EVALUATION
This problem has been caused by the fix of 6650759. In that fix, we augmented type-inference so that constraints generated during 15.12.2.7 were propagated to 15.12.2.8 (esp. upper bounds and equality constraints). Unfortunately, since javac sometimes produces 'nulltype' as the upper bound of an inference variable, it is sometime possible that the 'nulltype' itself is choosen as the inferred type during 15.12.2.8.
This behavior is not conformant w.r.t. the JLS and therefore should be addressed: nulltype constraints should never be propagated - or actively used by javac.
|