fix: Preserve lexical ordering on concat

crates/polars-core/src/chunked_array/logical/categorical/ops/full.rs

@@ -1,15 +1,20 @@
 use super::*;
 
 impl CategoricalChunked {
-    pub fn full_null(name: &str, is_enum: bool, length: usize) -> CategoricalChunked {
+    pub fn full_null(
+        name: &str,
+        is_enum: bool,
+        length: usize,
+        ordering: CategoricalOrdering,
+    ) -> CategoricalChunked {
         let cats = UInt32Chunked::full_null(name, length);
 
         unsafe {
             CategoricalChunked::from_cats_and_rev_map_unchecked(
                 cats,
                 Arc::new(RevMapping::default()),
                 is_enum,
-                Default::default(),
+                ordering,
             )
         }
     }

crates/polars-core/src/series/ops/null.rs

@@ -12,9 +12,13 @@ impl Series {
                 ArrayChunked::full_null_with_dtype(name, size, inner_dtype, *width).into_series()
             },
             #[cfg(feature = "dtype-categorical")]
-            dt @ (DataType::Categorical(rev_map, _) | DataType::Enum(rev_map, _)) => {
-                let mut ca =
-                    CategoricalChunked::full_null(name, matches!(dt, DataType::Enum(_, _)), size);
+            dt @ (DataType::Categorical(rev_map, ord) | DataType::Enum(rev_map, ord)) => {
+                let mut ca = CategoricalChunked::full_null(
+                    name,
+                    matches!(dt, DataType::Enum(_, _)),
+                    size,
+                    *ord,
+                );
                 // ensure we keep the rev-map of a cleared series
                 if let Some(rev_map) = rev_map {
                     unsafe { ca.set_rev_map(rev_map.clone(), false) }

py-polars/tests/unit/datatypes/test_categorical.py

@@ -812,3 +812,17 @@ def test_cast_from_cat_to_numeric() -> None:
 
     s = pl.Series(["1", "2", "3"], dtype=pl.Categorical)
     assert s.cast(pl.UInt8).sum() == 6
+
+
+def test_cat_preserve_lexical_ordering_on_clear() -> None:
+    s = pl.Series("a", ["a", "b"], dtype=pl.Categorical(ordering="lexical"))
+    s2 = s.clear()
+    assert s.dtype == s2.dtype
+
+
+def test_cat_preserve_lexical_ordering_on_concat() -> None:
+    dtype = pl.Categorical(ordering="lexical")
+
+    df = pl.DataFrame({"x": ["b", "a", "c"]}).with_columns(pl.col("x").cast(dtype))
+    df2 = pl.concat([df, df])
+    assert df2["x"].dtype == dtype