clock() behavior in ncurses-CodePudding

This is the reduced code to demonstrate the behavior.

#include <stdio.h>
#include <ncurses.h>
#include <time.h>

#define DELAY  2.0

int main ( void) {
    static clock_t start;
    static double delay = 0.0;
    double elapsed = 0.0;
    clock_t stop;

    initscr ( );
    timeout ( 20); // milliseconds getch waits for a character

    if ( delay < .0001) {
        start = clock();
        delay = DELAY;
    }
    do {
        // getch ( );
        stop = clock();
        elapsed = (double)( stop - start) / CLOCKS_PER_SEC;
        move ( 22, 0);
        printw ( "cps %ld elapsed %f\n", CLOCKS_PER_SEC, elapsed);
        refresh ( );
    } while ( elapsed < delay);
    endwin ( );
    return 0;
}

Compiled and linked with ncurses it runs for about two seconds.
Uncomment // getch ( ); recompile and link and it runs for about two hundred seconds.
The problem has been solved by using clock_gettime().
I am curious why clock() is behaving strangely.

CodePudding user response：

On POSIX systems (like Linux or macOS) the clock function reports the CPU time used for the process. If it doesn't use any CPU time, like when it's waiting a lot of time, then you will get very small differences.

If you use clock_gettime with either CLOCK_MONOTONIC or CLOCK_REALTIME those are based on the wall clock of the system.

Note that the behavior of clock is different on Windows where it's a representation of the wall clock.