Abdulahi and I gave another talk this year at Droidcon NYC. This one talks about how to scale Android development without a monorepo. The talk discusses various topics, including:
Recently, I spent some time trying to replicate a gesture I liked on iOS1. I had a simple button where, if you long press it, a menu shows up from which an item can be selected. On the iOS app, you could select an item from this menu in a single gesture. By long pressing the button and not letting go, you could drag your finger to the item and select a menu item in one gesture.
In this video, you can see this behavior in action:
Let’s start with a simple implementation of the button and the menu.
@Composable
fun DropDownExample(modifier: Modifier) {
val showMenu = remember { mutableStateOf(false) }
Column(
verticalArrangement = Arrangement.Center,
horizontalAlignment = Alignment.CenterHorizontally,
modifier = modifier
) {
Box {
Icon(
Icons.Default.MoreVert,
contentDescription = null,
tint = Color.DarkGray,
modifier = Modifier.combinedClickable(
onLongClick = { showMenu.value = true },
) { println("onClick") }
)
DropdownMenu(
expanded = showMenu.value,
onDismissRequest = { showMenu.value = false },
) {
(1..3).forEach { i ->
DropdownMenuItem(
text = { Text("Item $i") },
onClick = {
println("item $i clicked")
showMenu.value = false
}
)
}
}
}
}
}
Nothing particularly noteworthy here, and with just a few lines of code, we have the first half of the video mostly implemented.
Let’s start by adding a Modifier.pointerInput to the Column wrapping the button and its menu:
Column(
verticalArrangement = Arrangement.Center,
horizontalAlignment = Alignment.CenterHorizontally,
modifier = modifier
.pointerInput(Unit) {
detectDragGesturesAfterLongPress(
onDragStart = { offset -> println("onDragStart: $offset") },
onDrag = { change, offset -> println("onDrag: $change, $offset") },
onDragCancel = { println("onDragCancel") },
onDragEnd = { println("onDragEnd") }
)
}
) {
// ...
}
If we run this, we’ll see an onDragStart, followed by a onDragCancel. This is because we have a long press on the Icon itself that conflicts with this drag gesture. Since this gesture detects long press for us, we’ll move the long press handling here to avoid the conflict2.
We’ll switch Modifier.combinedClickable with just Modifier.clickable, and move the showMenu.value = true into onDragStart. If we do this, we’ll see what the drag events we expect to see after seeing the onDragStart, and we’ll see onDragEnd once the drag is complete.
Column(
verticalArrangement = Arrangement.Center,
horizontalAlignment = Alignment.CenterHorizontally,
modifier = modifier
.pointerInput(Unit) {
detectDragGesturesAfterLongPress(
onDragStart = { offset -> showMenu.value = true },
onDrag = { change, offset -> println("onDrag: $change, $offset") },
onDragCancel = { println("onDragCancel") },
onDragEnd = { println("onDragEnd") }
)
}
) {
Box {
Icon(
Icons.Default.MoreVert,
contentDescription = null,
tint = Color.DarkGray,
modifier = Modifier.clickable { /* some action on click */ }
)
DropdownMenu(/* ... */)
}
}
Ideally, we’d love to have the touch handling “transfer” to the DropdownMenu once we show it. Unfortunately, I couldn’t find a way to make this work, most likely due to the fact that the DropdownMenu is a Popup under the hood3. Consequently, we’ll have to handle the drag gesture calculations ourselves.
To do this, what we need is to know where the gesture is released, along with knowing the bounds of the DropdownMenuItems. With the combination of these, we can determine which item, if any, was selected, and perform the action.
We can add an onGloballyPositioned modifier to each item and to the box surrounding the menu itself. If we do this initially and just print the values, we’ll notice that the values we get for the position of the items are relative to the DropdownMenu itself, irrespective of whether we use boundsInRoot(), or boundsInWindow(). This is because the DropdownMenu is a Popup. In other words, the DropdownMenu’s origin is at (0, 0).
We know, however, that the DropdownMenu takes the position of the anchor it’s attached to. We can use this information to calculate the actual position on the screen to detect whether or not there is a hit. We do this by recording the position of the Box wrapping the anchor and the DropdownMenu.
val menuAnchorBounds = remember { mutableStateOf(Rect.Zero) }
val items = remember { mutableStateOf(arrayOf(Rect.Zero, Rect.Zero, Rect.Zero)) }
// <snip>
Box(
modifier = Modifier.onGloballyPositioned {
menuAnchorBounds.value = it.boundsInParent()
}
) {
Icon(/* ... */)
DropdownMenu(
expanded = showMenu.value,
onDismissRequest = { showMenu.value = false }
) {
(1..3).forEach { i ->
DropdownMenuItem(
text = { Text("Item $i") },
onClick = { println("item $i clicked"); showMenu.value = false },
modifier = Modifier.onGloballyPositioned {
items.value = items.value.apply { set(i - 1, it.boundsInParent()) }
}
)
}
}
}
With this, the remaining piece is to figure out, in onDragEnd, which item the drag ended at (if any). Since onDragEnd does not have any parameters about the ending position, we’ll have to store the position as it updates in onDrag. Afterwards, we can use this information in onDragEnd to check for a hit, in combination with the bounds of the items and the menu stored above.
@Composable
fun DropdownExample(modifier: Modifier) {
// <snip>
val lastPosition = remember { mutableStateOf(Offset.Unspecified) }
Column(
verticalArrangement = Arrangement.Center,
horizontalAlignment = Alignment.CenterHorizontally,
modifier = modifier
.pointerInput(Unit) {
detectDragGesturesAfterLongPress(
onDragStart = { offset -> showMenu.value = true },
onDrag = { change, offset -> lastPosition.value = change.position },
onDragCancel = { println("onDragCancel") },
onDragEnd = {
val upPosition = lastPosition.value
val adjustedTargetLocation = upPosition.minus(menuAnchorBounds.value.bottomLeft)
val match = items.value.indexOfFirst { adjustedTargetLocation in it }
if (match != -1) {
println("Item ${match + 1} clicked")
showMenu.value = false
} else {
println("No button clicked")
}
}
)
}
) {
// <snip>
}
}
We take the last position of the drag, and adjust it by the position of the menu’s anchor, in order to get positions relative to the DropdownMenu. In this case, because the menu opens anchoring to the bottom left of the anchor, we use that as the offset on the last drag position. We can then check for a match amongst our items. This works! 🎉
This works, but we no longer have any hover effect on each item when we drag over them. The DropdownMenuItem composable accepts an interactionSource, based on which it renders the hover effect. We can emit and memoize a PressInteraction.Press(pressPosition) when an item is hit, and emit a PressInteraction.Release(pressInteraction) when the item is no longer hovered over. We can do this within onDrag itself, though we need to explicitly map the position of the drag to the corresponding item, in a similar way to how we detected the item in onDragEnd.
// we need an interaction source per drop down menu item, since if we share them,
// we'd get all items selecting and unselecting together, irrespective of the position.
val interactionSources =
remember { arrayOf(MutableInteractionSource(), MutableInteractionSource(), MutableInteractionSource()) }
val lastInteractionItem = remember { mutableIntStateOf(-1) }
val lastInteractionPress = remember { mutableStateOf(Offset.Unspecified) }
// snip
onDrag = { change, dragAmount ->
lastPosition.value = change.position
val adjustedTargetLocation = change.position.minus(menuAnchorBounds.value.bottomLeft)
val currentItem = items.value.indexOfFirst { adjustedTargetLocation in it }
// if the item we are hovering over has changed...
if (currentItem != lastInteractionItem.intValue) {
// if it wasn't -1, we need to "release" the press on the old one
if (lastInteractionItem.intValue != -1) {
val interactionSource = interactionSources[lastInteractionItem.intValue]
interactionSource.tryEmit(
PressInteraction.Release(PressInteraction.Press(lastInteractionPress.value))
)
}
// if we have a chosen item, we need to "press" on the new one
if (currentItem != -1) {
val interactionSource = interactionSources[currentItem]
interactionSource.tryEmit(PressInteraction.Press(adjustedTargetLocation))
lastInteractionPress.value = adjustedTargetLocation
}
lastInteractionItem.intValue = currentItem
}
}
We’d then update our DropdownMenuItems to use the corresponding interactionSources[index].
If we try this out, we notice everything works great, with one exception: if we long press the anchor and don’t drag at all, the menu shows up, but the “click” handle is also fired on the button, which isn’t what we want. If we drag, however, everything is fine. How do we fix this?
As the documentation tells us, there are three PointerEventPasses. We are interested in the PointerEventPass.Initial pass, which, as the documentation says:
In the Initial pass, the event flows from the top of the UI tree to the bottom. This flow allows a parent to intercept an event before the child can consume it.
This seems to be what we want. Moreover, The code also links an example of this in the implementation of the Tooltip in Material 3. Notice how, in lines 228 to 229 of the implementation, PointerEventPass.Initial pointer events are consumed, preventing them from being actioned by the children. We can see something similar to this in the detectDragGesturesAfterLongPress code as well (though without the pass specification, and only after a drag, which is why this issue only appears if we don’t drag).
The easiest way to fix this is to copy the detectDragGesturesAfterLongPress code, and consume the PointerEventPass.Initial events after calling the onDragStart lambda by adding these two lines after the onDragStart.invoke(drag.position) call:
// consume the children's click handling
val event = awaitPointerEvent(pass = PointerEventPass.Initial)
event.changes.forEach { if (it.changedToUp()) it.consume() }
Early on, I came across this artice about multiselection within a photo grid, which pointed me in the right direction after getting stuck with trying to find a way to “hand off” events to the menu. This was a fun experiment for something that, admittedly, is not likely to be discovered or used much in my app. The journey did help me increase my understanding of how touch is handled in Compose.
Special thanks to Efeturi for reviewing this post.
Efeturi tells me this is a pattern that was there on Android in the native view days as well. ↩︎
The click will also conflict, but we’ll resolve that later by consuming those gestures in PointerEventPass.Initial so that the child does not receive nor act on them. ↩︎
A combination of PointerInputModifierNode combined with sharePointerInputWithSiblings seemed promising, but actual DropdownMenuItems don’t get pointer events, likely due to being a Popup. ↩︎
Compose on the web has always fascinated me. It’s always been really cool to take a working application on Android and have it run with minimal changes on various other platforms, including the web. While I’ve written about it in the past and show cased it in talks, it’s always been a toy of sorts for me.
This changed one day when I wanted to write a tool for visualizing some data at work. I decided to write a desktop app, and, I opted to use Kotlin and Compse, guessing it’d be the fastest way to accomplish what I want. After making it work, I started thinking of how I could share this tool with my colleagues. Sharing jars or instructions for building wasn’t as interesting, so what about just sharing a url? This made the utility of Compose for web much more apparent to me.
In this post, I want to share a few odds and ends and tidbits that I learned while working on various Compose for Web projects.
Generally speaking, Compose code written for Android will work on the web and in WebAssembly with minimal changes. There are still some differences, and often times, code that will compile on Android and JS (for example) won’t compile cleanly on WebAssembly. One work around for this is to use expect/actual to provide empty or no-op implementations for the offending platform and keep the real implementations where it compiles.
The development cycle is very fast, due to auto-reload whenever a change is made. Make sure to pass the -t flag to jsRun or wasmJsRun to ensure auto rebuilds and reloads of the page whenever changes are made. Without it, changes to the code seem to reload the page, but without reflecting any of the changes.
In a KMP project with web support, building the jsBrowserDistribution task (or wasmJsDistributionTask for WebAssembly) will output
a folder named productionExecutable under the build/dist/js or build/dist/wasm folder. These tasks will run optimizations (dead code removal, minification, etc). You can simply take the contents of that directory and deploy it to a server somewhere and it should just work. This makes for some compelling demos. As I learned from one of those repos, it’s quite easy to wire this up to GitHub Pages on every merge.
One trick I use to validate this locally is going to the productionExecutable directory, and running a simple HTTP server with Python:
python3 -m http.server
This allows me to test the production version locally, since using file:// urls doesn’t work due to security restrictions in the browser.
When declaring support for a web project, we add the type of web projects supported - either browser or nodejs. Code targeting browser runs in the browser, which means it runs on the client’s machine, whereas code targeting nodejs can run on the server.
This all seems obvious, so why mention it? Unfortunately, sometimes, developers forget this point. I saw an open source project put in some clever amount of work to hide an api key, wiring it through GitHub secrets to an environment variable in GitHub Actions, and then injecting it into the code at build time using BuildKonfig. The problem is that the code runs in the browser, so the api key is exposed to anyone who cared to look at the networking logs in the developer console.
As a workaround, the project could have a server side component that does the actual logic of talking to the third party API, and the web code can talk to this component. This keeps the API key from being exposed to the client, and allows the developer to control access to the API (add rate limiting, caching, etc).
Thanks to coil3, showing images from a URL is as easy as a few lines of code in Compose:
@Composable
fun WebImage(url: String, contentDescription: String, modifier: Modifier) {
AsyncImage(
model = ImageRequest.Builder(LocalPlatformContext.current)
.data(url)
.build(),
contentDescription = contentDescription,
modifier = modifier
)
}
This works well and looks great, but when you try it on web, you often find yourself running into images that don’t load. Looking at the logs, you see things like:
Access to fetch at '<url>' from origin 'http://localhost:8080' has been blocked
by CORS policy: No 'Access-Control-Allow-Origin' header is present on the requested resource.
Today, browsers have security checks that disallow loading resources from other domains, unless those domains explicitly allow it. This type of check helps prevent CSRF and various other attacks. This effectively means that the same URL an be read using curl or OkHttpClient, but not from the web due to these web browser security checks.
So what can you do?
Access-Control-Allow-Origin header to the response. This tells the browser that it’s okay to load the resource from this particular domain.I wrote before about using Compose/JS without WebAssembly. That having been said, the code for both is often very easy to share, since it’s almost identical. One of the exciting promises of WebAssembly today is the massive performance benefits in comparison to vanilla JavaScript. While today, wasm is supported out of the box (without special configurations) on Chrome and Firefox, it still doesn’t work on Safari just yet.
One thing I learned is that even Compose/JS uses WebAssembly under the hood for Skia drawing via Skiko. As of this writing, the size of the WebAssembly artifacts for this repository is around 28mb, whereas the size of the non-WebAssembly version is 13mb (uncompressed). Close to 8.5mb of these come from Skiko in both versions (with matching md5 sums of these files).
As of this writing, Ktor has a 3.0.0-wasm1 version that supports WebAssembly in its EAP repository. This version worked for me for both vanilla JavaScript and WebAssembly in debug, though release versions of JavaScript code failed to run. I worked around this for now by only using the 3.0.0-wasm1 version for wasm targets, and using the stable 2.3.8 version for JavaScript targets.
My coworker Abdulahi Osoble and I recently gave a talk at Droidcon NYC about how to use a design system to make building Server Driven UI easier. We go through the motivations and building blocks, along with architectural decisions that need to be made along the way.
The video is here, the slide deck is here, and the accompanying open source repository is here. The repository is my favorite part, since it showcases a fictional micro-design system running on multiple platforms, and showcases Server Driven UI working on all of these platforms with the same source json.
Not too long ago, I came across Kaushik’s blog post and repo about Kotlin scripting. Around the same time, I was trying to automate a process of cleaning up some images I had. Given a set of transparent pngs, I wanted to find out if the last line was a solid line, and, if so, and if nothing is touching that line, remove it. While typically, I turn to using Pillow and Python for image manipulation scripts, I decided to give writing a Kotlin script a try instead.
I wrote the script (using javax.imageio) and tested it and everything was great. I then proceeded to combine my Kotlin script with the shell to run it on the set of images like this:
# look at directories 351, 352, ... 360
for i in `seq 350 360`; do
DIR="/path/$i"
cd $DIR
# we have 15 images in each directory, sub_image_0.png, sub_image_1.png, ...
for j in `seq 0 14`; do
IMAGE=sub_image_$j.png
kotlin /Users/ahmedre/Documents/code/kotlin/line_finder/finder.main.kts $IMAGE
done
cd -
done
The script loops on a set of directories, each with a number, and on a set of 15 images inside that directory. For each image, I run the Kotlin script, passing it as a parameter. The surprise to me came when I found that running this took ~2.5 minutes! 150 images isn’t a massive amount, and these aren’t very large images, so why does it take so long?
At this point, I remembered having come across KScript at some point. It provides a wrapper around kotlinc, caching script compilation among other things. I replaced /usr/bin/env kotlin with /usr/bin/env kscript in the script, and replaced kotlin with kscript. This brought the time down to ~1 minute and 12 seconds.
I continued reading, after which I found that KScript has a package option used to deploy scripts as standalone binaries. I ran kscript --package finder.main.kts and took the finder.main binary and replaced it in the for loop above. This brought the time down to ~29 seconds.
Maybe it’s slow due to Kotlin Scripting, I thought. What if I make a compiled jar instead? I modified the script (adding a main method, etc), and used kotlinc to build a jar, using kotlinc finder.kt -include-runtime -d finder.jar to generate a new jar, and used java -jar finder.jar in place of the existing Kotlin command in the loop. Running this took ~28 seconds also.
This makes sense, since it seems that KScript is precompiling and caching the compiled script.
Bringing down the run to 28 seconds is great, but still felt way too long for a set of 150 images. As I was thinking about this while rerunning the script, I noticed something interesting - every few runs of the script, I’d get a dock icon in macOS, which would then disappear, followed by another one. This brought me to a realization - what if the reason this is so slow is that we are processing a single file per run, causing the jvm to spawn once for each image processed.
Going back to the original script and measuring the time of the entire method, I saw that processing an image takes roughly 75ms. Based on this, the expected time would be 75ms * 150 images =~ 11.25 seconds. Moreover, updating the initial shell loop to add a time in front of each run shows runs that are taking a bit longer (over 110ms) per run. The combined signal from these should have been enough to consider this optimization sooner.
What if we modified the script to run on each directory of images instead of on each image? I modified the script to handle a directory at a time, and re-ran the loop, without the inner loop from 0-14. The updated times were really surprising.
kotlin: 15 seconds.kscript: 16 seconds.kscript: 10 seconds.The real issue here was the cost of starting up a jvm for each file. Partially combining the files (to do 10 jvm process starts instead of 150), brought the time from ~28 seconds, to close to 10 seconds (or, with a vanilla Kotlin script, from ~2.5 minutes to 15 seconds). Using this information, combining the 10 runs down to a single run by further modifying the code to support nested directories brings the run time down to ~1.5 seconds (compiled), or ~5 seconds using Kotlin scripting 1. Not bad!
Kotlin Multiplatform is very powerful, and provides us the ability to compile Kotlin for non-JVM platforms (by going through LLVM). KMP could easily allow us to take our script and compile it for non-JVM platforms, therefore making it native. In other words, given something like finder.kt, I could do something like:
# install kotlin-native - on macOS, we can do:
brew install kotlin-native
# for the first time running kotlin-native on macOS, we'd have to clear the
# quarantine extended attribute.
xattr -d com.apple.quarantine '/opt/homebrew/Caskroom/kotlin-native/1.8.21/kotlin-native-macos-aarch64-1.8.21/konan/nativelib'/*
# now we can run it through the kotlin-native compiler
# won't work if we have any java.* or android.* imports
kotlinc-native finder.kt -o finder
This would give us a native command line application, without the jvm. We’d expect the performance to be better than what we’ve seen so far, due to not going through the jvm. I didn’t test this approach, however, due to the fact that ImageIO is a jvm only construct. To do this, I’d have to use skia, or expect/actual methods for reading and manipulating pixels on the various platforms.
There are three key takeaways here:
This is ~10ms per image, which is a lot less than the estimate of 75ms per image (since that 75ms was a measure of the entire execution). ↩︎
At KotlinConf this year, JetBrains announced that Compose for Web will be powered by WebAssembly. The example repository was very compelling, showing that a composable function can now be written on Android and shared across to iOS and web. I quickly got to work updating one of our internal projects to support this and was able to get it working fairly quickly. My coworker, Efeturi, said, “it’s nice but, too bad it’s still in alpha.” Little did I realize that trying to answer that simple comment would lead me to discovering something I never realized before - you can actually use the same Composable functions on the web today without WebAssembly (and thus without the alpha dependency).
But first, a bit of history…
In 2021, JetBrains announced a way to write Compose for the browser. This method, which allowed us to write Compose applications for web using something very similar to what we’d write using HTML and CSS, is now referred to as Compose HTML.
renderComposable(rootElementId = "root") {
Div({ style { padding(25.px) } }) {
Input(type = InputType.Text, attrs = {
onInput { inputText.value = it.value }
onKeyUp {
if (it.key == "Enter") {
searchText.value = inputText.value
}
}
})
Button(attrs = { onClick { } }) {
Text("Search")
}
}
}
This is really powerful for people developing for the web. For everyone else, it’d be a lot more convenient to be able to use the same @Composable functions we wrote for Android or iOS on the web.
While going through the example repository, I noticed that the web ImageViewer sample supported two platforms - WebAssembly and JavaScript. I asked about this on Kotlin Slack, and received an answer about this from Slava Kuzmich, who works on Wasm at JetBrains:
They both use Skia library compiled to WebAssembly. The difference is that
jstarget compiles your Kotlin code to JavaScript, whilewasmcompiles Kotlin code to WebAssembly.
We have two versions of Compose:
- “Compose for Web” is compatible with Android/Desktop/iOS but it doesn’t use HTML/DOM elements and renders everything on canvas using Skia.
- “Compose HTML” is the older version that uses different HTML-like APIs to generate DOM.
We currently focus more on the canvas version.
Therefore, Compose Multiplatform actually supports 2 primary flavors of Compose - Compose HTML and Compose for Web (which can target JS or WebAssembly). The latter is multiplatform and relies on Skia for rendering. The former is only available for usage in Kotlin/JS.
It ends up being pretty straight forward to adapt Compose on the Web in an app. I adopted the code in the wasm example repository’s jsMain directory and Gradle files to replace Compose HTML in my PrayerTimes KMP Example repository with Compose for Web targeting JavaScript in this pr.
Note that the official example, while a great guide for getting things running on the web targeting both JavaScript and WebAssembly, only runs with WebAssembly as of the writing of this post (i.e. :webApp:wasmRun worked for me, but :webApp:jsRun does not).
While it seems that targeting WebAssembly will be the most future proof way to have Compose on the web due to its performance and some massive work around optimizing sizes by the teams, there are three reasons to target JavaScript in the meanwhile today:
dev version of Compose todaySeveral people have already written about their journey migrating from Gson to Moshi. So why write this post? While migrating a fairly large project to Moshi, my team learned a few things (and came up with a few simple tools) that made it safer and easier to do this migration. Some of these points were either mentioned in passing in the aforementioned articles, or were skipped over altogether, and they are the focus of this post. Consequently, I’ll skip things like the motivation of moving to Moshi, since some of the aforementioned articles cover that pretty well.
One of the earliest concerns about migration is how to deal with Gson’s .setFieldNamingPolicy method, which allows the automatic mapping of something like created_at in json to createdAt in Java/Kotlin. One of the articles on this matter in specific that resonated with me when I read it long ago is Jesse Wilson’s article on naming conventions, which specifically addresses this point. Due to the arguments in that article, we decided to explicitly add annotations (@Json(name = "")) to the fields. While manually adding them is a lot of work, it can be done incrementally, and, as we discovered during the migration, it can help identify existing bugs that no one noticed before.
Slack’s moshi-gson-interop is a useful tool for slowly migrating from Gson to Moshi. It tries to intelligently decide whether a particular model should be deserialized using Gson or Moshi, based on some heuristics (which one can optionally tweak if needed). One such heuristic is that classes that are annotated with Moshi’s @JsonClass are given to Moshi to deserialize.
There’s also a logger parameter that lets us tell the library how to log various events.With this, we can easily watch logcat (for example) to know which classes are deserialized by Gson and which by Moshi and annotate more classes until the entire payload of a particular screen is all deserialized via Moshi.
To set this up, we do something like:
@Provides
fun provideGson(): Gson {
val originalGson = // ...
val originalMoshi = // ...
return if (ENABLE_MOSHI_INTEROP) {
val (moshi, gson) = originalMoshi.interopBuilder(originalGson)
.logger { Timber.d("moshi-gson: $it") }
.build()
gson
} else { originalGson }
}
Notice the nice property allowing us to gate this feature to a subset of people and safely continue using Gson while we build trust in our migration. This helps us stage our migration in a (mostly) safe and iterative manner.
At a high level, the migration strategy we used went a bit like this:
As mentioned above, moshi-gson-interop makes it easy to toggle this feature to make this migration incremental. Initially, for example, only a single developer might enable interop. Once a critical mass of screens has been migrated, we might enable this for all engineers, and only start rolling it out to employees after we’ve gained enough confidence on it.
Since most of our network calls use Retrofit, we wanted to build something for testing purposes to help us validate the correctness of the json data. Retrofit allows us to add a Converter.Factory while initializing it. Typically, this is the in-built GsonConverterFactory.create(gson) instance, for example. We can choose to either replace (or wrap) this instance to do some neat things.
For validation, for instance, we use a Converter.Factory that deserializes the data twice - once in Gson, and once in Moshi, and compares the outputs. This should only be used for development, since it’s very inefficient (both due to the double deserialization of the payload, and the one-shot reading of the entire response body). This looks something like this:
class GsonMoshiValidationConverterFactory(
private val gson: Gson,
private val moshi: Moshi
) : Converter.Factory() {
override fun responseBodyConverter(
type: Type,
annotations: Array<out Annotation>,
retrofit: Retrofit
): Converter<ResponseBody, *>? {
return InteropConverter<Any>(type, gson, moshi)
}
// TODO: override requestBodyConverter here if needed for migration
// in a similar way to what we did for the response body above. We'll
// need to write another Converter that takes the actual value and
// converts it to json, once for Moshi and once for Gson. Because the
// field ordering maybe different, we can do validation by converting
// the json strings back to objects and comparing the objects.
}
class InteropConverter<T>(
private val type: Type,
private val gson: Gson,
private val moshi: Moshi
): Converter<ResponseBody, T> {
override fun convert(value: ResponseBody): T? {
// WARNING: This is very inefficient, do NOT use this outside of development
val bodyAsString = value.string()
val moshiResult = try {
val adapter = moshi.adapter<T>(type)
adapter.fromJson(bodyAsString)
} catch (exception: Exception) {
// handle parsing exception via Moshi
}
val gsonResult = gson.fromJson<T>(bodyAsString, type)
if (moshiResult != gsonResult) {
// flag mismatch between Moshi and Gson
}
return gsonResult
}
}
Using this, we can clearly flag cases where Moshi and Gson deserialize something differently and can work to fix it. In order to work, however, the caveat is that the types being compared need to properly have an implemented .equals methods. One nice side effect of this is that whenever a particular item mismatches, we can also get the body as a string, and write a small validation unit test to iron out the cases.
This same trick of rolling our own Converter.Factory is useful for other things also, such as for very roughly measuring the performance of deserialization in Moshi versus Gson, and for surfacing exceptions at parsing time that are swallowed somewhere in the upstream code.
Note that if json is being sent, we should also override requestBodyConverter in the Converter.Factory similar to what we did for the rsponseBodyConverter. In this method, we can then convert the object to json for both Moshi and Gson. Note that if we compare them at this time, we’ll get a lot of noise due to the ordering of the fields being different. To work around this, we can re-serialize the json back to the type again, and check equality after the round trip.
While developing, we ran into several interesting problems and issues that we’ll go over here. Many of these are arguably working around bugs on the backend, and often times, the right solution to these problems is to reach out to the backend team to fix the result instead of working around it on mobile.
For sealed class hierarchies, we used moshi-sealed. Under the hood, moshi-sealed will create a PolymorphicJsonAdapterFactory, which will decide the flavor to create based on the parameter type. In case of an unknown type, a fallbackJsonAdapter can be passed in. One known issue is that if the backend sends no type (i.e. the field is absent), Moshi will throw an exception. This can likely be worked around with a custom JsonAdapter if necessary, but it would make more sense to ask backend to properly send a type in this case instead.
There are some scenarios where a payload will have a field coming back with any one of several json fields, such as in this example:
data class Person(
@SerializedName(value = "name", alternate = ["full_name"])
val actualName: String
)
In this case, the actualName field in our Person object should contain whichever the server sent back of name or full_name. One approach for solving this is mentioned in one of the articles mentioned earlier, but that approach doesn’t work well when the model has many fields (since it would otherwise result in a lot of duplication just for mapping).
When it wasn’t easy to use the above approach, we opted to take what we thought would be an easier approach to this. Note that the below code has a bug, we’ll get to that after:
data class Person(
val name: String?,
@Json(name = "full_name")
val fullName: String?
) {
// WARNING: This is actually wrong when using Gson and will result in a null
// value of actualName, even if one of name or fullName are passed in.
@Transient val actualName = (name ?: fullName)!!
}
We figured we’d make all the different names as nullable variables, and knowing that the backend will always return one, we force our result to be whichever one of those fields is non-null. Note that, without the @Transient, Gson will try to look for an actual_name in the json to set this to.
While this works well for Moshi with codegen, this doesn’t actually work for Gson (when the type doesn’t use a type adapter), where actualName will return null, irrespective of the value of name or fullName. This happens to be a side-effect of how Gson makes these values via reflection.
If we look at the decompiled bytecode for our Kotlin data class, we can see that this variable gets set as part of the constructor:
public Person(@Nullable String name, @Json(name = "full_name") @Nullable String fullName) {
this.name = name;
this.fullName = fullName;
String var10001 = this.name;
if (var10001 == null) {
var10001 = this.fullName;
}
Intrinsics.checkNotNull(var10001);
this.actualName = var10001;
}
Looking through these posts on StackOverflow, we find out that the reason is that without a custom deserializer, Gson uses Unsafe.allocateInstance to make an instance reflectively, bypassing the constructor. It reflectively sets the properties afterwards, resulting in our actualName never being set. To fix this, we can just change actualName to a getter instead:
data class Person(
val name: String?,
@Json(name = "full_name")
val fullName: String?
) {
val actualName: String
get() = (name ?: fullName)!!
}
Suppose we had a model with:
// case A
data class Widget(identifier: Int)
// case B
data class Widget(identifier: Int = 0)
In case A, if no identifier is set in the json, the Gson result would set it to 0, whereas the Moshi result would crash due to the field being missing. We can fix this by updating to case B, where we set a default value.
What if the json contained an identifier set to null? We’d expect the value to be defaulted to 0, and, in Gson, it is. However, in reality, this throws an error instead. While there are some suggested solutions in this thread about working around this, we opted for a similar approach to the above:
data class Widget(
@Json(name = "identifier")
@SerializedName("identifier")
internal val internalIdentifier: Int?
) {
val identifier: Int
get() = identifier ?: 0
}
One interesting note is that if the default in case B is not 0, Gson will still default to 0, which should make sense considering the note in the alternate keys section above.
This article offered a set of suggestions to make the migration from Gson to Moshi easier. By making the upgrade incremental, toggleable, and able to be validated, we make the migration a lot more achievable.
A little over a year ago, I gave a talk about building a PrayerTimes mobile application for iOS and Android using Kotlin Multiplatform some time ago. I recently gave a second talk about expanding this application to run on a plethora of other platforms (desktop, web, macOS, Linux, watchOS, iOS using Compose). You can watch it here and see the code on GitHub.
In a previous article, I talked about writing profiler-util, a tool I open sourced for visualizing build performance over time for personal projects (it interops the files generated by gradle-profiler and uploads the data to a Google Spreadsheet, along with providing tools for detecting regressions).
I typically run gradle-profiler on my personal projects every handful of PRs, especially those that I expect might affect build speeds (Gradle plugin versions, AGP versions, compiler plugins, etc).
The scenarios file I use for gradle-profiler consists of a number of cases, including abi and non-abi changes to app and non-app modules, changing resources, adding a composable function, clean builds using build cache, and configuration (see this page from the Android developer documentation and the gradle-profiler readme for more details).
Recently, I checked to see how the build performance for my app was doing, and was surprised to see this:
Looking closer, I could see the times greatly increasing for all but 3 metrics - configuration change, the change of a resource, and an incremental build without having changed any code. Given this, I looked and found that the change was upgrading AGP from 7.2.1 to 7.2.2 in this project. Surprised, I checked my other projects, and none of them had the same issue. A few weeks later, when I upgraded to AGP 7.3.0, the build times still didn’t improve, which surprised me.
I decided to look into this to try to figure out what was happening.
Given that this was only happening in this project and none of my other projects, I decided to look into one of the following initial potential explanations:
Going through these quickly:
clean_build_with_cache, and ran it without gradle-profiler (./gradlew clean; ./gradlew assembleDebug) with AGP 7.2.1 and again with AGP 7.2.2 to eyeball the results - once again, they seemed consistent with the profiler results, thus removing the mismatched properties as an explanation.Given that none of the above worked, I decided to try something else that
ultimately lead me to the problem - I ran the above clean_build_with_cache
commands locally with --profile for both AGP 7.2.1 and AGP 7.2.2:
git checkout <last commit with 7.2.1>
# warm up caches, etc due to AGP version change to 7.2.1
./gradlew clean
./gradlew assembleDebug
./gradlew clean
./gradlew assembleDebug --profile
git checkout <first commit with 7.2.2>
# note that profile html output files survive gradle clean
# warm up caches, etc due to AGP version change to 7.2.2
./gradlew clean
./gradlew assembleDebug
./gradlew clean
./gradlew assembleDebug --profile
I then compared the result files:
The delta in build times is the only thing that stands out. Going through the configuration, dependency resolution, and artifact transforms tabs, nothing stands out and all numbers are close to each other.
This leaves the task execution tab, which shows us something interesting:
Why is Realm taking 6 seconds when using 7.2.2 but not when using 7.2.1? And what’s Realm’s processor doing in app?
Realm objects have various annotations (@RealmModule on module classes, @PrimaryKey for primary keys, @Index for indices, etc), and requires an annotation processor, bundled within a realm-android plugin.
Historically, I had all said annotated classes in a separate, standalone module just for Realm models, since I know they’d rarely change and I didn’t want to pay the price of said processor every time. This is a good optimization in said cases.
Indeed, looking at app/build.gradle, I found the realm-android plugin in use there. So how did it get to app? Apparently, sometime back in May of 2016, io.realm imports were not resolving in the app module anymore (by not resolving, I am guessing this was in the IDE and not actually at compile time, though today, I have no way of being sure).
Removing the realm-android plugin brings back the compilation times to what they were before AGP 7.2.2 and above. This, however, doesn’t explain what AGP 7.2.2 had to do with this.
Warning - this is purely speculation at this point, and I am not certain about anything else in this paragraph. Looking at the release notes for 7.2.2, there are only 2 fixes there. One of them is this one, which has to do with fixing a bug related to the transform API when used with the ASM api. ASM is used for bytecode manipulation, and is in use by realm-android. I suspect that this bug caused the plugin to do nothing in app (which was fine since it’s technically not needed as mentioned above). The bug was introduced in 7.2 alpha, and wasn’t fixed until 7.2.2. The first commit in my profiling was using 7.2.0, so I never noticed any “improvement” between 7.1.x and 7.2.x, only to see it being lost again in 7.2.2. This is, however, only a guess.
Some takeaways here:
--profile flag is your friend.As an Android code base grows and has increasing code churn, unused resources are very likely to exist. While Android Studio has an option for cleaning up and removing unused resources, there are some reported bugs in this functionality that are still not fixed.
I started searching for another solution for finding and cleaning up unused resources in Android projects. I found this project, which has some Python 2 scripts to directly manipulate resources. Since it hadn’t been updated since 2018, I took the liberty to build something based on it, which fixes some of the issues I ran into while using it.
Introducing resource-cleanup - an open source project to help identify and clean unused resources (in retrospect, I wish I had named it “Mr. Kaplan”). It relies on several open source tools and heavily relies on the command line to accomplish its role.
The project has a set of shell scripts - one per resource type. It searches for usages of @[resourceType] or R.resourceType references. In the absence of these, it removes the given resource. The shell scripts take a single file and checks for usages accordingly. The documentation gives examples of how to combine this with tools like fd and shell loops.
These scripts have proven invaluable to us at work and helped us reduce our apk size by a sizable amount. Please give it a try!