Re: [HACKERS] HPUX Port

On Mon, 13 Jul 1998, Bruce Momjian wrote:
>>>> Applied. I see it is needed by configure so it finds rint on hpux. I
>>>> assume you have looked at makefiles/Makefile.hpux.

Yes, the HPUX makefile was fixed some time ago. The last piece of the
puzzle was to teach configure about it.

The Hermit Hacker <scrappy(at)hub(dot)org> writes:
> On Tue, 14 Jul 1998, Vince Vielhaber wrote:
>> What is rint? I just checked HP-UX 9.0 running on my 360 and it's not
>> there. Is it a special thing for PA?
> DESCRIPTION
> The rint() and the rintf() functions return the integral value
> (repre-sented as a double or float precision number) nearest to x
> according to the prevailing rounding mode.

rint() didn't use to be a standard part of libm, but I think it's
mandated by recent versions of the IEEE float math spec. In HPUX 10,
it's part of the standard math library libm. In HPUX 9, it's not in
the standard libm but is in the PA1.1-only libm that's kept in
/lib/pa1.1. The patches we're talking about have to do with configuring
Postgres to use that math library so it can use the native version of
rint().

I have no idea whether rint() is available anywhere for Series 300
machines. You don't have to worry too much if not; there is a
substitute routine available in the Postgres distribution.

BTW, does anyone have access to HPUX 9 running on a PA1.0 processor, ie
an old Series 800 machine? Is the /lib/pa1.1 directory even present in
such an installation? It suddenly occurs to me that we may need a
smarter approach to dealing with /lib/pa1.1, if that directory could be
present on machines that can't use the code in it. The current setup
in configure and Makefile.hpux will do the right thing if /lib/pa1.1
is not there at all, but if it is there on a 1.0 machine then you'd
end up with an unusable executable...

regards, tom lane