Add a __nowarn__ keyword

[Kaveh R. GHAZI]

On Wed, 8 Aug 2007, Paolo Bonzini wrote:

> > The warnings (messages!) that are not modifiable are those that don't
> > have the first argument of warn() appropriately set. So if that
> > message is emitted by warn(0, "cast discards...") instead of
> > warn(OPT_Wcast_qual, "cast discards...") that is a bug and it will be
> > really good to fix it.
>
> I believe these were preapproved, weren't they?
>
> >> In addition to what he said,
> >> you also can't use #pragma in a macro definition.
>
> You can, it's _Pragma("foo").
> Paolo

(Ok, imagine I just had a head-smacking "Doh!" moment.)

So this is encouraging, I added OPT_Wcast_qual to the warning statement
and now it obeys the #pragma (or _Pragma).  However I'm still having one
last problem.  The pragma interface is still on/off, not push/pop.  I.e.:

	#pragma GCC diagnostic ignored "-Wcast-qual"
	extern __inline void *CONST_CAST(const void *cv)
	{
	  return (void *)cv;
	}
	//#pragma GCC diagnostic error "-Wcast-qual"

	void foo (const void *cv)
	{
	  void *v = CONST_CAST(cv);  // WARNING1
	  void *v2 = (void *)cv;  // WARNING2
	}

Assuming you've corrected the missing OPT_Wcast_qual in c-typeck.c, if you
compile the above with -Wcast-qual -Werror, the code correctly avoids the
warnings on the line marked WARNING1.  However we should still get a
warning for WARNING2 but we don't.  I believe the pragma has changed the
behavior of gcc for the rest of the entire file.  There is no "pop" at the
end of the inline function CONST_CAST.

In order to see WARNING2, I have to uncomment the second pragma.  But this
is bad IMHO because it overrides the command line.  Assuming the inline
CONST_CAST function is defined in say system.h every file that includes it
will get -Wcast-qual hardcoded on.  I don't think this is desirable for
gcc or other project that start to use this mechanism.

		--Kaveh
--
Kaveh R. Ghazi			ghazi@caip.rutgers.edu