One repository per table or one per functional section?

Question

I am using ASP.NET MVC 2 and C# with Entity Framework 4.0 to code against a normalised SQL Server database. A part of my database structure contains a table of entries with foreign keys relating to sub-tables containing drivers, cars, engines, chassis etc.

I am following the Nerd Dinner tutorial which sets up a repository for dinners which is fair enough. Do I do one for drivers, one for engines, one for cars and so on or do I do one big one for entries?

Which is the best practise for this type of work? I am still new to this method of coding.

Answer 1

I guess there's really no single "best practice" for this - it depends on your coding style and the requirements for your app. You definitely could create a repository for each entity type in your system - that'll work just fine.

In your case here, I would probably at least consider having a repository for drivers, and possibly a second one for cars, engines, chassis (since those are kinda in the same area of expertise - they're interconnected, they "belong" together).

But: of course, if that single repository for cars, engine and chassis gets too bloated, you might consider breaking it up into three separate repositories.

I would try to find a balance between the number of repositories - try to group together what logically belongs together - and the number of methods on those repositories. Five, ten methods is okay - if you're talking 20, 30 or 50 methods, your repository might just be too big and unwieldy.

This is definitely an architectural decision, and as such, they're not really a lot of hard facts to guide you - it's more of a "gut feeling" and experience kind of thing. If you don't have that necessary experience yet - go with one approach, use it, and when you're done - look at it again with a critical eye and see: what worked about it? What about it didn't work? and then in your next project, try some other approach and question its validity, too, at the end of the project. Live and learn!

Answer 2

I am using generic (parametrized) repository for every entity. This repository contains basic CRUD function. Above that I have service for every functionality. Every service instantiates needed repositories. ObjectContext is common for all of them and there is one for every request. This is my repository interface:

public interface IRepository<T>
{
    //Retrieves list of items in table
    IQueryable<T> List();
    IQueryable<T> List(params string[] includes);

    //Creates from detached item
    void Create(T item);
    void Delete(int id);
    T Get(int id);
    T Get(int id, params string[] includes);
    void SaveChanges();
}

I have also generic implementation, that is long, but easy to implement. You won't have to create repository class for every type, just instantiate Repository<Type>.

Answer 3

Personally speaking, I find it best to create a separate repository for each table.

I then create a services layer where in one class I would run all the commands for a specific action (ex, change an already existing driver's car to a newly added car). The services layer is where I'll call multiple repository if an action I need done contains multiple objects that are interrelated.

I try my best to keep my repositories as "dumb" as possible and put all the "smart" stuff in the services layer. The extra services layer also helps me avoid bloating my controllers.

Answer 4

I usually take the diagramm of the database and draw circles around what I consisder a bigger unit or system.

That gets me something like "ProductSystem" , "OrderSystem" , "UserSystem", "PaymentSystem" in a shop.

If systems get too big I devide them.

If systems are too little I dont care: Everything will grow anyway.

If something belongs in some way to 2 systems I choose 1 and dont change it anymore, even if the next day the decission seams wrong: I know that the day will come when I want to change it back again.