Legacy fonts can hava a NameRecord not encoded in UTF-16BE

[moi15moi]

Currently, libass always decode family name with utf-16be:

libass/libass/ass_fontselect.c

Lines 283 to 284 in a48c98c

	ass_utf16be_to_utf8(buf, sizeof(buf), (uint8_t *)name.string,
	name.string_len);

But, Microsoft NameRecord don't always use utf-16be.
To know how libass should decode properly namerecord, see: MicrosoftDocs/typography-issues#956 (comment)

Something like this could be added in ass_utils.c:

char* get_name_encoding(FT_SfntName name) {
    if (name.platform_id == TT_PLATFORM_MICROSOFT)
    {
        switch (name.encoding_id)
        ​{
            case TT_MS_ID_PRC:
                return "windows-936";
            break;

            case TT_MS_ID_BIG_5:
                return (name.name_id == TT_NAME_ID_FONT_SUBFAMILY) ? "UTF-16BE" : "windows-950";
            break;

            case TT_MS_ID_WANSUNG:
                return (name.name_id == TT_NAME_ID_FONT_SUBFAMILY) ? "UTF-16BE" : "windows-949";
            break;

            default:
                return "UTF-16BE";
        }
    }
}

Finally, to decode byte into utf-8, libass could use ICU: https://unicode-org.github.io/icu/userguide/conversion/converters.html#1-single-string

PS: To test if it decode properly namerecord with BIG_5, download the font in this issue: MicrosoftDocs/typography-issues#956 (comment)

[astiob]

ICU is waaaay too massive for libass to use, if I’m not mixing anything up.

But the issue is, of course, real; thanks for creating a dedicated ticket for it. I already have name decoding code in https://github.com/astiob/libass/tree/debug-fonts. It should probably be adapted into mainline libass.

[moi15moi]

ICU is waaaay too massive for libass to use, if I’m not mixing anything up.

Ok. I don't have a good knowledge of C.

I already have name decoding code in https://github.com/astiob/libass/tree/debug-fonts.

I don't think it is a good idea to use mac platform id.

Here is what the Apple documentation mentions: Names with platformID 1 were required by earlier versions of macOS. Its use on modern platforms is discouraged.

[astiob]

Here is what the Apple documentation mentions: Names with platformID 1 were required by earlier versions of macOS. Its use on modern platforms is discouraged.

That means about as much as the Microsoft docs not mentioning Windows 95 quirks that GDI nevertheless emulates to this day. It’s “discouraged” to use non-Unicode fonts at all, but that’s exactly what we’re trying to do here. (And IIRC macOS itself still preferred Macintosh-platform names when I last checked.)

That branch is (as the name suggests) meant for debugging font issues, so it dumps all the information it can find in a font. What we actually want (ideally) is what VSFilter’s GDI calls use:

For TrueType fonts, uses names with the same platform and encoding as the first valid Microsoft-platform cmap (if any) or MacRoman cmap (otherwise). Never uses Unicode-platform names.

so all Microsoft-platform encodings, as well as Macintosh-platform MacRoman (whatever version of it is implemented in Windows).

[astiob]

Of course, we don’t currently support MacRoman cmaps, either, and I don’t remember if I’ve ever seen a font that lacked Microsoft-platform data and actually worked in VSFilter. (Zapfino lacks them, and it doesn’t work in VSFilter.) But anyway, just Microsoft-platform names for now would be plenty good, to match our support of Microsoft-platform cmaps.

[moi15moi]

Here is 2 font that should not use utf-16be to be decoded.
fonts.zip

[moi15moi]

I spoke with a Microsoft employee and he told me that GDI performed this processing:

# This NameRecord is from 文鼎中特廣告體 - PlatEncID 4.ttf
name_record = NameRecord()
name_record.nameID = 1
name_record.string = b"\x00\xa4\x00\xe5\x00\xb9\x00\xa9\x00\xa4\x00\xa4\x00\xaf\x00S\x00\xbc\x00s\x00\xa7\x00i\x00\xc5\x00\xe9"
name_record.platformID = 3
name_record.platEncID = 4
name_record.langID = 0

encoding = get_name_record_encoding(name_record)

if name_record.platformID == 3 and encoding != "utf_16_be":
    name_to_decode = name_record.string.replace(b"\x00", b"")
else:
    name_to_decode = name_record.string

decoded_name = name_to_decode.decode(encoding)