Actually it is. Or at least it's the best we have. When it comes to anything you can measure, the evidence points to program length as the best predictor of quality of code. And LOC is, for better or worse, the standard metric for code size.
We had a whole thread on HN about this a month ago. Specific studies are cited there. http://news.ycombinator.com/item?id=3037293
Speaking of which, someone later told me that Chapter 27 of McConnell's Code Complete cites studies according to which bug count grows as a mild exponential function of code size. I don't have that book handy. If anyone does, can they look at Chapter 27 and tell us what it says?
They report the following table
Project Size (LoC) Error Density (Bugs / kLoC)
0-2k 0-25
2k-16k 0-40
16k-64k 0.5-50
64k-512k 2-70
512k or more 4-100
So yes, the error density increases with LoC.In this case, I decided to implement it in a lisp after noticing that Pike's algorithm worked by consuming a character at a time off of both the pattern and the input text, lending itself naturally to a functional recursive implementation. Pike's version in C does this by advancing pointers, but it spoke equally well to an idiomatic car/cdr approach in a lisp (the recur/first/rest pattern you see in matchhere).
The LOC measure may be important when, for example, the goal is to show the number of operations to be as small as possible.