feat(sysiolib): add sysio::slug_name and make it an abigen builtin

[heifner]

Landing 3 of the slug_name carrier unification.

⚠️ MUST NOT MERGE BEFORE wire-sysio#619

This makes abigen stop describing slug_name — no typedef, no struct_def — which is only safe once the host knows the type intrinsically (#619's built_in_types entry).

Nothing breaks at the instant of merge: the committed .abis still carry a slug_name struct_def, so _is_type resolves it through the struct table. The failure arrives with the next contract-artifact regeneration on a host without #619, and then it is a hard throw. A regenerated ABI offers slug_name as neither builtin, typedef, struct, variant, enum, nor protobuf type, so _is_type (abi_serializer.cpp:401-411) misses on all six, validate() asserts invalid_type_inside_abi (:449), and set_abi propagates it (:228). get_table_rows builds a fresh abi_serializer per request (chain_plugin.cpp:2861, :3036), so it throws on every request — and validate() walks every struct field, so non-key fields fail like keys.

Scope: five contracts and all 103 pre-existing slug_name fields — reserv 48, uwrit 24, opreg 14, tokens 12, chains 5 — plus anything else building an abi_serializer over those ABIs, including JSON push_action.

The BE-key FC_ASSERT("Unsupported BE key type") is not the mechanism: it is caught and dlog'd (chain_plugin.cpp:2464-2474) with the default level at info, so alone it is silent and only degrades JSON key output to hex.

What changes

• sysio/slug_name.hpp (new) — the traits, a basic_name-derived slug_name struct, and the _s literal. This is the packed registry-code identifier the depot keys chains::chains, tokens::tokens, tokens::chaintokens, reserv::reserves and uwrit::locksums on. It lived only in wire-sysio's contracts/sysio.opp.common as a hand-rolled struct with its own packing loop, and belongs beside name — both are basic_name instantiations.
• plugins/sysio/gen.hpp — "slug_name" in the abigen builtins set.
• basic_name.hpp — CDT_REFLECT(value) on the template.

Nothing in CDT includes the new header: basic_name.hpp reaches every contract TU via name.hpp, so an alias reachable from there would be a hard redefinition against every wire-sysio TU still including the old copy.

slug_name is a DERIVED STRUCT — the alias rationale below was wrong

An earlier revision of this description argued for an alias, predicting that a derived
struct slug_name : basic_name<…> "would emit a base-carrying struct with zero fields plus a second
basic_name_slug_name_traits struct, because add_struct records one base and iterates only declared
fields". That diagnosis of add_struct was exactly right. The conclusion drawn from it was not.

Both forms leak, and the alias leaks worse:

declaration	abigen emits	host reaction
using slug_name = basic_name<…>	types: [slug_name -> basic_name_slug_name_traits] and the struct	rejects the ABI: duplicate_abi_type_def_exception: type already exists 'slug_name'
struct slug_name : basic_name<…>	slug_name (base, no fields) and the struct	shadows the struct_def; the orphan base is noise
derived + the add_struct guard	nothing	clean

The builtins entry never protected the alias: abigen resolves an alias to the underlying template
before the builtin match can apply. sysio::name escapes only because it is a derived struct, which
is why slug_name is one now.

This was unreachable until wire-sysio#619 deleted the contracts' own hand-rolled slug_name, because
no contract had ever fed CDT's type to abigen. It surfaced as 523 of 762 contract tests failing the
first time they did.

add_struct skips a builtin

The base walk in add_struct runs unconditionally, before any builtin check applies to the derived
type, so a builtin's base was described as an orphan struct_def that no field references — five ABIs
carried basic_name_slug_name_traits. add_type already skips builtins, so this is only reachable
where a caller force-adds a struct: today the kv-key path, which adds a table's key struct so clients
can reference it. Correct for a composite key, wrong for one that is already a builtin. name was
escaping by luck — no table is keyed on a bare name, and the moment one were, its base would leak
identically.

One validation algorithm, shared with the host

basic_name gains validity_error() — the single predicate both the constructor and is_valid_literal()
use — plus pack() and is_canonical(). validity_error and pack are token-identical with the
host-side fc::basic_name (same md5); only the throw mechanism differs (sysio::check with the traits'
message vs throw_invalid, which also names the offending input).

This gives the contracts the primitive they had no way to express before: "is this raw uint64 a code?"
is now slug_name::is_valid_literal, which is what wire-sysio#619's proto-boundary validation calls.

CDT is deliberately stricter than upstream CDT here — it exists only to interface with wire-sysio,
so the host's rules are the contract. In particular a trailing pad is rejected (upstream CDT accepts it,
upstream Spring rejects it; the two upstreams disagree, so "match upstream" could not settle it).

CDT_REFLECT on the template

to_key's generic arm reflects rather than consulting operator<<, so SYSLIB_SERIALIZE does not help it and an unreflected basic_name writes a silent zero-byte key.

This is defensive, not a live fix. to_key has no callers — core/sysio/key_utils.hpp is included by nothing here. The near-miss worth knowing: sysio::multi_index is a shim over kv_multi_index.hpp, which includes kv_utils.hpp (be_key_stream) — one letter from key_utils.hpp, and a different file. Both multi_index and kv::table encode through be_key_stream and never reach to_key.

Declaring it on basic_name covers slug_name, an alias with no declaration of its own. It does not cover sysio::name, a derived class whose own CDT_REFLECT(value) (name.hpp:213) hides the base's — so that one stays in effect and is not redundant.

Tests

tests/unit/slug_name_tests.cpp, 6 cases, registered in both places: tests/unit/ builds it, tests/CMakeLists.txt registers it with ctest. CDT tests are not globbed, so building without registering yields a test that compiles and never runs.

The centrepiece is byte identity with the host — slug_name{"LIQSOL"}.value == 53413609783296 plus six more constants taken from fc::slug_name's output. The two implementations live in separate toolchains and agree only because their traits agree, so these constants are the encoding's only mechanical guard: a failure means the traits diverged, so do not edit them. Also pinned: the traits policy, the 2^42 floor, and prefix grouping.

Two guards cover the lines the suite could not otherwise see:

• abigen-pass/slug_name_builtin pins that a sysio::slug_name field reaches the ABI as the bare builtin on both paths that can leak it — an action field type and a kv table's key_types — with no struct_def and no typedef. Verified by removing the builtins entry and rebuilding: with it gone, this fixture is the only one of 70 to fail.
• slug_name_to_key_writes_eight_bytes pins the CDT_REFLECT. Every other case here passes with the reflection deleted. It includes key_utils.hpp explicitly because nothing else does — itself why to_key had no coverage.

Verification

ctest -L unit_tests 32/32, toolchain suite 70/70. ctest counts binaries, not cases, so the deltas are inside them: slug_name_tests runs 6 where it ran 5, the toolchain suite 70 where it ran 69.

[huangminghuang]

Re-reviewed at head 5dba0330. wire-sysio#619 correctly fixes the host-side ABI/JSON support, so that dependency is not itself a finding. The issues below remain in CDT's generated ABI and public type implementation. I reproduced the ABI cases with packaged merge-base/HEAD compilers and the native serialization case with GCC 13 versus Clang 18.

[huangminghuang]

I accepted the rationale for keeping not_normalized_message unconditional. The exact serializer, action-method collision, and direct-key findings from the previous review are resolved. Two current-head issues remain below.

[huangminghuang]

The two findings from the previous review are fixed: the public header now compiles standalone under GCC 13, and direct record/enum collisions are diagnosed. I continue to accept the not_normalized_message rationale. The new collision guard still has three reproducible bypasses that produce the same missing-type ABI mismatch, detailed below.

[huangminghuang]

Re-reviewed at 0eb9eb9d. I accept reverting and deferring the builtin-name collision diagnostic; the three prior guard findings are resolved. The exact serializer, direct builtin KV-key shape, and self-contained header fixes remain valid. One separate public-template regression remains below.

[huangminghuang]

Re-reviewed at 9b3bbeb8. The convertible-alphabet regression is fixed comprehensively by normalizing every use through a constexpr std::string_view, with coverage for a policy exposing no find, size, or subscript operation. The serializer, direct builtin KV-key, and self-contained-header fixes remain valid. I continue to accept deferring builtin-name collision handling to a separate all-builtins change. Linux, package, and required checks pass; the documented wire-sysio#619 merge-order dependency remains. No remaining findings.

[huangminghuang]

Re-reviewed at 3b340d0f97338a1a96215d60a4ff1ac70a6163c0 using the open-code-review delegate workflow. No remaining actionable findings.

The action-method collision is now diagnosed before its synthetic ABI wrapper can be silently dropped. I rebuilt the current ABI-generation plugin and passed seven focused checks: the original reproduction, the committed failure fixture, collisions with name and asset, preservation of an explicit action name when the C++ method has a safe name, and both slug ABI fixtures. The library files are unchanged from the prior review, whose three affected unit-test binaries and GCC serialization/canonicality checks passed. The uniform traits requirement remains accepted.

Linux build/tests pass. Package verification stopped on an Ubuntu dependency download (libexpat1, HTTP 404), so that check remains unverified; macOS CI is still running. Full integration tests were not rerun locally. The documented merge-order dependency on Wire-Network/wire-sysio#619 remains in force.