Niko Matsakis

In my opinion the hero of this story is @frankmcsherry

Super impressive to see how much differential dataflow speeds stuff up in practice--it's too easy to forget how impressive it is when you're comparing it to microoptimized research database performance and not production software.

Now I'm wondering how hard it would be to write a procedural macro to automatically generate dataflow from Datalog style rules for simple stuff like this...

This could be pretty neat. My attempts at vanilla macros .. flailed. Datalog is pretty terse, and if you miss on any of its simplicity the resulting string can be hard to understand. E.g. I had to write repeated vars as `y1` and `y2` rather than the same name, which ... =/

I don't think vanilla macros could really do what you want here anyway (since you sort of need to be able to compare variable names). It would be interesting, though, if vanilla macros could understand binding sites, uses, and scopes, allowing shadowing like Rust let statements.

It might be easier to use macros if you were willing to use "SQL-flavored Datalog" (ignoring NATURAL JOIN and USING), since then you need to make the comparison explicit in syntax. That's a lot less terse, of course.