I no longer work at Microsoft, so please don't bother leaving a comment here or trying to contact me through my MSDN blog.
You can find my new blog at http://www.technologytoolbox.com/blog/jjameson. My new site also provides copies of all posts from my MSDN blog.
If you are using gated check-ins, be sure to also read my follow-up post:Bypassing a Gated Check-in in TFS 2010http://blogs.msdn.com/b/jjameson/archive/2010/12/03/bypassing-a-gated-check-in-in-tfs-2010.aspx
Earlier this year, I wrote a post that explains the process I use for incrementing the assembly version with each build in Team Foundation Server. However, the process was originally developed for TFS 2005 and as you probably know by now, the build process in TFS 2010 has changed significantly.
In TFS 2005 and 2008, the build process was defined entirely in MSBuild (i.e. TFSBuild.proj), whereas in TFS 2010 the bulk of the build process is based on Windows Workflow Foundation. TFS 2010 uses the workflow defined in DefaultTemplate.xaml for new projects created in TFS 2010, but uses the UpgradeTemplate.xaml workflow for TFS projects upgraded from a previous version.
The process I explained before generally works with TFS 2010 when using the UpgradeTemplate.xaml workflow (since that workflow is essentially just a wrapper around the legacy TFSBuild.proj file). In fact, I've been using that process (with a few tweaks) for several TFS projects since upgrading the "Jameson Datacenter" to TFS 2010 last May.
Nevertheless, I've been wanting to dive into the "standard" TFS 2010 build process workflow (i.e. DefaultTemplate.xaml) and determine a better way (going forward) to automatically increment the assembly version with each build.
Note that depending on your specific requirements, there may already be a solution out there for incrementing the assembly version with each build.
For example, Jim Lamb wrote a post back in February that describes how to create a custom workflow activity to achieve build numbers like "2009.11.18.1" -- and if this versioning scheme works for you, then great (you can stop reading this post and go read Jim's post instead)! Personally, I'm not a fan of date-based versioning schemes -- primarily because an assembly version like 2009.11.18.1 tells me nothing more than when the build was compiled. In other words, is this particular version a major release, a minor release, or perhaps just a patch? Looking just at the version number, I have no idea.
Also, as noted in Jim's post, John Robbins wrote a post about synchronizing the TFS build number with the assembly version. John's approach uses a versioning scheme that is similar -- if not identical -- to the one used by the Visual Studio team, in which the Major Version and Minor Version are fixed (e.g. 10.0), the Build Number is computed automatically from the current date, and the Revision is incremented for each build on a particular day.
While I like John's approach to assembly versioning (especially since it doesn't require anything to be installed on the build server), I still prefer to be able to control all four portions of the assembly version. Maybe it's just because I'm a "control freak" or perhaps it's simply because I've been doing it a certain way for the last ten years and it's hard to teach this old dog new tricks ;-)
As I've mentioned before, I increment the Build Number portion of the assembly version with each build on the "Main" branch, whereas I increment the Revision portion of the assembly version with each build on the "QFE" branch (for patches).
If this seems a little fuzzy, consider the project that I've been working on for a little over a year now. Our first build (off the Main branch) was 18.104.22.168. The second build was 22.214.171.124, and the third build was 126.96.36.199. By the time we were nearing the end of Sprint-3, we were up to build 188.8.131.52. However shortly before the Sprint-3 release to Production (or shortly after -- it's been so long that I honestly can't remember which) we needed to fix a couple of issues. Consequently, we checked in those code changes on the QFE branch. Looking at the Builds list that I maintain on the SharePoint team site, it appears that 184.108.40.206 was the last build for Sprint-3.
While I can't tell which iteration or milestone build 220.127.116.11 corresponds to (simply by looking at the assembly version), I can tell that it was a QFE/hotfix for the 18.104.22.168 release. From this, I can infer that the number of code changes between the 22.214.171.124 build and the 126.96.36.199 build are minimal. Contrast this with assembly versions where the Build Number portion is based on a date and the Revision is incremented based on the number of builds performed on that particular day. I can certainly see advantages and disadvantages to each approach.
Let's suppose that, like me, you want to keep using the assembly versioning scheme that I've used in the past. Here is how I recommend you increment the assembly version in TFS 2010. [Note: While I certainly hope you take the time to understand the changes made to the default workflow in order to increment the assembly version with each build, if you simply want to jump straight to the solution, refer to the attachment to this post.]
First, it is important to have a high-level understanding of the default workflow used in TFS 2010.
This "collapsed" view of the workflow illustrates the following high-level steps of the build process:
For the purposes of this post, the most interesting aspect of the build process is the separation of the "Update Drop Location" activities from the "Run On Agent" activities. Let's take a quick look at the details of "Update Drop Location":
As you can see, the portion of the workflow that updates the build number does not run on the build agent. In other words, it happens before the "Run On Agent" scope and therefore runs on the build controller -- not the build agent (although, in the case of the Jameson Datacenter -- and, I suspect, most TFS environments -- there is a single build agent running on the build controller).
Consequently, we are going to have to make some significant changes to the workflow if we want to increment the assembly version using a similar process to the one previously used for TFS 2005/2008. To understand why, let's review the high-level steps that I use for specifying the assembly version:
Note that you don't have to use the custom Version task from the MSBuild Community Tasks Project to increment the version number. If you'd rather write a custom workflow activity (á la Jim Lamb's post that I referred to earlier), go right ahead. In my case, (a) I've already installed the custom MSBuild Community Tasks on my build server, and (b) I know how to use the Version task and I know that it does what I need it to do. Yes, this does require that I run a little bit of "old school" MSBuild script inside my "shiny" new Team Build workflow, but I certainly don't have any objections with that. [The goal is to get all of the "goodness" of the new workflow-based build process, while still leveraging the ability to increment/control each of the four parts of the assembly version.]
With that review out of the way, let's turn our attention to modifying the workflow to automatically increment the assembly version. For the purposes of this walkthrough, I'll be using a brand new team TFS project that I chose to name foobar2010. [Hopefully you can come up with a better name for your TFS project ;-) ]
First, start by branching $/foobar2010/BuildProcessTemplates/DefaultTemplate.xaml to $/foobar2010/BuildProcessTemplates/CustomTemplate.xaml. [Whenever possible, we want to isolate our customizations, and personally I prefer Jim's recommendation of branching the OOTB workflow file rather than simply creating a copy (as recommended on MSDN) -- even if I never intend to merge the changes into DefaultTemplate.xaml.]
The next step is to move a couple of the existing workflow activities out of the Update Drop Location sequence into a new sequence named Update Build Number, as shown below.
Next, move Update Build Number and Update Drop Location inside the Run On Agent scope, as shown below:
Note that we need to first initialize the workspace (and hence be running on the build agent) before updating the build number so we can access the various assembly version files (e.g. AssemblyVersionInfo.txt and AssemblyVersionInfo.cs). Also note that we want to ensure that the label applied to the source code in TFS matches the build number. Consequently, the Update Build Number sequence is placed before the activity that labels the source code.
In the out-of-the-box workflow, the LabelName variable is scoped to Run On Agent (which is okay provided the build number is set prior to entering the Run On Agent scope). Since we are now updating the build number inside Run On Agent, we need to change the scope of the LabelName variable to ensure the label matches the updated build number.
To change the scope of the LabelName variable:
In addition to the source code label, we want the drop location on the Release Server to match the build number (e.g. \\dazzler\Builds\foobar2010\188.8.131.52). Therefore the Update Drop Location sequence needs to come after the Update Build Number sequence. [Whether the Update Drop Location sequence comes before or after the activity that labels the source code really doesn't matter. To me, it simply "feels better" to label the source code as early as possible during the build process.]
By default, the drop location is set to:
BuildDetail.DropLocationRoot + "\" + BuildDetail.BuildDefinition.Name + "\" + BuildDetail.BuildNumber
Personally, I don't really care which build definition was used to create a particular build. The truth is I can infer this from the build number. If the build number is something like 184.108.40.206, then the build was created by the "daily build" (i.e. a build definition named "Automated Build - Main"). If the build number is something like 220.127.116.11, then the build was created by a "QFE build" (e.g. a build definition named "QFE Build - v1.0").
More importantly, I want to make it as easy as possible for the Test and Release Management folks to find a specific build when deploying the solution. Consequently, remove the "extraneous" folder by updating the Set Drop Location activity (inside the sequence within If DropBuild And Build Reason is Triggered) so the DropLocation is set to:
BuildDetail.DropLocationRoot + "\" + BuildDetail.BuildNumber
Next, set the build number using the assembly version specified in the AssemblyVersionInfo.txt file. To do this, add a new InvokeProcess activity at the beginning of the Update Build Number for Triggered Builds activity in the Update Build Number sequence, and set the properties as follows:
All I'm doing here is using a little command-prompt "trickery" to read the contents of a file (using the type command). The file contains a single line of text that specifies the assembly version (e.g. "18.104.22.168" -- without the quotes). As a result, the assembly version is subsequently available using the stdOutput variable of the InvokeProcess activity.
Move the existing Update Build Number activity inside the InvokeProcess activity (below the stdOutput variable box) and change the BuilderNumberFormat property to stdOutput.
While I certainly don't expect any errors to occur with the InvokeProcess activity, it's still a good idea to ensure proper error handling in our build process. Therefore, add a Throw activity (below the errOutput variable box) and set the Exception property to:
Thus if any error happens to occur while reading the assembly version from the specified file, our build will immediately fail.
At this point, we've managed to set BuildDetail.BuildNumber (to whatever is specified in the AssemblyVersionInfo.txt file -- e.g. 22.214.171.124). Consequently, if we check-in the changes to CustomTemplate.xaml at this point, create a build definition using this new process template, and subsequently run a build, everything should work as expected.
However, what would happen if we started another build? Since we haven't yet implemented the pieces to actually increment the assembly version, the second build would fail (because, thankfully, TFS doesn't allow two builds to specify the same build number).
Let's modify the workflow to increment the assembly version...
Just below the InvokeProcess activity added earlier (inside the Update Build Number for Triggered Builds activity), add a new MSBuild activity, and set the properties as follows:
Next, create the actual MSBuild file to increment the assembly version (IncrementAssemblyVersion.proj):
<?xml version="1.0" encoding="utf-8"?>
<Project xmlns="http://schemas.microsoft.com/developer/msbuild/2003" ToolsVersion="3.5"
<Import Project="$(MSBuildExtensionsPath)\Microsoft\VisualStudio\TeamBuild\Microsoft.TeamFoundation.Build.targets" />
Text="Checking out version files from source control..." />
Command="$(TeamFoundationVersionControlTool) checkout AssemblyVersionInfo.txt AssemblyVersionInfo.cs"/>
Text="Incrementing the assembly version..." />
<Output TaskParameter="Major" PropertyName="Major" />
<Output TaskParameter="Minor" PropertyName="Minor" />
<Output TaskParameter="Build" PropertyName="Build" />
<Output TaskParameter="Revision" PropertyName="Revision" />
<Output TaskParameter="Value" PropertyName="IncrementedAssemblyVersion" />
Text="Updating version file ($(BuildProjectFolderPath)\AssemblyVersionInfo.cs) with incremented assembly version ($(IncrementedAssemblyVersion))..." />
Text="Checking in version files to source control..." />
Command="$(TeamFoundationVersionControlTool) checkin /override:"Check-in from automated build" /comment:"Increment assembly version ($(IncrementedAssemblyVersion)) $(NoCICheckinComment)" AssemblyVersionInfo.txt AssemblyVersionInfo.cs"/>
If you are familiar with the MSBuild customizations described in my earlier post, then this should seem very straightforward. In fact, to create this file, I simply copied one of my previous TFSBuild.proj files and started removing the parts that are no longer needed (because they are now implemented in the workflow). Hence the "3.5" version of this MSBuild file. If you want to update it to use MSBuild 4.0, go right ahead. Personally, I didn't see the need -- nor did I want to "tempt fate" since I know the MSBuild Community Tasks work as expected with the older version of MSBuild (specifically the Version task).
Be aware that the MSBuild file shown above is for the Main branch. For QFE branches, I modify two lines in the file in order increment the RevisionType instead of the BuildType (to generate assembly numbers like 126.96.36.199, 188.8.131.52, etc.).
Note that the DisplayName specified earlier for the new MSBuild activity is "Increment AssemblyVersion for next build" (as opposed to something like "Increment AssemblyVersion for this build"). This is an important point to understand and warrants further explanation.
When TFS 2010 starts a build, it uses a specific changeset to identify what version of the source code to get and compile. If, for example, a scheduled build starts at 5:00:00 AM on Tuesday, but one of the developers (say, Jeremy) happens to be working very early that morning and checks in code at 5:00:03 AM (3 seconds after the build started), then Jeremy's changes are not included in the build. In other words, we don't want any changesets included in the build after the changeset specified for the build (i.e. the GetVersion that is specified as an argument when starting the build).
Consequently, to ensure the incremented assembly version applies to the next build (and not the current build), we need to "rollback" the changeset created by the IncrementAssemblyVersion.proj MSBuild file. Fortunately, this is very easy to do -- simply by sync'ing the workspace again.
Copy the existing Get Workspace activity (the last activity in the Initialize Workspace sequence) and paste it below the new MSBuild activity added previously (Increment AssemblyVersion for next build). To clarify the purpose of this second SyncWorkspace activity, I recommend changing the DisplayName to Sync workspace to revert AssemblyVersion.
At this point, CustomTemplate.xaml should look like the following (in order to conserve space, only the Update Build Number portion is shown):
That's it -- we're done!
Check in the CustomTemplate.xaml file and queue a new build. Wait a few minutes for the build to complete and verify the new build appears as expected in the drop location.
Lastly, before finishing off this post, I want to discuss a couple of potential areas of concern. First, I've heard several people state that checking in files as part of the build process is not considered a "best practice." I've even seen some people go so far as to store their assembly version file (the equivalent of my AssemblyVersionInfo.txt file) on a network share somewhere simply to avoid checking out and subsequently checking in the updated file. Quite frankly, this scares me...a lot.
One could certainly argue that recovering this file is fairly easy and reliable (for example, in the event it is mistakenly deleted from the network share, you can easily "reverse engineer" the file from the most recent build number). However, just the mere idea of storing a file that is an integral part of your build anywhere outside of source control is enough to send me running down the hall yelling at the top of my lungs. It just doesn't feel right.
At the same time, however, creating a changeset as part of each "official" build certainly doesn't come without penalty. For example, let's suppose that your development project is fairly small and consequently the vast majority of your development is done on the Main branch. This means that your developers will likely be rebuilding the entire solution at least once per day (assuming you schedule a "daily build" that increments the assembly version). This assumes that each developer does a "Get Latest" each morning (which I certainly hope they do, simply out of habit), which subsequently downloads the incremented AssemblyVersionInfo.cs file (which is referenced by each and every project in the solution). Consequently all of the assemblies are compiled during the next build on the developer's local environment.
Honestly, in the ten years I've been using this approach, this "daily rebuild" has never presented a significant problem. It just becomes part of the expected "warmup" period each morning. In other words, it's a great excuse to go get a cup of coffee or discuss last night's episode of In Treatment with your colleagues hanging out in the break room ;-)
However, if you are building a very large solution that takes, say, something like 10 minutes (or more) to rebuild then this could definitely become an issue -- especially when you get close to a milestone and you start kicking off builds throughout the day. In that case, though, you probably don't have everyone working off your Main branch (that's certainly not how the DevDiv folks build the .NET Framework and Visual Studio). Instead, you likely have "Development" branches that people work on, and personally, I don't see any value in incrementing the assembly version automatically when build Dev branches.
The other concern that I've heard with regards to checking in an updated file as part of the build process is that it "pollutes" your merge information. For example, consider the example where the Revision is incremented as part of the QFE build process. Consequently, when performing a "reverse integration" (i.e. merging a hotfix from the QFE branch back through to the Main branch), you have to ignore the changes to the assembly version files. Again, at least in my experience, this doesn't introduce a significant level of pain. Your pain threshold may be much lower than mine in this particular area. If that's the case, then perhaps you should use a different assembly versioning scheme. [Don't worry, you are not going to hurt my feelings ;-) ]
In case you are wondering how I configure build definitions, here are the settings for the "daily build" as an example. If a setting is not listed in the following table, it means the default is used.