Category Archives: Swift

Dangerous Logging in Swift

I recently came across a perplexing crash in my unit tests that led me down a path of discovery, culminating in an awareness that “I was holding it wrong” when it comes to using NSLog in Swift.

Here is the source code to an extension on FileManager intended to make it easy to read the last modification date of a file:

@objc(rsLastModificationDateAtURL:)
func lastModifedDate(at url: URL) -> Date? {
    do {
        let resourceValues = try url.resourceValues(forKeys: [.contentModificationDateKey])
        return resourceValues.contentModificationDate
    } catch {
        NSLog("Failed to get modification date at URL: \(url) - \(error)")
        return nil
    }
}

I’ve highlighted the line of interest, an invocation of NSLog in the case where a modification date could not be read because of some exception. For example, if the file were not found this line of code would be reached. Looks innocent enough, right? Here’s what I came up against while running the unit tests for one of my apps:

highlighted source code line showing NSLog invocation crashing with EXC_BAD_ACCESS

At first I thought that one of the url or the error variables must be bad. After all, Swift string interpolation and NSLog are both battle-tested. Surely if there were a way to easily crash by logging good variables, it would be fixed a long time ago. But no, both of the variables in this instance were perfectly valid references. So what gives?

The Problem

The problem is rooted in the fact that NSLog from Swift calls through to the underlying C-based interface which supports template strings and variadic arguments. This is why, for example, you can invoke NSLog(@"Hi %@", name) in Objective-C and, if name represents the string “Daniel”, you get “Hi Daniel” printed to the console.

In the scenario of my crash, the interpolation of “url” and “error” results in surprise template placeholders, because of the presence of spaces in the path that the URL represents. In my tests, a file called “Open Program Ideas File” exists and is used to test some file manipulations. When the path to this file is encoded as a URL, the name of the file becomes “Open%20Program%20Ideas%20File”. Since the error’s failure message reiterates the name of the file it couldn’t find, we end up with another copy of the percent-escaped string in the parameter to NSLog. Each instance of “%20” followed by a letter is liable to be interpreted by NSLog as a printf-style format specifier. So as far as our filename’s string is concerned, we have “%20P”, “%20I”, and “%20F” threatening to cause trouble. Capital letters in printf formats are pretty uncommon, but the %F format specifier is documented as:

64-bit floating-point number (double), printed in decimal notation

In short, we have a situation where logging the URL and the error inadvertently asks NSLog to look for a 64-bit floating point number in the arguments list. A 64-bit floating point number of size 20, whatever that means in this case. So, whether the unit test, or app, crashes or not in a situation like this depends entirely on what data happen to be in the places where the variadic arguments would be, and whether that data causes a crash when attempting to interpret it as a type of the specified print format.

How to Fix It

So how do we fix it? Well, we need to make sure that the string that ultimately gets passed to NSLog doesn’t contain these surprise placeholder values. If this were Objective-C, we wouldn’t run into the problem because the parameters would need to be passed as variadic arguments:

NSLog(@"Failed to get modification date: %@ - %@", url, error);

But if we try that in Swift, we run into trouble. NSLog in Swift doesn’t support variadic arguments:

screenshot of build error indicating that variadic arguments are not allowed with NSLog in Swift

I’m pretty sure this is how I ended up in this situation in the first place: I probably tried to adapt my old Objective-C code to Swift, ran into this error, and obediently changed to using inline variable substitution instead. It just seemed like “the Swift thing to do.” But if we can’t use NSLog with inline substitution, and we can’t use variadic arguments, what are we supposed to do? Update: Thanks to Sebastian Celis for pointing out on Twitter that variadic arguments do work with NSLog. I think the failure above is specifically because I’m passing non-NSObjects to NSLog, which isn’t supported.

A few years ago Apple introduced a new logging framework called the Unified Logging System, which supports logging with variadic arguments from both Swift and Objective-C. It also supports a host of other features that allow you to categorize and prioritize your log message and, well, it sounds great, but it’s also quite a bit more complicated than just invoking NSLog. It’s complicated enough that I can’t offer a one-line fix for the situation I’ve described here, but I encourage you to investigate your options.

Audit Your Code

I also encourage you to audit your own code, and I’ll describe a method for searching your own code base to see if you might be vulnerable to the kind of unpredictable crash that inline interpolation can cause with NSLog. In Xcode, create a custom search scope that allows you to easily search just the Swift files in your code base, then search for instances of the problematic pattern. Obviously if you’ve got a 100%-Swift code base, creating the custom search scope is not necessary:

  1. Show the Find Navigator by selecting View -> Navigator -> Find from the menu bar.
  2. Click the “In Project” (by default) search scope label beneath the search field. This reveals the list of search scopes currently configured in your workspace.
  3. Click the “New Scope” button and configure a scope designed to match all files with the “swift” file extension”:

    screenshot of custom search scope popover showing title

  4. Click the Find type (“Text” by default) in the search parameters above the search field, and change it to “Regular Expression”.
  5. Type or paste in the search string NSLog\(.*\\\( — this is a regular expression that will find intances of the string “NSLog(” followed by any number of characters that includes the substring \(, the tell-tale sign of Swift string interpolation:

    screenshot of search results showing multiple files with

In my own code base, this audit results in an alarming 23 instances in 17 files, all of which I will soon be converting to a safer logging method. I encourage you to do the same in your code base. Hopefully you’ll have been followingn a safer pattern all along and won’t have any work to do, but if you’re anything like me, you might find you’re more vulnerable to the problem than you thought!

Completely Useful

Particularly as a developer coming from an Objective C background, it’s common to head into an override implementation in a subclass thinking of a property as a function. For this reason, I find myself commonly starting to implement a property override by typing something like “override func canBecome” when I want to override UIView’s “canBecomeFirstResponder”. A completion pops up like this:

Screenshot of Xcode offering a code completion with non-pertinent options.

Instead of the expected method, Xcode has twisted itself into knots to contrive that letters from the whole prototype for “addObserver” can be used to spell out “canBecome”. Not particularly useful.

It always takes me a few seconds, or longer, to realize that I haven’t misremembered the existence of a method or property that I want to override, I’ve simply disregarded whether it’s identified in Swift as a “var” or a “func”. One simple change to “override var canBecome” does the trick:

Screenshot of Xcode offering a useful code completion popup list.

In my not so very humble at all opinion, Xcode should go the extra mile here and look for matches in the “var” realm when you’ve type “func”, and vice-versa. If it finds something that perfectly matches everything you’ve typed except whether it’s a func or a var, then it should offer the completion and fix the mistyped designation when you accept it. I’ve filed Radar #42660012 requesting this enhancement.

Update: Several folks have suggested a good workaround: simply get in the habit of omitting the decorative terms, such as “override”, “func”, and “var”. It turns out if you leave out the narrowing term, Xcode does find the pertinent matches, and also inserts the required decoration upon inserting one.

My friend Chris Liscio pointed out that changing the behavior as I recommend might be a questionable move, because it defies a kind of rule for type-completion. In general, whatever the user has already typed should be considered what the user knows they want, and completion should only offer pertinent matches. It’s a compelling argument but I think that in cases like this the computer actually knows more about what I really want than I do.

Icon for File with UIKit

There’s a general consensus among many Mac and (mostly) iOS developers, that AppKit is “old and busted” and UIKit is “new and refined.” I am still fairly limited in my experience with UIKit, but in many ways I agree that in developing the framework, they left some of the more annoying baggage of AppKit behind. However, they also left off many conveniences that AppKit developers like myself have come to rely upon.

One of the easiest things in the world in AppKit is asking the framework for an image to represent a file’s icon. Say you’ve got a Swift source file handy and you want to show it in your app with an appropriate icon. It’s an easy one-liner on the Mac:

let swiftIcon = NSWorkspace.shared.icon(forFile: "/tmp/Test.swift")

Screenshot of Xcode showing code to request an icon from NSWorkspace.

On UIKit, as far as I can tell, it takes a bit more work. There is no UIWorkspace, which is probably fine, but there is also no UIImage.iconForFile, or similar method to make this quite as straightforward as it is on the Mac.

I came across “UIDocumentInteractionController” which seems to be the key to easily obtaining icons for arbitrary image types on iOS. After initializing it with a file URL, you can ask it for an array of icons, which it will reveal in order from smallest to largest. Tying this all together in an extension on UIImage, you could add a pretty handy helper method to your collection of code (sorry, one of these days I’ll get syntax highlighting working on this blog):

import UIKit

extension UIImage {
	public enum FileIconSize {
		case smallest
		case largest
	}

	public class func icon(forFileURL fileURL: URL, preferredSize: FileIconSize = .smallest) -> UIImage {
		let myInteractionController = UIDocumentInteractionController(url: fileURL)
		let allIcons = myInteractionController.icons

		// allIcons is guaranteed to have at least one image
		switch preferredSize {
		case .smallest: return allIcons.first!
		case .largest: return allIcons.last!
		}
	}
}

Now I’ve got basically the same functionality I had on AppKit:

Screenshot of invoking the helper method from code snippet above.

This is cool enough, but what’s even cooler is the UIKit method works on a URL irrespective of whether the underlying file actually exists! This, combined with the fact that a URL can be initialized with just about any String, means we can expand our icon utilities with a method for obtaining icons by file name alone:

extension UIImage {
	public class func icon(forFileNamed fileName: String, preferredSize: FileIconSize = .smallest) -> UIImage {
		return icon(forFileURL: URL(fileURLWithPath: fileName), preferredSize: preferredSize)
	}
}

Here we use it to get the icon for an arbitrary Pages file which may or may not exist:

Screenshot of calling the code snippet above to obtain a file icon for a Pages document.

Ah, now this is starting to feel even cooler than AppKit. But wait, NSWorkspace also supports another handy method:

let zipIcon = NSWorkspace.shared.icon(forFileType: "zip")

Well, we can achieve similar by taking advantage of the icon(forFileNamed:) method:

extension UIImage {
	public class func icon(forPathExtension pathExtension: String, preferredSize: FileIconSize = .smallest) -> UIImage {
		let baseName = "Generic"
		let fileName = (baseName as NSString).appendingPathExtension(pathExtension) ?? baseName
		return icon(forFileNamed: fileName, preferredSize: preferredSize)
	}
}

Now we’re really catching up:

Screenshot of code calling the iconForPathExtension example above.

But AppKit’s equivalent actually works on UTI type strings, as well. Wouldn’t that be cool to support on iOS? Seeing as MobileCoreServices gives us access to low level functions for converting UTI and file extensions, we can take a UTI, convert it, and let our existing helper methods take it from there:

import MobileCoreServices

extension FileManager {
	public func fileExtension(forUTI utiString: String) -> String? {
		guard
			let cfFileExtension = UTTypeCopyPreferredTagWithClass(utiString as CFString, kUTTagClassFilenameExtension)?.takeRetainedValue() else
		{
			return nil
		}

		return cfFileExtension as String
	}
}

extension UIImage {
	public class func icon(forUTI utiString: String, preferredSize: FileIconSize = .smallest) -> UIImage? {
		guard let fileExtension = FileManager.default.fileExtension(forUTI: utiString) else {
			return nil
		}
		return icon(forPathExtension: fileExtension, preferredSize: preferredSize)
	}
}

And voila!

Screenshot of code calling the iconForUTI code above.

It would be fairly easy to further extend this to support creating icon images from MIME types. Hope this helps some of you folks, especially coming from the Mac, who expected getting icons for file types to be slightly easier than it is.

Internal Typealias Promotion

In some scenarios it might be useful to declare a typealias internally to a module, to make it easier to implement the functionality of the module itself, but less useful to export that typealias to clients of the module. For example, consider an image manipulation framework that can work with NSImage or UIImage instances, depending on the platform. Internally to the module, I might define:

#if os(macOS)
	typealias RSPlatformNativeImage = NSImage
#else
	typealias RSPlatformNativeImage = UIImage
#endif

Then I can implement methods, including public facing methods like:

public func monochromeImage(fromImage image: RSPlatformNativeImage) -> RSPlatformNativeImage { ... }

Since I haven’t marked my typealiases as ‘public’, they won’t be exported to clients, but the above will also fail to compile. Swift requires that public methods work only with public types. This makes sense, because if the types aren’t public, how are clients expected to be able to work with them?

But if I mark the typealiases public, I impose a new type “RSPlatformNativeImage” on clients, when as far as they are concerned, this method operates on either an NSImage or UIImage. They might quickly get the idea that RSPlatformNativeImage is just a typealias, but it’s a bit of unwanted clutter on the public-facing API.

Obviously I can solve this by adding more platform-specific directives to the module so that whole functions are declared as working with either NSImage or UIImage, but it would be nice if Swift would help me out here. Instead of giving a compiler error, Swift could simply export the method using the public type that the typealias resolves to. In which case a client of the module for iOS would see the method as:

public func monochromeImage(fromImage image: UIImage) -> UIImage { ... }

And for Mac:

public func monochromeImage(fromImage image: NSImage) -> NSImage { ... }

Handling internal typealiases like this would cause the behavior for Swift clients to match what is already being done for Objective-C clients. The generated Module-Swift.h for this method in a Mac project is:

- (NSImage * _Nonnull) monochromeImageFromImage:(NSImage * _Nonnull)image SWIFT_WARN_UNUSED_RESULT;

Thus for Objective-C clients the clutter of the typelias definition is tidied away, but for Swift clients, it must still be dealt with. I filed a bug requesting this behavior in the Swift bug tracking system.

Getting a CFNumber’s Value in Swift

Recently, as a consequence of working with the CGImageSource API, I found myself in a situation where I had hold of a CFNumber and wanted to get its value, as a CGFloat, in Swift.

CFNumber wraps numeric values in such a way that, to get the value out, you have to specify both the desired type, and provide a pointer to the memory of the variable that will hold the value. This kind of direct memory manipulation is not particularly suited to Swift’s priorities for type safety and memory protection. Here’s the API I’d need to use in Swift:

func CFNumberGetValue(_ number: CFNumber!, 
                    _ theType: CFNumberType, 
                    _ valuePtr: UnsafeMutableRawPointer!) -> Bool

The first two parameters are straightforward, but whenever I see types like “UnsafeMutableRawPointer” in Swift, my brain melts down a little. I have never really sat down to truly understand the nuanced differences between these types, so I usually just try something and hope it works. Here I am hoping for a gift from Swift’s implicit bridging:

// myCFNumber is 30.5
var myFloat: CGFloat = 0
CFNumberGetValue(myCFNumber, .floatType, &myFloat)
print(myFloat) // "5.46688490824244e-315\n"

Welp. That didn’t work. Let’s see if we can refresh our memory about UnsafeMutableRawPointer. In the section titled “Raw, Unitialized Memory” or I read:

You can use methods like initializeMemory(as:from:) and moveInitializeMemory(as:from:count:) to bind raw memory to a type and initialize it with a value or series of values.

Oh jeez, am I really going to have to manually create an UnsafeMutableRawPointer? I’ll try anything:

var myFloat: CGFloat = 0
var myFloatPointer = UnsafeMutableRawPointer(mutating: &myFloat)
CFNumberGetValue(myCFNumber, .floatType, myFloatPointer)
print(myFloat) // "5.46688490824244e-315\n"

Alas, same problem. Surely somebody has figured this out? I try Googling for “CFNumberGetValue Swift GitHub” and find a promising result from an authoritative source. The Swift standard library itself!

var value: Float = 0
CFNumberGetValue(_cfObject, kCFNumberFloatType, &value)

Aha! Practically the same thing I was doing, except for one nuanced detail: the var value is declared as a Float instead of a CGFloat. But wait a minute, what file is this implementation in? NSNumber.swift? Oh, right. NSNumber and CFNumber are toll-free bridged, and Swift’s standard library fulfills that promise too:

let myFloat = (myCFNumber as NSNumber).floatValue
print(myFloat) // 30.5

In fact, Swift’s Float type is even cozier with CFNumber than I expected. What started as a confused mission to make use of CFNumberGetValue and its unsafe pointer argument culminated in a bit of sample code from GitHub that ultimately led me to the understanding that the way to get a CFNumber’s value in Swift is … simply to ask for it:

let myFloat = Float(myCFNumber)
print(myFloat) // 30.5

Helpless Help Menu

I was alerted by Christian Tietze of a pretty bad usability bug in macOS High Sierra. If you are running a Mac app, click the “Help” menu, and then dismiss it, whatever UI element you were focused on in the app loses its focus and does not regain it after dismissing the menu.

The problem is so bad that tabbing, clicking other UI elements, even switching to another app and back does not restore focus on the window’s responders. If the focus was on an NSTextView, such as the editor in MarsEdit, then the blinking cursor continues to animate, but keystrokes are ignored and simply cause the app to beep.

Christian filed a bug, and shared a workaround: set the delegate of the Help menu to your app’s delegate, and listen for the “menuDidClose” delegate method. If it’s the Help menu, restore focus manually.

I generalized this workaround to an approach that should work for whatever window, and whatever responder is currently focused when the Help menu is opened. By saving the window and the responder at “menuWillOpen” time, it can be precisely restored afterwards:

private weak var lastKeyWindow: NSWindow? = nil
private weak var lastResponder: NSResponder? = nil

func menuWillOpen(_ menu: NSMenu) {
   if menu == NSApp.helpMenu {
      if let activeWindow = NSApp.keyWindow {
         self.lastKeyWindow = activeWindow

         if let activeResponder = activeWindow.firstResponder {
            self.lastResponder = activeResponder
         }
      }

      // If the responder is a field editor, then save 
      // the delegate, which is e.g. the NSTextField being edited,
      // rather than the ephemeral NSTextView which will be 
      // removed when editing stops.
      if let textView = lastResponder as? NSTextView,
         textView.isFieldEditor {
         if let realTarget = textView.delegate as? NSResponder {
            lastResponder = realTarget
         }
      }
   }
}

func menuDidClose(_ menu: NSMenu) {
   if menu == NSApp.helpMenu {
      if doWorkaround {
         if let actualKeyWindow = self.lastKeyWindow {
            actualKeyWindow.makeKeyAndOrderFront(nil)
            actualKeyWindow.makeFirstResponder(self.lastResponder)
         }
      }
   }
}

Note that I didn’t clear out the weak var references to the lastKeyWindow and lastResponder. The reason is because part of the bug here involved NSMenu’s menuWillOpen and menuDidClose getting called more often than they probably should be. It’s probably the root issues that is causing the Help menu to excessively take control of the key window. It turns out we are going to get called on menuDidClose twice, so we need to be sure the desired target window and responder are still available the second time around.

As Christian points out, the workaround fixes the worst aspect of the bug: locking up the UI so that typing is ignored, but the focus ring around the target text field doesn’t always get redrawn as expected. My theory is that the focus ring animation is in the process of drawing when the second “menu will open” event is generated, causing the Help menu to reactive itself. The field being reactivate again very shortly after somehow doesn’t trigger the need to redraw the focus ring as you might expect.

I filed an additional bug, Radar #39436005, including a sample project that demonstrates both the bug and the workaround. Until Apple fixes this, Mac developers may want to implement a workaround along the lines demonstrated here. Given the horrible user experience associated with this bug, hopefully Apple will fix it promptly!

Optional Emptiness

Objective-C developers are comfortable with many idioms that fall out of the safety of messaging nil. For example, consider a chunk of Objective-C code that tests the “emptiness” of a UITextField string:

if (myTextField.text.length == 0) {
    // Do something
}

If myTextField.text is nil, what happens? In Objective-C, a message sent to nil will return nil, or zero, depending on the return type of the message. In this case “length” returns an integer, which happens to be zero when text is an empty string, and zero when text is nil. So this code block perfectly expresses “if the text is empty, do something.”

Adapting this code to Swift, you immediately run up against the language’s strict handling of optionals. Because myTextField.text might be nil, it has to be unwrapped before the length method can be called. This leads to less terse code such as:

if myTextField.text?.isEmpty != false {
    // Do something
}

This works! But it’s harder to read, and harder to reason. Similarly to the way the Objective-C version requires deep understanding of that language’s nil-messaging behavior, the Swift version requires deep understanding of optional chaining and comparison of optional and non-optional values. Here’s another example:

if (myTextField.text ?? "").isEmpty {
    // Do something
}

This is much easier to understand: use the non-nil String from myTextField, or else a constant string that is guaranteed to return true for isEmpty. It’s still more cumbersome than the original Objective-C, though.

In my own Swift adventures, I’ve addressed this using Swift’s powerful extension mechanism. It turns out that in Swift, any type that conforms to the “Collection” protocol implements an “isEmpty” method. String is one of these types. So with a small extension, we can add the “isEmpty” method not only to String? but to all optionals that wrap a collection:

extension Optional where Wrapped: Collection {
	public var isEmpty: Bool {
		switch(self) {
		case .none:
			return true
		case .some(let concreteSelf):
			return concreteSelf.isEmpty
		}
	}
}

With this extension in place, our test becomes:

if myTextField.text.isEmpty {
    // Do something
}

Which is both highly readable, behaves correctly when “text” is nil, and doesn’t require any deep language understanding to comprehend.

Thanks to Hwee-Boon Yar for the Objective-C scenario that motivated this post, and to Michel Fortin for putting forward the Swift equivalents cited above. This question came up in the Core Intuition Slack, where interesting discussions like this often take place. Join us!

Swift Integration Traps

In the nearly four years since Swift was announced at WWDC 2014, Mac and iOS developers have embraced the language with decreasing reluctance. As language features evolve, syntax stabilizes, and tooling improves, it’s easier than ever to leap into full-fledged Swift development.

Several months ago, I myself made this leap. Although the vast majority of my Mac source base consists of Objective-C files, I have enjoyed adding new source files in Swift, and even converting key files to Swift either as an exercise, or when I think I will gain specific advantages.

One remaining challenge in Swift is the lack of ABI stability. In layperson’s terms: the lack of ABI stability prevents compiled Swift code from one version of the Swift compiler and runtime from linking with and running in tandem with Swift code compiled for another version.

For most developers, this limitation simply means that the entire Swift standard library, along with glue libraries for linking to system frameworks, needs to be bundled with the application that is built with Swift. Although it’s a nuisance that several megabytes of libraries must be added to every single Swift app, in the big scheme of things, it’s not a big deal.

A worse consequence is the number of pitfalls that ABI instability present, that are difficult to understand intuitively, and in many cases impossible, or at least dangerous, to work around. These pitfalls lie mainly in areas where developer code is executed on behalf of a system service, in a system process. In this context, it is not possible for developers to ensure that the required version of Swift libraries will be available to support their code. Game over.

On the Mac, system integration plugins are a typical scenario for this problem. While iOS has evolved with a strong architecture for running developer code in standalone, sandboxed processes, on the Mac there are still many plugins that run in a shared system process alongside code from other developers. These plugins run the gamut from arcane, rarely used functionality, to very common, user-facing features where a plugin is effectively required in order to satisfy the platform behaviors prescribed by Apple and expected by end-users.

One example on the more arcane, or at least inessential, end of the spectrum, is the Screen Saver plugin interface. Create a new project in Xcode, and choose the “Screen Saver Plugin” template as your starting point. Notice how unlike most templates, Xcode doesn’t even offer a choice of language. Your source files will be Objective-C. At least they’re giving you a hint here.

On the more mainstream end of the spectrum are plugins such as System Preferences panels and QuickLook Plugins. Depending on the type of app you are developing, it may be essential, or at least very well-advised to implement one of these types of plugins. So what do you do if you have an existing Objective-C app that you want to port to Swift, or you are writing a Swift app from scratch, and need to support one of these plugin formats? In the case of System Preferences panels at least, you have a couple practical options:

  1. Implement the plugin code, and all supporting code in Objective-C.
  2. Move the functionality out of System Preferences and into the host app.

Each of these could be somewhat reasonable approaches for a System Preferences plugin. The content of these plugins is often fairly straightforward, standard UI, and the goal is usually to collect configuration data to convey to the host application. It’s also not unreasonable, and may even be preferable to move such configuration code out of System Preferences and into a native panel inside the host app.

QuickLook Plugins are another beast. Because the goal of a QuickLook Plugin is usually to convey a visual depiction of a native document type, it’s exceedingly common to take advantage of the very classes that present the document natively in the host app. Let’s say you’ve written an app in Swift, FancyGraphMaker. Apple encourages you to implement a QuickLook Plugin so that users will be able preview the appearance of your fancy graphs, both in the dedicated QuickLook interface, and by way of more unique looking icons in the Finder.

But once you’ve written the code to draw those fancy graphs in Swift, you’re locked out of using that code from a QuickLook Plugin. Worse? Finishing touches such as supporting Quick Look are liable to come later in the development of an app, so you’ve probably gone through the decision-making process of writing your app in Swift, before realizing that the decision effectively cuts you off from a key system feature. That’s a Swift Integration Trap.

Although the workarounds are not as straight-forward in this scenario as they are for a System Preferences pane, it is probably still technically possible to leverage Swift code in the implementation of a QuickLook Plugin. I have not tested this, but I imagine such a plugin could spawn an XPC process that is itself implemented in Swift and executes the bulk of the preview-generation work on behalf of the system-encumbered plugin code. The XPC process would be free to link to whatever bundled Swift libraries it requires, generate the desired preview data, and message it back to the host process. At least, I think that would work.

But I shouldn’t have to think that hard to get this to work, nor should any other developer. The problem with these Swift Integration Traps is twofold:

  1. If you don’t know about them, you end up stuck, potentially regretting the decision to move to Swift.
  2. If you do know about them, you might put off adopting Swift completely, or at least put off converting classes that are pertinent to QuickLook preview generation.

Each of these consequences is bad for developers, for users, and for Apple. Developers face a trickier decision process about whether to move to Swift, users face potential integration shortcomings for Swift-based apps, and Apple suffers either reduced adoption of Swift, reduced integration with system services, or both.

I filed Radar #38792518 requesting that QuickLook Plugins be supported by the App Extension model. Essentially, this would formalize the process of putting the generation code in a separate XPC process, as I speculated above would work around the problem. The App Extension system is designed to support, and in fact requires this approach. The faster Apple moves QuickLook Plugins, and other shared-process plugins to the App Extension model, the fast developers can embrace Swift with full knowledge that their efforts to integrate with the system will not be stymied.

Update: Thanks to a hint from Chris Liscio, I have learned that Apple has in fact made some progress on the QuickLook front, but it won’t help the vast majority of cases in which a QuickLook Plugin is used to provide previews for custom file types. It took me a while to hunt this down because it not very clearly documented, and Google searches do not lead to information about it.

At WWDC 2017, Apple announced support for a new QuickLook Preview Extension. It escaped my notice even while ardently searching for evidence of such a beast, because the news was shared in the What’s New in Core Spotlight session. Making matters worse, the term “QuickLook” does not appear once in the session transcript, although it turns out that “Quick Look” appears many times:

Core Spotlight is also coming to macOS and just like on iOS you can customize your preview. On macOS a preview is shown when you select a search result in the Spotlight window. Here you really do want to implement a Quick Look preview extension for your Core Spotlight item because Spotlight on macOS does not have a default preview.

Ooh, this sounds exciting! I’ve wondered over the years why such similar plugins, Spotlight importers, and QuickLook generators, shouldn’t be unified. Although the WWDC presentation emphasizes substantial parity in behavior for QuickLook Previews between iOS and macOS, there is a major gotcha:

Core Spotlight is great for databases and shoeboxes where your app has full control over the contents.
It’s not for items that the user monitors in the finder, for that the classic Spotlight API still exists and still works great.

I beg to differ with that “still works great” assessment, at least in the context of this post. Mac developers who want to integrate with QuickLook must still use a shared-process plugin. It’s still a Swift Integration Trap.

Xcode’s Secret Performance Tests

I was inspired today, by a question from another developer, to dig into Xcode’s performance testing. This developer had observed that XCTestCase exposes a property, defaultPerformanceMetrics, whose documentation strongly suggests can be used to add additional performance metrics:

This method returns XCTPerformanceMetric_WallClockTime by default. Subclasses of XCTestCase can override this method to change the behavior of measureBlock:.

If you’re not already familiar, the basic approach to using Xcode’s performance testing infrastructure is you add unit tests to your project that wrap code with instructions to measure performance. From the default unit test template:

func testPerformanceExample() {
	// This is an example of a performance test case.
	self.measure {
		// Put the code you want to measure the time of here.
	}
}

Depending on the application under test, one can imagine all manner of interesting things that might be useful to tabulate during the course of a critical length of code. As mentioned in the documentation, “Wall Clock Time” is the default performance metric. But what else can be measured?

Nothing.

At least, according to any header files, documentation, WWDC presentations, or blunt Googling that I have encountered. There is exactly one publicly documented Xcode performance testing metric, and it’s XCTPerformanceMetric_WallClockTime.

I was curious whether supporting additional, custom performance metrics might be possible but under-documented. To test this theory, I added “beansCounted” to the list of performance metrics returned from my XCTestCase subclass. For some reason I couldn’t get Swift to accept the XCTPerformanceMetric pseudo-type, but it allowed me to override as returning an array of String:

override static func defaultPerformanceMetrics() -> [String] {
	return ["beansCounted"]
}

When I build and test, this fails with a runtime exception “Unknown metric: beansCounted”. The location of an exception like this is a great clue about where to go hunting for information about whether an uknown metric can be made into a known one! If there’s a trick to implementing support for my custom “beansCounted” metric, the answer lies in the method XCTestCase’s “measureMetrics(_: automaticallyStartMeasuring: forBlock:)”, which is where the exception was thrown.

By setting a breakpoint on this method and stepping through the assembly in Xcode, I can watch as the logic unfolds. To simplify what happens: first, a list of allowable metrics is computed, and then the list of desired metrics is iterated. If any metric is not in the list? Bzzt! Throw an exception.

I determined that things are relatively hardcoded such that it’s not trivial to add support for a new metric. I was hoping I could implement some magic methods in my test case, like “startMeasuring_beansCounted” and “stopMeasuring_beansCounted”

but that doesn’t appear to be the case. The performance metrics are supported internally by a private Apple class called XCTPerformanceMetric, and the list of allowable metrics is derived from a few metrics hardcoded in the “measureMetrics…” method:

  • “com.apple.XCTPerformanceMetric_WallClockTime”
  • “com.apple.XCTPerformanceMetric_UserTime”
  • “com.apple.XCTPerformanceMetric_RunTime”
  • “com.apple.XCTPerformanceMetric_SystemTime”

As well as a bunch of others exposed by a private “knownMemoryMetrics” method:

  • “com.apple.XCTPerformanceMetric_TransientVMAllocationsKilobytes”
  • “com.apple.XCTPerformanceMetric_TemporaryHeapAllocationsKilobytes”
  • “com.apple.XCTPerformanceMetric_HighWaterMarkForVMAllocations”
  • “com.apple.XCTPerformanceMetric_TotalHeapAllocationsKilobytes”
  • “com.apple.XCTPerformanceMetric_PersistentVMAllocations”
  • “com.apple.XCTPerformanceMetric_PersistentHeapAllocations”
  • “com.apple.XCTPerformanceMetric_TransientHeapAllocationsKilobytes”
  • “com.apple.XCTPerformanceMetric_PersistentHeapAllocationsNodes”
  • “com.apple.XCTPerformanceMetric_HighWaterMarkForHeapAllocations”
  • “com.apple.XCTPerformanceMetric_TransientHeapAllocationsNodes”

How interesting! There are a lot more metrics defined than the single “wall clock time” exposed by Apple. So, should we use them? Official answer: no way! This is private, unsupported stuff, and can’t be relied upon. Punkass Daniel Jalkut answer? Why not? They’re your tests, and your the only one who will get hurt if they suddenly stop working. In my opinion taking advantage of private, undocumented system behavior for private, internal gain is much different than shipping public software that relies upon such undocumented behaviors.

I modified my unit test subclass to return a custom array of tests based on the discoveries above, just to test a few:

override static func defaultPerformanceMetrics() -> [String] {
	return [XCTPerformanceMetric_WallClockTime, "com.apple.XCTPerformanceMetric_TransientHeapAllocationsKilobytes", "com.apple.XCTPerformanceMetric_PersistentVMAllocations", "com.apple.XCTPerformanceMetric_UserTime"]
}

The tests build and run with no exception. That’s a good sign! But these “secret peformance tests” are only useful if they can be observed and tracked the way the wall clock time can be. How does Xcode hold up? I made my demonstration test purposefully impactful on some metrics:

func testPerformanceExample() {
	self.measure {
		for _ in 1..<100 {
			print("wasting time")
		}
		let _ = malloc(3000)
	}
}

Now when I build and test, look what shows up in the Test navigator’s editor pane:

Screenshot of performance metrics after reducing the size of allocations and length of run.

Look at all those extra columns! And if I click the “Set Baselines…” button, then tweak my function to make it substantially less performant:

func testPerformanceExample() {
	self.measure {
		for _ in 1..<10000 {
			print("wasting time")
		}
		let _ = malloc(300000)
	}
}

Now the columns have noticably larger numbers:

Screenshot of Xcode's test results after running tests with

But more importantly, the test fails:

Screenshot of test errors generated by failing to meet performance baselines.

I already mentioned that by any official standard, you should not take advantage of these secret metrics. They are clearly not supported by Apple, may be inaccurate or have bugs, and could outright stop working at any time. I also said that, in my humble opinion, you should feel free to use them if you can take advantage of them. The fact that they are supported so well in Xcode probably implies that groups internal to Apple are using them and benefiting from them. Your mileage may vary.

The only rule is this: if Apple does do anything to change their behavior, or you otherwise ruin your day by deciding to play with them, you shouldn’t blame Apple, and you can’t blame me!

Enjoy.

Accessible Frames

I love the macOS system-wide dictionary lookup feature. If you’re not familiar with this, just hold down the Control, Command, and D keys while hovering with the mouse cursor over a word, and the definition appears. What’s amazing about this feature is it works virtually everwyhere on the system, even in apps where the developers made no special effort to support it.

Occasionally, while solving a crossword puzzle in my own Black Ink app, I use this functionality to shed some light on one of the words in a clue. As alluded to above, it “just works” without any special work on my part:

Screen shot of Black Ink's interface, with a highlighted word being defined by macOS system-wide dictionary lookup.

Look carefully at the screenshot above, and you can see that although the dictionary definition for “Smidgen” appears, it seems to have highlighted the word in the wrong location. What’s going on here?

The view in Black Ink that displays the word is a custom NSTextField subclass called RSAutosizingTextField. It employs a custom NSTextFieldCell subclass in order to automatically adjust the font size and display position to best “fill up” the available space. In short: it behaves a lot like UITextField on iOS.

To achieve this behavior, one of the things my custom cell does is override NSTextFieldCell’s drawingRect(forBounds:). This allows me to take advantage of most of NSTextField’s default drawing behavior, but to nudge things a bit as I see fit. It’s this mix of customized and default behaviors that leads to the drawing bug seen above. I’ve overridden the drawing location, but haven’t done anything to override the content hierarchy as it’s reflected by the accessibility frameworks.

What do the accessibility frameworks have to do with macOS system-wide dictionary lookup? A lot. Apparently it’s the “accessibilityFrame” property that dictates not only where the dictionary lookup’s highlighting will be drawn, but also whether the mouse will even be considered “on top of” a visible word or not. So in the screenshot above, if the mouse is hovering over the lower half of the word “Smidgen”, then the dictionary lookup doesn’t even work.

The fix is to add an override to NSTextField’s accessibilityFrame method:

public override func accessibilityFrame() -> NSRect {
	let parentFrame = super.accessibilityFrame()
	guard let autosizingCell = self.cell as? RSAutosizingTextFieldCell else { return parentFrame }

	let horizontalOffset = autosizingCell.horizontalAdjustment(for: parentFrame)

	// If we're flipped (likely for NSTextField, then the adjustments will be inverted from what we
	// want them to be for the screen coordinates this method returns.
	let flippedMultiplier = self.isFlipped ? -1.0 as CGFloat : 1.0 as CGFloat
	let verticalOffset = flippedMultiplier * autosizingCell.verticalAdjustment(for: parentFrame)

	return NSMakeRect(parentFrame.origin.x + horizontalOffset, parentFrame.origin.y + verticalOffset, parentFrame.width, parentFrame.height)
}

Effectively I take the default accessibility frame, and nudge it by the same amount that my custom autosizing text cell is nudging the content during drawing. The result is only subtly difference, but makes a nice visual refinement, and a big improvement to usability:

Screen shot of dictionary lookup UI with properly aligned word focus.

I thought this was an interesting example of the accessibility frameworks being leveraged to provide a service that benefits a very wide spectrum of Mac users. There’s a conventional wisdom about accessibility that emphasizing the accessibility of apps will make the app more usable specifically for users who take advantage of screen readers and other accommodations, but more generally for everybody who uses the app. This is a pretty powerful example of that being the case!