MSE: Fix logic bugs with high precision duration

third_party/WebKit/Source/core/html/HTMLMediaElement.cpp

Index: third_party/WebKit/Source/core/html/HTMLMediaElement.cpp
diff --git a/third_party/WebKit/Source/core/html/HTMLMediaElement.cpp b/third_party/WebKit/Source/core/html/HTMLMediaElement.cpp
index 9e4d7251251f3d24dc3bbe4989ba190af632954f..8f36b5f8fbc8344b267aedbdcb42b868985efc5c 100644
--- a/third_party/WebKit/Source/core/html/HTMLMediaElement.cpp
+++ b/third_party/WebKit/Source/core/html/HTMLMediaElement.cpp
@@ -2006,16 +2006,6 @@ double HTMLMediaElement::duration() const {
   if (!m_webMediaPlayer || m_readyState < kHaveMetadata)
     return std::numeric_limits<double>::quiet_NaN();
 
-  // FIXME: Refactor so m_duration is kept current (in both MSE and
-  // non-MSE cases) once we have transitioned from kHaveNothing ->
-  // kHaveMetadata. Currently, m_duration may be out of date for at least MSE
-  // case because MediaSource and SourceBuffer do not notify the element
-  // directly upon duration changes caused by endOfStream, remove, or append
-  // operations; rather the notification is triggered by the WebMediaPlayer
-  // implementation observing that the underlying engine has updated duration
-  // and notifying the element to consult its MediaSource for current
-  // duration. See http://crbug.com/266644

[chcunningham, 2016/12/15 01:24:26]

Matt, I think this comment is stale (or I misunderstand). I checked all cases of
duration changing (decrease for eos, increase for append, user specified, init
segment) and found they all notify the pipeline which bubbles to WMPI and
ultimately HTMLMediaElement. If you agree we should close that old bug. 

[wolenetz, 2016/12/16 00:56:52]

These do eventually bubble, but in the interim time, m_duration is out of date
(though ::duration() with this change, will return current MSE duration). The
problem is that ::duration() is not always used; there's an m_duration which is
still not necessarily in agreement with WMPI, even with this change. At this
time, it could be a simple fix. WDYT?

[chcunningham, 2017/01/03 17:48:06]

At this point m_duration is only used in OnDurationChanged as a guard against
spamming that event. Having said that, I have taken this opportunity to clean
this up. 

HTMLMediaElement::duration() will now simply return m_duration. 

m_duration will be explicitly set to NaN during
invokeResourceSelectionAlgorithm() 

m_duration will be explicitly set to WMPI->duration() during transition to
HAVE_METADATA. 

m_duration will be updated (already is) whenever duration changes occur from the
lower layer

-
   if (m_mediaSource)
     return m_mediaSource->duration();