I’ve been using the Swift OpenAPI Generator to generate an interface to a REST server and it works quite well! There is soooo much less code!
But of course, we can’t have nice things. The OpenAPI generator is an Xcode plugin and once you enable plugins in Xcode 15, every other build crashes with “internal inconsistency error”. Apparently this happens across the board with Xcode plugins. Supposedly, the bug is fixed in Xcode 16, but that won’t be production ready for two months.
This is one of those ‘note to self’ blog posts. I’ve been working on the next round of features for GlowWorm including a desktop widget. The GlowWidget, like every widget ever, needs to share some data with the main app. Widgets are all sandboxed and locked down, so we need to use an App Group to enable sharing between the extension and the main app. This is where it gets weird.
The internet is full of info on how to set up app groups. It starts by going to the developer portal, creating a group and adding it to the app identifier. Groups created in the dev portal are always prefixed with group.. Once you’ve got that set up, you can add the App Group in the capabilities section of the Xcode project. Here’s the rub! For a Mac app, Xcode enforces that the app group begin with the developer’s Team ID which looks something like JQ49SEDXYZ. So… does it start with group. or does it start with JQ49SEDXYZ.?
The key info is summed up in this SO post and sort of in the docs. It turns out that Mac and iOS configs are completely different! The dev portal and the group. prefix is only for iOS. None of that needs to be done for Mac (like nothing at all in the dev portal). On the Mac, you just need to use the Team ID prefix and that’s it. The only setup is in the Capabilities section of Xcode, where you add an app group with your Team ID prefix, bundle ID and whatever suffix you’d like. For a bundle ID of com.example.app, you might make the app group JQ49SEDXYZ.com.example.app.shared.
Now that you’ve got an app group, make sure you’ve added that app group capability to all the relevant targets. For GlowWorm, it needs to be added to the main app as well as the widget. There are a couple of ways you can share data between the two targets now, but initially we’re only interested in sharing UserDefaults. To do that, you need to create an instance of UserDefaults explicitly associated with the app group. Using the example app group we’ve set up, you can create the shared instance with UserDefaults(suiteNamed: "JQ49SEDXYZ.com.example.app.shared").
And with that, we’re in business! Both the main app and the extension can access the data in the shared user defaults. Now we just have to make it look pretty. 🙈
My CI build failed today because it couldn’t find two files. After 10 minutes thinking I botched a git merge, I discover that no, git is fine. Xcode decided to create those files in my Downloads folder instead of the project directory. Thanks for that Xcode. You’re the best!
Interesting confirmation of a gut feeling I’ve had for a while. With UIKit on iOS, I see a lot of developers write code that updates bits of the user interface as inputs change. As the code grows, the execution path becomes a convoluted flow chart of branches and conditional updates. Secondary dependencies often get lost, giving rise to bugs that can take hours to track down. And then you’re not really sure that you didn’t break something else with the “fix”.
But you can ditch all this complexity by just updating the whole interface every time. When the view opens, update the interface. When the user checks the box, update the interface. When the network query returns data, update the interface. Now you’ve got a simple linear method that does all the work which is both less likely to have logical errors and easier to reason about if any errors do occur. Through no coincidence, this is also how SwiftUI works.
I’ve written more tests than just about… | justin․searls․co:
... the single most important thing programmers can do to improve their code is to minimize branching (e.g. if statements)...
I’m working on breaking components out of my Swift project into local SPM packages. I’m really starting to like the architecture. But I’m having trouble with testing. I’d like to ⌘-T in Xcode and run both the app unit tests and the SPM tests, but I can’t make it work. All the posts online redirect back to this answer: developer.apple.com/forums/th… But when I try to follow along, the SPM test targets just aren’t in the list. Anyone have a clue?
In general, I’m skeptical about AI code generation, especially in the hands of inexperienced engineers who think it’s smarter than they are. But when you need it to do a menial thing? Fantastic! For a Swift project, I needed a bunch of constants that I found on GitHub in a Python file. I pointed ChatGPT at the URL and asked it to generate the corresponding Swift enum, but change the keys from snake case to camel case. And… it just did it! I think it might even be right.
It’s about a week before WWDC 2023 kicks off in sunny California, so here’s a list of things I’m hoping to see come out of Cupertino. I’m skipping the AR/VR stuff, since it’s been speculated ad nauseam. I’m sure it’ll be cool and awesome and weird and probably not cheap. And maybe it’ll make me dizzy
TLDR; Better Xcode, Better Siri, Better Mac Catalyst
Mostly, I just want Xcode to be better. Don’t crash. Don’t be slow. Is that too much to ask?
We’ve all seen it. Switch branches, hit build and be faced with
Build operations are disabled: Package loading in progress. Please try again later.
I spend a lot of time working on Callisto, which is kind of a big app. We have dependencies split out into frameworks and it can take SPM a while to resolve all the packages when switching branches. So I hit this quite a lot. Usually, Xcode does a great job of queueing up actions when it’s busy. Like if you do a clean build and then run unit tests, it will finish the clean and then do the testing. That’s all I want for SPM resolution.
I haven’t written a lot of SwiftUI. We toyed with it early on building Callisto, but it wasn’t ready for a big project like that. A couple months ago, we thought about trying to build a screen here or there with SwiftUI, but ran into roadblocks. One of the key features of SwiftUI is the live preview. To make those work, Xcode compiles bits of your code behind the scenes and shows them in preview pane. Callisto is a Catalyst app but has an AppKit plugin for doing Mac system things. Xcode could not handle that when trying to make a SwiftUI preview. Xcode would try to compile the AppKit code with UIKit and become very upset when it didn’t work. That left our foray into SwiftUI dead in the water.
I’ve dabbled a bit with ChatGPT as a coding assistant. It’s great for small tasks with fiddly parameters. For instance, I needed to get the timestamp of some files. That’s the kind of thing that’s straightforward, but you don’t do it often, so you have to look up the specific API to stat the file and which attributes correspond to the creation date / modification date, etc. ChatGPT spit that code right out and I could move on to other things. But there’s friction there and the opportunity for a bespoke Xcode experience. I’m cautiously optimistic Apple will do something in this space, but I’m afraid it won’t be groundbreaking. Because, you know… Siri.
Since Callisto is built with Mac Catalyst, these are the sorts of updates we’d love to see as developers.
Dynamic Type on iOS has been around for 10+ years. That’s the bit in iOS Settings that lets you make the text on your device a little bigger or a lot bigger than standard. All the apps that support Dynamic Type will pick up the change and text across the whole device changes size. That’s great! A boon for aging eyes everywhere.
But it isn’t supported at all on macOS. This recent announcement about new accessibility features coming in iOS 17 mentions
For users with low vision, Text Size is now easier to adjust across Mac apps such as Finder, Messages, Mail, Calendar, and Notes.
That sounds a lot like Dynamic Text on the Mac, so fingers crossed.
Back in the day, lots of Mac apps used auxiliary windows that served as things like inspectors and floating tool panels. Apps like Photoshop had a bunch of these – paintbrushes, color panels, etc. You can still see these in AppKit apps like Preview and Quicktime Player, where the inspector is an auxiliary window. When you switch apps, it disappears to cut down on clutter. Right now, there’s nothing like that for Catalyst. The UIScene API for managing windows doesn’t have that level of distinction. Hopefully, at WWDC we’ll see a more mature Stage Manager implementation and at least some way of distinguishing main content windows and helper windows.
Callisto includes an embedded Python distribution for running the Jupyter Python Kernel. On the Mac, we just spawn a separate process and run Python normally. On iPad, we use a heavily modified Python to run in the same process space as the app to comply with App Store requirements. In practice, that hamstrings Callisto on the iPad a great deal. With a renewed interest in Pro apps on the iPad, it’d be incredible if we were allowed to run multiple processes and level the playing field between an M1 iPad and an M1 MacBook.
Yeah, never going to happen.
A lot of folks have called for a dedicated Passwords app instead of burying password management in Settings. I’m all for that, but I’d also love to see password sharing via your Apple Family. That’s really the last feature I’d need to leave 1Password behind. 1Password has been fantastic, but they’ve really pivoted to a corporate focus, so they’re less of good fit when I just want to share the Netflix password with my family. (Only for persons in my immediate family, residing in the same household. Netflix, if you’re reading this, I promise.)
I’m pretty heavily invested in the HomeKit ecosystem and I’m mostly happy with it. (I’m looking at you Siri.) I’ve got several third party apps that will stream video from my cameras, but they can only show live feeds. HKSV (HomeKit Secure Video) records events from these cameras and saves to iCloud, but scrolling back through time is only available via Apple’s Home app. Third party apps aren’t able to access the history of clips because there’s just no API for it. With HomeKit (hopefully) maturing a little more this year, Apple should make that available to developers.
Yes, macOS has Siri. But as underwhelming as Siri is on iOS, it’s worse on Mac.
In my notes, I wrote down “Not Brain Dead”. Siri just fails at the most basic things sometimes. When I try to open an app in my Applications folder with the phrase “Open AppName”, it fails maybe 2 out of 3 times. I’m not sure why it’s so bad and I don’t know of any way to debug it.
I usually use my laptop with the lid closed with external monitor, keyboard, webcam, etc. That means I can’t use “Hey Siri” for security reasons. I agree being able to turn off an always-on microphone is important, but if Siri is a serious feature, why can’t I explicitly grant permission for Siri to listen with an external microphone? It is interesting to note that “Hey Siri” does work with the ($1600) Mac Studio Display. That’s attributed to the onboard A13 chip.
Another bit of low hanging fruit for Siri is menu commands. At least it seems low hanging to me. If I’m using an app like Xcode, that has a menu command called ‘Build’, I would expect Siri to understand if I said “Hey Siri, Build”. But like the double meat burger, Siri is strictly off the menu.
But the real dream is a conversational Siri. We’ve all seen Iron Man. We’ve seen Tony talking to Jarvis like a person as he works. Could Siri ever do that in the context of Xcode? Playing with ChatGPT has teased this kind of reality. “Write a method to delete all the files in a given directory.” That’s a thing ChatGPT can do in a web browser window. Why can’t Siri to do that in Xcode? Imagine an Apple trained LLM that especially knew about Swift, UIKit, and all the Apple technologies. Add in the context of the project you’re currently working on. Talk about a developer accelerant. But that’s a big leap, so it’s doubtful we’ll see that this year, but maybe the first step?
Apple announced Apple Silicon for Mac at WWDC of 2020, three years ago, and the first commercial M1 Macs shipped in November of 2020. At the time, Apple estimated that the transition to Apple Silicon would take about two years. It’s been two and a half years since that first M1 MacBook Air shipped but the Mac Pro still sports an Intel chip. Something has clearly run amok.
But the rumor mill is pretty quiet on the Mac Pro front. Usually at this point, there would be at least some mention of the debut of a high profile, flagship Mac. It seems like we’ll be waiting until later in the year to see what kind of behemoth you can build out of lots of iPhone chips.
Years ago, HomePod ran its own fork of iOS called AudioOS. Apple merged AudioOS with tvOS several revisions back so now both these home devices run tvOS. Because they do similar things – played media and power HomeKit – that makes a lot of sense. But the two devices remained separate at a hardware level – a smart speaker and a TV streamer. I’ve been waiting for the hybrid devices for years now. There’s the HomePod with a screen, like an Amazon Echo Show, that does HomePod things, but augmented with a display. And there’s the AppleTV with a mic and speaker. The mic let’s you talk to the TV without holding the button on the remote and the speaker acts as a center channel for multiple HomePod sound stage.
Those are both consumer products, so not the kind of thing they’d squeeze into the WWDC Keynote, especially with the AR/VR announcement, but maybe this fall, just in time for Christmas.
Will we see any of this at WWDC? I certainly hope so. But there’s only one way to find out.
No sleep ‘til DubDub!
TL;DR -- In SwiftUI, use a fake testing `App` instead of your real `App` to make sure you're actually testing your code.
For unit testing, test coverage is an important metric. How much of your code base are you exercising during unit tests? Xcode is, unfortunately, not too bright when it comes to measuring coverage. It doesn’t have the intelligence to know if you’re “testing” a line of code, just that the line of code was executed. This is a problem right off the bat.
I’ve got a new app project in Xcode. The app does a little and I want to start adding tests before it gets too big. So I add a unit test target and Xcode plops in some empty tests. I run the unit tests, and they all pass since they’re empty. I check the coverage and it’s at 36%! How can that be, I didn’t test anything?
Xcode, the blissful idiot, is really reporting that 36% of the code was executed while running the unit tests. Usually, when an app starts up, it does some bootstrapping. You might set up your persistent storage, draw a couple of Views on screen, and maybe talk to the network. Xcode counts all that as “testing” because it ran during a unit test.
To make the test coverage more accurate, we need to do as little as possible outside of our actual unit tests. Jon Reid has a write-up of how to do this with a UIKit app by swapping out the app delegate during tests. But SwiftUI introduces a whole new startup sequence so we need a new approach for SwiftUI’s new app lifecycle.
After some futzing around, turns out it’s easy!
struct TestApp: App { // 1
var body: some Scene {
WindowGroup {
Text("I'm running tests!")
}
}
}
@main // 2
struct TestDriver {
static func main() {
if NSClassFromString("XCTestCase") != nil { // 3
TestApp.main()
} else {
MyRealApp.main()
}
}
}
There are three key points that make this work:
Now when I run unit tests, my coverage is at 0.8%! That’s more like it. In order to boost my test coverage, I now have to actually test code. And the code coverage metric really starts to mean something.
TLDR – If your unit tests crash with DocumentManager service tried to send a message to a deallocated host proxy, make sure you’re dismissing any presented instance of UIDocumentPickerViewController.
In our Callisto Xcode project, we’ve got a lot of unit tests. Like over a thousand. We’re at the point where if tests have a 0.1% chance of failing, then it happens every time. Our tests need to be really rock solid, or there’s no way we’ll get a clean run which is the only way CI will let a build through.
Some time ago, we started noticing the occasional test failure with an uncaught exception:
DocumentManager service tried to send a message to a deallocated host proxy
It would crop up when running test both locally and in CI. With so many tests, we get the weird edge case now and then, but they’re not worth the time to track down. We ignored it. With the recent update to Xcode 14.3 and macOS 13.3, the DocManager exception went from occasional annoyance to ‘omg, this happens every time I unit test on iOS’. So now we have to fix it.
But what’s a DocManager? IDK – there’s nothing with that name in our code and nothing in the docs about an Apple framework called DocManager. Looking through the traceback, it’s pretty obvious that it’s some kind of internal Apple thing. Surprisingly, a Google search for this error returns absolutely no results. That’s never a good sign.
But what’s a DocManager do? Callisto is UIDocument based, so maybe DocManager manages documents? A bunch of the unit tests open instances of a UIDocument and aren’t 100% about cleaning those up, so maybe some housekeeping will help. Cue the plumbing montage. We added a some bookkeeping to make sure that any open UIDocuments were closed at the end of each test, so now we’re sure there are no dangling UIDocuments. No impact – still the DocManager throws the exception.
This particular problem is a huge pain to track down. Somewhere in the tests, things get into a bad state. Later, while another test is running, some background thread discovers the bad state and throws an exception. The cause of the crash and the actual crashing are quite loosely coupled, making it hard to pin down the culprit. It also means that the offending tests will run just fine by itself, but only cause a crash when a large number of tests run, giving the background issue time to percolate to the surface.
After a couple hours of testing the tests and narrowing down which ones fail, it started to look like UIDocumentPickerViewController might be involved. We’ve got some ViewControllers that open a docPicker and get feedback via the docPicker’s delegate methods. To test those, we programmatically tap a button to cause the dockPicker to be presented, get a handle to that docPicker and call the delegate with the docPicker and some fake results. This works great for testing! Except for that pesky exception that gets thrown now and then.
If we take a closer look at the exception, we’ll see there’s a little more info attached in the userInfo dict, specifically a file and line number, DOCRemoteViewControlller.m line 42. Not that we have access to the source, but the name of the file, “Remote View Controller”, does offer a hint of its purpose. Those docPicker view controllers offer our app an escape hatch out of the sandbox and into the rest of the file system on the device. In my limited understanding, the docPickers interface with a separate process that manages file system access, so they actually represent some state from another (remote) long lived process. In the end, these dangling, un-dismissed docPickers were the root of the problem. Make sure all your UIDocumentPickerViewControllers are dismissed properly and the problem goes away.
Whew!