ncs() (defined in test/core/functions.sql) is a SQL function the test suite uses to introspect count_nulls' actual installed schema at runtime, via a pg_extension/pg_namespace join:
CREATE FUNCTION ncs() RETURNS name IMMUTABLE LANGUAGE sql AS $$
SELECT nspname FROM pg_namespace n JOIN pg_extension x ON n.oid = x.extnamespace WHERE extname = 'count_nulls'
$$;
It's used throughout test/core/functions.sql and test/sql/extension_tests.sql to schema-qualify calls into count_nulls regardless of which schema the extension landed in for a given test run.
Separately, the test suite is moving to a design where every install always uses a single, uniquely-prefixed, randomly generated schema name (see the in-progress PR implementing this), discoverable by any script via a new test/helpers/find_test_schema.sql psql-variable-based lookup (:"test_schema"). Now that a general-purpose discovery mechanism like that exists, ncs() may be redundant, overlapping machinery worth consolidating away in favor of one consistent way of finding "the schema count_nulls is installed in" - but this hasn't been evaluated in any depth, and is NOT part of that PR's scope.
This issue is just to track that as something worth investigating later - no decision has been made yet, and this should not block or be conflated with the random-schema PR itself.
ncs()(defined intest/core/functions.sql) is a SQL function the test suite uses to introspect count_nulls' actual installed schema at runtime, via apg_extension/pg_namespacejoin:It's used throughout
test/core/functions.sqlandtest/sql/extension_tests.sqlto schema-qualify calls into count_nulls regardless of which schema the extension landed in for a given test run.Separately, the test suite is moving to a design where every install always uses a single, uniquely-prefixed, randomly generated schema name (see the in-progress PR implementing this), discoverable by any script via a new
test/helpers/find_test_schema.sqlpsql-variable-based lookup (:"test_schema"). Now that a general-purpose discovery mechanism like that exists,ncs()may be redundant, overlapping machinery worth consolidating away in favor of one consistent way of finding "the schema count_nulls is installed in" - but this hasn't been evaluated in any depth, and is NOT part of that PR's scope.This issue is just to track that as something worth investigating later - no decision has been made yet, and this should not block or be conflated with the random-schema PR itself.