feat: replace mypy and pytype with pyrefly for static type analysis

.github/workflows/cicd_tests.yml

@@ -56,7 +56,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        opt: ["codeformat", "mypy"]  #  "pytype" omitted for being essentially deprecated, see #8865
+        opt: ["codeformat", "pyrefly"]
     steps:
     - name: Clean unused tools
       run: |
@@ -80,8 +80,7 @@ jobs:
       run: |
         # clean up temporary files
         $(pwd)/runtests.sh --build --clean
-        # Github actions have multiple cores, so parallelize pytype
-        $(pwd)/runtests.sh --build --${{ matrix.opt }} -j $(nproc --all)
+        $(pwd)/runtests.sh --build --${{ matrix.opt }}
 
   min-dep:  # Test with minumum dependencies installed for different OS, Python, and PyTorch combinations
     runs-on: ${{ matrix.os }}

.github/workflows/cron.yml

@@ -214,7 +214,7 @@ jobs:
         python -c "import torch; print(torch.__version__); print('{} of GPUs available'.format(torch.cuda.device_count()))"
         python -c 'import torch; print(torch.rand(5,3, device=torch.device("cuda:0")))'
         ngc --version
-        BUILD_MONAI=1 ./runtests.sh --build --coverage --unittests --disttests  # unit tests with pytype checks, coverage report
+        BUILD_MONAI=1 ./runtests.sh --build --coverage --pyrefly --unittests --disttests  # unit tests with pyrefly checks, coverage report
         BUILD_MONAI=1 ./runtests.sh --build --coverage --net  # integration tests with coverage report
         coverage xml --ignore-errors
         if pgrep python; then pkill python; fi

.github/workflows/weekly-preview.yml

@@ -12,7 +12,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        opt: ["codeformat", "mypy"]
+        opt: ["codeformat", "pyrefly"]
     steps:
     - name: Clean unused tools
       run: |

.gitignore

@@ -110,12 +110,17 @@ venv.bak/
 # pytype cache
 .pytype/
 
+# pyrefly cache
+.pyrefly_cache/
+
 # mypy
 .mypy_cache/
+.dmypy.json
+
 examples/scd_lvsegs.npz
 temp/
 .idea/
-.dmypy.json
+.plans/
 
 *~
 

CONTRIBUTING.md

@@ -38,7 +38,7 @@ Please note that, as per PyTorch, MONAI uses American English spelling. This mea
 ### Preparing pull requests
 
 To ensure the code quality, MONAI relies on several linting tools ([black](https://github.com/psf/black), [isort](https://github.com/timothycrosley/isort), [ruff](https://github.com/astral-sh/ruff)),
-static type analysis tools ([mypy](https://github.com/python/mypy), [pytype](https://github.com/google/pytype)), as well as a set of unit/integration tests.
+static type analysis tools ([pyrefly](https://github.com/facebook/pyrefly)), as well as a set of unit/integration tests.
 
 This section highlights all the necessary preparation steps required before sending a pull request.
 To collaborate efficiently, please read through this section and follow them.

monai/apps/auto3dseg/auto_runner.py

@@ -790,6 +790,7 @@ def _train_algo_in_nni(self, history: list[dict[str, Any]]) -> None:
             nni_config_filename = os.path.abspath(os.path.join(self.work_dir, f"{name}_nni_config.yaml"))
             ConfigParser.export_config_file(nni_config, nni_config_filename, fmt="yaml", default_flow_style=None)
 
+            # pyrefly: ignore [redundant-cast]
             max_trial = min(self.hpo_tasks, cast(int, default_nni_config["maxTrialNumber"]))
             cmd = "nnictl create --config " + nni_config_filename + " --port 8088"
 
@@ -805,6 +806,7 @@ def _train_algo_in_nni(self, history: list[dict[str, Any]]) -> None:
                 n_trainings = len(import_bundle_algo_history(self.work_dir, only_trained=True))
 
             cmd = "nnictl stop --all"
+            # pyrefly: ignore [bad-argument-type]
             run_cmd(cmd.split(), check=True)
             logger.info(f"NNI completes HPO on {name}")
             last_total_tasks = n_trainings

monai/apps/deepedit/transforms.py

@@ -434,6 +434,7 @@ def _randomize(self, d, key_label):
         else:
             logger.info(f"Not slice IDs for label: {key_label}")
             sid = None
+        # pyrefly: ignore [unsupported-operation]
         self.sid[key_label] = sid
 
     def __call__(self, data: Mapping[Hashable, np.ndarray]) -> dict[Hashable, np.ndarray]:
@@ -561,6 +562,7 @@ def __init__(
         self.guidance: dict[str, list[list[int]]] = {}
 
     def randomize(self, data=None):
+        # pyrefly: ignore [unsupported-operation]
         probability = data[self.probability]
         self._will_interact = self.R.choice([True, False], p=[probability, 1.0 - probability])
 
@@ -885,6 +887,7 @@ def _randomize(self, d, key_label):
         else:
             logger.info(f"Not slice IDs for label: {key_label}")
             sid = None
+        # pyrefly: ignore [unsupported-operation]
         self.sid[key_label] = sid
 
     def __call__(self, data: Mapping[Hashable, np.ndarray]) -> dict[Hashable, np.ndarray]:

monai/apps/deepgrow/dataset.py

@@ -175,6 +175,7 @@ def _save_data_2d(vol_idx, vol_image, vol_label, dataset_dir, relative_path):
             continue
 
         # For all Labels
+        # pyrefly: ignore [missing-attribute]
         unique_labels = np.unique(label.flatten())
         unique_labels = unique_labels[unique_labels != 0]
         unique_labels_count = max(unique_labels_count, len(unique_labels))

monai/apps/deepgrow/transforms.py

@@ -288,6 +288,7 @@ def __init__(self, guidance: str = "guidance", discrepancy: str = "discrepancy",
         self._will_interact = None
 
     def randomize(self, data=None):
+        # pyrefly: ignore [unsupported-operation]
         probability = data[self.probability]
         self._will_interact = self.R.choice([True, False], p=[probability, 1.0 - probability])
 

monai/apps/detection/networks/retinanet_detector.py

@@ -525,6 +525,7 @@ def forward(
             )
 
         # 4. Generate anchors and store it in self.anchors: List[Tensor]
+        # pyrefly: ignore [bad-argument-type]
         self.generate_anchors(images, head_outputs)
         # num_anchor_locs_per_level: List[int], list of HW or HWD for each level
         num_anchor_locs_per_level = [x.shape[2:].numel() for x in head_outputs[self.cls_key]]
@@ -535,6 +536,7 @@ def forward(
             # reshape to Tensor sized(B, sum(HWA), self.num_classes) for self.cls_key
             # or (B, sum(HWA), 2* self.spatial_dims) for self.box_reg_key
             # A = self.num_anchors_per_loc
+            # pyrefly: ignore [bad-argument-type]
             head_outputs[key] = self._reshape_maps(head_outputs[key])
 
         # 6(1). If during training, return losses

monai/apps/detection/transforms/array.py

@@ -257,10 +257,14 @@ def __call__(self, boxes: NdarrayTensor, src_spatial_size: Sequence[int] | int |
                 diff = od - zd
                 half = abs(diff) // 2
                 if diff > 0:  # need padding (half, diff - half)
+                    # pyrefly: ignore [bad-index, unsupported-operation]
                     zoomed_boxes[:, axis] = zoomed_boxes[:, axis] + half
+                    # pyrefly: ignore [bad-index, unsupported-operation]
                     zoomed_boxes[:, axis + spatial_dims] = zoomed_boxes[:, axis + spatial_dims] + half
                 elif diff < 0:  # need slicing (half, half + od)
+                    # pyrefly: ignore [bad-index, unsupported-operation]
                     zoomed_boxes[:, axis] = zoomed_boxes[:, axis] - half
+                    # pyrefly: ignore [bad-index, unsupported-operation]
                     zoomed_boxes[:, axis + spatial_dims] = zoomed_boxes[:, axis + spatial_dims] - half
         return zoomed_boxes
 

monai/apps/detection/transforms/box_ops.py

@@ -186,7 +186,9 @@ def flip_boxes(
     _flip_boxes: NdarrayTensor = boxes.clone() if isinstance(boxes, torch.Tensor) else deepcopy(boxes)  # type: ignore[assignment]
 
     for axis in flip_axes:
+        # pyrefly: ignore [bad-index, unsupported-operation]
         _flip_boxes[:, axis + spatial_dims] = spatial_size[axis] - boxes[:, axis] - TO_REMOVE
+        # pyrefly: ignore [bad-index, unsupported-operation]
         _flip_boxes[:, axis] = spatial_size[axis] - boxes[:, axis + spatial_dims] - TO_REMOVE
 
     return _flip_boxes

monai/apps/generation/maisi/networks/autoencoderkl_maisi.py

@@ -246,7 +246,9 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
 
         # update padding length if necessary
         padding = 3
+        # pyrefly: ignore [unsupported-operation]
         if padding % self.stride > 0:
+            # pyrefly: ignore [unsupported-operation]
             padding = (padding // self.stride + 1) * self.stride
         if self.print_info:
             logger.info(f"Padding size: {padding}")

monai/apps/nnunet/nnunetv2_runner.py

@@ -274,6 +274,7 @@ def convert_dataset(self):
                 modality = [modality]
 
             create_new_dataset_json(
+                # pyrefly: ignore [bad-argument-type]
                 modality=modality,
                 num_foreground_classes=num_foreground_classes,
                 num_input_channels=num_input_channels,

monai/apps/pathology/transforms/post/dictionary.py

@@ -403,6 +403,7 @@ def __call__(self, data):
         d = dict(data)
         for key in self.key_iterator(d):
             offset = d[self.offset_key] if self.offset_key else None
+            # pyrefly: ignore [bad-argument-type]
             centroid = self.converter(d[key], offset)
             key_to_add = f"{key}_{self.centroid_key_postfix}"
             if key_to_add in d:

monai/auto3dseg/operations.py

@@ -149,4 +149,5 @@ def evaluate(self, data: Any, **kwargs: Any) -> dict:
         Args:
             data: input data
         """
+        # pyrefly: ignore [missing-attribute]
         return {k: v(data[k], **kwargs).tolist() for k, v in self.data.items() if (callable(v) and k in data)}

monai/auto3dseg/seg_summarizer.py

@@ -208,6 +208,7 @@ def summarize(self, data: list[dict]) -> dict[str, dict]:
 
         for analyzer in self.summary_analyzers:
             if callable(analyzer):
+                # pyrefly: ignore [missing-attribute]
                 report.update({analyzer.stats_name: analyzer(data)})
 
         return report

monai/bundle/reference_resolver.py

@@ -254,6 +254,7 @@ def iter_subconfigs(cls, id: str, config: Any) -> Iterator[tuple[str, str, Any]]
         """
         for k, v in config.items() if isinstance(config, dict) else enumerate(config):
             sub_id = f"{id}{cls.sep}{k}" if id != "" else f"{k}"
+            # pyrefly: ignore [invalid-yield]
             yield k, sub_id, v
 
     @classmethod

monai/bundle/scripts.py

@@ -1966,6 +1966,7 @@ def create_workflow(
         )
 
     if config_file is not None:
+        # pyrefly: ignore [unexpected-keyword]
         workflow_ = workflow_class(config_file=config_file, **_args)
     else:
         workflow_ = workflow_class(**_args)

monai/bundle/workflows.py

@@ -224,6 +224,7 @@ def add_property(self, name: str, required: str, desc: str | None = None) -> Non
             desc: descriptions for the property.
         """
         if self.properties is None:
+            # pyrefly: ignore [bad-assignment]
             self.properties = {}
         if name in self.properties:
             logger.warning(f"property '{name}' already exists in the properties list, overriding it.")
@@ -329,6 +330,7 @@ def _get_property(self, name: str, property: dict) -> Any:
         elif name in self._props_vals:
             value = self._props_vals[name]
         elif name in self.parser.config[self.parser.meta_key]:  # type: ignore[index]
+            # pyrefly: ignore [missing-attribute]
             id = self.properties.get(name, None).get(BundlePropertyConfig.ID, None)
             value = self.parser[id]
         else:
@@ -621,6 +623,7 @@ def _check_optional_id(self, name: str, property: dict) -> bool:
         else:
             ref = self.parser.get(ref_id, None)
         # for reference IDs that not refer to a property directly but using expressions, skip the check
+        # pyrefly: ignore [unsupported-operation]
         if ref is not None and not ref.startswith(EXPR_KEY) and ref != ID_REF_KEY + id:
             return False
         return True

monai/data/dataset.py

@@ -1648,6 +1648,7 @@ def _cachecheck(self, item_transformed):
                             item_k = kvikio_numpy.fromfile(
                                 f"{hashfile}-{k}-{i}", dtype=meta_i_k["dtype"], like=cp.empty(())
                             )
+                            # pyrefly: ignore [missing-attribute]
                             item_k = convert_to_tensor(item[i].reshape(meta_i_k["shape"]), device=f"cuda:{self.device}")
                             item[i].update({k: item_k, f"{k}_meta_dict": meta_i_k})
                     return item

monai/data/grid_dataset.py

@@ -142,6 +142,7 @@ def __call__(
         self, data: Mapping[Hashable, NdarrayTensor]
     ) -> Generator[tuple[Mapping[Hashable, NdarrayTensor], np.ndarray], None, None]:
         d = dict(data)
+        # pyrefly: ignore [missing-attribute]
         original_spatial_shape = d[first(self.keys)].shape[1:]
 
         for patch in zip(*[self.patch_iter(d[key]) for key in self.keys]):

monai/data/image_reader.py

@@ -22,7 +22,7 @@
 from collections.abc import Callable, Iterable, Iterator, Sequence
 from dataclasses import dataclass
 from pathlib import Path
-from typing import TYPE_CHECKING, Any, TypeAlias
+from typing import TYPE_CHECKING, Any, TypeAlias  # pyrefly: ignore [missing-module-attribute]
 
 import numpy as np
 from torch.utils.data._utils.collate import np_str_obj_array_pattern

monai/data/wsi_datasets.py

@@ -250,8 +250,10 @@ def __init__(
                     self.offset_limits = None
                 elif isinstance(offset_limits, tuple):
                     if isinstance(offset_limits[0], int):
+                        # pyrefly: ignore [bad-assignment]
                         self.offset_limits = (offset_limits, offset_limits)
                     elif isinstance(offset_limits[0], tuple):
+                        # pyrefly: ignore [bad-assignment]
                         self.offset_limits = offset_limits
                     else:
                         raise ValueError(

monai/data/wsi_reader.py

@@ -319,6 +319,7 @@ def _get_metadata(
         }
         return metadata
 
+    # pyrefly: ignore [bad-override]
     def get_data(
         self,
         wsi,

monai/engines/evaluator.py

@@ -489,11 +489,13 @@ def _iteration(self, engine: EnsembleEvaluator, batchdata: dict[str, torch.Tenso
                 if engine.amp:
                     with torch.autocast("cuda", **engine.amp_kwargs):
                         if isinstance(engine.state.output, dict):
+                            # pyrefly: ignore [no-matching-overload]
                             engine.state.output.update(
                                 {engine.pred_keys[idx]: engine.inferer(inputs, network, *args, **kwargs)}
                             )
                 else:
                     if isinstance(engine.state.output, dict):
+                        # pyrefly: ignore [no-matching-overload]
                         engine.state.output.update(
                             {engine.pred_keys[idx]: engine.inferer(inputs, network, *args, **kwargs)}
                         )

monai/engines/trainer.py

@@ -774,4 +774,5 @@ def _compute_discriminator_loss() -> None:
             engine.state.output[AdversarialKeys.DISCRIMINATOR_LOSS].backward()
             engine.state.d_optimizer.step()
 
+        # pyrefly: ignore [bad-return]
         return engine.state.output

monai/fl/client/monai_algo.py

@@ -251,6 +251,7 @@ def _get_data_key_stats(self, data, data_key, hist_bins, hist_range, output_path
             dataroot=self.workflow.dataset_dir,  # type: ignore
             hist_bins=hist_bins,
             hist_range=hist_range,
+            # pyrefly: ignore [bad-argument-type]
             output_path=output_path,
             histogram_only=self.histogram_only,
         )

monai/handlers/mlflow_handler.py

@@ -234,6 +234,7 @@ def start(self, engine: Engine) -> None:
         self._log_params(attrs)
 
         if self.dataset_logger:
+            # pyrefly: ignore [bad-argument-type]
             self.dataset_logger(self.dataset_dict)
         else:
             self._default_dataset_log(self.dataset_dict)
@@ -257,6 +258,7 @@ def _set_experiment(self):
                     else:
                         raise e
 
+        # pyrefly: ignore [missing-attribute]
         if experiment.lifecycle_stage != mlflow.entities.LifecycleStage.ACTIVE:
             raise ValueError(f"Cannot set a deleted experiment '{self.experiment_name}' as the active experiment")
         self.experiment = experiment

monai/inferers/inferer.py

@@ -841,6 +841,7 @@ def network_wrapper(
 
         if isinstance(out, Mapping):
             for k in out.keys():
+                # pyrefly: ignore [unsupported-operation]
                 out[k] = out[k].unsqueeze(dim=self.spatial_dim + 2)
             return out
 

monai/losses/image_dissimilarity.py

@@ -241,6 +241,7 @@ def __init__(
         if self.kernel_type == "gaussian":
             self.register_buffer("preterm", 1 / (2 * sigma**2), persistent=False)
             self.register_buffer("bin_centers", bin_centers[None, None, ...], persistent=False)
+
         self.smooth_nr = float(smooth_nr)
         self.smooth_dr = float(smooth_dr)
 

monai/metrics/utils.py

@@ -217,6 +217,7 @@ def get_mask_edges(
         or_vol = seg_pred | seg_gt
         if not or_vol.any():
             pred, gt = lib.zeros(seg_pred.shape, dtype=bool), lib.zeros(seg_gt.shape, dtype=bool)
+            # pyrefly: ignore [bad-return]
             return (pred, gt) if spacing is None else (pred, gt, pred, gt)
         channel_first = [seg_pred[None], seg_gt[None], or_vol[None]]
         if spacing is None and not use_cucim:  # cpu only erosion

monai/networks/blocks/hyena.py

@@ -98,21 +98,21 @@ class _DepthwiseFFTForward:
     _spatial_dims: int  # set by subclasses
     fft_chunk_size: int = 0  # 0 = no chunking; set in subclass __init__
 
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        spatial = x.shape[2:]
+    def forward(self, input: torch.Tensor) -> torch.Tensor:
+        spatial = input.shape[2:]
         kernel_shape = self.weight.shape[2:]  # type: ignore[attr-defined]
         fft_dims = tuple(range(-self._spatial_dims, 0))
         fft_size = [s + k - 1 for s, k in zip(spatial, kernel_shape)]
-        in_dtype = x.dtype
+        in_dtype = input.dtype
 
         slices = (slice(None), slice(None)) + tuple(slice(k // 2, k // 2 + s) for s, k in zip(spatial, kernel_shape))
 
         chunk = getattr(self, "fft_chunk_size", 0)
-        if chunk > 0 and x.shape[1] > chunk:
+        if chunk > 0 and input.shape[1] > chunk:
             parts = []
-            for c0 in range(0, x.shape[1], chunk):
-                c1 = min(c0 + chunk, x.shape[1])
-                xc = x[:, c0:c1].float()
+            for c0 in range(0, input.shape[1], chunk):
+                c1 = min(c0 + chunk, input.shape[1])
+                xc = input[:, c0:c1].float()
                 kc = self.weight[c0:c1].squeeze(1).float()  # type: ignore[attr-defined]
                 kc = kc.flip(list(range(1, self._spatial_dims + 1)))
                 xc_fft = torch.fft.rfftn(xc, s=fft_size, dim=fft_dims)
@@ -125,7 +125,7 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
                 del out_c
             return torch.cat(parts, dim=1)
 
-        x_f32 = x.float()
+        x_f32 = input.float()
         k_f32 = self.weight.squeeze(1).float()  # type: ignore[attr-defined]
         # PyTorch ``F.conv*`` computes cross-correlation; FFT computes convolution.
         # Flip the kernel so the FFT output matches ``Conv{2,3}d`` exactly.