A deep technical argument for deprecating Rust's default `panic=unwind` behavior in favor of `panic=abort`. The author, a Rust core contributor, outlines multiple costs of unwinding: it forces all code to handle arbitrary interruption points, increases binary size (up to 10% measured by Fuchsia), limits compiler optimizations, makes the borrow checker more conservative (preventing moves out of `&mut` references), makes unsafe code extremely difficult to write correctly, and makes 'must move' types unworkable. The author notes that most production Rust services already use `panic=abort` or only use unwinding for cleanup rather than true recovery. The proposal is not to remove unwinding entirely (needed for C++ interop and backwards compatibility) but to make it opt-in rather than the default, allowing the borrow checker to be more permissive in unwind-free contexts.
