ddn.encoding.bid

ddn.encoding.bid — IEEE 754 binary integer decimal (BID) interchange for decimal64 and decimal128.

Field/bit and buffer codecs for the BID encoding only — **no IEEE arithmetic**, no ddn.num.decimal, no var, no binary-float bridge (design record doc/DDN-ENCODING.md D30–D36). Buffers are big-endian (interchange convention).

Semantics summary:

• Classification and field extraction are total: every bit

  pattern classifies (classify64/128), and fromBits64/128 preserve raw fields — including NaN payloads, signed zeros, infinity trailing bits and non-canonical finites (flagged).

• Field→bits and buffer operations are fallible Result

  operations: canonical writes reject out-of-range fields (INVALID_FIELD), short output is non-committing (INSUFFICIENT_CAPACITY), short input is TRUNCATED.

• Value interpretation of non-canonical finites (coefficient

  beyond 10^p−1) is the standard's "treat as zero" rule — this codec flags instead of erasing, and documents the rule.

decimal64 layout: sign (bit 63); 5-bit combination (bits 62..58): 11110 infinity, 11111 NaN (bit 57: 0 quiet / 1 signaling, bits 56..0 payload); otherwise finite — small form when bits 62..61 ≠ 11: E′ = bits 62..53 (10 bits, bias 398), C = bits 52..0; large form when 11: E′ top two = bits 60..59, low eight = bits 58..51, C = implicit 100 before bits 50..0. Value = (−1)^s × 10^(E′−398) × C; canonical: C ≤ 10^16 − 1 (which also bounds E′ ≤ 767 by construction).

Example:

import ddn.encoding.bid;

auto v = bid.fromBits64(0x31C0000000000001uL); // decimal64 1.0
assert(v.cls == BidClass.FINITE && !v.sign);
assert(v.biasedExponent == 398 && v.coefficient == 1);