ALNERNETIVE

Standing to Code

Nathan Allen-Wagner — Tue, 20 Mar 2012 19:12:00 GMT

Today I tried a Standing Desk for the day. I've heard about this a number of times, mostly on Twitter. The idea's not new, but it is gaining attention due to the health concerns around so much sitting.

I started standing at 7am. It's currently 3pm, and I'm still standing here. So far so good. My muscles are a little tired, but my mind feels more awake than most other days!

Another thing I'm noticing is that it's a lot easier to switch gears. If I need to step over for a cup of coffee, I'm already standing. No big deal. And getting back to coding is just as easy. Just walk up and get to it. No need to try and find that comfortable position in the chair, scoot it up to the desk, etc.

As for my desk… Well, it's the same desk I usually use… plus some boxes and a board to elevate my laptop to standing height.

Cost = $0. Adding years to my life and energy to my day… priceless.

(I'll report back soon to let you know how things are going after a week of standing)

Cheers!

SqlTransaction Ain't Always Transactional

Nathan Allen-Wagner — Sun, 04 Mar 2012 03:03:00 GMT

Q: "SQL run using a SqlTransaction is part of a SQL Server Transaction and can be rolled back."

A) True
B) False

The answer is of course "A", right? That's what I thought too.

Basic SqlTransaction Usage

Most of us assume that we can write this:

using (var con = new SqlConnection(".."))
{
con.Open();

using (var tran = con.BeginTransaction())
{
    var cmd = con.CreateCommand();
    cmd.CommandText = "Update dbo.Customer ...";
    cmd.Transaction = tran;

    try
    {
      cmd.ExecuteNonQuery();
      tran.Commit();
    }
    catch
    {
      tran.Rollback();
    }
}
}

This is plain and simple SqlTransaction based logic. Exec the command. If it completes without error, call Commit(). If it has an issue, then call Rollback();

But ... What Does This Do?

I challenge you to predict the behavior of the following "Retry" scenario. The code may seem a bit odd… Why would you execute more SQL after an exception? But it shows what one might do if the code implemented Retry semantics inside a transaction.

For retries, one might want to open a transaction, run a set of statements, and then retry any statements that had failed. The SQL transaction should remain in force in spite of any errors thrown by SQL server. For example, a foreign key violation does not automatically abort the transaction. Rather, the code would catch the FK violation and queue it for subsequent execution after completing the 1st pass.

This does in fact work… most of the time. I say most of the time, because deadlock errors in SQL Server cause some interesting behavior.

Here's a snippet that should easily run in LinqPad without too much work. The only pre-requisite is to create a DB in SqlExpress named "Scratch".

Pay close attention to lines 144-145. How will they behave after the deadlock? What is their impact on the database?

Shocked, Stupefied, and Generally Stumped

If you run the above code and look at the data that ends up in [dbo].[customer2], you'll notice a very strange thing:

The transactions were rolled back, but one of the "C" statement was still applied!

WHY???

Short Answer…

The Deadlock aborts the underlying transaction on SQL Server, but SqlTransaction and SqlCommand happily continue to execute more commands outside of any transaction.

Hindsight is 20/20

After diagnosing and fixing this issue, it almost makes sense, but it's not at all intuitive up front. I asked a couple colleagues to guess the outcome of the code, and both guessed wrong. I suspect that most of you will also be surprised by the results. Hindsight is 20/20, but up front, this one is nonsense.

Deadlocks in SQL server mean that SQL will kill one of the transactions and roll it back. This frees the locks and allows the other transaction to proceed. In the scenario above, we force a deadlock, meaning that one of the transactions is ROLLED BACK on the server. It's not just "zombied", which would cause "SELECT XACT_STATE()" to return "-1". No, in this case, the transaction is done. There is no more transaction.

So, how does one of those "C" statements get applied? Basically, SqlCommand doesn't know that there is a problem with the transaction on the server, nor does it ask the SqlTransaction object to check. Instead, it just sends the statement to the server, assuming that the server still has a transaction open on the current connection. But there is no transaction. As a result, those statements are executed with no transaction at all. The subsequent Rollback() has no means to undo their work.

Is This a Bug?

Calling something a bug is rather severe. In this case, I suspect that the framework can do better.

The system does some checks to ensure proper & safe usage of the SqlTransaction, but not enough. After a deadlock, the client still has a reference to a SqlTransaction object. This object still thinks that it has a transaction on the server. It will even allow you to call Rollback() on this transaction object after the deadlock. But it won't let you call Commit(). It seems that SqlTransaction knows that the transaction can't commit, and throws an error back at you. Additionally, if you use a SqlConnection with a SqlCommand, but don't pass the associated SqlTransaction, then it will throw an error.

So the system does some internal checks, but it doesn't check for the deadlock scenario. Well, I think it should do better. Before executing a command, it should check whether the associated transaction is gone. If the server transaction is gone, it shouldn't even run the command. It should just error. Otherwise, SqlCommand will continue to run the statements outside of any transaction.

I don't know if this is possible, or if it has some sever side-effects, but the current behavior of SqlTransaction is a very dangerous and sharp edge on the API surface of the .NET framework. A transaction that doesn't actually roll back the associated commands is not good.

Solution: Catch Deadlock Errors and Halt

The solution seems to require catching the SqlException, inspecting the Number to see if it's a deadlock, and then taking an alternate course of action. In the case of retry logic, the processing should immediately stop and rollback to prevent any subsequent commands from executing outside a transaction.

To catch a deadlock, look for SqlException.Number == 1205. This indicates a deadlock occurred.

Happy (Transaction) Coding!

Cheers!

Updated version of Monitored Undo Framework released

Nathan Allen-Wagner — Tue, 31 Jan 2012 15:24:00 GMT

Today I released an updated version of the Monitored Undo Framework.

Changes include:

A new parameter on the DefaultChangeFactory's methods for specifying the "description of the change". This can be helpful in cases where the UI shows a list of the undo / redo changes.
A new WPF sample that shows, in a simpler codebase, how to use the framework.
Updated build script that compiles all solutions, run the tests, and package the NuGet package.

These changes are also available via NuGet.

Cheers!

"tfpt unshelve /migrate" Doesn't Account for Renames

Nathan Allen-Wagner — Thu, 26 Jan 2012 21:14:00 GMT

A quick note about Team Foundation Power Tools (TFPT)'s "Unshelve" command…

tfpt unshelve has a /MIGRATE switch that will allow unshelving changes into a different branch / location than the one where they were originally shelved. I won't re-iterate the many articles describing how to use this command.

One issue that I ran into is unshelving a set of changes into a branch where some of the shelved files had been renamed / moved. It seems that the /migrate switch doesn't go back into TFS version history to locate the renames. Instead, it simply pulls out the changes and re-applies them to the new location.

For Example:

Let's say that I start with Branch A. One user (John Doe) get's latest and starts making changes.

Branch A
Src
File1.txt

Then… a different user (Jane Smith) renames File1.txt and checks that in.

Branch A
Src
File1_RENAMED.txt

Then… (for some reason), they need to move John's changes from Branch A to another branch. (Let's pretend that Branch A is the main branch, but the changes are too large for this iteration, so they want to move them to a separate branch for a future release.) So they create Branch B, which looks like this:

Branch B
Src
File1_RENAMED.txt

Then… John shelves his changes and they attempt to use tfpt unshelve /migrate to move the changes to Branch B.

But a strange thing happens. They end up with this:

Branch B
    Src
        File1.txt
        File1_RENAMED.txt

Huh? How'd that happen?

Someone who knows version control well could probably explain why, and suggest that this is obvious. But it wasn't to me, and as a result, some of the changes in the shelveset did not make it into the target solution.

How I should have done it…

If I was going to do this again, I think I'd:

Get the source and the target branches to the same version. (Merge changes as needed between the branches.)
In the source branch, get latest version so that the local workspace has the latest version.

This should propagate any pending renames / moves that were made in that source branch, or it's parent branches.

Shelve the changes that should be moved.
In the target branch, get latest version.
Run tfpt unshelve /migrate …

Cheers!

A Visual Studio Extesion to make Scrolling Better!

Nathan Allen-Wagner — Tue, 24 Jan 2012 03:37:54 GMT

Hi All,

A friend of mine (John Nastase) saw a tweet or two of mine complaining about scrolling in Visual Studio.

What did he do? He wrote a VS extension to make it happen!!!

Here's the extension that he just published in the VS extension gallery.

It basically adds a "cursor buffer" space to the top and bottom of the window so that as arrow down the page, the page starts to scrolling before the cursor hits the VERY BOTTOM OF THE PAGE. Instead, it starts scrolling a little earlier so that you can see what's coming. This saves you from scrolling too far and then having to arrow back up to where you want to be.

Brilliant!

Cheers!

EF Inserts Failing Because of Missing Association in SSDL

Nathan Allen-Wagner — Thu, 19 Jan 2012 21:37:00 GMT

I just found that Entity Framework (4.0) can stumble on One-To-One relationships if the EDMX's schema storage layer (SSDL) is missing the association that represents the Foreign Key in the database. I suppose this makes sense, but it's only obvious to me in hindsight.

I don't know how my EDMX ended up missing the SSDL association for the Foreign Key, but it was missing. However the the Conceptual Layer had the relationship, so the diagram made it look like it was associated. And because this is a supported scenario, it didn't complain at me that I was missing anything.

But…

When I went to save the records, EF stumbled on the insert. In my case, I had two entities that were related by a "1 to 0..1" relationship. I'm guessing that EF saw this, evaluated the SSDL layer's metadata and decided that order of insert didn't matter. As a result, it inserted them in alphabetical order… and failed. It failed because in my DB, I did have a foreign key.

The fix was to manually add the association into the SSDL layer.

Note: Editing EDMX files by hand SUCKS!

Microsoft's MVP program and Rob Eisenberg

Nathan Allen-Wagner — Thu, 05 Jan 2012 19:34:00 GMT

This is a long read, but it's got some detailed insight on the MVP program at Microsoft. I've heard these sentiments before, and I'm sure that it's not black and white. Regardless, Microsoft's actions in this case are unacceptable.

http://devlicio.us/blogs/rob_eisenberg/archive/2012/01/04/how-i-lost-regained-and-then-turned-down-an-mvp-award.aspx

If you don't know Rob Eisenberg, he's the creator of Caliburn and Caliburn Micro frameworks for XAML based platforms (WPF, Silverlight, WP7). If you want to have your mind blown, take a look at his source code for Caliburn Micro (and even more so for Caliburn). While you're at it, watch his "Build your own MVVM framework" presentation. I reviewed a lot of WPF frameworks, but Caliburn Micro stands alone in it's elegance, simplicity, and amazing use of the C# / .NET platform.

I seriously hope that MS addresses whatever is behind this story. We need people like Rob! They should be rewarded for their amazing contributions to the community.

Good Times at the .NET SIG…

Nathan Allen-Wagner — Wed, 14 Dec 2011 13:15:00 GMT

Had a great time at the .NET SIG last night where we talked about Software Craftsmanship and did Katas together. We had a good turn out (~40 attendees), and we spent a majority of the time doing Katas.

I was quite impressed by the group's willingness to try Pair Programming. Here's pic that Andrew Thorne snapped of one side of the room. You can see people working on a kata, mostly in pairs!

https://twitter.com/#!/Bennett_Adelson/status/146744206772281345/photo/1/large

Thanks to those of you who came out. Hope you enjoyed the time and got something out of it.

Cheers!

Signing EXEs and MSIs with Signtool via TFS Builds

Nathan Allen-Wagner — Fri, 18 Nov 2011 20:23:00 GMT

I won't repeat the "Signing your code with Signtool" walkthrough that is already detailed in 50,000 other posts. You can Bing that with Google as easily as I did. What I will share is a little issue that tripped me up for a bit.

I wanted the code signed as part of the TFS automated build. And, while I could run the command line myself, I could not get it to work under TFS builds.

The Scenario

I wanted the keys in the certificate store, rather than on disk. This makes it possible to use the certificate without storing the certificate's private key password on disk.
The build process is NOT running as local admin.
The code signing tasks happen as a "Post-Build" event, specified in the project's properties.
The certificate was installed in the Machine Store, not a specific account's store.

The Problem

I could run the Signtool command under my account and it worked. But if I ran under a limited account, it didn't work.

I tried copying the cert into the service account's personal store, but that didn't help.

The Solution - Permissions, of course!

It wasn't immediately obvious because of various error messages such as:

"No certificates were found that met all the given criteria"
"No provider was specified for the store or object"
… and others

I figured that I'd check to see what Signtool was looking for to see if I could isolate the issue. ProcessMonitor did the trick in a matter of minutes. Spinning it up, I found that it was getting access denied on the Machine Store's key files. I've seen this before related to IIS / ASP.NET (I think).

On Windows Server 2003 and earlier, the machine keys are stored under:
C:\Documents and Settings\All Users\Application Data\Microsoft\Crypto\RSA\MachineKeys

On Windows Server 2008, these are now under:
C:\ProgramData\Microsoft\Crypto\RSA\MachineKeys

I found the file related to the cert that I just installed and inspected permissions. Just "Administrators" and "System". I granted my build service account read access and, VOILA, the Signtool works!

If anyone knows how to manage cert permissions without mucking with the file system directly, please do tell! And if you know why this is the way it is, I'd love to hear it.

Cheers!

LINQ to Excel = Bliss!

Nathan Allen-Wagner — Wed, 09 Nov 2011 04:28:00 GMT

It's fun when an API is so awesome that it just makes you smile uncontrollably. That's how I felt after using Linq To Excel!

http://code.google.com/p/linqtoexcel/

I dreaded the thought of coding against Excel and decided to search around for the best approach. StackOverflow served up a couple options, including LINQ to Excel. Now I like LINQ, so it naturally attracted my eye. The project is completely awesome! It let me extract data from multiple sheets in an excel workbook with just a few lines of code, and no direct dependency on Excel to be installed on the machine!

Awesome!

So, if you need to pull data out of Excel, look no further. You can even pull the package into your project via NuGet!

Cheers!