Re: Fix PR rtl-optimization/42461

[Jan Hubicka <hubicka at ucw dot cz>]

> > If Ada language allows to remove exceptions and yet have
> > non-call-exceptions, I would strongly suggest to fork those two flags.  It
> > should not be _that_ hard to review existing flag_non_call_exceptions and
> > have flag_preserve_exceptions handle the cases where we avoid DCE (or code
> > motion or whatever) to preserve them.
> 
> Sure, we have had a patch to that effect for years in our internal tree.

I guess it would be nice to merge it to mainline (I.e. have something like
flag_preserve_exceptions).  Still we should resolve the issue with nothrow
calls and RTL land.

Honza
> 
> > I would expect that keeping in code all dead memory accesses and other mess
> > that might throw leads to quite suboptimal codegen.
> 
> Indeed, it brings about some interesting performance improvements for Ada.
> 
> -- 
> Eric Botcazou