nx_zigzag_varint.nx

nx_zigzag_varint.nx -- Protobuf zigzag encoding for signed integers. Distinct from nx_zigzag.nx (JPEG/MPEG/H.264 8x8 DCT scan-order primitive). Same word, completely different math. This one is the Protobuf signed-integer compression scheme. Problem: a naive varint encoding of a signed i64 wastes bytes on small negative numbers. -1 encoded as two's complement i64 is 0xFFFFFFFFFFFFFFFF, which a 7-bit-continuation varint emits as 10 full bytes. Zigzag remaps signed integers so small absolute values (positive OR negative) encode to small unsigned values: 0 -> 0 -1 -> 1 1 -> 2 -2 -> 3 2 -> 4 ... The mapping is bit-twiddled: encode(n) = (n << 1) ^ (n >> 63). (Note: this relies on >> being arithmetic-shift, which is correct behaviour for the encoding step per the cardinal on shift-arith in the parser-quirks meta-class memory.) After zigzag, a varint of (encode(n)) emits 1 byte for |n| < 64, 2 bytes for |n| < 8192, etc. For game deltas that cluster near zero (velocity changes, health deltas, small move offsets) this saves ~50% of bytes on typical streams. Composes with nx_state_delta_codec's varint codec to build a world-class binary protocol for poor-internet multiplayer. Source: developers.google.com/protocol-buffers/docs/encoding#signed-ints Introduced by Google Protocol Buffers (Sanjay Ghemawat + Jeff Dean 2001), now a de-facto standard for binary protocols. genealogy_id: protobuf_2001_zigzag_varint_encoding lineage_id: signed_integer_compact_encoding