xref: /original-bsd/usr.bin/make/dir.c (revision 065e4d72)
1 /*
2  * Copyright (c) 1988, 1989, 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * Copyright (c) 1989 by Berkeley Softworks
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Adam de Boor.
9  *
10  * %sccs.include.redist.c%
11  */
12 
13 #ifndef lint
14 static char sccsid[] = "@(#)dir.c	8.3 (Berkeley) 04/28/95";
15 #endif /* not lint */
16 
17 /*-
18  * dir.c --
19  *	Directory searching using wildcards and/or normal names...
20  *	Used both for source wildcarding in the Makefile and for finding
21  *	implicit sources.
22  *
23  * The interface for this module is:
24  *	Dir_Init  	    Initialize the module.
25  *
26  *	Dir_End  	    Cleanup the module.
27  *
28  *	Dir_HasWildcards    Returns TRUE if the name given it needs to
29  *	    	  	    be wildcard-expanded.
30  *
31  *	Dir_Expand	    Given a pattern and a path, return a Lst of names
32  *	    	  	    which match the pattern on the search path.
33  *
34  *	Dir_FindFile	    Searches for a file on a given search path.
35  *	    	  	    If it exists, the entire path is returned.
36  *	    	  	    Otherwise NULL is returned.
37  *
38  *	Dir_MTime 	    Return the modification time of a node. The file
39  *	    	  	    is searched for along the default search path.
40  *	    	  	    The path and mtime fields of the node are filled
41  *	    	  	    in.
42  *
43  *	Dir_AddDir	    Add a directory to a search path.
44  *
45  *	Dir_MakeFlags	    Given a search path and a command flag, create
46  *	    	  	    a string with each of the directories in the path
47  *	    	  	    preceded by the command flag and all of them
48  *	    	  	    separated by a space.
49  *
50  *	Dir_Destroy	    Destroy an element of a search path. Frees up all
51  *	    	  	    things that can be freed for the element as long
52  *	    	  	    as the element is no longer referenced by any other
53  *	    	  	    search path.
54  *	Dir_ClearPath	    Resets a search path to the empty list.
55  *
56  * For debugging:
57  *	Dir_PrintDirectories	Print stats about the directory cache.
58  */
59 
60 #include <stdio.h>
61 #include <sys/types.h>
62 #include <dirent.h>
63 #include <sys/stat.h>
64 #include "make.h"
65 #include "hash.h"
66 #include "dir.h"
67 
68 /*
69  *	A search path consists of a Lst of Path structures. A Path structure
70  *	has in it the name of the directory and a hash table of all the files
71  *	in the directory. This is used to cut down on the number of system
72  *	calls necessary to find implicit dependents and their like. Since
73  *	these searches are made before any actions are taken, we need not
74  *	worry about the directory changing due to creation commands. If this
75  *	hampers the style of some makefiles, they must be changed.
76  *
77  *	A list of all previously-read directories is kept in the
78  *	openDirectories Lst. This list is checked first before a directory
79  *	is opened.
80  *
81  *	The need for the caching of whole directories is brought about by
82  *	the multi-level transformation code in suff.c, which tends to search
83  *	for far more files than regular make does. In the initial
84  *	implementation, the amount of time spent performing "stat" calls was
85  *	truly astronomical. The problem with hashing at the start is,
86  *	of course, that pmake doesn't then detect changes to these directories
87  *	during the course of the make. Three possibilities suggest themselves:
88  *
89  *	    1) just use stat to test for a file's existence. As mentioned
90  *	       above, this is very inefficient due to the number of checks
91  *	       engendered by the multi-level transformation code.
92  *	    2) use readdir() and company to search the directories, keeping
93  *	       them open between checks. I have tried this and while it
94  *	       didn't slow down the process too much, it could severely
95  *	       affect the amount of parallelism available as each directory
96  *	       open would take another file descriptor out of play for
97  *	       handling I/O for another job. Given that it is only recently
98  *	       that UNIX OS's have taken to allowing more than 20 or 32
99  *	       file descriptors for a process, this doesn't seem acceptable
100  *	       to me.
101  *	    3) record the mtime of the directory in the Path structure and
102  *	       verify the directory hasn't changed since the contents were
103  *	       hashed. This will catch the creation or deletion of files,
104  *	       but not the updating of files. However, since it is the
105  *	       creation and deletion that is the problem, this could be
106  *	       a good thing to do. Unfortunately, if the directory (say ".")
107  *	       were fairly large and changed fairly frequently, the constant
108  *	       rehashing could seriously degrade performance. It might be
109  *	       good in such cases to keep track of the number of rehashes
110  *	       and if the number goes over a (small) limit, resort to using
111  *	       stat in its place.
112  *
113  *	An additional thing to consider is that pmake is used primarily
114  *	to create C programs and until recently pcc-based compilers refused
115  *	to allow you to specify where the resulting object file should be
116  *	placed. This forced all objects to be created in the current
117  *	directory. This isn't meant as a full excuse, just an explanation of
118  *	some of the reasons for the caching used here.
119  *
120  *	One more note: the location of a target's file is only performed
121  *	on the downward traversal of the graph and then only for terminal
122  *	nodes in the graph. This could be construed as wrong in some cases,
123  *	but prevents inadvertent modification of files when the "installed"
124  *	directory for a file is provided in the search path.
125  *
126  *	Another data structure maintained by this module is an mtime
127  *	cache used when the searching of cached directories fails to find
128  *	a file. In the past, Dir_FindFile would simply perform an access()
129  *	call in such a case to determine if the file could be found using
130  *	just the name given. When this hit, however, all that was gained
131  *	was the knowledge that the file existed. Given that an access() is
132  *	essentially a stat() without the copyout() call, and that the same
133  *	filesystem overhead would have to be incurred in Dir_MTime, it made
134  *	sense to replace the access() with a stat() and record the mtime
135  *	in a cache for when Dir_MTime was actually called.
136  */
137 
138 Lst          dirSearchPath;	/* main search path */
139 
140 static Lst   openDirectories;	/* the list of all open directories */
141 
142 /*
143  * Variables for gathering statistics on the efficiency of the hashing
144  * mechanism.
145  */
146 static int    hits,	      /* Found in directory cache */
147 	      misses,	      /* Sad, but not evil misses */
148 	      nearmisses,     /* Found under search path */
149 	      bigmisses;      /* Sought by itself */
150 
151 static Path    	  *dot;	    /* contents of current directory */
152 static Hash_Table mtimes;   /* Results of doing a last-resort stat in
153 			     * Dir_FindFile -- if we have to go to the
154 			     * system to find the file, we might as well
155 			     * have its mtime on record. XXX: If this is done
156 			     * way early, there's a chance other rules will
157 			     * have already updated the file, in which case
158 			     * we'll update it again. Generally, there won't
159 			     * be two rules to update a single file, so this
160 			     * should be ok, but... */
161 
162 
163 static int DirFindName __P((ClientData, ClientData));
164 static int DirMatchFiles __P((char *, Path *, Lst));
165 static void DirExpandCurly __P((char *, char *, Lst, Lst));
166 static void DirExpandInt __P((char *, Lst, Lst));
167 static int DirPrintWord __P((ClientData, ClientData));
168 static int DirPrintDir __P((ClientData, ClientData));
169 
170 /*-
171  *-----------------------------------------------------------------------
172  * Dir_Init --
173  *	initialize things for this module
174  *
175  * Results:
176  *	none
177  *
178  * Side Effects:
179  *	some directories may be opened.
180  *-----------------------------------------------------------------------
181  */
182 void
183 Dir_Init ()
184 {
185     dirSearchPath = Lst_Init (FALSE);
186     openDirectories = Lst_Init (FALSE);
187     Hash_InitTable(&mtimes, 0);
188 
189     /*
190      * Since the Path structure is placed on both openDirectories and
191      * the path we give Dir_AddDir (which in this case is openDirectories),
192      * we need to remove "." from openDirectories and what better time to
193      * do it than when we have to fetch the thing anyway?
194      */
195     Dir_AddDir (openDirectories, ".");
196     dot = (Path *) Lst_DeQueue (openDirectories);
197 
198     /*
199      * We always need to have dot around, so we increment its reference count
200      * to make sure it's not destroyed.
201      */
202     dot->refCount += 1;
203 }
204 
205 /*-
206  *-----------------------------------------------------------------------
207  * Dir_End --
208  *	cleanup things for this module
209  *
210  * Results:
211  *	none
212  *
213  * Side Effects:
214  *	none
215  *-----------------------------------------------------------------------
216  */
217 void
218 Dir_End()
219 {
220     dot->refCount -= 1;
221     Dir_Destroy((ClientData) dot);
222     Dir_ClearPath(dirSearchPath);
223     Lst_Destroy(dirSearchPath, NOFREE);
224     Dir_ClearPath(openDirectories);
225     Lst_Destroy(openDirectories, NOFREE);
226     Hash_DeleteTable(&mtimes);
227 }
228 
229 /*-
230  *-----------------------------------------------------------------------
231  * DirFindName --
232  *	See if the Path structure describes the same directory as the
233  *	given one by comparing their names. Called from Dir_AddDir via
234  *	Lst_Find when searching the list of open directories.
235  *
236  * Results:
237  *	0 if it is the same. Non-zero otherwise
238  *
239  * Side Effects:
240  *	None
241  *-----------------------------------------------------------------------
242  */
243 static int
244 DirFindName (p, dname)
245     ClientData    p;	      /* Current name */
246     ClientData	  dname;      /* Desired name */
247 {
248     return (strcmp (((Path *)p)->name, (char *) dname));
249 }
250 
251 /*-
252  *-----------------------------------------------------------------------
253  * Dir_HasWildcards  --
254  *	see if the given name has any wildcard characters in it
255  *
256  * Results:
257  *	returns TRUE if the word should be expanded, FALSE otherwise
258  *
259  * Side Effects:
260  *	none
261  *-----------------------------------------------------------------------
262  */
263 Boolean
264 Dir_HasWildcards (name)
265     char          *name;	/* name to check */
266 {
267     register char *cp;
268 
269     for (cp = name; *cp; cp++) {
270 	switch(*cp) {
271 	case '{':
272 	case '[':
273 	case '?':
274 	case '*':
275 	    return (TRUE);
276 	}
277     }
278     return (FALSE);
279 }
280 
281 /*-
282  *-----------------------------------------------------------------------
283  * DirMatchFiles --
284  * 	Given a pattern and a Path structure, see if any files
285  *	match the pattern and add their names to the 'expansions' list if
286  *	any do. This is incomplete -- it doesn't take care of patterns like
287  *	src / *src / *.c properly (just *.c on any of the directories), but it
288  *	will do for now.
289  *
290  * Results:
291  *	Always returns 0
292  *
293  * Side Effects:
294  *	File names are added to the expansions lst. The directory will be
295  *	fully hashed when this is done.
296  *-----------------------------------------------------------------------
297  */
298 static int
299 DirMatchFiles (pattern, p, expansions)
300     char	  *pattern;   	/* Pattern to look for */
301     Path	  *p;         	/* Directory to search */
302     Lst	    	  expansions;	/* Place to store the results */
303 {
304     Hash_Search	  search;   	/* Index into the directory's table */
305     Hash_Entry	  *entry;   	/* Current entry in the table */
306     Boolean 	  isDot;    	/* TRUE if the directory being searched is . */
307 
308     isDot = (*p->name == '.' && p->name[1] == '\0');
309 
310     for (entry = Hash_EnumFirst(&p->files, &search);
311 	 entry != (Hash_Entry *)NULL;
312 	 entry = Hash_EnumNext(&search))
313     {
314 	/*
315 	 * See if the file matches the given pattern. Note we follow the UNIX
316 	 * convention that dot files will only be found if the pattern
317 	 * begins with a dot (note also that as a side effect of the hashing
318 	 * scheme, .* won't match . or .. since they aren't hashed).
319 	 */
320 	if (Str_Match(entry->name, pattern) &&
321 	    ((entry->name[0] != '.') ||
322 	     (pattern[0] == '.')))
323 	{
324 	    (void)Lst_AtEnd(expansions,
325 			    (isDot ? strdup(entry->name) :
326 			     str_concat(p->name, entry->name,
327 					STR_ADDSLASH)));
328 	}
329     }
330     return (0);
331 }
332 
333 /*-
334  *-----------------------------------------------------------------------
335  * DirExpandCurly --
336  *	Expand curly braces like the C shell. Does this recursively.
337  *	Note the special case: if after the piece of the curly brace is
338  *	done there are no wildcard characters in the result, the result is
339  *	placed on the list WITHOUT CHECKING FOR ITS EXISTENCE.
340  *
341  * Results:
342  *	None.
343  *
344  * Side Effects:
345  *	The given list is filled with the expansions...
346  *
347  *-----------------------------------------------------------------------
348  */
349 static void
350 DirExpandCurly(word, brace, path, expansions)
351     char    	  *word;    	/* Entire word to expand */
352     char    	  *brace;   	/* First curly brace in it */
353     Lst	    	  path;	    	/* Search path to use */
354     Lst	    	  expansions;	/* Place to store the expansions */
355 {
356     char    	  *end;	    	/* Character after the closing brace */
357     char    	  *cp;	    	/* Current position in brace clause */
358     char    	  *start;   	/* Start of current piece of brace clause */
359     int	    	  bracelevel;	/* Number of braces we've seen. If we see a
360 				 * right brace when this is 0, we've hit the
361 				 * end of the clause. */
362     char    	  *file;    	/* Current expansion */
363     int	    	  otherLen; 	/* The length of the other pieces of the
364 				 * expansion (chars before and after the
365 				 * clause in 'word') */
366     char    	  *cp2;	    	/* Pointer for checking for wildcards in
367 				 * expansion before calling Dir_Expand */
368 
369     start = brace+1;
370 
371     /*
372      * Find the end of the brace clause first, being wary of nested brace
373      * clauses.
374      */
375     for (end = start, bracelevel = 0; *end != '\0'; end++) {
376 	if (*end == '{') {
377 	    bracelevel++;
378 	} else if ((*end == '}') && (bracelevel-- == 0)) {
379 	    break;
380 	}
381     }
382     if (*end == '\0') {
383 	Error("Unterminated {} clause \"%s\"", start);
384 	return;
385     } else {
386 	end++;
387     }
388     otherLen = brace - word + strlen(end);
389 
390     for (cp = start; cp < end; cp++) {
391 	/*
392 	 * Find the end of this piece of the clause.
393 	 */
394 	bracelevel = 0;
395 	while (*cp != ',') {
396 	    if (*cp == '{') {
397 		bracelevel++;
398 	    } else if ((*cp == '}') && (bracelevel-- <= 0)) {
399 		break;
400 	    }
401 	    cp++;
402 	}
403 	/*
404 	 * Allocate room for the combination and install the three pieces.
405 	 */
406 	file = emalloc(otherLen + cp - start + 1);
407 	if (brace != word) {
408 	    strncpy(file, word, brace-word);
409 	}
410 	if (cp != start) {
411 	    strncpy(&file[brace-word], start, cp-start);
412 	}
413 	strcpy(&file[(brace-word)+(cp-start)], end);
414 
415 	/*
416 	 * See if the result has any wildcards in it. If we find one, call
417 	 * Dir_Expand right away, telling it to place the result on our list
418 	 * of expansions.
419 	 */
420 	for (cp2 = file; *cp2 != '\0'; cp2++) {
421 	    switch(*cp2) {
422 	    case '*':
423 	    case '?':
424 	    case '{':
425 	    case '[':
426 		Dir_Expand(file, path, expansions);
427 		goto next;
428 	    }
429 	}
430 	if (*cp2 == '\0') {
431 	    /*
432 	     * Hit the end w/o finding any wildcards, so stick the expansion
433 	     * on the end of the list.
434 	     */
435 	    (void)Lst_AtEnd(expansions, file);
436 	} else {
437 	next:
438 	    free(file);
439 	}
440 	start = cp+1;
441     }
442 }
443 
444 
445 /*-
446  *-----------------------------------------------------------------------
447  * DirExpandInt --
448  *	Internal expand routine. Passes through the directories in the
449  *	path one by one, calling DirMatchFiles for each. NOTE: This still
450  *	doesn't handle patterns in directories...
451  *
452  * Results:
453  *	None.
454  *
455  * Side Effects:
456  *	Things are added to the expansions list.
457  *
458  *-----------------------------------------------------------------------
459  */
460 static void
461 DirExpandInt(word, path, expansions)
462     char    	  *word;    	/* Word to expand */
463     Lst	    	  path;	    	/* Path on which to look */
464     Lst	    	  expansions;	/* Place to store the result */
465 {
466     LstNode 	  ln;	    	/* Current node */
467     Path	  *p;	    	/* Directory in the node */
468 
469     if (Lst_Open(path) == SUCCESS) {
470 	while ((ln = Lst_Next(path)) != NILLNODE) {
471 	    p = (Path *)Lst_Datum(ln);
472 	    DirMatchFiles(word, p, expansions);
473 	}
474 	Lst_Close(path);
475     }
476 }
477 
478 /*-
479  *-----------------------------------------------------------------------
480  * DirPrintWord --
481  *	Print a word in the list of expansions. Callback for Dir_Expand
482  *	when DEBUG(DIR), via Lst_ForEach.
483  *
484  * Results:
485  *	=== 0
486  *
487  * Side Effects:
488  *	The passed word is printed, followed by a space.
489  *
490  *-----------------------------------------------------------------------
491  */
492 static int
493 DirPrintWord(word, dummy)
494     ClientData  word;
495     ClientData  dummy;
496 {
497     printf("%s ", (char *) word);
498 
499     return(dummy ? 0 : 0);
500 }
501 
502 /*-
503  *-----------------------------------------------------------------------
504  * Dir_Expand  --
505  *	Expand the given word into a list of words by globbing it looking
506  *	in the directories on the given search path.
507  *
508  * Results:
509  *	A list of words consisting of the files which exist along the search
510  *	path matching the given pattern.
511  *
512  * Side Effects:
513  *	Directories may be opened. Who knows?
514  *-----------------------------------------------------------------------
515  */
516 void
517 Dir_Expand (word, path, expansions)
518     char    *word;      /* the word to expand */
519     Lst     path;   	/* the list of directories in which to find
520 			 * the resulting files */
521     Lst	    expansions;	/* the list on which to place the results */
522 {
523     char    	  *cp;
524 
525     if (DEBUG(DIR)) {
526 	printf("expanding \"%s\"...", word);
527     }
528 
529     cp = strchr(word, '{');
530     if (cp) {
531 	DirExpandCurly(word, cp, path, expansions);
532     } else {
533 	cp = strchr(word, '/');
534 	if (cp) {
535 	    /*
536 	     * The thing has a directory component -- find the first wildcard
537 	     * in the string.
538 	     */
539 	    for (cp = word; *cp; cp++) {
540 		if (*cp == '?' || *cp == '[' || *cp == '*' || *cp == '{') {
541 		    break;
542 		}
543 	    }
544 	    if (*cp == '{') {
545 		/*
546 		 * This one will be fun.
547 		 */
548 		DirExpandCurly(word, cp, path, expansions);
549 		return;
550 	    } else if (*cp != '\0') {
551 		/*
552 		 * Back up to the start of the component
553 		 */
554 		char  *dirpath;
555 
556 		while (cp > word && *cp != '/') {
557 		    cp--;
558 		}
559 		if (cp != word) {
560 		    char sc;
561 		    /*
562 		     * If the glob isn't in the first component, try and find
563 		     * all the components up to the one with a wildcard.
564 		     */
565 		    sc = cp[1];
566 		    cp[1] = '\0';
567 		    dirpath = Dir_FindFile(word, path);
568 		    cp[1] = sc;
569 		    /*
570 		     * dirpath is null if can't find the leading component
571 		     * XXX: Dir_FindFile won't find internal components.
572 		     * i.e. if the path contains ../Etc/Object and we're
573 		     * looking for Etc, it won't be found. Ah well.
574 		     * Probably not important.
575 		     */
576 		    if (dirpath != (char *)NULL) {
577 			char *dp = &dirpath[strlen(dirpath) - 1];
578 			if (*dp == '/')
579 			    *dp = '\0';
580 			path = Lst_Init(FALSE);
581 			Dir_AddDir(path, dirpath);
582 			DirExpandInt(cp+1, path, expansions);
583 			Lst_Destroy(path, NOFREE);
584 		    }
585 		} else {
586 		    /*
587 		     * Start the search from the local directory
588 		     */
589 		    DirExpandInt(word, path, expansions);
590 		}
591 	    } else {
592 		/*
593 		 * Return the file -- this should never happen.
594 		 */
595 		DirExpandInt(word, path, expansions);
596 	    }
597 	} else {
598 	    /*
599 	     * First the files in dot
600 	     */
601 	    DirMatchFiles(word, dot, expansions);
602 
603 	    /*
604 	     * Then the files in every other directory on the path.
605 	     */
606 	    DirExpandInt(word, path, expansions);
607 	}
608     }
609     if (DEBUG(DIR)) {
610 	Lst_ForEach(expansions, DirPrintWord, (ClientData) 0);
611 	fputc('\n', stdout);
612     }
613 }
614 
615 /*-
616  *-----------------------------------------------------------------------
617  * Dir_FindFile  --
618  *	Find the file with the given name along the given search path.
619  *
620  * Results:
621  *	The path to the file or NULL. This path is guaranteed to be in a
622  *	different part of memory than name and so may be safely free'd.
623  *
624  * Side Effects:
625  *	If the file is found in a directory which is not on the path
626  *	already (either 'name' is absolute or it is a relative path
627  *	[ dir1/.../dirn/file ] which exists below one of the directories
628  *	already on the search path), its directory is added to the end
629  *	of the path on the assumption that there will be more files in
630  *	that directory later on. Sometimes this is true. Sometimes not.
631  *-----------------------------------------------------------------------
632  */
633 char *
634 Dir_FindFile (name, path)
635     char    	  *name;    /* the file to find */
636     Lst           path;	    /* the Lst of directories to search */
637 {
638     register char *p1;	    /* pointer into p->name */
639     register char *p2;	    /* pointer into name */
640     LstNode       ln;	    /* a list element */
641     register char *file;    /* the current filename to check */
642     register Path *p;	    /* current path member */
643     register char *cp;	    /* index of first slash, if any */
644     Boolean	  hasSlash; /* true if 'name' contains a / */
645     struct stat	  stb;	    /* Buffer for stat, if necessary */
646     Hash_Entry	  *entry;   /* Entry for mtimes table */
647 
648     /*
649      * Find the final component of the name and note whether it has a
650      * slash in it (the name, I mean)
651      */
652     cp = strrchr (name, '/');
653     if (cp) {
654 	hasSlash = TRUE;
655 	cp += 1;
656     } else {
657 	hasSlash = FALSE;
658 	cp = name;
659     }
660 
661     if (DEBUG(DIR)) {
662 	printf("Searching for %s...", name);
663     }
664     /*
665      * No matter what, we always look for the file in the current directory
666      * before anywhere else and we *do not* add the ./ to it if it exists.
667      * This is so there are no conflicts between what the user specifies
668      * (fish.c) and what pmake finds (./fish.c).
669      */
670     if ((!hasSlash || (cp - name == 2 && *name == '.')) &&
671 	(Hash_FindEntry (&dot->files, cp) != (Hash_Entry *)NULL)) {
672 	    if (DEBUG(DIR)) {
673 		printf("in '.'\n");
674 	    }
675 	    hits += 1;
676 	    dot->hits += 1;
677 	    return (strdup (name));
678     }
679 
680     if (Lst_Open (path) == FAILURE) {
681 	if (DEBUG(DIR)) {
682 	    printf("couldn't open path, file not found\n");
683 	}
684 	misses += 1;
685 	return ((char *) NULL);
686     }
687 
688     /*
689      * We look through all the directories on the path seeking one which
690      * contains the final component of the given name and whose final
691      * component(s) match the name's initial component(s). If such a beast
692      * is found, we concatenate the directory name and the final component
693      * and return the resulting string. If we don't find any such thing,
694      * we go on to phase two...
695      */
696     while ((ln = Lst_Next (path)) != NILLNODE) {
697 	p = (Path *) Lst_Datum (ln);
698 	if (DEBUG(DIR)) {
699 	    printf("%s...", p->name);
700 	}
701 	if (Hash_FindEntry (&p->files, cp) != (Hash_Entry *)NULL) {
702 	    if (DEBUG(DIR)) {
703 		printf("here...");
704 	    }
705 	    if (hasSlash) {
706 		/*
707 		 * If the name had a slash, its initial components and p's
708 		 * final components must match. This is false if a mismatch
709 		 * is encountered before all of the initial components
710 		 * have been checked (p2 > name at the end of the loop), or
711 		 * we matched only part of one of the components of p
712 		 * along with all the rest of them (*p1 != '/').
713 		 */
714 		p1 = p->name + strlen (p->name) - 1;
715 		p2 = cp - 2;
716 		while (p2 >= name && p1 >= p->name && *p1 == *p2) {
717 		    p1 -= 1; p2 -= 1;
718 		}
719 		if (p2 >= name || (p1 >= p->name && *p1 != '/')) {
720 		    if (DEBUG(DIR)) {
721 			printf("component mismatch -- continuing...");
722 		    }
723 		    continue;
724 		}
725 	    }
726 	    file = str_concat (p->name, cp, STR_ADDSLASH);
727 	    if (DEBUG(DIR)) {
728 		printf("returning %s\n", file);
729 	    }
730 	    Lst_Close (path);
731 	    p->hits += 1;
732 	    hits += 1;
733 	    return (file);
734 	} else if (hasSlash) {
735 	    /*
736 	     * If the file has a leading path component and that component
737 	     * exactly matches the entire name of the current search
738 	     * directory, we assume the file doesn't exist and return NULL.
739 	     */
740 	    for (p1 = p->name, p2 = name; *p1 && *p1 == *p2; p1++, p2++) {
741 		continue;
742 	    }
743 	    if (*p1 == '\0' && p2 == cp - 1) {
744 		if (DEBUG(DIR)) {
745 		    printf("must be here but isn't -- returing NULL\n");
746 		}
747 		Lst_Close (path);
748 		return ((char *) NULL);
749 	    }
750 	}
751     }
752 
753     /*
754      * We didn't find the file on any existing members of the directory.
755      * If the name doesn't contain a slash, that means it doesn't exist.
756      * If it *does* contain a slash, however, there is still hope: it
757      * could be in a subdirectory of one of the members of the search
758      * path. (eg. /usr/include and sys/types.h. The above search would
759      * fail to turn up types.h in /usr/include, but it *is* in
760      * /usr/include/sys/types.h) If we find such a beast, we assume there
761      * will be more (what else can we assume?) and add all but the last
762      * component of the resulting name onto the search path (at the
763      * end). This phase is only performed if the file is *not* absolute.
764      */
765     if (!hasSlash) {
766 	if (DEBUG(DIR)) {
767 	    printf("failed.\n");
768 	}
769 	misses += 1;
770 	return ((char *) NULL);
771     }
772 
773     if (*name != '/') {
774 	Boolean	checkedDot = FALSE;
775 
776 	if (DEBUG(DIR)) {
777 	    printf("failed. Trying subdirectories...");
778 	}
779 	(void) Lst_Open (path);
780 	while ((ln = Lst_Next (path)) != NILLNODE) {
781 	    p = (Path *) Lst_Datum (ln);
782 	    if (p != dot) {
783 		file = str_concat (p->name, name, STR_ADDSLASH);
784 	    } else {
785 		/*
786 		 * Checking in dot -- DON'T put a leading ./ on the thing.
787 		 */
788 		file = strdup(name);
789 		checkedDot = TRUE;
790 	    }
791 	    if (DEBUG(DIR)) {
792 		printf("checking %s...", file);
793 	    }
794 
795 
796 	    if (stat (file, &stb) == 0) {
797 		if (DEBUG(DIR)) {
798 		    printf("got it.\n");
799 		}
800 
801 		Lst_Close (path);
802 
803 		/*
804 		 * We've found another directory to search. We know there's
805 		 * a slash in 'file' because we put one there. We nuke it after
806 		 * finding it and call Dir_AddDir to add this new directory
807 		 * onto the existing search path. Once that's done, we restore
808 		 * the slash and triumphantly return the file name, knowing
809 		 * that should a file in this directory every be referenced
810 		 * again in such a manner, we will find it without having to do
811 		 * numerous numbers of access calls. Hurrah!
812 		 */
813 		cp = strrchr (file, '/');
814 		*cp = '\0';
815 		Dir_AddDir (path, file);
816 		*cp = '/';
817 
818 		/*
819 		 * Save the modification time so if it's needed, we don't have
820 		 * to fetch it again.
821 		 */
822 		if (DEBUG(DIR)) {
823 		    printf("Caching %s for %s\n", Targ_FmtTime(stb.st_mtime),
824 			    file);
825 		}
826 		entry = Hash_CreateEntry(&mtimes, (char *) file,
827 					 (Boolean *)NULL);
828 		Hash_SetValue(entry, (long)stb.st_mtime);
829 		nearmisses += 1;
830 		return (file);
831 	    } else {
832 		free (file);
833 	    }
834 	}
835 
836 	if (DEBUG(DIR)) {
837 	    printf("failed. ");
838 	}
839 	Lst_Close (path);
840 
841 	if (checkedDot) {
842 	    /*
843 	     * Already checked by the given name, since . was in the path,
844 	     * so no point in proceeding...
845 	     */
846 	    if (DEBUG(DIR)) {
847 		printf("Checked . already, returning NULL\n");
848 	    }
849 	    return(NULL);
850 	}
851     }
852 
853     /*
854      * Didn't find it that way, either. Sigh. Phase 3. Add its directory
855      * onto the search path in any case, just in case, then look for the
856      * thing in the hash table. If we find it, grand. We return a new
857      * copy of the name. Otherwise we sadly return a NULL pointer. Sigh.
858      * Note that if the directory holding the file doesn't exist, this will
859      * do an extra search of the final directory on the path. Unless something
860      * weird happens, this search won't succeed and life will be groovy.
861      *
862      * Sigh. We cannot add the directory onto the search path because
863      * of this amusing case:
864      * $(INSTALLDIR)/$(FILE): $(FILE)
865      *
866      * $(FILE) exists in $(INSTALLDIR) but not in the current one.
867      * When searching for $(FILE), we will find it in $(INSTALLDIR)
868      * b/c we added it here. This is not good...
869      */
870 #ifdef notdef
871     cp[-1] = '\0';
872     Dir_AddDir (path, name);
873     cp[-1] = '/';
874 
875     bigmisses += 1;
876     ln = Lst_Last (path);
877     if (ln == NILLNODE) {
878 	return ((char *) NULL);
879     } else {
880 	p = (Path *) Lst_Datum (ln);
881     }
882 
883     if (Hash_FindEntry (&p->files, cp) != (Hash_Entry *)NULL) {
884 	return (strdup (name));
885     } else {
886 	return ((char *) NULL);
887     }
888 #else /* !notdef */
889     if (DEBUG(DIR)) {
890 	printf("Looking for \"%s\"...", name);
891     }
892 
893     bigmisses += 1;
894     entry = Hash_FindEntry(&mtimes, name);
895     if (entry != (Hash_Entry *)NULL) {
896 	if (DEBUG(DIR)) {
897 	    printf("got it (in mtime cache)\n");
898 	}
899 	return(strdup(name));
900     } else if (stat (name, &stb) == 0) {
901 	entry = Hash_CreateEntry(&mtimes, name, (Boolean *)NULL);
902 	if (DEBUG(DIR)) {
903 	    printf("Caching %s for %s\n", Targ_FmtTime(stb.st_mtime),
904 		    name);
905 	}
906 	Hash_SetValue(entry, (long)stb.st_mtime);
907 	return (strdup (name));
908     } else {
909 	if (DEBUG(DIR)) {
910 	    printf("failed. Returning NULL\n");
911 	}
912 	return ((char *)NULL);
913     }
914 #endif /* notdef */
915 }
916 
917 /*-
918  *-----------------------------------------------------------------------
919  * Dir_MTime  --
920  *	Find the modification time of the file described by gn along the
921  *	search path dirSearchPath.
922  *
923  * Results:
924  *	The modification time or 0 if it doesn't exist
925  *
926  * Side Effects:
927  *	The modification time is placed in the node's mtime slot.
928  *	If the node didn't have a path entry before, and Dir_FindFile
929  *	found one for it, the full name is placed in the path slot.
930  *-----------------------------------------------------------------------
931  */
932 int
933 Dir_MTime (gn)
934     GNode         *gn;	      /* the file whose modification time is
935 			       * desired */
936 {
937     char          *fullName;  /* the full pathname of name */
938     struct stat	  stb;	      /* buffer for finding the mod time */
939     Hash_Entry	  *entry;
940 
941     if (gn->type & OP_ARCHV) {
942 	return Arch_MTime (gn);
943     } else if (gn->path == (char *)NULL) {
944 	fullName = Dir_FindFile (gn->name, dirSearchPath);
945     } else {
946 	fullName = gn->path;
947     }
948 
949     if (fullName == (char *)NULL) {
950 	fullName = strdup(gn->name);
951     }
952 
953     entry = Hash_FindEntry(&mtimes, fullName);
954     if (entry != (Hash_Entry *)NULL) {
955 	/*
956 	 * Only do this once -- the second time folks are checking to
957 	 * see if the file was actually updated, so we need to actually go
958 	 * to the file system.
959 	 */
960 	if (DEBUG(DIR)) {
961 	    printf("Using cached time %s for %s\n",
962 		    Targ_FmtTime((time_t)(long)Hash_GetValue(entry)), fullName);
963 	}
964 	stb.st_mtime = (time_t)(long)Hash_GetValue(entry);
965 	Hash_DeleteEntry(&mtimes, entry);
966     } else if (stat (fullName, &stb) < 0) {
967 	if (gn->type & OP_MEMBER) {
968 	    if (fullName != gn->path)
969 		free(fullName);
970 	    return Arch_MemMTime (gn);
971 	} else {
972 	    stb.st_mtime = 0;
973 	}
974     }
975     if (fullName && gn->path == (char *)NULL) {
976 	gn->path = fullName;
977     }
978 
979     gn->mtime = stb.st_mtime;
980     return (gn->mtime);
981 }
982 
983 /*-
984  *-----------------------------------------------------------------------
985  * Dir_AddDir --
986  *	Add the given name to the end of the given path. The order of
987  *	the arguments is backwards so ParseDoDependency can do a
988  *	Lst_ForEach of its list of paths...
989  *
990  * Results:
991  *	none
992  *
993  * Side Effects:
994  *	A structure is added to the list and the directory is
995  *	read and hashed.
996  *-----------------------------------------------------------------------
997  */
998 void
999 Dir_AddDir (path, name)
1000     Lst           path;	      /* the path to which the directory should be
1001 			       * added */
1002     char          *name;      /* the name of the directory to add */
1003 {
1004     LstNode       ln;	      /* node in case Path structure is found */
1005     register Path *p;	      /* pointer to new Path structure */
1006     DIR     	  *d;	      /* for reading directory */
1007     register struct dirent *dp; /* entry in directory */
1008 
1009     ln = Lst_Find (openDirectories, (ClientData)name, DirFindName);
1010     if (ln != NILLNODE) {
1011 	p = (Path *)Lst_Datum (ln);
1012 	if (Lst_Member(path, (ClientData)p) == NILLNODE) {
1013 	    p->refCount += 1;
1014 	    (void)Lst_AtEnd (path, (ClientData)p);
1015 	}
1016     } else {
1017 	if (DEBUG(DIR)) {
1018 	    printf("Caching %s...", name);
1019 	    fflush(stdout);
1020 	}
1021 
1022 	if ((d = opendir (name)) != (DIR *) NULL) {
1023 	    p = (Path *) emalloc (sizeof (Path));
1024 	    p->name = strdup (name);
1025 	    p->hits = 0;
1026 	    p->refCount = 1;
1027 	    Hash_InitTable (&p->files, -1);
1028 
1029 	    /*
1030 	     * Skip the first two entries -- these will *always* be . and ..
1031 	     */
1032 	    (void)readdir(d);
1033 	    (void)readdir(d);
1034 
1035 	    while ((dp = readdir (d)) != (struct dirent *) NULL) {
1036 #ifdef sun
1037 		/*
1038 		 * The sun directory library doesn't check for a 0 inode
1039 		 * (0-inode slots just take up space), so we have to do
1040 		 * it ourselves.
1041 		 */
1042 		if (dp->d_fileno == 0) {
1043 		    continue;
1044 		}
1045 #endif /* sun */
1046 		(void)Hash_CreateEntry(&p->files, dp->d_name, (Boolean *)NULL);
1047 	    }
1048 	    (void) closedir (d);
1049 	    (void)Lst_AtEnd (openDirectories, (ClientData)p);
1050 	    (void)Lst_AtEnd (path, (ClientData)p);
1051 	}
1052 	if (DEBUG(DIR)) {
1053 	    printf("done\n");
1054 	}
1055     }
1056 }
1057 
1058 /*-
1059  *-----------------------------------------------------------------------
1060  * Dir_CopyDir --
1061  *	Callback function for duplicating a search path via Lst_Duplicate.
1062  *	Ups the reference count for the directory.
1063  *
1064  * Results:
1065  *	Returns the Path it was given.
1066  *
1067  * Side Effects:
1068  *	The refCount of the path is incremented.
1069  *
1070  *-----------------------------------------------------------------------
1071  */
1072 ClientData
1073 Dir_CopyDir(p)
1074     ClientData p;
1075 {
1076     ((Path *) p)->refCount += 1;
1077 
1078     return ((ClientData)p);
1079 }
1080 
1081 /*-
1082  *-----------------------------------------------------------------------
1083  * Dir_MakeFlags --
1084  *	Make a string by taking all the directories in the given search
1085  *	path and preceding them by the given flag. Used by the suffix
1086  *	module to create variables for compilers based on suffix search
1087  *	paths.
1088  *
1089  * Results:
1090  *	The string mentioned above. Note that there is no space between
1091  *	the given flag and each directory. The empty string is returned if
1092  *	Things don't go well.
1093  *
1094  * Side Effects:
1095  *	None
1096  *-----------------------------------------------------------------------
1097  */
1098 char *
1099 Dir_MakeFlags (flag, path)
1100     char	  *flag;  /* flag which should precede each directory */
1101     Lst	    	  path;	  /* list of directories */
1102 {
1103     char	  *str;	  /* the string which will be returned */
1104     char	  *tstr;  /* the current directory preceded by 'flag' */
1105     LstNode	  ln;	  /* the node of the current directory */
1106     Path	  *p;	  /* the structure describing the current directory */
1107 
1108     str = strdup ("");
1109 
1110     if (Lst_Open (path) == SUCCESS) {
1111 	while ((ln = Lst_Next (path)) != NILLNODE) {
1112 	    p = (Path *) Lst_Datum (ln);
1113 	    tstr = str_concat (flag, p->name, 0);
1114 	    str = str_concat (str, tstr, STR_ADDSPACE | STR_DOFREE);
1115 	}
1116 	Lst_Close (path);
1117     }
1118 
1119     return (str);
1120 }
1121 
1122 /*-
1123  *-----------------------------------------------------------------------
1124  * Dir_Destroy --
1125  *	Nuke a directory descriptor, if possible. Callback procedure
1126  *	for the suffixes module when destroying a search path.
1127  *
1128  * Results:
1129  *	None.
1130  *
1131  * Side Effects:
1132  *	If no other path references this directory (refCount == 0),
1133  *	the Path and all its data are freed.
1134  *
1135  *-----------------------------------------------------------------------
1136  */
1137 void
1138 Dir_Destroy (pp)
1139     ClientData 	  pp;	    /* The directory descriptor to nuke */
1140 {
1141     Path    	  *p = (Path *) pp;
1142     p->refCount -= 1;
1143 
1144     if (p->refCount == 0) {
1145 	LstNode	ln;
1146 
1147 	ln = Lst_Member (openDirectories, (ClientData)p);
1148 	(void) Lst_Remove (openDirectories, ln);
1149 
1150 	Hash_DeleteTable (&p->files);
1151 	free((Address)p->name);
1152 	free((Address)p);
1153     }
1154 }
1155 
1156 /*-
1157  *-----------------------------------------------------------------------
1158  * Dir_ClearPath --
1159  *	Clear out all elements of the given search path. This is different
1160  *	from destroying the list, notice.
1161  *
1162  * Results:
1163  *	None.
1164  *
1165  * Side Effects:
1166  *	The path is set to the empty list.
1167  *
1168  *-----------------------------------------------------------------------
1169  */
1170 void
1171 Dir_ClearPath(path)
1172     Lst	    path; 	/* Path to clear */
1173 {
1174     Path    *p;
1175     while (!Lst_IsEmpty(path)) {
1176 	p = (Path *)Lst_DeQueue(path);
1177 	Dir_Destroy((ClientData) p);
1178     }
1179 }
1180 
1181 
1182 /*-
1183  *-----------------------------------------------------------------------
1184  * Dir_Concat --
1185  *	Concatenate two paths, adding the second to the end of the first.
1186  *	Makes sure to avoid duplicates.
1187  *
1188  * Results:
1189  *	None
1190  *
1191  * Side Effects:
1192  *	Reference counts for added dirs are upped.
1193  *
1194  *-----------------------------------------------------------------------
1195  */
1196 void
1197 Dir_Concat(path1, path2)
1198     Lst	    path1;  	/* Dest */
1199     Lst	    path2;  	/* Source */
1200 {
1201     LstNode ln;
1202     Path    *p;
1203 
1204     for (ln = Lst_First(path2); ln != NILLNODE; ln = Lst_Succ(ln)) {
1205 	p = (Path *)Lst_Datum(ln);
1206 	if (Lst_Member(path1, (ClientData)p) == NILLNODE) {
1207 	    p->refCount += 1;
1208 	    (void)Lst_AtEnd(path1, (ClientData)p);
1209 	}
1210     }
1211 }
1212 
1213 /********** DEBUG INFO **********/
1214 void
1215 Dir_PrintDirectories()
1216 {
1217     LstNode	ln;
1218     Path	*p;
1219 
1220     printf ("#*** Directory Cache:\n");
1221     printf ("# Stats: %d hits %d misses %d near misses %d losers (%d%%)\n",
1222 	      hits, misses, nearmisses, bigmisses,
1223 	      (hits+bigmisses+nearmisses ?
1224 	       hits * 100 / (hits + bigmisses + nearmisses) : 0));
1225     printf ("# %-20s referenced\thits\n", "directory");
1226     if (Lst_Open (openDirectories) == SUCCESS) {
1227 	while ((ln = Lst_Next (openDirectories)) != NILLNODE) {
1228 	    p = (Path *) Lst_Datum (ln);
1229 	    printf ("# %-20s %10d\t%4d\n", p->name, p->refCount, p->hits);
1230 	}
1231 	Lst_Close (openDirectories);
1232     }
1233 }
1234 
1235 static int DirPrintDir (p, dummy)
1236     ClientData	p;
1237     ClientData	dummy;
1238 {
1239     printf ("%s ", ((Path *) p)->name);
1240     return (dummy ? 0 : 0);
1241 }
1242 
1243 void
1244 Dir_PrintPath (path)
1245     Lst	path;
1246 {
1247     Lst_ForEach (path, DirPrintDir, (ClientData)0);
1248 }
1249