]> git.cameronkatri.com Git - mandoc.git/blobdiff - term.c
Reduce memory and time consumption on certain malformed input files
[mandoc.git] / term.c
diff --git a/term.c b/term.c
index c41482ca6b3f41cdb86d8adca8d9d9e0ff524c21..e6e3f641b7a5393b7ac08a5890658e9fbeef744f 100644 (file)
--- a/term.c
+++ b/term.c
@@ -1,6 +1,7 @@
-/*     $Id: term.c,v 1.151 2010/06/27 01:26:20 schwarze Exp $ */
+/*     $Id: term.c,v 1.242 2014/12/24 23:32:42 schwarze Exp $ */
 /*
- * Copyright (c) 2008, 2009 Kristaps Dzonsons <kristaps@bsd.lv>
+ * Copyright (c) 2008, 2009, 2010, 2011 Kristaps Dzonsons <kristaps@bsd.lv>
+ * Copyright (c) 2010-2014 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
  * 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 <ctype.h>
-#include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 
 #include "mandoc.h"
-#include "chars.h"
+#include "mandoc_aux.h"
 #include "out.h"
 #include "term.h"
 #include "main.h"
 
-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 size_t           cond_width(const struct termp *, int, int *);
+static void             adjbuf(struct termp *p, size_t);
+static void             bufferc(struct termp *, char);
+static void             encode(struct termp *, const char *, size_t);
+static void             encode1(struct termp *, int);
 
 
 void
 term_free(struct termp *p)
 {
 
-       if (p->buf)
-               free(p->buf);
-       if (p->symtab)
-               chars_free(p->symtab);
-
+       free(p->buf);
+       free(p->fontq);
        free(p);
 }
 
-
 void
-term_begin(struct termp *p, term_margin head, 
+term_begin(struct termp *p, term_margin head,
                term_margin foot, const void *arg)
 {
 
@@ -65,7 +58,6 @@ term_begin(struct termp *p, term_margin head,
        (*p->begin)(p);
 }
 
-
 void
 term_end(struct termp *p)
 {
@@ -73,110 +65,86 @@ term_end(struct termp *p)
        (*p->end)(p);
 }
 
-
-struct termp *
-term_alloc(enum termenc enc)
-{
-       struct termp    *p;
-
-       p = calloc(1, sizeof(struct termp));
-       if (NULL == p) {
-               perror(NULL);
-               exit(EXIT_FAILURE);
-       }
-
-       p->enc = enc;
-       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 (`Bl -tag' or `Bl -column'), which does
- * not have a trailing newline.
- *
+ * 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_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.
- *
- *  - TERMP_NOBREAK: this is the most important and is used when making
- *    columns.  In short: don't print a newline and instead pad to the
- *    right margin.  Used in conjunction with TERMP_NOLPAD.
- *
- *  - 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.
+ *  - 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_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.
  */
 void
 term_flushln(struct termp *p)
 {
-       int              i;     /* current input position in p->buf */
+       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 */
-       int              j;     /* temporary loop index */
-       int              jhy;   /* last hyphen before line overflow */
-       size_t           maxvis, mmax;
+       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 */
 
        /*
         * 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.
+        *
+        * 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.
         */
+       dv = p->rmargin > p->offset ? p->rmargin - p->offset : 0;
+       maxvis = (int)dv > p->overstep ? dv - (size_t)p->overstep : 0;
 
-       assert(p->offset < p->rmargin);
-
-       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;
+       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;
 
        /*
-        * Indent the first line of a paragraph.
-        */
-       vbl = p->flags & TERMP_NOLPAD ? 0 : p->offset;
-
-       /* 
-        * FIXME: if bp is zero, we still output the first word before
-        * breaking the line.
+        * Calculate the required amount of padding.
         */
+       vbl = p->offset + p->overstep > p->viscol ?
+             p->offset + p->overstep - p->viscol : 0;
 
-       vis = vend = i = 0;
-       while (i < (int)p->col) {
+       vis = vend = 0;
+       i = 0;
 
+       while (i < p->col) {
                /*
-                * Handle literal tab characters.
+                * Handle literal tab characters: collapse all
+                * subsequent tabs into a single huge set of spaces.
                 */
-               for (j = i; j < (int)p->col; j++) {
-                       if ('\t' != p->buf[j])
-                               break;
-                       vend = (vis/p->tabwidth+1)*p->tabwidth;
+               ntab = 0;
+               while (i < p->col && '\t' == p->buf[i]) {
+                       vend = (vis / p->tabwidth + 1) * p->tabwidth;
                        vbl += vend - vis;
                        vis = vend;
+                       ntab++;
+                       i++;
                }
 
                /*
@@ -186,17 +154,32 @@ term_flushln(struct termp *p)
                 * space is printed according to regular spacing rules).
                 */
 
-               /* LINTED */
-               for (jhy = 0; j < (int)p->col; j++) {
-                       if ((j && ' ' == p->buf[j]) || '\t' == p->buf[j])
+               for (j = i, jhy = 0; j < p->col; j++) {
+                       if (' ' == p->buf[j] || '\t' == p->buf[j])
                                break;
-                       if (8 != p->buf[j]) {
-                               if (vend > vis && vend < bp &&
-                                   ASCII_HYPH == p->buf[j])
-                                       jhy = j;
-                               vend++;
-                       } else
-                               vend--;
+
+                       /* Back over the the last printed character. */
+                       if (8 == p->buf[j]) {
+                               assert(j);
+                               vend -= (*p->width)(p, p->buf[j - 1]);
+                               continue;
+                       }
+
+                       /* 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;
+
+                       /*
+                        * Hyphenation now decided, put back a real
+                        * hyphen such that we get the correct width.
+                        */
+                       if (ASCII_HYPH == p->buf[j])
+                               p->buf[j] = '-';
+
+                       vend += (*p->width)(p, p->buf[j]);
                }
 
                /*
@@ -206,45 +189,50 @@ term_flushln(struct termp *p)
                if (vend > bp && 0 == jhy && vis > 0) {
                        vend -= vis;
                        (*p->endline)(p);
-                       if (TERMP_NOBREAK & p->flags) {
-                               p->viscol = p->rmargin;
-                               (*p->advance)(p, p->rmargin);
-                               vend += p->rmargin - p->offset;
-                       } else {
-                               p->viscol = 0;
+                       p->viscol = 0;
+                       if (TERMP_BRIND & p->flags) {
+                               vbl = p->rmargin;
+                               vend += p->rmargin;
+                               vend -= p->offset;
+                       } else
                                vbl = p->offset;
-                       }
 
-                       /* Remove the p->overstep width. */
+                       /* use pending tabs on the new line */
+
+                       if (0 < ntab)
+                               vbl += ntab * p->tabwidth;
 
-                       bp += (int)/* LINTED */
-                               p->overstep;
+                       /*
+                        * Remove the p->overstep width.
+                        * Again, if p->overstep is negative,
+                        * sign extension does the right thing.
+                        */
+
+                       bp += (size_t)p->overstep;
                        p->overstep = 0;
                }
 
-               /*
-                * Skip leading tabs, they were handled above.
-                */
-               while (i < (int)p->col && '\t' == p->buf[i])
-                       i++;
-
                /* Write out the [remaining] word. */
-               for ( ; i < (int)p->col; i++) {
+               for ( ; i < p->col; i++) {
                        if (vend > bp && jhy > 0 && i > jhy)
                                break;
                        if ('\t' == p->buf[i])
                                break;
                        if (' ' == p->buf[i]) {
-                               while (' ' == p->buf[i]) {
-                                       vbl++;
+                               j = i;
+                               while (i < p->col && ' ' == p->buf[i])
                                        i++;
-                               }
+                               dv = (i - j) * (*p->width)(p, ' ');
+                               vbl += dv;
+                               vend += dv;
                                break;
                        }
                        if (ASCII_NBRSP == p->buf[i]) {
-                               vbl++;
+                               vbl += (*p->width)(p, ' ');
                                continue;
                        }
+                       if (ASCII_BREAK == p->buf[i])
+                               continue;
 
                        /*
                         * Now we definitely know there will be
@@ -257,17 +245,24 @@ term_flushln(struct termp *p)
                                vbl = 0;
                        }
 
-                       if (ASCII_HYPH == p->buf[i])
-                               (*p->letter)(p, '-');
+                       (*p->letter)(p, p->buf[i]);
+                       if (8 == p->buf[i])
+                               p->viscol -= (*p->width)(p, p->buf[i-1]);
                        else
-                               (*p->letter)(p, p->buf[i]);
-
-                       p->viscol += 1;
+                               p->viscol += (*p->width)(p, p->buf[i]);
                }
-               vend += vbl;
                vis = vend;
        }
 
+       /*
+        * If there was trailing white space, it was not printed;
+        * so reset the cursor position accordingly.
+        */
+       if (vis > vbl)
+               vis -= vbl;
+       else
+               vis = 0;
+
        p->col = 0;
        p->overstep = 0;
 
@@ -278,45 +273,31 @@ term_flushln(struct termp *p)
        }
 
        if (TERMP_HANG & p->flags) {
-               /* We need one blank after the tag. */
-               p->overstep = /* LINTED */
-                       vis - maxvis + 1;
+               p->overstep += (int)(p->offset + vis - p->rmargin +
+                   p->trailspace * (*p->width)(p, ' '));
 
                /*
-                * 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 there is a request to keep the columns together,
+                * allow negative overstep when the column is not full.
                 */
-               if (p->overstep >= -1) {
-                       assert((int)maxvis + p->overstep >= 0);
-                       /* LINTED */
-                       maxvis += p->overstep;
-               } else
+               if (p->trailspace && p->overstep < 0)
                        p->overstep = 0;
+               return;
 
        } else if (TERMP_DANGLE & p->flags)
                return;
 
-       /* Right-pad. */
-       if (maxvis > vis + /* LINTED */
-                       ((TERMP_TWOSPACE & p->flags) ? 1 : 0)) {
-               p->viscol += maxvis - vis;
-               (*p->advance)(p, maxvis - vis);
-               vis += (maxvis - vis);
-       } else {        /* ...or newline break. */
+       /* If the column was overrun, break the line. */
+       if (maxvis < vis + p->trailspace * (*p->width)(p, ' ')) {
                (*p->endline)(p);
-               p->viscol = p->rmargin;
-               (*p->advance)(p, p->rmargin);
+               p->viscol = 0;
        }
 }
 
-
-/* 
+/*
  * A newline only breaks an existing line; it won't assert vertical
  * space.  All data in the output buffer is flushed prior to the newline
  * assertion.
@@ -326,15 +307,10 @@ term_newln(struct termp *p)
 {
 
        p->flags |= TERMP_NOSPACE;
-       if (0 == p->col && 0 == p->viscol) {
-               p->flags &= ~TERMP_NOLPAD;
-               return;
-       }
-       term_flushln(p);
-       p->flags &= ~TERMP_NOLPAD;
+       if (p->col || p->viscol)
+               term_flushln(p);
 }
 
-
 /*
  * Asserts a vertical space (a full, empty line-break between lines).
  * Note that if used twice, this will cause two blank spaces and so on.
@@ -347,34 +323,13 @@ term_vspace(struct termp *p)
 
        term_newln(p);
        p->viscol = 0;
-       (*p->endline)(p);
-}
-
-
-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);
+       if (0 < p->skipvsp)
+               p->skipvsp--;
+       else
+               (*p->endline)(p);
 }
 
-
+/* Swap current and previous font; for \fP and .ft P */
 void
 term_fontlast(struct termp *p)
 {
@@ -385,7 +340,7 @@ term_fontlast(struct termp *p)
        p->fontq[p->fonti] = f;
 }
 
-
+/* Set font, save current, discard previous; for \f, .ft, .B etc. */
 void
 term_fontrepl(struct termp *p, enum termfont f)
 {
@@ -394,43 +349,39 @@ term_fontrepl(struct termp *p, enum termfont f)
        p->fontq[p->fonti] = f;
 }
 
-
+/* Set font, save previous. */
 void
 term_fontpush(struct termp *p, enum termfont f)
 {
 
-       assert(p->fonti + 1 < 10);
        p->fontl = p->fontq[p->fonti];
-       p->fontq[++p->fonti] = f;
+       if (++p->fonti == p->fontsz) {
+               p->fontsz += 8;
+               p->fontq = mandoc_reallocarray(p->fontq,
+                   p->fontsz, sizeof(enum termfont *));
+       }
+       p->fontq[p->fonti] = f;
 }
 
-
-const void *
+/* Retrieve pointer to current font. */
+const enum termfont *
 term_fontq(struct termp *p)
 {
 
        return(&p->fontq[p->fonti]);
 }
 
-
-enum termfont
-term_fonttop(struct termp *p)
-{
-
-       return(p->fontq[p->fonti]);
-}
-
-
+/* Flush to make the saved pointer current again. */
 void
-term_fontpopq(struct termp *p, const void *key)
+term_fontpopq(struct termp *p, const enum termfont *key)
 {
 
-       while (p->fonti >= 0 && key != &p->fontq[p->fonti])
+       while (p->fonti >= 0 && key < p->fontq + p->fonti)
                p->fonti--;
        assert(p->fonti >= 0);
 }
 
-
+/* Pop one font off the stack. */
 void
 term_fontpop(struct termp *p)
 {
@@ -439,7 +390,6 @@ term_fontpop(struct termp *p)
        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
@@ -448,110 +398,125 @@ term_fontpop(struct termp *p)
 void
 term_word(struct termp *p, const char *word)
 {
-       const char      *sv, *seq;
-       int              sz;
+       const char       nbrsp[2] = { ASCII_NBRSP, 0 };
+       const char      *seq, *cp;
+       int              sz, uc;
        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:
-                       break;
-               }
+       enum mandoc_esc  esc;
 
        if ( ! (TERMP_NOSPACE & p->flags)) {
                if ( ! (TERMP_KEEP & p->flags)) {
-                       if (TERMP_PREKEEP & p->flags)
-                               p->flags |= TERMP_KEEP;
                        bufferc(p, ' ');
                        if (TERMP_SENTENCE & p->flags)
                                bufferc(p, ' ');
                } else
                        bufferc(p, ASCII_NBRSP);
        }
+       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;
-
-       /* FIXME: use strcspn. */
+       p->flags &= ~(TERMP_SENTENCE | TERMP_NONEWLINE);
 
-       while (*word) {
+       while ('\0' != *word) {
                if ('\\' != *word) {
-                       encode(p, word, 1);
-                       word++;
+                       if (TERMP_SKIPCHAR & p->flags) {
+                               p->flags &= ~TERMP_SKIPCHAR;
+                               word++;
+                               continue;
+                       }
+                       if (TERMP_NBRWORD & p->flags) {
+                               if (' ' == *word) {
+                                       encode(p, nbrsp, 1);
+                                       word++;
+                                       continue;
+                               }
+                               ssz = strcspn(word, "\\ ");
+                       } else
+                               ssz = strcspn(word, "\\");
+                       encode(p, word, ssz);
+                       word += (int)ssz;
                        continue;
                }
 
-               seq = ++word;
-               sz = a2roffdeco(&deco, &seq, &ssz);
+               word++;
+               esc = mandoc_escape(&word, &seq, &sz);
+               if (ESCAPE_ERROR == esc)
+                       continue;
 
-               switch (deco) {
-               case (DECO_RESERVED):
-                       res(p, seq, ssz);
+               switch (esc) {
+               case ESCAPE_UNICODE:
+                       uc = mchars_num2uc(seq + 1, sz - 1);
                        break;
-               case (DECO_SPECIAL):
-                       spec(p, seq, ssz);
+               case ESCAPE_NUMBERED:
+                       uc = mchars_num2char(seq, sz);
+                       if (uc < 0)
+                               continue;
                        break;
-               case (DECO_BOLD):
+               case ESCAPE_SPECIAL:
+                       if (p->enc == TERMENC_ASCII) {
+                               cp = mchars_spec2str(p->symtab,
+                                   seq, sz, &ssz);
+                               if (cp != NULL)
+                                       encode(p, cp, ssz);
+                       } else {
+                               uc = mchars_spec2cp(p->symtab, seq, sz);
+                               if (uc > 0)
+                                       encode1(p, uc);
+                       }
+                       continue;
+               case ESCAPE_FONTBOLD:
                        term_fontrepl(p, TERMFONT_BOLD);
-                       break;
-               case (DECO_ITALIC):
+                       continue;
+               case ESCAPE_FONTITALIC:
                        term_fontrepl(p, TERMFONT_UNDER);
-                       break;
-               case (DECO_ROMAN):
+                       continue;
+               case ESCAPE_FONTBI:
+                       term_fontrepl(p, TERMFONT_BI);
+                       continue;
+               case ESCAPE_FONT:
+                       /* FALLTHROUGH */
+               case ESCAPE_FONTROMAN:
                        term_fontrepl(p, TERMFONT_NONE);
-                       break;
-               case (DECO_PREVIOUS):
+                       continue;
+               case ESCAPE_FONTPREV:
                        term_fontlast(p);
-                       break;
+                       continue;
+               case ESCAPE_NOSPACE:
+                       if (TERMP_SKIPCHAR & p->flags)
+                               p->flags &= ~TERMP_SKIPCHAR;
+                       else if ('\0' == *word)
+                               p->flags |= (TERMP_NOSPACE | TERMP_NONEWLINE);
+                       continue;
+               case ESCAPE_SKIPCHAR:
+                       p->flags |= TERMP_SKIPCHAR;
+                       continue;
                default:
-                       break;
+                       continue;
                }
 
-               word += sz;
-               if (DECO_NOSPACE == deco && '\0' == *word)
-                       p->flags |= TERMP_NOSPACE;
-       }
+               /*
+                * Common handling for Unicode and numbered
+                * character escape sequences.
+                */
 
-       /* 
-        * 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('['):
-                       p->flags |= TERMP_NOSPACE;
-                       break;
-               default:
-                       break;
+               if (p->enc == TERMENC_ASCII) {
+                       cp = ascii_uc2str(uc);
+                       encode(p, cp, strlen(cp));
+               } else {
+                       if ((uc < 0x20 && uc != 0x09) ||
+                           (uc > 0x7E && uc < 0xA0))
+                               uc = 0xFFFD;
+                       encode1(p, uc);
                }
+       }
+       p->flags &= ~TERMP_NBRWORD;
 }
 
-
 static void
 adjbuf(struct termp *p, size_t sz)
 {
@@ -561,42 +526,62 @@ adjbuf(struct termp *p, size_t sz)
        while (sz >= p->maxcols)
                p->maxcols <<= 2;
 
-       p->buf = realloc(p->buf, p->maxcols);
-       if (NULL == p->buf) {
-               perror(NULL);
-               exit(EXIT_FAILURE);
-       }
+       p->buf = mandoc_reallocarray(p->buf, p->maxcols, sizeof(int));
 }
 
-
 static void
-buffera(struct termp *p, const char *word, size_t sz)
+bufferc(struct termp *p, char c)
 {
 
-       if (p->col + sz >= p->maxcols) 
-               adjbuf(p, p->col + sz);
+       if (p->col + 1 >= p->maxcols)
+               adjbuf(p, p->col + 1);
 
-       memcpy(&p->buf[(int)p->col], word, sz);
-       p->col += sz;
+       p->buf[p->col++] = c;
 }
 
-
+/*
+ * See encode().
+ * Do this for a single (probably unicode) value.
+ * Does not check for non-decorated glyphs.
+ */
 static void
-bufferc(struct termp *p, char c)
+encode1(struct termp *p, int c)
 {
+       enum termfont     f;
 
-       if (p->col + 1 >= p->maxcols)
-               adjbuf(p, p->col + 1);
+       if (TERMP_SKIPCHAR & p->flags) {
+               p->flags &= ~TERMP_SKIPCHAR;
+               return;
+       }
 
-       p->buf[(int)p->col++] = c;
-}
+       if (p->col + 6 >= p->maxcols)
+               adjbuf(p, p->col + 6);
 
+       f = *term_fontq(p);
+
+       if (TERMFONT_UNDER == f || TERMFONT_BI == f) {
+               p->buf[p->col++] = '_';
+               p->buf[p->col++] = 8;
+       }
+       if (TERMFONT_BOLD == f || TERMFONT_BI == f) {
+               if (ASCII_HYPH == c)
+                       p->buf[p->col++] = '-';
+               else
+                       p->buf[p->col++] = c;
+               p->buf[p->col++] = 8;
+       }
+       p->buf[p->col++] = c;
+}
 
 static void
 encode(struct termp *p, const char *word, size_t sz)
 {
-       enum termfont     f;
-       int               i;
+       size_t            i;
+
+       if (TERMP_SKIPCHAR & p->flags) {
+               p->flags &= ~TERMP_SKIPCHAR;
+               return;
+       }
 
        /*
         * Encode and buffer a string of characters.  If the current
@@ -604,27 +589,57 @@ encode(struct termp *p, const char *word, size_t sz)
         * character by character.
         */
 
-       if (TERMFONT_NONE == (f = term_fonttop(p))) {
-               buffera(p, word, sz);
+       if (*term_fontq(p) == TERMFONT_NONE) {
+               if (p->col + sz >= p->maxcols)
+                       adjbuf(p, p->col + sz);
+               for (i = 0; i < sz; i++)
+                       p->buf[p->col++] = word[i];
                return;
        }
 
-       for (i = 0; i < (int)sz; i++) {
-               if ( ! isgraph((u_char)word[i])) {
-                       bufferc(p, word[i]);
-                       continue;
-               }
+       /* Pre-buffer, assuming worst-case. */
 
-               if (TERMFONT_UNDER == f)
-                       bufferc(p, '_');
-               else
-                       bufferc(p, word[i]);
+       if (p->col + 1 + (sz * 5) >= p->maxcols)
+               adjbuf(p, p->col + 1 + (sz * 5));
 
-               bufferc(p, 8);
-               bufferc(p, word[i]);
+       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];
        }
 }
 
+void
+term_setwidth(struct termp *p, const char *wstr)
+{
+       struct roffsu    su;
+       size_t           width;
+       int              iop;
+
+       iop = 0;
+       width = 0;
+       if (NULL != wstr) {
+               switch (*wstr) {
+               case '+':
+                       iop = 1;
+                       wstr++;
+                       break;
+               case '-':
+                       iop = -1;
+                       wstr++;
+                       break;
+               default:
+                       break;
+               }
+               if (a2roffsu(wstr, &su, SCALE_MAX))
+                       width = term_hspan(p, &su);
+               else
+                       iop = 0;
+       }
+       (*p->setwidth)(p, iop, width);
+}
 
 size_t
 term_len(const struct termp *p, size_t sz)
@@ -633,91 +648,176 @@ term_len(const struct termp *p, size_t sz)
        return((*p->width)(p, ' ') * sz);
 }
 
+static size_t
+cond_width(const struct termp *p, int c, int *skip)
+{
+
+       if (*skip) {
+               (*skip) = 0;
+               return(0);
+       } else
+               return((*p->width)(p, c));
+}
 
 size_t
 term_strlen(const struct termp *p, const char *cp)
 {
-       size_t           sz;
+       size_t           sz, rsz, i;
+       int              ssz, skip, uc;
+       const char      *seq, *rhs;
+       enum mandoc_esc  esc;
+       static const char rej[] = { '\\', ASCII_NBRSP, ASCII_HYPH,
+                       ASCII_BREAK, '\0' };
+
+       /*
+        * Account for escaped sequences within string length
+        * calculations.  This follows the logic in term_word() as we
+        * must calculate the width of produced strings.
+        */
+
+       sz = 0;
+       skip = 0;
+       while ('\0' != *cp) {
+               rsz = strcspn(cp, rej);
+               for (i = 0; i < rsz; i++)
+                       sz += cond_width(p, *cp++, &skip);
+
+               switch (*cp) {
+               case '\\':
+                       cp++;
+                       esc = mandoc_escape(&cp, &seq, &ssz);
+                       if (ESCAPE_ERROR == esc)
+                               continue;
 
-       for (sz = 0; *cp; cp++)
-               sz += (*p->width)(p, *cp);
+                       rhs = NULL;
+
+                       switch (esc) {
+                       case ESCAPE_UNICODE:
+                               uc = mchars_num2uc(seq + 1, ssz - 1);
+                               break;
+                       case ESCAPE_NUMBERED:
+                               uc = mchars_num2char(seq, ssz);
+                               if (uc < 0)
+                                       continue;
+                               break;
+                       case ESCAPE_SPECIAL:
+                               if (p->enc == TERMENC_ASCII) {
+                                       rhs = mchars_spec2str(p->symtab,
+                                           seq, ssz, &rsz);
+                                       if (rhs != NULL)
+                                               break;
+                               } else {
+                                       uc = mchars_spec2cp(p->symtab,
+                                           seq, ssz);
+                                       if (uc > 0)
+                                               sz += cond_width(p, uc, &skip);
+                               }
+                               continue;
+                       case ESCAPE_SKIPCHAR:
+                               skip = 1;
+                               continue;
+                       default:
+                               continue;
+                       }
+
+                       /*
+                        * Common handling for Unicode and numbered
+                        * character escape sequences.
+                        */
+
+                       if (rhs == NULL) {
+                               if (p->enc == TERMENC_ASCII) {
+                                       rhs = ascii_uc2str(uc);
+                                       rsz = strlen(rhs);
+                               } else {
+                                       if ((uc < 0x20 && uc != 0x09) ||
+                                           (uc > 0x7E && uc < 0xA0))
+                                               uc = 0xFFFD;
+                                       sz += cond_width(p, uc, &skip);
+                                       continue;
+                               }
+                       }
+
+                       if (skip) {
+                               skip = 0;
+                               break;
+                       }
+
+                       /*
+                        * Common handling for all escape sequences
+                        * printing more than one character.
+                        */
+
+                       for (i = 0; i < rsz; i++)
+                               sz += (*p->width)(p, *rhs++);
+                       break;
+               case ASCII_NBRSP:
+                       sz += cond_width(p, ' ', &skip);
+                       cp++;
+                       break;
+               case ASCII_HYPH:
+                       sz += cond_width(p, '-', &skip);
+                       cp++;
+                       /* FALLTHROUGH */
+               case ASCII_BREAK:
+                       break;
+               default:
+                       break;
+               }
+       }
 
        return(sz);
 }
 
-
-size_t
+int
 term_vspan(const struct termp *p, const struct roffsu *su)
 {
        double           r;
+       int              ri;
 
        switch (su->unit) {
-       case (SCALE_CM):
-               r = su->scale * 2;
+       case SCALE_BU:
+               r = su->scale / 40.0;
+               break;
+       case SCALE_CM:
+               r = su->scale * 6.0 / 2.54;
+               break;
+       case SCALE_FS:
+               r = su->scale * 65536.0 / 40.0;
                break;
-       case (SCALE_IN):
-               r = su->scale * 6;
+       case SCALE_IN:
+               r = su->scale * 6.0;
                break;
-       case (SCALE_PC):
+       case SCALE_MM:
+               r = su->scale * 0.006;
+               break;
+       case SCALE_PC:
                r = su->scale;
                break;
-       case (SCALE_PT):
-               r = su->scale / 8;
+       case SCALE_PT:
+               r = su->scale / 12.0;
                break;
-       case (SCALE_MM):
-               r = su->scale / 1000;
+       case SCALE_EN:
+               /* FALLTHROUGH */
+       case SCALE_EM:
+               r = su->scale * 0.6;
                break;
-       case (SCALE_VS):
+       case SCALE_VS:
                r = su->scale;
                break;
        default:
-               r = su->scale - 1;
-               break;
+               abort();
+               /* NOTREACHED */
        }
-
-       if (r < 0.0)
-               r = 0.0;
-       return(/* LINTED */(size_t)
-                       r);
+       ri = r > 0.0 ? r + 0.4995 : r - 0.4995;
+       return(ri < 66 ? ri : 1);
 }
 
-
-size_t
+int
 term_hspan(const struct termp *p, const struct roffsu *su)
 {
-       double           r;
-
-       /* XXX: CM, IN, and PT are approximations. */
+       double           v;
 
-       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;
-       }
-
-       if (r < 0.0)
-               r = 0.0;
-       return((size_t)/* LINTED */
-                       r);
+       v = (*p->hspan)(p, su);
+       return(v > 0.0 ? v + 0.0005 : v - 0.0005);
 }
-
-