Makes sense overall. Not sure what to think TBH, but would be willing to merge in in some form.

@garretwilson WDYT?

Let me be honest—my original issue in #69 was several years and I don't readily remember all intricacies of the issue. However I'm now deep into LLMs, so just now I worked with the two top coding LLMs to analyze this pull request. The conclusion below is LLM generated, but it is a summary of a conversation that I guided, going over my original post, analyzing the pull request, looking for bugs, considering limitations, and discussing alternatives. I think you'll find it useful; be sure to check the limitations and potential bugs it points out. I provide my own note below after the summary.

Context

This PR addresses a problem I raised in issue #69 back in October 2023: the inability to convert ResolvedType back to java.lang.reflect.Type for use with libraries like Jackson. I've been reviewing this PR with LLM assistance to get back up to speed and evaluate the approach.

Alternative: Preserving the Original Type in ClassMate

One question worth considering: rather than reconstructing a Type externally, could ClassMate preserve the original during resolution?

// In ResolvedType
private final Type originalType; // null for programmatic construction
public Optional<Type> findOriginalType() {
 return Optional.ofNullable(originalType);
}

Trade-offs: Not all ResolvedType instances originate from a Type (some are constructed programmatically), so this would require nullable/Optional handling. There's also the philosophical question of whether ClassMate should carry unresolved type information. That said, it would avoid custom Type implementations entirely.

The mapper approach in this PR is a reasonable pragmatic alternative.

Potential Issues Found

1. Multi-dimensional generic arrays (High) — List<String>[][] loses its type parameter:

// Element type is GenericArrayType, not ParameterizedType, so falls through
if (elementType instanceof ParameterizedType) { ... }
return arrayType.getErasedType(); // Returns List[][] — String is lost

Fix: if (elementType instanceof ParameterizedType || elementType instanceof GenericArrayType)

2. Nested parameterized inner classes (Medium) — Owner type is raw:

// For Outer<String>.Inner<Integer>, owner becomes Outer.class, not Outer<String>
new ParameterizedTypeImpl(erasedType, ..., erasedType.getEnclosingClass());

The reconstructed type won't equals() a JVM-provided type in this case.

3. Array mutability (Low) — getActualTypeArguments() returns the internal array directly; JDK implementations return a clone.

Inherent Limitations (Not Bugs)

These stem from ClassMate's design, not this PR:

Minor Observations

• ResolvedTypeMapper is stateless—map() could be static. Is instance-based design intentional?
• Tests don't cover multi-dimensional generic arrays or nested parameterized inner classes.

Overall, this PR solves a real need. The multi-dimensional array issue should probably be addressed before merging.

Even though I pushed both LLMs to find problems, it looks like overall they both came away with a positive evaluation. I recommend checking into the bugs they mention, reviewing the limitations, and reconsidering whether you might want to just integrate this into ClassMate. In particular note the doubt about whether the facility should be stateless. I hope this summary is useful!

(Note that LLMs aren't all that innovative, so although they didn't find major errors with the PR, that doesn't indicate whether there are alternative creative solutions. Sorry I wasn't up-to-speed enough to provide more of my own direct evaluation at the moment.)

I'll keep the LLM conversation history around, so if you have any follow-up questions let me know and I'll report back the responses.