Thanks! Based on this, I created a file called "test.c" as follows:
#include <math.h>
float pg_sin(float *x) {return(sin(*x));}
and compiled it using: "gcc test.c -lm -shared" to create a shared object
file. I then tried the same steps as below (substituting
"/usr/lib/libm.so" with the path to my "a.out" file created by gcc), but
this still didn't work? I'm getting different wrong values, but they're
still wrong nonetheless?
On Tue, 1 Feb 2000, Sevo Stille wrote:
> Date: Tue, 01 Feb 2000 13:06:37 +0100
> From: Sevo Stille <sevo@ip23.net>
> To: Karel Zak - Zakkr <zakkr@zf.jcu.cz>
> Cc: mathprof@bigfoot.com, pgsql-general@postgreSQL.org
> Subject: Re: [GENERAL] Linking in sin() as a C function
>
> Karel Zak - Zakkr wrote:
> >
> > On Mon, 31 Jan 2000 mathprof@bigfoot.com wrote:
> >
> > > I tried the following, one at a time, to create sin() for PostgreSQL:
> > >
> > > CREATE FUNCTION sin(float8) RETURNS float8 AS '/usr/lib/libm.so' LANGUAGE 'c';
> > > CREATE FUNCTION sin(float4) RETURNS float4 AS '/usr/lib/libm.so' LANGUAGE 'c';
> > > CREATE FUNCTION sin(float4) RETURNS float8 AS '/usr/lib/libm.so' LANGUAGE 'c';
> > > CREATE FUNCTION sin(float8) RETURNS float4 AS '/usr/lib/libm.so' LANGUAGE 'c';
> > >
> > > Each of these gave different and odd results (and a 'segmentation fault'
> > > at one point), but none of them gave the right answer. What am I doing
> > > wrong?
> >
> > Yes, it is probably wrong. Very offen PG's buildin functions allocate memory
> > for result and IMHO your trial wrong mixing pointers. See a backend/utils/atd
> > in PG source as example.
>
> More specifically, only 1,2 and 4 byte integer values may be passed by
> call-by-value to userdefined C functions in PostgreSQL. That is, floats
> are passed by reference (pointers) while the math lib passes them by
> value - you have to write your own wrappers around the mathlib sin
> function.
>
> Sevo
>
> ************
>
