Refactor prepareToMeasureValuesAsync to startMeasureValuesAsyn which run test through callback

third_party/WebKit/PerformanceTests/resources/runner.js

 // There are tests for computeStatistics() located in LayoutTests/fast/harness/perftests
 
 if (window.testRunner) {
     testRunner.waitUntilDone();
     testRunner.dumpAsText();
 }
 
 (function () {
     var logLines = null;
     var completedIterations = -1;
@@ skipping 127 matching lines @@
 
     PerfTestRunner.forceLayout = function(doc) {
         doc = doc || document;
         if (doc.body)
             doc.body.offsetHeight;
         else if (doc.documentElement)
             doc.documentElement.offsetHeight;
     };
 
     function start(test, scheduler, runner) {
         if (!test) {

[Xianzhu, 2017/05/03 19:16:00]

Add "|| !runner" because with this CL, we should always have a runner.

[nednguyen, 2017/05/03 19:18:51]

Right.

[nednguyen, 2017/05/03 21:30:00]

Done.

             PerfTestRunner.logFatalError("Got a bad test object.");
             return;
         }
         currentTest = test;
 
         if (test.tracingCategories && !test.traceEventsToMeasure) {
             PerfTestRunner.logFatalError("test's tracingCategories is " +
                 "specified but test's traceEventsToMeasure is empty");
             return;
         }
 
         if (test.traceEventsToMeasure && !test.tracingCategories) {
             PerfTestRunner.logFatalError("test's traceEventsToMeasure is " +
                 "specified but test's tracingCategories is empty");
             return;
         }
         iterationCount = test.iterationCount || (window.testRunner ? 5 : 20);
         if (test.warmUpCount && test.warmUpCount > 0)
             completedIterations = -test.warmUpCount;
         logLines = PerfTestRunner.bufferedLog || window.testRunner ? [] : null;
         PerfTestRunner.log("Running " + iterationCount + " times");
         if (test.doNotIgnoreInitialRun)
             completedIterations++;
 
+        if (!scheduler) {
+          runner();
+          return;
+        }
+
         if (runner && test.tracingCategories && window.testRunner &&
             window.testRunner.supportTracing) {
             window.testRunner.traceEventsToMeasure = test.traceEventsToMeasure;
             window.testRunner.startTracing(test.tracingCategories, function() {
                 scheduleNextRun(scheduler, runner);
             });
         } else if (runner) {
             if (test.tracingCategories && !(window.testRuner &&

[Xianzhu, 2017/05/03 19:16:00]

There was a typo "testRuner" here.

How about changing old line 173-188 to:

        if (!test.tracingCategories) {
            scheduleNextRun(scheduler, runner);
            return;
        }

        if (window.testRunner && window.testRunner.supportTracing) {
            testRunner.startTracing(test.tracingCategories, function() {
                scheduleNextRun(scheduler, runner);
            });
            return;
        }

        PerfTestRunner.log("Tracing based metrics are specified but " +
            "tracing is not supported on this platform. To get those " +
            "metrics from this test, you can run the test using " +
            "tools/perf/run_benchmarks script.");

[nednguyen, 2017/05/03 19:18:50]

Thanks, I will make a separate CL for this.

                   window.testRunner.supportTracing)) {
                 PerfTestRunner.log("Tracing based metrics are specified but " +
                     "tracing is not supported on this platform. To get those " +
                     "metrics from this test, you can run the test using " +
                     "tools/perf/run_benchmarks script.");
             }
             scheduleNextRun(scheduler, runner);
         }
     }
 
@@ skipping 69 matching lines @@
                         testRunner.notifyDone();
                     });
             } else {
                 testRunner.notifyDone();
             }
         }
     }
 
     PerfTestRunner.prepareToMeasureValuesAsync = function (test) {
         PerfTestRunner.unit = test.unit;
-        start(test);
+        start(test, undefined, test.run);

[Xianzhu, 2017/05/03 17:14:25]

How about:
   start(test, function(task) { task(); }, function() { test.run(); });
?

I think this has the following benefits:
1. avoids change to start();
2. allows such tests to collect trace measurement.
3. allows test.run to access the test object through 'this';

[nednguyen, 2017/05/03 18:21:52]

I tried this first, but doing so lead to invalid test run. :-(

In particular, PerformanceTests/Parser/html-parser-threaded.html will output:

Ignoring warm-up run (undefined ms)
undefined ms
undefined ms
undefined ms
undefined ms
undefined ms
undefined ms
undefined ms
undefined ms
undefined ms
undefined ms
undefined ms
undefined ms
...

[Xianzhu, 2017/05/03 19:16:00]

I see the problem. scheduleNextRun() doesn't support async measurement.

We should have
  start(test, undefined, function() { test.run() });
here, and add
  if (!scheduler) {
    // This is an async measurement test which has its own scheduler.
    runner();
    return;
  }
in scheduleNextRun().

[nednguyen, 2017/05/03 19:18:50]

Acknowledged.

     }
 
     PerfTestRunner.measureValueAsync = function (measuredValue) {
         completedIterations++;
 
         try {
             ignoreWarmUpAndLog(measuredValue);
         } catch (exception) {
             PerfTestRunner.logFatalError("Got an exception while logging the result with name=" + exception.name + ", message=" + exception.message);
             return;
@@ skipping 146 matching lines @@
 
             iframe.contentDocument.close();
             document.body.removeChild(iframe);
         };
 
         PerfTestRunner.measureTime(test);
     }
 
     window.PerfTestRunner = PerfTestRunner;
 })();