Software engineering is in the process of turning itself into an evidence-based research discipline. This talk describes how that is happening, why it matters, and a few of the more interesting, surprising, and controversial results to date.
Last updated: March 2017. This talk is an update of one given at CUSEC 2010.
Since its start in 1998, Software Carpentry has evolved from a week-long training course at the US national laboratories into a worldwide volunteer effort to improve researchers' computing skills. This talk will explore the lessons we've learned along the way about applying open source software development techniques to teaching at scale, and about getting people and institutions to change the way they work.
Video from University of Illinois, Sept 15, 2016
Last updated: October 2016
Hundreds of books about writing compilers are currently on the market, but there are only three about writing debuggers. Everyone thinks we should teach children how to program, but undergraduate courses on computing education are practically nonexistent. This talk explores what these gaps and others in undergraduate Computer Science courses tell us about the state of computing today, and about how we could fix it.
Last updated: October 2016
Today's MOOC platforms use the Internet like television. What would they look like if they directly supported some of the techniques described in this useful book?
Allow people to create synchronized voiceovers for HTML slideshows. I've had several summer students take a run at this; the hard part is the authoring tool to add time marks, but as the demo linked in the title shows, the idea itself works.
Can you write a climate simulator in less than 500 lines of Python? What about constructing phylogenetic trees in less than 500 lines of R? This collection would show readers how science is turned into code across a broad range of disciplines. Each entry is 200-300 lines of scientific code, and another 200-300 lines showing readers how to test simple versions of a broad range of scientific applications.
In the spirit of Johnson's GUI Bloopers, this book teaches debugging by working through a hundred short examples, each showing a different kind of fault, a different method of diagnosis, or a different kind of fix.
How do programmers detect and handle errors? This companion to A Hundred Broken Programs would cover everything from type-checking during compilation to exception handling, rolling back transactions, and re-starting servers.
We tried to repeat Stefik et al's study of programming language syntax for languages commonly used in science, but weren't able to get enough subjects. I think it's worth trying again, both for its own sake and to show that this kind of work can and should be done.
Just as MOOCs are mistakenly treating the Internet like television, Caulfield's notion of choral explanations has me thinking that I've been mistaken in trying to treat lesson construction as software development. A "lesson" platform that uses Stack Overflow as its model rather than GitHub or Wikipedia would be fascinating to explore, as would collaborative choral software exegesis.
Jon Udell's Elm City calendar syndication project wasn't really about calendars: it was about teaching people how to think like the web. I have yet to see a better approach, and it would be exciting to resurrect this project and try again.
I firmly believe that the Kernighan Trilogy (Software Tools, The Unix Programming Environment, and The C Programming Language) are the main reason that Unix succeeded. I would really like an update that uses a modern operating system and a modern programming language (preferably functional), borrows heavily from PowerShell and the like, and takes Udell's "seven ways" to heart.
Your graduate degree is in ecology, but now you're running a three-person team responsible for building and maintaining a hundred thousand lines of code? This book (or course, or whatever) is everything you absolutely, positively need to know after you know how to program. (We've started work on this, but more hands would be welcome.)
I often use Sajaniemi et al's roles of variables in teaching, but like the classic design patterns, they were "discovered" by eyeballing novice code. I think that cluster analysis of patterns of class and variable use would uncover more patterns, and confirm my suspicion that some of the classics are really just different names for the same thing.
I would really like to run a one-semester course for first-year undergraduates whose subject was themselves. What happens in your brain when you learn something, and how can you learn more efficiently? What's the effect of sleep deprivation on the quality of work? How can you analyze the social dynamics of a classroom, or change them to make participation fairer? How can you run a small group project? Who decides what gets taught at universities and who gets to take part?