Re: multi-install PostgresNode fails with older postgres versions

On Wed, 7 Apr 2021 11:54:36 -0400
Andrew Dunstan <andrew@dunslane.net> wrote:

> On 4/7/21 10:37 AM, Jehan-Guillaume de Rorthais wrote:
> > Hi all,
> >
> > First, sorry to step in this discussion this late. I didn't noticed it
> > before :(
> >
> > I did some work about these compatibility issues in late 2020 to use
> > PostgresNode in the check_pgactivity TAP tests.
> >
> > See https://github.com/ioguix/check_pgactivity/tree/tests/t/lib
> >
> > PostgresNode.pm, TestLib.pm, SimpleTee.pm and RecursiveCopy.pm comes
> > unchanged from PostgreSQL source file (see headers and COPYRIGHT.pgsql).
> >
> > Then, I'm using the facet class PostgresNodeFacet to extend it with some
> > more methods. Finaly, I created one class per majpr version, each
> > inheriting from the next version. That means 13 inherits from
> > PostgresNodeFacet.pm, 12 inherits from 13, 11 inherits from 12, etc.
> >
> > When I'm creating a new node, I'm using the "pgaTester" factory class. It
> > relies on PATH to check the major version using pg_config, then loads the
> > appropriate class.
> >
> > That means some class overrides almost no methods but version(), which
> > returns the major version. Eg.:
> > https://github.com/ioguix/check_pgactivity/blob/tests/t/lib/PostgresNode12.pm
> >
> > From tests, I can check the node version using this method, eg.:
> >
> >   skip "skip non-compatible test on PostgreSQL 8.0 and before", 3
> >     unless $node->version <= 8.0;
> >
> > Of course, there's a lot of duplicate code between classes, but my main goal
> > was to keep PostgresNode.pm unchanged from upstream so I can easily update
> > it.
> >
> > And here is a demo test file:
> > https://github.com/ioguix/check_pgactivity/blob/tests/t/01-streaming_delta.t
> >
> > My limited set of tests are working with versions back to 9.0 so far.
> >
> > My 2¢  
> 
> 
> 
> I don't really want to create a multitude of classes. I think Mark is
> basically on the right track. I started off using a subclass of
> PostgresNode but was persuaded not to go down that route, and instead we
> have made some fairly minimal changes to PostgresNode itself. I think
> that was a good decision. If you take out the versioning subroutines,
> the actual version-aware changes Mark is proposing to PostgresNode are
> quite small - less that 200 lines supporting versions all the way back
> to 8.1. That's pretty awesome.


I took this road because as soon as you want to use some other methods like
enable_streaming, enable_archiving, etc, you find much more incompatibilities
on your way. My current stack of classes is backward compatible with much
more methods than just init(). But I admit it creates a multitude of class and
some duplicate code...

It's still possible to patch each methods in PostgresNode, but you'll end up
with a forest of conditional blocks depending on how far you want to go with old
PostgreSQL versions.

Regards,