Path: csiph.com!weretis.net!feeder9.news.weretis.net!panix!.POSTED.spitfire.i.gajendra.net!not-for-mail From: cross@spitfire.i.gajendra.net (Dan Cross) Newsgroups: comp.programming Subject: Re: Bun Zig Rust AI rewrite Date: Mon, 13 Jul 2026 09:57:42 -0000 (UTC) Organization: PANIX Public Access Internet and UNIX, NYC Message-ID: <1132cqm$4c8$1@reader1.panix.com> References: Injection-Date: Mon, 13 Jul 2026 09:57:42 -0000 (UTC) Injection-Info: reader1.panix.com; posting-host="spitfire.i.gajendra.net:166.84.136.80"; logging-data="4488"; mail-complaints-to="abuse@panix.com" X-Newsreader: trn 4.0-test77 (Sep 1, 2010) Originator: cross@spitfire.i.gajendra.net (Dan Cross) Xref: csiph.com comp.programming:16906 In article , Stefan Ram wrote: > [snip] >| Despite fears that the automated port would result in "AI slop," the >| performance metrics for the Rust-backed Bun v1.4.0 have proven highly >| successful: >| > [...] >| >| "Unsafe" Rust Concerns Early code leaks showed a heavy reliance on >| static mut and thousands of unsafe blocks to preserve Zig's original >| raw pointers, leading to arguments over whether the code is truly safe >| or idiomatic. > [snip] The bigger problem, pointed out externally, was not the use of `unsafe` (that can be ok), but rather that the rewrite had multple prominent instances of unsound memory accesses, and villated violating the _language's_ rules and memory safety generally. Critically, many were easily triggered from _safe_ Rust code; this is supposed to be a compile-time error. People misunderstand what "unsafe" means in Rust: it does not mean you can do whatever you want. It merely means that the responsibility for upholding the invariants required by the language shifts to the programmer, who for some reason knows that a given construct does not violate the language's rules, while the compiler cannot infer this. The Bun port ignored this, in part because the Bun authors (by their own admission) did not know Rust when they started. It's telling that they did not even bother to have an agent run the test suite with `miri`, which might have detected many of the problems. Does the rewrite "work", in so far as the bun test suite passes? Yes. Will it silently break and stop working on a minor toolchain update? Probably, also yes. Note that this also speaks to the limitations of testing, and far from being a success story of LLM use, is actually a good example of AI slop. - Dan C.