Neat writeup, and very interesting that they haven't found it to be enough of a performance win to turn it on in production yet.

I was actually most interested to hear about HHIR and their general compiler pipeline. I'm reminded of Rust's recent initiative to rearchitect the compiler with a Rust-specific IR of its own (MIR). Here's an overview of how the compiler devs expect rustc to operate once this work is complete: https://github.com/rust-lang/rfcs/pull/1211#issuecomment-128... and here's a comment expanding on this overview: https://github.com/rust-lang/rfcs/pull/1211#issuecomment-130...