> > The comment above the query in the official documentation is rather assertive
> > (even if may true to the letter) and may warrant some more cautionary
> > wording ? Added, perhaps, some variation of this:
> >
> > > For now, the only safe way to go is either reindex everything, or everything
> > > except some safe cases (non-partial indexes on plain-non-collatable datatypes
> > > only).
>
> I think the comment is very poorly worded, as it leads readers to believe that
> objects with a pg_depend dependency on a collation are the only one that would
> get corrupted in case of glibc/ICU upgrade.
>
> I agree that there should be a big fat red warning saying something like
> "reindex everything if there's any discrepancy between the recorded collation
> version and the currently reported one unless you REALLY know what you're
> doing."
Given that it does not seem straightforward to mechanically detect objects
in need of a collation-associated rebuild I would think that such a warning
would change matters for the better, documentation-wise.
Karsten
