Add cpplint and ruff linter to pre-commit and fix lint violations

[pstjohn]

Replace pylint with ruff for Python linting in pre-commit config and CI scripts. Keep black as the formatter. Fix all 14 ruff lint violations:

• F841: Remove unused variables
• F811: Remove unused import shadowed by local definition
• F821: Remove erroneous del of variables used later (bug fix)
• E721: Use is/is not for type comparisons
• PLR1714: Merge duplicate comparisons
• PLW0406: Add noqa for intentional self-imports

Also adds cpplint to pre-commit

[greptile-apps]

Greptile Summary

This PR replaces pylint with ruff for Python linting, adds cpplint as a pre-commit hook, deletes pylintrc, and fixes all 14 ruff violations across several files. The lint fixes are all semantically correct: PLR1714 transformations preserve logical equivalence, E721 is/is not replacements are appropriate, and the F811/F841 removals are clean-ups with no behavioral impact.

Confidence Score: 5/5

Safe to merge — all lint fixes are semantically equivalent transformations with no behavioral changes.

All PLR1714, E721, F841, F811, and F821 fixes have been verified as logically equivalent. No P0/P1 issues found. The one remaining comment (stale pylint disable in grouped_linear.py) is a P2 cleanup item that doesn't affect runtime behavior.

transformer_engine/pytorch/ops/basic/grouped_linear.py has a lingering stale comment; otherwise no files require special attention.

Vulnerabilities

No security concerns identified. Changes are limited to linter configuration and mechanical lint-driven code clean-ups.

Important Files Changed

Filename	Overview
.pre-commit-config.yaml	Adds ruff (Python lint) and cpplint (C++/CUDA lint) hooks; existing black formatter and clang-format hooks untouched. The cpplint exclude pattern aligns with CI; the --root flag omission (noted in a prior thread) only affects error-message path formatting.
pyproject.toml	Adds comprehensive ruff configuration mirroring the disabled pylint rules; per-file ignores for known re-export patterns (fp8.py, export.py, etc.) are appropriate.
transformer_engine/pytorch/distributed.py	Removes the shadowed import of _get_module_fsdp_state from torch.distributed.fsdp._common_utils; the local definition at line 1957 (used at line 2091) correctly takes precedence.
transformer_engine/pytorch/attention/dot_product_attention/context_parallel.py	Two PLR1714 fixes merge duplicate window_size equality comparisons into in-membership tests; both are semantically equivalent to the originals.
transformer_engine/pytorch/ops/basic/grouped_linear.py	E721 fix replaces type(weight.data) != weight_tensor_type with is not; the now-inert # pylint: disable=unidiomatic-typecheck comment was not removed.
transformer_engine/debug/pytorch/debug_quantization.py	PLR1714 fix merges two == API_CALL_MODIFY comparisons into a single in check; semantically equivalent. Several # pylint: disable= comments remain but are now inert.
transformer_engine/pytorch/ops/fused/forward_linear_bias_activation.py	F841 fix replaces the dead activation_op = None assignment with a pass comment; no behavioral change.
transformer_engine/debug/features/_test_dummy_feature.py	PLW0406 suppressed with # noqa: PLW0406 for the intentional self-import pattern; both noqa and legacy pylint comments coexist harmlessly.
transformer_engine/debug/features/api.py	E721 fix replaces type(ret) == torch.Tensor with is; the inline # pylint: disable=unidiomatic-typecheck was removed along with the refactored expression.
transformer_engine/debug/init.py	F841 fix removes unused as e from bare except ImportError clause; no behavioral change.
qa/L0_pytorch_lint/test.sh	Replaces pylint with ruff and adds cpplint; pins both to the same versions used in the pre-commit hook.
qa/L0_jax_lint/test.sh	Mirrors L0_pytorch_lint changes; installs same cpplint/ruff versions and replaces the pylint invocation with ruff.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Developer commits code] --> B[pre-commit hooks]

    B --> C[black\nPython formatter]
    B --> D[ruff\nPython linter]
    B --> E[cpplint\nC++/CUDA linter]
    B --> F[clang-format\nC++ formatter]
    B --> G[vermin\nPython version check]

    D --> D1["select: E, F, W, PL\nignore: E402, E501, PLR*, PLW*..."]
    E --> E1["files: transformer_engine/common|jax|pytorch\nexclude: build_tools/build/"]

    B --> H{All checks pass?}
    H -- Yes --> I[CI: qa/L0_pytorch_lint/test.sh\nqa/L0_jax_lint/test.sh]
    H -- No --> J[Block commit / auto-fix]

    I --> I1[cpplint + ruff\npinned to same versions\nas pre-commit]

Loading

_{Reviews (2): Last reviewed commit: "addressing greptile review" | Re-trigger Greptile}

[timmoon10]

LGTM, although is there a reason to prefer Ruff over Pylint? Speed?

[ksivaman]

We should remove all the pylint: disable comments

[pstjohn]

LGTM, although is there a reason to prefer Ruff over Pylint? Speed?

Yeah pretty much. If we're running it as a pre-commit check it's nice if it's fast. But I'm happy to check the pylint version speed that wouldn't require any changes 🤷