WebVTT: Support 'auto' text position

Source/core/html/track/vtt/VTTCue.h

Index: Source/core/html/track/vtt/VTTCue.h
diff --git a/Source/core/html/track/vtt/VTTCue.h b/Source/core/html/track/vtt/VTTCue.h
index 4d61aa61fff9afcd20a14b97c333537b8e6258f8..0e3afa27f4d79800140751f43c2638f2ff59ec9e 100644
--- a/Source/core/html/track/vtt/VTTCue.h
+++ b/Source/core/html/track/vtt/VTTCue.h
@@ -36,6 +36,7 @@
 namespace blink {
 
 class Document;
+class DoubleOrString;
 class ExecutionContext;
 class VTTCue;
 class VTTScanner;
@@ -79,8 +80,8 @@ public:
     int line() const { return m_linePosition; }
     void setLine(int, ExceptionState&);
 
-    int position() const { return m_textPosition; }
-    void setPosition(int, ExceptionState&);
+    void position(DoubleOrString&) const;
+    void setPosition(const DoubleOrString&, ExceptionState&);
 
     int size() const { return m_cueSize; }
     void setSize(int, ExceptionState&);
@@ -158,6 +159,8 @@ private:
     std::pair<double, double> getPositionCoordinates() const;
 
     void calculateDisplayParameters();
+    int calculateComputedTextPosition() const;
+    bool textPositionIsAuto() const;
 
     enum CueSetting {
         None,
@@ -173,7 +176,7 @@ private:
     String m_text;
     int m_linePosition;
     int m_computedLinePosition;
-    int m_textPosition;
+    float m_textPosition;

[philipj_slow, 2015/01/15 09:08:57]

Can you add a test that sets position to a value that can't be represented as
float and test which value is read back? I guess sneaky double->float is already
exposed to the Web a lot via CSSOM?

[fs, 2015/01/15 10:07:16]

Are you saying that you want me to convert this to double, or just a test that
verifies that round-tripping doesn't work? (99.100001 would be an example -
which also kind of illustrates the sillyness IMHO... that doesn't round-trip
exactly for doubles either though of course)

[philipj_slow, 2015/01/19 14:44:26]

I don't know what I want, in
https://www.w3.org/Bugs/Public/show_bug.cgi?id=19836#c8 I was worried about the
size and now I'm worried about the difference between float and double being
observable...

What do you think? double wherever obvserable and float for layout would be OK,
but it's quite pedantic and assume that CSSOM has the exact same "problem". I'll
sit on the fence and let you decide!

[fs, 2015/01/19 15:09:05]

I've gone with floats for now and added a test as "documentation" - will land
that separately (as requested elsewhere). I think it would be impossible to
avoid detection (you can always, say, add another 0 after the decimal point),
and wasting 12-16 bytes (~5% of VTTCue) on precision...
I'm considering splitting out display-state from VTTCue and allocate that
on-demand - maybe then the growth could be hidden in the reduction.

     int m_cueSize;
     WritingDirection m_writingDirection;
     CueAlignment m_cueAlignment;