gmp 2.0.2 test buglet

[der Mouse]

I've been using GMP 2.0.2 for multiprecision arithmetic and finally had
occasion to build it on a VAX.  (A MicroVAX-II running NetBSD 1.4T.)

First, gcc (which is "egcs-1.1.2") complained several times when
building the tests, about delayed evaluation of __builtin_constant_p,
saying to report it to some cygnus.com address that doesn't exist.

That's neither here nor there at the moment, though.  What prompted
this letter is, "make check" failed, and worse, failed spuriously.
t-get_d (mpq/tests) died with a SIGFPE; on investigation, it proved to
be simple overflow.  (VAX floats (both single and double) have a range
of roughly 1e-38 to 1e38; t-get_d was trying to generate a value on the
order of 1e130.)  I can't see any code there to limit the size of the
resulting numbers; I can't see why it doesn't overflow on IEEE-float
machines too, except luck.

I just touched st-get_d by hand and carried on.  But I thought I should
at least mention it, though I suspect 2.0.2 is by now probably old
enough that nobody really cares.

                                        der Mouse

                               address@hidden
                     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B