We now check on intersections of given path and existing to not only check types and contained property names but also properties full path which must not be present in already traversed paths.
Additionally we’ll now catch any CyclicPropertyReferenceExceptions on the root level to prevent cycle detection interfering with application startup.
Original pull request: #197.
If a DBObject was handed into MongoTemplate.remove(…) we previously failed to look up the id value to create a by-id-query. This commit adds explicit handling of DBObjects by looking up their _id field to obtain the id value.
We added an additional check on the fields used as key, so that TTL is ignored for CompoundIndex with more than one field (which effectively renders it useless on @CompoundIndex at all).
Prior to this change potentially invalid index structures would have been created for e.g. @CompoundIndex(def = "{'foo': 1, 'bar': 1}", expireAfterSeconds=10) leading to MongoDB not being able to clean up the indexes (logs: "ERROR: key for ttl index can only have 1 field")
This fix is related to https://jira.mongodb.org/browse/SERVER-10075.
Original pull request: #196.
PartTreeMongoQuery now explicitly check the presence of a manually defined field spec on the query method and creates a new Query if so.
Original pull request: #188.
We modified the regular expression used for cycle detection to match on the exact property name within the inspected path using word boundaries. This fix prevents sub sequences of an existing property (like ‘sub’ would have matched ‘substr’) from being matched.
Along the way we fixed the (false) assertion in one of the tests, as we create the +1 cycle reference index before actually breaking the operation.
Previously MongoTemplate.mapReduce(...) didn't translate nested objects, e.g. GeoCommand, within the given query. That could lead to exceptions during query serialization. We now pass the query and sort object of the given Query through the QueryMapper to avoid such problems.
Original pull request: #184.
We now guard cyclic non transient, non DBRef property references while inspecting domain types for potentially index structures. To do so we check on the properties path and owning type to determine potential cycles. If found we log a warn message pointing to path, entity and property potentially causing problems and skip processing for this path.
Original pull request: #180.
Index creation failed for @Indexed and @CompoundIndex as the resolved dotPath was not used for creation. We now not only resolve the dotPath but also use it within the key for index definition. In case of a nested compound index the key definition is enhanced by the provided path.
When leaving the key definition empty for nested compound index we'll create an index for the whole nested document. Trying to create a compound index on root level not providing key information leads to InvalidDataApiUsageException.
Original pull request: #179.
Only reject attribute setup if abbreviation is activated and a custom strategy is configured. Additional test cases for the rejection case and a custom, over-configuration (explicitly setting abbreviation to false, which is the default anyway).
Related pull request: #177.
The default for boolean attributes leaks into the evaluation of XML namespace attributes which causes us being unable to detect whether two attributes have been set in a conflicting way.
Fix the documentation on the field-naming-strategy-ref attribute.
Original pull request: #183.
Related pull request: #177.
Related ticket: DATAMONGO-925.
Additional test added to check if the issue has truly been resolved by DATAMONGO-888.
Original pull request: #181.
Related pull Request: #162.
Related ticket: DATAMONGO-888.
Previously we didn't support referring to aliased fields defined in former stages of an aggregation pipeline. We now also consider field aliases during field reference lookup.
Original pull request: #176.
We now support letting MongoDB generate index names by introducing the attribute "useGeneratedName" to the @Indexed, @GeoSpatialIndex, @CompoundIndex annotations.
With the overhaul of the index creation done in DATAMONGO-899 the CompoundIndex annotation is not longer just looked up at the concrete type but rather all its interfaces and super classes. So we just added an additional test to verify this behaviour.
Removed the introduction of the IndexDefinition being collection aware again. The collection an index is created in is now held in the IndexDefinitionHolder. This is mostly due to the fact that the IndexDefinition implementations can be used with MongoTemplate and the index opoerations take a collection alongside the index definition.
Made the IndexResolver API package protected so that we can further change it going forward. We should think about deprecating the collectionName attributes on index annotations as it doesn't make too much sense to manually configure the collection name for the indexes as the collection is predefined through the domain type setting here. This would allow us to remove the entire collection handling code inside the IndexResolver implementation.
Turned IndexDefinitionHolder into a value object.
Original pull request: #168.
Index creation did not consider the properties path when creating the index. This lead to broken index creation when nesting entities that might require index structures.
Off now index creation traverses the entities property path for all top level entities (namely those holding the @Document annotation) and creates the index using the full property path.
This is a breaking change as having an entity to carry the @Document annotation has not been required by now.
Original Pull Request: #168
In case we detect Query within a value used for an Update value we map the query itself to build the expression to use. This allows to form query statements for e.g. $pull using the same API as for the query itself.
Update update = new Update().pull("list", query(where("value").in("foo", "bar")));
Original Pull Request: #172.
We now check for the presence of DefaultParameterNameDiscoverer in order to determine if we are running with a Spring version later than 4.0 since this avoids potential NullPointerExceptions in cases where the package version information is not available e.g. in cases where the application was bundled into an uberjar e.g. via the maven-shade-plugin.
Original pull request: #173.
We now use the captured DBRef of a given LazyLoadingProxy in MappingMongoConverter.toDBRef(..) in order to avoid a new DBRef creation that would fail for the proxy.
Original pull request: #174.
We now properly resolve lazy-loading proxies for @DBRef's when an overridden equals or hash code method is called with Spring 4. We fall back to our old Objenesis proxy generation in order to circumvent the default handling for overridden hashcCode() and equals(…) methods in CglibAopProxies generated by Spring 4.
If we detect that we run with Spring 4 we use the repacked Objenesis that is included in Spring 4. Previously the generated proxy used some generic hashCode() or equals(…) logic that did not trigger a proper lazy loading in such cases.
Original pull request: #171.
We now consider custom mongo conversions in all stages of an aggregation pipeline. Previously we did this only for the first stage and returned object basically unmapped in later stages. We now pass the root AggregationOperationContext on to nested ExposedFieldsAggregationOperationContexts so that those can delegate any mongo Mapping to the root context.
Original pull request: #170.
We now actively pass on property type information to MetadataBackedField to ensure type hints get picked up correctly when converting a value to the according DBObject.
This has to be done as the fix for DATAMONGO-812 enforced proper writing of _class information for Updates, which caused trouble when querying documents by nested (complex) properties using an 'in' clause.
Original pull request: #169.
Mapping information is potentially required by multiple instances and thus must not be registered as nested bean. We now actively check for such an invalid scenario and explicitly reject it.
Original pull request: #165.
If an association property points to an interface not containing the id property QueryMapper threw a NullPointerException in isAssociationConversionNecessary(…) as the lookup of the id property fails.
We now check for the presence of an id property on the target type and check for assignability to indicated the need for conversion (usually in case when developers use raw ids in their update clauses, not the actual target instance.
Original pull request: #164.
We now allow referring to nested field expressions if the root segment of the nested field expression was exposed in earlier stages of the aggregation pipeline.
Original pull request: #167.
Use the camel case split logic from Spring Data Commons (introduced for DATACMNS-486) in a common CamelCaseSplittingFieldNamingStrategy super class.
MappingMongoConverterParser now also rejects the configuration if both abbreviate-field-names and field-naming-strategy-ref are configured.
