in reply to Advanced Sorting - GRT - Guttman Rosler Transform
This is a key point. I've seen it overlooked in two cases, both leading to subtle sorting problems of the "well, this stuff sorts most of the time" variety.
A naive read of Guttman-Rosler is "hey, let's cram the key parts together in a way we can uncram them later." That's necessary, but not sufficient. Simple concatenation won't do. You must pack(), or concatenate in such a way that the keys are aligned for a lexicographic sort. If you see someone use join in a Guttman-Rosler search, watch out!
Consider the difference between
If you don't align the keys (typically via pack or a carefully crafted sprintf), you risk getting composite keys that sort correctly most of the time. A variant of the problem can occur with non-numeric strings when joining them with a character like '|'.Naive join Aligned keys foo 47 foo 47 foo 103 foo 103