(j3.2006) (SC22WG5.3584) OPTIONAL arguments and C interop

[Bill Long]

At the last WG5 meeting we cut off debate on a couple of issues related 
to the C Interop TR due to time running out before convergence occurred.

One of the unresolved issues was the design for handling OPTIONAL 
arguments in interfaces with BIND(C).  The current design, which has 
been static for a couple of years, is that missing actual arguments 
corresponding to dummies with the OPTIONAL attribute are represented as 
actual arguments with the value NULL.   A side effect of this design is 
that OPTIONAL and VALUE cannot be both specified for a particular dummy 
argument.

The motivation behind this choice was twofold:

1) This convention is widely used in C programs to indicate a "not 
present" or "not supplied" argument, so it would seem a natural rule for 
the C programmer.

2) Many Fortran implementations already use this convention for 
Fortran-Fortran calls, and so implementation of this feature would be 
minimally disruptive (amounts to disabling an error message).


One vendor objected, based on the arguments that both OPTIONAL and VALUE 
should be allowed, and that this vendor does not currently use this 
convention for optional arguments, so there would be an implementation 
burden to implement a separate convention for BIND(C) routines.  The 
alternatives proposed were:

a) Change the specification to require an extra argument that would 
indicate whether an optional argument was present. This extra argument 
would be hidden in the Fortran interface, but be explicit in the 
corresponding C prototype.  (This matches that vendor's Fortran 
convention.)

b) Delete the capability of OPTIONAL arguments in BIND(C) interfaces 
entirely.


In the discussion so far, alternative (a) was viewed unfavorably for 
multiple reasons.  However, the choice between keeping the current 
proposal and removing it (alternative b) was not fully discussed.  
Neither were any other alternatives. 

Editorial comments on alternative (b):  The discussion of OPTIONAL 
arguments in the TR, and associated edits, are a very small fraction of 
the text involved, so removing it is editorially easy. On the other 
hand,  this feature was explicitly included in the original mandate for 
the TR, so there should be a significant technical reason for change at 
this point.

Comments and Discussion is welcome.  Feel free to solicit input from 
people not on the WG5 mail list as well.

Cheers,
Bill

-- 
Bill Long                                   longb at cray.com
Fortran Technical Support    &              voice: 651-605-9024
Bioinformatics Software Development         fax:   651-605-9142
Cray Inc., 1340 Mendota Heights Rd., Mendota Heights, MN, 55120