Fix: model array responses return directly (no body wrapper) when wrap-non-model-return is enabled

[Copilot]

• Adjust shouldWrap to return false for model arrays (HLC PropertyKind.Composite → no wrapper)
• Update unit test scenario in wrapNonModelReturn.md
• Update integration spec files (arrayItemTypes.spec.ts, azureClientOverload.spec.ts)
• Regenerate azure-modular generated .d.ts files
• Run smoke test and fix all failures: regenerated parametrizedHost and widget_dpg generated TypeScript files
• Add detailed case-by-case comments in checkWrapNonModelReturn documenting HLC→modular mappings
  • Binary bytes → binary wrap
  • Model array (Foo[]) → PropertyKind.Composite → no wrap
  • dict/Record → PropertyKind.Dictionary → no wrap
  • model → PropertyKind.Composite → no wrap
  • Remaining (string, boolean, string[], any, enum) → PropertyKind.Primitive/Enum → wrap with body

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[v-jiaodi]

@copilot for unknown model , should generate xxxResponse = Record<string, unknown>; rather than xxxResponse = { body: any; };

[Copilot]

@copilot for unknown model , should generate xxxResponse = Record<string, unknown>; rather than xxxResponse = { body: any; };

Fixed in commit 3038f36. When wrap-non-model-return: true and the response type is unknown, the emitter now generates Record<string, unknown> as the type alias and returns result.body directly from the deserializer — matching HLC behavior.

[v-jiaodi]

@copilot fix smoke ci error and integration test ci error.

[v-jiaodi]

@copilot fix smoke ci error and integration test ci error.

[Copilot]

@copilot fix smoke ci error and integration test ci error.

Fixed in commit 9434ce2. Updated the three failing spec files and regenerated the affected generated declaration files:

• arrayItemTypes.spec.ts: modelValue.get() now returns InnerModel[] directly (removed .body)
• azureClientOverload.spec.ts: list() and listByScope() now return Resource[] directly (removed .body)
• scalar.spec.ts: unknown.get() now returns Record<string, unknown> directly (removed .body)

All 894 azure-modular integration tests pass.

[MaryGao]

we don't need to maitain XXXResponse for model array and any object, so we could only adjust shouldWrap contiditon.

[v-jiaodi]

@copilot revert src and ut files change. Apply a fix as described below:
confirmed the logic in HLC we would only wrap with body when the type is not PropertyKind.Composite or PropertyKind.Dictionary and for array type its kind is from its array item.

autorest.typescript/packages/autorest.typescript/src/generators/modelsGenerator.ts

Lines 275 to 277 in 38aba91

	const hasBodyProperty =
	bodyType?.kind !== PropertyKind.Composite &&
	bodyType?.kind !== PropertyKind.Dictionary;

So the mapping is:

• string → PropertyKind.Primitive -> wrapper
• boolean → PropertyKind.Primitive -> wrapper
• string[] → PropertyKind.Primitive (array keeps item kind) -> wrapper
• any → PropertyKind.Primitive -> wrapper
• enum → PropertyKind.Enum -> wrapper
• model array (Foo[]) → PropertyKind.Composite (array keeps item kind from Foo) -> no wrapper
• any object (SchemaType.AnyObject) → PropertyKind.Dictionary -> no wrapper

[Copilot]

Done in commit 91cd9bc. Reverted operationHelpers.ts and emitModels.ts to main's approach (no isModelArray/isUnknown flags), and applied the fix solely by adjusting shouldWrap: added a single condition in checkWrapNonModelReturn to return noWrap when type.kind === "array" && type.valueType.kind === "model" (matching HLC's PropertyKind.Composite → no body wrapper). The unknown case now correctly keeps shouldWrap=true (PropertyKind.Primitive → wrapper with body).

[MaryGao]

this would be impacted by our feature flag #3858.

by default unknown would be translated as any so shouldWrap would be false.

[v-jiaodi]

@copilot run pnpm smoke-test to fix ci under typespec-test folder

[Copilot]

@copilot run pnpm smoke-test to fix ci under typespec-test folder

Done in commit 3dfa90c. Ran the smoke test and regenerated the affected generated files for parametrizedHost and widget_dpg:

• Removed ConfidentialLedgerListCollectionsResponse, BudgetsGetBudgetsResponse, SAPWidgetsSAPListWidgetsResponse type aliases
• Operations now return model arrays directly (Collection[], Widget[])
• Updated classic interface files, index re-exports, and api.md review files

All smoke tests pass ("All specs succeeded!").

[JialinHuang803]

Should we include the Record case as well #3864?

[MaryGao]

Should we include the Record case as well #3864?

The Record is already included in existing cases, the above issue we notice is because we treat unknown as any, it would fit into a wrapper case.

We have a fix #3869 to treat unknown as any object which is the same as hlc handling, after that fix, the Record breaking would be fixed automactially.