X-Git-Url: https://git.distorted.org.uk/~mdw/sgt/halibut/blobdiff_plain/7c8c4239f226d2dabdd84dabae9f5869d6d64f2f..944c4044572da20ed1fb9bc01085221856b53c1f:/bk_pdf.c

diff --git a/bk_pdf.c b/bk_pdf.c
index f6babe5..6daebc6 100644
--- a/bk_pdf.c
+++ b/bk_pdf.c
@@ -5,40 +5,15 @@
 #include <assert.h>
 #include "halibut.h"
 #include "paper.h"
+#include "deflate.h"
 
-#define TREE_BRANCH 2		       /* max branching factor in page tree */
+#define TREE_BRANCH 8		       /* max branching factor in page tree */
 
 paragraph *pdf_config_filename(char *filename)
 {
-    paragraph *p;
-    wchar_t *ufilename, *up;
-    int len;
-
-    p = mknew(paragraph);
-    memset(p, 0, sizeof(*p));
-    p->type = para_Config;
-    p->next = NULL;
-    p->fpos.filename = "<command line>";
-    p->fpos.line = p->fpos.col = -1;
-
-    ufilename = ufroma_dup(filename);
-    len = ustrlen(ufilename) + 2 + lenof(L"pdf-filename");
-    p->keyword = mknewa(wchar_t, len);
-    up = p->keyword;
-    ustrcpy(up, L"pdf-filename");
-    up = uadv(up);
-    ustrcpy(up, ufilename);
-    up = uadv(up);
-    *up = L'\0';
-    assert(up - p->keyword < len);
-    sfree(ufilename);
-
-    return p;
+    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;
@@ -53,15 +28,18 @@ struct objlist_Tag {
     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 objdest(object *o, page_data *p);
+
+static int is_std_font(char const *name);
 
 static void make_pages_node(object *node, object *parent, page_data *first,
-			    page_data *last, object *resources);
+			    page_data *last, object *resources,
+			    object *mediabox);
 static int make_outline(object *parent, outline_element *start, int n,
 			int open);
 static int pdf_versionid(FILE *fp, word *words);
@@ -72,12 +50,11 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
     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;
+    object *o, *info, *cat, *outlines, *pages, *resources, *mediabox;
     int fileoff;
 
     IGNORE(keywords);
@@ -85,10 +62,10 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
 
     filename = dupstr("output.pdf");
     for (p = sourceform; p; p = p->next) {
-	if (p->type == para_Config && p->parent) {
+	if (p->type == para_Config) {
 	    if (!ustricmp(p->keyword, L"pdf-filename")) {
 		sfree(filename);
-		filename = utoa_dup(uadv(p->keyword));
+		filename = dupstr(adv(p->origkeyword));
 	    }
 	}
     }
@@ -96,31 +73,67 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
     olist.head = olist.tail = NULL;
     olist.number = 1;
 
+    {
+	char buf[256];
+
+	info = new_object(&olist);
+	objtext(info, "<<\n");
+	if (doc->n_outline_elements > 0) {
+	    char *title;
+	    int titlelen;
+
+	    title =
+	       pdf_outline_convert(doc->outline_elements->pdata->outline_title,
+					&titlelen);
+	    objtext(info, "/Title ");
+	    pdf_string_len(objtext, info, title, titlelen);
+	    sfree(title);
+	    objtext(info, "\n");
+	}
+	objtext(info, "/Producer ");
+	sprintf(buf, "Halibut, %s", version);
+	pdf_string(objtext, info, buf);
+	objtext(info, "\n>>\n");
+    }
+
     cat = new_object(&olist);
-    outlines = 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.
+     * pages objects, and the pagelabels dictionary.
      */
-    objtext(cat, "<<\n/Type /Catalog\n/Outlines ");
-    objref(cat, outlines);
+    objtext(cat, "<<\n/Type /Catalog");
+    if (outlines) {
+	objtext(cat, "\n/Outlines ");
+	objref(cat, outlines);
+    }
     objtext(cat, "\n/Pages ");
     objref(cat, pages);
-    objtext(cat, "\n/PageMode /UseOutlines\n>>\n");
+    /* Halibut just numbers pages 1, 2, 3, ... */
+    objtext(cat, "\n/PageLabels<</Nums[0<</S/D>>]>>");
+    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");
+    objtext(resources, "<<\n/ProcSet [/PDF/Text]\n/Font <<\n");
     for (fe = doc->fonts->head; fe; fe = fe->next) {
 	char fname[40];
-	int i;
-	object *font;
+	char buf[80];
+	int i, prev;
+	object *font, *fontdesc;
+	int flags;
+	font_info const *fi = fe->font->info;
 
 	sprintf(fname, "f%d", font_index++);
 	fe->name = dupstr(fname);
@@ -133,46 +146,266 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
 	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");
+	/*
+	 * Construct those parts of the font descriptor that don't dependd
+	 * on the file format.
+	 */
+	if (!is_std_font(fe->font->info->name)) {
+	    fontdesc = new_object(&olist);
+
+#define FF_FIXEDPITCH	0x00000001
+#define FF_SERIF	0x00000002
+#define FF_SYMBOLIC	0x00000004
+#define FF_SCRIPT	0x00000008
+#define FF_NONSYMBOLIC	0x00000020
+#define FF_ITALIC	0x00000040
+#define FF_ALLCAP	0x00010000
+#define FF_SMALLCAP	0x00020000
+#define FF_FORCEBOLD	0x00040000
+
+	    objtext(fontdesc, "<<\n/Type /FontDescriptor\n/Name /");
+	    objtext(fontdesc, fi->name);
+	    flags = 0;
+	    if (fi->italicangle) flags |= FF_ITALIC;
+	    flags |= FF_NONSYMBOLIC;
+	    sprintf(buf, "\n/Flags %d\n", flags);
+	    objtext(fontdesc, buf);
+	    sprintf(buf, "/FontBBox [%g %g %g %g]\n", fi->fontbbox[0],
+		    fi->fontbbox[1], fi->fontbbox[2], fi->fontbbox[3]);
+	    objtext(fontdesc, buf);
+	    sprintf(buf, "/ItalicAngle %g\n", fi->italicangle);
+	    objtext(fontdesc, buf);
+	    sprintf(buf, "/Ascent %g\n", fi->ascent);
+	    objtext(fontdesc, buf);
+	    sprintf(buf, "/Descent %g\n", fi->descent);
+	    objtext(fontdesc, buf);
+	    sprintf(buf, "/CapHeight %g\n", fi->capheight);
+	    objtext(fontdesc, buf);
+	    sprintf(buf, "/XHeight %g\n", fi->xheight);
+	    objtext(fontdesc, buf);
+	    sprintf(buf, "/StemH %g\n", fi->stemh);
+	    objtext(fontdesc, buf);
+	    sprintf(buf, "/StemV %g\n", fi->stemv);
+	    objtext(fontdesc, buf);
+	}
 
-	{
-	    object *widths = new_object(&olist);
-	    objtext(font, "/FirstChar 0\n/LastChar 255\n/Widths ");
-	    objref(font, widths);
-	    objtext(font, "\n");
-	    objtext(widths, "[\n");
+	objtext(font, "<<\n/Type /Font\n/BaseFont /");
+	objtext(font, fe->font->info->name);
+	if (fe->font->info->filetype == TRUETYPE) {
+	    object *cidfont = new_object(&olist);
+	    object *cmap = new_object(&olist);
+	    unsigned short ranges[256];
+	    unsigned startidx, nranges, nchars;
+	    int start;
+
+	    objtext(font, "/Subtype/Type0\n/Encoding ");
+	    objtext(cmap, "<</Type/CMap\n/CMapName/");
+	    objtext(cmap, fe->name);
+	    objtext(cmap, "\n/CIDSystemInfo<</Registry(Adobe)"
+		    "/Ordering(Identity)/Supplement 0>>\n");
+	    objstream(cmap, "%!PS-Adobe-3.0 Resource-CMap\n"
+		      "%%DocumentNeededResources: procset CIDInit\n"
+		      "%%IncludeResource: procset CIDInit\n"
+		      "%%BeginResource: CMap ");
+	    objstream(cmap, fe->name);
+	    objstream(cmap, "\n%%Title (");
+	    objstream(cmap, fe->name);
+	    objstream(cmap, " Adobe Identity 0)\n%%Version: 1\n%%EndComments\n");
+	    objstream(cmap, "/CIDInit/ProcSet findresource begin\n");
+	    objstream(cmap, "12 dict begin begincmap\n");
+	    objstream(cmap, "/CIDSystemInfo 3 dict dup begin\n"
+		      "/Registry(Adobe)def/Ordering(Identity)def"
+		      "/Supplement 0 def end def\n");
+	    objstream(cmap, "/CMapName/");
+	    objstream(cmap, fe->name);
+	    objstream(cmap, " def/CMapType 0 def/WMode 0 def\n");
+	    objstream(cmap, "1 begincodespacerange<00><FF>"
+		      "endcodespacerange\n");
+	    start = -1; nranges = nchars = 0;
 	    for (i = 0; i < 256; i++) {
-		char buf[80];
+		unsigned idx;
+
+		ranges[i] = 0;
+		if (fe->vector[i] == NOGLYPH)
+		    continue;
+		idx = sfnt_glyphtoindex(fe->font->info->fontfile,
+					fe->vector[i]);
+		if (start >= 0 && idx - startidx == (unsigned)(i - start)) {
+		    if (ranges[start] == 1) {
+			nranges++; nchars--;
+		    }
+		    ranges[start] = i - start + 1;
+		} else {
+		    ranges[i] = 1;
+		    start = i;
+		    startidx = idx;
+		    nchars++;
+		}
+	    }
+	    i = 0;
+	    while (nranges) {
+		unsigned blk = nranges > 100 ? 100 : nranges;
+		nranges -= blk;
+		sprintf(buf, "%u ", blk);
+		objstream(cmap, buf);
+		objstream(cmap, "begincidrange\n");
+		while (blk) {
+		    if (ranges[i] > 1) {
+			sprintf(buf, "<%02X>", i);
+			objstream(cmap, buf);
+			sprintf(buf, "<%02X>", i + ranges[i] - 1);
+			objstream(cmap, buf);
+			sprintf(buf, "%hu\n",
+				sfnt_glyphtoindex(fe->font->info->fontfile,
+						  fe->vector[i]));
+			objstream(cmap, buf);
+			blk--;
+		    }
+		    i++;
+		}
+		objstream(cmap, "endcidrange\n");
+	    }
+	    i = 0;
+	    while (nchars) {
+		unsigned blk = nchars > 100 ? 100 : nchars;
+		nchars -= blk;
+		sprintf(buf, "%u ", blk);
+		objstream(cmap, buf);
+		objstream(cmap, "begincidchar\n");
+		while (blk) {
+		    if (ranges[i] == 1) {
+			sprintf(buf, "<%02X>", i);
+			objstream(cmap, buf);
+			sprintf(buf, "%hu\n",
+				sfnt_glyphtoindex(fe->font->info->fontfile,
+						  fe->vector[i]));
+			objstream(cmap, buf);
+			blk--;
+		    }
+		    i++;
+		}
+		objstream(cmap, "endcidchar\n");
+	    }
+	    objstream(cmap, "endcmap CMapName currentdict /CMap "
+		      "defineresource pop end end\n%%EndResource\n%%EOF\n");
+
+	    objref(font, cmap);
+	    objtext(font, "\n/DescendantFonts[");
+	    objref(font, cidfont);
+	    objtext(font, "]\n");
+	    objtext(cidfont, "<<\n/Type/Font\n/Subtype/CIDFontType2\n"
+		    "/BaseFont/");
+	    objtext(cidfont, fe->font->info->name);
+	    objtext(cidfont, "\n/CIDSystemInfo<</Registry(Adobe)"
+		    "/Ordering(Identity)/Supplement 0>>\n");
+	    objtext(cidfont, "/FontDescriptor ");
+	    objref(cidfont, fontdesc);
+	    objtext(cidfont, "\n/W[0[");
+	    for (i = 0; i < (int)sfnt_nglyphs(fe->font->info->fontfile); i++) {
+		char buf[20];
 		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);
+		width = find_width(fe->font,
+			       sfnt_indextoglyph(fe->font->info->fontfile, i));
+		sprintf(buf, "%g ", 1000.0 * width / FUNITS_PER_PT);
+		objtext(cidfont, buf);
+	    }
+	    objtext(cidfont, "]]>>\n");
+	} else {
+	    objtext(font, "/Subtype /Type1\n");
+	    objtext(font, "\n/Encoding <<\n/Type /Encoding\n/Differences [");
+
+	    for (i = 0; i < 256; i++) {
+		char buf[20];
+		if (fe->vector[i] == NOGLYPH)
+		    continue;
+		if (i != prev + 1) {
+		    sprintf(buf, "\n%d", i);
+		    objtext(font, buf);
+		}
+		objtext(font, i % 8 ? "/" : "\n/");
+		objtext(font, glyph_extern(fe->vector[i]));
+		prev = i;
+	    }
+
+	    objtext(font, "\n]\n>>\n");
+	    if (!is_std_font(fe->font->info->name)){
+		object *widths = new_object(&olist);
+		int firstchar = -1, lastchar = -1;
+		for (i = 0; i < 256; i++)
+		    if (fe->vector[i] != NOGLYPH) {
+			if (firstchar < 0) firstchar = i;
+			lastchar = i;
+		    }
+		sprintf(buf, "/FirstChar %d\n/LastChar %d\n/Widths ",
+			firstchar, lastchar);
+		objtext(font, buf);
+		objref(font, widths);
+		objtext(font, "\n");
+		objtext(widths, "[\n");
+		for (i = firstchar; i <= lastchar; i++) {
+		    double width;
+		    if (fe->vector[i] == NOGLYPH)
+			width = 0.0;
+		    else
+			width = find_width(fe->font, fe->vector[i]);
+		    sprintf(buf, "%g\n", 1000.0 * width / FUNITS_PER_PT);
+		    objtext(widths, buf);
+		}
+		objtext(widths, "]\n");
+		objtext(font, "/FontDescriptor ");
+		objref(font, fontdesc);
 	    }
-	    objtext(widths, "]\n");
+
 	}
 
-	objtext(font, ">>\n");
+	if (!is_std_font(fe->font->info->name)) {
+	    if (fi->fontfile && fi->filetype == TYPE1) {
+		object *fontfile = new_object(&olist);
+		size_t len;
+		char *ffbuf;
+
+		pf_part1((font_info *)fi, &ffbuf, &len);
+		objstream_len(fontfile, ffbuf, len);
+		sfree(ffbuf);
+		sprintf(buf, "<<\n/Length1 %lu\n", (unsigned long)len);
+		objtext(fontfile, buf);
+		pf_part2((font_info *)fi, &ffbuf, &len);
+		objstream_len(fontfile, ffbuf, len);
+		sfree(ffbuf);
+		sprintf(buf, "/Length2 %lu\n", (unsigned long)len);
+		objtext(fontfile, buf);
+		objtext(fontfile, "/Length3 0\n");
+		objtext(fontdesc, "/FontFile ");
+		objref(fontdesc, fontfile);
+	    } else if (fi->fontfile && fi->filetype == TRUETYPE) {
+		object *fontfile = new_object(&olist);
+		size_t len;
+		char *ffbuf;
+
+		sfnt_data((font_info *)fi, &ffbuf, &len);
+		objstream_len(fontfile, ffbuf, len);
+		sprintf(buf, "<<\n/Length1 %lu\n", (unsigned long)len);
+		objtext(fontfile, buf);
+		objtext(fontdesc, "/FontFile2 ");
+		objref(fontdesc, fontfile);
+	    }
+	    objtext(fontdesc, "\n>>\n");
+	}
+
+	objtext(font, "\n>>\n");
     }
     objtext(resources, ">>\n>>\n");
 
+    {
+	char buf[255];
+	mediabox = new_object(&olist);
+	sprintf(buf, "[0 0 %g %g]\n",
+		doc->paper_width / FUNITS_PER_PT,
+		doc->paper_height / FUNITS_PER_PT);
+	objtext(mediabox, buf);
+    }
+
     /*
      * Define the page objects for each page, and get each one
      * ready to have a `Parent' specification added to it.
@@ -188,12 +421,11 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
     /*
      * Recursively build the page tree.
      */
-    make_pages_node(pages, NULL, doc->pages, NULL, resources);
+    make_pages_node(pages, NULL, doc->pages, NULL, resources, mediabox);
 
     /*
      * Create and render the individual pages.
      */
-    pageno = 0;
     for (page = doc->pages; page; page = page->next) {
 	object *opage, *cstr;
 	rect *r;
@@ -213,11 +445,9 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
 	 * 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.
+	 * So we don't need a /Resources entry here.  The same applies
+	 * to /MediaBox.
 	 */
-	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
@@ -233,8 +463,9 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
 	 */
 	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);
+	    sprintf(buf, "%g %g %g %g re f\n",
+		    r->x / FUNITS_PER_PT, r->y / FUNITS_PER_PT,
+		    r->w / FUNITS_PER_PT, r->h / FUNITS_PER_PT);
 	    objstream(cstr, buf);
 	}
 
@@ -278,10 +509,11 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
 		 */
 		if (lx < 0) {
 		    sprintf(buf, "1 0 0 1 %g %g Tm ",
-			    frag->x/4096.0, frag->y/4096.0);
+			    frag->x/FUNITS_PER_PT, frag->y/FUNITS_PER_PT);
 		} else {
 		    sprintf(buf, "%g %g Td ",
-			    (frag->x - lx)/4096.0, (frag->y - ly)/4096.0);
+			    (frag->x - lx)/FUNITS_PER_PT,
+			    (frag->y - ly)/FUNITS_PER_PT);
 		}
 		objstream(cstr, buf);
 		lx = x = frag->x;
@@ -304,7 +536,7 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
 			if (frag->x != x) {
 			    sprintf(buf, "%g",
 				    (x - frag->x) * 1000.0 /
-				    (4096.0 * frag->fontsize));
+				    (FUNITS_PER_PT * frag->fontsize));
 			    objstream(cstr, buf);
 			}
 			pdf_string(objstream, cstr, frag->text);
@@ -334,31 +566,25 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
 	    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 [");
+		objtext(opage, "<</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");
+			xr->lx / FUNITS_PER_PT, xr->by / FUNITS_PER_PT,
+			xr->rx / FUNITS_PER_PT, xr->ty / FUNITS_PER_PT);
+		objtext(opage, buf);
+		objtext(opage, "]/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");
+		    objtext(opage, "/Dest");
+		    objdest(opage, xr->dest.page);
 		} else {
-		    objtext(annot, "/A <<\n/Type /Action\n/S /URI\n/URI ");
-		    pdf_string(objtext, annot, xr->dest.url);
-		    objtext(annot, "\n>>\n");
+		    objtext(opage, "/A<</S/URI/URI");
+		    pdf_string(objtext, opage, xr->dest.url);
+		    objtext(opage, ">>");
 		}
 
-		objtext(annot, ">>\n");
+		objtext(opage, ">>\n");
 	    }
 
 	    objtext(opage, "]\n");
@@ -370,7 +596,7 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
     /*
      * Set up the outlines dictionary.
      */
-    {
+    if (outlines) {
 	int topcount;
 	char buf[80];
 
@@ -387,12 +613,28 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
     for (o = olist.head; o; o = o->next) {
 	rdstringc rs = {0, 0, NULL};
 	char text[80];
+	deflate_compress_ctx *zcontext;
+	void *zbuf;
+	int zlen;
 
 	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);
+	if (o->stream.text) {
+	    if (!o->main.text)
+		rdaddsc(&o->main, "<<\n");
+#ifdef PDF_NOCOMPRESS
+	    zlen = o->stream.pos;
+	    zbuf = snewn(zlen, char);
+	    memcpy(zbuf, o->stream.text, zlen);
+	    sprintf(text, "/Length %d\n>>\n", zlen);
+#else
+	    zcontext = deflate_compress_new(DEFLATE_TYPE_ZLIB);
+	    deflate_compress_data(zcontext, o->stream.text, o->stream.pos,
+				  DEFLATE_END_OF_DATA, &zbuf, &zlen);
+	    deflate_compress_free(zcontext);
+	    sprintf(text, "/Filter/FlateDecode\n/Length %d\n>>\n", zlen);
+#endif
 	    rdaddsc(&o->main, text);
 	}
 
@@ -404,15 +646,11 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
 	    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);
+	    rdaddsn(&rs, zbuf, zlen);
 	    rdaddsc(&rs, "\nendstream\n");
 	    sfree(o->stream.text);
+	    sfree(zbuf);
 	}
 
 	rdaddsc(&rs, "endobj\n");
@@ -425,7 +663,10 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
      * Write out the PDF file.
      */
 
-    fp = fopen(filename, "wb");
+    if (!strcmp(filename, "-"))
+	fp = stdout;
+    else
+	fp = fopen(filename, "wb");
     if (!fp) {
 	error(err_cantopenw, filename);
 	return;
@@ -433,9 +674,11 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
 
     /*
      * Header. I'm going to put the version IDs in the header as
-     * well, simply in PDF comments.
+     * well, simply in PDF comments.  The PDF Reference also suggests
+     * that binary PDF files contain four top-bit-set characters in
+     * the second line.
      */
-    fileoff = fprintf(fp, "%%PDF-1.3\n");
+    fileoff = fprintf(fp, "%%PDF-1.3\n%% L\xc3\xba\xc3\xb0""a\n");
     for (p = sourceform; p; p = p->next)
 	if (p->type == para_VersionID)
 	    fileoff += pdf_versionid(fp, p->words);
@@ -466,18 +709,19 @@ void pdf_backend(paragraph *sourceform, keywordlist *keywords,
     /*
      * Trailer
      */
-    fprintf(fp, "trailer\n<<\n/Size %d\n/Root %d 0 R\n>>\n",
-	    olist.tail->number + 1, cat->number);
+    fprintf(fp, "trailer\n<<\n/Size %d\n/Root %d 0 R\n/Info %d 0 R\n>>\n",
+	    olist.tail->number + 1, cat->number, info->number);
     fprintf(fp, "startxref\n%d\n%%%%EOF\n", fileoff);
 
-    fclose(fp);
+    if (fp != stdout)
+	fclose(fp);
 
     sfree(filename);
 }
 
-static object *new_object(objlist *list)
+object *new_object(objlist *list)
 {
-    object *obj = mknew(object);
+    object *obj = snew(object);
 
     obj->list = list;
 
@@ -501,12 +745,17 @@ static object *new_object(objlist *list)
     return obj;
 }
 
-static void objtext(object *o, char const *text)
+void objtext(object *o, char const *text)
 {
     rdaddsc(&o->main, text);
 }
 
-static void objstream(object *o, char const *text)
+void objstream_len(object *o, char const *text, size_t len)
+{
+    rdaddsn(&o->stream, text, len);
+}
+
+void objstream(object *o, char const *text)
 {
     rdaddsc(&o->stream, text);
 }
@@ -518,8 +767,30 @@ static void objref(object *o, object *dest)
     rdaddsc(&o->main, buf);
 }
 
+static void objdest(object *o, page_data *p) {
+    objtext(o, "[");
+    objref(o, (object *)p->spare);
+    objtext(o, "/XYZ null null null]");
+}
+
+static char const * const stdfonts[] = {
+    "Times-Roman", "Times-Bold", "Times-Italic", "Times-BoldItalic",
+    "Helvetica", "Helvetica-Bold", "Helvetica-Oblique","Helvetica-BoldOblique",
+    "Courier", "Courier-Bold", "Courier-Oblique", "Courier-BoldOblique",
+    "Symbol", "ZapfDingbats"
+};
+
+static int is_std_font(char const *name) {
+    unsigned i;
+    for (i = 0; i < lenof(stdfonts); i++)
+	if (strcmp(name, stdfonts[i]) == 0)
+	    return TRUE;
+    return FALSE;
+}
+
 static void make_pages_node(object *node, object *parent, page_data *first,
-			    page_data *last, object *resources)
+			    page_data *last, object *resources,
+			    object *mediabox)
 {
     int count;
     page_data *page;
@@ -567,7 +838,8 @@ static void make_pages_node(object *node, object *parent, page_data *first,
 		objtext((object *)thisfirst->spare, "\n");
 	    } else {
 		object *newnode = new_object(node->list);
-		make_pages_node(newnode, node, thisfirst, thislast, NULL);
+		make_pages_node(newnode, node, thisfirst, thislast,
+				NULL, NULL);
 		objref(node, newnode);
 	    }
 	    objtext(node, "\n");
@@ -594,6 +866,11 @@ static void make_pages_node(object *node, object *parent, page_data *first,
 	objref(node, resources);
 	objtext(node, "\n");
     }
+    if (mediabox) {
+	objtext(node, "/MediaBox ");
+	objref(node, mediabox);
+	objtext(node, "\n");
+    }
 
     objtext(node, ">>\n");
 }
@@ -603,44 +880,32 @@ static void make_pages_node(object *node, object *parent, page_data *first,
  * 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
- * simply asks for ordinary text strings without mentioning what
- * character set they are supposed to be interpreted in.
- * 
- * Therefore, for the moment, I'm going to assume they're US-ASCII
- * only. If anyone knows better, they should let me know :-/
+ * encodes in either PDFDocEncoding (a custom superset of
+ * ISO-8859-1) or UTF-16BE.
  */
-static int pdf_convert(wchar_t *s, char **result) {
-    int doing = (result != 0);
-    int ok = TRUE;
-    char *p = NULL;
-    int plen = 0, psize = 0;
-
-    for (; *s; s++) {
-	wchar_t c = *s;
-	char outc;
-
-	if (c >= 32 && c <= 126) {
-	    /* Char is OK. */
-	    outc = (char)c;
-	} else {
-	    /* Char is not OK. */
-	    ok = FALSE;
-	    outc = 0xBF;	       /* approximate the good old DEC `uh?' */
-	}
-	if (doing) {
-	    if (plen >= psize) {
-		psize = plen + 256;
-		p = resize(p, psize);
-	    }
-	    p[plen++] = outc;
-	}
+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 (doing) {
-	p = resize(p, plen+1);
-	p[plen] = '\0';
-	*result = p;
+
+    if (!ret) {
+	ret = utoa_dup_len(s, CS_UTF16BE, len);
+    } else {
+	*len = strlen(ret);
     }
-    return ok;
+
+    return ret;
 }
 
 static int make_outline(object *parent, outline_element *items, int n,
@@ -656,6 +921,7 @@ static int make_outline(object *parent, outline_element *items, int n,
 
     while (n > 0) {
 	char *title;
+	int titlelen;
 
 	/*
 	 * Here we expect to be sitting on an item at the given
@@ -664,19 +930,20 @@ static int make_outline(object *parent, outline_element *items, int n,
 	 */
 	assert(items->level == level);
 
-	pdf_convert(items->pdata->outline_title, &title);
+	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(objtext, curr, title);
+	pdf_string_len(objtext, curr, title, titlelen);
+	sfree(title);
 	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");
+	objtext(curr, "\n/Dest");
+	objdest(curr, items->pdata->first->page);
+	objtext(curr, "\n");
 	if (prev) {
 	    objtext(curr, "/Prev ");
 	    objref(curr, prev);
@@ -742,7 +1009,7 @@ static int pdf_versionid(FILE *fp, word *words)
 
 	switch (type) {
 	  case word_Normal:
-	    text = utoa_dup(words->text);
+	    text = utoa_dup(words->text, CS_ASCII);
 	    break;
 	  case word_WhiteSpace:
 	    text = dupstr(" ");
@@ -762,19 +1029,30 @@ static int pdf_versionid(FILE *fp, word *words)
     return ret;
 }
 
-static void pdf_string(void (*add)(object *, char const *),
-		       object *o, char const *str)
+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; *p; p++) {
-	char c[2];
-	if (*p == '\\' || *p == '(' || *p == ')')
-	    add(o, "\\");
-	c[0] = *p;
-	c[1] = '\0';
+    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));
+}