there isn't any difference between Java and Kotlin for this though? aside from SAM syntax which the examples don't show anyway

Hey guys, justed wanted to give some Kudos for your work on various libraries. I'm currently experimenting with KotlinPoet to finally build a GTK GIR generator and KotlinPoet is a huge help!

With Redwood I saw demos that include hotreloading of changes. Would I be wrong to think that this mechanism would also allow for the initialization of a Composable with existing Nodes already structured in the tree? For example is there a way I could persist a state representation of a tree of Nodes to disk, and in the future start a Composition with the persisted state being the initial structure?

In theory yes. That would involve reusing the composition and calling setContent with the new stuff. Currently we do not support that through Treehouse, the reloading mechanism that we use in production, but nothing technical blocks it from a pure Redwood perspective.

The biggest problem might be Zipline/QuickJS. I'm not sure what its module unloading story is. I suspect it's doable, again, but it would require significant work.

I'm working mostly against just the core compose runtime. Would reusing the composition be something that could only happen during the same application session, or might this be possible to do from a cold start? My usage isn't really in a UI context and wanting to use compose to diff between states. My thinking was if I wanted to essentially resume these diffs between sessions what could be the cleanest way

Yes it would have to be within a single process. The slot table is not serializable so it cannot be persisted across process death. This also prevents isomorphic rendering for Compose Web where the first frame is rendered and produced as HTML and then the client-side JS picks up from that state and continues. It's not terrible in that specific case since you just redo the first state and it's not a big deal.

Ah that's too bad. Your example about Compose Web is essentially the same problem so makes perfect sense. I'll just need to find the best way to get that first frame rendered from that persisted state. Thank you

If you only want to persist state and not the actual nodes you can create a custom version of remember that writes to some persistent place, similar to rememberSavable on Android. When you resume you'll have the same state but you'll emit a full node tree rather than a diff. Although depending on what you're doing maybe you could simply ignore that first one?

That seems like a cool idea. I'll check out how rememberSaveable works.

Hello people, I have a question about SqlDelight (2.0 alpha). I have

Movie

and

TvShow

tables, I wanna query them together, so I'm planning to create a

View

and map it to a sealed class in my data layer. Before procede, I wanna make sure there isn't any native support to sealed classes in SqlDelight, in order to avoid to reinvent the wheel. 🙏

EDIT: nvm running from gradle was indeed the issue. TLDR dont try to shortcut the docs and just build the binaries and things will work. Leaving this here in case anyone else tries the same thing when using Mosaic is there any setup that needs to be done on the terminal to get Mosaics redrawing to work? Whenever i run the samples every recomposition gets emitted as new output rather than updating. Not sure if running from gradle is related

Yep. Gradle's process forwarding removes ANSI control sequences.

Side question, do you have a link to somewhere to help me understand whats going on in the input reading section of the robot sample? I poked around in the docs for JLine and couldnt find any references for the non ascii key codes e.g. enter space etc. Specifically unsure whats going on with the nested reads to get the arrow inputs

nope. it was mostly experimentally determined. eventually it'll work like Compose UI with simple callbacks but it'll take a while to get there.

Gotcha, will do some of my own experimenting then. Do you know if theres a way to handle the input on a individual composable level rather than at the top level inside

runMosaicBlocking

but outside

setContent

? My goal is to have multiple different “screen” analogs each self contained in a single Composable and handling the user input differently and independently. Using LaunchedEffect didnt do the trick im guessing for the same reason as the linked bug report. WOuld i just have to pull that out of the composable level entirely?

I know this is 😶 and hopefully I don’t get a paddlin’, but does square publish its design system anywhere? I see some square blog articles, and Maker on github, but I’m wondering if there’s documentation about the design system in general and/or the framework-agnostic tokens/components themselves. Thanks

Is there any prior art that lets me serialize Wire proto message with kotlinx.serialization? Want to encode a message as JSON in KMP 👀

I'm running into an issue where I can't use paparazzi in a source set that has Robolectric tests . I'm hoping I can make a

snapshotTest

source set to work around that, but would

recordPaparazzi

still depend on the Robolectric tests from the unit test source set? Any other issues that might come up with making a custom

snapshotTest

source set?

👋 Hey! I am curious about the workflow library. I have some questions: • What is your view on the lib as it relates with compose taking over the world not just on android in case anything has changed since this. • I've also seen cash app went the molecule route for presenters which is less constrained (which can be good and bad). Do you use both internally, if so how? • I like the explicitness of workflows but i'm wondering how you organize your code in modules? ie: do workflows in modules end up returning

Any

to avoid dependencies and then resolves at the factory level, you have shared interfaces, etc. Thanks!

Different apps use different libraries. Cash has its own navigator/presenter/renderer library and Molecule is one way you can write a presenter for it.

If I were starting a Compose project from scratch I wouldn't use workflow-core for state management / presenters (including navigation); but I would still use workflow-ui for binding view model structs emitted by presenter functions to UI composables. That's the subtext of my last droidcon talk. https://mastodon.social/@rjrjr/109082673687647930

I'm really hoping to find time to spin off a streamlined workflow-ui v2 as its own library, probably to be called Visualizer. But I'm a profoundly lazy person.

One glaring error in that talk: there's no reason to allow config changes to blow up your activity instances in a Compose app

We’re using okhttp as an http client on a ktor server that hits other services for some requests. Are there any best practices here for configuration? Because of okhttp being so prolific on Android it’s hard to find much about other use cases. I’m thinking it would mostly be around sane defaults for

Dispatcher

, but any other advice/docs would be appreciated.

The defaults should be fine. Otherwise they would be varied by the internal platform mechanism that knows on which platform it's running.

Probably the only configuration worth paying attention to is the Dispatcher’s

maxRequests

(default 64) and

maxRequestsPerHost

(default 5) . . . bumping these up is often a good tradeoff where you’ll spend more compute & network resources, and do more stuff concurrently

Hi! Is there a way to have

moleculeFlow

or

launchMolecule

only work when the resuting flow is being subscribed to? I’m trying to achieve a behavior similar to

SharingStarted.WhileSubscribed

so I can stop all processing when app is in background when collecting the resulting flow using

flowWithLifecycle

,

repeatOnLifecycle

or, in Compose UI,

collectAsStateWithLifecycle

.

That is the behavior of

moleculeFlow

It is cold.

launchMolecule

is hot.