Ignored Module tests: provide required input tensor by psiddh · Pull Request #19028 · pytorch/executorch

psiddh · 2026-04-21T16:44:49Z

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations.

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. This commit was authored with the help of Claude.

pytorch-bot · 2026-04-21T16:44:57Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19028

📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

❌ 1 New Failure, 43 Pending, 1 Unrelated Failure

As of commit b3b7564 with merge base 66e4656 ():

NEW FAILURE - The following job has failed:

pull / test-qnn-passes-linux / linux-job (gh)
backends/qualcomm/tests/test_passes.py::TestPasses::test_mha_to_sha

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

pull / test-qnn-wheel-packages-linux (3.12) / 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.

github-actions · 2026-04-21T16:45:40Z

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

Fixes previously ignored Android instrumentation tests for Module by providing the required MobilenetV2 input tensor so forward()/execute() are invoked with valid arguments.

Changes:

Remove @Ignore annotations from four instrumentation tests.
Add a dummyInput() helper that constructs a [1, 3, 224, 224] float tensor and pass it into forward() / execute().
Update the multi-threaded forward test to use the dummy input.

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

Copilot · 2026-04-21T16:48:49Z

+ val results = module.forward(EValue.from(dummyInput()))
 Assert.assertTrue(results[0].isTensor)


In the multi-thread test each worker allocates a new 1x3x224x224 float tensor via dummyInput(). With 100 threads this can create significant short-lived allocations (~60MB+), which can make the instrumentation test flaky on memory-constrained devices. Consider pre-creating a single Tensor (or even EValue) and reusing it across threads/tests to avoid repeated allocations.

JacobSzwejbka · 2026-04-21T19:13:41Z

@claude review this PR

claude · 2026-04-21T19:14:24Z

Claude finished @JacobSzwejbka's task in 52s —— View job

Review of PR #19028

Read the changed file and diff
Analyze the changes
Post review feedback

The fix is correct. The 4 tests were @Ignored because forward() was called with no arguments on mobilenet_v2, which requires a [1, 3, 224, 224] float input tensor. Adding the dummyInput() helper and passing it to each call is the right approach.

Minor items:

Naming convention (ModuleInstrumentationTest.kt:166): INPUT_SHAPE uses SCREAMING_CASE, which is conventional for const val in Kotlin. However, longArrayOf(...) cannot be const val (arrays aren't compile-time constants), so it's a private val. The sibling file ModuleE2ETest.kt:74 uses inline longArrayOf(1, 3, 224, 224) rather than a named constant, so there's no local precedent either way. This is a very minor style nit — not worth blocking on.
Multi-threaded allocations (ModuleInstrumentationTest.kt:141): Each of the 100 threads creates a new 1×3×224×224 float tensor (~600KB each, ~60MB total). This is fine for a test — Tensor.ones is cheap, the tensors are short-lived, and the test is verifying thread-safety of Module.forward(), not allocation performance. Pre-sharing a tensor across threads would actually weaken the test by introducing a shared mutable input. No change needed.
testForwardOnDestroyedModule (ModuleInstrumentationTest.kt:125): This test still calls module.forward() with no arguments — but that's intentional, since the test asserts IllegalStateException from calling forward on a destroyed module. The exception is thrown before argument validation, so the zero-arg call is fine here.

Verdict: This is a clean, minimal fix. LGTM.

JacobSzwejbka · 2026-04-21T20:55:39Z

@claude explain the ci failures on this PR

claude · 2026-04-21T20:56:26Z

Claude finished @JacobSzwejbka's task in 1m 47s —— View job

Get CI status overview
Investigate the new failure (test_mha_to_sha)
Check if failures are related to PR changes
Summarize findings

CI Failure Analysis

This PR only modifies a single file (ModuleInstrumentationTest.kt). None of the CI failures are caused by this PR's changes.

1. New Failure: `test-qnn-passes-linux` / `test_mha_to_sha`

Not related to this PR. The test fails with:

ModuleNotFoundError: No module named 'executorch.examples.qualcomm'

at backends/qualcomm/tests/test_passes.py:152, where it tries to import executorch.examples.qualcomm.oss_scripts.llama.masking_utils. This is a CI environment/packaging issue — the examples/qualcomm directory isn't being installed as a Python package in the test Docker container. This PR doesn't touch any Qualcomm backend code.

The Dr. CI bot flags this as a "NEW FAILURE" (not present on the merge base), which likely means this test was recently added or its dependencies changed in a recent trunk commit. It's unrelated to the Android test fix in this PR.

2. Cancelled Job: `unittest-editable / windows / windows-job`

This was simply cancelled — not a failure. CI jobs get cancelled for various infrastructure reasons (timeouts, resource contention, etc.). Retrying should resolve it.

3. Broken Trunk (3 failures)

These all existed on the merge base commit (66e4656) before this PR:

test_mv2_model (macOS + macOS-editable): export/tests/test_target_recipes.py::TestTargetRecipes::test_mv2_model — pre-existing failure on trunk
Windows unittest: cancelled with The operation was canceled — infrastructure issue on trunk

Recommendation: Rebasing onto viable/strict (as Dr. CI suggests) would clear the broken trunk failures. The test_mha_to_sha failure needs to be investigated separately by the Qualcomm backend team — it's a missing module issue in the CI test environment.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

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

This commit was authored with the help of Claude.

## Summary Reverts the following Android PRs: - #19099 — Android: consistent error types across all modules - #19124 — Android: Module implements Closeable - #19092 — Android: improve error diagnostics for LlmModule and exceptions - #19028 — Ignored Module tests: provide required input tensor Authored with Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ch#19028 (pytorch#19133) ## Summary Reverts the following Android PRs: - pytorch#19099 — Android: consistent error types across all modules - pytorch#19124 — Android: Module implements Closeable - pytorch#19092 — Android: improve error diagnostics for LlmModule and exceptions - pytorch#19028 — Ignored Module tests: provide required input tensor Authored with Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

psiddh · 2026-04-21T16:44:49Z

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations.

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. This commit was authored with the help of Claude.

pytorch-bot · 2026-04-21T16:44:57Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19028

📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

❌ 1 New Failure, 43 Pending, 1 Unrelated Failure

As of commit b3b7564 with merge base 66e4656 ():

NEW FAILURE - The following job has failed:

pull / test-qnn-passes-linux / linux-job (gh)
backends/qualcomm/tests/test_passes.py::TestPasses::test_mha_to_sha

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

pull / test-qnn-wheel-packages-linux (3.12) / 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.

github-actions · 2026-04-21T16:45:40Z

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

Fixes previously ignored Android instrumentation tests for Module by providing the required MobilenetV2 input tensor so forward()/execute() are invoked with valid arguments.

Changes:

Remove @Ignore annotations from four instrumentation tests.
Add a dummyInput() helper that constructs a [1, 3, 224, 224] float tensor and pass it into forward() / execute().
Update the multi-threaded forward test to use the dummy input.

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

Copilot · 2026-04-21T16:48:49Z

+ val results = module.forward(EValue.from(dummyInput()))
 Assert.assertTrue(results[0].isTensor)


In the multi-thread test each worker allocates a new 1x3x224x224 float tensor via dummyInput(). With 100 threads this can create significant short-lived allocations (~60MB+), which can make the instrumentation test flaky on memory-constrained devices. Consider pre-creating a single Tensor (or even EValue) and reusing it across threads/tests to avoid repeated allocations.

JacobSzwejbka · 2026-04-21T19:13:41Z

@claude review this PR

claude · 2026-04-21T19:14:24Z

Claude finished @JacobSzwejbka's task in 52s —— View job

Review of PR #19028

Read the changed file and diff
Analyze the changes
Post review feedback

The fix is correct. The 4 tests were @Ignored because forward() was called with no arguments on mobilenet_v2, which requires a [1, 3, 224, 224] float input tensor. Adding the dummyInput() helper and passing it to each call is the right approach.

Minor items:

Naming convention (ModuleInstrumentationTest.kt:166): INPUT_SHAPE uses SCREAMING_CASE, which is conventional for const val in Kotlin. However, longArrayOf(...) cannot be const val (arrays aren't compile-time constants), so it's a private val. The sibling file ModuleE2ETest.kt:74 uses inline longArrayOf(1, 3, 224, 224) rather than a named constant, so there's no local precedent either way. This is a very minor style nit — not worth blocking on.
Multi-threaded allocations (ModuleInstrumentationTest.kt:141): Each of the 100 threads creates a new 1×3×224×224 float tensor (~600KB each, ~60MB total). This is fine for a test — Tensor.ones is cheap, the tensors are short-lived, and the test is verifying thread-safety of Module.forward(), not allocation performance. Pre-sharing a tensor across threads would actually weaken the test by introducing a shared mutable input. No change needed.
testForwardOnDestroyedModule (ModuleInstrumentationTest.kt:125): This test still calls module.forward() with no arguments — but that's intentional, since the test asserts IllegalStateException from calling forward on a destroyed module. The exception is thrown before argument validation, so the zero-arg call is fine here.

Verdict: This is a clean, minimal fix. LGTM.

JacobSzwejbka · 2026-04-21T20:55:39Z

@claude explain the ci failures on this PR

claude · 2026-04-21T20:56:26Z

Claude finished @JacobSzwejbka's task in 1m 47s —— View job

Get CI status overview
Investigate the new failure (test_mha_to_sha)
Check if failures are related to PR changes
Summarize findings

CI Failure Analysis

This PR only modifies a single file (ModuleInstrumentationTest.kt). None of the CI failures are caused by this PR's changes.

1. New Failure: `test-qnn-passes-linux` / `test_mha_to_sha`

Not related to this PR. The test fails with:

ModuleNotFoundError: No module named 'executorch.examples.qualcomm'

at backends/qualcomm/tests/test_passes.py:152, where it tries to import executorch.examples.qualcomm.oss_scripts.llama.masking_utils. This is a CI environment/packaging issue — the examples/qualcomm directory isn't being installed as a Python package in the test Docker container. This PR doesn't touch any Qualcomm backend code.

The Dr. CI bot flags this as a "NEW FAILURE" (not present on the merge base), which likely means this test was recently added or its dependencies changed in a recent trunk commit. It's unrelated to the Android test fix in this PR.

2. Cancelled Job: `unittest-editable / windows / windows-job`

This was simply cancelled — not a failure. CI jobs get cancelled for various infrastructure reasons (timeouts, resource contention, etc.). Retrying should resolve it.

3. Broken Trunk (3 failures)

These all existed on the merge base commit (66e4656) before this PR:

test_mv2_model (macOS + macOS-editable): export/tests/test_target_recipes.py::TestTargetRecipes::test_mv2_model — pre-existing failure on trunk
Windows unittest: cancelled with The operation was canceled — infrastructure issue on trunk

Recommendation: Rebasing onto viable/strict (as Dr. CI suggests) would clear the broken trunk failures. The test_mha_to_sha failure needs to be investigated separately by the Qualcomm backend team — it's a missing module issue in the CI test environment.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

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

This commit was authored with the help of Claude.

## Summary Reverts the following Android PRs: - #19099 — Android: consistent error types across all modules - #19124 — Android: Module implements Closeable - #19092 — Android: improve error diagnostics for LlmModule and exceptions - #19028 — Ignored Module tests: provide required input tensor Authored with Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ch#19028 (pytorch#19133) ## Summary Reverts the following Android PRs: - pytorch#19099 — Android: consistent error types across all modules - pytorch#19124 — Android: Module implements Closeable - pytorch#19092 — Android: improve error diagnostics for LlmModule and exceptions - pytorch#19028 — Ignored Module tests: provide required input tensor Authored with Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

psiddh · 2026-04-21T16:44:49Z

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations.

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. This commit was authored with the help of Claude.

pytorch-bot · 2026-04-21T16:44:57Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19028

📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

❌ 1 New Failure, 43 Pending, 1 Unrelated Failure

As of commit b3b7564 with merge base 66e4656 ():

NEW FAILURE - The following job has failed:

pull / test-qnn-passes-linux / linux-job (gh)
backends/qualcomm/tests/test_passes.py::TestPasses::test_mha_to_sha

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

pull / test-qnn-wheel-packages-linux (3.12) / 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.

github-actions · 2026-04-21T16:45:40Z

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

Fixes previously ignored Android instrumentation tests for Module by providing the required MobilenetV2 input tensor so forward()/execute() are invoked with valid arguments.

Changes:

Remove @Ignore annotations from four instrumentation tests.
Add a dummyInput() helper that constructs a [1, 3, 224, 224] float tensor and pass it into forward() / execute().
Update the multi-threaded forward test to use the dummy input.

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

Copilot · 2026-04-21T16:48:49Z

+ val results = module.forward(EValue.from(dummyInput()))
 Assert.assertTrue(results[0].isTensor)


In the multi-thread test each worker allocates a new 1x3x224x224 float tensor via dummyInput(). With 100 threads this can create significant short-lived allocations (~60MB+), which can make the instrumentation test flaky on memory-constrained devices. Consider pre-creating a single Tensor (or even EValue) and reusing it across threads/tests to avoid repeated allocations.

JacobSzwejbka · 2026-04-21T19:13:41Z

@claude review this PR

claude · 2026-04-21T19:14:24Z

Claude finished @JacobSzwejbka's task in 52s —— View job

Review of PR #19028

Read the changed file and diff
Analyze the changes
Post review feedback

The fix is correct. The 4 tests were @Ignored because forward() was called with no arguments on mobilenet_v2, which requires a [1, 3, 224, 224] float input tensor. Adding the dummyInput() helper and passing it to each call is the right approach.

Minor items:

Naming convention (ModuleInstrumentationTest.kt:166): INPUT_SHAPE uses SCREAMING_CASE, which is conventional for const val in Kotlin. However, longArrayOf(...) cannot be const val (arrays aren't compile-time constants), so it's a private val. The sibling file ModuleE2ETest.kt:74 uses inline longArrayOf(1, 3, 224, 224) rather than a named constant, so there's no local precedent either way. This is a very minor style nit — not worth blocking on.
Multi-threaded allocations (ModuleInstrumentationTest.kt:141): Each of the 100 threads creates a new 1×3×224×224 float tensor (~600KB each, ~60MB total). This is fine for a test — Tensor.ones is cheap, the tensors are short-lived, and the test is verifying thread-safety of Module.forward(), not allocation performance. Pre-sharing a tensor across threads would actually weaken the test by introducing a shared mutable input. No change needed.
testForwardOnDestroyedModule (ModuleInstrumentationTest.kt:125): This test still calls module.forward() with no arguments — but that's intentional, since the test asserts IllegalStateException from calling forward on a destroyed module. The exception is thrown before argument validation, so the zero-arg call is fine here.

Verdict: This is a clean, minimal fix. LGTM.

JacobSzwejbka · 2026-04-21T20:55:39Z

@claude explain the ci failures on this PR

claude · 2026-04-21T20:56:26Z

Claude finished @JacobSzwejbka's task in 1m 47s —— View job

Get CI status overview
Investigate the new failure (test_mha_to_sha)
Check if failures are related to PR changes
Summarize findings

CI Failure Analysis

This PR only modifies a single file (ModuleInstrumentationTest.kt). None of the CI failures are caused by this PR's changes.

1. New Failure: `test-qnn-passes-linux` / `test_mha_to_sha`

Not related to this PR. The test fails with:

ModuleNotFoundError: No module named 'executorch.examples.qualcomm'

at backends/qualcomm/tests/test_passes.py:152, where it tries to import executorch.examples.qualcomm.oss_scripts.llama.masking_utils. This is a CI environment/packaging issue — the examples/qualcomm directory isn't being installed as a Python package in the test Docker container. This PR doesn't touch any Qualcomm backend code.

The Dr. CI bot flags this as a "NEW FAILURE" (not present on the merge base), which likely means this test was recently added or its dependencies changed in a recent trunk commit. It's unrelated to the Android test fix in this PR.

2. Cancelled Job: `unittest-editable / windows / windows-job`

This was simply cancelled — not a failure. CI jobs get cancelled for various infrastructure reasons (timeouts, resource contention, etc.). Retrying should resolve it.

3. Broken Trunk (3 failures)

These all existed on the merge base commit (66e4656) before this PR:

test_mv2_model (macOS + macOS-editable): export/tests/test_target_recipes.py::TestTargetRecipes::test_mv2_model — pre-existing failure on trunk
Windows unittest: cancelled with The operation was canceled — infrastructure issue on trunk

Recommendation: Rebasing onto viable/strict (as Dr. CI suggests) would clear the broken trunk failures. The test_mha_to_sha failure needs to be investigated separately by the Qualcomm backend team — it's a missing module issue in the CI test environment.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

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

This commit was authored with the help of Claude.

## Summary Reverts the following Android PRs: - #19099 — Android: consistent error types across all modules - #19124 — Android: Module implements Closeable - #19092 — Android: improve error diagnostics for LlmModule and exceptions - #19028 — Ignored Module tests: provide required input tensor Authored with Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ch#19028 (pytorch#19133) ## Summary Reverts the following Android PRs: - pytorch#19099 — Android: consistent error types across all modules - pytorch#19124 — Android: Module implements Closeable - pytorch#19092 — Android: improve error diagnostics for LlmModule and exceptions - pytorch#19028 — Ignored Module tests: provide required input tensor Authored with Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

psiddh · 2026-04-21T16:44:49Z

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations.

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. This commit was authored with the help of Claude.

pytorch-bot · 2026-04-21T16:44:57Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19028

📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

❌ 1 New Failure, 43 Pending, 1 Unrelated Failure

As of commit b3b7564 with merge base 66e4656 ():

NEW FAILURE - The following job has failed:

pull / test-qnn-passes-linux / linux-job (gh)
backends/qualcomm/tests/test_passes.py::TestPasses::test_mha_to_sha

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

pull / test-qnn-wheel-packages-linux (3.12) / 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.

github-actions · 2026-04-21T16:45:40Z

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

Fixes previously ignored Android instrumentation tests for Module by providing the required MobilenetV2 input tensor so forward()/execute() are invoked with valid arguments.

Changes:

Remove @Ignore annotations from four instrumentation tests.
Add a dummyInput() helper that constructs a [1, 3, 224, 224] float tensor and pass it into forward() / execute().
Update the multi-threaded forward test to use the dummy input.

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

Copilot · 2026-04-21T16:48:49Z

+ val results = module.forward(EValue.from(dummyInput()))
 Assert.assertTrue(results[0].isTensor)


In the multi-thread test each worker allocates a new 1x3x224x224 float tensor via dummyInput(). With 100 threads this can create significant short-lived allocations (~60MB+), which can make the instrumentation test flaky on memory-constrained devices. Consider pre-creating a single Tensor (or even EValue) and reusing it across threads/tests to avoid repeated allocations.

JacobSzwejbka · 2026-04-21T19:13:41Z

@claude review this PR

claude · 2026-04-21T19:14:24Z

Claude finished @JacobSzwejbka's task in 52s —— View job

Review of PR #19028

Read the changed file and diff
Analyze the changes
Post review feedback

The fix is correct. The 4 tests were @Ignored because forward() was called with no arguments on mobilenet_v2, which requires a [1, 3, 224, 224] float input tensor. Adding the dummyInput() helper and passing it to each call is the right approach.

Minor items:

Naming convention (ModuleInstrumentationTest.kt:166): INPUT_SHAPE uses SCREAMING_CASE, which is conventional for const val in Kotlin. However, longArrayOf(...) cannot be const val (arrays aren't compile-time constants), so it's a private val. The sibling file ModuleE2ETest.kt:74 uses inline longArrayOf(1, 3, 224, 224) rather than a named constant, so there's no local precedent either way. This is a very minor style nit — not worth blocking on.
Multi-threaded allocations (ModuleInstrumentationTest.kt:141): Each of the 100 threads creates a new 1×3×224×224 float tensor (~600KB each, ~60MB total). This is fine for a test — Tensor.ones is cheap, the tensors are short-lived, and the test is verifying thread-safety of Module.forward(), not allocation performance. Pre-sharing a tensor across threads would actually weaken the test by introducing a shared mutable input. No change needed.
testForwardOnDestroyedModule (ModuleInstrumentationTest.kt:125): This test still calls module.forward() with no arguments — but that's intentional, since the test asserts IllegalStateException from calling forward on a destroyed module. The exception is thrown before argument validation, so the zero-arg call is fine here.

Verdict: This is a clean, minimal fix. LGTM.

JacobSzwejbka · 2026-04-21T20:55:39Z

@claude explain the ci failures on this PR

claude · 2026-04-21T20:56:26Z

Claude finished @JacobSzwejbka's task in 1m 47s —— View job

Get CI status overview
Investigate the new failure (test_mha_to_sha)
Check if failures are related to PR changes
Summarize findings

CI Failure Analysis

This PR only modifies a single file (ModuleInstrumentationTest.kt). None of the CI failures are caused by this PR's changes.

1. New Failure: `test-qnn-passes-linux` / `test_mha_to_sha`

Not related to this PR. The test fails with:

ModuleNotFoundError: No module named 'executorch.examples.qualcomm'

at backends/qualcomm/tests/test_passes.py:152, where it tries to import executorch.examples.qualcomm.oss_scripts.llama.masking_utils. This is a CI environment/packaging issue — the examples/qualcomm directory isn't being installed as a Python package in the test Docker container. This PR doesn't touch any Qualcomm backend code.

The Dr. CI bot flags this as a "NEW FAILURE" (not present on the merge base), which likely means this test was recently added or its dependencies changed in a recent trunk commit. It's unrelated to the Android test fix in this PR.

2. Cancelled Job: `unittest-editable / windows / windows-job`

This was simply cancelled — not a failure. CI jobs get cancelled for various infrastructure reasons (timeouts, resource contention, etc.). Retrying should resolve it.

3. Broken Trunk (3 failures)

These all existed on the merge base commit (66e4656) before this PR:

test_mv2_model (macOS + macOS-editable): export/tests/test_target_recipes.py::TestTargetRecipes::test_mv2_model — pre-existing failure on trunk
Windows unittest: cancelled with The operation was canceled — infrastructure issue on trunk

Recommendation: Rebasing onto viable/strict (as Dr. CI suggests) would clear the broken trunk failures. The test_mha_to_sha failure needs to be investigated separately by the Qualcomm backend team — it's a missing module issue in the CI test environment.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

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

This commit was authored with the help of Claude.

## Summary Reverts the following Android PRs: - #19099 — Android: consistent error types across all modules - #19124 — Android: Module implements Closeable - #19092 — Android: improve error diagnostics for LlmModule and exceptions - #19028 — Ignored Module tests: provide required input tensor Authored with Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ch#19028 (pytorch#19133) ## Summary Reverts the following Android PRs: - pytorch#19099 — Android: consistent error types across all modules - pytorch#19124 — Android: Module implements Closeable - pytorch#19092 — Android: improve error diagnostics for LlmModule and exceptions - pytorch#19028 — Ignored Module tests: provide required input tensor Authored with Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

psiddh · 2026-04-21T16:44:49Z

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations.

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. This commit was authored with the help of Claude.

pytorch-bot · 2026-04-21T16:44:57Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19028

📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

❌ 1 New Failure, 43 Pending, 1 Unrelated Failure

As of commit b3b7564 with merge base 66e4656 ():

NEW FAILURE - The following job has failed:

pull / test-qnn-passes-linux / linux-job (gh)
backends/qualcomm/tests/test_passes.py::TestPasses::test_mha_to_sha

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

pull / test-qnn-wheel-packages-linux (3.12) / 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.

github-actions · 2026-04-21T16:45:40Z

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

Fixes previously ignored Android instrumentation tests for Module by providing the required MobilenetV2 input tensor so forward()/execute() are invoked with valid arguments.

Changes:

Remove @Ignore annotations from four instrumentation tests.
Add a dummyInput() helper that constructs a [1, 3, 224, 224] float tensor and pass it into forward() / execute().
Update the multi-threaded forward test to use the dummy input.

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

Copilot · 2026-04-21T16:48:49Z

+ val results = module.forward(EValue.from(dummyInput()))
 Assert.assertTrue(results[0].isTensor)


In the multi-thread test each worker allocates a new 1x3x224x224 float tensor via dummyInput(). With 100 threads this can create significant short-lived allocations (~60MB+), which can make the instrumentation test flaky on memory-constrained devices. Consider pre-creating a single Tensor (or even EValue) and reusing it across threads/tests to avoid repeated allocations.

JacobSzwejbka · 2026-04-21T19:13:41Z

@claude review this PR

claude · 2026-04-21T19:14:24Z

Claude finished @JacobSzwejbka's task in 52s —— View job

Review of PR #19028

Read the changed file and diff
Analyze the changes
Post review feedback

The fix is correct. The 4 tests were @Ignored because forward() was called with no arguments on mobilenet_v2, which requires a [1, 3, 224, 224] float input tensor. Adding the dummyInput() helper and passing it to each call is the right approach.

Minor items:

Naming convention (ModuleInstrumentationTest.kt:166): INPUT_SHAPE uses SCREAMING_CASE, which is conventional for const val in Kotlin. However, longArrayOf(...) cannot be const val (arrays aren't compile-time constants), so it's a private val. The sibling file ModuleE2ETest.kt:74 uses inline longArrayOf(1, 3, 224, 224) rather than a named constant, so there's no local precedent either way. This is a very minor style nit — not worth blocking on.
Multi-threaded allocations (ModuleInstrumentationTest.kt:141): Each of the 100 threads creates a new 1×3×224×224 float tensor (~600KB each, ~60MB total). This is fine for a test — Tensor.ones is cheap, the tensors are short-lived, and the test is verifying thread-safety of Module.forward(), not allocation performance. Pre-sharing a tensor across threads would actually weaken the test by introducing a shared mutable input. No change needed.
testForwardOnDestroyedModule (ModuleInstrumentationTest.kt:125): This test still calls module.forward() with no arguments — but that's intentional, since the test asserts IllegalStateException from calling forward on a destroyed module. The exception is thrown before argument validation, so the zero-arg call is fine here.

Verdict: This is a clean, minimal fix. LGTM.

JacobSzwejbka · 2026-04-21T20:55:39Z

@claude explain the ci failures on this PR

claude · 2026-04-21T20:56:26Z

Claude finished @JacobSzwejbka's task in 1m 47s —— View job

Get CI status overview
Investigate the new failure (test_mha_to_sha)
Check if failures are related to PR changes
Summarize findings

CI Failure Analysis

This PR only modifies a single file (ModuleInstrumentationTest.kt). None of the CI failures are caused by this PR's changes.

1. New Failure: `test-qnn-passes-linux` / `test_mha_to_sha`

Not related to this PR. The test fails with:

ModuleNotFoundError: No module named 'executorch.examples.qualcomm'

at backends/qualcomm/tests/test_passes.py:152, where it tries to import executorch.examples.qualcomm.oss_scripts.llama.masking_utils. This is a CI environment/packaging issue — the examples/qualcomm directory isn't being installed as a Python package in the test Docker container. This PR doesn't touch any Qualcomm backend code.

The Dr. CI bot flags this as a "NEW FAILURE" (not present on the merge base), which likely means this test was recently added or its dependencies changed in a recent trunk commit. It's unrelated to the Android test fix in this PR.

2. Cancelled Job: `unittest-editable / windows / windows-job`

This was simply cancelled — not a failure. CI jobs get cancelled for various infrastructure reasons (timeouts, resource contention, etc.). Retrying should resolve it.

3. Broken Trunk (3 failures)

These all existed on the merge base commit (66e4656) before this PR:

test_mv2_model (macOS + macOS-editable): export/tests/test_target_recipes.py::TestTargetRecipes::test_mv2_model — pre-existing failure on trunk
Windows unittest: cancelled with The operation was canceled — infrastructure issue on trunk

Recommendation: Rebasing onto viable/strict (as Dr. CI suggests) would clear the broken trunk failures. The test_mha_to_sha failure needs to be investigated separately by the Qualcomm backend team — it's a missing module issue in the CI test environment.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

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

This commit was authored with the help of Claude.

## Summary Reverts the following Android PRs: - #19099 — Android: consistent error types across all modules - #19124 — Android: Module implements Closeable - #19092 — Android: improve error diagnostics for LlmModule and exceptions - #19028 — Ignored Module tests: provide required input tensor Authored with Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ch#19028 (pytorch#19133) ## Summary Reverts the following Android PRs: - pytorch#19099 — Android: consistent error types across all modules - pytorch#19124 — Android: Module implements Closeable - pytorch#19092 — Android: improve error diagnostics for LlmModule and exceptions - pytorch#19028 — Ignored Module tests: provide required input tensor Authored with Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

psiddh · 2026-04-21T16:44:49Z

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations.

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. This commit was authored with the help of Claude.

pytorch-bot · 2026-04-21T16:44:57Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19028

📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

❌ 1 New Failure, 43 Pending, 1 Unrelated Failure

As of commit b3b7564 with merge base 66e4656 ():

NEW FAILURE - The following job has failed:

pull / test-qnn-passes-linux / linux-job (gh)
backends/qualcomm/tests/test_passes.py::TestPasses::test_mha_to_sha

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

pull / test-qnn-wheel-packages-linux (3.12) / 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.

github-actions · 2026-04-21T16:45:40Z

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

Fixes previously ignored Android instrumentation tests for Module by providing the required MobilenetV2 input tensor so forward()/execute() are invoked with valid arguments.

Changes:

Remove @Ignore annotations from four instrumentation tests.
Add a dummyInput() helper that constructs a [1, 3, 224, 224] float tensor and pass it into forward() / execute().
Update the multi-threaded forward test to use the dummy input.

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

Copilot · 2026-04-21T16:48:49Z

+ val results = module.forward(EValue.from(dummyInput()))
 Assert.assertTrue(results[0].isTensor)


In the multi-thread test each worker allocates a new 1x3x224x224 float tensor via dummyInput(). With 100 threads this can create significant short-lived allocations (~60MB+), which can make the instrumentation test flaky on memory-constrained devices. Consider pre-creating a single Tensor (or even EValue) and reusing it across threads/tests to avoid repeated allocations.

JacobSzwejbka · 2026-04-21T19:13:41Z

@claude review this PR

claude · 2026-04-21T19:14:24Z

Claude finished @JacobSzwejbka's task in 52s —— View job

Review of PR #19028

Read the changed file and diff
Analyze the changes
Post review feedback

The fix is correct. The 4 tests were @Ignored because forward() was called with no arguments on mobilenet_v2, which requires a [1, 3, 224, 224] float input tensor. Adding the dummyInput() helper and passing it to each call is the right approach.

Minor items:

Naming convention (ModuleInstrumentationTest.kt:166): INPUT_SHAPE uses SCREAMING_CASE, which is conventional for const val in Kotlin. However, longArrayOf(...) cannot be const val (arrays aren't compile-time constants), so it's a private val. The sibling file ModuleE2ETest.kt:74 uses inline longArrayOf(1, 3, 224, 224) rather than a named constant, so there's no local precedent either way. This is a very minor style nit — not worth blocking on.
Multi-threaded allocations (ModuleInstrumentationTest.kt:141): Each of the 100 threads creates a new 1×3×224×224 float tensor (~600KB each, ~60MB total). This is fine for a test — Tensor.ones is cheap, the tensors are short-lived, and the test is verifying thread-safety of Module.forward(), not allocation performance. Pre-sharing a tensor across threads would actually weaken the test by introducing a shared mutable input. No change needed.
testForwardOnDestroyedModule (ModuleInstrumentationTest.kt:125): This test still calls module.forward() with no arguments — but that's intentional, since the test asserts IllegalStateException from calling forward on a destroyed module. The exception is thrown before argument validation, so the zero-arg call is fine here.

Verdict: This is a clean, minimal fix. LGTM.

JacobSzwejbka · 2026-04-21T20:55:39Z

@claude explain the ci failures on this PR

claude · 2026-04-21T20:56:26Z

Claude finished @JacobSzwejbka's task in 1m 47s —— View job

Get CI status overview
Investigate the new failure (test_mha_to_sha)
Check if failures are related to PR changes
Summarize findings

CI Failure Analysis

This PR only modifies a single file (ModuleInstrumentationTest.kt). None of the CI failures are caused by this PR's changes.

1. New Failure: `test-qnn-passes-linux` / `test_mha_to_sha`

Not related to this PR. The test fails with:

ModuleNotFoundError: No module named 'executorch.examples.qualcomm'

at backends/qualcomm/tests/test_passes.py:152, where it tries to import executorch.examples.qualcomm.oss_scripts.llama.masking_utils. This is a CI environment/packaging issue — the examples/qualcomm directory isn't being installed as a Python package in the test Docker container. This PR doesn't touch any Qualcomm backend code.

The Dr. CI bot flags this as a "NEW FAILURE" (not present on the merge base), which likely means this test was recently added or its dependencies changed in a recent trunk commit. It's unrelated to the Android test fix in this PR.

2. Cancelled Job: `unittest-editable / windows / windows-job`

This was simply cancelled — not a failure. CI jobs get cancelled for various infrastructure reasons (timeouts, resource contention, etc.). Retrying should resolve it.

3. Broken Trunk (3 failures)

These all existed on the merge base commit (66e4656) before this PR:

test_mv2_model (macOS + macOS-editable): export/tests/test_target_recipes.py::TestTargetRecipes::test_mv2_model — pre-existing failure on trunk
Windows unittest: cancelled with The operation was canceled — infrastructure issue on trunk

Recommendation: Rebasing onto viable/strict (as Dr. CI suggests) would clear the broken trunk failures. The test_mha_to_sha failure needs to be investigated separately by the Qualcomm backend team — it's a missing module issue in the CI test environment.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

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

This commit was authored with the help of Claude.

## Summary Reverts the following Android PRs: - #19099 — Android: consistent error types across all modules - #19124 — Android: Module implements Closeable - #19092 — Android: improve error diagnostics for LlmModule and exceptions - #19028 — Ignored Module tests: provide required input tensor Authored with Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ch#19028 (pytorch#19133) ## Summary Reverts the following Android PRs: - pytorch#19099 — Android: consistent error types across all modules - pytorch#19124 — Android: Module implements Closeable - pytorch#19092 — Android: improve error diagnostics for LlmModule and exceptions - pytorch#19028 — Ignored Module tests: provide required input tensor Authored with Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

psiddh · 2026-04-21T16:44:49Z

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations.

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. This commit was authored with the help of Claude.

pytorch-bot · 2026-04-21T16:44:57Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19028

📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

❌ 1 New Failure, 43 Pending, 1 Unrelated Failure

As of commit b3b7564 with merge base 66e4656 ():

NEW FAILURE - The following job has failed:

pull / test-qnn-passes-linux / linux-job (gh)
backends/qualcomm/tests/test_passes.py::TestPasses::test_mha_to_sha

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

pull / test-qnn-wheel-packages-linux (3.12) / 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.

github-actions · 2026-04-21T16:45:40Z

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

Fixes previously ignored Android instrumentation tests for Module by providing the required MobilenetV2 input tensor so forward()/execute() are invoked with valid arguments.

Changes:

Remove @Ignore annotations from four instrumentation tests.
Add a dummyInput() helper that constructs a [1, 3, 224, 224] float tensor and pass it into forward() / execute().
Update the multi-threaded forward test to use the dummy input.

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

Copilot · 2026-04-21T16:48:49Z

+ val results = module.forward(EValue.from(dummyInput()))
 Assert.assertTrue(results[0].isTensor)


In the multi-thread test each worker allocates a new 1x3x224x224 float tensor via dummyInput(). With 100 threads this can create significant short-lived allocations (~60MB+), which can make the instrumentation test flaky on memory-constrained devices. Consider pre-creating a single Tensor (or even EValue) and reusing it across threads/tests to avoid repeated allocations.

JacobSzwejbka · 2026-04-21T19:13:41Z

@claude review this PR

claude · 2026-04-21T19:14:24Z

Claude finished @JacobSzwejbka's task in 52s —— View job

Review of PR #19028

Read the changed file and diff
Analyze the changes
Post review feedback

The fix is correct. The 4 tests were @Ignored because forward() was called with no arguments on mobilenet_v2, which requires a [1, 3, 224, 224] float input tensor. Adding the dummyInput() helper and passing it to each call is the right approach.

Minor items:

Naming convention (ModuleInstrumentationTest.kt:166): INPUT_SHAPE uses SCREAMING_CASE, which is conventional for const val in Kotlin. However, longArrayOf(...) cannot be const val (arrays aren't compile-time constants), so it's a private val. The sibling file ModuleE2ETest.kt:74 uses inline longArrayOf(1, 3, 224, 224) rather than a named constant, so there's no local precedent either way. This is a very minor style nit — not worth blocking on.
Multi-threaded allocations (ModuleInstrumentationTest.kt:141): Each of the 100 threads creates a new 1×3×224×224 float tensor (~600KB each, ~60MB total). This is fine for a test — Tensor.ones is cheap, the tensors are short-lived, and the test is verifying thread-safety of Module.forward(), not allocation performance. Pre-sharing a tensor across threads would actually weaken the test by introducing a shared mutable input. No change needed.
testForwardOnDestroyedModule (ModuleInstrumentationTest.kt:125): This test still calls module.forward() with no arguments — but that's intentional, since the test asserts IllegalStateException from calling forward on a destroyed module. The exception is thrown before argument validation, so the zero-arg call is fine here.

Verdict: This is a clean, minimal fix. LGTM.

JacobSzwejbka · 2026-04-21T20:55:39Z

@claude explain the ci failures on this PR

claude · 2026-04-21T20:56:26Z

Claude finished @JacobSzwejbka's task in 1m 47s —— View job

Get CI status overview
Investigate the new failure (test_mha_to_sha)
Check if failures are related to PR changes
Summarize findings

CI Failure Analysis

This PR only modifies a single file (ModuleInstrumentationTest.kt). None of the CI failures are caused by this PR's changes.

1. New Failure: `test-qnn-passes-linux` / `test_mha_to_sha`

Not related to this PR. The test fails with:

ModuleNotFoundError: No module named 'executorch.examples.qualcomm'

at backends/qualcomm/tests/test_passes.py:152, where it tries to import executorch.examples.qualcomm.oss_scripts.llama.masking_utils. This is a CI environment/packaging issue — the examples/qualcomm directory isn't being installed as a Python package in the test Docker container. This PR doesn't touch any Qualcomm backend code.

The Dr. CI bot flags this as a "NEW FAILURE" (not present on the merge base), which likely means this test was recently added or its dependencies changed in a recent trunk commit. It's unrelated to the Android test fix in this PR.

2. Cancelled Job: `unittest-editable / windows / windows-job`

This was simply cancelled — not a failure. CI jobs get cancelled for various infrastructure reasons (timeouts, resource contention, etc.). Retrying should resolve it.

3. Broken Trunk (3 failures)

These all existed on the merge base commit (66e4656) before this PR:

test_mv2_model (macOS + macOS-editable): export/tests/test_target_recipes.py::TestTargetRecipes::test_mv2_model — pre-existing failure on trunk
Windows unittest: cancelled with The operation was canceled — infrastructure issue on trunk

Recommendation: Rebasing onto viable/strict (as Dr. CI suggests) would clear the broken trunk failures. The test_mha_to_sha failure needs to be investigated separately by the Qualcomm backend team — it's a missing module issue in the CI test environment.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

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

This commit was authored with the help of Claude.

## Summary Reverts the following Android PRs: - #19099 — Android: consistent error types across all modules - #19124 — Android: Module implements Closeable - #19092 — Android: improve error diagnostics for LlmModule and exceptions - #19028 — Ignored Module tests: provide required input tensor Authored with Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ch#19028 (pytorch#19133) ## Summary Reverts the following Android PRs: - pytorch#19099 — Android: consistent error types across all modules - pytorch#19124 — Android: Module implements Closeable - pytorch#19092 — Android: improve error diagnostics for LlmModule and exceptions - pytorch#19028 — Ignored Module tests: provide required input tensor Authored with Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

psiddh · 2026-04-21T16:44:49Z

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations.

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. This commit was authored with the help of Claude.

pytorch-bot · 2026-04-21T16:44:57Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19028

📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

❌ 1 New Failure, 43 Pending, 1 Unrelated Failure

As of commit b3b7564 with merge base 66e4656 ():

NEW FAILURE - The following job has failed:

pull / test-qnn-passes-linux / linux-job (gh)
backends/qualcomm/tests/test_passes.py::TestPasses::test_mha_to_sha

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

pull / test-qnn-wheel-packages-linux (3.12) / 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.

github-actions · 2026-04-21T16:45:40Z

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

Fixes previously ignored Android instrumentation tests for Module by providing the required MobilenetV2 input tensor so forward()/execute() are invoked with valid arguments.

Changes:

Remove @Ignore annotations from four instrumentation tests.
Add a dummyInput() helper that constructs a [1, 3, 224, 224] float tensor and pass it into forward() / execute().
Update the multi-threaded forward test to use the dummy input.

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

Copilot · 2026-04-21T16:48:49Z

+ val results = module.forward(EValue.from(dummyInput()))
 Assert.assertTrue(results[0].isTensor)


In the multi-thread test each worker allocates a new 1x3x224x224 float tensor via dummyInput(). With 100 threads this can create significant short-lived allocations (~60MB+), which can make the instrumentation test flaky on memory-constrained devices. Consider pre-creating a single Tensor (or even EValue) and reusing it across threads/tests to avoid repeated allocations.

JacobSzwejbka · 2026-04-21T19:13:41Z

@claude review this PR

claude · 2026-04-21T19:14:24Z

Claude finished @JacobSzwejbka's task in 52s —— View job

Review of PR #19028

Read the changed file and diff
Analyze the changes
Post review feedback

The fix is correct. The 4 tests were @Ignored because forward() was called with no arguments on mobilenet_v2, which requires a [1, 3, 224, 224] float input tensor. Adding the dummyInput() helper and passing it to each call is the right approach.

Minor items:

Naming convention (ModuleInstrumentationTest.kt:166): INPUT_SHAPE uses SCREAMING_CASE, which is conventional for const val in Kotlin. However, longArrayOf(...) cannot be const val (arrays aren't compile-time constants), so it's a private val. The sibling file ModuleE2ETest.kt:74 uses inline longArrayOf(1, 3, 224, 224) rather than a named constant, so there's no local precedent either way. This is a very minor style nit — not worth blocking on.
Multi-threaded allocations (ModuleInstrumentationTest.kt:141): Each of the 100 threads creates a new 1×3×224×224 float tensor (~600KB each, ~60MB total). This is fine for a test — Tensor.ones is cheap, the tensors are short-lived, and the test is verifying thread-safety of Module.forward(), not allocation performance. Pre-sharing a tensor across threads would actually weaken the test by introducing a shared mutable input. No change needed.
testForwardOnDestroyedModule (ModuleInstrumentationTest.kt:125): This test still calls module.forward() with no arguments — but that's intentional, since the test asserts IllegalStateException from calling forward on a destroyed module. The exception is thrown before argument validation, so the zero-arg call is fine here.

Verdict: This is a clean, minimal fix. LGTM.

JacobSzwejbka · 2026-04-21T20:55:39Z

@claude explain the ci failures on this PR

claude · 2026-04-21T20:56:26Z

Claude finished @JacobSzwejbka's task in 1m 47s —— View job

Get CI status overview
Investigate the new failure (test_mha_to_sha)
Check if failures are related to PR changes
Summarize findings

CI Failure Analysis

This PR only modifies a single file (ModuleInstrumentationTest.kt). None of the CI failures are caused by this PR's changes.

1. New Failure: `test-qnn-passes-linux` / `test_mha_to_sha`

Not related to this PR. The test fails with:

ModuleNotFoundError: No module named 'executorch.examples.qualcomm'

at backends/qualcomm/tests/test_passes.py:152, where it tries to import executorch.examples.qualcomm.oss_scripts.llama.masking_utils. This is a CI environment/packaging issue — the examples/qualcomm directory isn't being installed as a Python package in the test Docker container. This PR doesn't touch any Qualcomm backend code.

The Dr. CI bot flags this as a "NEW FAILURE" (not present on the merge base), which likely means this test was recently added or its dependencies changed in a recent trunk commit. It's unrelated to the Android test fix in this PR.

2. Cancelled Job: `unittest-editable / windows / windows-job`

This was simply cancelled — not a failure. CI jobs get cancelled for various infrastructure reasons (timeouts, resource contention, etc.). Retrying should resolve it.

3. Broken Trunk (3 failures)

These all existed on the merge base commit (66e4656) before this PR:

test_mv2_model (macOS + macOS-editable): export/tests/test_target_recipes.py::TestTargetRecipes::test_mv2_model — pre-existing failure on trunk
Windows unittest: cancelled with The operation was canceled — infrastructure issue on trunk

Recommendation: Rebasing onto viable/strict (as Dr. CI suggests) would clear the broken trunk failures. The test_mha_to_sha failure needs to be investigated separately by the Qualcomm backend team — it's a missing module issue in the CI test environment.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

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

This commit was authored with the help of Claude.

## Summary Reverts the following Android PRs: - #19099 — Android: consistent error types across all modules - #19124 — Android: Module implements Closeable - #19092 — Android: improve error diagnostics for LlmModule and exceptions - #19028 — Ignored Module tests: provide required input tensor Authored with Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

@ignored

…t coverage Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All @DoNotStrip annotations preserved on JNI-called methods; new @DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Tests: fix 4 @ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent-close tests. Fix ModelRunner crash on loadMethod failure. This PR was authored with the help of Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@ignore

All 4 tests failed because they called forward() with zero arguments on mobilenet_v2 which expects a [1,3,224,224] float input. This was a test bug, not a runtime bug. Add a dummyInput() helper that creates a Tensor.ones with the correct shape, and remove all @ignore annotations. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ch#19028 (pytorch#19133) ## Summary Reverts the following Android PRs: - pytorch#19099 — Android: consistent error types across all modules - pytorch#19124 — Android: Module implements Closeable - pytorch#19092 — Android: improve error diagnostics for LlmModule and exceptions - pytorch#19028 — Ignored Module tests: provide required input tensor Authored with Claude.

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag). TrainingModule replaces Log.e silent failures with IllegalStateException. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods; new DoNotStrip added to renamed private native methods (generateNative, resetContextNative, loadNative). Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

@after

…t coverage (pytorch#19229) Summary: Combines all previously reverted Android improvement PRs (pytorch#18669, pytorch#19012, pytorch#19028, pytorch#19092, pytorch#19099, pytorch#19124) plus new Module lifecycle tests into a single atomic change. Error reporting: all sync errors throw exceptions instead of returning status codes or logging silently. Module.loadMethod() throws ExecutorchRuntimeException on failure. LlmModule.generate() and load() throw on error. Native methods renamed with "Native" suffix; public wrappers check status and throw. Lifecycle: Module, LlmModule, and TrainingModule implement Closeable for try-with-resources. LlmModule adds ReentrantLock to serialize access to non-thread-safe native state. stop() remains lock-free (C++ atomic flag) with volatile mDestroyed guard to prevent use-after-close. TrainingModule replaces Log.e silent failures with IllegalStateException. Thread safety: LlmModule adds checkNotReentrant() guard to all lock-acquiring public methods, preventing callbacks from re-entering module methods and corrupting native state. TrainingModule.mDestroyed made volatile for cross-thread visibility. Error consistency: ExecuTorchRuntime.validateFilePath throws IllegalArgumentException. SGD throws IllegalStateException. AsrModule throws ExecutorchRuntimeException. Cause-chaining constructor added to ExecutorchRuntimeException. JNI safety: Module and LlmModule constructors wrapped in try-catch to surface native initialization failures. LlmModule.load() uses throwExecutorchException with diagnostic detail. All DoNotStrip annotations preserved on JNI-called methods. std::move added for etdump_gen unique_ptr in profiling build path. LlmModelRunner generate path wrapped in try-catch to prevent HandlerThread death. Test Plan: fix 4 Ignored Module tests by providing required input tensor. Add 13 new lifecycle/API coverage tests. Add LlmModule use-after-close and idempotent close tests. Add @after tearDown to LlmModuleInstrumentationTest to prevent native resource leaks. Remove dead testMethodMetadata test. Fix testNonPteFile cleanup with assertThrows + finally. This diff updates callers of Module.loadMethod() to use try/catch instead of checking the return code, calling destroy() on failure. Co-authored-by: Claude <noreply@anthropic.com> Reviewed By: kirklandsign Differential Revision: D103233465 Pulled By: psiddh

psiddh requested a review from kirklandsign as a code owner April 21, 2026 16:44

CopilotAI review requested due to automatic review settings April 21, 2026 16:44

meta-claBot 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 Apr 21, 2026

psiddh changed the title ~~Fix 4 @Ignored Module tests: provide required input tensor~~Ignored Module tests: provide required input tensorApr 21, 2026

Copilotstarted reviewing on behalf of psiddh April 21, 2026 16:45 View session

psiddh mentioned this pull request Apr 21, 2026
Android ET APIs : Path to Stable (From Experimental) #18950
Open

CopilotAI reviewed Apr 21, 2026
View reviewed changes

Apply suggestions from code review
8fe6bec
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

CopilotAI review requested due to automatic review settings April 21, 2026 21:15

Copilotstarted reviewing on behalf of psiddh April 21, 2026 21:15 View session

CopilotAI reviewed Apr 21, 2026
View reviewed changes

Add module.destroy() to all tests to prevent native resource leaks
b3b7564
This commit was authored with the help of Claude.

kirklandsign approved these changes Apr 23, 2026
View reviewed changes

psiddh merged commit 3ec63f4 into pytorch:mainApr 23, 2026
163 of 169 checks passed

JacobSzwejbka mentioned this pull request Apr 25, 2026
Revert Android PRs #19099, #19124, #19092, #19028 #19133
Merged

psiddh mentioned this pull request Apr 30, 2026
Android: Closeable lifecycle, error reporting, thread safety, and test coverage (#19229) #19229
Merged