Revision control

Copy as Markdown

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at https://mozilla.org/MPL/2.0/. */
// This comment contains the central documentation for how we configured Jetpack Benchmark. Currently:
// - microbenchmark: configured differently than recommended (see inline notes below)
// - macrobenchmark: not configured
//
// To run our benchmarks, you need to set the "benchmark" gradle property. You can:
// - (preferred) Run via the command line (change the class you run on):
// ./gradlew -Pbenchmark app:connectedCheck -Pandroid.testInstrumentationRunnerArguments.class=org.mozilla.fenix.perf.SampleBenchmark
// - Use the IDE. Temporarily set the "benchmark" property in app/build.gradle with "ext.benchmark=true"
// near the top of the file. DO NOT COMMIT THIS.
// - (note: I was unable to get IDE run configurations working)
//
// To get the results, look at this file (we recommend using the median; results are in nanoseconds):
// app/build/outputs/connected_android_test_additional_output/nightlyAndroidTest/connected/<device>/org.mozilla.fenix-benchmarkData.json
//
// I was unable to get the results to print directly in Android Studio (perhaps it's my device).
//
// The official documentation suggests configuring microbenchmark in a separate module. This would
// require any benchmarked code to be in a library module, not the :app module (see below). To avoid
// this requirement, we created the "benchmark" gradle property.
//
// For the most accurate results, the documentation recommends running tests on rooted devices with
// the CPU clock locked.
//
// jetpack microbenchmark is a good fit.
// I think `android` represents this object:
ext.maybeConfigForJetpackBenchmark = { android ->
if (!project.hasProperty("benchmark")) {
return
}
// recommends setting up the Microbenchmark library in a separate module from your app: AFAICT,
// the reason for this is to prevent the benchmarks from being configured against debug
// builds. We chose not to do this because it's a lot of work to pull code out into a
// separate module just to benchmark it. We were able to replicate the outcome by setting
// this testBuildType property.
android.testBuildType "nightly"
// WARNING: our proguard configuration for androidTest is not set up correctly so the tests
// fail if we don't disable minification. DISABLING MINIFICATION PRODUCES BENCHMARKS THAT ARE
// LESS REPRESENTATIVE TO THE USER EXPERIENCE, however, so we made this tradeoff to reduce
// implementation time.
project.ext.disableOptimization = true
android.defaultConfig {
// WARNING: the benchmark framework warns you if you're running the test in a configuration
// that will compromise the accuracy of the results. Unfortunately, I couldn't get everything
// working so I had to suppress some things.
testInstrumentationRunnerArguments = [
// - ACTIVITY-MISSING: we're supposed to use the test instrumentation runner,
// "androidx.benchmark.junit4.AndroidBenchmarkRunner". However, when I do so, I get an error
// that we're unable to launch the activity. My understanding is that this runner will use an
// "IsolationActivity" to reduce the impact of other work on the device from affecting the benchmark
// and to opt into a lower-max CPU frequency on unrooted devices that support it
// - UNLOCKED: ./gradlew lockClocks, which locks the CPU frequency, fails on my device. See
// https://issuetracker.google.com/issues/176836267 for potential workarounds.
'androidx.benchmark.suppressErrors' : 'ACTIVITY-MISSING,UNLOCKED',
// The tests don't always output a JSON file with the data. To make sure it does, we have to
// set androidx.benchmark.output.enable to true.
'androidx.benchmark.output.enable' : 'true',
// We set the the output directory simply for simplicity since the benchmark_runner.py script
// can't know the name of the phone in the /build/outputs/ directory. The system defaults to
// {phone_name} which can be troublesome finding in some case.
//
// NOTE: Jetpack Benchmark outputs to Logcat too. However, the output in the logcat is
// the min of the several repeats, for more statistics. Therefore, to get more stats,
// we refer to the JSON file.
additionalTestOutputDir : '/storage/emulated/0/benchmark'
]
}
}