Motivation for Marmalade

Why is Programming So Difficult?
Overcoming Barriers with Tools
Our Approach

The research approach of Project Marmalade is to understand the causes software errors, both psychological on the part of the programmer and technological on the part of the programming system. By designing programming interfaces that prevent the causes of common software errors, we hope reduce the prevalence of software errors, while also helping programmers find and repair errors that slip through.

Why is Programming So Difficult?

Programming is the process of explaining how to perform some complex behavior. Programming is difficult because, in general, explaining any complex behavior requires careful attention to detail. For example, try describing in detail "how to drive." Is your description detailed enough for someone who's never driven a car? Could they follow your instructions without error? How much more detail would you need to help them avoid errors?

Of course, when we explain instructions to people, we don't always need to supply all of the details, since people share common knowledge about tasks, contexts, and meaning. As a result, people can generally deal with ambiguity and adapt to unexpected circumstances.

Computers, on the other hand, know almost nothing about our tasks, and they are horrible at adapting to novel situations. About all they know is how to do is read and write bits. If they encounter some unexpected circumstance, they tend to behave in unexpected ways. It is up to programmers to consider all of these circumstances, and prescribe some reasonable behavior for each and every one.

Overcoming Barriers with Tools

Expert programmers are usually up to the task. They learn to design and understand software architectures, algorithms, and data structures; they learn to read, write, and modify code written in numerous programming languages; and they acquire the skills necessary for determining when a program fails, what program states caused it to fail, and how to change a program to prevent failure.

But even expert programmers don't do this without help. They rely on a variety of tools to help them create software. Even a minimalist UNIX programmer uses gcc, make, emacs, grep, cvs, a shell, gdb, and pencil and paper, not to mention a lot of time and attention. Without all of these tools, UNIX programming would be nearly impossible.

Yet, even with these tools, most studies of programming activity suggest that 50% or more of expert programmers time is spent testing and debugging on average. A 2002 study commissioned by the National Institute of Standards and Technology found that industry software engineers spend up to 70-80% of their time finding and repairing errors.

Clearly, programming tools are important, if not essential to programming activity, but one may question how well current tools help to increase productivity and prevent errors.

Our Approach

In general, our research methodology involves four phases:

Understand how a programmer, the programmer's task, and the programming system interact. What problems do programmers have in trying to program and what's to blame?
Design new interaction styles and programming interfaces that intend to mitigate the problem. This involves lots of prototyping, brainstorming, and user testing.
Implement the solution. This is the computer science part of our research, where we design new architectures and algorithms that implement our design.
Verify the solution's effectiveness. This is the empirical part of our research, where we validate our solution's effectiveness in removing the problems identified in the first phase.

At each stage, we employ a number of different research tools. For phase (1), we use contextual inquiry, controlled studies, and qualitative observations to guide our understanding of the problem. In stage (2), we make use of paper prototypes, and frameworks such as the Cognitive Dimensions of Notations. In phase (3), we design software architectures and consider how components relate and communicate with each other. In phase (4) we perform controlled studies on our new tools to verify that the tool helped.

Motivation