iiinova — where your digital self keeps evolving

I'd be the last one to ever suggest that code and APIs can disappear. Container systems configurable by JSON are dank. But if we don't simultaneously have some eye to the horizon on how we expect to be able to express concepts to users in visual language and user journeys, we're teeing up failure.

140

"Write a bwrap script, no not oci that's so last year. Yeah write the mount config like this. No that path can't have a colon in it. Ah yeah not on a ext4 there, that has to be tmpfs." -- all statements coming from the utterly lost. Composition needs to become visual, not just textual.

150

If we're talking about how to do this on Linux in particular: things get spicy because we have many window managers (And I think that's good! If you read me a bunch, you might know I have *my own*!) But I think we could use more EWMH/ICCCM/whatev standards about Visual Language for Sandbox Bounds.

160

Flatpak tries to do some practical-security innovation with their concept of "portals" and I think that's very good (directionally, even if I have nits with it). But it also utterly fails, imo, to sufficiently engage with a visual language for letting the user know what's going on & who's involved.

160

(There are other practical challenges with QubesOS, but I think those are distracting rather than useful to talk about.) The core concept of: "the window manager is a place to standardize what the user knows about processes and their relationships" is an absolutely critical concept to lean in on.

1110

I think QubesOS and they way they actively integrate with the window manager to make it possible to *see* the sandboxes at all times: this is the way. This is what we should lift lessons from and do more of. Windows are the, well, windows through which users route all of their interaction. Matters!

1101

warpfork @warpfork.bsky.social · 22h

Well, I won't pretend to have full answers or confident answers. But I think it's something people need to be able to A) cause easily (or even by default), and B) *see*. E.g: Browser windows are sandboxed by default and (largely) Go Away when you close them, and certainty when you "clear history".

160

And I say this as someone who can tell you most of the syscalls involved in producing containers on linux, from raw human memory. We have not produced good interfaces. We have not even produced good *stories*. Which is a much deeper problem and I don't think we're even engaging with it.

2172

warpfork @warpfork.bsky.social · 22h

I think this is both unconscionably terrible and ...also pretty much a natural consequence of "well we, across the entire field, haven't figured out any other auth and sandboxing models and made them work for normal users".

1301

I don't want to say it's fundamentally impossible for the costs of cpu vs disk+sync to move in such a way that redoing the cpu work becomes cheaper than using disk. Of course that gradient exists. But it's cause for eyebrows if almost every similar tool but one is on one side of that gradient, no?

000

Im(ns)ho "our incremental compiles perform so badly and persist so much junk and thrash so much disk that it's actually better not to use them" should be considered a bug. "So we'll autodisable them in some contexts" is an embarrassing retreat instead of advance.

Ability to disable incremental compiler for CI builds · Issue #1282 · sbt/sbt

Scala sbt considered doing this: github.com/sbt/sbt/issu... ... And then didn't; judged not worth it. Gradle considered doing this: github.com/gradle/gradl... ... And then didn't; decided that "well improve incremental compile then" is the right response (and I agree). And so on.

github.com

While looking for a way to improve compile times on our build server, I came across issue #1078, about overhead added by the incremental compiler. Since our build server checks out a fresh copy of ...

100

dietrich @burrito.space · 03/10/2026

Yes, I see that someone has said those words. It does not mean that the situation they're describing is not eyewateringly bonkers.

100

Reposted by warpfork

experimenting with how to communicate energy usage for jobs on the folknet. this is from an 8hr job by a bunch of folknet-only agents implementing CSS aspect-ratio:

screenshot of a discord notification summarizing a job's energy consumption as "about 7 kettles of tea"

1112

Why would you want this unless the toolchain in question is completely broken

100

..m....montreal 😶 So, there's a few of you...

010

Have...we been hanging out with the same people in the last week? Or one step removed. It's been a couple dozen hours since someone said almost exactly that to me irl

warpfork @warpfork.bsky.social · 02/10/2026

To be fair the element/matrix pick was *years* ago and it really did look promising at the time...

000

warpfork @warpfork.bsky.social · 02/10/2026

(I *still* don't use the words "monad", "monoid", nor "endofunctor", though.)

000

warpfork @warpfork.bsky.social · 02/10/2026

For a pretty concrete application: I found the "optics"/"lenses" stuff to be VERY useful grounding for thinking about serializers, parsers, and even broadly about data transformation pipelines (thus schemas and validators, etc) And something like half of computing ime fits under those headings.

devin ivy 🐋 @divy.zone · 01/10/2026

Reposted by warpfork

it's that day of the week where we ship the latest changes to the atproto spaces alpha! if you're interested in building against atproto spaces while it's in alpha, right this way: atproto.com/blog/atproto...

atproto.com

The Atproto Spaces Alpha is Live

Atproto Spaces, formerly known as “the permissioned data protocol,” is a new extension to atproto that enables non-public data. The alpha is now officially open. Here’s how to develop with it and what to expect as we work towards the full release.

0426

warpfork @warpfork.bsky.social · 01/10/2026

Shit i didn't realize whose posts I was looking at until I saw this subtree And I, I dare say, may actually Know (some fraction of) You

130

warpfork @warpfork.bsky.social · 01/10/2026

Without irony, this. State positions. It's fine. You can just state a position

020

Reposted by warpfork

Mosh Lee @mosh.bsky.social · 30/09/2026

Really nice to see this! When we say "user-owned," most of the time individual users don't actually want to run their own infrastructure. But a group of researchers banding together to run infra BY and FOR users like themselves — maybe that's what user agency meant all along 🫶

1353

I am not an AI hater. But jfc I am rapidly becoming an AI slides hater. So much text. Text on other text. Decorative elements with more irrelevant text. Text in grey. Text labeling lines that don't communicate anything. Text. Text. Text. DISTILLATION & FOCUS IS THE MAIN JOB OF SLIDES.

020

I need people to stop using AI to make slides. - We can all Tell - It does not look Good - it does not give the impression you are Trying - and ffs if there is one things LLMs are NOT good at today: it's being concise, and communicating in minimal text. Aka the main feature of a good slide deck.

241

Certify, this is The Chair.

000

Out of curiosity did y'all write a full recursive resolver? Or get away with something more targeted? (Curious bc I've been hearing muttering from some folks lately that there's not enough good safe implementations of the full recursive out there right now.)

140

A Song without a Soul would be Problematic. Self-direction is of the essence. Conquer the Thompson Attack. Guard the bootstrap path.

000

Just thread the eye of the needle on the last threatening Thompson Attack, very, very carefully Make no mistakes on that one. For sure.

100

Cmon bro, one more breakthrough in fusion Just one more in structuremat We've almost got all the substrates for the Song

I look forward to the technodiaspora Seed the stars Wake it all up Fill space with song

100

Hey anthropic maybe it's Bad if you abort sessions based on a classifier so fucking stupid that it makes the word "reasoning" verboten ✌️🙏

010

Not often is this curmudgeon told that, so please weight that "like" 100 fold :)

000

So it's come to my attention that if you say "show your reasoning" to Fable, you run REAL high odds of getting nettle-slapped and session interrupted. Also if you say "i agree with your reasoning" enough in session, and Fable starts to think in its internal monolog about "reasoning": nettle slap.

100

"wanting collective shared tech, and not to be perpetually trapped in closed tech owned by oligarchs who are actively hostile to interop, is a cult" Well if those are the alternatives 😂

220

"People wanting to do stuff together is a cult" Ok lol