Re: v7.2b3 packaged, but not announced beyond here yet ...

On Friday 23 November 2001 11:08 am, Tom Lane wrote:
> Lamar Owen <lamar(dot)owen(at)wgcr(dot)org> writes:
> > By running the regression tests in a binary-only installation.

> Remind me to pay no attention whatsoever to regression test failure
> reports coming from people who use the RPM installation.

I typically field those anyway.

> I regard this setup as worse than useless, because it is guaranteed
> to cause regression failures that most people will not know how to
> interpret. We will be getting lots of complaints, and I for one am
> not going to waste my time scanning them closely to see if there is
> anything real there.

As this has been in the RPMset for, let's see, TWO YEARS now, I don't think
it's going to be that big of a problem. The locale deal has been there that
long -- and is something I can see a mile away -- which of course means that
I'll check those reports myself, and only will pass along the parts that are
real failures.

The discussion in a separate thread about being able to specify the collation
in queries sounds like it would solve a good deal of the problem, by
specifying the C collation in the regression queries. The other failures
involve a currency symbol, the presence of which still confuses me to a
certain extent.

> They are, when used as intended.

And it is my contention that the 'make check' method of running regression
testing is broken in itself, by assuming a source tree, by assuming a
development platform, and by assuming that regression testing is a
developer-only activity. I disagree with all those assumptions -- our
documentation even has stated (and may still state) that the regression
database and the regression queries are a good source of examples -- both for
the queries, and for the datamodels.

But those three assumptions run deep -- particularly the first two, which
underlie the whole package's philosophy in more than one area.

I'm just providing what I believe is a useful feature that has been requested
by RPM, non-development-platform, users.

If I must, I can patch the expected outputs to match the default en_US locale
-- but I am very loathe to do that! After all, the en_US locale may not be
what the user is running.

> In reality, it's extremely questionable that the regression tests
> will tell anything useful to a person who's installing prebuilt
> platform-specific binaries.

In reality, I think they are more useful than that to the enduser.

> The builder of the binaries should have
> run the regress tests.

I run them both in their 'intended' way as well as in the final prebuilt way
before releasing any RPMsets. There have been a couple of times that I
skipped that step, usually with bad results, but it is one of the things I
just do, now.

> So I'm not sure what the point is --- but I am
> sure that this setup will cause a lot more problems than it solves.

As the test subpackage has existed for two years, I would contend that the
package hasn't caused quite as many problems as you prophesy.

> I recommend forgetting about postgresql-test.

If the consensus of the steering committee is that we shouldn't distribute
the postgresql-test subpackage prebuilt, from ftp.postgresql.org, I will
disable it in the build. The possibility to build it will still be
available: it would just not be built by default.

Again, I have fielded questions about RPM issues for over two years, now, and
I fully intend to continue doing so. The standard 'failures' will be fully
documented in the README.rpm-dist (the fact that there are known failures is
already documented) in the RPMset, which is where someone who wants to run
the tests is going to have to go for information on running the tests
themselves. So, Tom, while I understand your concerns, I just want to make
sure it is realized that this is not a new thing, and that I volunteer to
field those results.
--
Lamar Owen
WGCR Internet Radio
1 Peter 4:11