]>
git.cameronkatri.com Git - mandoc.git/blob - apropos_db.c
1 /* $Id: apropos_db.c,v 1.11 2011/11/23 09:55:28 kristaps Exp $ */
3 * Copyright (c) 2011 Kristaps Dzonsons <kristaps@bsd.lv>
4 * Copyright (c) 2011 Ingo Schwarze <schwarze@openbsd.org>
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.
10 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR 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.
34 #include "apropos_db.h"
38 struct res res
; /* resulting record info */
40 * Maintain a binary tree for checking the uniqueness of `rec'
41 * when adding elements to the results array.
42 * Since the results array is dynamic, use offset in the array
43 * instead of a pointer to the structure.
47 int matched
; /* expression is true */
48 int *matches
; /* partial truth evaluations */
52 int regex
; /* is regex? */
53 int index
; /* index in match array */
54 uint64_t mask
; /* type-mask */
55 int cs
; /* is case-sensitive? */
56 int and; /* is rhs of logical AND? */
57 char *v
; /* search value */
58 regex_t re
; /* compiled re, if regex */
59 struct expr
*next
; /* next in sequence */
69 struct rec
*node
; /* record array for dir tree */
70 int len
; /* length of record array */
73 static const struct type types
[] = {
117 static DB
*btree_open(void);
118 static int btree_read(const DBT
*,
119 const struct mchars
*, char **);
120 static int expreval(const struct expr
*, int *);
121 static void exprexec(const struct expr
*,
122 const char *, uint64_t, struct rec
*);
123 static int exprmark(const struct expr
*,
124 const char *, uint64_t, int *);
125 static struct expr
*exprexpr(int, char *[], int *, int *, size_t *);
126 static struct expr
*exprterm(char *, int);
127 static DB
*index_open(void);
128 static int index_read(const DBT
*, const DBT
*, int,
129 const struct mchars
*, struct rec
*);
130 static void norm_string(const char *,
131 const struct mchars
*, char **);
132 static size_t norm_utf8(unsigned int, char[7]);
133 static void recfree(struct rec
*);
134 static int single_search(struct rectree
*, const struct opts
*,
135 const struct expr
*, size_t terms
,
136 struct mchars
*, int);
139 * Open the keyword mandoc-db database.
147 memset(&info
, 0, sizeof(BTREEINFO
));
150 db
= dbopen(MANDOC_DB
, O_RDONLY
, 0, DB_BTREE
, &info
);
158 * Read a keyword from the database and normalise it.
159 * Return 0 if the database is insane, else 1.
162 btree_read(const DBT
*v
, const struct mchars
*mc
, char **buf
)
165 /* Sanity: are we nil-terminated? */
169 if ('\0' != ((char *)v
->data
)[(int)v
->size
- 1])
172 norm_string((char *)v
->data
, mc
, buf
);
177 * Take a Unicode codepoint and produce its UTF-8 encoding.
178 * This isn't the best way to do this, but it works.
179 * The magic numbers are from the UTF-8 packaging.
180 * They're not as scary as they seem: read the UTF-8 spec for details.
183 norm_utf8(unsigned int cp
, char out
[7])
189 if (cp
<= 0x0000007F) {
192 } else if (cp
<= 0x000007FF) {
194 out
[0] = (cp
>> 6 & 31) | 192;
195 out
[1] = (cp
& 63) | 128;
196 } else if (cp
<= 0x0000FFFF) {
198 out
[0] = (cp
>> 12 & 15) | 224;
199 out
[1] = (cp
>> 6 & 63) | 128;
200 out
[2] = (cp
& 63) | 128;
201 } else if (cp
<= 0x001FFFFF) {
203 out
[0] = (cp
>> 18 & 7) | 240;
204 out
[1] = (cp
>> 12 & 63) | 128;
205 out
[2] = (cp
>> 6 & 63) | 128;
206 out
[3] = (cp
& 63) | 128;
207 } else if (cp
<= 0x03FFFFFF) {
209 out
[0] = (cp
>> 24 & 3) | 248;
210 out
[1] = (cp
>> 18 & 63) | 128;
211 out
[2] = (cp
>> 12 & 63) | 128;
212 out
[3] = (cp
>> 6 & 63) | 128;
213 out
[4] = (cp
& 63) | 128;
214 } else if (cp
<= 0x7FFFFFFF) {
216 out
[0] = (cp
>> 30 & 1) | 252;
217 out
[1] = (cp
>> 24 & 63) | 128;
218 out
[2] = (cp
>> 18 & 63) | 128;
219 out
[3] = (cp
>> 12 & 63) | 128;
220 out
[4] = (cp
>> 6 & 63) | 128;
221 out
[5] = (cp
& 63) | 128;
230 * Normalise strings from the index and database.
231 * These strings are escaped as defined by mandoc_char(7) along with
232 * other goop in mandoc.h (e.g., soft hyphens).
233 * This function normalises these into a nice UTF-8 string.
234 * Returns 0 if the database is fucked.
237 norm_string(const char *val
, const struct mchars
*mc
, char **buf
)
241 const char *seq
, *cpp
;
244 static const char res
[] = { '\\', '\t',
245 ASCII_NBRSP
, ASCII_HYPH
, '\0' };
247 /* Pre-allocate by the length of the input */
249 bsz
= strlen(val
) + 1;
250 *buf
= mandoc_realloc(*buf
, bsz
);
253 while ('\0' != *val
) {
255 * Halt on the first escape sequence.
256 * This also halts on the end of string, in which case
257 * we just copy, fallthrough, and exit the loop.
259 if ((sz
= strcspn(val
, res
)) > 0) {
260 memcpy(&(*buf
)[pos
], val
, sz
);
265 if (ASCII_HYPH
== *val
) {
269 } else if ('\t' == *val
|| ASCII_NBRSP
== *val
) {
273 } else if ('\\' != *val
)
276 /* Read past the slash. */
282 * Parse the escape sequence and see if it's a
283 * predefined character or special character.
286 esc
= mandoc_escape(&val
, &seq
, &len
);
287 if (ESCAPE_ERROR
== esc
)
291 * XXX - this just does UTF-8, but we need to know
292 * beforehand whether we should do text substitution.
296 case (ESCAPE_SPECIAL
):
297 if (0 != (u
= mchars_spec2cp(mc
, seq
, len
)))
305 * If we have a Unicode codepoint, try to convert that
306 * to a UTF-8 byte string.
310 if (0 == (sz
= norm_utf8(u
, utfbuf
)))
313 /* Copy the rendered glyph into the stream. */
318 *buf
= mandoc_realloc(*buf
, bsz
);
320 memcpy(&(*buf
)[pos
], cpp
, sz
);
328 * Open the filename-index mandoc-db database.
329 * Returns NULL if opening failed.
336 db
= dbopen(MANDOC_IDX
, O_RDONLY
, 0, DB_RECNO
, NULL
);
344 * Safely unpack from an index file record into the structure.
345 * Returns 1 if an entry was unpacked, 0 if the database is insane.
348 index_read(const DBT
*key
, const DBT
*val
, int index
,
349 const struct mchars
*mc
, struct rec
*rec
)
354 #define INDEX_BREAD(_dst) \
356 if (NULL == (np = memchr(cp, '\0', left))) \
358 norm_string(cp, mc, &(_dst)); \
359 left -= (np - cp) + 1; \
361 } while (/* CONSTCOND */ 0)
364 cp
= (char *)val
->data
;
366 rec
->res
.rec
= *(recno_t
*)key
->data
;
367 rec
->res
.volume
= index
;
369 INDEX_BREAD(rec
->res
.file
);
370 INDEX_BREAD(rec
->res
.cat
);
371 INDEX_BREAD(rec
->res
.title
);
372 INDEX_BREAD(rec
->res
.arch
);
373 INDEX_BREAD(rec
->res
.desc
);
378 * Search mandocdb databases in paths for expression "expr".
379 * Filter out by "opts".
380 * Call "res" with the results, which may be zero.
381 * Return 0 if there was a database error, else return 1.
384 apropos_search(int pathsz
, char **paths
, const struct opts
*opts
,
385 const struct expr
*expr
, size_t terms
, void *arg
,
386 void (*res
)(struct res
*, size_t, void *))
393 memset(&tree
, 0, sizeof(struct rectree
));
399 * Main loop. Change into the directory containing manpage
400 * databases. Run our expession over each database in the set.
403 for (i
= 0; i
< pathsz
; i
++) {
406 if ( ! single_search(&tree
, opts
, expr
, terms
, mc
, i
))
411 * Count matching files, transfer to a "clean" array, then feed
412 * them to the output handler.
415 for (mlen
= i
= 0; i
< tree
.len
; i
++)
416 if (tree
.node
[i
].matched
)
419 ress
= mandoc_malloc(mlen
* sizeof(struct res
));
421 for (mlen
= i
= 0; i
< tree
.len
; i
++)
422 if (tree
.node
[i
].matched
)
423 memcpy(&ress
[mlen
++], &tree
.node
[i
].res
,
426 (*res
)(ress
, mlen
, arg
);
431 for (i
= 0; i
< tree
.len
; i
++)
432 recfree(&tree
.node
[i
]);
440 single_search(struct rectree
*tree
, const struct opts
*opts
,
441 const struct expr
*expr
, size_t terms
,
442 struct mchars
*mc
, int vol
)
461 memset(&r
, 0, sizeof(struct rec
));
463 if (NULL
== (btree
= btree_open()))
466 if (NULL
== (idx
= index_open())) {
467 (*btree
->close
)(btree
);
471 while (0 == (ch
= (*btree
->seq
)(btree
, &key
, &val
, R_NEXT
))) {
472 if (key
.size
< 2 || sizeof(struct db_val
) != val
.size
)
474 if ( ! btree_read(&key
, mc
, &buf
))
482 * See if this keyword record matches any of the
483 * expressions we have stored.
485 if ( ! exprmark(expr
, buf
, mask
, NULL
))
489 * O(log n) scan for prior records. Since a record
490 * number is unbounded, this has decent performance over
491 * a complex hash function.
494 for (leaf
= root
; leaf
>= 0; )
495 if (rec
> rs
[leaf
].res
.rec
&&
498 else if (rec
< rs
[leaf
].res
.rec
&&
505 * If we find a record, see if it has already evaluated
506 * to true. If it has, great, just keep going. If not,
507 * try to evaluate it now and continue anyway.
510 if (leaf
>= 0 && rs
[leaf
].res
.rec
== rec
) {
511 if (0 == rs
[leaf
].matched
)
512 exprexec(expr
, buf
, mask
, &rs
[leaf
]);
517 * We have a new file to examine.
518 * Extract the manpage's metadata from the index
519 * database, then begin partial evaluation.
523 key
.size
= sizeof(recno_t
);
525 if (0 != (*idx
->get
)(idx
, &key
, &val
, 0))
529 if ( ! index_read(&key
, &val
, vol
, mc
, &r
))
532 /* XXX: this should be elsewhere, I guess? */
534 if (opts
->cat
&& strcasecmp(opts
->cat
, r
.res
.cat
))
536 if (opts
->arch
&& strcasecmp(opts
->arch
, r
.res
.arch
))
539 tree
->node
= rs
= mandoc_realloc
540 (rs
, (tree
->len
+ 1) * sizeof(struct rec
));
542 memcpy(&rs
[tree
->len
], &r
, sizeof(struct rec
));
543 rs
[tree
->len
].matches
=
544 mandoc_calloc(terms
, sizeof(int));
546 exprexec(expr
, buf
, mask
, &rs
[tree
->len
]);
547 /* Append to our tree. */
550 if (rec
> rs
[leaf
].res
.rec
)
551 rs
[leaf
].rhs
= tree
->len
;
553 rs
[leaf
].lhs
= tree
->len
;
557 memset(&r
, 0, sizeof(struct rec
));
561 (*btree
->close
)(btree
);
569 recfree(struct rec
*rec
)
574 free(rec
->res
.title
);
582 exprcomp(int argc
, char *argv
[], size_t *tt
)
590 e
= exprexpr(argc
, argv
, &pos
, &lvl
, tt
);
592 if (0 == lvl
&& pos
>= argc
)
600 * Compile an array of tokens into an expression.
601 * An informal expression grammar is defined in apropos(1).
602 * Return NULL if we fail doing so. All memory will be cleaned up.
603 * Return the root of the expression sequence if alright.
606 exprexpr(int argc
, char *argv
[], int *pos
, int *lvl
, size_t *tt
)
608 struct expr
*e
, *first
, *next
;
613 for ( ; *pos
< argc
; (*pos
)++) {
617 * Close out a subexpression.
620 if (NULL
!= e
&& 0 == strcmp(")", argv
[*pos
])) {
627 * Small note: if we're just starting, don't let "-a"
628 * and "-o" be considered logical operators: they're
629 * just tokens unless pairwise joining, in which case we
630 * record their existence (or assume "OR").
634 if (NULL
!= e
&& 0 == strcmp("-a", argv
[*pos
]))
636 else if (NULL
!= e
&& 0 == strcmp("-o", argv
[*pos
]))
639 if (log
> 0 && ++(*pos
) >= argc
)
643 * Now we parse the term part. This can begin with
644 * "-i", in which case the expression is case
648 if (0 == strcmp("(", argv
[*pos
])) {
651 next
= mandoc_calloc(1, sizeof(struct expr
));
653 next
->subexpr
= exprexpr(argc
, argv
, pos
, lvl
, tt
);
654 if (NULL
== next
->subexpr
) {
658 } else if (0 == strcmp("-i", argv
[*pos
])) {
659 if (++(*pos
) >= argc
)
661 next
= exprterm(argv
[*pos
], 0);
663 next
= exprterm(argv
[*pos
], 1);
668 next
->and = log
== 1;
669 next
->index
= (int)(*tt
)++;
671 /* Append to our chain of expressions. */
689 * Parse a terminal expression with the grammar as defined in
691 * Return NULL if we fail the parse.
694 exprterm(char *buf
, int cs
)
701 memset(&e
, 0, sizeof(struct expr
));
705 /* Choose regex or substring match. */
707 if (NULL
== (e
.v
= strpbrk(buf
, "=~"))) {
711 e
.regex
= '~' == *e
.v
;
715 /* Determine the record types to search for. */
719 while (NULL
!= (key
= strsep(&buf
, ","))) {
721 while (types
[i
].mask
&&
722 strcmp(types
[i
].name
, key
))
724 e
.mask
|= types
[i
].mask
;
728 e
.mask
= TYPE_Nm
| TYPE_Nd
;
731 i
= REG_EXTENDED
| REG_NOSUB
| cs
? 0 : REG_ICASE
;
732 if (regcomp(&e
.re
, e
.v
, i
))
736 e
.v
= mandoc_strdup(e
.v
);
738 p
= mandoc_calloc(1, sizeof(struct expr
));
739 memcpy(p
, &e
, sizeof(struct expr
));
744 exprfree(struct expr
*p
)
750 exprfree(p
->subexpr
);
761 exprmark(const struct expr
*p
, const char *cp
,
762 uint64_t mask
, int *ms
)
765 for ( ; p
; p
= p
->next
) {
767 if (exprmark(p
->subexpr
, cp
, mask
, ms
))
770 } else if ( ! (mask
& p
->mask
))
774 if (regexec(&p
->re
, cp
, 0, NULL
, 0))
777 if (NULL
== strstr(cp
, p
->v
))
780 if (NULL
== strcasestr(cp
, p
->v
))
794 expreval(const struct expr
*p
, int *ms
)
799 * AND has precedence over OR. Analysis is left-right, though
800 * it doesn't matter because there are no side-effects.
801 * Thus, step through pairwise ANDs and accumulate their Boolean
802 * evaluation. If we encounter a single true AND collection or
803 * standalone term, the whole expression is true (by definition
807 for (match
= 0; p
&& ! match
; p
= p
->next
) {
808 /* Evaluate a subexpression, if applicable. */
809 if (p
->subexpr
&& ! ms
[p
->index
])
810 ms
[p
->index
] = expreval(p
->subexpr
, ms
);
812 match
= ms
[p
->index
];
813 for ( ; p
->next
&& p
->next
->and; p
= p
->next
) {
814 /* Evaluate a subexpression, if applicable. */
815 if (p
->next
->subexpr
&& ! ms
[p
->next
->index
])
817 expreval(p
->next
->subexpr
, ms
);
818 match
= match
&& ms
[p
->next
->index
];
826 * First, update the array of terms for which this expression evaluates
828 * Second, logically evaluate all terms over the updated array of truth
830 * If this evaluates to true, mark the expression as satisfied.
833 exprexec(const struct expr
*p
, const char *cp
,
834 uint64_t mask
, struct rec
*r
)
837 assert(0 == r
->matched
);
838 exprmark(p
, cp
, mask
, r
->matches
);
839 r
->matched
= expreval(p
, r
->matches
);