The main advantage that I see is that his class transparently works with any function expecting a 'sequence', which spec only defines as either a vector (i.e.,

The main advantage that I see is that his class transparently works with any function expecting a 'sequence', which spec only defines as either a vector (i.e., 1-d array) or a list (i.e., linked list) while providing a some (extremely simple) lazy evaluation. I, for one, never realized that you could extend the sequence type in this way in SBCL. For example, if he created an instance of his interval-sequence class with (below 100000000), he wouldn't actually have to allocate the memory for a hundred million element vector (or a 100 million cons cells). Instead, it would just use a handful of bytes to represent the start of the sequence, and the length. By itself, it's fairly trivial. But, this sort of manipulation isn't part of the ansi spec so this post was just sort of letting us know what was possible (a 'hello world' of sorts). I already have a few interesting ideas on how one could use this same technique to provide slightly smarter lazy evaluation, and some sort of time/duration manipulation stuff.