c++code-coverageclang-cl

clang-cl code coverage only one level deep

I am trying to generate code coverage for a big C++ project using LLVM tools on windows, but all I can manage to get is one-level deep coverage, is that normal behavior or am I doing something wrong? Here's the setup: I have two classes:

class A
{
  public:
    bool foo() {return true;}
};

class B
{
  private:
    A kung;

  public:
    bool bar(){return kung.foo();}
};

And I use Google Test for unit testing everything:

TEST(Suite, Name)
{
  B ba;
  EXPECT_TRUE(ba.bar());
}

The project is configured and generated with CMake/Ninja and compiled with clang-cl with the following flags: -fprofile-instr-generate -fcoverage-mapping. Compilation is done in release, with all optimisations off (-Xclang -O0).

Running the test creates the expected profraw file which I then feed into grcov to generate html and markdown reports.

grcov ./bin --llvm --branch --llvm-path <MS_LLMV_PATH> -b ./bin -s ./Sources -t html,markdown --ignore-not-existing -o .

The generated report shows only 50% coverage, and shows that B::bar() is called, but that A::foo() is not.

Is this behavior correct and expected? Am I doing something wrong? Is there a bug in the tools I am using?

PS: I also tried the --coverage compiler option, which generates .gcno files, then .gcda files after running the tests. But I get even weirder results in the report, with a lot of blank lines in the code browser (so neither covered nor uncovered!).

Solution

  • Turns out I am an idiot :)

    I am generating both a static and a dynamic version for all my libraries, and I was only generation CC information for the dynamic libraries (and most of the Unit Tests are currently using the static libraries).

    All is well now.