Alternative: replace hand-written FullBlock Streamable with Option3<Program, Bytes>

[richardkiss]

Summary

This is an alternative design to #1456, proposed for discussion.

PR #1456 hand-writes 145 lines of Streamable for FullBlock to bit-pack a version flag into the Option prefix byte for transactions_generator. This PR proposes Option3<V1, V2> — a generic tri-state optional type in chia-traits — that lets FullBlock keep #[streamable] with zero hand-written serialization code.

Key Insight

PR #1456 appears to have four states (0b00/0b01/0b10/0b11), but there are really only three meaningful states:

1. No generator (v0 or v1 — identical to parse, since there's nothing there)
2. v0 Program generator (+ ref_list)
3. v1 raw bytes generator (no ref_list)

"v0 no generator" and "v1 no generator" are functionally identical at the wire level.

Changes

New type: Option3<V1, V2> in chia-traits

Wire encoding:

• None → 0x00
• Some1(V1) → 0x01 + V1
• Some2(V2) → 0x02 + V2

Backward-compatible with Option<T> for the None/Some cases. New nodes are the only ones that ever see 0x02, and they're hard-fork nodes.

FullBlock fields:

// Before (PR #1456):
transactions_generator: Option<Program>,
transactions_generator_ref_list: Vec<u32>,
transactions_generator_buffer: Option<Vec<u8>>,
version: u8,
// + 145 lines of hand-written Streamable

// After (this PR):
transactions_generator: Option3<Program, Bytes>,
transactions_generator_ref_list: Option<Vec<u32>>,  // None = v1, Some = v0
// + #[streamable] — zero hand-written code

Version detection: is_v1() ↔ transactions_generator_ref_list.is_none()

Wire Format

Differs from #1456 by +1 byte per block (the Option prefix on ref_list). For a hard fork protocol change, this overhead is negligible — typical blocks are KB to MB.

Trade-offs

	This PR	PR #1456
Hand-written serialization	0 lines	145 lines
Wire overhead vs current	+1 byte/block	0 bytes
Type safety	Can't set both Program + Bytes	Can set both (runtime validation needed)
Reusability	Option3 usable anywhere	One-off FullBlock impl

Tests

10 Rust tests covering v0/v1 roundtrips, prefix byte encoding, and ref_list presence. All passing.

Opened as draft for discussion — would love your thoughts on the 1-byte overhead trade-off vs. eliminating the hand-written impl.

Made with Cursor

[richardkiss]

Closing in favor of Arvid's #1456 (approved). The Option3 approach was an experiment to show we could avoid hand-rolled Streamable, but #1456 is the path forward for HF2 FullBlock wire format.