[sgt/halibut] / bk_pdf.c

/*
 * PDF backend for Halibut
 */

#include <assert.h>
#include "halibut.h"
#include "paper.h"

#define TREE_BRANCH 2		       /* max branching factor in page tree */

paragraph *pdf_config_filename(char *filename)
{
    return cmdline_cfg_simple("pdf-filename", filename, NULL);
}

typedef struct object_Tag object;
typedef struct objlist_Tag objlist;

struct object_Tag {
    objlist *list;
    object *next;
    int number;
    rdstringc main, stream;
    int size, fileoff;
    char *final;
};

struct objlist_Tag {
    int number;
    object *head, *tail;
};

static object *new_object(objlist *list);
static void objtext(object *o, char const *text);
static void objstream(object *o, char const *text);
static void pdf_string(void (*add)(object *, char const *),
		       object *, char const *);
static void pdf_string_len(void (*add)(object *, char const *),
			   object *, char const *, int);
static void objref(object *o, object *dest);

static void make_pages_node(object *node, object *parent, page_data *first,
			    page_data *last, object *resources);
static int make_outline(object *parent, outline_element *start, int n,
			int open);
static int pdf_versionid(FILE *fp, word *words);

void pdf_backend(paragraph *sourceform, keywordlist *keywords,
		 indexdata *idx, void *vdoc) {
    document *doc = (document *)vdoc;
    int font_index;
    font_encoding *fe;
    page_data *page;
    int pageno;
    FILE *fp;
    char *filename;
    paragraph *p;
    objlist olist;
    object *o, *cat, *outlines, *pages, *resources;
    int fileoff;

    IGNORE(keywords);
    IGNORE(idx);

    filename = dupstr("output.pdf");
    for (p = sourceform; p; p = p->next) {
	if (p->type == para_Config) {
	    if (!ustricmp(p->keyword, L"pdf-filename")) {
		sfree(filename);
		filename = dupstr(adv(p->origkeyword));
	    }
	}
    }

    olist.head = olist.tail = NULL;
    olist.number = 1;

    cat = new_object(&olist);
    if (doc->n_outline_elements > 0)
	outlines = new_object(&olist);
    else
	outlines = NULL;
    pages = new_object(&olist);
    resources = new_object(&olist);

    /*
     * The catalogue just contains references to the outlines and
     * pages objects.
     */
    objtext(cat, "<<\n/Type /Catalog");
    if (outlines) {
	objtext(cat, "\n/Outlines ");
	objref(cat, outlines);
    }
    objtext(cat, "\n/Pages ");
    objref(cat, pages);
    if (outlines)
	objtext(cat, "\n/PageMode /UseOutlines");
    objtext(cat, "\n>>\n");

    /*
     * Set up the resources dictionary, which mostly means
     * providing all the font objects and names to call them by.
     */
    font_index = 0;
    objtext(resources, "<<\n/Font <<\n");
    for (fe = doc->fonts->head; fe; fe = fe->next) {
	char fname[40];
	int i;
	object *font;

	sprintf(fname, "f%d", font_index++);
	fe->name = dupstr(fname);

	font = new_object(&olist);

	objtext(resources, "/");
	objtext(resources, fe->name);
	objtext(resources, " ");
	objref(resources, font);
	objtext(resources, "\n");

	objtext(font, "<<\n/Type /Font\n/Subtype /Type1\n/Name /");
	objtext(font, fe->name);
	objtext(font, "\n/BaseFont /");
	objtext(font, fe->font->name);
	objtext(font, "\n/Encoding <<\n/Type /Encoding\n/Differences [");

	for (i = 0; i < 256; i++) {
	    char buf[20];
	    if (!fe->vector[i])
		continue;
	    sprintf(buf, "\n%d /", i);
	    objtext(font, buf);
	    objtext(font, fe->vector[i] ? fe->vector[i] : ".notdef");
	}

	objtext(font, "\n]\n>>\n");

	{
	    object *widths = new_object(&olist);
	    objtext(font, "/FirstChar 0\n/LastChar 255\n/Widths ");
	    objref(font, widths);
	    objtext(font, "\n");
	    objtext(widths, "[\n");
	    for (i = 0; i < 256; i++) {
		char buf[80];
		double width;
		if (fe->indices[i] < 0)
		    width = 0.0;
		else
		    width = fe->font->widths[fe->indices[i]];
		sprintf(buf, "%g\n", 1000.0 * width / 4096.0);
		objtext(widths, buf);
	    }
	    objtext(widths, "]\n");
	}

	objtext(font, ">>\n");
    }
    objtext(resources, ">>\n>>\n");

    /*
     * Define the page objects for each page, and get each one
     * ready to have a `Parent' specification added to it.
     */
    for (page = doc->pages; page; page = page->next) {
	object *opage;

	opage = new_object(&olist);
	page->spare = opage;
	objtext(opage, "<<\n/Type /Page\n");
    }

    /*
     * Recursively build the page tree.
     */
    make_pages_node(pages, NULL, doc->pages, NULL, resources);

    /*
     * Create and render the individual pages.
     */
    pageno = 0;
    for (page = doc->pages; page; page = page->next) {
	object *opage, *cstr;
	rect *r;
	text_fragment *frag, *frag_end;
	char buf[256];
	int x, y, lx, ly;

	opage = (object *)page->spare;
	/*
	 * At this point the page dictionary is already
	 * half-written, with /Type and /Parent already present. We
	 * continue from there.
	 */

	/*
	 * The PDF spec says /Resources is required, but also says
	 * that it's inheritable and may be omitted if it's present
	 * in a Pages node. In our case it is: it's present in the
	 * topmost /Pages node because we carefully put it there.
	 * So we don't need a /Resources entry here.
	 */
	sprintf(buf, "/MediaBox [0 0 %g %g]\n",
		doc->paper_width / 4096.0, doc->paper_height / 4096.0);
	objtext(opage, buf);

	/*
	 * Now we're ready to define a content stream containing
	 * the actual text on the page.
	 */
	cstr = new_object(&olist);
	objtext(opage, "/Contents ");
	objref(opage, cstr);
	objtext(opage, "\n");

	/*
	 * Render any rectangles on the page.
	 */
	for (r = page->first_rect; r; r = r->next) {
	    char buf[512];
	    sprintf(buf, "%g %g %g %g re f\n", r->x / 4096.0,
		    r->y / 4096.0, r->w / 4096.0, r->h / 4096.0);
	    objstream(cstr, buf);
	}

	objstream(cstr, "BT\n");

	/*
	 * PDF tracks two separate current positions: the position
	 * given in the `line matrix' and the position given in the
	 * `text matrix'. We must therefore track both as well.
	 * They start off at -1 (unset).
	 */
	lx = ly = -1;
	x = y = -1;

	frag = page->first_text;
	while (frag) {
	    /*
	     * For compactness, I'm going to group text fragments
	     * into subsequences that use the same font+size. So
	     * first find the end of this subsequence.
	     */
	    for (frag_end = frag;
		 (frag_end &&
		  frag_end->fe == frag->fe &&
		  frag_end->fontsize == frag->fontsize);
		 frag_end = frag_end->next);

	    /*
	     * Now select the text fragment, and prepare to display
	     * the text.
	     */
	    objstream(cstr, "/");
	    objstream(cstr, frag->fe->name);
	    sprintf(buf, " %d Tf ", frag->fontsize);
	    objstream(cstr, buf);

	    while (frag && frag != frag_end) {
		/*
		 * Place the text position for the first piece of
		 * text.
		 */
		if (lx < 0) {
		    sprintf(buf, "1 0 0 1 %g %g Tm ",
			    frag->x/4096.0, frag->y/4096.0);
		} else {
		    sprintf(buf, "%g %g Td ",
			    (frag->x - lx)/4096.0, (frag->y - ly)/4096.0);
		}
		objstream(cstr, buf);
		lx = x = frag->x;
		ly = y = frag->y;

		/*
		 * See if we're going to use Tj (show a single
		 * string) or TJ (show an array of strings with
		 * x-spacings between them). We determine this by
		 * seeing if there's more than one text fragment in
		 * sequence with the same y-coordinate.
		 */
		if (frag->next && frag->next != frag_end &&
		    frag->next->y == y) {
		    /*
		     * The TJ strategy.
		     */
		    objstream(cstr, "[");
		    while (frag && frag != frag_end && frag->y == y) {
			if (frag->x != x) {
			    sprintf(buf, "%g",
				    (x - frag->x) * 1000.0 /
				    (4096.0 * frag->fontsize));
			    objstream(cstr, buf);
			}
			pdf_string(objstream, cstr, frag->text);
			x = frag->x + frag->width;
			frag = frag->next;
		    }
		    objstream(cstr, "]TJ\n");
		} else
		{
		    /*
		     * The Tj strategy.
		     */
		    pdf_string(objstream, cstr, frag->text);
		    objstream(cstr, "Tj\n");
		    frag = frag->next;
		}
	    }
	}
	objstream(cstr, "ET");

	/*
	 * Also, we want an annotation dictionary containing the
	 * cross-references from this page.
	 */
	if (page->first_xref) {
	    xref *xr;
	    objtext(opage, "/Annots [\n");

	    for (xr = page->first_xref; xr; xr = xr->next) {
		object *annot;
		char buf[256];

		annot = new_object(&olist);
		objref(opage, annot);
		objtext(opage, "\n");

		objtext(annot, "<<\n/Type /Annot\n/Subtype /Link\n/Rect [");
		sprintf(buf, "%g %g %g %g",
			xr->lx / 4096.0, xr->by / 4096.0,
			xr->rx / 4096.0, xr->ty / 4096.0);
		objtext(annot, buf);
		objtext(annot, "]\n/Border [0 0 0]\n");

		if (xr->dest.type == PAGE) {
		    objtext(annot, "/Dest [");
		    objref(annot, (object *)xr->dest.page->spare);
		    objtext(annot, " /XYZ null null null]\n");
		} else {
		    objtext(annot, "/A <<\n/Type /Action\n/S /URI\n/URI ");
		    pdf_string(objtext, annot, xr->dest.url);
		    objtext(annot, "\n>>\n");
		}

		objtext(annot, ">>\n");
	    }

	    objtext(opage, "]\n");
	}

	objtext(opage, ">>\n");
    }

    /*
     * Set up the outlines dictionary.
     */
    if (outlines) {
	int topcount;
	char buf[80];

	objtext(outlines, "<<\n/Type /Outlines\n");
	topcount = make_outline(outlines, doc->outline_elements,
				doc->n_outline_elements, TRUE);
	sprintf(buf, "/Count %d\n>>\n", topcount);
	objtext(outlines, buf);
    }

    /*
     * Assemble the final linear form of every object.
     */
    for (o = olist.head; o; o = o->next) {
	rdstringc rs = {0, 0, NULL};
	char text[80];

	sprintf(text, "%d 0 obj\n", o->number);
	rdaddsc(&rs, text);

	if (!o->main.text && o->stream.text) {
	    sprintf(text, "<<\n/Length %d\n>>\n", o->stream.pos);
	    rdaddsc(&o->main, text);
	}

	assert(o->main.text);
	rdaddsc(&rs, o->main.text);
	sfree(o->main.text);

	if (rs.text[rs.pos-1] != '\n')
	    rdaddc(&rs, '\n');

	if (o->stream.text) {
	    /*
	     * FIXME: If we ever start compressing stream data then
	     * it will have zero bytes in it, so we'll have to be
	     * more careful than this.
	     */
	    rdaddsc(&rs, "stream\n");
	    rdaddsc(&rs, o->stream.text);
	    rdaddsc(&rs, "\nendstream\n");
	    sfree(o->stream.text);
	}

	rdaddsc(&rs, "endobj\n");

	o->final = rs.text;
	o->size = rs.pos;
    }

    /*
     * Write out the PDF file.
     */

    fp = fopen(filename, "wb");
    if (!fp) {
	error(err_cantopenw, filename);
	return;
    }

    /*
     * Header. I'm going to put the version IDs in the header as
     * well, simply in PDF comments.
     */
    fileoff = fprintf(fp, "%%PDF-1.3\n");
    for (p = sourceform; p; p = p->next)
	if (p->type == para_VersionID)
	    fileoff += pdf_versionid(fp, p->words);

    /*
     * Body
     */
    for (o = olist.head; o; o = o->next) {
	o->fileoff = fileoff;
	fwrite(o->final, 1, o->size, fp);
	fileoff += o->size;
    }

    /*
     * Cross-reference table
     */
    fprintf(fp, "xref\n");
    assert(olist.head->number == 1);
    fprintf(fp, "0 %d\n", olist.tail->number + 1);
    fprintf(fp, "0000000000 65535 f \n");
    for (o = olist.head; o; o = o->next) {
	char entry[40];
	sprintf(entry, "%010d 00000 n \n", o->fileoff);
	assert(strlen(entry) == 20);
	fputs(entry, fp);
    }

    /*
     * Trailer
     */
    fprintf(fp, "trailer\n<<\n/Size %d\n/Root %d 0 R\n>>\n",
	    olist.tail->number + 1, cat->number);
    fprintf(fp, "startxref\n%d\n%%%%EOF\n", fileoff);

    fclose(fp);

    sfree(filename);
}

static object *new_object(objlist *list)
{
    object *obj = snew(object);

    obj->list = list;

    obj->main.text = NULL;
    obj->main.pos = obj->main.size = 0;
    obj->stream.text = NULL;
    obj->stream.pos = obj->stream.size = 0;

    obj->number = list->number++;

    obj->next = NULL;
    if (list->tail)
	list->tail->next = obj;
    else
	list->head = obj;
    list->tail = obj;

    obj->size = 0;
    obj->final = NULL;

    return obj;
}

static void objtext(object *o, char const *text)
{
    rdaddsc(&o->main, text);
}

static void objstream(object *o, char const *text)
{
    rdaddsc(&o->stream, text);
}

static void objref(object *o, object *dest)
{
    char buf[40];
    sprintf(buf, "%d 0 R", dest->number);
    rdaddsc(&o->main, buf);
}

static void make_pages_node(object *node, object *parent, page_data *first,
			    page_data *last, object *resources)
{
    int count;
    page_data *page;
    char buf[80];

    objtext(node, "<<\n/Type /Pages\n");
    if (parent) {
	objtext(node, "/Parent ");
	objref(node, parent);
	objtext(node, "\n");
    }

    /*
     * Count the pages in this stretch, to see if there are few
     * enough to reference directly.
     */
    count = 0;
    for (page = first; page; page = page->next) {
	count++;
	if (page == last)
	    break;
    }

    sprintf(buf, "/Count %d\n/Kids [\n", count);
    objtext(node, buf);

    if (count > TREE_BRANCH) {
	int i;
	page_data *thisfirst, *thislast;

	page = first;

	for (i = 0; i < TREE_BRANCH; i++) {
	    int number = (i+1) * count / TREE_BRANCH - i * count / TREE_BRANCH;
	    thisfirst = page;
	    while (number--) {
		thislast = page;
		page = page->next;
	    }

	    if (thisfirst == thislast) {
		objref(node, (object *)thisfirst->spare);
		objtext((object *)thisfirst->spare, "/Parent ");
		objref((object *)thisfirst->spare, node);
		objtext((object *)thisfirst->spare, "\n");
	    } else {
		object *newnode = new_object(node->list);
		make_pages_node(newnode, node, thisfirst, thislast, NULL);
		objref(node, newnode);
	    }
	    objtext(node, "\n");
	}

	assert(thislast == last || page == NULL);

    } else {
	for (page = first; page; page = page->next) {
	    objref(node, (object *)page->spare);
	    objtext(node, "\n");
	    objtext((object *)page->spare, "/Parent ");
	    objref((object *)page->spare, node);
	    objtext((object *)page->spare, "\n");
	    if (page == last)
		break;
	}
    }

    objtext(node, "]\n");

    if (resources) {
	objtext(node, "/Resources ");
	objref(node, resources);
	objtext(node, "\n");
    }

    objtext(node, ">>\n");
}

/*
 * In text on the page, PDF uses the PostScript font model, which
 * means that glyphs are identified by PS strings and hence font
 * encoding can be managed independently of the supplied encoding
 * of the font. However, in the document outline, the PDF spec
 * encodes in either PDFDocEncoding (a custom superset of
 * ISO-8859-1) or UTF-16BE.
 */
static char *pdf_outline_convert(wchar_t *s, int *len) {
    char *ret;

    ret = utoa_careful_dup(s, CS_PDF);

    /*
     * Very silly special case: if the returned string begins with
     * FE FF, then the PDF reader will mistake it for a UTF-16BE
     * string. So in this case we give up on PDFDocEncoding and
     * encode it in UTF-16 straight away.
     */
    if (ret && ret[0] == '\xFE' && ret[1] == '\xFF') {
	sfree(ret);
	ret = NULL;
    }

    if (!ret) {
	ret = utoa_dup_len(s, CS_UTF16BE, len);
    } else {
	*len = strlen(ret);
    }

    return ret;
}

static int make_outline(object *parent, outline_element *items, int n,
			int open)
{
    int level, totalcount = 0;
    outline_element *itemp;
    object *curr, *prev = NULL, *first = NULL, *last = NULL;

    assert(n > 0);

    level = items->level;

    while (n > 0) {
	char *title;
	int titlelen;

	/*
	 * Here we expect to be sitting on an item at the given
	 * level. So we start by constructing an outline entry for
	 * that item.
	 */
	assert(items->level == level);

	title = pdf_outline_convert(items->pdata->outline_title, &titlelen);

	totalcount++;
	curr = new_object(parent->list);
	if (!first) first = curr;
	last = curr;
	objtext(curr, "<<\n/Title ");
	pdf_string_len(objtext, curr, title, titlelen);
	objtext(curr, "\n/Parent ");
	objref(curr, parent);
	objtext(curr, "\n/Dest [");
	objref(curr, (object *)items->pdata->first->page->spare);
	objtext(curr, " /XYZ null null null]\n");
	if (prev) {
	    objtext(curr, "/Prev ");
	    objref(curr, prev);
	    objtext(curr, "\n");

	    objtext(prev, "/Next ");
	    objref(prev, curr);
	    objtext(prev, "\n>>\n");
	}
	prev = curr;

	items++, n--;
	for (itemp = items; itemp < items+n && itemp->level > level;
	     itemp++);

	if (itemp > items) {
	    char buf[80];
	    int count = make_outline(curr, items, itemp - items, FALSE);
	    if (!open)
		count = -count;
	    else
		totalcount += count;
	    sprintf(buf, "/Count %d\n", count);
	    objtext(curr, buf);
	}

	n -= itemp - items;
	items = itemp;
    }
    objtext(prev, ">>\n");

    assert(first && last);
    objtext(parent, "/First ");
    objref(parent, first);
    objtext(parent, "\n/Last ");
    objref(parent, last);
    objtext(parent, "\n");

    return totalcount;
}

static int pdf_versionid(FILE *fp, word *words)
{
    int ret;

    ret = fprintf(fp, "%% ");

    for (; words; words = words->next) {
	char *text;
	int type;

	switch (words->type) {
	  case word_HyperLink:
	  case word_HyperEnd:
	  case word_UpperXref:
	  case word_LowerXref:
	  case word_XrefEnd:
	  case word_IndexRef:
	    continue;
	}

	type = removeattr(words->type);

	switch (type) {
	  case word_Normal:
	    text = utoa_dup(words->text, CS_ASCII);
	    break;
	  case word_WhiteSpace:
	    text = dupstr(" ");
	    break;
	  case word_Quote:
	    text = dupstr("'");
	    break;
	}

	fputs(text, fp);
	ret += strlen(text);
	sfree(text);
    }

    ret += fprintf(fp, "\n");

    return ret;
}

static void pdf_string_len(void (*add)(object *, char const *),
			   object *o, char const *str, int len)
{
    char const *p;

    add(o, "(");
    for (p = str; len > 0; p++, len--) {
	char c[10];
	if (*p < ' ' || *p > '~') {
	    sprintf(c, "\\%03o", 0xFF & (int)*p);
	} else {
	    int n = 0;
	    if (*p == '\\' || *p == '(' || *p == ')')
		c[n++] = '\\';
	    c[n++] = *p;
	    c[n] = '\0';
	}
	add(o, c);
    }
    add(o, ")");
}

static void pdf_string(void (*add)(object *, char const *),
		       object *o, char const *str)
{
    pdf_string_len(add, o, str, strlen(str));
}
Commit	Line	Data
43341922	1	/*
	2	* PDF backend for Halibut
	3	*/
	4
	5	#include <assert.h>
	6	#include "halibut.h"
	7	#include "paper.h"
	8
	9	#define TREE_BRANCH 2 /* max branching factor in page tree */
	10
	11	paragraph pdf_config_filename(char filename)
	12	{
e4ea58f8	13	return cmdline_cfg_simple("pdf-filename", filename, NULL);
43341922	14	}
	15
	16	typedef struct object_Tag object;
	17	typedef struct objlist_Tag objlist;
	18
	19	struct object_Tag {
	20	objlist *list;
	21	object *next;
	22	int number;
	23	rdstringc main, stream;
	24	int size, fileoff;
	25	char *final;
	26	};
	27
	28	struct objlist_Tag {
	29	int number;
	30	object head, tail;
	31	};
	32
	33	static object new_object(objlist list);
	34	static void objtext(object o, char const text);
	35	static void objstream(object o, char const text);
7c8c4239	36	static void pdf_string(void (add)(object , char const *),
7c8c4239	37	object , char const );
ce1b04aa	38	static void pdf_string_len(void (add)(object , char const *),
ce1b04aa	39	object , char const , int);
43341922	40	static void objref(object o, object dest);
	41
	42	static void make_pages_node(object node, object parent, page_data *first,
	43	page_data last, object resources);
f0e51ce1	44	static int make_outline(object parent, outline_element start, int n,
f0e51ce1	45	int open);
09358aa7	46	static int pdf_versionid(FILE fp, word words);
43341922	47
	48	void pdf_backend(paragraph sourceform, keywordlist keywords,
	49	indexdata idx, void vdoc) {
	50	document doc = (document )vdoc;
	51	int font_index;
	52	font_encoding *fe;
	53	page_data *page;
	54	int pageno;
	55	FILE *fp;
	56	char *filename;
	57	paragraph *p;
	58	objlist olist;
	59	object o, cat, outlines, pages, *resources;
	60	int fileoff;
	61
	62	IGNORE(keywords);
	63	IGNORE(idx);
	64
	65	filename = dupstr("output.pdf");
	66	for (p = sourceform; p; p = p->next) {
7738a2fb	67	if (p->type == para_Config) {
43341922	68	if (!ustricmp(p->keyword, L"pdf-filename")) {
43341922	69	sfree(filename);
e4ea58f8	70	filename = dupstr(adv(p->origkeyword));
43341922	71	}
	72	}
	73	}
	74
	75	olist.head = olist.tail = NULL;
	76	olist.number = 1;
	77
	78	cat = new_object(&olist);
a615a778	79	if (doc->n_outline_elements > 0)
	80	outlines = new_object(&olist);
	81	else
	82	outlines = NULL;
43341922	83	pages = new_object(&olist);
	84	resources = new_object(&olist);
	85
	86	/*
43341922	87	* The catalogue just contains references to the outlines and
	88	* pages objects.
	89	*/
a615a778	90	objtext(cat, "<<\n/Type /Catalog");
	91	if (outlines) {
	92	objtext(cat, "\n/Outlines ");
	93	objref(cat, outlines);
	94	}
43341922	95	objtext(cat, "\n/Pages ");
43341922	96	objref(cat, pages);
a615a778	97	if (outlines)
	98	objtext(cat, "\n/PageMode /UseOutlines");
	99	objtext(cat, "\n>>\n");
43341922	100
	101	/*
	102	* Set up the resources dictionary, which mostly means
	103	* providing all the font objects and names to call them by.
	104	*/
	105	font_index = 0;
	106	objtext(resources, "<<\n/Font <<\n");
	107	for (fe = doc->fonts->head; fe; fe = fe->next) {
	108	char fname[40];
	109	int i;
	110	object *font;
	111
	112	sprintf(fname, "f%d", font_index++);
	113	fe->name = dupstr(fname);
	114
	115	font = new_object(&olist);
	116
	117	objtext(resources, "/");
	118	objtext(resources, fe->name);
	119	objtext(resources, " ");
	120	objref(resources, font);
	121	objtext(resources, "\n");
	122
	123	objtext(font, "<<\n/Type /Font\n/Subtype /Type1\n/Name /");
	124	objtext(font, fe->name);
	125	objtext(font, "\n/BaseFont /");
	126	objtext(font, fe->font->name);
	127	objtext(font, "\n/Encoding <<\n/Type /Encoding\n/Differences [");
	128
	129	for (i = 0; i < 256; i++) {
	130	char buf[20];
	131	if (!fe->vector[i])
	132	continue;
	133	sprintf(buf, "\n%d /", i);
	134	objtext(font, buf);
	135	objtext(font, fe->vector[i] ? fe->vector[i] : ".notdef");
	136	}
	137
	138	objtext(font, "\n]\n>>\n");
	139
	140	{
	141	object *widths = new_object(&olist);
	142	objtext(font, "/FirstChar 0\n/LastChar 255\n/Widths ");
	143	objref(font, widths);
	144	objtext(font, "\n");
	145	objtext(widths, "[\n");
	146	for (i = 0; i < 256; i++) {
	147	char buf[80];
	148	double width;
	149	if (fe->indices[i] < 0)
	150	width = 0.0;
	151	else
	152	width = fe->font->widths[fe->indices[i]];
	153	sprintf(buf, "%g\n", 1000.0 * width / 4096.0);
	154	objtext(widths, buf);
	155	}
	156	objtext(widths, "]\n");
	157	}
	158
	159	objtext(font, ">>\n");
	160	}
	161	objtext(resources, ">>\n>>\n");
	162
	163	/*
164	* Define the page objects for each page, and get each one
165	* ready to have a `Parent' specification added to it.
166	*/
167	for (page = doc->pages; page; page = page->next) {
168	object *opage;
169
170	opage = new_object(&olist);
171	page->spare = opage;
172	objtext(opage, "<<\n/Type /Page\n");
173	}
174
175	/*
176	* Recursively build the page tree.
177	*/
178	make_pages_node(pages, NULL, doc->pages, NULL, resources);
179
180	/*
181	* Create and render the individual pages.
182	*/
183	pageno = 0;
184	for (page = doc->pages; page; page = page->next) {
185	object opage, cstr;
23765aeb	186	rect *r;
7c8c4239	187	text_fragment frag, frag_end;
43341922	188	char buf[256];
7c8c4239	189	int x, y, lx, ly;
43341922	190
	191	opage = (object *)page->spare;
	192	/*
	193	* At this point the page dictionary is already
	194	* half-written, with /Type and /Parent already present. We
	195	* continue from there.
	196	*/
	197
	198	/*
	199	* The PDF spec says /Resources is required, but also says
	200	* that it's inheritable and may be omitted if it's present
	201	* in a Pages node. In our case it is: it's present in the
	202	* topmost /Pages node because we carefully put it there.
	203	* So we don't need a /Resources entry here.
	204	*/
	205	sprintf(buf, "/MediaBox [0 0 %g %g]\n",
	206	doc->paper_width / 4096.0, doc->paper_height / 4096.0);
	207	objtext(opage, buf);
	208
	209	/*
	210	* Now we're ready to define a content stream containing
	211	* the actual text on the page.
	212	*/
	213	cstr = new_object(&olist);
	214	objtext(opage, "/Contents ");
	215	objref(opage, cstr);
	216	objtext(opage, "\n");
	217
23765aeb	218	/*
	219	* Render any rectangles on the page.
	220	*/
	221	for (r = page->first_rect; r; r = r->next) {
	222	char buf[512];
	223	sprintf(buf, "%g %g %g %g re f\n", r->x / 4096.0,
	224	r->y / 4096.0, r->w / 4096.0, r->h / 4096.0);
	225	objstream(cstr, buf);
	226	}
	227
43341922	228	objstream(cstr, "BT\n");
43341922	229
7c8c4239	230	/*
	231	* PDF tracks two separate current positions: the position
	232	* given in the `line matrix' and the position given in the
	233	* `text matrix'. We must therefore track both as well.
	234	* They start off at -1 (unset).
	235	*/
	236	lx = ly = -1;
	237	x = y = -1;
	238
	239	frag = page->first_text;
	240	while (frag) {
	241	/*
	242	* For compactness, I'm going to group text fragments
	243	* into subsequences that use the same font+size. So
	244	* first find the end of this subsequence.
	245	*/
	246	for (frag_end = frag;
	247	(frag_end &&
	248	frag_end->fe == frag->fe &&
	249	frag_end->fontsize == frag->fontsize);
	250	frag_end = frag_end->next);
	251
	252	/*
	253	* Now select the text fragment, and prepare to display
	254	* the text.
	255	*/
43341922	256	objstream(cstr, "/");
43341922	257	objstream(cstr, frag->fe->name);
7c8c4239	258	sprintf(buf, " %d Tf ", frag->fontsize);
43341922	259	objstream(cstr, buf);
43341922	260
7c8c4239	261	while (frag && frag != frag_end) {
	262	/*
	263	* Place the text position for the first piece of
	264	* text.
	265	*/
	266	if (lx < 0) {
	267	sprintf(buf, "1 0 0 1 %g %g Tm ",
	268	frag->x/4096.0, frag->y/4096.0);
	269	} else {
	270	sprintf(buf, "%g %g Td ",
	271	(frag->x - lx)/4096.0, (frag->y - ly)/4096.0);
	272	}
43341922	273	objstream(cstr, buf);
7c8c4239	274	lx = x = frag->x;
	275	ly = y = frag->y;
	276
	277	/*
	278	* See if we're going to use Tj (show a single
	279	* string) or TJ (show an array of strings with
	280	* x-spacings between them). We determine this by
	281	* seeing if there's more than one text fragment in
	282	* sequence with the same y-coordinate.
	283	*/
	284	if (frag->next && frag->next != frag_end &&
	285	frag->next->y == y) {
	286	/*
	287	* The TJ strategy.
	288	*/
	289	objstream(cstr, "[");
	290	while (frag && frag != frag_end && frag->y == y) {
	291	if (frag->x != x) {
	292	sprintf(buf, "%g",
	293	(x - frag->x) * 1000.0 /
	294	(4096.0 * frag->fontsize));
	295	objstream(cstr, buf);
	296	}
	297	pdf_string(objstream, cstr, frag->text);
	298	x = frag->x + frag->width;
	299	frag = frag->next;
	300	}
	301	objstream(cstr, "]TJ\n");
	302	} else
	303	{
	304	/*
	305	* The Tj strategy.
	306	*/
	307	pdf_string(objstream, cstr, frag->text);
	308	objstream(cstr, "Tj\n");
	309	frag = frag->next;
	310	}
43341922	311	}
43341922	312	}
	313	objstream(cstr, "ET");
	314
138d7ffb	315	/*
	316	* Also, we want an annotation dictionary containing the
	317	* cross-references from this page.
	318	*/
	319	if (page->first_xref) {
	320	xref *xr;
	321	objtext(opage, "/Annots [\n");
	322
	323	for (xr = page->first_xref; xr; xr = xr->next) {
	324	object *annot;
	325	char buf[256];
	326
	327	annot = new_object(&olist);
	328	objref(opage, annot);
	329	objtext(opage, "\n");
	330
	331	objtext(annot, "<<\n/Type /Annot\n/Subtype /Link\n/Rect [");
	332	sprintf(buf, "%g %g %g %g",
	333	xr->lx / 4096.0, xr->by / 4096.0,
	334	xr->rx / 4096.0, xr->ty / 4096.0);
	335	objtext(annot, buf);
	336	objtext(annot, "]\n/Border [0 0 0]\n");
	337
	338	if (xr->dest.type == PAGE) {
	339	objtext(annot, "/Dest [");
	340	objref(annot, (object *)xr->dest.page->spare);
	341	objtext(annot, " /XYZ null null null]\n");
	342	} else {
7c8c4239	343	objtext(annot, "/A <<\n/Type /Action\n/S /URI\n/URI ");
	344	pdf_string(objtext, annot, xr->dest.url);
	345	objtext(annot, "\n>>\n");
138d7ffb	346	}
	347
	348	objtext(annot, ">>\n");
	349	}
	350
	351	objtext(opage, "]\n");
	352	}
	353
43341922	354	objtext(opage, ">>\n");
	355	}
	356
	357	/*
f0e51ce1	358	* Set up the outlines dictionary.
f0e51ce1	359	*/
a615a778	360	if (outlines) {
f0e51ce1	361	int topcount;
	362	char buf[80];
	363
	364	objtext(outlines, "<<\n/Type /Outlines\n");
	365	topcount = make_outline(outlines, doc->outline_elements,
	366	doc->n_outline_elements, TRUE);
	367	sprintf(buf, "/Count %d\n>>\n", topcount);
	368	objtext(outlines, buf);
	369	}
	370
	371	/*
43341922	372	* Assemble the final linear form of every object.
	373	*/
	374	for (o = olist.head; o; o = o->next) {
	375	rdstringc rs = {0, 0, NULL};
	376	char text[80];
	377
	378	sprintf(text, "%d 0 obj\n", o->number);
	379	rdaddsc(&rs, text);
	380
	381	if (!o->main.text && o->stream.text) {
	382	sprintf(text, "<<\n/Length %d\n>>\n", o->stream.pos);
	383	rdaddsc(&o->main, text);
	384	}
	385
	386	assert(o->main.text);
	387	rdaddsc(&rs, o->main.text);
	388	sfree(o->main.text);
	389
	390	if (rs.text[rs.pos-1] != '\n')
	391	rdaddc(&rs, '\n');
	392
	393	if (o->stream.text) {
	394	/*
	395	* FIXME: If we ever start compressing stream data then
	396	* it will have zero bytes in it, so we'll have to be
	397	* more careful than this.
	398	*/
	399	rdaddsc(&rs, "stream\n");
	400	rdaddsc(&rs, o->stream.text);
	401	rdaddsc(&rs, "\nendstream\n");
	402	sfree(o->stream.text);
	403	}
	404
	405	rdaddsc(&rs, "endobj\n");
	406
	407	o->final = rs.text;
	408	o->size = rs.pos;
	409	}
	410
	411	/*
	412	* Write out the PDF file.
	413	*/
	414
	415	fp = fopen(filename, "wb");
	416	if (!fp) {
	417	error(err_cantopenw, filename);
	418	return;
	419	}
	420
	421	/*
09358aa7	422	* Header. I'm going to put the version IDs in the header as
09358aa7	423	* well, simply in PDF comments.
43341922	424	*/
43341922	425	fileoff = fprintf(fp, "%%PDF-1.3\n");
09358aa7	426	for (p = sourceform; p; p = p->next)
	427	if (p->type == para_VersionID)
	428	fileoff += pdf_versionid(fp, p->words);
43341922	429
	430	/*
	431	* Body
	432	*/
	433	for (o = olist.head; o; o = o->next) {
	434	o->fileoff = fileoff;
	435	fwrite(o->final, 1, o->size, fp);
	436	fileoff += o->size;
	437	}
	438
	439	/*
	440	* Cross-reference table
	441	*/
	442	fprintf(fp, "xref\n");
	443	assert(olist.head->number == 1);
	444	fprintf(fp, "0 %d\n", olist.tail->number + 1);
	445	fprintf(fp, "0000000000 65535 f \n");
	446	for (o = olist.head; o; o = o->next) {
	447	char entry[40];
	448	sprintf(entry, "%010d 00000 n \n", o->fileoff);
	449	assert(strlen(entry) == 20);
	450	fputs(entry, fp);
	451	}
	452
	453	/*
	454	* Trailer
	455	*/
	456	fprintf(fp, "trailer\n<<\n/Size %d\n/Root %d 0 R\n>>\n",
	457	olist.tail->number + 1, cat->number);
	458	fprintf(fp, "startxref\n%d\n%%%%EOF\n", fileoff);
	459
	460	fclose(fp);
	461
	462	sfree(filename);
	463	}
	464
	465	static object new_object(objlist list)
	466	{
f1530049	467	object *obj = snew(object);
43341922	468
	469	obj->list = list;
	470
	471	obj->main.text = NULL;
	472	obj->main.pos = obj->main.size = 0;
	473	obj->stream.text = NULL;
	474	obj->stream.pos = obj->stream.size = 0;
	475
	476	obj->number = list->number++;
	477
	478	obj->next = NULL;
	479	if (list->tail)
	480	list->tail->next = obj;
	481	else
	482	list->head = obj;
	483	list->tail = obj;
	484
	485	obj->size = 0;
	486	obj->final = NULL;
	487
	488	return obj;
	489	}
	490
	491	static void objtext(object o, char const text)
	492	{
	493	rdaddsc(&o->main, text);
	494	}
	495
	496	static void objstream(object o, char const text)
	497	{
	498	rdaddsc(&o->stream, text);
	499	}
	500
	501	static void objref(object o, object dest)
	502	{
	503	char buf[40];
	504	sprintf(buf, "%d 0 R", dest->number);
	505	rdaddsc(&o->main, buf);
	506	}
	507
	508	static void make_pages_node(object node, object parent, page_data *first,
	509	page_data last, object resources)
	510	{
	511	int count;
	512	page_data *page;
	513	char buf[80];
	514
	515	objtext(node, "<<\n/Type /Pages\n");
	516	if (parent) {
	517	objtext(node, "/Parent ");
	518	objref(node, parent);
	519	objtext(node, "\n");
	520	}
	521
	522	/*
	523	* Count the pages in this stretch, to see if there are few
	524	* enough to reference directly.
	525	*/
	526	count = 0;
	527	for (page = first; page; page = page->next) {
	528	count++;
	529	if (page == last)
	530	break;
	531	}
532
533	sprintf(buf, "/Count %d\n/Kids [\n", count);
534	objtext(node, buf);
535
536	if (count > TREE_BRANCH) {
537	int i;
538	page_data thisfirst, thislast;
539
540	page = first;
541
542	for (i = 0; i < TREE_BRANCH; i++) {
543	int number = (i+1) * count / TREE_BRANCH - i * count / TREE_BRANCH;
544	thisfirst = page;
545	while (number--) {
546	thislast = page;
547	page = page->next;
548	}
549
550	if (thisfirst == thislast) {
551	objref(node, (object *)thisfirst->spare);
552	objtext((object *)thisfirst->spare, "/Parent ");
553	objref((object *)thisfirst->spare, node);
554	objtext((object *)thisfirst->spare, "\n");
555	} else {
556	object *newnode = new_object(node->list);
557	make_pages_node(newnode, node, thisfirst, thislast, NULL);
558	objref(node, newnode);
559	}
560	objtext(node, "\n");
561	}
562
563	assert(thislast == last \|\| page == NULL);
564
565	} else {
566	for (page = first; page; page = page->next) {
567	objref(node, (object *)page->spare);
568	objtext(node, "\n");
569	objtext((object *)page->spare, "/Parent ");
570	objref((object *)page->spare, node);
571	objtext((object *)page->spare, "\n");
572	if (page == last)
573	break;
574	}
575	}
576
577	objtext(node, "]\n");
578
579	if (resources) {
580	objtext(node, "/Resources ");
581	objref(node, resources);
582	objtext(node, "\n");
583	}
584
585	objtext(node, ">>\n");
586	}
f0e51ce1	587
	588	/*
	589	* In text on the page, PDF uses the PostScript font model, which
	590	* means that glyphs are identified by PS strings and hence font
	591	* encoding can be managed independently of the supplied encoding
	592	* of the font. However, in the document outline, the PDF spec
ce1b04aa	593	* encodes in either PDFDocEncoding (a custom superset of
ce1b04aa	594	* ISO-8859-1) or UTF-16BE.
f0e51ce1	595	*/
ce1b04aa	596	static char pdf_outline_convert(wchar_t s, int *len) {
	597	char *ret;
	598
	599	ret = utoa_careful_dup(s, CS_PDF);
	600
	601	/*
	602	* Very silly special case: if the returned string begins with
	603	* FE FF, then the PDF reader will mistake it for a UTF-16BE
	604	* string. So in this case we give up on PDFDocEncoding and
	605	* encode it in UTF-16 straight away.
	606	*/
	607	if (ret && ret[0] == '\xFE' && ret[1] == '\xFF') {
	608	sfree(ret);
	609	ret = NULL;
f0e51ce1	610	}
ce1b04aa	611
	612	if (!ret) {
	613	ret = utoa_dup_len(s, CS_UTF16BE, len);
	614	} else {
	615	*len = strlen(ret);
f0e51ce1	616	}
ce1b04aa	617
ce1b04aa	618	return ret;
f0e51ce1	619	}
f0e51ce1	620
f0e51ce1	621	static int make_outline(object parent, outline_element items, int n,
	622	int open)
	623	{
	624	int level, totalcount = 0;
	625	outline_element *itemp;
	626	object curr, prev = NULL, first = NULL, last = NULL;
f0e51ce1	627
	628	assert(n > 0);
	629
	630	level = items->level;
	631
	632	while (n > 0) {
7c8c4239	633	char *title;
ce1b04aa	634	int titlelen;
f0e51ce1	635
	636	/*
	637	* Here we expect to be sitting on an item at the given
	638	* level. So we start by constructing an outline entry for
	639	* that item.
	640	*/
	641	assert(items->level == level);
	642
ce1b04aa	643	title = pdf_outline_convert(items->pdata->outline_title, &titlelen);
f0e51ce1	644
	645	totalcount++;
	646	curr = new_object(parent->list);
	647	if (!first) first = curr;
	648	last = curr;
7c8c4239	649	objtext(curr, "<<\n/Title ");
ce1b04aa	650	pdf_string_len(objtext, curr, title, titlelen);
7c8c4239	651	objtext(curr, "\n/Parent ");
f0e51ce1	652	objref(curr, parent);
f0e51ce1	653	objtext(curr, "\n/Dest [");
be76d597	654	objref(curr, (object *)items->pdata->first->page->spare);
f0e51ce1	655	objtext(curr, " /XYZ null null null]\n");
	656	if (prev) {
	657	objtext(curr, "/Prev ");
	658	objref(curr, prev);
	659	objtext(curr, "\n");
	660
	661	objtext(prev, "/Next ");
	662	objref(prev, curr);
	663	objtext(prev, "\n>>\n");
	664	}
	665	prev = curr;
	666
	667	items++, n--;
	668	for (itemp = items; itemp < items+n && itemp->level > level;
	669	itemp++);
	670
	671	if (itemp > items) {
	672	char buf[80];
	673	int count = make_outline(curr, items, itemp - items, FALSE);
	674	if (!open)
	675	count = -count;
	676	else
	677	totalcount += count;
	678	sprintf(buf, "/Count %d\n", count);
	679	objtext(curr, buf);
	680	}
	681
	682	n -= itemp - items;
	683	items = itemp;
	684	}
	685	objtext(prev, ">>\n");
	686
	687	assert(first && last);
	688	objtext(parent, "/First ");
	689	objref(parent, first);
	690	objtext(parent, "\n/Last ");
	691	objref(parent, last);
	692	objtext(parent, "\n");
	693
	694	return totalcount;
	695	}
09358aa7	696
	697	static int pdf_versionid(FILE fp, word words)
	698	{
	699	int ret;
	700
	701	ret = fprintf(fp, "%% ");
	702
	703	for (; words; words = words->next) {
	704	char *text;
	705	int type;
	706
	707	switch (words->type) {
	708	case word_HyperLink:
	709	case word_HyperEnd:
	710	case word_UpperXref:
	711	case word_LowerXref:
	712	case word_XrefEnd:
	713	case word_IndexRef:
	714	continue;
	715	}
	716
	717	type = removeattr(words->type);
	718
	719	switch (type) {
	720	case word_Normal:
e4ea58f8	721	text = utoa_dup(words->text, CS_ASCII);
09358aa7	722	break;
	723	case word_WhiteSpace:
	724	text = dupstr(" ");
	725	break;
	726	case word_Quote:
	727	text = dupstr("'");
	728	break;
	729	}
	730
	731	fputs(text, fp);
	732	ret += strlen(text);
	733	sfree(text);
	734	}
	735
	736	ret += fprintf(fp, "\n");
	737
	738	return ret;
	739	}
7c8c4239	740
ce1b04aa	741	static void pdf_string_len(void (add)(object , char const *),
ce1b04aa	742	object o, char const str, int len)
7c8c4239	743	{
	744	char const *p;
	745
	746	add(o, "(");
ce1b04aa	747	for (p = str; len > 0; p++, len--) {
	748	char c[10];
	749	if (p < ' ' \|\| p > '~') {
	750	sprintf(c, "\\%03o", 0xFF & (int)*p);
	751	} else {
	752	int n = 0;
	753	if (p == '\\' \|\| p == '(' \|\| *p == ')')
	754	c[n++] = '\\';
	755	c[n++] = *p;
	756	c[n] = '\0';
	757	}
7c8c4239	758	add(o, c);
	759	}
	760	add(o, ")");
	761	}
ce1b04aa	762
	763	static void pdf_string(void (add)(object , char const *),
	764	object o, char const str)
	765	{
	766	pdf_string_len(add, o, str, strlen(str));
	767	}