Andrew Johnston billywhizz

path                                     type   inode                      size                       modified       perm    ver    dev      uid      gid  links suid sgid stky
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
.                                        dir    1                             0       2026-02-28T21:20:27.000Z  rwxrwxr-x    8.1    0.0        0        0      0    0    0    0
dev                                      dir    2                             0       2026-02-28T21:20:27.000Z  rwxrwxr-x    8.1    0.0        0        0      0    0    0    0
dev/null                                 chr    3                             0       2026-02-28T21:20:27.000Z  rw-rw-r--    8.1    1.3        0        0      0    0    0    0
dev/zero                                 chr    4                             0       2026-02-28T21:20:27.000Z  rw-rw-r--    8.1    1.5

Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz
#100~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC x86_64 GNU/Linux

$ node async-bench.js 
node_v24.13.1    sync time            time     1537 rate     19514068 rate/core     19514068 ns/iter        51.24 rss     58302464 usr 100.00 sys   0.00 tot 100.00
node_v24.13.1    sync time            time     1514 rate     19814339 rate/core     19814339 ns/iter        50.46 rss     58937344 usr 100.00 sys   0.00 tot 100.00
node_v24.13.1    sync time            time     1473 rate     20361091 rate/core     20361091 ns/iter        49.11 rss     58937344 usr 100.00 sys   0.00 tot 100.00
node_v24.13.1    sync time            time     1451 rate     20674993 rate/core     20883832 ns/iter        48.36 rss     58937344 usr  99.00 sys   0.00 tot  99.00
node_v24.13.1    sync time            time     1479 rate     20270737 rate/core     20270737 ns/iter        49.33 rss     58937344 usr 100.00 sys   0.00 tot 100.00

kevent with timeout = 0
rate 80986.22 nanos_op 12347
rate 81772.37 nanos_op 12229
rate 81650.72 nanos_op 12247
rate 81608.60 nanos_op 12253
rate 81304.12 nanos_op 12299
kevent64 with timeout = 0 and flags = 0
rate 81713.43 nanos_op 12237
rate 81648.45 nanos_op 12247

issue summary

setImmediate in node.js has a lot of overhead on macos compared to linux. attention was drawn to this in this tweet from jarred sumner.
in the referenced benchmark, node on macos was only able to achieve 80k ticks per second of the event loop versus millions per second for Bun & Deno on macos
on linux, difference between node, Bun and Deno are much less significant
the change implemented in Bun for MacOS uses the KEVENT_FLAG_IMMEDIATE flag to create a fast return from the syscall when there is no pending io.
the flag in question is only available when using the kevent64 syscall on macos
libuv uses the kevent syscall, which is common across bsd based platforms.
i built a version of node with [patches](https://github.com

https://github.com/microsoft/litebox https://hy.tencent.com/research/100025?langVersion=en https://crawshaw.io/blog/eight-more-months-of-agents https://mitchellh.com/writing/my-ai-adoption-journey https://red.anthropic.com/2026/zero-days/ https://www.luiscardoso.dev/blog/sandboxes-for-ai

Configuring a development environment for Bun can take 10-30 minutes depending on your internet connection and computer speed. You will need ~10GB of free disk space for the repository and build artifacts.

If you are using Windows, please refer to this guide

Using Nix (Alternative)

A Nix flake is provided as an alternative to manual dependency installation:

nix develop

	function getDatabase (name = 'ConversationsDatabase') {
	const req = window.indexedDB.open(name)
	return new Promise((resolve, reject) => {
	req.onsuccess = e => resolve(req.result)
	req.onerror = req.onabort = reject
	})
	}

	function getAllKeys (db, collection = 'conversations') {
	const req = db.transaction([collection], 'readonly').objectStore(collection).getAllKeys()

	"""
	The most atomic way to train and inference a GPT in pure, dependency-free Python.
	This file is the complete algorithm.
	Everything else is just efficiency.

	@karpathy
	"""

	import os # os.path.exists
	import math # math.log, math.exp

	import { api } from 'lib/sqlite/api.js'
	import { bindall, bind_custom } from 'lib/ffi.js'
	import { Bench } from 'lib/bench.js'

	const { assert, core, ptr, utf8_length } = lo
	const { dlopen, RTLD_NOW, RTLD_LOCAL } = core
	const sqlite_handle = assert(dlopen(lo.getenv('SQLITE_SO') \|\| 'libsqlite3.so', RTLD_NOW \| RTLD_LOCAL))
	const sqlite = bindall(api, sqlite_handle)

	const u32 = ptr(new Uint32Array(2))

	FROM scratch
	ADD sea /sea
	CMD ["/sea"]