Add cudf.options

docs/cudf/source/api_docs/index.rst

@@ -19,3 +19,4 @@ This page provides a list of all publicly accessible modules, methods and classe
     io
     subword_tokenize
     string_handling
+    options

docs/cudf/source/api_docs/options.rst

@@ -0,0 +1,12 @@
+.. _api.options:
+
+============
+cudf Options
+============
+
+.. autosummary::
+   :toctree: api/
+
+   cudf.get_option
+   cudf.set_option
+   cudf.describe_option

docs/cudf/source/developer_guide/index.md

@@ -5,3 +5,4 @@
 
 library_design
 documentation
+options

docs/cudf/source/developer_guide/options.md

@@ -0,0 +1,22 @@
+# Options
+
+The usage of options is explained in the [user guide](options_user_guide).
+This document provides more explanations on how developers work with options internally.
+
+Options are stored as a dictionary in the `cudf.options` module.
+Each option name is its key in the dictionary.
+The value of the option is an instance of an `Options` object.
+
+An `Options` object has the following attributes:
+- `value`: the current value of the option
+- `description`: a text description of the option
+- `validator`: a boolean function that returns `True` if `value` is valid,
+`False` otherwise.
+
+Developers can use `cudf.options._register_option` to add options to the dictionary.
+{py:func}`cudf.get_option` is provided to get option values from the dictionary.
+
+When testing the behavior of a certain option,
+it is advised to use [`yield` fixture](https://docs.pytest.org/en/7.1.x/how-to/fixtures.html#yield-fixtures-recommended) to set up and clean up the option.
+
+See the [API reference](api.options) for more details.

docs/cudf/source/user_guide/index.md

@@ -12,5 +12,6 @@ groupby
 guide-to-udfs
 cupy-interop
 dask-cudf
+options
 PandasCompat
 ```

docs/cudf/source/user_guide/options.md

@@ -0,0 +1,14 @@
+(options_user_guide)=
+
+# Options
+
+cuDF has an options API to configure and customize global behavior.
+This API complements the [pandas.options](https://pandas.pydata.org/docs/user_guide/options.html) API with features specific to cuDF.
+
+{py:func}`cudf.describe_option` will print the option's description,
+the current value, and the default value.
+When no argument is provided,
+all options are printed.
+To set value to a option, use {py:func}`cudf.set_option`.
+
+See the [API reference](api.options) for more details.

python/cudf/cudf/__init__.py

@@ -82,6 +82,12 @@
 from cudf.utils.dtypes import _NA_REP
 from cudf.utils.utils import set_allocator
 
+from cudf.options import (
+    get_option,
+    set_option,
+    describe_option,
+)
+
 try:
     from ptxcompiler.patch import patch_numba_codegen_if_needed
 except ImportError:
@@ -144,11 +150,13 @@
     "concat",
     "cut",
     "date_range",
+    "describe_option",
     "factorize",
     "from_dataframe",
     "from_dlpack",
     "from_pandas",
     "get_dummies",
+    "get_option",
     "interval_range",
     "isclose",
     "melt",
@@ -163,6 +171,7 @@
     "read_parquet",
     "read_text",
     "set_allocator",
+    "set_option",
     "testing",
     "to_datetime",
     "to_numeric",

python/cudf/cudf/options.py

@@ -0,0 +1,114 @@
+# Copyright (c) 2022, NVIDIA CORPORATION.
+
+from dataclasses import dataclass
+from typing import Any, Callable, Dict, Optional
+
+
+@dataclass
+class Option:
+    default: Any
+    value: Any
+    description: str
+    validator: Callable
+
+
+_OPTIONS: Dict[str, Option] = {}
+
+
+def _register_option(
+    name: str, default_value: Any, description: str, validator: Callable
+):
+    """Register an option.
+
+    Parameters
+    ----------
+    name : str
+        The name of the option.
+    default_value : Any
+        The default value of the option.
+    description : str
+        A text description of the option.
+    validator : Callable
+        Called on the option value to check its validity. Should raise an
+        error if the value is invalid.
+
+    Raises
+    ------
+    BaseException
+        Raised by validator if the value is invalid.
+    """
+    validator(default_value)
+    _OPTIONS[name] = Option(
+        default_value, default_value, description, validator
+    )
+
+
+def get_option(name: str) -> Any:
+    """Get the value of option.
+
+    Parameters
+    ----------
+    key : str
+        The name of the option.
+
+    Returns
+    -------
+    The value of the option.
+
+    Raises
+    ------
+    KeyError
+        If option ``name`` does not exist.
+    """
+    try:
+        return _OPTIONS[name].value
+    except KeyError:
+        raise KeyError(f'"{name}" does not exist.')
+
+
+def set_option(name: str, val: Any):
+    """Set the value of option.
+
+    Parameters
+    ----------
+    name : str
+        The name of the option.
+    val : Any
+        The value to set.
+
+    Raises
+    ------
+    KeyError
+        If option ``name`` does not exist.
+    BaseException
+        Raised by validator if the value is invalid.
+    """
+    try:
+        option = _OPTIONS[name]
+    except KeyError:
+        raise KeyError(f'"{name}" does not exist.')
+    option.validator(val)
+    option.value = val
+
+
+def _build_option_description(name, opt):
+    return (
+        f"{name}:\n"
+        f"\t{opt.description}\n"
+        f"\t[Default: {opt.default}] [Current: {opt.value}]"
+    )
+
+
+def describe_option(name: Optional[str] = None):
+    """Prints the description of an option.
+
+    If `name` is unspecified, prints the description of all available options.
+
+    Parameters
+    ----------
+    name : Optional[str]
+        The name of the option.
+    """
+    names = _OPTIONS.keys() if name is None else [name]
+    for name in names:
+        print(_build_option_description(name, _OPTIONS[name]))

python/cudf/cudf/tests/test_options.py

@@ -0,0 +1,77 @@
+# Copyright (c) 2022, NVIDIA CORPORATION.
+
+from contextlib import redirect_stdout
+from io import StringIO
+
+import pytest
+
+import cudf
+
+
+@pytest.fixture(scope="module", autouse=True)
+def empty_option_environment():
+    old_option_environment = cudf.options._OPTIONS
+    cudf.options._OPTIONS = {}
+    yield
+    cudf.options._OPTIONS = old_option_environment
+
+
+@pytest.fixture
+def odd_option(empty_option_environment):
+    def validator(x):
+        if not x % 2 == 1:
+            raise ValueError(f"Invalid option value {x}")
+
+    cudf.options._register_option(
+        "odd_option",
+        1,
+        "An odd option.",
+        validator,
+    )
+    yield
+    del cudf.options._OPTIONS["odd_option"]
+
+
+@pytest.fixture
+def even_option(empty_option_environment):
+    def validator(x):
+        if not x % 2 == 0:
+            raise ValueError(f"Invalid option value {x}")
+
+    cudf.options._register_option(
+        "even_option", 0, "An even option.", validator
+    )
+    yield
+    del cudf.options._OPTIONS["even_option"]
+
+
+def test_option_get_set(odd_option):
+    assert cudf.get_option("odd_option") == 1
+    cudf.set_option("odd_option", 101)
+    assert cudf.get_option("odd_option") == 101
+
+
+def test_option_set_invalid(odd_option):
+    with pytest.raises(ValueError, match="Invalid option value 0"):
+        cudf.set_option("odd_option", 0)
+
+
+def test_option_description(odd_option):
+    s = StringIO()
+    with redirect_stdout(s):
+        cudf.describe_option("odd_option")
+    s.seek(0)
+    expected = "odd_option:\n\tAn odd option.\n\t[Default: 1] [Current: 1]\n"
+    assert expected == s.read()
+
+
+def test_option_description_all(odd_option, even_option):
+    s = StringIO()
+    with redirect_stdout(s):
+        cudf.describe_option()
+    s.seek(0)
+    expected = (
+        "odd_option:\n\tAn odd option.\n\t[Default: 1] [Current: 1]\n"
+        "even_option:\n\tAn even option.\n\t[Default: 0] [Current: 0]\n"
+    )
+    assert expected == s.read()