Jetpack Compose is designed to be interoperable with an present View-based utility. This lets you take an incremental method to migrating your present app’s consumer interface to Compose.

Whereas your utility is in the midst of migration, and hasn’t but Composable on the house or touchdown display, customers might expertise crashes when navigating to a display created with Compose. It’s because compose library is loaded solely when first used. There’s a workaround that has been floating round so as to add a composable pixel on the house/touchdown display. This answer ensures that Compose elements are warmed up earlier than use. We don’t suggest taking this method, as making use of it in your utility leads to code that doesn’t contribute to the implementation of your perform and might result in unexpected errors. Quite, we suggest a mixture of the appliance startup library and customized reference profiles to resolve the crash problem. We are going to examine and visualize how the mixture will help enhance utility efficiency.

Baseline profiles assist enhance utility launch and runtime efficiency of Android purposes. Fundamental profiles are a listing of lessons and strategies included in an APK. They’re utilized by the Android Runtime (ART) throughout utility set up to precompile crucial paths into machine code. This can be a type of Profile Guided Optimization (PGO) that permits purposes to optimize startup, scale back crashes, and enhance efficiency for finish customers by lowering the quantity of code that should be compiled Simply-In-Time (JIT).

Jetpack Compose is distributed as a library. This permits for frequent updates to Compose and compatibility with older variations of Android. However distributing as a disaggregated library comes at a price; libraries should be loaded when the app begins and interpreted simply in time when the performance is required (for extra particulars, see Why it’s best to all the time efficiency take a look at Compose at launch). This could result in an extended app startup time or crash each time the app makes use of a library perform for the primary time. To resolve this problem, Compose gives a reference profile to construct Compose forward of time when the app is put in. Normally, the default reference profile supplied by Compose is ample to offer nice efficiency. However customizing the profile primarily based on widespread consumer interactions together with your app typically yields higher outcomes.

To research this efficiency enchancment, we wrote some Macrobenchmark assessments and measured efficiency on the pattern Sunflower utility. The app has been partially migrated to Compose and doesn’t have Composable on the house or touchdown display.

For this efficiency evaluation, we shall be interacting with three screens of the appliance:

• the House display which is constructed with a View-based format and has no Composables.
• the plant checklist The display is constructed with RecyclerView and every factor inside the view is constructed with Compose. It makes use of the ComposeView to host the Compose content material.
• the plant particulars A display that’s created with Compose and makes use of AndroidViewBinding to create and work together with Android format sources.

The Sunflower utility code has a BaselineProfileGenerator class that creates a baseline profile. We ran the take a look at and created a customized reference profile targeted on these Crucial Person Journeys (CUJs). This profile is out there inside the baseline-prof.txt file below the main font set. In comparison with the default baseline profile supplied by Compose, the customized baseline profile exhibits efficiency enhancements. Let’s check out the take a look at outcomes for a few situations:

• [Scenario 1]From the house display, the consumer navigates to the Plant Checklist display that’s partially built-in into Compose. Benchmarking is out there within the PlantListBenchmarks class.

Macrobenchmark runs your take a look at a number of instances and outputs the outcomes as a statistical distribution. The distribution is represented in percentiles. P50 — P99. The next have been the outcomes once we ran the benchmark assessments:

For the info proven within the desk above, P50 for CompilationMode.None implies that 50% of the frames have been rendered sooner than 7.1 milliseconds.

• [Scenario 2] The consumer navigates to a display fully constructed with Compose and makes use of the AndroidViewBinding to work together with the TextView. Benchmarking is out there within the PlantDetailBenchmarks class. The next have been the outcomes once we ran the benchmark assessments:

Be aware: Benchmark assessments have been run on Samsung Galaxy Fold, your benchmark outcomes might present completely different numbers, however the general efficiency affect needs to be comparable.

once you look intently P99 numbers, you’ll discover vital enhancements. These outliers are sometimes what trigger crashes.

To additional enhance utility efficiency, we are able to use the appliance startup library to heat up Compose elements earlier than they’re truly used. Let’s strive it out and see what we discover out.

The appliance startup library gives a structured option to initialize elements at utility startup. With the app startup library, you possibly can heat up the Compose library throughout app startup.

Observe these steps so as to add the Compose initializer utilizing the appliance startup library:

• To make use of Jetpack startup in your utility, add the dependency on the startup runtime in your utility’s configuration. construct.gradle proceedings.

To heat up the Compose part, create a brand new ComposeInitializer class as proven within the following code snippet.

Initializing the ComposeView on this approach is not going to add the content material to the view hierarchy, but it surely nonetheless helps to initialize the Compose part.

• To make ComposeInitializer discoverable by App Startup, go to AndroidManifest.xml file in your mission. add a <meta-data> low entry InitializationProvider manifest entry as proven within the following code:

InitializationProvider is a particular content material supplier that helps uncover and name the part initializer you simply outlined.

• Rebuild and replace the reference profile to account for the adjustments at startup. The benchmark profiler is a part of the macrobenchmark module. You simply must run the startPlantListPlantDetail take a look at BaselineProfileGenerator class.
• Rerun the macrobenchmark assessments from the PlantListBenchmarks Y PlantDetailBenchmarks class.

Here is what we found for every of the above situations:

• [Scenario 1]From the house display, the consumer navigates to the plant checklist display.

If you look intently, for the display that’s partially constructed with Compose, the outliers appear to have an enhancement of ~21%. Typically it’s the outlier frames that trigger the crash.

• [Scenario 2] The consumer navigates to the Plant Particulars display.

There are ~27% proportion enchancment in outliers.

In each circumstances, including the appliance startup initializer improved efficiency, particularly for outliers in P99.

Right here is the hyperlink to entry the code on GitHub.

Now the query is, do you have to all the time add the appliance startup library for such situations? Is it relevant to all of your initiatives? Properly, earlier than making such a choice, we strongly suggest that you simply:

1. Write benchmark assessments and measure utility efficiency.
2. implement the adjustments
3. Rerun your assessments to see if efficiency has truly improved.

This implementation method would allow you to in the long term to evaluate whether or not the solutions are actually relevant to your utility. Let statistical information drive your deployment choices to enhance efficiency.

• At all times add a baseline profile, no matter whether or not you want the appliance launcher library.
• The mix of the customized baseline profile and the app launcher library yields higher efficiency outcomes when your app has no composable components on the house or touchdown display, and the consumer navigates to a display that’s partially or absolutely constructed with Compose.
• As a substitute of including Composable to the splash/touchdown display for efficiency causes (and never design alternative), you need to use the appliance launcher library to heat up Compose elements.
• Simply because somebody suggests it does not imply you must observe that recommendation.. Write Macrobenchmark assessments to confirm if the appliance launcher library truly improves the efficiency of your utility. Let statistical information drive your deployment decisions to make sure they lead to improved efficiency.

Thanks Ben Trengrove, Florina Mutenescu, Ben Weiss, Ryan Mentley, and Rahul Ravikumar for serving to with revisions.