As If .NET Wasn't Confusing Enough

Sat, 10 Jun 2006 16:59:31 GMT

While it's evident that Microsoft is staffed with a lot of top-notch people, history has empirically demonstrated that they have quite a few dregs as well: Just recall how disastrously the whole .NET thing was handled circa-2000.

For those who forgot, suddenly every product (including those finished or on the verge of being finished) became a part of the .NET vision, even if they had absolutely no interaction with the .NET technology stack: Windows Server.NET, Exchange.NET, Messenger.NET, SQL Server -- all a part of the .NET generation -- just as Microsoft declared everything in the generation before a part of the DNA vision (I still hear developers talking about "Microsoft DNA", not really sure what they're talking about).

As a developer who was heavily involved with the betas of what we call .NET today - a runtime and a framework, and the associated tools, for building next generation solutions - I really had no idea what .NET was in Microsoft parlance. Just as ActiveX got muddled into a meaningless term, .NET was being hijacked to basically mean "buy whatever is new or coming out soon".

Eventually that insanity stopped, and .NET collapsed down to a sortof virtual-machine runtime, a framework, and a set of tools. .NET 1.0 was one runtime, one framework, and Visual Studio.NET 2002. .NET 1.1 was a new runtime, a new framework, and Visual Studio.NET 2003. .NET 2.0 was a new runtime, an expanded framework, and Visual Studio 2005 (note the dropping of .NET on the naming, given that Visual Studio, as always, also makes non-.NET applications). There are countless assemblies and extension libraries available targeting each of them, and of course I can make libraries tomorrow that target .NET 1.0, .NET 1.1, or .NET 2.0, and it doesn't magically evolve them into .NET 3.0.

Well it looks like Microsoft is at it again. They've decided that Vista's technology platform, WinFX (which will be partially backported), is so great that it can't be just a set of assemblies or systems that the .NET runtime interacts with. No, it must be .NET 3.0! So now if you have the .NET 2.0 runtime, the .NET 2.0 Framework, targeting it with Visual Studio 2005, and you add in the WinFX framework...voila, you have .NET 3.0.

Insanity. Absolute, unbelievable insanity. Perhaps there's some amazing explanation -- for instance that their April Fools project ran a little long, and they just got the output out -- but I suspect it is just more of the same that we saw circa-2000. Some short-term euphoria over a gonna-be-released-soon project has them screwing with the terminology yet again.

Already the boards are full of "So....does this mean WinFX comes with LINQ?" (LINQ is one of the technologies promised for the next real wave of .NET)

FxCop & Cyclomatic Complexity

Thu, 08 Jun 2006 00:24:48 GMT

[The static location of this piece can be found at this address]

FxCop As a Code Quality Tool

For the past while I've been using Visual Studio Team Edition for Software Developers, one of its benefits over the Professional Edition being the inclusion of static code analysis functionality right in the IDE.

This functionality comes via the FxCop codeset, which is an excellent -- albeit unpolished -- freely available tool for analyzing the probable code quality of Intermediate Language assemblies, testing code to ensure compliance with naming standards, best practices, and highlighting areas of code that are suspect. While it's less than pleasant starting FxCop analysis from scratch on long existing project -- to be met with hundreds upon hundreds of error messages -- it's a painless process if you add it to your quality checks early on.

The standalone FxCop is largely the same as the VSTE version, and in some ways is superior. For instance that it retains the ability to actually pass configuration settings to rules, rather than accepting whatever the defaults for the rule are.

Cyclomatic Complexity

One of the few differences between the standalone application and the VSTE-included version are the addition of several new maintenance checks in the Team Edition code, one of the most useful being the cyclomatic complexity checks. Cyclomatic complexity, for those who haven't come across it before, is often used to roughly gauge the complexity of a piece of code, to determine likely candidates for refactoring, and to identify what will likely become a maintenance problem in the future. Finding the most complex pieces of code often brings you to the buggiest code as well.

Given that I still use FxCop, both the .NET 1.1 and .NET 2.0 versions (not least because the integrated version offers no ability to configure settings for rules, instead only allowing you to wholesale enable or disable. This eliminates the ability to set thresholds for tests such as the cyclomatic complexity rules), the lack of consistency between the two versions was an annoying gap.

Introducing Cyclomatic Complexity Analysis For FxCop

So I implemented a simple cyclomatic counting rule for the standalone FxCop. While in there, I added checks for statement count (the number of intermediate language "statements", which can be indicative of overly complex methods), and callout count (e.g. callouts to other methods, again which can be an indicator of overly complex/convoluted methods).

As one added benefit, I added the ability to log all of these metrics to an SQL-capable OleDB destination (e.g. SQL Server, Access, etc). If you configured an OLEDB connection string, as detailed below, you can do data analysis after a run to create pretty reports of the complexity distributions of your projects, and so on.

Download Links

yafla FxCop Rules for .NET 1.1 (e.g. FxCop 1.32)
yafla FxCop Rules for .NET 2.0 (e.g. FxCop 1.35)

Caveats

Like any tool of this type, there is only a moderate correlation between the metrics measured and actual code quality or maintainability: It is entirely possible that the optimal implementation is a highly-complex, lengthy method. This tool only provides guidance, helping to determine which code should get a complexity analysis, however from there experience and good judgement have to be applied to determine if it's really a fault. If you're using the .NET 2.0 version of FxCop, make use of the SuppressMessage attribute on methods that are necessarily highly complex.

Instructions

Drop yaflaRules.dll in your FxCop Rules subdirectory (e.g. C:\\program files\\Microsoft FxCop 1.32\\Rules).

If you want more advanced settings, configure FxCop with your targets and selected rules and then save the project file. Open the newly created .FxCop file in an editor (for instance notepad) and find the element. Expand it to an opening and closing tag (e.g. ), and between it add

Between the Rule element add any of the following entries as Name attributes of an Entry element (as exampled following) -

Connection String - an OleDb connection string determining where it will log metrics. e.g. Provider=SQLNCLI;Server=(local);Database=Analysis;Trusted_Connection=yes;
Target Table - The target table for metric logging. Default - MethodComplexity
Cyclomatic Critical Error - Level at which a critical error is triggered. Default - 60
Cyclomatic Error - Level at which an error is triggered. Default - 50
Cyclomatic Critical Warning - Level at which a critical warning is triggered. Default - 45
Cyclomatic Warning - Level at which a warning is triggered. Default - 40
Cyclomatic Information - Level at which an infromation event is triggered. Default - 20
Cyclomatic Recommended - Recommended level. Default - 20
Statements Critical Error - Statement count at which a critical error is triggered. Default - 500
Statements Error - Statement count at which an error is triggered. Default - 350
Statements Critical Warning - Statement count at which a critical warning is triggered. Default - 250
Statements Warning - Statement count at which a warning is triggered. Default - 200
Statements Information - Statement count at which an information event is triggered. Default - 150
Statements Recommended - Recommended maximum statement count per method. Default - 100
Callouts Critical Error - Callout count at which a critical error is triggered. Default - 100
Callouts Error - Callout count at which an error is triggered. Default - 75
Callouts Critical Warning - Callout count at which a critical warning is triggered. Default - 50
Callouts Warning - Callout count at which a warning is triggered. Default - 40
Callouts Information - Callout count at which an information event is triggered. Default - 30
Callouts Recommended - Recommended maximum callout count per method. Default - 30

For instance, you might end up with a element that looks like the following:

Provider=SQLNCLI;Server=(local);Database=Analysis;Trusted_Connection=yes;100500

If you opt to take advantage of metrics logging, the destination table (which will be default will be MethodComplexity, unless overridden with the Target Table name entry) requires the following columns:

ContainingType - text (e.g. nvarchar(255))
MethodName - text (e.g. nvarchar(255))
Cyclomatic - int
Statements - int
Callouts - int

e.g.
CREATE TABLE [dbo].[MethodComplexity](
[ContainingType] [nvarchar](255) COLLATE SQL_Latin1_General_CP1_CI_AS NOT NULL,
[MethodName] [nvarchar](255) COLLATE SQL_Latin1_General_CP1_CI_AS NOT NULL,
[Cyclomatic] [int] NOT NULL,
[Statements] [int] NOT NULL,
[Callouts] [int] NOT NULL
) ON [PRIMARY]

Hopefully someone finds this interesting. It scratched my itch.

Dennis Forbes on Pragmatic Software Development