Cygwin PostgreSQL 7.4.1 Regression Test Issue

I was about to release Cygwin PostgreSQL 7.4.1-1. Unfortunately, I got
the attached regression test errors on the rules test. The diffs are
the following:

1. two missing "-" characters which seems to imply a field width
difference
2. occurrences of NULL::anyarray instead of NULL::"unknown"

Does anyone know why the above regressions are occurring? Note Cygwin
PostgreSQL 7.4-1 did *not* have any regressions (including rules).

Is it safe to release Cygwin PostgreSQL 7.4.1-1 with this issue?

Thanks,
Jason

--
PGP/GPG Key: http://www.tishler.net/jason/pubkey.asc or key servers
Fingerprint: 7A73 1405 7F2B E669 C19D 8784 1AFD E4CC ECF4 8EF6

Jason Tishler wrote:
> I was about to release Cygwin PostgreSQL 7.4.1-1. Unfortunately, I
> got the attached regression test errors on the rules test. The diffs
> are the following:
>
> 1. two missing "-" characters which seems to imply a field width
> difference
> 2. occurrences of NULL::anyarray instead of NULL::"unknown"

No, that shouldn't happen, especially since these issues are not
platform-dependent. If you can reproduce it, please send a bug report.

Peter,

On Sun, Jan 04, 2004 at 07:25:39PM +0100, Peter Eisentraut wrote:
> Jason Tishler wrote:
> > I was about to release Cygwin PostgreSQL 7.4.1-1. Unfortunately, I
> > got the attached regression test errors on the rules test. The
> > diffs are the following:
> >
> > 1. two missing "-" characters which seems to imply a field width
> > difference
> > 2. occurrences of NULL::anyarray instead of NULL::"unknown"
>
> No, that shouldn't happen, especially since these issues are not
> platform-dependent. If you can reproduce it,

Yes, I can reproduce it under PostgreSQL 7.4.1 and Cygwin 1.5.5-1 every
time. Note that I *cannot* reproduce it under PostgreSQL 7.4 and Cygwin
1.5.5-1. Of course, I cannot reproduce it under PostgreSQL 7.4.1 and
Red Hat Linux 8.0.

BTW, in order to help debug this faster, what are the minimal tests that
rules is dependent on? My current approach is to just run all 76 tests
listed before rules in serial_schedule. Unfortunately, this approach
take over 3 minutes per try. :,(

> please send a bug report.

OK, but isn't that like throwing a boomerang? :,)

Jason

--
PGP/GPG Key: http://www.tishler.net/jason/pubkey.asc or key servers
Fingerprint: 7A73 1405 7F2B E669 C19D 8784 1AFD E4CC ECF4 8EF6

Jason Tishler wrote:
> Yes, I can reproduce it under PostgreSQL 7.4.1 and Cygwin 1.5.5-1
> every time. Note that I *cannot* reproduce it under PostgreSQL 7.4
> and Cygwin 1.5.5-1. Of course, I cannot reproduce it under
> PostgreSQL 7.4.1 and Red Hat Linux 8.0.

I just tried out the 7.4 branch after a full Cygwin update (stable
packages), and all tests passed in each of several attempts. (Even
"make check" works now -- whoohoo!)

> BTW, in order to help debug this faster, what are the minimal tests
> that rules is dependent on? My current approach is to just run all
> 76 tests listed before rules in serial_schedule. Unfortunately, this
> approach take over 3 minutes per try. :,(

I managed to get the following to work:

./pg_regress create_table create_misc create_function_1 create_type
create_function_2 create_operator create_view rules

The create_table test will fail in this case, but the others should
pass.

Peter,

On Wed, Jan 07, 2004 at 12:23:12AM +0100, Peter Eisentraut wrote:
> Jason Tishler wrote:
> > Yes, I can reproduce it under PostgreSQL 7.4.1 and Cygwin 1.5.5-1
> > every time. Note that I *cannot* reproduce it under PostgreSQL 7.4
> > and Cygwin 1.5.5-1. Of course, I cannot reproduce it under
> > PostgreSQL 7.4.1 and Red Hat Linux 8.0.
>
> I just tried out the 7.4 branch after a full Cygwin update (stable
> packages), and all tests passed in each of several attempts. (Even
> "make check" works now -- whoohoo!)

7.4 works for me too. Please repeat the above with 7.4.1 and report
back to the list.

Thanks,
Jason

--
PGP/GPG Key: http://www.tishler.net/jason/pubkey.asc or key servers
Fingerprint: 7A73 1405 7F2B E669 C19D 8784 1AFD E4CC ECF4 8EF6

Jason Tishler wrote:
> > I just tried out the 7.4 branch after a full Cygwin update (stable
> > packages), and all tests passed in each of several attempts. (Even
> > "make check" works now -- whoohoo!)
>
> 7.4 works for me too. Please repeat the above with 7.4.1 and report
> back to the list.

Right now, the 7.4 branch *is* 7.4.1 (plus additional patches). I'll
try to get the exact 7.4.1 release tomorrow, but it would be highly
unusual if that made a difference. (Of course, the behavior you report
is also highly unusual, so yeah, I'll try it. :-) )

Peter,

On Wed, Jan 07, 2004 at 12:23:12AM +0100, Peter Eisentraut wrote:
> Jason Tishler wrote:
> (Even "make check" works now -- whoohoo!)

Sigh... Now I'm really confused!

The rules test passes under make check:

$ make MAX_CONNECTIONS=5 check
[snip]
rules ... ok
[snip]

but fails under make installcheck and your minimal test case:

> > BTW, in order to help debug this faster, what are the minimal tests
> > that rules is dependent on? My current approach is to just run all
> > 76 tests listed before rules in serial_schedule. Unfortunately, this
> > approach take over 3 minutes per try. :,(
>
> I managed to get the following to work:
>
> ./pg_regress create_table create_misc create_function_1 create_type
> create_function_2 create_operator create_view rules
>
> The create_table test will fail in this case, but the others should
> pass.

$ ./pg_regress create_table create_misc create_function_1 ... rules
[snip]
test rules ... FAILED
[snip]

Any ideas?

Thanks,
Jason

--
PGP/GPG Key: http://www.tishler.net/jason/pubkey.asc or key servers
Fingerprint: 7A73 1405 7F2B E669 C19D 8784 1AFD E4CC ECF4 8EF6

Peter,

On Wed, Jan 07, 2004 at 01:41:53AM +0100, Peter Eisentraut wrote:
> Jason Tishler wrote:
> > > I just tried out the 7.4 branch after a full Cygwin update (stable
> > > packages), and all tests passed in each of several attempts.
> > > (Even "make check" works now -- whoohoo!)
> >
> > 7.4 works for me too. Please repeat the above with 7.4.1 and report
> > back to the list.
>
> Right now, the 7.4 branch *is* 7.4.1 (plus additional patches).

Doh!

> I'll try to get the exact 7.4.1 release tomorrow, but it would be
> highly unusual if that made a difference. (Of course, the behavior
> you report is also highly unusual, so yeah, I'll try it. :-) )

Thanks. Or, you can try my package instead...

Jason

--
PGP/GPG Key: http://www.tishler.net/jason/pubkey.asc or key servers
Fingerprint: 7A73 1405 7F2B E669 C19D 8784 1AFD E4CC ECF4 8EF6

Jason Tishler wrote:
> Sigh... Now I'm really confused!
>
> The rules test passes under make check:
>
> $ make MAX_CONNECTIONS=5 check
> [snip]
> rules ... ok
> [snip]

It seems that the "make installcheck" case is actually hitting the wrong
server, namely a 7.4[.0] server. The rules test has indeed been
changed in 7.4.1 with exactly the diff that you are seeing in your
failed tests.

http://developer.postgresql.org/cvsweb.cgi/pgsql-server/src/test/regress/expected/rules.out

You should probably check your various path settings. It might be the
case that the wrong libpq gets used.

Peter,

On Wed, Jan 07, 2004 at 07:14:54PM +0100, Peter Eisentraut wrote:
> Jason Tishler wrote:
> > Sigh... Now I'm really confused!
> >
> > The rules test passes under make check:
> >
> > $ make MAX_CONNECTIONS=5 check
> > [snip]
> > rules ... ok
> > [snip]
>
> It seems that the "make installcheck" case is actually hitting the
> wrong server, namely a 7.4[.0] server. The rules test has indeed been
> changed in 7.4.1 with exactly the diff that you are seeing in your
> failed tests.
>
> http://developer.postgresql.org/cvsweb.cgi/pgsql-server/src/test/regress/expected/rules.out
>
> You should probably check your various path settings. It might be the
> case that the wrong libpq gets used.

AFAICT, I was bitten by the following:

This version, as with most minor versions, does not require a
dump/reload to put into place. That said, there are some changes to
the information schema that require a DROP/reload of that *specific*
schema ... please read the HISTORY file for instruction on how to do
this.

After rerunning initdb, the rules regression test now passes.

Thanks,
Jason

--
PGP/GPG Key: http://www.tishler.net/jason/pubkey.asc or key servers
Fingerprint: 7A73 1405 7F2B E669 C19D 8784 1AFD E4CC ECF4 8EF6