Luke Edwards

29 Apr 2025

Published empathic@1.1 🎉 (benchmarks below) Many packages are eagerly joining the @e18e.dev effort but (understandably) have existing support restrictions to maintain & so can't make as much progress as they'd like. This release helps with that~! (cont.)

29 Apr 2025

empathic@1.1 🔸Adds CommonJS support 🔹Allows Node 14 engines No code changes or bloating were needed on my end 💪 **The entire package is still less than 2kB and still fast.** The benchmarks compare `empathic/` submodules to their counterparts across the ecosystem. More: github.com/lukeed/empat...

Released `hrana` -- a 1kB libsql/hrana client 🔸runtime agnostic, works anywhere 🔹transaction and batch stmt support 🔸CI includes a cron check for Hrana spec updates 🔹dual JSR and npm publish, with provenance 🔸strictly typed 👉 github.com/lukeed/hrana /cc @tur.so

7 Jan 2025

error: TypeError: fetch failed at node:internal/deps/undici/undici:13178:13 at processTicksAndRejections (node:internal/process/task_queues:95:5) at runNextTicks (node:internal/process/task_queues:64:3) at process.processImmediate (node:internal/timers:454:9)

27 Dec 2024

Writing a new database client was easier than (still) trying to get its transitive `require` and `module` usage to transpile and/or polyfill correctly If anyone knows this fix I'd be happy to delete my client: ReferenceError: module is not defined

24 Dec 2024

clsx baked directly into svelte 🎄

Svelte@svelte.dev

Day 24 — Svelte now handles `class` attributes with github.com/lukeed/clsx. Much more powerful than the `class:` directive, but since this is Svelte you still get dead code elimination of unused styles! 🎁 docs svelte.dev/docs/svelte/... 🎁 demo svelte.dev/playground/6...

Luke Edwards reposted

Evan You

@evanyou.me

19 Dec 2024

Replying to @evanyou.me

As a result, Rolldown is now the fastest possible bundler you can run in the browser. Here are the numbers when bundling a benchmark app with 2.5k modules: - esbuild: 22.19s - Vite (via Rollup): 4.52s - Rolldown: ⚡️613.43 ms Try it out yourself on StackBlitz: stackblitz.com/~/github.com...

stackblitz.comStackBlitz

10 Dec 2024

are @rolldown.rs crates purposefully not published?

10 Dec 2024

/cc @boshen.github.io

Not strictly a 1:1 comparison... but still, I wish natives were optimized a bit more 😞 This spread is the same anywhere URLPattern is supported; eg esbench.com/bench/67537b...

These first images are actually BEST CASE for URLPattern since it was returning null (not a match) .. because user error The other images & benchmarks in this thread are corrected. URLPattern is about 100x (!!!) slower to yield “matching” results.

^ bsky.app/profile/luke... I’m still figuring out bsky, sorry for threading noise

Luke Edwards@lukeed.bsky.social

here's updated benchmark realistically, an app/router would *not* be doing `new URL` every time it's testing a match, but I added it here as regexparam's worst-case and to be apples-to-apples for full URL normalization. (same esbench link)

so... i was looking at some other JS routers/app servers 1) a LOT more exist than there was pre-Polka & pre-Worktop days 2) regexparam is everywhere 3) people generally inline regexparam (sometimes w/ link) ... I guess that's more appealing for 300b libs? 4) and, most surprisingly/worryingly...

(cont) You're paying for all this. Either: 1. Bigger-than-needed VMs 2. More-than-needed Lamdas/Workers/etc 3. The rest of your app isn't doing enough to encroach resource limits All cuz the README wants to show 2M req/sec instead of ~200k btw lol, no FaaS does 20k/s from a single instance~! 🙄😒

quasi-related tangent 🙃 bsky.app/profile/luke...

Luke Edwards@lukeed.bsky.social

Not strictly a 1:1 comparison... but still, I wish natives were optimized a bit more 😞 This spread is the same anywhere URLPattern is supported; eg esbench.com/bench/67537b...

(cont) which means your App has v significant memory leak. All for a mediocre README benchmark result. These routers may assume/require serverless environments, where "memory leaks dont matter" ...except that most FaaS dont kill/reset your App after every request.

(cont) They're generally kept alive & serve traffic until N idle timeout. And some snapshot memory for quicker resume, which means the cache is prefilled with the same MISS / volatile garbage. FaaS will spawn new instances according to demand; generally CPU or RAM*** load (sometimes raw req count)

(cont) Many routers have an internal match cache; which I guess is (cough, cheating) for Routing benchmarks? Cache Key is usually `method + req.pathname` which means Apps with dynamic routes (most do) using these routers have very infrequent cache HITs Very few of these actually *clear* the cache

send help, i wrote another router

Luke Edwards reposted

deno.comDeno 2.1: Wasm Imports and other enhancementsDeno 2.1 introduces first-class Wasm imports, built-in tracing with OpenTelemetry, faster `deno compile`, and improved package management.

1 Dec 2024

Has anyone had recent success with Paddle? They've been giving me this crap CSP error & docs say "localhost should be fine" but it isnt I'm using sandbox & checkout URL has been changed to any/all variations I can imagine Looks like bad HTTP header on their end.. deadend?

file system routers are a mistake

Luke Edwards reposted

Deno

@deno.land

21 Nov 2024

Deno 2.1 is out 🎉️ 🛩️ first class Wasm support 🌳️ Long Term Support branch ⭐️ Improved dependency management and much more. deno.com/blog/v2.1

8 Nov 2024

To help dev-twitter settle in, I made a quick little thing to help find your people here 🤓 It includes scripts to export your GitHub and/or Twitter follows to a file. And then in a localhost app, it allows you to automate searching & following of your social rolodex ✨ github.com/lukeed/blues...