On Thu, Oct 10, 2019 at 8:38 AM Peter Eisentraut
<peter.eisentraut@2ndquadrant.com> wrote:
> On 2019-10-09 21:19, Peter Eisentraut wrote:
> > On 2019-10-03 14:25, Thomas Munro wrote:
> >>> The only open question on this patch was whether it's a good version to
> >>> use. I think based on subsequent discussions, there was the realization
> >>> that this is the best we can do and better than nothing.
> >>>
> >>> In the patch, I would skip the configure test and just do
> >>>
> >>> #ifdef __GLIBC__
> >>>
> >>> directly.
> >>
> >> Ok. Here's one like that.
> >
> > Pushed that.
>
> Actually, I had to revert that because pg_dump and pg_upgrade tests need
> to be updated, but that seems doable.
[Returning from a couple of weeks mostly away from computers]
Right, sorry about that. Here is a new version that fixes that test,
and also gives credit to Christoph for the idea in the commit message.
While testing pg_upgrade scenarios I noticed that initdb-created
collations' versions are not preserved, potentially losing track of
information about corrupted indexes. That's a preexisting condition,
and probably well understood, but it made me realise that if we switch
to per-database object (for example: per index) version tracking as
mentioned up-thread, then we should probably preserve that information
across pg_upgrade.