===================================================================
RCS file: /cvs/mandoc/term.c,v
retrieving revision 1.80
retrieving revision 1.144
diff -u -p -r1.80 -r1.144
--- mandoc/term.c	2009/06/22 10:40:04	1.80
+++ mandoc/term.c	2010/06/08 09:20:08	1.144
@@ -1,4 +1,4 @@
-/*	$Id: term.c,v 1.80 2009/06/22 10:40:04 kristaps Exp $ */
+/*	$Id: term.c,v 1.144 2010/06/08 09:20:08 kristaps Exp $ */
 /*
  * Copyright (c) 2008, 2009 Kristaps Dzonsons <kristaps@kth.se>
  *
@@ -14,171 +14,331 @@
  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
  */
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <sys/types.h>
+
 #include <assert.h>
-#include <err.h>
+#include <ctype.h>
+#include <getopt.h>
+#include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <time.h>
 
+#include "mandoc.h"
+#include "chars.h"
+#include "out.h"
 #include "term.h"
 #include "man.h"
 #include "mdoc.h"
+#include "main.h"
 
-extern	int		  man_run(struct termp *, 
-				const struct man *);
-extern	int		  mdoc_run(struct termp *, 
-				const struct mdoc *);
+#define	PS_CHAR_WIDTH	  6
+#define	PS_CHAR_HEIGHT	  12
+#define	PS_CHAR_TOPMARG	 (792 - 24)
+#define	PS_CHAR_TOP	 (PS_CHAR_TOPMARG - 36)
+#define	PS_CHAR_LEFT	  36
+#define	PS_CHAR_BOTMARG	  24
+#define	PS_CHAR_BOT	 (PS_CHAR_BOTMARG + 36)
 
-static	struct termp	 *term_alloc(enum termenc);
+static	struct termp	 *alloc(char *, enum termenc, enum termtype);
 static	void		  term_free(struct termp *);
-static	void		  term_pword(struct termp *, const char *, int);
-static	void		  term_pescape(struct termp *, 
-				const char *, int *, int);
-static	void		  term_nescape(struct termp *,
-				const char *, size_t);
-static	void		  term_chara(struct termp *, char);
-static	void		  term_encodea(struct termp *, char);
-static	int		  term_isopendelim(const char *, int);
-static	int		  term_isclosedelim(const char *, int);
+static	void		  spec(struct termp *, const char *, size_t);
+static	void		  res(struct termp *, const char *, size_t);
+static	void		  buffera(struct termp *, const char *, size_t);
+static	void		  bufferc(struct termp *, char);
+static	void		  adjbuf(struct termp *p, size_t);
+static	void		  encode(struct termp *, const char *, size_t);
+static	void		  advance(struct termp *, size_t);
+static	void		  endline(struct termp *);
+static	void		  letter(struct termp *, char);
+static	void		  pageopen(struct termp *);
 
 
 void *
-ascii_alloc(void)
+ascii_alloc(char *outopts)
 {
 
-	return(term_alloc(TERMENC_ASCII));
+	return(alloc(outopts, TERMENC_ASCII, TERMTYPE_CHAR));
 }
 
 
-int
-terminal_man(void *arg, const struct man *man)
+void *
+ps_alloc(void)
 {
-	struct termp	*p;
 
-	p = (struct termp *)arg;
-	if (NULL == p->symtab)
-		p->symtab = term_ascii2htab();
+	return(alloc(NULL, TERMENC_ASCII, TERMTYPE_PS));
+}
 
-	return(man_run(p, man));
+
+void
+terminal_free(void *arg)
+{
+
+	term_free((struct termp *)arg);
 }
 
 
-int
-terminal_mdoc(void *arg, const struct mdoc *mdoc)
+static void
+term_free(struct termp *p)
 {
-	struct termp	*p;
 
-	p = (struct termp *)arg;
-	if (NULL == p->symtab)
-		p->symtab = term_ascii2htab();
+	if (p->buf)
+		free(p->buf);
+	if (p->symtab)
+		chars_free(p->symtab);
+	free(p);
+}
 
-	return(mdoc_run(p, mdoc));
+
+/*
+ * Push a single letter into our output engine.
+ */
+static void
+letter(struct termp *p, char c)
+{
+	
+	if (TERMTYPE_CHAR == p->type) {
+		/*
+		 * If using the terminal device, just push the letter
+		 * out into the screen.
+		 */
+		putchar(c);
+		return;
+	}
+
+	if ( ! (PS_INLINE & p->psstate)) {
+		/*
+		 * If we're not in a PostScript "word" context, then
+		 * open one now at the current cursor.
+		 */
+		printf("%zu %zu moveto\n", p->pscol, p->psrow);
+		putchar('(');
+		p->psstate |= PS_INLINE;
+	}
+
+	/*
+	 * We need to escape these characters as per the PostScript
+	 * specification.  We would also escape non-graphable characters
+	 * (like tabs), but none of them would get to this point and
+	 * it's superfluous to abort() on them.
+	 */
+
+	switch (c) {
+	case ('('):
+		/* FALLTHROUGH */
+	case (')'):
+		/* FALLTHROUGH */
+	case ('\\'):
+		putchar('\\');
+		break;
+	default:
+		break;
+	}
+
+	/* Write the character and adjust where we are on the page. */
+	putchar(c);
+	p->pscol += PS_CHAR_WIDTH;
 }
 
 
+/*
+ * Begin a "terminal" context.  Since terminal encompasses PostScript,
+ * the actual terminal, etc., there are a few things we can do here.
+ */
 void
-terminal_free(void *arg)
+term_begin(struct termp *p, term_margin head, 
+		term_margin foot, const void *arg)
 {
 
-	term_free((struct termp *)arg);
+	p->headf = head;
+	p->footf = foot;
+	p->argf = arg;
+
+	if (TERMTYPE_CHAR == p->type) {
+		/* Emit the header and be done. */
+		(*p->headf)(p, p->argf);
+		return;
+	}
+	
+	/*
+	 * Emit the standard PostScript prologue, set our initial page
+	 * position, then run pageopen() on the initial page.
+	 */
+
+	printf("%s\n", "%!PS");
+	printf("%s\n", "/Courier");
+	printf("%s\n", "10 selectfont");
+
+	p->pspage = 1;
+	p->psstate = 0;
+	pageopen(p);
 }
 
 
+/*
+ * Open a page.  This is only used for -Tps at the moment.  It opens a
+ * page context, printing the header and the footer.  THE OUTPUT BUFFER
+ * MUST BE EMPTY.  If it is not, output will ghost on the next line and
+ * we'll be all gross and out of state.
+ */
 static void
-term_free(struct termp *p)
+pageopen(struct termp *p)
 {
+	
+	assert(TERMTYPE_PS == p->type);
+	assert(0 == p->psstate);
 
-	if (p->buf)
-		free(p->buf);
-	if (TERMENC_ASCII == p->enc && p->symtab)
-		term_asciifree(p->symtab);
+	p->pscol = PS_CHAR_LEFT;
+	p->psrow = PS_CHAR_TOPMARG;
+	p->psstate |= PS_MARGINS;
 
-	free(p);
+	(*p->headf)(p, p->argf);
+	endline(p);
+
+	p->psstate &= ~PS_MARGINS;
+	assert(0 == p->psstate);
+
+	p->pscol = PS_CHAR_LEFT;
+	p->psrow = PS_CHAR_BOTMARG;
+	p->psstate |= PS_MARGINS;
+
+	(*p->footf)(p, p->argf);
+	endline(p);
+
+	p->psstate &= ~PS_MARGINS;
+	assert(0 == p->psstate);
+
+	p->pscol = PS_CHAR_LEFT;
+	p->psrow = PS_CHAR_TOP;
+
 }
 
 
-static struct termp *
-term_alloc(enum termenc enc)
+void
+term_end(struct termp *p)
 {
-	struct termp *p;
 
-	if (NULL == (p = malloc(sizeof(struct termp))))
-		err(1, "malloc");
-	bzero(p, sizeof(struct termp));
-	p->maxrmargin = 78;
-	p->enc = enc;
-	return(p);
+	if (TERMTYPE_CHAR == p->type) {
+		(*p->footf)(p, p->argf);
+		return;
+	}
+
+	printf("%s\n", "%%END");
 }
 
 
-static int
-term_isclosedelim(const char *p, int len)
+static void
+endline(struct termp *p)
 {
 
-	if (1 != len)
-		return(0);
+	if (TERMTYPE_CHAR == p->type) {
+		putchar('\n');
+		return;
+	}
 
-	switch (*p) {
-	case('.'):
-		/* FALLTHROUGH */
-	case(','):
-		/* FALLTHROUGH */
-	case(';'):
-		/* FALLTHROUGH */
-	case(':'):
-		/* FALLTHROUGH */
-	case('?'):
-		/* FALLTHROUGH */
-	case('!'):
-		/* FALLTHROUGH */
-	case(')'):
-		/* FALLTHROUGH */
-	case(']'):
-		/* FALLTHROUGH */
-	case('}'):
-		return(1);
-	default:
-		break;
+	if (PS_INLINE & p->psstate) {
+		printf(") show\n");
+		p->psstate &= ~PS_INLINE;
+	} 
+
+	if (PS_MARGINS & p->psstate)
+		return;
+
+	p->pscol = PS_CHAR_LEFT;
+	if (p->psrow >= PS_CHAR_HEIGHT + PS_CHAR_BOT) {
+		p->psrow -= PS_CHAR_HEIGHT;
+		return;
 	}
 
-	return(0);
+	/* 
+	 * XXX: can't run pageopen() until we're certain a flushln() has
+	 * occured, else the buf will reopen in an awkward state on the
+	 * next line.
+	 */
+	printf("showpage\n");
+	p->psrow = PS_CHAR_TOP;
 }
 
 
-static int
-term_isopendelim(const char *p, int len)
+/*
+ * Advance the output engine by a certain amount of whitespace.
+ */
+static void
+advance(struct termp *p, size_t len)
 {
+	size_t	 	i;
 
-	if (1 != len)
-		return(0);
+	if (TERMTYPE_CHAR == p->type) {
+		/* Just print whitespace on the terminal. */
+		for (i = 0; i < len; i++)
+			putchar(' ');
+		return;
+	}
 
-	switch (*p) {
-	case('('):
-		/* FALLTHROUGH */
-	case('['):
-		/* FALLTHROUGH */
-	case('{'):
-		return(1);
-	default:
-		break;
+	if (PS_INLINE & p->psstate) {
+		/* Dump out any existing line scope. */
+		printf(") show\n");
+		p->psstate &= ~PS_INLINE;
 	}
 
-	return(0);
+	p->pscol += len ? len * PS_CHAR_WIDTH : 0;
 }
 
 
+static struct termp *
+alloc(char *outopts, enum termenc enc, enum termtype type)
+{
+	struct termp	*p;
+	const char	*toks[2];
+	char		*v;
+	size_t		 width;
+
+	toks[0] = "width";
+	toks[1] = NULL;
+
+	p = calloc(1, sizeof(struct termp));
+	if (NULL == p) {
+		perror(NULL);
+		exit(EXIT_FAILURE);
+	}
+
+	p->type = type;
+	p->tabwidth = 5;
+	p->enc = enc;
+
+	width = 80;
+
+	while (outopts && *outopts)
+		switch (getsubopt(&outopts, UNCONST(toks), &v)) {
+		case (0):
+			width = (size_t)atoi(v);
+			break;
+		default:
+			break;
+		}
+
+	/* Enforce some lower boundary. */
+	if (width < 60)
+		width = 60;
+	p->defrmargin = width - 2;
+	return(p);
+}
+
+
 /*
  * Flush a line of text.  A "line" is loosely defined as being something
  * that should be followed by a newline, regardless of whether it's
  * broken apart by newlines getting there.  A line can also be a
- * fragment of a columnar list.
+ * fragment of a columnar list (`Bl -tag' or `Bl -column'), which does
+ * not have a trailing newline.
  *
- * Specifically, a line is whatever's in p->buf of length p->col, which
- * is zeroed after this function returns.
+ * The following flags may be specified:
  *
- * The variables TERMP_NOLPAD, TERMP_LITERAL and TERMP_NOBREAK are of
- * critical importance here.  Their behaviour follows:
- *
  *  - TERMP_NOLPAD: when beginning to write the line, don't left-pad the
  *    offset value.  This is useful when doing columnar lists where the
  *    prior column has right-padded.
@@ -187,52 +347,79 @@ term_isopendelim(const char *p, int len)
  *    columns.  In short: don't print a newline and instead pad to the
  *    right margin.  Used in conjunction with TERMP_NOLPAD.
  *
- *  - TERMP_NONOBREAK: don't newline when TERMP_NOBREAK is specified.
+ *  - TERMP_TWOSPACE: when padding, make sure there are at least two
+ *    space characters of padding.  Otherwise, rather break the line.
  *
+ *  - TERMP_DANGLE: don't newline when TERMP_NOBREAK is specified and
+ *    the line is overrun, and don't pad-right if it's underrun.
+ *
+ *  - TERMP_HANG: like TERMP_DANGLE, but doesn't newline when
+ *    overruning, instead save the position and continue at that point
+ *    when the next invocation.
+ *
  *  In-line line breaking:
  *
  *  If TERMP_NOBREAK is specified and the line overruns the right
  *  margin, it will break and pad-right to the right margin after
  *  writing.  If maxrmargin is violated, it will break and continue
- *  writing from the right-margin, which will lead to the above
- *  scenario upon exit.
- *
- *  Otherwise, the line will break at the right margin.  Extremely long
- *  lines will cause the system to emit a warning (TODO: hyphenate, if
- *  possible).
+ *  writing from the right-margin, which will lead to the above scenario
+ *  upon exit.  Otherwise, the line will break at the right margin.
  */
 void
 term_flushln(struct termp *p)
 {
-	int		 i, j;
-	size_t		 vsz, vis, maxvis, mmax, bp;
+	int		 i;     /* current input position in p->buf */
+	size_t		 vis;   /* current visual position on output */
+	size_t		 vbl;   /* number of blanks to prepend to output */
+	size_t		 vend;	/* end of word visual position on output */
+	size_t		 bp;    /* visual right border position */
+	int		 j;     /* temporary loop index */
+	int		 jhy;	/* last hyphen before line overflow */
+	size_t		 maxvis, mmax;
 
 	/*
 	 * First, establish the maximum columns of "visible" content.
 	 * This is usually the difference between the right-margin and
 	 * an indentation, but can be, for tagged lists or columns, a
-	 * small set of values.
+	 * small set of values. 
 	 */
 
 	assert(p->offset < p->rmargin);
-	maxvis = p->rmargin - p->offset;
-	mmax = p->maxrmargin - p->offset;
+
+	maxvis = (int)(p->rmargin - p->offset) - p->overstep < 0 ?
+		/* LINTED */ 
+		0 : p->rmargin - p->offset - p->overstep;
+	mmax = (int)(p->maxrmargin - p->offset) - p->overstep < 0 ?
+		/* LINTED */
+		0 : p->maxrmargin - p->offset - p->overstep;
+
 	bp = TERMP_NOBREAK & p->flags ? mmax : maxvis;
-	vis = 0;
 
 	/*
-	 * If in the standard case (left-justified), then begin with our
-	 * indentation, otherwise (columns, etc.) just start spitting
-	 * out text.
+	 * Indent the first line of a paragraph.
 	 */
+	vbl = p->flags & TERMP_NOLPAD ? 0 : p->offset;
 
-	if ( ! (p->flags & TERMP_NOLPAD))
-		/* LINTED */
-		for (j = 0; j < (int)p->offset; j++)
-			putchar(' ');
+	/* 
+	 * FIXME: if bp is zero, we still output the first word before
+	 * breaking the line.
+	 */
 
-	for (i = 0; i < (int)p->col; i++) {
+	vis = vend = i = 0;
+	while (i < (int)p->col) {
+
 		/*
+		 * Handle literal tab characters.
+		 */
+		for (j = i; j < (int)p->col; j++) {
+			if ('\t' != p->buf[j])
+				break;
+			vend = (vis/p->tabwidth+1)*p->tabwidth;
+			vbl += vend - vis;
+			vis = vend;
+		}
+
+		/*
 		 * Count up visible word characters.  Control sequences
 		 * (starting with the CSI) aren't counted.  A space
 		 * generates a non-printing word, which is valid (the
@@ -240,81 +427,132 @@ term_flushln(struct termp *p)
 		 */
 
 		/* LINTED */
-		for (j = i, vsz = 0; j < (int)p->col; j++) {
-			if (' ' == p->buf[j])
+		for (jhy = 0; j < (int)p->col; j++) {
+			if ((j && ' ' == p->buf[j]) || '\t' == p->buf[j])
 				break;
-			else if (8 == p->buf[j])
-				j += 1;
-			else
-				vsz++;
+			if (8 != p->buf[j]) {
+				if (vend > vis && vend < bp &&
+				    ASCII_HYPH == p->buf[j])
+					jhy = j;
+				vend++;
+			} else
+				vend--;
 		}
 
 		/*
-		 * Do line-breaking.  If we're greater than our
-		 * break-point and already in-line, break to the next
-		 * line and start writing.  If we're at the line start,
-		 * then write out the word (TODO: hyphenate) and break
-		 * in a subsequent loop invocation.
+		 * Find out whether we would exceed the right margin.
+		 * If so, break to the next line.
 		 */
+		if (vend > bp && 0 == jhy && vis > 0) {
+			vend -= vis;
+			endline(p);
+			if (TERMP_NOBREAK & p->flags) {
+				p->viscol = p->rmargin;
+				advance(p, p->rmargin);
+				vend += p->rmargin - p->offset;
+			} else {
+				p->viscol = 0;
+				vbl = p->offset;
+			}
 
-		if ( ! (TERMP_NOBREAK & p->flags)) {
-			if (vis && vis + vsz > bp) {
-				putchar('\n');
-				for (j = 0; j < (int)p->offset; j++)
-					putchar(' ');
-				vis = 0;
-			} 
-		} else if (vis && vis + vsz > bp) {
-			putchar('\n');
-			for (j = 0; j < (int)p->rmargin; j++)
-				putchar(' ');
-			vis = p->rmargin - p->offset;
+			/* Remove the p->overstep width. */
+
+			bp += (int)/* LINTED */
+				p->overstep;
+			p->overstep = 0;
 		}
 
 		/*
-		 * Prepend a space if we're not already at the beginning
-		 * of the line, then the word.
+		 * Skip leading tabs, they were handled above.
 		 */
+		while (i < (int)p->col && '\t' == p->buf[i])
+			i++;
 
-		if (0 < vis++)
-			putchar(' ');
-
+		/* Write out the [remaining] word. */
 		for ( ; i < (int)p->col; i++) {
-			if (' ' == p->buf[i])
+			if (vend > bp && jhy > 0 && i > jhy)
 				break;
-			putchar(p->buf[i]);
+			if ('\t' == p->buf[i])
+				break;
+			if (' ' == p->buf[i]) {
+				while (' ' == p->buf[i]) {
+					vbl++;
+					i++;
+				}
+				break;
+			}
+			if (ASCII_NBRSP == p->buf[i]) {
+				vbl++;
+				continue;
+			}
+
+			/*
+			 * Now we definitely know there will be
+			 * printable characters to output,
+			 * so write preceding white space now.
+			 */
+			if (vbl) {
+				advance(p, vbl);
+				p->viscol += vbl;
+				vbl = 0;
+			}
+
+			if (ASCII_HYPH == p->buf[i])
+				letter(p, '-');
+			else
+				letter(p, p->buf[i]);
+
+			p->viscol += 1;
 		}
-		vis += vsz;
+		vend += vbl;
+		vis = vend;
 	}
 
-	/*
-	 * If we've overstepped our maximum visible no-break space, then
-	 * cause a newline and offset at the right margin.
-	 */
+	p->col = 0;
+	p->overstep = 0;
 
-	if ((TERMP_NOBREAK & p->flags) && vis > maxvis) {
-		if ( ! (TERMP_NONOBREAK & p->flags)) {
-			putchar('\n');
-			for (i = 0; i < (int)p->rmargin; i++)
-				putchar(' ');
-		}
-		p->col = 0;
+	if ( ! (TERMP_NOBREAK & p->flags)) {
+		p->viscol = 0;
+		endline(p);
 		return;
 	}
 
-	/*
-	 * If we're not to right-marginalise it (newline), then instead
-	 * pad to the right margin and stay off.
-	 */
+	if (TERMP_HANG & p->flags) {
+		/* We need one blank after the tag. */
+		p->overstep = /* LINTED */
+			vis - maxvis + 1;
 
-	if (p->flags & TERMP_NOBREAK) {
-		if ( ! (TERMP_NONOBREAK & p->flags))
-			for ( ; vis <= maxvis; vis++)
-				putchar(' ');
-	} else
-		putchar('\n');
+		/*
+		 * Behave exactly the same way as groff:
+		 * If we have overstepped the margin, temporarily move
+		 * it to the right and flag the rest of the line to be
+		 * shorter.
+		 * If we landed right at the margin, be happy.
+		 * If we are one step before the margin, temporarily
+		 * move it one step LEFT and flag the rest of the line
+		 * to be longer.
+		 */
+		if (p->overstep >= -1) {
+			assert((int)maxvis + p->overstep >= 0);
+			/* LINTED */
+			maxvis += p->overstep;
+		} else
+			p->overstep = 0;
 
-	p->col = 0;
+	} else if (TERMP_DANGLE & p->flags)
+		return;
+
+	/* Right-pad. */
+	if (maxvis > vis + /* LINTED */
+			((TERMP_TWOSPACE & p->flags) ? 1 : 0)) {
+		p->viscol += maxvis - vis;
+		advance(p, maxvis - vis);
+		vis += (maxvis - vis);
+	} else {	/* ...or newline break. */
+		endline(p);
+		p->viscol = p->rmargin;
+		advance(p, p->rmargin);
+	}
 }
 
 
@@ -328,7 +566,7 @@ term_newln(struct termp *p)
 {
 
 	p->flags |= TERMP_NOSPACE;
-	if (0 == p->col) {
+	if (0 == p->col && 0 == p->viscol) {
 		p->flags &= ~TERMP_NOLPAD;
 		return;
 	}
@@ -348,224 +586,356 @@ term_vspace(struct termp *p)
 {
 
 	term_newln(p);
-	putchar('\n');
+	p->viscol = 0;
+	endline(p);
 }
 
 
-/*
- * Break apart a word into "pwords" (partial-words, usually from
- * breaking up a phrase into individual words) and, eventually, put them
- * into the output buffer.  If we're a literal word, then don't break up
- * the word and put it verbatim into the output buffer.
- */
+static void
+spec(struct termp *p, const char *word, size_t len)
+{
+	const char	*rhs;
+	size_t		 sz;
+
+	rhs = chars_a2ascii(p->symtab, word, len, &sz);
+	if (rhs) 
+		encode(p, rhs, sz);
+}
+
+
+static void
+res(struct termp *p, const char *word, size_t len)
+{
+	const char	*rhs;
+	size_t		 sz;
+
+	rhs = chars_a2res(p->symtab, word, len, &sz);
+	if (rhs)
+		encode(p, rhs, sz);
+}
+
+
 void
-term_word(struct termp *p, const char *word)
+term_fontlast(struct termp *p)
 {
-	int 		 i, j, len;
+	enum termfont	 f;
 
-	len = (int)strlen(word);
+	f = p->fontl;
+	p->fontl = p->fontq[p->fonti];
+	p->fontq[p->fonti] = f;
+}
 
-	if (p->flags & TERMP_LITERAL) {
-		term_pword(p, word, len);
-		return;
-	}
 
-	/* LINTED */
-	for (j = i = 0; i < len; i++) {
-		if (' ' != word[i]) {
-			j++;
-			continue;
-		} 
-		
-		/* Escaped spaces don't delimit... */
-		if (i && ' ' == word[i] && '\\' == word[i - 1]) {
-			j++;
-			continue;
-		}
+void
+term_fontrepl(struct termp *p, enum termfont f)
+{
 
-		if (0 == j)
-			continue;
-		assert(i >= j);
-		term_pword(p, &word[i - j], j);
-		j = 0;
-	}
-	if (j > 0) {
-		assert(i >= j);
-		term_pword(p, &word[i - j], j);
-	}
+	p->fontl = p->fontq[p->fonti];
+	p->fontq[p->fonti] = f;
 }
 
 
-/*
- * Determine the symbol indicated by an escape sequences, that is, one
- * starting with a backslash.  Once done, we pass this value into the
- * output buffer by way of the symbol table.
- */
-static void
-term_nescape(struct termp *p, const char *word, size_t len)
+void
+term_fontpush(struct termp *p, enum termfont f)
 {
-	const char	*rhs;
-	size_t		 sz;
-	int		 i;
 
-	if ((rhs = term_a2ascii(p->symtab, word, len, &sz))) 
-		for (i = 0; i < (int)sz; i++) 
-			term_encodea(p, rhs[i]);
+	assert(p->fonti + 1 < 10);
+	p->fontl = p->fontq[p->fonti];
+	p->fontq[++p->fonti] = f;
 }
 
 
-/*
- * Handle an escape sequence: determine its length and pass it to the
- * escape-symbol look table.  Note that we assume mdoc(3) has validated
- * the escape sequence (we assert upon badly-formed escape sequences).
- */
-static void
-term_pescape(struct termp *p, const char *word, int *i, int len)
+const void *
+term_fontq(struct termp *p)
 {
-	int		 j;
 
-	if (++(*i) >= len)
-		return;
+	return(&p->fontq[p->fonti]);
+}
 
-	if ('(' == word[*i]) {
-		(*i)++;
-		if (*i + 1 >= len)
-			return;
 
-		term_nescape(p, &word[*i], 2);
-		(*i)++;
-		return;
+enum termfont
+term_fonttop(struct termp *p)
+{
 
-	} else if ('*' == word[*i]) { 
-		(*i)++;
-		if (*i >= len)
-			return;
+	return(p->fontq[p->fonti]);
+}
 
-		switch (word[*i]) {
-		case ('('):
-			(*i)++;
-			if (*i + 1 >= len)
-				return;
 
-			term_nescape(p, &word[*i], 2);
-			(*i)++;
-			return;
-		case ('['):
+void
+term_fontpopq(struct termp *p, const void *key)
+{
+
+	while (p->fonti >= 0 && key != &p->fontq[p->fonti])
+		p->fonti--;
+	assert(p->fonti >= 0);
+}
+
+
+void
+term_fontpop(struct termp *p)
+{
+
+	assert(p->fonti);
+	p->fonti--;
+}
+
+
+/*
+ * Handle pwords, partial words, which may be either a single word or a
+ * phrase that cannot be broken down (such as a literal string).  This
+ * handles word styling.
+ */
+void
+term_word(struct termp *p, const char *word)
+{
+	const char	*sv, *seq;
+	int		 sz;
+	size_t		 ssz;
+	enum roffdeco	 deco;
+
+	sv = word;
+
+	if (word[0] && '\0' == word[1])
+		switch (word[0]) {
+		case('.'):
+			/* FALLTHROUGH */
+		case(','):
+			/* FALLTHROUGH */
+		case(';'):
+			/* FALLTHROUGH */
+		case(':'):
+			/* FALLTHROUGH */
+		case('?'):
+			/* FALLTHROUGH */
+		case('!'):
+			/* FALLTHROUGH */
+		case(')'):
+			/* FALLTHROUGH */
+		case(']'):
+			if ( ! (TERMP_IGNDELIM & p->flags))
+				p->flags |= TERMP_NOSPACE;
 			break;
 		default:
-			term_nescape(p, &word[*i], 1);
-			return;
+			break;
 		}
-	
-	} else if ('f' == word[*i]) {
-		(*i)++;
-		if (*i >= len)
-			return;
-		switch (word[*i]) {
-		case ('B'):
-			p->flags |= TERMP_BOLD;
+
+	if ( ! (TERMP_NOSPACE & p->flags)) {
+		bufferc(p, ' ');
+		if (TERMP_SENTENCE & p->flags)
+			bufferc(p, ' ');
+	}
+
+	if ( ! (p->flags & TERMP_NONOSPACE))
+		p->flags &= ~TERMP_NOSPACE;
+
+	p->flags &= ~TERMP_SENTENCE;
+
+	/* FIXME: use strcspn. */
+
+	while (*word) {
+		if ('\\' != *word) {
+			encode(p, word, 1);
+			word++;
+			continue;
+		}
+
+		seq = ++word;
+		sz = a2roffdeco(&deco, &seq, &ssz);
+
+		switch (deco) {
+		case (DECO_RESERVED):
+			res(p, seq, ssz);
 			break;
-		case ('I'):
-			p->flags |= TERMP_UNDER;
+		case (DECO_SPECIAL):
+			spec(p, seq, ssz);
 			break;
-		case ('P'):
+		case (DECO_BOLD):
+			term_fontrepl(p, TERMFONT_BOLD);
+			break;
+		case (DECO_ITALIC):
+			term_fontrepl(p, TERMFONT_UNDER);
+			break;
+		case (DECO_ROMAN):
+			term_fontrepl(p, TERMFONT_NONE);
+			break;
+		case (DECO_PREVIOUS):
+			term_fontlast(p);
+			break;
+		default:
+			break;
+		}
+
+		word += sz;
+		if (DECO_NOSPACE == deco && '\0' == *word)
+			p->flags |= TERMP_NOSPACE;
+	}
+
+	/* 
+	 * Note that we don't process the pipe: the parser sees it as
+	 * punctuation, but we don't in terms of typography.
+	 */
+	if (sv[0] && 0 == sv[1])
+		switch (sv[0]) {
+		case('('):
 			/* FALLTHROUGH */
-		case ('R'):
-			p->flags &= ~TERMP_STYLE;
+		case('['):
+			p->flags |= TERMP_NOSPACE;
 			break;
 		default:
 			break;
 		}
-		return;
+}
 
-	} else if ('[' != word[*i]) {
-		term_nescape(p, &word[*i], 1);
-		return;
+
+static void
+adjbuf(struct termp *p, size_t sz)
+{
+
+	if (0 == p->maxcols)
+		p->maxcols = 1024;
+	while (sz >= p->maxcols)
+		p->maxcols <<= 2;
+
+	p->buf = realloc(p->buf, p->maxcols);
+	if (NULL == p->buf) {
+		perror(NULL);
+		exit(EXIT_FAILURE);
 	}
+}
 
-	(*i)++;
-	for (j = 0; word[*i] && ']' != word[*i]; (*i)++, j++)
-		/* Loop... */ ;
 
-	if (0 == word[*i])
-		return;
+static void
+buffera(struct termp *p, const char *word, size_t sz)
+{
 
-	term_nescape(p, &word[*i - j], (size_t)j);
+	if (p->col + sz >= p->maxcols) 
+		adjbuf(p, p->col + sz);
+
+	memcpy(&p->buf[(int)p->col], word, sz);
+	p->col += sz;
 }
 
 
-/*
- * Handle pwords, partial words, which may be either a single word or a
- * phrase that cannot be broken down (such as a literal string).  This
- * handles word styling.
- */
 static void
-term_pword(struct termp *p, const char *word, int len)
+bufferc(struct termp *p, char c)
 {
-	int		 i;
 
-	if (term_isclosedelim(word, len))
-		if ( ! (TERMP_IGNDELIM & p->flags))
-			p->flags |= TERMP_NOSPACE;
+	if (p->col + 1 >= p->maxcols)
+		adjbuf(p, p->col + 1);
 
-	if ( ! (TERMP_NOSPACE & p->flags))
-		term_chara(p, ' ');
+	p->buf[(int)p->col++] = c;
+}
 
-	if ( ! (p->flags & TERMP_NONOSPACE))
-		p->flags &= ~TERMP_NOSPACE;
 
-	/* 
-	 * If ANSI (word-length styling), then apply our style now,
-	 * before the word.
+static void
+encode(struct termp *p, const char *word, size_t sz)
+{
+	enum termfont	  f;
+	int		  i;
+
+	/*
+	 * Encode and buffer a string of characters.  If the current
+	 * font mode is unset, buffer directly, else encode then buffer
+	 * character by character.
 	 */
 
-	for (i = 0; i < len; i++) 
-		if ('\\' == word[i]) 
-			term_pescape(p, word, &i, len);
+	if (TERMTYPE_PS == p->type) {
+		buffera(p, word, sz);
+		return;
+	} else if (TERMFONT_NONE == (f = term_fonttop(p))) {
+		buffera(p, word, sz);
+		return;
+	}
+
+	for (i = 0; i < (int)sz; i++) {
+		if ( ! isgraph((u_char)word[i])) {
+			bufferc(p, word[i]);
+			continue;
+		}
+
+		if (TERMFONT_UNDER == f)
+			bufferc(p, '_');
 		else
-			term_encodea(p, word[i]);
+			bufferc(p, word[i]);
 
-	if (term_isopendelim(word, len))
-		p->flags |= TERMP_NOSPACE;
+		bufferc(p, 8);
+		bufferc(p, word[i]);
+	}
 }
 
 
-/*
- * Insert a single character into the line-buffer.  If the buffer's
- * space is exceeded, then allocate more space by doubling the buffer
- * size.
- */
-static void
-term_chara(struct termp *p, char c)
+size_t
+term_vspan(const struct roffsu *su)
 {
-	size_t		 s;
+	double		 r;
 
-	if (p->col + 1 >= p->maxcols) {
-		if (0 == p->maxcols)
-			p->maxcols = 256;
-		s = p->maxcols * 2;
-		p->buf = realloc(p->buf, s);
-		if (NULL == p->buf)
-			err(1, "realloc");
-		p->maxcols = s;
+	switch (su->unit) {
+	case (SCALE_CM):
+		r = su->scale * 2;
+		break;
+	case (SCALE_IN):
+		r = su->scale * 6;
+		break;
+	case (SCALE_PC):
+		r = su->scale;
+		break;
+	case (SCALE_PT):
+		r = su->scale / 8;
+		break;
+	case (SCALE_MM):
+		r = su->scale / 1000;
+		break;
+	case (SCALE_VS):
+		r = su->scale;
+		break;
+	default:
+		r = su->scale - 1;
+		break;
 	}
-	p->buf[(int)(p->col)++] = c;
+
+	if (r < 0.0)
+		r = 0.0;
+	return(/* LINTED */(size_t)
+			r);
 }
 
 
-static void
-term_encodea(struct termp *p, char c)
+size_t
+term_hspan(const struct roffsu *su)
 {
+	double		 r;
 
-	if (TERMP_STYLE & p->flags) {
-		if (TERMP_BOLD & p->flags) {
-			term_chara(p, c);
-			term_chara(p, 8);
-		}
-		if (TERMP_UNDER & p->flags) {
-			term_chara(p, '_');
-			term_chara(p, 8);
-		}
+	/* XXX: CM, IN, and PT are approximations. */
+
+	switch (su->unit) {
+	case (SCALE_CM):
+		r = 4 * su->scale;
+		break;
+	case (SCALE_IN):
+		/* XXX: this is an approximation. */
+		r = 10 * su->scale;
+		break;
+	case (SCALE_PC):
+		r = (10 * su->scale) / 6;
+		break;
+	case (SCALE_PT):
+		r = (10 * su->scale) / 72;
+		break;
+	case (SCALE_MM):
+		r = su->scale / 1000; /* FIXME: double-check. */
+		break;
+	case (SCALE_VS):
+		r = su->scale * 2 - 1; /* FIXME: double-check. */
+		break;
+	default:
+		r = su->scale;
+		break;
 	}
-	term_chara(p, c);
+
+	if (r < 0.0)
+		r = 0.0;
+	return((size_t)/* LINTED */
+			r);
 }
+
+