Re: Range type adaptation implemented

On Mon, Sep 24, 2012 at 11:45:42AM +0200, Karsten Hilbert wrote:

> > The funniest surprise I've had during the development
> > was an error creating the object TextRange('#', '$') in the database.
> > It should be possible as
> >
> > >>> ord('#'), ord('$')
> > (35, 36)
> >
> > but PG insisted that "range lower bound must be less than or equal to
> > range upper bound". Well, turns out that:
> >
> > postgres=# select ascii('#') < ascii('$'), '#' < '$';
> > ?column? | ?column?
> > ----------+----------
> > t | f
> > (1 row)
>
> This does not seem intuitive but is probably quite logical:
>
> While
>
> ascii('#') < ascii('$')
>
> eventually compares numerical values (35 and 36) the comparison
>
> '#' < '$'
>
> operates on characters the sorting order (<) of which
> depends on the lcoale in use.

And can thus invalidate range bounds when collation rules
are changed per-query !?!

Karsten
--
GPG key ID E4071346 @ gpg-keyserver.de
E167 67FD A291 2BEA 73BD 4537 78B9 A9F9 E407 1346