Code coverage can be checked using clang‘s source-based coverage tools. You must use the GN argument use_coverage=true. It’s recommended to do this in a separate output directory since the added instrumentation will affect performance and generate an output file every time a binary is run. You can read more about this in clang's documentation but the bare minimum steps are also outlined below. You will also need to download the pre-built clang coverage tools, which are not downloaded by default. The easiest way to do this is to set a custom variable in your .gclient file. Under the “openscreen” solution, add:
"custom_vars": { "checkout_clang_coverage_tools": True, },
then run gclient runhooks. You can also run the python command from the clang_coverage_tools hook in //DEPS yourself or even download the tools manually (link).
Once you have your GN directory (we'll call it out/coverage) and have downloaded the tools, do the following to generate an HTML coverage report:
out/coverage/openscreen_unittests third_party/llvm-build/Release+Asserts/bin/llvm-profdata merge -sparse default.profraw -o foo.profdata third_party/llvm-build/Release+Asserts/bin/llvm-cov show out/coverage/openscreen_unittests -instr-profile=foo.profdata -format=html -output-dir=<out dir> [filter paths]
There are a few things to note here:
default.profraw is generated by running the instrumented code, but foo.profdata can be any path you want.<out dir> should be an empty directory for placing the generated HTML files. You can view the report at <out dir>/index.html.[filter paths] is a list of paths to which you want to limit the coverage report. For example, you may want to limit it to cast/ or even cast/streaming/. If this list is empty, all data will be in the report.The same process can be used to check the coverage of a fuzzer's corpus. Just add -runs=0 to the fuzzer arguments to make sure it only runs the existing corpus then exits.