This case seems to be another one where Eclipse's Java compiler crushes javac. The only question for me is whether it's a bug in JLS or javac.
interface EndoFunctor< C, FC extends EndoFunctor< C, FC > > { /*...*/ }
interface Algebra< C, FC extends EndoFunctor< ? extends C, FC > > { /*...*/ }
The second line compiles in Eclipse, but fails to compile in javac with the message that "type parameter FC is not within its bound".
FC is declared to extend EndoFunctor< ? extends C, FC >, and the bound on FC is that it extend EndoFunctor< D, FC > for the inferred D, which in this case is ? extends C. I think javac doesn't realize that the wildcard represents the same unknown type in both contexts. Eclipse does, though!
Apparently the following gets around the problem in javac:
interface EndoFunctor< C, FC extends EndoFunctor< ? extends C, FC > > { /*...*/ }
but this is a looser definition than I want for that interface.
I could also try
interface Algebra< C, D extends C, FC extends EndoFunctor< D, FC > >
but that approach forces me to carry that extra type parameter D through everywhere.
What to do?
What to do?
Here are a couple of pragmatic solutions.
Try using
javac
from the latest patch release of Java 7. I recall hearing of certain javac compiler bugs in Java 6 that were only fixed in Java 7 ... but I don't know of a list. (And the Java Bugs Database is hopeless at searching ...)Put up with it, and use one of the two alternatives that you've already found that "work".
来源:https://stackoverflow.com/questions/9937422/a-bad-interaction-between-self-referential-types-and-bounded-wildcards