How (Not) to Write an iOS SDK | Hacker News Reader

How (Not) to Write an iOS SDK

realm.io

112 pointsastigsen35 comments

I made it about 1/3rd of the way through this. Supporting all this guy's demands is probably much more time consuming than writing whatever the library originally intended to do.

Seems quite unreasonable to me. Perhaps a better way to write this would be describing how iOS's SDKs have too many variations in deployment to make them truly drop-in for everyone else.

I definitely laid out what I consider the "gold standard". Most open source projects hit some, but not all of the items on the list, for various reasons – and that's okay. It's certainly time consuming to work on some of these items, and the cost/benefit of each should be evaluated individually.

But there are quite a few developers that work on SDKs as a full time job, so hopefully they'll pick up on some of the items in the talk :^)

(as for the fragmentation in the iOS ecosystem in general… that's a topic in and of itself)

Writing your library in Objective-C is silly if you mean to use it from Swift. The languages are incredibly different (value types, powerful protocols, generics vs. just objects and simple C types), so you'll heavily limit yourself by writing in Objective-C. Nullability annotations don't really make up for that. If you're using Swift, don't limit yourself by writing your components in Objective-C.

I think the point is distributability. For an internal library, use whatever makes sense. If you want to make it available for all iOS developers, write it in ObjC, because many iOS apps are still using ObjC.

Maybe he had to deal with C/C++ APIs. He does flat out say the two main modern languages are Swift and Objective-C. I know using OpenCV from iOS is a real nightmare for this reason.

C libraries are fine in Swift, but I feel that dealing with a C++ library (and non-CF/system C libraries) is usually a rare exception when "writing an iOS SDK". Most iOS libraries are just written on top of Foundation and the appropriate platform interface kit (if it's a UI library).

You miss his point. It is much easier to consume Objective-C API from Swift then consuming swift API from Objective-C. Even more so if you need the library integrated at a C++ layer of your app.

I didn't miss his point. I disagree with it. Using Objective-C sacrifices so much that is great about using Swift, and the continually-shrinking Objective-C-only developer market doesn't make up for the loss of all of the Swiftyness.

C++ layers are rare for framework authors, so I don't think that the idea of them should be used to dissuade people from writing Swift frameworks. Obviously, if you need to integrate with C++, use a language that can.

There is still a massive, massive amount of enterprise software (see: everything at Apple) that has millions of lines of code in Objective-C. That code will never be rewritten. If you have millions of dollars riding on a framework/app in Objective-C the business needs outweigh any perceived language niceties you get in swift.

This aside - I also disagree with you, Swift is a worse language in a lot of ways, not to mention the extreme compile times and purely unfinished concepts, as well as bugs related to seemingly foundational things like auto-created initializers for structs not getting picked up properly by sourcekit.

If you already have millions of lines of code, "how to write a framework" is not very relevant in your case. You've already written one, obviously you won't rewrite it. The subject is new frameworks, if you're already using Swift in your app (and thus have paid the bundled frameworks price), why limit yourself to Objective-C because some other developers aren't using it yet?

Because the entire point of writing a framework is for other people to use it, which is a pretty strong pull towards the lowest common denominator.

I don't necessarily agree with that - many people/companies build frameworks for primary use in their own work that they open-source as a courtesy - with benefits of course, more eyes make shallow bugs, advertising, etc.

e.g. Github employees probably (originally) wrote ReactiveCocoa and Carthage because they were useful for developing Github apps, not necessarily to gift fantastic tools to the iOS world.

The macOS Dock and services handling infrastructure has already been rewritten in Swift.

More macOS components will be slowly rewritten in future OS versions.

Likewise some new Sierra APIs like Units are Swift only.

See this year's WWDC talk about Swift 3.0.

NSUnit looks to have an ObjC API?

https://developer.apple.com/reference/foundation/nsdimension...

I wasn't aware of it, thanks for pointing it out.

My point is that you don't sacrifice any of the awesomeness of Swift to the consumer of the SDK if you write a great Swift wrapper around your code.

Your code won't be as fast if your Swift types are copy-on-write wrappers around NSObject subclasses, instead of pure-Swift types.

I agree that you'll get the most coverage of potential developers by supporting both languages, but you're already doing the world a service by releasing open-source software. I don't think putting additional demands on maintainers (you must support CocoaPods, you must support Objective-C) is a reasonable request. If a consumer of the framework wants those features, they can submit a pull request (or just maintain the Podspec independently, as is done with ReactiveCocoa).

Performance is the least compelling reason IMO because if it's slow, you should optimize the slow bits. Otherwise it won't even register as a percentage of total time; looking at it this way is just arguing for premature optimization.

Using obj-c from swift is easier than using swift from obj-c in my experience. So if your goal is to distribute the project, then doing it in obj-c has real benefits.

If you use Swift and consume the library in an Objective-C project (which is perfectly possible) you will add the Swift ABI to your project automatically which adds about 10MB to your project. The ABI needs to be added because the ABI isn't stable yet.

For mobile apps that's still a big consideration.

The other way around is much easier, that's why he proposes Swift wrappers around Objective-C code that allow a Swifty interface with your libraries. But often using generics and nullability will already help a lot.

My presumption is that you're already using Swift in your app when you decide to write/break out a component framework and open-source it.

If you're using Objective-C, keep doing that, but then this isn't really a question to begin with.

There are a lot of demands/needs/wants in this press that really only make sense to me if I'm writing a commercial SDK. If I've written an SDK as a support library for one of my projects and I want to open-source it or release it in any way, it'll include the language and build system support for my project.

If anyone else find that helpful, great - that's why I released it. If you 'require' me to add cocoapod support, when I don't use, now, or understand the intricacies of cocoapods, then that' bad for both of us (I probably can't support you properly when it breaks).

If the audience for this is really the Dropbox-types, i.e. their products live and die on integrations and SDKs, then this makes more sense...

Another interesting possibility is that when you release the library with your preferred build system and language, people can contribute additional ones :)

Just wanted to say: This is an excellent way to present this talk. The videos and slides synchronized on the screen, the transcript (looks like a slightly edited one, or perhaps the "shooting script", but still...) with links that jump in the video to the correct time.

Really, really well done.

Nice shirt, too. :)

One small nit: Apple really doesn't want people using two-letter prefixes for their components. I know that's a pretty sore subject for a lot of iOS devs, but the example WFOAuth2 (where WF is the "Workflow" prefix) would be frowned upon.

The Workflow app itself, however, is spectacular, and I use it all the time. I use it to log arbitrary stuff to HealthKit (like when I drink a caffeinated beverage, for example).

You are right! I was considering using WFO, but decided to go with WF (which is what we use for Workflow).

Great to hear that you enjoy using Workflow :D

Seriously, thank you for writing this. I just started writing an iOS SDK days ago and had already made the mistake of starting off with Swift rather than Obj-C. Luckily I've only written 250 lines of code, you saved me hours of technical debt!!

>As such, it’s critical that you write your SDK in Objective-C so that the people that don’t ship Swift can actually use your SDK.

I'm not an iOS developer but aren't Swift and Objective C files interchangeable?

Almost all Objective-C APIs can be accessed from Swift. (The one exception I know of is that a type which is a class with protocol conformances, e.g. `UIViewController<UIPickerViewDataSource>`, cannot be represented exactly in Swift. You can still use APIs with such a type, but the type will be different in Swift.) So if you implement your API in Objective-C, you're guaranteed that it's usable in Swift (though perhaps not idiomatically).

There are multiple Swift constructs that cannot be exposed to Objective-C, such as enums with associated values, structs, and global variables and functions. So if you implement your API in Swift, you have to be careful if you want it to be usable from Objective-C.

Another problem with writing your API in Swift is that the Swift ABI isn't stable yet. It changes between compiler versions. This means that your API (if written in Swift) must be compiled with the same Swift compiler as the user's code. If you want to distribute your API as a binary-only library, you'll have to release a new version every time Apple releases a new version of Xcode, and your user will have to be sure to get the correct version of your library for their version of Xcode.

> The one exception I know of is that a type which is a class with protocol conformances, e.g. `UIViewController<UIPickerViewDataSource>`, cannot be represented exactly in Swift

Sort of doable with generics, but only from Swift code:

    <ViewController: UIViewController where ViewController: UIPickerViewDataSource>

Good information. Thanks for posting this.

The HN title lacks the iOS part of the headline - which is quite important because all the information is useless if you are not interested in iOS.

Agree. I closed the tab after reading iOS in the headline.

Me too.

I completely disagree, there is a lot of semantically valid information in the talk, like "make file paths configurable". Is that really only applicable to iOS? I think not.

Reply on news.ycombinator.com