January 2009 Blog Posts
Here’s a little known fact about T4 Templates…well besides that they exist in the first place. Visual Studio 2005 had support for T3 templates. T4 is the natural successor to T3. T3 is also the abbreviation used when marketing the movie Terminator 3: Rise of the Machines. I heard that the names are related, but I haven’t confirmed it beyond a doubt, which seems appropriate given the title of the movie. But you didn’t hear that from me, deal?
Although they’ve been around forever, I haven’t really dived into T4 templates until recently as I’ve been using them in Subtext. I’m excited that we’re now incorporating them into ASP.NET MVC.
The Visual Web Developer Team Blog just posted an in-depth look at T4 templates in ASP.NET MVC entitled, T4 Templates: A Quick-Start Guide for ASP.NET MVC Developers.
Be sure to check out their other post covering What’s New for MVC Tools in ASP.NET MVC 1.0 Release Candidate.
Hello there. :)
On Tuesday, we announced the release candidate for ASP.NET MVC. While there is much new in there to be excited about and many many bug fixes, there were two changes introduced in the RC that broke some scenarios which previously worked in the Beta, as reported by customers.
We’ve updated the Release Candidate with a refresh that addresses these two issues. You can use the recently released Microsoft Web Platform Installer 1.0 to install ASP.NET MVC RC Refresh. It happens to be a handy tool for installing not just ASP.NET MVC, but everything you might need to use ASP.NET MVC such as Visual Web Developer Express 2008 SP1.
The link on the official download page appears to be updated with the new MSI (we pushed it out yesterday), but we’ve experienced some odd proxy caching issues etc where some people were still getting the old MSI.
In order to be safe, you can get the download directly from this download link. I don’t anticipate any problems with that link, but being paranoid, the way to fully ensure you have the refresh is to right click on the downloaded file, select the Digital Signatures tab, and make sure the Time Stamp says Wednesday, January 28 and not Friday, January 23.
Don’t forget to fully uninstall the previous RC before installing this one.
Also note that this refresh does not address the Controls Collection Cannot Be Modified issue I reported recently. We will address that soon. The following describes the issues that this refresh does fix.
The first change was that we changed our helper methods that generate URLs and links to now generate relative paths instead of absolute paths. This caused problems with several AJAX scenarios.
The fix to this in the refresh was to roll back that check-in so the behavior went back to the way it was in the Beta. We will not use relative paths, and we have no intention of re-introducing the generation of relative paths. Because we rolled back this change to a known good state, we feel very confident in the current behavior. URL generation should work the same way it did back in the Beta.
The other regression is that in some cases, the RouteUrl (and thus RouteLink) methods return an empty string when you specify a route name, but the route has default parameters which are not parameters in the URL.
For example, if you have the following route:
routes.MapRoute("route-name",
"foo/bar",
new {controller="Home", action="index"});
Notice that controller has default value, but is not part of the URL. If you then specify:
<%= Url.RouteUrl("route-name") %>
You might expect that it would use that route to render the URL, but it doesn’t. This bug was introduced when we refactored all our url generating helpers to call into a common method. It turns out, however, that our RouteUrl methods (aka non-MVC specific) should have subtly different behavior than the MVC specific methods (such as Action). We added a flag to the common method so that this difference is taken into consideration. This was a fix that did not have a large surface area.
Another thing we did was update all our test cases (both unit tests and automated functional tests) with scenarios reported to us by customers. So now, we have these types of cases well covered.
If you downloaded the Release Candidate before today, you should check the digital signature timestamp as I described earlier and if it’s the old one, I recommend you go and download the refresh via the Web Platform Installer or directly at the URL I mentioned earlier.
As I mentioned before, we are very excited about this release and hope that you are enjoying writing code with it. :)
Earlier this morning, I posted on making a simple jQuery delete link which makes it easy to create a delete link that does a form post to a delete action. Commenters pointed out that my solution won’t work for down-level browsers such as some mobile phones, and they were right. I wasn’t really concerned about down-level browsers.
One solution for down-level browsers is to render a proper form with a submit button, and then hide the form with JavaScript. Of course this takes a bit more work. Here’s what I did. I made sure I had the following script in my master template.
<script type="text/javascript">
$("form.delete-link").css("display", "none");
$("a.delete-link").show();
$("a.delete-link").live('click', function(ev) {
ev.preventDefault();
$("form.delete-link").submit();
});
</script>
When the following HTML is rendered in the page…
<form method="post" action="/go/delete/1"
class="delete-link">
<input type="submit" value="delete" />
<input name="__RequestVerificationToken" type="hidden"
value="Jrcn83M7T...8Z6RkdIfMZIJ5mVb" />
</form>
<a class="delete-link" href="/go/delete/1"
style="display:none;">delete</a>
… the jQuery code shown above will hide the form, but display the link (notice the link is hidden by default). When the link is clicked, it posts the form. However, in cases where there is no JavaScript, the form will be displayed, but the link will not be because the JavaScript is the thing that hides the form.
To make this easier to use, I wrote the following helper:
public static string DeleteLink(this HtmlHelper html
, string linkText
, string routeName
, object routeValues) {
var urlHelper = new UrlHelper(html.ViewContext.RequestContext);
string url = urlHelper.RouteUrl(routeName, routeValues);
string format = @"<form method=""post"" action=""{0}""
class=""delete-link"">
<input type=""submit"" value=""{1}"" />
{2}
</form>";
string form = string.Format(format, html.AttributeEncode(url)
, html.AttributeEncode(linkText)
, html.AntiForgeryToken());
return form + html.RouteLink(linkText, routeName, routeValues
, new { @class = "delete-link", style = "display:none;" });
}
Notice that we’re using the AntiForgery helpers included with ASP.NET MVC. What this means is that I need to make one small change to my delete method on my controller. I need to add the ValidateAntiForgeryToken attribute to the method.
[ValidateAntiForgeryToken]
[AcceptVerbs(HttpVerbs.Post)]
public ActionResult Delete(int id) {
//Delete it
}
I’ve left out a bit. For example, I didn’t specify a callback to the jQuery code. So what should happen when this action method returns? I leave that as an exercise to the reader. I may address it in a future follow-up to this blog post. In my code, I’m just being cheesy and doing a full redirect, which works fine.
Technorati Tags:
aspnetmvc,
security
UPDATE: I have a followup to this post that works for down-level browsers.
In a recent post, Stephen Walther pointed out the dangers of using a link to delete data. Go read it as it provides very good coverage of the issues. The problem is not restricted to delete operations. Any time you allow a GET request to modify data, you’re asking for trouble. Read this story about something that happened to BackPack way back in the day to see what I mean.
The reason that delete operations deserve special attention is that it’s the most common case where you would use a link to change information. If you were editing a product record, for example, you would use a form. But a delete operation typically only needs one piece of information (the id) which is easy to encode in the URL of a GET request.
If you are using jQuery, one simple way to turn any link into a POST link is to add the following onclick attribute value:
$.post(this.href); return false;
For example
<a href="/go/delete/1" onclick="$.post(this.href); return false;">Delete</a>
Will now make a POST request to /go/delete/1 rather than a GET. Of course, you need to enforce this on the server side. This is pretty easy with ASP.NET MVC.
[AcceptVerbs(HttpVerbs.Post)]
public ActionResult Delete(int id) {
//Delete that stuff!
}
The AcceptVerbs attribute specifies that this action method only responds to POST requests, not GET requests.
At this point, you could easily write helpers specifically for delete links. I usually write very specific helper methods such as Html.DeleteProduct or Html.DeleteQuestion. Here’s an example of one I wrote for a sample app I’m building.
public static string DeleteAnswerLink(this HtmlHelper html, string linkText
, Answer answer) {
return html.RouteLink(linkText, "answer",
new { answerId = answer.Id, action = "delete" },
new { onclick="$.post(this.href); return false;" });
}
The nice thing about this approach is that you can leverage the existing helper methods by adding a minimal amount of extra information via the onclick attribute.
I hope the combination of Stephen’s post and this post will lead you to safer deleting.
In my last post, I announced the happy news that the Release Candidate for ASP.NET MVC is available. In this post, I say mea culpa for a known bug within this release.
This bug is a consequence of a change we made in our default template. We know have a content placeholder in the <head> section of the Site.master page.
<head runat="server">
<asp:ContentPlaceHolder ID="head" runat="server">
<title></title>
</asp:ContentPlaceHolder>
<link href="../../Content/Site.css" rel="stylesheet" type="text/css" />
</head>
The benefit here is that it makes it easy to specify view specific scripts, style sheets, and title tags from the specific view, like so.
<asp:Content ID="indexHead" ContentPlaceHolderID="head" runat="server">
<title>Home Page</title>
<style type="text/css">
/* Some style specific to this view */
</style>
</asp:Content>
Long story short, if the Header control (aka <head runat="server" />) doesn’t see a title tag as a direct child of itself, it renders an empty title tag in order to ensure the page has valid HTML. This results in having two title tags, the one you intended and an empty one.
We put in a fix for this whereby we modify the controls collection of the header control, but the fix itself causes a problem when you have code nuggets within the <head> section. For example, the following causes an exception.
<head runat="server">
<script src="<%= Url.Content("~/foo.js") %>" type="text/javascript">
</script>
<asp:ContentPlaceHolder ID="head" runat="server">
<title></title>
</asp:ContentPlaceHolder>
<link href="../../Content/Site.css" rel="stylesheet" type="text/css" />
</head>
The exception this causes is:
The Controls collection cannot be modified because the control contains code blocks (i.e. <% ... %>).
We unfortunately didn’t find this until very recently. The current workaround is simple. Place the script tag within a normal PlaceHolder control.
<head runat="server">
<asp:PlaceHolder runat="server" id="mainScripts">
<script src="<%= Url.Content("~/foo.js") %>" type="text/javascript">
</script>
</asp:PlaceHolder>
<asp:ContentPlaceHolder ID="head" runat="server">
<title></title>
</asp:ContentPlaceHolder>
<link href="../../Content/Site.css" rel="stylesheet" type="text/css" />
</head>
We have one simple solution to this we are bouncing around, and are investigating alternative solutions. We apologize for any inconveniences this may cause.
At long last I am happy, relieved, excited to announce the release candidate for ASP.NET MVC. Feel free to go download it now. I’ll wait right here patiently.
There have been a lot of improvements made since the Beta release so be sure to read the release notes. I’ve tried very hard to be thorough in the notes so do let me know if anything is lacking. We are also pushing new tutorials up to the ASP.NET MVC Website as I write this.
Also, don’t miss ScottGu’s usual epic blog post describing the many improvements. There’s also a post on the Web Tools team blog covering tooling aspects of this release in detail. As I mentioned when we released the Beta, we didn’t have plans for many new features in the runtime, but we did have a lot of tooling improvements to add. I’ve already described some of these changes in a previous blog post, as did ScottGu in his detailed look.
Our goal with this release was to fix all outstanding bugs which we felt were real showstoppers and try to address specific customer concerns. We worked hard to add a bit of spit and polish to this release.
Unfortunately, a few minor bugs did crop up at the last second, but we decided we could continue with this RC and fix the bugs afterwards as the impact appear to be relatively small and they all have workarounds. I wrote about one particular bug so that you’re aware of it.
For now, I want to share a few highlights. This is not an exhaustive list at all. For that, check out Scott’s post and the release notes.
Ajax
Yes, I do know that jQuery released a new version (1.3.1), and no, it is not in this release. :) We just didn’t have time to include it due to the timing of its release. However, we are performing due diligence right now and plan to include it with the RTM.
We did make some changes to our Ajax helpers to recognize the standard X-Requested-With HTTP header used by the major JavaScript libraries such as Prototype.js, jQuery, and Dojo. Thus the IsMvcAjaxRequest method was renamed to IsAjaxRequest and looks for this header rather than our custom one.
ControllerContext
ControllerContext no longer inherits from RequestContext, which will improve testing and extensibility scenarios. We would have liked to make changes to RequestContext, but it was introduced as part of the .NET Framework in ASP.NET 3.5 SP1. Thus we can’t change it in our out-of-band release.
Anti Forgery Helpers
These helpers were previously released in our “futures” assembly, but we’ve fixed a few bugs and moved them into the core ASP.NET MVC assembly.
These are helpers which help mitigate Cross-Site Request Forgery (CSRF) attacks. For a great description of these helpers, check out Steve Sanderson’s blog post on the topic.
MVC Futures
I added a couple of expression based helpers to the ASP.NET MVC futures assembly, Microsoft.Web.Mvc.dll. These are just samples to demonstrate how one could write such helpers. I’ll add a few more by the time we ship the RTM. Note, if you’re using the old futures assembly, it won’t work with the new RC. You’ll need to update to the new Futures assembly.
In case you missed it the first time, click here for the Download.
Technorati Tags:
aspnetmvc
A while back on a lark, I posted a prototype demonstrating how one could use Routing within Web Forms. This is something you can do today with ASP.NET 3.5 SP1, because of the work we did to separate Routing from ASP.NET MVC. I would have liked to include Web Form Routing as part of the Routing feature when we were working on SP1, but we didn’t have the time to do so in a robust manner before SP1 was locked down.
Since then, Scott Galloway, who just happens to be my office mate, has taken the reigns and is the PM in change of guiding the Web Form Routing feature to be included in ASP.NET 4.0 in a more integrated fashion.
He wrote a blog post earlier today describing some of the ways that routing will be more deeply integrated into ASP.NET such as new properties of the Page class, cool uses of Expression Builders, etc… That’s just a teaser as he promises to go a little more in depth soon.
Technorati Tags:
Routing,
ASP.NET
UPDATE: Be sure to read Peli’s post in which he explores all of these implementations using PEX. Apparently I have a lot more unit tests to write in order to define the expected behavior of the code.
I recently wrote a post in which I examined some existing implementations of named format methods and then described the fun I had writing a routine for implementing named formats.
In response to that post, I received two new implementations of the method that are worth calling out.
The first was sent to me by Atif Aziz, who I’ve corresponded with via email regarding Jayrock. He noticed that the slowest format method by James had one obvious performance flaw and fixed it up. He then sent me another version that passed all my unit tests. Here it is in full, and it’s much faster than before.
public static class JamesFormatter
{
public static string JamesFormat(this string format, object source)
{
return FormatWith(format, null, source);
}
public static string FormatWith(this string format
, IFormatProvider provider, object source)
{
if (format == null)
throw new ArgumentNullException("format");
List<object> values = new List<object>();
string rewrittenFormat = Regex.Replace(format,
@"(?<start>\{)+(?<property>[\w\.\[\]]+)(?<format>:[^}]+)?(?<end>\})+",
delegate(Match m)
{
Group startGroup = m.Groups["start"];
Group propertyGroup = m.Groups["property"];
Group formatGroup = m.Groups["format"];
Group endGroup = m.Groups["end"];
values.Add((propertyGroup.Value == "0")
? source
: Eval(source, propertyGroup.Value));
int openings = startGroup.Captures.Count;
int closings = endGroup.Captures.Count;
return openings > closings || openings % 2 == 0
? m.Value
: new string('{', openings) + (values.Count - 1)
+ formatGroup.Value
+ new string('}', closings);
},
RegexOptions.Compiled
| RegexOptions.CultureInvariant
| RegexOptions.IgnoreCase);
return string.Format(provider, rewrittenFormat, values.ToArray());
}
private static object Eval(object source, string expression)
{
try
{
return DataBinder.Eval(source, expression);
}
catch (HttpException e)
{
throw new FormatException(null, e);
}
}
}
The other version I got was from another Microsoftie named Henri Wiechers (no blog) who pointed out that a state machine of the sort that a parser or scanner would use, is well suited for this task. It also makes the code easier to understand if you’re accustomed to state machines.
public static class HenriFormatter
{
private static string OutExpression(object source, string expression)
{
string format = "";
int colonIndex = expression.IndexOf(':');
if (colonIndex > 0)
{
format = expression.Substring(colonIndex + 1);
expression = expression.Substring(0, colonIndex);
}
try
{
if (String.IsNullOrEmpty(format))
{
return (DataBinder.Eval(source, expression) ?? "").ToString();
}
return DataBinder.Eval(source, expression, "{0:" + format + "}")
?? "";
}
catch (HttpException)
{
throw new FormatException();
}
}
public static string HenriFormat(this string format, object source)
{
if (format == null)
{
throw new ArgumentNullException("format");
}
StringBuilder result = new StringBuilder(format.Length * 2);
using (var reader = new StringReader(format))
{
StringBuilder expression = new StringBuilder();
int @char = -1;
State state = State.OutsideExpression;
do
{
switch (state)
{
case State.OutsideExpression:
@char = reader.Read();
switch (@char)
{
case -1:
state = State.End;
break;
case '{':
state = State.OnOpenBracket;
break;
case '}':
state = State.OnCloseBracket;
break;
default:
result.Append((char)@char);
break;
}
break;
case State.OnOpenBracket:
@char = reader.Read();
switch (@char)
{
case -1:
throw new FormatException();
case '{':
result.Append('{');
state = State.OutsideExpression;
break;
default:
expression.Append((char)@char);
state = State.InsideExpression;
break;
}
break;
case State.InsideExpression:
@char = reader.Read();
switch (@char)
{
case -1:
throw new FormatException();
case '}':
result.Append(OutExpression(source, expression.ToString()));
expression.Length = 0;
state = State.OutsideExpression;
break;
default:
expression.Append((char)@char);
break;
}
break;
case State.OnCloseBracket:
@char = reader.Read();
switch (@char)
{
case '}':
result.Append('}');
state = State.OutsideExpression;
break;
default:
throw new FormatException();
}
break;
default:
throw new InvalidOperationException("Invalid state.");
}
} while (state != State.End);
}
return result.ToString();
}
private enum State
{
OutsideExpression,
OnOpenBracket,
InsideExpression,
OnCloseBracket,
End
}
}
As before, I wanted to emphasize that this was not a performance contest. I was interested in correctness with reasonable performance.
Now that the JamesFormatter is up to speed, I was able to crank up the number of iterations to 50,000 without testing my patience and tried it again. I also made the format string slightly longer. Here are the new results.
As you can see, they are all very close, though Henri’s method is the fastest.
I went ahead and updated the solution I had uploaded last time so you can try it yourself by downloading it now.
I’ve been having trouble getting to sleep lately, so I thought last night that I would put that to use and hack on Subtext a bit. While doing so, I ran into an old Asynchronous Fire and Forget helper method written way back by Mike Woodring which allows you to easily call a delegate asynchronously.
On the face of it, it seems like you could simply call BeginInvoke on the delegate and be done with it, Mike’s code addresses a concern with that approach:
Starting with the 1.1 release of the .NET Framework, the SDK docs now carry a caution that mandates calling EndInvoke on delegates you've called BeginInvoke on in order to avoid potential leaks. This means you cannot simply "fire-and-forget" a call to BeginInvoke without the risk of running the risk of causing problems.
The mandate he’s referring to, I believe, is this clause in the MSDN docs:
No matter which technique you use, always call EndInvoke to complete your asynchronous call.
Note that it doesn’t explicitly say “or you will get a memory leak”. But a little digging turns up the following comment in the MSDN forums.
The reason that you should call EndInvoke is because the results of the invocation (even if there is no return value) must be cached by .NET until EndInvoke is called. For example if the invoked code throws an exception then the exception is cached in the invocation data. Until you call EndInvoke it remains in memory. After you call EndInvoke the memory can be released. For this particular case it is possible the memory will remain until the process shuts down because the data is maintained internally by the invocation code. I guess the GC might eventually collect it but I don't know how the GC would know that you have abandoned the data vs. just taking a really long time to retrieve it. I doubt it does. Hence a memory leak can occur.
The thread continues to have some back and forth and doesn’t appear to be conclusive either way, but this post by Don Box gives a very pragmatic argument.
…the reality is that some implementations rely on the EndXXX call to clean up resources. Sometimes you can get away with it, but in the general case you can't.
In other words, why take the chance? In any case, much of this discussion is made redundant with the C# 3.0 Action class combined with ThreadPool.QueueUserWorkItem aka (QUWI)
Here is the code in Subtext for sending email, more or less. I have to define a delegate and then pass that to FireAndForget
// declaration
delegate bool SendEmailDelegate(string to,
string from,
string subject,
string body);
//... in a method body
SendEmailDelegate sendEmail = im.Send;
AsyncHelper.FireAndForget(sendEmail, to, from, subject, body);
This code relies on the FireAndForget method which I show here. Note I am not showing the full code. I just wanted to point out that the arguments to the delegate are not strongly typed. They are simply an array of objects which provide no guidance to how many arguments you need to pass.
public static void FireAndForget(Delegate d, params object[] args)
{
ThreadPool.QueueUserWorkItem(dynamicInvokeShim,
new TargetInfo(d, args));
}
Also notice that this implementation uses QUWI under the hood.
With C# 3.0, there is no need to abstract away the call to QUWI. Just pass in a lambda, which provides the benefit that you’re calling the actual method directly so you get Intellisense for the argumennts etc… So all that code gets replaced with:
ThreadPool.QueueUserWorkItem(callback => im.Send(to, from, subject, body));
Much cleaner and I get to get rid of more code! As I’ve said before, the only thing better than writing code is getting rid of code!
Rob pinged me today asking about how to respond to requests using different formats based on the extension in the URL. More specifically, he’d like to respond with HTML if there is no file extension, but with JSON if the URL ended with .json etc...
/home/index –> HTML
/home/index.json –> JSON
The first thing I wanted to tackle was writing a custom action invoker that would decide based on what’s in the route data, how to format the response.
This would allow the developer to simply return an object (the model) from an action method and the invoker would look for the format in the route data and figure out what format to send.
So I wrote a custom action invoker:
public class FormatHandlingActionInvoker : ControllerActionInvoker {
protected override ActionResult CreateActionResult(
ControllerContext controllerContext,
ActionDescriptor actionDescriptor,
object actionReturnValue) {
if (actionReturnValue == null) {
return new EmptyResult();
}
ActionResult actionResult = actionReturnValue as ActionResult;
if (actionResult != null) {
return actionResult;
}
string format = (controllerContext.RouteData.Values["format"]
?? "") as string;
switch (format) {
case "":
case "html":
var result = new ViewResult {
ViewData = controllerContext.Controller.ViewData,
ViewName = actionDescriptor.ActionName
};
result.ViewData.Model = actionReturnValue;
return result;
case "rss":
//TODO: RSS Result
break;
case "json":
return new JsonResult { Data = actionReturnValue };
}
return new ContentResult {
Content = Convert.ToString(actionReturnValue,
CultureInfo.InvariantCulture)
};
}
}
The key thing to note is that I overrode the method CreateActionResult. This method is responsible for examining the result returned from an action method (which can be any type) and figuring out what to do with it. In this case, if the result is already an ActionResult, we just use it. However, if it’s something else, we look at the format in the route data to figure out what to return.
For reference, here’s the code for the HomeController which simply returns an object.
[HandleError]
public class HomeController : Controller {
public object Index() {
return new {Title = "HomePage", Message = "Welcome to ASP.NET MVC" };
}
}
In order to make sure that all my controllers replaced the default invoker with this invoker, I wrote a controller factory that would set this invoker. I won’t show the code here, but I will include it in the download.
So at this point, we have everything in place, except for the fact that I haven’t dealt with how we get the format in the route data in the first place. Unfortunately, it ends up that this isn’t quite so straightforward. Consider the default route:
routes.MapRoute(
"Default",
"{controller}/{action}/{id}",
new { controller = "Home", action = "Index", id = "" }
);
Since this route allows for default values at each segment, this single route matches all the following URLs:
- /home/index/123
- /home/index
- /home
So the question becomes, if we want to support optional format extensions in the URL, would we have to support it for every segment? Making up a fictional syntax, maybe it would look like this:
routes.MapRoute(
"Default",
"{controller{.format}}/{action{.format}}/{id{.format}}",
new { controller = "Home", action = "Index", id = "", format = ""}
);
Where the {.format} part would be optional. Of course, we don’t have such a syntax available,
so I needed to put on my dirty ugly hacking hat and see what I could come up with.
I decided to do something we strongly warn people not to do, inheriting HttpRequestWrapper with my own HttpRequestBase
implementation and stripping off the extension before I try and match the routes.
Warning! Don’t do this at home! This is merely experimentation while I mull over a better approach. This approach relies on implementation details I should not be relying upon
public class HttpRequestWithFormat : HttpRequestWrapper
{
public HttpRequestWithFormat(HttpRequest request) : base(request) {
}
public override string AppRelativeCurrentExecutionFilePath {
get
{
string filePath = base.AppRelativeCurrentExecutionFilePath;
string extension = System.IO.Path.GetExtension(filePath);
if (String.IsNullOrEmpty(extension)) {
return filePath;
}
else {
Format = extension.Substring(1);
filePath = filePath.Substring(0, filePath.LastIndexOf(extension));
}
return filePath;
}
}
public string Format {
get;
private set;
}
}
I also had to write a custom route.
public class FormatRoute : Route
{
public FormatRoute(Route route) :
base(route.Url + ".{format}", route.RouteHandler) {
_originalRoute = route;
}
public override RouteData GetRouteData(HttpContextBase httpContext)
{
//HACK!
var context = new HttpContextWithFormat(HttpContext.Current);
var routeData = _originalRoute.GetRouteData(context);
var request = context.Request as HttpRequestWithFormat;
if (!string.IsNullOrEmpty(request.Format)) {
routeData.Values.Add("format", request.Format);
}
return routeData;
}
public override VirtualPathData GetVirtualPath(
RequestContext requestContext, RouteValueDictionary values)
{
var vpd = base.GetVirtualPath(requestContext, values);
if (vpd == null) {
return _originalRoute.GetVirtualPath(requestContext, values);
}
// Hack! Let's fix up the URL. Since "id" can be empty string,
// we want to check for this condition.
string format = values["format"] as string;
string funkyEnding = "/." + format as string;
if (vpd.VirtualPath.EndsWith(funkyEnding)) {
string virtualPath = vpd.VirtualPath;
int lastIndex = virtualPath.LastIndexOf(funkyEnding);
virtualPath = virtualPath.Substring(0, lastIndex) + "." + format;
vpd.VirtualPath = virtualPath;
}
return vpd;
}
private Route _originalRoute;
}
When matching incoming requests, this route replaces the HttpContextBase with my faked up one before calling GetRouteData. My faked up context returns a faked up request which strips the format extension from the URL.
Also, when generating URLs, this route deals with the cosmetic issue that the last segment of the URL has a default value of empty string. This makes it so that the URL might end up looking like /home/index/.json when I really wanted it to look like /home/index.json.
I’ve omitted some code from this blog post, but you can download the project here and try it out. Just navigate to /home/index and then try /home/index.json and you should notice the response format changes.
This is just experimental work. There’d be much more to do to make this useful. For example, would be nice if an action could specify which formats it would respond to. Likewise, it might be nice to respond based on accept headers rather than formats. I just wanted to see how automatic I could make it.
In any case, I was just having fun and didn’t have much time to put this together. The takeaway from this is really the CreateActionResult method of ControllerActionInvoker. That makes it very easy to create interesting default behavior so that your action methods can return whatever they want and you can implement your own conventions by overriding that method.
Download the hacky experiment here and play with it at your own risk. :)
DOUBLE UPDATE! Be sure to read Peli’s post in which he explores all of these implementations using PEX. Apparently I have a lot more unit tests to write in order to define the expected behavior of the code.
UPDATE: By the way, after you read this post, check out the post in which I revisit this topic and add two more implementations to check out.
Recently I found myself in a situation where I wanted to format a string using a named format string, rather than a positional one. Ignore for the moment the issue on whether this is a good idea or not, just trust me that I’ll be responsible with it.
The existing String.Format method, for example, formats values according to position.
string s = string.Format("{0} first, {1} second", 3.14, DateTime.Now);
But what I wanted was to be able to use the name of properties/fields, rather than position like so:
var someObj = new {pi = 3.14, date = DateTime.Now};
string s = NamedFormat("{pi} first, {date} second", someObj);
Looking around the internet, I quickly found three implementations mentioned in this StackOverflow question.
All three implementations are fairly similar in that they all use regular expressions for the parsing. Hanselman’s approach is to write an extension method of object (note that this won’t work in VB.NET until they allow extending object). James and Oskar wrote extension methods of the string class. James takes it a bit further by using DataBinder.Eval on each token, which allows you to have formats such as {foo.bar.baz} where baz is a property of bar which is a property of foo. This is something else I wanted, which the others do not provide.
He also makes good use of the MatchEvaluator delegate argument to the Regex.Replace method, perhaps one of the most underused yet powerful features of the Regex class. This ends up making the code for his method very succinct.
Handling Brace Escaping
I hade a chat about this sort of string parsing with Eilon recently and he mentioned that many developers tend to ignore or get escaping wrong. So I thought I would see how these methods handle a simple test of escaping the braces.
String.Format with the following:
Console.WriteLine(String.Format("{{{0}}}", 123));
produces the output (sans quotes) of “{123}”
So I would expect with each of these methods, that:
Console.WriteLine(NamedFormat("{{{foo}}}", new {foo = 123}));
Would produce the exact same output, “{123}”. However, only James’s method passed this test. But when I expanded the test to the following format, “{{{{{foo}}}}}”, all three failed. That should have produced “{{123}}”.
Certainly this is not such a big deal as this really is an edge case, but you never know when an edge case might bite you as Zune owners learned recently. More importantly, it poses an interesting problem - how do you handle this correctly? I thought it would be fun to try.
This is possible to handle correctly using regular expressions, but it’s challenging. Not only are you dealing with balanced matching, but the matching depends on whether the number of consecutive braces are odd or even.
For example, the following “{0}}” is not valid because the right end brace is escaped. However, “{0}}}” is valid. The expression is closed with the leftmost end brace, which is followed by an even number of consecutive braces, which means they are all escaped sequences.
Performance
As I mentioned earlier, only James’s method handles evaluation of sub-properties/fields via the use of the DataBinder.Eval method. Critics of his blog post point out that this is a performance killer.
Personally, until I’ve measured it in the scenarios in which I plan to use it, I doubt that the performance will really be an issue compared to everything else going on. But I thought I would check it out anyways, writing a simple console app which runs each method over 1000 iterations, and then divides by 1000 to get the number of milliseconds each method takes. Here’s the result:
Notice that James’s method is 43 times slower than Hanselman’s. Even so, it only takes 4.4 milliseconds. So if you don’t use it in a tight loop with a lot of iterations, it’s not horrible, but it could be better.
My Implementation
At this point, I thought it would be fun to write my own implementation using manual string parsing rather than regular expressions. I’m not sure my regex-fu is capable of handling the challenges I mentioned before. After implementing my own version, I ran the performance test and saw the following result.
Nice! by removing the overhead of using a regular expression in this particular case, my implementation is faster than the other implementations, despite my use of DataBinder.Eval. Hopefully my implementation is correct, because fast and wrong is even worse than slow and right.
One drawback to not using regular expressions is that the code for my implementation is a bit long. I include the entire source here. I’ve also zipped up the code for this solution which includes unit tests as well as the implementations of the other methods I tested, so you can see which tests they pass and which they don’t pass.
The core of the code is in two parts. One is a private method which parses and splits the string into an enumeration of segments represented by the ITextExpression interface. The method you call joins these segments together, evaluating any expressions against a supplied object, and returning the resulting string.
I think we could optimize the code even more by joining these operations into a single method, but I really liked the separation between the parsing and joining logic as it helped me wrap my head around it. Initially, I hoped that I could cache the parsed representation of the format string since strings are immutable thus I could re-use it. But it didn’t end up giving me any real performance gain when I measured it.
public static class HaackFormatter
{
public static string HaackFormat(this string format, object source)
{
if (format == null) {
throw new ArgumentNullException("format");
}
var formattedStrings = (from expression in SplitFormat(format)
select expression.Eval(source)).ToArray();
return String.Join("", formattedStrings);
}
private static IEnumerable<ITextExpression> SplitFormat(string format)
{
int exprEndIndex = -1;
int expStartIndex;
do
{
expStartIndex = format.IndexOfExpressionStart(exprEndIndex + 1);
if (expStartIndex < 0)
{
//everything after last end brace index.
if (exprEndIndex + 1 < format.Length)
{
yield return new LiteralFormat(
format.Substring(exprEndIndex + 1));
}
break;
}
if (expStartIndex - exprEndIndex - 1 > 0)
{
//everything up to next start brace index
yield return new LiteralFormat(format.Substring(exprEndIndex + 1
, expStartIndex - exprEndIndex - 1));
}
int endBraceIndex = format.IndexOfExpressionEnd(expStartIndex + 1);
if (endBraceIndex < 0)
{
//rest of string, no end brace (could be invalid expression)
yield return new FormatExpression(format.Substring(expStartIndex));
}
else
{
exprEndIndex = endBraceIndex;
//everything from start to end brace.
yield return new FormatExpression(format.Substring(expStartIndex
, endBraceIndex - expStartIndex + 1));
}
} while (expStartIndex > -1);
}
static int IndexOfExpressionStart(this string format, int startIndex) {
int index = format.IndexOf('{', startIndex);
if (index == -1) {
return index;
}
//peek ahead.
if (index + 1 < format.Length) {
char nextChar = format[index + 1];
if (nextChar == '{') {
return IndexOfExpressionStart(format, index + 2);
}
}
return index;
}
static int IndexOfExpressionEnd(this string format, int startIndex)
{
int endBraceIndex = format.IndexOf('}', startIndex);
if (endBraceIndex == -1) {
return endBraceIndex;
}
//start peeking ahead until there are no more braces...
// }}}}
int braceCount = 0;
for (int i = endBraceIndex + 1; i < format.Length; i++) {
if (format[i] == '}') {
braceCount++;
}
else {
break;
}
}
if (braceCount % 2 == 1) {
return IndexOfExpressionEnd(format, endBraceIndex + braceCount + 1);
}
return endBraceIndex;
}
}
And the code for the supporting classes
public class FormatExpression : ITextExpression
{
bool _invalidExpression = false;
public FormatExpression(string expression) {
if (!expression.StartsWith("{") || !expression.EndsWith("}")) {
_invalidExpression = true;
Expression = expression;
return;
}
string expressionWithoutBraces = expression.Substring(1
, expression.Length - 2);
int colonIndex = expressionWithoutBraces.IndexOf(':');
if (colonIndex < 0) {
Expression = expressionWithoutBraces;
}
else {
Expression = expressionWithoutBraces.Substring(0, colonIndex);
Format = expressionWithoutBraces.Substring(colonIndex + 1);
}
}
public string Expression {
get;
private set;
}
public string Format
{
get;
private set;
}
public string Eval(object o) {
if (_invalidExpression) {
throw new FormatException("Invalid expression");
}
try
{
if (String.IsNullOrEmpty(Format))
{
return (DataBinder.Eval(o, Expression) ?? string.Empty).ToString();
}
return (DataBinder.Eval(o, Expression, "{0:" + Format + "}") ??
string.Empty).ToString();
}
catch (ArgumentException) {
throw new FormatException();
}
catch (HttpException) {
throw new FormatException();
}
}
}
public class LiteralFormat : ITextExpression
{
public LiteralFormat(string literalText) {
LiteralText = literalText;
}
public string LiteralText {
get;
private set;
}
public string Eval(object o) {
string literalText = LiteralText
.Replace("{{", "{")
.Replace("}}", "}");
return literalText;
}
}
I mainly did this for fun, though I plan to use this method in Subtext for email fomatting.
Let me know if you find any situations or edge cases in which my version fails. I’ll probably be adding more test cases as I integrate this into Subtext. As far as I can tell, it handles normal formatting and brace escaping correctly.