C++ code coverage by tests

Code coverage by tests

We will continue on unit testing the code. But this time I'll show you how to check which part of code is covered by tests and which is not. Such information becomes useful when analyzing how well your code is tested. And the more code is tested, the more stable it can be considered.

In computer science, code coverage is a measure used to describe the degree to which the source code of a program is tested by a particular test suite. A program with high code coverage has been more thoroughly tested and has a lower chance of containing software bugs than a program with low code coverage.

Setup environment


These tools are all available on linux systems. On Debian:

$ sudo apt-get install g++ lcov cmake make

Create sample project

You can get a sample from github:

$ git clone https://github.com/povilasb-com/cpp-code-coverage.git

We'll use the same project structure as in C++ unit tests with googletest:

|__ googletest-cmake
  |__ lib
  | |__ googletest -> ~/dev-tools/gtest-1.7.0
  |__ src
  | |__ lib1.cpp
  | |__ lib1.hpp
  |__ test
  | |__ lib1_test.cpp
  |__ CMakeLists.txt
  |__ Makefile

But this time i'm not going to describe how to create the project itself, you can get it from github.

Although there are few differences in build scripts compared to C++ unit tests with googletest that I would like to mention:


cmake_minimum_required (VERSION 2.6)
project (Lib1 CXX)

set (CMAKE_CXX_FLAGS "-ggdb -fprofile-arcs -ftest-coverage")
set (CMAKE_EXE_LINKER_FLAGS "-fprofile-arcs -ftest-coverage")

include_directories ("${SRC_DIR}")


# Compiles static lib that will be linked with tests.
set (LIB_NAME "lib1")
add_library ("${LIB_NAME}" STATIC ${SRC_FILES})

# Include googletest.
set (GTEST_DIR "${CMAKE_CURRENT_SOURCE_DIR}/lib/googletest")
add_subdirectory (${GTEST_DIR})
include_directories ("${GTEST_DIR}/include")

# Build tests executable.
set (TEST_EXEC "${LIB_NAME}_test")

add_executable ("${TEST_EXEC}" ${TEST_SRC_FILES})
target_link_libraries ("${TEST_EXEC}" "${LIB_NAME}" "gtest" "gtest_main")

This CMake sets specific compiler (CMAKE_CXX_FLAGS) and linker (CMAKE_EXE_LINKER_FLAGS) flags "-fprofile-arcs" and "-ftest-coverage".

When these flags are set, compiler injects profiling code that collects data about program execution: e.g. which line how much times was executed.


BUILD_DIR = build

all: test test-run coverage

test: $(BUILD_DIR)
        cd $(BUILD_DIR); cmake $(CURDIR); make


        lcov --capture --directory $(BUILD_DIR) \
                --output-file $(BUILD_DIR)/coverage.info
        genhtml $(BUILD_DIR)/coverage.info \
                --output-directory $(BUILD_DIR)/coverage-report

        mkdir -p $@

        rm -rf $(BUILD_DIR)

.PHONY: all cmake clean coverage test test-run

This Makefile compared to unit test project Makefile contains target "coverage". This target runs lcov command that collects coverage data and generates HTML output.

Compile, run tests and build coverage report

All thsese steps are programmed in Makefile script, so simply invoke "make".

As a result it will

  • build tests in build directory;

  • run tests executable;

  • this executable will generate *.gcda and *.gcno files, which contain code execution information. Do not worry about these files, lcov tool will take care of them for us.

  • And finally it will run lcov tool that locates and parses execution data automatically for us and creates html output.

You can see the results in here.

Analysis report


In our sample code coverage by tests report is built in "./build/coverage-report". This directory contains "index.html". Simply open it in your web browser and you should be all set.

Do not worry if you see a lot of source files indicated by red color. This actually means that only a small part of this code was executed. But mostly these files are c++ standard libraries.

Simply navigate to src/lib1.cpp - this is the source file we were testing and we are interested in how many lines of this file our tests executed.

Analyzing our library sources


Analyzing further you'll see that the situation is not that bad.

Our sample tests cover 66% of lib1.cpp code. And this report show which lines in code were executed and which were not. So based on these results you can supplement your test cases.