My Thoughts on Refinements
I have followed the various Ruby 2.0 refinements discussion for a little while now. There is certainly a lot of talk in the air for and against the inclusion of this feature. For those tuning in, here are some notable links to get you up to speed:
- A presentation by Akira Matsuda at RubyConf 2012 introducing refinements along with other interesting new Ruby 2.0 (potential) features.
- A thorough discussion by Ruby runtime developers
- Another RubyConf 2012 talk by Brian Ford a prominent Rubinius developer.
- A blog post by Charles Oliver Nutter a prominent JRuby developer.
I want to start off by saying, I have a ton of respect for the incredible work all developers have put into making my job a reality. The community wouldn’t be where it is without all your hard work and sweat. And my career certainly depends on the fruits of your labor. Thank you.
There is an interesting mix of pros and cons for the inclusion of refinements in Ruby 2.0. I’m going to distil what I’ve read and my comments on the matter.
Implementation difficulties within MRI/JRuby/Rubinius/other
I’m just not qualified to weigh in on this area. I understand that there are some (many?) challenges in implementing this fully without breaking the principal of least surprise for developers. That may or may not sway everything else. It should certainly not be overlooked, but this isn’t a scenario where you have people proposing code without writing it. Runtime developers are currently debating the feasibility implementation, and it does look split.
Crazy use cases
Many are cited on Charles’ post such as:
What does the following do?
Josh Ballanco’s example, also from Charles’s post where
“the following code only refines the “bar” method, not the “foo” method.”
I strongly dislike this argument. In complete fairness, you have no idea
String#upcase is going to do. Someone before the call may or may
not have overriden the method. It could be
to email my grandmother stock quotes upon invocation. No one knows. But
we still love Ruby, we still get work done, and we still ship code.
Thinking of obtuse edgecases on the nastiness or confusion this could cause does not dismiss the potential possibilities. My initial reaction, and I think this is shared with many developers, is “Wow, this could really clean up some global-level monkey patching.” That doesn’t mean it will be used only for the forces of good.
Myron commented with some guidelines he uses when using monkey patching (which I think are great). But why wouldn’t the same apply to refinements?
Saying that refinements could lead to dangerous and unusuable code does not dimiss the potential benefits of an incredibly flexible feature. Describe some of the existing features to a non-Ruby developer and watch their face either light up in possibility or scowl in disbelief. Like re-opening classes
What’s wrong with monkey patching?
Brian Ford has asked this many times, and I don’t think this is a great question to lead to the dismissal of refinements. Monkey patching is global and refinements are local to the call stack. I have not seen many great arguments won based on the fact that we can accomplish the same behavior through global means.
Things we should be discussing
Aside from language-level implementation details which some language developers seem to think isn’t a problem and others do, I don’t think the many of the current questions and discussions are productive. These are the topics I would like to hear discussed.
Is Ruby too mature to include something so experimental?
This might seem like a trollish question, but I think needs serious consideration. Ruby is not a young language with a small userbase. Sure we might be small when compared to the sea of C code written and deployed. But there is much more Ruby code in active use now than 5 years ago.
Changes this large has an affect on both the users who choose to use it, as well as those who don’t. Are we reaching the point where this is too extreme? If so, where does an idea/implementation like this belong?
The implementation of refinements was first proposed and then implemented by Shugo Maeda in MRI. Both JRuby and Rubinius have been fighting for RubySpec for many years, and there has been no effort to spec this feature.
Think about that. A prominent MRI runtime developer thinks about an implements an interesting feature, it is brought into the most popular runtime used, and the developers tasked with keeping up compatible runtimes are now forced to implement the same feature in their respective environments.
I get why JRuby and Rubinius developers would not be terribly motivated to do that without an argument.
Here is where I think MRI and JRuby/Rubinius have competing goals. MRI is continuing to push the boundaries of how to express the Ruby language in new ways. Whereas JRuby/Rubinius are forced to play catch up with large features. I don’t mean that a knock against JRuby and Rubinius, because they certainly bring their own unique things to the table. JRuby has an amazing capacity for concurrency by leveraging the JVM and Rubinius has rethought even the simplest things such as a fully supported debugger.
On top of playing catch up, JRuby/Rubinius continues to advocate to developers to try their runtime and highlight successful production uses. It’s difficult to convince developers to use your runtime if you have to change a large swath of code in the process.
This doesn’t mean JRuby/Rubinius developers are fully biased against new ideas. It just means to me that they are heavily incentivized to fight highly experimental features coming from a different runtime. Like refinements.
The number one thing I think we should be discussing is “How did we get into this situation?” Many developers think refinements can be used in good/productive ways (myself included), but why are many smart, respected developers vehementely against it? Apart from language maturity or incentives for/against, I think this entire argument suffers from what countless arguments in the past suffer from: lack of communication.
MRI is largely written by Japanese developers, but has become heavily adopted in the United States. JRuby/Rubinius developers are predominately English speaking. There are many developers who speak multiple languages, but not enough.
Brian Ford mentioned this in the talk I listed above, though I think he could have presented that topic in a much more tactful, less rushed way. He proposes that we should all be discussing these features in English.
If I was a Japanese MRI developer, I would be offended by the way that was presented. First, there are many Japanese developers who have done a great job discussing very complex topics in English, just browse ruby-lang. I think they should be highlighted, or at least given some sort of props for their efforts.
Second, I can’t help but put myself in their shoes. If I built some awesome wizbat that I continued to care for but over the course of time happened to be largely adopted by Russian developers. I would be a bit upset if they criticized me for not discussing development issues in Russian. Nothing against Russian, I just don’t speak a lick of it.
On the other side of the coin, developers of JRuby and Rubinius are being forced to implement some very difficult features in their runtimes. Reading the dialog, it sounds like Charles’ and Brian’s complaints are being heard but perhaps not discussed as thoroughly as they probably should be.
I do not have any good suggestions for this problem. But I do think this is the crux of it. In fact, this is most likely a frustration that has been worked through for many years and refinements has now been the place to plant the flag and have it out.
Should refinements be in Ruby 2.0?
I think refinements offer some very interesting possibilities. It might turn out to be a disaster or the to key to triggering some amazing innovation in a caffeine-hyped developer. But instead of dividing the community by campaigning for or against its inclusion. I think we should be discussing matters that led us here, and will inevitably lead us to other disagreements if left unresolved.