Adjust video-durationchange unit test for new behavior

LayoutTests/media/video-durationchange-on-ended.html

Index: LayoutTests/media/video-durationchange-on-ended.html
diff --git a/LayoutTests/media/video-durationchange-on-ended.html b/LayoutTests/media/video-durationchange-on-ended.html
index f9a0d86a4561106b77c61221cf426adee50e5f61..089141cbf1759debffd249926a471348ae38d0fc 100644
--- a/LayoutTests/media/video-durationchange-on-ended.html
+++ b/LayoutTests/media/video-durationchange-on-ended.html
@@ -19,9 +19,10 @@
             });
 
             waitForEventAndEnd('ended', function() {
+                consoleWrite("initialReportedDuration=" + initialReportedDuration + " video.duration=" + video.duration);
                 // Verify that the actual duration is less than
                 // the initial reported duration
-                testExpected("video.duration < initialReportedDuration", true);
+                testExpected("video.duration <= initialReportedDuration", true);

[wolenetz, 2015/04/09 17:29:06]

Though I understand how this causes the test to pass, it contradicts the goal of
this layout test, which is to verify that durationchange fires on truncated
duration files, along with confirming the situation of an actually-truncated
duration.

IIUC, the Chromium CL (https://codereview.chromium.org/1069253004/) fixes the
ended firing behavior when the truncation is minimal, but regresses reporting
the actual duration in that case.

Ideally, the right fix would be to retain both correct ended firing behavior
(for both MSE and src=) as well as reporting the actual truncated duration (even
if the truncation is tiny). Any ideas how to fix this? Do I misunderstand? If I
don't, please at least file a crbug to fix the minimal truncation not being
reported correctly, and adjust this layout test's "content/truncated.webm" to
have > 250ms duration truncation and retain the strict "<" check here.

[servolk, 2015/04/29 21:31:01]

Sorry for the delay on this. Yes, your understanding is correct. I think in
order to fix this properly we need to:
a. Have an explicit 'pipeline ended' signal from media pipeline to blink,
instead of having blink rely on currentTime/duration comparison to detect eos
(crbug.com/438581)
b. When blink receives that 'pipeline ended' signal, it should re-check the
stream duration and fire 'durationchanged' event if the actual duration turns
out to be less than expected.

             });
         </script>
     </body>