Designing a Language (2017)
Mood
thoughtful
Sentiment
positive
Category
tech
Key topics
programming languages
language design
compiler design
A set of notes on designing a programming language, covering various aspects of language design and implementation.
Snapshot generated from the HN discussion
Discussion Activity
Very active discussionFirst comment
1h
Peak period
84
Day 1
Avg / period
84
Based on 84 loaded comments
Key moments
- 01Story posted
11/15/2025, 5:44:43 AM
4d ago
Step 01 - 02First comment
11/15/2025, 7:10:44 AM
1h after posting
Step 02 - 03Peak activity
84 comments in Day 1
Hottest window of the conversation
Step 03 - 04Latest activity
11/15/2025, 5:14:37 PM
3d ago
Step 04
Generating AI Summary...
Analyzing up to 500 comments to identify key contributors and discussion patterns
1. Is there a need for the programming language?
2.If the answer to the previous question is yes, can I find enough people to help and enough resources?
3. If the answer to the previous question is yes, can we release a MVPin a reasonable amount of time?
4. If the answer to the previous question is yes, what is the chance it will gather a reasonable amount of users?
There are literally tons of programming languages that didn't make it. I wouldn't want to waste my and other people resources.
I love working on software, architecture, design but only if I see some use.
Of course, for other people, the journey is more interesting than the destination and they have fun hacking stuff just for the sake of it. They discover things and learn new stuff they wouldn't have learned otherwise. And this is a path at least as valid as the other.
1. Yes, as long as there are new machines that need programming, new programming languages will be needed. Today's top languages were built for the machines of the 1970, 80s, and 90s. Tomorrow's languages will be built for machines of today and tomorrow. As Alan Kay put it, if you want to invent a new language, first invent the machine of the future and then build a language for it.
2. No, you cannot. First of all, PL devs are cats, it's very difficult collecting them without financial compensation. So if your plan is to post a language and hope that people will come help you, you'll likely be disappointed. The problem is that everyone else interested in building PLs has their own itch to scratch, and they're not going to scratch yours without some compensation.
You might think "Well I can just raise money to do this", and you would be wrong. First, it's very hard to raise money for PLs. Usually you have to have come sort of cred to do it. I know of only 3 projects to have raised VC money for a PL project, and they each had some success before they had done so: Chris Granger (Light Table), Paul Biggar (CircleCI), and Chris Lattner (Swift/LLVM). Granger's project Eve raised $2M and ran out of money after 3 years; Biggar's project Dark also raised money, then fired all the devs when he realized he was burning cash too fast, then he slow-burned development for years, then he gave up and handed development over to someone else; and Lattner raised almost $100M for Mojo, which is probably going to end much the same way as Eve and Dark, but I wish them the best.
Anyway, the point is that you personally (no offense) don't have the profile to raise $100M like Lattner. $2M is not enough for a PL project. Lattner is keeping Mojo closed source for now because there's no good answer for how they're going to make enough money as an open source language to justify raising $100M.
And the reason it's so hard to raise money is because there's no money to be made. No one pays for PLs. No one pays for PL dev tools. They have to be open source or they're rejected by the dev community. The only ones these days who can reasonably pay for all of this with no potential revenue stream are giant corporations, who use the lang as a hook into their ecosystem.
3. Even though the answer is no, you yourself can still get an MVP off the ground in a pretty reasonable amount of time. It's never been easier to make a PL. The problem with PLs is building them is kind of like measuring the coastline; language projects are fractals -- there's an infinite amount of detail you can work on in any given direction. It's very easy for a language project to become a language + editor project, and it's easy for that to turn into language + editor + operating system if you're not disciplined. Plenty of PL devs have fallen into that trap.
4. Rounds to 0% chance. You'll be lucky if you build something that even you will use. Rather, most PL devs end up working on their language in some other language, because working on languages is what they want to do!
That said, it's still important to write languages that you understand no one will use. First it allows you to try new things that may good but unpopular. If PL devs only did what was popular with devs, PLs would go nowhere as a field.
Consider the so called "Hornet's nest" of programming languages [1], which is the tightly related cluster of imperative programming languages which have been the most researched and used over the last 50 years. There is a vaaaaaaaast design space outside that nest, begging for more language development. No one will use most of them, but it's important to understand what those languages might look like to maybe find some new ideas that work.
Also "didn't make it" is kind of an unfair judgement. Gaining popularity doesn't have to be a goal. In fact, it shouldn't be a goal if you want to have any fun at all. There's an infinite amount of work to be done, and if you're not doing it for you, you won't get far at all. That's really the only way to fail at this.
Good luck!
[1] https://tomasp.net/techdims/#footer=index,navigation;left=ca...
Yeah, whenever I encounter a new language, to see how serious they are, I take a look at their github commit history. Usually they are all green every day, there's a sort of obsessive compulsion behind working on these projects.
It's pretty amazing how the boundary of what a PL actually is has expanded. It's really the story of "If you give a mouse a cookie"....
Used to be back in the day you didn't even have to implement the thing (ISWIM). But if you give the people a programming language they're going to expect a compiler to use it. Then devs started expecting a whole standard library sometime after the 70s. By the 80s and 90s IDEs were all the range -- you needed to provide at least syntax highlighting for sure. A breakpoint debugger was starting to become standard expectation.
In the 90s - 00s, open source rose to prominence and communities of open source developers works to create robust community-driven language ecosystems, which then became an expectation for new langs. Quite the paradox there -- how do you create a community around a new language if the new members expect a community??
But once you have a robust package ecosystem, devs start expecting ways to manage it. So now you not only need a package manager, but also a package repository and all the issues which come with that.
Now with all these packages you also need to provide a robust build system to download them all, build each one, link the binary, and it should be compatible with all major operating systems, all major architectures, and of course the web.
Today, LSPs were the most recent "must have", before "AI integration" took over and now you need to have AI assistants that know your language and all the libraries.
All that before you even start talking about the language specifics. To be popular, your language must a) be severely limited in its "weirdness budget" (the degree to which you break from tratitional languages must be a small delta or potential users complain) b) be imperative-first c) and most importantly, be open source and charge exactly $0 for all of this.
That's why the quickest way to build a new and different language is actually to create a cult around it. If you're gonna make any money at all, it'll be in selling plushies of your mascot. I wrote a whole novel about that route here a couple weeks ago. https://news.ycombinator.com/item?id=45806741
https://www.npmjs.com/package/wang-lang
- this new language looks and behaves exactly like javascript, except it doesnt have "eval" and "new Function", so it is CSP safe. That's the only difference. I wanted to execute dynamically generated code in chrome extension
- llm did most of the work of creating a nearley grammar and associated interpreter (whole thing is bundled, nearley is not a final dependency), elaborate tests make this quite sane to handle
- took me about total of 1 weeks for the initial mvp to try out, and then have been fixing bugs and inconsistencies with javascript behavior, about 1 day a month of effort
- mostly 0
The only reason to create was I couldnt find something similar and it was low effort thanks to llm
I also created another even smaller DSL you can say
https://www.npmjs.com/package/free-text-json-parser
It parses json embedded in plain text
> llm did most of the work
> it was low effort
I really wouldn’t trust its supposed safety.
linter would help me find and avoid usages of eval.
Right question is to design own linguistic language common between computer and across human.
- Raku has built in Grammars so it is a great place to do early iteration of your parser
- Raku is objects and type classes all the way down (as explained here https://gist.github.com/raiph/849a4a9d8875542fb86df2b2eda89296 )
- RakuAST development is well advanced (use v6.e.PREVIEW) with the Slangify module to accelerate development of sub languages (Slangs)
IIRC Perl 6 wanted to expand or morph into something better, spent a ton of time on it, and the community in general rejected it hard.
So now we have this dangling language that's shunned by its own community, regardless of its merits. Weird place to be in.
Well Raku is not shunned by the very warm and welcoming Raku Community … https://raku.org/community
I think modern RLHF schemes have models that train LLMs. LLMs teaching each other isn't new.
My knowledge is limited, just based on a read of https://huyenchip.com/2023/05/02/rlhf.html though.
You provide a goal as a big reward (eg test passing), and smaller rewards for any particular behaviours you want to encourage, and then leave the machine to figure out the best way to achieve those rewards through trial and error.
After a few million attempts, you generally either have a decent result, or more data around additional weights you need to apply before reiterating on the training.
It’s the other weights that are harder. You might want execution speed to be one metric. But how do you add weights to prevent cheating (eg hardcoding the results)? Or use of anti-patterns like global variables? (For example. Though one could argue that scoped variables aren’t something an AI-first language would need)
This is where the human feedback part comes into play.
It’s definitely not an easy problem. But it’s still more pragmatic than having a human curate the corpus. Particularly considering the end goal (no pun intended) is having an AI-first programming language.
I should close off by saying that I’m very skeptical that there’s any real value in an AI-first PL. so all of this is just a thought experiment rather than something I’d advocate.
With a AI-first language, I suspect the primitives to be more similar to assembly or WASM rather than something human readable like Rust or Python. So the amount of pre-training preparation would’ve a little easier since syntax errors due to parser constraints.
I’m not suggesting this would be easy though haha. I think it’s a solvable problem but that doesn’t mean it’s easy.
2. LLM generates python solution and seperate python test (as in python test calls code as black box process so it can test non python code)
3. Agent using skills etc. tries to write new language let's call it Shark.
4. Run Shark code against test. If fails use agentic flows to correct until test passes.
5. Now have list of challenges, working code (maybe not beautiful) for training.
A bit of human spot checking may not go amiss!
Fun
Even the successful ones are often pointless variations on a theme. Ruby, perl & python don't all need to exist for example, as they essentially do the same thing, about as poorly. Now python has won we should just drop the others
Different languages excel at different things. There shouldn’t be a “one size fits all” otherwise we’d be writing software in FORTRAN and assembly.
And designing a language is a good exercise if purely from an academic perspective. Eg you learn how to write parsers, and a bunch of PL theory that we take for granted when just being a consumer of a programming language.
Not everything needs to be done with global domination in mind.
I started programming assembly in 2025 for 6592 and Z80 cpus and believe me: it is fun and IMO actually easier then lets say learning Haskell or JS from scratch.
Assemblers with macros are amazingly simple.
I think that's fair. Even if you are just doing a hobby language there are plenty of unexplored niches, e.g. that compile-to-shell language I've forgotten the name of.
Which is why I said we’d still be using FORTRAN.
Languages that start out radically different don’t tend to gain momentum. Whereas languages that are familiar tend to grow and introduce new ideas.
Nothing is invented in a vacuum.
Also I completely disagree that one shouldn’t create a hobby project need to be innovative. Sometimes people do create things just because they can. And it’s a good thing too because otherwise we wouldn’t have half the open source software available to us today. Many of which was originally intended for personal use, including Linux.
The problem these days is we’re so brainwashed by stories of unicorn start ups pumped with VC money that now everyone thinks every hobby needs to has a viable business plan underneath. It’s like people have forgotten how to play for fun.
So people should go out and create new programming languages. The worst that would happen is they learn to be a better programmer in their day to day language.
The bad thing is the uncanny valley. Popular enough to fragment the niche and add tech debt, not big enough to win and defragment the niche, not innovative enough to make any real positive difference beyond personal tastes.
You think that it will be better for people to stop developing programming languages, but how do you think new programming languages will appear? Will they be better? To be better you need to try new ideas and to look how they work in practice. And very probably your ideas will fail when tested by reality. How to organize the activity to test new ideas without risks of fragmentation? We can't keep our results of developing an experimental language in secret, because then everyone will do the same, and we would need to test all ideas without any hope to learn from other's experience. U'see, the very activity of testing new ideas is almost a synonym for fragmentation.
I don't think you have any viable alternative to "every programmer should create its own programming language".
That said, I think it's okay, really, to allow people to think they are working on the next definitive thing; not because they are actually solving the "problem" of fragmentation once and for all, but rather bringing to bear the passion to see an idea to fruition. New concepts and approaches should not be considered to be inherently counterproductive or dangerous.
In broad terms, the spectrum of "what programming languages could look like" seems pretty well covered, yet I still have ideas that aren't represented. And then, that spectrum would still have looked pretty well covered, say for example, shortly before the dawn of Rust.
Build something for fun. Build it for yourself. If people want to use it then they’ll use it. But more likely they won’t.
The only thing I would advise against is building something expecting other people to use it.
As I said elsewhere, not enough people these days build things for their own pleasure without any expectation nor desire for it to be used by anyone else. Sometimes just doing something for yourself is its own reward.
And it’s far more likely that DIY projects will teach skills that you can then directly contribute to established projects with, rather than DIY projects fragmenting those established communities.
I just parse my language, translate it to C, and use C compiler errors.
I don't add new semantics, I just add many things like strings, map, etc to make it usable and fast.
I don't know if it's a good idea and how difficult this will be.
Its a valid approach.
It is a great idea, if you want to learn about languages!
(But if money is your goal, you may want to reconsider)
How does one create a new spoken/written language ?
I could and have written a few toy interpreters, but I have no academic or industrial background (on the matter of language design), so it is useful to know why they put some features into a language, and why they don't. It is actually one of the most confusing parts of writing an interpreter for a toy language -- in all of my projects I simply pick a subset of an existing language I know about, e.g. Python or C.
To get the best historical sense, pick a language feature that has evolved in several steps over a number of years, e.g. async/await, type annotations, the GIL, etc.
For ordered dicts, for instance, which is topical because of [1] posted two weeks ago, the relevant PEPs are:
- PEP372: The original collections.ordereddict proposal in Python 3.1 [2]
- PEP468: Making kwargs ordered in Python 3.6 [3], which also made standard dicts ordered.
[1] https://news.ycombinator.com/item?id=45756058
And that’s probably a good thing.
So you want to design a programming language (2017) - https://news.ycombinator.com/item?id=30481035 - Feb 2022 (58 comments)
Programming Language Design - https://news.ycombinator.com/item?id=27895124 - July 2021 (1 comment)
I'd also highly recommend that anyone interested in this kind of thing listen to all three of the Dynamic Languages Wizards Series panels from 2001: runtime [2], language design [3], and compilation [4]
Note that though these are videos, there isn't that much compelling in the visual portion, you could easily rip them to audio files and lose little.
[1] https://libarynth.org/fifty_questions_for_a_prospective_lang...
[2] https://www.youtube.com/watch?v=4LG-RtcSYUQ
1) You are NOT serious (in effort to be invested, resources, knowledge), then don't do it. 2) You are MEH serious, then probably design some DLC in Lua or similar, will serve your case 99%. 3) You ARE serious, then go for it. Chances are that you might even post it here one day, but also almost no one will ever use it apart from some crazy fans.
Adding construct like IF or variables is naturally next step but you will have code in place and idea where to put it and how approach it.
I learned a lot about JVM runtime, how Zig is parsing itself, how Lua represents values... Too many good rabbit holes to fall in.
I've recently found that moving from a linear memory model to a stack-based model creates a dramatic improvement in performance. The program tape is still linear, but the memory is a stack interface. It seems the search space is made prohibitively large by using pointer-based memory access. Stack based makes it a lot easier to stick arbitrary segments of programs together and have meaningful outcomes. Crossover of linear program tapes does not seem practical without constraining the memories in some way like this.
I believe there are far more interesting stuff to learn about these languages, like the whole category of runtimes could have been mentioned, which can directly affect the language design itself (e.g. having GC vs some language feature for managing memory, open vs closed world model, having an async feature in the language or let the runtime handle it, etc)
23 more comments available on Hacker News
Want the full context?
Jump to the original sources
Read the primary article or dive into the live Hacker News thread when you're ready.