Re: Postgres 7 on IRIX - Mailing list pgsql-ports

Any chance you can test our current PostgreSQL version and let us know
what needs changing for Irix?  I think it will work better now.

Content-Description: message body text

> Hernandez Alvarez Alfredo writes:
>  >     Thanks. I tried making a link from the cc to
>  > /usr/freeware/bin/gcc...  cant compile... I will try removing the mips
>  > compiler...
>
> That won't help.  Your problem is that configure has decided to use
> the MIPSpro compiler and set the CFLAGS for that.  gcc doesn't grok
> them.
>
> Try making another template file, e.g. template/irix6_gcc.  Merge
> template/irix5 and one or more of the _gcc templates.  Then configure
> with
>
>   ./configure --with-template=irix6_gcc
>
>
> For what it's worth I've just been building PostgreSQL 7.0.2 on IRIX
> 6.5.5m with the native MIPSpro 7.3 compilers.  Both -n32 and -64
> versions have been tried.  -o32 is obsolescent and was not considered.
>
> The -64 version gives 34 warnings about 64-bit pointers being stored
> in 32-bit integers.  That makes whatever is built unsafe IMHO.  How
> well does PostgreSQL build on Alpha and other 64-bit systems?
>
> One worrying problem is that when optimized -O2 both -n32 and -64
> versions failed regression tests such as oidjoins.  An example is
>
>     SELECT    oid, pg_aggregate.aggtransfn1
>     FROM    pg_aggregate
>     WHERE    pg_aggregate.aggtransfn1 != 0 AND
>         NOT EXISTS(SELECT * FROM pg_proc AS t1 WHERE t1.oid =
>       pg_aggregate.aggtransfn1);
>     ERROR:  getattproperties: no attribute tuple 1255 -2
>
> I have not investigated further.  Compiling without optimization seems
> okay.
>
>
> So the current state of affairs is that it is built -n32 -g and passes
> the regression tests.  To do this I've done the following.
>
> 1) Created a new template/irix6 containing
Content-Description: Template for IRIX 6.5

[ Attachment, skipping... ]
Content-Description: message body text

>
> This is picked up by "./configure --with-template=irix6" though
> presumably the configure script could be updated.
>
> Note that the LD entry gets overridden somewhere later.  My kludge for
> linker errors such as
>
>     ld -r -o SUBSYS.o heaptuple.o indextuple.o indexvalid.o printtup.o
>     scankey.o tupdesc.o
>     ld: FATAL   12 : Expecting o32 objects: heaptuple.o is n32.
>
> is to export the environment variable "SGI_ABI=-n32".
>
>
> 2) After running configure, edit config.h to get around the accept bug
>    http://www.postgresql.org/mhonarc/pgsql-ports/2000-05/msg00042.html
>
>
> 3) Edit tests/regress/resultmap to add entries for IRIX.
>
Content-Description: New resultmap entries for IRIX 6

[ Attachment, skipping... ]
Content-Description: message body and .signature

>
> Note that the geometry test still fails.  I've eyeballed the
> differences to be just due to FP precision but it needs to be checked
> thoroughly.
>
> (I will copy item (3) to pgsql-patches.)
>
> --
> Pete Forman                 -./\.- Disclaimer: This post is originated
> Western Geophysical           -./\.-  by myself and does not represent
> pete.forman@westgeo.com         -./\.-  the opinion of Baker Hughes or
> http://www.crosswinds.net/~petef  -./\.-  its divisions.


--
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 853-3000
  +  If your life is a hard drive,     |  830 Blythe Avenue
  +  Christ can be your backup.        |  Drexel Hill, Pennsylvania 19026