Rod

I'm inclined to think this suggests deletion of the traversal code then? At least until such time as we have a solid solution?

#236 implements by hand but I'm not certain that's what we want in general.

A prospective "real user" is likely to need to compile our code so I would think that rules out build dependencies? Why not continue the current fontations model - one...

> May I suggest comparative fuzzing, using FreeType as the oracle? How about we file a separate issue for that and keep this one to basic unit testing?

I agree it's basically the same as option 2, just removing the requirement to be a crate. For example, a wee python script or C++ tool become permissable.

How do we get from there might be to there definitely are and we do/don't have to care?

My fear is each has it's own nuances, but perhaps that's not true. As a quick check, does FreeType ever branch on tag other than to distinguish is/isn't a sfnt...

Hm. Also what about OTS?

SG, lets follow HB here then.

It's possible to hand-write code for which the tools work. I'm not clear what makes that intractable for machine-written code?