]> git.cameronkatri.com Git - mandoc.git/blobdiff - term.c
Commit and commit message by deraadt@:
[mandoc.git] / term.c
diff --git a/term.c b/term.c
index 266c49e732a4828f689e2f2017a9a0cc21fa87de..f4adf9522a4f5edf166a3a855a31ed8e6f005ff4 100644 (file)
--- a/term.c
+++ b/term.c
@@ -1,7 +1,7 @@
-/*     $Id: term.c,v 1.261 2017/06/01 19:05:37 schwarze Exp $ */
+/* $Id: term.c,v 1.284 2021/10/04 18:56:31 schwarze Exp $ */
 /*
+ * Copyright (c) 2010-2021 Ingo Schwarze <schwarze@openbsd.org>
  * Copyright (c) 2008, 2009, 2010, 2011 Kristaps Dzonsons <kristaps@bsd.lv>
- * Copyright (c) 2010-2017 Ingo Schwarze <schwarze@openbsd.org>
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -21,6 +21,7 @@
 
 #include <assert.h>
 #include <ctype.h>
+#include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include "main.h"
 
 static size_t           cond_width(const struct termp *, int, int *);
-static void             adjbuf(struct termp *p, size_t);
+static void             adjbuf(struct termp_col *, size_t);
 static void             bufferc(struct termp *, char);
 static void             encode(struct termp *, const char *, size_t);
 static void             encode1(struct termp *, int);
+static void             endline(struct termp *);
+static void             term_field(struct termp *, size_t, size_t);
+static void             term_fill(struct termp *, size_t *, size_t *,
+                               size_t);
 
 
 void
-term_free(struct termp *p)
+term_setcol(struct termp *p, size_t maxtcol)
 {
+       if (maxtcol > p->maxtcol) {
+               p->tcols = mandoc_recallocarray(p->tcols,
+                   p->maxtcol, maxtcol, sizeof(*p->tcols));
+               p->maxtcol = maxtcol;
+       }
+       p->lasttcol = maxtcol - 1;
+       p->tcol = p->tcols;
+}
 
-       free(p->buf);
+void
+term_free(struct termp *p)
+{
+       term_tab_free();
+       for (p->tcol = p->tcols; p->tcol < p->tcols + p->maxtcol; p->tcol++)
+               free(p->tcol->buf);
+       free(p->tcols);
        free(p->fontq);
        free(p);
 }
@@ -69,242 +88,344 @@ term_end(struct termp *p)
  * Flush a chunk of text.  By default, break the output line each time
  * the right margin is reached, and continue output on the next line
  * at the same offset as the chunk itself.  By default, also break the
- * output line at the end of the chunk.
- * The following flags may be specified:
- *
- *  - TERMP_NOBREAK: Do not break the output line at the right margin,
- *    but only at the max right margin.  Also, do not break the output
- *    line at the end of the chunk, such that the next call can pad to
- *    the next column.  However, if less than p->trailspace blanks,
- *    which can be 0, 1, or 2, remain to the right margin, the line
- *    will be broken.
- *  - TERMP_BRTRSP: Consider trailing whitespace significant
- *    when deciding whether the chunk fits or not.
- *  - TERMP_BRIND: If the chunk does not fit and the output line has
- *    to be broken, start the next line at the right margin instead
- *    of at the offset.  Used together with TERMP_NOBREAK for the tags
- *    in various kinds of tagged lists.
- *  - TERMP_DANGLE: Do not break the output line at the right margin,
- *    append the next chunk after it even if this one is too long.
- *    To be used together with TERMP_NOBREAK.
- *  - TERMP_HANG: Like TERMP_DANGLE, and also suppress padding before
- *    the next chunk if this column is not full.
+ * output line at the end of the chunk.  There are many flags modifying
+ * this behaviour, see the comments in the body of the function.
  */
 void
 term_flushln(struct termp *p)
 {
-       size_t           i;     /* current input position in p->buf */
-       int              ntab;  /* number of tabs to prepend */
-       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 */
-       size_t           dv;    /* temporary for visual pos calculations */
-       size_t           j;     /* temporary loop index for p->buf */
-       size_t           jhy;   /* last hyph before overflow w/r/t j */
-       size_t           maxvis; /* output position of visible boundary */
+       size_t   vbl;      /* Number of blanks to prepend to the output. */
+       size_t   vbr;      /* Actual visual position of the end of field. */
+       size_t   vfield;   /* Desired visual field width. */
+       size_t   vtarget;  /* Desired visual position of the right margin. */
+       size_t   ic;       /* Character position in the input buffer. */
+       size_t   nbr;      /* Number of characters to print in this field. */
 
        /*
-        * 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.
-        *
-        * The following unsigned-signed subtractions look strange,
-        * but they are actually correct.  If the int p->overstep
-        * is negative, it gets sign extended.  Subtracting that
-        * very large size_t effectively adds a small number to dv.
+        * Normally, start writing at the left margin, but with the
+        * NOPAD flag, start writing at the current position instead.
         */
-       dv = p->rmargin > p->offset ? p->rmargin - p->offset : 0;
-       maxvis = (int)dv > p->overstep ? dv - (size_t)p->overstep : 0;
-
-       if (p->flags & TERMP_NOBREAK) {
-               dv = p->maxrmargin > p->offset ?
-                    p->maxrmargin - p->offset : 0;
-               bp = (int)dv > p->overstep ?
-                    dv - (size_t)p->overstep : 0;
-       } else
-               bp = maxvis;
 
-       /*
-        * Calculate the required amount of padding.
-        */
-       vbl = p->offset + p->overstep > p->viscol ?
-             p->offset + p->overstep - p->viscol : 0;
+       vbl = (p->flags & TERMP_NOPAD) || p->tcol->offset < p->viscol ?
+           0 : p->tcol->offset - p->viscol;
+       if (p->minbl && vbl < p->minbl)
+               vbl = p->minbl;
+
+       if ((p->flags & TERMP_MULTICOL) == 0)
+               p->tcol->col = 0;
 
-       vis = vend = 0;
-       i = 0;
+       /* Loop over output lines. */
+
+       for (;;) {
+               vfield = p->tcol->rmargin > p->viscol + vbl ?
+                   p->tcol->rmargin - p->viscol - vbl : 0;
 
-       while (i < p->col) {
                /*
-                * Handle literal tab characters: collapse all
-                * subsequent tabs into a single huge set of spaces.
+                * Normally, break the line at the the right margin
+                * of the field, but with the NOBREAK flag, only
+                * break it at the max right margin of the screen,
+                * and with the BRNEVER flag, never break it at all.
                 */
-               ntab = 0;
-               while (i < p->col && p->buf[i] == '\t') {
-                       vend = term_tab_next(vis);
-                       vbl += vend - vis;
-                       vis = vend;
-                       ntab++;
-                       i++;
-               }
+
+               vtarget = (p->flags & TERMP_NOBREAK) == 0 ? vfield :
+                   p->maxrmargin > p->viscol + vbl ?
+                   p->maxrmargin - p->viscol - vbl : 0;
 
                /*
-                * 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
-                * space is printed according to regular spacing rules).
+                * Figure out how much text will fit in the field.
+                * If there is whitespace only, print nothing.
                 */
 
-               for (j = i, jhy = 0; j < p->col; j++) {
-                       if (' ' == p->buf[j] || '\t' == p->buf[j])
-                               break;
+               term_fill(p, &nbr, &vbr,
+                   p->flags & TERMP_BRNEVER ? SIZE_MAX : vtarget);
+               if (nbr == 0)
+                       break;
+
+               /*
+                * With the CENTER or RIGHT flag, increase the indentation
+                * to center the text between the left and right margins
+                * or to adjust it to the right margin, respectively.
+                */
+
+               if (vbr < vtarget) {
+                       if (p->flags & TERMP_CENTER)
+                               vbl += (vtarget - vbr) / 2;
+                       else if (p->flags & TERMP_RIGHT)
+                               vbl += vtarget - vbr;
+               }
 
-                       /* Back over the last printed character. */
-                       if (8 == p->buf[j]) {
-                               assert(j);
-                               vend -= (*p->width)(p, p->buf[j - 1]);
+               /* Finally, print the field content. */
+
+               term_field(p, vbl, nbr);
+
+               /*
+                * If there is no text left in the field, exit the loop.
+                * If the BRTRSP flag is set, consider trailing
+                * whitespace significant when deciding whether
+                * the field fits or not.
+                */
+
+               for (ic = p->tcol->col; ic < p->tcol->lastcol; ic++) {
+                       switch (p->tcol->buf[ic]) {
+                       case '\t':
+                               if (p->flags & TERMP_BRTRSP)
+                                       vbr = term_tab_next(vbr);
                                continue;
+                       case ' ':
+                               if (p->flags & TERMP_BRTRSP)
+                                       vbr += (*p->width)(p, ' ');
+                               continue;
+                       case '\n':
+                       case ASCII_BREAK:
+                               continue;
+                       default:
+                               break;
                        }
+                       break;
+               }
+               if (ic == p->tcol->lastcol)
+                       break;
 
-                       /* Regular word. */
-                       /* Break at the hyphen point if we overrun. */
-                       if (vend > vis && vend < bp &&
-                           (ASCII_HYPH == p->buf[j] ||
-                            ASCII_BREAK == p->buf[j]))
-                               jhy = j;
+               /*
+                * At the location of an automtic line break, input
+                * space characters are consumed by the line break.
+                */
 
-                       /*
-                        * Hyphenation now decided, put back a real
-                        * hyphen such that we get the correct width.
-                        */
-                       if (ASCII_HYPH == p->buf[j])
-                               p->buf[j] = '-';
+               while (p->tcol->col < p->tcol->lastcol &&
+                   p->tcol->buf[p->tcol->col] == ' ')
+                       p->tcol->col++;
 
-                       vend += (*p->width)(p, p->buf[j]);
-               }
+               /*
+                * In multi-column mode, leave the rest of the text
+                * in the buffer to be handled by a subsequent
+                * invocation, such that the other columns of the
+                * table can be handled first.
+                * In single-column mode, simply break the line.
+                */
+
+               if (p->flags & TERMP_MULTICOL)
+                       return;
+
+               endline(p);
+               p->viscol = 0;
 
                /*
-                * Find out whether we would exceed the right margin.
-                * If so, break to the next line.
+                * Normally, start the next line at the same indentation
+                * as this one, but with the BRIND flag, start it at the
+                * right margin instead.  This is used together with
+                * NOBREAK for the tags in various kinds of tagged lists.
                 */
-               if (vend > bp && 0 == jhy && vis > 0) {
-                       vend -= vis;
-                       (*p->endline)(p);
-                       p->viscol = 0;
 
-                       /* Use pending tabs on the new line. */
+               vbl = p->flags & TERMP_BRIND ?
+                   p->tcol->rmargin : p->tcol->offset;
+       }
 
-                       vbl = 0;
-                       while (ntab--)
-                               vbl = term_tab_next(vbl);
+       /* Reset output state in preparation for the next field. */
 
-                       /* Re-establish indentation. */
+       p->col = p->tcol->col = p->tcol->lastcol = 0;
+       p->minbl = p->trailspace;
+       p->flags &= ~(TERMP_BACKAFTER | TERMP_BACKBEFORE | TERMP_NOPAD);
 
-                       if (p->flags & TERMP_BRIND) {
-                               vbl += p->rmargin;
-                               vend += p->rmargin - p->offset;
-                       } else
-                               vbl += p->offset;
+       if (p->flags & TERMP_MULTICOL)
+               return;
 
-                       /*
-                        * Remove the p->overstep width.
-                        * Again, if p->overstep is negative,
-                        * sign extension does the right thing.
-                        */
+       /*
+        * The HANG flag means that the next field
+        * always follows on the same line.
+        * The NOBREAK flag means that the next field
+        * follows on the same line unless the field was overrun.
+        * Normally, break the line at the end of each field.
+        */
 
-                       bp += (size_t)p->overstep;
-                       p->overstep = 0;
-               }
+       if ((p->flags & TERMP_HANG) == 0 &&
+           ((p->flags & TERMP_NOBREAK) == 0 ||
+            vbr + term_len(p, p->trailspace) > vfield))
+               endline(p);
+}
+
+/*
+ * Store the number of input characters to print in this field in *nbr
+ * and their total visual width to print in *vbr.
+ * If there is only whitespace in the field, both remain zero.
+ * The desired visual width of the field is provided by vtarget.
+ * If the first word is longer, the field will be overrun.
+ */
+static void
+term_fill(struct termp *p, size_t *nbr, size_t *vbr, size_t vtarget)
+{
+       size_t   ic;        /* Character position in the input buffer. */
+       size_t   vis;       /* Visual position of the current character. */
+       size_t   vn;        /* Visual position of the next character. */
+       int      breakline; /* Break at the end of this word. */
+       int      graph;     /* Last character was non-blank. */
+
+       *nbr = *vbr = vis = 0;
+       breakline = graph = 0;
+       for (ic = p->tcol->col; ic < p->tcol->lastcol; ic++) {
+               switch (p->tcol->buf[ic]) {
+               case '\b':  /* Escape \o (overstrike) or backspace markup. */
+                       assert(ic > 0);
+                       vis -= (*p->width)(p, p->tcol->buf[ic - 1]);
+                       continue;
 
-               /* Write out the [remaining] word. */
-               for ( ; i < p->col; i++) {
-                       if (vend > bp && jhy > 0 && i > jhy)
+               case '\t':  /* Normal ASCII whitespace. */
+               case ' ':
+               case ASCII_BREAK:  /* Escape \: (breakpoint). */
+                       switch (p->tcol->buf[ic]) {
+                       case '\t':
+                               vn = term_tab_next(vis);
                                break;
-                       if ('\t' == p->buf[i])
+                       case ' ':
+                               vn = vis + (*p->width)(p, ' ');
                                break;
-                       if (' ' == p->buf[i]) {
-                               j = i;
-                               while (i < p->col && ' ' == p->buf[i])
-                                       i++;
-                               dv = (i - j) * (*p->width)(p, ' ');
-                               vbl += dv;
-                               vend += dv;
+                       case ASCII_BREAK:
+                               vn = vis;
                                break;
+                       default:
+                               abort();
                        }
-                       if (ASCII_NBRSP == p->buf[i]) {
-                               vbl += (*p->width)(p, ' ');
-                               continue;
+                       /* Can break at the end of a word. */
+                       if (breakline || vn > vtarget)
+                               break;
+                       if (graph) {
+                               *nbr = ic;
+                               *vbr = vis;
+                               graph = 0;
                        }
-                       if (ASCII_BREAK == p->buf[i])
-                               continue;
+                       vis = vn;
+                       continue;
+
+               case '\n':  /* Escape \p (break at the end of the word). */
+                       breakline = 1;
+                       continue;
 
+               case ASCII_HYPH:  /* Breakable hyphen. */
+                       graph = 1;
                        /*
-                        * Now we definitely know there will be
-                        * printable characters to output,
-                        * so write preceding white space now.
+                        * We are about to decide whether to break the
+                        * line or not, so we no longer need this hyphen
+                        * to be marked as breakable.  Put back a real
+                        * hyphen such that we get the correct width.
                         */
-                       if (vbl) {
-                               (*p->advance)(p, vbl);
-                               p->viscol += vbl;
-                               vbl = 0;
+                       p->tcol->buf[ic] = '-';
+                       vis += (*p->width)(p, '-');
+                       if (vis > vtarget) {
+                               ic++;
+                               break;
                        }
+                       *nbr = ic + 1;
+                       *vbr = vis;
+                       continue;
 
-                       (*p->letter)(p, p->buf[i]);
-                       if (8 == p->buf[i])
-                               p->viscol -= (*p->width)(p, p->buf[i-1]);
-                       else
-                               p->viscol += (*p->width)(p, p->buf[i]);
+               case ASCII_NBRSP:  /* Non-breakable space. */
+                       p->tcol->buf[ic] = ' ';
+                       /* FALLTHROUGH */
+               default:  /* Printable character. */
+                       graph = 1;
+                       vis += (*p->width)(p, p->tcol->buf[ic]);
+                       if (vis > vtarget && *nbr > 0)
+                               return;
+                       continue;
                }
-               vis = vend;
+               break;
        }
 
        /*
-        * If there was trailing white space, it was not printed;
-        * so reset the cursor position accordingly.
+        * If the last word extends to the end of the field without any
+        * trailing whitespace, the loop could not check yet whether it
+        * can remain on this line.  So do the check now.
         */
-       if (vis > vbl)
-               vis -= vbl;
-       else
-               vis = 0;
-
-       p->col = 0;
-       p->overstep = 0;
-       p->flags &= ~(TERMP_BACKAFTER | TERMP_BACKBEFORE);
 
-       if ( ! (TERMP_NOBREAK & p->flags)) {
-               p->viscol = 0;
-               (*p->endline)(p);
-               return;
+       if (graph && (vis <= vtarget || *nbr == 0)) {
+               *nbr = ic;
+               *vbr = vis;
        }
+}
 
-       if (TERMP_HANG & p->flags) {
-               p->overstep += (int)(p->offset + vis - p->rmargin +
-                   p->trailspace * (*p->width)(p, ' '));
+/*
+ * Print the contents of one field
+ * with an indentation of       vbl      visual columns,
+ * and an input string length of nbr     characters.
+ */
+static void
+term_field(struct termp *p, size_t vbl, size_t nbr)
+{
+       size_t   ic;    /* Character position in the input buffer. */
+       size_t   vis;   /* Visual position of the current character. */
+       size_t   dv;    /* Visual width of the current character. */
+       size_t   vn;    /* Visual position of the next character. */
+
+       vis = 0;
+       for (ic = p->tcol->col; ic < nbr; ic++) {
 
                /*
-                * If we have overstepped the margin, temporarily move
-                * it to the right and flag the rest of the line to be
-                * shorter.
-                * If there is a request to keep the columns together,
-                * allow negative overstep when the column is not full.
+                * To avoid the printing of trailing whitespace,
+                * do not print whitespace right away, only count it.
                 */
-               if (p->trailspace && p->overstep < 0)
-                       p->overstep = 0;
-               return;
 
-       } else if (TERMP_DANGLE & p->flags)
-               return;
+               switch (p->tcol->buf[ic]) {
+               case '\n':
+               case ASCII_BREAK:
+                       continue;
+               case '\t':
+                       vn = term_tab_next(vis);
+                       vbl += vn - vis;
+                       vis = vn;
+                       continue;
+               case ' ':
+               case ASCII_NBRSP:
+                       dv = (*p->width)(p, ' ');
+                       vbl += dv;
+                       vis += dv;
+                       continue;
+               default:
+                       break;
+               }
 
-       /* Trailing whitespace is significant in some columns. */
-       if (vis && vbl && (TERMP_BRTRSP & p->flags))
-               vis += vbl;
+               /*
+                * We found a non-blank character to print,
+                * so write preceding white space now.
+                */
 
-       /* If the column was overrun, break the line. */
-       if (maxvis < vis + p->trailspace * (*p->width)(p, ' ')) {
-               (*p->endline)(p);
-               p->viscol = 0;
+               if (vbl > 0) {
+                       (*p->advance)(p, vbl);
+                       p->viscol += vbl;
+                       vbl = 0;
+               }
+
+               /* Print the character and adjust the visual position. */
+
+               (*p->letter)(p, p->tcol->buf[ic]);
+               if (p->tcol->buf[ic] == '\b') {
+                       dv = (*p->width)(p, p->tcol->buf[ic - 1]);
+                       p->viscol -= dv;
+                       vis -= dv;
+               } else {
+                       dv = (*p->width)(p, p->tcol->buf[ic]);
+                       p->viscol += dv;
+                       vis += dv;
+               }
        }
+       p->tcol->col = nbr;
+}
+
+static void
+endline(struct termp *p)
+{
+       if ((p->flags & (TERMP_NEWMC | TERMP_ENDMC)) == TERMP_ENDMC) {
+               p->mc = NULL;
+               p->flags &= ~TERMP_ENDMC;
+       }
+       if (p->mc != NULL) {
+               if (p->viscol && p->maxrmargin >= p->viscol)
+                       (*p->advance)(p, p->maxrmargin - p->viscol + 1);
+               p->flags |= TERMP_NOBUF | TERMP_NOSPACE;
+               term_word(p, p->mc);
+               p->flags &= ~(TERMP_NOBUF | TERMP_NEWMC);
+       }
+       p->viscol = 0;
+       p->minbl = 0;
+       (*p->endline)(p);
 }
 
 /*
@@ -317,7 +438,7 @@ term_newln(struct termp *p)
 {
 
        p->flags |= TERMP_NOSPACE;
-       if (p->col || p->viscol)
+       if (p->tcol->lastcol || p->viscol)
                term_flushln(p);
 }
 
@@ -333,6 +454,7 @@ term_vspace(struct termp *p)
 
        term_newln(p);
        p->viscol = 0;
+       p->minbl = 0;
        if (0 < p->skipvsp)
                p->skipvsp--;
        else
@@ -404,27 +526,27 @@ term_word(struct termp *p, const char *word)
        const char       nbrsp[2] = { ASCII_NBRSP, 0 };
        const char      *seq, *cp;
        int              sz, uc;
-       size_t           ssz;
+       size_t           csz, lsz, ssz;
        enum mandoc_esc  esc;
 
-       if ( ! (TERMP_NOSPACE & p->flags)) {
-               if ( ! (TERMP_KEEP & p->flags)) {
-                       bufferc(p, ' ');
-                       if (TERMP_SENTENCE & p->flags)
+       if ((p->flags & TERMP_NOBUF) == 0) {
+               if ((p->flags & TERMP_NOSPACE) == 0) {
+                       if ((p->flags & TERMP_KEEP) == 0) {
                                bufferc(p, ' ');
-               } else
-                       bufferc(p, ASCII_NBRSP);
+                               if (p->flags & TERMP_SENTENCE)
+                                       bufferc(p, ' ');
+                       } else
+                               bufferc(p, ASCII_NBRSP);
+               }
+               if (p->flags & TERMP_PREKEEP)
+                       p->flags |= TERMP_KEEP;
+               if (p->flags & TERMP_NONOSPACE)
+                       p->flags |= TERMP_NOSPACE;
+               else
+                       p->flags &= ~TERMP_NOSPACE;
+               p->flags &= ~(TERMP_SENTENCE | TERMP_NONEWLINE);
+               p->skipvsp = 0;
        }
-       if (TERMP_PREKEEP & p->flags)
-               p->flags |= TERMP_KEEP;
-
-       if ( ! (p->flags & TERMP_NONOSPACE))
-               p->flags &= ~TERMP_NOSPACE;
-       else
-               p->flags |= TERMP_NOSPACE;
-
-       p->flags &= ~(TERMP_SENTENCE | TERMP_NONEWLINE);
-       p->skipvsp = 0;
 
        while ('\0' != *word) {
                if ('\\' != *word) {
@@ -444,9 +566,6 @@ term_word(struct termp *p, const char *word)
 
                word++;
                esc = mandoc_escape(&word, &seq, &sz);
-               if (ESCAPE_ERROR == esc)
-                       continue;
-
                switch (esc) {
                case ESCAPE_UNICODE:
                        uc = mchars_num2uc(seq + 1, sz - 1);
@@ -467,32 +586,56 @@ term_word(struct termp *p, const char *word)
                                        encode1(p, uc);
                        }
                        continue;
+               case ESCAPE_UNDEF:
+                       uc = *seq;
+                       break;
                case ESCAPE_FONTBOLD:
+               case ESCAPE_FONTCB:
                        term_fontrepl(p, TERMFONT_BOLD);
                        continue;
                case ESCAPE_FONTITALIC:
+               case ESCAPE_FONTCI:
                        term_fontrepl(p, TERMFONT_UNDER);
                        continue;
                case ESCAPE_FONTBI:
                        term_fontrepl(p, TERMFONT_BI);
                        continue;
                case ESCAPE_FONT:
+               case ESCAPE_FONTCR:
                case ESCAPE_FONTROMAN:
                        term_fontrepl(p, TERMFONT_NONE);
                        continue;
                case ESCAPE_FONTPREV:
                        term_fontlast(p);
                        continue;
+               case ESCAPE_BREAK:
+                       bufferc(p, '\n');
+                       continue;
                case ESCAPE_NOSPACE:
                        if (p->flags & TERMP_BACKAFTER)
                                p->flags &= ~TERMP_BACKAFTER;
                        else if (*word == '\0')
                                p->flags |= (TERMP_NOSPACE | TERMP_NONEWLINE);
                        continue;
+               case ESCAPE_DEVICE:
+                       if (p->type == TERMTYPE_PDF)
+                               encode(p, "pdf", 3);
+                       else if (p->type == TERMTYPE_PS)
+                               encode(p, "ps", 2);
+                       else if (p->enc == TERMENC_ASCII)
+                               encode(p, "ascii", 5);
+                       else
+                               encode(p, "utf8", 4);
+                       continue;
                case ESCAPE_HORIZ:
-                       if (a2roffsu(seq, &su, SCALE_EM) == 0)
+                       if (*seq == '|') {
+                               seq++;
+                               uc = -p->col;
+                       } else
+                               uc = 0;
+                       if (a2roffsu(seq, &su, SCALE_EM) == NULL)
                                continue;
-                       uc = term_hspan(p, &su) / 24;
+                       uc += term_hen(p, &su);
                        if (uc > 0)
                                while (uc-- > 0)
                                        bufferc(p, ASCII_NBRSP);
@@ -501,15 +644,65 @@ term_word(struct termp *p, const char *word)
                        else {
                                uc += p->col;
                                p->col = 0;
-                               if (p->offset > (size_t)(-uc)) {
+                               if (p->tcol->offset > (size_t)(-uc)) {
                                        p->ti += uc;
-                                       p->offset += uc;
+                                       p->tcol->offset += uc;
                                } else {
-                                       p->ti -= p->offset;
-                                       p->offset = 0;
+                                       p->ti -= p->tcol->offset;
+                                       p->tcol->offset = 0;
                                }
                        }
                        continue;
+               case ESCAPE_HLINE:
+                       if ((cp = a2roffsu(seq, &su, SCALE_EM)) == NULL)
+                               continue;
+                       uc = term_hen(p, &su);
+                       if (uc <= 0) {
+                               if (p->tcol->rmargin <= p->tcol->offset)
+                                       continue;
+                               lsz = p->tcol->rmargin - p->tcol->offset;
+                       } else
+                               lsz = uc;
+                       if (*cp == seq[-1])
+                               uc = -1;
+                       else if (*cp == '\\') {
+                               seq = cp + 1;
+                               esc = mandoc_escape(&seq, &cp, &sz);
+                               switch (esc) {
+                               case ESCAPE_UNICODE:
+                                       uc = mchars_num2uc(cp + 1, sz - 1);
+                                       break;
+                               case ESCAPE_NUMBERED:
+                                       uc = mchars_num2char(cp, sz);
+                                       break;
+                               case ESCAPE_SPECIAL:
+                                       uc = mchars_spec2cp(cp, sz);
+                                       break;
+                               case ESCAPE_UNDEF:
+                                       uc = *seq;
+                                       break;
+                               default:
+                                       uc = -1;
+                                       break;
+                               }
+                       } else
+                               uc = *cp;
+                       if (uc < 0x20 || (uc > 0x7E && uc < 0xA0))
+                               uc = '_';
+                       if (p->enc == TERMENC_ASCII) {
+                               cp = ascii_uc2str(uc);
+                               csz = term_strlen(p, cp);
+                               ssz = strlen(cp);
+                       } else
+                               csz = (*p->width)(p, uc);
+                       while (lsz >= csz) {
+                               if (p->enc == TERMENC_ASCII)
+                                       encode(p, cp, ssz);
+                               else
+                                       encode1(p, uc);
+                               lsz -= csz;
+                       }
+                       continue;
                case ESCAPE_SKIPCHAR:
                        p->flags |= TERMP_BACKAFTER;
                        continue;
@@ -529,10 +722,12 @@ term_word(struct termp *p, const char *word)
                                }
                        }
                        /* Trim trailing backspace/blank pair. */
-                       if (p->col > 2 &&
-                           (p->buf[p->col - 1] == ' ' ||
-                            p->buf[p->col - 1] == '\t'))
-                               p->col -= 2;
+                       if (p->tcol->lastcol > 2 &&
+                           (p->tcol->buf[p->tcol->lastcol - 1] == ' ' ||
+                            p->tcol->buf[p->tcol->lastcol - 1] == '\t'))
+                               p->tcol->lastcol -= 2;
+                       if (p->col > p->tcol->lastcol)
+                               p->col = p->tcol->lastcol;
                        continue;
                default:
                        continue;
@@ -557,25 +752,28 @@ term_word(struct termp *p, const char *word)
 }
 
 static void
-adjbuf(struct termp *p, size_t sz)
+adjbuf(struct termp_col *c, size_t sz)
 {
-
-       if (0 == p->maxcols)
-               p->maxcols = 1024;
-       while (sz >= p->maxcols)
-               p->maxcols <<= 2;
-
-       p->buf = mandoc_reallocarray(p->buf, p->maxcols, sizeof(int));
+       if (c->maxcols == 0)
+               c->maxcols = 1024;
+       while (c->maxcols <= sz)
+               c->maxcols <<= 2;
+       c->buf = mandoc_reallocarray(c->buf, c->maxcols, sizeof(*c->buf));
 }
 
 static void
 bufferc(struct termp *p, char c)
 {
-
-       if (p->col + 1 >= p->maxcols)
-               adjbuf(p, p->col + 1);
-
-       p->buf[p->col++] = c;
+       if (p->flags & TERMP_NOBUF) {
+               (*p->letter)(p, c);
+               return;
+       }
+       if (p->col + 1 >= p->tcol->maxcols)
+               adjbuf(p->tcol, p->col + 1);
+       if (p->tcol->lastcol <= p->col || (c != ' ' && c != ASCII_NBRSP))
+               p->tcol->buf[p->col] = c;
+       if (p->tcol->lastcol < ++p->col)
+               p->tcol->lastcol = p->col;
 }
 
 /*
@@ -588,31 +786,40 @@ encode1(struct termp *p, int c)
 {
        enum termfont     f;
 
-       if (p->col + 7 >= p->maxcols)
-               adjbuf(p, p->col + 7);
+       if (p->flags & TERMP_NOBUF) {
+               (*p->letter)(p, c);
+               return;
+       }
+
+       if (p->col + 7 >= p->tcol->maxcols)
+               adjbuf(p->tcol, p->col + 7);
 
        f = (c == ASCII_HYPH || c > 127 || isgraph(c)) ?
            p->fontq[p->fonti] : TERMFONT_NONE;
 
        if (p->flags & TERMP_BACKBEFORE) {
-               if (p->buf[p->col - 1] == ' ' || p->buf[p->col - 1] == '\t')
+               if (p->tcol->buf[p->col - 1] == ' ' ||
+                   p->tcol->buf[p->col - 1] == '\t')
                        p->col--;
                else
-                       p->buf[p->col++] = 8;
+                       p->tcol->buf[p->col++] = '\b';
                p->flags &= ~TERMP_BACKBEFORE;
        }
-       if (TERMFONT_UNDER == f || TERMFONT_BI == f) {
-               p->buf[p->col++] = '_';
-               p->buf[p->col++] = 8;
+       if (f == TERMFONT_UNDER || f == TERMFONT_BI) {
+               p->tcol->buf[p->col++] = '_';
+               p->tcol->buf[p->col++] = '\b';
        }
-       if (TERMFONT_BOLD == f || TERMFONT_BI == f) {
-               if (ASCII_HYPH == c)
-                       p->buf[p->col++] = '-';
+       if (f == TERMFONT_BOLD || f == TERMFONT_BI) {
+               if (c == ASCII_HYPH)
+                       p->tcol->buf[p->col++] = '-';
                else
-                       p->buf[p->col++] = c;
-               p->buf[p->col++] = 8;
+                       p->tcol->buf[p->col++] = c;
+               p->tcol->buf[p->col++] = '\b';
        }
-       p->buf[p->col++] = c;
+       if (p->tcol->lastcol <= p->col || (c != ' ' && c != ASCII_NBRSP))
+               p->tcol->buf[p->col] = c;
+       if (p->tcol->lastcol < ++p->col)
+               p->tcol->lastcol = p->col;
        if (p->flags & TERMP_BACKAFTER) {
                p->flags |= TERMP_BACKBEFORE;
                p->flags &= ~TERMP_BACKAFTER;
@@ -624,15 +831,24 @@ encode(struct termp *p, const char *word, size_t sz)
 {
        size_t            i;
 
-       if (p->col + 2 + (sz * 5) >= p->maxcols)
-               adjbuf(p, p->col + 2 + (sz * 5));
+       if (p->flags & TERMP_NOBUF) {
+               for (i = 0; i < sz; i++)
+                       (*p->letter)(p, word[i]);
+               return;
+       }
+
+       if (p->col + 2 + (sz * 5) >= p->tcol->maxcols)
+               adjbuf(p->tcol, p->col + 2 + (sz * 5));
 
        for (i = 0; i < sz; i++) {
                if (ASCII_HYPH == word[i] ||
                    isgraph((unsigned char)word[i]))
                        encode1(p, word[i]);
                else {
-                       p->buf[p->col++] = word[i];
+                       if (p->tcol->lastcol <= p->col ||
+                           (word[i] != ' ' && word[i] != ASCII_NBRSP))
+                               p->tcol->buf[p->col] = word[i];
+                       p->col++;
 
                        /*
                         * Postpone the effect of \z while handling
@@ -646,6 +862,8 @@ encode(struct termp *p, const char *word, size_t sz)
                        }
                }
        }
+       if (p->tcol->lastcol < p->col)
+               p->tcol->lastcol = p->col;
 }
 
 void
@@ -669,7 +887,7 @@ term_setwidth(struct termp *p, const char *wstr)
                default:
                        break;
                }
-               if (a2roffsu(wstr, &su, SCALE_MAX))
+               if (a2roffsu(wstr, &su, SCALE_MAX) != NULL)
                        width = term_hspan(p, &su);
                else
                        iop = 0;
@@ -721,12 +939,8 @@ term_strlen(const struct termp *p, const char *cp)
                switch (*cp) {
                case '\\':
                        cp++;
-                       esc = mandoc_escape(&cp, &seq, &ssz);
-                       if (ESCAPE_ERROR == esc)
-                               continue;
-
                        rhs = NULL;
-
+                       esc = mandoc_escape(&cp, &seq, &ssz);
                        switch (esc) {
                        case ESCAPE_UNICODE:
                                uc = mchars_num2uc(seq + 1, ssz - 1);
@@ -747,6 +961,24 @@ term_strlen(const struct termp *p, const char *cp)
                                                sz += cond_width(p, uc, &skip);
                                }
                                continue;
+                       case ESCAPE_UNDEF:
+                               uc = *seq;
+                               break;
+                       case ESCAPE_DEVICE:
+                               if (p->type == TERMTYPE_PDF) {
+                                       rhs = "pdf";
+                                       rsz = 3;
+                               } else if (p->type == TERMTYPE_PS) {
+                                       rhs = "ps";
+                                       rsz = 2;
+                               } else if (p->enc == TERMENC_ASCII) {
+                                       rhs = "ascii";
+                                       rsz = 5;
+                               } else {
+                                       rhs = "utf8";
+                                       rsz = 4;
+                               }
+                               break;
                        case ESCAPE_SKIPCHAR:
                                skip = 1;
                                continue;
@@ -858,7 +1090,7 @@ term_vspan(const struct termp *p, const struct roffsu *su)
 }
 
 /*
- * Convert a scaling width to basic units, rounding down.
+ * Convert a scaling width to basic units, rounding towards 0.
  */
 int
 term_hspan(const struct termp *p, const struct roffsu *su)
@@ -866,3 +1098,17 @@ term_hspan(const struct termp *p, const struct roffsu *su)
 
        return (*p->hspan)(p, su);
 }
+
+/*
+ * Convert a scaling width to basic units, rounding to closest.
+ */
+int
+term_hen(const struct termp *p, const struct roffsu *su)
+{
+       int bu;
+
+       if ((bu = (*p->hspan)(p, su)) >= 0)
+               return (bu + 11) / 24;
+       else
+               return -((-bu + 11) / 24);
+}