Rad wrote on Jan 24^th, 2009 at 4:35pm:

I hadn't heard of that particular text, although Matthias Felleisen and Shriram Krishnamurthi are well-known educators who have done a lot of good work in the Scheme community.

The overall tone of the paper is balanced, but think it's also notable that they don't mention some important things about why SICP is the way it is, and why today's students really struggle so much with it. MIT's student intake is truly exceptional; many of the "problems" talked about with SICP in the paper are complete non-issues, or elevate some modern fashion (viz, explicit methodologies) beyond where they should be. The simple truth is that other schools are dealing with a wider range of student abilities, and as I've talked about before, the level of ability of incoming students to software courses has been falling very substantially worldwide for many years.

Alternative texts like this exist as a practical compromise to broaden the appeal of SICP-like instruction in more universities, by lowering the barrier to entry; the advanced material from SICP thus elided is covered by texts like Essentials of Programming Languages so that in the end educators stretch the same core material into a) more stages to fill out three-year undergraduate courses, and b) be split over those years into different streams of ability, so that elite students can still get a proper classical CS education which will allow them to continue on, while the bulk of the undergraduates get at least some exposure to classical concepts (which most currently do not) during their basic vocational training.

Quote:

The same variant used by SICP and indeed created for the SICP course.

Quote:

While it is a poor decision to use Java for anything, it's rational in the context of the bargain being struck; to make some of the classical material available to institutions which have otherwise abandoned CS education and are just doing low-end vocational training. Those lower-tier institutions - which were the ones teaching Pascal as a programming language back in the 80's - pretty much all switched from Pascal to Java back from '95 through to '97, since that's the way the wind was blowing.

Such course providers have a real problem, incidentally, which is that the industry has split again, with web development now playing a huge role - despite all the dumbing down of courses based on Java, even that is something students are telling course providers is simply "too hard". So really, there needs to be even more streaming of students into different courses.

Which is not a problem itself; it has throughout my life always been the case that there is a huge intellectual gulf separating most "working" programmers from those aware of the intellectual foundations of the field. The only problem being the notion that everyone engaged in writing software is engaged in only minor variations of something called "programming", when that's not the case (although the belief nonetheless persists for equally obvious reasons).

The divide in abilities is real, which feeds into the hiring problem and thus to universities as a feedback loop since recruiters can't really distinguish between the ability of candidates to find the high-ability ones, employers want staff that are readily interchangeable and easily hired, while universities have to supply industry with candidates from a pool whose median ability is changing, but who can't really make that streaming of courses and candidates too explicit.

Quote:

Indeed, but that is because the authors are absolutely steeped in the tradition of the lambda calculus, and Scheme being essentially a vehicle for using it directly to write programs means that it is teaching the universal mathematical essence of computing. Other programming languages have abandoned that essence for the sake of convenience, and present a model of computing which is built around a small set of end tasks their user communities want to perform.

This is known as domain-specific languages; since Lisp (and Scheme) are such perfect vehicles for writing interpreters for other languages, their model of how programs are written is really built on the idea of bridging the gap between computer and problem by creating a tower of interpreters for languages that are ever closer to a particular problem domain.

[ Most of the things which would be held up as "general-purpose programming languages" really aren't general-purpose at all. C and C++ are particularly notable in this category. They are pretty terrible at everything except writing certain kinds of code where their semantic closeness to the way register machines work is useful for high performance, i.e. operating systems or run-time systems for other programming languages. ]

This is in contrast to the popular "modern" languages, none of which really aspire to being "universal". Instead, each of which was created close to some particular end problem space and was built to address only that (the cautionary exception being Java, where the designers did have the aspiration of it being able to be used for everything, but succeeded only in creating something never more than third-best at anything).

On the subject of programming languages, let’s not forget APL (A Programming Langauge), created by Ken Iverson in 1962. Arguably, you can do more ‘work’ with one line of APL code than with any other language.

Quote:

Why is this?

Quote:

Are you saying this is a watered-down version of SICP?

Quote:

which would suggest SICP is more compact, yes?

Quote:

Scheme was created specifically for SICP?

Quote:

Yes, I had this idea .. knew no better.

Quote:

interesting. I didn't know that.

Rad wrote on Jan 26^th, 2009 at 10:44pm:

I don't know; a chap I know with a Ph.D. in Computer Science Education doesn't know either. There are certainly a lot of plausible potential explanations, many of which may contribute to some extent. Bear in mind that this is not isolated to computing, but is affecting a great many engineering disciplines, and it is happening worldwide - it is more striking outside the U.S., but that is because many elite students from other countries study at US institutions (which tend to dominate the global league tables for postgraduate engineering) and so it's mainly apparent in the ratio of nationalities for, say, PhD's awarded by U.S. institutions (where the ratio of foreign-born engineers has been increasing for some time).

Rad wrote on Jan 26^th, 2009 at 10:44pm:

Not quite - it's totally different to SICP, as a matter of deliberate design, to teach different things to a different kind of student.

Those different design choices made by the authors are many; much is omitted outright, other things which SICP only touches on (largely via the exercises, letting readers discover rather than telling them outright) are addressed in extensive detail. That doesn't make it worse, since if it helps anyone not ready for SICP make progress that is a very, very good thing indeed; rather, that it isn't a substitute for SICP.

Rad wrote on Jan 26^th, 2009 at 10:44pm:

Yes, but it's more than that, more like ... the difference between showing and telling. SICP is content to guide; students are asked to perform tasks requiring astonishingly deep insights in the exercises, the confidence of the authors being that the problems cannot be solved without the student undergoing the "Aha!" moment and making those key insights for themselves. A text like HTDP aims to tell lots of simpler, more measurable things that no-one can miss, but in the process it avoids challenging students to reach those deeper understandings.

Again, it's not that one is necessarily worse; but they are very different in aims.

Rad wrote on Jan 26^th, 2009 at 10:44pm:

Actually, the folklore is that it was created to explore a thing called the Actor model, from a paper by a gentleman called Carl Hewitt which was a seminal step in the development of object-oriented programming. However, it's no accident that Scheme is not just a subset of the Lisp dialects of the day, but refounded the language on the classical Lambda Calculus (making it perfect for teaching), and it's no accident that Gerald Sussman was one of the designers and used it in the MIT course he taught.

You might notice that Carl Hewitt's Ph.D advisor at MIT was one Seymour Papert, collaborator with educational theorist Jean Piaget, and whose Logo programming language designed for teaching children is also a dialect of Lisp. Alan Kay also had as one of the motivating ideas for Smalltalk that it would be a vehicle for teaching children.

Rad wrote on Jan 26^th, 2009 at 10:50pm:

In the abstract, it's certainly likely; both languages have certain key characteristics like introspection underpinning them and which mean the language is hard to master without coming to understand how the constructs in the language could be used to create itself. But that understanding isn't inevitable. Of course the only people who learn either nowadays are those specifically interested in those deeper ideas anyway.

Rad wrote on Jan 26^th, 2009 at 10:50pm:

An implementation is the concrete realization of something which is an abstract idea. For a programming language, that abstract something is usually a specification of the behaviour of the language constructs; a program that takes in source code and arranges for it to execute according to the specification is an implementation of that specification.

Writing those programs - the compilers and interpreters for programming languages - involves making hundreds or thousands of tradeoffs, such as that between execution speed and memory consumption, or between execution speed and portability, or between execution speed and having debugging facilities, or many others. The sum of all these different tradeoffs an implementor can make in realizing the specification can add up to some big practical differences.

Quote:

Got me thinking. Trying to figure out why that may be. What are your intuitions?

If nothing else, that makes you skills even MORE valuable.

Quote:

But you would say "easier," no? *Lots* easier?

Quote:

Are there concepts in SICP that some programmers simply cannot grasp? I mean, I feel it's just a matter of time .. in that you can teach brain-surgery or rocket-science to someone of average intelligence .. but it will simply take that person LONGER to learn .. but they will still be able to learn .. given enuf time (and perseverance).

Quote:

I see. Good explanation. (Great, actually.) To be honest, I like someone to spell it out for me .. whenever possible. Eliminates ambiguity. Growing up on the East coast (near New York) taught me to value directness. [ "Rad, you stink," vs "Hmmm, what's that smell?" ] Tho now you have me curious about comparing the approaches used in the two courses.

Quote:

I know you have discussed L-C before, and I have even read the Wikipedia entry on L-C, but could you pls again explain WHY a language founded on the L-C is ideal for teaching???

Maybe its the term 'Calculus' that is throwing me?

Quote:

One of the reason I felt drawn toward HiDP over SICP, even tho my ego would prefer to tackle SICP (and it have the videos, too) .. is that I feel I might be able to teach the course to the Bug when he gets older, or even some local high school kids, after school, since Newport Beach has many wealthy families.

Just thoughts, tho .. but the teaching of children has been a recurring idea running thru my head lately, and I seem to have a knack for conveying complex ideas so folks can grasp them more easily.

Quote:

Can you pls summarize the diffs between Lisp, Scheme and Smalltalk?

Quote:

Uh, I've been reading that term .. just today .. CONSTRUCT(s). What is a construct?

That part about "creating itself" sounds deliciously intriguing .. like an inverse paradox or something.

Before I forget, I want to thank you for your help. I greatly appreciate it. Your knowledge/posts are so rich, I read them multiple times.

Quote:

You are referring to DrScheme, I suppose. I find it interesting that the book/text is dated Feb 2001:

http://www.amazon.com/How-Design-Programs-Introduction-Programming/dp/0262062186

.. with the online manual/version dated Sept 2003:

http://www.htdp.org/

.. but the language (and environment which comes WITH the language) is dated/updated January 21, 2009. (v4.1.4)

http://www.plt-scheme.org/

A date of Feb 2001 does not make the text obsolete?

Quote:

I am familiar with EOPL cuz there is a link to that website/info in the HELP menu of DrScheme, which I've been familiarizing myself with .. so I know it's beyond me right now.

Rad wrote on Jan 27^th, 2009 at 10:38pm:

It's a sort-of pun on the noun form of the English word we normally use as a verb: sense 4 in this reference. A construct (noun) is a constructed thing; programming language constructs (noun) are synthetic (i.e., constructed) ideas, built out of grammar and syntax and semantics. The self-referential joke is that those constructs (noun) exists in order to be put together in different ways to construct (verb) bigger things, i.e. programs.

Rad wrote on Jan 27^th, 2009 at 10:38pm:

Actually, I'd kinda like to skip that for now. The Lisp-vs-Scheme thing isn't that important, whereas a potted summary of how Smalltalk and Lisp relate wouldn't do Smalltalk justice (and wouldn't mean all that much to you if you don't yet know Lisp).

Rad wrote on Jan 27^th, 2009 at 10:46pm:

Lots. That's the entire point of the many textbooks like this or this or of course this are all about analysing their relative efficiency in terms of time (measured in some abstract way) and memory space.

Aside from that, some particular languages (C and C++) avoid taking any responsibility for memory management themselves, pushing it onto the program author (on the somewhat dubious basis that the program author can potentially figure out how to do this better than any computer can) with the result that just keeping track of what memory is in use can in fact dominate the amount of code written (yup, you read that right; it's not uncommon for programs in C or C++ to have more lines of code devoted to wrangling memory than to actually solving the end problem).

Rad wrote on Jan 27^th, 2009 at 10:56pm:

Yup. PLT Scheme is the umbrella for the project that DrScheme is part of.

Rad wrote on Jan 27^th, 2009 at 10:56pm:

Nope. The concepts in it are basically timeless.

Some of the HtDP author's views on development methodology are less timeless, since some of the development practices they talk about are really just matters of fashion, but they are practices which have not yet fallen out of fashion (although hopefully more people nowadays are aware of their limitations than in years past) and really for methodologies they don't become obsolete as much as they get overshadowed by the next Silver Bullet that gets seized upon and promoted by folks who never learned the lessons of the Mythical Man-Month.