Expand ruff checks, and apply non-formatting checks to CI

.github/workflows/lint.yml

@@ -143,8 +143,8 @@ jobs:
           set -xeuo pipefail
           bash admin/run_include_check.sh -d
 
-  ruff-format:
-    name: ruff-format
+  ruff:
+    name: ruff
     runs-on: ubuntu-24.04
     timeout-minutes: 30
 
@@ -164,7 +164,7 @@ jobs:
           python -m pip install --upgrade pip
           python -m pip install ruff==0.15.8
 
-      - name: Run ruff format check
+      - name: Run Ruff checks
         shell: bash
         run: |
           set -xeuo pipefail

admin/run_ruff_format.sh

@@ -15,17 +15,19 @@
 # limitations under the License.
 
 
-# Runs ruff format over the codebase.
+# Runs Ruff lint fixes and formatting over the codebase.
 # By default will modify files in-place. Use -d to do a dry-run.
 
 set -ex
 
 EXPECTED_RUFF_VERSION="0.15.8"
-RUFF_ARGS=()
+RUFF_CHECK_ARGS=(--fix)
+RUFF_FORMAT_ARGS=()
 while getopts ":d" opt; do
   case ${opt} in
     d )
-      RUFF_ARGS+=(--check)
+      RUFF_CHECK_ARGS=()
+      RUFF_FORMAT_ARGS+=(--check)
       ;;
     \? )
       echo "Usage: run_ruff_format.sh [-d]"
@@ -39,12 +41,15 @@ ROOT_DIR=$(git rev-parse --show-toplevel)
 if command -v ruff >/dev/null 2>&1; then
   ACTUAL_RUFF_VERSION=$(ruff --version | awk '{print $2}')
   if [ "$ACTUAL_RUFF_VERSION" != "$EXPECTED_RUFF_VERSION" ]; then
-    echo "Warning: expected ruff version $EXPECTED_RUFF_VERSION, found $ACTUAL_RUFF_VERSION. Formatting may not match CI checker." >&2
+    echo "Warning: expected ruff version $EXPECTED_RUFF_VERSION, found $ACTUAL_RUFF_VERSION. Results may not match CI." >&2
   fi
 else
   echo "Error: ruff is not installed; expected version $EXPECTED_RUFF_VERSION." >&2
   exit 1
 fi
 
+echo "Running ruff check:"
+ruff check "${RUFF_CHECK_ARGS[@]}" "$ROOT_DIR"
+
 echo "Running ruff format:"
-ruff format "${RUFF_ARGS[@]}" "$ROOT_DIR"
+ruff format "${RUFF_FORMAT_ARGS[@]}" "$ROOT_DIR"

benchmarks/cross_similarity_bench.py

@@ -24,7 +24,6 @@
 from nvmolkit.fingerprints import MorganFingerprintGenerator
 from nvmolkit.similarity import crossCosineSimilarity, crossTanimotoSimilarity
 
-
 SIZES = [2000, 4000, 6000, 8000, 10000, 12000, 14000, 16000, 20000, 24000, 28000, 32000]
 CPU_SINGLE_VALUE_ABOVE = 6000
 

benchmarks/tfd_prepare_mols.py

@@ -37,7 +37,6 @@
 import multiprocessing
 import os
 import pickle
-import sys
 import time
 from functools import partial
 

benchmarks/tfd_profile.py

@@ -174,7 +174,7 @@ def main():
     # === Warmup ===
     if args.warmup > 0:
         with nvtx.annotate("Warmup", color="red"):
-            first_mols = list(configs.values())[0]
+            first_mols = next(iter(configs.values()))
             warmup_mols = first_mols[: min(5, len(first_mols))]
             print(f"\nWarmup ({args.warmup} iteration(s)) with {len(warmup_mols)} molecules...")
             for _ in range(args.warmup):

nvmolkit/_mmff_bridge.py

@@ -39,7 +39,6 @@
 
 def default_rdkit_mmff_properties(mol: "Mol"):
     """Create default RDKit MMFF properties for ``mol``."""
-
     properties = rdForceFieldHelpers.MMFFGetMoleculeProperties(mol)
     if properties is None:
         raise ValueError("RDKit could not create MMFF properties for molecule")
@@ -58,7 +57,6 @@ def make_internal_mmff_properties(
     (variant, dielectric, per-term flags); the corresponding getters are not
     wrapped.  We read the settings through the C++ binding layer instead.
     """
-
     return _batchedForcefield.buildMMFFPropertiesFromRDKit(
         properties,
         float(non_bonded_threshold),

nvmolkit/autotune/_calibration.py

@@ -37,7 +37,7 @@ def auto_subsample(
     """
     if workload_size <= 0:
         raise ValueError("workload_size must be positive")
-    target = min(max_size, max(min_size, int(round(fraction * workload_size))))
+    target = min(max_size, max(min_size, round(fraction * workload_size)))
     target = min(target, workload_size)
     rng = random.Random(seed)
     indices = list(range(workload_size))
@@ -86,6 +86,6 @@ def shrink(indices: Sequence[int], factor: float = 0.5, *, min_size: int = 1) ->
     """
     if factor <= 0.0 or factor >= 1.0:
         raise ValueError("factor must be in (0, 1)")
-    new_size = max(min_size, int(round(len(indices) * factor)))
+    new_size = max(min_size, round(len(indices) * factor))
     new_size = min(new_size, len(indices))
     return list(indices[:new_size])

nvmolkit/autotune/_core.py

@@ -53,7 +53,7 @@ def _require_optuna():
     """Import optuna or raise an :class:`ImportError` with install instructions."""
     if not is_optuna_available():
         raise ImportError(OPTUNA_INSTALL_HINT)
-    import optuna  # noqa: PLC0415
+    import optuna
 
     return optuna
 
@@ -348,7 +348,7 @@ def collect_int_from_space(spec: Any) -> int:
         high_int = int(high)
         if low_int <= 0 or high_int <= 0:
             raise ValueError(f"Log-uniform range {spec!r} requires strictly positive bounds.")
-        midpoint = int(round(math.sqrt(low_int * high_int)))
+        midpoint = round(math.sqrt(low_int * high_int))
         return max(low_int, min(high_int, midpoint))
     if isinstance(spec, tuple) and len(spec) == 3 and all(isinstance(v, int) for v in spec):
         low, high, step = (int(v) for v in spec)

nvmolkit/autotune/_ff_common.py

@@ -100,7 +100,7 @@ def resolve_num_gpus(fixed_gpu_ids: list[int]) -> int:
     if fixed_gpu_ids:
         return max(1, len(fixed_gpu_ids))
     try:
-        import torch  # noqa: PLC0415
+        import torch
 
         return max(1, int(torch.cuda.device_count()))
     except Exception:

nvmolkit/batchedForcefield.py

@@ -87,8 +87,6 @@
 from nvmolkit._mmff_bridge import default_rdkit_mmff_properties, make_internal_mmff_properties
 from nvmolkit.types import CoordinateOutput, Device3DResult, FireOptions, HardwareOptions
 
-from nvmolkit import _batchedForcefield  # type: ignore
-
 if TYPE_CHECKING:
     from rdkit.Chem import Mol
     from rdkit.ForceField.rdForceField import MMFFMolProperties as RDKitMMFFMolProperties
@@ -456,8 +454,7 @@ def _minimize(
 
 
 class MMFFBatchedForcefield(_BatchedForcefieldBase):
-    """Evaluate MMFF energies and gradients, or run BFGS minimization, for a
-    batch of molecules with all their conformers.
+    """Evaluate MMFF energies and gradients or run BFGS minimization for molecule batches.
 
     Properties and constraints are per-molecule and are shared across all
     conformers of that molecule.  Results are nested as
@@ -515,6 +512,7 @@ def __init__(
         )
 
     def __getitem__(self, idx: int) -> MMFFBatchElement:
+        """Return the force-field element at ``idx``."""
         return super().__getitem__(idx)
 
     def _normalize_properties(
@@ -632,8 +630,7 @@ def minimize(
 
 
 class UFFBatchedForcefield(_BatchedForcefieldBase):
-    """Evaluate UFF energies and gradients, or run BFGS minimization, for a
-    batch of molecules with all their conformers.
+    """Evaluate UFF energies and gradients or run BFGS minimization for molecule batches.
 
     Constraints are per-molecule and are shared across all conformers of
     that molecule.  Results are nested as ``list[list[...]]`` — outer
@@ -679,6 +676,7 @@ def __init__(
         self._vdw_thresholds = _normalize_scalar_or_list(vdwThreshold, len(molecules), "vdwThreshold")
 
     def __getitem__(self, idx: int) -> UFFBatchElement:
+        """Return the force-field element at ``idx``."""
         return super().__getitem__(idx)
 
     def _build_native(self):

nvmolkit/mmffOptimization.py

@@ -33,8 +33,6 @@
     from rdkit.ForceField.rdForceField import MMFFMolProperties
 
 from nvmolkit import _mmffOptimization
-from nvmolkit._mmff_bridge import default_rdkit_mmff_properties, make_internal_mmff_properties
-from nvmolkit.types import CoordinateOutput, Device3DResult, HardwareOptions
 
 
 @overload

nvmolkit/substructure.py

@@ -41,10 +41,10 @@
 from nvmolkit._substructure import hasSubstructMatch as _hasSubstructMatch
 
 __all__ = [
-    "SubstructSearchConfig",
     "SubstructMatchResults",
-    "getSubstructMatches",
+    "SubstructSearchConfig",
     "countSubstructMatches",
+    "getSubstructMatches",
     "hasSubstructMatch",
 ]
 
@@ -79,6 +79,7 @@ def __init__(
         gpuIds: list[int] | None = None,
         algorithm: str = "dfs",
     ) -> None:
+        """Initialize a substructure-search configuration."""
         native = _NativeSubstructSearchConfig()
         native.batchSize = int(batchSize)
         native.workerThreads = int(workerThreads)
@@ -198,9 +199,11 @@ class SubstructMatchResults:
     shape: tuple[int, int]
 
     def __len__(self) -> int:
+        """Return the number of target molecules."""
         return self.shape[0]
 
     def __getitem__(self, target_idx: int) -> _SubstructTargetView:
+        """Return the match view for one target molecule."""
         return _SubstructTargetView(self, target_idx)
 
     def get_pair(self, target_idx: int, query_idx: int) -> list[np.ndarray]:

nvmolkit/tests/test_batched_forcefield.py

@@ -165,9 +165,10 @@ def assert_energy_and_gradient_close(got_energy, want_energy, got_grad, want_gra
 
 
 def _assert_batched_compute_matches_rdkit_mmff(mol_specs):
-    """Build a single MMFFBatchedForcefield from ``mol_specs`` and verify that
-    per-mol ``compute_energy``/``compute_gradients`` match RDKit's single-mol FF
-    for each mol, with its configured properties and (optionally) constraint.
+    """Verify batched MMFF energies and gradients against RDKit.
+
+    Build a single MMFFBatchedForcefield from ``mol_specs`` and compare each
+    molecule using its configured properties and optional constraint.
 
     When any mol has a constraint configured, additionally verify that the
     constraint has an observable effect on that mol's energy AND gradient — as
@@ -266,8 +267,11 @@ def _assert_batched_compute_matches_rdkit_mmff(mol_specs):
 
 
 def test_mmff_batched_forcefield_properties_match_rdkit():
-    """Batch of mols with varied per-mol property configurations (default, MMFF variant,
-    dielectric model, term toggles, fragmented+interfrag)."""
+    """Test varied per-molecule property configurations.
+
+    Cover defaults, MMFF variants, dielectric models, term toggles, and
+    fragmented molecules with interfragment interactions.
+    """
     _assert_batched_compute_matches_rdkit_mmff(
         [
             {"factory": load_reference_mol},
@@ -296,9 +300,11 @@ def test_mmff_batched_forcefield_properties_match_rdkit():
 
 
 def test_mmff_batched_forcefield_reads_externally_configured_properties():
-    """Configure RDKit MMFF properties via raw ``rdForceFieldHelpers.MMFFGetMoleculeProperties``
-    plus direct ``SetMMFF*Term``/``SetMMFFDielectricConstant`` calls — no nvmolkit helpers
-    in the path — then hand the object to ``MMFFBatchedForcefield``.
+    """Test externally configured RDKit MMFF properties.
+
+    Configure properties via raw ``rdForceFieldHelpers.MMFFGetMoleculeProperties``
+    plus direct ``SetMMFF*Term``/``SetMMFFDielectricConstant`` calls, with no
+    nvmolkit helpers in the path, then pass the object to ``MMFFBatchedForcefield``.
 
     Needed because of our workaround for RDKit bug https://github.com/rdkit/rdkit/issues/9253
     """
@@ -327,8 +333,11 @@ def test_mmff_batched_forcefield_reads_externally_configured_properties():
 
 
 def test_mmff_batched_forcefield_constraints_match_rdkit():
-    """Batch of mols with all 5 MMFF constraint types applied (one per mol), some also
-    carrying non-default property settings to exercise the properties+constraints path."""
+    """Test all five MMFF constraint types against RDKit.
+
+    Apply one constraint type per molecule, with some molecules also carrying
+    non-default property settings to exercise the properties-plus-constraints path.
+    """
     _assert_batched_compute_matches_rdkit_mmff(
         [
             {
@@ -515,8 +524,11 @@ def _build_constrained_mmff_batch(specs=_MMFF_BATCH_CONSTRAINT_SPECS, hardwareOp
 
 
 def _assert_batched_minimize_matches_rdkit(specs, mols, opt_energies, converged, make_ref_ff):
-    """Compare nvMolKit minimize() result to RDKit minimize per (mol, conformer), with each
-    mol carrying a different constraint from `specs`."""
+    """Compare nvMolKit and RDKit minimization results.
+
+    Compare each molecule and conformer while each molecule carries a different
+    constraint from ``specs``.
+    """
     for mol_idx, (mol, spec) in enumerate(zip(mols, specs)):
         assert len(opt_energies[mol_idx]) == mol.GetNumConformers()
         assert all(converged[mol_idx]), f"Mol {mol_idx} failed to converge"
@@ -531,9 +543,11 @@ def _assert_batched_minimize_matches_rdkit(specs, mols, opt_energies, converged,
 
 
 def test_mmff_batched_minimize_with_constraints_batch_matches_rdkit():
-    """Batch minimize with different constraint types on different-size mols
-    and different conformer counts, comparing each (mol, conformer) energy
-    to RDKit's minimize with the same constraint."""
+    """Test constrained batch minimization against RDKit.
+
+    Use different constraint types, molecule sizes, and conformer counts, then
+    compare each molecule and conformer energy to RDKit with the same constraint.
+    """
     mols, _, ff = _build_constrained_mmff_batch()
     opt_energies, converged = ff.minimize(maxIters=500)
 
@@ -544,9 +558,11 @@ def make_ref(mol, conf_id):
 
 
 def test_mmff_batched_minimize_respects_maxiters_and_forcetol():
-    """maxIters and forceTol must be plumbed through: a single-iteration minimize
-    should not converge and should leave energies closer to the starting point
-    than a generous-iteration minimize."""
+    """Test that maxIters and forceTol are passed through.
+
+    A single-iteration minimization should not converge and should leave energies
+    closer to the starting point than a generous-iteration minimization.
+    """
     perturbed_mols = [
         perturb_conformers(make_embedded_mol("CCCO", num_confs=2)),
         perturb_conformers(make_embedded_mol("c1ccccc1CCO", num_confs=2)),

nvmolkit/tests/test_mmff_optimization.py

@@ -23,8 +23,8 @@
 from rdkit.ForceField import rdForceField as _rdForceField  # noqa: F401
 from rdkit.Geometry import Point3D
 
-from nvmolkit.embedMolecules import EmbedMolecules
 import nvmolkit.mmffOptimization as nvmolkit_mmff
+from nvmolkit.embedMolecules import EmbedMolecules
 from nvmolkit.types import CoordinateOutput, Device3DResult, FireOptions, HardwareOptions
 
 
@@ -133,7 +133,11 @@ def calculate_rdkit_mmff_energies(
     """Calculate MMFF energies using RDKit for all conformers of all molecules.
 
     Args:
-        molecules: List of RDKit molecules with conformers
+        molecules: List of RDKit molecules with conformers.
+        maxIters: Maximum minimization iterations per conformer.
+        property_settings: Optional MMFF property overrides.
+        nonBondedThreshold: Non-bonded interaction cutoff.
+        ignoreInterfragInteractions: Whether to omit interactions between fragments.
 
     Returns:
         list: List of lists containing energies for each molecule's conformers

nvmolkit/types.py

@@ -21,9 +21,14 @@
 import numpy as np
 import torch
 
+# These imports populate Boost.Python's global converter registry. Keep this
+# bootstrap sequence developer-controlled so future binding dependencies are
+# not silently reordered by the formatter.
+# isort: off
 from nvmolkit import _arrayHelpers  # noqa: F401
 from nvmolkit import _embedMolecules  # type: ignore
 from nvmolkit import _types
+# isort: on
 
 
 class FireOptions:

nvmolkit/uffOptimization.py

@@ -27,7 +27,6 @@
     from rdkit.Chem import Mol
 
 from nvmolkit import _uffOptimization
-from nvmolkit.types import CoordinateOutput, Device3DResult, HardwareOptions
 
 
 @overload