Improve handling of fea-rs output #306
Conversation
rsheeter
force-pushed
the
os2_3
branch
3 times, most recently
from
f62b28e
to
82bfc69
Compare
rsheeter
changed the title
| let font = FontRef::new(&buf).unwrap(); | ||
| debug!( | ||
| "Built features, gpos? {} gsub? {}", | ||
| font.gpos().is_ok(), | ||
| font.gsub().is_ok() | ||
| ); | ||
| if let Ok(gpos) = font.gpos() { | ||
| context.set_gpos(gpos.to_owned_table()); | ||
| } | ||
| if let Ok(gsub) = font.gsub() { | ||
| context.set_gsub(gsub.to_owned_table()); | ||
| } |
There was a problem hiding this comment.
this is a very convoluted way of trying to do what we want, I think? Wouldn't it be nicer if we just had some way of retrieving the bytes of a specific table from the FontBuilder? (we'd need to add a method there, but nothing crazy)
There was a problem hiding this comment.
This way it's typed which is nice perhaps? - I agree it feels a bit clunky but ... it does express the intent in fairly straight forward terms
There was a problem hiding this comment.
Maybe this is a sign that fea-rs should be returning something other than a FontBuilder?
It would also be nice if fea-rs did not return a type from write-fonts directly, since this is a reason why every breaking change in write-fonts is currently a breaking change in fea-rs...
There was a problem hiding this comment.
Maybe this is a sign that fea-rs should be returning something other than a FontBuilder?
It would also be nice if fea-rs did not return a type from write-fonts directly, since this is a reason why every breaking change in write-fonts is currently a breaking change in fea-rs...
although maybe there's no good work around from this? it certainly doesn't change if I'm returning a set of write-fonts tables instead of a write-fonts TableBuilder. But it still might be a better API.
There was a problem hiding this comment.
Maybe this is a sign that fea-rs should be returning something other than a FontBuilder?
It would also be nice if fea-rs did not return a type from write-fonts directly, since this is a reason why every breaking change in write-fonts is currently a breaking change in fea-rs...
Returning any type from write-fonts has the same effect. Raw bytes, or something close to raw bytes (map of tag to raw bytes?) would work. I don't want to have multiple write-fonts in my dependency tree but perhaps it would bumping through versions simpler. Or we could just live with it; I would expect the rate of write-fonts breaks to decline over time as it stabilizes.
Split the single feature font from fea-rs into pieces and include those pieces in the final font instead of running fea-rs and then just abandoning the results if it succeeded.
This reveals that for Oswald GSUB doesn't match and we aren't ending up with a GPOS at all; filed #308.