The same story many experienced with SwiftUI vs UIKit. Apple has flubbed the move to declarative APIs hard. It's a scary experience committing to these APIs, because of the wild changes they experience, the performance issues, and the fact that they are borderline un-debuggable by the average mind compared to imperative APIs, so when you run into real trouble, you're often kind of just fucked. And it's not even like normal 1st party frameworks where the bug is virtually always in your code. With these, sometimes the bug really is Apple's.
I wish these things didn't have "Swift" in the name, since their problems have nothing to do with the language as far as I can tell, and they make Swift look bad.
That this is the best the richest company in the world can do with the API to their most valuable product is quite astonishing.
But then I hit a "Unable to infer complex types" error in SwiftUI, and think maybe the Google folks got it right after all.
These should really be open source: it's such a waste of time to have to experimentally poke around to see how the SwiftUI components interact with their UIKit counterparts, when these should really just be straightforward wrappers.
In many cases, the declarative APIs are just poorly thought out and/or clearly not "battle tested", and there is a lot of API churn in really crucial components which are also often not backported to older OS versions.
The worst offenders by far are the scroll & list views, which are basically useless. As a starting point, there is no way to get the current scroll position from a scroll view in iOS <18.
So yes, I often have to wrap the UIKit components myself, and deal with all the joys of UIKit I was hoping to finally avoid...
Apple did exactly the right thing with Swift itself: the Swift forums are excellent, and it's clear that Apple employees use these to prioritize the (very real) Swift bugs and limitations. Why does this not exist for SwiftUI?
I can’t imagine how it could be implemented more simply.
Even if you are rendering a complex chain of compute shaders with multistage heap fences onto a process shared IOSurfaceRef with several concurrent threads greedily realiasing your scratch buffer to queue up more work, it doesn’t matter. Eventually you’ll up with a UIView or CALayer and it’s still just one function call to bring it into SwiftUI.
It's funny, when SwiftUI and Compose were coming out we waited a bit but jumped in as soon as we could. With Compose, our early versions required us to make builds of Android Studio on our own machines that supported it. But even then Compose felt more polished in a lot of ways - and if we ran into issues we could _see the code_ and make our own components with the changes we needed if necessary.
SwiftUI is great, really, but has so many rough edges, gotchas, and bugs. They do get fixed, but you're still supporting 4 year old versions of iOS without those fixes, so you're screwed.
Just completely boneheaded not to open source SwiftUI from the start, especially with how great the open source Swift community is.
SwiftUI has the unfortunate combo of being closed-source, being platform-dependent, and being buggy in the most unusual ways. Which is a shame, because SwiftUI has a lot of good things going for it.
In my experience, the trick is making small tested and working views, and combine these into fullly-fledged screens.
In the present, Apple has beautiful hardware and sucky hardware.
No idea when they will get both beautiful hardware AND software.
(Frankly, I preferred good old, stable Objective-C - Apple ditched the KISS principle, when they adopted Swift).
Luckily, I had too many things to work on, so I didn't get around to it. I bookmarked this page for ideas the next time I have to deal with Core Data, thank you for the article!
No way.
At least right now, SwiftData shouldn't used for anything but quick-and-dirty PoC's, demos, or other limited-use, low-stakes things.
There are some good ideas in it, especially how it integrates with SwiftUI. But it's alpha quality right now.
The article actually oversells SwiftData, IMO. I don't think it wasn't usable in iOS 17.
And its concurrency story isn't great. E.g. if you want to do queries or other data access in the background, you'll need to create and maintain separate, Sendable variants of your models. Now, I get that SD is really a memory-based database (typically backed by sqlite), so you can probably do a lot right on the main thread, but you'll end up wanting to move some things for various reasons, and it's awkward and boiler-platey. Not to mention, models have a magical connection to containers. You can't see it directly, but try to instantiate a model without a container (for, you know, preview -- a core feature of SwiftUI).
There are also some creaky design decisions. On one hand, it acts like an ORM, that wants to persist an object graph. But it also leaks relational concepts all over the place. They probably should have picked one approach or the other... as it is, at every turn you have to figure out if they went relational or object graph, and program accordingly.
Now, CoreData has it's own issues, and, I assume, Apple is no longer going to focus on it, so I'm not sure it's a good idea to use it either.
Personally, I've just started using sqlite3. With that you've got all the speed, power, and reliability, and don't have to hope Apple doesn't blow up your project on each release.
(I use some utilities to take the boiler plate out of passing Swift values as parameters and receiving Swift values as results. Other than that, you can just call the sqlite3 C API directly.)