A minor release deadline is only a couple of weeks away and
unfortunately we still have no confirmation from a Windows user that
this technique can be used to unbreak a database cluster that actually
goes though that exact upgrade sequence. So I spent today figuring
out how to write a TAP test to try to convince myself of that in the
abstract. No C code change from v2. Does the test show a useful
enough pathway that is likely to help people in future cases at least?
I learned two new things:
1. With a remapped locale, you can log into your existing databases
again, but CREATE DATABASE doesn't work (see test). That must have
been the case for the historical transitions that were hard-coded
(Macau etc); this is no different, it's just that the mapping table is
now user-changeable. In other words, this just gives an administrator
a way to log into a cluster again after such a change, but then they
really need to update the names in catalogs, at least for any database
that might be used as a template, to get a 100% functional system. So
I have added a paragraph to the documentation piece to explain that.
2. Including the encoding suffix eg "en-US.1252" doesn't actually
work after all (discovered by CI). You have to use just "en-US". (I
guess ".UTF-8" must be a special case.) So I have updated the
documentation example.