A couple of days ago, I released a native macOS application called Remember. It is a small, keyboard-driven application for stashing away notes/reminders for later. One of the cool things about it from a programming nerd perspective is that, while it is a completely native Cocoa application whose frontend is built with Swift, the core business logic is all in Racket!

Why not use racket/gui?

I started out with a proof of concept that used Racket for the GUI, but I realized that I'd have to write a bunch of Objective-C FFI code to get the UI to look the way I wanted (a carbon copy of Spotlight) and it seemed like it would be a pain to try and integrate DDHotKey and to add support for launching at login into a package that is easy to distribute. I was also unsure about how I could notarize the distribution for macOS Catalina (more on this later).

Why not do it all in Swift?

I don't know Swift particularly well, nor do I like it very much. I find Apple's documentation lackluster and Xcode is surprisingly buggy (renaming a class and its associated file fails to rename the file on disk, which causes Xcode to fail silently, for example). I wouldn't mind the documentation being bad if the core cocoa code was open source/source available; at least then I could look at the implementation to try and understand what's going on. /rant

More importantly, I plan to support Windows and Linux which means that writing the core in a portable language is going to minimize the amount of work I have to do as well as the differences between the implementations on each platform.

How it Works

The Racket core runs a custom JSONRPC server that listens for commands on stdin and sends responses on stdout. Using Racket's raco exe and raco distribute commands, that core gets built into a native executable and copied into the Swift app's resources folder. The Swift application runs the core as a subprocess on startup and communicates with it via pipes.

RPC commands are asynchronously handled by the core and the core may also send asynchronous notifications to the frontend to let it know when entries are due.

Notarization

This gave me an actual migraine. It took me a couple of hours to figure out how to get everything notarized. I had to enable App Sandboxing for both the frontend and the core application, figure out which (via trial and error) entitlements were necessary, realize that I needed a separate set of entitlements for the core application and that the com.apple.security.inherit entitlement, for whatever reason, doesn't let the subprocess inherit its parents entitlements, meaning that I had to also explicitly assign the core application the “Allow JIT” and “Allow Unsigned Executable Memory” entitlements, else the process would get killed with a SIGINT and a red herring error message about how the executable doesn't have a valid bundle identifier.

Would I do this again?

I'll have to see how porting to other platforms goes, but so far I'm very happy with this approach. The result is fast and now that I've built the RPC infrastructure I can easily copy all that code into other projects. Writing the business logic in Racket means that I can iterate very quickly and writing the GUI code using the native tools for each platform is advantageous in terms of look and feel and distribution.

What about iOS?

Unfortunately, the RPC approach breaks down on iOS where you're not allowed to run subprocesses. An approach that could work there is building the app into a shared library, linking against it and doing the RPC in-process. I think that approach could work, but Racket would have to be able to target arm64 for that to be feasible. Fortunately, now that Racket is able to run on top of Chez Scheme, which already has backends for many platforms, including arm32le, that might be a possibility in the future. myfreeweb pointed out on lobste.rs that this is already supported!