Summary
SSZ spans ~9,100 LOC across 22 files in three stacks that don't share types, hashing, or error handling:
ethereum_ssz + tree_hash derives for consensus-spec containers (crates/eth2api/src/spec/, 3,498 LOC). Every container repeats a 5-derive line plus per-field #[serde_as] attributes — 93 derive lines, 155 Hex0x occurrences. ssz_types is not a dependency, so crates/ssz/types.rs reimplements SszList/SszVector/BitList/BitVector with 12 manual Encode/Decode/TreeHash impls (844 LOC).
- A hand-ported fastssz
Hasher (crates/ssz/src/hasher.rs, 640 LOC) driving cluster/src/ssz.rs — 29 versioned free functions selected via function-pointer tables, plus an SSZError<H> that duplicates pluto_ssz::Error — and a second locally-declared SszHashable trait in app/src/obolapi/exit.rs#L30-L39.
- Hand-written byte-level codecs for Charon wire formats (
core/src/ssz_codec.rs, 1,261 LOC), including deliberate deviations from what a derive would emit (documented in unsigneddata.rs: Charon emits an offset table even for fixed-size containers).
Consequences: two incompatible hash-tree-root paths, invisible semantic differences between types (some deliberately omit TreeHash and nothing flags it), and duplicated list/vector plumbing that upstream ssz_types already provides.
Proposed change
- Before touching any stack: pin the current hasher against Charon/fastssz with golden-vector tests (known input → known 32-byte root, expected values captured from the Go side). The existing
hasher.rs tests are self-consistency checks (manual-tree comparisons, bitlist guards) — nothing cross-implementation. Add binary round-trip tests for the zero-test codec helpers (encode.rs, decode.rs) at the same time; they are part of the surface being consolidated. Same for the crate's other untested seams: helpers.rs put_byte_list/put_bytes_n IncorrectListSize, serde_utils.rs ssz_list_u64_string_serde (string/int parse, MAX rejection, round-trip), the binary Encode/Decode bound checks of SszList/SszVector (only the JSON path is tested today), and the hasher's CountGreaterThanLimit and get_depth/next_power_of_two/calculate_limit boundaries.
- Adopt
ssz_types for List/Vector/Bitfield and delete most of crates/ssz/types.rs.
- Merge the two hasher-side interfaces (
SszHashable in obolapi vs the free-function tables in cluster) into one, e.g. a version-aware trait in pluto_ssz; collapse the duplicated error enum.
- Keep the Charon wire-format codecs (
ssz_codec.rs) as explicit, documented exceptions — they encode incompatibilities that derives cannot express.
Open question
The cluster config/definition/lock hashes must byte-match Charon's fastssz output. Before replacing the custom Hasher with tree_hash, prove equivalence on the existing fixture corpus (the verify_hashes golden files) — if fastssz semantics differ (they do for e.g. bitlist mixins and string hashing), the custom hasher stays and only the interface around it is consolidated.
Summary
SSZ spans ~9,100 LOC across 22 files in three stacks that don't share types, hashing, or error handling:
ethereum_ssz+tree_hashderives for consensus-spec containers (crates/eth2api/src/spec/, 3,498 LOC). Every container repeats a 5-derive line plus per-field#[serde_as]attributes — 93 derive lines, 155Hex0xoccurrences.ssz_typesis not a dependency, socrates/ssz/types.rsreimplementsSszList/SszVector/BitList/BitVectorwith 12 manualEncode/Decode/TreeHashimpls (844 LOC).Hasher(crates/ssz/src/hasher.rs, 640 LOC) drivingcluster/src/ssz.rs— 29 versioned free functions selected via function-pointer tables, plus anSSZError<H>that duplicatespluto_ssz::Error— and a second locally-declaredSszHashabletrait inapp/src/obolapi/exit.rs#L30-L39.core/src/ssz_codec.rs, 1,261 LOC), including deliberate deviations from what a derive would emit (documented inunsigneddata.rs: Charon emits an offset table even for fixed-size containers).Consequences: two incompatible hash-tree-root paths, invisible semantic differences between types (some deliberately omit
TreeHashand nothing flags it), and duplicated list/vector plumbing that upstreamssz_typesalready provides.Proposed change
hasher.rstests are self-consistency checks (manual-tree comparisons, bitlist guards) — nothing cross-implementation. Add binary round-trip tests for the zero-test codec helpers (encode.rs,decode.rs) at the same time; they are part of the surface being consolidated. Same for the crate's other untested seams:helpers.rsput_byte_list/put_bytes_nIncorrectListSize,serde_utils.rsssz_list_u64_string_serde(string/int parse, MAX rejection, round-trip), the binaryEncode/Decodebound checks ofSszList/SszVector(only the JSON path is tested today), and the hasher'sCountGreaterThanLimitandget_depth/next_power_of_two/calculate_limitboundaries.ssz_typesfor List/Vector/Bitfield and delete most ofcrates/ssz/types.rs.SszHashableinobolapivs the free-function tables incluster) into one, e.g. a version-aware trait inpluto_ssz; collapse the duplicated error enum.ssz_codec.rs) as explicit, documented exceptions — they encode incompatibilities that derives cannot express.Open question
The cluster config/definition/lock hashes must byte-match Charon's fastssz output. Before replacing the custom
Hasherwithtree_hash, prove equivalence on the existing fixture corpus (theverify_hashesgolden files) — if fastssz semantics differ (they do for e.g. bitlist mixins and string hashing), the custom hasher stays and only the interface around it is consolidated.