A common approach is to iterate over the records, and collect and organize the unique values.

{define-proc public {unique-values
                        rs:RecordSet,
                        f:RecordField
                    }:Array
    || field info
    def name = f.name
    def domain = f.domain
    || collect unique values
    def vals = {Set}
    {for r in rs do
        {vals.insert r[name]}}
    || sort according to domain
    def sorted-vals = {Array efficient-size = vals.size}
    {for val in vals do
        {sorted-vals.append val}}
    {sorted-vals.sort comparison-proc =
        {proc {x:any, y:any}:bool
            {return {domain.compare x, y} <= 0}}}
    || result
    {return sorted-vals}    
}

Of course, you could do this in SQL, with 'SELECT DISTINCT', if you're using the Curl Data Kit Project which makes the SQLite database engine available for use in Curl applications.

I don't mean to suggest that is necessary just for collecting unique values, but the more complex the data processing (aggregation, joins, indexes, etc), the more code you can save by using SQL instead.