Thanks for the clarification, but giving up performance is a no-go for us.
Also I have my concerns about shemaqualifying each and every use of the -> operator, there are really a lot of them in my functions and it would severely impact readability.
Paul van der Linden <paul.doskabouter@gmail.com> writes: > during maintenance I saw a lot of lines in my postgreslog saying: > CONTEXT: SQL function "line_function" during inlining > automatic analyze of table "osm.planet_osm_line" > ERROR: operator does not exist: public.hstore -> unknown at character 45
It sounds like line_function is careless about its search path assumptions. auto-analyze will run index expressions with the search_path set to empty (i.e., only pg_catalog is accessible) and hstore isn't normally installed in pg_catalog.
The easy fix would be to attach "SET search_path = public" to that function, but I believe that destroys the ability to inline it, which might be a performance problem for you. Alternatively you could schema-qualify the operator name, that is "foo OPERATOR(public.->) bar".