Possible bug in ASP.NET MVC with form values being replaced

Question

Yes, this behavior is currently by design. Even though you're explicitly setting values, if you post back to the same URL, we look in model state and use the value there. In general, this allows us to display the value you submitted on postback, rather than the original value.

There are two possible solutions:

Solution 1

Use unique names for each of the fields. Note that by default we use the name you specify as the id of the HTML element. It's invalid HTML to have multiple elements have the same id. So using unique names is good practice.

Solution 2

Do not use the Hidden helper. It seems like you really don't need it. Instead, you could do this:

<input type="hidden" name="the-name" 
  value="<%= Html.AttributeEncode(Model.Value) %>" />

Of course, as I think about this more, changing the value based on a postback makes sense for Textboxes, but makes less sense for hidden inputs. We can't change this for v1.0, but I'll consider it for v2. But we need to think through carefully the implications of such a change.

Same as others I would have expected the ModelState to be used to fill the Model and as we explicitly use the Model in expressions in the view, it should use the Model and not ModelState.

It's a design choice and I do get why: if validations fail, the input value might not be parseable to the datatype in the model and you still want to render whatever wrong value the user typed, so it's easy to correct it.

The only thing I don't understand is: why isn't it by design that the Model is used, which is set explicitly by the developer and if a validation error occurred, the ModelState is used.

I have seen many people using workarounds like

• ModelState.Clear(): Clears all ModelState values, but basically disables usage of default validation in MVC
• ModelState.Remove("SomeKey"): Same as ModelState.Clear() but needs micromanagement of ModelState keys, which is too much work and it doesn't feel right with the auto binding feature from MVC. Feels like 20 years back when we were also managing Form and QueryString keys.
• Rendering HTMLthemselves: too much work, detail and throws away the HTML Helper methods with the additional features. An example: Replace @Html.HiddenFor by <input type="hidden" name="@NameFor(m => m.Name)" id="@Html.IdFor(m=>m.Name)" value="@Html.AttributeEncode(Model.Name)">. Or replace @Html.DropDownListFor by ...
• Create custom HTML Helpers to replace default MVC HTML Helpers to avoid the by-design issue. This is a more generic approach then rendering your HTML, but still requires more HTML+MVC knowledge or decompiling System.Web.MVC to still keep all other features but disable ModelState precedence over Model.
• Apply the POST-REDIRECT-GET Pattern: this is easy in some environments, but harder in the ones with more interaction/complexity. This pattern has it's pros and cons and you shouldn't be forced to apply this pattern because of a by-design choice of ModelState over Model.

Issue

So the issue is that the Model is filled from ModelState and in the view, we set explicitly to use the Model. Everybody expects the Model value (in case it changed) to be used unless there's a validation error; then the ModelState can be used.

Currently, in the MVC Helper extensions, the ModelState value gets precedence over the Model value.

Solution

So the actual fix for this issue should be: for each expression to pull the Model value the ModelState value should be removed if there is no validation error for that value. If there's a validation error for that input control the ModelState value shouldn't be removed and it will be used like normal. I think this solves the issue exactly, which is better than most workarounds.

The code is here:

    /// <summary>
    /// Removes the ModelState entry corresponding to the specified property on the model if no validation errors exist. 
    /// Call this when changing Model values on the server after a postback, 
    /// to prevent ModelState entries from taking precedence.
    /// </summary>
    public static void RemoveStateFor<TModel, TProperty>(this HtmlHelper helper,  
        Expression<Func<TModel, TProperty>> expression)
    {
        //First get the expected name value. This is equivalent to helper.NameFor(expression)
        string name = ExpressionHelper.GetExpressionText(expression);
        string fullHtmlFieldName = helper.ViewContext.ViewData.TemplateInfo.GetFullHtmlFieldName(name);

        //Now check whether modelstate errors exist for this input control
        ModelState modelState;
        if (!helper.ViewData.ModelState.TryGetValue(fullHtmlFieldName, out modelState) ||
            modelState.Errors.Count == 0)
        {
            //Only remove ModelState value if no modelstate error exists,
            //so the ModelState will not be used over the Model
            helper.ViewData.ModelState.Remove(name);
        }
    }

And then we create our own HTML Helper extensions todo this before calling the MVC extensions:

    public static MvcHtmlString TextBoxForModel<TModel, TProperty>(this HtmlHelper<TModel> htmlHelper,
        Expression<Func<TModel, TProperty>> expression,
        string format = "",
        Dictionary<string, object> htmlAttributes = null)
    {
        RemoveStateFor(htmlHelper, expression);
        return htmlHelper.TextBoxFor(expression, format, htmlAttributes);
    }

    public static IHtmlString HiddenForModel<TModel, TProperty>(this HtmlHelper<TModel> htmlHelper,
        Expression<Func<TModel, TProperty>> expression)
    {
        RemoveStateFor(htmlHelper, expression);
        return htmlHelper.HiddenFor(expression);
    }

This solution removes the issue but doesn't require you to decompile, analyze, and rebuild whatever MVC is offering you normally (don't forget also managing changes over-time, browser differences, etc.).

I think the logic of "Model value unless validation error then ModelState" should have been by-design. If it was, it wouldn't have bitten so many people, but still covered what MVC was intended todo.

I just ran into same issue. Html helpers like TextBox() precedence for passed values appear to behave exactly opposite what I inferred from the Documentation where it says:

The value of the text input element. If this value is null reference (Nothing in Visual Basic), the value of the element is retrieved from the ViewDataDictionary object. If no value exists there, the value is retrieved from the ModelStateDictionary object.

To me, I read that the value, if passed is used. But reading TextBox() source:

string attemptedValue = (string)htmlHelper.GetModelStateValue(name, typeof(string));
tagBuilder.MergeAttribute("value", attemptedValue ?? ((useViewData) ? htmlHelper.EvalString(name) : valueParameter), isExplicitValue);

seems to indicate that the actual order is the exact opposite of what is documented. Actual order seems to be:

1. ModelState
2. ViewData
3. Value (passed into TextBox() by caller)

Heads-up - this bug still exists in MVC 3. I'm using the Razor markup syntax (like that really matters), but I encountered the same bug with a foreach loop that produced the same value for an object property every single time.

This would be the expected behavoir - MVC doesn't use a viewstate or other behind your back tricks to pass extra information in the form, so it has no idea which form you submitted (the form name is not part of the data submitted, only a list of name/value pairs).

When MVC renders the form back, it is simply checking to see if a submitted value with the same name exists - again, it has no way of knowing which form a named value came from, or even what type of control it was (whether you use a radio, text or hidden, it's all just name=value when its submitted through HTTP).

foreach (var s in ModelState.Keys.ToList())
                if (s.StartsWith("detalleProductos"))
                    ModelState.Remove(s);

ModelState.Remove("TimeStamp");
ModelState.Remove("OtherOfendingHiddenFieldNamePostedToSamePage1");
ModelState.Remove("OtherOfendingHiddenFieldNamePostedToSamePage2");

return View(model);