Better perf scoping

Often it is need to troubleshoot performance issues at a specific region of a program rather than the entire program. To achive profiling of exact region the following snippets of code are helpful:

#include <signal.h>
#include <sys/wait.h>

pid_t run_perf() {
    pid_t pid = fork();
    if (pid != 0)
        return pid;
    const auto parent_pid = std::to_string(getppid());
    std::cout << "Running perf on parent process " << parent_pid << std::endl;
    execlp("perf",
           "perf",
           "record",
           "-F999",
           "-g",
           "--call-graph",
           "dwarf",
           "-p",
           parent_pid.c_str(),
           (char*)nullptr);
    throw std::runtime_error(std::string("perf failed: ") + std::strerror(errno));
}

void stop_perf(pid_t perf_pid) {
    printf("Sent SIGTERM to the perf\n");
    kill(perf_pid, SIGTERM);
    int status = 0;
    printf("Waiting the perf\n");
    waitpid(perf_pid, &status, 0);
    printf("The perf has terminated\n");
}

void benchmarked() {
    auto const perf_pid = run_perf();
    
    // the code
    
    stop_pid(perf_id);
}

What happens here is that instead of running the perf from the terminal we run it from the benchmarked code. We run the perf right before the code of interest and kill it right after it.

Visitor Counter