Re: pgsql: Fix pg_dump so pg_upgrade'ing an extension with simple opfamilie

Stephen Frost <sfrost(at)snowman(dot)net> writes:
> * Tom Lane (tgl(at)sss(dot)pgh(dot)pa(dot)us) wrote:
>> I see this has broken the buildfarm's pg_upgrade checks, though only in
>> HEAD which makes me suspicious that it's exposing a pre-existing bug.
>> Odd, because I definitely tested pg_upgrade here. Will dig into it
>> later this evening.

> My test shows, near the end:

> ERROR: zero-length delimited identifier at or near """" at character 40
> STATEMENT: CREATE OPERATOR FAMILY "box_ops" USING "";

> Not immediately seeing why it's unable to find the amname for box_ops
> though.

It looks like there is indeed a pre-existing bug: create_am.sql creates
an index AM, creates an opfamily for it, and then drops the index AM ...
but yet the opfamily still remains. Somebody (possibly me) missed the
boat on dependency creation. This is likely a *very* longstanding bug,
but it never mattered until we had the ability to create/drop index AMs.

Should have a fix in a bit.

regards, tom lane