On Tue, Jan 3, 2012 at 7:40 PM, Thiago Braga Nobre <thiagobraganobre@gmail.com> wrote: > A practical example is the scheme substring (text, integer, integer), if > I step substring (row ['name'], start, end) but for some reason they start > and end string causes an exception, thesubstring function should not perform > the conversion from string to integer?
Eh... maybe the other way around?
It has long bothered me that, beginning in 8.3, we can't figure out that substring(123456, 2, 2) isn't intended to cast 123456 to text. There just aren't that many other sensible things for that to mean. It's not exactly clear what to do about it, though, since as far as the backend knows we could have intended to call substring(bit, integer, integer) or substring(bytea, integer, integer) rather than substring(text, integer, integer). How do we make a principled decision to prefer text? I don't know, but I wish we had a way.
