...and precompiled headers!
I've used CMake for around a year or so and programmed in C++ for maybe half of that time as of right now, and recently I've finally had a chance to write unit tests for a project of mine. Previous projects were either too simple or too reliant on third party software to warrant the effort but this one really lent itself to it.
I wanted the workflow for building and testing the project to be
cd build cmake .. make make test
and I wanted to see the test output when running
make test. This disqualified utilities like
CTest since they hide the ouput of the test driver without proving an option to show it.
I didn't realize this until I actually ran a test for the first time, but I also want test compiles and runs to be fast. This is why I spent quite a bit of time on figuring out how to precompile the Catch2 header.
The unit test folder is split up in two groups: (1) the Catch2
main function and (2) the actual tetst.
└── tests ├── catch.h ├── CMakeLists.txt ├── main.cpp └── tests.cpp
Directory listing of the `tests` folder in my project.
main.cpp contains the following two lines
#define CATCH_CONFIG_MAIN #include "catch.h"
tests.cpp contains the
CMakeLists.txt in the
tests/ folder contains the following code:
# This exclusively contains the catch2 main define, separated because precompiling # headers would be a problem otherwise. add_library(tests-main STATIC main.cpp) # This contains the actual tests. add_executable(tests-run tests.cpp) target_link_libraries(tests-run tests-main) target_compile_definitions(tests-run PRIVATE CATCH_CONFIG_FAST_COMPILE CATCH_CONFIG_DISABLE_MATCHERS) # Important if you don't want the test compile to take >5s every time. target_precompile_headers(tests-run PRIVATE catch.h) add_custom_target(test "tests-run" "-d yes")
The comments in the first couple of lines reveal that the Catch2
main has to be its own file in order to allow headers to be precompiled. The reason for this is sneaky and (in my opinion) borders on being a bug: If
CATCH_CONFIG_MAIN is defined,
#include "catch.h" expands with the implementation code (which would usually reside in
.cpp files), otherwise it expands without it.
Precompiling would therefore cause the same implementation code to be expanded for all instances of
#include "catch.h" and not just the one time it should usually be expanded, leading to linking errors.
In the above CMake code, this problem is sidestepped by compiling Catch2's
main as a library which is then linked to the actual tests. I got this idea from a blog post by Mochan Shrestha, and it works beautifully so far.
The last line looks simple, and it really is simple: it adds a custom target called
target which builds and runs
make test works as expected. Test compilation is slow for the first compile but really quick afterwards.