add deprecation warnings for various IO methods

python/dask_cudf/dask_cudf/__init__.py

@@ -1,7 +1,9 @@
 # Copyright (c) 2018-2024, NVIDIA CORPORATION.
 
-from dask import config
+import warnings
+from importlib import import_module
 
+from dask import config
 import dask.dataframe as dd
 from dask.dataframe import from_delayed  # noqa: E402
 
@@ -34,26 +36,44 @@ def read_parquet(*args, **kwargs):
         return dd.read_parquet(*args, **kwargs)
 
 
-def raise_not_implemented_error(attr_name):
+def _deprecated_api(old_api, new_api=None, rec=None):
     def inner_func(*args, **kwargs):
+        if new_api:
+            # Use alternative
+            msg = f"{old_api} is now deprecated. "
+            msg += rec or f"Please use {new_api} instead."
+            warnings.warn(msg, FutureWarning)
+            new_attr = new_api.split(".")
+            module = import_module(".".join(new_attr[:-1]))
+            return getattr(module, new_attr[-1])(*args, **kwargs)
+
+        # No alternative - raise an error
         raise NotImplementedError(
-            f"Top-level {attr_name} API is not available for dask-expr."
+            f"{old_api} is no longer supported. " + (rec or "")
         )
 
     return inner_func
 
 
 if QUERY_PLANNING_ON:
     from ._expr.expr import _patch_dask_expr
+    from . import io  # noqa: F401
 
-    groupby_agg = raise_not_implemented_error("groupby_agg")
+    groupby_agg = _deprecated_api("dask_cudf.groupby_agg")
     read_text = DataFrame.read_text
-    to_orc = raise_not_implemented_error("to_orc")
     _patch_dask_expr()
 
 else:
     from ._legacy.groupby import groupby_agg  # noqa: F401
-    from ._legacy.io import read_text, to_orc  # noqa: F401
+    from ._legacy.io import read_text  # noqa: F401
+    from . import io  # noqa: F401
+
+
+to_orc = _deprecated_api(
+    "dask_cudf.to_orc",
+    new_api="dask_cudf._legacy.io.to_orc",
+    rec="Please use DataFrame.to_orc instead.",
+)
 
 
 __all__ = [

python/dask_cudf/dask_cudf/_expr/collection.py

@@ -153,7 +153,10 @@ def groupby(
         )
 
     def to_orc(self, *args, **kwargs):
-        return self.to_legacy_dataframe().to_orc(*args, **kwargs)
+        from dask_cudf._legacy.io import to_orc
+
+        return to_orc(self, *args, **kwargs)
+        # return self.to_legacy_dataframe().to_orc(*args, **kwargs)
 
     @staticmethod
     def read_text(*args, **kwargs):

python/dask_cudf/dask_cudf/_legacy/core.py

@@ -1,7 +1,6 @@
 # Copyright (c) 2018-2024, NVIDIA CORPORATION.
 
 import math
-import textwrap
 import warnings
 
 import numpy as np
@@ -57,9 +56,6 @@ def __repr__(self):
         return s % (type(self).__name__, len(self.dask), self.npartitions)
 
 
-concat = dd.concat
-
-
 normalize_token.register(_Frame, lambda a: a._name)
 
 
@@ -688,50 +684,6 @@ def reduction(
     return dd.core.new_dd_object(graph, b, meta, (None, None))
 
 
-@_dask_cudf_performance_tracking
-def from_cudf(data, npartitions=None, chunksize=None, sort=True, name=None):
-    from dask_cudf import QUERY_PLANNING_ON
-
-    if isinstance(getattr(data, "index", None), cudf.MultiIndex):
-        raise NotImplementedError(
-            "dask_cudf does not support MultiIndex Dataframes."
-        )
-
-    # Dask-expr doesn't support the `name` argument
-    name = {}
-    if not QUERY_PLANNING_ON:
-        name = {
-            "name": name
-            or ("from_cudf-" + tokenize(data, npartitions or chunksize))
-        }
-
-    return dd.from_pandas(
-        data,
-        npartitions=npartitions,
-        chunksize=chunksize,
-        sort=sort,
-        **name,
-    )
-
-
-from_cudf.__doc__ = (
-    textwrap.dedent(
-        """
-        Create a :class:`.DataFrame` from a :class:`cudf.DataFrame`.
-
-        This function is a thin wrapper around
-        :func:`dask.dataframe.from_pandas`, accepting the same
-        arguments (described below) excepting that it operates on cuDF
-        rather than pandas objects.\n
-        """
-    )
-    # TODO: `dd.from_pandas.__doc__` is empty when
-    # `DASK_DATAFRAME__QUERY_PLANNING=True`
-    # since dask-expr does not provide a docstring for from_pandas.
-    + textwrap.dedent(dd.from_pandas.__doc__ or "")
-)
-
-
 for name in (
     "add",
     "sub",

python/dask_cudf/dask_cudf/core.py

@@ -1,6 +1,12 @@
 # Copyright (c) 2020-2024, NVIDIA CORPORATION.
 
+import textwrap
+
 import dask.dataframe as dd
+from dask.tokenize import tokenize
+
+import cudf
+from cudf.utils.performance_tracking import _dask_cudf_performance_tracking
 
 # This module provides backward compatibility for legacy import patterns.
 if dd.DASK_EXPR_ENABLED:
@@ -12,4 +18,49 @@
 else:
     from dask_cudf._legacy.core import DataFrame, Index, Series  # noqa: F401
 
-from dask_cudf._legacy.core import concat, from_cudf  # noqa: F401
+
+concat = dd.concat  # noqa: F401
+
+
+@_dask_cudf_performance_tracking
+def from_cudf(data, npartitions=None, chunksize=None, sort=True, name=None):
+    from dask_cudf import QUERY_PLANNING_ON
+
+    if isinstance(getattr(data, "index", None), cudf.MultiIndex):
+        raise NotImplementedError(
+            "dask_cudf does not support MultiIndex Dataframes."
+        )
+
+    # Dask-expr doesn't support the `name` argument
+    name = {}
+    if not QUERY_PLANNING_ON:
+        name = {
+            "name": name
+            or ("from_cudf-" + tokenize(data, npartitions or chunksize))
+        }
+
+    return dd.from_pandas(
+        data,
+        npartitions=npartitions,
+        chunksize=chunksize,
+        sort=sort,
+        **name,
+    )
+
+
+from_cudf.__doc__ = (
+    textwrap.dedent(
+        """
+        Create a :class:`.DataFrame` from a :class:`cudf.DataFrame`.
+
+        This function is a thin wrapper around
+        :func:`dask.dataframe.from_pandas`, accepting the same
+        arguments (described below) excepting that it operates on cuDF
+        rather than pandas objects.\n
+        """
+    )
+    # TODO: `dd.from_pandas.__doc__` is empty when
+    # `DASK_DATAFRAME__QUERY_PLANNING=True`
+    # since dask-expr does not provide a docstring for from_pandas.
+    + textwrap.dedent(dd.from_pandas.__doc__ or "")
+)

python/dask_cudf/dask_cudf/io/__init__.py

@@ -1 +1,32 @@
-# Copyright (c) 2018-2024, NVIDIA CORPORATION.
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+from dask_cudf import _deprecated_api
+
+from . import csv, orc, json, parquet, text  # noqa: F401
+
+
+read_csv = _deprecated_api(
+    "dask_cudf.io.read_csv", new_api="dask_cudf.read_csv"
+)
+read_json = _deprecated_api(
+    "dask_cudf.io.read_json", new_api="dask_cudf.read_json"
+)
+read_orc = _deprecated_api(
+    "dask_cudf.io.read_orc", new_api="dask_cudf.read_orc"
+)
+to_orc = _deprecated_api(
+    "dask_cudf.io.to_orc",
+    new_api="dask_cudf._legacy.io.to_orc",
+    rec="Please use the DataFrame.to_orc method instead.",
+)
+read_text = _deprecated_api(
+    "dask_cudf.io.read_text", new_api="dask_cudf.read_text"
+)
+read_parquet = _deprecated_api(
+    "dask_cudf.io.read_parquet", new_api="dask_cudf.read_parquet"
+)
+to_parquet = _deprecated_api(
+    "dask_cudf.io.to_parquet",
+    new_api="dask_cudf._legacy.io.parquet.to_parquet",
+    rec="Please use the DataFrame.to_parquet method instead.",
+)

python/dask_cudf/dask_cudf/io/csv.py

@@ -0,0 +1,8 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+from dask_cudf import _deprecated_api
+
+read_csv = _deprecated_api(
+    "dask_cudf.io.csv.read_csv",
+    new_api="dask_cudf.read_csv",
+)

python/dask_cudf/dask_cudf/io/json.py

@@ -0,0 +1,8 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+from dask_cudf import _deprecated_api
+
+read_json = _deprecated_api(
+    "dask_cudf.io.json.read_json",
+    new_api="dask_cudf.read_json",
+)

python/dask_cudf/dask_cudf/io/orc.py

@@ -0,0 +1,13 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+from dask_cudf import _deprecated_api
+
+read_orc = _deprecated_api(
+    "dask_cudf.io.orc.read_orc",
+    new_api="dask_cudf.read_orc",
+)
+to_orc = _deprecated_api(
+    "dask_cudf.io.orc.to_orc",
+    new_api="dask_cudf._legacy.io.orc.to_orc",
+    rec="Please use the DataFrame.to_orc method instead.",
+)

python/dask_cudf/dask_cudf/io/parquet.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2018-2024, NVIDIA CORPORATION.
+# Copyright (c) 2024, NVIDIA CORPORATION.
 import functools
 
 import pandas as pd
@@ -7,6 +7,8 @@
 
 import cudf
 
+from dask_cudf import _deprecated_api
+
 # Dask-expr imports CudfEngine from this module
 from dask_cudf._legacy.io.parquet import CudfEngine  # noqa: F401
 
@@ -115,3 +117,19 @@ def _tune_up(self, parent):
         if isinstance(parent, CudfFusedParquetIO):
             return
         return parent.substitute(self, CudfFusedParquetIO(self))
+
+
+read_parquet = _deprecated_api(
+    "dask_cudf.io.parquet.read_parquet",
+    new_api="dask_cudf.read_parquet",
+)
+to_parquet = _deprecated_api(
+    "dask_cudf.io.parquet.to_parquet",
+    new_api="dask_cudf._legacy.io.parquet.to_parquet",
+    rec="Please use the DataFrame.to_parquet method instead.",
+)
+create_metadata_file = _deprecated_api(
+    "dask_cudf.io.parquet.create_metadata_file",
+    new_api="dask_cudf._legacy.io.parquet.create_metadata_file",
+    rec="Please raise an issue if this feature is needed.",
+)

python/dask_cudf/dask_cudf/io/text.py

@@ -0,0 +1,8 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+from dask_cudf import _deprecated_api
+
+read_text = _deprecated_api(
+    "dask_cudf.io.text.read_text",
+    new_api="dask_cudf.read_text",
+)