Fix stack overflow on circular reference after satisfies

[Copilot]

const f = () => 42 satisfies typeof f causes a stack overflow due to infinite recursion between checkSatisfiesExpression → error reporting → typeToString → serializeReturnTypeForSignature → pseudoTypeToType → getRegularTypeOfExpression → checkExpressionEx → checkSatisfiesExpression.

The satisfies check fails (number vs () => any), triggers error reporting, which serializes the function type, which re-evaluates the satisfies expression, which fails again, ad infinitum.

• Pass checkMode through to checkSatisfiesExpression and skip the assignability check when CheckModeTypeOnly is set (the mode used by getTypeOfExpression in the serialization path). The check is purely diagnostic—exprType is always returned regardless of the result.

Now correctly produces the expected errors:

TS7023: 'f' implicitly has return type 'any' because it does not have a return type annotation and is referenced directly or indirectly in one of its return expressions.
TS1360: Type 'number' does not satisfy the expected type '() => any'.

[jakebailey]

@copilot is this a porting bug? If not why doesn't Strada hit this?

[Copilot]

@copilot is this a porting bug? If not why doesn't Strada hit this?

Yes, this is a porting bug. Strada doesn't hit this because it doesn't have the pseudochecker path.

The cycle is: serializeReturnTypeForSignature → pseudoTypeEquivalentToType → pseudoTypeToType (for PseudoTypeKindInferred) → getRegularTypeOfExpression → checkExpressionEx → checkSatisfiesExpression → error reporting → typeToString → serializeReturnTypeForSignature → ∞

In Strada, serializeReturnTypeForSignature uses canReuseTypeNodeAnnotation/typeNodeIsEquivalentToType which don't call getRegularTypeOfExpression during return type serialization. The Go port replaces this with the pseudochecker approach where pseudoTypeToType for inferred types calls getRegularTypeOfExpression, creating the re-entrant path that doesn't exist in Strada.