Open in StackBlitz

npm i https://pkg.pr.new/@typespec/http-client-csharp@11008

commit: bb87f5e

No changes needing a change description found.

No regen diff in Azure/azure-sdk-for-net#59998

Review notes (non-blocking)

LGTM overall — correct, well-scoped cycle-safety fix. CI is green and the change is approved. A few minor observations:

1. BuildConstructorParameters — dropped .Signature != null: The change from BaseModelProvider?.FullConstructor.Signature != null to BaseModelProvider is not null && !HasBaseModelProviderCycle() is safe — ConstructorProvider.Signature is a non-nullable ConstructorSignature, so the old null check only guarded BaseModelProvider. No regression; just flagging since the null-guard removal isn''t obvious from the diff.

2. Logic duplication: The cycle-safe enumeration now exists in three forms (ModelProvider.EnumerateBaseModelProviders, plus two in MrwSerializationTypeDefinition), and HasBaseModelProviderCycle re-walks the same chain EnumerateBaseModelProviders already guards. Consider consolidating to a single shared helper to avoid drift, though the private/cross-class boundaries make the current split understandable.

3. No regression test: Since the repo favors TDD, a unit test constructing a cyclic ModelProvider (custom base chain looping back) would lock in the fix and document the scenario. Cycles only originate from custom code, so it''s awkward to set up but worthwhile.

4. Description drift: The PR body lists a 4th change — "Emit new on explicit response conversion operators from ClientModel" — that isn''t present in the diff (3 files, none touching conversion operators). Worth trimming the description to match.

--generated by Copilot

Added a regression test (BaseModelProviderCycleDoesNotRecurseInfinitely) in ModelProviderTests.cs to cover the base-model-provider cycle fix.

It wires two ModelProvider instances into a cycle (A → B → A) via an overridden BuildBaseModelProvider, then exercises FullConstructor, Constructors, and Fields — the base-traversal paths that previously recursed infinitely. Verified the test passes on this branch and that reverting the HasBaseModelProviderCycle guard aborts the run with a stack overflow, confirming it locks in the fix.

--generated by Copilot