Folks who are running Tailscale for their own networks, what provider are you using? One of the big 4 they offer (arguably 3, given GitHub is owned by Microsoft, but GitHub auth is much less of a hellscape than Microsoft auth 😂) or a custom OIDC provider?
Sure do wish I could rerun an individual task in a GitHub Actions flow while the other tasks are still running. Looking at one that was just a transient network failure and everything else is ✅ and… I’m just sitting here waiting because the UI won’t let me do anything till everything else is done. 🙃
Love to try to upload a video to Google Slides. 🙃
“Video format unsupported.” Cool: why? Who knows. Google won’t tell you!
Said to @dfreeman.io a few days ago:
> I believe that the phrase “I need a language-native way to extend/embed with other languages” is the actual wording for the Summon Greater Lisp Weenie spell.
I’m back (though probably at low bandwidth for the foreseeable future: just too much going on in real life to spend much time on social media). Happy Easter!
Selection from the 2019 Book of Common Prayer which reads, “Almighty and everlasting God, you hate nothing you have made, and you forgive the sins of all who are penitent: Create and make in us new and contrite hearts, that we, worthily lamenting our sins and acknowledging our wretchedness, may obtain of you, the God of all mercy, perfect remission and forgiveness; through Jesus Christ our Lord, who lives and reigns with you and the Holy Spirit, one God, for ever and ever. Amen.”
As is my long habit, I am signing out of social media for Lent. I’ll (presumably!) be back after Easter Sunday. In the meantime, I’ll leave you with the Collect for Ash Wednesday—one of my favorite of the Collects of the whole Christian year.
Grace and peace to all of you.
I’ve said it before and I’ll say it again: Apple shouldn’t need to *trick* people into installing Tahoe, and yet its Settings > General > Software Update UI seems intent on doing just that.
Astonishing to me—though at this point it shouldn’t be—that there’s still no way to filter to unread comments on Google Docs. Just unbelievably user-hostile design.
Yep.
Yeah, we track the compilation times for each package in CI. You can do this whether you’re doing it “all at once” by running `--build` or via a build orchestrator, though the latter is definitely easier to get metrics for and is what we do.
Type checking time for the packages in our monorepo.
It minimally sped up type checking time by 1.5× in the packages least affected. In many cases, it was 2–4× faster. In a few, glorious cases—notably also the largest and slowest—it was more like 5–6× improvements. Huge win for CI, possibly even huger for local development experience.
That’s correct, yeah—the Go rewrite of the compiler, which we’re adopting because it is significantly faster for type checking and emitting type declarations. (We use other tools to compile TS to JS.) This graph shows how long it takes to run the type checker on each package.
Are you using `--incremental`? If so, you may be seeing it be able to just say “nothing to do here”. If not… not sure!
It’s still something we may evaluate/trial, but it’s basically down to whether we want to take on the risk and migration churn/cost of adopting an alpha. So far, the answer is no, but that could change eventually!
I say “(correctly)” because the alpha version oxlint shipped is interesting but at present we would rather wait to use a version that works using an official approach rather than the (very smart!) hacks they’re currently using.
Slow. 😂 We’re still using typescript-eslint, and waiting for tsgo to have the relevant hooks for ESLint/oxlint/etc. to plug in (correctly 😝), and that’s actually our longest pole CI-time-wise right now, though we’re also working on fixing things about our codebase itself to strongly mitigate that.
Noticed the top-level shortcuts for the (vertical) “tabs”. Thanks, and keep it up!
Less crappy version of the image here (Datadog happened to be refreshing its view when I snagged the screenshot and I didn’t notice)!
A line chart showing many lines, each representing the time take to type check a package, all decreasing over a one-month span.
Our tsgo adoption at Vanta has been pretty high-impact. Everything is so, so much faster now.
New issue of my newsletter: “Where Should Scholars Draw the Line on AI?” — Between the poles of Zero AI and AI For Everything lies a vast, poorly mapped middle ground newsletter.dancohen.org/archive/wher...
TypeScript excitement 😉
Congrats to Renegade334 on landing type declarations for @tc39.es Temporal - the new JS Date-Time API - in upstream TypeScript 🎉
Heading for TS 6.0 Beta next week 👍
Temporal is available in Firefox & Chrome today.
github.com/microsoft/Ty...
It’s rare either to have so many runtimes *or* for the language authoring and semantics to be so thoroughly decoupled from the target runtime. We’ve all internalized this after many years, but I needed to explain the details to some very sharp folks this week because it’s very non-obvious. (2/2)
I also think it’s hugely important in the context of people just getting started with the ecosystem—many of whom have deep expertise in *other* ecosystems but haven’t had to understand the particular and (by comparison) peculiar dynamics of *this* ecosystem. (1/2)
…the mistake most folks have made one way or another are down to exactly this, plus in many cases a deep (deep!) ignorance of human cognition and real neuroscience.
(I very often doubt any of them have ever just paid attention to a child growing from infanthood through age five.)
…understand the “stochastic parrots” claim, though, very little of the last few years has been particularly surprising! Certain jumps have been impressive—the recent Gemini 3/Opus 4.5 leap in quality and effectiveness, for example—but none of them have ultimate surprised me in *shape*. I think that…
You said sometime in the last few days something to the effect that people often get confused both ways by the “stochastic parrots” point: that it means they cannot be capable of startling things, or that they cannot actually be stochastic parrots, and both of those are wrong.
If you actually…
Yes, that’s exactly what I had in mind—sorry for the imprecision!
I want to emphasize that I’m really delighted by the app. It feels *good*.
Yeah, I would default to *never* yelling at someone that way; we had a unique working relationship where that made sense.
Guardrails and ways it’s afe to fail are definitely a big part of it. Another is using the positional and earned authority you have to help change the *EMs* defaults.
A very quick weekend read: