MNT modernise linters

.gitignore

@@ -4,6 +4,11 @@ hub/
 __pycache__/
 .pytest_cache/
 .mypy_cache/
+.ruff_cache/
+
+# pixi environments
+.pixi/*
+!.pixi/config.toml
 *.py[cod]
 *$py.class
 py36-64/

.pre-commit-config.yaml

@@ -9,20 +9,17 @@ repos:
         exclude: skops/card/tests/examples
     -   id: check-case-conflict
     -   id: check-merge-conflict
--   repo: https://github.com/psf/black
-    rev: 23.9.1
-    hooks:
-    -   id: black
 -   repo: https://github.com/astral-sh/ruff-pre-commit
     # Ruff version.
-    rev: v0.0.292
+    rev: v0.16.6
     hooks:
-    -   id: ruff
-        args: ["--fix", "--show-source"]
--   repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.6.0
+    -   id: ruff-check
+        args: ["--fix", "--output-format=full"]
+    -   id: ruff-format
+-   repo: https://github.com/facebook/pyrefly-pre-commit
+    # Pyrefly version. The hook runs `pyrefly check` on the whole project, using
+    # the [tool.pyrefly] section in pyproject.toml.
+    rev: 1.2.0
     hooks:
-    -   id: mypy
-        args: [--config-file=pyproject.toml]
-        exclude: "spaces/"
-        additional_dependencies: [types-requests>=2.28.5]
+    -   id: pyrefly-check
+        name: pyrefly (type checking)

CONTRIBUTING.rst

@@ -72,8 +72,11 @@ And you can choose an environment to run the tests with:
 
    pixi run -e ci-sklearn15 tests
 
-In order to setup ``pre-commit`` hooks, you'd need to run the linter once, ignoring
-the outputs:
+We use `ruff <https://docs.astral.sh/ruff/>`_ for formatting and linting, and
+`pyrefly <https://pyrefly.org/>`_ for type checking. Both are run through
+``pre-commit``, with their versions pinned in ``.pre-commit-config.yaml``. In order
+to setup the ``pre-commit`` hooks, you'd need to run the linter once, ignoring the
+outputs:
 
 .. code:: bash
 

README.rst

@@ -1,6 +1,6 @@
 .. -*- mode: rst -*-
 
-|readthedocs| |github-actions| |Codecov| |PyPI| |Black|
+|readthedocs| |github-actions| |Codecov| |PyPI| |Ruff|
 
 .. |readthedocs| image:: https://readthedocs.org/projects/skops/badge/?version=latest&style=flat
     :target: https://skops.readthedocs.io/en/latest/
@@ -18,9 +18,9 @@
     :target: https://pypi.org/project/skops
     :alt: PyPi
 
-.. |Black| image:: https://img.shields.io/badge/code%20style-black-000000.svg
-    :target: https://github.com/psf/black
-    :alt: Black
+.. |Ruff| image:: https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json
+    :target: https://github.com/astral-sh/ruff
+    :alt: Ruff
 
 .. image:: https://raw.githubusercontent.com/skops-dev/skops/main/docs/images/logo.png
   :width: 500

examples/plot_california_housing.py

@@ -1010,8 +1010,8 @@
 # %%
 
 fig, ax = plt.subplots()
-for weight in params["weights"]:  # type: ignore
-    for p in params["p"]:  # type: ignore
+for weight in params["weights"]:
+    for p in params["p"]:
         query = f"param_weights=='{weight}' & param_p=={p}"
         df_subset = df_cv.query(query)
         df_subset.plot(

pyproject.toml

@@ -57,15 +57,25 @@ Repository = "http://github.com/skops-dev/skops"
 Issues = "http://github.com/skops-dev/skops/issues"
 Changelog = "https://skops.readthedocs.io/en/stable/changes.html"
 
-[tool.black]
+[tool.ruff]
 line-length = 88
-target_version = ['py310', 'py311', 'py312']
-preview = true
+extend-exclude = [
+    # locally installed pixi environments
+    ".pixi",
+    # These markdown files are data (model card template and test fixtures) and
+    # must be kept verbatim; ruff would otherwise format the python code blocks in
+    # them.
+    "skops/card/default_template.md",
+    "skops/card/tests/examples",
+]
 
-[tool.ruff]
-# all rules can be found here: https://beta.ruff.rs/docs/rules/
+[tool.ruff.lint]
+# all rules can be found here: https://docs.astral.sh/ruff/rules/
 select = ["E", "F", "W", "I"]
 
+[tool.ruff.lint.isort]
+known-first-party = ["skops"]
+
 [tool.pytest.ini_options]
 doctest_optionflags = "NORMALIZE_WHITESPACE ELLIPSIS"
 filterwarnings = [
@@ -88,10 +98,13 @@ omit = [
     "skops/conftest.py",
 ]
 
-[tool.mypy]
-exclude = "(\\w+/)*test_\\w+\\.py$|spaces/skops_model_card_creator|old"
-ignore_missing_imports = true
-no_implicit_optional = true
+[tool.pyrefly]
+project-includes = ["skops", "examples", "scripts", "docs/conf.py"]
+# Replace imports that cannot be resolved (e.g. numpy, scikit-learn, which are not
+# installed in the pre-commit environment) with `Any` instead of erroring.
+ignore-missing-imports = ["*"]
+# The `legacy` preset is the one pyrefly recommends for projects migrating from mypy.
+preset = "legacy"
 
 [tool.pixi.workspace]
 channels = ["conda-forge"]
@@ -124,8 +137,6 @@ fairlearn = ">=0.7.0"
 [tool.pixi.feature.tests.dependencies]
 pytest = ">=7"
 pytest-cov = ">=2.9.0"
-flake8 = ">=3.8.2"
-types-requests = ">=2.28.5"
 flaky = ">=3.7.0"
 rich = ">=12"
 matplotlib = ">=3.3"

skops/card/_markup.py

@@ -287,7 +287,7 @@ def _parse_div(self, item) -> str:
                     kvparts.append(k)
                 else:
                     kvparts.append(f'{k}="{v}"')
-            tags.append(f' {" ".join(kvparts)}')
+            tags.append(f" {' '.join(kvparts)}")
         tags.append(">")
 
         start = "".join(tags)

skops/card/_model_card.py

@@ -193,7 +193,7 @@ def _check_table(self) -> None:
             self._is_pandas_df = False
 
         if self._is_pandas_df:
-            ncols = len(self.table.columns)  # type: ignore
+            ncols = len(self.table.columns)  # pyrefly: ignore[missing-attribute]
         else:
             ncols = len(self.table)
         if ncols == 0:
@@ -218,7 +218,7 @@ def format(self) -> str:
 
     def __repr__(self) -> str:
         if self._is_pandas_df:
-            nrows, ncols = self.table.shape  # type: ignore
+            nrows, ncols = self.table.shape  # pyrefly: ignore[missing-attribute]
         else:
             # table cannot be empty, so no checks needed here
             ncols = len(self.table)

skops/card/tests/test_card.py

@@ -220,7 +220,7 @@ def test_model_caching(skops_model_card, iris_skops_file, destination_path):
         assert mock_load_model.call_count == 1  # cached call
 
 
-CUSTOM_TEMPLATES = [None, {}, {"A Title", "Another Title", "A Title/A Section"}]  # type: ignore
+CUSTOM_TEMPLATES = [None, {}, {"A Title", "Another Title", "A Title/A Section"}]
 
 
 class TestAddModelPlot:

skops/card/tests/test_parser.py

@@ -6,7 +6,7 @@
 from unittest.mock import Mock, patch
 
 import pytest
-import yaml  # type: ignore
+import yaml
 
 from skops.card import parse_modelcard
 from skops.card._parser import PandocParser, check_pandoc_installed

skops/io/_audit.py

@@ -287,8 +287,8 @@ def __init__(
         self.trusted = self._get_trusted(trusted, default=[])
         # TODO: deal with case that __id__ is unknown or prevent it from
         # happening
-        self.cached = load_context.get_object(state.get("__id__"))  # type: ignore
-        self.children = {}  # type: ignore
+        self.cached = load_context.get_object(state.get("__id__"))  # pyrefly: ignore[bad-argument-type]
+        self.children = {}
 
     def _construct(self):
         # TODO: FIXME This causes a recursion error when loading a cached

skops/io/_general.py

@@ -51,7 +51,7 @@ def dict_get_state(obj: Any, save_context: SaveContext) -> dict[str, Any]:
             continue
         if np.isscalar(key) and hasattr(key, "item"):
             # convert numpy value to python object
-            key = key.item()  # type: ignore
+            key = key.item()
         content[key] = get_state(value, save_context)
     res["content"] = content
     res["key_types"] = key_types
@@ -228,7 +228,7 @@ def isnamedtuple(self, t) -> bool:
         # This is needed since namedtuples need to have the args when
         # initialized.
         b = t.__bases__
-        if len(b) != 1 or b[0] != tuple:
+        if len(b) != 1 or b[0] is not tuple:
             return False
         f = getattr(t, "_fields", None)
         if not isinstance(f, tuple):
@@ -315,7 +315,7 @@ def _construct(self):
         namespace = self.children["namespace"].construct()
         instance = partial(func, *args, **kwds)  # always use partial, not a subclass
         # partial always has __setstate__
-        instance.__setstate__((func, args, kwds, namespace))  # type: ignore
+        instance.__setstate__((func, args, kwds, namespace))
         return instance
 
 
@@ -492,7 +492,7 @@ def _construct(self):
         # bypasses the __init__, and then we set the attributes. This solves the
         # issue of required init arguments. Note that the instance created here
         # might not be valid until all its attributes have been set below.
-        instance = cls.__new__(cls)  # type: ignore
+        instance = cls.__new__(cls)
 
         if not self.children["attrs"]:
             # nothing more to do
@@ -553,9 +553,9 @@ def get_unsafe_set(self) -> set[str]:
         res = super().get_unsafe_set()
         obj_node = self.children["obj"]
         res.add(
-            obj_node.module_name  # type: ignore
+            obj_node.module_name  # pyrefly: ignore[missing-attribute]
             + "."
-            + obj_node.class_name  # type: ignore
+            + obj_node.class_name  # pyrefly: ignore[missing-attribute]
             + "."
             + self.children["func"]
         )

skops/io/_numpy.py

@@ -69,9 +69,7 @@ def __init__(
     ) -> None:
         super().__init__(state, load_context, trusted)
         self.type = state["type"]
-        self.trusted = self._get_trusted(
-            trusted, [np.ndarray] + NUMPY_DTYPE_TYPE_NAMES  # type: ignore
-        )
+        self.trusted = self._get_trusted(trusted, [np.ndarray] + NUMPY_DTYPE_TYPE_NAMES)
         if self.type == "numpy":
             self.children = {
                 "content": io.BytesIO(load_context.src.read(state["file"]))
@@ -349,11 +347,11 @@ def _construct(self):
     # a C wrapper: https://github.com/numpy/numpy/pull/23020
     try:
         # numpy>=2
-        from numpy._core._multiarray_umath import (  # type: ignore
+        from numpy._core._multiarray_umath import (
             _ArrayFunctionDispatcher,
         )
     except ImportError:
-        from numpy.core._multiarray_umath import (  # type: ignore
+        from numpy.core._multiarray_umath import (
             _ArrayFunctionDispatcher,
         )
 

skops/io/_persist.py

@@ -225,8 +225,8 @@ def get_untrusted_types(
     if data:
         content = io.BytesIO(data)
     else:
-        # mypy doesn't understand that file cannot be None here, thus ignore
-        content = file  # type: ignore
+        # the type checker doesn't understand that file cannot be None here
+        content = file  # pyrefly: ignore[bad-assignment]
 
     with ZipFile(content, "r") as zip_file:
         schema = json.loads(zip_file.read("schema.json"))

skops/io/_protocol.py

@@ -23,4 +23,5 @@
 version Y instead.
 
 """
+
 PROTOCOL = 2

skops/io/_sklearn.py

@@ -234,10 +234,10 @@ def __init__(
 
 def loss_get_state(obj: Any, save_context: SaveContext) -> dict[str, Any]:
     reduce = obj.__reduce__()
-    if type(obj) == reduce[0]:
+    if type(obj) is reduce[0]:
         state = reduce_get_state(obj, save_context)
         state["__loader__"] = "LossNode"
-    elif type(obj) == reduce[1][0]:
+    elif type(obj) is reduce[1][0]:
         # The output is commonly of the form:
         # >>> CyPinballLoss(1).__reduce__()
         # (<cyfunction __pyx_unpickle_CyPinballLoss at 0x7b1d00099ff0>,
@@ -363,10 +363,11 @@ def _construct(self):
 # TODO: remove once support for sklearn<1.2 is dropped.
 # Starting from sklearn 1.2, _DictWithDeprecatedKeys is removed as it's no
 # longer needed for GraphicalLassoCV, see #187.
-if _DictWithDeprecatedKeys is not None:
+# skops requires scikit-learn>=1.2, so this block never runs in the test matrix.
+if _DictWithDeprecatedKeys is not None:  # pragma: no cover
     GET_STATE_DISPATCH_FUNCTIONS.append(
         (_DictWithDeprecatedKeys, _DictWithDeprecatedKeys_get_state)
     )
-    NODE_TYPE_MAPPING[
-        ("_DictWithDeprecatedKeysNode", PROTOCOL)
-    ] = _DictWithDeprecatedKeysNode  # type: ignore
+    NODE_TYPE_MAPPING[("_DictWithDeprecatedKeysNode", PROTOCOL)] = (
+        _DictWithDeprecatedKeysNode
+    )

skops/io/tests/_utils.py

@@ -55,7 +55,7 @@ def _is_builtin(val):
             _assert_generic_objects_equal(subval1, subval2, path=f"{path}[]")
             return
 
-    assert type(val1) == type(val2), f"Path: type({path})"
+    assert type(val1) is type(val2), f"Path: type({path})"
     if hasattr(val1, "__dict__"):
         assert_params_equal(val1.__dict__, val2.__dict__, path=f"{path}.__dict__")
     elif _is_builtin(val1):
@@ -84,7 +84,7 @@ def _assert_vals_equal(val1, val2, path=""):
         # Some objects return a tuple of parameters, others a dict.
         state1 = val1.__getstate__()
         state2 = val2.__getstate__()
-        assert type(state1) == type(state2), f"Path: {path}"
+        assert type(state1) is type(state2), f"Path: {path}"
         if isinstance(state1, tuple):
             _assert_tuples_equal(state1, state2, path=path)
         else:
@@ -98,12 +98,12 @@ def _assert_vals_equal(val1, val2, path=""):
             # for arrays with at least 2 dimensions, check that contiguity is
             # preserved, but only if the array is not a view
             if val1.squeeze().ndim > 1 and val1.flags["OWNDATA"]:
-                assert (
-                    val1.flags["C_CONTIGUOUS"] is val2.flags["C_CONTIGUOUS"]
-                ), f"Path: {path}.flags"
-                assert (
-                    val1.flags["F_CONTIGUOUS"] is val2.flags["F_CONTIGUOUS"]
-                ), f"Path: {path}.flags"
+                assert val1.flags["C_CONTIGUOUS"] is val2.flags["C_CONTIGUOUS"], (
+                    f"Path: {path}.flags"
+                )
+                assert val1.flags["F_CONTIGUOUS"] is val2.flags["F_CONTIGUOUS"], (
+                    f"Path: {path}.flags"
+                )
             if val1.dtype == object:
                 assert val2.dtype == object, f"Path: {path}.dtype"
                 assert val1.shape == val2.shape, f"Path: {path}.shape"
@@ -166,7 +166,7 @@ def assert_params_equal(params1, params2, path=""):
             warnings.filterwarnings("ignore", category=FutureWarning, module="sklearn")
             val1, val2 = params1[key], params2[key]
             subpath = f"{path}[{key}]"
-        assert type(val1) == type(val2), f"Path: type({subpath})"
+        assert type(val1) is type(val2), f"Path: type({subpath})"
 
         if _is_steps_like(val1):
             # Deal with Pipeline.steps, FeatureUnion.transformer_list, etc.