For some time now (I'd guess almost a year 🤷), the coverage
merging on Windows has been failing, because the reports have been
generated in a different folder than expected. Our merge script did
not report failure because it was not checking the returned error
code from OpenCppCoverage, and for some reason, the `codecov` tool
happily returned 0 even though it did not find the file it was
supposed to upload...
The former is also fixed by this commit.
Parsing --list-tests is broken, as Catch automatically line wraps the
line when it gets too long, stripping any whitespace in the process.
This means that it's impossible to reproduce the exact name of the
test if the test's name is long enough to line-wrap.
Furthermore, overwriting the LABELS property with the discovered labels
breaks users who manually added custom ctest labels.
Rolling back to using --list-test-names-only for now, as it does not
wrap lines even on very long test names.
We may be able parse the output of --list-tags to produce the ctest labels.
However, the straightforward way of doing this is to use CMake's
get_property(TEST ...) and set_property(TEST ... APPEND ...), which don't
work if the test name has spaces or other special characters. We would
need to mangle the test name to a valid CMake identifier to do it that way.
- The current setup tries to detect USE_CPP14/USE_CPP17 and sets the
CXX_STANDARD property for the SelfTest target. This is not ideal, since
CMAKE_CXX_STANDARD can be provided by the toolchain file or as command line
option and should be used by the library internally correctly. Hence, the
whole set of the relevant lines from `projects/CMakeLists.txt` have been
removed.
- The above can also cause subtle issues where the user is expecting the tests
to compile with C++17 after setting CMAKE_CXX_STANDARD and then getting
results of compilation with C++11 as USE_CPP17 has not been set.
- The current build matrix used the above code to run the tests. So, even
though the it should not required anymore to build Catch2, it was still
required to send correct options to build matrix. In that respect,
.travis.yml has been modified to send correct options to the build command
in the new setup.
This keeps it out of the main include path when benchmarking is
enabled, somewhat reducing the compilation-time penalty.
Also moved some other functions into the .cpp file, especially
helpers that could be given internal linkage, and concretized some
iterator-templated code that only ever used
`std::vector<double>::iterator`.
This allows us to move <stdexcept> out of the common path, and replace
it with just <exception>. The difference between these two headers is
~13k lines after preprocessing on libstdc++ (16k vs 3k) and ~17k lines
for MS's STL(33k vs 16k).
Note that this is only beneficial if no other stdlib header we use
includes <stdexcept>. AFAIK this is true for the newest MS's STL,
but I have no idea of the applicability for libstdc++ and libc++.
* Units from <ratio> are no longer redeclared in our own namespace
* The default clock is `steady_clock`, not `high_resolution_clock`,
because, as HH says "high_resolution_clock is useless. If you want
measure the passing of time, use steady_clock. If you want user
friendly time, use system_clock".
* Benchmarking support is opt-in, not opt-out, to avoid the large
(~10%) compile time penalty.
* Benchmarking-related options in CLI are always present, to decrease
the amount of code that is only compiled conditionally and making
the whole shebang more maintainble.
Changes done to Nonius:
* Moved things into "Catch::Benchmark" namespace
* Benchmarks were integrated with `TEST_CASE`/`SECTION`/`GENERATE` macros
* Removed Nonius's parameters for benchmarks, Generators should be used instead
* Added relevant methods to the reporter interface (default-implemented, to avoid
breaking existing 3rd party reporters)
* Async processing is guarded with `_REENTRANT` macro for GCC/Clang, used by default
on MSVC
* Added a macro `CATCH_CONFIG_DISABLE_BENCHMARKING` that removes all traces of
benchmarking from Catch
