Refactored JavaConfig base class to reflect the XML namespace a bit more closely (esp. regarding configuration of custom converters and thus registering "simple" types). Prevent duplicate invocation of getUserCredentials(). Added JavaDoc to explain which configuration methods use which other ones to ease detailed configuration.
Updated Sonargraph architecture description file to include GridFS subsystem. Moved QueryMapper into convert subsystem. Polished MongoRepository interface. Let SimpleMongoRepository implement MongoRepository.
Fixed a ClassCastException that occurred because we didn't consider index information of geo indexes (they return "2d" as direction). Introduced new IndexField abstraction that supersedes the fieldSpec Map in IndexInfo.
The Spring container does not check nested generic types of the type parameter of ApplicationEvent<T>. As T is parameterized in our case as well (PersistentEntity<…, …>) we can code an event listener against that fully parameterized type but might run into ClassCastExceptions as we might get other implementations handed into the method at runtime. We now do an instanceof check to safely invoke checkForIndexes(…) only in case we get the correct event type.
When using a PropertyPlaceHolderConfigurer to set WriteConcerns on a MongoFactoryBean just like this:
<bean class="….mongodb.core.MongoFactoryBean">
<property name="writeConcern" value="${mongodb.writeConcern}"/>
</bean>
we might create invalid WriteConcerns as the BeanFactory will use the WriteConcern's constructor taking a String to create the instance by default. To make Spring use the valueOf(…) method in advance one needs to register either our already existing WriteConcernPropertyEditor or the newly introduced StringToWriteConcernConverter in Springs ConversionService.
MappingMongoConverter now rejects objects that would result in field keys containing a dot as we cannot reliably escape and unescape them without potentially wrecking correct keys on reading. However I added a property mapKeyReplacement that can be set to e.g. ~ to have all dots in map keys replaced with ~. This will of course cause ~ to be transformed into dots when reading. If further customization is necessary override potentiallyEscapeMapKey(…) and potentiallyUnescapeMapKey(…).
The reference in MongoRepositoryFactoryBean to MongoTemplate was unnecessary on the one hand and could cause problems in case the MongoTemplate is proxied as it can't be wired into the factory anymore then.
Polished pom.xml files. Upgraded to Log4J 1.2.16 as it excludes unnecessary dependencies. Consolidated test dependencies into parent pom. Remove dependency-management sections. Exclude commons-logging in favor of SLF4J.
If you have a property of type object and it contains a collection we didn't property read it back in as creating the collection instance failed due to an invalid call to CollectionFactory. We now default the parameter handed to that call to List in case the property type is not a Collection at all.
We now provide a GridFsTemplate for storing, retrieving and deleting documents from GridFS. Introduced GridFsResource implementation to integrate with Spring's Resource abstraction. Beyond that GridFsOperations extends ResourcePatternResolver and thus allows to retrieve GridFsResources via Ant-style filename patterns.
Huge refactoring of the way MappingMongoConverter instantiates entities. The constructor arguments now have to mirror a property exactly in terms of name. Thus we can pick up mapping information from the property to lookup the correct value from the source document. The @Value annotation can be used to either inject completely arbitrary values into the instance (e.g. by referring to a Spring bean) or simply define an expression against DBObject's fields:
class Sample {
String foo;
String bar;
Sample(String foo, @Value("#root._bar") String bar) {
this.foo = foo;
this.bar = bar;
}
}
trying to create an instance of this class from
{ "foo" : "FOO" } -> new Sample("FOO", null)
{ "_bar" : "BAR" } -> new Sample(null, "BAR").
QueryMapper assumed finding a BasicBSONList for $(n)or operators. This is generally true if the DBObject was created through our Query abstraction. If you use the MongoDB driver QueryBuilder this will fail. We're now only insisting on an Iterable which fixes the issue.
Updated changelog, changed reference documentation inclusion links to point to SD Commons 1.0.0.RELEASE documentation. Updated dependency information in reference documentation.
Remonde removal of generic ObjectToStringConverter as it doesn't break our converter lookup due to the changed algorithm how to involve Spring Converters in the conversion process.
The execution of MongoTemplate.geoNear(…) potentially caused NullPointerExceptions in case the actual query does not return any results. The wrapping return object returns null for the result list and general statistics which we didn't shield against.
Documented classpath scanning feature of custom converters (DATAMONGO-301). Documented converter disambiguation using @ReadingConverter, @WritingConverter (DATACMNS-113, DATAMONGO-342). Fixed some code formatting on the way.
CustomConversions now evaluates @ReadingConverter / @WritingConverter when adding Converter implementations. See DATACMNS-113 and the appropriate commit for details. Added unit test to verify StringToBigIntegerConverter does not get added as writing converter.
In case a query referencing an entity's id needs massaging before being executed (e.g. Strings that can be ObjectID) the massaging failed in case the query was built using _id already as we did not detect that one being an id reference as we compared it to the entity's id property name. We now also compare against it's field name.
Changed the setter parameters for ServerAddresses to use arrays instead of List. We now register the ServerAddressPropertyEditor to convert a ServerAddress[] and thus don't register a PropertyEditor for List which caused unwanted side effects before.
Oliver Gierke
Mark Pollack
Thomas Risberg