[2/N] Apply py39 ruff fixes (pytorch#141938)

Fixes #ISSUE_NUMBER

Pull Request resolved: pytorch#141938
Approved by: https://github.com/ezyang

Author: cyyever

tools/pyi/gen_pyi.py

@@ -1063,13 +1063,6 @@ def replace_special_case(hint: str) -> str:
         # NB: Keep this in sync with enum in aten/src/ATen/core/Reduction.h
         hint = hint.replace("at::Reduction::Mean", "1")
         hint = hint.replace(": Tensor = None", ": Optional[Tensor] = None")
-        # Match both:
-        # ": Union[Tensor, Tuple[Tensor, ...], List[Tensor]] = None"
-        # ": Union[Tuple[Tensor, ...], List[Tensor]] = None"
-        hint = hint.replace(
-            "Tuple[Tensor, ...], List[Tensor]] = None",
-            "Tuple[Tensor, ...], List[Tensor], None] = None",
-        )
         return hint
 
     docstrs = gather_docstrs()

torchgen/_autoheuristic/ah_tree.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any, Optional
 
 import numpy as np
 from sklearn.tree import _tree  # type: ignore[import-untyped]
@@ -34,10 +34,10 @@ class DecisionTree:
     does not seem to be easy with sklearn.
     """
 
-    def __init__(self, sklearn_tree: Any, feature_names: List[str]) -> None:
+    def __init__(self, sklearn_tree: Any, feature_names: list[str]) -> None:
         self.feature_names = feature_names
         self.root = self._convert_sklearn_tree(sklearn_tree.tree_)
-        self.classes_: List[str] = sklearn_tree.classes_
+        self.classes_: list[str] = sklearn_tree.classes_
 
     def _convert_sklearn_tree(
         self, sklearn_tree: Any, node_id: int = 0
@@ -193,9 +193,9 @@ def _apply_single(self, x: Any) -> int:
 
     def codegen(
         self,
-        dummy_col_2_col_val: Dict[str, Tuple[str, Any]],
-        lines: List[str],
-        unsafe_leaves: List[int],
+        dummy_col_2_col_val: dict[str, tuple[str, Any]],
+        lines: list[str],
+        unsafe_leaves: list[int],
     ) -> None:
         # generates python code for the decision tree
         def codegen_node(node: DecisionTreeNode, depth: int) -> None:
@@ -223,7 +223,7 @@ def codegen_node(node: DecisionTreeNode, depth: int) -> None:
                 codegen_node(node.right, depth + 1)
 
         def handle_leaf(
-            node: DecisionTreeNode, indent: str, unsafe_leaves: List[int]
+            node: DecisionTreeNode, indent: str, unsafe_leaves: list[int]
         ) -> str:
             """
             This generates the code for a leaf node in the decision tree. If the leaf is unsafe, the learned heuristic

torchgen/_autoheuristic/benchmark_runner.py

@@ -2,7 +2,7 @@
 import random
 import time
 from abc import abstractmethod
-from typing import Any, Tuple
+from typing import Any
 
 from tqdm import tqdm  # type: ignore[import-untyped]
 
@@ -71,7 +71,7 @@ def run(self) -> None:
     def run_benchmark(self, *args: Any) -> None: ...
 
     @abstractmethod
-    def create_input(self) -> Tuple[Any, ...]: ...
+    def create_input(self) -> tuple[Any, ...]: ...
 
     def main(self, num_samples: int, num_reps: int) -> None:
         for _ in tqdm(range(num_samples)):

torchgen/_autoheuristic/benchmark_utils.py

@@ -1,10 +1,10 @@
 import random
-from typing import Any, Tuple
+from typing import Any
 
 import torch
 
 
-def transpose_tensors(p_transpose_both: float = 0.05) -> Tuple[bool, bool]:
+def transpose_tensors(p_transpose_both: float = 0.05) -> tuple[bool, bool]:
     transpose_both = random.choices(
         [True, False], [p_transpose_both, 1 - p_transpose_both]
     )[0]
@@ -31,7 +31,7 @@ def get_mm_tensors(
     transpose_right: bool,
     dtype_left: Any,
     dtype_right: Any,
-) -> Tuple[Any, Any]:
+) -> tuple[Any, Any]:
     if transpose_left:
         a = torch.randn(k, m, dtype=dtype_left).t()
     else:

torchgen/_autoheuristic/merge_data.py

@@ -1,14 +1,13 @@
 import sys
-from typing import List
 
 
-def merge_txt_files(file_list: List[str], output_file: str) -> None:
+def merge_txt_files(file_list: list[str], output_file: str) -> None:
     if not file_list:
         print("No input files provided.")
         return
 
-    metadata: List[str] = []
-    content: List[str] = []
+    metadata: list[str] = []
+    content: list[str] = []
 
     # Read metadata and content from all files
     for file_path in file_list:

torchgen/_autoheuristic/mixed_mm/gen_data_mixed_mm.py

@@ -6,7 +6,7 @@
 
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
-from typing import Any, Tuple
+from typing import Any
 
 from benchmark_runner import BenchmarkRunner  # type: ignore[import-not-found]
 from benchmark_utils import (  # type: ignore[import-not-found]
@@ -33,7 +33,7 @@ class BenchmarkRunnerMixedMM(BenchmarkRunner):  # type: ignore[misc, no-any-unim
     def __init__(self) -> None:
         super().__init__("mixed_mm")
 
-    def create_input(self) -> Tuple[Any, ...]:
+    def create_input(self) -> tuple[Any, ...]:
         dtype1, dtype2 = self.get_dtypes()
         m, k, n = self.get_m_k_n(dtype1)
         transpose_left, transpose_right = False, True
@@ -109,7 +109,7 @@ def get_random_num_small(self) -> int:
         else:
             return get_random_between_pow2(1, 7)
 
-    def get_m_k_n(self, dtype: Any) -> Tuple[int, int, int]:
+    def get_m_k_n(self, dtype: Any) -> tuple[int, int, int]:
         numel_max = 2**31
 
         # repeat until tensors fit in memory

torchgen/_autoheuristic/mixed_mm/test_mixed_mm.py

@@ -24,7 +24,7 @@ def test_mixedmm_a100(self) -> None:
 # fmt: off
 # This file was generated by AutoHeuristic. Do not modify it manually!
 # To regenerate this file, take a look at the steps in the README.md file inside torchgen/_autoheuristic/mixed_mm/
-from typing import List, Optional, Tuple
+from typing import Optional
 
 from torch._inductor.autoheuristic.autoheuristic_utils import (
     AHContext,
@@ -39,7 +39,7 @@ def test_mixedmm_a100(self) -> None:
 class MixedMMA100(LearnedHeuristicDecision):
 
     def __init__(self) -> None:
-        self.choices: List[Choice] = []
+        self.choices: list[Choice] = []
         self.fill_choices()
 
     def check_precondition(self, metadata: AHMetadata, context: AHContext,) -> bool:
@@ -84,7 +84,7 @@ def fill_choices(self) -> None:
     def get_name(self) -> str:
         return 'mixed_mm'
 
-    def get_best_choices(self, context: AHContext) -> Optional[List[Tuple[float, int]]]:
+    def get_best_choices(self, context: AHContext) -> Optional[list[tuple[float, int]]]:
         if str(context.get_value('1LEQmLEQ16')) != 'True':
             if context.get_value('m') <= 32.5:
                 if context.get_value('n') <= 6976.0:
@@ -186,7 +186,7 @@ def test_mixedmm_h100(self) -> None:
 # fmt: off
 # This file was generated by AutoHeuristic. Do not modify it manually!
 # To regenerate this file, take a look at the steps in the README.md file inside torchgen/_autoheuristic/mixed_mm/
-from typing import List, Optional, Tuple
+from typing import Optional
 
 from torch._inductor.autoheuristic.autoheuristic_utils import (
     AHContext,
@@ -201,7 +201,7 @@ def test_mixedmm_h100(self) -> None:
 class MixedMMH100(LearnedHeuristicDecision):
 
     def __init__(self) -> None:
-        self.choices: List[Choice] = []
+        self.choices: list[Choice] = []
         self.fill_choices()
 
     def check_precondition(self, metadata: AHMetadata, context: AHContext,) -> bool:
@@ -245,7 +245,7 @@ def fill_choices(self) -> None:
     def get_name(self) -> str:
         return 'mixed_mm'
 
-    def get_best_choices(self, context: AHContext) -> Optional[List[Tuple[float, int]]]:
+    def get_best_choices(self, context: AHContext) -> Optional[list[tuple[float, int]]]:
         if context.get_value('arith_intensity') <= 15.988086223602295:
             if context.get_value('n') <= 25280.0:
                 if context.get_value('n') <= 1344.0:

torchgen/_autoheuristic/mm/gen_data_mm.py

@@ -6,7 +6,7 @@
 
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
-from typing import Any, Tuple
+from typing import Any
 
 from benchmark_runner import BenchmarkRunner  # type: ignore[import-not-found]
 from benchmark_utils import (  # type: ignore[import-not-found]
@@ -28,7 +28,7 @@ class BenchmarkRunnerMM(BenchmarkRunner):  # type: ignore[misc, no-any-unimporte
     def __init__(self) -> None:
         super().__init__("mm")
 
-    def create_input(self) -> Tuple[Any, ...]:
+    def create_input(self) -> tuple[Any, ...]:
         dtype = random.choices([torch.float32, torch.float16, torch.bfloat16])[0]
         set_precision(dtype)
         m, k, n = self.get_m_k_n(dtype)
@@ -100,7 +100,7 @@ def get_random_dim(self) -> int:
         print(f"random_type {distr_type} not supported")
         sys.exit(1)
 
-    def get_m_k_n(self, dtype: Any) -> Tuple[int, int, int]:
+    def get_m_k_n(self, dtype: Any) -> tuple[int, int, int]:
         numel_max = 2**31
 
         # repeat until tensors fit in memory

torchgen/_autoheuristic/pad_mm/gen_data_pad_mm.py

@@ -5,7 +5,7 @@
 
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
-from typing import Any, Tuple
+from typing import Any
 
 from benchmark_runner import BenchmarkRunner  # type: ignore[import-not-found]
 from benchmark_utils import (  # type: ignore[import-not-found]
@@ -30,7 +30,7 @@ class BenchmarkRunnerPadMM(BenchmarkRunner):  # type: ignore[misc, no-any-unimpo
     def __init__(self) -> None:
         super().__init__("pad_mm")
 
-    def create_input(self) -> Tuple[Any, ...]:
+    def create_input(self) -> tuple[Any, ...]:
         dtype = self.get_dtype()
         set_precision(dtype)
         m, k, n = self.get_m_k_n(dtype)
@@ -113,7 +113,7 @@ def get_random_dim(
     def is_aligned(self, dim: int, align_size: int) -> bool:
         return dim % align_size == 0
 
-    def get_m_k_n(self, dtype: Any) -> Tuple[int, int, int]:
+    def get_m_k_n(self, dtype: Any) -> tuple[int, int, int]:
         uniform = random.choices([True, False])[0]
         align_size = get_alignment_size_dtype(dtype)
 

torchgen/_autoheuristic/train_decision.py

@@ -531,7 +531,7 @@ def gen_predict_fn_def(self):
         """
         Generates the definition of the predict function.
         """
-        return "def get_best_choices(self, context: AHContext) -> Optional[List[Tuple[float, int]]]:"
+        return "def get_best_choices(self, context: AHContext) -> Optional[list[tuple[float, int]]]:"
 
     def codegen_boilerplate(
         self, heuristic_name, opt_name, threshold, shared_memory, device_capa, classes
@@ -545,7 +545,7 @@ def codegen_boilerplate(
 # fmt: off
 # This file was generated by AutoHeuristic. Do not modify it manually!
 # To regenerate this file, take a look at the steps in the README.md file inside torchgen/_autoheuristic/{opt_name}/
-from typing import List, Optional, Tuple
+from typing import Optional
 
 from torch._inductor.autoheuristic.autoheuristic_utils import (
     AHContext,
@@ -560,7 +560,7 @@ def codegen_boilerplate(
 class {heuristic_name}(LearnedHeuristicDecision):
 
     def __init__(self) -> None:
-        self.choices: List[Choice] = []
+        self.choices: list[Choice] = []
         self.fill_choices()
 
 {self.gen_precondition(opt_name, shared_memory, device_capa)}

torchgen/api/lazy.py

@@ -80,7 +80,7 @@ def process_ir_type(
      (3) making cpp-reference types into cpp-value types (e.g. vector instead of IntArrayRef)
 
     (1) converts at::Tensors to lazy::Values (which wrap lazy::Nodes, with which Lazy IR represents tensors.)
-    There is special handling for Optional[Tensor] or List[Tensor], etc- hence 'tensor-like'
+    There is special handling for Optional[Tensor] or list[Tensor], etc- hence 'tensor-like'
 
     This is incomplete- there are assertions in places that it's expected to need to add
     more types as the codegen is used with more operators.

torchgen/api/python.py

@@ -956,14 +956,13 @@ def argument_type_str_pyi(t: Type) -> str:
             ret = "Union[_int, _size]" if t.size is not None else "_size"
         elif t.is_tensor_like():
             # TODO: this doesn't seem right...
-            # Tensor?[] currently translates to Optional[Union[Tuple[Tensor, ...], List[Tensor]]]
-            # It should probably translate to   Union[Tuple[Optional[Tensor], ...], List[Optional[Tensor]]]
-            if isinstance(t.elem, OptionalType):
-                add_optional = True
+            # Tensor?[] currently translates to Optional[Union[tuple[Tensor, ...], list[Tensor]]]
+            # It should probably translate to   Union[tuple[Optional[Tensor], ...], list[Optional[Tensor]]]
+            add_optional = True
             ret = (
-                "Union[Tensor, Tuple[Tensor, ...], List[Tensor]]"
+                "Union[Tensor, tuple[Tensor, ...], list[Tensor]]"
                 if t.size is not None
-                else "Union[Tuple[Tensor, ...], List[Tensor]]"
+                else "Union[tuple[Tensor, ...], list[Tensor]]"
             )
         elif str(t.elem) == "float":
             ret = "Sequence[_float]"
@@ -1001,7 +1000,7 @@ def return_type_str_pyi(t: Type) -> str:
 
     if isinstance(t, ListType):
         inner = return_type_str_pyi(t.elem)
-        return f"Tuple[{inner}, ...]"
+        return f"tuple[{inner}, ...]"
 
     return argument_type_str_pyi(t)
 
@@ -1014,7 +1013,7 @@ def returns_structseq_pyi(signature: PythonSignature) -> tuple[str, str] | None:
         # These types are structseq objects which act like named NamedTuples, but
         # the constructor acts like the constructor of tuple. Using typing.NamedTuple
         # does not allow us to override __init__.
-        seq_type = f"Tuple[{', '.join(python_returns)}]"
+        seq_type = f"tuple[{', '.join(python_returns)}]"
         structseq_def_lines = [
             f"class {structseq_name}({seq_type}):",
         ]
@@ -1038,12 +1037,12 @@ def returns_structseq_pyi(signature: PythonSignature) -> tuple[str, str] | None:
         structseq_def = "\n".join(structseq_def_lines)
         # Example:
         # structseq_def = (
-        #     "class max(Tuple[Tensor, Tensor]):\n"
+        #     "class max(tuple[Tensor, Tensor]):\n"
         #     "    @property\n"
         #     "    def values(self) -> Tensor: ...\n"
         #     "    @property\n"
         #     "    def indices(self) -> Tensor: ...\n"
-        #     "    def __new__(cls, sequence: Tuple[Tensor, Tensor]): ...\n"
+        #     "    def __new__(cls, sequence: tuple[Tensor, Tensor]): ...\n"
         #     "    n_fields: _int = 2",
         #     "    n_sequeunce_fields: _int = 2",
         #     "    n_unnamed_fields: _int = 0",
@@ -1060,7 +1059,7 @@ def returns_str_pyi(signature: PythonSignature) -> str:
 
     python_returns = [return_type_str_pyi(r.type) for r in signature.returns.returns]
     if len(python_returns) > 1:
-        return "Tuple[" + ", ".join(python_returns) + "]"
+        return "tuple[" + ", ".join(python_returns) + "]"
     if len(python_returns) == 1:
         return python_returns[0]
     return "None"

torchgen/context.py

@@ -2,7 +2,7 @@
 
 import contextlib
 import functools
-from typing import Any, Callable, List, Optional, Tuple, TYPE_CHECKING, TypeVar, Union
+from typing import Any, Callable, Optional, TYPE_CHECKING, TypeVar, Union
 
 import torchgen.local as local
 from torchgen.model import (
@@ -39,7 +39,7 @@
     str,
 )
 
-F3 = TypeVar("F3", Tuple[NativeFunction, Any], List[NativeFunction])
+F3 = TypeVar("F3", tuple[NativeFunction, Any], list[NativeFunction])
 
 
 @contextlib.contextmanager

torchgen/executorch/model.py

@@ -1,5 +1,5 @@
 # Represents all kernels used by an Executorch model.
-# It maintains a Dict[OperatorName, Dict[ETKernelKey, BackendMetadata]] structure.
+# It maintains a dict[OperatorName, dict[ETKernelKey, BackendMetadata]] structure.
 
 from __future__ import annotations
 

torchgen/gen_schema_utils.py

@@ -1,4 +1,4 @@
-from typing import Any, Optional, Tuple, Union
+from typing import Any, Optional, Union
 
 from torchgen.model import (
     Annotation,
@@ -80,8 +80,8 @@ class FunctionSchemaGen:
     @staticmethod
     def from_example(
         op_name: str,
-        example_inputs: Tuple[Tuple[str, Any], ...],
-        example_outputs: Tuple[Any, ...],
+        example_inputs: tuple[tuple[str, Any], ...],
+        example_outputs: tuple[Any, ...],
     ) -> FunctionSchema:
         args = []
         for name, inp in example_inputs: