[tomtourwe]

What

1014d98 casts return Optional.empty(); into an Optional<T> as
(Optional.empty() as Optional<T>), and Optional.of(value) likewise; kotlinc cannot infer T
from a cast, and Kotlin's Optional.of takes a non-null T.

Now: (Optional.empty<T>() as Optional<T>) and (Optional.of(value!!) as Optional<T>); the !!
changes nothing, Java's Optional.of(null) throws too (reported as ASSERT_AT_DEREFERENCE).
Test: TestJavaToKotlinStructure.optionalEmptyAndOfIntoTypeVariable (output compiles with kotlinc).

How it was found

A mutation-testing run (PIT) over the lines that maddi's latest 100 commits changed listed the changed code no
existing test checks. Tests written for those lines, from what each commit says the code should do, exposed this
bug in the lines 1014d98 changed, which lists empty and of but handled only ofNullable.

Tests

On devel 07b926b: the new test fails without the fix and passes with it.
With this fix and the nine others of this series applied together, every maddi test passes
(run on 7bc974b: 2,743 tests, 0 failures).

🤖 Generated with Claude Code

[bnaudts]

Thanks. Devel had already fixed this independently in another form (ba7a889, the optionalFactory rule), so the fix itself isn't taken. Your test passes without it on devel, and it is kept as a regression test in 00e31a8, with you as author. Closing this PR.

The whole batch (#146–#155) passes the Kotlin, C#, printer, front-end, inspection and bytecode suites on JDK 26 and JDK 27. It goes into devel with the next merge of the ws/maddi-bug branch.

🤖 Generated with Claude Code