Richard Maine wrote:
> Related question. I know that there were f77 compilers that had RECL
> in units of bytes, and other compilers that used units of 32-bit
> words. I don't recall whether this has carried over to f90 in
> practice or not, though. Anyone know if there are any f90 compilers
> that actually use anything other than bytes for the units of RECL by
> default? (I'm not counting compatability mode switches to match older
> conventions - such switches could still remain). If not, perhaps we
> can just eliminate some of the waffling and define that to be the
> units?
I continue to prefer that the units of positioning for direct access and
stream I/O are the units of RECL in the OPEN statement, which can be
gotten using INQUIRE by IOLENGTH.
Although there are not now any systems of interest to Fortran 2000 that
use word sizes that are not multiples of 8 bits, that doesn't mean that
there never again will be.
The difficulty of providing C compilers for Unisys 1100 systems is an
example of what happens when a language specifies too much of the
computer's architecture.
Unisys architecture was attractive to the navigation teams at JPL because
single-precision arithmetic got one more digit than IEEE, while double
precision got 2+ digits more than IEEE. It would be unfortunate if the
design of Fortran were eventually blamed for a computer designer's
failure to provide an improvement on existing practice.
Allowing problem-defined record size units was a good idea that ought to
be continued. I quibble with putting the I/O list on INQUIRE instead
of OPEN, but that's exactly that -- a quibble.
Best regards,
Van
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|