Full logical patch: and, or, subexpressions.

Full logical patch: and, or, subexpressions.

[Kristaps Dzonsons]

[-- Attachment #1: Type: text/plain, Size: 1214 bytes --]

Hi,

Enclosed is a polished patch for full logical subexpressions, including 
manual updates for the syntax.  This allows the following (from the manual):

      An expression consists of search terms joined by logical
      operators -a (and) and -o (or).  The -a operator has precedence
      over -o and both are evaluated left-to-right.

      ( expr )
              True if the subexpression expr is true.

      expr1 -a expr2
              True if both expr1 and expr2 are true (logical "and").

      expr1 [-o] expr2
              True if expr1 and/or expr2 evaluate to true (logical
              "or").

      term    True if term is satisfied.  This has syntax
              [key[,key]*(=~)]?val, where operand key is an mdoc(7)
              macro to query and val is its value.  Operator =
              evaluates a substring, while ~ evaluates a regular
              expression.

      -i term
              Same as term, but term is evaluated case-insensitively.

This patch also hides away backend "struct rec" components (lhs, rhs, 
matches) to apropos_db.c and adds lots of in-line documentation.

Oks?  If so, I'll check it in and get to work merging Ingo's excellent work.

Thanks,

Kristaps

[-- Attachment #2: patch.txt --]
[-- Type: text/plain, Size: 21105 bytes --]

Index: apropos.1
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/apropos.1,v
retrieving revision 1.3
diff -u -r1.3 apropos.1
--- apropos.1	9 Nov 2011 10:53:48 -0000	1.3
+++ apropos.1	17 Nov 2011 14:57:26 -0000
@@ -22,15 +22,17 @@
 .Nd search the manual page database
 .Sh SYNOPSIS
 .Nm
-.Op Fl I
 .Op Fl S Ar arch
 .Op Fl s Ar section
-.Op Fl t Ar types
-.Ar pattern
+.Ar expression...
 .Sh DESCRIPTION
 The
 .Nm
-utility queries the manual page database.
+utility queries a manual page database generated by
+.Xr mandocdb 8 , 
+evaluating on
+.Ar expression
+for each file in the database.
 Its arguments are as follows:
 .Bl -tag -width Ds
 .It Fl S Ar arch
@@ -40,52 +42,76 @@
 See
 .Xr man 1
 for a listing of manual sections.
-.It Fl I
-Case-insensitive pattern matching.
-.It Fl t Ar types
-Consider only types of keyword match, where
-.Ar types
-is a comma-separated list consisting of the following:
-.Ar name ,
-manual names;
-.Ar func ,
-function names;
-.Ar utility ,
-utility names;
-.Ar incl ,
-include files;
-.Ar var ,
-variable names;
-.Ar stand ,
-standards ;
-.Ar auth ,
-authors;
-.Ar conf ,
-configuration strings;
-.Ar desc ,
-descriptive text;
-.Ar xref ,
-cross-references.
-.Ar path ,
-file pathnames;
-.Ar env ,
-environment variables; or
-.Ar err ,
-error codes.
-Specifying
-.Ar all
-will search for all types.
-.It Ar key
-The search key.
+.El
+.Pp
+An
+.Ar expression
+consists of search terms joined by logical operators
+.Fl a
+.Pq and
+and
+.Fl o
+.Pq or .
+The
+.Fl a
+operator has precedence over
+.Fl o
+and both are evaluated left-to-right.
+.Pp
+.Bl -tag -width Ds
+.It \&( Ar expr No \&)
+True if the subexpression
+.Ar expr
+is true.
+.It Ar expr1 Fl a Ar expr2
+True if both
+.Ar expr1
+and
+.Ar expr2
+are true (logical
+.Qq and ) .
+.It Ar expr1 Oo Fl o Oc Ar expr2
+True if
+.Ar expr1
+and/or
+.Ar expr2
+evaluate to true (logical
+.Qq or ) .
+.It Ar term
+True if
+.Ar term
+is satisfied.
+This has syntax
+.Li [key[,key]*(=~)]?val ,
+where operand
+.Li key
+is an
+.Xr mdoc 7
+macro to query and
+.Li val
+is its value.
+Operator
+.Li \&=
+evaluates a substring, while
+.Li \&~
+evaluates a regular expression.
+.It Fl i Ar term
+Same as
+.Ar term ,
+but
+.Ar term
+is evaluated case-insensitively.
 .El
 .Pp
 By default,
 .Nm
-searches for pattern matches of
-.Ar pattern
-in manual names and descriptions
-and displays results sorted by manual title.
-Output is formatted as
+parses terms as case-sensitive regular expressions
+.Pq the Li \&~ operator
+over manual names and descriptions
+.Pq the Li \&Nm No and Li \&Nd No macros .
+Multiple terms imply pairwise
+.Fl o .
+Results are sorted by manual title, with output formatted as
 .Pp
 .D1 title(sec) \- description
 .Pp
@@ -119,16 +145,34 @@
 .Ex -std
 .Sh EXAMPLES
 Search for
+.Qq mdoc
+within the manual name and description:
+.Pp
+.Dl $ apropos mdoc
+.Pp
+Two variants of searching for
+.Qq mdoc ,
+.Qq roff ,
+or
+.Qq man
+within manual names and descriptions:
+.Pp
+.Dl $ apropos mdoc roff man
+.Dl $ apropos mdoc \-o roff \-o man
+.Pp
+Search for
 .Qq optind
-as a variable name in the library category:
+and
+.Qq optarg
+as variable names in the library category:
 .Pp
-.Dl $ apropos \-tvar \-s 3 optind
+.Dl $ apropos \-s 3 \(dqVa~^optind$\(dq -a \(dqVa~^optarg$\(dq
 .Pp
-Search for all manuals referencing the term
+Search for all manuals referencing
 .Qq POSIX
 in any letter case:
 .Pp
-.Dl $ apropos \-tall \-I posix
+.Dl $ apropos \-\- \-i posix
 .\" .Sh DIAGNOSTICS
 .\" For sections 1, 4, 6, 7, & 8 only.
 .\" .Sh ERRORS
Index: apropos.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/apropos.c,v
retrieving revision 1.13
diff -u -r1.13 apropos.c
--- apropos.c	14 Nov 2011 10:07:06 -0000	1.13
+++ apropos.c	17 Nov 2011 14:57:26 -0000
@@ -29,7 +29,7 @@
 #include "mandoc.h"
 
 static	int	 cmp(const void *, const void *);
-static	void	 list(struct rec *, size_t, void *);
+static	void	 list(struct res *, size_t, void *);
 static	void	 usage(void);
 
 static	char	*progname;
@@ -38,8 +38,7 @@
 main(int argc, char *argv[])
 {
 	int		 ch;
-	size_t		 sz;
-	char		*buf;
+	size_t		 terms;
 	struct opts	 opts;
 	struct expr	*e;
 	extern int	 optind;
@@ -72,32 +71,11 @@
 	if (0 == argc) 
 		return(EXIT_SUCCESS);
 
-	/* 
-	 * Collapse expressions into a single string.  
-	 * First count up the contained strings, adding a space at the
-	 * end of each (plus nil-terminator).  Then merge.
-	 */
-
-	for (sz = 0, ch = 0; ch < argc; ch++)
-		sz += strlen(argv[ch]) + 1;
-
-	buf = mandoc_malloc(++sz);
-
-	for (*buf = '\0', ch = 0; ch < argc; ch++) {
-		strlcat(buf, argv[ch], sz);
-		strlcat(buf, " ", sz);
-	}
-
-	buf[sz - 2] = '\0';
-
-	if (NULL == (e = exprcomp(buf))) {
+	if (NULL == (e = exprcomp(argc, argv, &terms))) {
 		fprintf(stderr, "Bad expression\n");
-		free(buf);
 		return(EXIT_FAILURE);
 	}
 
-	free(buf);
-
 	/*
 	 * Configure databases.
 	 * The keyword database is a btree that allows for duplicate
@@ -105,18 +83,20 @@
 	 * The index database is a recno.
 	 */
 
-	apropos_search(&opts, e, NULL, list);
+	ch = apropos_search(&opts, e, terms, NULL, list);
 	exprfree(e);
-	return(EXIT_SUCCESS);
+	if (0 == ch)
+		fprintf(stderr, "%s: Database error\n", progname);
+	return(ch ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
 /* ARGSUSED */
 static void
-list(struct rec *res, size_t sz, void *arg)
+list(struct res *res, size_t sz, void *arg)
 {
 	int		 i;
 
-	qsort(res, sz, sizeof(struct rec), cmp);
+	qsort(res, sz, sizeof(struct res), cmp);
 
 	for (i = 0; i < (int)sz; i++)
 		printf("%s(%s%s%s) - %s\n", res[i].title, 
@@ -130,18 +110,14 @@
 cmp(const void *p1, const void *p2)
 {
 
-	return(strcmp(((const struct rec *)p1)->title,
-				((const struct rec *)p2)->title));
+	return(strcmp(((const struct res *)p1)->title,
+				((const struct res *)p2)->title));
 }
 
 static void
 usage(void)
 {
 
-	fprintf(stderr, "usage: %s "
-			"[-I] "
-			"[-S arch] "
-			"[-s section] "
-			"EXPR\n", 
-			progname);
+	fprintf(stderr, "usage: %s [-S arch] [-s section] "
+			"expression...\n", progname);
 }
Index: apropos_db.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/apropos_db.c,v
retrieving revision 1.4
diff -u -r1.4 apropos_db.c
--- apropos_db.c	14 Nov 2011 10:07:06 -0000	1.4
+++ apropos_db.c	17 Nov 2011 14:57:26 -0000
@@ -32,11 +32,30 @@
 #include "apropos_db.h"
 #include "mandoc.h"
 
+struct	rec {
+	struct res	 res; /* resulting record info */
+	/*
+	 * Maintain a binary tree for checking the uniqueness of `rec'
+	 * when adding elements to the results array.
+	 * Since the results array is dynamic, use offset in the array
+	 * instead of a pointer to the structure.
+	 */
+	int		 lhs;
+	int		 rhs;
+	int		 matched; /* expression is true */
+	int		*matches; /* partial truth evaluations */
+};
+
 struct	expr {
-	int		 regex;
-	int	 	 mask;
-	char		*v;
-	regex_t	 	 re;
+	int		 regex; /* is regex? */
+	int		 index; /* index in match array */
+	int	 	 mask; /* type-mask */
+	int		 cs; /* is case-sensitive? */
+	int		 and; /* is rhs of logical AND? */
+	char		*v; /* search value */
+	regex_t	 	 re; /* compiled re, if regex */
+	struct expr	*next; /* next in sequence */
+	struct expr	*subexpr;
 };
 
 struct	type {
@@ -64,14 +83,22 @@
 };
 
 static	DB	*btree_open(void);
-static	int	 btree_read(const DBT *, const struct mchars *, char **);
-static	int	 exprexec(const struct expr *, char *, int);
+static	int	 btree_read(const DBT *, 
+			const struct mchars *, char **);
+static	int	 expreval(const struct expr *, int *);
+static	void	 exprexec(const struct expr *, 
+			const char *, int, struct rec *);
+static	int	 exprmark(const struct expr *, 
+			const char *, int, int *);
+static	struct expr *exprexpr(int, char *[], int *, int *, size_t *);
+static	struct expr *exprterm(char *, int);
 static	DB	*index_open(void);
 static	int	 index_read(const DBT *, const DBT *, 
 			const struct mchars *, struct rec *);
 static	void	 norm_string(const char *,
 			const struct mchars *, char **);
 static	size_t	 norm_utf8(unsigned int, char[7]);
+static	void	 recfree(struct rec *);
 
 /*
  * Open the keyword mandoc-db database.
@@ -300,13 +327,13 @@
 	left = val->size;
 	cp = (char *)val->data;
 
-	rec->rec = *(recno_t *)key->data;
+	rec->res.rec = *(recno_t *)key->data;
 
-	INDEX_BREAD(rec->file);
-	INDEX_BREAD(rec->cat);
-	INDEX_BREAD(rec->title);
-	INDEX_BREAD(rec->arch);
-	INDEX_BREAD(rec->desc);
+	INDEX_BREAD(rec->res.file);
+	INDEX_BREAD(rec->res.cat);
+	INDEX_BREAD(rec->res.title);
+	INDEX_BREAD(rec->res.arch);
+	INDEX_BREAD(rec->res.desc);
 	return(1);
 }
 
@@ -314,38 +341,37 @@
  * Search the mandocdb database for the expression "expr".
  * Filter out by "opts".
  * Call "res" with the results, which may be zero.
+ * Return 0 if there was a database error, else return 1.
  */
-void
+int
 apropos_search(const struct opts *opts, const struct expr *expr,
-		void *arg, void (*res)(struct rec *, size_t, void *))
+		size_t terms, void *arg, 
+		void (*res)(struct res *, size_t, void *))
 {
-	int		 i, len, root, leaf;
+	int		 i, rsz, root, leaf, mask, mlen, rc, ch;
 	DBT		 key, val;
 	DB		*btree, *idx;
 	struct mchars	*mc;
-	int		 ch;
 	char		*buf;
 	recno_t		 rec;
-	struct rec	*recs;
-	struct rec	 srec;
+	struct rec	*rs;
+	struct res	*ress;
+	struct rec	 r;
 
+	rc	= 0;
 	root	= -1;
 	leaf	= -1;
 	btree	= NULL;
 	idx	= NULL;
 	mc	= NULL;
 	buf	= NULL;
-	recs	= NULL;
-	len	= 0;
-
-	memset(&srec, 0, sizeof(struct rec));
+	rs	= NULL;
+	rsz	= 0;
 
-	/* XXX: error out with bad regexp? */
+	memset(&r, 0, sizeof(struct rec));
 
 	mc = mchars_alloc();
 
-	/* XXX: return fact that we've errored? */
-
 	if (NULL == (btree = btree_open())) 
 		goto out;
 	if (NULL == (idx = index_open())) 
@@ -362,7 +388,13 @@
 		if ( ! btree_read(&key, mc, &buf))
 			break;
 
-		if ( ! exprexec(expr, buf, *(int *)val.data))
+		mask = *(int *)val.data;
+
+		/*
+		 * See if this keyword record matches any of the
+		 * expressions we have stored.
+		 */
+		if ( ! exprmark(expr, buf, mask, NULL))
 			continue;
 
 		memcpy(&rec, val.data + 4, sizeof(recno_t));
@@ -374,19 +406,31 @@
 		 */
 
 		for (leaf = root; leaf >= 0; )
-			if (rec > recs[leaf].rec && recs[leaf].rhs >= 0)
-				leaf = recs[leaf].rhs;
-			else if (rec < recs[leaf].rec && recs[leaf].lhs >= 0)
-				leaf = recs[leaf].lhs;
+			if (rec > rs[leaf].res.rec && 
+					rs[leaf].rhs >= 0)
+				leaf = rs[leaf].rhs;
+			else if (rec < rs[leaf].res.rec && 
+					rs[leaf].lhs >= 0)
+				leaf = rs[leaf].lhs;
 			else 
 				break;
 
-		if (leaf >= 0 && recs[leaf].rec == rec)
+		/*
+		 * If we find a record, see if it has already evaluated
+		 * to true.  If it has, great, just keep going.  If not,
+		 * try to evaluate it now and continue anyway.
+		 */
+
+		if (leaf >= 0 && rs[leaf].res.rec == rec) {
+			if (0 == rs[leaf].matched)
+				exprexec(expr, buf, mask, &rs[leaf]);
 			continue;
+		}
 
 		/*
-		 * Now we actually extract the manpage's metadata from
-		 * the index database.
+		 * We have a new file to examine.
+		 * Extract the manpage's metadata from the index
+		 * database, then begin partial evaluation.
 		 */
 
 		key.data = &rec;
@@ -395,52 +439,62 @@
 		if (0 != (*idx->get)(idx, &key, &val, 0))
 			break;
 
-		srec.lhs = srec.rhs = -1;
-		if ( ! index_read(&key, &val, mc, &srec))
+		r.lhs = r.rhs = -1;
+		if ( ! index_read(&key, &val, mc, &r))
 			break;
 
-		if (opts->cat && strcasecmp(opts->cat, srec.cat))
+		/* XXX: this should be elsewhere, I guess? */
+
+		if (opts->cat && strcasecmp(opts->cat, r.res.cat))
 			continue;
-		if (opts->arch && strcasecmp(opts->arch, srec.arch))
+		if (opts->arch && strcasecmp(opts->arch, r.res.arch))
 			continue;
 
-		recs = mandoc_realloc
-			(recs, (len + 1) * sizeof(struct rec));
+		rs = mandoc_realloc
+			(rs, (rsz + 1) * sizeof(struct rec));
 
-		memcpy(&recs[len], &srec, sizeof(struct rec));
+		memcpy(&rs[rsz], &r, sizeof(struct rec));
+		rs[rsz].matches = mandoc_calloc(terms, sizeof(int));
 
+		exprexec(expr, buf, mask, &rs[rsz]); 
 		/* Append to our tree. */
 
 		if (leaf >= 0) {
-			if (rec > recs[leaf].rec)
-				recs[leaf].rhs = len;
+			if (rec > rs[leaf].res.rec)
+				rs[leaf].rhs = rsz;
 			else
-				recs[leaf].lhs = len;
+				rs[leaf].lhs = rsz;
 		} else
-			root = len;
+			root = rsz;
 		
-		memset(&srec, 0, sizeof(struct rec));
-		len++;
+		memset(&r, 0, sizeof(struct rec));
+		rsz++;
 	}
+	
+	/*
+	 * If we haven't encountered any database errors, then construct
+	 * an array of results and push them to the caller.
+	 */
 
-	if (1 == ch)
-		(*res)(recs, len, arg);
+	if (1 == ch) {
+		for (mlen = i = 0; i < rsz; i++)
+			if (rs[i].matched)
+				mlen++;
+		ress = mandoc_malloc(mlen * sizeof(struct res));
+		for (mlen = i = 0; i < rsz; i++)
+			if (rs[i].matched)
+				memcpy(&ress[mlen++], &rs[i].res, 
+						sizeof(struct res));
+		(*res)(ress, mlen, arg);
+		free(ress);
+		rc = 1;
+	}
 
-	/* XXX: else?  corrupt database error? */
 out:
-	for (i = 0; i < len; i++) {
-		free(recs[i].file);
-		free(recs[i].cat);
-		free(recs[i].title);
-		free(recs[i].arch);
-		free(recs[i].desc);
-	}
+	for (i = 0; i < rsz; i++)
+		recfree(&rs[i]);
 
-	free(srec.file);
-	free(srec.cat);
-	free(srec.title);
-	free(srec.arch);
-	free(srec.desc);
+	recfree(&r);
 
 	if (mc)
 		mchars_free(mc);
@@ -450,23 +504,148 @@
 		(*idx->close)(idx);
 
 	free(buf);
-	free(recs);
+	free(rs);
+	return(rc);
+}
+
+static void
+recfree(struct rec *rec)
+{
+
+	free(rec->res.file);
+	free(rec->res.cat);
+	free(rec->res.title);
+	free(rec->res.arch);
+	free(rec->res.desc);
+
+	free(rec->matches);
 }
 
 struct expr *
-exprcomp(char *buf)
+exprcomp(int argc, char *argv[], size_t *tt)
+{
+	int		 pos, lvl;
+	struct expr	*e;
+
+	pos = lvl = 0;
+	*tt = 0;
+
+	e = exprexpr(argc, argv, &pos, &lvl, tt);
+
+	if (0 == lvl && pos >= argc)
+		return(e);
+
+	exprfree(e);
+	return(NULL);
+}
+
+/*
+ * Compile an array of tokens into an expression.
+ * An informal expression grammar is defined in apropos(1).
+ * Return NULL if we fail doing so.  All memory will be cleaned up.
+ * Return the root of the expression sequence if alright.
+ */
+static struct expr *
+exprexpr(int argc, char *argv[], int *pos, int *lvl, size_t *tt)
+{
+	struct expr	*e, *first, *next;
+	int		 log;
+
+	first = next = NULL;
+
+	for ( ; *pos < argc; (*pos)++) {
+		e = next;
+
+		/*
+		 * Close out a subexpression.
+		 */
+
+		if (NULL != e && 0 == strcmp(")", argv[*pos])) {
+			if (--(*lvl) < 0)
+				goto err;
+			break;
+		}
+
+		/*
+		 * Small note: if we're just starting, don't let "-a"
+		 * and "-o" be considered logical operators: they're
+		 * just tokens unless pairwise joining, in which case we
+		 * record their existence (or assume "OR").
+		 */
+		log = 0;
+
+		if (NULL != e && 0 == strcmp("-a", argv[*pos]))
+			log = 1;			
+		else if (NULL != e && 0 == strcmp("-o", argv[*pos]))
+			log = 2;
+
+		if (log > 0 && ++(*pos) >= argc)
+			goto err;
+
+		/*
+		 * Now we parse the term part.  This can begin with
+		 * "-i", in which case the expression is case
+		 * insensitive.
+		 */
+
+		if (0 == strcmp("(", argv[*pos])) {
+			++(*pos);
+			++(*lvl);
+			next = mandoc_calloc(1, sizeof(struct expr));
+			next->cs = 1;
+			next->subexpr = exprexpr(argc, argv, pos, lvl, tt);
+			if (NULL == next->subexpr) {
+				free(next);
+				next = NULL;
+			}
+		} else if (0 == strcmp("-i", argv[*pos])) {
+			if (++(*pos) >= argc)
+				goto err;
+			next = exprterm(argv[*pos], 0);
+		} else
+			next = exprterm(argv[*pos], 1);
+
+		if (NULL == next)
+			goto err;
+
+		next->and = log == 1;
+		next->index = (int)(*tt)++;
+
+		/* Append to our chain of expressions. */
+
+		if (NULL == first) {
+			assert(NULL == e);
+			first = next;
+		} else {
+			assert(NULL != e);
+			e->next = next;
+		}
+	}
+
+	return(first);
+err:
+	exprfree(first);
+	return(NULL);
+}
+
+/*
+ * Parse a terminal expression with the grammar as defined in
+ * apropos(1).
+ * Return NULL if we fail the parse.
+ */
+static struct expr *
+exprterm(char *buf, int cs)
 {
-	struct expr	*p;
 	struct expr	 e;
+	struct expr	*p;
 	char		*key;
-	int		 i, icase;
+	int		 i;
 
-	if ('\0' == *buf)
-		return(NULL);
+	memset(&e, 0, sizeof(struct expr));
 
-	/*
-	 * Choose regex or substring match.
-	 */
+	e.cs = cs;
+
+	/* Choose regex or substring match. */
 
 	if (NULL == (e.v = strpbrk(buf, "=~"))) {
 		e.regex = 0;
@@ -476,18 +655,11 @@
 		*e.v++ = '\0';
 	}
 
-	/*
-	 * Determine the record types to search for.
-	 */
+	/* Determine the record types to search for. */
 
-	icase = 0;
 	e.mask = 0;
 	if (buf < e.v) {
 		while (NULL != (key = strsep(&buf, ","))) {
-			if ('i' == key[0] && '\0' == key[1]) {
-				icase = REG_ICASE;
-				continue;
-			}
 			i = 0;
 			while (types[i].mask &&
 					strcmp(types[i].name, key))
@@ -498,9 +670,11 @@
 	if (0 == e.mask)
 		e.mask = TYPE_Nm | TYPE_Nd;
 
-	if (e.regex &&
-	    regcomp(&e.re, e.v, REG_EXTENDED | REG_NOSUB | icase))
-		return(NULL);
+	if (e.regex) {
+		i = REG_EXTENDED | REG_NOSUB | cs ? 0 : REG_ICASE;
+		if (regcomp(&e.re, e.v, i))
+			return(NULL);
+	}
 
 	e.v = mandoc_strdup(e.v);
 
@@ -512,26 +686,96 @@
 void
 exprfree(struct expr *p)
 {
+	struct expr	*pp;
+	
+	while (NULL != p) {
+		if (p->subexpr)
+			exprfree(p->subexpr);
+		if (p->regex)
+			regfree(&p->re);
+		free(p->v);
+		pp = p->next;
+		free(p);
+		p = pp;
+	}
+}
 
-	if (NULL == p)
-		return;
+static int
+exprmark(const struct expr *p, const char *cp, int mask, int *ms)
+{
+
+	for ( ; p; p = p->next) {
+		if (p->subexpr) {
+			if (exprmark(p->subexpr, cp, mask, ms))
+				return(1);
+			continue;
+		} else if ( ! (mask & p->mask))
+			continue;
 
-	if (p->regex)
-		regfree(&p->re);
+		if (p->regex) {
+			if (regexec(&p->re, cp, 0, NULL, 0))
+				continue;
+		} else if (p->cs) {
+			if (NULL == strstr(cp, p->v))
+				continue;
+		} else {
+			if (NULL == strcasestr(cp, p->v))
+				continue;
+		}
+
+		if (NULL == ms)
+			return(1);
+		else
+			ms[p->index] = 1;
+	}
 
-	free(p->v);
-	free(p);
+	return(0);
 }
 
 static int
-exprexec(const struct expr *p, char *cp, int mask)
+expreval(const struct expr *p, int *ms)
 {
+	int		 match;
 
-	if ( ! (mask & p->mask))
-		return(0);
+	/*
+	 * AND has precedence over OR.  Analysis is left-right, though
+	 * it doesn't matter because there are no side-effects.
+	 * Thus, step through pairwise ANDs and accumulate their Boolean
+	 * evaluation.  If we encounter a single true AND collection or
+	 * standalone term, the whole expression is true (by definition
+	 * of OR).
+	 */
+
+	for (match = 0; p && ! match; p = p->next) {
+		/* Evaluate a subexpression, if applicable. */
+		if (p->subexpr && ! ms[p->index])
+			ms[p->index] = expreval(p->subexpr, ms);
+
+		match = ms[p->index];
+		for ( ; p->next && p->next->and; p = p->next) {
+			/* Evaluate a subexpression, if applicable. */
+			if (p->next->subexpr && ! ms[p->next->index])
+				ms[p->next->index] = 
+					expreval(p->next->subexpr, ms);
+			match = match && ms[p->next->index];
+		}
+	}
+
+	return(match);
+}
+
+/*
+ * First, update the array of terms for which this expression evaluates
+ * to true.
+ * Second, logically evaluate all terms over the updated array of truth
+ * values.
+ * If this evaluates to true, mark the expression as satisfied.
+ */
+static void
+exprexec(const struct expr *p, const char *cp, int mask, struct rec *r)
+{
 
-	if (p->regex)
-		return(0 == regexec(&p->re, cp, 0, NULL, 0));
-	else
-		return(NULL != strcasestr(cp, p->v));
+	assert(0 == r->matched);
+	exprmark(p, cp, mask, r->matches);
+	r->matched = expreval(p, r->matches);
 }
Index: apropos_db.h
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/apropos_db.h,v
retrieving revision 1.4
diff -u -r1.4 apropos_db.h
--- apropos_db.h	14 Nov 2011 10:07:06 -0000	1.4
+++ apropos_db.h	17 Nov 2011 14:57:26 -0000
@@ -17,23 +17,13 @@
 #ifndef APROPOS_H
 #define APROPOS_H
 
-struct	rec {
+struct	res {
 	char		*file; /* file in file-system */
 	char		*cat; /* category (3p, 3, etc.) */
 	char		*title; /* title (FOO, etc.) */
 	char		*arch; /* arch (or empty string) */
 	char		*desc; /* description (from Nd) */
 	unsigned int	 rec; /* record in index */
-	/*
-	 * By the time the apropos_search() callback is called, these
-	 * are superfluous.
-	 * Maintain a binary tree for checking the uniqueness of `rec'
-	 * when adding elements to the results array.
-	 * Since the results array is dynamic, use offset in the array
-	 * instead of a pointer to the structure.
-	 */
-	int		 lhs;
-	int		 rhs;
 };
 
 struct	opts {
@@ -45,11 +35,10 @@
 
 struct	expr;
 
-void	 	 apropos_search(const struct opts *, 
-			const struct expr *, void *, 
-			void (*)(struct rec *, size_t, void *));
-
-struct	expr	*exprcomp(char *);
+int	 	 apropos_search(const struct opts *, 
+			const struct expr *, size_t, void *, 
+			void (*)(struct res *, size_t, void *));
+struct	expr	*exprcomp(int, char *[], size_t *);
 void		 exprfree(struct expr *);
 
 __END_DECLS

Re: Full logical patch: and, or, subexpressions.

[Ingo Schwarze]

Hi Kristaps,

Kristaps Dzonsons wrote on Thu, Nov 17, 2011 at 04:01:56PM +0100:

> Enclosed is a polished patch for full logical subexpressions,

This looks good, thanks!
I have merged and committed it to OpenBSD.
As expected, the conflicts were massive,
i basically did a line-by-line manual merge
without ever using patch(1) at all.

Considering the time of the night, I did only light testing,
but put it in anyway such that you have a reference merge
when merging back to bsd.lv.  I'd be grateful if you could
point out all my merge errors to me...  >:o)

Yours,
  Ingo
--
 To unsubscribe send an email to tech+unsubscribe@mdocml.bsd.lv