ibatis-user-java mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stephen Friedrich <stephen.friedr...@fortis-it.eu>
Subject AW: regresssion (?) in beta 9: No automapping if some properties are explicitly mapped
Date Tue, 02 Feb 2010 20:28:26 GMT
Thanks for the quick response. Yes, I was looking for release notes somewhere close to the
download link, but haven't found any.
Setting FULL did the trick.

I had just started to evaluate performance anyway because first results on the integration
server were not so great:
A simple web service method (JAX-RS) that reads a thousand rows, then serializes that to JSON
(no collections involved) takes two seconds.
Now looking at some profiling results in YourKit I see that FastResulSetHandler.applyAutomaticMappings()
alone takes up more than half of the total web service time:

[I tried to send a screenshot here, but the mail was rejected due to suspected spam]
56% FastResultSetHandler.applyAutomaticMappings()
18%    MetaObject.findProperty()
16%       MetaClass.buildProperty()
13%           Reflector.findPropertyName()
10%              String.toUpperCase()
...

I already patched Reflector.findPropertyName() so that the calls to toUpperCase() are avoided,
using an additional map:
  public String findPropertyName(String name) {
      //  return caseInsensitivePropertyMap.get(name.toUpperCase());
      String propName = propertyMap.get(name);
      if (propName == null) {
          propName = caseInsensitivePropertyMap.get(name.toUpperCase());
          if (propName != null) {
              propertyMap.put(name, propName);
          }
      }
      return propName;
  }
That is noticeably better - most of all on the real web server, which is a quite dated Sparc
T2000 with horrible single-thread performance.
However I feel that much better performance gains should be reachable on a higher abstraction
level.

Is there a conceptual reason why automatic mapping should be noticeably slower than explicit
mapping?
That should only make a difference for the first request, shoudn't it?

Stephen

Von: Clinton Begin [mailto:clinton.begin@gmail.com]
Gesendet: Dienstag, 2. Februar 2010 16:47
An: user-java@ibatis.apache.org
Betreff: Re: regresssion (?) in beta 9: No automapping if some properties are explicitly mapped

Sorry, this was a change. See the documentation for <settings> in the user guide. Unfortunately
I poorly advertised the change.

You can now configure the setting autoMappingBehavior="NONE|PARTIAL|FULL"

NONE and FULL are pretty self explanatory.  FULL is how it worked in previous betas.  However,
the new default is PARTIAL, which means that automapping only happens on simple, flat queries.
 If you use associations or collections, PARTIAL mode will not automap fields.

The primary reason was that I found when doing complex joins, it was hard to tell where the
value would end up, especially in cases where the column matched multiple property names in
parent and child objects.  So I chose PARTIAL to be a happy medium between simplicity and
expressiveness.  But if you're confident that your unit testing will deal with any uncertainty,
then feel free to set it to FULL.  Otherwise, I suggest sticking with PARTIAL.  There is also
an obvious performance benefit to setting it to PARTIAL or NONE.

Cheers,
Clinton
On Tue, Feb 2, 2010 at 7:26 AM, Stephen Friedrich <stephen.friedrich@fortis-it.eu<mailto:stephen.friedrich@fortis-it.eu>>
wrote:
I wonder if this was an intentional change:
My mapping config looks like this:

    <resultMap id="partSetMap" type="PartSetDto">
        <id property="id" column="id"/>
        <collection property="parts" ofType=" PartDto">
            <result column="part_id" property="id" javaType="Long"/>
            <result column="serial_no" property="serialNo" javaType="String"/>
            <result column="is_reserved" property="isReserved" javaType="boolean"/>
        </collection>
    </resultMap>

    <select id="selectPage" resultMap="partSetMap">
        ...

PartSetDto has lots of other properties than "id".
In beta 8 those were automatically filled by naming convention.
In beta 9 all other fields are empty.



Mime
View raw message