@leofcoin/binary

@leofcoin/binary is a dependency-free shared implementation of unsigned base-128 integers and length-prefixed byte frames. Numeric values retain the existing wire bytes. No chain field, codec identifier, hash algorithm or supply representation changes as part of this migration.

It is independent of @leofcoin/codecs, which remains the registry of codec identifiers. This package does not import or re-export that registry.

import { BinaryReader, BinaryWriter, varUintLength } from '@leofcoin/binary'

const signature = new Uint8Array(64)
const writer = new BinaryWriter(varUintLength(1) + varUintLength(129) + signature.length)
writer.writeVarUint(1).writeVarUint(129).writeBytes(signature)
const reader = new BinaryReader(writer.finish())
const version = reader.readVarUint()
const codec = reader.readVarUint()
const decodedSignature = reader.readBytes(reader.remaining)

• writeVarUint / readVarUint: unsigned safe-integer Numbers, at most 8 bytes.
• writeVarBigInt / readVarBigInt: unsigned BigInts through 2^256-1, at most 37 bytes.
• writeFrame / readFrame: Number length prefix plus bytes. A failed operation does not advance the cursor. Writes check capacity before changing bytes.
• encodeFrames / decodeFrames: a sequence of those frames; optional raw prefix bytes on encode, corresponding prefix length on decode. Encoding allocates one output buffer, without allocating a varint array for every field. Decoding is iterative, so many fields do not exhaust the call stack.

Readers return views, not copies; modifying a view modifies the input buffer. writer.finish() likewise returns a view. Use finish(true) to retain an independent copy before resetting/reusing a writer. Writers have fixed capacity; the caller supplies the capacity or an existing Uint8Array. There is no hidden buffer growth or mutable process-global .bytes state.

New reader calls reject padded/non-canonical integers by default. Existing block validator framing explicitly uses canonical: false to preserve legacy padded prefix acceptance. Tightening that consensus rule needs a separate protocol decision. Overflow and truncated fields still fail closed. decodeFrames accepts an optional maxFrames policy; network size/membership limits remain the caller's responsibility. BigInt support does not automatically migrate existing fields.

Checks: npm test -w @leofcoin/binary, npm test -w @leofcoin/messages. Benchmark: node test/bench/binary-codec.js. It compares the actual installed legacy framing packages and verifies equal bytes before measuring. Results are local CPU timings, not blockchain throughput or transport latency claims.