Enable Pyrefly in fbcode/executorch (#21045) by maggiemoss · Pull Request #21045 · pytorch/executorch

maggiemoss · 2026-07-20T19:57:31Z

Summary:

Automated migration to enable Pyrefly type checking for fbcode/executorch.

Added python.set_pyrefly(True) to PACKAGE file
Suppressed pre-existing type errors

Pyrefly is Meta's next-generation Python type checker, replacing Pyre.

If you encounter issues, you can revert the PACKAGE change by removing
the python.set_pyrefly(True) line.
#pyreupgrade

Differential Revision: D112856308

pytorch-bot · 2026-07-20T19:57:34Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21045

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (1 Unrelated Failure)

As of commit adad0ed with merge base 2524691 ():

FLAKY - The following job failed but was likely due to flakiness present on trunk:

pull / test-arm-backend-no-driver (test_pytest_models_tosa) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

linux-foundation-easycla · 2026-07-20T19:57:37Z

❌ The email address for the commit (adad0ed) is not linked to the GitHub account, preventing the EasyCLA check. Consult this Help Article and GitHub Help to resolve. (To view the commit's email address, add .patch at the end of this PR page's URL.) For further assistance with EasyCLA, please visit our EasyCLA portal and chat with our support bot.

meta-codesync · 2026-07-20T19:57:43Z

@maggiemoss has exported this pull request. If you are a Meta employee, you can view the originating Diff in D112856308.

github-actions · 2026-07-20T19:58:24Z

This PR needs a `release notes:` label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

Copilot

Pull request overview

This PR is an automated migration to enable Pyrefly type checking for fbcode/executorch by adding targeted # pyrefly: ignore [...] suppressions for pre-existing type errors across Vulkan and Cadence backend code (plus backend test utilities).

Changes:

Added Pyrefly suppressions in Vulkan pattern-rewrite code paths to unblock type checking in FX node manipulation and metadata access.
Added Pyrefly suppressions in Cadence AOT passes/quantizer code paths where dynamic FX structures and heterogeneous argument types trip strict typing.
Added Pyrefly suppressions in backend test utilities/tests to accommodate type-checker limitations around verifier types and FX node metadata.

Reviewed changes

Copilot reviewed 24 out of 24 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
backends/vulkan/patterns/select_as_symint.py	Add Pyrefly suppressions for FX node/metadata typing in a pattern replacement.
backends/vulkan/patterns/sdpa.py	Add Pyrefly suppressions for FX `users`/metadata usage in SDPA rewrite.
backends/vulkan/patterns/rms_norm.py	Add Pyrefly suppressions for decorator typing and FX metadata propagation.
backends/vulkan/patterns/quantized_unary.py	Add Pyrefly suppressions around FX node target checks / attributes.
backends/vulkan/patterns/quantized_pixel_shuffle.py	Add Pyrefly suppressions for PatternMatch attribute typing and decorator usage.
backends/vulkan/patterns/quantized_linear.py	Add Pyrefly suppressions across quantized linear matching/rewrite and param mutation utilities.
backends/vulkan/patterns/quantized_embedding.py	Add Pyrefly suppressions for param extraction/mutation and FX metadata propagation.
backends/vulkan/patterns/quantized_convolution.py	Add Pyrefly suppressions for symbolic/int-like arithmetic and param update calls.
backends/vulkan/patterns/quantized_binary.py	Add Pyrefly suppressions for FX node target checks on binary quantized ops.
backends/vulkan/partitioner/vulkan_partitioner.py	Add Pyrefly suppression for calling a feature-provided predicate callback.
backends/test/program_builder.py	Add Pyrefly suppression for verifier list return typing.
backends/test/graph_builder.py	Add Pyrefly suppressions for attribute overrides in a helper ExportPass subclass.
backends/cadence/utils/facto_util.py	Add Pyrefly suppressions for constraint/spec mutation typing mismatches.
backends/cadence/aot/type_dispatch.py	Add Pyrefly suppressions for FX arg/shape access typing in operator dispatch.
backends/cadence/aot/tests/test_replace_ops_passes.py	Add Pyrefly suppression for expected-op-counts typing in a test helper call.
backends/cadence/aot/tests/test_quantizer_ops.py	Add Pyrefly suppression for a complex typed test-case table assignment.
backends/cadence/aot/tests/test_memory_passes.py	Add Pyrefly suppressions around optional metadata access in memory-allocation assertions.
backends/cadence/aot/tests/test_fusion_ops_passes.py	Add Pyrefly suppression for generator/sort specialization typing in a test.
backends/cadence/aot/replace_ops.py	Add Pyrefly suppressions for mixed FX arg types and slice/index computations in passes.
backends/cadence/aot/quantizer/quantizer.py	Add Pyrefly suppressions for pattern anchor extraction and quantizer list construction.
backends/cadence/aot/quantizer/patterns.py	Add Pyrefly suppressions across PartitionAnchors construction and FX-arg structure checks.
backends/cadence/aot/fuse_ops.py	Add Pyrefly suppressions for getattr-based packed-weight extraction typing.
backends/cadence/aot/compiler_utils.py	Add Pyrefly suppressions for getattr-based param access typing.
backends/cadence/aot/compiler_funcs.py	Add Pyrefly suppressions for numeric casts from heterogeneous FX arg tuples.

Comments suppressed due to low confidence (1)

backends/cadence/aot/tests/test_memory_passes.py:245

Same pattern again: arg_spec may be None but is used as if it’s always present. Add a direct assertion and drop the pyrefly: ignore [missing-attribute] suppressions so failures are clearer and type checking can narrow correctly.

        arg = cast(torch.fx.Node, node.args[0])
        arg_spec = arg.meta.get("spec", None)
        # pyrefly: ignore [missing-attribute]
        self.assertEqual(arg_spec.mem_id, spec.mem_id)
        self.assertEqual(
            spec.mem_offset,
            # pyrefly: ignore [missing-attribute]
            arg_spec.mem_offset + index * inner_dim_elements,
            # pyrefly: ignore [missing-attribute]
            f"{arg=} for node {node=} has wrong memory offset: {arg_spec.mem_offset=} for select on {dim=} {index=}, "
            f"but output has {spec.mem_offset=}"
            f"{spec=} {arg_spec=}",
        )

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

        for arg in cast(list[torch.fx.Node], node.args[0]):
            arg_spec = arg.meta.get("spec", None)
+            # pyrefly: ignore [missing-attribute]
            self.assertEqual(arg_spec.mem_id, node_spec.mem_id)
            actual_offset = node_spec.mem_offset + dim_offset * inner_dim_elements
            self.assertEqual(
+                # pyrefly: ignore [missing-attribute]
                arg_spec.mem_offset,
                actual_offset,
+                # pyrefly: ignore [missing-attribute]
                f"{arg=} of node {node=} has wrong memory offset: expected {arg_spec.mem_offset=}, but got {actual_offset=} = {node_spec.mem_offset=} + {dim_offset=} * {inner_dim_elements=}",
            )
+            # pyrefly: ignore [missing-attribute]
            dim_offset += arg_spec.shape[dim]


        arg = cast(torch.fx.Node, node.args[0])
        arg_spec = arg.meta.get("spec", None)
+        # pyrefly: ignore [missing-attribute]
        self.assertEqual(arg_spec.mem_id, spec.mem_id)
        self.assertEqual(
            spec.mem_offset,
+            # pyrefly: ignore [missing-attribute]
            arg_spec.mem_offset + start * inner_dim_elements,
+            # pyrefly: ignore [missing-attribute]
            f"{arg=} for node {node=} has wrong memory offset: {arg_spec.mem_offset=} {start=} for slice on {dim=}, but output has {spec.mem_offset=}",
        )


+        # pyrefly: ignore [not-callable]
        if not features.are_node_inputs_supported_fn(node):
            self.log_skip(node, "op args not supported")
            return False


    def get_verifiers(self) -> Optional[list[Verifier]]:
        if self.mode == IrMode.ATEN:
            return None
+        # pyrefly: ignore [bad-return]
        return [


    if bias_tensor is not None:
+        # pyrefly: ignore [bad-argument-type]
        utils.align_width_and_update_state_dict(ep, match.bias_node, bias_tensor)


Summary: Pull Request resolved: pytorch#21045 Automated migration to enable Pyrefly type checking for `fbcode/executorch`. - Added `python.set_pyrefly(True)` to PACKAGE file - Suppressed pre-existing type errors Pyrefly is Meta's next-generation Python type checker, replacing Pyre. If you encounter issues, you can revert the PACKAGE change by removing the `python.set_pyrefly(True)` line. #pyreupgrade Differential Revision: D112856308

Copilot

Pull request overview

Copilot reviewed 24 out of 24 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (2)

backends/test/program_builder.py:100

get_verifiers() advertises Optional[list[Verifier]] but it returns the result of EXIREdgeDialectVerifier(..., class_only=True), which is a verifier class (type[Verifier]) rather than a Verifier instance. The added pyrefly ignore hides a real type/API mismatch and makes it harder to reason about what ExportedProgram.verifiers expects.

    def get_verifiers(self) -> Optional[list[Verifier]]:
        if self.mode == IrMode.ATEN:
            return None
        # pyrefly: ignore [bad-return]
        return [
            EXIREdgeDialectVerifier(

backends/vulkan/partitioner/vulkan_partitioner.py:269

features.are_node_inputs_supported_fn is typed as Optional[Callable] (see op_registry), so calling it unconditionally can be a runtime error if it is ever set to None. The pyrefly ignore masks this; adding an assert keeps behavior the same today (default is non-None) while making the assumption explicit and satisfying the type checker.

        # pyrefly: ignore [not-callable]
        if not features.are_node_inputs_supported_fn(node):
            self.log_skip(node, "op args not supported")

+        # pyrefly: ignore [unsupported-operation]
        if (self.groups > 1 and self.groups < out_channels) and (
            out_channels / self.groups
        ) % 4 != 0:


Copilot AI review requested due to automatic review settings July 20, 2026 19:57

maggiemoss requested a review from SS-JIA as a code owner July 20, 2026 19:57

meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Jul 20, 2026

meta-codesync Bot added the meta-exported label Jul 20, 2026

meta-codesync Bot had a problem deploying to cadence July 20, 2026 19:57 Failure

Copilot started reviewing on behalf of maggiemoss July 20, 2026 19:58 View session

Copilot AI reviewed Jul 20, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings July 20, 2026 20:37

meta-codesync Bot changed the title ~~Enable Pyrefly in fbcode/executorch~~ Enable Pyrefly in fbcode/executorch (#21045) Jul 20, 2026

maggiemoss force-pushed the export-D112856308 branch from 9d05368 to adad0ed Compare July 20, 2026 20:37

Copilot started reviewing on behalf of maggiemoss July 20, 2026 20:38 View session

Copilot AI reviewed Jul 20, 2026

View reviewed changes

Comment thread backends/vulkan/patterns/quantized_convolution.py

Comment on lines +110 to 113

# pyrefly: ignore [unsupported-operation]

if (self.groups > 1 and self.groups < out_channels) and (

out_channels / self.groups

) % 4 != 0:

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Enable Pyrefly in fbcode/executorch (#21045)#21045

Enable Pyrefly in fbcode/executorch (#21045)#21045
maggiemoss wants to merge 1 commit into
pytorch:mainfrom
maggiemoss:export-D112856308

maggiemoss commented Jul 20, 2026 •

edited by meta-codesync Bot

Loading

Uh oh!

pytorch-bot Bot commented Jul 20, 2026 •

edited

Loading

Uh oh!

linux-foundation-easycla Bot commented Jul 20, 2026 •

edited

Loading

Uh oh!

meta-codesync Bot commented Jul 20, 2026

Uh oh!

github-actions Bot commented Jul 20, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

maggiemoss commented Jul 20, 2026 • edited by meta-codesync Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot Bot commented Jul 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21045

✅ You can merge normally! (1 Unrelated Failure)

Uh oh!

linux-foundation-easycla Bot commented Jul 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

meta-codesync Bot commented Jul 20, 2026

Uh oh!

github-actions Bot commented Jul 20, 2026

This PR needs a release notes: label

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

maggiemoss commented Jul 20, 2026 •

edited by meta-codesync Bot

Loading

pytorch-bot Bot commented Jul 20, 2026 •

edited

Loading

linux-foundation-easycla Bot commented Jul 20, 2026 •

edited

Loading

This PR needs a `release notes:` label