Short version: here is the link to the uploaded .NET regular expression. (look in the comment for the updated version)

I noticed that the javascript code that I am using to parse PGN chess games and display it is rather slow and I wanted to create my own PGN parser, one that would be optimal in speed. "It should be easy", I thought, as I was imagining getting the BNF syntax for PGN, copy pasting it into a parser generator and effortlessly getting the Javascript parser that would spit out all secrets of the game. It wasn't easy.

First of all, the BNF notation for Portable Game Notation was not complete. Sure, text was used to explain the left overs, but there was no real information about it in any of the "official" PGN pages or Wikipedia. Software and chess related FTPs and websites seemed to be terrible obsolete or missing altogether.

Then there was the parser generator. Wikipedia tells me that ANTLR is pretty good, as it can spew Javascript code on the other end. I downloaded it (a .jar Java file - ugh!), ran it, pasted BNF into it... got a generic error. 10 minutes later I was learning that ANTLR does not support BNF, but only its own notation. Searches for tools that would do the conversion automatically led me to smartass RTFM people who explained how easy it is to do it manually. Maybe they should have done for me, then.

After all this (and many fruitless searches on Google) I decided to use regular expressions. After all, it might make a lot of sense to have a parser in a language like C#, but the difference in speed between a Javascript implementation and a native regular expression should be pretty large, no matter how much they optimize the engine. Ok, let's define the rules of a PGN file then.

In a PGN file, a game always starts with some tags, explaining what the event is, who played, when, etc. The format of a tag is [name "value"]. There are PGN files that do not have this marker, but then there wouldn't be more than one game inside. The regular expression for a tag is: (\[\s*(?<tagName>\w+)\s*"(?<tagValue>[^"]*)"\s*\]\s*)+. Don't be scared, it only means some empty space maybe, then a word, some empty space again, then a quoted string that does not contain quotes, then some empty space again, all in square brackets and maybe followed by more empty space, all of this appearing at least once.

So far so good, now comes the list of moves. The simplest possible move looks like 1. e4, so a move number and a move. But there are more things that can be added to a move. For starters, the move for black could be following next (1. e4 e5) or a bit after, maybe if there are commentaries or variations for the move of the white player (1... e5). The move itself has a variety of possible forms:

• e4 - pawn moved to e4
• Nf3 - knight moved to f3
• Qxe5 - queen captured on e5
• R6xf6 - the rook on the 6 rank captured on f6
• Raa8 - The rook on file a moved to a8
• Ka1xc2 - the knight at a1 captured on c2
• f8=Q - pawn moved to f8 and promoted to queen
• dxe8=B - pawn on the d file captured on e8 and promoted to bishop

There is more information about the moves. If you give check, you must end it with a + sign, if you mate you end with #, if the move is weird, special, very good, very bad, you can end it with stuff like !!, !?, ?, !, etc which are the PGN version of WTF?!. And if that is not enough, there are some numbers called NAG which are supposed to represent a numeric, language independent, status code. Also, the letters that represent the pieces are not language independent, so a French PGN might look completely different from an English one. So let's attempt a regular expression for the move only. I will not implement NAG or other pieces for non-English languages: (?:[PNBRQK]?[a-h]?[1-8]?x?[a-h][1-8](?:\=[PNBRQK])?|O(-?O){1,2})[\+#]?(\s*[\!\?]+)?). I know, scary. But it means a letter in the list PNBRQK, one for each possible type of chess piece, which may appear or it may not, then a letter between a and h, which would represent a file, then a number between 1 and 8 which would represent a rank. Both letter and number might not appear, since they represent hints on where the piece that moved was coming from. Then there is a possible letter x, indicating a capture, then, finally, the destination coordinates for the move. There follows an equal sign and another piece, in case of promotion. An astute reader might observe that this also matches a rook that promotes to something else, for example. This is not completely strict. If this long expression is not matched, maybe something that looks like OO, O-O, OOO or O-O-O could be matched, representing the two possible types of castling, when a rook and a king move at the same time around each other, provided neither had not moved yet. And to top it off, we allow for some empty space and the characters ! and ? in order to let chess annotators express their feelings.

It's not over yet. PGN notation allows for commentaries, which are bits of text inside curly brackets {what an incredibly bad move!} and also variations. The variations show possible outcomes from the main branch. They are lists of moves that are enclosed in round brackets. The branches can be multiple and they can branch themselves! Now, this is a problem, as regular expressions are not recursive. But we only need to match variations and then reparse them in code when found. So, let's attempt a regular expression. It is getting quite big already, so let's add some tokens that can represent already discussed bits. I will use a @ sign to enclose the tokens. Here we go:

• @tags@ - we will use this as a marker for one or more tags
• @move@ - we will use this as a marker for the complicated move syntax explained above
• (?<moveNumber>\d+)(?<moveMarker>\.|\.{3})\s*(?<moveValue>@move@)(?:\s*(?<moveValue2>@move@))?\s* - the move number, 1 or 3 dots, some empty space, then a move. It can be followed directly by another move, for black. Lets call this @line@
• (?:\{(?<varComment>[^\}]*?)\}\s*)? - this is a comment match, something enclosed in curly brackets; we'll call it @comment@
• (?:@line@@variations@@comment@)* - wow, so simple! Multiple lines, each maybe followed by variations and a comment. This would be a @list@ of moves.
• (?<endMarker>1\-?0|0\-?1|1/2\-?1/2|\*)?\s* - this is the end marker of a game. It should be there, but in some cases it is not. It shows the final score or an unfinished match. We'll call it @ender@
• (?<pgnGame>\s*@tags@@list@@ender@) - The final tokenised regular expression, containing an entire PGN game.

But it is not over yet. Remember @variations@ ? We did not define it and with good reason. A good approximation would be (?:\((?<variation>.*)\)\s*)*, which defines something enclosed in parenthesis. But it would not work well. Regular expressions are greedy by default, so it would just get the first round bracket and everything till the last found in the file! Using the non greedy marker ? would not work either, as the match will stop after the first closing bracket inside a variation. Comments might contain parenthesis characters as well.

The only solution is to better match a variation so that some sort of syntax checking is being performed. We know that a variation contains a list of moves, so we can use that, by defining @variations@ as (?:\((?<variation>@list@)\)\s*)*. @list@ already contains @variations@, though, so we can do this a number of times, to the maximum supported branch depth, then replace the final variation with the generic "everything goes" approximation from above. When we read the results of the match, we just take the variation matches and reparse them with the list subexpression, programatically, and check extra syntax features, like the number of moves being subsequent.

It is no wonder that at the Regular Expressions Library site there was no expression for PGN. I made the effort to upload it, maybe other people refine it and make it even better. Here is the link to the uploaded regular expression. The complete regular expression is here:

(?<pgnGame>\s*(?:\[\s*(?<tagName>\w+)\s*"(?<tagValue>[^"]*)"\s*\]\s*)+(?:(?<moveNumber>\d+)(?<moveMarker>\.|\.{3})\s*(?<moveValue>(?:[PNBRQK]?[a-h]?[1-8]?x?[a-h][1-8](?:\=[PNBRQK])?|O(-?O){1,2})[\+#]?(\s*[\!\?]+)?)(?:\s*(?<moveValue2>(?:[PNBRQK]?[a-h]?[1-8]?x?[a-h][1-8](?:\=[PNBRQK])?|O(-?O){1,2})[\+#]?(\s*[\!\?]+)?))?\s*(?:\(\s*(?<variation>(?:(?<varMoveNumber>\d+)(?<varMoveMarker>\.|\.{3})\s*(?<varMoveValue>(?:[PNBRQK]?[a-h]?[1-8]?x?[a-h][1-8](?:\=[PNBRQK])?|O(-?O){1,2})[\+#]?(\s*[\!\?]+)?)(?:\s*(?<varMoveValue2>(?:[PNBRQK]?[a-h]?[1-8]?x?[a-h][1-8](?:\=[PNBRQK])?|O(-?O){1,2})[\+#]?(\s*[\!\?]+)?))?\s*(?:\((?<varVariation>.*)\)\s*)?(?:\{(?<varComment>[^\}]*?)\}\s*)?)*)\s*\)\s*)*(?:\{(?<comment>[^\}]*?)\}\s*)?)*(?<endMarker>1\-?0|0\-?1|1/2\-?1/2|\*)?\s*)

Note: the flavour of the regular expression above is .Net. Javascript does not support named tags, the things between the angle brackets, so if you want to make it work for js, remove ?<name> constructs from it.

Now to work on the actual javascript (ouch!)

Update: I took my glorious regular expression and used it in a javascript code only to find out that groups in Javascript do not act like collections of found items, but only the last match. In other words, if you match 'abc' with (.)* (match as many characters in a row, and capture each character in part) you will get an array that contains 'abc' as the first item and 'c' as the second. That's insane!

Update: As per Matty's suggestion, I've added the less used RxQ move syntax (I do have a hunch that it is not complete, for example stuff like RxN2, RxNa or RxNa2 might also be accepted, but they are not implemented in the regex). I also removed the need for at least one PGN tag. To avoid false positives you might still want to use the + versus the * notation after the tagName/tagValue construct. The final version is here:

(?<pgnGame>\s*(?:\[\s*(?<tagName>\w+)\s*"(?<tagValue>[^"]*)"\s*\]\s*)*(?:(?<moveNumber>\d+)(?<moveMarker>\.|\.{3})\s*(?<moveValue>(?:[PNBRQK]?[a-h]?[1-8]?x?(?:[a-h][1-8]|[NBRQK])(?:\=[PNBRQK])?|O(-?O){1,2})[\+#]?(\s*[\!\?]+)?)(?:\s*(?<moveValue2>(?:[PNBRQK]?[a-h]?[1-8]?x?(?:[a-h][1-8]|[NBRQK])(?:\=[PNBRQK])?|O(-?O){1,2})[\+#]?(\s*[\!\?]+)?))?\s*(?:\(\s*(?<variation>(?:(?<varMoveNumber>\d+)(?<varMoveMarker>\.|\.{3})\s*(?<varMoveValue>(?:[PNBRQK]?[a-h]?[1-8]?x?(?:[a-h][1-8]|[NBRQK])(?:\=[PNBRQK])?|O(-?O){1,2})[\+#]?(\s*[\!\?]+)?)(?:\s*(?<varMoveValue2>(?:[PNBRQK]?[a-h]?[1-8]?x?(?:[a-h][1-8]|[NBRQK])(?:\=[PNBRQK])?|O(-?O){1,2})[\+#]?(\s*[\!\?]+)?))?\s*(?:\((?<varVariation>.*)\)\s*)?(?:\{(?<varComment>[^\}]*?)\}\s*)?)*)\s*\)\s*)*(?:\{(?<comment>[^\}]*?)\}\s*)?)*(?<endMarker>1\-?0|0\-?1|1/2\-?1/2|\*)?\s*)

The Regexlib version has also been updated in a comment (I don't know how - or if it is possible - to edit the original).

I was debugging an application when I've noticed that there was an exception thrown in one of the legacy DAL modules. A method was using regular ADO.Net to run a stored procedure, fill a DataSet, then proceeded on getting an IDataReader for the set, using the CreateDataReader method. On reader.Read() the exception DataTableReader is invalid for current DataTable 'Table' was thrown.



I've investigated the issue only to notice that the DataSet was correctly retrieved from the database, the only problem came when creating the reader. Any meaningful property threw this error. I've found a thread that discussed this problem here, but the answer was not there. Instead, I read an obscure line somewhere that said this exception is thrown when the DataSet has changes and tried that solution: before running CreateDataReader, I ran DataSet.AcceptChanges. And it worked!



The strange part comes now: I did the AcceptChanges bit in the Watch window during debug, not as a permanent change to the code. From then on, the code worked, no matter how many times I ran iisreset or restarted the browser. I've added the solution in the code for good measure, but I am still not sure if this is the solution or simply some fluke of the universe. One possible answer is the "race condition" described in this discussion, which also suggests this happens in debug mode only. Strange, innit?



Update: AcceptChanges did not solve the issue on the production server. I am still investigating, but if you know what this is about, please share :)

I needed to pass an array of IDs to a stored procedure on SQL Server 2008. This version of the server supports user defined table types and a way to access it from .Net, of course. A comprehensive resource for sending arrays to any version of SQL Server can be found here.



Long story short, for 2008 you first define a user table type that has a single int column (we are talking about an array of integers here, obviously), then a stored procedure that takes a parameter of that type. A way to send the array from .Net code is detailed here. As you can see, you create an array of something called SqlMetaData, holding the information of each column as defined in the user defined type, then you use an SqlParameter of SqlDbType Structured and with the TypeName the name of the user defined table in SQL Server. The parameter will have a list of SqlDataRecord instances that have the integer values in their first columns. Yes, there is an even longer story and I consider this short :-P



All nice and easy, but there is a caveat, something that is not immediately obvious from the code. The column metadata is set as a property value for any of the records that are added to the sql parameter value list. What if the list is empty? In this case it appears that there is a bug somewhere. The stored procedure fails, I guess because it does not receive the structure of the user defined table declared in the metadata and cannot map it to the user defined type.



A solution for this is to add a dummy SqlDataRecord with no values and then, in the stored procedure, check for NULL. A very ugly solution. The solution on Erland Sommarskog's blog did not say anything about this specifically, but I did find this: There are a few peculiarities, though. This does not work:

EXEC get_product_names NULL

but results in this error message:

Operand type clash: void type is incompatible with integer_list_tbltype

It is quite logical when you think of it: NULL is a scalar value, and not a table value. But what do you think about this:

EXEC get_product_names

You may expect this to result in an error about missing parameters, but instead this runs and produces an empty result set!. Therefore the solution I used was to check in code if the .Net list of integers was empty and, in that case, do not send a parameter to the stored procedure. And it worked.

I've met a very interesting WPF bug today, something that is hard to explain or reproduce, but might give terrible headaches if you don't know its source.

I had a WPF UserControl, with its xaml and cs files. Now, I know that in MVVM I shouldn't really use those much, but it was out of my control. The control had a section of resources (UserControl.Resources) in which there was a ResourceDictionary with some stuff in it. Considering that I'd removed all the merged dictionaries from this, I thought that I had no need of the dictionary tags, after all the Resources property of an element is already a ResourceDictionary. So it was something like this:

<UserControl ... >
  <UserControl.Resources>
    <!-- <ResourceDictionary> with these tags commented the error occurs -->
      ... stuff ...
    <!-- </ResourceDictionary> -->
  </UserControl.Resources>
</UserControl>

The error itself is that, during compilation, the partial user control class defines in the code behind doesn't seem to find things from the xaml. Probably, the compiler fails building the xaml into a class, but fails silently, while the codebehind is completely disconnected from the xaml because it is the only partial file for that class name.

By selectively removing items in the resources I've narrowed it down to one of the converters. It was creating using the MarkupExtension trick, but it was also declared as a resource for some reason. I do not see why that should matter, but still.

Bottom line: when the partial codebehind class for a WPF user control (or maybe for windows as well) fails to connect to the xaml, it means it silently fails the compilation of the XAML and you should try checking the resources of the elements therein.

I've finally finished reading Pro ASP.Net MVC Framework by Steven Sanderson. The book is slightly dated, since it discusses the technology used in Visual Studio 2008 and without any mention of the new Razor engine, but these are details that are not important to the content of the book anyway. I can say that it is a very nice book and it was worth reading, especially the first part.

There are two parts to this, the first being a TDD ASP.Net MVC web shop application built step by step and explained line by line. It goes through some Domain Driven Design concepts as well, it does unit testing and mocking, even shows off a little dependency injection via Castle Windsor. What I liked most, though, is how painstakingly thorough Sanderson was explaining every single detail. He didn't assume anything as he documented every step of the way, down to what lambda expressions are and what .Net features he was using.

The second part of the book is a little less readable, as it goes through the classes and features of ASP.Net MVC, complete with methods, properties and small samples. I highly recommend reading this part while actually experimenting with the framework on the computer. Even if you do not, this part of the book remains a very valuable reference for when you do. In this section of the book you can learn about data entry, Ajax and partial updates, application security and deployment, even how to mix classic ASP.Net with MVC, though not really recommended.

The bottom line is that Pro ASP.Net MVC Framework is a must read for a developer learning ASP.Net MVC. There is an updated version of the book for VS2010 and .Net 4 that I think that I will also read (the book was so good). Here is the link for Pro ASP.NET MVC 2 Framework.

A colleague of mine started using a control I made in which there was a Hyperlink. Well, the purpose of it was not complex, I just needed a text that can be clicked. A Hyperlink sounded like the only solution out of the box, since there is no LinkButton control in the standard WPF controls. That aside, after I finish writing this blog post, I will write myself a LinkButton control to use in these situations instead, since Hyperlink seems to have several design flaws. It's not even a control, but a flow element.

What my colleague reported is that the Hyperlink would not get enabled in certain situations and we've come to the conclusion that after the initialization of the control, the IsEnabled property on the Hyperlink does not change when an ancestor control changes its enabledness. The only way to force it is to actually bind it to an ancestor IsEnabled.

Here is the scenario: You place several items in a XAML file. They are a text in a Hyperlink, in a TextBlock (since Hyperlinks cannot be directly part of a Panel, first design flaw), in a StackPanel. The text in the text block will appear as a clickable link. Set the StackPanel to IsEnabled="False" and the text will appear as disabled. Now, add a ToggleButton and bind its IsChecked property to the StackPanel IsEnabled property. Click the button and the StackPanel will get disabled, but not the hyperlink. Start with a disabled StackPanel, the link will be disabled, click the button and the hyperlink will stay disabled. The solution: set on the Hyperlink, inline or via a style, IsEnabled="{Binding IsEnabled,RelativeSource={RelativeSource AncestorType={x:FrameworkElement}}}". Now that is ugly.

As a sideline, whenever you see a WPF element inexplicably disabled and you use Snoop on it and try to set IsEnabled to true and you can't, there is probably one of two situations:

1. A parent of the control is disabled
2. The control is implementing ICommandSource and its Command property is set on an ICommand that returns false on its CanExecute method

Well, I have been kind of absent from the blog lately and that is for several reasons. One is that I have been waiting for some news that would determine my direction as a professional developer. The other is that I have re-acquired a passion for chess. So, between work at the office, watching chess videos, playing chess with my PDA and watching all seven seasons of Star Trek Deep Space 9, I haven't had much time for blogging.

Also, when you think about it, the last period of my programming life has been in some sort of a limbo: switched from ASP.Net to WPF, then to ASP.Net again (while being promised it would be temporary), then back to WPF (but in a mere executive position). Meanwhile, Microsoft didn't do much to help me, and thus saw their profits plummet. Well, maybe it was a coincidence, but what if it wasn't?

I am complaining about Microsoft because I was so sold into the whole WPF/Silverlight concept, while I was totally getting fed up with web work. Yet WPF is slow, with no clear development pathway when using it, while Silverlight is essentially something else, supported by only a few platforms, and I haven't even gotten around to use it yet. And now the Internet Explorer 9/Windows 8 duo come in force placing Javascript and HTML5 in the forefront again. Check out this cool ArsTechnica blog post about Microsoft's (re)new(ed) direction.

All of this, plus the mysterious news I have been waiting for that I won't detail (don't want to jinx it :-S), but which could throw me back into the web world, plus the insanity with the mobile everything that has only one common point: web. Add to it the not too enthusiastic reaction of my blog readers when starting talking about WPF. So the world either wants web or I just have been spouting one stupid thing after another and blew my readers away.

All these shining signs pointing me towards web development also say that I should be relearning web dev with ASP.Net MVC, getting serious about Javascript, relearning HTML in its 5th incarnation and finally making some sense of CSS. Exciting and crazy at the same time. Am I getting too old for this shit or am I ready for the challenge? We'll just have to see, won't we?

This is a case of a bug fix that I made work, but I can't understand why the solution works. Basically, the story was that some internal component of a third part control forced WPF to throw an exception on the UI thread. As it was impossible to plug the hole in the third party library, and since its next version is supposed to solve the issue, I've opted for a somewhat ugly hack: I've handled the DispatcherUnhandledException event of the Application class and I've basically said to it to ignore that specific UI error.

I will get into details of what the error and where it came from was and how to handle it, but I want to focus on the fact that, since this was a fix for a specific class, I've inherited from that class and used a static method in it to do the above handling of the event. Well, it worked most of the cases, but not all. Some code that involved moving the focus of WPF elements programmatically would cause the bug to reappear.

At first I thought it was a matter of a change in the policy of exception handling from .Net 1.0 to 2.0 and above. So I've set the <legacyUnhandledExceptionPolicy enabled="1"/> option in the app.config runtime section, but it didn't help.

I've tried everything, from using the control instance Dispatcher in the constructor or in the Loaded event, to moving the code directly to the point after the application was instantiated and before the application was run. Bingo, it worked! I thought that was it. I've again encapsulated the entire behavior in the inherited control and ... watched it fail.

Let me simplify the situation: static code that doesn't work when encapsulated in a static class works perfectly when the same code is inlined in the calling code. Can you explain that? I cannot!

The code is simple:

    application.DispatcherUnhandledException +=
      application_DispatcherUnhandledException;

    static void application_DispatcherUnhandledException
           (object sender, DispatcherUnhandledExceptionEventArgs e)
    {
      if (e.Exception.Message.Contains("Hover") 
          && e.Exception.Message
                 .Contains("System.Windows.Controls.ControlTemplate"))
      {
        e.Handled = true;
      }
    }

Move that in a static class and execute it as MyClass.RegisterFix(application); and it doesn't catch all the exceptions thrown, although it works in most cases.

Can anyone explain this to me? Why does it matter where the code is?

As I was blogging before, RedGate are assholes. They bought Reflector, promised to keep it free, then asked money for it. But every crisis can be turned into an opportunity. JetBrains promised a free decompiler tool and they have kept their word as they have released an early build. A total news to me, but not really a surprise, other software companies decided to build their own version in order to boost their visibility in the developer world. Telerik, for example, has just released JustDecompile, beta version.

It is no secret that JetBrains is a company that I respect a lot, as they made ReSharper, the coolest tool I've ever had the pleasure to work with, but I will try to be as unbiased as possible in comparing the options. I have tried dotPeek on WPF's PresentationFramework.dll from the .NET framework 4.0, as I often need to check the sources in order to understand functionality or bugs.

As a footnote, Reflector, just before it went commercial, could not decompile some of the code there. Not only it did not decompile it, but it presented empty methods like that was all there was in the code, with no warnings or errors or explicative comments. So, even if free, I bet Reflector would have sucked in the end, after getting into the money grabbing hands of RedGate.

dotPeek has seen and decompiled the code that Reflector did not. Also, I have to say that similar functionality like in ReSharper, finding usages, going to declaration, etc are making dotPeek a very nice tool to work with. What I did not quite like is that it doesn't have yet the functionality to save the sources to text files. But I am sure this is just a detail that was not implemented yet. Hopefully, they will provide a rich plugin model like old Reflector did.

Unfortunately, to download JustDecompile, they need you to have a Telerik login in order to download, which, as everyone knows, simply sucks. No one likes a registration form, folks! Especially one that presents you with wonderful prechecked checkboxes for permission for Telerik to send you all kind of stupid promotions and newsletters. Also, the download is of a .msi file. Most developers like to see what they are installing and preferably just copy it from a file archive. Running the .msi took forever, including the mandatory 100% CPU utilization bit that I will never understand in installation products. (coming from the .NET runtime optimization service, mscorsvw, called by ngen) But that's just the delivery system. Let's check out the actual thing.

JustDecompile starts reasonably fast and it also has a nice look, being build with Telerik controls and what not. The decompilation is a bit weird at first, since it shows only the method names and for a second there I thought it was as bad as Reflector was, but then I noticed the Expand All Members button. The context menu is not nearly as useful as dotPeek's, but there are a lot of options in the top toolbar and the navigation via links is fast and intuitive. It also has no text saving options yet.

As the decompiled sources were, I noticed these differences:

• JustDecompiler places inline member declarations in constructors, dotPeek shows it inline. It might not seem an important thing, but an internal class gains a weird public constructor in order to place the declaration there, instead of using the only internal constructor that the class had. It looks strange too as its last line is base(); which is not even legal.
• dotPeek seems to want to cast everything in the source code. For example List list = (List) null;
• JustDecompiler shows a Dictionary TryGetValue method with a ref parameter, dotPeek shows the correct out.
• dotPeek creates really simple names for local scope variables like listand list1, JustDecompiler seems to create more meaningful names like attachedAnnotations
• JustDecompiler shows a class internal as dotPeek shows it as internal abstract.
• JustDecompiler seems to fail to decompile correctly indexer access.
• JustDecompiler doesn't seem to handle explicit interface implementations.
• JustDecompiler doesn't seem to decompile readonly fields.
• JustDecompiler transforms a piece of code into an if with a return in it and then some other code, dotPeek decompiles it into an if/else.
• JustDecompiler doesn't seem to handle Unicode characters. dotPeek correctly encodes them in source like "\x001B".
• dotPeek seems to join nested ifs in a single one, as opposed to JustDecompiler.
• JustDecompiler uses base. in order to access members coming from base classes, while dotPeek uses this.

I will stop here. I am sure there are many other differences. My conclusion is that dotPeek could do with the naming algorithm JustDecompiler seems to use for local scope variables, but in most other ways is superior to JustDecompiler for now. As both programs are in beta, this could quickly change. I do hope that healthy competition between these two products (and, why not, shady developer meetings in bars over tons of beer and pizza, in order to compare ideas intercompanies) will result in great products. My only wish is that one of these products would become open source, but as both use proprietary bits from commercial products, I doubt it will happen.

Have fun, devs!

Update 23 Feb 2012:
Spurred by a comment from Telerik, I again tried a (quick and dirty, mind you) comparison of the two .Net decompilation tools: JetBrains dotPeek and Telerik JustDecompile. Here are my impressions:

First of all, the Telerik tool has a really cute installer. I am certainly annoyed with the default Windows one and its weird error codes and inexplicable crashes. Now, that doesn't mean the Telerik installer does better in the error section, since I had none, but how could it not? The problem with the installation of JustDecompile is that it also tried to install (option checkboxes set by default) JustCode and JustTrace. The checkboxes themselves were something really custom, graphically, so I almost let them checked, since they looked as part of the background picture. If it weren't for my brain spam detector which went all red lights and alarm bells when seeing a really beautiful installer for a free tool, I might have installed the two applications.

Now for the decompilation itself. I was trying to see what the VisualBasic Strings.FormatNumber method contained. The results:

• dotPeek showed xml documentation comments, JustDecompile did not
• dotPeek showed default values for method parameters, JustDecompile did not
• JustDecompile could decompile the source in C#, VB and IL, dotPeek did only C#
• JustDecompile showed the source closer to the original source (I can say this because it also shows VB, which is probably the language in which Microsoft.VisualBasic was written), dotPeek shows an equivalent source, but heavily optimized, with things like ternary operators, inversions of if blocks and even removals of else sections if the if block can directly return from the method
• There are some decorative attributes that dotPeek shows, while JustDecompile does not (like MethodImplAttribute)
• dotPeek has a tabbed interface that allows the opening of more than a single file, JustDecompile has only a code view window
• dotPeek shows the code of a class in a window, in order to see a method, it scrolls to where the method is; JustDecompile shows a class as a stub, one needs to click on a method to see the implementation of only that method in the code window

My conclusion remains that dotPeek is a lot more usable than JustDecompile. As a Resharper user, I can accept that I am biased, but one of the major functionalities of a .Net decompiler is to show you usable code. While I can take individual methods or properties with JustDecompile and paste them in my code, I can take entire classes with dotPeek, which makes me choose dotPeek for the moment, no matter all the other points above. Of course, if any of the two tools would give me a button that would allow me to take a dll and see it as a Visual Studio project, it would quickly rise to the top of my choices.

Update 26 Apr 2013:
I've again compared the two .Net decompilers. JustDecompile 1.404.2 versus dotPeek EAP 1.1.1.511. You might ask why I am comparing with the Early Access Program version. It is because JustDecompile now has the option to export the assembly to a Visual Studio project (yay!), but dotPeek only has this in the EAP version so far.
I have this to report:
Telerik's JustDecompile:

• the installer is just as cute as before, only it is for a suite called DevCraft, of which one of the products is JustDecompile
• something that seemed a bit careless is the "trial" keyword appearing in both download page and installer. If installing just JD, it is not trial
• again the checkboxes for JustCode and JustTrace are checked by default, but at least they are more visible in the list of products in the suite
• a Help Improve the Telerik Installer Privacy Policy checkbox checked by default appeared and it is not that visible
• the same need to have an account to Telerik in order to download JD
• when installing JD, it also installs the Telerik Control Panel a single place to download and manage Telerik products, which is not obvious from the installer
• the install takes about two minutes on my computer to a total size of 31MB, including the control panel
• if a class is in a multipart namespace like Net.Dns, it uses folders named Net.Dns if there is no class in the Net namespace
• not everything goes smoothly, sometimes the decompiler throws exceptions that are then logged in the code as comments, with the request to mail to JustDecompilePublicFeedback@telerik.com
• it creates the AssemblyInfo.cs file in a Properties folder, just like when creating a project
• resolves string concatenation with string.Concat, rather than using the '+' operator as in the original code
• resolves foreach loops into while(true) loops with breaks when a condition is met
• uses private static methods in a class with the qualified class name
• resolves inline variables, leaving the code readable
• overall it has a nicer decompiled code structure than dotPeek
• adds explicit default constructors to classes
• places generic class constraint at the end of constraints list, generating an exception
• it doesn't catch all reference assemblies, sometimes you have to manually add them to the list
• decompiles enum values to integer in method optional parameters default values, generating compilation errors
• decompiles default(T) to null in method optional parameters default values, generating compilation errors
• decompiles class destructors to Finalize methods which are not valid, generating compilation errors
• types of parameters in calls to base constructors are sometimes wrong
• places calls to base/this constructors at the end of constructor code blocks, which of course does not work, when you place more complex code in the calls
• doesn't understand cast to ValueType (which is somewhat obscure, I agree)
• really fucks up expressions trees like FluentNHibernate mapping classes, but I hate NHibernate anyway
• resolves if blocks with return in them to goto/label sometimes
• resolves readonly fields instantiated from a constructor to a mess that uses a local variable to set the field (which is not valid)
• doesn't resolve corectly a class name if it conflicts with the name of a local method or field
• inlines constants (although I don't think they can solve this)
• switch/case statements on Enum values sometimes gain weird extra case blocks
• sometimes it uses safe casting with value types (x as bool)

JetBrain's dotPeek:

• the EAP version has a standalone executable version which doesn't need installation
• the whole install is really fast and installs around 46MB
• as I said above, it does not have the Export to Project option until version 1.1
• the decompilation process is slower than JustDecompile's
• if a class is in a multipart namespace like Net.Dns, it uses a folder structure like Net/Dns
• sometimes things don't go well and it marks this with // ISSUE: comments, describing the problem. Note: these are not code exceptions, but issues with the decompiled code
• it inlines a lot of local variables, making the code more compact and less readable
• weird casting of items in string concatenations
• a tendency to strong typed casting, making the code less readable and generating compilation errors at times
• the AssemblyInfo.cs file is not created in a Properties folder
• when there are more classes in a single file, it creates a file for each, named as the original file, but prefixed with a number, instead of using the name of the class
• it has an option to create the solution for the project as well
• it creates types for anonymous types, and creates files with weird names for them, which are not really valid, screwing the project.
• it has problems with base constructor calls and constructor inheritance
• it has problems with out parameters, it makes a complete mess of them
• tries to create a type for Linq IQueryable results, badly
• it has problems with class names that are the same as names of namespaces (this is an issue of ReSharper as well, when it doesn't present the option to choose between a class name and a namespace name)
• resolves while(method) to invalid for loops
• it doesn't resolve corectly a class name if it conflicts with the name of a local method or field
• problems with explicit interface implementations: ISomething a=new Something(); a.Method(); (it declares a as Something, not ISomething)
• problems with decompiling linq method chains
• I found a situation where it resolved Decimal.op_Increment(d) for 1+d
• indirectly used assemblies are not added to the reference list
• it sometimes creates weird local variables like local_0, which are not declared, so not valid
• adds a weird [assembly: Extension] in the AssemblyInfo file, which is not valid
• a lot of messed up bool values resolved as (object) (bool) (value ? 1 : 0), which doesn't even work
• inlines constants (although I don't think they can solve this)
• __Null local1 = null; - really?

After decompiling, solving the issues and compiling again an assembly in the project I am working on I got these sizes:
JustDecompile: 409088
dotPeek: 395776
The original: 396288

Of course, this is not really a scientific comparison between the two. I was excited by the implementation of Export to Project in both products and I focused mainly on that. The navigation between types and methods is vastly improved in JustDecompile and, to my chagrin, I have to admit that it may be easier and safer to use than dotPeek at this time. Good job, Telerik! Oh, and no, they have NOT paid me to do this research :-)

As usual when I stumble into a problem that I can't solve only to see that it has a simple explanation and that I am not sufficiently informed on the matter, I had some misgivings on blogging about it. But these are the best possible blog entries, the "Oh, I am so stupid!" moments, because other people are sure to make the same mistake and you wouldn't wish for them the same humiliation, would you? Well, I wouldn't :-P

So here it is: I was having a ToggleButton and a ContextMenu in a custom control. I wanted to have the IsChecked property bound in two-way mode with the IsOpen property of the ContextMenu. So I did what most people would do, I created a Setter on the IsChecked property with a Binding on DropDown.IsOpen as a value (DropDown is the property of the control holding the ContextMenu). And it worked, of course. My custom control would inherit from ToggleButton and use the style with the Setter in it.

But now comes the tricky part: I wanted that when a certain condition was met, the button be checked and the menu open. So I added a Trigger to the Style on the condition that I wanted, with a Setter on the IsChecked property to True. And from then on, nothing made any sense. I would click the button and it would not open the menu anymore.

Well, if you think about it, you have a Setter in the Style and then another Setter on the same property in the Trigger. It makes a sort of a sense for them to interfere with each other, but I also know that setting a Binding as value to a DependencyProperty is like using SetBinding on the owner of the property. And I thought it was normal for the IsChecked property to be set to true from the trigger and, in turn, change the value of IsOpen. But it didn't happen. Let it be a lesson to other bozos like myself that this thing does make sense logically (or as wishful thinking), but not in WPF. And here is why:

Let's try to evaluate the values of IsChecked and IsOpen. First case scenario: clicking on the button. That changes the status of IsChecked from true to false and viceversa. Internally, what does happen is WPF finds any bindings associated with the value and refreshes them. In this case, it finds a binding to IsOpen and so the ContextMenu also opens up. Second case scenario: the condition in the ViewModel is met, the trigger is fired and the value of IsChecked is set. It should do the same thing, right? Find the bindings and refresh them. And it does! But in that fateful moment, it sees that the IsChecked property has a setter associated with it, from the trigger in the style, that sets it to True. It does not go further, because the trigger setter overrides the style setter. I personally think this is closer to a bug than a reasonable behavior, but I am biased here :) I mean, you set the value of the property directly in the code IsChecked=true;, and it works, but you set it in the trigger and it overrides the binding?

There are more solutions to this problem. One solution would be to replace the setter value in the trigger with another binding also to the IsOpen property of the ContextMenu, but with a nifty converter. I haven't tried it, because I think that, if it worked, it would add too much weird complexity and even if I love weird, this is a little bit too much for me. Of course, a programmatic solution is also possible, adding handlers for the Open and Close events of the ContextMenu and setting IsChecked, as well as setting IsOpen based on the value sent to the IsChecked property change callback. But I wanted to do something that is as WPFish as possible. Another solution is to set a binding to the IsOpen property, since it is two way, and this is what I did. Unfortunately, the ContextMenu is a variable of the control, so I had to manually set the binding in that property's PropertyChangedCallback. A more elegant solution, I believe, is to have another element present that would have two-way bindings for both IsChecked and IsOpen properties to two of its own properties. Internally, when one changes, the other is synchronized. This leaves both ToggleButton and ContextMenu free to have any setters on IsChecked and IsOpen without interference.

The pattern of using a separate control to link properties from other controls together is called Binding Hub. Here is an article detailing it. I disagree with the naming of properties as I believe for each connection a separate hub should be created with properties that actually make sense :), but the concept is powerful and I like it.

Now what else could I have been? I have been doing this damn job for at least 6 years, depending on what you call programming and what not, and even in the worst moments of depression and disappointment I still can't imagine myself doing anything else.

The point is that I have taken the 70-518 exam: Designing and Developing Windows Applications Using Microsoft .NET Framework 4 with 950 points of 1000. The same old story: pointless questions with some ridiculous answers and focus on technologies that nobody has really heard about like Microsoft Sync Framework or using DTOs or planning testing strategies (you only have to know how they are named, though).

Even worse, after getting almost all the questions and answers from the web in the form of a helpful .vce file, I started researching them on MSDN. I really wanted to know exactly what each question meant and why were those the correct answers. I really expected to find tons of documentation from Microsoft about all of this, but no! Not only did I not find what I was looking for most of the time, but when I did it was either obsolete information, meaning there were from 2007 or having specific message on the pages that the information is no longer valid, or formatted in a weird and unfriendly way that was not conducive to any type of training or learning.

Go ahead, try finding documentation on Microsoft Sync Framework, a Hands on Lab or anything that can be read from top to bottom and give you the ability to use that technology. Nada! There is no book for the MCPD exam, not even one that is going to be published in the future. Sure, I can download anything, install it, try it out, google for error messages and fix things up when they don't work, but I would have done that anyway!

I leave this whole experience behind with a vague sense of disappointment. I've learned a lot about WCF and I am going to read Julie Lerman's Entity Framework book now, but none of these four exams did really push me to learn anything new or (something that I feel I deserved from Microsoft) to find the best ways of doing things. It is all a jumbled, bureaucratic mess.

A shameful affair, this. I barely got 820 of 1000 for the Microsoft TS exam 70-516: Accessing Data with Microsoft .NET Framework 4. I could invoke my birthday and some other circumstances to motivate my low result, but let's face it: I could have done a lot better.

What I can tell you about the exam is that most of the questions regard Entity Framework, but also Linq to SQL, classic ADO.Net, WCF Data Services. There are a few questions regarding security and the Microsoft Sync Framework thrown for good measure, as well. Also, there is a .vce file on the net from 2010-07-29, by Carina, and, even if there is some overlap, those are mostly NOT the questions in the exam, while the .vce from 2011-12-09 by acarum is a fake.

As preparation materials I've used Julie Lerman's excellent book Programming Entity Framework Second Edition, which is a thorough and well written book, but kind of huge! Try to allocate at least a month to read through it. I'd really intended to make a post with MSDN resources for the exam, but I didn't have the time.

Onwards, comrades! The MCPD 70-518 exam awaits!

I was doing some tests with localization in a WPF application and I've found that the way to change dynamically the language for a control (and all of its descendants) is to set the Language property. In my application at least, the propagation of change in the entire subtree took a few seconds. As I found out, the way to refresh the language is to actually remove and readd children to control trees. This is probably why simply changing the Thread culture does not work, what the Language property is for and explains why the error was thrown.

During the Language change on the Window object a validation from one of the custom controls there threw a weird Exception: Cannot modify the logical children for this node at this time because a tree walk is in progress.. What could it mean?

This link on StackOverflow says something vague about a bug in the Silverlight charting toolkit, but I think that is wrong, or at least not the root of the problem. I looked through the Reflectored .Net sources and found that this exception is thrown in AddLogicalChild and RemoveLogicalChild methods in FrameworkElement and FrameworkContentElement when the internal property IsLogicalChildrenIterationInProgress is true. Apparently, this property is set to true whenever am internal class DescendentsWalker<T> is "walking" the visual and/or the logical trees. What is this class?

DescendentsWalker<T> seems to be a class used by the WPF framework to find resources, resolve property values, find ancestors/descendants in the control trees, propagate events, etc, so it is a very important low level class. Either because of a bug or exactly in order to prevent one, an error is thrown if someone is trying to change the structure of the tree while something is walking the tree. That is what the error represents.

Now, the only question is how to solve it. The validation code was trying to change the content property of a control, thus modifying the control tree. As such, I do believe this is a slight bug in the framework, as changing the tree could just wait until the tree walk is over, but still. A simple try/catch solve it locally, but what should I do to prevent such errors in the future for all of my code? I can't pad my code with try/catch blocks.

I've tried using Dispatcher.Invoke for the change, it didn't work. The IsLogicalChildrenIterationInProgress property itself does a this.ReadInternalFlag(System.Windows.InternalFlags.IsLogicalChildrenIterationInProgress); and _flags is obviously a private field of the FrameworkElement or FrameworkContentElement class that is never encapsulated into a public property. So I can't check for it in a simple way. Of course, one can always catch Dispatcher.UnhandledException and just set e.Handled to true if this error occurs (Oops, a minor worldwide tremor caused by developer shudder!), but I don't see another global solution.

Well, at least I've thoroughly examined the problem and its causes and possible solutions, however bad or annoying they may be. If you know of a more elegant solution for it, please share it. As this is the only time I ever got this error, I believe it is something that can be handled locally, but only if one knows how to test for it. So, change the Language of your windows or do other lengthy tree walks in order to protect your application from this bug.

Well, I wanted to convey something like the scene where Mr. Fancy Pants gets the icecream, but still, I passed the MCTS 70-511 exam: Windows Application Development. The grade is a shameful 982, because of a question regarding a DataGridView and a DataSet with related tables (as well as a microwave oven and non-dairy creamer), but it is still a decent pass.

Even if I knew most of the stuff already, some materials for study were: the awful MCTS Self-Paced Training Kit for 70-511 and a post from James J. Foster's blog.

Today I went and certified myself as a Windows Communication Foundation developer in .Net 4.0 (exam 70-513). I passed with 1000/1000, which makes me "strong" in all areas of the exam. It was difficult, too, as there is no official preparation material for the exam. You have the old 3.5 preparation book for the 70-503 exam and then you have to guess about the things that are in 4.0. Basically, you can read the book, then study a little about the router services, service discovery and simplified configuration. I found a nice post on James J. Foster's blog detailing the material that needs study, with links. Also, here is a link on MSDN about what is new in WCF 4.0.

The problem with the exam is that I had found the questions (the real ones) on the web. You can download them and simulate exams until you remember all of them and their answers. These are called brain dumps (.vce format) and apparently it is the norm rather than the exception; Cisco exams are just the same. It was too weird to be true and, frankly, the whole point of taking these exams was to actually learn something in a structured way, so I was both hoping that those were the real questions and also that they were not and I will have the opportunity to test myself on what I have learned. Alas, it was not the case. I knew all the questions already and this has soiled the entire experience and grade.

Actually, when you think about it, it makes sense. Microsoft needs to show how serious they are about the whole test thing and ignore the elephant in the room altogether, pretending the answers are not out there. Meanwhile, the companies that organize the tests need those answers public, otherwise nobody would dare come take them and thus pay for them. I mean, really... 50 questions of increasing obscureness and a minimum passing grade of 800/1000 ? Who would even consider testing themselves against sets of four questions where one of them would throw a compiler error, one of them would be completely off topic and the other two were competing on which one gets the most hard to remember, one google-second away configuration setting that would make your brain explode. The ugliest thing is that people learning this stuff normally have little to no chance of getting every question right, even if in the official preparation books you find stuff like "Exam tip: beware questions in this format, as they will always be incorrect, but we will ask them anyway so we can trap you if you didn't read our material".

So, bottom line. I am proud of myself for actually studying for this. I feel I know a lot more about WCF and the powerful ways in which I can make it work for me. But the fact that any kid can download a file, do a dozen test runs on their home computer and then take the test and achieve the same perfect score soils it a little.

Well, this makes me a MCTS. I have 3 more exams to take and then I will become a MCPD. It's like achievements in games, only they are worth something.