Autowire RelationalEntityInformation instance into CustomRepository

[chinshaw]

Apologies I don't know if this is a feature request or if I am completely missing something, however I am stuck trying to come up with a solution to getting the domain entity type during custom repository initialization (RepositoryFragment). It looks like the SimpleR2dbcRepsitory is initialized with RelationalEntityInformation<T,ID> but it seems that the beanfactories used between SimpleR2dbcRepository and the RepositoryFragmentFactory bean have a different contexts.

I went through pretty much all the spring docs and tried quite a few things including trying to use the newish feature for RepositoryMethodContext.getContext() during initialization but was hesitant to try it during function callback because of the note on the performance hit.

I could probably add a workaround to initialize my "tablemetadata" during the first upsert call but this feels dirty to me.

Appreciate any feedback, I may be missing something completely or my implementation maybe covered by some other functionality all together.

Here is a sample of what I am trying to accomplish and let me know if I am completely off base and there is a much cleaner way to do this.

interface SpringCoroutineRepositoryExample : CoroutineCrudRepository<SampleEntity,Xid>

// Spring Custom Repository 
interface UpsertRepository<T : IDomainEntity<Xid>> {
    fun upsert(values: List<T>): Flow<T>
}

// NOTE: Construction fails due to missing RelationalEntityInformation
class UpsertRepositoryImpl<T : IDomainEntity<Xid>>(
    entity: RelationalEntityInformation<T, Any>,
    r2dbcEntityTemplate: R2dbcEntityTemplate,
    auditorAware: ReactiveAuditorAware<String>
) : UpsertRepository<T> {
        
    init {
        // initialize table meta data using 
        val kclass = entity.javaType.kotlin
        // Using kclass initialize my custom table data
        initializeTableMetaData(kclass)
    }
    
    // Custom reflection to introspect columns
    private fun initializeTableMetaData(kClass: KClass<T>) {}

    // For brevity ... use my table meta data and the DatabaseClient to construct an upsert
    // 
    //  INSERT INTO ${tableMetaData.tableName} ($insertColumns)
    //     VALUES :tuples
    //   ON CONFLICT (${tableMetaData.uniqueKey})
    //      DO UPDATE SET
    //      $updateColumns
    //   RETURNING *;
    override fun upsert(values: List<T>): Flow<T> = emptyFlow()
}

[mp911de]

That isn't possible because the lifecycle of RelationalEntityInformation is different than the one of a fragment. A fragment is being registered in the context once and used for all repositories that want to use the fragment. We can provide you with thread-local details through RepositoryMethodContext. The documentation explains how this works.

Metadata access needs to be enabled, either by letting your fragment implement RepositoryMetadataAccess as marker or by setting setExposeMetadata(true) on the actual repository factory.

[chinshaw]

@mp911de Apologies for taking longer to respond than it took you. I spent a few days digging through the code base which is littered with your commits and I wanted to say thank you for the incredible work and attention to detail.

I was dumbfounded by how Fragments were stitched into repositories but your explanation clarified what I was missing. I was stuck with the idea that each repository would own it's own fragment.