assert.timeout() can fail on simple synchronous tests

[Krinkle]

Tell us about your runtime:

• QUnit version: 2.13.0-dev (latest git)
• What environment are you running QUnit in?: Browser

What are you trying to do?

Code that reproduces the problem:

The following test fails intermittently in IE Opera (Chromium-based) and IE 10 on Windows (probably other browsers as well). However, it passes consistently in Firefox, Chrome, and Safari on macOS:

	QUnit.test( "does not push a failure if test is synchronous", function( assert ) {
		assert.timeout( 1 );

		var wait = Date.now() + 10;
		while ( Date.now() < wait ) {} // eslint-disable-line no-empty

		assert.true( true );
	} );

Failure

Test took longer than 1ms; test timed out.

[Krinkle]

I suspect the issue might be with this assumption in assert.timeout() which is meant to catch cases where assert.timeout() is called twice in the same test, but that is not actually happening in our reproduction case:

qunit/src/assert.js

Lines 18 to 33 in 9f16eed

	timeout( duration ) {
	if ( typeof duration !== "number" ) {
	throw new Error( "You must pass a number as the duration to assert.timeout" );
	}
	this.test.timeout = duration;
	// If a timeout has been set, clear it and reset with the new duration
	if ( config.timeout ) {
	clearTimeout( config.timeout );
	if ( config.timeoutHandler && this.test.timeout > 0 ) {
	resetTestTimeout( this.test.timeout );
	}
	}
	}

I added the following to confirm the issue:

  	QUnit.test( "does not push a failure if test is synchronous", function( assert ) {
+		if ( QUnit.config.timeout ) {
+			throw new Error( "QUnit.config.timeout: " + QUnit.config.timeout );
+		}
 		assert.timeout( 1 );
 

With this patch in place, I confirmed that indeed this condition is being intemittently. About every other test run for me. And while it doesn't cause a test failure normally for me in Firefox or Chrome, I do find this condition being met met half the time in those browsers.

So it would seem we have a race condition where about half the time, config.timeout ends up retaining some kind of value. And only on Windows does it cause the test to fail as a result of that.