fix: SanityCheckPlan error with window functions and NVL filter (#20231)

## Which issue does this PR close?

Closes #20194

## Rationale for this change

A query with `ROW_NUMBER() OVER (... ORDER BY CASE WHEN col='0' THEN 1
ELSE 0 END)` combined with a filter `nvl(t2.value_2_3,'0')='0'` fails
with a `SanityCheckPlan` error. This worked in 50.3.0 but broke in
52.1.0.

## What changes are included in this PR?

**Root cause**: `collect_columns_from_predicate_inner` was extracting
equality pairs where neither side was a `Column` (e.g. `nvl(col, '0') =
'0'`), creating equivalence classes between complex expressions and
literals. `normalize_expr`'s deep traversal would then replace the
literal `'0'` inside unrelated sort/window CASE WHEN expressions with
the complex NVL expression, corrupting the sort ordering and causing a
mismatch between `SortExec`'s reported output ordering and
`BoundedWindowAggExec`'s expected ordering.

**Fix** (two changes in `filter.rs`):
1. **`collect_columns_from_predicate_inner`**: Only extract equality
pairs where at least one side is a `Column` reference. This matches the
function's documented intent ("Column-Pairs") and prevents
complex-expression-to-literal equivalence classes from being created.
2. **`extend_constants`**: Recognize `Literal` expressions as inherently
constant (previously only checked `is_expr_constant` on the input's
equivalence properties, which doesn't know about literals). This ensures
constant propagation still works for `complex_expr = literal` predicates
— e.g. `nvl(col, '0')` is properly marked as constant after the filter.

## How was this tested?

- Unit test `test_collect_columns_skips_non_column_pairs` verifying the
filtering logic
- Sqllogictest reproducing the exact query from the issue
- Full test suites: equivalence tests (51 passed), physical-plan tests
(1255 passed), physical-optimizer tests (20 passed)
- Manual verification with datafusion-cli running the reproduction query

## Test plan
- [x] Unit test for `collect_columns_from_predicate_inner` column
filtering
- [x] Sqllogictest regression test for #20194
- [x] Existing test suites pass
- [x] Manual reproduction query succeeds

---------

Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>

Author: EeshanBembi

datafusion/physical-plan/src/filter.rs

@@ -58,12 +58,12 @@ use datafusion_common::{
 use datafusion_execution::TaskContext;
 use datafusion_expr::Operator;
 use datafusion_physical_expr::equivalence::ProjectionMapping;
-use datafusion_physical_expr::expressions::{BinaryExpr, Column, lit};
+use datafusion_physical_expr::expressions::{BinaryExpr, Column, Literal, lit};
 use datafusion_physical_expr::intervals::utils::check_support;
 use datafusion_physical_expr::utils::{collect_columns, reassign_expr_columns};
 use datafusion_physical_expr::{
-    AcrossPartitions, AnalysisContext, ConstExpr, ExprBoundaries, PhysicalExpr, analyze,
-    conjunction, split_conjunction,
+    AcrossPartitions, AnalysisContext, ConstExpr, EquivalenceProperties, ExprBoundaries,
+    PhysicalExpr, analyze, conjunction, split_conjunction,
 };
 
 use datafusion_physical_expr_common::physical_expr::fmt_sql;
@@ -349,6 +349,20 @@ impl FilterExec {
         })
     }
 
+    /// Returns the `AcrossPartitions` value for `expr` if it is constant:
+    /// either already known constant in `input_eqs`, or a `Literal`
+    /// (which is inherently constant across all partitions).
+    fn expr_constant_or_literal(
+        expr: &Arc<dyn PhysicalExpr>,
+        input_eqs: &EquivalenceProperties,
+    ) -> Option<AcrossPartitions> {
+        input_eqs.is_expr_constant(expr).or_else(|| {
+            expr.as_any()
+                .downcast_ref::<Literal>()
+                .map(|l| AcrossPartitions::Uniform(Some(l.value().clone())))
+        })
+    }
+
     fn extend_constants(
         input: &Arc<dyn ExecutionPlan>,
         predicate: &Arc<dyn PhysicalExpr>,
@@ -361,18 +375,24 @@ impl FilterExec {
             if let Some(binary) = conjunction.as_any().downcast_ref::<BinaryExpr>()
                 && binary.op() == &Operator::Eq
             {
-                // Filter evaluates to single value for all partitions
-                if input_eqs.is_expr_constant(binary.left()).is_some() {
-                    let across = input_eqs
-                        .is_expr_constant(binary.right())
-                        .unwrap_or_default();
+                // Check if either side is constant — either already known
+                // constant from the input equivalence properties, or a literal
+                // value (which is inherently constant across all partitions).
+                let left_const = Self::expr_constant_or_literal(binary.left(), input_eqs);
+                let right_const =
+                    Self::expr_constant_or_literal(binary.right(), input_eqs);
+
+                if let Some(left_across) = left_const {
+                    // LEFT is constant, so RIGHT must also be constant.
+                    // Use RIGHT's known across value if available, otherwise
+                    // propagate LEFT's (e.g. Uniform from a literal).
+                    let across = right_const.unwrap_or(left_across);
                     res_constants
                         .push(ConstExpr::new(Arc::clone(binary.right()), across));
-                } else if input_eqs.is_expr_constant(binary.right()).is_some() {
-                    let across = input_eqs
-                        .is_expr_constant(binary.left())
-                        .unwrap_or_default();
-                    res_constants.push(ConstExpr::new(Arc::clone(binary.left()), across));
+                } else if let Some(right_across) = right_const {
+                    // RIGHT is constant, so LEFT must also be constant.
+                    res_constants
+                        .push(ConstExpr::new(Arc::clone(binary.left()), right_across));
                 }
             }
         }
@@ -1012,6 +1032,19 @@ fn collect_columns_from_predicate_inner(
     let predicates = split_conjunction(predicate);
     predicates.into_iter().for_each(|p| {
         if let Some(binary) = p.as_any().downcast_ref::<BinaryExpr>() {
+            // Only extract pairs where at least one side is a Column reference.
+            // Pairs like `complex_expr = literal` should not create equivalence
+            // classes — the literal could appear in many unrelated expressions
+            // (e.g. sort keys), and normalize_expr's deep traversal would
+            // replace those occurrences with the complex expression, corrupting
+            // sort orderings. Constant propagation for such pairs is handled
+            // separately by `extend_constants`.
+            let has_direct_column_operand =
+                binary.left().as_any().downcast_ref::<Column>().is_some()
+                    || binary.right().as_any().downcast_ref::<Column>().is_some();
+            if !has_direct_column_operand {
+                return;
+            }
             match binary.op() {
                 Operator::Eq => {
                     eq_predicate_columns.push((binary.left(), binary.right()))
@@ -2164,6 +2197,48 @@ mod tests {
         Ok(())
     }
 
+    /// Regression test for https://github.com/apache/datafusion/issues/20194
+    ///
+    /// `collect_columns_from_predicate_inner` should only extract equality
+    /// pairs where at least one side is a Column. Pairs like
+    /// `complex_expr = literal` must not create equivalence classes because
+    /// `normalize_expr`'s deep traversal would replace the literal inside
+    /// unrelated expressions (e.g. sort keys) with the complex expression.
+    #[test]
+    fn test_collect_columns_skips_non_column_pairs() -> Result<()> {
+        let schema = test::aggr_test_schema();
+
+        // Simulate: nvl(c2, 0) = 0  →  (c2 IS DISTINCT FROM 0) = 0
+        // Neither side is a Column, so this should NOT be extracted.
+        let complex_expr: Arc<dyn PhysicalExpr> = binary(
+            col("c2", &schema)?,
+            Operator::IsDistinctFrom,
+            lit(0u32),
+            &schema,
+        )?;
+        let predicate: Arc<dyn PhysicalExpr> =
+            binary(complex_expr, Operator::Eq, lit(0u32), &schema)?;
+
+        let (equal_pairs, _) = collect_columns_from_predicate_inner(&predicate);
+        assert_eq!(
+            0,
+            equal_pairs.len(),
+            "Should not extract equality pairs where neither side is a Column"
+        );
+
+        // But col = literal should still be extracted
+        let predicate: Arc<dyn PhysicalExpr> =
+            binary(col("c2", &schema)?, Operator::Eq, lit(0u32), &schema)?;
+        let (equal_pairs, _) = collect_columns_from_predicate_inner(&predicate);
+        assert_eq!(
+            1,
+            equal_pairs.len(),
+            "Should extract equality pairs where one side is a Column"
+        );
+
+        Ok(())
+    }
+
     /// Columns with Absent min/max statistics should remain Absent after
     /// FilterExec.
     #[tokio::test]

datafusion/sqllogictest/test_files/window.slt

@@ -6081,3 +6081,49 @@ WHERE acctbal > (
 );
 ----
 1
+
+# Regression test for https://github.com/apache/datafusion/issues/20194
+# Window function with CASE WHEN in ORDER BY combined with NVL filter
+# should not trigger SanityCheckPlan error from equivalence normalization
+# replacing literals in sort expressions with complex filter expressions.
+statement ok
+CREATE TABLE issue_20194_t1 (
+  value_1_1 decimal(25) NULL,
+  value_1_2 int NULL,
+  value_1_3 bigint NULL
+);
+
+statement ok
+CREATE TABLE issue_20194_t2 (
+  value_2_1 bigint NULL,
+  value_2_2 varchar(140) NULL,
+  value_2_3 varchar(140) NULL
+);
+
+statement ok
+INSERT INTO issue_20194_t1 (value_1_1, value_1_2, value_1_3) VALUES (6774502793, 10040029, 1120);
+
+statement ok
+INSERT INTO issue_20194_t2 (value_2_1, value_2_2, value_2_3) VALUES (1120, '0', '0');
+
+query RII
+SELECT
+  t1.value_1_1, t1.value_1_2,
+  ROW_NUMBER() OVER (
+    PARTITION BY t1.value_1_1, t1.value_1_2
+    ORDER BY
+      CASE WHEN t2.value_2_2 = '0' THEN 1 ELSE 0 END ASC,
+      CASE WHEN t2.value_2_3 = '0' THEN 1 ELSE 0 END ASC
+  ) AS ord
+FROM issue_20194_t1 t1
+INNER JOIN issue_20194_t2 t2
+  ON t1.value_1_3 = t2.value_2_1
+  AND nvl(t2.value_2_3, '0') = '0';
+----
+6774502793 10040029 1
+
+statement ok
+DROP TABLE issue_20194_t1;
+
+statement ok
+DROP TABLE issue_20194_t2;