On Tue, Mar 2, 2021, at 06:22, Isaac Morland wrote: > On Tue, 2 Mar 2021 at 00:06, Joel Jacobson <j...@compiler.org> wrote: >> I find it strange two ranges of zero-length with different bounds are >> considered equal: >> >> SELECT '[7,7)'::int4range = '[8,8)'::int4range; >> ?column? >> ---------- >> t >> (1 row) >> >> This seems like a bug to me. What am I missing here? >> >> Unless fixed, then the way I see it, I don't think we can use int4range[] >> for regexp_positions(), >> if we want to allow returning the positions for zero-length matches, which >> would be nice. > > Ranges are treated as sets. As such equality is defined by membership. > > That being said, I agree that there may be situations in which it would be > convenient to have empty ranges at specific locations. Doing this would > introduce numerous questions which would have to be resolved. For example, > where/when is the empty range resulting from an intersection operation?
Hmm, I think I would assume the intersection of two non-overlapping ranges to be isempty()=TRUE, and for lower() and upper() to continue to return NULL. But I think a zero-length range created with actual bounds should return the lower() and upper() values during creation, instead of NULL. I tried to find some other programming environments with range types. The first one I found was Ada. The below example is similar to int4range(7,6,'[]') which is invalid in PostgreSQL: with Ada.Text_IO; use Ada.Text_IO; procedure Hello is type Foo is range 7 .. 6; begin Put_Line ( Foo'Image(Foo'First) ); Put_Line ( Foo'Image(Foo'Last) ); end Hello; $ ./gnatmake hello.adb $ ./hello 7 6 I Ada, the 'Range of the Empty_String is 1 .. 0 https://en.wikibooks.org/wiki/Ada_Programming/Types/array#Array_Attributes I think there is a case for allowing access to the the lower/upper vals instead of returning NULL, since we can do so without changing what isempty() would return for the same values,. /Joel
