Merge lp://staging/~justinmcp/oxide/mediahub into lp://staging/~oxide-developers/oxide/oxide.trunk

WebPreferences is leaked if we delete a WebView without ever accessing the preferences

Get rid of ContentBrowserClient::CreateWebPreferences(), clean up preferences ownership a bit and add some more tests to check the behaviour of WebView.preferences (including testing that WebViews inherit their opener's preferences)

Proxy OverrideSystemLocationProvider through PlatformIntegration

Get rid of the ContentMainDelegate and ContentBrowserClient implementations from qt/

Rename PlatformIntegration to BrowserPlatformIntegration (we'll almost certainly have a RendererPlatformIntegration at some point), and introduce PlatformDelegate

Get rid of some uses of ContentClient::GetInstance from Oxide code

Remove the last user of ContentClient::GetInstance()

Bump Chromium rev to 40.0.2209.0

Fix return type for flags propertes on SecurityStatus (doesn't affect the QML API)

Add an experimental API for changing the process model (currently untested - it only exists for unbreaking single process mode. Please don't use it yet)

Allow other Chromium process models to be selected, add an environment variable (OXIDE_PROCESS_MODEL), and add a warning for unsupported process models

Add missing enums

Fix arm build

Fix a shutdown hang introduced by r838

Make the UserScriptSlave destructor private

In multi-process mode, a RenderProcessHost is deleted when there are no more objects in the browser process using it. However, in single-process mode there is a single RPH representing the in-process render-thread, which lives until the process dies. This means that we need to keep the BrowserContext alive too, to avoid use-after-free crashes

Fix a nullptr deref on mobile

Fix a crash when typing

Get rid of a reinterpret_cast

Make Oxide.processModel authoritative (ie, OXIDE_PROCESS_MODEL only sets the default)

Get rid of GetBrowserProcessMainInstance

Consolidate the startup-related state in to one class

Make WebView.context read-only in single-process mode. In single-process mode, it is not possible to have mutliple WebContexts, so all WebViews will use the default context. However, if single-process is selected from the OXIDE_PROCESS_MODEL environment variable, allow the first application created WebContext to become the default context (so that "WebView { context: WebContext {} }" would still work

Add revision metadata for recently added APIs

Bump Chromium rev to 40.0.2214.5

Disable WebView.incognito in single-process mode - it won't work anyway

Don't create an OTR context unnecessarily

OxideQQuickGlobal shouldn't be a real singleton - each QQmlEngine assumes ownership of it

Don't run the API tests twice (with/without a profile directory). Instead, add a "persistent" property to TestWebContext and allow individual tests to opt out of persistent storage

Don't subclass PowerSaveBlockerImpl, as our implementation is never created

Monitor application state changes, and update PowerSaveBlocker state accordingly
(remove the screen dim lock when the app goes into the background, and restore it when it comes to the foreground).

Bump Chromium rev to 40.0.2214.6

Don't require cups-config

Limit the maximum decoded image size on mobile

Revert r866, as it had some other unfinished changes

Limit the maximum decoded image size on mobile

Gesture detector settings are in DIP units, so don't apply a scale to them. Also, tweak the gesture fling curve (probably needs further refinement)

Bump Chromium rev to 40.0.2214.10

Do not assume the cookie returned by com.canonical.Unity.Screen.keepDisplayOn is > 0
(the very first call actually returns 0, and each subsequent call increments the cookie by 1).

Expose an API to save/restore the current state of a webview.

Fix issue with devtools target type

Fix pt-BR.po and oxide mo inconsistent naming

Default to pinch virtual-viewport on mobile (same as Android now)

Drop the explicit --disable-gpu-rasterization flag - the blacklist takes care of disabling this globally on linux. Add "OXIDE_IGNORE_GPU_BLACKLIST" environment variable for disabling the blacklist - useful for testing GPU rasterization on the phone, which is totally awesome

Do not notify the IM extension if the current input method type is none.

Update qt version to 1.5.0

Bump Chromium rev to 41.0.2224.3

Bump Chromium rev to 41.0.2236.0

Fix the condition for setting WEBRTC_DETECT_ARM_NEON

Remove fuzz from power_save_blocker.patch

Ship v8 natives data files

Tidy up LoadEvent's for failed loads

Call QDesktopServices::openUrl on the UI thread. Even though QDesktopServices is thread safe, it uses QMetaObject::invokeMethod in direct mode making it quite difficult to interact with correctly when using setUrlHandler. Also, fix up tests to not provide dummy handlers in order to force load failures

Version LoadEvent.isError

Bump Chromium rev to 41.0.2243.0

Rename CompositorSoftwareOutputDevice::current_frame_ to CompositorSoftwareOutputDevice::backing_frame_

Collapse CompositorThreadProxy and CompositorThreadProxyBase

Drop the refcount check from CompositorThreadProxy::DidSwapCompositorFrame - always reclaim the resources for returned frames. It should be a bug for consumer to return frames that are still in use

Handle the frontbuffer being deleted on the compositor thread when posting the swap frame to the main thread. This can happen when, eg, hiding a webview (when we delete the LayerTreeHost)

r890 requires that CompositorFrameHandle is a RefCountedThreadSafe

Actually, add back the refcount check dropped from r890. As CompositorFrameHandle isn't meant to be thread safe, we should ensure that the compositor owns the object at this point before passing it to the compositor thread

Add a typedef for the frame handle vector in CompositorThreadProxy

Stop using non-const references in CompositorThreadProxy

Move the CompositorThreadProxy check from the compositor thread to the UI thread, as it is already for ReclaimResourcesForFrame

Stack allocate CompositorFrameAck on the compositor thread in the ReclaimResourcesForFrame path

Clean up the GPU thread shim

Add a UA stylesheet that sets a minimum viewport width (active on non-desktop form factors), set to the same width as Chrome for Android

Only inject the viewport UA override if there is no viewport meta tag, else we end up setting min-width: 980px everywhere (which breaks sites that behave correctly). There is still an edge case that is broken here - if the viewport meta tag is added or removed dynamically then the layout will break

Add LocationBarController API, which allows embedders to opt in to having the renderer compositor calculate the UI position. This will be used by the browser to animate the locationbar when scrolling

Various fixes in CompositorUtils:
- Don't pass TextureRef between threads, as it's not thread safe. In particular, don't add a reference to it on the GPU thread and then post it to the UI thread with a task that may not ever run (eg, at shutdown).
- Keep all TextureRef's in a map in CompositorUtils, as this allows us to release them all at shutdown in the case where we create a new one after the UI thread has stopped processing tasks. This and the above fix should fix bug 1337506.
- The client ID passed to TextureRef is not the ID of the client process, but the client-side resource ID of the texture (which we don't have, because we don't use the client API). Just pass 0 to this

Bump Chromium rev to 41.0.2267.0

Make CertificateError.cancelled work correctly

Don't fire TypeCommitted LoadEvents for subframes

Fix build with Qt >= 5.4

Revert r906, as it's incomplete

Fix build with Qt >= 5.4

Fix a build failure on new checkouts

Bump Chromium rev to 41.0.2272.2

Bump Chromium rev to 41.0.2272.3

Refactor WebFrame around RenderFrameHost rather than FrameTreeNode, although we can't get rid of our use of FrameTreeNode just yet. Plus other fixes

Drop the NULL checks on root_frame_

Ensure V8 snapshot is loaded in the browser process, so single-process mode works

Add experimental support for playing audio via mediahub on mobile