[PATCH] Fix PR tree-optimization/18663

[Eric Botcazou]

> This is OK for mainline.  I guess having a fixed lower bound and
> a variable upper bound (and vice versa) is rare enough not to
> be worth optimizing?

I think so, at least for 4.0.x, judging from the ACATS results with 
tree-checking enabled.  But Richard is holding back a better patch so this 
will likely be revisited for 4.1.

I've installed the fix and closed the PR.

-- 
Eric Botcazou