Rust
Rust — Technology. drawback_SDD: Borrow Checker vs probabilistic nature of LLMs · role: Target language; safe Rust turns UAF/double-free/leaks into compile errors, auto cleanup via Drop (RAII); ~4% unsafe code after porting
Jarred Sumner rewrote Bun from Zig to Rust in eleven days, 3 to 14 May 2026, using around fifty dynamic workflows in Claude Code. The motivation was one bug class: use-after-free, double-free and leaks born from mixing JavaScriptCore's garbage-collected memory with manually managed Zig memory. In safe Rust those become compilation errors, with cleanup handled by Drop. The port ran to 6,502 commits and roughly 16,000 compilation errors treated as a work queue, cost about $165,000 in tokens, and left 4% unsafe code.
The same strictness reads as a liability elsewhere. A research report on language fit for Spec Drive Development places Rust at the high-friction end: the borrow checker conflicts with the probabilistic nature of LLMs, producing correction loops expensive in tokens and in time. Kent Beck hit the concrete version of this while building BPlusTree3, abandoning two versions as Rust's ownership model compounded complexity, then unblocking the work by writing Python first and transliterating it into Rust through Augment's Remote Agent.
Rust also appears as infrastructure rather than as subject. Home Mixer, the orchestrator of X's For You feed, and Thunder, its Kafka-fed in-memory candidate store, are both written in it.
What stays unresolved is whether the compiler is the agent's most reliable reviewer or its heaviest tax. The Bun rewrite made both true at once: 16,000 errors to grind through, and a class of memory bugs that could no longer ship.
- Type
- Technology
- drawback_SDD
- Borrow Checker vs probabilistic nature of LLMs
- role
- Target language; safe Rust turns UAF/double-free/leaks into compile errors, auto cleanup via Drop (RAII); ~4% unsafe code after porting
- relations
- 6
- Cited in
- 2 fiches
Neighborhood
← uses
→ reduces
→ replaces