I wrote:
> Hmm, this seems like a plpgsql deficiency. It feels it can generate a
> separate parameter symbol ($n) for each occurrence of each variable it
> passes into a SQL query. But for this query to be legal, the two
> instances of IntervalMinutes have to be represented by the *same*
> parameter symbol (notice they are not in the regurgitated query).
> It would be more efficient anyway to not generate multiple parameters
> for the same value, so we oughta fix this.
Patch applied to HEAD and 8.1 branches.
regards, tom lane