> It might be possible to get away with having only one pg_type column, > pointing at the parse-analysis function. That function would generate > a SubscriptingRef tree node containing the OID of the appropriate > execution function, which execQual.c could call without ever knowing > its name explicitly.
Btw, is it acceptable if such generated SubscriptingRef will contain just a
function pointer to the appropriate execution function instead of an OID (e.g.
like `ExprStateEvalFunc`)? It will help to avoid problems in case of extensions.