- the great concurrency model
- the REPL
- being able to edit with Paredit,
- it's a great language
- It's also amazing for hiring devs: there is a niche pool of devs who want to use clojure, and it's easy to hire them since demand for them is low. (Vs the thousands of Rails companies who are competing for a much larger pool)
- being able to use clojurescript, in particular using Om (we use a pattern which is similar to the Elm Architecture, and it's great for writing frontends with).
The downsides are:
- mostly related to Java. Startup time is shitty (which is why you use the REPL a lot).
- Libraries are often unmaintained (though often it doesn't matter, since they are typically very composable and don't need changing).
- The web framework story is poor too - it's not nearly as advanced as the rails and nodejs worlds - you have to compose different libraries which is sometimes good and sometimes bad.
- Finally, Rails encourages good web-app best practices: you can only really scale Rails horizontally, so you start with queues and horizontal scaling and workers. With Clojure, it's tempting to run things in threads and pools since it's so easy to do. But it scales poorly, and leads to concurrency bugs you wouldn't get in the Rails world.
Now, because you're using AWS, or a new deploy, or something else, the server dies. What happened all that stuff running in the background? Because it was built with clojure primitives, it never got saved anywhere, like it would if you had used a queue and workers.
The summary is: Clojure's awesome stuff encourages you to use it, but it's actually a bad pattern for horizontally scalable apps. Rails - lacking the cool concurrency stuff - forces you to have better habits, and makes them easy.
We also use Clojure for the "matcher," which is a webserver that receives a request every time someone visits one of our publishers' websites. The matcher must decide very quickly which ads are relevant and decide which one to show. There is some online machine learning involved. Porting this latency-sensitive service to Clojure (from Python) has been great: the Clojure version is much more performant, it is easier to change, scale, and monitor, and it fails more gracefully.
Our data pipeline moves events for pageviews, ad impressions, clicks, etc. through kafka queues between AWS regions, archives them, and indexes them into ES. It's all Clojure.
Finally, we use Clojure and Datomic for the database where campaigns, creatives, etc., are configured.
Oh, and we're hiring.
- Dependency management is so easy and portable.
- I can write really concise code while keeping it readable (and dare I say aesthetically pleasing) with ParEdit. Compare this to Scala which tempts you to write gnarly and dense one liners. In clojure these are so unreadable that you feed compelled to indent them properly.
- Clojure feels like playing with Legos. Each individual piece is small and easy to understand, and it's easy to build something bigger by stacking them together.
- I have never once missed OOP
- ClojureScript is cool but for a production app I would rather not use it. Not enough availability of libraries and JS interop is not intuitive. When I'm debugging CLJS, half the time the stack trace is unreadable transpiled JS
- Proto Repl (atom plugin) does not consistently work right for me. I still have not found a good REPL workflow so I am frequently restarting REPLs which is just sad.
For me, deploying production Clojure has been fantastic. Is it a panacea? No, definitely not. If you like typing, stay clear. Spec is cool. Very cool. But it's not static typing. And, if thinking in s-expressions doesn't come natural within a reasonable amount of time, consider Elixir instead.
I do, however, prefer dynamic languages, and I frankly LOVE LISP, so it suits me. The JVM has never been a hindrance; in fact, I appreciate the comfort of knowing there's a deep ecosystem of Java that can be tapped, which I have (easily!), in extreme situations like PDF manipulation, or whatever.
Clojure also meets its performance claims – it's fast. My company doesn't carry behemoth level request loads, so I can hardly provide impressive metrics, but decently considered code will execute plenty fast with room to spare. Although we're light users, Clojure's concurrency features are nice too. For most async stuff, forget background job processes – use futures. core.async is cool, but overkill for most web stuff. We've built a bare-minimum job processor for must-finish tasks, but that got done in about 100 lines of code.
My major piece of advice, if you plan to use Clojure for the web, is to get comfortable building custom homes for your web apps. Not in the sense that you'll HTTP stuff from scratch – no, there are excellent libraries for all of the critical stuff. Redis, AMQP, ORMs and SQL libraries of all flavors, caching, routers – everything you'd want for the web is there.
Just be mindful of the fact that your web app will look nothing like your neighbor's. There are NO integrated web frameworks. Not successful ones, at least.
And, I think, for good reason. Clojure is not the average Clojure developer's first language. There was a significant influx of Rails devs several years ago. Clojure is bereft of black magic. A lot of Clojure libraries are so small, and so sharp, you can easily unearth their features by reading a few source files. Which, of course, is good and bad. The side effect, being, very little of the documentation in the community is geared towards beginner programmers. Starting out, my single biggest frustration in the Clojure community was the knowledge library authors assumed you brought to the library.
I'd compare the ecosystem – loosely – to NPM's, but without the cruft. You build your web app up from small components (Component and Mount are awesome!... eh, except for the fact that they are, as are so many Clojure things, original material that takes time to digest...) that easily fit together. It's a breath of fresh air knowing almost all libraries live in harmony.
So, the backend Clojure story, in my view, is great – if you don't mind being on your own once in a while.
The front-end story, however... well, is a mixed bag, I think. Yes, ClojureScript is making huge strides in being futuristic and compiling to negative kilobytes an' all. But some of its highest profile projects are sadly proving impenetrable. Om and Om.Next are total non-starters for small to mid-sized projects (YMMV) as their documentation is so beyond comprehensible. I _have_ made a real project in Om.Next – with tremendous effort – but I've been developing in Clojure for nearly five years. And, when you can't remember how something works a day after you wrote it... that's trouble.
Back to Clojure proper, one thing I enjoy immensely is seeing how far I can take a program in a single file. Not always practical, but it's fun knowing you can scan a file (often starting at the bottom) and quickly grok it. Do know, however, reading Clojure (or any LISP, that is) is sometimes harder than writing it, especially in a large codebase. It's here, I think, where statically typed languages have an edge on Clojure – refactoring is more dangerous: you can never fully tell the shape of data passing between functions.
Spec has helped, for sure. For simple, obvious tasks, take a risk. On mission-critical-can't-break-type stuff, however, you'll be grateful for spec. No, it's not the anxiety-free experience you hear about in Elm or Haskell, but geez, at least you can get fresh code working fast. It's here where I continue to believe dynamic languages like Clojure have a permanent place in history – solving problems iteratively, with little pomp and ceremony, without knowing the end from the beginning. And saying to yourself, "Yes, this works, and while not perfect, I made it quickly, and it's extensible and maintainable."
The most beautiful thing about writing a production app in Clojure is this: you can build a big app from tinier programs. I don't mean microservices. I mean small, easily divisible namespaces in a single project that talk to each other via messages. Like, really, you can easily treat separate namespaces as if they're different actors in a plug and play network.
A talk I heard recently (can't remember where) suggested writing your big program from small programs, small enough that any individual program can be rewritten in a week. Clojure does this for us. When I started out, being unfamiliar with functional paradigms in general, I brought a lot of tightly coupled OO practices that I had to unlearn. Unfortunately, "real world" Clojure tutorials are scarcer than they ought to be. But, unlike in my Ruby and Java days, there are _very few_ things that need to be completely rewritten in Clojure. The core language is simple enough that almost all Clojure code resembles itself.
Back then, you did lein new compojure app then you had to figure out what to change in your handler file to get the thing up and running. Today, build, run, and everything just works.
Security has seen a massive improvement, ring comes with CRSF by default, and the libraries have improved dramatically. The Bedra talk was on point, but no longer as relevant.
With all that said, I don't mind using it in production because it is just me alone working on my own projects. I've interviewed with a few places that use it in prod and they are a) very happy using Clojure and b) are on polyglot stacks. I'm not aware of anyone using Clojure completely alone.
The inherent problem with Clojure is that, rightly or wrongly, the perception from many programmers is that it is too hard to learn, so there is no selection of programmers, and the perception matches from the hiring companies. Companies will be much more picky on who they even talk to. You also have to consider that the common stacks are large framework / ORM / a lot of things baked in / etc. Clojure strips away all of this magic. It appears that many programmers just aren't comfortable working this close to the metal (have no opinion on this and don't want to get into it).
Since there is a dearth of programmers, I would never build a project that I intended to hire more people to work with, so really, I'd choose any of the more-known languages to build very large projects.
My recent project is here: https://www.butternotes.com/
I'm using Clojure / Compojure and using a lot of XML generating code for this. I'm happy with the speed of development and happy with how easy it is to make major changes and expand them globally. So, what you lose in defaults you gain in flexibility, which is the trade-off you are required to take on when working with Clojure. There is no analog to Rails, Django, etc.