Re: Tsearch2 removal bit me - how to correct?

On 10/15/2017 14:49, Karl Denninger wrote:
> On 10/15/2017 10:47, Tom Lane wrote:
>> Karl Denninger <karl(at)denninger(dot)net> writes:
>>> Attempting to migrate to 10.0 from 9.6, which has had "tsearch2" loaded
>>> for backward compatibility since "forever" (which I suspect is no longer
>>> really required as the capability is internal and has been for a long
>>> time) results in this blowup logged in loadable_libraries.txt:
>>> could not load library "$libdir/tsearch2": ERROR:  could not access file
>>> "$libdir/tsearch2": No such file or directory
>>> So what's causing this to be requested on startup and how do I kill it? 
>> Look for '$libdir/tsearch2' in the pg_proc.probin column of each
>> DB in the installation ...
>>
>> regards, tom lane
> That worked on one of my clusters, but on another I'm seeing entries
> in pg_catalog, and can't remove them.
>
> pgsql=# \c fapforum
> You are now connected to database "fapforum" as user "pgsql".
> fapforum=# select proname from pg_proc where probin like '%tsearch2%';
>     proname
> ----------------
>  prsd_end
>  prsd_lextype
>  prsd_start
>  thesaurus_init
> (4 rows)
>
> fapforum=# \df prsd_end;
>                             List of functions
>    Schema   |   Name   | Result data type | Argument data types |  Type
> ------------+----------+------------------+---------------------+--------
>  pg_catalog | prsd_end | void             | internal            | normal
> (1 row)
>
> fapforum=# drop function prsd_end(internal);
> ERROR:  cannot drop function prsd_end(internal) because it is required
> by the database system
>
> The others I can drop, but I wind up with these in each database...
> any idea where the piece is that the original package stuffed in that
> I need to get rid of so I can kill these?
>
This has also left me with an interesting other issue that is likely
related.

The removal of the functions of course necessitated the removal of the
gin indices I was using.  Ok, I understand that.  But the 9.6
documentation says this is valid:

SELECT to_tsvector('english','in the list of stop words');

But....

ticker=# select to_tsvector('english', 'in the list of stop words');
ERROR:  invalid input syntax for type oid: "english"
LINE 1: select to_tsvector('english', 'in the list of stop words');

Interestingly enough if I don't specify the language....
ticker=# select to_tsvector('in the list of stop words');
        to_tsvector
----------------------------
 'list':3 'stop':5 'word':6
(1 row)

Which is correct.  So it appears the internal functions are both there
and working.

But, attempting to re-create the indices, omitting the language so it
defaults, fails with a complaint that the operand function must be
immutable.

ticker=# create index idx_message on post using gin(to_tsvector( message
|| ' ' || message2 || ' ' || message3));
ERROR:  functions in index expression must be marked IMMUTABLE

That's not so good.....

Yes, the default is specified as english in postgresql.conf

default_text_search_config = 'pg_catalog.english'

I suspect the "remnants" of the previous module are doing this.... any
ideas?

--
Karl Denninger
karl(at)denninger(dot)net <mailto:karl(at)denninger(dot)net>
/The Market Ticker/
/[S/MIME encrypted email preferred]/