Question about blockdedup and call to count()

[yeikel]

In the method blockdedup ,

We have the following code :

 val distinctPartitions = blocked.select(partColumn).distinct().count()
    val hashPart = new HashPartitioner(distinctPartitions.toInt)

val blockedRDD = blocked.rdd
      .keyBy(x => x.getString(x.fieldIndex(partColumn)))
      .partitionBy(hashPart)

If I understand correctly , calling .count() will evaluate the dataframe. Wouldn't it be beneficial to persist the dataframe , do the count and then do the keyBy ?

Also , why can't we just pass partitionBy with the field name?