x86-64

Positions are u32 bytes by design (matching reference-runtime coordinates), while usize is used for in-memory indexing — so 64-bit builds address large pools without conflating file offsets with pointer widths. Byte-length math uses checked conversions (castUsizeToU32) and saturating helpers where overflow is meaningful. Large sources (100k-node benchmark corpus) parse in ~10 ms ReleaseFast on x86_64.

Related: ARM64, Performance.