(j3.2006) TYPE(*) and scalars

[Bill Long]

On 3/21/12 5:58 PM, Rasmussen, Craig E wrote:
>
> On Mar 21, 2012, at 12:59 PM, Bill Long wrote:
>
>>
>>
>> On 3/21/12 1:35 PM, Bader, Reinhold wrote:
>>> Hello,
>>>
>>>
>>> [...]
>>>>
>>>>
>>>> It should be possible to work around this.  If the "known interface
>>>> name" is foo, then
>>>>
>>>> interface foo
>>>>    subroutine bar1 (x) bind(C, name="cbar")
>>>>      type(*) :: x
>>>>    end subroutine bar1
>>>>    subroutine bar2 (x) bind(C, name="cbar")
>>>>      type(*),dimension(*) :: x
>>>>    end subroutine bar2
>>>> end interface
>>>>
>>>> will work.  The specifics are distinguishable because of the rank
>>>> distinction, independent of the types.  Since you can map both of them
>>>> to the same C routine, there is no need to have extra C wrappers.  On
>>>> the C side, the corresponding formal parameter is (void *), which is
>>>> valid for either specific.
>>>
>>> Unfortunately, this will not quite solve the problem, because it now is not possible any more
>>> to process rank 2 and higher arrays - as soon as you put a DIMENSION(*) dummy into a generic
>>> interface, exact rank matching of actual with dummy will be required. I guess one could add
>>> interfaces with
>>>
>>> type(*),dimension(1,*) :: x
>>> type(*),dimension(1,1,*) :: x
>>>
>>
>> OK, but this is unrelated to whether TYPE(*) can be used without using assumed rank.  And all of these specifics can map to the same C function.  Admittedly, assumed rank is a simpler alternative.
>>
>> Cheers,
>> Bill
>>
>
> Bill,
>
> I don't understand.  It seems that you are contradicting Reinhold as he says you can't use a generic interface (as you do in your foo example).

No, Reinhold said ( I think ) that you can use the generic interface if 
you do a lot of typing.  I'm agreeing with him on that. Both the "can" 
and "lot".

>  Also the fact that two different procedures have the same binding name seems problematic.

Why? That seems exactly what you need.  There is only one of the C 
functions in the MPI library that you want to call in each of the cases.

> Suppose you try to implement bar1 and bar2 in Fortran, how does the C programmer have access  to either routine?

The point here is that you never actually implement bar1 and bar2.  C 
programmers would not access these routines - they would take the easy 
way and call the C functions directly like they always have.

> Especially since the symbol would have the same name and would lead to a conflict.
>

No conflict.  Aleks argued for this capability for some time, and 
exactly for this reason.

> I assume it does too much damage to the standard to allow TYPE(*) to take both scalar and array actual arguments?

Completely out of the question for the TS.  It is WAAAAAY to late for a 
change like this.  Apart from it not making sense to begin with.

Cheers,
Bill


> After all a scalar is now just an array of rank 0, yes (I can just hear Malcolm starting to wind up at this point so I'll not say any more :-)
>
> -craig
>
>
> _______________________________________________
> J3 mailing list
> J3 at j3-fortran.org
> http://j3-fortran.org/mailman/listinfo/j3

-- 
Bill Long                                           longb at cray.com
Fortran Technical Support    &                 voice: 651-605-9024
Bioinformatics Software Development            fax:   651-605-9142
Cray Inc./Cray Plaza, Suite 210/380 Jackson St./St. Paul, MN 55101