fix explode offsets for empty lists (#4005)

polars/polars-core/src/chunked_array/ops/explode.rs

@@ -229,8 +229,9 @@ pub(crate) fn offsets_to_indexes(offsets: &[i64], capacity: usize) -> Vec<IdxSiz
         }
         previous_empty = previous_offset == *offset;
     }
-    // last appended index.
-    let last_idx = idx[idx.len() - 1];
+    // undo latest increment
+    last_idx -= 1;
+
     for _ in 0..(capacity - count as usize) {
         idx.push(last_idx);
     }

polars/polars-core/src/frame/explode.rs

@@ -49,7 +49,7 @@ impl DataFrame {
 
         for (i, s) in columns.iter().enumerate() {
             // Safety:
-            // offsets are not take longer than the Series.
+            // offsets don't have indices exceeding Series length.
             if let Ok((exploded, offsets)) = get_exploded(s) {
                 let col_idx = self.check_name_to_idx(s.name())?;
 

polars/polars-core/src/series/comparison.rs

@@ -8,7 +8,8 @@ use crate::series::arithmetic::coerce_lhs_rhs;
     feature = "dtype-duration",
     feature = "dtype-datetime",
     feature = "dtype-date",
-    feature = "dtype-time"
+    feature = "dtype-time",
+    feature = "dtype-struct"
 ))]
 use std::ops::Deref;
 

polars/polars-lazy/src/logical_plan/optimizer/predicate_pushdown/mod.rs

@@ -531,7 +531,7 @@ impl PredicatePushDown {
         lp_arena: &mut Arena<ALogicalPlan>,
         expr_arena: &mut Arena<AExpr>,
     ) -> Result<ALogicalPlan> {
-        let acc_predicates = PlHashMap::with_capacity(32);
+        let acc_predicates = PlHashMap::new();
         self.push_down(logical_plan, acc_predicates, lp_arena, expr_arena)
     }
 }

py-polars/tests/test_explode.py

@@ -29,3 +29,17 @@ def test_explode_empty_df_3902() -> None:
         }
     )
     assert df.explode("second").frame_equal(expected)
+
+
+def test_explode_empty_list_4003() -> None:
+    df = pl.DataFrame(
+        [
+            {"id": 1, "nested": []},
+            {"id": 2, "nested": [1]},
+            {"id": 3, "nested": [2]},
+        ]
+    )
+    assert df.explode("nested").to_dict(False) == {
+        "id": [1, 2, 3],
+        "nested": [None, 1, 2],
+    }