float8 regression test failure in head

Christopher Kings-Lynne <chriskl@familyhealth.com.au> writes:

Attached are the test failures I'm currently getting.

It looks like Neil didn't update expected/float8-small-is-zero.out
for his recent changes (for which, shame on him). Would you get
together to verify the correct regression outputs for your platform
and commit the updated expected file?

regards, tom lane

This has not yet been fixed...

Chris

Tom Lane wrote:

Christopher Kings-Lynne wrote:

This has not yet been fixed...

Attached is a patch for this issue. I've checked with Chris, and this
patch allows the regression tests to pass on his machine. I also
updated float8-exp-three-digits expected file, but I wasn't able to
test these changes (I don't have access to a QNX machine).

I intend to apply this patch tomorrow.

Cheers,

Neil

Neil Conway <neilc@samurai.com> writes:

Attached is a patch for this issue. I've checked with Chris, and this
patch allows the regression tests to pass on his machine. I also
updated float8-exp-three-digits expected file, but I wasn't able to
test these changes (I don't have access to a QNX machine).

FWIW, my procedure for handling updates in regression tests that have
multiple expected variants is to apply the same diff that occurs on the
machines I can test to the variants I can't test. It's possible that
this will be wrong, but 99% of the time it's right and it saves followup
work.

An easy way to apply such changes is to feed a default-format diff
output to patch(1):

diff test.myplatform.orig test.myplatform | patch

patch then bleats about not knowing which file to patch, and you
tell it "test.otherplatform". Works every time ...

regards, tom lane

On 22-Mar-04, at 11:03 PM, Neil Conway wrote:

Attached is a patch for this issue.

Patch applied to CVS HEAD.

-Neil

On Mon, Mar 22, 2004 at 11:03:38PM -0500, Neil Conway wrote:

Christopher Kings-Lynne wrote:

This has not yet been fixed...

Attached is a patch for this issue. I've checked with Chris, and this
patch allows the regression tests to pass on his machine. I also
updated float8-exp-three-digits expected file, but I wasn't able to
test these changes (I don't have access to a QNX machine).

Mystery: why I haven't I seen this problem before in the last 2 years...
Anyway, on NetBSD-3.99.17/i386, I need the included patch to pass float8.
float8-small-is-zero_1.out probably needs it too, float8.out and
float8-exp-three-digits-win32.out already look OK.

Cheers,

Patrick

Patrick Welche wrote:

On Mon, Mar 22, 2004 at 11:03:38PM -0500, Neil Conway wrote:

Christopher Kings-Lynne wrote:

This has not yet been fixed...

Attached is a patch for this issue. I've checked with Chris, and this
patch allows the regression tests to pass on his machine. I also
updated float8-exp-three-digits expected file, but I wasn't able to
test these changes (I don't have access to a QNX machine).

Mystery: why I haven't I seen this problem before in the last 2 years...
Anyway, on NetBSD-3.99.17/i386, I need the included patch to pass float8.
float8-small-is-zero_1.out probably needs it too, float8.out and
float8-exp-three-digits-win32.out already look OK.

You realize that all these platforms use the same file:

float8/i.86-.*-freebsd[234]=float8-small-is-zero
float8/i.86-.*-openbsd=float8-small-is-zero
float8/i.86-.*-netbsd=float8-small-is-zero
float8/m68k-.*-netbsd=float8-small-is-zero

Are we going to have to make a NetBSD/i386-specific file to fix this?
Don't we have some of these platforms on the build farm. Are they
failing?

--
Bruce Momjian http://candle.pha.pa.us
EnterpriseDB http://www.enterprisedb.com

+ If your life is a hard drive, Christ can be your backup. +

Bruce Momjian <pgman@candle.pha.pa.us> writes:

Don't we have some of these platforms on the build farm. Are they
failing?

canary, gazelle, and osprey all pass this test just fine. Before
accepting any patches in this area we'd better find out the difference
between the OP's netbsd system and those machines.

regards, tom lane

Tom Lane wrote:

Bruce Momjian <pgman@candle.pha.pa.us> writes:

Don't we have some of these platforms on the build farm. Are they
failing?

canary, gazelle, and osprey all pass this test just fine. Before
accepting any patches in this area we'd better find out the difference
between the OP's netbsd system and those machines.

Agreed.

--
Bruce Momjian http://candle.pha.pa.us
EnterpriseDB http://www.enterprisedb.com

+ If your life is a hard drive, Christ can be your backup. +

On Thu, Apr 20, 2006 at 11:56:29AM -0400, Bruce Momjian wrote:

Tom Lane wrote:

Bruce Momjian <pgman@candle.pha.pa.us> writes:

Don't we have some of these platforms on the build farm. Are they
failing?

canary, gazelle, and osprey all pass this test just fine. Before
accepting any patches in this area we'd better find out the difference
between the OP's netbsd system and those machines.

Agreed.

Quoting from my original post "NetBSD-3.99.17/i386" - so what are you
running?

.. and just to check:

patrimoine=# select '1e-400'::float;
ERROR: "1e-400" is out of range for type double precision
patrimoine=# select version();
version
-------------------------------------------------------------------------------------------------------------
PostgreSQL 8.2devel on i386-unknown-netbsdelf3.99.17, compiled by GCC gcc (GCC) 3.3.3 (NetBSD nb3 20040520)
(1 row)

1e-400 wasn't rounded down to 0.

Cheers,

Patrick

Patrick Welche <prlw1@newn.cam.ac.uk> writes:

canary, gazelle, and osprey all pass this test just fine. Before
accepting any patches in this area we'd better find out the difference
between the OP's netbsd system and those machines.

Quoting from my original post "NetBSD-3.99.17/i386" - so what are you
running?

http://www.pgbuildfarm.org/cgi-bin/show_members.pl

The entries in the buildfarm list say things like "NetBSD 3.0". Is 3.99
newer? If so, has its underflow handling changed?

regards, tom lane