Entity Framework, Code First and Full Text Search

I realize that a lot of questions have been asked relating to full text search and Entity Framework, but I hope this question is a bit different.

I am using Entity Framework, Code First and need to do a full text search. When I need to perform the full text search, I will typically have other criteria/restrictions as well - like skip the first 500 rows, or filter on another column, etc.

I see that this has been handled using table valued functions - see http://sqlblogcasts.com/blogs/simons/archive/2008/12/18/LINQ-to-SQL---Enabling-Fulltext-searching.aspx. And this seems like the right idea.

Unfortunately, table valued functions are not supported until Entity Framework 5.0 (and even then, I believe, they are not supported for Code First).

My real question is what are the suggestions for the best way to handle this, both for Entity Framework 4.3 and Entity Framework 5.0. But to be specific:

1. Other than dynamic SQL (via System.Data.Entity.DbSet.SqlQuery, for example), are there any options available for Entity Framework 4.3?

2. If I upgrade to Entity Framework 5.0, is there a way I can use table valued functions with code first?

Thanks, Eric

Solution 1:

Using interceptors introduced in EF6, you could mark the full text search in linq and then replace it in dbcommand as described in http://www.entityframework.info/Home/FullTextSearch:

public class FtsInterceptor : IDbCommandInterceptor
{
    private const string FullTextPrefix = "-FTSPREFIX-";

    public static string Fts(string search)
    {
        return string.Format("({0}{1})", FullTextPrefix, search);
    }

    public void NonQueryExecuting(DbCommand command, DbCommandInterceptionContext<int> interceptionContext)
    {
    }

    public void NonQueryExecuted(DbCommand command, DbCommandInterceptionContext<int> interceptionContext)
    {
    }

    public void ReaderExecuting(DbCommand command, DbCommandInterceptionContext<DbDataReader> interceptionContext)
    {
        RewriteFullTextQuery(command);
    }

    public void ReaderExecuted(DbCommand command, DbCommandInterceptionContext<DbDataReader> interceptionContext)
    {
    }

    public void ScalarExecuting(DbCommand command, DbCommandInterceptionContext<object> interceptionContext)
    {
        RewriteFullTextQuery(command);
    }

    public void ScalarExecuted(DbCommand command, DbCommandInterceptionContext<object> interceptionContext)
    {
    }

    public static void RewriteFullTextQuery(DbCommand cmd)
    {
        string text = cmd.CommandText;
        for (int i = 0; i < cmd.Parameters.Count; i++)
        {
            DbParameter parameter = cmd.Parameters[i];
            if (parameter.DbType.In(DbType.String, DbType.AnsiString, DbType.StringFixedLength, DbType.AnsiStringFixedLength))
            {
                if (parameter.Value == DBNull.Value)
                    continue;
                var value = (string)parameter.Value;
                if (value.IndexOf(FullTextPrefix) >= 0)
                {
                    parameter.Size = 4096;
                    parameter.DbType = DbType.AnsiStringFixedLength;
                    value = value.Replace(FullTextPrefix, ""); // remove prefix we added n linq query
                    value = value.Substring(1, value.Length - 2);
                    // remove %% escaping by linq translator from string.Contains to sql LIKE
                    parameter.Value = value;
                    cmd.CommandText = Regex.Replace(text,
                        string.Format(
                            @"\[(\w*)\].\[(\w*)\]\s*LIKE\s*@{0}\s?(?:ESCAPE N?'~')",
                            parameter.ParameterName),
                        string.Format(@"contains([$1].[$2], @{0})",
                                    parameter.ParameterName));
                    if (text == cmd.CommandText)
                        throw new Exception("FTS was not replaced on: " + text);
                    text = cmd.CommandText;
                }
            }
        }
    }

}
static class LanguageExtensions
{
    public static bool In<T>(this T source, params T[] list)
    {
        return (list as IList<T>).Contains(source);
    }
}

For example, if you have class Note with FTS-indexed field NoteText:

public class Note
{
    public int NoteId { get; set; }
    public string NoteText { get; set; }
}

and EF map for it

public class NoteMap : EntityTypeConfiguration<Note>
{
    public NoteMap()
    {
        // Primary Key
        HasKey(t => t.NoteId);
    }
}

and context for it:

public class MyContext : DbContext
{
    static MyContext()
    {
        DbInterception.Add(new FtsInterceptor());
    }

    public MyContext(string nameOrConnectionString) : base(nameOrConnectionString)
    {
    }

    public DbSet<Note> Notes { get; set; }

    protected override void OnModelCreating(DbModelBuilder modelBuilder)
    {
        modelBuilder.Configurations.Add(new NoteMap());
    }
}

you can have quite simple syntax to FTS query:

class Program
{
    static void Main(string[] args)
    {
        var s = FtsInterceptor.Fts("john");

        using (var db = new MyContext("CONNSTRING"))
        {
            var q = db.Notes.Where(n => n.NoteText.Contains(s));
            var result = q.Take(10).ToList();
        }
    }
}

That will generate SQL like

exec sp_executesql N'SELECT TOP (10) 
[Extent1].[NoteId] AS [NoteId], 
[Extent1].[NoteText] AS [NoteText]
FROM [NS].[NOTES] AS [Extent1]
WHERE contains([Extent1].[NoteText], @p__linq__0)',N'@p__linq__0 char(4096)',@p__linq__0='(john)   

Please notice that you should use local variable and cannot move FTS wrapper inside expression like

var q = db.Notes.Where(n => n.NoteText.Contains(FtsInterceptor.Fts("john")));

Solution 2:

I have found that the easiest way to implement this is to setup and configure full-text-search in SQL Server and then use a stored procedure. Pass your arguments to SQL, allow the DB to do its job and return either a complex object or map the results to an entity. You don't necessarily have to have dynamic SQL, but it may be optimal. For example, if you need paging, you could pass in PageNumber and PageSize on every request without the need for dynamic SQL. However, if the number of arguments fluctuates per query, it will be the optimal solution.