Efcore: Support Polymorphic Relationships

[bjorn-ali-goransson]

For example: I have two Entity Types Page and Post, and both implement the interface ITeaser. Let's say that ITeaser has a Title.

Now, I want my Post to have a new property RecommendedNextRead that is of type ITeaser, and can thus be of both type Post and Page.

What would be the best way to implement this? Is it even feasible supported?

[rowanmiller]

It's not currently supported, unless Post/Tag have a common base type (but that means they would need to be stored in the same table with TPH... presumably not what you want 😄). We have discussed supporting this, but it's not something we're working on at the moment or in the immediate future. Moving to the backlog.

[bdebaere]

Copied from my StackOverflow post.

Consider the following table design...

      Id UNIQUEIDENTIFIER

    OrderLine
      Id UNIQUEIDENTIFIER
      OrderId UNIQUEIDENTIFIER

    Connection
      Id UNIQUEIDENTIFIER
      FromEntityId UNIQUEIDENTIFIER
      ToEntityId UNIQUEIDENTIFIER

It allows linking any table to any other table via the Connection table. The alternative being to create a ConnectionFromTxToTy table for every single combination. Which in our schema means creating 70! (! for factorial) tables just for this functionality. On top of this that would mean adding 70 navigation properties to every class. This is far from easy to implement or consume.

It is possible to add a navigation property Connections to my models. Which does allow me to view all Connections linked to a single entity (provided I double them up).

    order
        .HasMany(entity => entity.Connections)
        .WithOne()
        .HasForeignKey(entity => entity.FromEntityId)

However, I'd like to go one step further and actually be able to include the linked entity, like any other entity. The downside of course is that this is not supported by EntityFrameworkCore. I cannot create a class like below and expect it to work out of the box.

    public class Connection
    {
        public Guid Id { get; set; }

        public Guid FromEntityId { get; set; }

        public Guid ToEntityId { get; set; }

        public virtual object FromEntity { get; set; }

        public virtual object ToEntity { get; set; }
    }

And as soon I would try to include on Orders.Include(order => order.Connections).ThenInclude(connection => connection.ToEntity) I would lose all typed functionality. But this I could live with if something like the following would function: Orders.Include(order => order.Connections).ThenInclude(connection => connection.ToEntity).ThenInclude(orderLine => ((OrderLine)orderLine).Order).

Currently I've included the linked entity as JSON, but that obviously means I lose all include functionality after that. But worse than that, any ValueConverters I assigned to that type aren't being executed.

How can I tackle this challenge? I'm not sure what's the best path: the virtual IDictionary<string, object> route or the virtual object route or something else. I feel like I might potentially get there with a custom QueryModelGenerator and a custom SqlServerQuerySqlGeneratorFactory, although I don't know if that alone would be enough. But EntityFrameworkCore stops me already at the navigation property with type object.

[bjorn-ali-goransson]

Hey,

Sounds like a relational database isn't the right tool for the job, honestly ...

[bricelam]

See also #757

[atrauzzi]

I'd definitely like to see polymorphic relations supported and if the number of times it's asked for even in this project's bug tracker is any indicator... It's quite high priority.

[AndriySvyryd]

@atrauzzi This issue currently only has 9 👍 votes, so it's actually rather low on the feature request list

[atrauzzi]

@AndriySvyryd - I've been scanning this repository as well as stack overflow (as well as others) and the combined organic demand is a much different story.

I get you're trying to be objective, but that methodology leans into a disingenuous extreme. It takes empathy to truly understand what users want, not the expectation that people find the right place to put a "thumbs up" as well as to self-select in such a specific way.

Hopefully feature priority for this project can be more visionary, not reactionary...

[AndriySvyryd]

Yes, we take more thing into account for the release planning process, this is just the most accessible metric.

[atrauzzi]

Right, but I don't want it to become a bad-faith distraction to my point which is that I think demand for this in particular is more than _the most accessible metric_.

My apologies, but also in my defence, I do have a knack for these things.

[AndriySvyryd]

We are gathering feedback to make sure that we don't miss any gaps not currently reflected in the issues. We'll adjust our plans if necessary according to the data.

[atrauzzi]

Hmm. That's really no different to waiting for +1s. It substitutes analysis & intuition with statistics and intellect. This goes back to my previous point about being visionary and not reactionary.

I get it, the next thing to say is that _if people don't speak up, they can't be helped_... But then what is the long term impact of that and what blind spots does it create? Current sentiment towards EF core is hurting .net in general and quite a few long-term asks have been sitting in the freezer since 2014.

The impression given here is that _the data_ is bad due to a bad methodology and it's pulling focus away from the things people actually want.

[roji]

@atrauzzi we aren't just looking at data - analysis and intuition are also telling us that certain other issues are probably more important to our users (the thing about intuition is that it's a bit hard to substantiate :rofl:). I also don't believe in purely data- and metric-driven decisions, but I don't think this project does that in general. It's easy to say that a certain feature is important or has been in the backlog for a long time. What's hard is to gauge its importance compared to everything else that users want.