Problem
Inlining forward-compat machinery (additionalProperties, validate/validity, dual accessors) into every generated class doesn't scale — the reference SDK is ~850k lines across ~1,100 model files, which would blow our coverage floor and binary-compat baseline.
Proposed change
Push the invariant machinery into the runtime (the four-state field model) and emit only the field list + accessors per model (≈<100 lines each). Decide up front to exclude generated modules from the aggregate Kover floor and give them a separate .api baseline.
Acceptance
Depends on the four-state field model.
Dependencies
Priority: highest · Effort: large
Problem
Inlining forward-compat machinery (additionalProperties, validate/validity, dual accessors) into every generated class doesn't scale — the reference SDK is ~850k lines across ~1,100 model files, which would blow our coverage floor and binary-compat baseline.
Proposed change
Push the invariant machinery into the runtime (the four-state field model) and emit only the field list + accessors per model (≈<100 lines each). Decide up front to exclude generated modules from the aggregate Kover floor and give them a separate
.apibaseline.Acceptance
Depends on the four-state field model.
Dependencies
Priority: highest · Effort: large