🚀 go-pugleaf

I FTPd the "paranoia" math testing program from research.att.com and gave
it try on my OS/2 2.0 machine.  I have both Microsoft FORTRAN 5.1 and
WATCOM F77/386 9.01b (that is, the B-level patches have been applied).

First of all, neither the double nor single precision version crashes
OS/2, regardless of which compiler was used.  I wonder why others have
experienced crashes.

Second, WATCOM passes with flying colors on the double precision test.
On the single precision test, as well as both precisions with the Microsoft
compiler, the program is unable to properly determine the radix, so the
slew of defect messages that follow are presumably consequences of this.
It looks to me like the program is fooled into thinking there are more
significant digits than there really are, because of the wrong radix, so
many of the defect messages seem to be triggered by inappropriate
comparisons.

Third, I didn't try either compiler with optimization disabled.