To my mind though, Clojure brought clarity to the JVM on issues of concurrency, state, mutability, object orientation and persistence and for that I will always love it.
https://www.reddit.com/r/Clojure/comments/3tha86/transducers...
Likewise Spec ideas are not new to old timer Lispers, many forms of Lisp type annotations are similar to it.
What Clojure has brought was making those ideas "modern" again.
What might be revolutionary is making more "common programmers" aware of them.
And SIGPLAN papers don't really count. A revolutionary product is not the same as a revolutionary idea. Implementing ideas that have never been productized is still revolutionary.
ETS tables are usually out-of-(Erlang-)process. You are sending messages to them to update them and query them, and those processes are each functional, so it's still functional. IIRC there's a mode where it runs "in process" with you, but it's still conceptually behind the same interface. (Which raises interesting questions about what exactly is "mutable" or "not mutable", which IMHO ultimately get best answered by Rust.)
What Erlang is not is pure. You can query an ETS table, send it a command to mutate, then run the exact same query again and get a difference response. The message passing makes things impure, even though every individual component of the system is immutable-functional. And there are some other impure functions in Erlang as well, of course, like the functions to get the current time, but the message passing is in some sense the fundamental introduction of impurity at the architectural level.
That's how you think about it, but allowing mutation and allowing blocking interaction with processes are exactly the same, as blocking interaction with processes is the same having as continuations, and continuations alone are all you need for mutation, even if you imagine things to be immutable. Like monads, continuations are a universal effect. An otherwise pure language with continuations is indistinguishable from an imperative language, i.e., it is imperative.
Both Clojure and Erlang are functional-imperative languages, but unlike other imperative-functional languages (like ML), they have been designed with concurrency in mind, and the mutations they support are transactional (or at least atomic(.
Hyperbole much?
Data is a first class citizen of the language: data literal is part of its syntax, and it uses powerful persistent data structures behind that.
Programming with Clojure is what has made me realise that data is really what matters after all. None of that FP or OO discourse should deviate us from that reality check that computing is all about data-in/data-out. Pull the plug and/or battery of your computer and see what remains next time you power it up.
So yes, Clojure may not be revolutionary in itself, it does rely on the shoulders of giants. It is however setting a new standard in terms of being successful in bringing together a very well thought out set of features and philosophy to bring the focus of programming back towards the real complexity, instead of the accidental one created by how we think.
I wrote Clojure full-time for a few years. I was lucky to do so and get paid to learn such an interesting language and paradigm of software development.
Aside from the points you mention, of which I agree, there is one other point I took away from the experience that many would disagree with: I really, really missed static typing.
So that's the main reason I don't use Clojure as much any more and went back to other languages with a different model of typing. But I learned so much and it changed the way I work in other languages.
It is not pure, but it is much closer to this than other lisps and it is immutable by default with efficient functional data structures. CL and Schemes are not this way.
Source: I played for a while around 2008 with common lisp and web apps, and while the tooling (hunchentoot, elephant, etc.) was solid, my completely subjective assessment was that the community languished and it was very hard for newbies.
EDIT: Why was I downvoted? Explanations enable enhancement more than punishment... Is it for the reddit-like "source:"? I really dont get it.
Anyone is free to create a native AOT compiler for Clojure.
In fact many attempted to do so, but have given up due to the effort it means bringing a full eco-system up to speed to what Clojure already enjoys in JVM, .NET and JS runtimes, libraries and tooling.
Specially in quality of generated native code and GC implementations.
That it doesn't compile to Go, that's because Go isn't a good compilation target. Besides being a subpar language that ignored the last 30 years of research at least, it wasn't designed to be a platform for other languages. If you're looking for disappointments, you aren't looking in the right direction ;-)
That is not really fair. As an example, Clojure itself is dynamically typed, which means it ignores a large chunk of research itself.
It does not. PL research is about how to construct type systems if you choose to construct them. There's plenty of research in untyped languages, too (for a recent, pretty exciting example, see Dedalus, which is the language semantics at the core of Eve).
(1) "I love functional programming! Therefore, newer languages that aren't following in the mold of Haskell are conceptually antiquated."
... yet...
(2) "I don't care about static typing. Therefore, it's cool to ignore the trend of most major languages moving in that direction over the past decade."
Then of course the Rust and Swift guys would tell you that Clojure is an anachronism for relying on garbage collection. Etc, etc...
All of these trade-offs, at the level on which they're commonly discussed on HN and other forums, are matters of subjective preference. And that's okay. Disrespecting other communities for their subjective preferences, while trying to give it the objective veneer of "You've ignored the latest 'advances'!", is just pretentiousness.
I assure you, we would very much not. (The Rust team, anyway. I haven't seen that kind of nonsense out of the Swift team either, and it'd make even less sense, as Swift is technically garbage collected.)
But that doesn't mean that you can't discuss the tradeoffs, or that they're just subjective preferences. Someone who really understands Rust would never say that GCs are an anachronism, as they would be very much aware of the cost Rust pays for the ability to maintain safety when a GC can't be used. But that's not a subjective preference, either. A language like Rust knowingly sacrifices convenience and development speed to be able to target certain kinds of applications.
Can beef that up with some arguments? The creators of Go are not ignorant, they just decided consciously to create a language easy enough for recent graduates to be productive quickly.
From a science point of view, the only valid reason why clojure cannot compile to Go is that nobody cared to implement that.[1] Well, either that, or the people who considered doing that realized that compiling a programming language to a Chinese board game is equivalent to using an abacus and abandoned the idea .... ;)
Please. They didn't ignore the research they just decided against it. You might say that they were paying more attention to the last 30 years of experience.
I get that that the inventors of Go are respected people, but even they can mistakes.
It would be cool if Clojure was one and the same thing and then had a backend to support generic platforms - this way you would have ClojureJVM, ClojureNET, ClojureGO, whatever, with different levels of support and speed, but at least you could share the code. For example, maybe the Go version could have only :aot and write Golang sources.
You could make an analogous argument for C not being portable, since it uses an ecosystem of tools and libraries that might not be portable to all systems.
Of course, you will always have a set of points at which the language needs to call lower-level functionality, but you should keep that set as small as possible, to make the ecosystem easier to port.
And in the modern "client/server" world, you really need this, because systems are often heterogeneous, and you often need to move one piece of code to another part of the system which runs on a different runtime.