One of the blogs that I have on my feed list recently reposted this tidbit:
There are only two debugging techniques in the universe:
- /* */
Since I recently posted my own little list of some debugging techniques, I can’t resist weighing in on this assertion. While their list is a bit flip, changing behavior and running experiments to see the changes is, of course, a core debugging technique. However, the claim that debuggers are just extensions of this is a bit too reductionist for my tastes – it’s like claiming that a car is just a horse that takes longer to get tired, or that the printing press is just a scribe that works faster.
I also feel that “debugging techniques” shouldn’t be limited to “things you do to the program,” either – thought experiments, code change analyses, and the like are all valid in my book as well. My definition includes any form of investigation that gives you more insight into the problem.
Completely separate from the article linked above, I had a glance at Wikipedia’s article on debugging, and…well, there’s a lot of stuff in there that gets my back up:
- Citing language choice as having an impact on the debugging process is silly – your choice of language may make it more difficult to write buggy code, but once a bug is in there, I’m hard pressed to think of a reason why language choice would have a substantive impact on the actual debugging process. (Saying that C++ makes debugging easier than C because it has single-line comments is not funny.)
- “Generally, high-level programming languages, such as Java, make debugging easier, because they have features such as exception handling that make real sources of erratic behaviour easier to spot. In lower-level programming languages such as C or assembly, bugs may cause silent problems such as memory corruption, and it is often difficult to see where the initial problem happened.”
1) Exception handling is only as useful as the exceptions that are thrown in the code.
2) The claim that “high-level programming languages” like Java don’t suffer from “memory corruption” is misleading. Any imperative language with side effects is going to be capable of bugs that look like “memory corruption,” and can effectively be treated the same way with regards to the debugging process. (As an aside, it’s interesting to see how the definition of what qualifies as a “low-level language” have shifted over the years…yikes.)
- Static code analysis tools are meant to be used to fix code problems before you run into the bugs they cause. Using them as an example of a debugging tool is kind of missing the point of using them entirely. lint isn’t going to help you debug why something is busted – it’s only going to tell you that you’re using an uninitialized variable, and that’s something that shouldn’t have been in your compiled code in the first place.
Addendum to the earlier debugging post: One important quality of the printf and commenting techniques that I didn’t mention in my earlier post, though, is that they are techniques that will work even if you can’t actually attach a debugger to the process (or if there is no debugger for the environment in which you’re working). Sometimes this characteristic can really save your bacon – I remember having to change the screen background color register (the poor man’s printf, which doesn’t even need the C standard library) on a certain console in order to debug the DVD boot loading code of a game. Each line of code was prefaced with a call to set the background color to a distinct value, and the color remaining on the screen when the console froze allowed me to determine the location of the crash (and, eventually, the solution).