This is the tale of Gloss, my first major contribution to the open source iOS world. Gloss is a JSON parsing library that the world didn’t need, released in 2015.
Let me take you back there. 2015 was a time when the Swift language was young and intrepid Swift developers didn’t have a first-class option for JSON parsing. This caused many third-party JSON parsing frameworks to pop up. It became an in-joke in the community, What’s the JSON parsing flavor of the week? Each library saw some amount of success, each with its own tradeoffs, and none categorically better than the rest.
Gloss itself wasn’t even a standout on the JSON parsing playing field. I’m convinced it survives to this day only because a Ray Wenderlich tutorial once used it (fun fact!) which caused it to be adopted by apps and SDKs that otherwise wouldn’t. Just out of the blue one day, I saw my GitHub stars rising astronomically (for me) - and that’s when things changed.
Once its adoption reached a critical mass, it was no longer responsible to consider it a toy project I could toss aside. For the first year or so, I didn’t even want to toss it aside: I was having a blast! For a developer young in my career, it was validating and illuminating to maintain open-source software that other people relied on. I learned a ton about writing usable libraries and about communicating with other developers in the open. But…it got old once I wanted to use my coding energy elsewhere. Yet, I’ve been maintaining Gloss for 5 years.
Times have changed for Swift developers. There is now an excellent first-class option for JSON parsing in Swift, and it should absolutely be your first choice. Of course, I’m referring to Codable. So this weekend, I finally made a big move to help people stop using my code and start using something better.
The audience for this post is two-fold. First, it’s for anyone using Gloss today to get a more colorful backstory to the Migration Guide. Second, it’s for the developers who nod their heads knowingly at maintaining a library with a superior successor. Probably some Swift developers who maintain JSON parsing libraries. For these folks, I’ll share some tactical things I did to still feel accountable to the developers that rely on my code.
Where Do We Begin?
I’ve been thinking about doing this for quite a while. I’ve started many branches only to have abandoned them. What brought me to getting serious about it was what brings a lot of humans to change: pain.
It wasn’t the pain of maintaining
Gloss. After years of
Codable floating around, I now rarely get GitHub Issues or Pull Requests. No, it was the pain of using a (different) third-party JSON parsing framework at my workplace and wanting so much to use
Codable, having to create my own workarounds for doing so, and thinking what a relief it would have been if that library could give me a hand. It was realizing that I’m potentially putting users of my library through the same pain.
So, instead of my stagnation giving others a reason to stagnate - I’m going to offer a boost in the right direction.
Once I set my sights on doing this, I realized that I didn’t know how to “do” it. There isn’t a guide (that I know of) of how to gracefully help users of your library get over the hump. Is it purely a code solution? A single document? Do I make breaking changes? Do I simply archive the repo without a word?
I very often fiddle in code when the outline of a solution is there but the details are still primordial goop. And, this case was no different. I asked myself: if I was a user of Gloss trying to move to
Codable piecemeal, what splint might I write in between the two? With some fiddling later, a path forward was lit.
I made the decision to first release a backwards-compatible version of the library that includes helper methods safe to place in current implementations. These helpers aid the migrator in uncovering where their
Codable definitions would go awry. Indeed, that’s exactly what I’ve done in my own workplace code while migrating to
Codable. Given it’s a tactic I’ve proved out on real-world applications at scale, it’s something I feel good suggesting.
What helped immensely is that I had a Demo project waiting that uses Gloss, as well as a suite of tests verifying its functionality. I was able to work out many kinks in my splint by migrating my Demo project and I feel much more confident my work is backwards-compatible with my tests all green.
One of the many lessons Gloss taught me is what elements should be included in a professional open source library. That topic alone is enough for another post - but, to say the least, a good Demo project and test suite are it.
The Migration Guide
Once the implementation approach was figured out, writing a Migration Guide became much simpler. The point of it is to illustrate just enough to help the dedicated migrator get busy.
What I came to realize is that the Migration Guide cannot cover every scenario out there. It is those invested in migrating and in understanding how Codable expresses itself (which is a boon to any iOS developer!) that will lean heavily on this document. I did extra diligence to make starting on migration as copy-pastable as possible. With this document in hand, I’ll no longer entertain feature requests or troubleshooting for folks not on the migration path.
With Code and Guide ready, the last piece was to figure out how to get it all out to the world. Of course publishing said Code and Guide are part of the rollout - but what else?
Early on I determined that part of the point was to keep new developers from picking up
Gloss instead of
Codable. So, the first place to deter them is the
README. I’ll be honest, it was hard to take down a README that I worked on for over 5 years, but it’s the right thing to do. I kept a backup of the former README for reference, but the new README serves one purpose: a stop sign with only directions on how to start migrating.
A couple other good leverage points to keep folks from picking up Gloss or attempting feature development are at Issues and Pull Requests. So a new Issue Template and Pull Request Template were in order.
Lastly, I’m a big fan of writing down those things that exist between the technical and non-technical. It becomes a reference for myself if I ever need to think about something similar again, and hopefully proves useful to others. So I chose to write a blog post as part of this rollout.
In working through my own solution, it became apparent that breaking changes would likely both help ease migration and enable Gloss to be more opinionated in the direction of Codable. The jury is still out on whether I’ll pursue those breaking changes, or leave things as they are and await any feedback as folks work their way through migrating. I’ll be waiting expectantly to see if I get any new Issues post-release.
Then, once enough time has passed, I’ll finally hit the button that marks a repo as
read-only and close the chapter of my life that included Gloss. Let’s say, by the end of 2020.
Steps to Enable Other Developers to Stop Using Your Code Too
What follows is a little reference of just the steps I took toward getting my library deprecation-ready:
- If you have a release in progress, finish up that release and ship it: that’ll be your last release on anything that doesn’t involve migration (example: Gloss 3.1.1)
- Write a backwards-compatible splint that helps migrators try out the destination framework without harming their current integration (example: PR/363)
- Test your splint with realistic usage! Hopefully you have tests (and a Demo project) for your library and can confirm the splint doesn’t affect expected behavior or make for a nightmarish integration
- Prepare a Migration Guide once the shape of the splint is down
- Release a specific version of the library only containing that splint. Given it’s backwards-compatible, this should be a minor release. You can now think of your library as before-migration and after-migration. New changes are made only to support migration; only changes compatible with that version are accepted; users on lower versions with version-specific fixes must be gently turned down and urged to upgrade (example: Gloss 3.2.0)
- Move your current README to an archive of that README for reference only (example: Gloss/README_ARCHIVE.md)
- Create a new README specifying that the library is Deprecated and the succeeding library should be used in place of yours. Be sure to thank everyone who’s used and contributed to your library to date! (example: Gloss/README.md)
- Change the description of the repo to clearly indicate it’s Deprecated
- Update your Issue Template and Pull Request Template to push people towards migrating
- Write a blog post :)
After an amount of time has passed addressing migration issues, it’s time to officially deprecate. At that point you can:
- Move to only having a singular branch on your repo (no more develop branch)
- If on GitHub, use the Archive feature to make the repo read-only
- Hands off keyboard
You may end up not following every single step here, some of the steps might not work at all for your library, you may add a step of your own. Regardless, my hope is it’s helpful to see how one maintainer puzzled their way through this process while still feeling accountable.
Amidst preparing this piece, I stumbled upon the remains of a JSON parsing library written by an iOS developer whose code and communication I admire: JohnSundell/Unbox. I came to find that John too made the call to push developers towards Codable, and with a Migration Guide to ease the journey. I took finding this as a sign I’m making the right move.
So, while John & I released our libraries at nearly the same time in 2015, one of us deprecated 1 (one) whole year ago and the other is writing this now. As for the Ray Wenderlich tutorial that ushered Gloss here? Well, it’s been updated to Codable too. The time has come.
As of writing, Gloss has 1659 stars, 28 contributors, 36 releases…and 666 commits. Hm. I’ll miss you too, Gloss - thanks for 5 years.
This blog is licensed under Attribution-Noncommercial-Share Alike 3.0 Unported license.
comments powered by Disqus