[wide-int] Handle zero-precision INTEGER_CSTs again

[Marek Polacek]

On Mon, May 05, 2014 at 07:32:31PM +0200, Richard Biener wrote:
> > void_zero_node is used for ubsan too, and survives into gimple.
> > I did hit this in real tests, it wasn't just theoretical.
> 
> Ugh - for what does it use that ... :/

It's used like this:
t = fold_build3 (COND_EXPR, void_type_node, t, tt, void_zero_node);
so if condition T isn't true, nothing should happen.  I remember that
initially I used something similar to void_zero_node, but that
resulted in ICEs.

	Marek