Android: unified error reporting — all sync errors throw, no silent f… by psiddh · Pull Request #18669 · pytorch/executorch

psiddh · 2026-04-02T07:52:25Z

…ailures

Replace the mixed error-reporting pattern (int return codes, empty arrays, silent no-ops) with a consistent exception-based contract across Module, LlmModule, and LlmCallback:

Module: all public methods acquire mLock and check destroyed state; loadMethod() throws ExecutorchRuntimeException on native error; execute()/forward() throw IllegalStateException on destroyed module; destroy() throws if lock unavailable (concurrent execution)
LlmModule: all generate/load/prefill methods check destroyed state via volatile flag and throw on native errors; resetNative() deprecated in favor of future close(); stop() intentionally unguarded for interrupt-during-generate; prefill methods return void instead of long
LlmCallback: onError() default logs via android.util.Log with try/catch fallback for JVM unit test environments
ExecutorchRuntimeException: added ALREADY_LOADED (0x04) error code, javadoc on all 19 error codes, "ExecuTorch" casing in error messages
JNI: renamed registrations to match Java native method names (generateNative, loadNative, resetContextNative); removed double exception throw from C++ load(); unknown input typeCode now throws
Tests: updated for void returns and assertThrows; all @ignore preserved
Benchmark: ModelRunner and LlmModelRunner adapted to try/catch pattern
All APIs are still experimental.

pytorch-bot · 2026-04-02T07:52:30Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18669

📄 Preview Python docs built from this PR

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

❌ 4 New Failures, 2 Unrelated Failures

As of commit 795e56c with merge base 60d57e5 ():

NEW FAILURES - The following jobs have failed:

Lint / link-check / lint-urls (gh)
pull / android / build-android (gh)
Process completed with exit code 1.
pull / test-openvino-linux / linux-job (gh)
RuntimeError: Command docker exec -t a2b7f90d645e86fb969532b3de0a60eb4dec2a83d4f10e8dd4bd865c9343b80b /exec failed with exit code 1
pull / unittest-arm-backend-with-no-deps (test_pytest_ops_tosa) / linux-job (gh)
RuntimeError: Command docker exec -t 3e44221772e67f9d39b24dd953aed584979958d7a05ea0d0c4117125cd8ba7b0 /exec failed with exit code 1

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest / windows / windows-job (gh) (trunk failure)
##[error]The operation was canceled.
pull / unittest-editable / windows / windows-job (gh) (trunk failure)
##[error]The operation was canceled.

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

github-actions · 2026-04-02T07:58:09Z

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.

@ignore

…ailures Replace the mixed error-reporting pattern (int return codes, empty arrays, silent no-ops) with a consistent exception-based contract across Module, LlmModule, and LlmCallback: - Module: all public methods acquire mLock and check destroyed state; loadMethod() throws ExecutorchRuntimeException on native error; execute()/forward() throw IllegalStateException on destroyed module; destroy() throws if lock unavailable (concurrent execution) - LlmModule: all generate/load/prefill methods check destroyed state via volatile flag and throw on native errors; resetNative() deprecated in favor of future close(); stop() intentionally unguarded for interrupt-during-generate; prefill methods return void instead of long - LlmCallback: onError() default logs via android.util.Log with try/catch fallback for JVM unit test environments - ExecutorchRuntimeException: added ALREADY_LOADED (0x04) error code, javadoc on all 19 error codes, "ExecuTorch" casing in error messages - JNI: renamed registrations to match Java native method names (generateNative, loadNative, resetContextNative); removed double exception throw from C++ load(); unknown input typeCode now throws - Tests: updated for void returns and assertThrows; all @ignore preserved - Benchmark: ModelRunner and LlmModelRunner adapted to try/catch pattern Breaking change under @experimental — all APIs are still experimental. Co-authored-by: Claude <noreply@anthropic.com>

Copilot

Pull request overview

This PR standardizes Android-side error reporting by shifting ExecuTorch Java APIs from mixed return-code patterns to an exception-based contract, and aligning JNI registrations with the updated native method names.

Changes:

Updated Module and LlmModule public APIs to throw on errors/destroyed-state instead of returning status codes or empty results.
Renamed/adjusted JNI native method registrations to match updated Java native method names and added stricter input-type validation.
Adapted Android benchmark and instrumentation tests to the new void/exception-based APIs.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
extension/benchmark/android/benchmark/app/src/main/java/org/pytorch/minibench/ModelRunner.java	Adapts benchmarking to exception-based `loadMethod()` while preserving a numeric status metric.
extension/benchmark/android/benchmark/app/src/main/java/org/pytorch/minibench/LlmModelRunner.java	Converts `load()` to try/catch error-code extraction and logs generation failures.
extension/android/jni/jni_layer.cpp	Throws on unsupported input `EValue` type codes; renames registered natives for `etdump`/`getMethods`.
extension/android/jni/jni_layer_llama.cpp	Simplifies `load()` to return error codes only; renames registered natives to `*Native` variants.
extension/android/executorch_android/src/main/java/org/pytorch/executorch/Module.java	Enforces locking + destroyed checks across public methods; converts `loadMethod()` to `void` that throws on native errors; wraps `getMethods`/`etdump`.
extension/android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmModule.java	Adds destroyed-state guard; converts `generate/load/prefill` methods to throw exceptions and return `void`.
extension/android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmCallback.java	Adds a default `onError()` implementation that logs to Logcat with a fallback path.
extension/android/executorch_android/src/main/java/org/pytorch/executorch/ExecutorchRuntimeException.java	Adds `ALREADY_LOADED`, expands error-code Javadoc, and standardizes “ExecuTorch” casing in formatted messages.
extension/android/executorch_android/src/androidTest/java/org/pytorch/executorch/ModuleInstrumentationTest.kt	Updates tests for exception-based `loadMethod()` and destroyed-module behavior.
extension/android/executorch_android/src/androidTest/java/org/pytorch/executorch/LlmModuleInstrumentationTest.kt	Updates LLM instrumentation test to the new `void load()` API.

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

...droid/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmCallback.java

extension/android/jni/jni_layer_llama.cpp

.../executorch_android/src/androidTest/java/org/pytorch/executorch/ModuleInstrumentationTest.kt

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

…pytorch/executorch/ModuleInstrumentationTest.kt Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…/executorch/extension/llm/LlmCallback.java Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 10 out of 10 changed files in this pull request and generated 5 comments.

Comments suppressed due to low confidence (3)

extension/android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmModule.java:276

These native methods are registered by name via fbjni. Without @DoNotStrip (or an explicit consumer ProGuard/R8 keep rule), their names may be obfuscated, causing JNI registration to fail at runtime. Please annotate generateNative() (and other registered native methods) with @DoNotStrip or add appropriate keep rules for native members.

      int numEos) {
    checkNotDestroyed();
    int err = generateNative(prompt, seqLen, llmCallback, echo, temperature, numBos, numEos);
    if (err != 0) {
      throw ExecutorchRuntimeException.makeExecutorchException(err, "Failed to generate");
    }
  }

  private native int generateNative(
      String prompt,
      int seqLen,
      LlmCallback llmCallback,
      boolean echo,

extension/android/executorch_android/src/androidTest/java/org/pytorch/executorch/ModuleInstrumentationTest.kt:117

After Module.destroy(), loadMethod() throws IllegalStateException. This assertion uses the broad RuntimeException type, which could hide regressions (e.g., wrong exception type). Prefer asserting IllegalStateException specifically here.

    module.loadMethod(FORWARD_METHOD)
  }

  @Test
  @Throws(IOException::class)
  fun testLoadOnDestroyedModule() {
    val module = Module.load(getTestFilePath(TEST_FILE_NAME))

extension/benchmark/android/benchmark/app/src/main/java/org/pytorch/minibench/ModelRunner.java:43

If loadMethod("forward") throws, this code records an errorCode but then still runs warmup/benchmark forward() calls and etdump(). With the new exception-based contract, those calls can also throw and abort the benchmark, preventing metrics from being recorded. Consider returning early (or wrapping forward()/etdump() in try/catch and reporting an appropriate load_status) when preloading fails.

    Module module = Module.load(model.getPath());
    int errorCode = 0;
    try {
      module.loadMethod("forward");
    } catch (Exception e) {
      errorCode =
          (e instanceof org.pytorch.executorch.ExecutorchRuntimeException)
              ? ((org.pytorch.executorch.ExecutorchRuntimeException) e).getErrorCode()
              : -1;
    }
    long loadEnd = System.nanoTime();

    for (int i = 0; i < numWarmupIter; i++) {
      module.forward();
    }

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

Copilot · 2026-04-02T13:33:43Z

...droid/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmCallback.java

+  default void onError(int errorCode, String message) {
+    try {
+      android.util.Log.e("ExecuTorch", "LLM error " + errorCode + ": " + message);
+    } catch (Throwable t) {
+      System.err.println("ExecuTorch LLM error " + errorCode + ": " + message);
+    }


LlmCallback.onError() intends to fall back in JVM unit test environments, but catching RuntimeException won’t catch NoClassDefFoundError/LinkageError thrown when android.util.Log isn’t on the classpath. Catch Throwable (or at least Exception + LinkageError) so the System.err fallback reliably runs off-Android.

Copilot · 2026-04-02T13:33:43Z

...android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmModule.java

+    checkNotDestroyed();
+    resetContextNative();
+  }
+


resetContextNative() is registered from JNI by method name; without @DoNotStrip or keep rules, R8 can rename it and break native registration. Add @DoNotStrip (and consider applying consistently to all JNI-registered native members in this class).

Suggested change

@DoNotStrip

Copilot · 2026-04-02T13:33:43Z

...android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmModule.java

+      throw ExecutorchRuntimeException.makeExecutorchException(err, "Failed to load model");
+    }
+  }
+


loadNative() is registered from JNI by method name; without @DoNotStrip or keep rules, R8 can rename it and break native registration. Add @DoNotStrip (and consider applying consistently to all JNI-registered native members in this class).

Suggested change

@DoNotStrip

Copilot · 2026-04-02T13:33:44Z

.../executorch_android/src/androidTest/java/org/pytorch/executorch/ModuleInstrumentationTest.kt

  @Test
  @Throws(IOException::class)
  fun testModuleLoadMethodNonExistantMethod() {
    val module = Module.load(getTestFilePath(TEST_FILE_NAME))

-    val loadMethod = module.loadMethod(NONE_METHOD)
-    Assert.assertEquals(loadMethod.toLong(), INVALID_ARGUMENT.toLong())
+    val exception =
+        Assert.assertThrows(ExecutorchRuntimeException::class.java) {


This test currently asserts only that some RuntimeException is thrown. Since Module.loadMethod() now throws a specific ExecutorchRuntimeException on native load failures, assert that type (and ideally the expected errorCode) to ensure the exception contract stays correct.

Copilot · 2026-04-02T13:33:44Z

.../executorch_android/src/androidTest/java/org/pytorch/executorch/ModuleInstrumentationTest.kt

+    Assert.assertThrows(IllegalStateException::class.java) { module.loadMethod(FORWARD_METHOD) }
  }

  @Test
  @Throws(IOException::class)
  fun testForwardOnDestroyedModule() {
    val module = Module.load(getTestFilePath(TEST_FILE_NAME))

-    val loadMethod = module.loadMethod(FORWARD_METHOD)
-    Assert.assertEquals(loadMethod.toLong(), OK.toLong())
+    module.loadMethod(FORWARD_METHOD)


After destroy(), forward()/execute() now throw IllegalStateException. This assertion uses the broad RuntimeException type, which could hide regressions. Prefer asserting IllegalStateException specifically (and optionally validate the message).

meta-cla bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Apr 2, 2026

psiddh force-pushed the unified_reporting branch from 2b96224 to 65e793b Compare April 2, 2026 08:10

psiddh marked this pull request as ready for review April 2, 2026 08:11

psiddh requested review from Gasoonjia and kirklandsign as code owners April 2, 2026 08:11

psiddh requested review from GregoryComer and Copilot April 2, 2026 08:11

Copilot started reviewing on behalf of psiddh April 2, 2026 08:12 View session

Copilot AI reviewed Apr 2, 2026

View reviewed changes

Update extension/android/jni/jni_layer_llama.cpp

4338e13

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

Copilot AI review requested due to automatic review settings April 2, 2026 13:28

psiddh and others added 2 commits April 2, 2026 06:28

Update extension/android/executorch_android/src/androidTest/java/org/…

e734382

…pytorch/executorch/ModuleInstrumentationTest.kt Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Update extension/android/executorch_android/src/androidTest/java/org/…

28241aa

…pytorch/executorch/ModuleInstrumentationTest.kt Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot started reviewing on behalf of psiddh April 2, 2026 13:28 View session

Update extension/android/executorch_android/src/main/java/org/pytorch…

795e56c

…/executorch/extension/llm/LlmCallback.java Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot AI reviewed Apr 2, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Android: unified error reporting — all sync errors throw, no silent f…#18669

Android: unified error reporting — all sync errors throw, no silent f…#18669
psiddh wants to merge 5 commits intopytorch:mainfrom
psiddh:unified_reporting

psiddh commented Apr 2, 2026

Uh oh!

pytorch-bot bot commented Apr 2, 2026 •

edited

Loading

Uh oh!

github-actions bot commented Apr 2, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 2, 2026

Uh oh!

Copilot AI Apr 2, 2026

Uh oh!

Copilot AI Apr 2, 2026

Uh oh!

Copilot AI Apr 2, 2026

Uh oh!

Copilot AI Apr 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

psiddh commented Apr 2, 2026

Uh oh!

pytorch-bot bot commented Apr 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18669

❌ 4 New Failures, 2 Unrelated Failures

Uh oh!

github-actions bot commented Apr 2, 2026

This PR needs a release notes: label

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

pytorch-bot bot commented Apr 2, 2026 •

edited

Loading

This PR needs a `release notes:` label