Author | Lightnews

david barsky

david barsky

@davidbarsky.com

950 followers 420 following 480 posts

i like cooking and reading books. my day job is to work on ersc.io, but before, it was rust-analyzer. he/they is fine.

Posts Media Videos Starter Packs

david barsky @davidbarsky.com · 4h

i hit this issue in the last two weeks! i’m glad the oxide folks wrote this up

Adam Leventhal @ahl.bsky.social · 7h

My colleague Dave Pacheco wrote up a great description of a new (to us?) Rust async pathology--"futurelock"--another extremely sharp edge to watch out for, with no particular guardrails. Of course, we'll be talking about it with @bcantrill.bsky.social on Monday's Oxide and Friends

609 - Futurelock / RFD / Oxide

rfd.shared.oxide.computer

david barsky @davidbarsky.com · 4h

wait, how so? genuinely curious!

david barsky @davidbarsky.com · 4h

so this is how anthropic will get me to spend $200/month, huh

a screenshot of my claude usage, saying that I’ve maxed out my quota until 9 PM, but used 12% of my weekly quota.

david barsky @davidbarsky.com · 1d

i think one of more surreal moments in the last few weeks was seeing @steveklabnik.com's announcement be featured by @quinnypig.com in Last Week in AWS. like, i started that jj startup! we're not even on AWS at the moment!

From Last Week In AWS:

Steve Klabnik leaving Oxide for a jj startup feels like 2012 all over again: betting early on infrastructure tooling that makes engineers' lives measurably better. He's notable in the Rust world, so of course the post spends the first third talking about Rust in accordance with that community's tenets.

david barsky @davidbarsky.com · 1d

we also won't require forks; you'll just submit a set of commits for contribution atop of `main`/`trunk`.

david barsky @davidbarsky.com · 1d

sorry for the delay. for full disclosure: we're building a proprietary SaaS product, raised venture capital, and find "open core" software to be distasteful. we have not considered what federation entails, but strongly believe that this is *your* data.

david barsky @davidbarsky.com · 3d

oh, i was making a dumb joke about how half the polycules i know have an SSO setup.

as for federation: whatcha mean? activitypub-style federation, something more like tangled, or…?

david barsky @davidbarsky.com · 3d

the latter is going to need to pay the sso tax on ersc.io, unfortunately

Reposted by david barsky

Lukas Wirth @lukaswirth.dev · 4d

rust-analyzer now fully uses the new trait solver! 🎉🎉 rust-analyzer.github.io/thisweek/202...

Commit: 049767eRelease: 2025-10-27 (v0.3.2658)

rust-analyzer.github.io

david barsky @davidbarsky.com · 5d

oh so DLMs speak heptapod b

david barsky @davidbarsky.com · 7d

i think this analogy is more revealing than intended, given the infamous personalities of many surgeons

david barsky @davidbarsky.com · 8d

they’re very proud of the fact that they don’t have any SREs there

(source: i worked there)

david barsky @davidbarsky.com · 9d

(i’m pretty sure we bank with your employer, fwiw)

david barsky @davidbarsky.com · 9d

1. go-to-def being slow is interesting/weird, ty. i should profile it sooner than later. do you have cache priming enabled?
2. give it a try. doesn’t require nightly, just a clone and a cargo xtask. it’s been a big improvement for things like autocomplete, IME

david barsky @davidbarsky.com · 9d

(for one, if you have many, try disabling rust-analyzer’s native diagnostics. we lock on a per-open file basics, which causes some severe degradation in performance)

david barsky @davidbarsky.com · 9d

ah. the single core performance there is pretty darn good, and ra’s performance is bottlenecked on single-core. two follow-up questions:
1. how many tabs of buck2 code do you have open at any given time
2. have you tired using HEAD (building from source) of ra with the new trait solver

david barsky @davidbarsky.com · 9d

what’s your hardware, outta curiosity?

david barsky @davidbarsky.com · 9d

you should put in _just_ your email; none of the `ersc subscribe --newsletter`. sorry about that UI not being clear!

Reposted by david barsky

Steve Klabnik @steveklabnik.com · 9d

TL;DR: I'm going to be leaving @oxide.computer next month, which I'm very sad about. But it's to join @ersc.io , which I'm very excited about!

Reposted by david barsky

Steve Klabnik @steveklabnik.com · 9d

I see a future in #jj-vcs: steveklabnik.com/writing/i-se...

I see a future in jj

Blog post: I see a future in jj by Steve Klabnik

steveklabnik.com

david barsky @davidbarsky.com · 10d

wait, what’s happening

david barsky @davidbarsky.com · 10d

i think for ra to take advantage of it, it needs to be statically enforced

david barsky @davidbarsky.com · 10d

this is a limited sample, but the serial nature of proc macro evaluation has a big impact on rust-analyzer’s performance:

bsky.app/profile/davi...

david barsky @davidbarsky.com · 12d

like, this isn’t a fair, head-to-head comparison because macros introduce new items to resolve, but disabling proc macro expansion in rust-analyzer on buck2 (a 500k LoC codebase) caused ra’s startup time to go from 30 seconds to *11* seconds. there was only a 20% reduction in items!

david barsky @davidbarsky.com · 10d

as for data, nothing in public. i know of some stuff from my time at facebook, but i can’t disclose that.

david barsky @davidbarsky.com · 10d

yes, but:
1. from-scratch build times are pretty common IME. get several per day because i switch between commits!
2. they’ll only get more common as workflows make use of worktree-like features
3. CI build times are *very* legible
4. proc macros have to rerun on every keystroke in an IDE!