Re: Regression tests versus the buildfarm environment

On 08/12/2010 10:22 AM, Tom Lane wrote:
> Peter Eisentraut<peter_e(at)gmx(dot)net> writes:
>> On ons, 2010-08-11 at 16:17 -0400, Tom Lane wrote:
>>> We should have the buildfarm configuration such that any one run uses
>>> the same port number for both installed and uninstalled regression
>>> tests.
>> I'm getting lost here what the actual requirements are. The above is
>> obviously not going to work as a default for pg_regress, because the
>> port number for an installed test is determined by the user and is
>> likely to be in the public range, whereas the uninstalled test should
>> use something from the private range.
> Certainly, but the buildfarm's "installed" test doesn't try to start on
> 5432. It starts on whatever branch_port the buildfarm owner has
> specified for that animal and that branch. It's the owner's
> responsibility to make that nonconflicting across the services and
> buildfarm critters he has running on a given machine. What I'm saying
> is that *in the buildfarm* we want the "make check" case to also use
> this predetermined safe port number. That has nothing whatever to do
> with what is good practice for other cases.
>
>

Well, I think the steps to do that are:

* change src/test/GNUmakefile to provide for a TMP_PORT setting that
gets passed to pg_regress
* change src/tools/msvc/vcregress.pl to match
* backpatch these changes to all live branches
* change the buildfarm script to use the new options.

I don't think this should preclude changing the way we calculate the
default port for pg_regress, for the reasons mentioned upthread.

cheers

andrew