perf(selectors): speed up the c selector by caching the column exis…

…tence check

ibis/selectors.py

@@ -362,13 +362,17 @@ def c(*names: str | ir.Column) -> Predicate:
     """Select specific column names."""
     names = frozenset(col if isinstance(col, str) else col.get_name() for col in names)
 
-    def func(col: ir.Value) -> bool:
-        schema = col.op().rel.schema
-        if extra_cols := (names - schema.keys()):
+    @functools.cache
+    def check_delta(schema):
+        if extra_cols := names - schema._name_locs.keys():
             raise exc.IbisInputError(
                 f"Columns {extra_cols} are not present in {schema.names}"
             )
-        return col.get_name() in names
+
+    def func(col: ir.Value) -> bool:
+        op = col.op()
+        check_delta(op.rel.schema)
+        return op.name in names
 
     return where(func)
 

ibis/tests/benchmarks/test_benchmarks.py

@@ -22,9 +22,10 @@
 import ibis.expr.datatypes as dt
 import ibis.expr.operations as ops
 import ibis.expr.types as ir
+import ibis.selectors as s
 from ibis.backends import _get_backend_names
 
-pytestmark = [pytest.mark.benchmark, pytest.mark.timeout(30)]
+pytestmark = [pytest.mark.benchmark]
 
 
 def make_t():
@@ -967,3 +968,11 @@ def test_duckdb_timestamp_conversion(benchmark):
     con = ibis.duckdb.connect()
     series = benchmark(con.execute, expr)
     assert series.size == (stop - start).total_seconds()
+
+
+@pytest.mark.parametrize("cols", [1_000, 10_000])
+def test_selectors(benchmark, cols):
+    t = ibis.table(name="t", schema={f"col{i}": "int" for i in range(cols)})
+    n = cols - cols // 10
+    sel = s.across(s.c(*[f"col{i}" for i in range(n)]), lambda c: c.cast("str"))
+    benchmark(sel.expand, t)