Object to object mapping verification - is this extension method useful?

In our .NET tests we use NSubstitute & ExpectedObjects.

Testing object expectations involves hand crafting large anonymous objects and when new properties are added we need to go back to these anonymous objects & update them.

Attempting below to get a fluent Object to DTO builder - which fails when a property is missed.

Here is the implementation:

//CustomerCreatedEvent has all below properties 
        var exp1 = @event.ToDto(
            x=> x.AggregateId.As("CustomerId"), 
            x => x.Email, // comment this out and test FAILS -> CustomerDetail.Email required
            x => x.FirstName, 
            x => x.Surname);

        var exp2 = new {
                CustomerId= @event.AggregateId,
                @event.Email,      // comment this out and test still passes
                @event.FirstName,
                @event.Surname};

        exp1.ToExpectedObject().ShouldMatch(actual);
        exp2.ToExpectedObject().ShouldMatch(actual);

I have 2 questions:

• Is my code just adding 'noise'?

-
Is the implementation code below sound?

public static TResult ToDto(this TSource obj, params Expression>[] items) where TSource : class
{
var eo = new ExpandoObject();
var props = eo as IDictionary;

```
foreach (var item in items)
{
var member = item.Body as MemberExpression;
var unary = item.Body as UnaryExpression;
var body = member ?? (unary != null ? unary.Operand as MemberExpression : null);

if (member != null && body.Member is PropertyInfo)
{
var property = body.Member as PropertyInfo;
props[property.Name] = obj.GetType()
.GetProperty(property.Name)
.GetValue(obj, null);
}
else
{
var property = unary.Operand as MemberExpression;
if (property != null)
{
props[property.Member.Name] = obj.GetType()
.GetProperty(property.Member.Name)

Minor nitpick first:

public static T2 ToDto(this T1 obj, params Expression>[] items) where T1 : class
{

I think putting generic type constraints on a new line makes them easier to spot and adds to readability (especially with long signatures) - consider:

public static T2 ToDto(this T1 obj, params Expression>[] items)
    where T1 : class
{

I really don't like T1 and T2 for generic type parameter names. T is ok when it's the only parameter, but I'd much, much rather like to see TSource and TResult - it makes it much clearer which one serves which purpose, at a glance.

This also instantly makes it easy to find source a much more meaningful name than obj. I'm all for readable identifiers, so I would like to see props renamed to properties, and eo to expandoObject, or perhaps just value - I will be using these identifiers for the remainder of this post.

I like that you're using var - I don't understand why you're not using it here, since JsonConvert.SerializeObject returns a String:

string json = JsonConvert.SerializeObject(value);  // need json.net

That said, the comment isn't necessary. If you referenced Json.net from a NuGet package, the IDE will tell you when/if a package is missing. If you referenced Json.net from an assembly, the project will fail to build and under "references", Newtonsoft.Json will appear problematic - the comment really doesn't help with anything, it's just noise.

In this block:

if (member != null && body.Member is PropertyInfo)
{
    var property = body.Member as PropertyInfo;
    if (property != null)
        properties[property.Name] = source.GetType().GetProperty(property.Name).GetValue(source, null);
}

The condition if (property != null) will always be true, because body.Member is PropertyInfo is necessarily true in that branch. Hence it can be simplified to this:

if (member != null && body.Member is PropertyInfo)
{
    var property = body.Member as PropertyInfo;
    properties[property.Name] = source.GetType().GetProperty(property.Name).GetValue(source, null);
}

The else branch has a redundant cast - you defined unary as item.Body as UnaryExpression; hence, unary already has the reference you're getting in this superfluous declaration:

var ubody = (UnaryExpression)item.Body;

Just use unary, you don't need ubody at all.

I haven't tested it, but the type parameter in DeserializeAnonymousType seems superfluous as well - Activator.CreateInstance() will create a TResult object.. that you can readily return - so instead of this:

var json = JsonConvert.SerializeObject(value);
var anon = JsonConvert.DeserializeAnonymousType(json, Activator.CreateInstance());
return ((JObject)anon).ToObject();

You'd have that:

var json = JsonConvert.SerializeObject(value);
return JsonConvert.DeserializeAnonymousType(json, Activator.CreateInstance());

Notice I'm letting the compiler infer the generic type parameter here.

Lastly, the way chained reflection methods are laid out isn't consistent between the if and the else branch - one has it like this:

properties[property.Name] = source.GetType().GetProperty(property.Name).GetValue(source, null);

The other has it like that:

properties[property.Member.Name] = source.GetType()
    .GetProperty(property.Member.Name)
    .GetValue(source, null);

I like how putting each instruction on a new line makes it easier to see what's going on. Some may argue that it's even easier when the dots line up:

properties[property.Member.Name] = source.GetType()
                                         .GetProperty(property.Member.Name)
                                         .GetValue(source, null);

StackExchange Code Review Q#69359, answer score: 6