On 19/08/14 23:28, Patrick Alken wrote:
One issue would be that if the code were rewritten any code that passed
a const gsl_vector* argument would cause a compilation failure. The
workaround might be to use a
gsl_vector const* const v
argument. But I don’t know if that would cause problems, for example
with older compilers.
I don't have a solution to this problem, but with my gcc (4.4.7),
changing to gsl_vector* const doesn't fix the issue. Compiling with
-Wall -W produces no warnings and the program runs as before.
Actually my suggested solution won’t work. A function declaration like
void f( gsl_vector *const v );
is treated by the compiler as equivalent to
void f( gsl_vector * v );
and it’s possible to use the second form in a declaration and the first
in a definition.
The only compiler advantage of the first form is in a function
definition: you guarantee that v (though not what it points to) won’t be
changed. In a declaration it does no more than hint to the user of the
function that the data won’t be changed.
Although the gsl_vector functions can’t offer a guarantee not to change
a const gsl_vector * argument, I’m not sure this is a problem. In
practice they don’t; so there are no surprises for the user.
I think the root problem is this. Neither C nor C++ has a way to declare
a struct const so that the data pointed to by any pointer in the struct
is also const. There are workarounds. But I don’t know of any that don’t
involve copying data, which is an unnecessary overhead when the only
real requirement is to reassure a function user that nothing pointed to
by a function parameter will be changed.
--
John D Lamb
