> Usability is similarly ill-defined and hard to measure.
Human factors are very well studied and standardized, and there is a well-established discipline called "Human Factors Engineering", which also provides established test and evaluation methods. Human Factors research is considered solid and well-established because it has been built on rigorous experimental psychology and engineering principles developed over more than a century, with systematic methodology and empirical validation. Even if much of it is unknown or ignored by computer science or programming language design, there are many disciplines where Human Factors Engineering is critical (see e.g. ANSI/AAMI HE75).
Usability is therefore neither ill-defined nor hard to measure. Several ISO 9241 series standards address textual and command-based interaction directly relevant to programming language assessment. ETSI EG 202 116 provides extensive human factors guidelines for command language style. ITU-T Recommendation Z.361 provides guidelines for Human-Computer Interfaces in telecommunications management, incorporating the ISO 9241-10 dialogue principles. ISO/IEC TR 24772 addresses programming language vulnerabilities specifically from a human factors perspective.
E.g. Ada did have substantial human factors considerations documented in its design rationale, directly contradicting the notion that such principles don't apply to professional developers or programming languages. It's rather that computer science seems to continue ignoring established fields ("Not invented here"). Human factors in software development have been overlooked by researchers in the software engineering and development research areas, despite their obvious relevance. So what is lacking is primarily interest or willingness, not the fundamentals and means.
Honest question: given the relatively niche status of Ada compared to other systems programming languages, and similar diminished levels of popularity for other languages mentioned in this thread (Pascal, Perl), does this really prove that these principles are particularly effective for programming languages? I understand that this is a relatively small sample size, but I feel like that could be just as well used to argue that other factors might have been more important to the successes of languages designed like Ada than the more rigorous approach to human usability. It doesn't feel particularly obvious to me that we know these principles actually work well for programming languages but choose to ignore them rather than them not fitting the domain particularly well at all, or maybe something in between where we don't really know whether they apply well yet or not, and more attempts at using them might not actually be as successful as one might hope.
I didn't claim that usability features were "important to the successes of languages designed like Ada". But Ada at least explicitly considered human factors in their design (even if mostly based on expert judgment and established principles, not practical studies), which also seems very appropriate given the criticality of most applications written in Ada. But as my ergonomics professor at ETH Zurich, Helmut Krüger (successor to the renowned Étienne Grandjean), used to say: people get used to even the most ergonomically terrible systems. The "level of suffering" experienced by most people is probably simply not great enough to systematically take such aspects into account. But there are still industries where it is important to reduce the human tendency to make mistakes by taking appropriate measures. Ada was created for such an industry from the very beginning.
> I didn't claim that usability features were "important to the successes of languages designed like Ada".
This kind of seems like it's focusing too much on my exact word choice and less the actual intent of my question behind it. The question I have is why following established principles should matter; I don't think it should be particularly surprising that someone might assume that making a language more usable for humans would be related to the number of humans who end up deciding to use it, and if that's not the case, I wanted to understand why my intuition is wrong.
> there are still industries where it is important to reduce the human tendency to make mistakes by taking appropriate measures. Ada was created for such an industry from the very beginning
This is a good point that I hadn't considered; it definitely makes sense to me that some domains might be less tolerant to human errors than others, and those domains would better reflect how well-designed a language is for humans.
> The "level of suffering" experienced by most people is probably simply not great enough to systematically take such aspects into account. But there are still industries where it is important to reduce the human tendency to make mistakes by taking appropriate measures.
Reading this part a couple of times, I think this might be where the nuance lies. My colloquial understanding of what it means for something to be ergonomic (and even by the idea of what"level of suffering" would mean) isn't quite the same as the measurement of how likely it is for something to induce human error. This might just be a case where the common use of the term isn't the same as how it's used inside the field of study, but I would have expected that the ergonomics of a language and measurement of the "level of suffering" would be with respect to the programmer, not the one experiencing the use of the software that's developed as a result. That isn't to say I disagree with the idea that the end-user experience should ultimately be more important, but I think that might account for the disconnect between what you're describing here and what I would have expected from a discussion around "programming language ergonomics" (which also might explain the difference between Ada and the other languages mentioned in this thread).
> The question I have is why following established principles should matter
Apparently I still don't understand your question, sorry. For what I understand, following established principles is part of the engineering profession; it has proven to be the right thing to do over decades, and it is part of engineering education.
> I would have expected that the ergonomics of a language and measurement of the "level of suffering" would be with respect to the programmer, not the one experiencing the use of the software that's developed as a result.
Usually not the "level of suffering" is measured in human factors engineering, but the time needed and degree of fulfillment of typical tasks a typical representative of a test group is suppost to perform. You can do that with different designs and can then conclude which one meets the performance requirements best. Human factors typically enter a specification as performance requirements (what functions shall the system implement and how well). Given a programming language, you could measure how long a typical programmer requires to complete a specific task and how many errors the implementation has in the first version.
> Apparently I still don't understand your question, sorry. For what I understand, following established principles is part of the engineering profession; it has proven to be the right thing to do over decades, and it is part of engineering education.
I agree that following established principles is important, but my understanding is that the principles get established because they're better at leading to desirable outcomes. I'm trying to understand what the outcomes are that the principles you describe are intended to lead to. From your most recent two replies, my best interpretation is that it leads to fewer errors overall in the programs produced, but that wasn't as apparent to me from your first comment. I do think I understand now though.
> I'm trying to understand what the outcomes are that the principles you describe are intended to lead to.
The application of the principles of Human factors engineering to the design of systems reduces human errors, increases productivity, and enhances safety, health and comfort when interacting with these systems. For a programming language, taking human factors into account appropriately means that the target group of language users (i.e. programmers) is sufficiently capable of performing their tasks in all phases of a program's life cycle, e.g., they are not cognitively overwhelmed, and the likelihood of misunderstandings or mistakes is reduced. However, they should neither be unnecessarily restricted or hindered in their work, because also this creates unnecessary extraneous cognitive load that exhausts the programmer's limited working memory capacity. Human working memory can hold only 3-5 "chunks" of information simultaneously. This is a well-documented, hard biological constraint; when programming languages impose excessive formalism, they force programmers to juggle more mental "chunks" than working memory can handle. Self-explanatory code (which includes avoiding incomprehensible abbreviations or confusing syntax) reduces the cognitive load on the programmer. Ada's explicit human factors principle states: "Code is read more than written"; over a program's lifetime, especially in large, long-lived systems, code is read orders of magnitude more often than it's written; Ada's formalism optimizes for the more frequent activity (reading and maintenance) at the expense of the less frequent activity (initial writing). As a language designer, you therefore have to find the right balance, which of course is a function of your target audience, and the primary activities they will perform with the language.
ADA was designed with a goal of improving software reliability. Potentially at the cost of other factors, like programming speed. Real life projects using it demonstrated that the error rate of large ADA projects was around half of equivalent C or FORTRAN projects.
However popularity is determined by other factors. Such as personal productivity, and accessibility for novices. These other factors are often in direct opposition to long-term maintainability. It is good for productivity to be able to do things in whatever way is convenient. But that flexibility is a burden for the maintenance programmer. Likewise novices frequently create useful code, which becomes hard to maintain down the line. (Excel spreadsheets are a classic example of this.)
The result is that ADA was a good fit if you have a large project, time to market was not your top issue, and reliability and maintainability were top concerns. Many defense projects have these exact characteristics. Which is why ADA was developed and used there.
But consider a startup. Projects are small. The top concern is time to market. Maintainability will only become an issue if your product launches, gets to market, and succeeds. That's a future problem that can take care of itself.
That's why, when you look at startups, you find a high density of scripting languages. Which language has changed over time. For example Amazon used Perl, Facebook used PHP, and Instagram used Python. But all scripting languages share the similar characteristic of having fast initial development times, and poor long-term maintainability. (Yes, even Python. Internal data showing that is why Google began walking away from Python around 15 years ago.)
Niklaus Wirth designed Pascal (1970) with explicit pedagogical and human factors goals documented in his seminal paper "On the Design of Programming Languages". Wirth explicitly stated his belief that "insights gained from educational considerations could benefit programming language design in general, and that the simplicity and clarity he was striving for should be a guiding principle for all language design, serving equally pedagogical and development purposes". But - as with Ada - I'm not aware of any notable human factors engineering studies to validate design decisions.
Also Alan Kay and the Xerox PARC team designed Smalltalk (as Papert did before with Logo) with profound human-centered considerations, and they even "tested" their early concepts with children.
Also some other languages explicitly state human-centered design goals (e.g. Python, Eiffel), but as with Pascal or Ada the approach was more based on expert judgment, formal analysis, and established principles, not practical studies.
Human factors studies on programming languages are really hard to do right. It is easy to study someone seeing a language for the first time. However programming well requires a high level of expertise and so the real question isn't how easy it is for a beginner, it is how easy it is for someone who has been doing it for years. Or maybe how much different is someone after a week vs month vs year (that is at what point do the experience gains plateau)- in all this you have to be careful to ask different tasks - since good programmers will abstract when they need to do something twice.
You make a very good point, and maybe a problem here is that “human factors studies” are set up like market research rather than anthropology.
People who’ve spent a long time programming have spent a long time optimizing everything about their work, and they’re willing to talk about it (most of them won’t shut up about it, even).
I think it's not more difficult than identifying representative user groups (at least five members per group) and defining typical tasks for a usability test, focussing on the specific features in question. It's just a lot of work and requires experience and training.
Perl is one obvious one, as anyone will realize, if one has read some of Larry Wall's articles about his (and later the team's) motivations for the (syntax and semantics) design decisions they made about the language. For example, in his annual State of The Onion talks, he often discusses these points.
They may not have used standards such as the gp comment mentions, but they definitely considered human factors a lot.
E.g. TIMTOWTDI - There Is More Than One Way To Do It.
But that's not the only area in which they applied it.
Perl presents a fascinating counterexample: Larry Wall, trained as a linguist, explicitly cared about human factors, but his linguistic philosophy produced a language that empirical tests show performs poorly on readability and learnability measures (see e.g. http://dx.doi.org/10.1145/2534973 or https://doi.org/10.1145/2089155.2089159).
I haven't used Perl much, but my impression it is much easier to learn if you already know awk and sed - his original target was people who knew awk and sed well but were running into limitations using the two as separate tools. However the language quickly spread to people who don't know either of the previous tools and then it is difficult to learn.
>I haven't used Perl much, but my impression it is much easier to learn if you already know awk and sed - his original target was people who knew awk and sed well but were running into limitations using the two as separate tools.
Yes. In fact, in the early days, there were tools with names like sed2perl and awk2perl, or similar, IIRC. And those could convert code between their respective source and target languages.
I'm gonna read your links but I assume that 90% of the programming population doesn't grasp linguistic approaches to representing knowledge and logic (not saying this as a superiority thing, more as a cultural gap too wide)
I'm sorry, but both of those studies are a joke. They are taking novices who can't program, and are asking them to perform basic programming tasks in languages that they haven't learned. They are essentially testing how steep the initial learning curve is when you first see a language.
Once you understand a thing, you know what it’s capable of.
A lot of my early expertise in performance analysis was heavily informed by my SIGPLAN membership. Many of the improvements showing up in compilers and interpreters would show up in some form there, and of course those developers were either involved in the papers or had access to the same research. So when some new version came out with a big explanation of what it did, I already had a reasonably good notion of how it worked.
It was a dark day when they got rid of the paper proceedings.
Many new languages are still recycling ideas from the 1970's research labs.
Outside affine types, all the praise for Rust's type system traces back to Standard ML from 1976.
The heretic of doing systems programming in GC enabled programming languages (GC in the CS sense, including RC), goes back to research at Xerox PARC, DEC and ETHZ, late 1970's, early 1980's.
Other things that we know, like dependent types, effects, formal proofs, capabilities, linear and affine type systems, are equally a few decades old from 1980's, early 1990's.
Unfortunely while we have progressed, it seems easier to sell stuff like ChatGPT than better ways to approach software development.
Earlier than the sixties? All elements of OOP were known before 1967, but their combination, which we still use today under the title OOP, appeared in Simula 67 for the first time. I think the first appearance of a GC in literature was in 1960.
I think, the first mark‑and‑sweep collector was published in McCarthy's 1960 Communications of the ACM paper "Recursive Functions of Symbolic Expressions and Their Computation by Machine, Part I". It's resonable to assume, that they already had it when Steve Russell implemented the first Lisp evaluator, but we don't know exactly when it was added.
No, that paper doesn't describe the garbage collector. I do think it is true that before it was published Slug Russell had implemented the GC, but I think it's correct that we don't have listings from that early.
Edit: yes, yes it does describe the garbage collector.
Isn't the "Free-Storage List" in section 4c, starting on page 26, a mark & sweep collector? They didn't use the term then, but I think it describes one. I'm not aware of any earlier publication.
The paper only has 12 pages, but there is a section 4c with that title starting on page 192 of https://dl.acm.org/doi/pdf/10.1145/367177.367199, and you are correct that it describes a mark-and-sweep collector. I was mistaken, and I appreciate the correction.
Yes, but ideas are (mostly) worthless. I mean, they are necessary, but that's the easy part, building the technical foundation that make it possible is the hard part.
The internet needs wires and routers, distributed computing need a good network (i.e. the internet), current-day AI needs GPUs and GPUs need silicon chips that defy the laws of physics. Really, looking at the EUV lithography process makes all of computer science feel insignifiant by comparison, everything about it is absurd.
The real progress is that now, we can implement the ideas from the 70s, the good ones at least. I don't want to diminish the work of the founders of computer science, there is real genius here, but out of the billions of people on this planet, individual geniuses are not in short supply, but the real progress come from the millions of people that worked on the industrial complex, supply chains and trade that lead to modern GPUs, among everything that define modern computing.
"Fully specified" is the important part here, and the work that went into making the idea fully specified is where the value lies.
For this, we could look at intellectual property laws. Ideas are not protected. Neither by patents, nor copyright, nor trademark. If you want to make your idea worthy with regard to the law, you have to "fully specify" it, turning it into an invention (patent), or code (copyright).
Aside from bringing in the groundbreaking feature, Rust doesn't bring in any groundbreaking changes. It could be argued that bringing in lesser known features to more people is a good thing in it's own right.
I'm not a 'native' programmer, but even I can agree with that.
As it became easier to abstract away from pure assembly, people did just that. Some ideas stuck, some converged, and the ones with the most tenacious, adaptable and attractive communities remained.
Before I learned to code, no programming language was even remotely readable to me. But the more I learned, the more I could shed the notion that this was purely my fault, and accept that sometimes things are a certain way because someone found it interesting or useful. Applies to natural languages and mathematics, too.
There is something to the existence of fads and fundamentals. When I started, it was Object-Oriented-Programming (with multiple-inheritance and operator overloading, of course), Round-Trip Engineering (RTE), XML, and UML.
IMHO, not the ideas were bad, but the execution of them was. Ideas were too difficult/unfinished/not battle-tested at the time. A desire for premature optimisation without a full understanding of the problem space. The problem is that most programmers are beginners, and many teachers are intermediate programmers at best, and managers don't understand what programmers actually do. Skill issues abound. "Drive a nail with a screwdriver" indeed.
Nowadays, Round-Trip Engineering might be ready for a new try.
I always recommend people to learn at least one of the following: scripting language, compiled language, bytecode compiled language (C# or Java are industry giants), and at least one front-end web language either TS or JS. If they're still hungry I explain Erlang / Elixir / Gleam and tell them to try that out.
> Java and its OO relatives capture a communications-oriented model of computation where the funda-
mental computational elements are stateful agents that compute by sending one another messages;
I wish even only half the OOP world actually understood it as the above.
The main answer is that we have only a limited ability to modernize existing programming languages. For example, most languages are not null safe, because most languages are old and we can't make them null safe without breaking backward compatibility with most existing code. And we can't break backward compatibility for practical reasons. So Java will never be null safe, PHP will never be strongly or statically typed, etc.
So for fundamental language features, replacing older languages is the only way to achieve progress. Unfortunately that's a very slow process. Python, the currently most popular language, is already over 30 years old.
It's not just 30 years old. It's 30 years of people building libraries of useful code. If you "modernize" it so that much of that 30 years of work is thrown away, that's really expensive - maybe more expensive than continuing to build on a less-than-perfect foundation.
But that turns into the trap of short-term thinking - eventually you reach the point where you would have been better off throwing it away and starting over. You don't reach that in the year you throw it away, though, nor in the year after.
That CAN turn into the trap of short term thinking. However if you keep investing in modernizing over time you can get most of the benefits of updated code without having to scrap everything.
The catch is you will also have a bit of old code that cannot be modernized reasonably, and new code has to somehow interoperate with it. Which means languages can't break anything because it might be the one thing you can't figure out how to not use anymore even though you know better and would do it different if you started today. Worse often the problem is an early design decision and so the bad practice is everywhere and you can't get rid of it in any one place because everything depends on it.
For me, I find reviewing and analyzing programming languages to be a fun activity. Writing esolangs is also fun, you can write it without needing to care about things like backward compatibility or practicality.
This reminds me of recreational math & gamedev, you simply do whatever you feel is fun and design it exactly as you'd like it to be.
When studying programming languages you really have to focus on the why's.
When I was learning Rust I started out just associating patterns with lib types. Need to dynamically hold items? Vec. Need a global mutex? install lazy_static.
This is fine if you're beginning, but at some point you need to read about why people choose this. 9/10 times there's a more elegant option you didn't know about because you just did what everyone else does. This separates programmers from coders.
The only reason I learned this was because my current company has programmers, not coders. I learned a ton from them
I think that one of the things that they neglect to mention, on why we invent new languages, and it's probably the most important thing - people want new ways to express concepts.
It's super important because those concepts get measured, and absorbed into existing languages (as best they can), but that wouldn't have happened without the new languages
New concepts like Rust's "ownership model", Smalltalk's "Object Orientation", Lisp's "Functional programming", Haskell's "Lazy evaluation", Java's "Green threads"
And while Smalltalk may be the first quite "pure" OO language, it by no means invented the concept (you could argue it has a purer implementation of messaging but even some of that was in Simula 67, five years earlier).
At the right time (or just before the right time). Spotting the right audience (or stumbling onto what a significant audience needs).
And, I think, being better for what that audience is trying to do than existing tools. (But maybe that was implied in your statement.) This also implies adequate tooling and libraries.
And publicity, to reach that audience (though viral is better than corporate).
I don't think we have a lack of concepts, but a lack of concept connectivity.
To avoid feature bloat of unconnected pieces of ad-hoc syntax Java does the right thing and focuses on expressing easily-composable language building blocks.
As a last adopter language, Java has the luxory of cherry-picking good features of other languages but I think their line of thinking should be the reference moving forward.
> New concepts like Rust's "ownership model", Smalltalk's "Object Orientation", Lisp's "Functional programming", Haskell's "Lazy evaluation", Java's "Green threads"
I sincerely want to ask if this was an ironic comment given the topic? Because obviously none of these core concepts were really new to the languages you ascribe them to.
The appeal of experimenting with languages varies widely between academics and industrial practitioners. When you’re in the business of creating languages, exploration often takes on the character of an art project.
However, as a trench-line coder, I enjoy dabbling in languages to learn different techniques for achieving a similar set of goals without sacrificing pragmatism. In that sense, I rarely have the luxury to explore purely for exploration’s sake. So I wouldn’t describe abstraction, performance, or usability as “aesthetics,” nor would I spend time on a frivolous language that I know won’t gain much traction outside academia.
I like reading the perspectives of academics just to see how wildly different they are from those of the people I work with in the industry. This is probably a good thing.
>Common answers to this question will include words like abstraction, performance, convenience, usability etc. The problem with these answers is that apart from the measurable, they are all subjective, aesthetic choices.
That just is not true at all. These are all legitimate engineering tradeoffs, which any serious project has to balance. Calling this "aesthetics" is completely dishonest. These aren't arbitrary categories, these are meaningful distinctions engineers use when evaluating tools to write software. I think the students better understand what programming languages are than the teacher.
If you accept that a programming language is a tool and not just an academic game of terms, then all these questions have clear answers.
Agree, and we actually have both the standards and established methods to conduct representative tradeoff studies. But this knowledge is mostly ignored by CS and programming language designs. Even for Ada, there was little empirical evidence for specific design decision. A systematic survey discovered only 22 randomized controlled trials of textual programming language features conducted between the early 1950s through 2012, across six decades. This staggering scarcity explains why language designers rely heavily on intuition rather than evidence (see e.g. https://www.cs.cmu.edu/~NatProg/programminglanguageusability...).
The answer to (2) is, IMO, "to make computing cheaper". It's interesting to me that this is not the obvious, default answer (it may not be the most actionable answer but IMO it should at least be noted as a way to frame discussions). I think we're at the tail end of computing's artisanal, pre-industrial era where researchers and programmers alike have this latent, tacit view of computing as a kind of arcana.
I think it is more interesting to see which languages are still used today and how popular these are. Because this is also tied to the human user/developer.
For instance, I used BASIC when I was young - not as a professional but as a hobbyist. I liked it too. I wouldn't use BASIC today because it would be entirely useless and inefficient.
But is that a problem of the language itself, or is it just a problem of available toolchains? E.g. if the gcc compiler collection would come with BASIC support and you just could type something like "gbasic -O3 foobar.bas -o foobar" to get a properly optimised executable out of your BASIC source code file then some may would still use BASIC today, I guess?
I started with BASIC too. Also enjoyed BlitzBasic2 for a long time on the Amiga. That's where I learned programming… back then when programming was still fun.
Which basic. I remember Atari BASIC - which was a terrible language since the organization was line numbers. Visual BASIC looked a lot nicer the few times I saw it, but I have a total of 10 minutes of my life looking at visual BASIC and so I have no clue how it is in the real world. I do know Visual BASIC was used by a lot of untrained programmers who never learned why we have best practices and so there is a lot of bad BASIC out there - this isn't the fault of the language just the users.
Realistically? No, not at all. The reason there are no toolchains for BASIC is because nobody uses BASIC (because it's not functional in our modern world), not the other way round.
Why do you think BASIC is not functional? Our modern world does not differ from the 1980 world at all. Variables are variables, subroutines are subroutines.
It fall out of fashion, along with Pascal, Perl, Ruby, but that's just fashion.
I have even seen pretty darn impressive things done with VisualBasic back in the day. And that were not hobby things. I've seen it used in very important mission critical telecommunication equipment. The compiler was custom, not the one from Microsoft. After all, the language had pretty much anything other languages had.
How can a language be "inefficient"? You can say it lacked on expressiveness. Maybe was too verbose? But I would not place BASIC into the "verbose" category.
Because BASIC simply doesn't have first-class functions, and they would be quite hard to represent in a BASIC-like syntax while keeping the language idiomatic. Even the unreasonably clunky C pattern of having a pointer to a function taking void* as its first argument (to account for closure captures) gets you a whole lot closer to functional programming than even the fanciest BASICs.
I was never more productive than with REALBasic, it was a true RAD tool, and I'd happily use Xojo nowadays if it wasn't so expensive. That's a "modern" structured BASIC so I guess not what you had in mind.
One thing I've learned over the years is that the language is almost irrelevant, the tooling and 3rd-party library support are much more important.
One of PGs essays hit on this when he talked about “blub”. Building a language for a specific solution/domain was/is expensive, but using a language to build abstractions or an internal domain specific language has been much more tractable. As such some languages are more or less suited for this activity; Lisp being the king then Ruby then etc.
So to me the study languages was interesting from this DSL perspective.
(2022) This is particularly important here as the essay makes no mention of LLMs or coding agents (which were still in their infancy in development environments; this article is post original copilot/codex but pre ChatGPT).
In at least two ways, which you might see as alternatively positive and negative wrt whether expertise in programming languages is useful.
Firstly, when using them to create software it's pretty obvious that experienced devs and people who understand theory have a greater ability to guide, curate and control them.
Secondly, as they improve in ability we can see a paradigm change for people using them at least as significant as the jump from assembly to high level languages. Most programmers would have no need to study assembly these days.
Either way, their omission (while appropriate for the year, if somewhat lacking in foresight) is a significant one that renders it somewhat dated already.
Yeah. Now the trend goes in an entirely different direction: telling an LLM in natural language what to achieve using those programming languages. How quickly times change.
Edit: I assume this comment gets downvoted because people don't like where we are heading, not because they really think LLM programming capabilities won't continue to improve at a staggering pace.
> LLM programming capabilities won't continue to improve at a staggering pace.
The error rate of models make language design, tooling, testing methodology and human review more important than ever before. This demands language evolution. You could get faar with lax testing and language tooling with enough caution and skill. but when LLMs enter the picture, that no longer flies.
we need tooling, static analysis, testing paradigms, language design that restrict how dangerous the LLM is allowed to act.
natural language is faar to fuzzy to replace programming (system specification is already famously impossible thing to do right). If you think it truly will replace code, i highly suspect you work om webbdesign, where testing and reliability was always a secondary concern.
And even then, I think were already on the convergence platoe of LLM code. The companies are raising prices as diminishing improvement and balooning compute costs.
There absolutely lies value in studying programming language, but maybe not when reinventing ideas of the past.
> I encourage everyone to create the most absurd, implausible, and impractical languages. Chasing the measurable is often useful, expressing the expressible is insightful, but never forget the true goal of language design: to explore and create what isn’t.
Sorry, but this sounds more like an artsclass to me. Don't get me wrong, there was a point in time where exploration of the unknown was the only way to move forward. But these days we would need greater insights into higher-level language semantics and inherent tradeoffs to guide language-design and language evolution.
There is plenty to choose from and one can learn already so much just by reading up on the Java-EG mailing lists. Brian Goetz has a true academic mindset and I frequently feel inspired when I read his reasoning which is both highly structured and accessible.
Otherwise we would just be left with another compiler class. Compiler basics really aren't that difficult.
Indeed, it is, and that's the point! Being interfaces to computers for humans, programming languages sit at the intersection of computer science and humanities. Lots of people like to treat programming languages like they're math class, but that's only half the picture. The other half is usability, ergonomics, learnability, and especially community. Not to mention the form of the language is all about aesthetics. How many times has someone on Hacker News called a language "beautiful" or "ugly" referring to the way it looks? When people praise Python they talk about how easy it is to read and how pleasant it is to look at compared to C++. Or look at what people say about Elm error messages versus C++ template errors. Actually a lot of what's wrong with C++ could have been averted if the designers had paid more attention in art class.
> But these days we would need greater insights into higher-level language semantics and inherent tradeoffs to guide language-design and language evolution.
Here's a talk that argues there's much more fertile languages ground for ideas outside of the "programming languages are math" area, which has been thoroughly strip-mined for decades:
This author takes the perspective that programming languages are much greater than the sum of the syntax + semantics + toolchain + libraries, and treating them as such is limiting their potential.
Programming languages are obsolete in the LLM era. What current generation AI has revealed is that English is actually the ultimate representation of computer programs and systems, which is both sufficiently terse and precise to economically describe the operation of arbitrarily complex programs.
There is no reason to study programming languages in 2025, other than as a historical curiosity - the same way one may study languages equally as pitiable as e.g. COBOL, Lisp, or MIPS assembly.
I think it's the exact opposite -- LLMs have revealed the precise utility of programming languages. For decades the "English as programming language" has been the holy grail of language designers. From COBOL to SQL to AppleScript, it was the hope that one day we'll be able to program a computer just as easily as we can instruct a person.
Well LLMs finally offer that, and what they are proving is what programmers have known for decades -- natural language is a terrible way to specify a program to a computer. So what is happening in the LLM world is they are reinventing programming languages and software engineering. They're just calling it "prompt engineering" and "context engineering".
What this tell us is that natural languages are not only not sufficient for the task of programming, to make them sufficient you need to bring back all the properties you lost by ditching the programming language. Things like reliability, reproducibility, determinism, unambiguity are thrown away when you use an LLM, and context engineering / prompt engineering are ways of trying to get that back. They won't work well. What you really want is a programming language.
> natural languages are not only not sufficient for the task of programming,
Downthread there is an example of an ICPC problem statement, [0] given as natural language, (modulo some inequalities and example program inputs/outputs) which was sufficient for Gemini to program & implement the correct solution where no other human could.
> problem statement, [0] given as natural language, (modulo some inequalities and example program inputs/outputs)
I also see two graphics, and several formal mathematical expressions. You can't modulo away all the not-natural language and then claim natural language alone was sufficient. I presume these things were added by the authors to increase clarity of the problem statement, and I agree with them. They used formal languages to specify all the important parameters in an unambiguous way, which was the right call! Otherwise we would all be left wondering at the semantics.
Anyway, I don't think this really responds to my point, because competition prompts are designed to be self-contained problem statements that are as clear and unambiguous as possible for competition purposes. And in this case, they switched to speaking in a formal language when being precise and unambiguous was most important.
On the other hand, my statement was about the task of programming, which typically involves solving ill-defined problems with specifications and requirements that shift over time. I've been in programming competitions, I've also been a programmer, and I don't find one to be really related to the other.
Good luck debugging and fixing the multitude of errors introduced by your LLM. ESPECIALLY if you're outputting some low-level assembler or machine code. At least when we have our AI outputting Java/Typescript/Python/etc, it's easy (for a programmer) to see where it's run afoul.
I have every confidence that even current generation state-of-the-art artificial intelligence will have far superior facilities at debugging and managing the complexity of even its own bare metal programming output than humans.
Why? Consider Gemini's recent performance at the International Collegiate Programming Contest [0], in which it solved a problem that no other human team was able to solve.
Wetware intelligence is itself obsolete, at least as concerns the domain of computing.
> Usability is similarly ill-defined and hard to measure.
Human factors are very well studied and standardized, and there is a well-established discipline called "Human Factors Engineering", which also provides established test and evaluation methods. Human Factors research is considered solid and well-established because it has been built on rigorous experimental psychology and engineering principles developed over more than a century, with systematic methodology and empirical validation. Even if much of it is unknown or ignored by computer science or programming language design, there are many disciplines where Human Factors Engineering is critical (see e.g. ANSI/AAMI HE75).
Usability is therefore neither ill-defined nor hard to measure. Several ISO 9241 series standards address textual and command-based interaction directly relevant to programming language assessment. ETSI EG 202 116 provides extensive human factors guidelines for command language style. ITU-T Recommendation Z.361 provides guidelines for Human-Computer Interfaces in telecommunications management, incorporating the ISO 9241-10 dialogue principles. ISO/IEC TR 24772 addresses programming language vulnerabilities specifically from a human factors perspective.
E.g. Ada did have substantial human factors considerations documented in its design rationale, directly contradicting the notion that such principles don't apply to professional developers or programming languages. It's rather that computer science seems to continue ignoring established fields ("Not invented here"). Human factors in software development have been overlooked by researchers in the software engineering and development research areas, despite their obvious relevance. So what is lacking is primarily interest or willingness, not the fundamentals and means.
Honest question: given the relatively niche status of Ada compared to other systems programming languages, and similar diminished levels of popularity for other languages mentioned in this thread (Pascal, Perl), does this really prove that these principles are particularly effective for programming languages? I understand that this is a relatively small sample size, but I feel like that could be just as well used to argue that other factors might have been more important to the successes of languages designed like Ada than the more rigorous approach to human usability. It doesn't feel particularly obvious to me that we know these principles actually work well for programming languages but choose to ignore them rather than them not fitting the domain particularly well at all, or maybe something in between where we don't really know whether they apply well yet or not, and more attempts at using them might not actually be as successful as one might hope.
I didn't claim that usability features were "important to the successes of languages designed like Ada". But Ada at least explicitly considered human factors in their design (even if mostly based on expert judgment and established principles, not practical studies), which also seems very appropriate given the criticality of most applications written in Ada. But as my ergonomics professor at ETH Zurich, Helmut Krüger (successor to the renowned Étienne Grandjean), used to say: people get used to even the most ergonomically terrible systems. The "level of suffering" experienced by most people is probably simply not great enough to systematically take such aspects into account. But there are still industries where it is important to reduce the human tendency to make mistakes by taking appropriate measures. Ada was created for such an industry from the very beginning.
> I didn't claim that usability features were "important to the successes of languages designed like Ada".
This kind of seems like it's focusing too much on my exact word choice and less the actual intent of my question behind it. The question I have is why following established principles should matter; I don't think it should be particularly surprising that someone might assume that making a language more usable for humans would be related to the number of humans who end up deciding to use it, and if that's not the case, I wanted to understand why my intuition is wrong.
> there are still industries where it is important to reduce the human tendency to make mistakes by taking appropriate measures. Ada was created for such an industry from the very beginning
This is a good point that I hadn't considered; it definitely makes sense to me that some domains might be less tolerant to human errors than others, and those domains would better reflect how well-designed a language is for humans.
> The "level of suffering" experienced by most people is probably simply not great enough to systematically take such aspects into account. But there are still industries where it is important to reduce the human tendency to make mistakes by taking appropriate measures.
Reading this part a couple of times, I think this might be where the nuance lies. My colloquial understanding of what it means for something to be ergonomic (and even by the idea of what"level of suffering" would mean) isn't quite the same as the measurement of how likely it is for something to induce human error. This might just be a case where the common use of the term isn't the same as how it's used inside the field of study, but I would have expected that the ergonomics of a language and measurement of the "level of suffering" would be with respect to the programmer, not the one experiencing the use of the software that's developed as a result. That isn't to say I disagree with the idea that the end-user experience should ultimately be more important, but I think that might account for the disconnect between what you're describing here and what I would have expected from a discussion around "programming language ergonomics" (which also might explain the difference between Ada and the other languages mentioned in this thread).
> The question I have is why following established principles should matter
Apparently I still don't understand your question, sorry. For what I understand, following established principles is part of the engineering profession; it has proven to be the right thing to do over decades, and it is part of engineering education.
> I would have expected that the ergonomics of a language and measurement of the "level of suffering" would be with respect to the programmer, not the one experiencing the use of the software that's developed as a result.
Usually not the "level of suffering" is measured in human factors engineering, but the time needed and degree of fulfillment of typical tasks a typical representative of a test group is suppost to perform. You can do that with different designs and can then conclude which one meets the performance requirements best. Human factors typically enter a specification as performance requirements (what functions shall the system implement and how well). Given a programming language, you could measure how long a typical programmer requires to complete a specific task and how many errors the implementation has in the first version.
> Apparently I still don't understand your question, sorry. For what I understand, following established principles is part of the engineering profession; it has proven to be the right thing to do over decades, and it is part of engineering education.
I agree that following established principles is important, but my understanding is that the principles get established because they're better at leading to desirable outcomes. I'm trying to understand what the outcomes are that the principles you describe are intended to lead to. From your most recent two replies, my best interpretation is that it leads to fewer errors overall in the programs produced, but that wasn't as apparent to me from your first comment. I do think I understand now though.
> I'm trying to understand what the outcomes are that the principles you describe are intended to lead to.
The application of the principles of Human factors engineering to the design of systems reduces human errors, increases productivity, and enhances safety, health and comfort when interacting with these systems. For a programming language, taking human factors into account appropriately means that the target group of language users (i.e. programmers) is sufficiently capable of performing their tasks in all phases of a program's life cycle, e.g., they are not cognitively overwhelmed, and the likelihood of misunderstandings or mistakes is reduced. However, they should neither be unnecessarily restricted or hindered in their work, because also this creates unnecessary extraneous cognitive load that exhausts the programmer's limited working memory capacity. Human working memory can hold only 3-5 "chunks" of information simultaneously. This is a well-documented, hard biological constraint; when programming languages impose excessive formalism, they force programmers to juggle more mental "chunks" than working memory can handle. Self-explanatory code (which includes avoiding incomprehensible abbreviations or confusing syntax) reduces the cognitive load on the programmer. Ada's explicit human factors principle states: "Code is read more than written"; over a program's lifetime, especially in large, long-lived systems, code is read orders of magnitude more often than it's written; Ada's formalism optimizes for the more frequent activity (reading and maintenance) at the expense of the less frequent activity (initial writing). As a language designer, you therefore have to find the right balance, which of course is a function of your target audience, and the primary activities they will perform with the language.
There is a giant https://www.dreamsongs.com/WorseIsBetter.html phenomena here. The word "good" should never be accepted at face value. It implies a question. Good for what?
ADA was designed with a goal of improving software reliability. Potentially at the cost of other factors, like programming speed. Real life projects using it demonstrated that the error rate of large ADA projects was around half of equivalent C or FORTRAN projects.
However popularity is determined by other factors. Such as personal productivity, and accessibility for novices. These other factors are often in direct opposition to long-term maintainability. It is good for productivity to be able to do things in whatever way is convenient. But that flexibility is a burden for the maintenance programmer. Likewise novices frequently create useful code, which becomes hard to maintain down the line. (Excel spreadsheets are a classic example of this.)
The result is that ADA was a good fit if you have a large project, time to market was not your top issue, and reliability and maintainability were top concerns. Many defense projects have these exact characteristics. Which is why ADA was developed and used there.
But consider a startup. Projects are small. The top concern is time to market. Maintainability will only become an issue if your product launches, gets to market, and succeeds. That's a future problem that can take care of itself.
That's why, when you look at startups, you find a high density of scripting languages. Which language has changed over time. For example Amazon used Perl, Facebook used PHP, and Instagram used Python. But all scripting languages share the similar characteristic of having fast initial development times, and poor long-term maintainability. (Yes, even Python. Internal data showing that is why Google began walking away from Python around 15 years ago.)
Apart from ADA, what are languages / stacks with most aspects of "Human Factors Engineering" considered?
Niklaus Wirth designed Pascal (1970) with explicit pedagogical and human factors goals documented in his seminal paper "On the Design of Programming Languages". Wirth explicitly stated his belief that "insights gained from educational considerations could benefit programming language design in general, and that the simplicity and clarity he was striving for should be a guiding principle for all language design, serving equally pedagogical and development purposes". But - as with Ada - I'm not aware of any notable human factors engineering studies to validate design decisions.
Also Alan Kay and the Xerox PARC team designed Smalltalk (as Papert did before with Logo) with profound human-centered considerations, and they even "tested" their early concepts with children.
Also some other languages explicitly state human-centered design goals (e.g. Python, Eiffel), but as with Pascal or Ada the approach was more based on expert judgment, formal analysis, and established principles, not practical studies.
Human factors studies on programming languages are really hard to do right. It is easy to study someone seeing a language for the first time. However programming well requires a high level of expertise and so the real question isn't how easy it is for a beginner, it is how easy it is for someone who has been doing it for years. Or maybe how much different is someone after a week vs month vs year (that is at what point do the experience gains plateau)- in all this you have to be careful to ask different tasks - since good programmers will abstract when they need to do something twice.
You make a very good point, and maybe a problem here is that “human factors studies” are set up like market research rather than anthropology.
People who’ve spent a long time programming have spent a long time optimizing everything about their work, and they’re willing to talk about it (most of them won’t shut up about it, even).
I think it's not more difficult than identifying representative user groups (at least five members per group) and defining typical tasks for a usability test, focussing on the specific features in question. It's just a lot of work and requires experience and training.
Perl is one obvious one, as anyone will realize, if one has read some of Larry Wall's articles about his (and later the team's) motivations for the (syntax and semantics) design decisions they made about the language. For example, in his annual State of The Onion talks, he often discusses these points.
They may not have used standards such as the gp comment mentions, but they definitely considered human factors a lot.
E.g. TIMTOWTDI - There Is More Than One Way To Do It.
But that's not the only area in which they applied it.
Perl presents a fascinating counterexample: Larry Wall, trained as a linguist, explicitly cared about human factors, but his linguistic philosophy produced a language that empirical tests show performs poorly on readability and learnability measures (see e.g. http://dx.doi.org/10.1145/2534973 or https://doi.org/10.1145/2089155.2089159).
I haven't used Perl much, but my impression it is much easier to learn if you already know awk and sed - his original target was people who knew awk and sed well but were running into limitations using the two as separate tools. However the language quickly spread to people who don't know either of the previous tools and then it is difficult to learn.
>I haven't used Perl much, but my impression it is much easier to learn if you already know awk and sed - his original target was people who knew awk and sed well but were running into limitations using the two as separate tools.
Yes. In fact, in the early days, there were tools with names like sed2perl and awk2perl, or similar, IIRC. And those could convert code between their respective source and target languages.
And even easier if you know C and Unix.
Just like human dialects, across the same main language, :)
I'm gonna read your links but I assume that 90% of the programming population doesn't grasp linguistic approaches to representing knowledge and logic (not saying this as a superiority thing, more as a cultural gap too wide)
I'm sorry, but both of those studies are a joke. They are taking novices who can't program, and are asking them to perform basic programming tasks in languages that they haven't learned. They are essentially testing how steep the initial learning curve is when you first see a language.
A far more important measure is something like function points per month of developer time. See page 49 of https://www.ifpug.org/wp-content/uploads/2017/04/IYSM.-Thirt... for some data on that. ADA 95 did pretty well, around 11. Perl beat it with 15.
Of course that measure isn't perfect. Excel trounced both with over 30. But Excel is only appropriate for some kinds of projects.
Obviously, COBOL as a human-friendly improvement over Assembler, and BASIC as a more natural (English-friendly) improvement over COBOL, qualify.
I think Eiffel would be one of them.
Once you understand a thing, you know what it’s capable of.
A lot of my early expertise in performance analysis was heavily informed by my SIGPLAN membership. Many of the improvements showing up in compilers and interpreters would show up in some form there, and of course those developers were either involved in the papers or had access to the same research. So when some new version came out with a big explanation of what it did, I already had a reasonably good notion of how it worked.
It was a dark day when they got rid of the paper proceedings.
> Once you understand a thing, you know what it’s capable of.
Did you.... just quote Blade? :-)
I just confessed to reading the paper versions of SIGPLAN, what do you think?
But my favorite will always be
“Some motherfuckers are always trying to ice skate uphill.”
I tries to answer the question "Why do we design new programming languages?" but it forgets the simplest of answers:
Because we can. Because a compiler is nothing more than a fancy text translator.
Because what we know about programming has progressed, and new languages appear to take advantage of that.
Many new languages are still recycling ideas from the 1970's research labs.
Outside affine types, all the praise for Rust's type system traces back to Standard ML from 1976.
The heretic of doing systems programming in GC enabled programming languages (GC in the CS sense, including RC), goes back to research at Xerox PARC, DEC and ETHZ, late 1970's, early 1980's.
Other things that we know, like dependent types, effects, formal proofs, capabilities, linear and affine type systems, are equally a few decades old from 1980's, early 1990's.
Unfortunely while we have progressed, it seems easier to sell stuff like ChatGPT than better ways to approach software development.
It seams like most things boil down to ideas from the 70's. The internet, distributed computing, AI, etc...
> most things boil down to ideas from the 70's
Rather from the sixties. E.g. OOP including dynamic dispatch, late binding, GC etc. appeared 1967 with Simula.
GC was a bit earlier. :)
Earlier than the sixties? All elements of OOP were known before 1967, but their combination, which we still use today under the title OOP, appeared in Simula 67 for the first time. I think the first appearance of a GC in literature was in 1960.
Did Lisp have GC from the beginning? If so, that would be 1958.
I think, the first mark‑and‑sweep collector was published in McCarthy's 1960 Communications of the ACM paper "Recursive Functions of Symbolic Expressions and Their Computation by Machine, Part I". It's resonable to assume, that they already had it when Steve Russell implemented the first Lisp evaluator, but we don't know exactly when it was added.
No, that paper doesn't describe the garbage collector. I do think it is true that before it was published Slug Russell had implemented the GC, but I think it's correct that we don't have listings from that early.
Edit: yes, yes it does describe the garbage collector.
Isn't the "Free-Storage List" in section 4c, starting on page 26, a mark & sweep collector? They didn't use the term then, but I think it describes one. I'm not aware of any earlier publication.
The paper only has 12 pages, but there is a section 4c with that title starting on page 192 of https://dl.acm.org/doi/pdf/10.1145/367177.367199, and you are correct that it describes a mark-and-sweep collector. I was mistaken, and I appreciate the correction.
Ok, I see. Just downloaded the first hit from Google, which has 34 pages and is nicely printed: https://www-formal.stanford.edu/jmc/recursive.pdf.
The concept of automatic memory management dates back to the 1950s.
However, John McCarthy, the creator of Lisp, introduced the first widely recognized garbage collection mechanism around 1959/1960.
If so, he didn't write about it in that paper, and I don't think he introduced it at all; I think Slug Russell did.
Correction: as Rochus pointed out, he certainly did write about it in that paper, giving a complete description.
Yes, but ideas are (mostly) worthless. I mean, they are necessary, but that's the easy part, building the technical foundation that make it possible is the hard part.
The internet needs wires and routers, distributed computing need a good network (i.e. the internet), current-day AI needs GPUs and GPUs need silicon chips that defy the laws of physics. Really, looking at the EUV lithography process makes all of computer science feel insignifiant by comparison, everything about it is absurd.
The real progress is that now, we can implement the ideas from the 70s, the good ones at least. I don't want to diminish the work of the founders of computer science, there is real genius here, but out of the billions of people on this planet, individual geniuses are not in short supply, but the real progress come from the millions of people that worked on the industrial complex, supply chains and trade that lead to modern GPUs, among everything that define modern computing.
As for this measure, Software is worthless as well, which is basically fully specified ideas.
"Fully specified" is the important part here, and the work that went into making the idea fully specified is where the value lies.
For this, we could look at intellectual property laws. Ideas are not protected. Neither by patents, nor copyright, nor trademark. If you want to make your idea worthy with regard to the law, you have to "fully specify" it, turning it into an invention (patent), or code (copyright).
Indeed, everything old is new again.
Aside from bringing in the groundbreaking feature, Rust doesn't bring in any groundbreaking changes. It could be argued that bringing in lesser known features to more people is a good thing in it's own right.
Hey Walter, while we have you. Have you ever had the itch to make a newer D that takes all the fancy constructs while retaining the good parts?
We're doing just that by introducing an "Editions" feature that will streamline the feature set.
Really fancy one, when it gets to what translation actually means in implementation effort.
Hehe, it was tongue-in-cheek. It takes a buttload of efforts to make a good one.
[dead]
I'm not a 'native' programmer, but even I can agree with that.
As it became easier to abstract away from pure assembly, people did just that. Some ideas stuck, some converged, and the ones with the most tenacious, adaptable and attractive communities remained.
Before I learned to code, no programming language was even remotely readable to me. But the more I learned, the more I could shed the notion that this was purely my fault, and accept that sometimes things are a certain way because someone found it interesting or useful. Applies to natural languages and mathematics, too.
Because we can, because it's fun, and because some of us are compulsively obsessed.
linguistic constructs will direct your brain when solving problems and may help you discover solutions spaces you wouldn't with another set of idioms
not to promote FP but imperative stateful vs closures/function oriented is quite a strong example of that
a different paradigm can really be a massive intellectual tool
I still like Olin Shiver's take on this: https://www.ccs.neu.edu/home/shivers/papers/why-teach-pl.pdf
There is something to the existence of fads and fundamentals. When I started, it was Object-Oriented-Programming (with multiple-inheritance and operator overloading, of course), Round-Trip Engineering (RTE), XML, and UML.
IMHO, not the ideas were bad, but the execution of them was. Ideas were too difficult/unfinished/not battle-tested at the time. A desire for premature optimisation without a full understanding of the problem space. The problem is that most programmers are beginners, and many teachers are intermediate programmers at best, and managers don't understand what programmers actually do. Skill issues abound. "Drive a nail with a screwdriver" indeed.
Nowadays, Round-Trip Engineering might be ready for a new try.
Olin also wrote the greatest acknowledgment section ever: https://scsh.net/docu/html/man.html
I always recommend people to learn at least one of the following: scripting language, compiled language, bytecode compiled language (C# or Java are industry giants), and at least one front-end web language either TS or JS. If they're still hungry I explain Erlang / Elixir / Gleam and tell them to try that out.
> Java and its OO relatives capture a communications-oriented model of computation where the funda- mental computational elements are stateful agents that compute by sending one another messages;
I wish even only half the OOP world actually understood it as the above.
> Why do we design new programming languages?
The main answer is that we have only a limited ability to modernize existing programming languages. For example, most languages are not null safe, because most languages are old and we can't make them null safe without breaking backward compatibility with most existing code. And we can't break backward compatibility for practical reasons. So Java will never be null safe, PHP will never be strongly or statically typed, etc.
So for fundamental language features, replacing older languages is the only way to achieve progress. Unfortunately that's a very slow process. Python, the currently most popular language, is already over 30 years old.
It's not just 30 years old. It's 30 years of people building libraries of useful code. If you "modernize" it so that much of that 30 years of work is thrown away, that's really expensive - maybe more expensive than continuing to build on a less-than-perfect foundation.
But that turns into the trap of short-term thinking - eventually you reach the point where you would have been better off throwing it away and starting over. You don't reach that in the year you throw it away, though, nor in the year after.
That CAN turn into the trap of short term thinking. However if you keep investing in modernizing over time you can get most of the benefits of updated code without having to scrap everything.
The catch is you will also have a bit of old code that cannot be modernized reasonably, and new code has to somehow interoperate with it. Which means languages can't break anything because it might be the one thing you can't figure out how to not use anymore even though you know better and would do it different if you started today. Worse often the problem is an early design decision and so the bad practice is everywhere and you can't get rid of it in any one place because everything depends on it.
Python is one language that famously broke backwards compatibility with the 2.x/3.x split.
For me, I find reviewing and analyzing programming languages to be a fun activity. Writing esolangs is also fun, you can write it without needing to care about things like backward compatibility or practicality.
This reminds me of recreational math & gamedev, you simply do whatever you feel is fun and design it exactly as you'd like it to be.
When studying programming languages you really have to focus on the why's.
When I was learning Rust I started out just associating patterns with lib types. Need to dynamically hold items? Vec. Need a global mutex? install lazy_static.
This is fine if you're beginning, but at some point you need to read about why people choose this. 9/10 times there's a more elegant option you didn't know about because you just did what everyone else does. This separates programmers from coders.
The only reason I learned this was because my current company has programmers, not coders. I learned a ton from them
I think that one of the things that they neglect to mention, on why we invent new languages, and it's probably the most important thing - people want new ways to express concepts.
It's super important because those concepts get measured, and absorbed into existing languages (as best they can), but that wouldn't have happened without the new languages
New concepts like Rust's "ownership model", Smalltalk's "Object Orientation", Lisp's "Functional programming", Haskell's "Lazy evaluation", Java's "Green threads"
While those languages made the concepts mainstream, they weren't the ones coming up with them.
Rust's "ownership model", is a simplification of Cyclone, AT&T's research on a better C, based on mix of affine and linear type systems.
https://en.wikipedia.org/wiki/Cyclone_(programming_language)
Haskell's "Lazy evaluation" was present in Miranda, before all related researchers came up with Haskell as common playground.
https://en.wikipedia.org/wiki/Miranda_(programming_language)
"History of Haskell"
https://www.microsoft.com/en-us/research/wp-content/uploads/...
Java's "Green threads" go back to systems like Concurrent Pascal.
https://dl.acm.org/doi/10.1145/775332.775335
And while Smalltalk may be the first quite "pure" OO language, it by no means invented the concept (you could argue it has a purer implementation of messaging but even some of that was in Simula 67, five years earlier).
Cyclone was indeed a very interesting language. It's always surprising what ultimately prevails.
I guess it is always a matter of being there at the right time, or having the luck to spot the right audience.
At the right time (or just before the right time). Spotting the right audience (or stumbling onto what a significant audience needs).
And, I think, being better for what that audience is trying to do than existing tools. (But maybe that was implied in your statement.) This also implies adequate tooling and libraries.
And publicity, to reach that audience (though viral is better than corporate).
I don't think we have a lack of concepts, but a lack of concept connectivity.
To avoid feature bloat of unconnected pieces of ad-hoc syntax Java does the right thing and focuses on expressing easily-composable language building blocks.
As a last adopter language, Java has the luxory of cherry-picking good features of other languages but I think their line of thinking should be the reference moving forward.
> New concepts like Rust's "ownership model", Smalltalk's "Object Orientation", Lisp's "Functional programming", Haskell's "Lazy evaluation", Java's "Green threads"
I sincerely want to ask if this was an ironic comment given the topic? Because obviously none of these core concepts were really new to the languages you ascribe them to.
I'm pretty sure they mentioned
>we create programming languages to experience new ideas; ideas that would have remained inaccessible had we stayed with the old languages.
The appeal of experimenting with languages varies widely between academics and industrial practitioners. When you’re in the business of creating languages, exploration often takes on the character of an art project.
However, as a trench-line coder, I enjoy dabbling in languages to learn different techniques for achieving a similar set of goals without sacrificing pragmatism. In that sense, I rarely have the luxury to explore purely for exploration’s sake. So I wouldn’t describe abstraction, performance, or usability as “aesthetics,” nor would I spend time on a frivolous language that I know won’t gain much traction outside academia.
I like reading the perspectives of academics just to see how wildly different they are from those of the people I work with in the industry. This is probably a good thing.
>Common answers to this question will include words like abstraction, performance, convenience, usability etc. The problem with these answers is that apart from the measurable, they are all subjective, aesthetic choices.
That just is not true at all. These are all legitimate engineering tradeoffs, which any serious project has to balance. Calling this "aesthetics" is completely dishonest. These aren't arbitrary categories, these are meaningful distinctions engineers use when evaluating tools to write software. I think the students better understand what programming languages are than the teacher.
If you accept that a programming language is a tool and not just an academic game of terms, then all these questions have clear answers.
> These are all legitimate engineering tradeoffs
Agree, and we actually have both the standards and established methods to conduct representative tradeoff studies. But this knowledge is mostly ignored by CS and programming language designs. Even for Ada, there was little empirical evidence for specific design decision. A systematic survey discovered only 22 randomized controlled trials of textual programming language features conducted between the early 1950s through 2012, across six decades. This staggering scarcity explains why language designers rely heavily on intuition rather than evidence (see e.g. https://www.cs.cmu.edu/~NatProg/programminglanguageusability...).
> I encourage everyone to create the most absurd, implausible, and impractical languages.
Been there, done that: https://esolangs.org/wiki/Ziim
The answer to (2) is, IMO, "to make computing cheaper". It's interesting to me that this is not the obvious, default answer (it may not be the most actionable answer but IMO it should at least be noted as a way to frame discussions). I think we're at the tail end of computing's artisanal, pre-industrial era where researchers and programmers alike have this latent, tacit view of computing as a kind of arcana.
Old languages died or are barely in use anymore.
I think it is more interesting to see which languages are still used today and how popular these are. Because this is also tied to the human user/developer.
For instance, I used BASIC when I was young - not as a professional but as a hobbyist. I liked it too. I wouldn't use BASIC today because it would be entirely useless and inefficient.
But is that a problem of the language itself, or is it just a problem of available toolchains? E.g. if the gcc compiler collection would come with BASIC support and you just could type something like "gbasic -O3 foobar.bas -o foobar" to get a properly optimised executable out of your BASIC source code file then some may would still use BASIC today, I guess?
I started with BASIC too. Also enjoyed BlitzBasic2 for a long time on the Amiga. That's where I learned programming… back then when programming was still fun.
Which basic. I remember Atari BASIC - which was a terrible language since the organization was line numbers. Visual BASIC looked a lot nicer the few times I saw it, but I have a total of 10 minutes of my life looking at visual BASIC and so I have no clue how it is in the real world. I do know Visual BASIC was used by a lot of untrained programmers who never learned why we have best practices and so there is a lot of bad BASIC out there - this isn't the fault of the language just the users.
Realistically? No, not at all. The reason there are no toolchains for BASIC is because nobody uses BASIC (because it's not functional in our modern world), not the other way round.
Why do you think BASIC is not functional? Our modern world does not differ from the 1980 world at all. Variables are variables, subroutines are subroutines.
It fall out of fashion, along with Pascal, Perl, Ruby, but that's just fashion.
I have even seen pretty darn impressive things done with VisualBasic back in the day. And that were not hobby things. I've seen it used in very important mission critical telecommunication equipment. The compiler was custom, not the one from Microsoft. After all, the language had pretty much anything other languages had.
How can a language be "inefficient"? You can say it lacked on expressiveness. Maybe was too verbose? But I would not place BASIC into the "verbose" category.
> Why do you think BASIC is not functional?
Because BASIC simply doesn't have first-class functions, and they would be quite hard to represent in a BASIC-like syntax while keeping the language idiomatic. Even the unreasonably clunky C pattern of having a pointer to a function taking void* as its first argument (to account for closure captures) gets you a whole lot closer to functional programming than even the fanciest BASICs.
Here, "functional" is being used to mean "ablity to function", not "relating to the functional programming paradigm".
I hate that languages have become fads. The concepts have not changed, but there is a constant churn of languages.
I don't have to relearn natural language every 5-10 years, but for some reason I'm expected to when it comes to programming.
Except there are,
https://learn.microsoft.com/en-us/dotnet/visual-basic/
https://www.xojo.com/
https://www.mikroe.com/mikrobasic-pic
Just the three that come to my mind, of BASIC toolchains that people actually pay real money to use.
I was never more productive than with REALBasic, it was a true RAD tool, and I'd happily use Xojo nowadays if it wasn't so expensive. That's a "modern" structured BASIC so I guess not what you had in mind.
One thing I've learned over the years is that the language is almost irrelevant, the tooling and 3rd-party library support are much more important.
Depends pretty much on what BASIC compiler one is talking about.
One of PGs essays hit on this when he talked about “blub”. Building a language for a specific solution/domain was/is expensive, but using a language to build abstractions or an internal domain specific language has been much more tractable. As such some languages are more or less suited for this activity; Lisp being the king then Ruby then etc.
So to me the study languages was interesting from this DSL perspective.
The author says:
“This class is about the study of programming languages.”
Where is that class?
Related: https://cs.wellesley.edu/~cs251/f16/assignments/antics/Stefi...
(2022) This is particularly important here as the essay makes no mention of LLMs or coding agents (which were still in their infancy in development environments; this article is post original copilot/codex but pre ChatGPT).
> This is particularly important here
No actually. Why is that important? I dont quite see why that is relevant. Could you elaborate?
In at least two ways, which you might see as alternatively positive and negative wrt whether expertise in programming languages is useful.
Firstly, when using them to create software it's pretty obvious that experienced devs and people who understand theory have a greater ability to guide, curate and control them.
Secondly, as they improve in ability we can see a paradigm change for people using them at least as significant as the jump from assembly to high level languages. Most programmers would have no need to study assembly these days.
Either way, their omission (while appropriate for the year, if somewhat lacking in foresight) is a significant one that renders it somewhat dated already.
Yeah. Now the trend goes in an entirely different direction: telling an LLM in natural language what to achieve using those programming languages. How quickly times change.
Edit: I assume this comment gets downvoted because people don't like where we are heading, not because they really think LLM programming capabilities won't continue to improve at a staggering pace.
> LLM programming capabilities won't continue to improve at a staggering pace.
The error rate of models make language design, tooling, testing methodology and human review more important than ever before. This demands language evolution. You could get faar with lax testing and language tooling with enough caution and skill. but when LLMs enter the picture, that no longer flies.
we need tooling, static analysis, testing paradigms, language design that restrict how dangerous the LLM is allowed to act.
natural language is faar to fuzzy to replace programming (system specification is already famously impossible thing to do right). If you think it truly will replace code, i highly suspect you work om webbdesign, where testing and reliability was always a secondary concern.
And even then, I think were already on the convergence platoe of LLM code. The companies are raising prices as diminishing improvement and balooning compute costs.
There absolutely lies value in studying programming language, but maybe not when reinventing ideas of the past.
> I encourage everyone to create the most absurd, implausible, and impractical languages. Chasing the measurable is often useful, expressing the expressible is insightful, but never forget the true goal of language design: to explore and create what isn’t.
Sorry, but this sounds more like an artsclass to me. Don't get me wrong, there was a point in time where exploration of the unknown was the only way to move forward. But these days we would need greater insights into higher-level language semantics and inherent tradeoffs to guide language-design and language evolution.
There is plenty to choose from and one can learn already so much just by reading up on the Java-EG mailing lists. Brian Goetz has a true academic mindset and I frequently feel inspired when I read his reasoning which is both highly structured and accessible.
Otherwise we would just be left with another compiler class. Compiler basics really aren't that difficult.
> this sounds more like an artsclass to me.
Indeed, it is, and that's the point! Being interfaces to computers for humans, programming languages sit at the intersection of computer science and humanities. Lots of people like to treat programming languages like they're math class, but that's only half the picture. The other half is usability, ergonomics, learnability, and especially community. Not to mention the form of the language is all about aesthetics. How many times has someone on Hacker News called a language "beautiful" or "ugly" referring to the way it looks? When people praise Python they talk about how easy it is to read and how pleasant it is to look at compared to C++. Or look at what people say about Elm error messages versus C++ template errors. Actually a lot of what's wrong with C++ could have been averted if the designers had paid more attention in art class.
> But these days we would need greater insights into higher-level language semantics and inherent tradeoffs to guide language-design and language evolution.
Here's a talk that argues there's much more fertile languages ground for ideas outside of the "programming languages are math" area, which has been thoroughly strip-mined for decades:
https://medium.com/bits-and-behavior/my-splash-2016-keynote-...
This author takes the perspective that programming languages are much greater than the sum of the syntax + semantics + toolchain + libraries, and treating them as such is limiting their potential.
[dead]
[flagged]
Programming languages are obsolete in the LLM era. What current generation AI has revealed is that English is actually the ultimate representation of computer programs and systems, which is both sufficiently terse and precise to economically describe the operation of arbitrarily complex programs.
There is no reason to study programming languages in 2025, other than as a historical curiosity - the same way one may study languages equally as pitiable as e.g. COBOL, Lisp, or MIPS assembly.
I think it's the exact opposite -- LLMs have revealed the precise utility of programming languages. For decades the "English as programming language" has been the holy grail of language designers. From COBOL to SQL to AppleScript, it was the hope that one day we'll be able to program a computer just as easily as we can instruct a person.
Well LLMs finally offer that, and what they are proving is what programmers have known for decades -- natural language is a terrible way to specify a program to a computer. So what is happening in the LLM world is they are reinventing programming languages and software engineering. They're just calling it "prompt engineering" and "context engineering".
What this tell us is that natural languages are not only not sufficient for the task of programming, to make them sufficient you need to bring back all the properties you lost by ditching the programming language. Things like reliability, reproducibility, determinism, unambiguity are thrown away when you use an LLM, and context engineering / prompt engineering are ways of trying to get that back. They won't work well. What you really want is a programming language.
> natural languages are not only not sufficient for the task of programming,
Downthread there is an example of an ICPC problem statement, [0] given as natural language, (modulo some inequalities and example program inputs/outputs) which was sufficient for Gemini to program & implement the correct solution where no other human could.
[0] https://worldfinals.icpc.global/problems/2025/finals/problem...
> problem statement, [0] given as natural language, (modulo some inequalities and example program inputs/outputs)
I also see two graphics, and several formal mathematical expressions. You can't modulo away all the not-natural language and then claim natural language alone was sufficient. I presume these things were added by the authors to increase clarity of the problem statement, and I agree with them. They used formal languages to specify all the important parameters in an unambiguous way, which was the right call! Otherwise we would all be left wondering at the semantics.
Anyway, I don't think this really responds to my point, because competition prompts are designed to be self-contained problem statements that are as clear and unambiguous as possible for competition purposes. And in this case, they switched to speaking in a formal language when being precise and unambiguous was most important.
On the other hand, my statement was about the task of programming, which typically involves solving ill-defined problems with specifications and requirements that shift over time. I've been in programming competitions, I've also been a programmer, and I don't find one to be really related to the other.
> Programming languages are obsolete in the LLM era.
Wouldn't LLM need to directly output machine code for that to be true?
They can, trivially, with tool [0] usage.
[0] https://modelcontextprotocol.io/specification/2025-06-18/ser...
Ok let me put in the missing implicit words.
> Wouldn't LLM need to directly output the correct machine code for that to be true?
zzz..
Good luck debugging and fixing the multitude of errors introduced by your LLM. ESPECIALLY if you're outputting some low-level assembler or machine code. At least when we have our AI outputting Java/Typescript/Python/etc, it's easy (for a programmer) to see where it's run afoul.
I have every confidence that even current generation state-of-the-art artificial intelligence will have far superior facilities at debugging and managing the complexity of even its own bare metal programming output than humans.
Why? Consider Gemini's recent performance at the International Collegiate Programming Contest [0], in which it solved a problem that no other human team was able to solve.
Wetware intelligence is itself obsolete, at least as concerns the domain of computing.
[0] https://deepmind.google/discover/blog/gemini-achieves-gold-l...
> solved a problem that no other human team was able to solve.
That problem: https://worldfinals.icpc.global/problems/2025/finals/problem...
Sarcasm, right?
The best way to get a strong answer is to post boldly incorrect statements. Then I get other people to write my actually-desired arguments for me.
I call it, "wetware LLM prompt engineering".
And the wetware often refuses to do that and simply downvotes you. How dare they.
Like real LLMs it's a bit of a slot machine, but my comment history will assure you that the jackpots are far more attainable.