bk_text and bk_info both need to know the on-screen width of

[sgt/halibut] / bk_info.c

/*
 * info backend for Halibut
 * 
 * Possible future work:
 * 
 *  - configurable indentation, bullets, emphasis, quotes etc?
 * 
 *  - configurable choice of how to allocate node names?
 *     + possibly a template-like approach, choosing node names to
 *       be the full section title or perhaps the internal keyword?
 *     + neither of those seems quite right. Perhaps instead a
 *       Windows Help-like mechanism, where a magic config
 *       directive allows user choice of name for every node.
 *     + Only trouble with that is, now what happens to the section
 *       numbers? Do they become completely vestigial and just sit
 *       in the title text of each node? Or do we keep them in the
 *       menus somehow? I think people might occasionally want to
 *       go to a section by number, if only because all the _other_
 *       formats of the same document will reference the numbers
 *       all the time. So our menu lines could look like one of
 *       these:
 *        * Nodename: Section 1.2. Title of section.
 *        * Section 1.2: Nodename. Title of section.
 * 
 *  - might be helpful to diagnose duplicate node names!
 */

#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
#include "halibut.h"

typedef struct {
    char *filename;
    int maxfilesize;
    int charset;
} infoconfig;

typedef struct {
    rdstringc output;
    int charset;
    charset_state state;
    int wcmode;
} info_data;
#define EMPTY_INFO_DATA { { 0, 0, NULL }, 0, CHARSET_INIT_STATE, FALSE }
static const info_data empty_info_data = EMPTY_INFO_DATA;

typedef struct node_tag node;
struct node_tag {
    node *listnext;
    node *up, *prev, *next, *lastchild;
    int pos, started_menu, filenum;
    char *name;
    info_data text;
};

typedef struct {
    char *text;
    int length;
    int nnodes, nodesize;
    node **nodes;
} info_idx;

static int info_rdadd(info_data *, wchar_t);
static int info_rdadds(info_data *, wchar_t const *);
static int info_rdaddc(info_data *, char);
static int info_rdaddsc(info_data *, char const *);

static void info_heading(info_data *, word *, word *, int);
static void info_rule(info_data *, int, int);
static void info_para(info_data *, word *, wchar_t *, word *, keywordlist *,
                      int, int, int);
static void info_codepara(info_data *, word *, int, int);
static void info_versionid(info_data *, word *);
static void info_menu_item(info_data *, node *, paragraph *);
static word *info_transform_wordlist(word *, keywordlist *);
static int info_check_index(word *, node *, indexdata *);

static int info_rdaddwc(info_data *, word *, word *, int);

static node *info_node_new(char *name, int charset);
static char *info_node_name(paragraph *p, int charset);

static infoconfig info_configure(paragraph *source) {
    infoconfig ret;

    /*
     * Defaults.
     */
    ret.filename = dupstr("output.info");
    ret.maxfilesize = 64 << 10;
    ret.charset = CS_ASCII;

    for (; source; source = source->next) {
        if (source->type == para_Config) {
            if (!ustricmp(source->keyword, L"info-filename")) {
                sfree(ret.filename);
                ret.filename = dupstr(adv(source->origkeyword));
            } else if (!ustricmp(source->keyword, L"info-charset")) {
                char *csname = utoa_dup(uadv(source->keyword), CS_ASCII);
                ret.charset = charset_from_localenc(csname);
                sfree(csname);
            } else if (!ustricmp(source->keyword, L"info-max-file-size")) {
                ret.maxfilesize = utoi(uadv(source->keyword));
            }
        }
    }

    return ret;
}

paragraph *info_config_filename(char *filename)
{
    return cmdline_cfg_simple("info-filename", filename, NULL);
}

void info_backend(paragraph *sourceform, keywordlist *keywords,
                  indexdata *idx, void *unused) {
    paragraph *p;
    infoconfig conf;
    word *prefix, *body, *wp;
    word spaceword;
    wchar_t *prefixextra;
    int nesting, nestindent;
    int indentb, indenta;
    int filepos;
    int has_index;
    info_data intro_text = EMPTY_INFO_DATA;
    node *topnode, *currnode;
    word bullet;
    FILE *fp;

    /*
     * FIXME: possibly configurability?
     */
    int width = 70, listindentbefore = 1, listindentafter = 3;
    int indent_code = 2, index_width = 40;

    IGNORE(unused);

    conf = info_configure(sourceform);

    /*
     * Go through and create a node for each section.
     */
    topnode = info_node_new("Top", conf.charset);
    currnode = topnode;
    for (p = sourceform; p; p = p->next) switch (p->type) {
        /*
         * Chapter titles.
         */
      case para_Chapter:
      case para_Appendix:
      case para_UnnumberedChapter:
      case para_Heading:
      case para_Subsect:
        {
            node *newnode, *upnode;
            char *nodename;

            nodename = info_node_name(p, conf.charset);
            newnode = info_node_new(nodename, conf.charset);
            sfree(nodename);

            p->private_data = newnode;

            if (p->parent)
                upnode = (node *)p->parent->private_data;
            else
                upnode = topnode;
            assert(upnode);
            newnode->up = upnode;

            currnode->next = newnode;
            newnode->prev = currnode;

            currnode->listnext = newnode;
            currnode = newnode;
        }
        break;
    }

    /*
     * Set up the display form of each index entry.
     */
    {
        int i;
        indexentry *entry;

        for (i = 0; (entry = index234(idx->entries, i)) != NULL; i++) {
            info_idx *ii = mknew(info_idx);
            info_data id = EMPTY_INFO_DATA;

            id.charset = conf.charset;

            ii->nnodes = ii->nodesize = 0;
            ii->nodes = NULL;

            ii->length = info_rdaddwc(&id, entry->text, NULL, FALSE);

            ii->text = id.output.text;

            entry->backend_data = ii;
        }
    }

    /*
     * An Info file begins with a piece of introductory text which
     * is apparently never shown anywhere. This seems to me to be a
     * good place to put the copyright notice and the version IDs. 
     * Also, Info directory entries are expected to go here.
     */
    intro_text.charset = conf.charset;

    info_rdaddsc(&intro_text,
            "This Info file generated by Halibut, ");
    info_rdaddsc(&intro_text, version);
    info_rdaddsc(&intro_text, "\n\n");

    for (p = sourceform; p; p = p->next)
        if (p->type == para_Config &&
            !ustricmp(p->keyword, L"info-dir-entry")) {
            wchar_t *section, *shortname, *longname, *kw;
            char *s;

            section = uadv(p->keyword);
            shortname = *section ? uadv(section) : NULL;
            longname = *shortname ? uadv(shortname) : NULL;
            kw = *longname ? uadv(longname) : NULL;

            if (!*longname) {
                error(err_infodirentry, &p->fpos);
                continue;
            }

            info_rdaddsc(&intro_text, "INFO-DIR-SECTION ");
            info_rdadds(&intro_text, section);
            info_rdaddsc(&intro_text, "\nSTART-INFO-DIR-ENTRY\n* ");
            info_rdadds(&intro_text, shortname);
            info_rdaddsc(&intro_text, ": (");
            s = dupstr(conf.filename);
            if (strlen(s) > 5 && !strcmp(s+strlen(s)-5, ".info"))
                s[strlen(s)-5] = '\0';
            info_rdaddsc(&intro_text, s);
            sfree(s);
            info_rdaddsc(&intro_text, ")");
            if (*kw) {
                keyword *kwl = kw_lookup(keywords, kw);
                if (kwl && kwl->para->private_data) {
                    node *n = (node *)kwl->para->private_data;
                    info_rdaddsc(&intro_text, n->name);
                }
            }
            info_rdaddsc(&intro_text, ".   ");
            info_rdadds(&intro_text, longname);
            info_rdaddsc(&intro_text, "\nEND-INFO-DIR-ENTRY\n\n");
        }

    for (p = sourceform; p; p = p->next)
        if (p->type == para_Copyright)
            info_para(&intro_text, NULL, NULL, p->words, keywords,
                      0, 0, width);

    for (p = sourceform; p; p = p->next)
        if (p->type == para_VersionID)
            info_versionid(&intro_text, p->words);

    if (intro_text.output.text[intro_text.output.pos-1] != '\n')
        info_rdaddc(&intro_text, '\n');

    /* Do the title */
    for (p = sourceform; p; p = p->next)
        if (p->type == para_Title)
            info_heading(&topnode->text, NULL, p->words, width);

    nestindent = listindentbefore + listindentafter;
    nesting = 0;

    currnode = topnode;

    /* Do the main document */
    for (p = sourceform; p; p = p->next) switch (p->type) {

      case para_QuotePush:
        nesting += 2;
        break;
      case para_QuotePop:
        nesting -= 2;
        assert(nesting >= 0);
        break;

      case para_LcontPush:
        nesting += nestindent;
        break;
      case para_LcontPop:
        nesting -= nestindent;
        assert(nesting >= 0);
        break;

        /*
         * Things we ignore because we've already processed them or
         * aren't going to touch them in this pass.
         */
      case para_IM:
      case para_BR:
      case para_Biblio:                /* only touch BiblioCited */
      case para_VersionID:
      case para_NoCite:
      case para_Title:
        break;

        /*
         * Chapter titles.
         */
      case para_Chapter:
      case para_Appendix:
      case para_UnnumberedChapter:
      case para_Heading:
      case para_Subsect:
        currnode = p->private_data;
        assert(currnode);
        assert(currnode->up);

        if (!currnode->up->started_menu) {
            info_rdaddsc(&currnode->up->text, "* Menu:\n\n");
            currnode->up->started_menu = TRUE;
        }
        info_menu_item(&currnode->up->text, currnode, p);

        has_index |= info_check_index(p->words, currnode, idx);
        info_heading(&currnode->text, p->kwtext, p->words, width);
        nesting = 0;
        break;

      case para_Rule:
        info_rule(&currnode->text, nesting, width - nesting);
        break;

      case para_Normal:
      case para_Copyright:
      case para_DescribedThing:
      case para_Description:
      case para_BiblioCited:
      case para_Bullet:
      case para_NumberedList:
        has_index |= info_check_index(p->words, currnode, idx);
        if (p->type == para_Bullet) {
            bullet.next = NULL;
            bullet.alt = NULL;
            bullet.type = word_Normal;
            bullet.text = L"-";        /* FIXME: configurability */
            prefix = &bullet;
            prefixextra = NULL;
            indentb = listindentbefore;
            indenta = listindentafter;
        } else if (p->type == para_NumberedList) {
            prefix = p->kwtext;
            prefixextra = L".";        /* FIXME: configurability */
            indentb = listindentbefore;
            indenta = listindentafter;
        } else if (p->type == para_Description) {
            prefix = NULL;
            prefixextra = NULL;
            indentb = listindentbefore;
            indenta = listindentafter;
        } else {
            prefix = NULL;
            prefixextra = NULL;
            indentb = indenta = 0;
        }
        if (p->type == para_BiblioCited) {
            body = dup_word_list(p->kwtext);
            for (wp = body; wp->next; wp = wp->next);
            wp->next = &spaceword;
            spaceword.next = p->words;
            spaceword.alt = NULL;
            spaceword.type = word_WhiteSpace;
            spaceword.text = NULL;
        } else {
            wp = NULL;
            body = p->words;
        }
        info_para(&currnode->text, prefix, prefixextra, body, keywords,
                  nesting + indentb, indenta,
                  width - nesting - indentb - indenta);
        if (wp) {
            wp->next = NULL;
            free_word_list(body);
        }
        break;

      case para_Code:
        info_codepara(&currnode->text, p->words,
                      nesting + indent_code,
                      width - nesting - 2 * indent_code);
        break;
    }

    /*
     * Create an index node if required.
     */
    if (has_index) {
        node *newnode;
        int i, j, k;
        indexentry *entry;

        newnode = info_node_new("Index", conf.charset);
        newnode->up = topnode;

        currnode->next = newnode;
        newnode->prev = currnode;
        currnode->listnext = newnode;

        info_rdaddsc(&newnode->text, "Index\n-----\n\n");

        info_menu_item(&topnode->text, newnode, NULL);

        for (i = 0; (entry = index234(idx->entries, i)) != NULL; i++) {
            info_idx *ii = (info_idx *)entry->backend_data;

            for (j = 0; j < ii->nnodes; j++) {
                /*
                 * When we have multiple references for a single
                 * index term, we only display the actual term on
                 * the first line, to make it clear that the terms
                 * really are the same.
                 */
                if (j == 0)
                    info_rdaddsc(&newnode->text, ii->text);
                for (k = (j ? 0 : ii->length); k < index_width; k++)
                    info_rdaddc(&newnode->text, ' ');
                info_rdaddsc(&newnode->text, "   *Note ");
                info_rdaddsc(&newnode->text, ii->nodes[j]->name);
                info_rdaddsc(&newnode->text, "::\n");
            }
        }
    }

    /*
     * Finalise the text of each node, by adding the ^_ delimiter
     * and the node line at the top.
     */
    for (currnode = topnode; currnode; currnode = currnode->listnext) {
        char *origtext = currnode->text.output.text;
        currnode->text = empty_info_data;
        currnode->text.charset = conf.charset;
        info_rdaddsc(&currnode->text, "\037\nFile: ");
        info_rdaddsc(&currnode->text, conf.filename);
        info_rdaddsc(&currnode->text, ",  Node: ");
        info_rdaddsc(&currnode->text, currnode->name);
        if (currnode->prev) {
            info_rdaddsc(&currnode->text, ",  Prev: ");
            info_rdaddsc(&currnode->text, currnode->prev->name);
        }
        info_rdaddsc(&currnode->text, ",  Up: ");
        info_rdaddsc(&currnode->text, (currnode->up ?
                                       currnode->up->name : "(dir)"));
        if (currnode->next) {
            info_rdaddsc(&currnode->text, ",  Next: ");
            info_rdaddsc(&currnode->text, currnode->next->name);
        }
        info_rdaddsc(&currnode->text, "\n\n");
        info_rdaddsc(&currnode->text, origtext);
        /*
         * Just make _absolutely_ sure we end with a newline.
         */
        if (currnode->text.output.text[currnode->text.output.pos-1] != '\n')
            info_rdaddc(&currnode->text, '\n');

        sfree(origtext);
    }    

    /*
     * Compute the offsets for the tag table.
     */
    filepos = intro_text.output.pos;
    for (currnode = topnode; currnode; currnode = currnode->listnext) {
        currnode->pos = filepos;
        filepos += currnode->text.output.pos;
    }

    /*
     * Split into sub-files.
     */
    if (conf.maxfilesize > 0) {
        int currfilesize = intro_text.output.pos, currfilenum = 1;
        for (currnode = topnode; currnode; currnode = currnode->listnext) {
            if (currfilesize > intro_text.output.pos &&
                currfilesize + currnode->text.output.pos > conf.maxfilesize) {
                currfilenum++;
                currfilesize = intro_text.output.pos;
            }
            currnode->filenum = currfilenum;
            currfilesize += currnode->text.output.pos;
        }
    }

    /*
     * Write the primary output file.
     */
    fp = fopen(conf.filename, "w");
    if (!fp) {
        error(err_cantopenw, conf.filename);
        return;
    }
    fputs(intro_text.output.text, fp);
    if (conf.maxfilesize == 0) {
        for (currnode = topnode; currnode; currnode = currnode->listnext)
            fputs(currnode->text.output.text, fp);
    } else {
        int filenum = 0;
        fprintf(fp, "\037\nIndirect:\n");
        for (currnode = topnode; currnode; currnode = currnode->listnext)
            if (filenum != currnode->filenum) {
                filenum = currnode->filenum;
                fprintf(fp, "%s-%d: %d\n", conf.filename, filenum,
                        currnode->pos);
            }
    }
    fprintf(fp, "\037\nTag Table:\n");
    if (conf.maxfilesize > 0)
        fprintf(fp, "(Indirect)\n");
    for (currnode = topnode; currnode; currnode = currnode->listnext)
        fprintf(fp, "Node: %s\177%d\n", currnode->name, currnode->pos);
    fprintf(fp, "\037\nEnd Tag Table\n");
    fclose(fp);

    /*
     * Write the subfiles.
     */
    if (conf.maxfilesize > 0) {
        int filenum = 0;
        fp = NULL;

        for (currnode = topnode; currnode; currnode = currnode->listnext) {
            if (filenum != currnode->filenum) {
                char *fname;

                filenum = currnode->filenum;

                if (fp)
                    fclose(fp);
                fname = mknewa(char, strlen(conf.filename) + 40);
                sprintf(fname, "%s-%d", conf.filename, filenum);
                fp = fopen(fname, "w");
                if (!fp) {
                    error(err_cantopenw, fname);
                    return;
                }
                sfree(fname);
                fputs(intro_text.output.text, fp);
            }
            fputs(currnode->text.output.text, fp);
        }

        if (fp)
            fclose(fp);
    }
}

static int info_check_index(word *w, node *n, indexdata *idx)
{
    int ret = 0;

    for (; w; w = w->next) {
        if (w->type == word_IndexRef) {
            indextag *tag;
            int i;

            tag = index_findtag(idx, w->text);
            if (!tag)
                break;

            for (i = 0; i < tag->nrefs; i++) {
                indexentry *entry = tag->refs[i];
                info_idx *ii = (info_idx *)entry->backend_data;

                if (ii->nnodes > 0 && ii->nodes[ii->nnodes-1] == n) {
                    /*
                     * If the same index term is indexed twice
                     * within the same section, we only want to
                     * mention it once in the index. So do nothing
                     * here.
                     */
                    continue;
                }

                if (ii->nnodes >= ii->nodesize) {
                    ii->nodesize += 32;
                    ii->nodes = resize(ii->nodes, ii->nodesize);
                }

                ii->nodes[ii->nnodes++] = n;

                ret = 1;
            }
        }
    }

    return ret;
}

static word *info_transform_wordlist(word *words, keywordlist *keywords)
{
    word *ret = dup_word_list(words);
    word *w;
    keyword *kwl;

    for (w = ret; w; w = w->next) {
        w->private_data = NULL;
        if (w->type == word_UpperXref || w->type == word_LowerXref) {
            kwl = kw_lookup(keywords, w->text);
            if (kwl) {
                if (kwl->para->type == para_NumberedList ||
                    kwl->para->type == para_BiblioCited) {
                    /*
                     * In Info, we do nothing special for xrefs to
                     * numbered list items or bibliography entries.
                     */
                    break;
                } else {
                    /*
                     * An xref to a different section has its text
                     * completely replaced.
                     */
                    word *w2, *w3, *w4;
                    w2 = w3 = w->next;
                    w4 = NULL;
                    while (w2) {
                        if (w2->type == word_XrefEnd) {
                            w4 = w2->next;
                            w2->next = NULL;
                            break;
                        }
                        w2 = w2->next;
                    }
                    free_word_list(w3);

                    /*
                     * Now w is the UpperXref / LowerXref we
                     * started with, and w4 is the next word after
                     * the corresponding XrefEnd (if any). The
                     * simplest thing is just to stick a pointer to
                     * the target node structure in the private
                     * data field of the xref word, and let
                     * info_rdaddwc and friends read the node name
                     * out from there.
                     */
                    w->next = w4;
                    w->private_data = kwl->para->private_data;
                    assert(w->private_data);
                }
            }
        }
    }

    return ret;
}

static int info_rdaddwc(info_data *id, word *words, word *end, int xrefs) {
    int ret = 0;

    for (; words && words != end; words = words->next) switch (words->type) {
      case word_HyperLink:
      case word_HyperEnd:
      case word_XrefEnd:
      case word_IndexRef:
        break;

      case word_Normal:
      case word_Emph:
      case word_Code:
      case word_WeakCode:
      case word_WhiteSpace:
      case word_EmphSpace:
      case word_CodeSpace:
      case word_WkCodeSpace:
      case word_Quote:
      case word_EmphQuote:
      case word_CodeQuote:
      case word_WkCodeQuote:
        assert(words->type != word_CodeQuote &&
               words->type != word_WkCodeQuote);
        if (towordstyle(words->type) == word_Emph &&
            (attraux(words->aux) == attr_First ||
             attraux(words->aux) == attr_Only))
            ret += info_rdadd(id, L'_');      /* FIXME: configurability */
        else if (towordstyle(words->type) == word_Code &&
                 (attraux(words->aux) == attr_First ||
                  attraux(words->aux) == attr_Only))
            ret += info_rdadd(id, L'`');      /* FIXME: configurability */
        if (removeattr(words->type) == word_Normal) {
            if (cvt_ok(id->charset, words->text) || !words->alt)
                ret += info_rdadds(id, words->text);
            else
                ret += info_rdaddwc(id, words->alt, NULL, FALSE);
        } else if (removeattr(words->type) == word_WhiteSpace) {
            ret += info_rdadd(id, L' ');
        } else if (removeattr(words->type) == word_Quote) {
            ret += info_rdadd(id, quoteaux(words->aux) == quote_Open ? L'`' : L'\'');
                                       /* FIXME: configurability */
        }
        if (towordstyle(words->type) == word_Emph &&
            (attraux(words->aux) == attr_Last ||
             attraux(words->aux) == attr_Only))
            ret += info_rdadd(id, L'_');     /* FIXME: configurability */
        else if (towordstyle(words->type) == word_Code &&
                 (attraux(words->aux) == attr_Last ||
                  attraux(words->aux) == attr_Only))
            ret += info_rdadd(id, L'\'');     /* FIXME: configurability */
        break;

      case word_UpperXref:
      case word_LowerXref:
        if (xrefs && words->private_data) {
            /*
             * This bit is structural and so must be done in char
             * rather than wchar_t.
             */
            ret += info_rdaddsc(id, "*Note ");
            ret += info_rdaddsc(id, ((node *)words->private_data)->name);
            ret += info_rdaddsc(id, "::");
        }
        break;
    }

    return ret;
}

static int info_width_internal(word *words, int xrefs, int charset);

static int info_width_internal_list(word *words, int xrefs, int charset) {
    int w = 0;
    while (words) {
        w += info_width_internal(words, xrefs, charset);
        words = words->next;
    }
    return w;
}

static int info_width_internal(word *words, int xrefs, int charset) {
    switch (words->type) {
      case word_HyperLink:
      case word_HyperEnd:
      case word_XrefEnd:
      case word_IndexRef:
        return 0;

      case word_Normal:
      case word_Emph:
      case word_Code:
      case word_WeakCode:
        return (((words->type == word_Emph ||
                  words->type == word_Code)
                 ? (attraux(words->aux) == attr_Only ? 2 :
                    attraux(words->aux) == attr_Always ? 0 : 1)
                 : 0) +
                (cvt_ok(charset, words->text) || !words->alt ?
                 ustrwid(words->text, charset) :
                 info_width_internal_list(words->alt, xrefs, charset)));

      case word_WhiteSpace:
      case word_EmphSpace:
      case word_CodeSpace:
      case word_WkCodeSpace:
      case word_Quote:
      case word_EmphQuote:
      case word_CodeQuote:
      case word_WkCodeQuote:
        assert(words->type != word_CodeQuote &&
               words->type != word_WkCodeQuote);
        return (((towordstyle(words->type) == word_Emph ||
                  towordstyle(words->type) == word_Code)
                 ? (attraux(words->aux) == attr_Only ? 2 :
                    attraux(words->aux) == attr_Always ? 0 : 1)
                 : 0) + 1);

      case word_UpperXref:
      case word_LowerXref:
        if (xrefs && words->private_data) {
            /* "*Note " plus "::" comes to 8 characters */
            return 8 + strwid(((node *)words->private_data)->name, charset);
        }
        break;
    }
    return 0;                          /* should never happen */
}

static int info_width_noxrefs(void *ctx, word *words)
{
    return info_width_internal(words, FALSE, *(int *)ctx);
}
static int info_width_xrefs(void *ctx, word *words)
{
    return info_width_internal(words, TRUE, *(int *)ctx);
}

static void info_heading(info_data *text, word *tprefix,
                         word *words, int width) {
    int length;
    int firstlinewidth, wrapwidth;
    wrappedline *wrapping, *p;

    length = 0;
    if (tprefix) {
        length += info_rdaddwc(text, tprefix, NULL, FALSE);
        length += info_rdadds(text, L": ");/* FIXME: configurability */
    }

    wrapwidth = width;
    firstlinewidth = width - length;

    wrapping = wrap_para(words, firstlinewidth, wrapwidth,
                         info_width_noxrefs, &text->charset, 0);
    for (p = wrapping; p; p = p->next) {
        length += info_rdaddwc(text, p->begin, p->end, FALSE);
        info_rdadd(text, L'\n');
        while (length--)
            info_rdadd(text, L'-');  /* FIXME: configurability */
        info_rdadd(text, L'\n');
        length = 0;
    }
    wrap_free(wrapping);
    info_rdadd(text, L'\n');
}

static void info_rule(info_data *text, int indent, int width) {
    while (indent--) info_rdadd(text, L' ');
    while (width--) info_rdadd(text, L'-');
    info_rdadd(text, L'\n');
    info_rdadd(text, L'\n');
}

static void info_para(info_data *text, word *prefix, wchar_t *prefixextra,
                      word *input, keywordlist *keywords,
                      int indent, int extraindent, int width) {
    wrappedline *wrapping, *p;
    word *words;
    int e;
    int i;
    int firstlinewidth = width;

    words = info_transform_wordlist(input, keywords);

    if (prefix) {
        for (i = 0; i < indent; i++)
            info_rdadd(text, L' ');
        e = info_rdaddwc(text, prefix, NULL, FALSE);
        if (prefixextra)
            e += info_rdadds(text, prefixextra);
        /* If the prefix is too long, shorten the first line to fit. */
        e = extraindent - e;
        if (e < 0) {
            firstlinewidth += e;       /* this decreases it, since e < 0 */
            if (firstlinewidth < 0) {
                e = indent + extraindent;
                firstlinewidth = width;
                info_rdadd(text, L'\n');
            } else
                e = 0;
        }
    } else
        e = indent + extraindent;

    wrapping = wrap_para(words, firstlinewidth, width, info_width_xrefs,
                         &text->charset, 0);
    for (p = wrapping; p; p = p->next) {
        for (i = 0; i < e; i++)
            info_rdadd(text, L' ');
        info_rdaddwc(text, p->begin, p->end, TRUE);
        info_rdadd(text, L'\n');
        e = indent + extraindent;
    }
    wrap_free(wrapping);
    info_rdadd(text, L'\n');

    free_word_list(words);
}

static void info_codepara(info_data *text, word *words,
                          int indent, int width) {
    int i;

    for (; words; words = words->next) if (words->type == word_WeakCode) {
        for (i = 0; i < indent; i++)
            info_rdadd(text, L' ');
        if (info_rdadds(text, words->text) > width) {
            /* FIXME: warn */
        }
        info_rdadd(text, L'\n');
    }

    info_rdadd(text, L'\n');
}

static void info_versionid(info_data *text, word *words) {
    info_rdadd(text, L'[');                    /* FIXME: configurability */
    info_rdaddwc(text, words, NULL, FALSE);
    info_rdadds(text, L"]\n");
}

static node *info_node_new(char *name, int charset)
{
    node *n;

    n = mknew(node);
    n->text = empty_info_data;
    n->text.charset = charset;
    n->up = n->next = n->prev = n->lastchild = n->listnext = NULL;
    n->name = dupstr(name);
    n->started_menu = FALSE;

    return n;
}

static char *info_node_name(paragraph *par, int charset)
{
    info_data id = EMPTY_INFO_DATA;
    char *p, *q;

    id.charset = charset;
    info_rdaddwc(&id, par->kwtext ? par->kwtext : par->words, NULL, FALSE);
    info_rdaddsc(&id, NULL);

    /*
     * We cannot have commas or colons in a node name. Remove any
     * that we find, with a warning.
     */
    p = q = id.output.text;
    while (*p) {
        if (*p == ':' || *p == ',') {
            error(err_infonodechar, &par->fpos, *p);
        } else {
            *q++ = *p;
        }
        p++;
    }
    *p = '\0';

    return id.output.text;
}

static void info_menu_item(info_data *text, node *n, paragraph *p)
{
    /*
     * FIXME: Depending on how we're doing node names in this info
     * file, we might want to do
     * 
     *   * Node name:: Chapter title
     * 
     * _or_
     * 
     *   * Chapter number: Node name.
     * 
     * This function mostly works in char rather than wchar_t,
     * because a menu item is a structural component.
     */
    info_rdaddsc(text, "* ");
    info_rdaddsc(text, n->name);
    info_rdaddsc(text, "::");
    if (p) {
        info_rdaddc(text, ' ');
        info_rdaddwc(text, p->words, NULL, FALSE);
    }
    info_rdaddc(text, '\n');
}

/*
 * These functions implement my wrapper on the rdadd* calls which
 * allows me to switch arbitrarily between literal octet-string
 * text and charset-translated Unicode. (Because no matter what
 * character set I write the actual text in, I expect info readers
 * to treat node names and file names literally and to expect
 * keywords like `*Note' in their canonical form, so I have to take
 * steps to ensure that those structural elements of the file
 * aren't messed with.)
 */
static int info_rdadds(info_data *d, wchar_t const *wcs)
{
    if (!d->wcmode) {
        d->state = charset_init_state;
        d->wcmode = TRUE;
    }

    if (wcs) {
        char buf[256];
        int len, width, ret;

        width = ustrwid(wcs, d->charset);

        len = ustrlen(wcs);
        while (len > 0) {
            int prevlen = len;

            ret = charset_from_unicode(&wcs, &len, buf, lenof(buf),
                                       d->charset, &d->state, NULL);

            assert(len < prevlen);

            if (ret > 0) {
                buf[ret] = '\0';
                rdaddsc(&d->output, buf);
            }
        }

        return width;
    } else
        return 0;
}

static int info_rdaddsc(info_data *d, char const *cs)
{
    if (d->wcmode) {
        char buf[256];
        int ret;

        ret = charset_from_unicode(NULL, 0, buf, lenof(buf),
                                   d->charset, &d->state, NULL);
        if (ret > 0) {
            buf[ret] = '\0';
            rdaddsc(&d->output, buf);
        }

        d->wcmode = FALSE;
    }

    if (cs) {
        rdaddsc(&d->output, cs);
        return strlen(cs);
    } else
        return 0;
}

static int info_rdadd(info_data *d, wchar_t wc)
{
    wchar_t wcs[2];
    wcs[0] = wc;
    wcs[1] = L'\0';
    return info_rdadds(d, wcs);
}

static int info_rdaddc(info_data *d, char c)
{
    char cs[2];
    cs[0] = c;
    cs[1] = '\0';
    return info_rdaddsc(d, cs);
}