Sergey Sidorov

@_Felipe @yminsky It may come as a surprise but engineers at Facebook/Meta also spend a good amount of time cleaning up feature flags. The good thing is that you have metrics for everything including flag usage so you can just put your agent at it and focus on important stuff

@yminsky This paper from Facebook describes many ideas that I haven’t seen being adopted. research.facebook.com/publications/h…

This is true, but the feature flag problem is solved badly across the industry. If done well, leaving feature flags in the code indefinitely wouldn’t be a big issue.

@thorstenball My favourite!

New keyboard is good! It's a Lofree Flow 2 84.

@valyala @sean_j_roberts WITHOUT ENUMS OR UNIONS

@valyala @sean_j_roberts For the past decade I've been paid to code in C, C++, Go, and Java. None of these are perfect (maybe I just suck at programming lol), but Go is dumb enough that I'm finally allowed to stop being clever and just write code that works and works pretty fast.

hot take: Go would be the greatest programming language on earth if it had enums and union types. they simplified their way into having to do really weird stuff just to represent reality. you should still probably use it though.

P.S. This is part of Elastic' broader push, involving some fairly bold changes to make logs and metrics faster, cheaper, and easier to put into action. There's a lot more under the hood: lnkd.in/ep8CCWn2

🔥 First-class PromQL support in Elasticsearch Many teams already rely on PromQL in their day-to-day work. We're making PromQL a first-class experience in Elasticsearch. Blog: elastic.co/observability-…

I don't get the backlash against Anthropic enforcing DMCA on Claude Code. It’s their IP. "They trained on OSS" isn’t an argument imo, everyone does! That doesnt cancel copyright. We all benefit from what they ship anyways

@niklas_wortmann @jetbrains I almost forgot what a good IDE is like. I used VSCode for years at Facebook. Now in the Java world again, IntelliJ made a huge step up. It's fast, and it feels nice. That feature is a good example of tiny yet really helpful stuff u use day to day

We’re experimenting with a different UX using AI in @jetbrains IDEs. Most AI features wait for a prompt or a trigger action. Instead, we are exploring a different direction: AI that automatically surfaces helpful context. We’ve just released an experimental plugin with two such features. Details in the thread 🧵

@niklas_wortmann @martinbeentjes @jetbrains Please don't. This will just take focus to a different place. What u do with native IDEs is great and we want more of this. I find it a good example of small QoL improvements that actually make difference

@martinbeentjes @jetbrains Not as of now. Long term we could maybe evaluate using ACP but that could potentially also increase the cost

@ahmetb usgraphics.com/products/berke…

@ahmetb 100% - used this for a couple of years now, super tasty

now that you are spending all your time in the terminal, treat yourself to a nice font like Berkeley Mono.

@ryanlpeterman Spent 5 years building distributed systems at FB across Core Systems and Data Infra, and nearly as much before in HFT. It was hard to go deep while grinding the ladder. This summer I paused the grind to build OSS search infrastructure at Elastic.

@serasid Cool, what're you building?

@mitchellh Why does the terminal app need any of these? Honest, question, not that I'm trying to challenge it. I've been using iterm + tmux for ages. Do I miss anything folks? :)

Been working on a new developer overlay system in Ghostty and majorly improving the inspector. Here is an example of the "semantic prompt" overlay in the renderer, plus the right pane shows that this screen has seen semantic content at least once (showing proper integration). The overlay system works by using CPU-rendered 2D graphics and compositing them on top of the terminal screen via the GPU using the same image shaders we use for features like Kitty Graphics Protocol. Choosing to use a CPU renderer for the overlay is a pragmatic one: it lets us iterate more quickly on these debug tools, its immediately cross-platform (no per-platform shaders), and the debug overlay is not typically on so it doesn't need to be as optimized as everything else. Enabling these debug overlays does "blow up" our frame draw times from ~30us to ~600us but just remember there are 8300us in a 120fps frame. So, we're going to be okay.

@samswoora What exactly the refactor? Things you mentioned existed for ages (at least at facebook). Safeguards - sure, but I doubt site infra is going to change much

Rumor is FAANG style co’s are refactoring their monorepos to scale in preparation for infinite agent code

@ForrestTheWoods @mitchellh @JustJake 100%. It's such a joke git became the de facto tool for source control

@mitchellh @JustJake Git is so bad. So very very bad. It's always been bad. But sadly it's all an entirely generation of programmers knows :(

If you haven’t done this already It’s going to get very, VERY painful very VERY soon

@kerckhove_ts @karanjagtiani04 It's default port indeed: man7.org/linux/man-page…. Although I'm not sure if it is Linux specific.

@karanjagtiani04 Let's call that one magicPortValueThatMeansDynamicPortAssignmentButIsn'tAValidPortItself :: PortNumber instead.

Ah yes, the default port number! That seems like a good port number to use by default.

You asked for it, so here it is. Visualizing CPU cache speeds relative to RAM. Cache optimization is important too!

WOW! Check out this panel to learn about hidden tax affecting your uptime #SLO, how SLICK: SLO Reviews at @Meta works, and why @awscloud is sponsoring SLOconf.💜 @gregarnette @imaya @serasid @AWS_Partners

SPEAKER ANNOUNCEMENT: @serasid and David Bartok of @Meta will be speaking at #SLOconf! 🙌 This talk provides a structured workflow to have meaningful discussions about SLOs and identify follow-ups, enabling teams to get the most value out of #SLOs.🔥 sloconf.com