Ian

@iangrunert.bsky.social
@[email protected]

[Ian @iangrunert]

Another good day. 78 files changed, 79 insertions(+), 8294 deletions(-)

[Oliver Medhurst @CanadaHonk]

https://t.co/mGIHWeoONf is now served by a Porffor-compiled TS native binary!

[Jacob Miller @pwnies]

I'm excited to share / launch an early preview of https://t.co/K1rCfFUvda.

Diffui takes a different approach to AI design, leveraging diffusion as the first step. Take it for a spin - I'd love your thoughts.

Demos in thread:

[Ian @iangrunert]

@mitchellh I wonder if the other models would get there if the GTK source was available via the filesystem, instead of requiring a web search.

[Ian @iangrunert]

@awesomekling @ladybirdbrowser Is this in reaction to bottlenecks on JS execution found on particular sites, or just a fun project / exploration?

[Ian @iangrunert]

@paulgb Yep, "Review this branch / commit", and then describe what you don't like about it / want fixed. It's pretty effective (particularly with plan mode); probably helps that it's getting a fresh context window to work with.

[Ian @iangrunert]

https://t.co/RvbSxiBM9l

Read more about writing a JS Engine using Claude Code here:

https://t.co/ElsvxBfvWr

[Ian @iangrunert]

Claude Code is great at "duct tape" build systems and refactor work. Here's an example, commit link below.

Input: Could we compile quickjs libregexp to wasm, call it via wazero, and use it as a replacement for regexp2 which can actually handle some of the unicode edge cases?

[Ian @iangrunert]

https://t.co/ElsvxBfvWr

[Ian @iangrunert]

I wrote a JS Engine in Go using Claude Code. This has been my side project for the past two months.

It's passing ~44000 Test262 tests, almost reaching parity with QuickJS on Test262.

The code is available on Github (MIT License), and I wrote a short blog about it.

[Ian @iangrunert]

@pwnies https://t.co/WKsaxQ72wk ? /cc @davidcrawshaw

[Ian @iangrunert]

@pwnies Sleep; or catch up on the N marvel movies you missed.

[nolen @itseieio]

made a hook that adds a bouncing dvd logo to claude code whenever it's thinking

[Ian @iangrunert]

@awesomekling @ladybirdbrowser Nice! Yeah the classic JS benchmarks don't spend that much time in parse / compile. It might come more into play with multi-megabyte JS bundle sizes on the web.

Those generated property-escapes tests in Test262 are really brutal.

[Ian @iangrunert]

@awesomekling @ladybirdbrowser Ah that makes lot more sense, it's a nice isolated subsystem to start off with. That means you don't have to deal with the GC integration yet, either.

Any change on Test262 runtime, or do you run that with cached bytecode?

[Ian @iangrunert]

@davidcrawshaw It's unclear to me how easy it is to escalate from those to an exploit given the use-after-free protections in the allocator, but still deeply annoying. There's also a bunch of type confusion footguns around downcasting.

https://t.co/wMPw6I6fD4

[Ian @iangrunert]

@davidcrawshaw "Too dangerous" depends on the threat model.

For web browsers, a successful exploit is extremely valuable; a likely target for nation-state level actors.

C++ with smart pointers has a bunch of sharp edges which are easy to make mistakes that can lead to memory safety issues.

[Ian @iangrunert]

@Zephraph Zig Day is also happening on Saturday:

https://t.co/JFwYbrPMJE

[Ian @iangrunert]

@davidcrawshaw Why not an iPhone 17 or a 17 Pro? They all have MIE.

[brandon @burcs]

wild rebrand...