# Fix #629: Incorrect method reference results when return value requires boxing

[alwaysalearner1234]

Description

Fixes #629.

Method references that return a primitive value were returning incorrect values when the functional interface expected an object.

For example:

Function<String, Integer> f = String::length;

The primitive int result was not being boxed before being returned as an object.

Fix

• Detect primitive-to-reference return type conversions.
• Box primitive values using the appropriate wrapper valueOf() method.
• Keep existing primitive and reference return behavior unchanged.

Example

Before:

Function<String, Integer> f = String::length;
→ incorrect value

After:

Function<String, Integer> f = String::length;
→ 5

Testing

Tested with method references including:

• String::length
• String::isEmpty
• Integer::parseInt
• Bound instance methods returning primitives
• ToIntFunction<String> to ensure existing behavior still works

[cyrille-artho]

On my Mac laptop with OpenJDK 11.0.32, 20 unit tests fail.
The result is the same for the current version of jpf-core, branch master, in the main repository.
On Darshan's system (OpenJDK 11.0.31 on IntelliJ on Windows), all unit tests pass.

[alwaysalearner1234]

Thanks for taking the time to test this and verify the behavior on your end!

I investigated the test failures locally and compared the PR branch against a clean checkout of upstream/master. The failures are also reproducible on the current master branch, so they do not appear to be introduced by the changes in this PR.

The failures seem to be environment-specific, particularly given the difference between the macOS/OpenJDK 11.0.32 and Windows/OpenJDK 11.0.31 environments.

For this PR, I focused on the method-reference boxing issue described in #629 and verified that the relevant regression cases behave correctly after the change.

I'll keep the PR focused on #629 and leave the pre-existing test failures as a separate issue.

[cyrille-artho]

Two unit tests fail.