Rulebearing Playground Coverage Schemas GitHub

ADR-0009: dependency-cruiser's and ArchUnitNET's test suites are the specification

Context

"Superset" is a claim that must be measured. dependency-cruiser (MIT) ships 546 extraction fixtures, 25 validation specs, graph-utility specs, one directory per reporter, and whole-cruise expectations. ArchUnitNET (Apache 2.0) ships TestAssembly and ArchUnitNETTests, whose fluent assertions specify every predicate and condition. Both are pinned to versions.

Decision

Two conformance gates are required checks from the first pull request and both ratchet.

Gate 1, dependency-cruiser 18.2.0, in five layers: (1) every test/extract fixture byte-compared through rb-extract-ts; (2) the original test/validate and test/graph-utl specs run unmodified under a Node harness with the #validate and #graph-utl imports remapped to a shim that calls rulebearing validate --rules - --module -; (3) every test/report/<reporter> fixture byte-compared through rulebearing fmt --from dependency-cruiser, version string normalised; (4) every emitted json validates against the pinned cruise-result schema with extensions stripped, and every accepted config validates against the configuration schema; (5) zero-diff on the TypeScript oracle repositories at pinned commits, plus a mutation branch with twelve deliberate violations both tools must report.

Gate 2, ArchUnitNET 0.13.4: TestAssembly built with a portable PDB and committed as a fixture; each test in ArchUnitNETTests/Fluent/Syntax/Elements/** ported to a data-driven case with the same selection and expected pass or fail set; NetArchTest's test project treated the same way; the .NET oracle repositories' tests, imported by import archunit, must agree with dotnet test.

Ratchets: conformance/excluded.json (with a reason per entry) may only shrink; the unported-test count may only fall. The two coverage tabs in docs/artifacts/ are the ledger: a row may not say Parity until the pinned suite says so.

Test beds: oracle, greenfield and scale repositories are cloned at pinned SHAs in a nightly run that publishes a zero-diff table, the init fixtures and a timing table; a timing regression over 20% fails the nightly run.

Consequences

Alternatives considered