On Tue, Aug 19, 2025 at 12:52:24PM +0000, Hayato Kuroda (Fujitsu) wrote:
> I have run the tests on my CI and found that windows cannot accept the test.
> Per attached output, starting the postgres with the single-user mode was failed
> because the user had admin permissions [1]. However, pg_ctl start command could be
> done by the same user.
> I'm not familiar with Windows, but according to your blogpost [2], Windows seems
> to allows running some server command by generating the restricted context for
> running postgres commands. In case of single-user mode, however, we directly run
> the given command with the current user. So, there is a possibility that only
> instance can boot only by pg_ctl.
> Based on that, I want to skip the test on windows platform rather than modifying
> the ci environment.
>
> IIUC, other tests which uses --single cannot work on windows as well.
This has reminded me of 1a9d80282811, where I have used a trick with
pg_ctl to allow single-user mode executions to bypass the privileged
account permission failure, but I don't see what we could do here as
we should pass down commands to postgres for execution via stdin for
IPC::Run. We don't really have anything WIN32-specific, so skipping
the tests on Windows sounds fine by me.
One thing that you have forgotten is to update EXTRA_INSTALL to add
test_decoding, or a make check in the module fails. I have expanded a
bit more the tests, as for example we have paths based on active_pid
for temporary slots, which could matter at some point. I have added a
few more things with physical slots.
With all that in mind, I don't really see the point of 0002. There
was also a mention of replication origins upthread, but we don't
really have anything specific to a shared state or a validation in
this area AFAIK, so I don't think that test additions are worth it in
this case.
--
Michael
