Fix OptionalToObjectConverter applicability check

[junhyeong9812]

Overview

OptionalToObjectConverter (registered by DefaultConversionService for Optional -> Object) unwraps an Optional and delegates the contained value to the ConversionService.

Problem

matches() derived the contained type with TypeDescriptor.getElementTypeDescriptor():

return ConversionUtils.canConvertElements(sourceType.getElementTypeDescriptor(), targetType, this.conversionService);

TypeDescriptor.getElementTypeDescriptor() only resolves element types for arrays, streams and collections, so for an Optional it returns null. ConversionUtils.canConvertElements() then treats a null source element type as "maybe" and returns true unconditionally. As a result matches() always matched, and ConversionService.canConvert(Optional<X>, target) returned true even when X cannot be converted to target — violating the canConvert contract (a true result implies convert is capable of converting), since the subsequent conversion fails.

For example, with the out-of-the-box DefaultConversionService:

TypeDescriptor optionalInteger = new TypeDescriptor(
        ResolvableType.forClassWithGenerics(Optional.class, Integer.class), null, null);
cs.canConvert(optionalInteger, TypeDescriptor.valueOf(LocalDate.class)); // true (incorrect)
cs.convert(Optional.of(42), optionalInteger, TypeDescriptor.valueOf(LocalDate.class)); // fails

Fix

Resolve the Optional's element type from its generic and check it against the target, mirroring the sibling ObjectToOptionalConverter. A raw Optional (or otherwise unresolved element type) stays permissive, since the element type is unknown. A regression test is added to DefaultConversionServiceTests.

Note on impact

This is a contract-correctness fix rather than a fix for data corruption: the conversion itself already fails loudly (the subsequent convert throws), so nothing is silently mis-converted. The value is that canConvert/matches is a predicate callers rely on to decide whether to attempt a conversion or take a different path. Returning an incorrect true defeats that — it turns what should be a clean negative answer into a deferred exception. Restoring an accurate matches() lets callers get the correct answer up front. Impact is bounded, since it only affects Optional used as a conversion source whose element type is not convertible to the target.

[sbrannen]

Good catch, and thanks for submitting the PR to fix it! 👍

Can you please add tests that cover the Optional<?> and Optional<? extends Number> use cases in addition to the raw Optional case?

[junhyeong9812]

Thanks for the review! I have added the requested tests, plus two closely related cases:

• canConvertOptionalWithUnboundedWildcardRemainsPermissive: Optional<?> has no usable element type (the Object upper bound resolves to null), so the converter stays permissive even for a target with no matching converter.
• canConvertOptionalWithBoundedWildcardReflectsUpperBound: Optional<? extends Number> is decided against the Number upper bound — convertible to String (asserted end-to-end down to the "42" result), not convertible to LocalDate, with convert() consistently reporting ConverterNotFoundException.
• canConvertRawOptionalRemainsPermissive: the raw Optional case as its own test for visibility.
• canConvertOptionalWithUnresolvedTypeVariableRemainsPermissive: an unresolved Optional<T>, covering the third shape mentioned in the code comment.

The wildcard descriptors are built from fixture fields via TypeDescriptor(Field), since wildcard types cannot be expressed with ResolvableType.forClassWithGenerics. The branch has also been rebased onto the latest main.

[sbrannen]

Thanks for making the requested changes.

[sbrannen]

This has been merged into main.

Thanks