I think the rough plan for OkHttp multiplatform JS support would be. omit the streaming data accessors, and only have one providing completed data. String, bytes. But follow okio on this. Not sure how that applies to natural streaming concepts like sockets.

Low confidence on this prediction.

In the one place I use it I buffer the whole body before wrapping

I think the rough plan for OkHttp multiplatform JS support would be. omit the streaming data accessors, and only have one providing completed data. String, bytes.

Buffering the entire contents of a large response in memory isn't going to work for large/streaming responses. I'd imagine what I'll have to end up doing is creating an async version of Pipe and then maybe wrapping it in sum type we actually expose like:

sealed class ByteStream {
     data class Source(val source: okio.Source): ByteStream(), okio.Source by source
     
     data class Streaming(val source: AsyncStream): ByteStream()
}

// convenience functions for most common operations that deal with the variant differences 
suspend fun ByteStream.writeToFile(file: File): Unit
suspend fun ByteStream.toByteArray(): ByteArray
suspend fun ByteStream.toBuffer(): okio.Buffer

For JVM/Native it will be relatively simple, just expose the underlying contents of the response as a

Source

and have the convenience suspend functions block on

<http://Dispatchers.IO|Dispatchers.IO>

. On JS expose an async abstraction that introduces one layer of buffering between the consumer and producer.

Buffering the entire contents of a large response in memory isn't going to work for large/streaming responses.

We might choose to not support these on JS.

Interesting. Ok well I appreciate the insights into how you guys are considering supporting (or not supporting) this use case. We haven't gotten to KMP support yet but this question on JS is a big "known unknown" for us as well. We won't be able to not support streaming/large responses though so JS I/O will likely be a different abstraction, time will tell I suppose

For KotlinPoet, if I got a

KSFunctionDeclaration

which I know I want to call inside of the body of a function I am generating, is there a way for me to do that in a better way that just doing something like this:

fun FileSpec.Builder.foo() {
  addImport(ksFunctionDeclaration.packageName.asString() + ksFunctionDeclaration.simpleName.asString())
  addFunction(
    FunSpec
      .builder()
      // stuff
      .addStatement("${ksFunctionDeclaration.simpleName}()")
      .build()
  )
}

I feel like I must be missing some part of the API that simplifies this. or am I wrong?

You're looking for a

MemberName

which will hold a reference to a function and automatically emit an import when used with

%M

details of

%M

are documented on

CodeBlock

https://square.github.io/kotlinpoet/1.x/kotlinpoet/kotlinpoet/com.squareup.kotlinpoet/-code-block/index.html

Hey, is the

redwood-layout-uiview

not being published to snapshots yet? I'm unable to find it, but I might be looking wrong

I have no idea, but I can look later

We may have just forgot to apply the publish plugin to it

Thanks. I just checked and you're right the publish plugin is not being applied. Should I create a github issue for that?

Sure!

Is there a way to get Redwood working without going through the serialize/deserialize route and instead have the Compose code drive the Widget code directly?

Not currently, there's a big TODO. There shouldn't be anything really blocking it except some work

Cool, thanks. I might dig into that then. Is there some docs/info about how you'd want to implement it? So I don't go and reinvent the wheel?

It's been a long time since I've thought about it, but the design is such that the Compose code already talks directly to the Widgets. The only thing is that we only support the protocol implementation of widgets because of the applier in use which appends operations directly to the protocol. The change that would have to be made is that the applier would need to talk to the Widget.Children instance on the current widget. Then, for the protocol, we would just use an implementation that appended to the protocol. I can look a bit at it today and maybe write up some of what is needed.

I see. No worries, let's not waste your time now. I checked with my team and we should be good with just using the protocol implementation for now.

This was originally tracked by https://github.com/cashapp/redwood/issues/14 but like 95% of the work has been done

Hi, I'm trying to use molecule in KMP, but I can't build it for ios. I'm stuck with the following error. Has any one faced the same problem and found a solution? Just updated to Molecule 0.6.0.

> Task :client:linkPodDebugFrameworkIosArm64
e: Compilation failed: Unbound symbols not allowed

Unbound public symbol IrSimpleFunctionPublicSymbolImpl: presenters/MoleculePresenter.models|4816226463707555740[0]

 * Source files: 
 * Compiler version info: Konan: 1.7.20 / Kotlin: 1.7.20
 * Output kind: FRAMEWORK

e: java.lang.AssertionError: Unbound symbols not allowed

Unbound public symbol IrSimpleFunctionPublicSymbolImpl: presenters/MoleculePresenter.models|4816226463707555740[0]
        at org.jetbrains.kotlin.ir.util.SymbolTableKt.noUnboundLeft(SymbolTable.kt:1218)
        at org.jetbrains.kotlin.psi2ir.Psi2IrTranslator.generateModuleFragment(Psi2IrTranslator.kt:96)
        at org.jetbrains.kotlin.psi2ir.Psi2IrTranslator.generateModuleFragment$default(Psi2IrTranslator.kt:75)
        at org.jetbrains.kotlin.backend.konan.PsiToIrKt.psiToIr(PsiToIr.kt:187)
        at org.jetbrains.kotlin.backend.konan.ToplevelPhasesKt$psiToIrPhase$1.invoke(ToplevelPhases.kt:120)
        at org.jetbrains.kotlin.backend.konan.ToplevelPhasesKt$psiToIrPhase$1.invoke(ToplevelPhases.kt:118)
        at org.jetbrains.kotlin.backend.common.phaser.PhaseBuildersKt$namedOpUnitPhase$1.invoke(PhaseBuilders.kt:96)
        at org.jetbrains.kotlin.backend.common.phaser.PhaseBuildersKt$namedOpUnitPhase$1.invoke(PhaseBuilders.kt:94)
        at org.jetbrains.kotlin.backend.common.phaser.NamedCompilerPhase.invoke(CompilerPhase.kt:96)
        at org.jetbrains.kotlin.backend.common.phaser.CompositePhase.invoke(PhaseBuilders.kt:29)
        at org.jetbrains.kotlin.backend.common.phaser.NamedCompilerPhase.invoke(CompilerPhase.kt:96)
        at org.jetbrains.kotlin.backend.common.phaser.CompilerPhaseKt.invokeToplevel(CompilerPhase.kt:43)
        at org.jetbrains.kotlin.backend.konan.KonanDriverKt.runTopLevelPhases(KonanDriver.kt:40)
        at org.jetbrains.kotlin.cli.bc.K2Native.doExecute(K2Native.kt:94)
        at org.jetbrains.kotlin.cli.bc.K2Native.doExecute(K2Native.kt:37)
        at org.jetbrains.kotlin.cli.common.CLICompiler.execImpl(CLICompiler.kt:99)
        at org.jetbrains.kotlin.cli.common.CLICompiler.execImpl(CLICompiler.kt:47)
        at org.jetbrains.kotlin.cli.common.CLITool.exec(CLITool.kt:101)
        at org.jetbrains.kotlin.cli.common.CLITool.exec(CLITool.kt:79)
        at org.jetbrains.kotlin.cli.common.CLITool.exec(CLITool.kt:43)
        at org.jetbrains.kotlin.cli.common.CLITool$Companion.doMainNoExit(CLITool.kt:179)
        at org.jetbrains.kotlin.cli.bc.K2Native$Companion$mainNoExitWithGradleRenderer$1.invoke(K2Native.kt:412)
        at org.jetbrains.kotlin.cli.bc.K2Native$Companion$mainNoExitWithGradleRenderer$1.invoke(K2Native.kt:411)
        at org.jetbrains.kotlin.util.UtilKt.profileIf(Util.kt:22)
        (...)

Can you add https://github.com/cashapp/molecule/blob/trunk/gradle.properties#L37-L39 to your gradle properties and try?

Why does molecule do not find the MonotonicFrameClock on Compose web (html dom api)? I playing with molecule and on android this frame sync works nice, but compose web fails with an exception. I thought this frame clock is mandatory for compose runtime. And the dom render uses DefaultMonotonicFrameClock 🤔 https://github.com/JetBrains/compose-jb/blob/685e305c5ee62bfd1c6c004163a2cd80c6017c98/web/internal-web-core-runtime/src/jsMain/kotlin/org/jetbrains/compose/web/renderComposable.kt#L40 Edit: found a workaround, adding the DefaultMonotonicFrameClock manually to the coroutine Scope on JS

are you using it in a scope whose coroutine context has a clock in it?

I’m looking at a pattern shown in the image below from a very helpful talk on Molecule by Ash Davies and Bill Phillips. I understand these snippets aren’t meant to be taken literally, but I’m having difficulty filling in some of the blanks. 1. I get

A MonotonicFrameClock is not available in this CoroutineContext...

crash when I use

viewModelScope

to launch Molecule as it is bound to

Dispatchers.Main.immediate

. My solve was to simply launch with

AndroidUiDispatcher.Main

, since having the ViewModel hold the Presenter is enough to make it survive configuration changes. Am I missing something else here? 2. What is the recommendation for emitting events within a Composable screen (or any view, for that matter)? The snippet has

events

as a SharedFlow, but it does not successfully emit an event when I call

events.tryEmit(LoginEvent)

and calling

emit

directly won’t work outside a coroutine.

I’m looking to use paparazzi in an Android project that also has Kotlin/Native code building for androidNative* targets. Because paparazzi uses android unit tests, it’s having trouble finding my .so files (and if it could, they’d be built for android NDK). Do I need to add a macosArm64 target to my ndk module, or is there a way to make this work with the .so files I already build for androidNative*?

Hey, I'm trying to use sqldelight in multiplatform to connect to a cloud database on AWS from android and ios, but can't work out where to put the URL configuration for this when initialising the drivers, does anyone here know how I would go about this or if it is possible?

Is it okay to run multiple ProtocolRedwoodCompositions in parallel (like embedding in different parts of the app)? Or is there a better way to do it?

Yep, should be fine. No different than running two ComposeViews next to each other on screen with Compose UI. Our composition setup is nearly identical to theirs.