Yeah, that's a very similar approach to dova's from the comments of the original article (I mentioned his solution briefly in my last post), but performs even a bit better than that one.

Like dova's, it does perform better under the right circumstances, like with an actual table of dates, but when used with an iTVF to generate dates the performance gets very bad.

Still, in terms of raw performance, that approach with a table of dates is the top of the heap so far.

Cheers!