Last month I covered a Special Islands challenge. The task was to identify periods of activity for each service ID, tolerating a gap of up to an input number of seconds (
@allowedgap). The caveat was that the solution had to be pre-2012 compatible, so you couldn’t use functions like LAG and LEAD, or aggregate window functions with a frame. I got a number of very interesting solutions posted in the comments by Toby Ovod-Everett, Peter Larsson, and Kamil Kosno. Make sure to go over their solutions since they’re all quite creative.
Curiously, a number of the solutions ran slower with the recommended index than without it. In this article I propose an explanation for this.
Even though all solutions were interesting, here I wanted to focus on the solution by Kamil Kosno, who’s an ETL developer with Zopa. In his solution, Kamil used a very creative technique to emulate LAG and LEAD without LAG and LEAD. You will probably find the technique handy if you need to perform LAG/LEAD-like calculations using code that is pre-2012 compatible.
Kamil’s solution was quite clever.