Re: Building 7.1 under Linux for S/390

From: Peter Eisentraut <peter_e(at)gmx(dot)net>
To: "Ferguson, Neale" <NEALE(dot)FERGUSON(at)ca(dot)com>
Cc: Lamar Owen <lamar(dot)owen(at)wgcr(dot)org>, pgsql-ports(at)postgresql(dot)org
Subject: Re: Building 7.1 under Linux for S/390
Date: 2000-07-01 21:24:04
Message-ID: Pine.LNX.4.21.0007011953210.13037-100000@localhost.localdomain
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-ports

Ferguson, Neale writes:

> Just built the v7.1.0 snapshot I downloaded yesterday. No problems
> other than building the perl stuff. The Makefile uses CC=cc whereas
> all the others pick up CC=gcc.

The Perl stuff uses the ExtUtils::MakeMaker module to build. This has
worked reasonably well so far for the interface module (Pg.pm), but the
PL/Perl build doesn't seem to work for anyone except for some of the
latest Linux distributions (which doesn't include me). Apparently,
MakeMaker causes things to be build with the same compiler and the same
switches that Perl was originally build with, which is a provably flawed
approach but nothing we can do much about.

> Attached are the updated patches,

You should send patches to pgsql-patches(at)postgresql(dot)org so that everyone
can see them. Discussions to pgsql-ports(at)postgresql(dot)org for that matter.

> the regress.out file and the regression.diffs. The failures seem to be
> in floating point related tests owing to my level of hardware
> requiring software emulation of some of the fp operations.

Geometry tests mixing up +0 and -0 and the nth decimal place are common,
although mixing up +-0 indicates a failure to comply to IEEE 754.

But the float8 test worries me: 1.2345678901234e+200 ^ 1e200 should
certainly not give -3.09434604738258e-17 on any platform. Seems like
someone ought to enable the overflow trap.

You can also provide your own platform specific expected results. This is
documented somewhere, I forgot where. Look at src/test/regress/resultmap
for a start.

> In addition, I've not changed config.guess exactly what configure is
> expecting (it returns s390-ibm-linux) so that it doesn't automatically
> find my linux_s390 template.

You should send changes to config.{guess,sub} to config-patches(at)gnu(dot)org(dot)
We'd rather not maintain our own copy of these. :) If config.guess fails
to recognize the platform then it should still work to use `./configure
--host=s390-ibm-linux' and the template should be picked up via the
src/templates/.similar file.

If you get stuck somewhere, just repost the patches and I'll be glad to
help.

--
Peter Eisentraut Sernanders väg 10:115
peter_e(at)gmx(dot)net 75262 Uppsala
http://yi.org/peter-e/ Sweden

Browse pgsql-ports by date

  From Date Subject
Next Message Peter Eisentraut 2000-07-01 21:24:20 Linux templates
Previous Message Bruce Momjian 2000-07-01 21:17:26 Re: Linux templates