Re: [HACKERS] Problems with LSEGs

David Smith wrote:
>   Architecture (example: Intel Pentium)         :   DEC Alpha
>   Operating System (example: Linux 1.3.42 ELF)  :   Digital UNIX 4.0
>   PostgreSQL version (example: Postgres95-1.01) :   PostgreSQL-6.1 (3 June)
>   Compiler used (example:  gcc 2.7.0)           :   DEC C V5.2-023
> ... the slope of the segment does appear to be an
> issue. If the point is on the segment, that too causes a problem.

I can't get your test case to fail on my i686/Linux machine. I'm very
familiar with Alpha boxes, and haven't noticed anything obvious in the
lseg code causing a problem.

The routines are in src/backend/utils/adt/geo_ops.c, and if you start
putting in printf() statements to narrow down the problem we might have
some luck fixing it. Another possibility is to compile with the flag
"-ieee_noinexact" to suppress floating point traps on underflow and
overflow, but I don't know if that will help.

------------------------------