(j3.2006) TKR for C Interop

Aleksandar Donev donev1
Wed Feb 28 12:55:28 EST 2007


On Wednesday 28 February 2007 08:06, Bill Long wrote:

>  know you want everyone to change their dope vectors to be just like NAGs
I want no such thing---I want a consistent non-Cray centered but 
programmer-centered solution. I am a user and care about the use of the API, 
not which compiler it is going to fit better than others.

What you want is NAG and Sun and others that do not contain TKR information in 
the same way Cray does to have to do work to use different descriptors for 
Interoperable procedures. If the goal is to enable *all* vendors to not 
change anything about the way they handle assumed-shape etc. arguments and 
just slap BIND(C) on the routine, then here is a NEW idea:

The routines in the API for creating descriptors can include TKR information 
as input. If the compiler uses these in the descriptor, it can put them in it 
to get a "valid" descriptor. If it doesn't, it *can* throw them away. That 
is, the descriptor does not need to store this information. There should be 
no API routine to extract TKR info from a descriptor. We need a new API 
routine to clone (copy) a descriptor, and than we are done.

This is at least consistent, IMO.

Aleks



More information about the J3 mailing list