PR java/27025: ICE on simple initializer

[David Daney]

Andrew Haley wrote:
> gcj assumes that in the operation (A + B), where A and B are
> constants, the result must be a constant.  It ain't necessaarily so.


Can you tell my how to reconcile your claim with JLS (v2 or v3) section 
15.28?  Especially in relation to the testcase in the PR.

I am not saying that your fix is incorrect or in conflict with the JLS, 
but it seems like a surprising claim.

David Daney