A quick follow up to this morning's mail:
I discussed this issue with Denny for a while, and we came up with this:
* I'll explore the possibility of using a BadValue object instead of a BadSnak, that is, model the error on the DataValue level. My initial impression was that this would be more work, but I'm no longer sure, and will just try and see.
* We will represent the error as a string inside the BadValue/BadSnak object. There seem to be no immediate benefits or obvious use cases for wrapping that in an Error object. (This in reply to an earlier discussion on Gerrit).
-- daniel