A Ruby Design Process - Talking Points
At RubyConf 2012, I gave a talk proposing a Ruby design process. Following the conference, I detailed the process in a blog post. I submitted a ticket on the MRI bug tracker asking Matz to consider my proposal. The proposed design process is summarized on the RubySpec website. If you support the proposed process, please add your voice by signing at the above link.
The discussion of a Ruby design process is emotionally charged. People focus on their own ideas and illusions far more than facts. Conversations resemble political debates, not technical discussions. Consequently, ten times more energy is invested in clarifying misunderstandings and misinformation than in substantive discussions about the merits of the proposal.
To help keep the discussion focused on the merits, the following points amplify and further explain the proposal.
Every reasonably complex endeavor has elements that compete. Examples are all around us. "You can have it fast, cheap, good; pick any two." The CAP theorem. Enough time for recreation but enough money to pay the bills. The only way to resolve such tensions is to prioritize. The same is true of a Ruby design process.
The Ruby programming language is complex. Any implementation of the language is unavoidably complex. There are half a dozen significant implementations of Ruby. Those all represent different interests in Ruby. Attempting to implement a unified definition of Ruby is undeniably complex and difficult.
It's easy to imagine that Matz would have different priorities than another team implementing Ruby. The purpose of the proposed design process is to prioritize those things that create a unified Ruby programming language.
For businesses and developers whose salaries derive from writing Ruby, the value of a unified Ruby programming language should be obvious. If someone believes that businesses, customers, and developer salaries would be better served by a fragmented Ruby language, they are encouraged to give their argument in support of it.
There is no definition of the Ruby programming language that is complete and accurate enough to provide unity to the Ruby community.
The proposed Ruby design process prioritizes the following things:
- A precise and clear description of what exactly is the Ruby programming language.
- Efficient use of the time and attention of implementers.
- Fairness to all implementations in deciding Ruby programming language features.
The Design Process Is Focused On The Definition Of Ruby
The product of the proposed design process is a precise, verifiable, and unified definition of the Ruby programming language. The definition is independent of any particular implementation. There are numerous programming languages used to implement Ruby. None of those languages, or implementation artifacts that result from the use of those languages, should be given special preference in the definition of Ruby semantics.
The RubySpec project is the only project that has the explicit goal of a complete, executable description of the Ruby programming language. The product of the proposed design process is a more complete and more accurate RubySpec exactly describing the behavior of the Ruby language in a way that gives developers and the rest of the Ruby community clarity and confidence in understanding Ruby.
The Design Process Is As Simple As Possible, But No Simpler
The proposed design process does not impose any restriction on any implementation with respect to experimenting with Ruby language features. This appears to be one of the most misunderstood parts of the proposal.
Every implementation is completely free to experiment with language features. They can use whatever development model they choose, whatever source control they like, write or not write tests in any framework they fancy. They may collaborate with other implementations in any way they wish. They can discuss features in any language, on any mailing list they choose, or by means of any medium they wish.
In other words, the MRI developers can continue doing exactly what they have been doing.
The design process is only concerned with what becomes the official definition of Ruby. Experimental features may or may not ultimately have value. The design process is optimized for reaching consensus on the features that will be part of a unified definition of Ruby with the least amount of work or ceremony.
The Design Process Is Open To The Community
Right now there is no formal design process for determining what features are officially Ruby. There has never been a design process. People sometimes propose a feature to Matz and sometimes he accepts it. Most of the time, language features result from Matz's experimentation with the language. Matz has repeatedly said he has the absolute authority to decide what features can be called Ruby. If this ad hoc process is "open" to the community, then so is the proposed design process.
Under the proposed design process, anyone, whether from the Ruby community or not, can propose a feature to Matz, just like they can now. If Matz agrees with the proposed feature he, or other MRI developers, can bring it to the Ruby design council for discussion and approval. Hence, the proposed process is just as "open" to the community.
However, the proposed process is even more open to the actual Ruby community. There are Ruby developers that are entering the community from Java and may never use MRI at all. If they have an idea for a Ruby feature, they may be able to implement it in Java but not C. They would be able to work with JRuby to propose a feature that has been implemented, tested, and validated with actual experience.
The Design Process Is Efficient
The proposed design process requires that language experiments are clearly documented and described by precise specs before being proposed as official Ruby features. The discussion then focuses on concrete issues. This approach makes efficient use of the attention of Ruby implementers. It also limits communication about irrelevant details, further reducing the burden on Ruby implementers when reviewing proposals.
Most important, precise specs makes the effort of implementing the feature as small as possible. This also makes the most sense: Who is more qualified to describe precisely how a feature works by writing specs but the people proposing the feature?
While the design process requires documentation and RubySpecs, it does not prevent an implementation from writing fully functional code for the feature. The design process only lists what is required in a proposal. It's doubtful any feature could be proposed with sufficiently detailed documentation and RubySpecs without any code.
The Design Process Concerns Language Evolution, Not Experimentation
Under the proposed process, Matz is as free to experiment with the Ruby language as he ever has been. So is every other Ruby implementation. However, developers, businesses, and Ruby implementers get clarity, visibility, and transparency for what is officially Ruby. The proposed process is only concerned with features that will be part of the unified definition of the Ruby language.
The proposed design process is not concerned with how implementations experiment with language features, with what VCS they use, or what bug tracking software, whether they practice BDD/TDD or write tests at all, what language they use while discussing Ruby over lunch, or any of a myriad other activities not explicitly defined by the process.
English Is A Reasonable Language For Collaboration
The proposed design process attempts to reduce as much as possible the need for all implementers to discuss proposed language features. The discussion occurs after clear documentation is written, after precise RubySpecs are written, and after everyone implements the feature so that it passes RubySpec. The discussions then focus on concrete facts about the impact of the proposed feature to existing and future code, whether it is in libraries, frameworks, or applications.
For the limited discussion that is required, English is a reasonable language. It is the only language that is likely to be used to some extent by all the Ruby implementations. It is the language used by international communities in computing, science, mathematics, and other fields.
In fact, English is the language in which the Ruby ISO standard is written. And that standard was written by MRI Japanese developers.
The Design Process is Fair
Matz and other MRI developers have put a tremendous amount of effort into the Ruby language. However, so have all the developers of other implementations of Ruby. In fact, the combined effort on other implementations of Ruby very likely far exceeds all the effort dedicated to MRI. Every implementation of Ruby is helping Ruby and helping Ruby developers.
Most importantly, other implementations of Ruby are helping Ruby developers in ways that MRI and Matz are not. Furthermore, while Matz may know a lot about Ruby, he doesn't know everything about the many challenges Ruby faces. Matz is not a concurrency guy, or a business guy. He's said that himself. He's also not likely a Java guy, or C# guy, or crypto guy.
The Ruby language needs a design process that fairly distributes authority over what features become officially Ruby. There is no process more fair than consensus among all the Ruby implementations. These are people who have dedicated tremendous effort to understand and implement the many complex idiosyncratic Ruby behaviors and support developers in many difficult environments.
If all the implementations believe that a feature is bad for Ruby, Matz should not ignore their collective wisdom and dedication to Ruby. The proposed design process merely formalizes that fact. Finally, if Matz disagrees, he is still free to include the feature in MRI and demonstrate its importance. The same is true of every other Ruby implementation.
A Design Process Is Critical At This Stage Of Ruby's Development
At this stage of Ruby's development, a design process is critical. Point out a single other widely used programming language with a half-dozen implementations with a unified definition of the language and no formal process for agreeing on that definition.
Ruby has no formal design process, despite Ruby being an industrial strength programming language, having hundreds of millions of dollars invested in businesses built around the language. There is no process to maintain unity of the Ruby language. There is no definitive resource defining Ruby. Developers can only support multiple Ruby implementations with trial and error.
It is short-sighted and irresponsible to advocate for the ease of experimenting with random language features over the stability and unity of the Ruby language. Ruby is not a toy language. It is almost twenty years old. It powers many businesses and pays tens of thousands of developer salaries.
Matz created the language. His creation has been a gift to the world. The Ruby community has been and continues to be immensely grateful to Matz and deferential to his opinions and there is every reason to believe that will continue. Insisting that one's ownership of a language is paramount at this stage of its development is unreasonable.
The proposed design process seeks to create these three things:
- A definition of the Ruby language that is independent of any particular implementation of Ruby.
- A definition of Ruby that is explicit and verifiable by running RubySpec.
- A process of deciding what features are in the Ruby language that is fair to all implementations of Ruby.
If you disagree with those goals, please explain why the goals are not good for Ruby. If you believe the proposed process would not promote those goals, please explain why.
If you support the proposed process, please talk to people about it and why you support it. Also, please add your voice in support.