Re: pg_tablespace_location() failure with allow_in_place_tablespaces

At Wed, 16 Mar 2022 15:42:52 +0900, Michael Paquier <michael(at)paquier(dot)xyz> wrote in
> On Wed, Mar 16, 2022 at 10:34:15AM +0900, Kyotaro Horiguchi wrote:
> > +1. Desn't the doc need to mention that?
>
> Yes, I agree that it makes sense to add a note, even if
> allow_in_place_tablespaces is a developer option. I have added the
> following paragraph in the docs:
> + A full path of the symbolic link in <filename>pg_tblspc/</filename>
> + is returned. A relative path to the data directory is returned
> + for tablespaces created with
> + <xref linkend="guc-allow-in-place-tablespaces"/> enabled.

I'm not sure that the "of the symbolic link in pg_tblspc/" is
needed. And allow_in_place_tablespaces alone doesn't create in-place
tablesapce. So this might need rethink at least for the second point.

> Another thing that was annoying in the first version of the patch is
> the useless call to lstat() on Windows, not needed because it is
> possible to rely just on pgwin32_is_junction() to check if readlink()
> should be called or not.

Agreed. And v2 looks cleaner.

The test detects the lack of the feature.
It successfully builds and runs on Rocky8 and Windows11.

> This leads me to the revised version attached. What do you think?

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center