Look-ahead types about fathom

[mikeday]

That's clever! It could be used to handle the hmtx table dependencies as well, by doing a limited lookahead parse that skips over everything to grab the num_glyphs field and then a full reparse that uses that value.

One issue is that the spec author might find it a little tedious to duplicate the definitions, and there is a risk of introducing error if the lookahead and full definitions are inconsistent.

In some ways it feels like this is the output of a compiler that has analysed the data dependencies in a spec that appears circular due to forward references and flattened it to a sequence of multiple linear passes. Maybe this is something that can be done semi-automatically? :)

[markbrown]

The code that's been arranged into multiple passes would want to be filling in the same host data structure over those passes, whereas the look-ahead type would produce two separate data structures, with the first presumably being much smaller, and the second being the complete version. So I don't think this is quite suitable as a target for that analysis :(

I've just realised that the link type from #28 can also achieve the same result for the GSUB case, so I think I'll take it up there and close this issue for now.