Chris Banes

Nice. What’s the backend stack look like?

Nice summary! I didn’t know about the R8 prop

I don’t know about banks in France, so can’t comment. Most of the ‘app’ banks in the UK are great (Monzo, Starling, etc).

For forex I now use Wise, and have had zero problems.

I seriously do *not* recommend Revolut for any serious amount of money.

Used to use them for large sums from the US (RSUs). One got triggered by some automated check, and my money was held for 2 months whilst they ‘investigated’. Customer services were useless and didn’t help.

I like Cloudflare. It hosts my DNS anyway so makes everything very simple.

It’s also cost price, so cheap.

It shouldn’t matter. Zen syncs with Firefox Sync, although I don’t use FF Mobile so can’t say for sure that it works.

Nice post!

I've migrated to Zen Browser (zen-browser.app) for simplicity from Arc. Highly recommended!

There was an example in one of the many splinter threads.

Either way, I’d still argue that the defaults are the safest option for all.

If you want something else, you could code-gen bindings (which we do a lot of for convenience purposes).

I think that this is very codebase specific. If anything, having singletons everywhere feels like a warning sign to me.

Why are view specific things heavy enough to warrant being created only once?

IMO scoping is for controlling lifecycle, and nothing else.

I migrated our entire codebase over to it last week, and we’re now fully on the Metro train. 💺

It has pretty much everything y’all will need, right now.

Ah true true, I forgot that it was an optional experimental feature :skeletor_mirror:

Have you found a way to enforce the usage of path dependencies?

I wonder if there’s a way for a `check` like task could inspect the dependencies across all configurations.

From a Compose point of view I think it’s better to talk about GraphicsLayers, rather than RNs.

RNs are cool, but they’re an Android-only implementation for a GraphicsLayer. In fact, GLs on older versions of Android use a View (since RN wasn’t public then). On CMP, GLs use a SkPicture.

302 bar would have a terrible business plan

I don’t know how they expect anyone to know the location of this bar…

I’m not sure how they really could fix it. They’d need to split the compilation somehow, but you’d then lose the automatic ability to reference any `projects.foo` from any module.

I’d love to see their metrics on how many perform the call gesture, and then actually proceed to speak to someone. I’d bet that it’s a tiny percentage.

What is it with Facebook adding pull-up gestures for the most random actions:

Instagram DMs: Pull up to toggle disappearing messages.

WhatsApp: Pull up to call.

The amount of times I’ve triggered these actions: too many.
The amount of times that I’ve meant to perform them: absolutely zero.

Also, I read this and thought: I like to do both of these kind of things.

Does that make me a ‘bastard’ engineer? I quite like that title tbh.

K/N build speed, especially when running tests. When every module needs to be linked just to run tests, it gets very slow, very quickly.

Ah yes. I have both JB AI and Junie installed so wasn’t sure what was coming from what.

I haven’t tried Cursor tbh, but from what I’ve seen it’s quite similar

GitHub - cashapp/redwood: Multiplatform reactive UI for Android, iOS, and web using Kotlin and Jetpack Compose Multiplatform reactive UI for Android, iOS, and web using Kotlin and Jetpack Compose - cashapp/redwood

That's basically what Compose for Web is: a composable DSL over the native UI toolkit.

It's also reminds of what the CashApp folks are doing with Redwood: github.com/cashapp/redw...

I've been trying out @jetbrains.com Junie over the past few few days, and have pretty blown away with it.

I've managed to do everything from a complex refactoring of Haze (with follow up tasks), to fairly complex string manipulation (pretty printing data class toString).

Fine for one off use cases, like MapView, WebView, etc, etc. I wouldn't build my entire UI around it though.

Bit different (again). RN doesn't try to try to draw to it's own canvas, it's (mostly) wrappers around native components.

UiKitView is that too, but the rest of CMP on iOS _does_ have it's own Canvas. UiKitView works by punching a hole through the canvas (like SurfaceView). That comes with issues.

Very different. Calf will be wrapping the native SwiftUI views through UIKitView or similar.

Personally I’m not a big fan of building UIs like that, but YMMV.

In fact, someone has already contributed a shader: github.com/chrisbanes/h...

My plan is to add support for custom shaders into Haze as public API, and then add a materials class which uses it for LG.

Seems to be a lot of talk about Haze and Liquid Glass…

At its core, Haze is a background content transform library, and the ‘visual effect’ (blurring, etc) is just an implementation detail.

Liquid Glass will be a runtime shader like others which Haze already uses.

I know and jest.

But whilst Material might be the design language for Android, it’s also a design language for other platforms. It’s a weird dual role.

Also, very few people watch talks. This needs to be better documented.