The case of the non-enforced foreign key relationship

While working with the TicketDesk 2.0 MVC project, I came across a really unusual situation within the relational database that TicketDesk uses.

The resulting voyage of discovery lead to me finding my first legitimate use of a "non-enforced foreign key relationship" in a relational database... The situation required an explicitly defined foreign key relationship, but also prohibited the server being able to enforce the relationship.

Here is how it happened...

I'm trying to make as few changes to the TicketDesk 1.x data model as possible. The focus for the 2.0 project is the shift to the new MVC platform and sticking with the old data model makes it easier to support upgrades. Plus the existing model is pretty decent as it is.

In TicketDesk, attachments are stored in the database rather than the file system. This prevents the need for write access to the web server's file system. It also allows you to copy or move ALL of the data by just moving the database. Simpler all-around.

We could have a huge discussion about the drawbacks and advantages of this design. The wisdom of using a RDBMS to store binary file data is very debatable. But the end-analysis here is that TicketDesk isn't expected to handle a lot of files, they are not likely to be large files, and the advantages of having all the ticket data in the database outweighs the disadvantages in most environments.

Anyway, the big problem in TicketDesk 1.x is that the client-side file upload mechanism is terrible. It does gets the job done, but it requires that files be uploaded together in a single HTTP POST. This can be slow, the process is subject to request timeouts and size limits, and it uses a LOT of resources on the server. Additionally, if the server rejects the submission then the user has to re-upload everything all over again.

For TicketDesk 2.x I wanted to switch to using a flash based uploader. I personally HATE flash, but when it comes to file uploading plain HTML and javascript fail to provide any good solutions.

The flash uploader will send the files up to the server using separate out-of-band connections to the web server. This means that the server has to "put" the files somewhere until the user has finished submitting the rest of the form.

For new tickets, the files will get uploaded before there is a ticket in the database to associate them with. For existing tickets, the uploads need to be held in a pending status until the user has finished supplying the meta-data  (descriptions, file names, and comments for the activity history log).

I really didn't want the server to dump pending attachments to the file system or hold them in memory, so I needed to modify the database to provide a place for the server to put the uploaded files as soon as they arrived.

The original data model looked like this:

As you can see, there are two tables. Tickets and TicketAttachments. The contents of the file are directly stored in TicketAttachments along with meta-data (file size, name, description, file type, etc). The old primary key was on both TicketId and FileId... not my best primary key definition ever, but it made sense at the time. TicketId is foreign keyed to the TicketId column in parent Tickets table.

Not a remarkable design really.

At first I toyed with creating a new "PendingFiles" table. But there was already a table with the right kind of structure in the database. It seemed to be a waste to duplicate table structure and I've never been a fan of "staging" tables anyway.

So I started by figuring out how to use the existing TicketAttachments table to store the "pending" files as well as "real" files.

For pending attachments to an existing ticket, the table is already pretty good. All that is needed is a column to flag files as pending instead of real. So I added an "IsPending" bit column to the table. New attachments have IsPending flagged. Once the user commits the changes, we just flip the bit to make the file a real attachment.

Here is the revision.

Now, to handle attachments for new tickets. I figured that I could put the files in the same table, but we'd need to leave the TicketId column null... we don't know the TicketId yet.

I didn't like the primary key as it was anyway. FileId is an identity column and can serve as a primary key all by itself.

So, I changed the primary key to FileId then marked TicketId column to allow nulls and got this:

Great! Except...

Because of the foreign key relationship, we can't add a row to TicketAttachments with a null TicketId value unless there is a row in the Tickets table that also has a null TicketId. We can't make TicketId in the Tickets table nullable because it is the primary key (and an identity field to boot).

Ok fine... the DBA in me screamed bloody murder, but I deleted the foreign key relationship. I've been doing web development long enough to know when the advantages of denormalization outweigh good relational design. I hate having to manage relationships purely in code, but sometimes that really is the best way.

So we get to this:

But! LINQ to SQL needs the foreign key to generate the relationship between the entities. I could manually define the relationship in the LINQ to SQL mapping file, but I *know* that this is going to bite me in the ass later.... it always does. Combine that with the uncomfortable absence of a foreign key in the database itself and I just couldn't bring myself to let that slide.

Then I remembered that there is this odd setting in the SQL Management Tools. The designer for editing foreign keys has a true/false flag for "enforce relationship". I never paid it much attention... I mean, what crazy bastard would go to all the effort to define a foreign key, then tell the server to ignore it?

That just never made any sense to me. Until now...

So I did some testing and research. If you set a foreign key up in SQL, but tell it not to enforce it (this is called the NOCHECK option) then the SQL server will behave as if the foreign key doesn't exist. Thus, we can add rows to TicketAttachments that don't correspond to a row in the parent table in direct violation of the foreign key.

Tools that read the database schema do see that there is a relationship though, and so LINQ to SQL will still generate a relationship in the mapping file automatically. From the point of view of external tools it is as if there actually is a regular foreign key relationship in place.

So the final design ended up looking like this:

The foreign key shown here is setup with the NOCHECK option and is not enforced by the server.

So now I know why you'd use the NOCHECK option... I don't expect to need it often, but now I know why such a bizarre option exists and have finally, after years of working with databases, had a good reason to use it.

TicketDesk 1.2.3 Released on CodePlex

I've formally packaged a new version of TicketDesk at Codeplex.

TicketDesk 1.2.3 is a minor update. The most significant change is that the HTML editor has been replaced with the markitUp! editor using the markdown syntax.

This should hopefully work around some problems many users were reporting with adding comments in IE8 as well as some display problems that could occur when a user cut/paste content from a word processor or other sources with embedded rich formatting.

TicketDesk 2.0 MVC - alpha demo now available

As I've mentioned before, I'm working on the next major version of TicketDesk 2.0 on the  ASP.NET  MVC framework. The project is still a little early in development, but is starting to resemble a real application now.

I've put demo site up to give the public a preview...

The demo of TicketDesk 2.0 MVC alpha is now online. I'll be updating it from time to time as I reach different milestones. When the project gets closer to a beta state I'll likely check it into source control over at the TicketDesk codeplex project, but for now I'm working offline on it.

Implemented so far:

• TicketCenter with the default list views


• TicketEditor displays ticket information
  
  
  
  • The visual formatting is very rough
  
  
  • Attachments cannot be downloaded (this is on purpose... my demo is not an FTP server for the public's warez  :P)
  
  
• New Ticket feature should be fully functional 


• TicketEditor activity panel should support most activities
  
  
  
  • "edit ticket" and "add attachments" remain incomplete.   
  
  
• Account management is borrowed mostly from the "sample" MVC app, but has been customized for ticketdesk.

Stuff that isn't done:

• Notifications and RSS are not implemented


• Visual Styling and formatting is very "stock MVC sample"  for now
  
  
  
  • I like the general layout, but the text formatting needs lots of work
  
  
• Ticket Search is absent


• None of the admin tools are complete

Go poke at the demo and see what you think.

I welcome any comments, observations, or questions you might have, but don't go reporting bugs yet...this is an alpha demo so I already know it doesn't work very well yet :)

Enjoy...

Fixing markitUp! 1.1.5 - bug in IE8 when closing preview iframe

[UPDATE - 1/12/2010]: MarkItUp! version 1.1.6 inclues a fix for this issue (thanks Jay!)

I've been working with the wonderful markitUp! editor by Jay Salvat. Specifically, I'm using markitUp! as a markdown editor for TicketDesk and a few other apps I'm working on. I'll probably post more about using markitUp! as a markdown editor later, but for now I wanted to address a specific bug that markitUp! exhibits in IE8.

By default, markitUp! uses an iframe element for a preview window.

In IE 8, closing the preview iframe will cause IE 8 to try to close the entire hosting window or tab. IE 8 will prompt the user before it does this, but if you click yes when prompted it will indeed kill the window/tab... which is not good.

After some digging, I've located the problem...

Here is the relevant code with the part that is called when closing the preview window in bold:

// open preview window
// open preview window
function preview() {
 if (!previewWindow || previewWindow.closed) {
     if (options.previewInWindow) {
  previewWindow = window.open('', 'preview', options.previewInWindow);
     } else {
  iFrame = $('');
  if (options.previewPosition == 'after') {
      iFrame.insertAfter(footer);
  } else {
      iFrame.insertBefore(header);
  }
  previewWindow = iFrame[iFrame.length - 1].contentWindow || frame[iFrame.length - 1];
     }
 } else if (altKey === true) {
       if (iFrame) {
   iFrame.remove();
      }
      previewWindow.close();
      previewWindow = iFrame = false;
 }
 if (!options.previewAutoRefresh) {
     refreshPreview();
 }
}

What is supposed to happen is that the code removes the iframe element, then calls the close method on previewWindow variable. That variable would normally have a reference to the content window within the iframe (you can see where that is set in blue in the code excerpt above). So calling close on the variable would normally just try to close that sub-window... or maybe it would do nothing at all because the iframe containing the sub-window would have already been removed. The behavior is internal to the browser and I suspect that specific mechanics are probably a little different from one browser to another. But either way, this works fine on all the browsers I've tested with except IE 8.

With IE8, this code appears to invoke the the close() call on the containing window instead (which is your page's main window in most cases). If you make the close call before the iframe is removed, IE8 will behave like the other browsers do, but when the close call happens after the iframe is removed IE 8 starts asking you if you want to close the whole browser window. for some reason, the contents of the previewWindow variable change after you remove the iframe.

Not a problem! my solution was to simply alter the code to only call the close method when there isn't an iframe being used. That way, close is called for cases where you are using the pop-up window, but doesn't get called when you are using an iframe.

// open preview window
 function preview() {
  if (!previewWindow || previewWindow.closed) {
      if (options.previewInWindow) {
   previewWindow = window.open('', 'preview', options.previewInWindow);
      } else {
   iFrame = $('');
   if (options.previewPosition == 'after') {
       iFrame.insertAfter(footer);
   } else {
       iFrame.insertBefore(header);
   }
   previewWindow = iFrame[iFrame.length - 1].contentWindow || frame[iFrame.length - 1];
      }
  } else if (altKey === true) {
          if (iFrame) {
     iFrame.remove();
        }        else {
     //SMR - else block added here to prevent this call when preview is in iframe
     //      IE8 incorrectly tries to close the hosting window if you call it when using iframe
     previewWindow.close();
        }
          previewWindow = iFrame = false;
  }
  if (!options.previewAutoRefresh) {
      refreshPreview();
  }
}

 This seems to fix the problem in IE 8, while not breaking anything in the other browsers I'm testing with (chrome, firefox, etc). I could have just moved the close call so it happened before the iFrame gets removed (which also seems to work), but I'm a little concerned that closing the window before removing the iframe might have unexpected results in browsers I'm not testing with... but it would probably be fine either way.

If you want, you can download my modified versions of the markitup! source. I have included a standard and minified one both.

Please note that this version also contains my own killPreview() function. By default markitUp! has only one toolbar button for the preview. If you click it, the preview opens and if you ALT + Click it the preview closes. But in my own implementations I prefer to have a separate toolbar button for closing the preview... users don't magically "know" about the ALT+Click trick and I get tired of people reporting "I can't close the preview window" as a bug in the apps that use markitUp!.

TicketDesk 2.0 and the ASP.NET MVC Framework

Now that the ASP.NET MVC Framework is out, I've decided to tackle learning the new platform the same way I usually do... by writing a real application for the new platform.

TicketDesk 1.0 was originally just a playground application to help me get up to speed during the last round of new-tech releases from Microsoft... so it seemed natural to explore the MVC Framework with a re-write of the same application. TicketDesk is just small enough to be workable by a lone part-time programmer, and it is just big enough to provide a decent proving ground for the new technologies.

So let's discuss MVC and how it relates to TicketDesk 2.0...

One of the ironies of my life is that I've been primarily an ASP.NET developer ever since it was first released and I've also been working with MVC and MVC-like development patterns nearly that entire time too.

MVC patterns just makes sense for web apps seeing as the nature of HTTP itself matches that pattern so cleanly. In other environments, MVC has been a formally accepted pattern for years and years.

But ASP.NET Webforms was initially designed to make programming for the web feel more like windows programming with an event driven programming model. Microsoft excels at event driven programming techniques, and the resulting webforms framework was a fantastic adaptation of the pattern into web development space. Webforms allows you to mostly ignore all that messy HTTP stuff and code pages just like you would in a persistent windows environment.

But like most abstractions, webforms tends to break-down when you try to do stuff at the edges. So it wasn't uncommon for platform developers to find problems that just didn't map well to the abstractions provided by webforms. So many of us ended up spending amazing amounts of time hacking into the gap between the webforms model and the raw HTTP pipeline itself.

If you look at the architectures behind most of the larger and more successful ASP.NET application platforms (sharepoint, the 1.x starter kits, IBuySpy, DotNetNuke, CommunityServer, etc.) you will usually find elaborate examples these kinds of hacks. All of them are just variations on a theme... use MVC-like patterns to gain some control over the HTTP request/response pipeline.

With the rise of modern AJAX techniques and technologies, the need for a new approach has become very apparent. Ajax mucks around with the request pipeline in ways that the webforms framework does not tolerate elegantly. If you've tried to do any significant Ajax stuff in webforms, you've probably noticed how quickly things get messy.

Fortunately Microsoft recognized this and decided to formally embrace the MVC pattern. The result is the ASP.NET MVC Framework which was delivered a few months ago.  

Which brings me back to TicketDesk....

I originally built TicketDesk 1.x  as a way to experiment with .NET technologies that were new at the time (Ajax, EF, and LINQ to SQL). So I thought it would be fitting to do the same thing again now to get my hands dirty with the Microsoft MVC Framework.

I didn't port the existing TicketDesk 1.x code though. Instead, I've started with a clean solution and am re-implementing the same set of features as TicketDesk 1.x using all fresh code written for the MVC framework.

I suspected all-along that TicketDesk would probably map very well to the MVC Framework, and I'm no stranger to the MVC design pattern itself. I had also hoped that the MVC design pattern might eliminate many of the obstacles I had encountered especially with the Ajax parts of TicketDesk 1.x.

The experiment is about 3 months old now, and has been very challenging. The MVC Framework itself has a lot of room for improvement, but is a solid foundation on which to start. Some of the most obvious drawbacks are the slim Visual Studio IDE support, sparse documentation, and poor examples of how to do ASP.NET MVC "the right way".

The biggest challenge for me has been the steep learning curve. I've been writing web apps for over 12 years, most of that working with ASP.NET, but the ASP.NET MVC framework really requires an entirely different way of thinking. I'm also just now learning my way around JQuery too which has further slowed me down.

Currently Microsoft is providing only basic Ajax functionality within the MVC framework, but they have encouraged the use of JQuery. JQuery gives you a rich and very successful source for all those fancy UI components that Microsoft doesn't provide on the MVC framework. While the ASP.NET MVC Framework doesn't help you much with JQuery, it also doesn't interfere any.  Future versions of the framework promise to further embrace JQuery head-on. I've not been impressed with Microsoft's own ability to deliver decent Ajax libraries so far, but JQuery has a very large 3rd party community developing high quality code... and most of it is some kind of open source to boot.

While I've found that writing against the MVC Framework takes significantly longer and requires much more effort, the quality and usability of the resulting application is many orders of magnitude better.

So I've formally decided to re-write the official TicketDesk application on the ASP.NET MVC Framework.

The initial 2.0 release will not contain very much new functionality compared to 1.x, but I hope to provide a significantly better user experience and a much more compartmentalized code-base.

Currently TicketDesk 2.0 is targeting the ASP.NET MVC Framework 1.0 on the .NET 3.5 stack. I did experiment with the RTM release of the Entity Framework this time, but I still find that EF is just not ready... so I'll be sticking with LINQ to SQL for a while longer.  I'm confident that I can switch back to EF should the next version resolve my remaining concerns. It is likely that the next version of the MVC Framework will be released before I am done with TicketDesk 2.0, so it will likely shift to target that version before the final release.

Here are some early goals for the TicketDesk 2.0 project:

• Implement 100% of the functionality from TicketDesk 1.x
    
• Upgrade Tools for 1.x to 2.x migration
    
• Improve Application Settings and Administration (more and better online admin tools)
    
• Improve formatting for RSS and Email notifications
    
• Enable full functionality for browsers without JavaScript
    
• Enable smoother Ajax UI features for browsers that do support JavaScript
    
• Use a Markup editor instead of a WYSIWYG HTML editor (too many problems with raw HTML data entry). I'm currently working with MarkItUp! using Markdown syntax.
    
• Unit testing for controllers and business/entity logic (using VS Test Project)
    
• A cleaner separation between the web application and model/business/entity logic
    
• Fully W3C compliant XHTML 1.0 Strict Output
  

I have no real time-frame for a 2.0 delivery as this is still a part-time project for me. Currently I have implemented most of the functionality for the TicketCenter, new ticket creation, and have just started work on the Ticket Viewer/Editor.

I have not marked out a potential 1.3 upgrade of the older code-base either, but my primary focus will be on the 2.0 MVC version.

TicketDesk - Design Philosophies Explained

It has been just over a year since I introduced TicketDesk over at CodePlex. While it hasn't taken the world by storm or anything, it did generate a lot more interest than I would have expected. There are several companies using TicketDesk in production environments, and there have been a few thousand downloads from other people that may be using it too.

While TicketDesk isn't generating the kind of download numbers that I'd want to base a software startup on, for an open source project it is what you might call "wildly successful".

If there was a major failure on my part with bringing TicketDesk to the public, it would be that I didn't do a good job explaining the ideas behind the overall design. So let me take a stab at explaining the philosophy behind TicketDesk.

The general idea behind TicketDesk was to take my 15 years or so of experience, much of it spent being frustrated by help desk issue trackers, and use that experience to design a different kind of help desk system; one that avoids those problems.

And believe me, I have a very long list of complaints with help desk systems!

I suppose the best way to explain it is to discuss the fundamental design idea then illustrate how TicketDesk implements them.

TicketDesk is an issue tracker for help desks... and that is all:

The help desk at most organizations will have many considerations aside from issue tracking. There are internal rank structures, chains of command, political issues, business practices, and financial considerations of all kinds. Unfortunately, the help desk is deeply involved in all of these things.

The mission of TicketDesk is to allow the help desk keep track of issues, and that is all it does.

• TicketDesk does not attempt to understand your org chart.


• It doesn't recognize user rank, status, or departmental affiliations.


• It doesn't act as a time tracker.


• It doesn't do billing.


• It doesn't do project management.


• It doesn't manage your inventory.


• It doesn't handle your business process.


• It doesn't do inner-departmental accounting.


• And it absolutely does NOT care about your internal politics.

TicketDesk is made for internal help desks:

TicketDesk was designed exclusively for use by help desks supporting users within the same organization. It assumes there is a decent level of trust between all participants.

TicketDesk can be used in other environments, and there are plans for future versions to better enable external user scenarios.

You should carefully evaluate TicketDesk's features before attempting to use it in a customer-facing capacity. Also, you may find the features insufficient for organizations performing contracted support for users external to your organization.

Have as few data fields as possible for any given ticket:

This the most basic design ideas behind TicketDesk.

In most help desk systems there are just too many fields, and few of them turn out to be useful. During planning management is hyped-up about the advantages all those fields will bring, but it doesn't take long for the staff to learn that the free-text description field is the only reliable source of information (and even that is a dubious assumption).

So I've spent a lot of time thinking about the various fields common to similar systems.

There are many reasons why different fields fail, but it boils down to just three overall trends:

1. The fields may not relate to the user's specific problem. For example, Questions like what OS are you using aren't useful when the user is reporting a problem with their phone.


2. The end user is incapable of answering some questions. It isn't their fault, they aren't IT professionals so they just don't know the answers, especially to the more technical and detailed questions like "what is your OS version?" or "what is the printer model number?".


3. The end user is not qualified to answer some questions. This isn't a lack of skill, but just a lack of enough information. The classic example here is the priority field, which users cannot provide a meaningful answer. They don't know how their problem stacks up in relation to other issues; only IT can provide a useful answer here.

After exploring the problems I came to the conclusion that these just cannot be solved by software and it is unlikely that training, threat, or corporate policy would help either. The only solution is for the system to expect these problems, embrace them, and concentrate on helping the humans work around them on a case by case basis.

TicketDesk follows important philosophies:

1. Avoid asking any question where the user cannot be reasonably expected to answer with 100% accuracy no matter what kind of problem they are reporting.


2. Avoid asking questions that don't apply to nearly every possible situation being reported.

Thus TicketDesk asks as little from the user as possible. The system expects that the only useful field will be the free-text details field. Other fields do exist, but they are designed to be general in nature, optional, or are answered by the staff rather than the end-user.

Tickets should evolve as a natural conversation between the help desk and the end-user:

As discussed above, TicketDesk does not attempt gather a lot of detailed and quantifiable information up front. Instead it expects that help desk may have to ask for additional information.

Tickets are designed to be an ongoing two-way conversation with the user by borrowing heavily from web 2.0 and social networking concepts.

The activity area of tickets acts as a forum-style discussion board combined with an activity and history log. Every action that can be performed with a ticket solicits additional comments that also become part of the ongoing conversation.

The notifications system (and RSS feeds) ensures that both staff and users remain informed as the ticket progresses to completion. And TicketDesk makes it very simple to perform actions or add comments which encourages the staff to actually make frequent updates as they work through an issue.

The result should be a constant stream of information flowing between the user who submitted the ticket and the help desk staffer assigned to deal with it. Either party, as well as interested 3rd parties, can jump in at any time to add to the conversation.

Avoid Workflow & Routing Hell:

This is one of the more controversial of TicketDesk's design philosophies.

Most help desk systems have customizable and dynamic workflows with rule-based routing. This allows for a lot of control over how a ticket moves through the system.

There is no inherent "problem" with this kind of system in my experience. I have had the misfortune of working with system where the workflow customizations were insanely over-engineered to create horridly inefficient routes with many unnecessary steps, but when used wisely these features don't exactly present a "problem" directly.

Avoiding advanced workflow and routing is a design philosophy based mostly on technical considerations.

Workflow and routing is a nightmare to code, especially for a small development team with limited resources. The advantage of this kind of feature set though is rather limited. Other than making managers happy by having the system act as a policy-cop, there isn't much added value to the feature set.

Additionally, TicketDesk is designed to collect a very minimal set of fields, and doesn't expect end users to necessarily fill them in meaningfully so in TicketDesk there aren't many fields that can participate usefully with advanced workflows.

Instead I designed TicketDesk to use a static state-based workflow that should be valid in just about any organization. While simple, it is also unobtrusive and frictionless for the most part.

There have been some requests for workflow options that require only simple workflow customization options or a limited set of pre-defined optional rules. I plan to explore those ideas for inclusion in future versions of TicketDesk, but I have no plans to introduce a full-featured workflow customization or rule-based routing engine.

Allow organic categorization:

Most issue tracker systems provide the end user several with cascading category lists with context sensitive sub-categories. The options in sub-cats adjust according to previous selections to produce granular categorizations. As described before though, this just doesn't work that well because users don't get these selections right very often or the selections themselves are incomplete or outdated.

By omitting detailed categorization in TicketDesk, the searchability of tickets does become a little degraded and it can be more difficult to locate related tickets.

To give TicketDesk decent searchability without re-producing all the problems of traditional over-categorization; TicketDesk includes a web 2.0 style tagging mechanism. This allows users and staff both to organically add keywords to tickets as they desire.

Anyone can tag, but it is only really successful as a substitute for categorization if the help desk takes it on themselves to ensure that tickets are tagged well before being resolved. This takes some discipline and effort, but the up-side is that it produces a degree of searchability that can far exceed traditional categorization mechanisms. And best of all, there isn't a lot of administrative overhead to tagging since the system evolves and adapts all by itself over time.

Tagging is optional though, and many shops (mine included) choose not to make much good use of it. That's OK as TicketDesk doesn't rely on tagging, and a lack of it doesn't degrade the system's ability to perform the primary mission.

Email Notifications should not spam users:

This is a major problem in a lot of different software systems. There is a need to keep users informed of changes in a timely manner, but if you send notifications too frequently the system will overwhelm users.

When this happens people tend to ignore notifications and the important ones get lost in the noise.

To combat this problem, TicketDesk puts an enormous amount of effort into reducing the number of notifications sent to ensure that notification always conveys useful new information.

Here are the basic rules behind the email system:

• Do not notify users about changes that they have made themselves. You know what you just did right?


• Wait a few minutes before sending a notification to see if additional events involving the same ticket happen. If so, wait until changes slow down a bit, then consolidate the events into a single message.


• Convey all of the information about the ticket in the message so users do not have to log in to see what is going on.


• Attempt to guarantee delivery by supporting an intelligent re-try mechanism.

Depsite the fact that this system took a while to get implemented, it has proven good at keeping down the number of messages sent as well as eliminating unnecessary notifications.

The actual format of the notification message is still a little rough around the edges, but that will be worked out in future releases.

TicketDesk will not provide performance reporting:

This is also a controversial philosophy, but one that is absolutely essential to the success of the system.

TicketDesk will not implement any reports or data collection features assist management in measuring employee performance, or that could be used this way.

Anytime the issue tracker becomes a tool by which management measures employee performance the system ceases to have value. Instead it becomes an enemy of the users. Users will manipulating the data in the system to protect themselves and inflate their performance numbers. Anything that would make them "look bad" will be deliberately obscured or omitted from the system.

Researchers call this "management dysfunction", and it is a well established and thoroughly vetted reality. Despite that though, managers around the world still insist on attempting to automate the measurement of employee performance... which is ironic. If they were successful what would be the point in having managers on staff?

Your help desk is probably staffed by very smart people. People that love figuring things out and whose job is to be very good at figuring things out. How long will take them to learn how to game the system?

Even if you have some honest staffers that don't manipulate the system... it will punish those honest users while rewarding users that do manipulate the data to their advantage.

The purpose of TicketDesk is to facilitate honest and open communication between users and help desk. If the system is used to gather performance metrics then it cannot provide honesty nor openness and fails the primary mission.

To complete the failure, any performance metrics you "thought" the system was gathering turn out to be inaccurate and distorted, resulting in a system that can't measuring actual performance nor perform the other tasks it is designed for.

I first learned about this issue from Joel Spolsky, creator of the popular FogBugz bug tracking system, but have witnessed this same phenomenon in nearly every help desk environment I've ever worked with. .

You can read Joel's take on the issue yourself if you wish, he explains it better than I can.

Now... there are ways to do useful reporting in a way that doesn't lead to management dysfunction. But it takes very careful design where you deliberately create reports that cannot be used to show individual or group performance metrics. That is a slippery slope, and I have not yet had time to do the design for such reports yet.

I do have plans to add some reporting in the future, but the reporting will be carefully designed to prevent such abuses.

ASP.NET MVC - After RedirectToAction call, the target action fails to render a partial view correctly

Another undocumented "feature" I ran into when playing with the MVC.

I have an ajax action link that sorts a list on the page. To accomplish this, the link performs an ajax request to the controller's Sort action and in the end redraws the part of the page contianing the list's data with the new sort settings.

No biggie... except that when using Mozilla Firefox, it never worked right. The list would be redrawn, but would always contain the entire page's content including the menus, headers, and all the other stuff.

This was another one of those things I had hoped would magically go away with the RTM of the MVC framework... but when it didn't I had to go figure it out.

The process here was a little more complex than what you'd see in the examples and tutorial apps, but not by much. Here is how it worked.

The controller has two actions: List and Sort

The List action reads the user's profile and gets data based on the user's sort and filter preferences. It will render a partial view containing just the list's data if the request is an ajax request, otherwise it renders the entire page view.

The Sort action simply updates the user's profile with their new sort preferences. Then it uses RedirectToAction to tell the browser to call the controller's List action again.... and the list action would do the actual list building using the updated settings.

Simple enough... 

After some debugging though, I found that some browsers don't resend the necessary headers on a subsequent ajax request when redirected this way. So the list action would think the request was a non-ajax request and it would thus render the whole page view.

There was a lot of discussion about some late beta changes to the MVC framework's IsAjaxRequest() method, so I had hoped it was just a beta bug... but apparently not.

In order to get this to work reliably, I had to have the sort method add a setting to TempData if it was called from ajax, then have the list method check both the TempData setting AND the IsAjaxRequest() method.

Annoying, but at least there is a workaround.