[bug-gnulib] Re: [PATCH]: fix warning in the hash module

[Ian Abbott]

On 18/05/2005 19:29, Yoann Vandoorselaere wrote:
> On Wed, 2005-05-18 at 16:16 +0100, Ian Abbott wrote:
>
> > On 17/05/2005 10:20, Yoann Vandoorselaere wrote:
> >
> > > However, I still consider this version of the patch to be a hack, when
> > > the real bug here is that you should not use const since you permit the
> > > user to free() the data from the interface callback. The interface
> > > should not use const at all.
> >
> > Can you show how the Hash_data_freer callback is called from 
> > hash_insert, because I can't see it!
>
> It is not. The point was that the interface allow the user to specify a
> callback for freeing the data, and thus the pointer to theses data
> should not be marked as const.

In that case, I agree with Jim that the current prototype declaration is 
correct.  The function does not modify the data through the pointer, 
hence the thing it is pointing to should be declared const within the 
function.

There is an unavoidable removal of the const qualifier because the 
function returns a non-const qualified pointer to the const qulaified 
data (similar to strstr).  That could be avoided by making the function 
return a 'const void *', but that would be the wrong thing to do, as the 
caller might want to modify the data and would have to cast away the 
const qualifier itself to do that.

Your workaround using a union would hide the warning about removal of 
the qualifier.  Whether or not it is worth implementing is a matter of 
opinion.  Presumably, the same sort of trick would have to be done in 
gnulib's strstr implementation and similar functions too.

--
-=( Ian Abbott @ MEV Ltd.    E-mail: <address@hidden>        )=-
-=( Tel: +44 (0)161 477 1898   FAX: +44 (0)161 718 3587         )=-