spoonbill is failing citext test

On Oct 14, 2008, at 12:36, Alvaro Herrera wrote:

I don't know if you have noticed, but the spoonbill buildfarm member
is
failing the citext test. Unfortunately the regression diff is not
very
helpful:

Binary files /home/pgbuild/pgbuildfarm/HEAD/pgsql.2397/contrib/
citext/expected/citext.out and /home/pgbuild/pgbuildfarm/HEAD/pgsql.
2397/contrib/citext/results/citext.out differ

http://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=spoonbill&dt=2008-10-14%20100509&stg=contrib-install-check

Maybe we should pass -a to diff so that it displays the difference
even
if it thinks the file is binary.

What does the diff look like now?

Thanks,

David

Alvaro Herrera <alvherre@commandprompt.com> writes:

Maybe we should pass -a to diff so that it displays the difference even
if it thinks the file is binary.

Unfortunately not portable:

$ /usr/bin/diff -a foo bar
/usr/bin/diff: illegal option -- a
usage: diff [ -C n ] [ -S name ] [ -bcefhilnrstw ] dir1 dir2
diff [-C n ] [ -bcefhintw ] file1 file2
diff [ -D string ] [ -biw ] file1 file2

Perhaps Stefan can run the test manually and get us a readable diff.

regards, tom lane

* Tom Lane <tgl@sss.pgh.pa.us> [081014 16:08]:

Alvaro Herrera <alvherre@commandprompt.com> writes:

Maybe we should pass -a to diff so that it displays the difference even
if it thinks the file is binary.

Unfortunately not portable:

$ /usr/bin/diff -a foo bar
/usr/bin/diff: illegal option -- a
usage: diff [ -C n ] [ -S name ] [ -bcefhilnrstw ] dir1 dir2
diff [-C n ] [ -bcefhintw ] file1 file2
diff [ -D string ] [ -biw ] file1 file2

I'll bet /usr/bin/make pukes on PostgreSQL make files there too ;-)

a.

--
Aidan Van Dyk Create like a god,
aidan@highrise.ca command like a king,
http://www.highrise.ca/ work like a slave.

Aidan Van Dyk wrote:

* Tom Lane <tgl@sss.pgh.pa.us> [081014 16:08]:

Alvaro Herrera <alvherre@commandprompt.com> writes:

Maybe we should pass -a to diff so that it displays the difference even
if it thinks the file is binary.

Unfortunately not portable:

$ /usr/bin/diff -a foo bar
/usr/bin/diff: illegal option -- a
usage: diff [ -C n ] [ -S name ] [ -bcefhilnrstw ] dir1 dir2
diff [-C n ] [ -bcefhintw ] file1 file2
diff [ -D string ] [ -biw ] file1 file2

I'll bet /usr/bin/make pukes on PostgreSQL make files there too ;-)

So what? We have an explicit dependency on GNU make, but not on GNU
diffutils. We try quite hard to keep the list of dependencies down.

cheers

andrew

Tom Lane wrote:

Alvaro Herrera <alvherre@commandprompt.com> writes:

Maybe we should pass -a to diff so that it displays the difference even
if it thinks the file is binary.

Unfortunately not portable:

$ /usr/bin/diff -a foo bar
/usr/bin/diff: illegal option -- a
usage: diff [ -C n ] [ -S name ] [ -bcefhilnrstw ] dir1 dir2
diff [-C n ] [ -bcefhintw ] file1 file2
diff [ -D string ] [ -biw ] file1 file2

Perhaps Stefan can run the test manually and get us a readable diff.

well what we are looking at here are actually two issues - one is the
regression failure - diff -a produces:

http://www.kaltenbrunner.cc/files/citext.diff.txt

the other problem is that each of those failures leaves spoonbill with a
running postgresql instance that causes a follow up failure with the
next buildfarm run because there is not enough shared memory left for
another instance ...

Stefan

On Oct 14, 2008, at 14:00, Stefan Kaltenbrunner wrote:

well what we are looking at here are actually two issues - one is
the regression failure - diff -a produces:

http://www.kaltenbrunner.cc/files/citext.diff.txt

Am I reading that right? Is there really just an issue of a different
number of spaces before the caret in that error message? Seems kinda
weird…

Best,

David

Stefan Kaltenbrunner wrote:

Tom Lane wrote:

Alvaro Herrera <alvherre@commandprompt.com> writes:

Maybe we should pass -a to diff so that it displays the difference even
if it thinks the file is binary.

Unfortunately not portable:

$ /usr/bin/diff -a foo bar
/usr/bin/diff: illegal option -- a
usage: diff [ -C n ] [ -S name ] [ -bcefhilnrstw ] dir1 dir2
diff [-C n ] [ -bcefhintw ] file1 file2
diff [ -D string ] [ -biw ] file1 file2

Perhaps Stefan can run the test manually and get us a readable diff.

well what we are looking at here are actually two issues - one is the
regression failure - diff -a produces:

http://www.kaltenbrunner.cc/files/citext.diff.txt

the other problem is that each of those failures leaves spoonbill with
a running postgresql instance that causes a follow up failure with
the next buildfarm run because there is not enough shared memory left
for another instance ...

Why should it leave a running postgresql instance?

cheers

andrew

David E. Wheeler wrote:

On Oct 14, 2008, at 14:00, Stefan Kaltenbrunner wrote:

well what we are looking at here are actually two issues - one is the
regression failure - diff -a produces:

http://www.kaltenbrunner.cc/files/citext.diff.txt

Am I reading that right? Is there really just an issue of a different
number of spaces before the caret in that error message? Seems kinda weird�

well it seems to report the error location different from what
regression output says it should (though I would argue that the
regression output actually looks more correct/usable than the expected
output).

Stefan

Stefan Kaltenbrunner <stefan@kaltenbrunner.cc> writes:

Tom Lane wrote:

Perhaps Stefan can run the test manually and get us a readable diff.

well what we are looking at here are actually two issues - one is the
regression failure - diff -a produces:
http://www.kaltenbrunner.cc/files/citext.diff.txt

Huh. This is blowback from the change I made recently to make "cannot
cast to insert target type" errors point at the source expression
instead of the target column name. Easily fixed, but I wonder why
spoonbill was complaining? Apparently it thought the output file was
not textual (thus defeating the --ignore-spaces option), but why?

the other problem is that each of those failures leaves spoonbill with a
running postgresql instance that causes a follow up failure with the
next buildfarm run

I think this is a known buildfarm-script problem --- it forgets to shut
down the installed postmaster if there's a failure in this step.

regards, tom lane

Tom Lane wrote:

the other problem is that each of those failures leaves spoonbill with a
running postgresql instance that causes a follow up failure with the
next buildfarm run

I think this is a known buildfarm-script problem --- it forgets to shut
down the installed postmaster if there's a failure in this step.

That bug has been fixed (in Script version 1.93 / Release 3.1, released
back in June). Spoonbill is unfortunately running an older version.

cheers

andrew

Andrew Dunstan wrote:

Tom Lane wrote:

the other problem is that each of those failures leaves spoonbill
with a running postgresql instance that causes a follow up failure
with the next buildfarm run

I think this is a known buildfarm-script problem --- it forgets to shut
down the installed postmaster if there's a failure in this step.

That bug has been fixed (in Script version 1.93 / Release 3.1, released
back in June). Spoonbill is unfortunately running an older version.

*grmbl* 3.0 in fact - will update ...

Stefan