UDAFSumWhere.java

package com.facebook.hive.udf;

import org.apache.hadoop.hive.ql.exec.UDAF;
import org.apache.hadoop.hive.ql.exec.UDAFEvaluator;

/**
 * Compute the SUM of row items for which a condition is true;
 * "true" means "not false and also not null."
 *
 * This sounds like it is the same as SUM(col) with a WHERE or GROUP BY, but
 * it allows for multiple columns to be tracked separately within a single
 * aggregation.  This is faster and cleaner than
 * SUM(col * CAST(example = foo AS INT)). Also very useful for computing percentages,
 * for example SUM_WHERE(col_x, thing='blah') / SUM(col_x).
 *
 * If either the conditional or the item to sum is null, we return null
 * because we don't know what the sum is.  This seems extreme, and is
 * slightly inconsistent with SUM, but it can be easily escaped by adding
 * the appropriate IS NULL clause into the conditional.
 */
public final class UDAFSumWhere extends UDAF {

  public static class UDAFCountWhereEvaluator implements UDAFEvaluator {

    Double s = null;
    
    public UDAFCountWhereEvaluator() {
      super();
      init();
    }

    public void init() {
      s = null;
    }

    public boolean iterate(Double item, Boolean ThisBool) {
      if (item != null && ThisBool != null && ThisBool) {
        if (s == null) {
          s = item;
        } else {
          s += item;
        }
      }
      return true;
    }

    public Double terminatePartial() {
      return s;
    }

    public boolean merge(Double s2) {
      if (s2 != null) {
        if (s == null) {
          s = s2;
        } else {
          s += s2;
        }
      }
      return true;
    }

    public Double terminate() {
      return s;
    }
  }
}