Why teach PhD core courses, with programming languages

17 Jan 2012

In early 2010, a senior colleague challenged the idea of teaching a required course on programming languages in the PhD core curriculum. Now said senior colleague is beginning to challenge the very idea of a core curriculum.

This raises the question whether

intellectual disciplines, such as CS, have a core and, if so, how to define it.

I will rephrase this question as follows to make the answer obvious:

Is there fundamental knowledge in CS that, in principle, is needed in every other field.

Conversely, such collections of knowledge make up the core of the discipline. Once we ask the question this way, we have a simple test for core knowledge. We point at another field within CS, say graphics or security, and question how some candidate for ’core’ness may contribute.

For programming languages, in particular, this procedure shows how deeply it is connected to almost everything we do. Take security and languages. For decades, many programmers and researchers have failed to understand type safety and memory safety, and, as a result, we have suffered exploits due to out-of-bounds errors and null pointer accesses. Safety follows from the main theorem of PL, also called the type soundness theorem. Any self-respecting PL course should introduce the theorem and sketch a proof. How about graphics and languages? Years ago I sat through a colloquium talk at Rice where Tony DeRose explained the design of a small language for describing geometric surfaces. From what I understand DeRose went to Hollywood and turned his research into award-winning animations at Pixar. Program verification? If you don’t understand the semantics of languages, how are you going to verify the behavior of programs in this language? Hardware anyone? A wide spectrum of programming languages ideas, especially ideas from the functional world, have been turned into hardware description languages and some became the basis for successful companies. Go on, pick your field and you will see that knowledge from programming languages can help.

For algorithms, systems, and theory, we can obviously make similar arguments. No need to waste space on these.

For other fields, say graphics, it is impossible to show such universal usefulness to other areas in CS. On occasion, however, researchers in graphics encounter problems that challenge people in core areas. It would be foolish of programming language researchers to ignore such application problems. They re-prove how important a core area is and how many contributions it can make. Quite the opposite, the day application areas no longer pose problems for a core area, the latter is doomed to the dustbin of history.

Now that we can distinguish core areas from others, we may raise the more important question of

whether the common PhD curriculum should focus on core areas.

In theory, the answer is obviously affirmative. A discipline that doesn’t expect its PhD students to know the basics of the common core is intellectually bankrupt. How else can a graduate of a PhD program defend the integrity of CS itself? Good core courses will enable our PhD students to articulate the scientific nature of our discipline. The opposite is to short-sell CS as a random collection of hacks and tricks that make nice little entertaining machines work smoothly and securely. This is a goal worth working for, and a core is what is needed.

Even in theory, teaching the basics of the core also has practical value. Given that (almost) every other field can benefit from knowledge of a core area, it is likely that, over the course of a career, a researcher will encounter a situation where this core knowledge helps solve a domain-specific problem. While it is unlikely that a core course teaches exactly the needed knowledge, a well-trained researcher knows where to start the search from a shred of a cross connection.

In practice, the answer is a function of many different variables and depends on the individual department. — If the department has capable and inspiring researchers in the agreed-upon core areas, the theoretical answer is the practical answer. I’d rather not address the problem for a context where the antecedent isn’t true. — If the department is particularly strong in an application area, say robotics, the question should not become whether students should see robotics in lieu of a core area but whether there should be room for a required course in a specific application area. As I have argued above, application areas keep core areas honest by challenging them with fundamental problems, and I consider it obvious that PhD students should therefore be exposed to application areas. When the department has strong application areas, students should probably give a choice of application area courses.

An alternative practical answer is that PhD students should always come from sufficiently strong undergraduate programs. In other words, whatever undergraduate programs teach in a core area is good enough, and when students arrive at their new program, we might as well get on with the business of training them in our favorite fields. This answer overlooks a major problem, however. Many undergraduate programs cannot teach core areas at the research level, because they lack suitable faculty. Let’s make this concrete. No, this situation is not ideal. It places under-prepared PhD students along side well-prepared students and thus creates tension within a class of PhD students. But life isn’t fair. We can only try to give people a chance. A typical ACM-style undergraduate course teaches a "survey of programming language" course. In such a course, students get to see a smattering of programming languages and write a few hundred lines of code in languages that are not taught in the introductory course. Sadly this kind of course is completely inadequate to convey the underlying principles of PL; it’s as if we were to teach engineering-calculus to math majors and say that’s good enough for an understanding of topology. In short, if we do not wish to reject the graduates of such programs out of hand, we must offer them courses on the core of CS once they arrive in our PhD programs.