Annotation of src/usr.bin/make/dir.c, Revision 1.5
1.5 ! briggs 1: /* $OpenBSD: dir.c,v 1.4 1996/06/26 05:36:29 deraadt Exp $ */
! 2: /* $NetBSD: dir.c,v 1.11 1996/08/13 16:42:02 christos Exp $ */
1.1 deraadt 3:
4: /*
5: * Copyright (c) 1988, 1989, 1990 The Regents of the University of California.
6: * Copyright (c) 1988, 1989 by Adam de Boor
7: * Copyright (c) 1989 by Berkeley Softworks
8: * All rights reserved.
9: *
10: * This code is derived from software contributed to Berkeley by
11: * Adam de Boor.
12: *
13: * Redistribution and use in source and binary forms, with or without
14: * modification, are permitted provided that the following conditions
15: * are met:
16: * 1. Redistributions of source code must retain the above copyright
17: * notice, this list of conditions and the following disclaimer.
18: * 2. Redistributions in binary form must reproduce the above copyright
19: * notice, this list of conditions and the following disclaimer in the
20: * documentation and/or other materials provided with the distribution.
21: * 3. All advertising materials mentioning features or use of this software
22: * must display the following acknowledgement:
23: * This product includes software developed by the University of
24: * California, Berkeley and its contributors.
25: * 4. Neither the name of the University nor the names of its contributors
26: * may be used to endorse or promote products derived from this software
27: * without specific prior written permission.
28: *
29: * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30: * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31: * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32: * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33: * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34: * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35: * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36: * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37: * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38: * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39: * SUCH DAMAGE.
40: */
41:
42: #ifndef lint
43: #if 0
44: static char sccsid[] = "@(#)dir.c 5.6 (Berkeley) 12/28/90";
45: #else
1.5 ! briggs 46: static char rcsid[] = "$OpenBSD: dir.c,v 1.4 1996/06/26 05:36:29 deraadt Exp $";
1.1 deraadt 47: #endif
48: #endif /* not lint */
49:
50: /*-
51: * dir.c --
52: * Directory searching using wildcards and/or normal names...
53: * Used both for source wildcarding in the Makefile and for finding
54: * implicit sources.
55: *
56: * The interface for this module is:
57: * Dir_Init Initialize the module.
58: *
59: * Dir_End Cleanup the module.
60: *
61: * Dir_HasWildcards Returns TRUE if the name given it needs to
62: * be wildcard-expanded.
63: *
64: * Dir_Expand Given a pattern and a path, return a Lst of names
65: * which match the pattern on the search path.
66: *
67: * Dir_FindFile Searches for a file on a given search path.
68: * If it exists, the entire path is returned.
69: * Otherwise NULL is returned.
70: *
71: * Dir_MTime Return the modification time of a node. The file
72: * is searched for along the default search path.
73: * The path and mtime fields of the node are filled
74: * in.
75: *
76: * Dir_AddDir Add a directory to a search path.
77: *
78: * Dir_MakeFlags Given a search path and a command flag, create
79: * a string with each of the directories in the path
80: * preceded by the command flag and all of them
81: * separated by a space.
82: *
83: * Dir_Destroy Destroy an element of a search path. Frees up all
84: * things that can be freed for the element as long
85: * as the element is no longer referenced by any other
86: * search path.
87: * Dir_ClearPath Resets a search path to the empty list.
88: *
89: * For debugging:
90: * Dir_PrintDirectories Print stats about the directory cache.
91: */
92:
93: #include <stdio.h>
94: #include <sys/types.h>
95: #include <dirent.h>
96: #include <sys/stat.h>
97: #include "make.h"
98: #include "hash.h"
99: #include "dir.h"
100:
101: /*
102: * A search path consists of a Lst of Path structures. A Path structure
103: * has in it the name of the directory and a hash table of all the files
104: * in the directory. This is used to cut down on the number of system
105: * calls necessary to find implicit dependents and their like. Since
106: * these searches are made before any actions are taken, we need not
107: * worry about the directory changing due to creation commands. If this
108: * hampers the style of some makefiles, they must be changed.
109: *
110: * A list of all previously-read directories is kept in the
111: * openDirectories Lst. This list is checked first before a directory
112: * is opened.
113: *
114: * The need for the caching of whole directories is brought about by
115: * the multi-level transformation code in suff.c, which tends to search
116: * for far more files than regular make does. In the initial
117: * implementation, the amount of time spent performing "stat" calls was
118: * truly astronomical. The problem with hashing at the start is,
119: * of course, that pmake doesn't then detect changes to these directories
120: * during the course of the make. Three possibilities suggest themselves:
121: *
122: * 1) just use stat to test for a file's existence. As mentioned
123: * above, this is very inefficient due to the number of checks
124: * engendered by the multi-level transformation code.
125: * 2) use readdir() and company to search the directories, keeping
126: * them open between checks. I have tried this and while it
127: * didn't slow down the process too much, it could severely
128: * affect the amount of parallelism available as each directory
129: * open would take another file descriptor out of play for
130: * handling I/O for another job. Given that it is only recently
131: * that UNIX OS's have taken to allowing more than 20 or 32
132: * file descriptors for a process, this doesn't seem acceptable
133: * to me.
134: * 3) record the mtime of the directory in the Path structure and
135: * verify the directory hasn't changed since the contents were
136: * hashed. This will catch the creation or deletion of files,
137: * but not the updating of files. However, since it is the
138: * creation and deletion that is the problem, this could be
139: * a good thing to do. Unfortunately, if the directory (say ".")
140: * were fairly large and changed fairly frequently, the constant
141: * rehashing could seriously degrade performance. It might be
142: * good in such cases to keep track of the number of rehashes
143: * and if the number goes over a (small) limit, resort to using
144: * stat in its place.
145: *
146: * An additional thing to consider is that pmake is used primarily
147: * to create C programs and until recently pcc-based compilers refused
148: * to allow you to specify where the resulting object file should be
149: * placed. This forced all objects to be created in the current
150: * directory. This isn't meant as a full excuse, just an explanation of
151: * some of the reasons for the caching used here.
152: *
153: * One more note: the location of a target's file is only performed
154: * on the downward traversal of the graph and then only for terminal
155: * nodes in the graph. This could be construed as wrong in some cases,
156: * but prevents inadvertent modification of files when the "installed"
157: * directory for a file is provided in the search path.
158: *
159: * Another data structure maintained by this module is an mtime
160: * cache used when the searching of cached directories fails to find
161: * a file. In the past, Dir_FindFile would simply perform an access()
162: * call in such a case to determine if the file could be found using
163: * just the name given. When this hit, however, all that was gained
164: * was the knowledge that the file existed. Given that an access() is
165: * essentially a stat() without the copyout() call, and that the same
166: * filesystem overhead would have to be incurred in Dir_MTime, it made
167: * sense to replace the access() with a stat() and record the mtime
168: * in a cache for when Dir_MTime was actually called.
169: */
170:
171: Lst dirSearchPath; /* main search path */
172:
173: static Lst openDirectories; /* the list of all open directories */
174:
175: /*
176: * Variables for gathering statistics on the efficiency of the hashing
177: * mechanism.
178: */
179: static int hits, /* Found in directory cache */
180: misses, /* Sad, but not evil misses */
181: nearmisses, /* Found under search path */
182: bigmisses; /* Sought by itself */
183:
184: static Path *dot; /* contents of current directory */
185: static Hash_Table mtimes; /* Results of doing a last-resort stat in
186: * Dir_FindFile -- if we have to go to the
187: * system to find the file, we might as well
188: * have its mtime on record. XXX: If this is done
189: * way early, there's a chance other rules will
190: * have already updated the file, in which case
191: * we'll update it again. Generally, there won't
192: * be two rules to update a single file, so this
193: * should be ok, but... */
194:
195:
196: static int DirFindName __P((ClientData, ClientData));
197: static int DirMatchFiles __P((char *, Path *, Lst));
198: static void DirExpandCurly __P((char *, char *, Lst, Lst));
199: static void DirExpandInt __P((char *, Lst, Lst));
200: static int DirPrintWord __P((ClientData, ClientData));
201: static int DirPrintDir __P((ClientData, ClientData));
202:
203: /*-
204: *-----------------------------------------------------------------------
205: * Dir_Init --
206: * initialize things for this module
207: *
208: * Results:
209: * none
210: *
211: * Side Effects:
212: * some directories may be opened.
213: *-----------------------------------------------------------------------
214: */
215: void
216: Dir_Init ()
217: {
218: dirSearchPath = Lst_Init (FALSE);
219: openDirectories = Lst_Init (FALSE);
220: Hash_InitTable(&mtimes, 0);
221:
222: /*
223: * Since the Path structure is placed on both openDirectories and
224: * the path we give Dir_AddDir (which in this case is openDirectories),
225: * we need to remove "." from openDirectories and what better time to
226: * do it than when we have to fetch the thing anyway?
227: */
228: Dir_AddDir (openDirectories, ".");
229: dot = (Path *) Lst_DeQueue (openDirectories);
230:
231: /*
232: * We always need to have dot around, so we increment its reference count
233: * to make sure it's not destroyed.
234: */
235: dot->refCount += 1;
236: }
237:
238: /*-
239: *-----------------------------------------------------------------------
240: * Dir_End --
241: * cleanup things for this module
242: *
243: * Results:
244: * none
245: *
246: * Side Effects:
247: * none
248: *-----------------------------------------------------------------------
249: */
250: void
251: Dir_End()
252: {
253: dot->refCount -= 1;
254: Dir_Destroy((ClientData) dot);
255: Dir_ClearPath(dirSearchPath);
256: Lst_Destroy(dirSearchPath, NOFREE);
257: Dir_ClearPath(openDirectories);
258: Lst_Destroy(openDirectories, NOFREE);
259: Hash_DeleteTable(&mtimes);
260: }
261:
262: /*-
263: *-----------------------------------------------------------------------
264: * DirFindName --
265: * See if the Path structure describes the same directory as the
266: * given one by comparing their names. Called from Dir_AddDir via
267: * Lst_Find when searching the list of open directories.
268: *
269: * Results:
270: * 0 if it is the same. Non-zero otherwise
271: *
272: * Side Effects:
273: * None
274: *-----------------------------------------------------------------------
275: */
276: static int
277: DirFindName (p, dname)
278: ClientData p; /* Current name */
279: ClientData dname; /* Desired name */
280: {
281: return (strcmp (((Path *)p)->name, (char *) dname));
282: }
283:
284: /*-
285: *-----------------------------------------------------------------------
286: * Dir_HasWildcards --
287: * see if the given name has any wildcard characters in it
288: *
289: * Results:
290: * returns TRUE if the word should be expanded, FALSE otherwise
291: *
292: * Side Effects:
293: * none
294: *-----------------------------------------------------------------------
295: */
296: Boolean
297: Dir_HasWildcards (name)
298: char *name; /* name to check */
299: {
300: register char *cp;
301:
302: for (cp = name; *cp; cp++) {
303: switch(*cp) {
304: case '{':
305: case '[':
306: case '?':
307: case '*':
308: return (TRUE);
309: }
310: }
311: return (FALSE);
312: }
313:
314: /*-
315: *-----------------------------------------------------------------------
316: * DirMatchFiles --
317: * Given a pattern and a Path structure, see if any files
318: * match the pattern and add their names to the 'expansions' list if
319: * any do. This is incomplete -- it doesn't take care of patterns like
320: * src / *src / *.c properly (just *.c on any of the directories), but it
321: * will do for now.
322: *
323: * Results:
324: * Always returns 0
325: *
326: * Side Effects:
327: * File names are added to the expansions lst. The directory will be
328: * fully hashed when this is done.
329: *-----------------------------------------------------------------------
330: */
331: static int
332: DirMatchFiles (pattern, p, expansions)
333: char *pattern; /* Pattern to look for */
334: Path *p; /* Directory to search */
335: Lst expansions; /* Place to store the results */
336: {
337: Hash_Search search; /* Index into the directory's table */
338: Hash_Entry *entry; /* Current entry in the table */
339: Boolean isDot; /* TRUE if the directory being searched is . */
340:
341: isDot = (*p->name == '.' && p->name[1] == '\0');
342:
343: for (entry = Hash_EnumFirst(&p->files, &search);
344: entry != (Hash_Entry *)NULL;
345: entry = Hash_EnumNext(&search))
346: {
347: /*
348: * See if the file matches the given pattern. Note we follow the UNIX
349: * convention that dot files will only be found if the pattern
350: * begins with a dot (note also that as a side effect of the hashing
351: * scheme, .* won't match . or .. since they aren't hashed).
352: */
353: if (Str_Match(entry->name, pattern) &&
354: ((entry->name[0] != '.') ||
355: (pattern[0] == '.')))
356: {
357: (void)Lst_AtEnd(expansions,
1.5 ! briggs 358: (isDot ? estrdup(entry->name) :
1.1 deraadt 359: str_concat(p->name, entry->name,
360: STR_ADDSLASH)));
361: }
362: }
363: return (0);
364: }
365:
366: /*-
367: *-----------------------------------------------------------------------
368: * DirExpandCurly --
369: * Expand curly braces like the C shell. Does this recursively.
370: * Note the special case: if after the piece of the curly brace is
371: * done there are no wildcard characters in the result, the result is
372: * placed on the list WITHOUT CHECKING FOR ITS EXISTENCE.
373: *
374: * Results:
375: * None.
376: *
377: * Side Effects:
378: * The given list is filled with the expansions...
379: *
380: *-----------------------------------------------------------------------
381: */
382: static void
383: DirExpandCurly(word, brace, path, expansions)
384: char *word; /* Entire word to expand */
385: char *brace; /* First curly brace in it */
386: Lst path; /* Search path to use */
387: Lst expansions; /* Place to store the expansions */
388: {
389: char *end; /* Character after the closing brace */
390: char *cp; /* Current position in brace clause */
391: char *start; /* Start of current piece of brace clause */
392: int bracelevel; /* Number of braces we've seen. If we see a
393: * right brace when this is 0, we've hit the
394: * end of the clause. */
395: char *file; /* Current expansion */
396: int otherLen; /* The length of the other pieces of the
397: * expansion (chars before and after the
398: * clause in 'word') */
399: char *cp2; /* Pointer for checking for wildcards in
400: * expansion before calling Dir_Expand */
401:
402: start = brace+1;
403:
404: /*
405: * Find the end of the brace clause first, being wary of nested brace
406: * clauses.
407: */
408: for (end = start, bracelevel = 0; *end != '\0'; end++) {
409: if (*end == '{') {
410: bracelevel++;
411: } else if ((*end == '}') && (bracelevel-- == 0)) {
412: break;
413: }
414: }
415: if (*end == '\0') {
416: Error("Unterminated {} clause \"%s\"", start);
417: return;
418: } else {
419: end++;
420: }
421: otherLen = brace - word + strlen(end);
422:
423: for (cp = start; cp < end; cp++) {
424: /*
425: * Find the end of this piece of the clause.
426: */
427: bracelevel = 0;
428: while (*cp != ',') {
429: if (*cp == '{') {
430: bracelevel++;
431: } else if ((*cp == '}') && (bracelevel-- <= 0)) {
432: break;
433: }
434: cp++;
435: }
436: /*
437: * Allocate room for the combination and install the three pieces.
438: */
439: file = emalloc(otherLen + cp - start + 1);
440: if (brace != word) {
441: strncpy(file, word, brace-word);
442: }
443: if (cp != start) {
444: strncpy(&file[brace-word], start, cp-start);
445: }
446: strcpy(&file[(brace-word)+(cp-start)], end);
447:
448: /*
449: * See if the result has any wildcards in it. If we find one, call
450: * Dir_Expand right away, telling it to place the result on our list
451: * of expansions.
452: */
453: for (cp2 = file; *cp2 != '\0'; cp2++) {
454: switch(*cp2) {
455: case '*':
456: case '?':
457: case '{':
458: case '[':
459: Dir_Expand(file, path, expansions);
460: goto next;
461: }
462: }
463: if (*cp2 == '\0') {
464: /*
465: * Hit the end w/o finding any wildcards, so stick the expansion
466: * on the end of the list.
467: */
468: (void)Lst_AtEnd(expansions, file);
469: } else {
470: next:
471: free(file);
472: }
473: start = cp+1;
474: }
475: }
476:
477:
478: /*-
479: *-----------------------------------------------------------------------
480: * DirExpandInt --
481: * Internal expand routine. Passes through the directories in the
482: * path one by one, calling DirMatchFiles for each. NOTE: This still
483: * doesn't handle patterns in directories...
484: *
485: * Results:
486: * None.
487: *
488: * Side Effects:
489: * Things are added to the expansions list.
490: *
491: *-----------------------------------------------------------------------
492: */
493: static void
494: DirExpandInt(word, path, expansions)
495: char *word; /* Word to expand */
496: Lst path; /* Path on which to look */
497: Lst expansions; /* Place to store the result */
498: {
499: LstNode ln; /* Current node */
500: Path *p; /* Directory in the node */
501:
502: if (Lst_Open(path) == SUCCESS) {
503: while ((ln = Lst_Next(path)) != NILLNODE) {
504: p = (Path *)Lst_Datum(ln);
505: DirMatchFiles(word, p, expansions);
506: }
507: Lst_Close(path);
508: }
509: }
510:
511: /*-
512: *-----------------------------------------------------------------------
513: * DirPrintWord --
514: * Print a word in the list of expansions. Callback for Dir_Expand
515: * when DEBUG(DIR), via Lst_ForEach.
516: *
517: * Results:
518: * === 0
519: *
520: * Side Effects:
521: * The passed word is printed, followed by a space.
522: *
523: *-----------------------------------------------------------------------
524: */
525: static int
526: DirPrintWord(word, dummy)
527: ClientData word;
528: ClientData dummy;
529: {
530: printf("%s ", (char *) word);
531:
532: return(dummy ? 0 : 0);
533: }
534:
535: /*-
536: *-----------------------------------------------------------------------
537: * Dir_Expand --
538: * Expand the given word into a list of words by globbing it looking
539: * in the directories on the given search path.
540: *
541: * Results:
542: * A list of words consisting of the files which exist along the search
543: * path matching the given pattern.
544: *
545: * Side Effects:
546: * Directories may be opened. Who knows?
547: *-----------------------------------------------------------------------
548: */
549: void
550: Dir_Expand (word, path, expansions)
551: char *word; /* the word to expand */
552: Lst path; /* the list of directories in which to find
553: * the resulting files */
554: Lst expansions; /* the list on which to place the results */
555: {
556: char *cp;
557:
558: if (DEBUG(DIR)) {
559: printf("expanding \"%s\"...", word);
560: }
561:
562: cp = strchr(word, '{');
563: if (cp) {
564: DirExpandCurly(word, cp, path, expansions);
565: } else {
566: cp = strchr(word, '/');
567: if (cp) {
568: /*
569: * The thing has a directory component -- find the first wildcard
570: * in the string.
571: */
572: for (cp = word; *cp; cp++) {
573: if (*cp == '?' || *cp == '[' || *cp == '*' || *cp == '{') {
574: break;
575: }
576: }
577: if (*cp == '{') {
578: /*
579: * This one will be fun.
580: */
581: DirExpandCurly(word, cp, path, expansions);
582: return;
583: } else if (*cp != '\0') {
584: /*
585: * Back up to the start of the component
586: */
587: char *dirpath;
588:
589: while (cp > word && *cp != '/') {
590: cp--;
591: }
592: if (cp != word) {
593: char sc;
594: /*
595: * If the glob isn't in the first component, try and find
596: * all the components up to the one with a wildcard.
597: */
598: sc = cp[1];
599: cp[1] = '\0';
600: dirpath = Dir_FindFile(word, path);
601: cp[1] = sc;
602: /*
603: * dirpath is null if can't find the leading component
604: * XXX: Dir_FindFile won't find internal components.
605: * i.e. if the path contains ../Etc/Object and we're
606: * looking for Etc, it won't be found. Ah well.
607: * Probably not important.
608: */
609: if (dirpath != (char *)NULL) {
610: char *dp = &dirpath[strlen(dirpath) - 1];
611: if (*dp == '/')
612: *dp = '\0';
613: path = Lst_Init(FALSE);
614: Dir_AddDir(path, dirpath);
615: DirExpandInt(cp+1, path, expansions);
616: Lst_Destroy(path, NOFREE);
617: }
618: } else {
619: /*
620: * Start the search from the local directory
621: */
622: DirExpandInt(word, path, expansions);
623: }
624: } else {
625: /*
626: * Return the file -- this should never happen.
627: */
628: DirExpandInt(word, path, expansions);
629: }
630: } else {
631: /*
632: * First the files in dot
633: */
634: DirMatchFiles(word, dot, expansions);
635:
636: /*
637: * Then the files in every other directory on the path.
638: */
639: DirExpandInt(word, path, expansions);
640: }
641: }
642: if (DEBUG(DIR)) {
643: Lst_ForEach(expansions, DirPrintWord, (ClientData) 0);
644: fputc('\n', stdout);
645: }
646: }
647:
648: /*-
649: *-----------------------------------------------------------------------
650: * Dir_FindFile --
651: * Find the file with the given name along the given search path.
652: *
653: * Results:
654: * The path to the file or NULL. This path is guaranteed to be in a
655: * different part of memory than name and so may be safely free'd.
656: *
657: * Side Effects:
658: * If the file is found in a directory which is not on the path
659: * already (either 'name' is absolute or it is a relative path
660: * [ dir1/.../dirn/file ] which exists below one of the directories
661: * already on the search path), its directory is added to the end
662: * of the path on the assumption that there will be more files in
663: * that directory later on. Sometimes this is true. Sometimes not.
664: *-----------------------------------------------------------------------
665: */
666: char *
667: Dir_FindFile (name, path)
668: char *name; /* the file to find */
669: Lst path; /* the Lst of directories to search */
670: {
671: register char *p1; /* pointer into p->name */
672: register char *p2; /* pointer into name */
673: LstNode ln; /* a list element */
674: register char *file; /* the current filename to check */
675: register Path *p; /* current path member */
676: register char *cp; /* index of first slash, if any */
677: Boolean hasSlash; /* true if 'name' contains a / */
678: struct stat stb; /* Buffer for stat, if necessary */
679: Hash_Entry *entry; /* Entry for mtimes table */
680:
681: /*
682: * Find the final component of the name and note whether it has a
683: * slash in it (the name, I mean)
684: */
685: cp = strrchr (name, '/');
686: if (cp) {
687: hasSlash = TRUE;
688: cp += 1;
689: } else {
690: hasSlash = FALSE;
691: cp = name;
692: }
693:
694: if (DEBUG(DIR)) {
695: printf("Searching for %s...", name);
696: }
697: /*
698: * No matter what, we always look for the file in the current directory
699: * before anywhere else and we *do not* add the ./ to it if it exists.
700: * This is so there are no conflicts between what the user specifies
701: * (fish.c) and what pmake finds (./fish.c).
702: */
703: if ((!hasSlash || (cp - name == 2 && *name == '.')) &&
704: (Hash_FindEntry (&dot->files, cp) != (Hash_Entry *)NULL)) {
705: if (DEBUG(DIR)) {
706: printf("in '.'\n");
707: }
708: hits += 1;
709: dot->hits += 1;
1.5 ! briggs 710: return (estrdup (name));
1.1 deraadt 711: }
712:
713: if (Lst_Open (path) == FAILURE) {
714: if (DEBUG(DIR)) {
715: printf("couldn't open path, file not found\n");
716: }
717: misses += 1;
718: return ((char *) NULL);
719: }
720:
721: /*
722: * We look through all the directories on the path seeking one which
723: * contains the final component of the given name and whose final
724: * component(s) match the name's initial component(s). If such a beast
725: * is found, we concatenate the directory name and the final component
726: * and return the resulting string. If we don't find any such thing,
727: * we go on to phase two...
728: */
729: while ((ln = Lst_Next (path)) != NILLNODE) {
730: p = (Path *) Lst_Datum (ln);
731: if (DEBUG(DIR)) {
732: printf("%s...", p->name);
733: }
734: if (Hash_FindEntry (&p->files, cp) != (Hash_Entry *)NULL) {
735: if (DEBUG(DIR)) {
736: printf("here...");
737: }
738: if (hasSlash) {
739: /*
740: * If the name had a slash, its initial components and p's
741: * final components must match. This is false if a mismatch
742: * is encountered before all of the initial components
743: * have been checked (p2 > name at the end of the loop), or
744: * we matched only part of one of the components of p
745: * along with all the rest of them (*p1 != '/').
746: */
747: p1 = p->name + strlen (p->name) - 1;
748: p2 = cp - 2;
749: while (p2 >= name && p1 >= p->name && *p1 == *p2) {
750: p1 -= 1; p2 -= 1;
751: }
752: if (p2 >= name || (p1 >= p->name && *p1 != '/')) {
753: if (DEBUG(DIR)) {
754: printf("component mismatch -- continuing...");
755: }
756: continue;
757: }
758: }
759: file = str_concat (p->name, cp, STR_ADDSLASH);
760: if (DEBUG(DIR)) {
761: printf("returning %s\n", file);
762: }
763: Lst_Close (path);
764: p->hits += 1;
765: hits += 1;
766: return (file);
767: } else if (hasSlash) {
768: /*
769: * If the file has a leading path component and that component
770: * exactly matches the entire name of the current search
771: * directory, we assume the file doesn't exist and return NULL.
772: */
773: for (p1 = p->name, p2 = name; *p1 && *p1 == *p2; p1++, p2++) {
774: continue;
775: }
776: if (*p1 == '\0' && p2 == cp - 1) {
777: if (DEBUG(DIR)) {
778: printf("must be here but isn't -- returing NULL\n");
779: }
780: Lst_Close (path);
781: return ((char *) NULL);
782: }
783: }
784: }
785:
786: /*
787: * We didn't find the file on any existing members of the directory.
788: * If the name doesn't contain a slash, that means it doesn't exist.
789: * If it *does* contain a slash, however, there is still hope: it
790: * could be in a subdirectory of one of the members of the search
791: * path. (eg. /usr/include and sys/types.h. The above search would
792: * fail to turn up types.h in /usr/include, but it *is* in
793: * /usr/include/sys/types.h) If we find such a beast, we assume there
794: * will be more (what else can we assume?) and add all but the last
795: * component of the resulting name onto the search path (at the
796: * end). This phase is only performed if the file is *not* absolute.
797: */
798: if (!hasSlash) {
799: if (DEBUG(DIR)) {
800: printf("failed.\n");
801: }
802: misses += 1;
803: return ((char *) NULL);
804: }
805:
806: if (*name != '/') {
807: Boolean checkedDot = FALSE;
808:
809: if (DEBUG(DIR)) {
810: printf("failed. Trying subdirectories...");
811: }
812: (void) Lst_Open (path);
813: while ((ln = Lst_Next (path)) != NILLNODE) {
814: p = (Path *) Lst_Datum (ln);
815: if (p != dot) {
816: file = str_concat (p->name, name, STR_ADDSLASH);
817: } else {
818: /*
819: * Checking in dot -- DON'T put a leading ./ on the thing.
820: */
1.5 ! briggs 821: file = estrdup(name);
1.1 deraadt 822: checkedDot = TRUE;
823: }
824: if (DEBUG(DIR)) {
825: printf("checking %s...", file);
826: }
827:
828:
829: if (stat (file, &stb) == 0) {
830: if (DEBUG(DIR)) {
831: printf("got it.\n");
832: }
833:
834: Lst_Close (path);
835:
836: /*
837: * We've found another directory to search. We know there's
838: * a slash in 'file' because we put one there. We nuke it after
839: * finding it and call Dir_AddDir to add this new directory
840: * onto the existing search path. Once that's done, we restore
841: * the slash and triumphantly return the file name, knowing
842: * that should a file in this directory every be referenced
843: * again in such a manner, we will find it without having to do
844: * numerous numbers of access calls. Hurrah!
845: */
846: cp = strrchr (file, '/');
847: *cp = '\0';
848: Dir_AddDir (path, file);
849: *cp = '/';
850:
851: /*
852: * Save the modification time so if it's needed, we don't have
853: * to fetch it again.
854: */
855: if (DEBUG(DIR)) {
856: printf("Caching %s for %s\n", Targ_FmtTime(stb.st_mtime),
857: file);
858: }
859: entry = Hash_CreateEntry(&mtimes, (char *) file,
860: (Boolean *)NULL);
861: Hash_SetValue(entry, (long)stb.st_mtime);
862: nearmisses += 1;
863: return (file);
864: } else {
865: free (file);
866: }
867: }
868:
869: if (DEBUG(DIR)) {
870: printf("failed. ");
871: }
872: Lst_Close (path);
873:
874: if (checkedDot) {
875: /*
876: * Already checked by the given name, since . was in the path,
877: * so no point in proceeding...
878: */
879: if (DEBUG(DIR)) {
880: printf("Checked . already, returning NULL\n");
881: }
882: return(NULL);
883: }
884: }
885:
886: /*
887: * Didn't find it that way, either. Sigh. Phase 3. Add its directory
888: * onto the search path in any case, just in case, then look for the
889: * thing in the hash table. If we find it, grand. We return a new
890: * copy of the name. Otherwise we sadly return a NULL pointer. Sigh.
891: * Note that if the directory holding the file doesn't exist, this will
892: * do an extra search of the final directory on the path. Unless something
893: * weird happens, this search won't succeed and life will be groovy.
894: *
895: * Sigh. We cannot add the directory onto the search path because
896: * of this amusing case:
897: * $(INSTALLDIR)/$(FILE): $(FILE)
898: *
899: * $(FILE) exists in $(INSTALLDIR) but not in the current one.
900: * When searching for $(FILE), we will find it in $(INSTALLDIR)
901: * b/c we added it here. This is not good...
902: */
903: #ifdef notdef
904: cp[-1] = '\0';
905: Dir_AddDir (path, name);
906: cp[-1] = '/';
907:
908: bigmisses += 1;
909: ln = Lst_Last (path);
910: if (ln == NILLNODE) {
911: return ((char *) NULL);
912: } else {
913: p = (Path *) Lst_Datum (ln);
914: }
915:
916: if (Hash_FindEntry (&p->files, cp) != (Hash_Entry *)NULL) {
1.5 ! briggs 917: return (estrdup (name));
1.1 deraadt 918: } else {
919: return ((char *) NULL);
920: }
921: #else /* !notdef */
922: if (DEBUG(DIR)) {
923: printf("Looking for \"%s\"...", name);
924: }
925:
926: bigmisses += 1;
927: entry = Hash_FindEntry(&mtimes, name);
928: if (entry != (Hash_Entry *)NULL) {
929: if (DEBUG(DIR)) {
930: printf("got it (in mtime cache)\n");
931: }
1.5 ! briggs 932: return(estrdup(name));
1.1 deraadt 933: } else if (stat (name, &stb) == 0) {
934: entry = Hash_CreateEntry(&mtimes, name, (Boolean *)NULL);
935: if (DEBUG(DIR)) {
936: printf("Caching %s for %s\n", Targ_FmtTime(stb.st_mtime),
937: name);
938: }
939: Hash_SetValue(entry, (long)stb.st_mtime);
1.5 ! briggs 940: return (estrdup (name));
1.1 deraadt 941: } else {
942: if (DEBUG(DIR)) {
943: printf("failed. Returning NULL\n");
944: }
945: return ((char *)NULL);
946: }
947: #endif /* notdef */
948: }
949:
950: /*-
951: *-----------------------------------------------------------------------
952: * Dir_MTime --
953: * Find the modification time of the file described by gn along the
954: * search path dirSearchPath.
955: *
956: * Results:
957: * The modification time or 0 if it doesn't exist
958: *
959: * Side Effects:
960: * The modification time is placed in the node's mtime slot.
961: * If the node didn't have a path entry before, and Dir_FindFile
962: * found one for it, the full name is placed in the path slot.
963: *-----------------------------------------------------------------------
964: */
965: int
966: Dir_MTime (gn)
967: GNode *gn; /* the file whose modification time is
968: * desired */
969: {
970: char *fullName; /* the full pathname of name */
971: struct stat stb; /* buffer for finding the mod time */
972: Hash_Entry *entry;
973:
974: if (gn->type & OP_ARCHV) {
975: return Arch_MTime (gn);
976: } else if (gn->path == (char *)NULL) {
977: fullName = Dir_FindFile (gn->name, dirSearchPath);
978: } else {
979: fullName = gn->path;
980: }
981:
982: if (fullName == (char *)NULL) {
1.5 ! briggs 983: fullName = estrdup(gn->name);
1.1 deraadt 984: }
985:
986: entry = Hash_FindEntry(&mtimes, fullName);
987: if (entry != (Hash_Entry *)NULL) {
988: /*
989: * Only do this once -- the second time folks are checking to
990: * see if the file was actually updated, so we need to actually go
991: * to the file system.
992: */
993: if (DEBUG(DIR)) {
994: printf("Using cached time %s for %s\n",
995: Targ_FmtTime((time_t)(long)Hash_GetValue(entry)), fullName);
996: }
997: stb.st_mtime = (time_t)(long)Hash_GetValue(entry);
998: Hash_DeleteEntry(&mtimes, entry);
999: } else if (stat (fullName, &stb) < 0) {
1000: if (gn->type & OP_MEMBER) {
1001: if (fullName != gn->path)
1002: free(fullName);
1003: return Arch_MemMTime (gn);
1004: } else {
1005: stb.st_mtime = 0;
1006: }
1007: }
1008: if (fullName && gn->path == (char *)NULL) {
1009: gn->path = fullName;
1010: }
1011:
1012: gn->mtime = stb.st_mtime;
1013: return (gn->mtime);
1014: }
1015:
1016: /*-
1017: *-----------------------------------------------------------------------
1018: * Dir_AddDir --
1019: * Add the given name to the end of the given path. The order of
1020: * the arguments is backwards so ParseDoDependency can do a
1021: * Lst_ForEach of its list of paths...
1022: *
1023: * Results:
1024: * none
1025: *
1026: * Side Effects:
1027: * A structure is added to the list and the directory is
1028: * read and hashed.
1029: *-----------------------------------------------------------------------
1030: */
1031: void
1032: Dir_AddDir (path, name)
1033: Lst path; /* the path to which the directory should be
1034: * added */
1035: char *name; /* the name of the directory to add */
1036: {
1037: LstNode ln; /* node in case Path structure is found */
1038: register Path *p; /* pointer to new Path structure */
1039: DIR *d; /* for reading directory */
1040: register struct dirent *dp; /* entry in directory */
1041:
1042: ln = Lst_Find (openDirectories, (ClientData)name, DirFindName);
1043: if (ln != NILLNODE) {
1044: p = (Path *)Lst_Datum (ln);
1045: if (Lst_Member(path, (ClientData)p) == NILLNODE) {
1046: p->refCount += 1;
1047: (void)Lst_AtEnd (path, (ClientData)p);
1048: }
1049: } else {
1050: if (DEBUG(DIR)) {
1051: printf("Caching %s...", name);
1052: fflush(stdout);
1053: }
1054:
1055: if ((d = opendir (name)) != (DIR *) NULL) {
1056: p = (Path *) emalloc (sizeof (Path));
1.5 ! briggs 1057: p->name = estrdup (name);
1.1 deraadt 1058: p->hits = 0;
1059: p->refCount = 1;
1060: Hash_InitTable (&p->files, -1);
1061:
1062: /*
1063: * Skip the first two entries -- these will *always* be . and ..
1064: */
1065: (void)readdir(d);
1066: (void)readdir(d);
1067:
1068: while ((dp = readdir (d)) != (struct dirent *) NULL) {
1.3 niklas 1069: #if defined(sun) && defined(d_ino) /* d_ino is a sunos4 #define for d_fileno */
1.1 deraadt 1070: /*
1071: * The sun directory library doesn't check for a 0 inode
1072: * (0-inode slots just take up space), so we have to do
1073: * it ourselves.
1074: */
1075: if (dp->d_fileno == 0) {
1076: continue;
1077: }
1.3 niklas 1078: #endif /* sun && d_ino */
1.1 deraadt 1079: (void)Hash_CreateEntry(&p->files, dp->d_name, (Boolean *)NULL);
1080: }
1081: (void) closedir (d);
1082: (void)Lst_AtEnd (openDirectories, (ClientData)p);
1083: (void)Lst_AtEnd (path, (ClientData)p);
1084: }
1085: if (DEBUG(DIR)) {
1086: printf("done\n");
1087: }
1088: }
1089: }
1090:
1091: /*-
1092: *-----------------------------------------------------------------------
1093: * Dir_CopyDir --
1094: * Callback function for duplicating a search path via Lst_Duplicate.
1095: * Ups the reference count for the directory.
1096: *
1097: * Results:
1098: * Returns the Path it was given.
1099: *
1100: * Side Effects:
1101: * The refCount of the path is incremented.
1102: *
1103: *-----------------------------------------------------------------------
1104: */
1105: ClientData
1106: Dir_CopyDir(p)
1107: ClientData p;
1108: {
1109: ((Path *) p)->refCount += 1;
1110:
1111: return ((ClientData)p);
1112: }
1113:
1114: /*-
1115: *-----------------------------------------------------------------------
1116: * Dir_MakeFlags --
1117: * Make a string by taking all the directories in the given search
1118: * path and preceding them by the given flag. Used by the suffix
1119: * module to create variables for compilers based on suffix search
1120: * paths.
1121: *
1122: * Results:
1123: * The string mentioned above. Note that there is no space between
1124: * the given flag and each directory. The empty string is returned if
1125: * Things don't go well.
1126: *
1127: * Side Effects:
1128: * None
1129: *-----------------------------------------------------------------------
1130: */
1131: char *
1132: Dir_MakeFlags (flag, path)
1133: char *flag; /* flag which should precede each directory */
1134: Lst path; /* list of directories */
1135: {
1136: char *str; /* the string which will be returned */
1137: char *tstr; /* the current directory preceded by 'flag' */
1138: LstNode ln; /* the node of the current directory */
1139: Path *p; /* the structure describing the current directory */
1140:
1.5 ! briggs 1141: str = estrdup ("");
1.1 deraadt 1142:
1143: if (Lst_Open (path) == SUCCESS) {
1144: while ((ln = Lst_Next (path)) != NILLNODE) {
1145: p = (Path *) Lst_Datum (ln);
1146: tstr = str_concat (flag, p->name, 0);
1147: str = str_concat (str, tstr, STR_ADDSPACE | STR_DOFREE);
1148: }
1149: Lst_Close (path);
1150: }
1151:
1152: return (str);
1153: }
1154:
1155: /*-
1156: *-----------------------------------------------------------------------
1157: * Dir_Destroy --
1158: * Nuke a directory descriptor, if possible. Callback procedure
1159: * for the suffixes module when destroying a search path.
1160: *
1161: * Results:
1162: * None.
1163: *
1164: * Side Effects:
1165: * If no other path references this directory (refCount == 0),
1166: * the Path and all its data are freed.
1167: *
1168: *-----------------------------------------------------------------------
1169: */
1170: void
1171: Dir_Destroy (pp)
1172: ClientData pp; /* The directory descriptor to nuke */
1173: {
1174: Path *p = (Path *) pp;
1175: p->refCount -= 1;
1176:
1177: if (p->refCount == 0) {
1178: LstNode ln;
1179:
1180: ln = Lst_Member (openDirectories, (ClientData)p);
1181: (void) Lst_Remove (openDirectories, ln);
1182:
1183: Hash_DeleteTable (&p->files);
1184: free((Address)p->name);
1185: free((Address)p);
1186: }
1187: }
1188:
1189: /*-
1190: *-----------------------------------------------------------------------
1191: * Dir_ClearPath --
1192: * Clear out all elements of the given search path. This is different
1193: * from destroying the list, notice.
1194: *
1195: * Results:
1196: * None.
1197: *
1198: * Side Effects:
1199: * The path is set to the empty list.
1200: *
1201: *-----------------------------------------------------------------------
1202: */
1203: void
1204: Dir_ClearPath(path)
1205: Lst path; /* Path to clear */
1206: {
1207: Path *p;
1208: while (!Lst_IsEmpty(path)) {
1209: p = (Path *)Lst_DeQueue(path);
1210: Dir_Destroy((ClientData) p);
1211: }
1212: }
1213:
1214:
1215: /*-
1216: *-----------------------------------------------------------------------
1217: * Dir_Concat --
1218: * Concatenate two paths, adding the second to the end of the first.
1219: * Makes sure to avoid duplicates.
1220: *
1221: * Results:
1222: * None
1223: *
1224: * Side Effects:
1225: * Reference counts for added dirs are upped.
1226: *
1227: *-----------------------------------------------------------------------
1228: */
1229: void
1230: Dir_Concat(path1, path2)
1231: Lst path1; /* Dest */
1232: Lst path2; /* Source */
1233: {
1234: LstNode ln;
1235: Path *p;
1236:
1237: for (ln = Lst_First(path2); ln != NILLNODE; ln = Lst_Succ(ln)) {
1238: p = (Path *)Lst_Datum(ln);
1239: if (Lst_Member(path1, (ClientData)p) == NILLNODE) {
1240: p->refCount += 1;
1241: (void)Lst_AtEnd(path1, (ClientData)p);
1242: }
1243: }
1244: }
1245:
1246: /********** DEBUG INFO **********/
1247: void
1248: Dir_PrintDirectories()
1249: {
1250: LstNode ln;
1251: Path *p;
1252:
1253: printf ("#*** Directory Cache:\n");
1254: printf ("# Stats: %d hits %d misses %d near misses %d losers (%d%%)\n",
1255: hits, misses, nearmisses, bigmisses,
1256: (hits+bigmisses+nearmisses ?
1257: hits * 100 / (hits + bigmisses + nearmisses) : 0));
1258: printf ("# %-20s referenced\thits\n", "directory");
1259: if (Lst_Open (openDirectories) == SUCCESS) {
1260: while ((ln = Lst_Next (openDirectories)) != NILLNODE) {
1261: p = (Path *) Lst_Datum (ln);
1262: printf ("# %-20s %10d\t%4d\n", p->name, p->refCount, p->hits);
1263: }
1264: Lst_Close (openDirectories);
1265: }
1266: }
1267:
1268: static int DirPrintDir (p, dummy)
1269: ClientData p;
1270: ClientData dummy;
1271: {
1272: printf ("%s ", ((Path *) p)->name);
1273: return (dummy ? 0 : 0);
1274: }
1275:
1276: void
1277: Dir_PrintPath (path)
1278: Lst path;
1279: {
1280: Lst_ForEach (path, DirPrintDir, (ClientData)0);
1281: }
![[BACK]](/icons/back.gif){kind=icon}
Return to dir.c CVS log ![[TXT]](/icons/text.gif){kind=icon}
![[DIR]](/icons/dir.gif){kind=icon}
Up to [local] / src / usr.bin / make