Lots of "make check" regressions

[Alexandre Petit-Bianco]

Bryce McKinlay writes:

> Upon inspection, gcj is never even generating a call to
> _Jv_IsInstanceOf() (arguably it shouldn't anyway because the type of
> the array is provable, but I was not aware that gcj had gained this
> optimzation)

Gcj does try to optimize the generation of Jv_InstanceOf but doesn't
try to do this particular optimization. For this particular test case,
gcj now decides that the answer should be `false' (in the past, it
would issue a call Jv_IsInstanceOf.) So yes, there's something
wrong. I'll look into it shortly. The branch is affected too.

./A