BeanListHandler#handle(ResultSet) is not optimal

• Key: DBUTILS-37
• Type: Improvement
• Status: Closed
• Resolution: Fixed
• Priority: Minor
• Assignee: Dan Fabulich
• Reporter: Julien Aymé
• Fix Version/s: 1.2

Description

I use the BeanListHandler for huge ResultSets (about 1000000 rows), and I searched through the code to see if I could gain little time.

It appeared to me that the following code - in BeanProcessor.class - was executed too many times:

PropertyDescriptor[] props = this.propertyDescriptors(type);
        ResultSetMetaData rsmd = rs.getMetaData();
        int[] columnToProperty = this.mapColumnsToProperties(rsmd, props);

for the following reason.
Since BeanListHandler extends GenericListHandler, the method #handle(ResultSet) calls #handleRow(ResultSet) for each row in the ResultSet,
which in the case of a BeanListHandler, calls RowProcessor#toBean(ResultSet, Class),
which itself calls BeanProcessor#toBean(ResultSet, Class).

A very simple way to make the BeanListHandler#handle(ResultSet) method faster is to override the GenericListHandler#handle(ResultSet) method by this code:

public Object handle(ResultSet rs) throws SQLException {
        return this.convert.toBeanList(rs, type);
    }

This way, the code I showed would be called only once, as it would not call BeanProcessor#toBean(ResultSet, Class) for each row but BeanProcessor#toBeanList(ResultSet, Class).

Comments

The new OptimalBeanListHandler class (whole file and patch).

With the change I propose, the BeanListHandler#handleRow(ResultSet) method is never called anymore. So, any subclass of BeanListHandler which would override this method would not work properly anymore.

In order to prevent this from happening, I preferred to create a new class (named OptimalBeanListHandler - created from BeanListHandler with svn copy), which would implements ResultSetHandler directly rather than extends GenericListHandler.

[Dan Fabulich]

I don't think it's necessary to create a separate OptimalBeanListHandler... I think nobody should be using the old implementation.

I checked in a change to the "bugfixing" branch in revision 743097 that just replaces the old implementation with this new faster implementation.

[Henri Yandell]

svn ci -m "Merging in Dab Fabulich's work on https://svn.apache.org/repos/asf/commons/sandbox/dbutils/bugfixing from -r741987:747723. Resolving DBUTILS-34 - DBUTILS-37 - DBUTILS-29 - DBUTILS-14 - DBUTILS-31 - DBUTILS-39 - DBUTILS-41 - DBUTILS-44 - DBUTILS-33 - DBUTILS-42 - DBUTILS-40"

Sending pom.xml
Sending src/java/org/apache/commons/dbutils/BasicRowProcessor.java
Sending src/java/org/apache/commons/dbutils/BeanProcessor.java
Sending src/java/org/apache/commons/dbutils/QueryRunner.java
Adding src/java/org/apache/commons/dbutils/handlers/AbstractListHandler.java
Sending src/java/org/apache/commons/dbutils/handlers/ArrayListHandler.java
Sending src/java/org/apache/commons/dbutils/handlers/BeanListHandler.java
Sending src/java/org/apache/commons/dbutils/handlers/ColumnListHandler.java
Deleting src/java/org/apache/commons/dbutils/handlers/GenericListHandler.java
Sending src/java/org/apache/commons/dbutils/handlers/MapListHandler.java
Sending src/test/org/apache/commons/dbutils/BaseTestCase.java
Adding src/test/org/apache/commons/dbutils/QueryRunnerTest.java
Transmitting file data .........
Committed revision 747724.