[lint] add flake8-comment linter

.lintrunner.toml

@@ -403,6 +403,20 @@ command = [
 ]
 is_formatter = true
 
+[[linter]]
+code = 'FLAKE8_COMMENT'
+include_patterns = ['**/*.py', '**/*.pyi']
+exclude_patterns = [
+    'torch/onnx/_internal/diagnostics/infra/sarif/**'
+]
+command = [
+    'python3',
+    'tools/linter/adapters/flake8_comment_linter.py',
+    '--',
+    '@{{PATHSFILE}}',
+]
+is_formatter = true
+
 [[linter]]
 code = 'SPACES'
 include_patterns = ['**']

benchmarks/dynamo/huggingface.py

@@ -1,4 +1,7 @@
 #!/usr/bin/env python3
+# Disable the flake warnings for the imports. Flake8 does not provide a way to
+# disable just warning for the entire file. Disabling flake8 entirely.
+# flake8: noqa
 
 import importlib
 import logging
@@ -46,9 +49,6 @@ def pip_install(package):
     subprocess.check_call([sys.executable, "-m", "pip", "install", package])
 
 
-# Disable the flake warnings for the imports. Flake8 does not provide a way to
-# disable just warning for the entire file. Disabling flake8 entirely.
-# flake8: noqa
 imports = [
     "AlbertForPreTraining",
     "AutoConfig",

functorch/examples/dp_cifar10/cifar10_opacus.py

@@ -124,7 +124,6 @@ def test(args, model, test_loader, device):
     return np.mean(top1_acc)
 
 
-# flake8: noqa: C901
 def main():
     args = parse_args()
 

functorch/examples/dp_cifar10/cifar10_transforms.py

@@ -190,7 +190,6 @@ def test(args, model, test_loader, device):
     return np.mean(top1_acc)
 
 
-# flake8: noqa: C901
 def main():
     args = parse_args()
 

test/dynamo/test_misc.py

@@ -1,4 +1,5 @@
 # Owner(s): ["module: dynamo"]
+# flake8: noqa
 # ruff: noqa: F841
 import abc
 import collections
@@ -2772,7 +2773,7 @@ def test_dtypes_no_graphbreaks(self):
             "int",
             np.intp,
             np.int32,
-            np.uint8
+            np.uint8,
             # np.dtype('int')       # XXX: as above
         ]
 
@@ -7578,7 +7579,6 @@ def fn():
     @unittest.skipIf(sys.version_info >= (3, 13), "feature landed in 3.13")
     def test_get_instruction_source_311(self):
         def f():
-            # flake8: noqa
             # fmt: off
             # test binary ops
             a = ( b   )   +   c
@@ -7593,7 +7593,7 @@ def f():
             << (
 
                 c  # test
-                \
+
             )  # test
             )
 
@@ -8091,8 +8091,7 @@ def run_fn():
 
     def test_torch_compile_ctx_on_forward_and_training_step(self):
         class MyModel(torch.nn.Module):
-            def forward(self):
-                ...
+            def forward(self): ...
 
             def training_step(self):
                 self()

test/functorch/test_aotdispatch.py

@@ -5750,8 +5750,7 @@ def fn(x):
 
 class GradsNoForceContiguousContextManager(ContextDecorator):
     def __enter__(self):
-        # flake8: noqa: TOR901
-        self.lib = torch.library.Library("_mylib", "FRAGMENT")
+        self.lib = torch.library.Library("_mylib", "FRAGMENT")  # noqa: TOR901
         self.d = {
             torch.channels_last: 0,
             torch.contiguous_format: 0,

tools/linter/adapters/flake8_comment_linter.py

@@ -0,0 +1,112 @@
+"""
+FLAKE8_COMMENT: Checks files to make sure # flake8: noqa is only at the top of files
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import logging
+import re
+import sys
+import tokenize
+from enum import Enum
+from io import StringIO
+from typing import NamedTuple
+
+
+LINTER_CODE = "FLAKE8_COMMENT"
+
+
+class LintSeverity(str, Enum):
+    ERROR = "error"
+    WARNING = "warning"
+    ADVICE = "advice"
+    DISABLED = "disabled"
+
+
+class LintMessage(NamedTuple):
+    path: str | None
+    line: int | None
+    char: int | None
+    code: str
+    severity: LintSeverity
+    name: str
+    original: str | None
+    replacement: str | None
+    description: str | None
+
+
+def check_file(filename: str) -> LintMessage | None:
+    logging.debug("Checking file %s", filename)
+
+    pattern = r"^# flake8:\s*noqa"
+    is_start_of_file = True
+
+    with open(filename, encoding="utf-8") as f:
+        original = f.read()
+
+        for token in tokenize.generate_tokens(StringIO(original).readline):
+            if (
+                token.type != tokenize.COMMENT and token.type != tokenize.NL
+            ) and is_start_of_file:
+                is_start_of_file = False
+
+            if (
+                token.type == tokenize.COMMENT
+                and not is_start_of_file
+                and re.search(pattern, token.string)
+            ):
+                replacement_lines = original.splitlines()
+                replacement_lines[token.start[0] - 1] = ""
+                replacement_lines.insert(0, "# flake8: noqa")
+                return LintMessage(
+                    path=filename,
+                    line=None,
+                    char=None,
+                    code=LINTER_CODE,
+                    severity=LintSeverity.ERROR,
+                    name="mid-file '# flake8: noqa'",
+                    original=original,
+                    replacement="\n".join(replacement_lines),
+                    description="'# flake8: noqa' in the middle of the file ",
+                )
+    return None
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="native functions linter",
+        fromfile_prefix_chars="@",
+    )
+    parser.add_argument(
+        "--verbose",
+        action="store_true",
+        help="location of native_functions.yaml",
+    )
+    parser.add_argument(
+        "filenames",
+        nargs="+",
+        help="paths to lint",
+    )
+
+    args = parser.parse_args()
+
+    logging.basicConfig(
+        format="<%(threadName)s:%(levelname)s> %(message)s",
+        level=logging.NOTSET
+        if args.verbose
+        else logging.DEBUG
+        if len(args.filenames) < 1000
+        else logging.INFO,
+        stream=sys.stderr,
+    )
+
+    lint_messages = []
+    for filename in args.filenames:
+        lint_message = check_file(filename)
+        if lint_message is not None:
+            lint_messages.append(lint_message)
+
+    for lint_message in lint_messages:
+        print(json.dumps(lint_message._asdict()), flush=True)

torch/_logging/_internal.py

@@ -1,4 +1,5 @@
 # mypy: allow-untyped-defs
+
 import functools
 import hashlib
 import importlib.util
@@ -655,7 +656,7 @@ def pad_to(s, length=30):
 
   TORCH_LOGS_OUT=/tmp/output.txt will output the logs to /tmp/output.txt as
   well. This is useful when the output is long.
-"""  # flake8: noqa: B950
+"""
     msg = f"""
 TORCH_LOGS Info
 {examples}

torch/ao/quantization/quantize_pt2e.py

@@ -227,7 +227,7 @@ def convert_pt2e(
         # for detailed explanation of output quantized model
         quantized_model = convert_pt2e(prepared_model)
 
-    """  # flake8: noqa
+    """
     torch._C._log_api_usage_once("quantization_api.quantize_pt2e.convert_pt2e")
     if not isinstance(use_reference_representation, bool):
         raise ValueError(

torch/jit/_shape_functions.py

@@ -1,10 +1,10 @@
+# flake8: noqa
 # mypy: allow-untyped-defs
 import math
 from typing import Any, Callable, Dict, List, Optional, Tuple, Union
 
 
 number = Union[int, float]
-# flake8: noqa
 
 ###
 # There are generated files that depend on this file

torch/onnx/_internal/diagnostics/_rules.py

@@ -1,4 +1,5 @@
 # mypy: allow-untyped-defs
+# flake8: noqa
 """
 GENERATED CODE - DO NOT EDIT DIRECTLY
 This file is generated by gen_diagnostics.py.
@@ -10,7 +11,6 @@
 import dataclasses
 from typing import Tuple
 
-# flake8: noqa
 from torch.onnx._internal.diagnostics import infra
 
 

torch/onnx/_internal/diagnostics/infra/sarif/version.py

@@ -1,6 +1,6 @@
+# flake8: noqa
 from typing import Final
 
 
 SARIF_VERSION: Final = "2.1.0"
 SARIF_SCHEMA_LINK: Final = "https://docs.oasis-open.org/sarif/sarif/v2.1.0/cs01/schemas/sarif-schema-2.1.0.json"
-# flake8: noqa

torch/onnx/_internal/exporter/_analysis.py

@@ -1,7 +1,7 @@
-"""Compatibility analyzer for PyTorch models."""
-
 # mypy: allow-untyped-defs
 # flake8: noqa: B950 We do not need flake8 as it complains line length
+"""Compatibility analyzer for PyTorch models."""
+
 from __future__ import annotations
 
 import dataclasses