[None][infra] expand Python mypy coverage

.pre-commit-config.yaml

@@ -1,3 +1,6 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
 # <<<< BEGIN AUTO-GENERATED by scripts/legacy_utils.py gen-configs >>>>
 # Source of truth: legacy-files.txt
 # To modify this block: edit legacy-files.txt, then run:
@@ -1457,10 +1460,8 @@ legacy-files: &legacy_files |
     )$
 # <<<< END AUTO-GENERATED >>>>
 
-# The set of files subjected to mypy static analysis is single-sourced in
-# pyproject.toml under [tool.mypy] `files`. The type-check hook below runs mypy
-# with no explicit file arguments (pass_filenames: false) so mypy reads that
-# list; the hook's own `files:` pattern only gates *when* the hook triggers.
+# Full-project mypy scans use the file set in pyproject.toml. The pre-commit
+# hook passes changed paths to the script's default changed-file mode.
 
 # Global exclude: vendored code + trtllm-gen FMHA artifacts (cubin pointers, export headers, cuda_ptx)
 exclude: '(^tensorrt_llm/_torch/attention/kernels/blackwell/kimi_k3_kda/kda_chunked_bt16(_small_bh)?\.py$|^cpp/tensorrt_llm/common/sha256/|^triton_kernels/|^tensorrt_llm/_torch/attention/backends/prims_ts/|^tensorrt_llm/_torch/visual_gen/cute_dsl_kernels/blackwell/sol_attn/|trtllmGenKernels/fmha/cubin/kernelMetaInfo\.h$|cubin\.cpp$|cubin\.h$|trtllmGenKernels/fmha/trtllmGen_fmha_export/|trtllmGenKernels/fmha/cuda_ptx/|trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelMetaInfo\.h$|trtllmGenKernels/gemm/trtllmGen_gemm_export/KernelMetaInfo\.h$|\.cubin\.tar\.zst$)'
@@ -1612,14 +1613,13 @@ repos:
         files: >-
           ^(\.claude/skills/[^/]+/SKILL\.md|\.claude/agents/.*\.md|scripts/check_skill_naming_convention\.py)$
     -   id:  type-check
-        name: Run static analysis of type correctness using mypy
-        entry: scripts/run_mypy.sh
+        name: Record advisory mypy type-check report
+        entry: scripts/run_mypy.sh --diff-base HEAD
         language: script
         verbose: true
         require_serial: true  # because of install_types=true
-        # Run mypy with no explicit paths so it type-checks the set declared in
-        # pyproject.toml [tool.mypy] `files` (single source of truth). This
-        # `files:` pattern only decides when the hook triggers — any change under
-        # the sampler source/test dirs.
-        pass_filenames: false
-        files: ^(tensorrt_llm/_torch/pyexecutor/sampler/|tests/unittest/_torch/sampler/)
+        # Check changed paths and retain diagnostics only on added diff lines.
+        # HEAD is the pre-commit base for staged changes.
+        # Configuration changes trigger a full-project scan without diff filtering.
+        pass_filenames: true
+        files: ^(tensorrt_llm/.*\.pyi?$|pyproject\.toml$|scripts/(mypy_exclusion\.json|mypy_baseline\.txt|run_mypy\.sh|mypy_exclusion\.py|diff_line_filter\.py)$|\.pre-commit-config\.yaml$)

AGENTS.md

@@ -190,6 +190,13 @@ See [CI overview](docs/source/developer-guide/ci-overview.md) for full details.
 | Test waives | `tests/integration/test_lists/waives.txt` | Skip known-failing tests with NVBug links |
 | Performance | See [benchmarking guide](docs/source/developer-guide/perf-benchmarking.md) | `trtllm-bench` and `trtllm-serve` benchmarks |
 
+Run changed-file Python static checks with `scripts/run_mypy.sh`, or pass
+`--full-project` to scan all product Python files. The `[tool.mypy]` section in
+`pyproject.toml` defines the product and test roots and the diagnostic allowlist.
+Keep its enabled/disabled code lists aligned with the pinned mypy version in
+`requirements-dev.txt`. Missing imports are reported except for PyTorch and the
+generated `tensorrt_llm.bindings` modules, which may be absent on development hosts.
+
 New C++ unit tests should use `add_tllm_gtest` with explicit `SOURCES`,
 `LINK_LIBRARIES`, `OWNER`, and scheduling `LABELS`. Request `FULL_STACK` explicitly
 for integration tests and include `gpu` in `LABELS` for tests that need a GPU.

jenkins/Build.groovy

@@ -572,7 +572,7 @@ def runLLMBuild(
         withEnv(["MYPY_REQUIRE_BINDINGS=1"]) {
             trtllm_utils.llmExecStepWithRetry(
                 pipeline,
-                script: "cd ${LLM_ROOT} && bash scripts/run_mypy.sh",
+                script: "cd ${LLM_ROOT} && bash scripts/run_mypy.sh --full-project && cat mypy_final_report.txt",
                 shortCommondRunTimeMax: 240  // always retry on failure if the cmd runs less than 240s
             )
         }

pyproject.toml

@@ -1,3 +1,6 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
 ####################################################################################################
 ###############################  BUILD CONFIGURATION  ##############################################
 ####################################################################################################
@@ -844,71 +847,101 @@ follow_imports = "silent"
 explicit_package_bases = true
 namespace_packages = true
 strict = true
+check_untyped_defs = true
 show_error_codes = true
-# Files checked when mypy runs with no explicit paths. This is the single source
-# of truth for the type-check file set. Both the CI build stage (which invokes
-# scripts/run_mypy.sh directly) and the local pre-commit hook run mypy without
-# explicit paths (the hook sets `pass_filenames: false`), so both read this list.
-files = [
-    "tensorrt_llm/_torch/pyexecutor/sampler/sampler.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/sampler_common.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/sampling_resolution.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/logprobs.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/sampler_features.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/seed_manager.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/finish_reasons.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/top_p_decay.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/penalties.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/token_ban.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/ops/flashinfer.py",
-    "tensorrt_llm/_torch/pyexecutor/sampler/ops/vanilla.py",
-    "tests/unittest/_torch/sampler/test_torch_sampler.py",
-    "tests/unittest/_torch/sampler/test_token_ban.py",
-    "tests/unittest/_torch/sampler/test_beam_search.py",
-    "tests/unittest/_torch/sampler/test_beam_search_util.py",
-]
-# NB:
-#   .   -> 'tensorrt_llm'
-#   tests/unittest   -> 'utils'
-#   tests/unittest/_torch/sampler   -> for 'test_beam_search_util'
-mypy_path = ".:tests/unittest:tests/unittest/_torch/sampler"
-
-[[tool.mypy.overrides]]
-module = ["tensorrt_llm.bindings.*"]
-ignore_errors = true
-
-# Generated pybind11 stub; DeepGEMM returns a bare py::dict from
-# get_bf16_mega_gate_config, which stubgen renders without type parameters.
-[[tool.mypy.overrides]]
-module = ["tensorrt_llm.deep_gemm_cpp_tllm", "deep_gemm_cpp_tllm"]
-disallow_any_generics = false
-
-[[tool.mypy.overrides]]
-module = [
-    "tensorrt_llm._torch.pyexecutor.sampler",
-    "tensorrt_llm._torch.pyexecutor.sampler.*",
+# Diagnostic allowlist for product code. Keep the complementary
+# disabled list in sync with the mypy version pinned in requirements-dev.txt.
+enable_error_code = [
+    "syntax",
+    "import-not-found",
+    "name-defined",
+    "used-before-def",
+    "call-arg",
 ]
 disable_error_code = [
+    "abstract",
+    "annotation-unchecked",
+    "arg-type",
+    "assert-type",
+    "assignment",
+    "attr-defined",
+    "await-not-async",
+    "call-overload",
+    "comparison-overlap",
+    "deprecated",
+    "dict-item",
+    "empty-body",
+    "exhaustive-match",
+    "exit-return",
+    "explicit-any",
+    "explicit-override",
+    "func-returns-value",
+    "has-type",
+    "ignore-without-code",
+    "import",
+    "import-untyped",
+    "index",
+    "list-item",
+    "literal-required",
+    "maybe-unrecognized-str-typeform",
+    "metaclass",
+    "method-assign",
+    "misc",
+    "mutable-override",
+    "name-match",
+    "narrowed-type-not-subtype",
+    "no-any-return",
+    "no-any-unimported",
+    "no-overload-impl",
+    "no-redef",
     "no-untyped-call",
-    # TODO(TRTLLM-15310): strict mode's warn_redundant_casts flags several cast(...) calls
-    # in the sampler package (introduced by #16620). Whether a cast is "redundant" is
-    # binding-dependent -- the full build check (compiled bindings + stubs) flags 2,
-    # the bindings-less local/pre-commit check flags ~20 -- so removing casts is a
-    # fragile WAR. Suppress redundant-cast for the sampler module until the casts are
-    # audited and the annotations tightened, rather than churn the source.
+    "no-untyped-def",
+    "operator",
+    "overload-cannot-match",
+    "overload-overlap",
+    "override",
+    "possibly-undefined",
+    "prop-decorator",
     "redundant-cast",
+    "redundant-expr",
+    "redundant-self",
+    "return",
+    "return-value",
+    "safe-super",
+    "str-bytes-safe",
+    "str-format",
+    "truthy-bool",
+    "truthy-function",
+    "truthy-iterable",
+    "top-level-await",
+    "type-abstract",
+    "type-arg",
+    "type-var",
+    "typeddict-item",
+    "typeddict-readonly-mutated",
+    "typeddict-unknown-key",
+    "unreachable",
+    "untyped-decorator",
+    "unused-awaitable",
+    "unused-coroutine",
+    "unused-ignore",
+    "unimported-reveal",
+    "union-attr",
+    "valid-newtype",
+    "valid-type",
+    "var-annotated",
 ]
+# Files checked by scripts/run_mypy.sh --full-project. Pre-commit passes changed
+# product Python files explicitly to the script's default changed-file mode.
+files = ["tensorrt_llm"]
+# scripts/run_mypy.sh reads path and symbol exclusions from scripts/mypy_exclusion.json.
+mypy_path = "."
 
 [[tool.mypy.overrides]]
-module = [
-    "tests.*",
-    "_torch.*",  # tests/unittest/_torch/*, cf. mypy_path
-    "test_torch_sampler",  # tests/unittest/_torch/sampler/..., cf. mypy_path
-    "test_token_ban",  # tests/unittest/_torch/sampler/..., cf. mypy_path
-    "test_beam_search",  # tests/unittest/_torch/sampler/..., cf. mypy_path
-    "test_beam_search_util",  # tests/unittest/_torch/sampler/..., cf. mypy_path
-]
-disable_error_code = [
-    "no-untyped-def",
-    "no-untyped-call",
-]
+# TensorRT-LLM bindings are generated by the C++ build and may be absent in a source checkout.
+module = ["torch.*", "tensorrt_llm.bindings", "tensorrt_llm.bindings.*"]
+ignore_missing_imports = true
+
+[[tool.mypy.overrides]]
+module = ["tensorrt_llm.llmapi.tracer"]
+disable_error_code = ["import-not-found"]

scripts/diff_line_filter.py

@@ -0,0 +1,100 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+"""Match source locations to added lines in a Git diff."""
+
+import re
+import subprocess
+from pathlib import Path
+
+DIFF_HUNK_PATTERN = re.compile(r"^@@ -\d+(?:,\d+)? \+(\d+)(?:,(\d+))? @@")
+
+
+class AddedLineFilter:
+    """Check whether source locations fall on lines added relative to a Git ref."""
+
+    def __init__(self, project_root: Path, diff_base: str) -> None:
+        self.project_root = project_root.resolve()
+        self.added_lines_by_path: dict[str, set[int]] = {}
+
+        git_root = subprocess.run(
+            ["git", "rev-parse", "--show-toplevel"],
+            cwd=self.project_root,
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        if git_root.returncode != 0:
+            raise ValueError(f"cannot read diff lines: {git_root.stderr.strip()}")
+
+        base = subprocess.run(
+            ["git", "rev-parse", "--verify", f"{diff_base}^{{commit}}"],
+            cwd=self.project_root,
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        if base.returncode != 0:
+            raise ValueError(f"invalid diff base {diff_base!r}: {base.stderr.strip()}")
+        self.base_commit = base.stdout.strip()
+
+    def contains(self, source_path: str, line_number: int) -> bool:
+        """Return whether a path and line identify added or replaced source code."""
+        path = Path(source_path)
+        resolved_path = (
+            path.resolve() if path.is_absolute() else (self.project_root / path).resolve()
+        )
+        try:
+            relative_path = resolved_path.relative_to(self.project_root)
+        except ValueError:
+            return False
+
+        path_key = relative_path.as_posix()
+        if path_key not in self.added_lines_by_path:
+            self.added_lines_by_path[path_key] = self._load_added_lines(relative_path)
+        return line_number in self.added_lines_by_path[path_key]
+
+    def _load_added_lines(self, relative_path: Path) -> set[int]:
+        path = relative_path.as_posix()
+        tracked = subprocess.run(
+            ["git", "ls-files", "--error-unmatch", "--", path],
+            cwd=self.project_root,
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        source_path = self.project_root / relative_path
+        if tracked.returncode == 1 and source_path.is_file():
+            return set(range(1, len(source_path.read_text().splitlines()) + 1))
+        if tracked.returncode != 0:
+            raise ValueError(
+                f"cannot check Git tracking state for {path}: {tracked.stderr.strip()}"
+            )
+
+        diff = subprocess.run(
+            [
+                "git",
+                "diff",
+                "--no-ext-diff",
+                "--no-renames",
+                "--unified=0",
+                self.base_commit,
+                "--",
+                path,
+            ],
+            cwd=self.project_root,
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        if diff.returncode != 0:
+            raise ValueError(f"cannot read diff for {path}: {diff.stderr.strip()}")
+
+        lines: set[int] = set()
+        for diff_line in diff.stdout.splitlines():
+            match = DIFF_HUNK_PATTERN.match(diff_line)
+            if match is None:
+                continue
+            start = int(match[1])
+            count = int(match[2] or 1)
+            lines.update(range(start, start + count))
+        return lines