Thread: 7.1.3 with source NAMEDATALEN change fails name regression
Tom,
The installation on my Solaris 2.6 box includes one change to source code that seems to make Postgresql fail the name regression test.
The change is for a line in file /src/include/postgres_ext.h
from #define NAMEDATALEN 32
to #define NAMEDATALEN 64
(A php application that I'm planning to run needs this change.)
When I initially got Postgres to configure and compile I was operating on a cleanly unpacked archive on which I had forgotten to make this change. This initial installation passed all regression tests.
I then remembered the needed source code edit, wiped the installation and started afresh. The build went cleanly this time but the name regression test failed. This isn't completely out of left field given the change, but I need to know
a) if it is a significant failure and
b) if it should fail. If not, is this a bug in the source or the test?
I have posted regression.diffs to http://pgsetup.awarehouse.com/regression.11-26-01.diffs.htm
and regression.out to http://pgsetup.awarehouse.com/regression.11-26-01.out.htm
It looks like my build is returning some longer strings than the expect result. This could be OK if the test depends on the standard value for NAMEDATALEN, and the larger value in my build causes longer strings to be returned. I also noticed that the 'f1' result text is typically three characters to the right of its position in the expected text.
A quick read of the src/test/regress/README file doesn't shed light on this issue, nor does a search of the pgsql-general archives.
Actually, the more I look at the diffs file, the more this looks like a data validation-like failure appropriate to the source code edit. In other words, harmless. Confirmation or any other info would be much appreciated.
Thanks for any help,
David De Graff
"David De Graff" <postgresql@awarehouse.com> writes:
> I then remembered the needed source code edit, wiped the installation and s=
> tarted afresh. The build went cleanly this time but the name regression tes=
> t failed. This isn't completely out of left field given the change, but I n=
> eed to know
> a) if it is a significant failure and
> b) if it should fail. If not, is this a bug in the source or the test?
I'd expect that; the regression test tests truncation of names, and
naturally it's set up for 32-byte names. You should be able to validate
the results by eyeball.
regards, tom lane