Fine-tuned the page breaking algorithm by adding penalties and

[sgt/halibut] / bk_paper.c

/*
 * Paper printing pre-backend for Halibut.
 * 
 * This module does all the processing common to both PostScript
 * and PDF output: selecting fonts, line wrapping and page breaking
 * in accordance with font metrics, laying out the contents and
 * index pages, generally doing all the page layout. After this,
 * bk_ps.c and bk_pdf.c should only need to do linear translations
 * into their literal output format.
 */

/*
 * To be done:
 * 
 *  - implement some simple graphics
 *     * I had an underline below chapter headings in the original
 *       Perl version, and I thought it looked rather nice
 *     * also we need para_Rule.
 * 
 *  - set up contents section now we know what sections begin on
 *    which pages
 * 
 *  - do PDF outline
 * 
 *  - index
 * 
 *  - header/footer? Page numbers at least would be handy. Fully
 *    configurable footer can wait, though.
 * 
 * That should bring us to the same level of functionality that
 * original-Halibut had, and the same in PDF plus the obvious
 * interactive navigation features. After that, in future work:
 * 
 *  - linearised PDF, perhaps?
 * 
 *  - I'm uncertain of whether I need to include a ToUnicode CMap
 *    in each of my font definitions in PDF. Currently things (by
 *    which I mean cut and paste out of acroread) seem to be
 *    working fairly happily without it, but I don't know.
 * 
 *  - configurability
 * 
 *  - title pages
 */

#include <assert.h>
#include <stdio.h>

#include "halibut.h"
#include "paper.h"

static font_data *make_std_font(font_list *fontlist, char const *name);
static void wrap_paragraph(para_data *pdata, word *words,
                           int w, int i1, int i2);
static page_data *page_breaks(line_data *first, line_data *last,
                              int page_height);
static void render_line(line_data *ldata, int left_x, int top_y,
                        xref_dest *dest, keywordlist *keywords);
static int paper_width_simple(para_data *pdata, word *text);
static void code_paragraph(para_data *pdata,
                           font_data *fn, font_data *fi, font_data *fb,
                           int font_size, int indent, word *words);

void *paper_pre_backend(paragraph *sourceform, keywordlist *keywords,
                        indexdata *idx) {
    paragraph *p;
    document *doc;
    int indent, extra_indent, firstline_indent, aux_indent;
    para_data *pdata;
    line_data *ldata, *firstline, *lastline;
    font_data *tr, *ti, *hr, *hi, *cr, *co, *cb;
    page_data *pages;
    font_list *fontlist;
    word *aux, *aux2;

    /*
     * FIXME: All these things ought to become configurable.
     */
    int paper_width = 595 * 4096;
    int paper_height = 841 * 4096;
    int left_margin = 72 * 4096;
    int top_margin = 72 * 4096;
    int right_margin = 72 * 4096;
    int bottom_margin = 108 * 4096;
    int indent_list_bullet = 6 * 4096;
    int indent_list = 24 * 4096;
    int indent_quote = 18 * 4096;
    int base_leading = 4096;
    int base_para_spacing = 10 * 4096;
    int chapter_top_space = 72 * 4096;
    int sect_num_left_space = 12 * 4096;

    int base_width = paper_width - left_margin - right_margin;
    int page_height = paper_height - top_margin - bottom_margin;

    IGNORE(idx);                       /* FIXME */

    /*
     * First, set up some font structures.
     */
    fontlist = mknew(font_list);
    fontlist->head = fontlist->tail = NULL;
    tr = make_std_font(fontlist, "Times-Roman");
    ti = make_std_font(fontlist, "Times-Italic");
    hr = make_std_font(fontlist, "Helvetica-Bold");
    hi = make_std_font(fontlist, "Helvetica-BoldOblique");
    cr = make_std_font(fontlist, "Courier");
    co = make_std_font(fontlist, "Courier-Oblique");
    cb = make_std_font(fontlist, "Courier-Bold");

    /*
     * Go through and break up each paragraph into lines.
     */
    indent = 0;
    firstline = lastline = NULL;
    for (p = sourceform; p; p = p->next) {
        p->private_data = NULL;

        switch (p->type) {
            /*
             * These paragraph types are either invisible or don't
             * define text in the normal sense. Either way, they
             * don't require wrapping.
             */
          case para_IM:
          case para_BR:
          case para_Rule:
          case para_Biblio:
          case para_NotParaType:
          case para_Config:
          case para_VersionID:
          case para_NoCite:
            break;

            /*
             * These paragraph types don't require wrapping, but
             * they do affect the line width to which we wrap the
             * rest of the paragraphs, so we need to pay attention.
             */
          case para_LcontPush:
            indent += indent_list; break;
          case para_LcontPop:
            indent -= indent_list; assert(indent >= 0); break;
          case para_QuotePush:
            indent += indent_quote; break;
          case para_QuotePop:
            indent -= indent_quote; assert(indent >= 0); break;

            /*
             * This paragraph type is special. Process it
             * specially.
             */
          case para_Code:
            pdata = mknew(para_data);
            code_paragraph(pdata, cr, co, cb, 12, indent, p->words);
            p->private_data = pdata;
            if (pdata->first != pdata->last) {
                pdata->first->penalty_after += 100000;
                pdata->last->penalty_before += 100000;
            }
            break;

            /*
             * All of these paragraph types require wrapping in the
             * ordinary way. So we must supply a set of fonts, a
             * line width and auxiliary information (e.g. bullet
             * text) for each one.
             */
          case para_Chapter:
          case para_Appendix:
          case para_UnnumberedChapter:
          case para_Heading:
          case para_Subsect:
          case para_Normal:
          case para_BiblioCited:
          case para_Bullet:
          case para_NumberedList:
          case para_DescribedThing:
          case para_Description:
          case para_Copyright:
          case para_Title:
            pdata = mknew(para_data);

            /*
             * Choose fonts for this paragraph.
             * 
             * FIXME: All of this ought to be completely
             * user-configurable.
             */
            switch (p->type) {
              case para_Title:
                pdata->fonts[FONT_NORMAL] = hr;
                pdata->sizes[FONT_NORMAL] = 24;
                pdata->fonts[FONT_EMPH] = hi;
                pdata->sizes[FONT_EMPH] = 24;
                pdata->fonts[FONT_CODE] = cb;
                pdata->sizes[FONT_CODE] = 24;
                break;

              case para_Chapter:
              case para_Appendix:
              case para_UnnumberedChapter:
                pdata->fonts[FONT_NORMAL] = hr;
                pdata->sizes[FONT_NORMAL] = 20;
                pdata->fonts[FONT_EMPH] = hi;
                pdata->sizes[FONT_EMPH] = 20;
                pdata->fonts[FONT_CODE] = cb;
                pdata->sizes[FONT_CODE] = 20;
                break;

              case para_Heading:
              case para_Subsect:
                pdata->fonts[FONT_NORMAL] = hr;
                pdata->fonts[FONT_EMPH] = hi;
                pdata->fonts[FONT_CODE] = cb;
                pdata->sizes[FONT_NORMAL] =
                    pdata->sizes[FONT_EMPH] =
                    pdata->sizes[FONT_CODE] =
                    (p->aux == 0 ? 16 : p->aux == 1 ? 14 : 13);
                break;

              case para_Normal:
              case para_BiblioCited:
              case para_Bullet:
              case para_NumberedList:
              case para_DescribedThing:
              case para_Description:
              case para_Copyright:
                pdata->fonts[FONT_NORMAL] = tr;
                pdata->sizes[FONT_NORMAL] = 12;
                pdata->fonts[FONT_EMPH] = ti;
                pdata->sizes[FONT_EMPH] = 12;
                pdata->fonts[FONT_CODE] = cr;
                pdata->sizes[FONT_CODE] = 12;
                break;
            }

            /*
             * Also select an indentation level depending on the
             * paragraph type (list paragraphs other than
             * para_DescribedThing need extra indent).
             * 
             * (FIXME: Perhaps at some point we might even arrange
             * for the user to be able to request indented first
             * lines in paragraphs.)
             */
            if (p->type == para_Bullet ||
                p->type == para_NumberedList ||
                p->type == para_Description) {
                extra_indent = firstline_indent = indent_list;
            } else {
                extra_indent = firstline_indent = 0;
            }

            /*
             * Find the auxiliary text for this paragraph.
             */
            aux = aux2 = NULL;
            aux_indent = 0;

            switch (p->type) {
              case para_Chapter:
              case para_Appendix:
              case para_Heading:
              case para_Subsect:
                /*
                 * For some heading styles (FIXME: be able to
                 * configure which), the auxiliary text contains
                 * the chapter number and is arranged to be
                 * right-aligned a few points left of the primary
                 * margin. For other styles, the auxiliary text is
                 * the full chapter _name_ and takes up space
                 * within the (wrapped) chapter title, meaning that
                 * we must move the first line indent over to make
                 * space for it.
                 */
                if (p->type == para_Heading || p->type == para_Subsect) {
                    int len;

                    aux = p->kwtext2;
                    len = paper_width_simple(pdata, p->kwtext2);
                    aux_indent = -len - sect_num_left_space;
                } else {
                    aux = p->kwtext;
                    aux2 = mknew(word);
                    aux2->next = NULL;
                    aux2->alt = NULL;
                    aux2->type = word_Normal;
                    aux2->text = ustrdup(L": ");
                    aux2->breaks = FALSE;
                    aux2->aux = 0;
                    aux_indent = 0;

                    firstline_indent += paper_width_simple(pdata, aux);
                    firstline_indent += paper_width_simple(pdata, aux2);
                }
                break;

              case para_Bullet:
                /*
                 * Auxiliary text consisting of a bullet. (FIXME:
                 * configurable bullet.)
                 */
                aux = mknew(word);
                aux->next = NULL;
                aux->alt = NULL;
                aux->type = word_Normal;
                aux->text = ustrdup(L"\x2022");
                aux->breaks = FALSE;
                aux->aux = 0;
                aux_indent = indent + indent_list_bullet;
                break;

              case para_NumberedList:
                /*
                 * Auxiliary text consisting of the number followed
                 * by a (FIXME: configurable) full stop.
                 */
                aux = p->kwtext;
                aux2 = mknew(word);
                aux2->next = NULL;
                aux2->alt = NULL;
                aux2->type = word_Normal;
                aux2->text = ustrdup(L".");
                aux2->breaks = FALSE;
                aux2->aux = 0;
                aux_indent = indent + indent_list_bullet;
                break;

              case para_BiblioCited:
                /*
                 * Auxiliary text consisting of the bibliography
                 * reference text, and a trailing space.
                 */
                aux = p->kwtext;
                aux2 = mknew(word);
                aux2->next = NULL;
                aux2->alt = NULL;
                aux2->type = word_Normal;
                aux2->text = ustrdup(L" ");
                aux2->breaks = FALSE;
                aux2->aux = 0;
                aux_indent = indent;
                firstline_indent += paper_width_simple(pdata, aux);
                firstline_indent += paper_width_simple(pdata, aux2);
                break;
            }

            wrap_paragraph(pdata, p->words, base_width,
                           indent + firstline_indent,
                           indent + extra_indent);

            p->private_data = pdata;

            pdata->first->aux_text = aux;
            pdata->first->aux_text_2 = aux2;
            pdata->first->aux_left_indent = aux_indent;

            /*
             * Line breaking penalties.
             */
            switch (p->type) {
              case para_Chapter:
              case para_Appendix:
              case para_Heading:
              case para_Subsect:
              case para_UnnumberedChapter:
                /*
                 * Fixed and large penalty for breaking straight
                 * after a heading; corresponding bonus for
                 * breaking straight before.
                 */
                pdata->first->penalty_before = -500000;
                pdata->last->penalty_after = 500000;
                for (ldata = pdata->first; ldata; ldata = ldata->next)
                    ldata->penalty_after = 500000;
                break;

              case para_DescribedThing:
                /*
                 * This is treated a bit like a small heading:
                 * there's a penalty for breaking after it (i.e.
                 * between it and its description), and a bonus for
                 * breaking before it (actually _between_ list
                 * items).
                 */
                pdata->first->penalty_before = -200000;
                pdata->last->penalty_after = 200000;
                break;

              default:
                /*
                 * Most paragraph types: widow/orphan control by
                 * discouraging breaking one line from the end of
                 * any paragraph.
                 */
                if (pdata->first != pdata->last) {
                    pdata->first->penalty_after = 100000;
                    pdata->last->penalty_before = 100000;
                }
                break;
            }

            break;
        }

        if (p->private_data) {
            pdata = (para_data *)p->private_data;

            /*
             * Set the line spacing for each line in this paragraph.
             */
            for (ldata = pdata->first; ldata; ldata = ldata->next) {
                if (ldata == pdata->first)
                    ldata->space_before = base_para_spacing / 2;
                else
                    ldata->space_before = base_leading / 2;
                if (ldata == pdata->last)
                    ldata->space_after = base_para_spacing / 2;
                else
                    ldata->space_after = base_leading / 2;
                ldata->page_break = FALSE;
            }

            /*
             * Some kinds of section heading do require a page
             * break before them.
             */
            if (p->type == para_Title ||
                p->type == para_Chapter ||
                p->type == para_Appendix ||
                p->type == para_UnnumberedChapter) {
                pdata->first->page_break = TRUE;
                pdata->first->space_before = chapter_top_space;
            }

            /*
             * Link all line structures together into a big list.
             */
            if (pdata->first) {
                if (lastline) {
                    lastline->next = pdata->first;
                    pdata->first->prev = lastline;
                } else {
                    firstline = pdata->first;
                    pdata->first->prev = NULL;
                }
                lastline = pdata->last;
            }
        }
    }

    /*
     * Now we have an enormous linked list of every line of text in
     * the document. Break it up into pages.
     */
    pages = page_breaks(firstline, lastline, page_height);

    /*
     * Now we're ready to actually lay out the pages. We do this by
     * looping over _paragraphs_, since we may need to track cross-
     * references between lines and even across pages.
     */
    for (p = sourceform; p; p = p->next) {
        pdata = (para_data *)p->private_data;

        if (pdata) {
            xref_dest dest;
            dest.type = NONE;
            for (ldata = pdata->first; ldata; ldata = ldata->next) {
                render_line(ldata, left_margin, paper_height - top_margin,
                            &dest, keywords);
                if (ldata == pdata->last)
                    break;
            }
        }
    }

    doc = mknew(document);
    doc->fonts = fontlist;
    doc->pages = pages;
    doc->paper_width = paper_width;
    doc->paper_height = paper_height;
    return doc;
}

static font_encoding *new_font_encoding(font_data *font)
{
    font_encoding *fe;
    int i;

    fe = mknew(font_encoding);
    fe->next = NULL;

    if (font->list->tail)
        font->list->tail->next = fe;
    else
        font->list->head = fe;
    font->list->tail = fe;

    fe->font = font;
    fe->free_pos = 0x21;

    for (i = 0; i < 256; i++) {
        fe->vector[i] = NULL;
        fe->indices[i] = -1;
        fe->to_unicode[i] = 0xFFFF;
    }

    return fe;
}

static font_data *make_std_font(font_list *fontlist, char const *name)
{
    const int *widths;
    int nglyphs;
    font_data *f;
    font_encoding *fe;
    int i;

    widths = ps_std_font_widths(name);
    if (!widths)
        return NULL;

    for (nglyphs = 0; ps_std_glyphs[nglyphs] != NULL; nglyphs++);

    f = mknew(font_data);

    f->list = fontlist;
    f->name = name;
    f->nglyphs = nglyphs;
    f->glyphs = ps_std_glyphs;
    f->widths = widths;
    f->subfont_map = mknewa(subfont_map_entry, nglyphs);

    /*
     * Our first subfont will contain all of US-ASCII. This isn't
     * really necessary - we could just create custom subfonts
     * precisely as the whim of render_string dictated - but
     * instinct suggests that it might be nice to have the text in
     * the output files look _marginally_ recognisable.
     */
    fe = new_font_encoding(f);
    fe->free_pos = 0xA1;               /* only the top half is free */
    f->latest_subfont = fe;

    for (i = 0; i < (int)lenof(f->bmp); i++)
        f->bmp[i] = 0xFFFF;

    for (i = 0; i < nglyphs; i++) {
        wchar_t ucs;
        ucs = ps_glyph_to_unicode(f->glyphs[i]);
        assert(ucs != 0xFFFF);
        f->bmp[ucs] = i;
        if (ucs >= 0x20 && ucs <= 0x7E) {
            fe->vector[ucs] = f->glyphs[i];
            fe->indices[ucs] = i;
            fe->to_unicode[ucs] = ucs;
            f->subfont_map[i].subfont = fe;
            f->subfont_map[i].position = ucs;
        } else {
            /*
             * This character is not yet assigned to a subfont.
             */
            f->subfont_map[i].subfont = NULL;
            f->subfont_map[i].position = 0;
        }
    }

    return f;
}

static int string_width(font_data *font, wchar_t const *string, int *errs)
{
    int width = 0;

    if (errs)
        *errs = 0;

    for (; *string; string++) {
        int index;

        index = font->bmp[(unsigned short)*string];
        if (index == 0xFFFF) {
            if (errs)
                *errs = 1;
        } else {
            width += font->widths[index];
        }
    }

    return width;
}

static int paper_width_internal(void *vctx, word *word, int *nspaces);

struct paper_width_ctx {
    int minspacewidth;
    para_data *pdata;
};

static int paper_width_list(void *vctx, word *text, word *end, int *nspaces) {
    int w = 0;
    while (text && text != end) {
        w += paper_width_internal(vctx, text, nspaces);
        text = text->next;
    }
    return w;
}

static int paper_width_internal(void *vctx, word *word, int *nspaces)
{
    struct paper_width_ctx *ctx = (struct paper_width_ctx *)vctx;
    int style, type, findex, width, errs;
    wchar_t *str;

    switch (word->type) {
      case word_HyperLink:
      case word_HyperEnd:
      case word_UpperXref:
      case word_LowerXref:
      case word_XrefEnd:
      case word_IndexRef:
        return 0;
    }

    style = towordstyle(word->type);
    type = removeattr(word->type);

    findex = (style == word_Normal ? FONT_NORMAL :
              style == word_Emph ? FONT_EMPH :
              FONT_CODE);

    if (type == word_Normal) {
        str = word->text;
    } else if (type == word_WhiteSpace) {
        if (findex != FONT_CODE) {
            if (nspaces)
                (*nspaces)++;
            return ctx->minspacewidth;
        } else
            str = L" ";
    } else /* if (type == word_Quote) */ {
        if (word->aux == quote_Open)
            str = L"\x2018";           /* FIXME: configurability! */
        else
            str = L"\x2019";           /* FIXME: configurability! */
    }

    width = string_width(ctx->pdata->fonts[findex], str, &errs);

    if (errs && word->alt)
        return paper_width_list(vctx, word->alt, NULL, nspaces);
    else
        return ctx->pdata->sizes[findex] * width;
}

static int paper_width(void *vctx, word *word)
{
    return paper_width_internal(vctx, word, NULL);
}

static int paper_width_simple(para_data *pdata, word *text)
{
    struct paper_width_ctx ctx;

    ctx.pdata = pdata;
    ctx.minspacewidth =
        (pdata->sizes[FONT_NORMAL] *
         string_width(pdata->fonts[FONT_NORMAL], L" ", NULL));

    return paper_width_list(&ctx, text, NULL, NULL);
}

static void wrap_paragraph(para_data *pdata, word *words,
                           int w, int i1, int i2)
{
    wrappedline *wrapping, *p;
    int spacewidth;
    struct paper_width_ctx ctx;
    int line_height;

    /*
     * We're going to need to store the line height in every line
     * structure we generate.
     */
    {
        int i;
        line_height = 0;
        for (i = 0; i < NFONTS; i++)
            if (line_height < pdata->sizes[i])
                line_height = pdata->sizes[i];
        line_height *= 4096;
    }

    spacewidth = (pdata->sizes[FONT_NORMAL] *
                  string_width(pdata->fonts[FONT_NORMAL], L" ", NULL));
    if (spacewidth == 0) {
        /*
         * A font without a space?! Disturbing. I hope this never
         * comes up, but I'll make a random guess anyway and set my
         * space width to half the point size.
         */
        spacewidth = pdata->sizes[FONT_NORMAL] * 4096 / 2;
    }

    /*
     * I'm going to set the _minimum_ space width to 3/5 of the
     * standard one, and use the standard one as the optimum.
     */
    ctx.minspacewidth = spacewidth * 3 / 5;
    ctx.pdata = pdata;

    wrapping = wrap_para(words, w - i1, w - i2, paper_width, &ctx, spacewidth);

    /*
     * Having done the wrapping, we now concoct a set of line_data
     * structures.
     */
    pdata->first = pdata->last = NULL;

    for (p = wrapping; p; p = p->next) {
        line_data *ldata;
        word *wd;
        int len, wid, spaces;

        ldata = mknew(line_data);

        ldata->pdata = pdata;
        ldata->first = p->begin;
        ldata->end = p->end;
        ldata->line_height = line_height;

        ldata->xpos = (p == wrapping ? i1 : i2);

        if (pdata->last) {
            pdata->last->next = ldata;
            ldata->prev = pdata->last;
        } else {
            pdata->first = ldata;
            ldata->prev = NULL;
        }
        ldata->next = NULL;
        pdata->last = ldata;

        spaces = 0;
        len = paper_width_list(&ctx, ldata->first, ldata->end, &spaces);
        wid = (p == wrapping ? w - i1 : w - i2);
        wd = ldata->first;

        ldata->hshortfall = wid - len;
        ldata->nspaces = spaces;
        /*
         * This tells us how much the space width needs to
         * change from _min_spacewidth. But we want to store
         * its difference from the _natural_ space width, to
         * make the text rendering easier.
         */
        ldata->hshortfall += ctx.minspacewidth * spaces;
        ldata->hshortfall -= spacewidth * spaces;
        /*
         * Special case: on the last line of a paragraph, we
         * never stretch spaces.
         */
        if (ldata->hshortfall > 0 && !p->next)
            ldata->hshortfall = 0;

        ldata->aux_text = NULL;
        ldata->aux_text_2 = NULL;
        ldata->aux_left_indent = 0;
        ldata->penalty_before = ldata->penalty_after = 0;
    }

}

static page_data *page_breaks(line_data *first, line_data *last,
                              int page_height)
{
    line_data *l, *m;
    page_data *ph, *pt;

    /*
     * Page breaking is done by a close analogue of the optimal
     * paragraph wrapping algorithm used by wrap_para(). We work
     * backwards from the end of the document line by line; for
     * each line, we contemplate every possible number of lines we
     * could put on a page starting with that line, determine a
     * cost function for each one, add it to the pre-computed cost
     * function for optimally page-breaking everything after that
     * page, and pick the best option.
     * 
     * Since my line_data structures are only used for this
     * purpose, I might as well just store the algorithm data
     * directly in them.
     */

    for (l = last; l; l = l->prev) {
        int minheight, text = 0, space = 0;
        int cost;

        l->bestcost = -1;
        for (m = l; m; m = m->next) {
            if (m != l && m->page_break)
                break;                 /* we've gone as far as we can */

            if (m != l)
                space += m->prev->space_after;
            if (m != l || m->page_break)
                space += m->space_before;
            text += m->line_height;
            minheight = text + space;

            if (m != l && minheight > page_height)
                break;

            /*
             * Compute the cost of this arrangement, as the square
             * of the amount of wasted space on the page.
             * Exception: if this is the last page before a
             * mandatory break or the document end, we don't
             * penalise a large blank area.
             */
            if (m->next && !m->next->page_break)
            {
                int x = page_height - minheight;
                int xf;

                xf = x & 0xFF;
                x >>= 8;

                cost = x*x;
                cost += (x * xf) >> 8;
            } else
                cost = 0;

            if (m->next && !m->next->page_break) {
                cost += m->penalty_after;
                cost += m->next->penalty_before;
            }

            if (m->next && !m->next->page_break)
                cost += m->next->bestcost;
            if (l->bestcost == -1 || l->bestcost > cost) {
                /*
                 * This is the best option yet for this starting
                 * point.
                 */
                l->bestcost = cost;
                if (m->next && !m->next->page_break)
                    l->vshortfall = page_height - minheight;
                else
                    l->vshortfall = 0;
                l->text = text;
                l->space = space;
                l->page_last = m;
            }
        }
    }

    /*
     * Now go through the line list forwards and assemble the
     * actual pages.
     */
    ph = pt = NULL;

    l = first;
    while (l) {
        page_data *page;
        int text, space;

        page = mknew(page_data);
        page->next = NULL;
        page->prev = pt;
        if (pt)
            pt->next = page;
        else
            ph = page;
        pt = page;

        page->first_line = l;
        page->last_line = l->page_last;

        page->first_text = page->last_text = NULL;

        page->first_xref = page->last_xref = NULL;

        /*
         * Now assign a y-coordinate to each line on the page.
         */
        text = space = 0;
        for (l = page->first_line; l; l = l->next) {
            if (l != page->first_line)
                space += l->prev->space_after;
            if (l != page->first_line || l->page_break)
                space += l->space_before;
            text += l->line_height;

            l->page = page;
            l->ypos = text + space +
                space * (float)page->first_line->vshortfall /
                page->first_line->space;

            if (l == page->last_line)
                break;
        }

        l = page->last_line->next;
    }

    return ph;
}

static void add_string_to_page(page_data *page, int x, int y,
                               font_encoding *fe, int size, char *text)
{
    text_fragment *frag;

    frag = mknew(text_fragment);
    frag->next = NULL;

    if (page->last_text)
        page->last_text->next = frag;
    else
        page->first_text = frag;
    page->last_text = frag;

    frag->x = x;
    frag->y = y;
    frag->fe = fe;
    frag->fontsize = size;
    frag->text = dupstr(text);
}

/*
 * Returns the updated x coordinate.
 */
static int render_string(page_data *page, font_data *font, int fontsize,
                         int x, int y, wchar_t *str)
{
    char *text;
    int textpos, textwid, glyph;
    font_encoding *subfont = NULL, *sf;

    text = mknewa(char, 1 + ustrlen(str));
    textpos = textwid = 0;

    while (*str) {
        glyph = font->bmp[*str];

        if (glyph == 0xFFFF)
            continue;                  /* nothing more we can do here */

        /*
         * Find which subfont this character is going in.
         */
        sf = font->subfont_map[glyph].subfont;

        if (!sf) {
            int c;

            /*
             * This character is not yet in a subfont. Assign one.
             */
            if (font->latest_subfont->free_pos >= 0x100)
                font->latest_subfont = new_font_encoding(font);

            c = font->latest_subfont->free_pos++;
            if (font->latest_subfont->free_pos == 0x7F)
                font->latest_subfont->free_pos = 0xA1;

            font->subfont_map[glyph].subfont = font->latest_subfont;
            font->subfont_map[glyph].position = c;
            font->latest_subfont->vector[c] = font->glyphs[glyph];
            font->latest_subfont->indices[c] = glyph;
            font->latest_subfont->to_unicode[c] = *str;

            sf = font->latest_subfont;
        }

        if (!subfont || sf != subfont) {
            if (subfont) {
                text[textpos] = '\0';
                add_string_to_page(page, x, y, subfont, fontsize, text);
                x += textwid;
            } else {
                assert(textpos == 0);
            }
            textpos = 0;
            subfont = sf;
        }

        text[textpos++] = font->subfont_map[glyph].position;
        textwid += font->widths[glyph] * fontsize;

        str++;
    }

    if (textpos > 0) {
        text[textpos] = '\0';
        add_string_to_page(page, x, y, subfont, fontsize, text);
        x += textwid;
    }

    return x;
}

/*
 * Returns the updated x coordinate.
 */
static int render_text(page_data *page, para_data *pdata, line_data *ldata,
                       int x, int y, word *text, word *text_end, xref **xr,
                       int shortfall, int nspaces, int *nspace,
                       keywordlist *keywords)
{
    while (text && text != text_end) {
        int style, type, findex, errs;
        wchar_t *str;
        xref_dest dest;

        switch (text->type) {
            /*
             * Start a cross-reference.
             */
          case word_HyperLink:
          case word_UpperXref:
          case word_LowerXref:

            if (text->type == word_HyperLink) {
                dest.type = URL;
                dest.url = utoa_dup(text->text);
                dest.page = NULL;
            } else {
                keyword *kwl = kw_lookup(keywords, text->text);
                para_data *pdata;

                if (kwl) {
                    assert(kwl->para->private_data);
                    pdata = (para_data *) kwl->para->private_data;
                    dest.type = PAGE;
                    dest.page = pdata->first->page;
                    dest.url = NULL;
                } else {
                    /*
                     * Shouldn't happen, but *shrug*
                     */
                    dest.type = NONE;
                    dest.page = NULL;
                    dest.url = NULL;
                }
            }
            if (dest.type != NONE) {
                *xr = mknew(xref);
                (*xr)->dest = dest;    /* structure copy */
                if (page->last_xref)
                    page->last_xref->next = *xr;
                else
                    page->first_xref = *xr;
                page->last_xref = *xr;

                /*
                 * FIXME: Ideally we should have, and use, some
                 * vertical font metric information here so that
                 * our cross-ref rectangle can take account of
                 * descenders and the font's cap height. This will
                 * do for the moment, but it isn't ideal.
                 */
                (*xr)->lx = (*xr)->rx = x;
                (*xr)->by = y;
                (*xr)->ty = y + ldata->line_height;
            }
            goto nextword;
            
            /*
             * Finish extending a cross-reference box.
             */
          case word_HyperEnd:
          case word_XrefEnd:
            *xr = NULL;
            goto nextword;

          case word_IndexRef:
            goto nextword;
            /*
             * FIXME: we should do something with this.
             */
        }

        style = towordstyle(text->type);
        type = removeattr(text->type);

        findex = (style == word_Normal ? FONT_NORMAL :
                  style == word_Emph ? FONT_EMPH :
                  FONT_CODE);

        if (type == word_Normal) {
            str = text->text;
        } else if (type == word_WhiteSpace) {
            x += pdata->sizes[findex] *
                string_width(pdata->fonts[findex], L" ", NULL);
            if (nspaces && findex != FONT_CODE) {
                x += (*nspace+1) * shortfall / nspaces;
                x -= *nspace * shortfall / nspaces;
                (*nspace)++;
            }
            goto nextword;
        } else /* if (type == word_Quote) */ {
            if (text->aux == quote_Open)
                str = L"\x2018";               /* FIXME: configurability! */
            else
                str = L"\x2019";               /* FIXME: configurability! */
        }

        (void) string_width(pdata->fonts[findex], str, &errs);

        if (errs && text->alt)
            x = render_text(page, pdata, ldata, x, y, text->alt, NULL,
                            xr, shortfall, nspaces, nspace, keywords);
        else
            x = render_string(page, pdata->fonts[findex],
                              pdata->sizes[findex], x, y, str);

        if (*xr)
            (*xr)->rx = x;

        nextword:
        text = text->next;
    }

    return x;
}

static void render_line(line_data *ldata, int left_x, int top_y,
                        xref_dest *dest, keywordlist *keywords)
{
    int nspace;
    xref *xr;
    
    if (ldata->aux_text) {
        int x;
        xr = NULL;
        nspace = 0;
        x = render_text(ldata->page, ldata->pdata, ldata,
                        left_x + ldata->aux_left_indent,
                        top_y - ldata->ypos,
                        ldata->aux_text, NULL, &xr, 0, 0, &nspace, keywords);
        if (ldata->aux_text_2)
            render_text(ldata->page, ldata->pdata, ldata,
                        x, top_y - ldata->ypos,
                        ldata->aux_text_2, NULL, &xr, 0, 0, &nspace, keywords);
    }
    nspace = 0;

    /*
     * There might be a cross-reference carried over from a
     * previous line.
     */
    if (dest->type != NONE) {
        xr = mknew(xref);
        xr->dest = *dest;    /* structure copy */
        if (ldata->page->last_xref)
            ldata->page->last_xref->next = xr;
        else
            ldata->page->first_xref = xr;
        ldata->page->last_xref = xr;
        xr->lx = xr->rx = left_x + ldata->xpos;
        xr->by = top_y - ldata->ypos;
        xr->ty = top_y - ldata->ypos + ldata->line_height;
    } else
        xr = NULL;

    render_text(ldata->page, ldata->pdata, ldata, left_x + ldata->xpos,
                top_y - ldata->ypos, ldata->first, ldata->end, &xr,
                ldata->hshortfall, ldata->nspaces, &nspace, keywords);

    if (xr) {
        /*
         * There's a cross-reference continued on to the next line.
         */
        *dest = xr->dest;
    } else
        dest->type = NONE;
}

static void code_paragraph(para_data *pdata,
                           font_data *fn, font_data *fi, font_data *fb,
                           int font_size, int indent, word *words)
{
    /*
     * For code paragraphs, I'm going to hack grievously and
     * pretend the three normal fonts are the three code paragraph
     * fonts.
     */
    pdata->fonts[FONT_NORMAL] = fb;
    pdata->fonts[FONT_EMPH] = fi;
    pdata->fonts[FONT_CODE] = fn;
    pdata->sizes[FONT_NORMAL] =
        pdata->sizes[FONT_EMPH] =
        pdata->sizes[FONT_CODE] = font_size;

    pdata->first = pdata->last = NULL;

    for (; words; words = words->next) {
        wchar_t *t, *e, *start;
        word *lhead = NULL, *ltail = NULL, *w;
        line_data *ldata;
        int prev = -1, curr;

        t = words->text;
        if (words->next && words->next->type == word_Emph) {
            e = words->next->text;
            words = words->next;
        } else
            e = NULL;

        start = t;

        while (*start) {
            while (*t) {
                if (!e || !*e)
                    curr = 0;
                else if (*e == L'i')
                    curr = 1;
                else if (*e == L'b')
                    curr = 2;
                else
                    curr = 0;

                if (prev < 0)
                    prev = curr;

                if (curr != prev)
                    break;

                t++;
                if (e && *e)
                    e++;
            }

            /*
             * We've isolated a maximal subsequence of the line
             * which has the same emphasis. Form it into a word
             * structure.
             */
            w = mknew(word);
            w->next = NULL;
            w->alt = NULL;
            w->type = (prev == 0 ? word_WeakCode :
                      prev == 1 ? word_Emph : word_Normal);
            w->text = mknewa(wchar_t, t-start+1);
            memcpy(w->text, start, (t-start) * sizeof(wchar_t));
            w->text[t-start] = '\0';
            w->breaks = FALSE;

            if (ltail)
                ltail->next = w;
            else
                lhead = w;
            ltail = w;

            start = t;
            prev = -1;
        }

        ldata = mknew(line_data);

        ldata->pdata = pdata;
        ldata->first = lhead;
        ldata->end = NULL;
        ldata->line_height = font_size * 4096;

        ldata->xpos = indent;

        if (pdata->last) {
            pdata->last->next = ldata;
            ldata->prev = pdata->last;
        } else {
            pdata->first = ldata;
            ldata->prev = NULL;
        }
        ldata->next = NULL;
        pdata->last = ldata;

        ldata->hshortfall = 0;
        ldata->nspaces = 0;
        ldata->aux_text = NULL;
        ldata->aux_text_2 = NULL;
        ldata->aux_left_indent = 0;
        /* General opprobrium for breaking in a code paragraph. */
        ldata->penalty_before = ldata->penalty_after = 50000;
    }
}