Share via


Using Code Coverage to Determine How Much Code is being Tested

To determine what proportion of your project's code is actually being tested by coded tests such as unit tests, you can use the code coverage feature of Visual Studio. To guard effectively against bugs, your tests should exercise or 'cover' a large proportion of your code.

Code coverage analysis can be applied to both managed (CLI) and unmanaged (native) code.

Code coverage is an option when you run test methods using Test Explorer. The results table shows the percentage of the code that was run in each assembly, class, and method. In addition, the source editor shows you which code has been tested.

Code coverage results with coloring

Requirements

  • Visual Studio Ultimate, Visual Studio Premium

To analyze code coverage on unit tests in Test Explorer

  1. On the Test menu, choose Analyze Code Coverage.

  2. To see which lines have been run, choose Show Code Coverage Coloring IconShow Code Coverage Coloring.

    To alter the colors, or to use bold face, choose Tools, Options, Environment, Fonts and Colors, Show settings for: Text Editor. Under Display Items, adjust the Coverage items.

  3. If the results show low coverage, investigate which parts of the code are not being exercised, and write more tests to cover them. Development teams typically aim for about 80% code coverage. In some situations, lower coverage is acceptable. For example, lower coverage is acceptable where some code is generated from a standard template.

Tip

To get accurate results:

  • Make sure that compiler optimization is turned off.

    If you are working with unmanaged (native) code, use a debug build.

  • Make sure that you are generating .pdb (symbol) files for each assembly.

If you don’t get the results you expect, see Troubleshooting Code Coverage. Don’t forget to run code coverage again after updating your code. Coverage results and code coloring are not automatically updated after you modify your code or when you run tests.

Reporting in blocks or lines

Code coverage is counted in blocks. A block is a piece of code with exactly one entry and exit point. If the program’s control flow passes through a block during a test run, that block is counted as covered. The number of times the block is used has no effect on the result.

You can also have the results displayed in terms of lines by choosing Add/Remove Columns in the table header. If the test run exercised all the code blocks in any line of code, it is counted as one line. Where a line contains some code blocks that were exercised and some that were not, that is counted as a partial line.

Some users prefer a count of lines because the percentages correspond more closely to the size of the fragments that you see in the source code. A long block of calculation would count as a single block even if it occupies many lines.

Managing code coverage results

The Code Coverage Results window usually shows the result of the most recent run. The results will vary if you change your test data, or if you run only some of your tests each time.

The code coverage window can also be used to view previous results, or results obtained on other computers.

You can merge the results of several runs, for example from runs that use different test data.

  • To view a previous set of results, select it from the drop-down menu. The menu shows a temporary list that is cleared when you open a new solution.

  • To view results from a previous session, choose Import Code Coverage Results, navigate to the TestResults folder in your solution, and import a .coverage file.

    The coverage coloring might be incorrect if the source code has changed since the .coverage file was generated.

  • To make results readable as text, choose Export Code Coverage Results. This generates a readable .coveragexml file which you could process with other tools or send easily in mail.

  • To send results to someone else, send either a .coverage file or an exported .coveragexml file. They can then import the file. If they have the same version of the source code, they can see coverage coloring.

Merging results from different runs

In some situations, different blocks in your code will be used depending on the test data. Therefore, you might want to combine the results from different test runs.

For example, suppose that when you run a test with input "2", you find that 50% of a particular function is covered. When you run the test a second time with the input "-2" you see in the coverage coloring view that the other 50% of the function is covered. Now you merge the results from the two test runs, and the report and coverage coloring view show that 100% of the function was covered.

Use Icon for Merge button in Code Coverage windowMerge Code Coverage Results to do this. You can choose any combination of recent runs or imported results. If you want to combine exported results, you must import them first.

Use Export Code Coverage Results to save the results of a merge operation.

Limitations in merging

  • If you merge coverage data from different versions of the code, the results are shown separately, but they are not combined. To get fully combined results, use the same build of the code, changing only the test data.

  • If you merge a results file that has been exported and then imported, you can only view the results by lines, not by blocks. Use the Add/Remove Columns command to show the line data.

  • If you merge results from tests of an ASP.NET project, the results for the separate tests are displayed, but not combined. This applies only to the ASP.NET artifacts themselves: results for any other assemblies will be combined.

Excluding elements from the code coverage results

You might want to exclude specific elements in your code from the coverage scores, for example if the code is generated from a text template. Add the attribute System.Diagnostics.CodeAnalysis.ExcludeFromCodeCoverage to any of the following code elements: class, struct, method, property, property setter or getter, event. Note that excluding a class does not exclude its derived classes.

For example:

using System.Diagnostics.CodeAnalysis; 
...
public class ExampleClass1
{ 
    [ExcludeFromCodeCoverage]
    void ExampleMethod() {...}

    [ExcludeFromCodeCoverage] // exclude property
    int ExampleProperty1 
    { get {...} set{...}}

    int ExampleProperty2
    {
        get
        {
            ...
        }
        [ExcludeFromCodeCoverage] // exclude setter
        set
        {
            ...
        }
    }
 
}
[ExcludeFromCodeCoverage]
class ExampleClass2 { ... }
Imports System.Diagnostics.CodeAnalysis


Class ExampleClass1        
    <ExcludeFromCodeCoverage()>
    Public Sub ExampleSub1()
        ...
    End Sub

    ' Exclude property
    < ExcludeFromCodeCoverage()>
    Property ExampleProperty1 As Integer
        ...
    End Property

    ' Exclude setter
    Property ExampleProperty2 As Integer
        Get
            ...
        End Get
        <ExcludeFromCodeCoverage()>
        Set(ByVal value As Integer)
            ...
        End Set
    End Property
End Class

<ExcludeFromCodeCoverage()>
Class ExampleClass2
...
End Class
// A .cpp file compiled as managed (CLI) code.
using namespace System::Diagnostics::CodeAnalysis;
...
public ref class ExampleClass1
{
  public:
    [ExcludeFromCodeCoverage]
    void ExampleFunction1() { ... }
    
    [ExcludeFromCodeCoverage]
    property int ExampleProperty2 {...}

    property int ExampleProperty2 {
      int get() { ... }
     [ExcludeFromCodeCoverage]
      void set(int value) { ...  }
   }

}

[ExcludeFromCodeCoverage]
public ref class ExampleClass2
{ ... }

Excluding elements in Native C++ code

To exclude unmanaged (native) elements in C++ code:

#include <CodeCoverage\CodeCoverage.h>
...

// Exclusions must be compiled as unmanaged (native):
#pragma managed(push, off)

// Exclude a particular function:
ExcludeFromCodeCoverage(Exclusion1, L"MyNamespace::MyClass::MyFunction");

// Exclude all the functions in a particular class:
ExcludeFromCodeCoverage(Exclusion2, L"MyNamespace::MyClass2::*");

// Exclude all the functions generated from a particular template: 
ExcludeFromCodeCoverage(Exclusion3, L"*::MyFunction<*>"); 


// Exclude all the code from a particular .cpp file:
ExcludeSourceFromCodeCoverage(Exclusion4, L"*\\unittest1.cpp");

// After setting exclusions, restore the previous managed/unmanaged state:
#pragma managed(pop)

Use the following macros:

-
ExcludeFromCodeCoverage(ExclusionName, L"FunctionName");

ExcludeSourceFromCodeCoverage(ExclusionName, L"SourceFilePath");
  • ExclusionName is any unique name.

  • FunctionName is a fully qualified function name. It may contain wildcards. For example, to exclude all the functions of a class, write MyNamespace::MyClass::*

  • SourceFilePath is the local or UNC path of a .cpp file. It may contain wildcards. The following example excludes all files in a particular directory: \\MyComputer\Source\UnitTests\*.cpp

  • #include <CodeCoverage\CodeCoverage.h>

  • Place calls to the exclusion macros in the global namespace, not within any namespace or class.

  • You can place the exclusions either in the unit test code file or the application code file.

  • The exclusions must be compiled as unmanaged (native) code, either by setting the compiler option or by using #pragma managed(off).

Note

To exclude functions in C++/CLI code, apply the attribute [System::Diagnostics::CodeAnalysis::ExcludeFromCodeCoverage] to the function. This is the same as for C#.

Including or excluding additional elements

Code coverage analysis is performed only on assemblies that are loaded and for which a .pdb file is available in the same directory as the .dll or .exe file. Therefore in some circumstances, you can extend the set of assemblies that is included by getting copies of the appropriate .pdb files.

You can exercise more control over which assemblies and elements are selected for code coverage analysis by writing a .runsettings file. For example, you can exclude assemblies of particular kinds without having to add attributes to their classes. For more information, see Customizing Code Coverage Analysis.

Analyzing code coverage in the build service

When you check in your code, your tests will run on the build server, along with all the other tests from other team members. (If you haven’t already set this up, see Run Tests in Your Build Process.) It’s useful to analyze code coverage on the build service, because that gives the most up-to-date and comprehensive picture of coverage in the whole project. It will also include automated system tests and other coded tests that you don’t usually run on the development machines.

  1. In Team Explorer, open Builds, and then add or edit a build definition.

  2. On the Process page, expand Automated Tests, Test Source, Run Settings. Set Type of Run Settings File to Code Coverage Enabled.

    If you have more than one Test Source definition, repeat this step for each one.

    • But there is no field named Type of Run Settings File.

      Under Automated Tests, select Test Assembly and choose the ellipsis button [...] at the end of the line. In the Add/Edit Test Run dialog box, under Test Runner, choose Visual Studio Test Runner.

Setting the build definition for code coverage

After the build runs, the code coverage results are attached to the test run and appear in the build summary.

Analyzing Code Coverage in a Command Line

To run tests from the command line, use vstest.console.exe. Code coverage is an option of this utility. For more information, see VSTest.Console.exe Command-Line Options.

  1. Launch the Visual Studio Developer Command Prompt:

    On the Windows Start menu, choose All Programs, Microsoft Visual Studio, Visual Studio Tools, Developer Command Prompt.

  2. Run:

    vstest.console.exe MyTestAssembly.dll /EnableCodeCoverage

Troubleshooting

If you do not see code coverage results, see Troubleshooting Code Coverage.

External resources

Guidance

Testing for Continuous Delivery with Visual Studio 2012 – Chapter 2: Unit Testing: Testing the Inside

See Also

Concepts

Customizing Code Coverage Analysis

Troubleshooting Code Coverage

Verifying Code by Using Unit Tests