I have a very simple macro here that makes use of ClassSymbol.knownDirectSubclasses

The macro is invoked in the test (EnumerateSpec.scala). It works as expected when you pass it the type of trait Foo, which is defined in src/main. But when you pass it the type of trait Goo, which is defined in src/test, it cannot find any of the trait's subclasses. (ClassSymbol.knownDirectSubclasses returns an empty Set, as you can see from the println output.)

If you checkout the bugreport branch of https://github.com/cb372/enumerate and run sbt test you can see for yourself.

The weird thing is, if you then make an edit to src/main/scala/enumerate/package.scala, triggering recompilation, then the macro starts working as expected and the tests pass. So I'm guessing this is somehow related to compilation order.

At first I thought it might be a thread-safety issue similar to #6240, but as far as I can tell the Symbol has already been initialized. Manually calling initialize, info, etc. before knownDirectSubclasses did not fix the problem.