95 lines
3.1 KiB
OCaml
95 lines
3.1 KiB
OCaml
(* Embed a font *)
|
|
open Pdfutil
|
|
|
|
type t = Pdftext.font list * (int, int * int) Hashtbl.t (* Table returns font number and charcode for given unicode codepoint *)
|
|
|
|
type cpdffont =
|
|
PreMadeFontPack of t
|
|
| EmbedInfo of {fontfile : Pdfio.bytes; fontname : string; encoding : Pdftext.encoding}
|
|
| ExistingNamedFont
|
|
|
|
let fontpack_of_standardfont sf =
|
|
let te = Pdftext.text_extractor_of_font_real sf in
|
|
let table = null_hash () in
|
|
for x = 0 to 255 do
|
|
let u = hd (Pdftext.codepoints_of_text te (string_of_char (char_of_int x))) in
|
|
Hashtbl.add table u (0, x)
|
|
done;
|
|
([sf], table)
|
|
|
|
let get_char (fonts, table) u =
|
|
match Hashtbl.find table u with
|
|
| (n, charcode) -> Some (charcode, n, List.nth fonts n)
|
|
| exception Not_found -> None
|
|
|
|
let fontnum = ref 0
|
|
|
|
let basename () =
|
|
incr fontnum;
|
|
"AAAAA" ^ string_of_char (char_of_int (!fontnum + 65))
|
|
|
|
let make_single_font ~fontname ~encoding pdf f =
|
|
let name_1 = basename () in
|
|
let module TT = Cpdftruetype in
|
|
let fontfile =
|
|
let len = Pdfio.bytes_size f.TT.subset_fontfile in
|
|
Pdf.Stream
|
|
{contents =
|
|
(Pdf.Dictionary
|
|
[("/Length", Pdf.Integer len); ("/Length1", Pdf.Integer len)],
|
|
Pdf.Got f.TT.subset_fontfile)}
|
|
in
|
|
let fontfile_num = Pdf.addobj pdf fontfile in
|
|
let open Pdftext in
|
|
let fontmetrics =
|
|
let a = Array.make 256 0. in
|
|
for x = f.TT.firstchar to f.TT.lastchar do
|
|
a.(x) <- float_of_int (f.TT.widths.(x - f.TT.firstchar))
|
|
done;
|
|
a
|
|
in
|
|
(f.TT.subset,
|
|
SimpleFont
|
|
{fonttype = Truetype;
|
|
basefont = Printf.sprintf "/%s+%s" name_1 fontname;
|
|
fontmetrics = Some fontmetrics;
|
|
firstchar = f.TT.firstchar;
|
|
lastchar = f.TT.lastchar;
|
|
widths = f.TT.widths;
|
|
fontdescriptor = Some
|
|
{ascent = float_of_int f.TT.ascent;
|
|
descent = float_of_int f.TT.descent;
|
|
avgwidth = float_of_int f.TT.avgwidth;
|
|
maxwidth = float_of_int f.TT.maxwidth;
|
|
flags = f.TT.flags;
|
|
italicangle = float_of_int f.TT.italicangle;
|
|
capheight = float_of_int f.TT.capheight;
|
|
xheight = float_of_int f.TT.xheight;
|
|
stemv = float_of_int f.TT.stemv;
|
|
fontbbox = (float_of_int f.TT.minx, float_of_int f.TT.miny,
|
|
float_of_int f.TT.maxx, float_of_int f.TT.maxy);
|
|
fontfile = Some (FontFile2 fontfile_num);
|
|
charset = None;
|
|
tounicode = f.TT.tounicode};
|
|
encoding})
|
|
|
|
let make_fontpack_hashtable fs =
|
|
let indexes = indx0 fs in
|
|
let table = null_hash () in
|
|
iter2
|
|
(fun i (subset, f) ->
|
|
let charcode_extractor = Pdftext.charcode_extractor_of_font_real f in
|
|
iter
|
|
(fun u ->
|
|
match charcode_extractor u with
|
|
| Some x -> Hashtbl.add table u (i, x)
|
|
| None -> Printf.printf "charcode_extractor could not find char in make_fontpack_hashtable\n")
|
|
subset)
|
|
indexes fs;
|
|
table
|
|
|
|
let embed_truetype pdf ~fontfile ~fontname ~codepoints ~encoding =
|
|
let found_codepoints, fs = Cpdftruetype.parse ~subset:codepoints fontfile encoding in
|
|
let subsets_and_their_fonts = map (make_single_font ~fontname ~encoding pdf) fs in
|
|
(map snd subsets_and_their_fonts, make_fontpack_hashtable subsets_and_their_fonts)
|