Re: Using AF_UNIX sockets always for tests on Windows

On 2022-12-01 Th 21:10, Andres Freund wrote:
> Hi,
>
> On 2022-12-01 20:56:18 -0500, Tom Lane wrote:
>> Andres Freund <andres@anarazel.de> writes:
>>> On 2022-12-01 20:30:36 -0500, Tom Lane wrote:
>>>> If we remove that, won't we have a whole lot of code that's not
>>>> tested at all on any platform, ie all the TCP-socket code?
>>> There's some coverage via the auth and ssl tests. But I agree it's an
>>> issue. But to me the fix for that seems to be to add a dedicated test for
>>> that, rather than relying on windows to test our socket code - that's quite a
>>> few separate code paths from the tcp support of other platforms.
>> IMO that's not the best way forward, because you'll always have
>> nagging questions about whether a single-purpose test covers
>> everything that needs coverage.
> Still seems better than not having any coverage in our development
> environments...
>
>
>> I think the best place to be in would be to be able to run the whole test
>> suite using either TCP or UNIX sockets, on any platform (with stuff like the
>> SSL test overriding the choice as needed).
> I agree that that's useful. But it seems somewhat independent from the
> majority of the proposed changes. To be able to test force-tcp-everywhere we
> don't need e.g.  code for setting sspi auth in pg_regress etc - it's afaik
> just needed so there's a secure way of running tests at all on windows.
>
> I think 0003 should be "trimmed" to only change the default for
> $use_unix_sockets on windows and to remove PG_TEST_USE_UNIX_SOCKETS. Whoever
> wants to, can then add a new environment variable to force tap tests to use
> tcp.
>

Not sure if it's useful here, but a few months ago I prepared patches to
remove the config-auth option of pg_regress, which struck me as more
than odd, and replace it with a perl module. I didn't get around to
finishing them, but the patches as of then are attached.

I agree that having some switch that says "run everything with TCP" or
"run (almost) everything with Unix sockets" would be good.


cheers


andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com