xref: /openbsd/usr.bin/mandoc/mansearch.c (revision d9a51c35)
1 /* $OpenBSD: mansearch.c,v 1.67 2022/12/26 19:16:02 jmc Exp $ */
2 /*
3  * Copyright (c) 2012 Kristaps Dzonsons <kristaps@bsd.lv>
4  * Copyright (c) 2013-2018 Ingo Schwarze <schwarze@openbsd.org>
5  *
6  * Permission to use, copy, modify, and distribute this software for any
7  * purpose with or without fee is hereby granted, provided that the above
8  * copyright notice and this permission notice appear in all copies.
9  *
10  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHORS DISCLAIM ALL WARRANTIES
11  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR
13  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17  */
18 
19 #include <sys/mman.h>
20 #include <sys/types.h>
21 
22 #include <assert.h>
23 #include <err.h>
24 #include <errno.h>
25 #include <fcntl.h>
26 #include <glob.h>
27 #include <limits.h>
28 #include <regex.h>
29 #include <stdio.h>
30 #include <stdint.h>
31 #include <stddef.h>
32 #include <stdlib.h>
33 #include <string.h>
34 #include <unistd.h>
35 
36 #include "mandoc_aux.h"
37 #include "mandoc_ohash.h"
38 #include "manconf.h"
39 #include "mansearch.h"
40 #include "dbm.h"
41 
42 struct	expr {
43 	/* Used for terms: */
44 	struct dbm_match match;   /* Match type and expression. */
45 	uint64_t	 bits;    /* Type mask. */
46 	/* Used for OR and AND groups: */
47 	struct expr	*next;    /* Next child in the parent group. */
48 	struct expr	*child;   /* First child in this group. */
49 	enum { EXPR_TERM, EXPR_OR, EXPR_AND } type;
50 };
51 
52 const char *const mansearch_keynames[KEY_MAX] = {
53 	"arch",	"sec",	"Xr",	"Ar",	"Fa",	"Fl",	"Dv",	"Fn",
54 	"Ic",	"Pa",	"Cm",	"Li",	"Em",	"Cd",	"Va",	"Ft",
55 	"Tn",	"Er",	"Ev",	"Sy",	"Sh",	"In",	"Ss",	"Ox",
56 	"An",	"Mt",	"St",	"Bx",	"At",	"Nx",	"Fx",	"Lk",
57 	"Ms",	"Bsx",	"Dx",	"Rs",	"Vt",	"Lb",	"Nm",	"Nd"
58 };
59 
60 
61 static	struct ohash	*manmerge(struct expr *, struct ohash *);
62 static	struct ohash	*manmerge_term(struct expr *, struct ohash *);
63 static	struct ohash	*manmerge_or(struct expr *, struct ohash *);
64 static	struct ohash	*manmerge_and(struct expr *, struct ohash *);
65 static	char		*buildnames(const struct dbm_page *);
66 static	char		*buildoutput(size_t, struct dbm_page *);
67 static	size_t		 lstlen(const char *, size_t);
68 static	void		 lstcat(char *, size_t *, const char *, const char *);
69 static	int		 lstmatch(const char *, const char *);
70 static	struct expr	*exprcomp(const struct mansearch *,
71 				int, char *[], int *);
72 static	struct expr	*expr_and(const struct mansearch *,
73 				int, char *[], int *);
74 static	struct expr	*exprterm(const struct mansearch *,
75 				int, char *[], int *);
76 static	void		 exprfree(struct expr *);
77 static	int		 manpage_compare(const void *, const void *);
78 
79 
80 int
mansearch(const struct mansearch * search,const struct manpaths * paths,int argc,char * argv[],struct manpage ** res,size_t * sz)81 mansearch(const struct mansearch *search,
82 		const struct manpaths *paths,
83 		int argc, char *argv[],
84 		struct manpage **res, size_t *sz)
85 {
86 	char		 buf[PATH_MAX];
87 	struct dbm_res	*rp;
88 	struct expr	*e;
89 	struct dbm_page	*page;
90 	struct manpage	*mpage;
91 	struct ohash	*htab;
92 	size_t		 cur, i, maxres, outkey;
93 	unsigned int	 slot;
94 	int		 argi, chdir_status, getcwd_status, im;
95 
96 	argi = 0;
97 	if ((e = exprcomp(search, argc, argv, &argi)) == NULL) {
98 		*sz = 0;
99 		return 0;
100 	}
101 
102 	cur = maxres = 0;
103 	if (res != NULL)
104 		*res = NULL;
105 
106 	outkey = KEY_Nd;
107 	if (search->outkey != NULL)
108 		for (im = 0; im < KEY_MAX; im++)
109 			if (0 == strcasecmp(search->outkey,
110 			    mansearch_keynames[im])) {
111 				outkey = im;
112 				break;
113 			}
114 
115 	/*
116 	 * Remember the original working directory, if possible.
117 	 * This will be needed if the second or a later directory
118 	 * is given as a relative path.
119 	 * Do not error out if the current directory is not
120 	 * searchable: Maybe it won't be needed after all.
121 	 */
122 
123 	if (getcwd(buf, PATH_MAX) == NULL) {
124 		getcwd_status = 0;
125 		(void)strlcpy(buf, strerror(errno), sizeof(buf));
126 	} else
127 		getcwd_status = 1;
128 
129 	/*
130 	 * Loop over the directories (containing databases) for us to
131 	 * search.
132 	 * Don't let missing/bad databases/directories phase us.
133 	 * In each, try to open the resident database and, if it opens,
134 	 * scan it for our match expression.
135 	 */
136 
137 	chdir_status = 0;
138 	for (i = 0; i < paths->sz; i++) {
139 		if (chdir_status && paths->paths[i][0] != '/') {
140 			if ( ! getcwd_status) {
141 				warnx("%s: getcwd: %s", paths->paths[i], buf);
142 				continue;
143 			} else if (chdir(buf) == -1) {
144 				warn("%s", buf);
145 				continue;
146 			}
147 		}
148 		if (chdir(paths->paths[i]) == -1) {
149 			warn("%s", paths->paths[i]);
150 			continue;
151 		}
152 		chdir_status = 1;
153 
154 		if (dbm_open(MANDOC_DB) == -1) {
155 			if (errno != ENOENT)
156 				warn("%s/%s", paths->paths[i], MANDOC_DB);
157 			continue;
158 		}
159 
160 		if ((htab = manmerge(e, NULL)) == NULL) {
161 			dbm_close();
162 			continue;
163 		}
164 
165 		for (rp = ohash_first(htab, &slot); rp != NULL;
166 		    rp = ohash_next(htab, &slot)) {
167 			page = dbm_page_get(rp->page);
168 
169 			if (lstmatch(search->sec, page->sect) == 0 ||
170 			    lstmatch(search->arch, page->arch) == 0 ||
171 			    (search->argmode == ARG_NAME &&
172 			     rp->bits <= (int32_t)(NAME_SYN & NAME_MASK)))
173 				continue;
174 
175 			if (res == NULL) {
176 				cur = 1;
177 				break;
178 			}
179 			if (cur + 1 > maxres) {
180 				maxres += 1024;
181 				*res = mandoc_reallocarray(*res,
182 				    maxres, sizeof(**res));
183 			}
184 			mpage = *res + cur;
185 			mandoc_asprintf(&mpage->file, "%s/%s",
186 			    paths->paths[i], page->file + 1);
187 			if (access(chdir_status ? page->file + 1 :
188 			    mpage->file, R_OK) == -1) {
189 				warn("%s", mpage->file);
190 				warnx("outdated mandoc.db contains "
191 				    "bogus %s entry, run makewhatis %s",
192 				    page->file + 1, paths->paths[i]);
193 				free(mpage->file);
194 				free(rp);
195 				continue;
196 			}
197 			mpage->names = buildnames(page);
198 			mpage->output = buildoutput(outkey, page);
199 			mpage->bits = search->firstmatch ? rp->bits : 0;
200 			mpage->ipath = i;
201 			mpage->sec = *page->sect - '0';
202 			if (mpage->sec < 0 || mpage->sec > 9)
203 				mpage->sec = 10;
204 			mpage->form = *page->file;
205 			free(rp);
206 			cur++;
207 		}
208 		ohash_delete(htab);
209 		free(htab);
210 		dbm_close();
211 
212 		/*
213 		 * In man(1) mode, prefer matches in earlier trees
214 		 * over matches in later trees.
215 		 */
216 
217 		if (cur && search->firstmatch)
218 			break;
219 	}
220 	if (res != NULL && cur > 1)
221 		qsort(*res, cur, sizeof(struct manpage), manpage_compare);
222 	if (chdir_status && getcwd_status && chdir(buf) == -1)
223 		warn("%s", buf);
224 	exprfree(e);
225 	*sz = cur;
226 	return res != NULL || cur;
227 }
228 
229 /*
230  * Merge the results for the expression tree rooted at e
231  * into the the result list htab.
232  */
233 static struct ohash *
manmerge(struct expr * e,struct ohash * htab)234 manmerge(struct expr *e, struct ohash *htab)
235 {
236 	switch (e->type) {
237 	case EXPR_TERM:
238 		return manmerge_term(e, htab);
239 	case EXPR_OR:
240 		return manmerge_or(e->child, htab);
241 	case EXPR_AND:
242 		return manmerge_and(e->child, htab);
243 	default:
244 		abort();
245 	}
246 }
247 
248 static struct ohash *
manmerge_term(struct expr * e,struct ohash * htab)249 manmerge_term(struct expr *e, struct ohash *htab)
250 {
251 	struct dbm_res	 res, *rp;
252 	uint64_t	 ib;
253 	unsigned int	 slot;
254 	int		 im;
255 
256 	if (htab == NULL) {
257 		htab = mandoc_malloc(sizeof(*htab));
258 		mandoc_ohash_init(htab, 4, offsetof(struct dbm_res, page));
259 	}
260 
261 	for (im = 0, ib = 1; im < KEY_MAX; im++, ib <<= 1) {
262 		if ((e->bits & ib) == 0)
263 			continue;
264 
265 		switch (ib) {
266 		case TYPE_arch:
267 			dbm_page_byarch(&e->match);
268 			break;
269 		case TYPE_sec:
270 			dbm_page_bysect(&e->match);
271 			break;
272 		case TYPE_Nm:
273 			dbm_page_byname(&e->match);
274 			break;
275 		case TYPE_Nd:
276 			dbm_page_bydesc(&e->match);
277 			break;
278 		default:
279 			dbm_page_bymacro(im - 2, &e->match);
280 			break;
281 		}
282 
283 		/*
284 		 * When hashing for deduplication, use the unique
285 		 * page ID itself instead of a hash function;
286 		 * that is quite efficient.
287 		 */
288 
289 		for (;;) {
290 			res = dbm_page_next();
291 			if (res.page == -1)
292 				break;
293 			slot = ohash_lookup_memory(htab,
294 			    (char *)&res, sizeof(res.page), res.page);
295 			if ((rp = ohash_find(htab, slot)) != NULL) {
296 				rp->bits |= res.bits;
297 				continue;
298 			}
299 			rp = mandoc_malloc(sizeof(*rp));
300 			*rp = res;
301 			ohash_insert(htab, slot, rp);
302 		}
303 	}
304 	return htab;
305 }
306 
307 static struct ohash *
manmerge_or(struct expr * e,struct ohash * htab)308 manmerge_or(struct expr *e, struct ohash *htab)
309 {
310 	while (e != NULL) {
311 		htab = manmerge(e, htab);
312 		e = e->next;
313 	}
314 	return htab;
315 }
316 
317 static struct ohash *
manmerge_and(struct expr * e,struct ohash * htab)318 manmerge_and(struct expr *e, struct ohash *htab)
319 {
320 	struct ohash	*hand, *h1, *h2;
321 	struct dbm_res	*res;
322 	unsigned int	 slot1, slot2;
323 
324 	/* Evaluate the first term of the AND clause. */
325 
326 	hand = manmerge(e, NULL);
327 
328 	while ((e = e->next) != NULL) {
329 
330 		/* Evaluate the next term and prepare for ANDing. */
331 
332 		h2 = manmerge(e, NULL);
333 		if (ohash_entries(h2) < ohash_entries(hand)) {
334 			h1 = h2;
335 			h2 = hand;
336 		} else
337 			h1 = hand;
338 		hand = mandoc_malloc(sizeof(*hand));
339 		mandoc_ohash_init(hand, 4, offsetof(struct dbm_res, page));
340 
341 		/* Keep all pages that are in both result sets. */
342 
343 		for (res = ohash_first(h1, &slot1); res != NULL;
344 		    res = ohash_next(h1, &slot1)) {
345 			if (ohash_find(h2, ohash_lookup_memory(h2,
346 			    (char *)res, sizeof(res->page),
347 			    res->page)) == NULL)
348 				free(res);
349 			else
350 				ohash_insert(hand, ohash_lookup_memory(hand,
351 				    (char *)res, sizeof(res->page),
352 				    res->page), res);
353 		}
354 
355 		/* Discard the merged results. */
356 
357 		for (res = ohash_first(h2, &slot2); res != NULL;
358 		    res = ohash_next(h2, &slot2))
359 			free(res);
360 		ohash_delete(h2);
361 		free(h2);
362 		ohash_delete(h1);
363 		free(h1);
364 	}
365 
366 	/* Merge the result of the AND into htab. */
367 
368 	if (htab == NULL)
369 		return hand;
370 
371 	for (res = ohash_first(hand, &slot1); res != NULL;
372 	    res = ohash_next(hand, &slot1)) {
373 		slot2 = ohash_lookup_memory(htab,
374 		    (char *)res, sizeof(res->page), res->page);
375 		if (ohash_find(htab, slot2) == NULL)
376 			ohash_insert(htab, slot2, res);
377 		else
378 			free(res);
379 	}
380 
381 	/* Discard the merged result. */
382 
383 	ohash_delete(hand);
384 	free(hand);
385 	return htab;
386 }
387 
388 void
mansearch_free(struct manpage * res,size_t sz)389 mansearch_free(struct manpage *res, size_t sz)
390 {
391 	size_t	 i;
392 
393 	for (i = 0; i < sz; i++) {
394 		free(res[i].file);
395 		free(res[i].names);
396 		free(res[i].output);
397 	}
398 	free(res);
399 }
400 
401 static int
manpage_compare(const void * vp1,const void * vp2)402 manpage_compare(const void *vp1, const void *vp2)
403 {
404 	const struct manpage	*mp1, *mp2;
405 	const char		*cp1, *cp2;
406 	size_t			 sz1, sz2;
407 	int			 diff;
408 
409 	mp1 = vp1;
410 	mp2 = vp2;
411 	if ((diff = mp2->bits - mp1->bits) ||
412 	    (diff = mp1->sec - mp2->sec))
413 		return diff;
414 
415 	/* Fall back to alphabetic ordering of names. */
416 	sz1 = strcspn(mp1->names, "(");
417 	sz2 = strcspn(mp2->names, "(");
418 	if (sz1 < sz2)
419 		sz1 = sz2;
420 	if ((diff = strncasecmp(mp1->names, mp2->names, sz1)))
421 		return diff;
422 
423 	/* For identical names and sections, prefer arch-dependent. */
424 	cp1 = strchr(mp1->names + sz1, '/');
425 	cp2 = strchr(mp2->names + sz2, '/');
426 	return cp1 != NULL && cp2 != NULL ? strcasecmp(cp1, cp2) :
427 	    cp1 != NULL ? -1 : cp2 != NULL ? 1 : 0;
428 }
429 
430 static char *
buildnames(const struct dbm_page * page)431 buildnames(const struct dbm_page *page)
432 {
433 	char	*buf;
434 	size_t	 i, sz;
435 
436 	sz = lstlen(page->name, 2) + 1 + lstlen(page->sect, 2) +
437 	    (page->arch == NULL ? 0 : 1 + lstlen(page->arch, 2)) + 2;
438 	buf = mandoc_malloc(sz);
439 	i = 0;
440 	lstcat(buf, &i, page->name, ", ");
441 	buf[i++] = '(';
442 	lstcat(buf, &i, page->sect, ", ");
443 	if (page->arch != NULL) {
444 		buf[i++] = '/';
445 		lstcat(buf, &i, page->arch, ", ");
446 	}
447 	buf[i++] = ')';
448 	buf[i++] = '\0';
449 	assert(i == sz);
450 	return buf;
451 }
452 
453 /*
454  * Count the buffer space needed to print the NUL-terminated
455  * list of NUL-terminated strings, when printing sep separator
456  * characters between strings.
457  */
458 static size_t
lstlen(const char * cp,size_t sep)459 lstlen(const char *cp, size_t sep)
460 {
461 	size_t	 sz;
462 
463 	for (sz = 0; *cp != '\0'; cp++) {
464 
465 		/* Skip names appearing only in the SYNOPSIS. */
466 		if (*cp <= (char)(NAME_SYN & NAME_MASK)) {
467 			while (*cp != '\0')
468 				cp++;
469 			continue;
470 		}
471 
472 		/* Skip name class markers. */
473 		if (*cp < ' ')
474 			cp++;
475 
476 		/* Print a separator before each but the first string. */
477 		if (sz)
478 			sz += sep;
479 
480 		/* Copy one string. */
481 		while (*cp != '\0') {
482 			sz++;
483 			cp++;
484 		}
485 	}
486 	return sz;
487 }
488 
489 /*
490  * Print the NUL-terminated list of NUL-terminated strings
491  * into the buffer, separating strings with sep.
492  */
493 static void
lstcat(char * buf,size_t * i,const char * cp,const char * sep)494 lstcat(char *buf, size_t *i, const char *cp, const char *sep)
495 {
496 	const char	*s;
497 	size_t		 i_start;
498 
499 	for (i_start = *i; *cp != '\0'; cp++) {
500 
501 		/* Skip names appearing only in the SYNOPSIS. */
502 		if (*cp <= (char)(NAME_SYN & NAME_MASK)) {
503 			while (*cp != '\0')
504 				cp++;
505 			continue;
506 		}
507 
508 		/* Skip name class markers. */
509 		if (*cp < ' ')
510 			cp++;
511 
512 		/* Print a separator before each but the first string. */
513 		if (*i > i_start) {
514 			s = sep;
515 			while (*s != '\0')
516 				buf[(*i)++] = *s++;
517 		}
518 
519 		/* Copy one string. */
520 		while (*cp != '\0')
521 			buf[(*i)++] = *cp++;
522 	}
523 
524 }
525 
526 /*
527  * Return 1 if the string *want occurs in any of the strings
528  * in the NUL-terminated string list *have, or 0 otherwise.
529  * If either argument is NULL or empty, assume no filtering
530  * is desired and return 1.
531  */
532 static int
lstmatch(const char * want,const char * have)533 lstmatch(const char *want, const char *have)
534 {
535         if (want == NULL || have == NULL || *have == '\0')
536                 return 1;
537         while (*have != '\0') {
538                 if (strcasestr(have, want) != NULL)
539                         return 1;
540                 have = strchr(have, '\0') + 1;
541         }
542         return 0;
543 }
544 
545 /*
546  * Build a list of values taken by the macro im in the manual page.
547  */
548 static char *
buildoutput(size_t im,struct dbm_page * page)549 buildoutput(size_t im, struct dbm_page *page)
550 {
551 	const char	*oldoutput, *sep, *input;
552 	char		*output, *newoutput, *value;
553 	size_t		 sz, i;
554 
555 	switch (im) {
556 	case KEY_Nd:
557 		return mandoc_strdup(page->desc);
558 	case KEY_Nm:
559 		input = page->name;
560 		break;
561 	case KEY_sec:
562 		input = page->sect;
563 		break;
564 	case KEY_arch:
565 		input = page->arch;
566 		if (input == NULL)
567 			input = "all\0";
568 		break;
569 	default:
570 		input = NULL;
571 		break;
572 	}
573 
574 	if (input != NULL) {
575 		sz = lstlen(input, 3) + 1;
576 		output = mandoc_malloc(sz);
577 		i = 0;
578 		lstcat(output, &i, input, " # ");
579 		output[i++] = '\0';
580 		assert(i == sz);
581 		return output;
582 	}
583 
584 	output = NULL;
585 	dbm_macro_bypage(im - 2, page->addr);
586 	while ((value = dbm_macro_next()) != NULL) {
587 		if (output == NULL) {
588 			oldoutput = "";
589 			sep = "";
590 		} else {
591 			oldoutput = output;
592 			sep = " # ";
593 		}
594 		mandoc_asprintf(&newoutput, "%s%s%s", oldoutput, sep, value);
595 		free(output);
596 		output = newoutput;
597 	}
598 	return output;
599 }
600 
601 /*
602  * Compile a set of string tokens into an expression.
603  * Tokens in "argv" are assumed to be individual expression atoms (e.g.,
604  * "(", "foo=bar", etc.).
605  */
606 static struct expr *
exprcomp(const struct mansearch * search,int argc,char * argv[],int * argi)607 exprcomp(const struct mansearch *search, int argc, char *argv[], int *argi)
608 {
609 	struct expr	*parent, *child;
610 	int		 needterm, nested;
611 
612 	if ((nested = *argi) == argc)
613 		return NULL;
614 	needterm = 1;
615 	parent = child = NULL;
616 	while (*argi < argc) {
617 		if (strcmp(")", argv[*argi]) == 0) {
618 			if (needterm)
619 				warnx("missing term "
620 				    "before closing parenthesis");
621 			needterm = 0;
622 			if (nested)
623 				break;
624 			warnx("ignoring unmatched right parenthesis");
625 			++*argi;
626 			continue;
627 		}
628 		if (strcmp("-o", argv[*argi]) == 0) {
629 			if (needterm) {
630 				if (*argi > 0)
631 					warnx("ignoring -o after %s",
632 					    argv[*argi - 1]);
633 				else
634 					warnx("ignoring initial -o");
635 			}
636 			needterm = 1;
637 			++*argi;
638 			continue;
639 		}
640 		needterm = 0;
641 		if (child == NULL) {
642 			child = expr_and(search, argc, argv, argi);
643 			continue;
644 		}
645 		if (parent == NULL) {
646 			parent = mandoc_calloc(1, sizeof(*parent));
647 			parent->type = EXPR_OR;
648 			parent->next = NULL;
649 			parent->child = child;
650 		}
651 		child->next = expr_and(search, argc, argv, argi);
652 		child = child->next;
653 	}
654 	if (needterm && *argi)
655 		warnx("ignoring trailing %s", argv[*argi - 1]);
656 	return parent == NULL ? child : parent;
657 }
658 
659 static struct expr *
expr_and(const struct mansearch * search,int argc,char * argv[],int * argi)660 expr_and(const struct mansearch *search, int argc, char *argv[], int *argi)
661 {
662 	struct expr	*parent, *child;
663 	int		 needterm;
664 
665 	needterm = 1;
666 	parent = child = NULL;
667 	while (*argi < argc) {
668 		if (strcmp(")", argv[*argi]) == 0) {
669 			if (needterm)
670 				warnx("missing term "
671 				    "before closing parenthesis");
672 			needterm = 0;
673 			break;
674 		}
675 		if (strcmp("-o", argv[*argi]) == 0)
676 			break;
677 		if (strcmp("-a", argv[*argi]) == 0) {
678 			if (needterm) {
679 				if (*argi > 0)
680 					warnx("ignoring -a after %s",
681 					    argv[*argi - 1]);
682 				else
683 					warnx("ignoring initial -a");
684 			}
685 			needterm = 1;
686 			++*argi;
687 			continue;
688 		}
689 		if (needterm == 0)
690 			break;
691 		if (child == NULL) {
692 			child = exprterm(search, argc, argv, argi);
693 			if (child != NULL)
694 				needterm = 0;
695 			continue;
696 		}
697 		needterm = 0;
698 		if (parent == NULL) {
699 			parent = mandoc_calloc(1, sizeof(*parent));
700 			parent->type = EXPR_AND;
701 			parent->next = NULL;
702 			parent->child = child;
703 		}
704 		child->next = exprterm(search, argc, argv, argi);
705 		if (child->next != NULL) {
706 			child = child->next;
707 			needterm = 0;
708 		}
709 	}
710 	if (needterm && *argi)
711 		warnx("ignoring trailing %s", argv[*argi - 1]);
712 	return parent == NULL ? child : parent;
713 }
714 
715 static struct expr *
exprterm(const struct mansearch * search,int argc,char * argv[],int * argi)716 exprterm(const struct mansearch *search, int argc, char *argv[], int *argi)
717 {
718 	char		 errbuf[BUFSIZ];
719 	struct expr	*e;
720 	char		*key, *val;
721 	uint64_t	 iterbit;
722 	int		 cs, i, irc;
723 
724 	if (strcmp("(", argv[*argi]) == 0) {
725 		++*argi;
726 		e = exprcomp(search, argc, argv, argi);
727 		if (*argi < argc) {
728 			assert(strcmp(")", argv[*argi]) == 0);
729 			++*argi;
730 		} else
731 			warnx("unclosed parenthesis");
732 		return e;
733 	}
734 
735 	if (strcmp("-i", argv[*argi]) == 0 && *argi + 1 < argc) {
736 		cs = 0;
737 		++*argi;
738 	} else
739 		cs = 1;
740 
741 	e = mandoc_calloc(1, sizeof(*e));
742 	e->type = EXPR_TERM;
743 	e->bits = 0;
744 	e->next = NULL;
745 	e->child = NULL;
746 
747 	if (search->argmode == ARG_NAME) {
748 		e->bits = TYPE_Nm;
749 		e->match.type = DBM_EXACT;
750 		e->match.str = argv[(*argi)++];
751 		return e;
752 	}
753 
754 	/*
755 	 * Separate macro keys from search string.
756 	 * If needed, request regular expression handling.
757 	 */
758 
759 	if (search->argmode == ARG_WORD) {
760 		e->bits = TYPE_Nm;
761 		e->match.type = DBM_REGEX;
762 		mandoc_asprintf(&val, "[[:<:]]%s[[:>:]]", argv[*argi]);
763 		cs = 0;
764 	} else if ((val = strpbrk(argv[*argi], "=~")) == NULL) {
765 		e->bits = TYPE_Nm | TYPE_Nd;
766 		e->match.type = DBM_REGEX;
767 		val = argv[*argi];
768 		cs = 0;
769 	} else {
770 		if (val == argv[*argi])
771 			e->bits = TYPE_Nm | TYPE_Nd;
772 		if (*val == '=') {
773 			e->match.type = DBM_SUB;
774 			e->match.str = val + 1;
775 		} else
776 			e->match.type = DBM_REGEX;
777 		*val++ = '\0';
778 		if (strstr(argv[*argi], "arch") != NULL)
779 			cs = 0;
780 	}
781 
782 	/* Compile regular expressions. */
783 
784 	if (e->match.type == DBM_REGEX) {
785 		e->match.re = mandoc_malloc(sizeof(*e->match.re));
786 		irc = regcomp(e->match.re, val,
787 		    REG_EXTENDED | REG_NOSUB | (cs ? 0 : REG_ICASE));
788 		if (irc) {
789 			regerror(irc, e->match.re, errbuf, sizeof(errbuf));
790 			warnx("regcomp /%s/: %s", val, errbuf);
791 		}
792 		if (search->argmode == ARG_WORD)
793 			free(val);
794 		if (irc) {
795 			free(e->match.re);
796 			free(e);
797 			++*argi;
798 			return NULL;
799 		}
800 	}
801 
802 	if (e->bits) {
803 		++*argi;
804 		return e;
805 	}
806 
807 	/*
808 	 * Parse out all possible fields.
809 	 * If the field doesn't resolve, bail.
810 	 */
811 
812 	while (NULL != (key = strsep(&argv[*argi], ","))) {
813 		if ('\0' == *key)
814 			continue;
815 		for (i = 0, iterbit = 1; i < KEY_MAX; i++, iterbit <<= 1) {
816 			if (0 == strcasecmp(key, mansearch_keynames[i])) {
817 				e->bits |= iterbit;
818 				break;
819 			}
820 		}
821 		if (i == KEY_MAX) {
822 			if (strcasecmp(key, "any"))
823 				warnx("treating unknown key "
824 				    "\"%s\" as \"any\"", key);
825 			e->bits |= ~0ULL;
826 		}
827 	}
828 
829 	++*argi;
830 	return e;
831 }
832 
833 static void
exprfree(struct expr * e)834 exprfree(struct expr *e)
835 {
836 	if (e->next != NULL)
837 		exprfree(e->next);
838 	if (e->child != NULL)
839 		exprfree(e->child);
840 	free(e);
841 }
842