Google Plus Logo
Twitter Logo
Facebook Logo

Reduce Catch tests compilation time by another 16%

No, it's not the same post as two days! I've been able to reduce the compilation time of my test cases by another 16%!

Two days ago, I posted an article about how I reduced the compilation time of my tests by 13%, by bypassing the expression deduction from Catch. I came up with the macro REQUIRE_EQUALS:

template<typename L, typename R>
void evaluate_result(Catch::ResultBuilder&& __result, L lhs, R rhs){
    __result.setResultType(lhs == rhs);

#define REQUIRE_EQUALS(lhs, rhs) \
    evaluate_result(Catch::ResultBuilder( "REQUIRE", CATCH_INTERNAL_LINEINFO, #lhs " == " #rhs, Catch::ResultDisposition::Normal ), lhs, rhs);

This has the advantage that the left and right hand sides are directly set, not deduced with templates and operator overloading. This still has exactly the same features has the original macro, but it is a bit less nice in the test code. I was quite happy with that optimization, but it turned out, I was not aggressive enough in my optimizations.

Even though it seems simple, the macro is still bloated. There are two constructors calls: ResultBuilder and SourceLineInfo (hidden behind CATCH_INTERNAL_LINEINFO). That means that if you test case has 100 assertions, 200 constructor calls will need to be processed by the compiler. Considering that I have some test files with around 400 assertions, this is a lot of overhead for nothing. Moreover, two parameters have always the same value, no need to repeat them every time.

Simplifying the macro to the minimum led me to this:

template<typename L, typename R>
void evaluate_result(const char* file, std::size_t line, const char* exp, L lhs, R rhs){
    Catch::ResultBuilder result("REQUIRE", {file, line}, exp, Catch::ResultDisposition::Flags::Normal);
    result.setResultType(lhs == rhs);

#define REQUIRE_EQUALS(lhs, rhs) \
    evaluate_result(__FILE__, __LINE__, #lhs " == " #rhs, lhs, rhs);

The macro is now a simple function call. Even though the function is a template function, it will only be compiled for a few types (double and float in my case), whereas the code of the macro would be unconditionally compiled for each invocation.

With this new macro and function, the compilation time went down from 664 seconds to 554 seconds! This is more than 16% reduction in compilation time. When comparing against the original compilation time (without both optimizations) of 764 seconds, this is a 27% reduction! And there are absolutely no difference in features.

This is a really great result, in my opinion. I don't think this can be cut down more. However, there is still some room for optimization regarding the includes that Catch need. Indeed, it is very bloated as well. A new test framework, doctest follows the same philosophy, but has much smaller include overhead. Once all the necessary features are in doctest, I may consider adapting my macros for it and using it in place of Catch is there is some substantial reduction in compilation time.

If you want to take a look at the code, you can find the adapted code on Github.


Comments powered by Disqus