On Mon, Nov 26, 2012 at 04:24:41PM +0100, Richard Biener wrote:
> On Thu, Nov 15, 2012 at 9:09 PM, Jakub Jelinek <ja...@redhat.com> wrote:
> > On the following testcase substitute_and_fold ICEs because memmove
> > of length 1 on an empty class is optimized away, and this function wasn't
> > prepared to handle that.
> >
> > Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for
> > trunk/4.7?
>
> I think the bug is that the stmt is removed. fold_stmt is not supposed to
> "remove" the stmt in any case - it may replace it with a gimple_nop at most.
>
> Thus, gimplify_and_update_call_from_tree is at fault here.
>
> I am testing a patch that fixes it.
Note that fold_stmt_1 also has code to handle a call folding into nothing,
so perhaps that could be removed too if you change it to fold into a nop
instead.
Jakub
