v3.16.0

Fixes

• Multiple fixes in catch_discover_tests:
  • Fixed <target>_TESTS variable accumulating JSON fragments alongside test names.
    • This was introduced during the refactoring in last release.
  • Fixed <target>_TESTS variable from catch_discover_tests not escaping test names to be properly parsed by CMake.
    • This means that e.g. test names with semicolons will not be split into multiple partial test names.
    • This bug has existed since the first version of the script.
  • Fixed TEST_PREFIX/TEST_SUFFIX args having leading/trailing whitespace stripped.
• Added workaround for Clang 20-21 compile error with TEMPLATE_PRODUCT_TEST_CASE (#3115, #3173)

Improvements

• Verbosity option is now handled per reporter.
  • The standalone --verbosity flag is propagated to all reporters as default, just like --colour-mode.
• The JSON reporter considers verbosity when listing tests.
• Another set of performance improvements for catch_discover_tests performance
  • The newest version can register about 3k tests in 1 second, up from 1k previously.
• The initial TEST_CASE registration is slightly faster.
• Reduced overhead from first instantiating BENCHMARK machinery in a TU.
• Improved compilation speed when multiple templated test case macros use the same types.
• Rewrote implementation of the string matchers
  • Case-sensitive matching (the default) is significantly faster.
  • Case-insensitive matching is faster in most cases.
• Added optimizer barrier to calls into benchmarks without return values
  • This limits the optimizations compiler can perform between calls into the benchmarked function, improving the accuracy.
• Added warning that checks for using sharding without deterministic test order (#3186)