Ask HN: Why Isn't Go Functional?
One of the things I keep reading about functional languages is how they make reasoning about code easier and how this is particularly useful for distributed systems. Given that Go was built by Google specifically for the purposes of building distributed systems, why isn't it functional? As discussed in the Go FAQ relating to the design principles of Go, the language was designed with the idea of using it in situations as an alternative to older object-oriented/procedural languages like C, C++, and Java.[1] So they chose to make a language with less "book keeping and repetition", and with a focus on ease of entry to the language itself. Creating a purely functional or even mainly functional language to replace or start new services and programs written or designed by people familiar with the OO/procedural paradigm would be to add another entry barrier or level of complexity to adopting the language. I believe this would be counter to their goals for the language. > So they chose to make a language with less "book keeping and repetition" Yes, less repetition and more I like that they aren’t trying to hide that it’s a beginner language. Go was built by Rob Pike and based upon the work that was done for the distributed operating system Plan 9 and the Limbo programming language that was built along with it. When these things were being developed, functional programming techniques were not in vogue in the mainstream software engineering world. It likely might not have been practical. Go was designed by some of the most gifted engineers of our time.
In particular, Rob Pike, and Ken Thompson had been working in distributed system for more than 40 years, and before Google they worked for one of the most renewed research facilities in the whole world. Murray Hill gathered together people focused on solving real problems for the most distributed industry in the last 150 years. They had the opportunity to witness technologies and solutions that worked. And I quite sure that they suffer those that doesn’t. From my humble point of view, Go is a compromise of what works in real life, the first time I see the go compiler, I traced it back to the good things of Limbo, C, object oriented programming and the adding of a simple syntax to create CSP concurrency made it a killer application. The idea that functional languages are overall superior is just not accepted by the majority of software developers. In practice, most popular languages borrow many concepts that originated in functional languages, but don't restrict themselves to being purely functional. They definitely make automated reasoning about code easier. But perhaps they don't always make it easier for humans. See for yourself by comparing functional and non-functional versions of various algorithms on this site: https://rosettacode.org/wiki/Sorting_algorithms/Merge_sort I think Rob Pike, one of the inventors doesn't see the value of the functional style, at least that is what I gather from the readme of https://github.com/robpike/filter :P Personally, I would love if the language design was a step closer to I.e. ocaml, especiall if I see the 'this is a struct, but only one of the fields should be not-null' hack to implement sum-types, or when I see how closely select over a channel resembles pattern-matching. But the library ecosystem is good and 'just using forloops' is fine. There is no silver bullet for any kind of problem in software engineering. And, distributed systems is no exception. Also, when designing a large system e.g. a distributed system, the fact of selecting a language for its functional, procedural or object oriented capabilities is not much important afaik. Hence, I think the first sentence of your question has no effect on the question. For the question of >why isn't it functional? I think current state of Go is performing enough for Google's problems. It isn't purely functional, I don't think, but functions are a first-class thing in Go, aren't they? I mean, I've returned functions from functions, and passed functions as parameters.. and there's no requirement that a function modify state anywhere.. what else is required to be purely functional? In a purely functional language, functions apply inputs to a single expression to produce an output (though there is usually a way to add some declarations inside the function to keep the expression from getting to hairy). Side effects are possible, but they are wrapped up as values that still have to be passed around as functions and possible errors must be handled. "Purely functional" means that pure functions are not only possible, but that they are the only kinds of functions in the language. While it might be theoretically possible to write purely functional code in Go, it doesn't have mechanisms for treating side effects as values, and the heavy reliance on arrays and channels and things would make it unnatural. Pure functions don't contain loops because a loop is only used to generate side effects. That would be hard in Go, but it's pleasant in languages that support pure FP. If this explanation sounds weird, learn some Haskell to get an idea of what pure FP is like. It's fun stuff. Go isn't functional because if it was, it wouldn't be Go; it'd be Haskell, ML or Ocaml. Google didn't want to make a clone of Ocaml or the like; they wanted something different. Hence, Go. Because it's a language build by engineers with a pragmatical mindset ? Because they wanted it to be actually usable by the vast majority of programmers developing real world solutions? Not saying that functional programming doesn't have a place but if it was as wonder as some people say it is it would take over the world because the shops that used it would way outperform those who don't but for most things that doesn't appear to happen. That's a pretty poor argument. Functional programming isn't as popular because it's more difficult to learn, not because it's not as good. Sometimes the easier, almost as good solution is better than the technically superior option, especially if you need a ton of people to learn. Functional programming is certainly better in a lot of ways, but Google wasn't trying to build a "good" language, they were trying to build a "good enough" language, which is why it doesn't really innovate, but merely pulls in familiar features into a simple package. Go isn't the best at anything, it's just good enough at several things, and very simple to learn, and that's exactly why Google uses it. No, they were trying to build a good language for their set of problems. Their set of problems include huge numbers of engineers, and code bases in the tens of millions of lines. They wanted short compile times, and ease of reading code. They weren't going to get either of those by something in the direction of Haskell. And the result is not "just good enough". It's actually quite good for their problem set. >One of the things I keep reading about functional languages is how they make reasoning about code easier and how this is particularly useful for distributed systems. Given that Go was built by Google specifically for the purposes of building distributed systems, why isn't it functional? 1) Who said that what you "keep reading" about functional language is true? 2) Even if true, who said that what you "keep reading" about functional language is true in all cases? 3) Who said other factors don't come at play, and might be even more important? 4) Who said a language has to adopt functional principles wholesale, and isn't it enough that it has some (as languages increasingly get)? 5) Doesn't the IT industry always find some methodology to hype as the silver bullet, much akin to the diet industry? Maybe functional programming is one of them? 6) If you don't have personal experience with functional languages or can't personally appreciate whether they really do make "reasoning about code easier", why take what you read about them at face value? Perhaps they miss something? 7) To argue on the other side, who said that when a large company builds something, they get it right? Often it's the inverse. 8) Who said Go was built "by Google" in the strong sense? Go was a rogue project from a small team AT Google. It wasn't an official Google effort (the way Java was for Sun, .NET for MS, and so on) to make an official Google language. It's just that that group was concerned with "what language features would be a good fit for the work with do here at Google" -- but the inverse is not true, Google as a company/management/etc wasn't concerned with building such a language. They just went ahead with people devoting time to the project. 9) Who said the Go team has kept with the times (or even with older but still relevant PL research)? What I write above are not answers and are not meant to
be "the truth", they're meant to make one see the situation more critically. Re 1: Who said? Those who wrote the stuff being read, of course. (You're muttering "self referential" like that's some kind of a problem... /s) Re 2 and 3: I think this is the big one. Functional language are great in some situations. Non-functional languages are great in some other situations. The problem is, someone who doesn't know functional languages gets into a situation where they're good, finds out about functional languages, starts using them, and a large number of their problems go away. They experience functional languages being wonderful. But they conclude "functional languages are wonderful" rather than "functional languages are wonderful in that situation", and they proceed to try to sell the world on how wonderful functional languages are. (Not just functional - ditto for other language styles and development techniques.) The trick is, given the problem, to pick the right languages, tools, and techniques, rather than to have One Right Answer. Re 9: I'd say, given Pike's background, that he almost certainly knew at least the older PL research. Ditto Kernighan (who didn't design the language, but did write the book).
Truly an excellent design. x, err := dostuff();
if err!=nil { return nil, err; }
y, err := otherstuff(x);
if err!=nil { return nil, err; }
z, err := morestuff(y);
if err!=nil { return nil, err; }
w, err := alsodo(x,y);
if err!=nil { return nil, err; }