LINQ to Entities: Why can't I use Split method as condition?

Question

I have the following LINQ query:

var aKeyword = "ACT";
var results = from a in db.Activities
              where a.Keywords.Split(',').Contains(aKeyword) == true
              select a;

Keywords is a comma delimited field.

Everytime I run this query I get the following error:

"LINQ to Entities does not recognize the method 'Boolean Contains[String](System.Collections.Generic.IEnumerable`1[System.String], System.String)' method, and this method cannot be translated into a store expression."

What is the alternative for what I am trying to do?

Answer 1

Your problem is that LINQ-to-Entites has to translate everything you give it into SQL to send to the database.

If that is really what you need to do, you'll have to force LINQ-to-Entities to pull back all the data and LINQ-to-Objects to evaluate the condition.

Ex:

var aKeyword = "ACT";
var results = from a in db.Activities.ToList()
              where a.Keywords.Split(',').Contains(aKeyword) == true
              select a;

Be aware though, that this will pull back all the objects from the Activities table. An alternative may be to let the DB do a bit of an initial filter, and filter down the rest of the way afterwards:

var aKeyword = "ACT";
var results = (from a in db.Activities
              where a.Keywords.Contains(aKeyword)
              select a).ToList().Where(a => a.KeyWords.Split(',').Contains(aKeyword));

That will let LINQ-to-Entities do the filter it understands (string.Contains becomes a like query) that will filter down some of the data, then apply the real filter you want via LINQ-to-Objects once you have the objects back. The ToList() call forces LINQ-to-Entities to run the query and build the objects, allowing LINQ-to-Objects to be the engine that does the second part of the query.

Answer 2

In response to your performance considerations on a big dataset:

You are going to be doing non indexed wildcard string matching on the client, so yes, there will be performance loss.

Is there a reason why you have multiple keywords in one table field? You could normalize that out, to have a ActivityKeywords table where for each Activity you have a number of Keyword records.

Activities(activity_id, ... /* remove keywords field */) ---> ActivityKeywords(activity_id, keyword_id) ---> Keywords(keyword_id, value)

Check out Non-first normal form: http://en.wikipedia.org/wiki/Database_normalization

EDIT: Also even if you were to stick with the one column, there is a way to do everything serverside (if you have a strict syntax: 'keyword1, keyword2, ..., keywordN'):

var aKeyword = "ACT";
var results = (from a in db.Activities
              where a.Keywords.Contains("," + aKeyword) || a.Keywords.Contains(aKeyword + ",")
              select a;

Answer 3

My guess is the way you are calling Split. It should take an array. Maybe there is another Split in Linq it is finding and giving you an unusual error:

This works for Linq to Objects:

 var dataStore = new List<string>
                    {
                        "foo,bar,zoo",
                        "yelp,foo",
                        "fred",
                        ""
                    };
 var results = from a in dataStore
               where a.Split(new[] {','}).Contains("foo")
               select a;

 foreach (var result in results)
 {
     Console.WriteLine("Match: {0}", result);
 }

Outputs the following:

Match: foo,bar,zoo
Match: yelp,foo

Actually, thinking about it, do you need the split at all? a.Contains("foo") may be enough for you (unless you don't want to hit foobar).

Answer 4

You may want to look at this question about L2E and .Contains for a solution that should be more efficient than guessing at a superset before filtering client side.