c

crates/polars-plan/src/plans/conversion/dsl_to_ir.rs

@@ -374,6 +374,39 @@ pub fn to_alp_impl(lp: DslPlan, ctxt: &mut DslConversionContext) -> PolarsResult
             output_schema: None,
             filter: None,
         },
+        // `select(pl.all().filter())` -> `filter()`
+        // `select((pl.all().<exprs>).filter())` -> `select(pl.all().<exprs>).filter()`
+        DslPlan::Select {
+            expr,
+            input,
+            options,
+        } if matches!(expr.as_slice(), &[Expr::Filter { .. }]) => {
+            let Expr::Filter {
+                input: filter_input,
+                by: predicate,
+            } = expr.into_iter().next().unwrap()
+            else {
+                unreachable!()
+            };
+
+            let input = match filter_input.as_ref() {
+                // If it's just a wildcard we don't need the `Select` node.
+                Expr::Wildcard => input,
+                _ => Arc::new(DslPlan::Select {
+                    expr: vec![Arc::unwrap_or_clone(filter_input)],
+                    input,
+                    options,
+                }),
+            };
+
+            return to_alp_impl(
+                DslPlan::Filter {
+                    input,
+                    predicate: Arc::unwrap_or_clone(predicate),
+                },
+                ctxt,
+            );
+        },
         DslPlan::Select {
             expr,
             input,

py-polars/tests/unit/expr/test_exprs.py

@@ -672,9 +672,18 @@ def test_filter_all() -> None:
     )
 
     q = df.lazy().select(pl.all().filter(~pl.col("p")))
-    assert "__POLARS_CSER" in q.explain()
+    # Ensure this is re-written to a `Filter` node during IR conversion.
+    assert r'SELECTION: col("p").not()' in q.explain()
 
     assert_frame_equal(
         q.collect(),
         pl.DataFrame({"a": [2, 5], "b": ["q", "t"], "p": False}),
     )
+
+    q = df.lazy().select((pl.all().reverse()).filter(~pl.col("p")))
+    assert r'FILTER col("p").not()' in q.explain()
+
+    assert_frame_equal(
+        q.collect(),
+        pl.DataFrame({"a": [5, 2], "b": ["t", "q"], "p": False}),
+    )