Struct LcgRng

Source

pub struct LcgRng(/* private fields */);

Expand description

32-bit integer LCG. Step once per next() / signed() call.

Implementations§

Source §

impl LcgRng

Source

pub const fn new(seed: u32) -> Self

Build with an explicit seed. Same seed → same sequence.

Source

pub fn next(&mut self) -> u32

One LCG step. Multiplier + increment are glibc’s constants. Returns the fresh internal state.

§Take the high bits, not the low ones

The low bits of any power-of-two LCG are very weak, and bit 0 of this one is not random at all. The multiplier and the increment are both odd, so x' = x*m + c gives x'0 = x0 ^ 1: bit 0 strictly alternates, period two. Bit k has period at most 2^(k+1).

So next() & 1 ping-pongs, next() & 127 cycles inside 128 draws, and next() % 25 leans on the same weak bits. Use LcgRng::next_mixed for anything that masks or takes a remainder, or shift the high half down yourself as LcgRng::signed does.

Source

pub fn next_mixed(&mut self) -> u32

One step with the strong high half folded down over the weak low half, for callers that mask or take a remainder.

VoXide worked this out and carried it as a private wrapper: its callers lean on & 1, % 4 and % 120, and the raw low bits cycle with tiny periods. It belongs on the generator rather than in one game.

Source

pub fn below(&mut self, max: u32) -> u32

Uniform-ish value in [0, max), or 0 when max is 0.

Sourced from LcgRng::next_mixed, so it is safe against the low-bit weakness a bare next() % max walks into.

Source

pub fn signed(&mut self, range: i16) -> i16

Signed integer in [-range, +range], symmetric about zero, sourced from five bits of the LCG: thirty-two evenly spaced values from -range to +range, rounded toward zero. Any range is fine, negative included (the result is then mirrored); the arithmetic is 32-bit, so it does not overflow for the whole i16 range.

Source