Re: A test for replay of regression tests

Hi,

On 2022-01-27 17:16:17 -0500, Andrew Dunstan wrote:
> On crake (slowish fedora 34), a normal check run took 95s, and this test
> took 114s.

That's roughly what I see on msys after the fix.

> On my windows test instance where I noticed this (w10,
> msys2/ucrt), check took 516s and this test took 685s.

Hm. That's both excruciatingly slow. Way way slower than what I see here, also
w10, msys2/ucrt. Any chance the test instance has windows defender running,
without a directory exclusion? I saw that trash performance to a near
standstill.

Does it get better with the attached patch?

I was confused why this didn't fail fatally on CI, which uses
PG_TEST_USE_UNIX_SOCKETS. I think he reason is that pg_regress' use of PGHOST
is busted, btw. It says it'll use PGHOST if --host isn't specified, but it
doesn't work.

* When testing an existing install, we honor existing environment
* variables, except if they're overridden by command line options.
*/
if (hostname != NULL)
{
setenv("PGHOST", hostname, 1);
unsetenv("PGHOSTADDR");
}

but hostname is initialized in the existing-install case:

#if !defined(HAVE_UNIX_SOCKETS)
use_unix_sockets = false;
#elif defined(WIN32)

/*
* We don't use Unix-domain sockets on Windows by default, even if the
* build supports them. (See comment at remove_temp() for a reason.)
* Override at your own risk.
*/
use_unix_sockets = getenv("PG_TEST_USE_UNIX_SOCKETS") ? true : false;
#else
use_unix_sockets = true;
#endif

if (!use_unix_sockets)
hostname = "localhost";

Greetings,

Andres Freund