Re: [PATCH] Fix overzealous folding

[Mark Mitchell <mark at codesourcery dot com>]

Richard Guenther wrote:
> > Why?  TYPE_{MIN,MAX}_VALUE seems like it's giving useful information to the
> > optimizer.  That is a way in which enums and ints are different; why not let
> > the optimizer take advantage?
> >
> > The situation that concerns me is that, if I understand correctly, Eric's
> > patch is going to make:
> >
> >  enum E { low = INT_MIN, high = INT_MAX };
> >
> > behave different from "int" -- and I don't think it should.
>
> But the only difference between this enum and int is their TYPE_MIN/MAX_VALUE.
> So either you want the optimizer to take advantage of that difference or not.

For the enum above, there's no difference in TYPE_{MIN,MAX}_VALUE from 
int.  So, I want them treated the same.

For:

  enum E { low = 0, high = 1 };

there is a difference, and I want the optimizer to take advantage of 
that fact.

Eric's patch makes all ENUMERAL_TYPEs be treated different from 
INTEGER_TYPEs.  I think that's wrong because whether a type is declared 
as an enum or not should not in itself make a difference.  What does 
make a difference is the range of the type.  If Eric's patch was 
checking the ranges in some logical way, then I wouldn't object.

--
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713