fix: Align str.to_titlecase with polars v1.35.0 behavior (#3238)

* fix: Align str.to_titlecase

* Update docstrings

Author: FBruzzesi

narwhals/_duckdb/expr_str.py

@@ -35,7 +35,7 @@ def to_titlecase(self) -> DuckDBExpr:
 
         def _to_titlecase(expr: Expression) -> Expression:
             extract_expr = F(
-                "regexp_extract_all", F("lower", expr), lit(r"[a-z0-9]*[^a-z0-9]*")
+                "regexp_extract_all", F("lower", expr), lit(r"[a-z]*[^a-z]*")
             )
             elem = col("_")
             capitalize = lambda_expr(

narwhals/_polars/expr.py

@@ -393,10 +393,10 @@ class PolarsExprStringNamespace(
     def to_titlecase(self) -> PolarsExpr:
         native_expr = self.native
 
-        if BACKEND_VERSION < (1, 5):
+        if BACKEND_VERSION < (1, 35):
             native_result = (
                 native_expr.str.to_lowercase()
-                .str.extract_all(r"[a-z0-9]*[^a-z0-9]*")
+                .str.extract_all(r"[a-z]*[^a-z]*")
                 .list.eval(pl.element().str.to_titlecase())
                 .list.join("")
             )

narwhals/_spark_like/expr_str.py

@@ -54,7 +54,7 @@ def _to_titlecase(expr: Column) -> Column:
             F = self.compliant._F
             lower_expr = F.lower(expr)
             extract_expr = F.regexp_extract_all(
-                lower_expr, regexp=F.lit(r"[a-z0-9]*[^a-z0-9]*"), idx=0
+                lower_expr, regexp=F.lit(r"[a-z]*[^a-z]*"), idx=0
             )
             capitalized_expr = F.transform(extract_expr, f=F.initcap)
             return F.array_join(capitalized_expr, delimiter="")

narwhals/expr_str.py

@@ -446,21 +446,8 @@ def to_titlecase(self) -> ExprT:
             This is a form of case transform where the first letter of each word is
             capitalized, with the rest of the word in lowercase.
 
-        Warning:
-            Different backends might follow different rules to determine what a "word" is:
-
-            - duckdb, polars and spark-like use non-**alphanumeric** characters to
-                define the word boundaries.
-            - pandas-like, pyarrow and dask use non-**alphabetic** characters to define
-                the word boundaries, matching the behavior of
-                [`str.title`](https://docs.python.org/3/library/stdtypes.html#str.title).
-
-            We can observe the difference with the string `"with123numbers"`:
-
-            - non-**alphanumeric** -> `"With123numbers"`
-                - notice lowercase **n** after the digits
-            - non-**alphabetic** -> `"With123Numbers"`
-                - notice uppercase **N** after the digits
+            Word boundaries are defined by non-**alphabetic** characters, matching the
+            behavior of [`str.title`](https://docs.python.org/3/library/stdtypes.html#str.title)
 
         Examples:
             >>> import polars as pl

narwhals/series_str.py

@@ -399,17 +399,8 @@ def to_titlecase(self) -> SeriesT:
             This is a form of case transform where the first letter of each word is
             capitalized, with the rest of the word in lowercase.
 
-        Warning:
-            Different backends might follow different rules to determine what a "word" is:
-
-            - polars uses **non-alphanumeric** characters to define the word boundaries.
-            - pandas-like and pyarrow use **non-alphabetic** characters to define
-                the word boundaries, matching the behavior of
-                [`str.title`](https://docs.python.org/3/library/stdtypes.html#str.title).
-
-            As an example of such difference, in the former case the string `"with123numbers"`
-            is mapped to `"With123numbers"` (notice lowercase **n** after the digits), while
-            in the latter to `"With123Numbers"` (notice uppercase **N** after the digits).
+            Word boundaries are defined by non-**alphabetic** characters, matching the
+            behavior of [`str.title`](https://docs.python.org/3/library/stdtypes.html#str.title)
 
         Examples:
             >>> import pyarrow as pa

tests/expr_and_series/str/to_titlecase_test.py

@@ -15,26 +15,7 @@
     ]
 }
 
-expected_non_alphabetic = {
-    "a": [
-        "E.T. Phone Home",
-        "They'Re Bill'S Friends From The Uk",
-        "To Infinity,And Beyond!",
-        "With123Numbers",
-        "__Dunder__Score_A1_.2B ?Three",
-    ]
-}
-expected_non_alphanumeric = {
-    "a": [
-        "E.T. Phone Home",
-        "They'Re Bill'S Friends From The Uk",
-        "To Infinity,And Beyond!",
-        "With123numbers",
-        "__Dunder__Score_A1_.2b ?Three",
-    ]
-}
-
-NON_ALPHANUMERIC_BACKENDS = ("duckdb", "polars", "pyspark")
+expected = {"a": [s.title() for s in data["a"]]}
 
 
 def test_str_to_titlecase_expr(
@@ -47,25 +28,13 @@ def test_str_to_titlecase_expr(
     if "ibis" in str(constructor):
         request.applymarker(pytest.mark.xfail)
 
-    expected = (
-        expected_non_alphanumeric
-        if any(x in str(constructor) for x in NON_ALPHANUMERIC_BACKENDS)
-        else expected_non_alphabetic
-    )
-
     df = nw.from_native(constructor(data))
     result_frame = df.select(nw.col("a").str.to_titlecase())
 
     assert_equal_data(result_frame, expected)
 
 
 def test_str_to_titlecase_series(constructor_eager: ConstructorEager) -> None:
-    expected = (
-        expected_non_alphanumeric
-        if any(x in str(constructor_eager) for x in NON_ALPHANUMERIC_BACKENDS)
-        else expected_non_alphabetic
-    )
-
     df = nw.from_native(constructor_eager(data), eager_only=True)
     result_series = df["a"].str.to_titlecase()
 

tests/utils_test.py

@@ -524,7 +524,7 @@ def func3(
 
 
 def test_requires() -> None:
-    class SomeAccesssor:
+    class SomeAccessor:
         _accessor: ClassVar[Accessor] = "str"
 
         def __init__(self, compliant: ProbablyCompliant) -> None:
@@ -565,8 +565,8 @@ def repeat(self, n: int) -> str:
             return self.native * n
 
         @property
-        def str(self) -> SomeAccesssor:
-            return SomeAccesssor(self)
+        def str(self) -> SomeAccessor:
+            return SomeAccessor(self)
 
     v_05 = ProbablyCompliant("123", (0, 5))
     v_201 = ProbablyCompliant("123", (2, 0, 1))