使用多个 OR 快速查询并使用 Entity Framework Core 对连接表进行过滤

Question

Here is my model

class Parent 
{ 
   int Id; 
   string Name; 
   List<Child> Childs; 
} // name is unique

class Child 
{ 
    int Id; 
    int ParentId; 
    string Name; 
    Parent Parent; 
} // couple (id, name) is unique

With a given list of couples (parent name, child name) I'd like to get the couples (parent, child) where child can be null if the parent with the given name exists but not the child. The SQL query would look like this:

SELECT * 
FROM parents p
LEFT JOIN childs c ON c.parent_id = p.id
WHERE p.name = 'parent1' AND (c.name IS NULL OR c.name = 'child1')
   OR p.name = 'parent2' AND (c.name IS NULL OR c.name = 'child2')
   OR p.name = 'parent3' AND (c.name IS NULL OR c.name = 'child3')
   OR p.name = 'parent4' AND (c.name IS NULL OR c.name = 'child4');

I've tried expressing this query with Entity Framework Core using PredicateBuilder for the Or and False methods

var predicate = PredicateBuilder.False<Parent>()
    .Or(p => p.Name == "parent1" && p.Childs.Any(c => c.Name == "child1"))
    .Or(p => p.Name == "parent2" && p.Childs.Any(c => c.Name == "child2"))
    .Or(p => p.Name == "parent3" && p.Childs.Any(c => c.Name == "child3"))
    .Or(p => p.Name == "parent4" && p.Childs.Any(c => c.Name == "child4"));

var p = await _db.Parents
    .Include(p => p.Childs)
    .Where(predicate)
    .ToArrayAsync();

This is the closest I could get but this doesn't get the expected the result:

• if the child doesn't exist the parent is not present in the result set
• Parent.Childs contains all children of the parent instead of only the wanted one

Is my query expressible with Entity Framework Core?

Answer 1

As per your comment, the requirement now is: give me all parents, specified by name, and only one specific child per parent, if present. That is: parents having other children will appear in the result, but without children.

That sounds rather trivial, but it isn't. The gotcha is that it requires two filters, one on parents and one on children, in which the child filter is even parent-specific. A SQL query would look like this:

SELECT * 
FROM parents p1
LEFT JOIN 
(
    SELECT ch.*
    FROM children ch
    JOIN parents p2 ON ch.parentid = p2.id
    WHERE (p2.name = 'parent1' AND ch.name = 'child1')
       OR (p2.name = 'parent2' AND ch.name = 'child2')
       OR (p2.name = 'parent3' AND ch.name = 'child3')
       OR (p2.name = 'parent4' AND ch.name = 'child4') -- filter 2
) fc ON fc.parentid = p1.id
WHERE p1.name IN ('parent1','parent2','parent3','parent4') -- filter 1

For an EF LINQ query the parent predicate can be a simple Contains , but you'd want to build the predicate using a predicate builder. Here, for reason following later, I use LINQkit.core .

To be able to build the predicates from one source I use a temporary structure (but I guess you already have something similar):

var filters = new[]
{
    new { ParentName = "parent1", ChildName = "child1" },
    new { ParentName = "parent2", ChildName = "child2" },
    new { ParentName = "parent3", ChildName = "child3" },
    new { ParentName = "parent4", ChildName = "child5" },
};

And prepare the predicates:

using LinqKit;
...
var parentNames = filters.Select(f => f.ParentName).ToList();
var childPredicateStarter = PredicateBuilder.New<Child>();
foreach (var filter in filters)
{
    childPredicateStarter = childPredicateStarter
        .Or(c => c.Parent.Name == filter.ParentName && c.Name == filter.ChildName);
}

Now, ideally, the LINQ query would look like this ( db is a context), working around the lack of filtering in Include :

var p = db.Parents
    .Where(p => parentNames.Contains(p.Name))
    .Select(p => new
    {
        Parent = p,
        Children = p.Children.Where(childPredicateStarter)
    })
    .AsEnumerable()
    .Select(p => p.Parent);

But that doesn't run because p.Children is IEnumerable , so childPredicateStarter implicitly converts to a Func instead of the required Expression<Func>> . See here for an in-depth explanation.

The actual working version is:

// Convert to expression:
Expression<Func<Child, bool>> childPredicate = childPredicateStarter;

var p = db.Parents.AsExpandable() // <-- LINQKit's AsExpandable()
    .Where(p => parentNames.Contains(p.Name))
    .Select(p => new
    {
        Parent = p,
        Children = p.Children.Where(c => childPredicate.Invoke(c))
    })
    .AsEnumerable()
    .Select(p => p.Parent);

The AsExpandable call converts the Invoke back into a proper expression tree that EF can translate into SQL.

Answer 2

Parent.Childs contains all childs of the parent instead of only the wanted one

Filtered include is coming, but not implemented yet. I am a little at a loss on why you think it would actually filter at all, given that your code CLEARLY states to incldue ALL children...

.Include(p => p.Childs)

Means include the children (btw., childs is bad english - the plural is Children). There is no filter there.

Regarding filtered include:

Filtering on Include in EF Core

Quoting the relevant part for here:

"Finally, this feature has been implemented starting with EF Core preview version 5.0.0-preview.3.20181.2 and will be GA in EF Core version 5.0.0"

But even then, you will have to filter (ie ptut a where into the include, not just tell it to get all of them).