Yuvi Panda

JupyterHub | MyBinder | Kubernetes | Open Culture

Paper notes: ‘The impact of syntax colouring on program comprehension’

I’ve recently started reading more academic papers, and thought it’d be useful to write notes about them and publish them as I go along! This one is for The impact of syntax colouring on program comprehension

  • I was amazed at the amount of prior research it is citing. Why have I not been reading these for the last 10 years of my life?
  • Apparently it is ok to report findings with a sample size of 10 people. I do not know how I feel about this.
  • The fact that there’s a large amount of thought put into the design of the experiment is quite nice, and surprisingly different from environments I’ve worked in the past where product managers designed ‘experiments’
  • To avoid datatype-related confusion, a uniform variable naming scheme was adopted in the tasks. For example, integers were named x, y, etc. and lists were named list1, list2, etc.. As someone pretty used to Python, I would have found this annoying – but I’m curious what the effect of identifier names is in program comprehension. It also reminded me I haven’t written any code in a stronger typed language in a while (I don’t think Java counts)
  • They used Solarized Color Scheme, which has a lot of fans although I’ve never been one.
  • Lots of self-reporting for ‘programming proficiency’. This is the ‘we give up!’ answer to measuring programming proficiency, I guess :)
  • We gathered data from 10 graduate computer science students at the University of Cambridge. This too seems fairly common, but I’ve no idea if such an un-diverse group of student group being studied affects the results at all?
  • They also discarded data from 3 of the students because they wore glasses and their eye-tracking hardware could not really deal with that. So this entire paper is from data from 7 students doing one particular course from one particular university.
  • We use the Shapiro-Wilk test to establish normality. We use the Wilcoxon signed rank test (WSRT) for paired nonparametric comparisons. I know some of these words!
  • As the data was not normally distributed, a 2-way ANNOVA could not be used to investigate the interaction of experience with highlighting on task times I know most of the words, but still can not make sense of this sentence.
  • Currently feeling very illiterate, but am sure this is just a feeling that will pass.
  • . The median difference in task completion time was 8.4s in favour of highlighting. To my untrained brain, that does not seem that much to me.
  • The presence of syntax highlighting significantly reduces task completion time, but the magnitude of this effect decreases as programming experience increase – this is their primary conclusion, which I can totally believe. But would I have believed it if they had come to a different conclusion? Would they have published it if it had? Would they have if there was more data? I don’t fully understand / know Academia enough to know.
  • I wonder if there has been research into richer forms of syntax highlighting – not just keyword based ones, but more contextual. Perhaps based on types (autodetected?), or scope, or usage frequency, or source, or whatever.

Overall, I enjoyed reading it – good paper! Thought provoking in some forms, but could’ve aimed higher, I suppose. I hope they continue doing good work!


Share