musteresel's blog

Running Golden tests with CMake


tagged: c++, c, cmake

“Golden tests” are tests which run a tool and compare it’s output to files with expected outputs, the so called golden files.

To run such a golden test, the following steps are necessary:

  1. Compile the tool, if it’s not already existing.
  2. Run the tool with some input, capturing its output.
  3. Compare the output with the expected output.

To show how this can be done with CMake I use a small example project:

/// src/
bool CountLines(std::istream & in, unsigned & lines)
  char c;
  while (in.get(c))
    if (c == '\n') ++lines;
  return ! c.bad();

To test this function I write a “tool” which uses it:

/// test/
int main(int argc, char **argv)
  // Error handling ommitted
  std::ifstream in(argv[1]);
  std::ofstream out(argv[2]);
  unsigned lines = 0;
  if (! CountLines(in, lines))
    out << "Error\n";
    return 1;
  out << lines << "\n";

Now for the CMake part, there’s a top level CMakeLists.txt which creates a library from all the “implementation” files (if the end result should be an executable, then using a static library and linking against that works fine):

# top level CMakeLists.txt
# example for executable as end result
add_library(lib STATIC
  # ... probably more
add_executable(exe src/
target_link_libraries(exe lib)


So much for the setup, now to the important test/CMakeLists.txt which generates tool from test/ and the lib library and runs it:

# test/CMakeLists.txt

# 1) Create tool
add_executable(tool EXCLUDE_FROM_ALL
target_link_libraries(tool lib)

# 2) Run tool with an input to produce some output
add_test(NAME golden-1-run
  tool ${CMAKE_CURRENT_SOURCE_DIR}/golden-1-in golden-1-out)

# 3) Compare output with expected output
  ${CMAKE_COMMAND} -E compare_files

# A) cmake extra; run compare (3) only when running the tool
#    worked (2), and run that (2) only when the tool is actually
#    built (1)
  --build "${CMAKE_BINARY_DIR}"
  --config $<CONFIG>
  --target tool

There’s a few things going on …

  1. The tool exectuable is created. It links to the (static in this case) lib library with the implementation and will thus directly use (and test) that compiled code.

    One important thing: I’ve specified EXCLUDE_FROM_ALL which means the tool executable won’t be built by make all or even make test. This requires some boiler plate code (explained further down in “A)”). Removing the EXCLUDE_FROM_ALL means the tool will be built by make / make all … but still not by make test. This is a known bug / short comming of CMake, see this answer and the questions / other answers for more.

  2. Specify a golden test (“golden-1-run”) which runs the tool with an input file (golden-1-in) to produce some output (golden-1-out).

    I use the add_test(NAME .. COMMAND tool ..) version of add_test because that uses the full path to the tool exectuable (from the tool target) to run the test and thus does not interfere with possibly available other commands (think of test …).

    This reads the input file from the current source directory (the directory where that CMakeLists.txt is) and writes the output into the current binary directory.

  3. Specify the compare step of the golden test (“golden-1-cmp”). This uses the CMake command compare_files to compare the output produced by the previous test to the expected output from the golden file (golden-1-expected).

    The expected output file comes from the current source directory, whereas the generated output file is in the current binary directory.

    One short comming of using the CMake compare_files command: it cannot show what’s different, it just compares the files. For that using diff would be an option.

And last but not least, the boiler plate code from A)

Fixtures in CMake are modelled with 3 properties: FIXTURES_SETUP, FIXTURES_CLEANUP and FIXTURES_REQUIRED. The last one - FIXTURES_REQUIRED - is set on a test which needs a fixture. The other two define the fixture: Tests marked with FIXTURES_SETUP “setup” the fixture … so they’re run before any test which needs the fixture. Test marked with FIXTURES_CLEANUP are run after any tests which needs the fixture and do “cleanup” tasks.

In the example, there are two fixtures - tool_fixture and golden-1_fixture - which each have a single setup test associated with them. The relationship is probably best to understand when visualized, so I’ve made an ASCII drawing:

| tool_build |
     | calls for setup
|--------------|               vvvvvvvvvvvvvvvv
| tool_fixture | <<---needs--- | golden-1-run |
|--------------|               ^^^^^^^^^^^^^^^^
                                    | calls for setup
vvvvvvvvvvvvvvvv               |------------------|
| golden-1-cmp | ---needs--->> | golden-1_fixture |
^^^^^^^^^^^^^^^^               |------------------|

So running ctest -R golden-1-cmp now runs first the tool_build test (building the tool exectuable), if that succeeds it runs golden-1-run (which produces the output) and only if that also succeeds it runs the requested golden-1-cmp test.

All that boilerplate CMake code can be put into a function to make it easier to define golden test. I’ve written one adjusted for my own project … which I’ll generalize and publish if I have the time to do so.