be safely free'd.
932  *
933  * Side Effects:
934  *	If the file is found, it is added in the modification times hash
935  *	table.
936  *-----------------------------------------------------------------------
937  */
938 static char *
939 DirLookupSubdir(Path *p, const char *name)
940 {
941     struct make_stat mst;
942     char *file;			/* the current filename to check */
943 
944     if (p != dot) {
945 	file = str_concat3(p->name, "/", name);
946     } else {
947 	/*
948 	 * Checking in dot -- DON'T put a leading ./ on the thing.
949 	 */
950 	file = bmake_strdup(name);
951     }
952 
953     DIR_DEBUG1("checking %s ...\n", file);
954 
955     if (cached_stat(file, &mst) == 0) {
956 	nearmisses += 1;
957 	return file;
958     }
959     free(file);
960     return NULL;
961 }
962 
963 /*-
964  *-----------------------------------------------------------------------
965  * DirLookupAbs  --
966  *	Find if the file with the given name exists in the given path.
967  *
968  * Results:
969  *	The path to the file, the empty string or NULL. If the file is
970  *	the empty string, the search should be terminated.
971  *	This path is guaranteed to be in a different part of memory
972  *	than name and so may be safely free'd.
973  *
974  * Side Effects:
975  *	None.
976  *-----------------------------------------------------------------------
977  */
978 static char *
979 DirLookupAbs(Path *p, const char *name, const char *cp)
980 {
981     char *p1;			/* pointer into p->name */
982     const char *p2;		/* pointer into name */
983 
984     DIR_DEBUG1("   %s ...\n", p->name);
985 
986     /*
987      * If the file has a leading path component and that component
988      * exactly matches the entire name of the current search
989      * directory, we can attempt another cache lookup. And if we don't
990      * have a hit, we can safely assume the file does not exist at all.
991      */
992     for (p1 = p->name, p2 = name; *p1 && *p1 == *p2; p1++, p2++) {
993 	continue;
994     }
995     if (*p1 != '\0' || p2 != cp - 1) {
996 	return NULL;
997     }
998 
999     if (Hash_FindEntry(&p->files, cp) == NULL) {
1000 	DIR_DEBUG0("   must be here but isn't -- returning\n");
1001 	/* Return empty string: terminates search */
1002 	return bmake_strdup("");
1003     }
1004 
1005     p->hits += 1;
1006     hits += 1;
1007     DIR_DEBUG1("   returning %s\n", name);
1008     return bmake_strdup(name);
1009 }
1010 
1011 /*-
1012  *-----------------------------------------------------------------------
1013  * DirFindDot  --
1014  *	Find the file given on "." or curdir
1015  *
1016  * Results:
1017  *	The path to the file or NULL. This path is guaranteed to be in a
1018  *	different part of memory than name and so may be safely free'd.
1019  *
1020  * Side Effects:
1021  *	Hit counts change
1022  *-----------------------------------------------------------------------
1023  */
1024 static char *
1025 DirFindDot(Boolean hasSlash MAKE_ATTR_UNUSED, const char *name, const char *cp)
1026 {
1027 
1028     if (Hash_FindEntry(&dot->files, cp) != NULL) {
1029 	DIR_DEBUG0("   in '.'\n");
1030 	hits += 1;
1031 	dot->hits += 1;
1032 	return bmake_strdup(name);
1033     }
1034     if (cur && Hash_FindEntry(&cur->files, cp) != NULL) {
1035 	DIR_DEBUG1("   in ${.CURDIR} = %s\n", cur->name);
1036 	hits += 1;
1037 	cur->hits += 1;
1038 	return str_concat3(cur->name, "/", cp);
1039     }
1040 
1041     return NULL;
1042 }
1043 
1044 /*-
1045  *-----------------------------------------------------------------------
1046  * Dir_FindFile  --
1047  *	Find the file with the given name along the given search path.
1048  *
1049  * Input:
1050  *	name		the file to find
1051  *	path		the Lst of directories to search
1052  *
1053  * Results:
1054  *	The path to the file or NULL. This path is guaranteed to be in a
1055  *	different part of memory than name and so may be safely free'd.
1056  *
1057  * Side Effects:
1058  *	If the file is found in a directory which is not on the path
1059  *	already (either 'name' is absolute or it is a relative path
1060  *	[ dir1/.../dirn/file ] which exists below one of the directories
1061  *	already on the search path), its directory is added to the end
1062  *	of the path on the assumption that there will be more files in
1063  *	that directory later on. Sometimes this is true. Sometimes not.
1064  *-----------------------------------------------------------------------
1065  */
1066 char *
1067 Dir_FindFile(const char *name, Lst path)
1068 {
1069     LstNode ln;			/* a list element */
1070     char *file;			/* the current filename to check */
1071     Path *p;			/* current path member */
1072     const char *cp;		/* Terminal name of file */
1073     Boolean hasLastDot = FALSE;	/* true we should search dot last */
1074     Boolean hasSlash;		/* true if 'name' contains a / */
1075     struct make_stat mst;	/* Buffer for stat, if necessary */
1076     const char *trailing_dot = ".";
1077 
1078     /*
1079      * Find the final component of the name and note whether it has a
1080      * slash in it (the name, I mean)
1081      */
1082     cp = strrchr(name, '/');
1083     if (cp) {
1084 	hasSlash = TRUE;
1085 	cp += 1;
1086     } else {
1087 	hasSlash = FALSE;
1088 	cp = name;
1089     }
1090 
1091     DIR_DEBUG1("Searching for %s ...", name);
1092 
1093     if (path == NULL) {
1094 	DIR_DEBUG0("couldn't open path, file not found\n");
1095 	misses += 1;
1096 	return NULL;
1097     }
1098 
1099     Lst_Open(path);
1100     if ((ln = Lst_First(path)) != NULL) {
1101 	p = LstNode_Datum(ln);
1102 	if (p == dotLast) {
1103 	    hasLastDot = TRUE;
1104 	    DIR_DEBUG0("[dot last]...");
1105 	}
1106     }
1107     DIR_DEBUG0("\n");
1108 
1109     /*
1110      * If there's no leading directory components or if the leading
1111      * directory component is exactly `./', consult the cached contents
1112      * of each of the directories on the search path.
1113      */
1114     if (!hasSlash || (cp - name == 2 && *name == '.')) {
1115 	/*
1116 	 * We look through all the directories on the path seeking one which
1117 	 * contains the final component of the given name.  If such a beast
1118 	 * is found, we concatenate the directory name and the final
1119 	 * component and return the resulting string. If we don't find any
1120 	 * such thing, we go on to phase two...
1121 	 *
1122 	 * No matter what, we always look for the file in the current
1123 	 * directory before anywhere else (unless we found the magic
1124 	 * DOTLAST path, in which case we search it last) and we *do not*
1125 	 * add the ./ to it if it exists.
1126 	 * This is so there are no conflicts between what the user
1127 	 * specifies (fish.c) and what pmake finds (./fish.c).
1128 	 */
1129 	if (!hasLastDot && (file = DirFindDot(hasSlash, name, cp)) != NULL) {
1130 	    Lst_Close(path);
1131 	    return file;
1132 	}
1133 
1134 	while ((ln = Lst_Next(path)) != NULL) {
1135 	    p = LstNode_Datum(ln);
1136 	    if (p == dotLast)
1137 		continue;
1138 	    if ((file = DirLookup(p, name, cp, hasSlash)) != NULL) {
1139 		Lst_Close(path);
1140 		return file;
1141 	    }
1142 	}
1143 
1144 	if (hasLastDot && (file = DirFindDot(hasSlash, name, cp)) != NULL) {
1145 	    Lst_Close(path);
1146 	    return file;
1147 	}
1148     }
1149     Lst_Close(path);
1150 
1151     /*
1152      * We didn't find the file on any directory in the search path.
1153      * If the name doesn't contain a slash, that means it doesn't exist.
1154      * If it *does* contain a slash, however, there is still hope: it
1155      * could be in a subdirectory of one of the members of the search
1156      * path. (eg. /usr/include and sys/types.h. The above search would
1157      * fail to turn up types.h in /usr/include, but it *is* in
1158      * /usr/include/sys/types.h).
1159      * [ This no longer applies: If we find such a beast, we assume there
1160      * will be more (what else can we assume?) and add all but the last
1161      * component of the resulting name onto the search path (at the
1162      * end).]
1163      * This phase is only performed if the file is *not* absolute.
1164      */
1165     if (!hasSlash) {
1166 	DIR_DEBUG0("   failed.\n");
1167 	misses += 1;
1168 	return NULL;
1169     }
1170 
1171     if (*cp == '\0') {
1172 	/* we were given a trailing "/" */
1173 	cp = trailing_dot;
1174     }
1175 
1176     if (name[0] != '/') {
1177 	Boolean checkedDot = FALSE;
1178 
1179 	DIR_DEBUG0("   Trying subdirectories...\n");
1180 
1181 	if (!hasLastDot) {
1182 	    if (dot) {
1183 		checkedDot = TRUE;
1184 		if ((file = DirLookupSubdir(dot, name)) != NULL)
1185 		    return file;
1186 	    }
1187 	    if (cur && (file = DirLookupSubdir(cur, name)) != NULL)
1188 		return file;
1189 	}
1190 
1191 	Lst_Open(path);
1192 	while ((ln = Lst_Next(path)) != NULL) {
1193 	    p = LstNode_Datum(ln);
1194 	    if (p == dotLast)
1195 		continue;
1196 	    if (p == dot) {
1197 		if (checkedDot)
1198 		    continue;
1199 		checkedDot = TRUE;
1200 	    }
1201 	    if ((file = DirLookupSubdir(p, name)) != NULL) {
1202 		Lst_Close(path);
1203 		return file;
1204 	    }
1205 	}
1206 	Lst_Close(path);
1207 
1208 	if (hasLastDot) {
1209 	    if (dot && !checkedDot) {
1210 		checkedDot = TRUE;
1211 		if ((file = DirLookupSubdir(dot, name)) != NULL)
1212 		    return file;
1213 	    }
1214 	    if (cur && (file = DirLookupSubdir(cur, name)) != NULL)
1215 		return file;
1216 	}
1217 
1218 	if (checkedDot) {
1219 	    /*
1220 	     * Already checked by the given name, since . was in the path,
1221 	     * so no point in proceeding...
1222 	     */
1223 	    DIR_DEBUG0("   Checked . already, returning NULL\n");
1224 	    return NULL;
1225 	}
1226 
1227     } else { /* name[0] == '/' */
1228 
1229 	/*
1230 	 * For absolute names, compare directory path prefix against the
1231 	 * the directory path of each member on the search path for an exact
1232 	 * match. If we have an exact match on any member of the search path,
1233 	 * use the cached contents of that member to lookup the final file
1234 	 * component. If that lookup fails we can safely assume that the
1235 	 * file does not exist at all.  This is signified by DirLookupAbs()
1236 	 * returning an empty string.
1237 	 */
1238 	DIR_DEBUG0("   Trying exact path matches...\n");
1239 
1240 	if (!hasLastDot && cur &&
1241 	    ((file = DirLookupAbs(cur, name, cp)) != NULL)) {
1242 	    if (file[0] == '\0') {
1243 		free(file);
1244 		return NULL;
1245 	    }
1246 	    return file;
1247 	}
1248 
1249 	Lst_Open(path);
1250 	while ((ln = Lst_Next(path)) != NULL) {
1251 	    p = LstNode_Datum(ln);
1252 	    if (p == dotLast)
1253 		continue;
1254 	    if ((file = DirLookupAbs(p, name, cp)) != NULL) {
1255 		Lst_Close(path);
1256 		if (file[0] == '\0') {
1257 		    free(file);
1258 		    return NULL;
1259 		}
1260 		return file;
1261 	    }
1262 	}
1263 	Lst_Close(path);
1264 
1265 	if (hasLastDot && cur &&
1266 	    ((file = DirLookupAbs(cur, name, cp)) != NULL)) {
1267 	    if (file[0] == '\0') {
1268 		free(file);
1269 		return NULL;
1270 	    }
1271 	    return file;
1272 	}
1273     }
1274 
1275     /*
1276      * Didn't find it that way, either. Sigh. Phase 3. Add its directory
1277      * onto the search path in any case, just in case, then look for the
1278      * thing in the hash table. If we find it, grand. We return a new
1279      * copy of the name. Otherwise we sadly return a NULL pointer. Sigh.
1280      * Note that if the directory holding the file doesn't exist, this will
1281      * do an extra search of the final directory on the path. Unless something
1282      * weird happens, this search won't succeed and life will be groovy.
1283      *
1284      * Sigh. We cannot add the directory onto the search path because
1285      * of this amusing case:
1286      * $(INSTALLDIR)/$(FILE): $(FILE)
1287      *
1288      * $(FILE) exists in $(INSTALLDIR) but not in the current one.
1289      * When searching for $(FILE), we will find it in $(INSTALLDIR)
1290      * b/c we added it here. This is not good...
1291      */
1292 #ifdef notdef
1293     if (cp == traling_dot) {
1294 	cp = strrchr(name, '/');
1295 	cp += 1;
1296     }
1297     cp[-1] = '\0';
1298     (void)Dir_AddDir(path, name);
1299     cp[-1] = '/';
1300 
1301     bigmisses += 1;
1302     ln = Lst_Last(path);
1303     if (ln == NULL) {
1304 	return NULL;
1305     } else {
1306 	p = LstNode_Datum(ln);
1307     }
1308 
1309     if (Hash_FindEntry(&p->files, cp) != NULL) {
1310 	return bmake_strdup(name);
1311     } else {
1312 	return NULL;
1313     }
1314 #else /* !notdef */
1315     DIR_DEBUG1("   Looking for \"%s\" ...\n", name);
1316 
1317     bigmisses += 1;
1318     if (cached_stat(name, &mst) == 0) {
1319 	return bmake_strdup(name);
1320     }
1321 
1322     DIR_DEBUG0("   failed. Returning NULL\n");
1323     return NULL;
1324 #endif /* notdef */
1325 }
1326 
1327 
1328 /*-
1329  *-----------------------------------------------------------------------
1330  * Dir_FindHereOrAbove  --
1331  *	search for a path starting at a given directory and then working
1332  *	our way up towards the root.
1333  *
1334  * Input:
1335  *	here		starting directory
1336  *	search_path	the path we are looking for
1337  *	result		the result of a successful search is placed here
1338  *	result_len	the length of the result buffer
1339  *			(typically MAXPATHLEN + 1)
1340  *
1341  * Results:
1342  *	0 on failure, 1 on success [in which case the found path is put
1343  *	in the result buffer].
1344  *
1345  * Side Effects:
1346  *-----------------------------------------------------------------------
1347  */
1348 Boolean
1349 Dir_FindHereOrAbove(const char *here, const char *search_path,
1350 		    char *result, int result_len)
1351 {
1352     struct make_stat mst;
1353     char dirbase[MAXPATHLEN + 1], *dirbase_end;
1354     char try[MAXPATHLEN + 1], *try_end;
1355 
1356     /* copy out our starting point */
1357     snprintf(dirbase, sizeof(dirbase), "%s", here);
1358     dirbase_end = dirbase + strlen(dirbase);
1359 
1360     /* loop until we determine a result */
1361     while (TRUE) {
1362 
1363 	/* try and stat(2) it ... */
1364 	snprintf(try, sizeof(try), "%s/%s", dirbase, search_path);
1365 	if (cached_stat(try, &mst) != -1) {
1366 	    /*
1367 	     * success!  if we found a file, chop off
1368 	     * the filename so we return a directory.
1369 	     */
1370 	    if ((mst.mst_mode & S_IFMT) != S_IFDIR) {
1371 		try_end = try + strlen(try);
1372 		while (try_end > try && *try_end != '/')
1373 		    try_end--;
1374 		if (try_end > try)
1375 		    *try_end = '\0';	/* chop! */
1376 	    }
1377 
1378 	    snprintf(result, result_len, "%s", try);
1379 	    return TRUE;
1380 	}
1381 
1382 	/*
1383 	 * nope, we didn't find it.  if we used up dirbase we've
1384 	 * reached the root and failed.
1385 	 */
1386 	if (dirbase_end == dirbase)
1387 	    break;		/* failed! */
1388 
1389 	/*
1390 	 * truncate dirbase from the end to move up a dir
1391 	 */
1392 	while (dirbase_end > dirbase && *dirbase_end != '/')
1393 	    dirbase_end--;
1394 	*dirbase_end = '\0';	/* chop! */
1395 
1396     } /* while (TRUE) */
1397 
1398     return FALSE;
1399 }
1400 
1401 /*-
1402  *-----------------------------------------------------------------------
1403  * Dir_MTime  --
1404  *	Find the modification time of the file described by gn along the
1405  *	search path dirSearchPath.
1406  *
1407  * Input:
1408  *	gn		the file whose modification time is desired
1409  *
1410  * Results:
1411  *	The modification time or 0 if it doesn't exist
1412  *
1413  * Side Effects:
1414  *	The modification time is placed in the node's mtime slot.
1415  *	If the node didn't have a path entry before, and Dir_FindFile
1416  *	found one for it, the full name is placed in the path slot.
1417  *-----------------------------------------------------------------------
1418  */
1419 int
1420 Dir_MTime(GNode *gn, Boolean recheck)
1421 {
1422     char *fullName;		/* the full pathname of name */
1423     struct make_stat mst;	/* buffer for finding the mod time */
1424 
1425     if (gn->type & OP_ARCHV) {
1426 	return Arch_MTime(gn);
1427     } else if (gn->type & OP_PHONY) {
1428 	gn->mtime = 0;
1429 	return 0;
1430     } else if (gn->path == NULL) {
1431 	if (gn->type & OP_NOPATH)
1432 	    fullName = NULL;
1433 	else {
1434 	    fullName = Dir_FindFile(gn->name, Suff_FindPath(gn));
1435 	    if (fullName == NULL && gn->flags & FROM_DEPEND &&
1436 		!Lst_IsEmpty(gn->implicitParents)) {
1437 		char *cp;
1438 
1439 		cp = strrchr(gn->name, '/');
1440 		if (cp) {
1441 		    /*
1442 		     * This is an implied source, and it may have moved,
1443 		     * see if we can find it via the current .PATH
1444 		     */
1445 		    cp++;
1446 
1447 		    fullName = Dir_FindFile(cp, Suff_FindPath(gn));
1448 		    if (fullName) {
1449 			/*
1450 			 * Put the found file in gn->path
1451 			 * so that we give that to the compiler.
1452 			 */
1453 			gn->path = bmake_strdup(fullName);
1454 			if (!Job_RunTarget(".STALE", gn->fname))
1455 			    fprintf(stdout,
1456 				    "%s: %s, %d: ignoring stale %s for %s, "
1457 				    "found %s\n", progname, gn->fname,
1458 				    gn->lineno,
1459 				    makeDependfile, gn->name, fullName);
1460 		    }
1461 		}
1462 	    }
1463 	    DIR_DEBUG2("Found '%s' as '%s'\n",
1464 		       gn->name, fullName ? fullName : "(not found)");
1465 	}
1466     } else {
1467 	fullName = gn->path;
1468     }
1469 
1470     if (fullName == NULL) {
1471 	fullName = bmake_strdup(gn->name);
1472     }
1473 
1474     if (cached_stats(&mtimes, fullName, &mst, recheck ? CST_UPDATE : 0) < 0) {
1475 	if (gn->type & OP_MEMBER) {
1476 	    if (fullName != gn->path)
1477 		free(fullName);
1478 	    return Arch_MemMTime(gn);
1479 	} else {
1480 	    mst.mst_mtime = 0;
1481 	}
1482     }
1483 
1484     if (fullName && gn->path == NULL) {
1485 	gn->path = fullName;
1486     }
1487 
1488     gn->mtime = mst.mst_mtime;
1489     return gn->mtime;
1490 }
1491 
1492 /*-
1493  *-----------------------------------------------------------------------
1494  * Dir_AddDir --
1495  *	Add the given name to the end of the given path. The order of
1496  *	the arguments is backwards so ParseDoDependency can do a
1497  *	Lst_ForEach of its list of paths...
1498  *
1499  * Input:
1500  *	path		the path to which the directory should be
1501  *			added
1502  *			XXX: Why would this ever be NULL, and what does
1503  *			that mean?
1504  *	name		the name of the directory to add
1505  *
1506  * Results:
1507  *	none
1508  *
1509  * Side Effects:
1510  *	A structure is added to the list and the directory is
1511  *	read and hashed.
1512  *-----------------------------------------------------------------------
1513  */
1514 Path *
1515 Dir_AddDir(Lst path, const char *name)
1516 {
1517     LstNode ln = NULL;		/* node in case Path structure is found */
1518     Path *p = NULL;		/* pointer to new Path structure */
1519     DIR *d;			/* for reading directory */
1520     struct dirent *dp;		/* entry in directory */
1521 
1522     if (path != NULL && strcmp(name, ".DOTLAST") == 0) {
1523 	ln = Lst_Find(path, DirFindName, name);
1524 	if (ln != NULL)
1525 	    return LstNode_Datum(ln);
1526 
1527 	dotLast->refCount++;
1528 	Lst_Prepend(path, dotLast);
1529     }
1530 
1531     if (path != NULL)
1532 	ln = Lst_Find(openDirectories, DirFindName, name);
1533     if (ln != NULL) {
1534 	p = LstNode_Datum(ln);
1535 	if (Lst_FindDatum(path, p) == NULL) {
1536 	    p->refCount += 1;
1537 	    Lst_Append(path, p);
1538 	}
1539 	return p;
1540     }
1541 
1542     DIR_DEBUG1("Caching %s ...", name);
1543 
1544     if ((d = opendir(name)) != NULL) {
1545 	p = bmake_malloc(sizeof(Path));
1546 	p->name = bmake_strdup(name);
1547 	p->hits = 0;
1548 	p->refCount = 1;
1549 	Hash_InitTable(&p->files);
1550 
1551 	while ((dp = readdir(d)) != NULL) {
1552 #if defined(sun) && defined(d_ino) /* d_ino is a sunos4 #define for d_fileno */
1553 	    /*
1554 	     * The sun directory library doesn't check for a 0 inode
1555 	     * (0-inode slots just take up space), so we have to do
1556 	     * it ourselves.
1557 	     */
1558 	    if (dp->d_fileno == 0) {
1559 		continue;
1560 	    }
1561 #endif /* sun && d_ino */
1562 	    (void)Hash_CreateEntry(&p->files, dp->d_name, NULL);
1563 	}
1564 	(void)closedir(d);
1565 	Lst_Append(openDirectories, p);
1566 	if (path != NULL)
1567 	    Lst_Append(path, p);
1568     }
1569     DIR_DEBUG0("done\n");
1570     return p;
1571 }
1572 
1573 /*-
1574  *-----------------------------------------------------------------------
1575  * Dir_CopyDir --
1576  *	Callback function for duplicating a search path via Lst_Copy.
1577  *	Ups the reference count for the directory.
1578  *
1579  * Results:
1580  *	Returns the Path it was given.
1581  *-----------------------------------------------------------------------
1582  */
1583 void *
1584 Dir_CopyDir(void *p)
1585 {
1586     ((Path *)p)->refCount += 1;
1587 
1588     return p;
1589 }
1590 
1591 /*-
1592  *-----------------------------------------------------------------------
1593  * Dir_MakeFlags --
1594  *	Make a string by taking all the directories in the given search
1595  *	path and preceding them by the given flag. Used by the suffix
1596  *	module to create variables for compilers based on suffix search
1597  *	paths.
1598  *
1599  * Input:
1600  *	flag		flag which should precede each directory
1601  *	path		list of directories
1602  *
1603  * Results:
1604  *	The string mentioned above. Note that there is no space between
1605  *	the given flag and each directory. The empty string is returned if
1606  *	Things don't go well.
1607  *
1608  * Side Effects:
1609  *	None
1610  *-----------------------------------------------------------------------
1611  */
1612 char *
1613 Dir_MakeFlags(const char *flag, Lst path)
1614 {
1615     Buffer buf;
1616     LstNode ln;			/* the node of the current directory */
1617 
1618     Buf_Init(&buf, 0);
1619 
1620     if (path != NULL) {
1621 	Lst_Open(path);
1622 	while ((ln = Lst_Next(path)) != NULL) {
1623 	    Path *p = LstNode_Datum(ln);
1624 	    Buf_AddStr(&buf, " ");
1625 	    Buf_AddStr(&buf, flag);
1626 	    Buf_AddStr(&buf, p->name);
1627 	}
1628 	Lst_Close(path);
1629     }
1630 
1631     return Buf_Destroy(&buf, FALSE);
1632 }
1633 
1634 /*-
1635  *-----------------------------------------------------------------------
1636  * Dir_Destroy --
1637  *	Nuke a directory descriptor, if possible. Callback procedure
1638  *	for the suffixes module when destroying a search path.
1639  *
1640  * Input:
1641  *	pp		The directory descriptor to nuke
1642  *
1643  * Results:
1644  *	None.
1645  *
1646  * Side Effects:
1647  *	If no other path references this directory (refCount == 0),
1648  *	the Path and all its data are freed.
1649  *
1650  *-----------------------------------------------------------------------
1651  */
1652 void
1653 Dir_Destroy(void *pp)
1654 {
1655     Path *p = (Path *)pp;
1656     p->refCount -= 1;
1657 
1658     if (p->refCount == 0) {
1659 	LstNode ln;
1660 
1661 	ln = Lst_FindDatum(openDirectories, p);
1662 	Lst_Remove(openDirectories, ln);
1663 
1664 	Hash_DeleteTable(&p->files);
1665 	free(p->name);
1666 	free(p);
1667     }
1668 }
1669 
1670 /*-
1671  *-----------------------------------------------------------------------
1672  * Dir_ClearPath --
1673  *	Clear out all elements of the given search path. This is different
1674  *	from destroying the list, notice.
1675  *
1676  * Input:
1677  *	path		Path to clear
1678  *
1679  * Results:
1680  *	None.
1681  *
1682  * Side Effects:
1683  *	The path is set to the empty list.
1684  *
1685  *-----------------------------------------------------------------------
1686  */
1687 void
1688 Dir_ClearPath(Lst path)
1689 {
1690     while (!Lst_IsEmpty(path)) {
1691 	Path *p = Lst_Dequeue(path);
1692 	Dir_Destroy(p);
1693     }
1694 }
1695 
1696 
1697 /*-
1698  *-----------------------------------------------------------------------
1699  * Dir_Concat --
1700  *	Concatenate two paths, adding the second to the end of the first.
1701  *	Makes sure to avoid duplicates.
1702  *
1703  * Input:
1704  *	path1		Dest
1705  *	path2		Source
1706  *
1707  * Results:
1708  *	None
1709  *
1710  * Side Effects:
1711  *	Reference counts for added dirs are upped.
1712  *
1713  *-----------------------------------------------------------------------
1714  */
1715 void
1716 Dir_Concat(Lst path1, Lst path2)
1717 {
1718     LstNode ln;
1719     Path *p;
1720 
1721     for (ln = Lst_First(path2); ln != NULL; ln = LstNode_Next(ln)) {
1722 	p = LstNode_Datum(ln);
1723 	if (Lst_FindDatum(path1, p) == NULL) {
1724 	    p->refCount += 1;
1725 	    Lst_Append(path1, p);
1726 	}
1727     }
1728 }
1729 
1730 static int
1731 percentage(int num, int den)
1732 {
1733     return den != 0 ? num * 100 / den : 0;
1734 }
1735 
1736 /********** DEBUG INFO **********/
1737 void
1738 Dir_PrintDirectories(void)
1739 {
1740     LstNode ln;
1741 
1742     fprintf(debug_file, "#*** Directory Cache:\n");
1743     fprintf(debug_file,
1744 	    "# Stats: %d hits %d misses %d near misses %d losers (%d%%)\n",
1745 	    hits, misses, nearmisses, bigmisses,
1746 	    percentage(hits, hits + bigmisses + nearmisses));
1747     fprintf(debug_file, "# %-20s referenced\thits\n", "directory");
1748 
1749     Lst_Open(openDirectories);
1750     while ((ln = Lst_Next(openDirectories)) != NULL) {
1751 	Path *p = LstNode_Datum(ln);
1752 	fprintf(debug_file, "# %-20s %10d\t%4d\n", p->name, p->refCount,
1753 		p->hits);
1754     }
1755     Lst_Close(openDirectories);
1756 }
1757 
1758 static int
1759 DirPrintDir(void *p, void *dummy MAKE_ATTR_UNUSED)
1760 {
1761     fprintf(debug_file, "%s ", ((Path *)p)->name);
1762     return 0;
1763 }
1764 
1765 void
1766 Dir_PrintPath(Lst path)
1767 {
1768     Lst_ForEach(path, DirPrintDir, NULL);
1769 }
1770