Stopwatch is inconsistent when guarding against negative Elapsed durations

[MihaZupan]

Stopwatch has the following code to detect negative elapsed durations, but this logic only runs during the Stopwatch.Stop call.

runtime/src/libraries/System.Private.CoreLib/src/System/Diagnostics/Stopwatch.cs

Lines 63 to 72 in a1f1ab8

	if (_elapsed < 0)
	{
	// When measuring small time periods the Stopwatch.Elapsed*
	// properties can return negative values. This is due to
	// bugs in the basic input/output system (BIOS) or the hardware
	// abstraction layer (HAL) on machines with variable-speed CPUs
	// (e.g. Intel SpeedStep).
	_elapsed = 0;
	}

If this behavior is worth preserving for stopped Stopwatch instances, I don't see why the responsibility should be shifted to the caller for "active" ones.

I believe this logic should either be removed entirely, or apply to all overloads regardless of IsRunning, possibly including the new GetElapsedTime helpers.

I wouldn't be surprised if some of our own logic is not accounting for the possibility of negative durations.

[ghost]

Tagging subscribers to this area: @tommcdon
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Stopwatch has the following code to detect negative elapsed durations, but this logic only runs during the Stopwatch.Stop call.

runtime/src/libraries/System.Private.CoreLib/src/System/Diagnostics/Stopwatch.cs

Lines 63 to 72 in a1f1ab8

	if (_elapsed < 0)
	{
	// When measuring small time periods the Stopwatch.Elapsed*
	// properties can return negative values. This is due to
	// bugs in the basic input/output system (BIOS) or the hardware
	// abstraction layer (HAL) on machines with variable-speed CPUs
	// (e.g. Intel SpeedStep).
	_elapsed = 0;
	}

If this behavior is worth preserving for stopped Stopwatch instances, I don't see why the responsibility should be shifted to the caller for "active" ones.

I believe this logic should either be removed entirely, or apply to all overloads regardless of IsRunning, possibly including the new GetElapsedTime helpers.

Author:	MihaZupan
Assignees:	-
Labels:	bug, area-System.Diagnostics
Milestone:	-

[stephentoub]

If this was "due to bugs", is this even still an issue?

I agree we should either delete the logic or make it apply everywhere. It doesn't make sense to only do this for stopped, if it's even needed.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Stopwatch has the following code to detect negative elapsed durations, but this logic only runs during the Stopwatch.Stop call.

runtime/src/libraries/System.Private.CoreLib/src/System/Diagnostics/Stopwatch.cs

Lines 63 to 72 in a1f1ab8

	if (_elapsed < 0)
	{
	// When measuring small time periods the Stopwatch.Elapsed*
	// properties can return negative values. This is due to
	// bugs in the basic input/output system (BIOS) or the hardware
	// abstraction layer (HAL) on machines with variable-speed CPUs
	// (e.g. Intel SpeedStep).
	_elapsed = 0;
	}

If this behavior is worth preserving for stopped Stopwatch instances, I don't see why the responsibility should be shifted to the caller for "active" ones.

I believe this logic should either be removed entirely, or apply to all overloads regardless of IsRunning, possibly including the new GetElapsedTime helpers.

I wouldn't be surprised if some of our own logic is not accounting for the possibility of negative durations.

Author:	MihaZupan
Assignees:	-
Labels:	bug, area-System.Runtime
Milestone:	8.0.0

[ericstj]

Adjusted the label on this since Stopwatch lives in System.Runtime.

[tannergooding]

We've switched to using exclusively monotonic timers for quite a while now and so I think this code is effectively "dead".

We might need to confirm that same is true for Mono, but I believe its mostly all shared atm.

[tannergooding]

Moving to future. The implementation is calling exclusively into 64-bit monotonic timers as of right now.

It would be good to remove this now unnecessary logic and document the reasoning. The overflow rate for 1ns ticks is approx. 292 years, which is more than reasonable.