Bun 1.4 Rust rewrite is not looking good

I attraction astir Bun. I person been rooting for it since the first merchandise successful 2022. I switched each my improvement from Node to Bun. I utilized it successful the improvement of the Nue framework and now pinch my caller task Hertta.

The past 3 months person not looked bully for Bun. It started arsenic 1 of the astir awesome individual engineering projects I person seen, but has now turned into this weird AI-powered animal pinch continuous mendacious promises and an progressively disappointment community.

In the adjacent type of Bun

In the adjacent type of Bun utilized to beryllium a affirmative tweet to watch for. For years it meant a characteristic had been implemented, tested, and would vessel successful a fewer days. This changed aft the Rust rewrite. Now the posts are mendacious promises astir the upcoming release:

It’s now 3 months and counting since the past unchangeable release, the longest spread successful Bun’s history since 2022. Nothing different there. Software slips, that’s normal. It’s conscionable that an relationship which utilized to pass pinch existent dates and existent numbers has switched to vibing. And the personification guidance is what you’d expect aft changeless mendacious promises:

Bun connected GitHub

The Bun 1.4 rewrite is simply a large stake connected AI. In the past month, 15.8k commits came from robobun, 1.6k commits from autofix-ci[bot], and 790 commits from Jarred.

6 months ago, astir of Bun’s PRs came from group prompting Claude. Nowadays, astir of Bun’s PRs travel from Claude prompting Claude.

The task has complete 5k unfastened propulsion requests, which is the largest number of propulsion requests I’ve seen. For comparison, OpenClaw has 2.2k, and React has 441. GitHub recommends staying nether 1,000 unfastened PRs against a azygous branch earlier mergeability checks commencement timing out.

The biggest interest is, of course, the codification itself. In the early days Jarred’s activity was inspirational. I thought he was a existent Zig talent, until I publication Zig creator Andrew Kelley’s thoughts connected the Bun rewrite:

We became progressively horrified astatine the programming practices we saw successful Bun’s codebase. Hacks connected apical of hacks. Abuse of assertions. Jarred was already penning slop good earlier he had entree to LLMs.

What was the problem pinch Zig?

This rewrite is the astir intimately watched real-world trial of whether AI agents tin return complete a accumulation codebase pinch a quality mostly directing alternatively than reading. Anthropic’s ain estimation is besides connected the line: if this goes well, it is existent impervious of what agentic coding tin do. If it goes badly, it will nonstop a awesome successful the other direction.

The number of unsafe blocks successful the Rust codification suggests the rewrite did not present the representation information that was fixed arsenic the logic for doing the rewrite successful the first place. Instead this rewrite feels much for illustration an Anthropic ad.

And was Zig really the problem? Bun’s early personality was built connected Zig: its performance, its accelerated compile times, its debased friction, its nonstop representation power pinch a mini team.

It feels for illustration Jarred and Anthropic decided early connected that this was going to beryllium written successful Rust, and utilized Zig’s representation issues arsenic the excuse to fto the world cognize really powerful Claude is. A rewrite for illustration this would make awesome headlines, and it surely did. Now we’re looking astatine the agelong tail of issues from the rewrite they didn’t hole for.

Maybe Bun should person put that aforesaid AI-assisted effort into disciplined, human-understood Zig alternatively of a afloat connection change. I ne'er saw Jarred earnestly prosecute pinch this option.

And ‘tomorrow’ has travel and gone. Still nary v1.4.

¯\_(ツ)_/¯