On Tue, Jan 23, 2018 at 10:48:07PM -0500, David Steele wrote:
> Sorry - it means "level of effort". I was trying to get an idea if it
> is something that could be available in the PG11 development cycle, or
> if I should be looking at other ways to get the testing for this patch
> completed.
I don't think that it would be more than a couple of hours digging
things out, but that may be optimistic. What only needs to be done is
extending get_new_node with an optional argument to define the bin
directory of a PostgresNode and register the path. If the binary
directory is undefined, just rely on PATH and call pg_config --bindir to
get the information, then register it. What we need to be careful at is
that all the binary calls of PostgresNode.pm need to be changed to use
the binary path, which is not really complicated, but it can be easy to
miss some.
There is one small issue with TestLib::check_pg_config but I think that
we can live with the existing version relying on PATH. Once this
refactoring is done, you need the patch set I sent previously here with
some tiny modifications:
https://www.postgresql.org/message-id/CAB7nPqRJXz0sEuUL36eBsF7iZtOQGMJoJPGFWxHLuS6TYPxf5w%40mail.gmail.com
Those modifications involve just looking at the environment variables
specified in pg_upgrade's TESTING and change the node binaries if those
are defined. It is also necessary to tweak probin's paths for the dump
consistency checks, which is something that my last patch set was not
doing.
It is tiring to see this topic come up again and again, so you know
what, let's bite the bullet. I commit myself into coding this thing with
a patch set for the next commit fest. the only thing I want to be sure
about is if folks on this list are fine with the plan of this email.
--
Michael