🚀 go-pugleaf

Submitted-by: johnl@iecc.cambridge.ma.us (John R. Levine)

>the GKS LIS proved you could write Fortran programs in any language.

There's another kind or program?  Uh oh.

I think we need to step back a little and consider what appears to me to
be the root of all the standardization trouble: standards should codify
existing practice, or at worse choose from among a few existing versions
of existing practice, perhaps with some minimal glue where the selected
versions don't match up perfectly.  Invented standards usually fail.  ANSI
C came close to the existing practice ideal except for the locale glop
which, not surprisingly, is the least satisfactory part of the C standard.

What the GKS LIS really proved, of course, is that we don't know how to
design usable libraries other than by experimentation, and that a suitable
design for one language is probably not a suitable design for another.
These are interesting and useful lessons, but not very helpful for
practical standards design except as a warning of what not to do.

POSIX .1 and .2 did a reasonable job of codifying existing practice,
because there was (not by accident) a lot of existing practice to codify
so it looks like they'll be fairly successful.

The problems that people have pointed out with the POSIX tests demonstrate
that it is not appropriate to standardize test suites, particularly not
test suites that haven't themselves been tested by a significant amount of
use.

I'm all in favor of test suites.  I think there should be lots of them,
since they make the implementor's design so much easier, and raise the
user's confidence that a product that passed the tests probably works.
But test suites make poor standards.  When I was writing an early F77
compiler in 1978-79 (INfort, if anyone remembers if) the compiler passed
most of the tests long before I would have asked anyone to use it on real
code.  The tests were useful, but hardly definitive.  There were places
where the test suites were just wrong, even though they'd been in use for
many years to validate F66 compilers.

The argument may be made that various pressures require various sorts of
standards do allow competitive procurement and the like.  I don't believe
it.  A really bad Fortran POSIX binding would in practice mean that nobody
would write POSIX interface code in Fortran, but would more likely write
the interface in C and use some C to Fortran hack.  Ada is admittedly a
special case, given its dedicated customers, but even there a bad enough
standard would force people to circumvent it.

Regards,
John Levine, johnl@iecc.cambridge.ma.us, {spdcc|ima|world}!iecc!johnl

Volume-Number: Volume 29, Number 98