Hi all, I suggest to improve the documentation related to the upper() function for discrete sets. It should be useful to report some examples using discrete type ranges like datarange or int8range.
I found this discussion on the bug mailing list: http://www.postgresql.org/message-id/20150116152713.2582.10...@wrigleys.postgresql.org <http://www.postgresql.org/message-id/20150116152713.2582.10...@wrigleys.postgresql.org> It states that the mathematical definition used for upper bound is actually wrong for discrete sets when explicitly defining ranges as closed sets and I agree. I was working a lot with dates when I first found this “bug”. Users as me expect that upper(‘[1,2]’::int8range) returns 2 but it doesn’t. From the documentation it’s hard to find why PostgreSQL returns 3 instead of 2, so an example with a brief and mathematically clear explanation should be provided (the explanation provided is not clear at all). Also, from the answers given in the link above I suspect that this function will be never modified to work as expected but I would like to ask if it could be done in the near future. As for now the only way to return the actual mathematical upper bound of a discrete range is to change if from being right closed to right opened. This will return the expected value: SELECT upper(‘[1,2)’::int8range) but this is more a hack than a real solution because that range should contain only the number 1. Given integer numbers, the interval [1,2) translates to {1} but what the user really wants is {1,2}. Again, the upper bound of {1} is 1 while the upper bound of {1,2} is 2. The definition described in the documentation also contrasts with the lower() function. Users like me expect a similar behavior for this function, but it doesn’t and instead is mathematically correct. Regards Pietro Pugni
