It really isn't so big a deal; the author of this post is spreading fear, in a way. For example, saying that you need "real world data" to catch "edge cases" l

It really isn't so big a deal; the author of this post is spreading fear, in a way. For example, saying that you need "real world data" to catch "edge cases" like this... no, proper unit tests will do just fine. Also, the problem of splitting different-sized keys in the way the author first tries to do does not strike me as a mistake anyone would fall into or that is hard to get out of.