I've been running some experiments based on <the e...

# compose

[Mark Murphy]

I've been running some experiments based on the excellent post by @Zach Klippenstein (he/him) [MOD] regarding scope of recomposition. I feel like I'm getting different results, and I'm not sure where I'm going wrong in my analysis. Details in 🧵 .

[Mark Murphy]

I have this in a scrap project:

@Composable
private fun RecomposeTest() {
  val switchState = remember { mutableStateOf(false) }

  Log.d("wut?", "RecomposeTest()")

  Row(modifier = Modifier.padding(8.dp)) {
    Switch(
      checked = switchState.value,
      onCheckedChange = { switchState.value = it },
      modifier = Modifier.padding(end = 4.dp)
    )
    Text(stringResource(R.string.app_name))
  }
}

I had expected that only the Switch() would get recomposed when the Switch() itself is clicked, as the call to Switch() is the only place where I seem to reference that switchState value.

Yet, I get the log message on every click of the Switch(). Unless I am misunderstanding something, that means that RecomposeTest() is being recomposed, not just the Switch(), or even the Row().

FWIW, I see this with beta01 and beta04.

[Mark Murphy]

The RecomposeTest() is being called from the setContent() call in the activity:

class MainActivity : ComponentActivity() {
  override fun onCreate(savedInstanceState: Bundle?) {
    super.onCreate(savedInstanceState)
    setContent {
      MyApplicationTheme {
        // A surface container using the 'background' color from the theme
        Surface(color = MaterialTheme.colors.background) {
          RecomposeTest()
        }
      }
    }
  }
}

Otherwise, the project is standard Arctic Fox Canary 12 "Empty Compose Activity", except for bumping things up to beta04 after realizing that the template was at beta01 .

[Mark Murphy]

So, the tactical question is: why is that Log.d() call being made for every Switch() click?

The more interesting question is: how do folks like me figure out what is going on here? IOW, is there a way that I can determine why the Compose runtime is deciding that RecomposeTest() needs to be called again?

[Sean McQuillan [G]]

Short answer: Row is inline

[Sean McQuillan [G]]

Longer answer, parameters are resolved to values in the parent recompose scope, so any parameter that reads a state means the calling site is what recomposes, not the callee

[Sean McQuillan [G]]

In this case you're passing:

checked = switchState.value

and the read happens there (when resolving the function parameters)

[Mark Murphy]

right, and with Row() being inline, that's actually in the body of RecomposeTest()

🤦🏻‍♂️

[Sean McQuillan [G]]

yep

[Mark Murphy]

OK, that's going to take some getting used to, but it makes sense. Now I just gotta figure out how to explain that sort of behavior in an introductory Compose book, without breaking anyone's 🧠 ...

Thanks for the clarification!

[Sean McQuillan [G]]

hah

[Mark Murphy]

I suspect strongly that I will wind up with the The Sean McQuillan Approach™️ of "don't worry, Compose will figure out the right scope" 😁

[Sean McQuillan [G]]

HAHA

[Sean McQuillan [G]]

I was just writing that 🙂

[Sean McQuillan [G]]

But really, I'd advise to be cautious about what you show in a book as this runtime behavior is not guaranteed to be stable at all (and future optimizations are likely)

[Sean McQuillan [G]]

If you do want to show an example, I'd suggest doing something you have full control over and do the state read clearly in the composition scope such that it could never realistically be optimized

[Sean McQuillan [G]]

e.g.

val aColorForSomeReason = if (switchedState.value) { Color.Red } else { Color.Blue }

[Mark Murphy]

Understood.

At the same time, Compose's compiler magic makes logging more... interesting, in the "may you live in interesting times" sort of interesting. For example, logging a state value affects the scope of recomposition, if your logging is outside what might ordinarily have been recomposed.

At some point, I want to warn developers about these sorts of impacts.

[Sean McQuillan [G]]

Yea

[Sean McQuillan [G]]

I think it's a good chapter actually

[Sean McQuillan [G]]

I'd just avoid (1) any builtins [they're very likely to change behavior] (2) anything that could be optimized due to only being used once with no side effect from the value

[Mark Murphy]

how are you defining "builtins"? do you mean the standard composables like Row()?

[Sean McQuillan [G]]

yea

[Sean McQuillan [G]]

Defining a regular Composable fun is gonna get you a restart scope 🙂

[Sean McQuillan [G]]

And the composable lambda behavior is probably the big trick to show off

[Mark Murphy]

OK, thanks again!

[Zach Klippenstein (he/him) [MOD]]

Late to this thread but I do call out the inline caveat at the end of the post. Perhaps it should be more prominent? It’s easy to miss, and since some of the most common composable wrappers are inline maybe it’s worth bringing it forward a bit

[julioromano]

Here’s the link to a relatively old convo where I was asking a similar question to which @Adam Powell and @Chuck Jazdzewski [G] answered with very knowledgeable insights.

https://kotlinlang.slack.com/archives/CJLTWPH7S/p1611232921023400

[Mark Murphy]

I do call out the inline caveat at the end of the post

I saw that, both before and after I started this thread. In my case, it was mostly a matter of "not putting two and two together" with what you had there and the symptoms that I was seeing. IOW, 🤦🏻‍♂️ . Again, the post was great!

[Zach Klippenstein (he/him) [MOD]]

I pulled the paragraph about inline functions into its own heading and added some more detail and an example to hopefully make it easier to catch.