Re: pgsql: Fix collprovider of predefined collations

I wrote:
> Peter Eisentraut <peter_e(at)gmx(dot)net> writes:
>> Fix collprovider of predefined collations
>> An earlier version of the patch had collprovider as an integer and thus
>> set these to 0, but the correct setting is now null.

> Surely this is not right. Neither collprovider nor the other fixed-length
> fields following it in pg_collation are marked nullable, and most of them
> are accessed as struct fields so that's not an easy thing to change.

Ah, on looking closer, I see what the patch actually changes to null is
collversion not collprovider. So the patch is ok, the commit message
not so much.

> The fact that the buildfarm didn't blow up is a bit odd. Maybe we're
> missing some enforcement somewhere?

As penance for the false alarm, I looked into that angle and indeed we
didn't have any cross-check against storing NULL from the BKI data into
a putatively not null column. We do now.

regards, tom lane