Fontforge can't handle Adobe feature files correctly
Probably not our fault, but keeping this ticket open to track any progress on getting FontForge to work. If compiled with double or long-double arithmetic, it dies trying to apply the kern feature file to the font. If compiled with float (default, but not what I'd been using) it applies the feature, saves the font, and then dies when loading the resulting font. Looks like it has a lot of trouble with the very large kerning table - which is a problem now for the Korean fonts with ~11000 glyphs, but will probably also become a problem for the Japanese fonts which will pass ~5000 glyphs within the planned scope of this project.
In the short term, select.hl has been hacked to simply disable generation of kerned hangul fonts.
Mere possibility of kerning seems to be fixed by inserting our own subtable breaks in the kerning feature file. FoFo is supposed to do that automatically, but evidently not.
This does not solve everything related to kerning, feature files, and OpenType tables. In particular, some combinations of mark-to-base and context-substitution result in FoFo putting garbage rules in the substitution table. And at the best of times FoFo doesn't seem to correctly interpret context substitution rules in its own Metrics window.
For the moment things are working pretty well, with the inserted subtable breaks and taking great care not to do kerning on glyphs that are "marks" for the purposes of mark-to-base. It's not clear to what extent this is FoFo's fault and to what extent it's Adobe's; there are poorly documented (and Williams claims unimplemented) parts of the feature file spec relating to these subjects, but FontForge is definitely observed to do things that can't possibly be right.
Leaving the bug open for the moment because I expect further breakage, even though the particular issues that inspired its creation seem to be worked around at present.