On 2021-02-12 22:00, Oleg Bartunov wrote:
> We don't have (and really it's impossible) regression test for stemmers, so
> maybe we should warn users about possible inconsistencies of old
> tsvectors and new stemmers ?
Yeah, it's analogous to collation and Unicode updates. We could invent
a versioning mechanism; we have some of the infrastructure for that now.
But until we do that, perhaps some more elaborate guidance in the
major version release notes would be appropriate.