Posted by smartmic 11 hours ago
One thing I built: defun https://github.com/killme2008/defun -- a macro for defining Clojure functions with pattern matching, Elixir-style. Still probably my favorite thing I've open sourced.
I find myself missing Clojure-style multimethods in most languages that aren't Clojure (or Erlang); once multimethods clicked for me, it seemed so blatantly obvious to me that it's the "correct" way to do modular programming that I get a little annoyed at languages that don't support it. And core.async is simply wonderful. There are lots of great concurrency libraries in lots of languages that give you CSP-style semantics (e.g. Tokio in Rust) but none of them have felt quite as natural to me as core.async.
I haven't had a chance to touch Clojure in serious capacity in awhile, though I have been tempted to see if I can get Claude to generate decent bindings for Vert.x to port over a personal project I've been doing in Java.
I had an idea about writing something similar, but for multimethods, but never got around thinking it through and trying it out.
The way defmulti and defmethod work is that they do a concurrency safe operation on a data structure, which is used to dispatch to the right method when you call the function.
My hunch is that it should be possible to do something similar by using core match. What I don't know is whether it's a good idea or a terrible one though. When you're already doing pattern matching, then you likely want to see everything in one place like with your library.
- syntax is hard to read unless you spend a lot time getting used to it
- convention for short var names makes it even harder
- function definition order makes it even harder
- too dynamic for most people's taste
- no type safety
- the opposite of boring
- no clear use case to show it clearly beating other languages
- niche with small community and job market
- JVM
For all those reasons its a hard sell for most imo.
We mostly hired people with no previous Clojure experience. Majority of hires could pick up and get productive quickly. People fresh out of college picked it up faster. I even had a case of employee transitioning careers to S.E., with no previous programming experience, and the language was a non issue.
I can't remember an instance where the language was a barrier to ship something. Due to reduced syntax surface and lack of exotic features, the very large codebase followed the same basic idioms. It was often easy to dive into any part of the codebase and contribute. Due to the focus on data structures and REPL, understanding the codebase was simply a process of running parts of a program, inspecting its state, making a change, and repeat. Following this process naturally lead to having a good test suite, and we would rely on that.
Running on the JVM is the opposite of a problem. Being able to leverage the extensive JVM ecosystem is an enormous advantage for any real business, and the runtime performance itself is top tier and always improving.
The only hurdle I could say I observed in practice was not having a lot of compile time guarantees, but since it was a large codebase anyway, static guarantees would only matter in a local context, and we had our own solution to check types against service boundaries, so in the end it would've been a small gain regardless.
I have to push back on this one, respectfully.
Clojure is easily the most boring, stable language ecosystem I’ve used. The core team is obsessed with the stability of the language, often to the detriment of other language values.
This attitude also exists among library authors to a significant degree. There is a lot of old Clojure code out there that just runs, with no tweaks needed regardless of language version.
Also, you have access to tons of battle tested Java libraries, and the JVM itself is super stable now.
I won’t comment on or argue with your other points, but Clojure has been stable and boring for more than a decade now, in my experience.
Have you worked for a company that hasn’t created its own, as you put it “mini language”?
Have you worked for a company that doesn’t indulge in over engineering, over abstraction and hidden cost?
Do you actually do programming for a job at all?
because programmers suck we should make tools that make it easier for them to suck?
Creating these mini DSLs is something that requires a lot of thought and good design. There is a danger here as you pointed out sharply.
But I have some caveats and counter examples:
I would say the danger is greater when using macros and far less dangerous when using data DSLs. The Clojure community has been moving towards the latter since a while.
There are some _very good_ examples of (data-) DSLs provided by libraries, such as hiccup (and derived libraries), reitit, malli, honeysql, core match, spec and the datalog flavor of Clojure come to mind immediately (there are more that I forget).
In many cases they can even improve performance, because they can optimize what you put into them behind the scenes.
There’s a rule of thumb: write a macro as a last resort.
It’s not hard to stick to it. In general, you can go a long, long way with HOFs, transducers, and standard macros before a hand-rolled macro would serve you better.
That’s pretty much exactly the opposite of how I always felt. Perhaps because I’m not a programmer by education, I always struggle to remember the syntax of programming languages, unless I’m working in them all the time. After I return to a language after working in other languages for a while, I always have difficulties remembering the syntax, and I spend some time feeling very frustrated.
Clojure and Lisps more generally are the exception. There is very little syntax, and therefore nothing to remember. I can pick it up and feel at home immediately, no matter how long I’ve been away from the language.
I agree with the short variable name convention, that's annoying and I wish people would stop that.
Everyone complains about a lack of type safety, but honestly I really just don't find that that is as much of an issue as people say it is. I dunno, I guess I feel like for the things I write in Clojure, type issues manifest pretty early and don't really affect production systems.
The clearest use-case I have for Clojure is how much easier it is to get correct concurrent software while still being able to use your Java libraries. The data structures being persistent gives you a lot of thread safety for free, but core.async can be a really nice way to wrangle together tasks, atoms are great for simple shared memory, and for complicated shared memory you have Haskell-style STM available. I don't remember the last time I had to reach for a raw mutex in Clojure.
Good concurrency constructs is actually how I found Clojure; I was looking for a competent port of Go-style concurrency on the JVM and I saw people raving about core.async, in addition to the lovely persistent maps, and immediately fell in love with the language.
Also, I really don't think the JVM is a downside; everyone hates on Java but the fact that you can still import any Java library means you're never blocked on language support. Additionally, if you're willing to use GraalVM, you can get native AOT executables that launch quickly (though you admittedly might need to do a bit of forward-declaration of reflection to get it working).
That's fair if you're looking at it from a performance perspective.
Not entirely fair if you look at it from a perspective of wanting fast feedback loops and correctness. In Clojure you get the former via the REPL workflow and the latter through various other means that in many cases go beyond what a typical type system provides.
> the opposite of boring
It's perhaps one of the most "boring in a good way" languages I ever used.
this is what i meant by that: https://news.ycombinator.com/item?id=47614353
While it's amazing once you've learned it, and you're slurp/barfing while making huge structural edits to your code, it's a tall order.
I used Clojure for a long time, but I can't go back to dynamic typing. I cringe at the amount of time I spent walking through code with paper and pencil to track things like what are the exact keyvals in the maps that can reach this function that are solved with, say, `User = Guest | LoggedIn` + `LogIn(Guest, Password) -> LoggedIn | LogInError`.
Though I'm glad it exists for the people who prefer it.
you absolutely do NOT need to learn paredit to write lisp, any modern vim/emacs/vscode plugin will just handle parentheses for you automatically.
that said, if you do learn paredit style workflow - nobody in any language in any ide will come even close to how quickly you can manipulate the codebase.
It's like seeing that a movie is playing at the theater so you show up only to sit down next to people to explain your qualms with it, lolz. Sometimes you need to let others enjoy the show.
The OP of this thread even said all that needed to be said "The learning curve is steep but very much worth it" yet we're trapped in this cycle because someone had to embellish it with a listicle.
I take my post back.
Plus to be fair we're having this discussion in the context of an article from 2021 that just rose to front page of HN, only to repeat the same set of pros we've been hearing about Clojure for ages (code as data, repl, etc).
Probably should expect some dissenting opinions.
it's not even in the top 50 here: https://www.tiobe.com/tiobe-index/. Lisp is 26.
For some reason I doubt this is in any way representative of the real world. Scratch, which is a teaching language for children, bigger than PHP? Which is smaller than Rust? Yeah, these are results you get when you look at the Internet, alright.
Besides, the community and ecosystem is large enough that there are multiple online spaces for you to get help, and personally I've been a "professional" (employed + freelancing) Clojure/Script developer for close to 7 years now, never had any issues finding new gigs or positions, also never had issues hiring for Clojure projects either.
Sometimes "big enough" is just that, big enough :)
Every problem people face is "not a problem" or "actually a good thing" or, maybe if all else fails we can make users feel bad about themselves. Clojure is intended for "well experienced, very smart developers". Don't you know, our community skews towards very senior developers! So if you don't like something, maybe the problem is just that you're not well experienced enough? Or, maybe what you work on is just too low-brow for our very smart community!
How about just "different"? Turtle want to teach everyone to program, that's fine, just another way of building and maintaining a language. Clojure is clearly not trying to cater to the "beginner programmer" crowd, and while you might see it as "unhealthy attitude", I'd personally much prefer to realize having many different languages for different people is way better than every language trying to do the same thing for the same people. Diversity in languages is a benefit in my eyes, rather than a bad thing.
Type's are for compilers ;) jk. I'm fully lover or type's but removing the constraint is easy in clojure. teams resist.
<3 the opposite of boring.
Not to mention in a post AI world, cost of code generation is cheap, so orgs even need even fewer devs, combine all this with commonly used languages and frameworks and you need not worry about - "too valuable to replace or fire".
Having said that - there may be a (very) small percentage of orgs which care about people, code crafting and quality and may look at Clojure as a good option.
This is only true if you assume C-like syntax is the "default."
But regardless of that, I'd argue that there's much less syntax to learn in LISPy languages. The core of it is really just one single syntactic concept.
Even among lisps this has been problematic, you can look at common lisp's LOOP macro as an attempt to squeeze more structural meaning into a non-S-expression format.
The syntax argument is such a tired argument. With LISPy language there is almost zero syntax, it's pretty much executable AST.
Because of this, formatting matters a lot, but I don't think that's too different than other languages.
If you think LISP is hard to read, you are someone who could most benefit from branching out to a non-Algol lineage language.
Also, the little syntax present is pretty much timeless. Learn once and its yours for the next 50 years.
To be pedantic, this isn't quite correct. Syntax isn't countable like that. What S-expressions are light on is production rules. At their most basic they have IIRC 7 production rules but there are absolutely 0 languages based on s-expressions which are that simple, since it doesn't give you anything like quasiquotes, vectors, Lisp 2 function resolution, etc. Reader macros make matters much worse.
What we can say is that they are constructively simple, but not particularly unique in that. Once you get into real sexpr languages they aren't simpler than horn clauses, and are constructively more complex than languages like Brainfuck and Forth.
It's repeated a lot because it's true. The collective developer world has decided that LISP syntax is not the preference. Good if you prefer it, but you're the in the overwhelming minority.
Random example i just found via github explore: https://github.com/replikativ/datahike/blob/main/src/datahik...
You probably love it but to me it looks like a wall of text. Sure I can figure out what it does, but it gives me a headache.
To use the right words: it’s not a syntax issue, it just looks unfamiliar to you.
To this day I have to look up whenever I get back into clojure what the "syntax" is of ns, require, import, etc.
but i bet if you sat down a junior developer not yet entrenched in any style yet, they'd be able to grok lisp code MUCH faster than the intricacies of syntax of the other alternatives ¯\_(ツ)_/¯
That is, prose is good for entertainment, but less so for conveying information, even less so for exactness.
I nominate this sentence for "best inadvertent freestyle rap on Hackernews".
The syntax is the best in the world (how computer's really operate?) but it's always been a pain to setup the tooling for me. I'm dumb like that. Now with AI it's become super easy to get back into the REPL and I'm in heaven.
Totally moving it back into workflow and proposing to bring it back into the dayjob.
I don't know exactly what you mean by this, but Clojure syntax is not really anywhere close to how computers actually process instructions.
Clojure is very nice though.
I love the idea of an AI integrated repl (like what Jeremy Howard and team have done with solveit), it's far more in line with my preferred vision of the AI augmented future of coding. Less "swarm of agent" more, "learn with the agent".
Setup is:
- nvim --listen /tmp/nvim — starts Neovim with a socket Claude can connect to
- /mcp in Claude Code — enables the Neovim MCP server, gives Claude direct control of Neovim
- lein repl in the nvim terminal
- Claude reads .nrepl-port, runs :ConjureConnect — REPL is live!
The loop is so dope:
- Claude writes code directly into my .clj files
- Then evals it into the running process via Conjure
- Sees the result in the REPL, iterates if wrong, all in the same conversation turn
- wrap-reload middleware means the web server hot-reloads changed namespaces on the next request
Feel like a holy grail to me to back in the hot seat with repl driven but I can drop in and figure things out... all on the JVM. Madness!
Something where I feel Kotlin did better.
For me the best way to introduce something like this is that I can actually start with small software increments on a Spring Java project.
no amount of ide smartness or agentic shenanigans is going to replace the feeling of having development process in sync with your thought process
or toyota releasing a game engine: https://www.theverge.com/games/875995/toyota-fluorite-game-e...
(okay it's a design system, not so much a framework, but still)
We have a codebase at work that was "stuck." We've consistently done minor library upgrades, but no major upgrades in several years, and was recognized as a major piece of technical debt / minor disaster for almost two years, in that we urgently needed to dedicate an engineer to it for a month or more to bring it up to date. We also suspected that framework upgrades would improve performance enough to save us a little bit in operating costs. I got curious, created a branch, and threw Claude at it. Claude knocked it out in a couple of days while I mostly worked on other things. Then we dedicated several engineer days to doing extra manual testing. Done and deployed. Now we're ready to experiment with giving it less resources to see if the performance improvement holds up in practice.
This codebase was only about 200k lines of code, so probably smaller than yours. Really curious how it would go with a larger codebase.
EDIT: Claude may only have taken a couple of days because I was only checking in occasionally to give it further instructions. I don't know how fast it would have been with my complete attention.
and the benefit of lisp comes from the fact that every expression is fully delineated by parentheses, so you don't need to select any text, find the start or the end of any syntactic construction, you just eval the form you're standing at and see the result
or just as easily you can wrap that form, bind the locals to test values and eval that to see what changes
I think debuggers are just better anyways. When I got into the weeds trying to do this interactive REPL workflow on an actually running webapp it was a big mess. You have to write custom code all the time to basically capture data and save it into global variables, so that you can inspect them (as opposed to a debugger where you can just set a breakpoint and inspect all the variables in the middle of the precise spot in the request-response cycle).
I think maybe this was ahead of its time when it came out in 2008 or got popular in 2010-2015, but nowadays I am not seeing what's specifically more productive about the Clojure REPL than the interactive development experiences available in Laravel, Python, JS, etc.
If it was like a choice, Clojure has my favourite reverse debugger flowstorm which also exposes the runtime information of your program programmatically
So you can code cool visual programs against the runtime values of your first program
Like I built an emulator at work that simulates our program in production as a flowstorm plugin then you can step through the program frame by frame
I love taking the guess work out of production issues, just get the computer to show you exactly how everything went down
It's like a rewindable movie made up of thousands or millions of frames of the execution of your program
That's where "code is data" shines and you can manipulate it at will. For the rest, I'm glad I have "code as text" in my editor, where I can see and inspect it far more easily than one line at a time.
All that said, 90% of the time you still just eval a bit of a code to see what happens and that's the same between the two languages.
(you can theoretically pass "reload": true (or similar option) in launch.json for auto reload, tho I haven't felt the need to use that in my workflows.)
The cycle is:
1. Write production code.
2. Write some dummy code in the same file (fake data, setup).
3. Evaluate that dummy code. See what happens.
4. Modify code until satisfied.
"Typical REPL" workflow: Have one editor open, have one REPL open, have one terminal open that runs the application. One change is typically: Experiment in the REPL window, copy-paste into your editor, write tests, restart application (lose all state), setup reproduction state, test change. Or something like this.
In a Clojure REPL workflow, you'd do something like: Have one editor open, this starts the REPL and often the application in the background too. One change is typically: Edit code, evaluate that snippet of code (which sends it to the REPL and the running application), write tests, evaluate them too in the editor, if you're happy, hit CTRL+S and you're done. Application still has the existing state, no restarts needed and you essentially never have to leave the editor window/pane.
Of course, others might have slightly different workflows, but for myself and many (most?) other Clojure developers I've observed in the wild, this is pretty much the standard.
* Clojure is built different in terms of hot code reloading
* the REPL is its own application process in languages Ruby or Python, but in Clojure it’s sortof a client for the system
Is that right? Is there more to it?
So in a JavaScript server, you might have the database connection set in some config/thing you pass around to request handlers to use, and if you change a request handler, you typically stop the entire application, then start it again. Then the database connection (and the entire config in fact) would need to again connect.
In a Clojure application, first you either start the repl from the server, or start the server from the repl, then you might instead keep the database connection in an "atom" that you keep around for as long as you have the repl/server running. And instead of restarting the process when you change a handler, you'd only change the handler.
And yeah, the libraries and ecosystem at large is built with this (mostly) in mind for everything. The language also supports this by letting you "redefine" basically anything in a running Clojure application, which would be harder in other languages.
I've done some experiments years ago for doing the same in JavaScript, and it kind of works, but every library/framework in the ecosystem is not built with this in mind, so you'll be writing a lot of your own code. Which, now with LLMs, maybe is feasible, but can't say it was at the time exactly.
The fun way to get a feel for lisp machines is emacs, it's so easy to fall of a language and especially hand-coding in a language if you don't have to.
Personally, I find waiting more than 200ms unacceptable and really < 50ms is ideal. When the feedback is very small, it becomes practical to save the file on every keystroke and get nearly instantaneous results with every input char.
An example of me solving an Advent of Code with clojure and repl. You can see i never interact with the repl directly, I just send code to it via my editor and get results inline.
1. REPL is automatically compiled into running systems 2. Great hot-reloading support
So it's generally very easy to "poke" at a running system, and the whole dev process assumes you will do this.
TBH, these days it is largely possible in a C++ debugger. Less so 10 years ago, though.
there's a detailed explanation here: https://youtu.be/Djsg33AN7CU?t=659
https://clojure.org/news/2020/02/20/state-of-clojure-2020
The most recent report: