Stroustrup: What Should We Teach New Software Developers? | Hacker News Reader

Stroustrup: What Should We Teach New Software Developers?

cacm.acm.org

61 pointszzygan40 comments

I can't believe no one's mentioned:

  Computer science is no more about computers than astronomy is about telescopes.
  -- Edsger Dijkstra

Contrary to the beliefs or desires of some, colleges and universities are not vocational training institutions. In the actuarial world, there is a rigorous credentialing process involving a series of exams usually taking a minimum of 4.5 years to complete. However, even with all of the exams passed, without experience you'd be put into an entry-level position. If your experience is in one field of insurance and you want to get into another, that experience won't count for much in the new field and you'll probably be starting near entry-level again.

That's for good reason. Academia is for teaching you known facts and theories, and how to think for yourself. Experience in an industry teaches you how to do a particular type of work so that other people are willing to give you money in exchange for your labor. These are inherently different activities, though the former can weed out some inappropriate candidates for the latter.

The purists complaining that CS isn't about programming are correct but misguided: the vast majority of CS majors don't _want_ to study "real," abstract CS.

IMO the right thing is to offer degrees in Software Development, which would include some classic CS curriculum but by no means all, and we could have discussions about the right way to teach programming without people getting bent out of shape that Programming Is Not CS.

I'm a CS major, and while I admittedly wasn't fully aware of what "computer science" entailed, I knew enough to know I wanted to study it. And I don't regret it.

Yeah, same here, except I graduated in 2004. I took CS from an engineering school, so I had a little less theoretical background (that was replaced with EE / CompE basics). But I agree -- I had exposure to big-O in h.s., but didn't realize that CS would involve so little programming (I probably wrote under 2k lines of code in 4 years, to be honest... but I was a slacker too ;) ).

IMO, if one is to choose between theoretical background and EE/CompE, the latter is significantly more valuable. Of course, both are important.

But CS is about programming. However, in order to really understand programming (that is, making computers do things), it's important to understand "algorithms, data structures, machine architecture, programming (principled), some math (primarily to teach proof-based and quantitative reasoning), and systems (such as operating systems and databases)". Similarly, a real artist needs to know the different constructions of canvas, consistency and production of paint, brushes, etc., in order to use his tools and medium to the fullest extent.

CS seems to have the strange belief that CS is pure and SE is just vocational stuff. If that's true then everyone should do maths instead.

It's like only teaching general relativity and then making someone a civil engineer - after all they know the fundementals of gravity so the bridge should stay up!

What he says sounds good in the abstract, but I worry about how it will be implemented. For instance he raises the issue of licensing people to practice computer science. However as he acknowledges that it is difficult to even identify who should be in charge of deciding what should be tested for such a license.

And that is a big issue. A license created to meet the needs of business 40 years ago would certainly have required learning COBOL. A license created 20 years ago would have no relevance to web development. One created today would likely be geared towards the needs of large organizations with large code bases - which would put most agile development techniques, languages, and practices off the table.

For better or worse the programming world has diverse communities with rapidly changing, diverse needs, and no possible solution has any hope of fitting all of them.

Licensing is an afterthought to his main thesis. His main point is that we have to remove the disconnect between industry and academia. Both parties take pride in being completely separate from one another. This has to end, and that is a cultural change, something that is completely within the realm of doing in less than 10 years.

As for the relevancy of licensing, I think there is a certain degree to which one could start certifying software engineers on a project management perspective, because the principals of how one approaches a project and brings it to completion aren't that different from any other project in any other industry. If you're running your project correctly, programming is maybe 10% of the project. That other 90% has a lot to learn from the rest of the industrial fields to catch up and start adopting standard practices.

A user's requirements are going to be the same no matter what programming language you use. The way you test the software is going to be the same no matter what programming language you use. Within a specific programming paradigm, how you build the software is a shear matter of syntax, all the parts are going to have to be the same. Just as you can have a suspension bridge versus a stone bridge, I think you could study object oriented programming versus functional programming, and I don't see any reason why we shouldn't expect professional, licensed software developers to know both.

COBOL, C, C#, Clojure, C-whatever-next, it doesn't matter. Wood, stone, and iron. The principles haven't changed much at all.

I think there is a certain degree to which one could start certifying software engineers on a project management perspective, because the principals of how one approaches a project and brings it to completion aren't that different from any other project in any other industry.

Huh?

Project management on a website that has an incremental 1 week release cycle has no relationship to project management of an internal database upgrade that is planned for 6 months or project management for developing a shrink wrapped game over 2 years.

The problems, documents, workflow, problems, types of people you need involved, etc are radically different in these three cases.

I disbelieve.

I've been in a situation where I was creating a reporting system, and was sitting in a group of people building a website. There was absolutely no similarity in requirements, testing, rollout procedures, and turnaround expectations between me and the other developers. This despite the fact that we were working in the same organization, on projects specified by the same product managers!

As for a specific programming paradigm comment, the statement seems to me to be vacuously true because you define "paradigm" to include the internal structure of the software, or else false. Because I've seen many cases where two people deliver essentially the same functionality using the same basic toolkit, but the internal organization, pieces, etc are completely different.

COBOL, C, C#, Clojure, C-whatever-next, it doesn't matter. Wood, stone, and iron. The principles haven't changed much at all.

Huh? COBOL doesn't even have recursion. C has function pointers but no closures. C# has closures but no Lisp style macros. Clojure has all of the above. The capabilities of the languages are different, and therefore the proper design and structure of software is as well.

Also where principles have not changed, people's understanding of them may. For instance take test driven development. It isn't new - when Larry Wall delivered Perl 1 in the mid-80s it came with a reasonably complete test suite. When CPAN was founded in the mid-90s the whole perl Makefile.pl, make, make test, make install routine was already standard. And the result was that Perl got a good name for being portable in practice.

However as recently as a decade ago you'd have been hard pressed to find many advocating a test driven style of development. Problems existed that it would work well for. Useful enough tools existed for people to take advantage of them. But there wasn't wide spread acceptance of that knowledge.

Today if you are a programmer and don't know the basics of unit testing, you need some remedial re-education. And a lot of organizations don't use it.

In short unit testing today is where source control was in the late 90s which is where structured programming was in the 80s. It is a good idea. It is widely recognized to be a good idea. But it is a good idea that is more honored by the breach than the observance in your average programming shop.

The problems, documents, workflow, problems, types of people you need involved, etc are radically different in these three cases.

How so? In the web site case I need a person who can code and person to do art/design. In the instance of a game guess what I need a person who can code and someone to take care of the art and design. The only real difference seems to be a matter of scale. Every project in the software development world consists of two things content and code.

The capabilities of the languages are different, and therefore the proper design and structure of software is as well. Hey guess what the properties of wood iron and stone are just as different and require different designs or do you think the sears tower could have been made of wood with out a change in its design?

You really think that the only thing you need to produce a shrink wrapped game that is in production for 2 years is someone to code and someone to take care of art and design? And you really don't understand why I would say that project management on that project is very different than project management on a website that delivers incremental revisions every week?

Sorry, but you clearly have some things to learn. Go pick up http://www.amazon.com/Software-Project-Survival-Guide-Practi... and learn what a stand alone project on the scale of that game needs to do just to have a reasonable chance of success. Virtually none of that book applies to the incrementally developed website.

Oh, and don't forget that the game has no chance of success without a lot of non-software work such as licensing and marketing.

The only problem that he actually mentions is:

Also, many students fail to connect what they learn in one class to what they learn in another. Thus, we often see students with high grades in algorithms, data structures, and software engineering who nevertheless hack solutions in an operating systems class with total disregard for data structures, algorithms, and the structure of the software.

That is to say pass fail grading of programs, grading should be similar to an English class how you express it is just as important as what you express.

I wholeheartedly agree with you on the grading aspect. Even in classes where professors or TAs effectively grade and comment the code, I think most students have a tendency to see the grade they received and ignore the professor's feedback. The program worked, they got the grade, onto the next assignment.

I had two classes this past quarter taught by a professor with a very different grading system. He used a program he wrote called the "code critic" to have us submit exercises into his grading queue. Only working code could be submitted and we could only have three assignments in the queue at once. He would then use his program to return the code with criticism added on any lines not to his liking and mark the assignment as "Not done," "Almost done," or "Done." If he didn't mark it done, we had to fix the code in response to criticism. These criticisms were very picky, even sometimes dealing with the naming of functions and variables.

At the end of the quarter we were graded on how many assignments we were completed, and how much we grew as programmers. If our assignments at the end of the quarter did not require any resubmissions, this was reflected in our grade.

I think this method was an extremely effective alternative to traditional grading. I learned more about programming and style in both C++ and Lisp in these two classes than I have in any other classes to date.

even sometimes dealing with the naming of functions and variables

Ugh. Asking a CS major to adhere to naming conventions is like asking an art major to color within the lines. It has very little to do with computing.

Your analogy does not make sense to me. The professor clearly wanted his students to become not just better at hacking together solutions, but presenting readable, consistent code. This is a noble goal, and you're correct it has little to do with computing. It has to do with communicating to other people what you have done.

Good point the analogy definitely doesn't completely account for the adjacent field of software engineering, where coding styles are certainly important. I guess that's mostly because artists rarely work in large groups, and when they do, I'd imagine the work is mostly parallelize-able: several people don't need to go through the details of any single piece of work.

Animation. Easy to work in parallel but output is very constrained stylistically. Multiple author books tend to have stylistic constraints as well otherwise the book suffers from an uneven feeling.

I find it interesting the he suggests requiring a minor in another field. I largely agree with this, but at my school, you can't get a minor if you're a CS major. For instance, by the time I graduate, I will have the required credits for a minor in math. That minor won't be recognized by the university, and I can only say that I've completed the courses for a math minor without actually earning one.

The same is true for double-majoring. I have a friend that wanted to double major in CS and linguistics to do AI, but he was told that he couldn't. Instead, he has to stick around for an extra year to get two distinct degrees. He's had to waste time because some of the courses he needed to take required you to have declared a major in that field.

This is from a school that probably has a more vocationally applied view of a CS degree. We are, for instance, required to take two semesters of software engineering. They're obviously interested in giving a student an education that will be useful in the industry, but they also have inane rules like the one described above.

It seems unproductive to respond to this discussing how difficult it would be to fix/implement. I think Stroustrup probably knows that better than most anyone else, given, as he noted, his unique perspective.

What is productive is gaining acknowledgement that this problem exists, as that is the (cliche) first step. I believe this is a serious problem. I've seen the sub-par software, code, and the people writing it for years - it's not changing through academia or industry. Both sides continue to address their needs independent of one another, and that is far from optimal.

Many of the finest professionals I have met in the industry cannot say more than a few positive words about their CS programs. Most of them will also recommend against pursuing the degree if you actually want to excel with code.

Agreed, the fix(es) will not be an easy find, but it is far from impossible. This needs to be taken care of - I'd love to be able to tell friends that want to be a good coder that they can go to school for that.

This reminds me of Andrei Sakharov, the father of soviet nuclear bomb who became an anti-nuclear activist later in his life. OOP makes writing bloatware so easy, you don't need to think, just reuse someone else's classes, and this is what they teach in CS 101. Ironically the author is one of the creators of the problem.

I don't think these problems are unique to computer science.

No, they aren't. What is unique though is that we have such a complete reliance on software without any sense of how to do it well. We know how to build skyscrapers without them falling down when the first real "user" steps through the front door. We don't know any sort of thing for software development.

We can't afford a single skyscraper collapse, but we can usually afford to use bad software.

The article sounds like Dijkstra in 1988. My bet is another 22 years won't change anything.

...and, let's face it, the relative complexity of a skyscraper compared to a computer software program, given equal cost, is very, very low. A skyscraper costs hundreds of millions to construct. An equally costly software project would be something like the entire lifespan of the .NET Framework including the CLR, Standard Library, C#, VB.NET, Silverlight, all of the "Enterprise" features, etc.

Also, software isn't really construction. It's built by the people who design it, for the most part. Imagine a skyscraper (or car, bridge, etc) built entirely by engineers. Yikes.

We can't afford a single skyscraper collapse, but we can usually afford to use bad software.

In fact techniques exist to produce software with near-zero defect rates (primarily relying on extremely thorough code inspection, IIRC), but they're far too expensive for commercial use.

Except it's is the only field that believes CS an SE are the same thing.

You don't expect a mechanical engineer to know the chemistry of a steel melt, and you don't expect a chemist to know about engines.

Yet in CS we either teach the detailed chemistry and expect the student to pick up the engineering themselves - or we teach just engineering without explaining what a metal is

You are never going to be able to license software developers as a whole. Maybe some small subset, but the barriers to entry are just too low for a widespread licensing system to be imposed. Software is too fluid for some standards body to say 'you must know XYZ', we are constantly inventing new and better ways to write software.

Also remember, this is the guy who gave the world C++

"We are constantly inventing new and better ways to write software". Name one "new" thing in the last 10 years. Whatever you come up with, I bet there is a reasonable analog from 20 or even 30 years ago. We know practically nothing concrete about software development. Even the concepts of RAD and Agile which are so popular today have been around for 30 years.

"Also remember, this is the guy who gave the world C++". Yes, one of the most popular object oriented languages that is still suitable for systems programming. What of it? For all the complaining about C++ as an ideal tool, it's been incredibly successful as a practical tool. I don't use it, but I have a lot of respect for what it has done.

I currently make my living with Linux, and yet I get this nagging feeling that if programmers were licensed, I would have to have learned any number of Microsoft technologies (OLE, COM, DCOM, ActiveX, .Net) depending on when I got licensed.

I entered college in 1987. Windows 2.x had just come out (I think)---OS/2 wasn't even being designed as of yet. The Intel 386 had just come out, and major application design in PCs in assembly had just passed in favor of C or Pascal. I couldn't even have comprehended my job today (programming wise---web based applications written in PHP) back then.

Yes, there hasn't really been anything "new" in the past 20 years (except maybe for ubiquitous networking) but the details sure have changed a lot.

Whether or not an appropriate, relevant licensing scheme would be developed is a different issue from whether or not it could be developed.

As for the details changing, yes, programming has changed a lot, but programming is such a small part of any project.

Bang! Pow! Super-Stroustrup defeats an entire platoon of evil Straw-Men, single-handed!

I think this article was point on. I wish more people in our industry (software developers) understood and tried to fix these problems.

Surely there is a better source of "What Should We Teach..." than the fellow who pulled C++ out of Pandora's box?

That's sort of ad hominem. Would DHH be a better choice? I mean honestly, I find it hard to take seriously that Bjarne Stroustrup is not credentialed to be a voice in this discussion.

Another CS professor: "I never code."

Sort of like a linguistics professor saying "I never talk".

Really? How so?

Linguistics is the study of language, but CS at its core is the study of algorithms and theory. A CS prof never coding is more akin to a Physics professor never designing his own bridge. Sure, he technically could if he wanted to, but that's probably still better left to the engineers.

Dijkstra said (rightly, IMO) "Computer science is no more about computers than astronomy is about telescopes." I think a corollary to that is that it's also not about programming/coding per se, but the theoretical study of techniques we use when building programs.

I think that's just going too far. I think the analogy to a Physics professor never building a bridge is a bit silly. You might as well say a "Physics professor that never swims or enjoys a warm breeze on a hot summer's day."

A Computer Science theoretician's work is somewhat more closely tied to programming. I mean, let's be serious.

What do you have in mind when you say "computer science theoretician" ? Do you mean someone who publishes in the ACM Symposium on the Theory of Computing or do you mean something else?

Regarding computer science as a discipline, please read:

http://news.ycombinator.com/item?id=968013

http://news.ycombinator.com/item?id=690798

Reply on news.ycombinator.com