]> git.cameronkatri.com Git - mandoc.git/blobdiff - html.c
clarify the difference between .Cm and .Ic;
[mandoc.git] / html.c
diff --git a/html.c b/html.c
index 96702d0fa2b1d9b2a1b5deed7eb849fd69962095..1302972a4e13a65bab4314d53bb548c7f4521475 100644 (file)
--- a/html.c
+++ b/html.c
@@ -1,7 +1,7 @@
-/*     $Id: html.c,v 1.196 2017/01/18 19:22:21 schwarze Exp $ */
+/*     $Id: html.c,v 1.254 2019/03/03 13:02:11 schwarze Exp $ */
 /*
  * Copyright (c) 2008-2011, 2014 Kristaps Dzonsons <kristaps@bsd.lv>
- * Copyright (c) 2011-2015, 2017 Ingo Schwarze <schwarze@openbsd.org>
+ * Copyright (c) 2011-2015, 2017-2019 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
 #include "config.h"
 
 #include <sys/types.h>
+#include <sys/stat.h>
 
 #include <assert.h>
 #include <ctype.h>
 #include <stdarg.h>
+#include <stddef.h>
 #include <stdio.h>
 #include <stdint.h>
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
 
-#include "mandoc.h"
 #include "mandoc_aux.h"
+#include "mandoc_ohash.h"
+#include "mandoc.h"
+#include "roff.h"
 #include "out.h"
 #include "html.h"
 #include "manconf.h"
@@ -58,6 +62,8 @@ static        const struct htmldata htmltags[TAG_MAX] = {
        {"meta",        HTML_NOSTACK | HTML_AUTOCLOSE | HTML_NLALL},
        {"title",       HTML_NLAROUND},
        {"div",         HTML_NLAROUND},
+       {"div",         0},
+       {"section",     HTML_NLALL},
        {"h1",          HTML_NLAROUND},
        {"h2",          HTML_NLAROUND},
        {"span",        0},
@@ -65,8 +71,6 @@ static        const struct htmldata htmltags[TAG_MAX] = {
        {"br",          HTML_NOSTACK | HTML_AUTOCLOSE | HTML_NLALL},
        {"a",           0},
        {"table",       HTML_NLALL | HTML_INDENT},
-       {"tbody",       HTML_NLALL | HTML_INDENT},
-       {"col",         HTML_NOSTACK | HTML_AUTOCLOSE | HTML_NLALL},
        {"tr",          HTML_NLALL | HTML_INDENT},
        {"td",          HTML_NLAROUND},
        {"li",          HTML_NLAROUND | HTML_INDENT},
@@ -75,8 +79,10 @@ static       const struct htmldata htmltags[TAG_MAX] = {
        {"dl",          HTML_NLALL | HTML_INDENT},
        {"dt",          HTML_NLAROUND},
        {"dd",          HTML_NLAROUND | HTML_INDENT},
-       {"blockquote",  HTML_NLALL | HTML_INDENT},
+       {"p",           HTML_NLAROUND | HTML_INDENT},
        {"pre",         HTML_NLALL | HTML_NOINDENT},
+       {"var",         0},
+       {"cite",        0},
        {"b",           0},
        {"i",           0},
        {"code",        0},
@@ -85,6 +91,7 @@ static        const struct htmldata htmltags[TAG_MAX] = {
        {"math",        HTML_NLALL | HTML_INDENT},
        {"mrow",        0},
        {"mi",          0},
+       {"mn",          0},
        {"mo",          0},
        {"msup",        0},
        {"msub",        0},
@@ -100,27 +107,19 @@ static    const struct htmldata htmltags[TAG_MAX] = {
        {"mover",       0},
 };
 
-static const char      *const roffscales[SCALE_MAX] = {
-       "cm", /* SCALE_CM */
-       "in", /* SCALE_IN */
-       "pc", /* SCALE_PC */
-       "pt", /* SCALE_PT */
-       "em", /* SCALE_EM */
-       "em", /* SCALE_MM */
-       "ex", /* SCALE_EN */
-       "ex", /* SCALE_BU */
-       "em", /* SCALE_VS */
-       "ex", /* SCALE_FS */
-};
+/* Avoid duplicate HTML id= attributes. */
+static struct ohash     id_unique;
+
+static void     html_reset_internal(struct html *);
+static void     print_byte(struct html *, char);
+static void     print_endword(struct html *);
+static void     print_indent(struct html *);
+static void     print_word(struct html *, const char *);
 
-static void     a2width(const char *, struct roffsu *);
-static void     html_endline(struct html *);
-static void     html_indent(struct html *);
 static void     print_ctag(struct html *, struct tag *);
-static int      print_escape(char);
+static int      print_escape(struct html *, char);
 static int      print_encode(struct html *, const char *, const char *, int);
 static void     print_href(struct html *, const char *, const char *, int);
-static void     print_metaf(struct html *, enum mandoc_esc);
 
 
 void *
@@ -130,30 +129,54 @@ html_alloc(const struct manoutput *outopts)
 
        h = mandoc_calloc(1, sizeof(struct html));
 
-       h->tags.head = NULL;
+       h->tag = NULL;
        h->style = outopts->style;
-       h->base_man = outopts->man;
+       if ((h->base_man1 = outopts->man) == NULL)
+               h->base_man2 = NULL;
+       else if ((h->base_man2 = strchr(h->base_man1, ';')) != NULL)
+               *h->base_man2++ = '\0';
        h->base_includes = outopts->includes;
        if (outopts->fragment)
                h->oflags |= HTML_FRAGMENT;
+       if (outopts->toc)
+               h->oflags |= HTML_TOC;
+
+       mandoc_ohash_init(&id_unique, 4, 0);
 
        return h;
 }
 
-void
-html_free(void *p)
+static void
+html_reset_internal(struct html *h)
 {
        struct tag      *tag;
-       struct html     *h;
+       char            *cp;
+       unsigned int     slot;
 
-       h = (struct html *)p;
-
-       while ((tag = h->tags.head) != NULL) {
-               h->tags.head = tag->next;
+       while ((tag = h->tag) != NULL) {
+               h->tag = tag->next;
                free(tag);
        }
+       cp = ohash_first(&id_unique, &slot);
+       while (cp != NULL) {
+               free(cp);
+               cp = ohash_next(&id_unique, &slot);
+       }
+       ohash_delete(&id_unique);
+}
 
-       free(h);
+void
+html_reset(void *p)
+{
+       html_reset_internal(p);
+       mandoc_ohash_init(&id_unique, 4, 0);
+}
+
+void
+html_free(void *p)
+{
+       html_reset_internal(p);
+       free(p);
 }
 
 void
@@ -162,31 +185,44 @@ print_gen_head(struct html *h)
        struct tag      *t;
 
        print_otag(h, TAG_META, "?", "charset", "utf-8");
+       if (h->style != NULL) {
+               print_otag(h, TAG_LINK, "?h??", "rel", "stylesheet",
+                   h->style, "type", "text/css", "media", "all");
+               return;
+       }
 
        /*
-        * Print a default style-sheet.
+        * Print a minimal embedded style sheet.
         */
 
        t = print_otag(h, TAG_STYLE, "");
        print_text(h, "table.head, table.foot { width: 100%; }");
-       html_endline(h);
+       print_endline(h);
        print_text(h, "td.head-rtitle, td.foot-os { text-align: right; }");
-       html_endline(h);
+       print_endline(h);
        print_text(h, "td.head-vol { text-align: center; }");
-       html_endline(h);
-       print_text(h, "table.foot td { width: 50%; }");
-       html_endline(h);
-       print_text(h, "table.head td { width: 33%; }");
-       html_endline(h);
-       print_text(h, "div.spacer { margin: 1em 0; }");
+       print_endline(h);
+       print_text(h, "div.Pp { margin: 1ex 0ex; }");
+       print_endline(h);
+       print_text(h, "div.Nd, div.Bf, div.Op { display: inline; }");
+       print_endline(h);
+       print_text(h, "span.Pa, span.Ad { font-style: italic; }");
+       print_endline(h);
+       print_text(h, "span.Ms { font-weight: bold; }");
+       print_endline(h);
+       print_text(h, "dl.Bl-diag ");
+       print_byte(h, '>');
+       print_text(h, " dt { font-weight: bold; }");
+       print_endline(h);
+       print_text(h, "code.Nm, code.Fl, code.Cm, code.Ic, "
+           "code.In, code.Fd, code.Fn,");
+       print_endline(h);
+       print_text(h, "code.Cd { font-weight: bold; "
+           "font-family: inherit; }");
        print_tagq(h, t);
-
-       if (h->style)
-               print_otag(h, TAG_LINK, "?h??", "rel", "stylesheet",
-                   h->style, "type", "text/css", "media", "all");
 }
 
-static void
+void
 print_metaf(struct html *h, enum mandoc_esc deco)
 {
        enum htmlfont    font;
@@ -204,12 +240,15 @@ print_metaf(struct html *h, enum mandoc_esc deco)
        case ESCAPE_FONTBI:
                font = HTMLFONT_BI;
                break;
+       case ESCAPE_FONTCW:
+               font = HTMLFONT_CW;
+               break;
        case ESCAPE_FONT:
        case ESCAPE_FONTROMAN:
                font = HTMLFONT_NONE;
                break;
        default:
-               abort();
+               return;
        }
 
        if (h->metaf) {
@@ -231,84 +270,147 @@ print_metaf(struct html *h, enum mandoc_esc deco)
                h->metaf = print_otag(h, TAG_B, "");
                print_otag(h, TAG_I, "");
                break;
+       case HTMLFONT_CW:
+               h->metaf = print_otag(h, TAG_SPAN, "c", "Li");
+               break;
        default:
                break;
        }
 }
 
-int
-html_strlen(const char *cp)
+void
+html_close_paragraph(struct html *h)
 {
-       size_t           rsz;
-       int              skip, sz;
-
-       /*
-        * Account for escaped sequences within string length
-        * calculations.  This follows the logic in term_strlen() as we
-        * must calculate the width of produced strings.
-        * Assume that characters are always width of "1".  This is
-        * hacky, but it gets the job done for approximation of widths.
-        */
+       struct tag      *t;
 
-       sz = 0;
-       skip = 0;
-       while (1) {
-               rsz = strcspn(cp, "\\");
-               if (rsz) {
-                       cp += rsz;
-                       if (skip) {
-                               skip = 0;
-                               rsz--;
-                       }
-                       sz += rsz;
+       for (t = h->tag; t != NULL && t->closed == 0; t = t->next) {
+               switch(t->tag) {
+               case TAG_P:
+               case TAG_PRE:
+                       print_tagq(h, t);
+                       break;
+               case TAG_A:
+                       print_tagq(h, t);
+                       continue;
+               default:
+                       continue;
                }
-               if ('\0' == *cp)
+               break;
+       }
+}
+
+/*
+ * ROFF_nf switches to no-fill mode, ROFF_fi to fill mode.
+ * TOKEN_NONE does not switch.  The old mode is returned.
+ */
+enum roff_tok
+html_fillmode(struct html *h, enum roff_tok want)
+{
+       struct tag      *t;
+       enum roff_tok    had;
+
+       for (t = h->tag; t != NULL; t = t->next)
+               if (t->tag == TAG_PRE)
                        break;
-               cp++;
-               switch (mandoc_escape(&cp, NULL, NULL)) {
-               case ESCAPE_ERROR:
-                       return sz;
-               case ESCAPE_UNICODE:
-               case ESCAPE_NUMBERED:
-               case ESCAPE_SPECIAL:
-               case ESCAPE_OVERSTRIKE:
-                       if (skip)
-                               skip = 0;
-                       else
-                               sz++;
+
+       had = t == NULL ? ROFF_fi : ROFF_nf;
+
+       if (want != had) {
+               switch (want) {
+               case ROFF_fi:
+                       print_tagq(h, t);
                        break;
-               case ESCAPE_SKIPCHAR:
-                       skip = 1;
+               case ROFF_nf:
+                       html_close_paragraph(h);
+                       print_otag(h, TAG_PRE, "");
                        break;
-               default:
+               case TOKEN_NONE:
                        break;
+               default:
+                       abort();
+               }
+       }
+       return had;
+}
+
+char *
+html_make_id(const struct roff_node *n, int unique)
+{
+       const struct roff_node  *nch;
+       char                    *buf, *bufs, *cp;
+       unsigned int             slot;
+       int                      suffix;
+
+       for (nch = n->child; nch != NULL; nch = nch->next)
+               if (nch->type != ROFFT_TEXT)
+                       return NULL;
+
+       buf = NULL;
+       deroff(&buf, n);
+       if (buf == NULL)
+               return NULL;
+
+       /*
+        * In ID attributes, only use ASCII characters that are
+        * permitted in URL-fragment strings according to the
+        * explicit list at:
+        * https://url.spec.whatwg.org/#url-fragment-string
+        */
+
+       for (cp = buf; *cp != '\0'; cp++)
+               if (isalnum((unsigned char)*cp) == 0 &&
+                   strchr("!$&'()*+,-./:;=?@_~", *cp) == NULL)
+                       *cp = '_';
+
+       if (unique == 0)
+               return buf;
+
+       /* Avoid duplicate HTML id= attributes. */
+
+       bufs = NULL;
+       suffix = 1;
+       slot = ohash_qlookup(&id_unique, buf);
+       cp = ohash_find(&id_unique, slot);
+       if (cp != NULL) {
+               while (cp != NULL) {
+                       free(bufs);
+                       if (++suffix > 127) {
+                               free(buf);
+                               return NULL;
+                       }
+                       mandoc_asprintf(&bufs, "%s_%d", buf, suffix);
+                       slot = ohash_qlookup(&id_unique, bufs);
+                       cp = ohash_find(&id_unique, slot);
                }
+               free(buf);
+               buf = bufs;
        }
-       return sz;
+       ohash_insert(&id_unique, slot, buf);
+       return buf;
 }
 
 static int
-print_escape(char c)
+print_escape(struct html *h, char c)
 {
 
        switch (c) {
        case '<':
-               printf("&lt;");
+               print_word(h, "&lt;");
                break;
        case '>':
-               printf("&gt;");
+               print_word(h, "&gt;");
                break;
        case '&':
-               printf("&amp;");
+               print_word(h, "&amp;");
                break;
        case '"':
-               printf("&quot;");
+               print_word(h, "&quot;");
                break;
        case ASCII_NBRSP:
-               printf("&nbsp;");
+               print_word(h, "&nbsp;");
                break;
        case ASCII_HYPH:
-               putchar('-');
+               print_byte(h, '-');
                break;
        case ASCII_BREAK:
                break;
@@ -321,16 +423,18 @@ print_escape(char c)
 static int
 print_encode(struct html *h, const char *p, const char *pend, int norecurse)
 {
-       size_t           sz;
-       int              c, len, nospace;
+       char             numbuf[16];
        const char      *seq;
+       size_t           sz;
+       int              c, len, breakline, nospace;
        enum mandoc_esc  esc;
-       static const char rejs[9] = { '\\', '<', '>', '&', '"',
+       static const char rejs[10] = { ' ', '\\', '<', '>', '&', '"',
                ASCII_NBRSP, ASCII_HYPH, ASCII_BREAK, '\0' };
 
        if (pend == NULL)
                pend = strchr(p, '\0');
 
+       breakline = 0;
        nospace = 0;
 
        while (p < pend) {
@@ -340,36 +444,50 @@ print_encode(struct html *h, const char *p, const char *pend, int norecurse)
                        continue;
                }
 
-               sz = strcspn(p, rejs);
-               if (p + sz > pend)
-                       sz = pend - p;
+               for (sz = strcspn(p, rejs); sz-- && p < pend; p++)
+                       print_byte(h, *p);
 
-               fwrite(p, 1, sz, stdout);
-               p += (int)sz;
+               if (breakline &&
+                   (p >= pend || *p == ' ' || *p == ASCII_NBRSP)) {
+                       print_otag(h, TAG_BR, "");
+                       breakline = 0;
+                       while (p < pend && (*p == ' ' || *p == ASCII_NBRSP))
+                               p++;
+                       continue;
+               }
 
                if (p >= pend)
                        break;
 
-               if (print_escape(*p++))
+               if (*p == ' ') {
+                       print_endword(h);
+                       p++;
                        continue;
+               }
 
-               esc = mandoc_escape(&p, &seq, &len);
-               if (ESCAPE_ERROR == esc)
-                       break;
+               if (print_escape(h, *p++))
+                       continue;
 
+               esc = mandoc_escape(&p, &seq, &len);
                switch (esc) {
                case ESCAPE_FONT:
                case ESCAPE_FONTPREV:
                case ESCAPE_FONTBOLD:
                case ESCAPE_FONTITALIC:
                case ESCAPE_FONTBI:
+               case ESCAPE_FONTCW:
                case ESCAPE_FONTROMAN:
-                       if (0 == norecurse)
+                       if (0 == norecurse) {
+                               h->flags |= HTML_NOSPACE;
                                print_metaf(h, esc);
+                               h->flags &= ~HTML_NOSPACE;
+                       }
                        continue;
                case ESCAPE_SKIPCHAR:
                        h->flags |= HTML_SKIPCHAR;
                        continue;
+               case ESCAPE_ERROR:
+                       continue;
                default:
                        break;
                }
@@ -394,6 +512,15 @@ print_encode(struct html *h, const char *p, const char *pend, int norecurse)
                        if (c <= 0)
                                continue;
                        break;
+               case ESCAPE_UNDEF:
+                       c = *seq;
+                       break;
+               case ESCAPE_DEVICE:
+                       print_word(h, "html");
+                       continue;
+               case ESCAPE_BREAK:
+                       breakline = 1;
+                       continue;
                case ESCAPE_NOSPACE:
                        if ('\0' == *p)
                                nospace = 1;
@@ -409,10 +536,11 @@ print_encode(struct html *h, const char *p, const char *pend, int norecurse)
                if ((c < 0x20 && c != 0x09) ||
                    (c > 0x7E && c < 0xA0))
                        c = 0xFFFD;
-               if (c > 0x7E)
-                       printf("&#%d;", c);
-               else if ( ! print_escape(c))
-                       putchar(c);
+               if (c > 0x7E) {
+                       (void)snprintf(numbuf, sizeof(numbuf), "&#x%.4X;", c);
+                       print_word(h, numbuf);
+               } else if (print_escape(h, c) == 0)
+                       print_byte(h, c);
        }
 
        return nospace;
@@ -421,14 +549,26 @@ print_encode(struct html *h, const char *p, const char *pend, int norecurse)
 static void
 print_href(struct html *h, const char *name, const char *sec, int man)
 {
+       struct stat      sb;
        const char      *p, *pp;
+       char            *filename;
+
+       if (man) {
+               pp = h->base_man1;
+               if (h->base_man2 != NULL) {
+                       mandoc_asprintf(&filename, "%s.%s", name, sec);
+                       if (stat(filename, &sb) == -1)
+                               pp = h->base_man2;
+                       free(filename);
+               }
+       } else
+               pp = h->base_includes;
 
-       pp = man ? h->base_man : h->base_includes;
        while ((p = strchr(pp, '%')) != NULL) {
                print_encode(h, pp, p, 1);
                if (man && p[1] == 'S') {
                        if (sec == NULL)
-                               putchar('1');
+                               print_byte(h, '1');
                        else
                                print_encode(h, sec, NULL, 1);
                } else if ((man && p[1] == 'N') ||
@@ -446,36 +586,36 @@ struct tag *
 print_otag(struct html *h, enum htmltag tag, const char *fmt, ...)
 {
        va_list          ap;
-       struct roffsu    mysu, *su;
        struct tag      *t;
        const char      *attr;
-       char            *s;
-       double           v;
-       int              i, have_style, tflags;
+       char            *arg1, *arg2;
+       int              style_written, tflags;
 
        tflags = htmltags[tag].flags;
 
-       /* Push this tags onto the stack of open scopes. */
+       /* Push this tag onto the stack of open scopes. */
 
        if ((tflags & HTML_NOSTACK) == 0) {
                t = mandoc_malloc(sizeof(struct tag));
                t->tag = tag;
-               t->next = h->tags.head;
-               h->tags.head = t;
+               t->next = h->tag;
+               t->refcnt = 0;
+               t->closed = 0;
+               h->tag = t;
        } else
                t = NULL;
 
        if (tflags & HTML_NLBEFORE)
-               html_endline(h);
-       if (h->flags & HTML_NLDONE)
-               html_indent(h);
+               print_endline(h);
+       if (h->col == 0)
+               print_indent(h);
        else if ((h->flags & HTML_NOSPACE) == 0) {
                if (h->flags & HTML_KEEP)
-                       printf("&#160;");
+                       print_word(h, "&#x00A0;");
                else {
                        if (h->flags & HTML_PREKEEP)
                                h->flags |= HTML_KEEP;
-                       putchar(' ');
+                       print_endword(h);
                }
        }
 
@@ -486,19 +626,17 @@ print_otag(struct html *h, enum htmltag tag, const char *fmt, ...)
 
        /* Print out the tag name and attributes. */
 
-       printf("<%s", htmltags[tag].name);
+       print_byte(h, '<');
+       print_word(h, htmltags[tag].name);
 
        va_start(ap, fmt);
 
-       have_style = 0;
-       while (*fmt != '\0') {
-               if (*fmt == 's') {
-                       printf(" style=\"");
-                       have_style = 1;
-                       fmt++;
-                       break;
-               }
-               s = va_arg(ap, char *);
+       while (*fmt != '\0' && *fmt != 's') {
+
+               /* Parse attributes and arguments. */
+
+               arg1 = va_arg(ap, char *);
+               arg2 = NULL;
                switch (*fmt++) {
                case 'c':
                        attr = "class";
@@ -510,115 +648,75 @@ print_otag(struct html *h, enum htmltag tag, const char *fmt, ...)
                        attr = "id";
                        break;
                case '?':
-                       attr = s;
-                       s = va_arg(ap, char *);
+                       attr = arg1;
+                       arg1 = va_arg(ap, char *);
                        break;
                default:
                        abort();
                }
-               printf(" %s=\"", attr);
+               if (*fmt == 'M')
+                       arg2 = va_arg(ap, char *);
+               if (arg1 == NULL)
+                       continue;
+
+               /* Print the attributes. */
+
+               print_byte(h, ' ');
+               print_word(h, attr);
+               print_byte(h, '=');
+               print_byte(h, '"');
                switch (*fmt) {
-               case 'M':
-                       print_href(h, s, va_arg(ap, char *), 1);
+               case 'I':
+                       print_href(h, arg1, NULL, 0);
                        fmt++;
                        break;
-               case 'I':
-                       print_href(h, s, NULL, 0);
+               case 'M':
+                       print_href(h, arg1, arg2, 1);
                        fmt++;
                        break;
                case 'R':
-                       putchar('#');
+                       print_byte(h, '#');
+                       print_encode(h, arg1, NULL, 1);
                        fmt++;
-                       /* FALLTHROUGH */
+                       break;
                default:
-                       print_encode(h, s, NULL, 1);
+                       print_encode(h, arg1, NULL, 1);
                        break;
                }
-               putchar('"');
+               print_byte(h, '"');
        }
 
-       /* Print out styles. */
-
-       s = NULL;
-       su = &mysu;
-       while (*fmt != '\0') {
-
-               /* First letter: input argument type. */
-
-               switch (*fmt++) {
-               case 'h':
-                       i = va_arg(ap, int);
-                       SCALE_HS_INIT(su, i);
-                       break;
-               case 's':
-                       s = va_arg(ap, char *);
-                       break;
-               case 'u':
-                       su = va_arg(ap, struct roffsu *);
-                       break;
-               case 'v':
-                       i = va_arg(ap, int);
-                       SCALE_VS_INIT(su, i);
-                       break;
-               case 'w':
-                       s = va_arg(ap, char *);
-                       a2width(s, su);
-                       break;
-               default:
-                       abort();
-               }
-
-               /* Second letter: style name. */
-
-               switch (*fmt++) {
-               case 'b':
-                       attr = "margin-bottom";
-                       break;
-               case 'h':
-                       attr = "height";
-                       break;
-               case 'i':
-                       attr = "text-indent";
-                       break;
-               case 'l':
-                       attr = "margin-left";
-                       break;
-               case 't':
-                       attr = "margin-top";
-                       break;
-               case 'w':
-                       attr = "width";
-                       break;
-               case 'W':
-                       attr = "min-width";
-                       break;
-               case '?':
-                       printf("%s: %s;", s, va_arg(ap, char *));
+       style_written = 0;
+       while (*fmt++ == 's') {
+               arg1 = va_arg(ap, char *);
+               arg2 = va_arg(ap, char *);
+               if (arg2 == NULL)
                        continue;
-               default:
-                       abort();
+               print_byte(h, ' ');
+               if (style_written == 0) {
+                       print_word(h, "style=\"");
+                       style_written = 1;
                }
-               v = su->scale;
-               if (su->unit == SCALE_MM && (v /= 100.0) == 0.0)
-                       v = 1.0;
-               else if (su->unit == SCALE_BU)
-                       v /= 24.0;
-               printf("%s: %.2f%s;", attr, v, roffscales[su->unit]);
+               print_word(h, arg1);
+               print_byte(h, ':');
+               print_byte(h, ' ');
+               print_word(h, arg2);
+               print_byte(h, ';');
        }
-       if (have_style)
-               putchar('"');
+       if (style_written)
+               print_byte(h, '"');
 
        va_end(ap);
 
        /* Accommodate for "well-formed" singleton escaping. */
 
        if (HTML_AUTOCLOSE & htmltags[tag].flags)
-               putchar('/');
+               print_byte(h, '/');
 
-       putchar('>');
+       print_byte(h, '>');
 
        if (tflags & HTML_NLBEGIN)
-               html_endline(h);
+               print_endline(h);
        else
                h->flags |= HTML_NOSPACE;
 
@@ -635,50 +733,77 @@ print_ctag(struct html *h, struct tag *tag)
 {
        int      tflags;
 
-       /*
-        * Remember to close out and nullify the current
-        * meta-font and table, if applicable.
-        */
-       if (tag == h->metaf)
-               h->metaf = NULL;
-       if (tag == h->tblt)
-               h->tblt = NULL;
-
-       tflags = htmltags[tag->tag].flags;
-
-       if (tflags & HTML_INDENT)
-               h->indent--;
-       if (tflags & HTML_NOINDENT)
-               h->noindent--;
-       if (tflags & HTML_NLEND)
-               html_endline(h);
-       html_indent(h);
-       printf("</%s>", htmltags[tag->tag].name);
-       if (tflags & HTML_NLAFTER)
-               html_endline(h);
-
-       h->tags.head = tag->next;
-       free(tag);
+       if (tag->closed == 0) {
+               tag->closed = 1;
+               if (tag == h->metaf)
+                       h->metaf = NULL;
+               if (tag == h->tblt)
+                       h->tblt = NULL;
+
+               tflags = htmltags[tag->tag].flags;
+               if (tflags & HTML_INDENT)
+                       h->indent--;
+               if (tflags & HTML_NOINDENT)
+                       h->noindent--;
+               if (tflags & HTML_NLEND)
+                       print_endline(h);
+               print_indent(h);
+               print_byte(h, '<');
+               print_byte(h, '/');
+               print_word(h, htmltags[tag->tag].name);
+               print_byte(h, '>');
+               if (tflags & HTML_NLAFTER)
+                       print_endline(h);
+       }
+       if (tag->refcnt == 0) {
+               h->tag = tag->next;
+               free(tag);
+       }
 }
 
 void
 print_gen_decls(struct html *h)
 {
+       print_word(h, "<!DOCTYPE html>");
+       print_endline(h);
+}
 
-       puts("<!DOCTYPE html>");
-       h->flags |= HTML_NLDONE;
+void
+print_gen_comment(struct html *h, struct roff_node *n)
+{
+       int      wantblank;
+
+       print_word(h, "<!-- This is an automatically generated file."
+           "  Do not edit.");
+       h->indent = 1;
+       wantblank = 0;
+       while (n != NULL && n->type == ROFFT_COMMENT) {
+               if (strstr(n->string, "-->") == NULL &&
+                   (wantblank || *n->string != '\0')) {
+                       print_endline(h);
+                       print_indent(h);
+                       print_word(h, n->string);
+                       wantblank = *n->string != '\0';
+               }
+               n = n->next;
+       }
+       if (wantblank)
+               print_endline(h);
+       print_word(h, " -->");
+       print_endline(h);
+       h->indent = 0;
 }
 
 void
 print_text(struct html *h, const char *word)
 {
-       if ((h->flags & (HTML_NLDONE | HTML_NOSPACE)) == 0) {
+       if (h->col && (h->flags & HTML_NOSPACE) == 0) {
                if ( ! (HTML_KEEP & h->flags)) {
                        if (HTML_PREKEEP & h->flags)
                                h->flags |= HTML_KEEP;
-                       putchar(' ');
+                       print_endword(h);
                } else
-                       printf("&#160;");
+                       print_word(h, "&#x00A0;");
        }
 
        assert(NULL == h->metaf);
@@ -693,8 +818,11 @@ print_text(struct html *h, const char *word)
                h->metaf = print_otag(h, TAG_B, "");
                print_otag(h, TAG_I, "");
                break;
+       case HTMLFONT_CW:
+               h->metaf = print_otag(h, TAG_SPAN, "c", "Li");
+               break;
        default:
-               html_indent(h);
+               print_indent(h);
                break;
        }
 
@@ -717,48 +845,112 @@ print_text(struct html *h, const char *word)
 void
 print_tagq(struct html *h, const struct tag *until)
 {
-       struct tag      *tag;
+       struct tag      *this, *next;
 
-       while ((tag = h->tags.head) != NULL) {
-               print_ctag(h, tag);
-               if (until && tag == until)
-                       return;
+       for (this = h->tag; this != NULL; this = next) {
+               next = this == until ? NULL : this->next;
+               print_ctag(h, this);
        }
 }
 
+/*
+ * Close out all open elements up to but excluding suntil.
+ * Note that a paragraph just inside stays open together with it
+ * because paragraphs include subsequent phrasing content.
+ */
 void
 print_stagq(struct html *h, const struct tag *suntil)
 {
-       struct tag      *tag;
+       struct tag      *this, *next;
 
-       while ((tag = h->tags.head) != NULL) {
-               if (suntil && tag == suntil)
-                       return;
-               print_ctag(h, tag);
+       for (this = h->tag; this != NULL; this = next) {
+               next = this->next;
+               if (this == suntil || (next == suntil &&
+                   (this->tag == TAG_P || this->tag == TAG_PRE)))
+                       break;
+               print_ctag(h, this);
        }
 }
 
-void
-print_paragraph(struct html *h)
+
+/***********************************************************************
+ * Low level output functions.
+ * They implement line breaking using a short static buffer.
+ ***********************************************************************/
+
+/*
+ * Buffer one HTML output byte.
+ * If the buffer is full, flush and deactivate it and start a new line.
+ * If the buffer is inactive, print directly.
+ */
+static void
+print_byte(struct html *h, char c)
 {
-       struct tag      *t;
+       if ((h->flags & HTML_BUFFER) == 0) {
+               putchar(c);
+               h->col++;
+               return;
+       }
 
-       t = print_otag(h, TAG_DIV, "c", "spacer");
-       print_tagq(h, t);
+       if (h->col + h->bufcol < sizeof(h->buf)) {
+               h->buf[h->bufcol++] = c;
+               return;
+       }
+
+       putchar('\n');
+       h->col = 0;
+       print_indent(h);
+       putchar(' ');
+       putchar(' ');
+       fwrite(h->buf, h->bufcol, 1, stdout);
+       putchar(c);
+       h->col = (h->indent + 1) * 2 + h->bufcol + 1;
+       h->bufcol = 0;
+       h->flags &= ~HTML_BUFFER;
 }
 
 /*
  * If something was printed on the current output line, end it.
- * Not to be called right after html_indent().
+ * Not to be called right after print_indent().
  */
-static void
-html_endline(struct html *h)
+void
+print_endline(struct html *h)
 {
-       if (h->flags & HTML_NLDONE)
+       if (h->col == 0)
                return;
 
+       if (h->bufcol) {
+               putchar(' ');
+               fwrite(h->buf, h->bufcol, 1, stdout);
+               h->bufcol = 0;
+       }
        putchar('\n');
-       h->flags |= HTML_NLDONE | HTML_NOSPACE;
+       h->col = 0;
+       h->flags |= HTML_NOSPACE;
+       h->flags &= ~HTML_BUFFER;
+}
+
+/*
+ * Flush the HTML output buffer.
+ * If it is inactive, activate it.
+ */
+static void
+print_endword(struct html *h)
+{
+       if (h->noindent) {
+               print_byte(h, ' ');
+               return;
+       }
+
+       if ((h->flags & HTML_BUFFER) == 0) {
+               h->col++;
+               h->flags |= HTML_BUFFER;
+       } else if (h->bufcol) {
+               putchar(' ');
+               fwrite(h->buf, h->bufcol, 1, stdout);
+               h->col += h->bufcol + 1;
+       }
+       h->bufcol = 0;
 }
 
 /*
@@ -768,30 +960,28 @@ html_endline(struct html *h)
  * but do not use print_otag() for producing it.
  */
 static void
-html_indent(struct html *h)
+print_indent(struct html *h)
 {
-       int      i;
+       size_t   i;
 
-       if ((h->flags & HTML_NLDONE) == 0)
+       if (h->col)
                return;
 
-       if (h->noindent == 0)
-               for (i = 0; i < h->indent * 2; i++)
+       if (h->noindent == 0) {
+               h->col = h->indent * 2;
+               for (i = 0; i < h->col; i++)
                        putchar(' ');
-       h->flags &= ~(HTML_NLDONE | HTML_NOSPACE);
+       }
+       h->flags &= ~HTML_NOSPACE;
 }
 
 /*
- * Calculate the scaling unit passed in a `-width' argument.  This uses
- * either a native scaling unit (e.g., 1i, 2m) or the string length of
- * the value.
+ * Print or buffer some characters
+ * depending on the current HTML output buffer state.
  */
 static void
-a2width(const char *p, struct roffsu *su)
+print_word(struct html *h, const char *cp)
 {
-       if (a2roffsu(p, su, SCALE_MAX) < 2) {
-               su->unit = SCALE_EN;
-               su->scale = html_strlen(p);
-       } else if (su->scale < 0.0)
-               su->scale = 0.0;
+       while (*cp != '\0')
+               print_byte(h, *cp++);
 }