bigwidth — the trait of big-number widths

Source

lib/bigwidth.low

Layer

L0 — pure computation

Capabilities

none

It holds one trait, width. bigint16 · bigint32 · bigint64 · bigint each give a type limb that satisfies it — bigint16.limb is a value meaning «compute in 16-bit limbs». When a library op that uses big numbers takes this trait, the caller picks the width on each call. The ops that take it are rsa’s verify_pss_by · verify_pkcs1_by and, since 2026-10-24, the _by twin of every curve op that computes with big numbers (p256 · p384 · ed25519 · ecdsa — generated from the hand-written ops by scripts/gen-curve-by.py).

proc verify_pss_by input comptime wt type . input lw wt . … requires bigwidth.width wt. . .
  … method lw. mod_exp. ow. sw. e. nw. mb. r2. acc. mw. tw. . …

let w lit bigint16.limb do tag 0 . end . .
rsa.verify_pss_by bigint16.limb. w. sig. n. …     ← only this call uses 16-bit limbs

The methods have the names and inputs of the big-number ops — limbs · zero · from_bytes · to_bytes · from_bytes_le · to_bytes_le · ge_mod · dbl_mod · mont_mul · n0inv · mod_exp · add_mod · sub_mod · is_zero · equal · bit. The width is fixed at translation, so choosing costs nothing (300 RSA-2048 exponentiations · native -O2 · x86-64: direct call 0.101 s, through the trait 0.102 s — measured 2026-10-23).

Why a module of its own — the four width modules use it, and bigint uses three width modules. With the trait inside one of them, they would call each other in a cycle.

There is a separate way to change the width for the whole program — build limb_bits in pkg.low or lowentc --limb-bits (bigint).