Equivalent to HarfBuzz clusters?

[khaledhosny]

I don’t see an equivalent to HarfBuzz cluster or any way to associate output glyphs with input code points, or did I miss it?

[JimmyLefevre]

Is a Harfbuzz cluster a grapheme cluster, or is it something different?

I've been thinking of adding either a codepoint (or grapheme?) index or a user data pointer to the glyph struct, although I personally don't need think kind of thing, so I would like to hear what would be useful for you.

[khaledhosny]

HarfBuzz clusters are explained here https://harfbuzz.github.io/clusters.html. Basically they allow mapping output glyphs back to input codepoint indices (including indicating what glyphs got ligated together and what glyphs got decomposed). This is needed for hit testing etc (or even non-interactive uses like PDF export will need such info).

I don’t have an immediate need for it (I don’t use kb, I’m only interested in testing it and finding shaping issues since I make complex fonts that often produce different results in different implementations). To make my life simpler, I’m adding a kbts shaper to HarfBuzz so I can use hb-shape/hb-view instead of my crude cli tool. HarfBuzz needs the cluster information, and I’m currently setting them to 0.

[JimmyLefevre]

Thanks for the information. I am getting around to this now.

Aside from some inevitable reordering shenanigans in the normalization and/or cluster processing passes, it seems like providing level 2 clusters and letting people use the included grapheme breaker if they so choose would be a good compromise.