On 2015-05-18 19:23:59 -0400, Tom Lane wrote:
> OK, now I'm on the warpath, because I went to fix this and discovered
> that since that discussion, somebody named Freund committed yet another
> shared catalog with a collation-dependent index. This time, at least,
> we can fix it *before* it gets into the wild.
Hrmpf, good point.
> Is it okay to change pg_replication_origin.roname to type "name",
> and if not what do you want to do instead?
It was turned into text after it initially was name, because of length
concerns.
Hm, just forcing a collation and restricting the input to ascii should
work, right? What I'm wondering is how we easily can do the collation
forcing part. The best seems to be to force the collation on the column
itself. We could add BKI_COLLATION(). Or we could invent a alias
'systext' or something that's intended to be used in catalogs?
> While I'm looking at it, why in the world have roident and not just a
> standard system OID column? This catalog seems willfully ignorant of
> Postgres conventions.
There's a comment: * Needs to fit into an uint16, so we don't waste too much space in WAL * records. For this reason we
don'tuse a normal Oid column here, since * we need to handle allocation of new values manually.
I mean it could use the standard oid, but given it's allocated
differently...