feat(duckdb): implement approx_median

Author: cpcloud

ibis/backends/duckdb/registry.py

@@ -167,6 +167,11 @@ def _regex_extract(t, expr):
         ops.RegexExtract: _regex_extract,
         ops.RegexReplace: fixed_arity(sa.func.regexp_replace, 3),
         ops.StringContains: fixed_arity(sa.func.contains, 2),
+        ops.CMSMedian: reduction(
+            # without inline text, duckdb fails with
+            # RuntimeError: INTERNAL Error: Invalid PhysicalType for GetTypeIdSize # noqa: E501
+            lambda arg: sa.func.approx_quantile(arg, sa.text(str(0.5)))
+        ),
         ops.HLLCardinality: reduction(sa.func.approx_count_distinct),
     }
 )

ibis/backends/tests/test_client.py

@@ -337,6 +337,7 @@ def test_verify(ddl_backend, ddl_con):
         assert ddl_backend.api.verify(expr)
 
 
+@pytest.mark.never(["duckdb"], reason="duckdb supports approximate median")
 def test_not_verify(alchemy_con, alchemy_backend):
     # There is no expression that can't be compiled to any backend
     # Testing `not verify()` only for an expression not supported in postgres