* Alvaro Herrera (alvherre@2ndquadrant.com) wrote:
> Actually, there was a bug in the changes of the rule for ALTER EXTENSION
> ADD OPERATOR CLASS. I noticed by chance only, and upon testing it
> manually I realized I had made a mistake. I then remembered I made the
> same bug previously, fixed by 5c5ffee80f35, and I'm not wondering why do
> we not have any test for ALTER EXTENSION ADD other than pg_upgrading
> some database that contains an extension which uses each command. This
> seems pretty dangerous to me, generally speaking ... we should
> definitely be testing all these ALTER EXTENSION commands.
It'd certainly be great to have more testing in general, but we're not
going to be able to include complete code coverage tests in the normal
set of regression tests which are run.. What I've been thinking for a
while (probably influenced by other discussion) is that we should have
the buildfarm running tests for code coverage as those are async from
the development process. That isn't to say we shouldn't add more tests
to the main regression suite when it makes sense to, we certainly
should, but we really need to be looking at code coverage tools and
adding tests to improve our test coverage which can be run by the
buildfarm animals (or even just a subset of them, if we feel that having
all the animals running them would be excessive).
Thanks!
Stephen
