Thanks to everyone who made it to the discussion yesterday - notes are here (additional thanks for those who helped take notes!). Some key takeaways from my perspective:
- We don’t see the two approaches as being mutually exclusive - there’s definitely interest from the call in reviewing both patchsets and look to merging given the perf benefits noted so far (or expected through further work in the case of Elizaveta’s work)
- As usual, merging behind a flag to allow incremental work upstream is a good strategy. But the ultimate goal is of course to enable by default.
- There’s concern about the “ecosystem impact” in terms of tooling having issues handling code that had more aggressive shrink wrapping enabled, but this isn’t understood to be a blocker so far. Before turning on by default someone would need to do some targeted investigation into the impact (i.e. testing), and ideally it would be turned on nearer the beginning of a release cycle to give plenty of time for unexpected problems to be found.