Bump minimum Polars version to 1.35

.pre-commit-config.yaml

@@ -35,7 +35,7 @@ repos:
     hooks:
       - id: mypy
         additional_dependencies: [
-          "polars>=1.30,<1.39",
+          "polars>=1.35,<1.39",
           "numpy",
           pyarrow-stubs,
           "pyarrow<24.0.0",  # https://github.com/rapidsai/cudf/issues/22229

ci/test_wheel_cudf_polars.sh

@@ -71,7 +71,7 @@ for version in "${VERSIONS[@]}"; do
         COVERAGE_ARGS=(--no-cov)
     fi
 
-    timeout 1h ./ci/run_cudf_polars_pytests.sh \
+    timeout 15m ./ci/run_cudf_polars_pytests.sh \
         "${COVERAGE_ARGS[@]}" \
         --numprocesses=8 \
         --dist=worksteal \

conda/environments/all_cuda-129_arch-aarch64.yaml

@@ -67,7 +67,7 @@ dependencies:
 - packaging
 - pandas>=2.0,<2.4.0
 - pandoc
-- polars>=1.30,<1.39
+- polars>=1.35,<1.39
 - pre-commit
 - pyarrow>=19.0.0,<24
 - pydata-sphinx-theme>=0.15.4

conda/environments/all_cuda-129_arch-x86_64.yaml

@@ -67,7 +67,7 @@ dependencies:
 - packaging
 - pandas>=2.0,<2.4.0
 - pandoc
-- polars>=1.30,<1.39
+- polars>=1.35,<1.39
 - pre-commit
 - pyarrow>=19.0.0,<24
 - pydata-sphinx-theme>=0.15.4

conda/environments/all_cuda-131_arch-aarch64.yaml

@@ -67,7 +67,7 @@ dependencies:
 - packaging
 - pandas>=2.0,<2.4.0
 - pandoc
-- polars>=1.30,<1.39
+- polars>=1.35,<1.39
 - pre-commit
 - pyarrow>=19.0.0,<24
 - pydata-sphinx-theme>=0.15.4

conda/environments/all_cuda-131_arch-x86_64.yaml

@@ -67,7 +67,7 @@ dependencies:
 - packaging
 - pandas>=2.0,<2.4.0
 - pandoc
-- polars>=1.30,<1.39
+- polars>=1.35,<1.39
 - pre-commit
 - pyarrow>=19.0.0,<24
 - pydata-sphinx-theme>=0.15.4

conda/recipes/cudf-polars/recipe.yaml

@@ -37,7 +37,7 @@ requirements:
     - python
     - pylibcudf =${{ version }}
     - rapidsmpf =${{ minor_version }}
-    - polars>=1.30,<1.39
+    - polars>=1.35,<1.39
     - packaging
     - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="x") }}
     - if: cuda_major == "12"

dependencies.yaml

@@ -821,7 +821,7 @@ dependencies:
           # 'nvidia-ml-py' provides the 'pynvml' module
           - &nvidia_ml_py nvidia-ml-py>=12
           - packaging
-          - polars>=1.30,<1.39
+          - polars>=1.35,<1.39
           - typing_extensions>=4.0.0
   run_cudf_polars_experimental:
     common:

python/cudf_polars/cudf_polars/containers/datatype.py

@@ -45,13 +45,9 @@ def _dtype_to_header(dtype: pl.DataType) -> DataTypeHeader:
     if name in SCALAR_NAME_TO_POLARS_TYPE_MAP:
         return {"kind": "scalar", "name": name}
     if isinstance(dtype, pl.Decimal):
-        # TODO: Add version guard once we support polars 1.34
-        # Also keep in mind the typing change in polars:
-        # https://github.com/pola-rs/polars/pull/25227
-        precision = dtype.precision if dtype.precision is not None else 38
         return {
             "kind": "decimal",
-            "precision": precision,
+            "precision": dtype.precision,
             "scale": dtype.scale,
         }
     if isinstance(dtype, pl.Datetime):

python/cudf_polars/cudf_polars/dsl/expressions/string.py

@@ -12,7 +12,7 @@
 from enum import IntEnum, auto
 from typing import TYPE_CHECKING, Any, ClassVar, cast
 
-from polars import Struct as pl_Struct, polars  # type: ignore[attr-defined]
+from polars import Struct as pl_Struct
 from polars.exceptions import InvalidOperationError
 
 import pylibcudf as plc
@@ -23,14 +23,15 @@
 from cudf_polars.dsl.utils.reshape import broadcast
 from cudf_polars.utils.dtypes import make_empty_column
 from cudf_polars.utils.versions import (
-    POLARS_VERSION_LT_132,
     POLARS_VERSION_LT_136,
     POLARS_VERSION_LT_138,
 )
 
 if TYPE_CHECKING:
     from typing import Self
 
+    from polars import polars  # type: ignore[attr-defined]
+
     from cudf_polars.containers import DataFrame, DataType
 
 __all__ = ["StringFunction"]
@@ -292,21 +293,6 @@ def _validate_input(self) -> None:
                 raise NotImplementedError(
                     "strip operations only support scalar patterns"
                 )
-        elif self.name is StringFunction.Name.ZFill:
-            if isinstance(self.children[1], Literal):
-                _, width = self.children
-                assert isinstance(width, Literal)
-                if (
-                    POLARS_VERSION_LT_132
-                    and width.value is not None
-                    and width.value < 0
-                ):  # pragma: no cover
-                    dtypestr = polars.dtype_str_repr(width.dtype.polars_type)
-                    raise InvalidOperationError(
-                        f"conversion from `{dtypestr}` to `u64` "
-                        f"failed in column 'literal' for 1 out of "
-                        f"1 values: [{width.value}]"
-                    ) from None
 
     @staticmethod
     def _create_regex_program(
@@ -437,24 +423,6 @@ def do_evaluate(
             else:
                 col_width = self.children[1].evaluate(df, context=context)
                 assert isinstance(col_width, Column)
-                all_gt_0 = plc.binaryop.binary_operation(
-                    col_width.obj,
-                    plc.Scalar.from_py(
-                        0, plc.DataType(plc.TypeId.INT64), stream=df.stream
-                    ),
-                    plc.binaryop.BinaryOperator.GREATER_EQUAL,
-                    plc.DataType(plc.TypeId.BOOL8),
-                    stream=df.stream,
-                )
-
-                if POLARS_VERSION_LT_132 and not plc.reduce.reduce(
-                    all_gt_0,
-                    plc.aggregation.all(),
-                    plc.DataType(plc.TypeId.BOOL8),
-                    stream=df.stream,
-                ).to_py(stream=df.stream):  # pragma: no cover
-                    raise InvalidOperationError("fill conversion failed.")
-
                 return Column(
                     plc.strings.padding.zfill_by_widths(
                         column.obj, col_width.obj, stream=df.stream
@@ -981,21 +949,14 @@ def do_evaluate(
                 dtype=self.dtype,
             )
         elif self.name is StringFunction.Name.PadStart:
-            if POLARS_VERSION_LT_132:  # pragma: no cover
-                (column,) = columns
-                width_arg, char = self.options
-                pad_width = cast(int, width_arg)
-            else:
-                (column, width_col) = columns
-                (char,) = self.options
-                # TODO: Maybe accept a string scalar in
-                # cudf::strings::pad to avoid DtoH transfer
-                # See https://github.com/rapidsai/cudf/issues/20202
-                width_py = width_col.obj.to_scalar(stream=df.stream).to_py(
-                    stream=df.stream
-                )
-                assert width_py is not None
-                pad_width = int(width_py)
+            (column, width_col) = columns
+            (char,) = self.options
+            # TODO: Maybe accept a string scalar in
+            # cudf::strings::pad to avoid DtoH transfer
+            # See https://github.com/rapidsai/cudf/issues/20202
+            width_py = width_col.obj.to_scalar(stream=df.stream).to_py(stream=df.stream)
+            assert width_py is not None
+            pad_width = int(width_py)
 
             return Column(
                 plc.strings.padding.pad(
@@ -1008,20 +969,13 @@ def do_evaluate(
                 dtype=self.dtype,
             )
         elif self.name is StringFunction.Name.PadEnd:
-            if POLARS_VERSION_LT_132:  # pragma: no cover
-                (column,) = columns
-                width_arg, char = self.options
-                pad_width = cast(int, width_arg)
-            else:
-                (column, width_col) = columns
-                (char,) = self.options
-                # TODO: Maybe accept a string scalar in
-                # cudf::strings::pad to avoid DtoH transfer
-                width_py = width_col.obj.to_scalar(stream=df.stream).to_py(
-                    stream=df.stream
-                )
-                assert width_py is not None
-                pad_width = int(width_py)
+            (column, width_col) = columns
+            (char,) = self.options
+            # TODO: Maybe accept a string scalar in
+            # cudf::strings::pad to avoid DtoH transfer
+            width_py = width_col.obj.to_scalar(stream=df.stream).to_py(stream=df.stream)
+            assert width_py is not None
+            pad_width = int(width_py)
 
             return Column(
                 plc.strings.padding.pad(

python/cudf_polars/cudf_polars/dsl/ir.py

@@ -49,8 +49,6 @@
     join_cuda_streams,
 )
 from cudf_polars.utils.versions import (
-    POLARS_VERSION_LT_131,
-    POLARS_VERSION_LT_134,
     POLARS_VERSION_LT_136,
     POLARS_VERSION_LT_137,
     POLARS_VERSION_LT_138,
@@ -505,16 +503,7 @@ def __init__(
             raise NotImplementedError(
                 "Read from cloud storage"
             )  # pragma: no cover; no test yet
-        if (
-            any(str(p).startswith("https:/") for p in self.paths)
-            and POLARS_VERSION_LT_131
-        ):  # pragma: no cover; polars passed us the wrong URI
-            # https://github.com/pola-rs/polars/issues/22766
-            raise NotImplementedError("Read from https")
-        if any(
-            str(p).startswith("file:/" if POLARS_VERSION_LT_131 else "file://")
-            for p in self.paths
-        ):
+        if any(str(p).startswith("file://") for p in self.paths):
             raise NotImplementedError("Read from file URI")
         if self.typ == "csv":
             if any(
@@ -1980,19 +1969,15 @@ def _strip_predicate_casts(node: expr.Expr) -> expr.Expr:
         ):
             return child
 
-        if (
-            not POLARS_VERSION_LT_134
-            and isinstance(child, expr.ColRef)
-            and (
-                (
-                    plc.traits.is_floating_point(src.plc_type)
-                    and plc.traits.is_floating_point(dst.plc_type)
-                )
-                or (
-                    plc.traits.is_integral(src.plc_type)
-                    and plc.traits.is_integral(dst.plc_type)
-                    and src.plc_type.id() == dst.plc_type.id()
-                )
+        if isinstance(child, expr.ColRef) and (
+            (
+                plc.traits.is_floating_point(src.plc_type)
+                and plc.traits.is_floating_point(dst.plc_type)
+            )
+            or (
+                plc.traits.is_integral(src.plc_type)
+                and plc.traits.is_integral(dst.plc_type)
+                and src.plc_type.id() == dst.plc_type.id()
             )
         ):
             return child
@@ -3040,16 +3025,6 @@ def __init__(self, schema: Schema, name: str, options: Any, df: IR):
                 # same sub-shapes
                 raise NotImplementedError("Explode with more than one column")
             self.options = (tuple(to_explode),)
-        elif POLARS_VERSION_LT_131 and self.name == "rename":  # pragma: no cover
-            # As of 1.31, polars validates renaming in the IR
-            old, new, strict = self.options
-            if len(new) != len(set(new)) or (
-                set(new) & (set(df.schema.keys()) - set(old))
-            ):
-                raise NotImplementedError(
-                    "Duplicate new names in rename."
-                )  # pragma: no cover
-            self.options = (tuple(old), tuple(new), strict)
         elif self.name == "unpivot":
             indices, pivotees, variable_name, value_name = self.options
             value_name = "value" if value_name is None else value_name
@@ -3123,11 +3098,6 @@ def do_evaluate(
             # No-op in our data model
             # Don't think this appears in a plan tree from python
             return df  # pragma: no cover
-        elif POLARS_VERSION_LT_131 and name == "rename":  # pragma: no cover
-            # final tag is "swapping" which is useful for the
-            # optimiser (it blocks some pushdown operations)
-            old, new, _ = options
-            return df.rename_columns(dict(zip(old, new, strict=True)))
         elif name == "explode":
             ((to_explode,),) = options
             index = df.column_names.index(to_explode)