I'm well aware of how it's "done right", but as I say in the documentation, this is the fifth generation and the contrast of doing the normalized queries and the denormalized queries was quite strong - the denormalized table allows me to formulate the queries in a very simple fashion, which was not possible with the normalized data. As the file types get saved in a separate column anyway (mime_type), querying by filetype doesn't imply mucking around with the extension either ;-). But most of that isn't obvious from what I said above, nor from the (lacking) documentation ...