On Fri, Feb 25, 2005 at 02:21:59PM -0800, Kevin M. Kilbride wrote:
> >I agree. Adding a const attribute to system interfaces will _never_
> >break calling code, since it simply advertises a guarantee that the
> >code will not change what is supplied. The code either does or it
> >doesn't. If it truly doesn't, then why not advertise it as such? It
> >can hardly be argued as "breaking" the POSIX interface. There is no
> >real hope of making all of userland truly WARNS=6 compliant without it.
What about code which correctly uses -Wbad-function-cast or
-Wcast-qual in combination with -Werror? That code breaks.
const char ** and char ** are not type compatibel in ISO C.
That's IMO a shortcoming in ISO C, but we can't fix it.
Changing the interface adds problems. It is possible to use
e.g. __DECONST for exactly this occurences, but it has to be
done explicitly and it is easy to find all users of such
macros to verify the correctness of those.
Joerg