my favorite code snippet in bun v1.2 video

hint: it ain’t markup language.

Bun’s 10th parser coming next week.

claude started hallucinating `—reporter=verbose` after this sometimes and now I wonder if that should be the way to enable verbose output to disable this.

particularly servers that use lots of async/await involving multithreaded I/o like fs or fetch.

this week we’re making bun use up to 10x less memory in long-running servers.

I’ve also seen candidates blindly copy paste from terminal to ChatGPT in their browser and back repeatedly without reading or trying to understand. this is slow and doesnt work very well. It needs direct terminal access and you need to read & understand closely what it’s doing.

“what does properly used mean?”. does it know how to build & test the code? can it do that without you pressing approve each time? are errors compact enough to avoid bloating context window? did you tell it which files to look at and provide examples in your initial prompt? Etc.

imagine how ridiculous it would seem today to refuse to send email and only send physical mail.

seeing properly used AI coding tools make junior engineers massively more productive . not convinced companies will hire fewer junior engineers going forward. i do think companies will hire fewer engineers who don’t use ai tools though.

filming for Bun v1.3 next week.

It’s also a good way for us to fix the shebang issue with —bun. If you install node with bun then something like `bun node bun` would rewrite the shebang everywhere globally.

I’ve been somewhat against this but I’m coming around to the idea. I also think it’s really easy for us to add.

should we add a node version manager to bun.

the team fixed issues addressing over 1,000 👍 reactions . this is a big release for `bun install`.

RT @bunjavascript: Bun now comes pre-installed in Jules!

bugs that only reproduce in CI happen too often right now. I suspect it’s because the disks we’re using in EC2 are a lot slower than our local machines.

a bug with —linker=isolated is blocking the release. Only reproduces on Windows. Hasn’t reproduced locally, only CI.

If no test files are found, it prints a message more helpful to agents

If all tests pass then it prints the count and number of files that ran, without printing individual tests that ran