Improved support of hstore for case when hstore extension is created inside a PG schema

[romul]

Assume we have hstore extension created in such the way:
CREATE EXTENSION IF NOT EXISTS hstore WITH SCHEMA test;
So we will get the following type info:

%Postgrex.TypeInfo{
  array_elem: 0, 
  base_type: 0, 
  comp_elems: [], 
  input: "test.hstore_in", 
  oid: 21066, 
  output: "test.hstore_out", 
  receive: "test.hstore_recv", 
  send: "test.hstore_send", 
  type: "hstore"
}

B/c of schema name is not a constant, only type field could be matched for sure.

[ericmj]

See my reply on the mailing list [1]. Please send a PR that parses out the schema name from the function name.

[1] https://groups.google.com/d/msg/elixir-ecto/1UmWWZ0_xZE/j5t_Jt9gPNYJ

[romul]

Do you mean removing schema prefix for TypeInfo fields in Types.build_types/1?

[ericmj]

Yes

[romul]

Unfortunately it isn't enough.
Types are preloaded with Types.bootstrap_query/2, which also use send matchers in SQL query like

SELECT t.oid, t.typname, t.typsend, t.typreceive, t.typoutput, t.typinput,
       t.typelem, coalesce(r.rngsubtype, 0), ARRAY (
  SELECT a.atttypid
  FROM pg_attribute AS a
  WHERE a.attrelid = t.typrelid AND a.attnum > 0 AND NOT a.attisdropped
  ORDER BY a.attnum
)
FROM pg_type AS t
LEFT JOIN pg_range AS r ON r.rngtypid = t.oid
WHERE
  t.typname::text = ANY ((ARRAY[])::text[]) OR
  t.typsend::text = ANY ((ARRAY['array_send', 'range_send', 'record_send', 'cidsend', 'xidsend', 'regtypesend', 'regclasssend', 'regoperatorsend', 'regopersend', 'regproceduresend', 'regprocsend', 'oidsend', 'hstore_send', 'unknownsend', 'tidsend', 'enum_send', 'interval_send', 'timestamptz_send', 'timestamp_send', 'timetz_send', 'time_send', 'date_send', 'uuid_send', 'numeric_send', 'float8send', 'float4send', 'int8send', 'int4send', 'int2send', 'byteasend', 'varcharsend', 'citextsend', 'textsend', 'bpcharsend', 'boolsend', 'void_send'])::text[]) OR
  t.typreceive::text = ANY ((ARRAY[])::text[]) OR
  t.typoutput::text = ANY ((ARRAY[])::text[]) OR
  t.typinput::text = ANY ((ARRAY[])::text[])

I think, we should admit, that hstore could not be matched by typsend, else we will end up with ugly workarounds like removing schema from t.typsend::text directly in SQL.
Here is a complete fix for hstore: romul@52f095d, which eliminates need in writing own Postgrex extension.