[PATCH 1/3] travis: test against postgresql 10

[Daniel Black]

On Fri, 13 Jul 2018 18:31:09 +0100
Stephen Finucane <stephen at that.guru> wrote:

> On Tue, 2018-07-03 at 16:37 +1000, Daniel Black wrote:
> > Signed-off-by: Daniel Black <daniel at linux.ibm.com>
>
> [Forgive me if this comes across rather silly - I'm no Travis expert]
> 
> I initially read this and thought...aren't we already doing this?
> Looking into it more, it seems we're not quite. What we are doing is
> running a test matrix of supported Python (and Python package)
> versions (currently Python 2.7, 3.4, 3.5, and 3.6) against the
> supported database backends (currently MariaDB/MySQL  and
> PostgreSQL). That gives us 8 variations.

> This seems to be adding
> three more - one using MariaDB 10 and Python 2.7, one using
> PostgreSQL 9.6 and Python 3.6 and one using PostgreSQL 10 and Python
> 3.6. Is that all correct so far?

yes.

> 
> If so, I have the following comments/questions:
> 
>  * The addtional PostgreSQL 9.6 environment seems unnecessary, given
>    that we're already testing this with multiple Python versions

Quite right. I'll remove or replace with a PG 11 release (currently at
beta2).

>  * Is there any real reason we need to keep testing the older versions
>    of the databases?

I'm assuming there is a lower version of database that you want to
maintain compatibility with. Or at least knowing if a break in
compatibility is made so it can be noted in release notes.

>  * If so, is there any reason we need to run multiple the tests for
>    multiple Python versions against both PostgreSQL and MariaDB?
>    Couldn't we just run them against e.g. PostgreSQL 10 and then
> simply run the Python 3.6 tests against the other three possible
> backends?

In practice its making sure the underlying libraries of django,
psycopg2 and mysqlclient are doing the right thing for each of the
server versions. Sure, this is mostly testing more than just patchwork.
I think it puts the development in a good place to start using newer
CTEs queries depending on database server version.

I think all combinations fall within the scope of a supported system so
may as well incur the few extra minutes of testing for the benefit of
users.