Zig redefines @bitCast on logical bit layout and gains a 5% compiler speedup

Category: Languages
Status: confirmed
Sources: Zig devlog, HN
Summary: A 2026-06-25 Zig devlog entry redefines @bitCast in terms of a type's logical bit layout (an ordered sequence of bits) rather than raw memory reinterpretation, making aggregate bitcasts endian-agnostic and enabling casts like [2]u3 to @Vector(3, u2). Separately, the LLVM backend now stores arbitrary bit-width integers (such as u4, i13) as ABI-sized types in memory while keeping bit-int types in SSA form, mirroring Clang's _BitInt handling; the change produced about a 5% speedup in the Zig compiler itself. The work landed in master and targets 0.17.0.
Why it matters: Endian-agnostic @bitCast removes a class of portability bugs, and better arbitrary-width integer lowering improves codegen quality across the language.
Follow-up: Track the 0.17.0 release and any migration notes for code relying on the old byte-reinterpretation semantics.