You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit a893038
Browse filesBrowse the repository at this point in the historyBrowse files
Addresses the UNSURE items from the #56 audit — own-only reads that break under
attr-level (ADR-0029) inheritance, where the codegen counterpart already reads effective:
- C# MetaIdentity: Fields/Unique/Generation/References/Enforce switched OwnAttr → Attr.
Same ADR-0029 identity edge fixed in TS (#57) — an identity that inherits @fields via
node-level extends without restating it now resolves its key (was: empty → no PK).
+ a regression gate (InheritWithoutRestateGateTests) mirroring the TS pk-resolver gate.
- Python validation: validate_discriminator @values and validation_passes
@filterable/@db.indexed reads switched attr() → attrs().get() — an inherited enum's
members no longer false-reject a discriminator value, and an inherited @filterable field
is validated consistently with the (already-effective) codegen.
- Java ValidationPhase: @filterable/@db.indexed field reads and the warnFilterableWithoutIndex
index-collection identity @fields read switched (.., false) → (.., true). (validateIdentityNode
was already FR-024-correct — conditional effective-when-super-present — and is left as-is.)
Left own-only deliberately: the validation reference-walk (validate-at-declaration) and the
M:N junction identity.reference reads (the junction must declare its references locally).
Verified, no regressions: Java metadata 1058, C# conformance 665 + codegen 12 (+1 gate),
Python 1207.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
0 commit comments