lcov: A Good HTML Generator for gcov Results

Writing test cases is a good way to prevent and detect problems or bugs in source code. They improve understanding of the difficult parts by requiring deeper thought into how to test of those areas. By rerunning the same test sequences one can assure that the code still produces the same results even after making significant changes. cppunit provides a C++ test framework which is sort of over-bloated. However reduced to a set of reusable template files the framework gets quite handy.

To measure how much of the code is tested, gcov provides a way to determine which lines are executed during a test suite run. Note that this simple line-is-touched metric is only one aspect of how well a piece of code is tested. However gcov's results are printed in text mode and it cannot merge the results from multiple coverage files, so multi-file test suites cannot be measured as a whole.

Yesterday I finally found a good open-source Linux tool to get correct coverage results: lcov. It was designed to measure coverage in the Linux kernel, but works very well on user-space programs as well. lcov builds on gcov's data files and generates HTML report files. It even highlights the untested source code lines.

I uploaded the test coverage results of the STX B+ Tree test suite. It shows 89.2% coverage of the main, most difficult header file implementing the insert and erase algorithms.