Alvaro,
* Alvaro Herrera (alvherre@2ndquadrant.com) wrote:
> Stephen Frost wrote:
> > I don't think we've made use of it either. If the targets/code are
> > already there to make it happen and it's just a matter of having a
> > system running which is generating the website then I can probably get
> > that going. I was under the impression that there was more to it than
> > that though.
>
> "configure --enable-coverage"; install the resulting tree; then run
> whatever tests you want, then "make coverage". That's enough to get the
> HTML reports.
Ok, cool, I'll take a look at that.
> > > We had someone else trying to submit patches to improve coverage of the
> > > regression tests, but (probably due to wrong stars alignment) they
> > > started with CREATE DATABASE which made the tests a lot slower, which
> > > got the patches turned down -- the submitter disappeared after that
> > > IIRC, probably discouraged by the lack of results.
> >
> > Agreed, and I think that's unfortunate. It's an area which we could
> > really improve in and would be a good place for someone new to the
> > community to be able to contribute- but we need to provide the right way
> > for those tests to be added and that way isn't to include them in the
> > main suite of tests which are run during development.
>
> Well, I don't disagree that we could do with some tests that are run
> additionally to the ones we currently have, but some basic stuff that
> takes almost no time to run would be adequate to have in the basic
> regression tests. The one I'm thinking is something like generate a
> VALUES of all the supported object types, then running
> "pg_get_object_address" on them to make sure we support all object types
> (or at least that we're aware which object types are not supported.)
Agreed, that sounds perfectly reasonable to me. I didn't mean to imply
that we shouldn't add tests where they make sense, including to the core
regression suites (particularly coverage tests like that one you're
suggesting here), just pointing out that we need a way to address code
coverage based tested also which is done outside of the main regression
suite.
> For instance, Petr Jelinek's patch to add the TABLESAMPLE clause adds a
> new object type which needs support in get_object_address, but I'm not
> sure it's added to the stuff in the object_address test. It's a very
> easy omission to make.
Agreed.
Thanks!
Stephen