Re: Unifying the escape-sequence parser.

[Kristaps Dzonsons <kristaps@bsd.lv>]

[-- Attachment #1: Type: text/plain, Size: 1474 bytes --]

> Enclosed are my efforts to unify the escaped-sequence functions in out.c
> (a2roffdeco()) and mandoc.c (mandoc_special()).
>
> This handles, as far as I can see, all syntaxes of the groff(7) escapes.
>
> When called during libmandoc validation, it will check for GLYPH_ERROR
> and be followed by a search-and-replace of ASCII_HYPH for `-' in the
> substring. When invoked from term.c or html.c, it will switch on the
> returned type and substring value.
>
> This will clear up a nice big chunk of code, but it's a pretty delicate
> area, so please look it over!
>
> If you compile this file, you can test escapes by running, e.g.,
>
> % ./a.out s+\'\(\\f\[asdf\]\)\'123
>
> to see the values of "start" and "end".

Step 2.

This finishes off the new escape-sequence parser and puts it into 
mandoc_escape (mandoc.c, mandoc.h), then makes it the underlying engine 
for a2roffdeco (out.c) (requiring a tiny change to term.c and html.c for 
bailing out on bad sequences) and roff_res (roff.c).

Now all escape-sequences are being parsed with the same engine!  This 
logic was being repeated in THREE different places, earlier (mandoc.c 
for validation, out.c for output, and roff.c for predefined escapes).

I've run this over all manuals I know of without problems, but it can 
really use a close look-over with border cases.

The next step is to clean out the out.c code, completely removing enum 
roffdeco (putting that logic into mandoc.c, perhaps).

Thoughts?

Kristaps

[-- Attachment #2: patch.escapes.txt --]
[-- Type: text/plain, Size: 21158 bytes --]

Index: html.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/html.c,v
retrieving revision 1.131
diff -u -r1.131 html.c
--- html.c	22 Mar 2011 14:05:45 -0000	1.131
+++ html.c	8 Apr 2011 11:16:10 -0000
@@ -337,7 +337,8 @@
 			break;
 
 		seq = ++p;
-		len = a2roffdeco(&deco, &seq, &sz);
+		if (0 == (len = a2roffdeco(&deco, &seq, &sz)))
+			break;
 
 		switch (deco) {
 		case (DECO_NUMBERED):
Index: libmandoc.h
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/libmandoc.h,v
retrieving revision 1.17
diff -u -r1.17 libmandoc.h
--- libmandoc.h	28 Mar 2011 23:52:13 -0000	1.17
+++ libmandoc.h	8 Apr 2011 11:16:10 -0000
@@ -73,7 +73,6 @@
 			int, int, const char *);
 void		 mandoc_vmsg(enum mandocerr, struct mparse *, 
 			int, int, const char *, ...);
-int		 mandoc_special(char *);
 char		*mandoc_strdup(const char *);
 char		*mandoc_getarg(struct mparse *, char **, int, int *);
 char		*mandoc_normdate(struct mparse *, char *, int, int);
Index: man_validate.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/man_validate.c,v
retrieving revision 1.67
diff -u -r1.67 man_validate.c
--- man_validate.c	22 Mar 2011 15:30:30 -0000	1.67
+++ man_validate.c	8 Apr 2011 11:16:11 -0000
@@ -54,7 +54,7 @@
 static	int	  check_part(CHKARGS);
 static	int	  check_root(CHKARGS);
 static	int	  check_sec(CHKARGS);
-static	int	  check_text(CHKARGS);
+static	void	  check_text(CHKARGS);
 
 static	int	  post_AT(CHKARGS);
 static	int	  post_fi(CHKARGS);
@@ -151,7 +151,8 @@
 
 	switch (m->last->type) {
 	case (MAN_TEXT): 
-		return(check_text(m, m->last));
+		check_text(m, m->last);
+		return(1);
 	case (MAN_ROOT):
 		return(check_root(m, m->last));
 	case (MAN_EQN):
@@ -204,43 +205,48 @@
 	return(1);
 }
 
-
-static int
+static void
 check_text(CHKARGS) 
 {
-	char		*p;
-	int		 pos, c;
+	char		*p, *pp, *cpp;
+	int		 pos;
 	size_t		 sz;
 
-	for (p = n->string, pos = n->pos + 1; *p; p++, pos++) {
-		sz = strcspn(p, "\t\\");
-		p += (int)sz;
+	p = n->string;
+	pos = n->pos + 1;
 
-		if ('\0' == *p)
-			break;
+	while ('\0' != *p) {
+		sz = strcspn(p, "\t\\");
 
+		p += (int)sz;
 		pos += (int)sz;
 
 		if ('\t' == *p) {
-			if (MAN_LITERAL & m->flags)
-				continue;
-			man_pmsg(m, n->line, pos, MANDOCERR_BADTAB);
+			if ( ! (MAN_LITERAL & m->flags))
+				man_pmsg(m, n->line, pos, MANDOCERR_BADTAB);
+			p++;
+			pos++;
 			continue;
-		}
+		} else if ('\0' == *p)
+			break;
 
-		/* Check the special character. */
+		pos++;
+		pp = ++p;
 
-		c = mandoc_special(p);
-		if (c) {
-			p += c - 1;
-			pos += c - 1;
-		} else
+		if (ESCAPE_ERROR == mandoc_escape
+				((const char **)&pp, NULL, NULL)) {
 			man_pmsg(m, n->line, pos, MANDOCERR_BADESCAPE);
-	}
+			break;
+		}
 
-	return(1);
-}
+		cpp = p;
+		while (NULL != (cpp = memchr(cpp, ASCII_HYPH, pp - cpp)))
+			*cpp = '-';
 
+		pos += pp - p;
+		p = pp;
+	}
+}
 
 #define	INEQ_DEFINE(x, ineq, name) \
 static int \
Index: mandoc.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/mandoc.c,v
retrieving revision 1.44
diff -u -r1.44 mandoc.c
--- mandoc.c	28 Mar 2011 23:52:13 -0000	1.44
+++ mandoc.c	8 Apr 2011 11:16:11 -0000
@@ -35,198 +35,315 @@
 
 static	int	 a2time(time_t *, const char *, const char *);
 static	char	*time2a(time_t);
+static	int	 numescape(const char *);
 
-int
-mandoc_special(char *p)
+/*
+ * Pass over recursive numerical expressions.  This context of this
+ * function is important: it's only called within character-terminating
+ * escapes (e.g., \s[xxxyyy]), so all we need to do is handle initial
+ * recursion: we don't care about what's in these blocks. 
+ * This returns the number of characters skipped or -1 if an error
+ * occurs (the caller should bail).
+ */
+static int
+numescape(const char *start)
 {
-	int		 len, i;
-	char		 term;
-	char		*sv;
-	
-	len = 0;
+	int		 i;
+	size_t		 sz;
+	const char	*cp;
+
+	i = 0;
+
+	/* The expression consists of a subexpression. */
+
+	if ('\\' == start[i]) {
+		cp = &start[++i];
+		/*
+		 * Read past the end of the subexpression.
+		 * Bail immediately on errors.
+		 */
+		if (ESCAPE_ERROR == mandoc_escape(&cp, NULL, NULL))
+			return(-1);
+		return(i + cp - &start[i]);
+	} 
+
+	if ('(' != start[i++])
+		return(0);
+
+	/*
+	 * A parenthesised subexpression.  Read until the closing
+	 * parenthesis, making sure to handle any nested subexpressions
+	 * that might ruin our parse.
+	 */
+
+	while (')' != start[i]) {
+		sz = strcspn(&start[i], ")\\");
+		i += (int)sz;
+
+		if ('\0' == start[i])
+			return(-1);
+		else if ('\\' != start[i])
+			continue;
+
+		cp = &start[++i];
+		if (ESCAPE_ERROR == mandoc_escape(&cp, NULL, NULL))
+			return(-1);
+		i += cp - &start[i];
+	}
+
+	/* Read past the terminating ')'. */
+	return(++i);
+}
+
+/*
+ * Handle an escaped sequeence.  This should be called with any
+ * string subsequent a `\'.  Pass a pointer to this substring as "end";
+ * it will be set to the supremum of the parsed escape sequence.  If
+ * this returns ESCAPE_ERROR, the string is bogus and should be thrown
+ * away.  If not ESCAPE_ERROR or ESCAPE_IGNORE, "start" is set to the
+ * first relevant character of the substring (font, glyph, whatever) of
+ * length sz.  Both "start" and "sz" may be NULL.
+ */
+enum mandoc_esc
+mandoc_escape(const char **end, const char **start, int *sz)
+{
+	char		 c, term, numeric;
+	int		 i, lim, ssz;
+	const char	*cp;
+	enum mandoc_esc	 gly; 
+
+	cp = *end;
+	if (start)
+		*start = cp;
+	i = 0;
+	gly = ESCAPE_ERROR;
 	term = '\0';
-	sv = p;
+	numeric = 0;
 
-	assert('\\' == *p);
-	p++;
+	switch ((c = cp[i++])) {
+	/*
+	 * First the glyphs.  There are several different forms of
+	 * these, but each eventually returns a substring of the glyph
+	 * name.
+	 */
+	case ('('):
+		gly = ESCAPE_SPECIAL;
+		lim = 2;
+		break;
+	case ('['):
+		gly = ESCAPE_SPECIAL;
+		term = ']';
+		break;
+	case ('C'):
+		if ('\'' != cp[i])
+			return(ESCAPE_ERROR);
+		gly = ESCAPE_SPECIAL;
+		term = '\'';
+		break;
 
-	switch (*p++) {
-#if 0
-	case ('Z'):
-		/* FALLTHROUGH */
-	case ('X'):
-		/* FALLTHROUGH */
-	case ('x'):
-		/* FALLTHROUGH */
-	case ('S'):
-		/* FALLTHROUGH */
-	case ('R'):
-		/* FALLTHROUGH */
-	case ('N'):
-		/* FALLTHROUGH */
-	case ('l'):
-		/* FALLTHROUGH */
-	case ('L'):
-		/* FALLTHROUGH */
-	case ('H'):
+	/*
+	 * Handle all triggers matching \X(xy, \Xx, and \X[xxxx], where
+	 * 'X' is the trigger.  These have opaque sub-strings.
+	 */
+	case ('g'):
 		/* FALLTHROUGH */
-	case ('h'):
+	case ('k'):
 		/* FALLTHROUGH */
-	case ('D'):
+	case ('M'):
 		/* FALLTHROUGH */
-	case ('C'):
+	case ('m'):
 		/* FALLTHROUGH */
-	case ('b'):
+	case ('n'):
 		/* FALLTHROUGH */
-	case ('B'):
+	case ('V'):
 		/* FALLTHROUGH */
-	case ('a'):
+	case ('Y'):
+		if (ESCAPE_ERROR == gly)
+			gly = ESCAPE_IGNORE;
 		/* FALLTHROUGH */
-	case ('A'):
-		if (*p++ != '\'')
-			return(0);
-		term = '\'';
-		break;
-#endif
-	case ('h'):
+	case ('*'):
+		if (ESCAPE_ERROR == gly)
+			gly = ESCAPE_PREDEF;
 		/* FALLTHROUGH */
-	case ('v'):
+	case ('F'):
+		if (ESCAPE_ERROR == gly)
+			gly = ESCAPE_FONTFAM;
 		/* FALLTHROUGH */
-	case ('s'):
-		if (ASCII_HYPH == *p)
-			*p = '-';
+	case ('f'):
+		if (ESCAPE_ERROR == gly)
+			gly = ESCAPE_FONT;
 
-		i = 0;
-		if ('+' == *p || '-' == *p) {
-			p++;
-			i = 1;
-		}
+		if (start) 
+			*start = &cp[i];
 
-		switch (*p++) {
+		switch (cp[i++]) {
 		case ('('):
-			len = 2;
+			lim = 2;
 			break;
 		case ('['):
 			term = ']';
 			break;
-		case ('\''):
-			term = '\'';
-			break;
-		case ('0'):
-			i = 1;
-			/* FALLTHROUGH */
 		default:
-			len = 1;
-			p--;
+			lim = 1;
+			i--;
 			break;
 		}
+		break;
 
-		if (ASCII_HYPH == *p)
-			*p = '-';
-		if ('+' == *p || '-' == *p) {
-			if (i)
-				return(0);
-			p++;
-		} 
-		
-		/* Handle embedded numerical subexp or escape. */
-
-		if ('(' == *p) {
-			while (*p && ')' != *p)
-				if ('\\' == *p++) {
-					i = mandoc_special(--p);
-					if (0 == i)
-						return(0);
-					p += i;
-				}
-
-			if (')' == *p++)
-				break;
-
-			return(0);
-		} else if ('\\' == *p) {
-			if (0 == (i = mandoc_special(p)))
-				return(0);
-			p += i;
-		}
-
+	/*
+	 * These escapes are of the form \X'Y', where 'X' is the trigger
+	 * and 'Y' is any string.  These have opaque sub-strings.
+	 */
+	case ('A'):
+		/* FALLTHROUGH */
+	case ('b'):
+		/* FALLTHROUGH */
+	case ('D'):
+		/* FALLTHROUGH */
+	case ('o'):
+		/* FALLTHROUGH */
+	case ('R'):
+		/* FALLTHROUGH */
+	case ('X'):
+		/* FALLTHROUGH */
+	case ('Z'):
+		if ('\'' != cp[i++])
+			return(ESCAPE_ERROR);
+		gly = ESCAPE_IGNORE;
+		term = '\'';
 		break;
-#if 0
-	case ('Y'):
+
+	/*
+	 * These escapes are of the form \X'N', where 'X' is the trigger
+	 * and 'N' resolves to a numerical expression.
+	 */
+	case ('B'):
 		/* FALLTHROUGH */
-	case ('V'):
+	case ('h'):
 		/* FALLTHROUGH */
-	case ('$'):
+	case ('H'):
 		/* FALLTHROUGH */
-	case ('n'):
+	case ('L'):
 		/* FALLTHROUGH */
-#endif
-	case ('k'):
+	case ('l'):
 		/* FALLTHROUGH */
-	case ('M'):
+	case ('N'):
+		if (ESCAPE_ERROR == gly)
+			gly = ESCAPE_NUMBERED;
 		/* FALLTHROUGH */
-	case ('m'):
+	case ('S'):
 		/* FALLTHROUGH */
-	case ('f'):
+	case ('v'):
 		/* FALLTHROUGH */
-	case ('F'):
+	case ('w'):
 		/* FALLTHROUGH */
-	case ('*'):
-		switch (*p++) {
+	case ('x'):
+		if (ESCAPE_ERROR == gly)
+			gly = ESCAPE_IGNORE;
+		if ('\'' != cp[i++])
+			return(ESCAPE_ERROR);
+		term = numeric = '\'';
+		break;
+
+	/* 
+	 * Sizes get a special category of their own.
+	 */
+	case ('s'):
+		gly = ESCAPE_IGNORE;
+
+		if (start) 
+			*start = &cp[i];
+
+		/* See +/- counts as a sign. */
+		c = cp[i];
+		if ('+' == c || '-' == c || ASCII_HYPH == c)
+			++i;
+
+		switch (cp[i++]) {
 		case ('('):
-			len = 2;
+			lim = 2;
 			break;
 		case ('['):
-			term = ']';
+			term = numeric = ']';
+			break;
+		case ('\''):
+			term = numeric = '\'';
 			break;
 		default:
-			len = 1;
-			p--;
+			lim = 1;
+			i--;
 			break;
 		}
+
+		/* See +/- counts as a sign. */
+		c = cp[i];
+		if ('+' == c || '-' == c || ASCII_HYPH == c)
+			++i;
+
 		break;
-	case ('('):
-		len = 2;
-		break;
-	case ('['):
-		term = ']';
-		break;
-	case ('z'):
-		len = 1;
-		if ('\\' == *p) {
-			if (0 == (i = mandoc_special(p)))
-				return(0);
-			p += i;
-			return(*p ? (int)(p - sv) : 0);
-		}
-		break;
-	case ('o'):
-		/* FALLTHROUGH */
-	case ('w'):
-		if ('\'' == *p++) {
-			term = '\'';
-			break;
-		}
-		/* FALLTHROUGH */
+
+	/*
+	 * Anything else is assumed to be a glyph.
+	 */
 	default:
-		len = 1;
-		p--;
+		gly = ESCAPE_SPECIAL;
+		lim = 1;
+		i--;
 		break;
 	}
 
-	if (term) {
-		for ( ; *p && term != *p; p++)
-			if (ASCII_HYPH == *p)
-				*p = '-';
-		return(*p ? (int)(p - sv) : 0);
+	assert(ESCAPE_ERROR != gly);
+
+	if (start)
+		*start = &cp[i];
+
+	/*
+	 * If a terminating block has been specified, we need to
+	 * handle the case of recursion, which could have their
+	 * own terminating blocks that mess up our parse.  This, by the
+	 * way, means that the "start" and "size" values will be
+	 * effectively meaningless.
+	 */
+
+	ssz = 0;
+	if (numeric && -1 == (ssz = numescape(&cp[i])))
+		return(ESCAPE_ERROR);
+
+	i += ssz;
+
+	/*
+	 * We have a character terminator.  Try to read up to that
+	 * character.  If we can't (i.e., we hit the nil), then return
+	 * an error; if we can, calculate our length, read past the
+	 * terminating character, and exit.
+	 */
+
+	if ('\0' != term) {
+		*end = strchr(&cp[i], term);
+		if ('\0' == *end)
+			return(ESCAPE_ERROR);
+		if (sz)
+			*sz = *end - &cp[i];
+		(*end)++;
+		return(gly);
 	}
 
-	for (i = 0; *p && i < len; i++, p++)
-		if (ASCII_HYPH == *p)
-			*p = '-';
-	return(i == len ? (int)(p - sv) : 0);
-}
+	assert(lim > 0);
 
+	/*
+	 * We have a numeric limit.  If the string is shorter than that,
+	 * stop and return an error.  Else adjust our endpoint, length,
+	 * and return the current glyph.
+	 */
+
+	if ((size_t)lim > strlen(&cp[i]))
+		return(ESCAPE_ERROR);
+
+	if (sz)
+		*sz = lim;
+	*end = &cp[i] + lim;
+	return(gly);
+}
 
 void *
 mandoc_calloc(size_t num, size_t size)
Index: mandoc.h
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/mandoc.h,v
retrieving revision 1.69
diff -u -r1.69 mandoc.h
--- mandoc.h	28 Mar 2011 21:49:42 -0000	1.69
+++ mandoc.h	8 Apr 2011 11:16:11 -0000
@@ -288,6 +288,16 @@
 	MPARSE_MAN /* assume -man */
 };
 
+enum	mandoc_esc {
+	ESCAPE_ERROR = 0,
+	ESCAPE_IGNORE, /* escape to be ignored */
+	ESCAPE_SPECIAL, /* a regular special character */
+	ESCAPE_PREDEF, /* a predefined special character */
+	ESCAPE_FONT, /* a font mode */
+	ESCAPE_FONTFAM, /* a font family */
+	ESCAPE_NUMBERED /* a numbered glyph */
+};
+
 typedef	void	(*mandocmsg)(enum mandocerr, enum mandoclevel,
 			const char *, int, int, const char *);
 
@@ -309,6 +319,8 @@
 void		 *mandoc_calloc(size_t, size_t);
 void		 *mandoc_malloc(size_t);
 void		 *mandoc_realloc(void *, size_t);
+
+enum mandoc_esc	  mandoc_escape(const char **, const char **, int *);
 
 __END_DECLS
 
Index: out.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/out.c,v
retrieving revision 1.39
diff -u -r1.39 out.c
--- out.c	17 Mar 2011 08:49:34 -0000	1.39
+++ out.c	8 Apr 2011 11:16:11 -0000
@@ -178,237 +178,70 @@
 int
 a2roffdeco(enum roffdeco *d, const char **word, size_t *sz)
 {
-	int		 i, j, lim;
-	char		 term, c;
-	const char	*wp;
-	enum roffdeco	 dd;
+	const char	*cp, *start;
+	int		 ssz;
+	enum mandoc_esc	 esc;
 
 	*d = DECO_NONE;
-	lim = i = 0;
-	term = '\0';
-	wp = *word;
 
-	switch ((c = wp[i++])) {
-	case ('('):
+	cp = start = *word;
+
+	esc = mandoc_escape(&cp, word, &ssz);
+
+	switch (esc) {
+	case (ESCAPE_ERROR):
+		return(0);
+	case (ESCAPE_IGNORE):
+		break;
+	case (ESCAPE_NUMBERED):
+		*d = DECO_NUMBERED;
+		break;
+	case (ESCAPE_FONT):
+		*d = DECO_FONT;
+		break;
+	case (ESCAPE_FONTFAM):
+		*d = DECO_FFONT;
+		break;
+	case (ESCAPE_SPECIAL):
 		*d = DECO_SPECIAL;
-		lim = 2;
 		break;
-	case ('F'):
-		/* FALLTHROUGH */
-	case ('f'):
-		*d = 'F' == c ? DECO_FFONT : DECO_FONT;
-
-		switch (wp[i++]) {
-		case ('('):
-			lim = 2;
-			break;
-		case ('['):
-			term = ']';
-			break;
+	case (ESCAPE_PREDEF):
+		*d = DECO_RESERVED;
+		break;
+	}
+
+	assert(ssz >= 0);
+	*sz = (size_t)ssz;
+	ssz = cp - start;
+
+	if (1 == *sz && (DECO_FONT == *d || DECO_FFONT == *d))
+		switch (**word) {
 		case ('3'):
 			/* FALLTHROUGH */
 		case ('B'):
 			*d = DECO_BOLD;
-			return(i);
+			break;
 		case ('2'):
 			/* FALLTHROUGH */
 		case ('I'):
 			*d = DECO_ITALIC;
-			return(i);
+			break;
 		case ('P'):
 			*d = DECO_PREVIOUS;
-			return(i);
+			break;
 		case ('1'):
 			/* FALLTHROUGH */
 		case ('R'):
 			*d = DECO_ROMAN;
-			return(i);
-		default:
-			i--;
-			lim = 1;
-			break;
-		}
-		break;
-	case ('k'):
-		/* FALLTHROUGH */
-	case ('M'):
-		/* FALLTHROUGH */
-	case ('m'):
-		/* FALLTHROUGH */
-	case ('*'):
-		if ('*' == c)
-			*d = DECO_RESERVED;
-
-		switch (wp[i++]) {
-		case ('('):
-			lim = 2;
-			break;
-		case ('['):
-			term = ']';
-			break;
-		default:
-			i--;
-			lim = 1;
-			break;
-		}
-		break;
-
-	case ('N'):
-
-		/*
-		 * Sequence of characters:  backslash,  'N' (i = 0),
-		 * starting delimiter (i = 1), character number (i = 2).
-		 */
-
-		*word = wp + 2;
-		*sz = 0;
-
-		/*
-		 * Cannot use a digit as a starting delimiter;
-		 * but skip the digit anyway.
-		 */
-
-		if (isdigit((int)wp[1]))
-			return(2);
-
-		/*
-		 * Any non-digit terminates the character number.
-		 * That is, the terminating delimiter need not
-		 * match the starting delimiter.
-		 */
-
-		for (i = 2; isdigit((int)wp[i]); i++)
-			(*sz)++;
-
-		/*
-		 * This is only a numbered character
-		 * if the character number has at least one digit.
-		 */
-
-		if (*sz)
-			*d = DECO_NUMBERED;
-
-		/*
-		 * Skip the terminating delimiter, even if it does not
-		 * match, and even if there is no character number.
-		 */
-
-		return(++i);
-
-	case ('h'):
-		/* FALLTHROUGH */
-	case ('v'):
-		/* FALLTHROUGH */
-	case ('s'):
-		j = 0;
-		if ('+' == wp[i] || '-' == wp[i]) {
-			i++;
-			j = 1;
-		}
-
-		switch (wp[i++]) {
-		case ('('):
-			lim = 2;
-			break;
-		case ('['):
-			term = ']';
 			break;
-		case ('\''):
-			term = '\'';
-			break;
-		case ('0'):
-			j = 1;
-			/* FALLTHROUGH */
 		default:
-			i--;
-			lim = 1;
-			break;
-		}
-
-		if ('+' == wp[i] || '-' == wp[i]) {
-			if (j)
-				return(i);
-			i++;
-		} 
-
-		/* Handle embedded numerical subexp or escape. */
-
-		if ('(' == wp[i]) {
-			while (wp[i] && ')' != wp[i])
-				if ('\\' == wp[i++]) {
-					/* Handle embedded escape. */
-					*word = &wp[i];
-					i += a2roffdeco(&dd, word, sz);
-				}
-
-			if (')' == wp[i++])
-				break;
-
-			*d = DECO_NONE;
-			return(i - 1);
-		} else if ('\\' == wp[i]) {
-			*word = &wp[++i];
-			i += a2roffdeco(&dd, word, sz);
-		}
-
-		break;
-	case ('['):
-		*d = DECO_SPECIAL;
-		term = ']';
-		break;
-	case ('c'):
-		*d = DECO_NOSPACE;
-		return(i);
-	case ('z'):
-		*d = DECO_NONE;
-		if ('\\' == wp[i]) {
-			*word = &wp[++i];
-			return(i + a2roffdeco(&dd, word, sz));
-		} else
-			lim = 1;
-		break;
-	case ('o'):
-		/* FALLTHROUGH */
-	case ('w'):
-		if ('\'' == wp[i++]) {
-			term = '\'';
 			break;
-		} 
-		/* FALLTHROUGH */
-	default:
-		*d = DECO_SSPECIAL;
-		i--;
-		lim = 1;
-		break;
-	}
-
-	assert(term || lim);
-	*word = &wp[i];
-
-	if (term) {
-		j = i;
-		while (wp[i] && wp[i] != term)
-			i++;
-		if ('\0' == wp[i]) {
-			*d = DECO_NONE;
-			return(i);
 		}
 
-		assert(i >= j);
-		*sz = (size_t)(i - j);
-
-		return(i + 1);
-	}
-
-	assert(lim > 0);
-	*sz = (size_t)lim;
-
-	for (j = 0; wp[i] && j < lim; j++)
-		i++;
-	if (j < lim)
-		*d = DECO_NONE;
+	if (1 == *sz && DECO_SPECIAL == *d)
+		*d = 'c' == **word ? DECO_NOSPACE : DECO_SSPECIAL;
 
-	return(i);
+	return(ssz);
 }
 
 /*
Index: read.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/read.c,v
retrieving revision 1.11
diff -u -r1.11 read.c
--- read.c	4 Apr 2011 23:04:38 -0000	1.11
+++ read.c	8 Apr 2011 11:16:11 -0000
@@ -142,7 +142,7 @@
 	"tab in non-literal context",
 	"end of line whitespace",
 	"bad comment style",
-	"unknown escape sequence",
+	"bad escape sequence",
 	"unterminated quoted string",
 	
 	"generic error",
Index: roff.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/roff.c,v
retrieving revision 1.131
diff -u -r1.131 roff.c
--- roff.c	5 Apr 2011 22:22:33 -0000	1.131
+++ roff.c	8 Apr 2011 11:16:11 -0000
@@ -382,9 +382,10 @@
 	const char	*stnam;	/* start of the name, after "[(*" */
 	const char	*cp;	/* end of the name, e.g. before ']' */
 	const char	*res;	/* the string to be substituted */
-	int		 i, maxl;
+	int		 i;
 	size_t		 nsz;
 	char		*n;
+	enum mandoc_esc	 esc;
 
 	/* Search for a leading backslash and save a pointer to it. */
 
@@ -392,48 +393,11 @@
 	while (NULL != (cp = strchr(cp, '\\'))) {
 		stesc = cp++;
 
-		/*
-		 * The second character must be an asterisk.
-		 * If it isn't, skip it anyway:  It is escaped,
-		 * so it can't start another escape sequence.
-		 */
-
-		if ('\0' == *cp)
-			return(1);
-		if ('*' != *cp++)
-			continue;
-
-		/*
-		 * The third character decides the length
-		 * of the name of the string.
-		 * Save a pointer to the name.
-		 */
-
-		switch (*cp) {
-		case ('\0'):
-			return(1);
-		case ('('):
-			cp++;
-			maxl = 2;
-			break;
-		case ('['):
-			cp++;
-			maxl = 0;
-			break;
-		default:
-			maxl = 1;
+		esc = mandoc_escape(&cp, &stnam, &i);
+		if (ESCAPE_ERROR == esc)
 			break;
-		}
-		stnam = cp;
-
-		/* Advance to the end of the name. */
-
-		for (i = 0; 0 == maxl || i < maxl; i++, cp++) {
-			if ('\0' == *cp)
-				return(1); /* Error. */
-			if (0 == maxl && ']' == *cp)
-				break;
-		}
+		else if (ESCAPE_PREDEF != esc)
+			continue;
 
 		/*
 		 * Retrieve the replacement string; if it is
@@ -442,10 +406,8 @@
 
 		res = roff_getstrn(r, stnam, (size_t)i);
 
-		if (NULL == res) {
-			cp -= maxl ? 1 : 0;
+		if (NULL == res)
 			continue;
-		}
 
 		/* Replace the escape sequence by the string. */
 
@@ -454,7 +416,7 @@
 
 		strlcpy(n, *bufp, (size_t)(stesc - *bufp + 1));
 		strlcat(n, res, nsz);
-		strlcat(n, cp + (maxl ? 0 : 1), nsz);
+		strlcat(n, cp, nsz);
 
 		free(*bufp);
 
Index: term.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/term.c,v
retrieving revision 1.183
diff -u -r1.183 term.c
--- term.c	4 Apr 2011 21:14:12 -0000	1.183
+++ term.c	8 Apr 2011 11:16:11 -0000
@@ -457,6 +457,7 @@
 term_word(struct termp *p, const char *word)
 {
 	const char	*seq;
+	int		 sz;
 	size_t		 ssz;
 	enum roffdeco	 deco;
 
@@ -487,7 +488,9 @@
 			continue;
 
 		seq = ++word;
-		word += a2roffdeco(&deco, &seq, &ssz);
+		if (0 == (sz = a2roffdeco(&deco, &seq, &ssz)))
+			break;
+		word += sz;
 
 		switch (deco) {
 		case (DECO_NUMBERED):