[PATCH] Fix PR ada/33788

[Richard Guenther]

On Jan 14, 2008 12:15 PM, Richard Kenner <kenner@vlsi1.ultra.nyu.edu> wrote:
> > > > I wonder if we will ever see BLKmode or VOIDmode -
> > >
> > > I don't think either is valid for NOP_EXPR.
> >
> > But we nowhere check this
>
> That may be, but as far as I can tell, lots of things will fail if such
> a NOP_EXPR were to show up.
>
> > If you extract from a bitfield and have BIT_FIELD_REF_UNSIGNED unset
> > then it better should.  Even though it is not documented.  ...which we also
> > should do.
>
> In the NOP_EXPR or as part of *generating* the NOP_EXPR?  Don't lots of the
> compiler assume that a non-mode-changing NOP_EXPR doesn't generate code?

something like NOP_EXPR<int, BIT_FIELD_REF <int, ...>>.  And yes,
probably there are inconsistencies regarding the assumption that a
NOP_EXPR that doesn't change mode creates (or creates not) code.

See the various bitfield related wrong-code bugs we have.  See the interesting
and inconsistently used reduce-bit-field-operations langhook.

It's a mess and I like to not add to it in stage3.

Richard.