Post History

 > p.s. I'm not sure whether the criticism applies only to generic repositories or not.

This applies to generic repositories as most of their operations (e.g. get entity by id, update the entity, delete entity) are already being done by the DbSet<T>).

 > consuming the ORM directly from the Application layer (..) the first doesn't give me the flexibility of implementing my business logic without polluting it with query logic, particularly considering that more than one database will most likely be required in this application.

My approach is hybrid in this case. If your application is mainly using a relational database for the operational parts (CRUD) and some others for querying external information, notifying other APIs etc. I would take the following approach:

- **consuming the ORM directly from the Application layer** for the use cases that involve the primary relational database. That is, your application layer might inject the `ICustomDbContext` which exposes the minimum required to get/update the data:

   DbSet<TEntity> Set<TEntity>() where TEntity : class;
   IQueryable<TEnt> ReadSet<TEnt>() where TEnt : class;
   Task<int> SaveChangesAsync(CancellationToken cancellationToken);

This indeed couples the application to the ORM and you have some queries in the application. However, queries are actually part of the business logic and abstractions tend to complicate things (i.e. poor performance in some cases).

It is very unlikely that you will need to change the ORM in the future, but it still works if you change the relational database.

An exception for this rule is the cases when you need to write a direct query due to performance reasons. This should be placed in another class (repository) because changing the provider might need revisiting the query.

- **interacting with other databases** - if your application needs to talk to other databases, the logic for this can be included in a repository class for each case. This is required because other databases come with their own specifics and you do not want to mix those specifics with the application business logic

- **consuming APIs or other services** - this can be put in a separate service class. In a clean architecture, these services are implemented in the Infrastructure layer and injected (dependency inversion) in the application layer. 

Note: for extensive information about why generic repositories are not fashionable anymore and how to replace them with "generic services", check out [this article](https://www.thereformedprogrammer.net/is-the-repository-pattern-useful-with-entity-framework-core/).