Re: [PATCH]: Fix problematic -Wcast-qual cases using new CONST_CAST macro

["Kaveh R. GHAZI" <ghazi@caip.rutgers.edu>]

On Sun, 5 Aug 2007, Gabriel Dos Reis wrote:

> On Sun, 5 Aug 2007, Mark Mitchell wrote:
>
> | Gabriel Dos Reis wrote:
> |
> | > | Richard, do you have any remaining objections to this patch?
> | > | http://gcc.gnu.org/ml/gcc-patches/2007-07/msg01993.html
> |
> | I'm not very keen on the use of a union here, just as a way to outwit
> | our own warning machinery.
>
> A good point that the union is overkill.

I used a union because it's the only solution I found that works.  I would
be happy to use something else more elegant if someone suggests something
concrete.


> |  For grep-ability, just using CONST_CAST with
> | a definition that is a traditional C-style cast, would work fine.
>
> The compiler will "see" through the macro expansion and will start
> warning when -Wcast-qual is on.

Right.



> | Or,
> | can we go through an inline function, with warnings disabled on the
> | inline function?


How do we disable warnings on the fly inside an inline function?  Did
someone implement "#pragma nowarn foo" when I wasn't looking?



> I don't know whether this works; Kaveh can you try it?
> (You would still have to write the function name all caps).
> -- Gaby

Sure, but try what exactly?

I attempted various inline function permutations and the only thing that
works is putting a union inside the inline function.  (By "works" I mean
silences -Wcast-qual warning.)  I don't see how putting the union inside
an inline function as any better than using a macro, in fact the inline
function version is slightly worse.  It can't use __typeof() like Richard
suggested because the function prototype hides the arguments actual type.

In light of this, I'd like permission to install my existing patch.  If
someone can suggest a better definition for CONST_CAST() and/or write an
inline function to replace it, it's merely a one-line change to system.h
as a follow-up.  Until then, the union hack is the only thing I can think
of that actually does the job.

Okay?

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