Three Dollar Quill

Okay, I've got a bit of time tonight, let's see what I can make of it. The big issue right now is that I have a lot of code that's making assumptions about the in-memory representation of the lexicon, and that's going to need to change. Currently, the values in the mapping are tuples of frozensets of pronuciations, and "metadata" objects. Now, thinking about this, I may actually be able to avoid thinking about this, because the children certainly seem to be part of the metadata. If there's metadata, there can be children, and covice coversa.

Currently, the children mapping goes from gloss extensions to pronunciation sets, but just to make sure I've future-proofed this, let's look ahead. In the event that a derivation creates "sort of a new word", then the choice of type consistent with current usage would be... a tuple of pronunciation sets and metadata.

...

Okay, I've updated metadata serialization and deserialization, and now that part of it should just work. Now I need to get the paradigm code working, because I've just thrown a big ol' wrench in all of that. ... To be fair, I just realized this code was sort of buggy anyway. Really buggy, actually.

...

All right, I've updated the code. I believe it works now, but I believed that before, and I was clearly wrong, so I'm going to wrap things up for tonight, and make sure I'm well-rested when I have to debug this nonsense.

Good night.