[Feature request] Profiler: reset timings

[questor]

First thanks for you great project, it's really great to have single-header-libs!

Describe the problem you're solving

I'm measuring a specific function which is NOT called per frame, but sometimes multiple times on specific requests and I do not want to have the accumulated timings but the timings of each run.

Describe the solution you'd like

I want to have a way to reset all the timings back to zero so that I can do another measurements with clean (0) timings.

[questor]

I should add that I tried to add this functionality, but (so far) failed because I do not fully understand the structure of the classes :/

[DmitriBogdanov]

Thank you for the feedback!

To reset the timings one would need to clear Profiler::call_graph_info and reset the ThreadCallGraph, default constructor should work well enough. Should look roughly like this:

// In 'Profiler' class
void clear() {
    const std::lock_guard lock(this->call_graph_mutex);
    
    this->call_graph_info.clear();
}

// Free function in the 'utl::profiler' namespace
void clear_results() {
    profiler.clear();
    thread_call_graph = ThreadCallGraph{};
}

Won't clear detached threads that are currently in-flight, but there is nothing we could do about those in a general case anyways. I will look into adding an API for explicit state control such as this.

Ultimately it seems what you're looking for is tracing which records individual call timings and gathers the results later on. This is also possible to incorporate and would be a great improvement in terms of available information, however it requires some careful thought in terms of design seeing its significant performance implications when enabled. It is likely to be added once I get to working on the v2.0.0 release.

[questor]

in a (very fast and rough) test your proposal crashes with an assertion in NodeMatrix::next_id... when it doesn't ring a bell I will wait for your api-change and will work until then with your library as it is... only have to modify my tests a bit, should still work for my (small) usecase. again, thanks for releasing your stuff :)