Frame timings and non-monotonic clocks

[dmit]

I was looking at using this library in my project that uses Zig 0.16 (nightly), so I set out to see how much effort it would take to update a local fork to the new Zig 0.16 std.Io interface. While that work is ongoing, I might have discovered an issue with the frame timing code in src/core/program.zig:

zigzag/src/core/program.zig

Lines 80 to 81 in b2610e0

	.start_time = std.time.nanoTimestamp(),
	.last_frame_time = std.time.nanoTimestamp(),

zigzag/src/core/program.zig

Lines 190 to 210 in b2610e0

	/// Execute a single frame: poll input, process events, render.
	pub fn tick(self: *Self) !void {
	const now = std.time.nanoTimestamp();
	const delta = @as(u64, @intCast(now - self.last_frame_time));
	// Enforce framerate limit
	const min_frame_time_ns: u64 = if (self.options.fps > 0)
	@divFloor(std.time.ns_per_s, self.options.fps)
	else
	16_666_666; // ~60fps default
	if (delta < min_frame_time_ns) {
	std.Thread.sleep(min_frame_time_ns - delta);
	}
	self.last_frame_time = std.time.nanoTimestamp();
	const actual_delta = @as(u64, @intCast(self.last_frame_time - now + @as(i128, @intCast(delta))));
	self.context.delta = actual_delta;
	self.context.elapsed = @intCast(self.last_frame_time - self.start_time);
	self.context.frame += 1;

In 0.15 std.time.nanoTimestamp uses the system's real clock, so it's not monotonic (may return an earlier timestamp than before after a daylight savings clock jump, or after an NTP adjustment), and will have large forward jumps in case of sleep/resume. The latter is not much of a problem (aside from tanking the reported frame rate?), but the former might cause an @intCast() from a negative signed integer value to an unsigned integer, which is Illegal Behavior.

It appears that this code should instead be using std.time.Timer in 0.15, and in the future std.Io.Clock.awake in 0.16.

[meszmate]

Good catch, thanks for the detailed writeup. Will look into it.

[meszmate]

Fixed on latest main with monotonic timing and 0.16 compatible sleep path.
If it still reproduces, please reopen the issue.