Joe Conway <mail@joeconway.com> writes:
> The only remaining problem is that this causes opr_sanity to fail based
> on this query: ...
> This fails because I implemented text_substr() and bytea_substr() to
> take either 2 or 3 args. This was necessary for SQL92 spec compliance.
Rather than loosening the opr_sanity test, I'd suggest setting this
up as two separate builtin functions. They can call a common
implementation routine if you like. But a runtime test on the number
of arguments doesn't offer any attractive improvement.
> I'm planning to take on the replace function next.
Isn't Gavin on that already?
regards, tom lane