Following a conversation on #8:

...the way Commandant is set up makes it kind of painful where every error needs to be the same type so I avoided doing anything drastic, but I think with enough effort one could make a generic wrapper error and each command could have its own error which coalesces into the generic wrapper automatically, satisfying logical soundness and type constraints.

Each command instance should have its own specialized error type which is transformed into the generic wrapper to satisfy the error type constraint on the actual main running instance.

Right now, the tool assumes all strings should go into a .strings file. This isn't necessarily true. Plural strings need added to a .stringsdict plist. It should be possible to allow some sort of flagging that a desired function is plural (not sure the most elegant way to do this). Furthermore, the tool will need to parse out existing .stringsdicts and do the same sort of behaviors that it does for plain .strings files.

For LocalizableNameString, I think that's what it's called, you can use a name that doesn't exist and the compiler allows it to compile. Only at run time do you get a fatalError crash. We should implement a phantom type to resolve this and add compile time support.