I'm using Dynamic Linq (source code). Sadly there is little documentation about how to use it :-) In a fun boomerang effect, there is an "evolved" version. The boomerang effect is because the code for generating the dynamic class is based on one of my responses :-) The remaining code seems to be very beautiful... And there is a full suit of unit tests with code samples!!! Note that this second library is a superset of the first library, so you can probably apply many examples to the first one! :-)
I'm adding some static methods to translate the result of a a Dynamic Linq query to a IEnumerable<object[]>.... Example code:
using (var ctx = new Model1())
{
var result = ctx.MyTable
.Take(100)
.SimpleSelect(new[] { "ID", "Col1", "Col2" })
.ToObjectArray();
foreach (var row in result)
{
Console.WriteLine(string.Join(", ", row));
}
}
More complex example:
var columnsNames = new[] { "SomeNullableInt32", "SomeNonNullableDateTimeColumn" };
// One for each column!
var formatters = new Func<object, string>[]
{
x => x != null ? x.ToString() : null,
x => ((DateTime)x).ToShortDateString()
};
var result = ctx.MyTable.Take(100).SimpleSelect(columnsNames).ToObjectArray();
foreach (var row in result)
{
var stringRow = new string[row.Length];
for (int i = 0; i < row.Length; i++)
{
stringRow[i] = formatters[i](row[i]);
}
Console.WriteLine(string.Join(", ", stringRow));
}
And the classes... One (SimpleSelect) produces the Dynamic SQL Select, and "anonymizes" the field names. I do this because for each type of return the Dynamic Linq will generate at runtime a class. This class won't be unloaded until the program ends. By anonymizing the columns (I rename them to Item1, Item2, Item3...) I increase the possibility that the same class will be reused. Note that different type of columns will generate different classes! (int Item1, string Item2 will be a different class from int Item1, DateTime Item2), the other (ToObjectArray) returns a IEnumerable<object[]>, something easier to parse.
public static class DynamicLinqTools
{
private static ConcurrentDictionary<Type, Func<object, object[]>> Converters = new ConcurrentDictionary<Type, Func<object, object[]>>();
public static IQueryable SimpleSelect(this IQueryable query, string[] fields)
{
// With a little luck, "anonymizing" the field names we should
// reduce the number of types created!
// new (field1 as Item1, field2 as Item2)
return query.Select(string.Format("new ({0})", string.Join(", ", fields.Select((x, ix) => string.Format("{0} as Item{1}", x, ix + 1)))));
}
public static IEnumerable<object[]> ToObjectArray(this IQueryable query)
{
Func<object, object[]> converter;
Converters.TryGetValue(query.ElementType, out converter);
if (converter == null)
{
var row = Expression.Parameter(typeof(object), "row");
// ElementType row2;
var row2 = Expression.Variable(query.ElementType, "row2");
// (ElementType)row;
var cast = Expression.Convert(row, query.ElementType);
// row2 = (ElementType)row;
var assign = Expression.Assign(row2, cast);
var properties = query.ElementType.GetProperties(BindingFlags.Public | BindingFlags.Instance)
.Where(x => x.CanRead && x.GetIndexParameters().Length == 0)
.ToArray();
// (object)row2.Item1, (object)row2.Item2, ...
var properties2 = Array.ConvertAll(properties, x => Expression.Convert(Expression.Property(row2, x), typeof(object)));
// new object[] { row2.Item1, row2.Item2 }
var array = Expression.NewArrayInit(typeof(object), properties2);
// row2 = (ElementType)row; (return) new object[] { row2.Item1, row2.Item2 }
var body = Expression.Block(typeof(object[]), new[] { row2 }, assign, array);
var exp = Expression.Lambda<Func<object, object[]>>(body, row);
converter = exp.Compile();
Converters.TryAdd(query.ElementType, converter);
}
foreach (var row in query)
{
yield return converter(row);
}
}
}
IEnumerable<object[]>. That would be the perfect solutionDbDataReaderwork perfectly well. EF isn't built for doing these things. With NHibernate it would be quite easy (using the CriteriaAPI and a projection toobject[]).