I just spent a long time trying to work this one out, so I'm posting here in case anyone else makes the same mistake as me.
So, to replicate the situation, I just had a few classes with basic inheritance:
public abstract class Foo
{
public string Name { get; set; }
}
public class Bar : Foo
{
public int SomethingSpecial { get; set; }
}
public class Baz : Foo
{
public string SomethingMundane { get; set; }
}
Now, I want to be able to take a json string, and parse back a concrete implementation of Foo, without knowing beforehand which type the json represents. JSON.Net handles this, by using a $type variable to look up the type of the object:
{
"$type": "MyNamespace.Bar",
"Name": "Resources",
"SomethingSpecial": 42
}
When deserializing the object, you may want to specify the type handling options, and a custom binder, in the JsonSerializerSettings, but that's optional.
The problem I ran into was that when posting the json from a web client, the object was not being deserialized, instead either throwing an exception, or in the case of a POST parameter in ASP.Net, the value was just coming in as null.
The problem here is that Newtonsoft.Json (at least, the version I'm using - 9.0.1) expects the $type element to be the first element of the object, and will ignore it when it is in any other position. So the following json, while technically equivalent to the json in the question, will not work:
{
"Name": "Resources",
"$type": "MyNamespace.Bar",
"SomethingSpecial": 42
}
There is a setting for this, although it may have a performance impact:
new JsonSerializerSettings
{
// $type no longer needs to be first
MetadataPropertyHandling = MetadataPropertyHandling.ReadAhead
}
It's worth noting that if you're using JavaScript's JSON.stringify function, properties are generally written out in the order they were first assigned to the object, so you could also make sure the first thing you do is assign $type.
Related
Given an Activity (created via the designer) that has several OutArgument properties, is it possible to get their strongly-typed value from a property after invoking the workflow?
The code looks like this:
// generated class
public partial class ActivityFoo : System.Activities.Activity....
{
....
public System.Activities.OutArgument<decimal> Bar { ... }
public System.Activities.OutArgument<string> Baz { ... }
}
// my class
var activity = new ActivityFoo();
var result = WorkflowInvoker.Invoke(activity);
decimal d = activity.Bar.Get(?)
string s = activity.Baz.Get(?)
The T Get() method on OutArgument<T> that requires an ActivityContext which I'm not sure how to obtain in code.
I also realize it's possible to get the un-typed values from result["Bar"] and result["Baz"] and cast them, but I'm hoping there's another way.
Updated to make it clear there are multiple Out values, although the question would still apply even if there was only one.
If you look at workflows as code, an Activity is no more than a method that receives input arguments and (potentially) returns output arguments.
It happens that Activities allows one to return multiple output arguments, something that C# methods, for example, don't (actually that's about to change with C# 7 and tuples).
That's why you've an WorkflowInvoker.Invoke() overload which returns a Dictionary<string, object> because the framework obviously doesn't know what\how many\of what type output arguments you have.
Bottom line, the only way for you to do it fully strong-typed is exactly the same way you would be doing on a normal C# method - return one OutArgument of a custom type:
public class ActivityFooOutput
{
public decimal Bar { get; set }
public decimal Baz { get; set; }
}
// generated class
public partial class ActivityFoo : System.Activities.Activity....
{
public System.Activities.OutArgument<ActivityFooOutput> Result { ... }
}
// everything's strongly-typed from here on
var result = WorkflowInvoker.Invoke<ActivityFooOutput>(activity);
decimal d = result.Bar;
string s result.Baz;
Actually, if you don't want to create a custom type for it, you can use said tuples:
// generated
public System.Activities.OutArgument<Tuple<decimal, string>> Result { ... }
// everything's strongly-typed from here on
var result = WorkflowInvoker.Invoke<Tuple<decimal, string>>(activity);
decimal d = result.Item1;
string s result.Item2;
Being the first option obviously more scalable and verbose.
If we have the following controller action in Web API
public async Task<HttpResponseMessage> GetRoutes(
[FromUri] MapExtentQuery extent,
[FromUri] PagingQuery paging)
{
...
}
with
class MapExtentQuery {
public int X { get;set; }
public int Y { get; set; }
}
class PagingQuery {
public int Skip { get; set; }
public int Top { get; set; }
}
and we make a GET request to /routes both parameters (extent and paging) will be null.
If the request contains at least one querystring parameter, for instance
/routes?x=45
then both complex parameters will get initialized, so in the case of the 2nd route
extent.X = 45
extent.Y = 0
paging != null (but Skip and Top will be 0 of course).
Why does the [FromUri] binder work this way? It makes little or no sense.
I would understand if it initialized only the parameter that contains a property that matched at least one of the querystring values.
The problem is, this behaviour requires us to check when parameters are null (which happens only in the case that no querystring parameter was provided) and then initiliaze them ourselves.
Because obviously those complex params might have constructors which would set some property values to default.
It is because when you do not have query string and the parameters have FromUri attribute, the parser for the query string does not run and all values are not binded - you receive null.
If you have query string, the binder runs and instantiates all FromUri parameters and primitive types with their default values and then tries to fill in the properties from the query string values. This leaves you with two instantiated parameters but only one having the populated value because that is all the query string has.
As for why it works this way - most likely performance - this way they will not need to find which parameters they need to instantiate - they instantiate all of them and then find the properties. It is faster and performance is critical in the action and model binding.
I have a WCF webservice, that exposes these classes:
[DataContract]
public class TemplatesFormat
{
List<DynAttribute> _dynsattributes = new List<DynAttribute>();
[DataMember]
public List<DynAttribute> DynsAttributes
{
get { return _dynsattributes; }
set { _dynsattributes = value; }
}
}
[DataContract]
public class DynAttribute
{
string _key = "";
string _val = "";
[DataMember]
public string Key
{
get { return _key; }
set { _key = value; }
}
[DataMember]
public string Value
{
get { return _val; }
set { _val = value; }
}
}
Basically, 2 classes. DynAttribute with 2 string attributes and TemplatesFormat, with an attribute that is a List of DynAttribute class.
So far, so good.
But, when I reference the web service from an ASP.NET web page and try to use the TemplatesFormat, I can't see the List attribute.
I mean, I actually "see" it, but it is not a list (does not contain an "Add()") and I don't know how to use it.
I think I am missing something related with de [DataContrat] and the fact that it is a custom type, since, I don't have the same problem with DynAttribute class (I see the Key and Value attributes because they are strings) but, I can't get it right for the List...
Any idea???
When you add reference to wcf service you need to change Collection Type to Generic List.
Please see my post wcf-proxy-returning-array-instead-of-list-even-though-collection-type-generic for more details and snipp picture.
WCF is meant to support consumption by many other platforms. Because List<DynAttribute> is not a primitive type, it is likely converting it to DynAttribute[].
In your consuming application. Try taking your variable and seeing if you can .ToList() it to turn it back into the List<DynAttribute> you're expecting.
In my ASP.NET MVC 2 web application, I allow users to create custom input fields of different data types to extend our basic input form. While tricky, building the input form from a collection of custom fields is straight-forward enough.
However, I'm now to the point where I want to handle the posting of this form and I'm not certain what the best way to handle this would be. Normally, we'd use strongly-typed input models that get bound from the various statically-typed inputs available on the form. However, I'm at a loss for how to do this with a variable number of input fields that represent different data types.
A representative input form might look something like:
My date field: [ date time input
control ]
My text field: [ text input
field ]
My file field: [ file upload
control ]
My number field: [ numerical input control ]
My text field 2: [text input field ]
etc...
Ideas I've thought about are:
Sending everything as strings (except for the file inputs, which would need to be handled specially).
Using a model with an "object" property and attempting to bind to that (if this is even possible).
Sending a json request to my controller with the data encoded properly and attempting to parse that.
Manually processing the form collection in my controller post action - certainly an option, but I'd love to avoid this.
Has anyone tackled an issue like this before? If so, how did you solve it?
Update:
My "base" form is handled on another input area all together, so a solution doesn't need to account for any sort of inheritence magic for this. I'm just interested in handling the custom fields on this interface, not my "base" ones.
Update 2:
Thank you to ARM and smartcaveman; both of you provided good guidance for how this could be done. I will update this question with my final solution once its been implemented.
This is how I would begin to approach the issue. A custom model binder would be pretty easy to build based on the FormKey property (which could be determined by the index and/or label, depending).
public class CustomFormModel
{
public string FormId { get; set; }
public string Label { get; set; }
public CustomFieldModel[] Fields { get; set; }
}
public class CustomFieldModel
{
public DataType DateType { get; set; } // System.ComponentModel.DataAnnotations
public string FormKey { get; set; }
public string Label { get; set; }
public object Value { get; set; }
}
public class CustomFieldModel<T> : CustomFieldModel
{
public new T Value { get; set; }
}
Also, I noticed one of the comments below had a filtered model binder system. Jimmy Bogard from Automapper made a really helpful post about this method at http://www.lostechies.com/blogs/jimmy_bogard/archive/2009/03/17/a-better-model-binder.aspx , and later revised in, http://www.lostechies.com/blogs/jimmy_bogard/archive/2009/11/19/a-better-model-binder-addendum.aspx . It has been very helpful for me in building custom model binders.
Update
I realized that I misinterpreted the question, and that he was specifically asking how to handle posting of the form "with a variable number of input fields that represent different data types". I think the best way to do this is to use a structure similar to above but leverage the Composite Pattern. Basically, you will need to create an interface like IFormComponent and implement it for each datatype that would be represented. I wrote and commented an example interface to help explain how this would be accomplished:
public interface IFormComponent
{
// the id on the html form field. In the case of a composite Id, that doesn't have a corresponding
// field you should still use something consistent, since it will be helpful for model binding
// (For example, a CompositeDateField appearing as the third field in the form should have an id
// something like "frmId_3_date" and its child fields would be "frmId_3_date_day", "frmId_3_date_month",
// and "frmId_3_date_year".
string FieldId { get; }
// the human readable field label
string Label { get; }
// some functionality may require knowledge of the
// Parent component. For example, a DayField with a value of "30"
// would need to ask its Parent, a CompositeDateField
// for its MonthField's value in order to validate
// that the month is not "February"
IFormComponent Parent { get; }
// Gets any child components or null if the
// component is a leaf component (has no children).
IList<IFormComponent> GetChildren();
// For leaf components, this method should accept the AttemptedValue from the value provider
// during Model Binding, and create the appropriate value.
// For composites, the input should be delimited in someway, and this method should parse the
// string to create the child components.
void BindTo(string value);
// This method should parse the Children or Underlying value to the
// default used by your business models. (e.g. a CompositeDateField would
// return a DateTime. You can get type safety by creating a FormComponent<TValue>
// which would help to avoid issues in binding.
object GetValue();
// This method would render the field to the http response stream.
// This makes it easy to render the forms simply by looping through
// the array. Implementations could extend this for using an injected
// formatting
void Render(TextWriter writer);
}
I am assuming that the custom forms can be accessed via some sort of id which can be contained as a form parameter. With that assumption, the model binder and provider could look something like this.
public interface IForm : IFormComponent
{
Guid FormId { get; }
void Add(IFormComponent component);
}
public interface IFormRepository
{
IForm GetForm(Guid id);
}
public class CustomFormModelBinder : IModelBinder
{
private readonly IFormRepository _repository;
public object BindModel(ControllerContext controllerContext, ModelBindingContext bindingContext)
{
ValueProviderResult result;
if(bindingContext.ValueProvider.TryGetValue("_customFormId", out result))
{
var form = _repository.GetForm(new Guid(result.AttemptedValue));
var fields = form.GetChildren();
// loop through the fields and bind their values
return form;
}
throw new Exception("Form ID not found.");
}
}
Obviously, all the code here is just to get the point across, and would need to be completed and cleaned up for actual use. Also, even if completed this would only bind to an implementation of the IForm interface, not a strongly typed business object. (It wouldn't be a huge step to convert it to a dictionary and build a strongly typed proxy using the Castle DictionaryAdapter, but since your users are dynamically creating the forms on the site, there is probably no strongly typed model in your solution and this is irrelevant). Hope this helps more.
Take a peek at what I did here: MVC2 Action to handle multiple models and see if can get you on the right track.
If you use a FormCollection as one of your parameters to your action, you can then go thru that form collection looking for bits of data here or there in order to bind those values to whatever an then save the data. You are most likely going to need to take advantage of both strategy and command patterns to get this to work.
Best of luck, feel free to ask follow-up questions.
Edit:
Your method which does the work should look something like this:
private/public void SaveCustomFields(var formId, FormCollection collection) //var as I don't know what type you are using to Id the form.
{
var binders = this.binders.select(b => b.CanHandle(collection)); //I used IOC to get my list of IBinder objects
// Method 1:
binders.ForEach(b => b.Save(formId, collection)); //This is the execution implementation.
// Method 2:
var commands = binders.Select(b => b.Command(formId, collection));
commands.ForEach(c => c.Execute());
}
public DateBinder : IBinder //Example binder
{
public bool CanHandle(FormCollection collection)
{
return (null != collection["MyDateField"]); //Whatever the name of this field is.
}
//Method 1
public void Save(var formId, FormCollection collection)
{
var value = DateTime.Parse(collection["MyDateField"]);
this.someLogic.Save(formId, value); //Save the value with the formId, or however you wish to save it.
}
//Method 2
public Command Command(var formId, FormCollection collection)
{
//I haven't done command pattern before so I'm not sure exactly what to do here.
//Sorry that I can't help further than that.
}
}
I would think one of the best options is to create a custom model binder, which makes it possible to have custom logic behind the scenes and still very customizable code behind.
Maybe these articles can help you:
http://www.gregshackles.com/2010/03/templated-helpers-and-custom-model-binders-in-asp-net-mvc-2/
http://www.singingeels.com/Articles/Model_Binders_in_ASPNET_MVC.aspx
More specifically I would probably take as the controller argument a custom class with all "base" properties included. The class could then for example include a dictionary linking the name of each field to either just an object or an interface which you implement once for each data-type making it simple to process the data later.
/Victor
I have several variables that I need to send from page to page...
What is the best way to do this?
Just send them one by one:
string var1 = Session["var1"] == null ? "" : Session["var1"].ToString();
int var2 = Session["var2"] == null ? 0 : int.Parse(Session["var2"].ToString());
and so on...
Or put them all in some kind of container-object?
struct SessionData
{
public int Var1 { get; set; }
public string Var2 { get; set; }
public int Var3 { get; set; }
}
--
SessionData data = Session["data"] as SessionData;
What is the best solution? What do you use?
A hybrid of the two is the most maintainable approach. The Session offers a low-impedance, flexible key-value pair store so it would be wasteful not to take advantage of that. However, for complex pieces of data that are always related to each other - for example, a UserProfile - it makes sense to have a deeply nested object.
If all the data that you're storing in the Session is related, then I would suggest consolodating it into a single object like your second example:
public class UserData
{
public string UserName { get; set; }
public string LastPageViewed { get; set; }
public int ParentGroupId { get; set; }
}
And then load everything once and store it for the Session.
However, I would not suggest bundling unrelated Session data into a single object. I would break each seperate group of related items into their own. The result would be something of a middleground between the two hardline approaches you provided.
I use a SessionHandler, which is a custom rolled class that looks like this
public static class SessionHandler
{
public static string UserId
{
get
{
return Session["UserId"];
}
set
{
Session["UserId"] = value;
}
}
}
And then in code I do
var user = myDataContext.Users.Where(u => u.UserId = SessionHandler.UserId).FirstOrDefault();
I don't think I've every created an object just to bundle other objects for storage in a session, so I'd probably go with the first option. That said, if you have such a large number of objects that you need to bundle them up to make it easier to work with, you might want to re-examine your architecture.
I've used both. In general, many session variable names leads to a possibility of collisions, which makes collections a litte more reliable. Make sure the collection content relates to a single responsibility, just as you would for any object. (In fact, business objects make excellent candidates for session objects.)
Two tips:
Define all session names as public static readonly variables, and make it a coding standard to use only these static variables when naming session data.
Second, make sure that every object is marked with the [Serializable] attribute. If you ever need to save session state out-of-process, this is essential.
The big plus of an object: properties are strongly-typed.