On 03/02/21 14:20, Joel Jacobson wrote: > This would effectively mean there would be absolutely no semantic changes at > all. > > I will work on a new patch to try out this idea.
I may have been assuming a degree of orthogonality in SQL that isn't really there ... only in a few situations (like creating an index) do you easily get to specify a non-default operator class to use when comparing or hashing a value. So perhaps a solution could be built on the same range machinery, but requiring some new syntax in CREATE TYPE ... AS RANGE, something like WITH POSITIONS. A new concrete range type created that way would not be a whole different kind of a thing, and would share most machinery with other range types, but would have the position-remembering behavior. Given a range type created over int4 in that way, maybe named int4prange, the regexp_positions function could return one of those. Regards, -Chap
