Hi!

The OpenJDK uses hb_font_set_funcs to customize both the nominal glyph function and the variation glyph function for fonts. It also uses hb_buffer_set_invisible_glyph to set the expected invisible glyph ID (0xFFFF). Finally, the JDK also returns this invisible glyph ID to HarfBuzz from the two custom font glyph functions when the queried code point is default-ignorable (e.g. ZWJ or ZWNJ).

Is this best practice? Or are the two custom font glyph functions expected to always report the exact glyph IDs in the font file, without worrying about invisible glyphs and default-ignorable categorization? The reason I ask is that we have run into an issue [1] where the custom nominal glyph function reports back to HarfBuzz that ZWJ maps to the invisible glyph, which leads to an emoji sequence [2] not running through the expected GSUB substitution.

I think I know the answer, but I want to make sure I'm not missing anything :-)

Thanks!

Daniel

[1] https://bugs.openjdk.org/browse/JDK-8353230
[2] \ud83d\udc71\ud83c\udffd\u200d\u2640\ufe0f, which is (Person with Blonde Hair) + (Emoji Modifier Fitzpatrick Type-4) + (Zero-Width Joiner) + (Female Sign) + (Variation Selector-16)