[sentry-junior]

📜 Description

Adds an io.sentry.robolectric-test-split build-logic plugin and applies it to sentry-android-core and sentry-android-replay. It splits each module's unit tests into two tasks:

• testReleaseUnitTest (the AGP task) runs only the Robolectric tests. It gets maxHeapSize = "2g" and maxParallelForks = availableProcessors / 2 (at least 1). The plugin now sets core's old unitTests.all { maxHeapSize = "2g" }.
• testReleaseJvmUnitTest (new) runs the other tests with Gradle's defaults: one fork and a 512m heap. It uses the same test classes and classpath as the AGP task, and test/check depend on it.

A Robolectric test is a class whose own bytecode has @RunWith(AndroidJUnit4 | RobolectricTestRunner | ParameterizedRobolectricTestRunner). Both modules use only @RunWith(AndroidJUnit4::class) and put it directly on the test class. The bytecode is read with ASM when the task runs, so the split is config-cache safe and needs no per-file list.

When you pass --tests to testReleaseUnitTest, the split is skipped, so running a plain JVM test from Android Studio still works. Gradle has no public getter for the --tests patterns, so this reads DefaultTestFilter.commandLineIncludePatterns, which is an internal API.

💡 Motivation and Context

Robolectric tests take most of the unit-test time, and the build scans show that each module's unit tests run in a single executor (maxParallelForks was never set). This tries giving only the Robolectric tests more forks and heap, while the plain JVM tests run separately with default settings. Discussed in #team-sdk-java-android.

Caching note: both tasks share the same classpath (all test classes), so changing any test class still reruns both. This PR doesn't make caching more granular; that would take separate source sets.

💚 How did you test it?

Locally (JDK 21, 4 CPUs):

• :sentry-android-replay:testReleaseJvmUnitTest + :sentry-android-replay:testReleaseUnitTest: 79 tests in 4 classes + 210 tests in 18 classes, no failures. There's no overlap, and the split matches the @RunWith(AndroidJUnit4::class) source files exactly. Two Robolectric executors ran with -Xmx2g and the JVM executor with -Xmx512m. A second run reused the configuration cache.
• :sentry-android-core:testReleaseJvmUnitTest: 364 tests in 41 classes, no failures. The only non-Robolectric file with @Test it doesn't run is the abstract ApplicationExitIntegrationTestBase, whose tests run through its Robolectric subclasses.
• :sentry-android-replay:testReleaseUnitTest --tests DefaultReplayBreadcrumbConverterTest (a JVM test) runs it.
• I did not run core's Robolectric task locally; CI covers it.

Timing in my sandbox doesn't tell us much. Compare the CI build scan for this PR with main.

📝 Checklist

• I added GH Issue ID & Linear ID
• I added tests to verify the changes.
• No new PII added or SDK only sends newly added PII if sendDefaultPII is enabled.
• I updated the docs if needed.
• I updated the wizard if needed.
• Review from the native team if needed.
• No breaking change or entry added to the changelog.
• No breaking change for hybrid SDKs or communicated to hybrid SDKs.
• Public API changes reviewed by another Mobile SDK team member or implemented according to the develop docs spec.

🔮 Next steps

• Compare the sentry-android-core / sentry-android-replay test task durations and CI memory against main, and tune maxParallelForks (CI runners have 4 CPUs, so this gives 2 forks).
• If it helps, apply the plugin to the other Robolectric modules (navigation, navigation3, distribution).
• Open question: is the internal DefaultTestFilter use for the --tests passthrough acceptable, or would we rather drop IDE support for JVM tests through testReleaseUnitTest?

#skip-changelog

via roman.

--

View Session [Sentry]

[sentry]

📲 Install Builds

Android

🔗 App Name	App ID	Version	Configuration
SDK Size	io.sentry.tests.size	8.62.0 (1)	release

⚙️ sentry-android Build Distribution Settings