Post

@jorropo.net burst my bubble; this could actually be observable by a very goofy case of a boxed nil interface call, something that the Go test suite did not test! But, it's still salvageable; just generate one wrapper instead of zero, which is just a simplification of my second CL. redone

go-review.googlesource.comgo-review.googlesource.comGerrit Code Review

Jake Bailey@jakebailey.dev

This is a fun one: I just sent go.dev/cl/790740, which elides method wrappers for embedded types at offset zero, common for mixins/inherited data types. Doing this removes 60 _thousand_ method wrappers from tsgo's AST package, dropping the compilation of our main exe from 49s to 28s on my machine.

15/6/26, 9:59 pm0 reposts14 likes

Jorropo

@jorropo.bsky.social

16 June

What is the new benchmark result now that you've removed the elision ?

Still just as good!

the proof

But it does change stack traces, but thankfully just for that edge case, which, again did not have a test 🫠 That or, maybe the spec should change or be reinterpreted in such a way that we can avoid the wrappers entirely...?