Moving RevEng provider code down into Relational.Design (where possible) #2741
Conversation
Fix issue 2713 - allow providers to only provide the relational model of the DB and have the inherited base class provide the enhanced model for code-generation. Also fix up some error messages.
| } | ||
| } | ||
|
|
||
| public virtual EntityTypeCodeGeneratorHelper EntityTypeCodeGeneratorHelper(EntityTypeGeneratorModel model) |
There was a problem hiding this comment.
@ajcvickers It feels odd to me that the DbContextCodeGeneratorHelper method is abstract whereas the EntityTypeCodeGeneratorHelper method is virtual. That arises because DbContextCodeGeneratorHelper has the IRelationalMetadataExtensionProvider we discussed. I can make the EntityTypeCodeGeneratorHelper method abstract and simply override it SqlServerMetadataModelProvider to return new EntityTypeCodeGeneratorHelper() but that seems unnecessary?
| if (((EntityType)otherEntityType) | ||
| .FindAnnotation(ReverseEngineeringMetadataModelProvider.AnnotationNameEntityTypeError) == null) | ||
| { | ||
| if (!foreignKey.IsUnique) |
There was a problem hiding this comment.
Why does this check matter?
There was a problem hiding this comment.
If the foreign key is unique you have a 1:1 (or 0..1:0..1 etc) relationship. NavPropInitializers are used to create a new empty HashSet for a NavigationProperty which represents a 1:N relationship.
There was a problem hiding this comment.
👍 Thanks for clarifying.
|
2 thoughts:
|
|
Yep - we should talk. |
|
|
(with the TODO's we discussed in person)
|
Merged with 5c30958 |
Fix issue #2465 - making code generation templates non-provider specific.
Fix issue #2713 - allow providers to only provide the relational model of the DB and have the inherited base class provide the enhanced model for code-generation.
Also fix up some error messages.
Also updated to produce AddGeneratedNever() on a property which the KeyConvention would normally set up as AddGeneratedOnAdd() but where the underlying database column does not have identity set.