[gregjotau]

Ordinary fields and properties currently trigger KotlinReflectionUtils.isDataClass even though their names cannot match the delegate-field pattern. Reject names without a $delegate suffix before inspecting Kotlin metadata. Candidate delegate names still pass through the existing data-class and regex checks, preserving their matching behavior.

Includes regression tests for both overloads and a JMH cold-start benchmark using the existing SimpleDataClass fixture. On upstream main 677c139, JDK 27 and the default Kotlin 2.4.21, ten fresh JVM forks with SingleShotTime, no warmup and -prof gc gave:

Filtering the fixture's two ordinary fields	Main	Patched
Mean ± JMH error	170.576 ± 20.560 ms	1.940 ± 0.424 ms
Allocation	22,011,113 B/op	590,817 B/op

This measures the filter's first use, including reflection initialization it triggers. Other application components can initialize the same metadata, so these figures are not a whole-application speedup claim. A separate Kotlin 2.5.0-Beta2 reproduction also improves with cold and cached metadata.

Validation: ./mvnw test passes (3,899 tests, zero failures/errors, 13 skipped). Eight fresh extracted Spring Boot application startups pass health, login-render and public API document checks. A separate instrumented startup recorded 4,019 calls to isDataClass from the filter across 328 declaring classes before the change and zero afterwards. Shared-host load prevents a reliable end-to-end timing claim. ./mvnw -Pjmh -DskipTests test-compile also passes.

Closes #3567.

[gregjotau]

@mp911de would be great to have this one for 4.2.0 spring boot :) We are already using the RC1 spring boot and benefit from the recently added performance improvements 🚀

FYI. this is just me asking gpt-6.1 sol:

can we deeply benchmark startup performance of reai app
Profile deeply and see if there is anything we can do in our own code to make it faster.
And check also if there is anything we can contribute upstream to make it faster

And then after the analysis:

Try to find at least one more worthwhile contribution.
But don't open it if you are not confident it is clearly an improvement/better

That was literally my prompt.