I have pushed the change for
#define XXXType char* to typedef const char* XXXType; 2f1984a65b4f
Please fix or report any problems to petsc-dev at mcs.anl.gov
Congratulations to Karl for his first removal of undesirable #defines in PETSc
:-)
Onward to a CPP free PETSc :-)
Barry
BTW: the nightly builds are terrible, people are pushing incompletely tested
code.
On Sep 30, 2012, at 8:46 AM, Karl Rupp <rupp at mcs.anl.gov> wrote:
> Hi,
>
> >>> ^^^ Note that this const is superfluous.
>>> Why is it superfluous? Isn't the second argument type const char* const
>>> this way?
>>>
>>> It's superfluous for the same reason we don't "set" by passing "const
>>> PetscInt". The const is irrelevant to the caller. All it means is that the
>>> implementation doesn't change the *its* copy (pass by value) and even that
>>> isn't type checked with respect to the public declaration. It's just
>>> clutter and suggests that the person who wrote it doesn't understand types.
>>
>> Which clearly I don't :-(
>>
>> So do we just go with typedef const char* VecType and then all
>> signatures are VecType?
>
> yes, Jed is absolutely right on that. The important thing to keep in mind
> here is that typedef is atomic, i.e. it is not a 'string replacement' as
> opposed to what the C preprocessor does.
>
> Example:
> #define VecType char*
> int Set(const VecType)
> expands to
> int Set(const char*)
>
> However, with
> typedef const char* VecType
> int Set(const VecType)
> there is no immediate string replacement, i.e. it is NOT the same as
> int Set(const const char*)
> with the compiler ignoring one of the const keywords. Instead, denoting the
> precedence using brackets, the type passed to the Set() is
> const {const char*}
> which is a const copy to a const char*. Now, as 'const char*' is copied when
> passed to the function anyways, the first 'const' is superfluous as Jed
> pointed out. As a consequence, we can simply use
> Set(VecType);
> Get(VecType*);
> which is pretty much the standard prototype for any pair of Getter/Setter and
> also the way arguments of type PetscInt are handled.
>
> In conclusion, we have not only eliminated the preprocessor magic, we are
> also able to provide a cleaned-up interface for XYZSetType() and
> XYZGetType(). Thanks, Barry, for making the decision in favor of a typedef :-)
>
> Best regards,
> Karli
>
>
