Why Programming Language Matters
Choice of programming language can be a source of religious wars between software engineers and developers. It is one of the many areas where people build a subjective opinion based on how “nice” or “elegant” something is and whether they get joy working with it or whether they find themselves fighting the language or tool every step of the way.
Clearly, this is going to vary person-to-person as people’s mental models vary, as their idea of elegance varies and whether a technology differs greatly from other tools with which the developer has a lot of experience.
Due to the strong influence of subjectively, it is common for people to put the programming language debate into the same camp as arguing over tabs versus spaces for indentation or reoccurring arguments around editors. All programming languages can do the same things, but vary in their approach, so some feel it is a waste of time trying to argue as if there are objective measures of quality.
I recently saw a strong example of this stance in the opening session of QCon London 2018. Someone running a track on using various languages in the back end asked the room if anyone felt that choice of language mattered in back end applications and only minority of people raised their hand who were then told they were “wrong” if they felt it mattered.
I am someone who raised their hand and I would like to explain why it is not “wrong” to care about choice of programming language in our applications.
This is not meant to be a counter-argument or strong disagreement though. My intention here is to expand on a deliberately blanket statement like “it doesn’t matter” as I think there is much value in exploring the nuances on where that statement is true and where it is false.
Firstly, I want to acknowledge the things that absolutely do not matter (most of the time) so I do not give the impression I am defending religious or flame wars. Then I want to remind the reasons why subjectivity is important at times before finally unpacking what I think are truly objective things that matter about a programming language.
Things that do not matter
I fully understand people who take the “it doesn’t matter” stance since most debates or discussion on language choice focus on trivial aspects, things that people just do not “like” and other subjective things.
Yes, Python uses whitespace semantically and other languages do not. No, it does not really matter. I personally like it as it removes the redundancy of using braces, for example, to determine scope and then also indenting for readability — i.e. when I indent I achieve both scope and readability at once. If you do not like it, then that’s fine too and it doesn’t affect your application at the end of the day if you just get used to it or have appropriate tooling.
Yes, Java is verbose and Scala is not. Does it really matter for your project? Only if that boilerplate soaks up real time, which is rare with appropriate tooling. Personally, I prefer to get going more quickly with Python but that’s because I like to do rapid prototyping where I think there are some objective arguments that you can move faster in more expressive languages. However, in that case you are probably always better prototyping in languages you know well, which makes this a subjective point again.
Typing (sort of)
With appropriate testing, you do get away with achieving the same results with or without static or dynamic typing. There are debates around bug rates and I have even personally seen bugs in Node.js applications that any type checking would have trivially caught. I’ll leave it as an exercise for the reader to determine if there is objective evidence around type checking and bugs.
For now, I’ll assume this to be a subjective matter for the purposes of my argument.
It was common maybe 10 years ago to hear lots of arguments that Python was demonstrably slower than Java (or other languages) in a few cases. I’ve not been following performance benchmarks recently, so maybe the argument went away because Python is faster now. However, I heard a very convincing argument from a speaker at PyCon UK around that time in that Python (in her view) saves developer time — which is expensive — while trading off CPU time, which is arguably a lot cheaper than developer salaries now.
So, while you could choose a language for optimal performance on the back end of a web application, you might be delaying delivery (which is what you get paid for) for the sake of performance tuning.
With the modern landscape of cloud computing, autoscaling and serverless architectures we can paper over any need to micro-manage performance with caching, horizontal scaling or breaking up monolithic services. Of course, I am not advocating for throwing performance out the window, but we can obsess too much about milliseconds here and there when really we’re paid to deliver value not necessarily on CPU cycle efficiency.
I hope it goes without saying that there are clearly embedded use cases where high efficiency is key, but my focus here is on enterprise and web applications.
Sometimes subjective things do matter
I know this is not necessarily the intent when someone states that programming language choice “doesn’t matter”, but it does avoid the human side of the issue in that developers’ happiness is important too.
I think the assertion that all programming languages are equally capable for a web back end comes from a very technical viewpoint where, yes, a Ruby service could be rewritten in Rust and would work equally well.
So, technically, yes, language does not matter if all that matters to you is the technical capability.
Software engineering teams are made up of people who will have personal feelings about languages, who will have various levels of experience with languages and might even have various levels of general software development experience. There’s a reason Python gets suggested to new programmers.
If a team is about to work on a new project, then factoring in people’s knowledge, opinions and, yes, even feelings is not wholly irrational. In this context, it is objectively the case that a team whose subjective needs are met will deliver the project sooner than a team who is forced to work with tools they are unfamiliar with.
I realise that means I could be interpreted as supporting the argument that “PHP is good because so many things are written in it and it’s easy to deploy.” I do not feel good about this and can only apologise.
Objective differences that do matter
Does all this mean that there is no right choice of programming language for a given application? Is it just about choosing something you like or are some tools a better fit for certain jobs?
The argument gets thrown around that all languages are equally capable. They are, after all, all Turing Complete (or we would not likely call them programming languages at all) and for every functional solution to a problem, you can rewrite it in an object-oriented or imperative way. We know this is the case as there are whole collections of solutions to things like the Towers of Hanoi in multiple languages.
So, in this sense, all languages are capable of solving (nearly) all problems, but are some objectively better at certain problems? What things make a language better?
A language is more than just a language
As with the subjective differences, I do actually agree the choice “doesn’t matter” in the very technical sense that you could build anything in any language (assuming you had people that knew it well or time for them to learn). And again I disagree with that viewpoint on the basis that it’s missing a wider context that is important for actual engineering of business systems and web applications.
The key piece missing here is considering a language in terms of its syntax and semantics only, but in the real, applied discipline of software engineering we do more than just write code. We also build and deploy that code. We pull in dependencies, write tests and build on top of frameworks.
It is important to consider the effectiveness of a language not just in terms of the language, but the ecosystem and community built up around it.
Use cases where one language dominates
I personally only learnt Ruby because of Cucumber and Capybara. I learnt some Scala because of Apache Camel and Groovy due to Jenkins pipelines. Some of these can be done in other languages, but there is a clear “right choice” for many of these applications.
It is this capability in the available libraries and frameworks that I think is one of the strongest objective considerations around language choice. It doesn’t have to be the only consideration as you can still use less mature libraries in a language your team is familiar with if you understand what you’re trading off.
A lot of experimental projects I do deal with Linked Data and the Semantic Web for which I find RDFLib — written in Python — to be extremely powerful. It is useful for building applications that deal with RDF which much of the details around the various RDF formats abstracted away. In fact, I am yet to find an equivalent in any other language where a common Graph model can be parsed out of various formats. Most implementations elsewhere only deal with one format at a time, so I always come back to Python when I want to move quickly and not write objectively more boilerplate code.
Alternatives can be less mature
I should expand a bit more on the Cucumber and Apache Camel examples as someone is bound to point out that CucumberJVM and Cucumber.js exist. Also, there are other behaviour testing frameworks such as Lettuce, Nightwatch and Behat.
Apache Camel is just one implementation of message routing and there are other ways to solve some of the problems it offers to solve.
However, it is this exact trap I want to highlight: a lot of the time the tool in your preferred language could be less mature or playing catch-up to the more established player. This might be ok for your use case as the core features are enough for you or the familiarity of the language is a strong enough consideration that you are willing to take on the extra cost of implementing missing functionality, but so long as you understand the choice you have made.
I have anecdotally seen too many people reach for alternatives to Cucumber simply because they “don’t like” Ruby, but without even a moment’s consideration of the sheer maturity of the ecosystem around Cucumber, Capybara and RSpec. In this case, they are allowing a subjective choice trump objective considerations and sometimes twist the “language doesn’t matter” argument to support going with their preferred choice.
Node.js has matured recently and big, well-known frameworks like Express.js have a lot of traction, but I have see these technologies adopted in large, high-traffic enterprise systems back when they were far less mature.
I struggled with memory management in Node.js as if I were dealing with the JVM in the 90s and encountered really basic bugs like lack of connecting pooling or socket reuse in various NPM libraries. I still struggle to find good library support for Cache-Control and Vary headers in HTTP client libraries and spend objectively more time on things I simply do not think about when I do the same tasks in a Java or Python project.
One thing that I think has really helped Node.js mature is having more of a community form in that area. Having conferences dedicated to the technology and online discussion areas really makes a language better to work with in the enterprise.
Most enterprise engineers will rely heavily on Github issues, Stack Overflow and other online forums to do their jobs. If we encounter errors, we need to be able to put the error message into Google. If we are trying to make two technologies work together, we can find complete solutions in blogs on how to configure them (e.g. if you search for “nginx node.js” you get countless examples of configuration you can lift to set nginx up as a reverse proxy in front of a Node.js app).
The help is only there if a community reaches a critical mass such that 1) they have probably encountered most of your problems before you and 2) a culture of sharing, blogging and helping out learners can emerge.
I think only certain kinds of enterprises (perhaps brave ones!) are really in a position to adopt newer things like Go or Rust before the community is really established. There’s a reason the JVM is still a safe choice for the more cautious enterprise. I’m not saying bravery is bad and I think many enterprises could do with being more experimental and forward-thinking at times, but everything in moderation or you won’t actually deliver anything.
One of the things that keeps bring me back to Python is the community. I can see newer users coming to the language for data science or as children and there is a culture of nurturing and helping inexperienced users.
Programming language matters if you to do more than just “program”
I have played with semantics a bit here, but I think it’s an important message. Programming language doesn’t matter if all you are doing is programming — i.e. just writing code — but software engineering is more than that.
As engineers, we are solving problems through code. We might want to reuse solved problems in the form of libraries or we might want to draw upon frameworks that abstract away lower-level concerns so we can focus on the higher-level problems.
We want to create good teams that solve problems more quickly and more effectively that one engineer can do alone.
Choosing a programming language will affect your ability to solve these problems. Choosing different languages will have different effects on your teams.
If you are truly a software engineer, I think it does matter to consider your language carefully.