[patch] for PR 19333

[Richard Henderson]

On Tue, Jan 11, 2005 at 09:57:44PM +0100, Zdenek Dvorak wrote:
> well, in the testcase it is basically cast to (short [][4]) and then
> worked with.  I agree that it does not make much sense to me, but
> apparently someone writes code like this and expects it to work.

I'm convinced that one of two things should have happened.

Either "short a[][]" should get

  error: elements of array ' have incomplete type

or we shouldn't have removed the cast to (short [][4]).  There's
absolutely nothing sensible we can do in the tree optimizers with
a dereferenced incomplete type.


r~