Ignoring the numerous typos and the lack of a working implementation, this stands out:
"Give the last 20 years experience with C, Java, and other “real” languages, perhaps it is time to reconsider, and instead create a language designed for novices."
It's the most mind-expanding, most diverse in the paradigms it supports, and certainly one of the most interesting languages out there. And it's a tiny kernel language the size of Scheme that one can learn piece-meal. You can use it like Pascal, or Scheme, or Prolog, or Self, or Java, or Erlang ..
Mozart/Oz can teach newbies, no doubt. But it can't be taught by anyone. It requires an open mind and someone who is a competent programmer, not just an assignment pusher. I don't think most profs and TAs will be able to teach it as well as they should, but students can learn it on their own.
By the way, I know a few of the names there and I know how completely different their computing tastes and personalities are. I would love to see the archives for the discussion mailing-list for this one, will be very entertaining .. and informative :-)
I don't understand how "suitability for teaching" and "suitability for industry" differ, except that industry naturally wants to stick with whatever it already has.
But if you can design a language that serves as an excellent medium for teaching all of the important ideas in Computer Science, why wouldn't you go on to build the libraries and infrastructure you need to make it viable in industry?
> I don't understand how "suitability for teaching" and "suitability for industry" differ
Industry is made up of people who have by necessity cultivated a patience with excessive complexity.
In particular, I suspect that many unnecessarily complex or counterintuitive language features arise in imitation of what the designers have learned to live with in other languages, even as their focus is on "improving" some aspect that they think is important for industry.
If you did design a language which favored learnability at every opportunity you might not end up with something that flattered everyone's preconceived notions of what is suitable for industry, either.
I hesitate to name examples here, as there is a significant risk of trolling...
It's simple. Designing a language that clearly illustrates CS fundamentals and a designing a language that supports maintenance of huge codebases by teams of programmers of varying ability are two very, very different problems. It's no accident that CS profs love Scheme and banks love Java.
What really matters is how well your language supports your primary level of abstraction. The turtles-all-the-way-down nature of elegant languages like Smalltalk or Lisp doesn't really make that much difference when you're mostly manipulating high-level third party APIs and where mature tools and predictable, easily readable code count for more than the ability to roll your own new object system in the core language.
First, I am under the impression that a great many CS professors teach CS using Java.
Second, are you suggesting that since banks love Java, then Java is a language that supports maintenance of huge codebases by teams of programmers of varying ability to a greater extent than Scheme or any other language?
Third, I am confused by the suggestion that "elegant" languages are somehow unsuitable for writing "predictable, easily readable code." Can you expand on this?
Update: Fourth, I forgot the essential point I was implying in my original question. What is there about "maintenance of huge codebases by teams of programmers of varying ability" that isn't a CS fundamental?
Sure, if you are talking about Computer Science as a study of the properties of a formal system, this is irrrelevant and lost amidst talk about category theory and turing equivalence. However if you look at CS in a wider context you have professors teaching things like encapsulation, which has certain important implications for the study of formal systems but also has important implications for "maintenance of huge codebases by teams of programmers of varying ability."
This is part of the motivation for Wirth expanding Pascal into the Modula family of languages: To create a language that was useful for teaching and practical for the construction of the Lilith workstation.
I can't imagine anything about "maintenance of huge codebases by teams of programmers of varying ability" that isn't a suitable topic for teaching and especially research.
My impression is that most CS profs teach Java because they've been forced to and that they're not very happy about it. Which makes sense, because Java is a terrible language for teaching fundamentals.
And yes, I do think Java is much better than Scheme for the kind of software a bank wants. The rigid syntax, static type system, and built-in OO, and lack of macros make it easily to build very powerful tools, which you need when you're wading through a huge, unfamiliar codebase. These same features make it pretty easy to jump into an arbitrary codeblock and understand what it's doing. Scheme, for example, is so flexible that you can design your own exception-handling and iteration constructs, which nobody but you can read.
As far as elegance goes, it counts mostly in the small. Quicksort in Haskell is a thing of beauty but most production code looks nothing like this. The vast majority of real production code is just API glue and simple iterations and elegance just doesn't enter into the discussion.
Your arguments appear to be of your own construction, not of mine. I never said Scheme is an ideal language for teaching, you did. Yet in the same thread, you deride "elegant" languages as only being elegant in the small. This thread is about devising a new language for teaching. If the Scheme is only elegant in the small, why would we use it for teaching? We would devise a language that isn't anything like Scheme if we believed what you profess to believe.
If "rigid syntax, static type system, and built-in OO, and lack of macros make it easily to build very powerful tools," why wouldn't we teach with a language featuring "rigid syntax, static type system, and built-in OO, and lack of macros?"
Overall, I get the strong impression that you like Java and don't like Scheme. Have at it, you are not alone. But there is no reason for me to debate that with you, since at no time have I said that I prefer Scheme to Java for teaching or for industry.
My claim is, essentially, that if we are designing a single new language, then whatever characteristics we think are appropriate for demonstrating CS concepts will also be appropriate for business. I voiced no objection to static typing or rigid syntax or lack of macros. You did.
If "rigid syntax, static type system, and built-in OO, and lack of macros make it easily to build very powerful tools," why wouldn't we teach with a language featuring "rigid syntax, static type system, and built-in OO, and lack of macros?
Because the point of CS education is to teach concepts, not tools, and all of those features actually obscure the core ideas. It's the same reason that code examples in textbooks aren't littered with the kinds of error handling code that real production code requires. It's the same reason that medical students work on cadavers before getting anywhere close to a real operating room and all the extra risks and distractions that entails.
Overall, I get the strong impression that you like Java and don't like Scheme.
No, I'm just not one of those people that tends to assume that industry programmers are either stupid and/or ignorant for hesitating to build commercial apps in research languages.
You equate "teaching language" with "research language." You believe that a language with rigid syntax, static typing, and built-in OO would obscure the core ideas of computer science. Fine. You believe that. I'm sure you grasp that others, especially the list of people mentioned in the post, do not share your views.
So, what we have here is that you have certain beliefs about what makes a language suitable for teaching and what makes a language suitable for industry, and on the basis of your views, you do not believe that the same language would be suitable for both purposes. I accept logically that your conclusions are compatible with your beliefs.
I don't understand your last sentence. It seems to be some sort of axe grinding, as I certainly don't recall saying anything about industry programmers being ignorant or stupid.
So all together, I can't think of anything new to add to this thread. I feel like a foil used for you to set up and demolish your own arguments.
Having taught/TA'd first-year university programming courses in both Scheme and Java, I find the idea of a teachability-oriented imperative OO programming language quite appealing.
Our first-semester course used DrScheme, and the simplicity of language and IDE helped us focus on teaching a structured approach to programming, with unit tests first etc. Also, whenever the program did not behave as expected, we could go through the program with Scheme's eyes and mentally apply the evaluation rules.
In the second semester, students are supposed to learn Java, and that was challenging because you really have to understand static methods and String arrays just to write a Helloworld program of which you can explain every line. Also, Java rewards using public field accesses instead of getters&setters (in the short term), which is just the habit you don't want your students to adopt. I could go on with arrays vs. collections, classes in the io package which should be interfaces etc...
So if there was a clean statically-typed modern OO language, I'd look forward to using it.
> New languages such as C and Scala have learned lessons from Java's design (... blah blah ...), and also retain or duplicate many features for backwards compatibility with C or Java.
I think that somehow a few special characters disappeared here; maybe they were eaten by some kind of text processor or something. I think the first line should be "C or C++", the second one "C# and Scala".
BTW, I'm just being a messenger here - I got this link off Twitter and thought it'd be of relevance here.
I can't muster much enthusiasm for the initiative myself. My gut reaction is "oh no, as if we needed one more", and I winced at the blog post which said "the most important thing in a new language is that you can println 'Hello world'". I can't find any evidence of design thinking that strives to achieve the lofty objectives set out in the PDF paper.
I'm not a fan of Teaching languages, teach algorithms and clean concepts. It takes much time to get good at a language you might as well be using one to learn your algorithms with.
And not just the language but the Version Control, Libraries, Documentation Tricks, Culture, and Relationships that come with truly learning a Language.
"Give the last 20 years experience with C, Java, and other “real” languages, perhaps it is time to reconsider, and instead create a language designed for novices."
Such language exists, and it's called Oz.
http://en.wikipedia.org/wiki/Oz_(programming_language)
It's the most mind-expanding, most diverse in the paradigms it supports, and certainly one of the most interesting languages out there. And it's a tiny kernel language the size of Scheme that one can learn piece-meal. You can use it like Pascal, or Scheme, or Prolog, or Self, or Java, or Erlang ..
Mozart/Oz can teach newbies, no doubt. But it can't be taught by anyone. It requires an open mind and someone who is a competent programmer, not just an assignment pusher. I don't think most profs and TAs will be able to teach it as well as they should, but students can learn it on their own.
http://www.info.ucl.ac.be/~pvr/paradigms.html
Click on the PDF poster there and see how broad the language is.