🚀 go-pugleaf

> i've a problem in the libg++-2.2 test suite (rationals test) -
> the problem occurs only on vax-ultrix.

> what is the error and how can i solve it??

As far as I know, there is no error (except perhaps in the test itself).

The problem is that VAX floating-point uses a non-IEEE representation.
The problemantical tests involve converting (double) to Rationals.
The exact result will depend somewhat on floating-point representation.

Consider:
	8c8
	< onePointTwo = 5404319552844595/4503599627370496
	---
	> onePointTwo = 21617278211378381/18014398509481984

Using my 'Q' interpreter, I get:

	Q4> 5404319552844595/4503599627370496 * 1.0
	1.2
	Q5> 21617278211378381/18014398509481984 * 1.0
	1.2

So both results are actually correct.

Arguably, the tests should be re-written to be less sensitive
to floating-point representations.  I.e. one could take the
Rational, multiply by 1000, round to an integer, and then
divide by 1000, before printing the result.  I don't care
enough to do this myself, but I would consider patches to do so.

	--Per Bothner
Cygnus Support     bothner@cygnus.com