Bug in regression tests - Mailing list pgsql-patches

postgreSQL 7.0.2 downloaded from the UofSaskatchewan mirror.

RedHat Linux 6.1

-----------
The -L parameter needs to be added to the regress.sh script on the
createlang command.  The script can not be run without this modification.

It appears to me the script was not tested before the distro was shipped.
----------

I HIGHLY recomend the following:  As a software developer with close to 25
years experiance I have encountered the _same_ failing with programmer
after programmer and vendor after vendor.  This is the ommission of a
"Last Step QA test"  Call is an LSQAT if you like.

All this needs to be is a simple shell script that does an installation
and runs the regression tests ... full regression.

You can have level 1, level 2, level 3, etc. if you like.  But you want a
totally brain dead tester that a 6 year old child can run by typing in say
"doitall" or "lsqat" or whatever command you wish.

This script should make zero assumptions about the environment,
configurations, you name it.

So in my case I basically wrote this and if I need to re-install or try to
figure out what I did a year from now - I can go back and look.

----------

This script, COMPLETE with its output for _each_ platform should be
included in an installation examples section with the distro.

Now - I would set up a "clean" machine.  In my case it is RedHat 6.1  I
run VMware... I can create, copy and blow away these machines at will.
VMware is cheap.

Just put 100% of the commands in a file like an old mainframe batch job
and let it rip!!!.  Go have a nap if you like.  When you wake up it should
have worked.  If it does not work - then the installation is flawed and it
will not work for your users either.

Now - this will ENSURE that your users have a workign model to fall back
on.  With the 7.0.2 distro - it is clear that there never was a fully
working version because it is impossible as far as I know to run
createlang without the -L parameter.

----------

Finally, the docs are out of sync with the code base.  This can be
forgiven somewhat.

Final finally:  I'll volunteer to create the LSQAT if a couple people will
give me a bit of a hand.  Since I have only spent one day so far with
PostgreSQL I'm not able to test all the functions.  I'd need basically a
list of things to test.

I think the best way is to run two (2) separate installations.  In the one
case we run a test intallation - nothing is assumed.  We create our own
directories for EVERYTHING with the exception of course of /etc/ld.so.conf
and a few things like that which probably _should_ be left as is.  I'd
call this the gypsey version.  When done the user gets a /pgsql_gypsey
tree that will run...  and can be blown away at will.  This is real good
for development and testing.

The 2nd method would provide a standard default production environment...
and make assumtions that for instance there IS a /usr/local directory that
can be used.

I'd want to write it in bash using redhat 6.1  It can serve as a template
and others can do what they will with it.



So - if this is of interest, let me know.

BTW - it looks like a damn fine product and I'm really glad that Oracle
corp does not have a LSQAT for Oracle 8i for Linux.  Had I been able to
install Oracle I probably wouldn't been writing to you now  :-)

Terrell Larson
Calgary