Bug 29987 – Refactor RenderThemeChromiumMac and RenderThemeChromiumSkia to render media controls using GraphicsContext.

• Component: WebCore Misc.
• Status: RESOLVED FIXED

[Andrew Scherkus]

We were using a mix of GraphicsContext, CG and Skia canvas code depending on the port.  No more!!

This follows the pattern for RenderThemeSafari and RenderThemeWin where a common class RenderMediaControls is used to handle drawing the media controls.  For Chromium I created RenderMediaControlsChromium.

No need for additional layout tests as this is pure refactor/cosmetic change and does not address bugs or add new functionality.

[Andrew Scherkus]

Created attachment 40483 [details]
Round 1

[Eric Seidel (no email)]

The WebKit video folks should be cc'd.

[Eric Seidel (no email)]

Oh, I read that as RenderThemeMac initially.  Now I see that this is chromium only.

[Andrew Scherkus]

It's still ok to get video input :)

I'm hoping that one day we can get rid of the special paintMediaFoo() methods in RenderTheme and switch to paintMedia(ControlPart part) instead (you'll notice in RenderTheme::paint() we do the case handling bit to pass off to methods)

[Eric Seidel (no email)]

Comment on attachment 40483 [details]
Round 1

Looks like you're just moving code.

Style:
 39     static bool shouldRenderMediaControlPart(ControlPart part, Element* e);

I'm not sure I understand this change:
 141         && (!style->hasAppearance() || document()->page()->theme()->shouldRenderMediaControlPart(style->appearance(), m_mediaElement));

I guess that allows all shouldRenderMediaControlPart implementations to ASSERT(appearance)?

Seems we could pretty easily consolidate this now, no?
// FIXME: this is copied from RenderMediaControls.cpp, try to consolidate.
 63 static HTMLMediaElement* parentMediaElement(RenderObject* o)

Seems we should just use a HashMap for all these now that we have so many:
79     static Image* soundFull = Image::loadPlatformResource("mediaSoundFull").releaseRef();
 80     static Image* soundNone = Image::loadPlatformResource("mediaSoundNone").releaseRef();
 81     static Image* soundDisabled = Image::loadPlatformResource("mediaSoundDisabled").releaseRef();

platformResource("name")

which looks it up and sticks it in the hash if it doesn't already exists.  I could ASSERT_NOT_REACHED that the loadPlatformResource ever fails.

Not required part of this change, but something to think about for the future.

Bug number?
 115     // FIXME: this should be a rounded rect but need to fix GraphicsContextSkia first.

We generally avoid default: for enums:
35         return true;
 236     default:
 237         return false;
Otherwise the compiler can't help us make sure we haven't missed a case.  Maybe ControlPart has a bunch more types than we're seeing here, in which case default: would maek sense.

Seems we should unify all the ASSERT_NOT_REACHED cases:
253     case MediaSeekBackButton:
 254         ASSERT_NOT_REACHED();
 255         break;
 256     case MediaSeekForwardButton:
 257         ASSERT_NOT_REACHED();
 258         break;

Seems this needs one more round.

[Andrew Scherkus]

Created attachment 40750 [details]
Round 2

no appearance means it will pass in NoControlPart enum, which doesn't help the callee figure out which part it is and they can't make a good decision whether they should render it or not

[Andrew Scherkus]

Comment on attachment 40750 [details]
Round 2

gah forgot to prepare an updated change log

[Andrew Scherkus]

Created attachment 40752 [details]
Round 2 + ChangeLog

[Andrew Scherkus]

MediaControlElements seemed like the most logical place to put parentMediaElement (both RenderMediaControls and RenderMediaControlsChromium use that file), but I made the function a global -- if that's not cool just let me know.

RenderMedia didn't feel like as good as a fit since we're dealing with RenderOjects.  HTMLMediaElement might work.. thoughts appreciated.

[Eric Carlson]

Comment on attachment 40752 [details]
Round 2 + ChangeLog


> +HTMLMediaElement* parentMediaElement(RenderObject* o)
> +{
> +    Node* node = o->node();
> +    Node* mediaNode = node ? node->shadowAncestorNode() : 0;
> +    if (!mediaNode || (!mediaNode->hasTagName(HTMLNames::videoTag) && !mediaNode->hasTagName(HTMLNames::audioTag)))
> +        return 0;
> +
> +    return static_cast<HTMLMediaElement*>(mediaNode);

I suggest naming this "toParentMediaElement" to match the pattern used elsewhere in WebCore (toInputElement, toRenderXXXX, etc). This file seems like the right place for the function.


> +class RenderMediaControlsChromium {
> +public:
> +    static bool shouldRenderMediaControlPart(ControlPart part, Element* e);

Don't need the parameter names here.

r=me with these minor changes.

[Andrew Scherkus]

Created attachment 40815 [details]
Round 3

Renamed parentMediaElement -> toParentMediaElement

Removed parameters from function signature in RenderMediaControlsChromium

[Eric Carlson]

Comment on attachment 40815 [details]
Round 3

r=me

Thanks!

[WebKit Commit Bot]

Comment on attachment 40815 [details]
Round 3

Clearing flags on attachment: 40815

Committed r49259: <http://trac.webkit.org/changeset/49259>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.