7cc3858a4c
content during ViewTransition. Without a fallback surface, the restore page could draw the a new frame using the restored page's primary surface before the VT animation runs. The primary surface is the last surface drew before entering the BFCache, hence introducing a visual glitch. To ensure a smooth view transition: 1. We need to ensure that drawing content from the restored page starts with its first main frame (i.e a commit from the renderer main thread). That requires updating the SurfaceID when the page is restored. Generally updating the parent sequence number is sufficient for this but we want to prevent Viz from reusing a cached CompositorFrame for the restored page. This happens because the logic here favours using any frame from the primary Surface's SurfaceAllocationGroup: https://source.chromium.org/chromium/chromium/src/+/refs/heads/main:components/viz/service/surfaces/surface_manager.cc;l=392;drc=73265a30d5b5a3873dbf3b92f01abc5c2892ece3. In the case of BFCache activation, the primary surface is captured as the last frame before the page entered BFCache. So we also need to change the allocation group, which is done by updating the embed token. 2. We need to ensure we keep drawing the content from the previous page until the restored page draws a frame. This is done by reusing the existing code for retaining a fallback surface when swapping RWHVs. This CL explicitly tells the View/ui::Layer/CC/Viz to not use the primary surface of the BFC-restored page. Instead, set the fallback surface of the restored page to the primary surface of the current view, and draw using the fallback. need tests. Low-Coverage-Reason: render_widget_host_view_child_frame.cc does not Bug: 1415340 Change-Id: I00da69d9988ca269668d6748fb99760658843fa7 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/4563504 Reviewed-by: Khushal Sagar <khushalsagar@chromium.org> Reviewed-by: Bo Liu <boliu@chromium.org> Commit-Queue: William Liu <liuwilliam@chromium.org> Reviewed-by: Nico Weber <thakis@chromium.org> Reviewed-by: Jonathan Ross <jonross@chromium.org> Reviewed-by: Vladimir Levin <vmpstr@chromium.org> Cr-Commit-Position: refs/heads/main@{#1182211}
Content module
High-level overview
The "content" module is located in src/content, and is the core code needed to
render a page using a multi-process sandboxed browser. It includes all the web
platform features (i.e. HTML5) and GPU acceleration. It does not include Chrome
features, e.g. extensions/autofill/spelling etc.
Motivation
As the Chromium code has grown, features inevitably hooked into the wrong
places, causing layering violations and dependencies that shouldn't exist. It's
been hard for developers to figure out what the "best" way is because the APIs
(when they existed) and features were together in the same directory. To avoid
this happening, and to add a clear separation between the core pieces of the
code that render a page using a multi-process browser, consensus was reached to
move the core Chrome code into src/content (content not
chrome :) ).
content vs chrome
content should only contain code that is required to implement the web
platform. Generally, a feature belongs in this category if and only if all of
the following are true:
- Its launch is tracked on the https://chromestatus.com/ dashboard.
- It has an associated spec.
- It is going through the feature development lifecycle.
In contrast, many features that are common to modern web browsers do not satisfy
these criteria and thus, are not implemented in content. A non-exhaustive
list:
- Extensions
- NaCl
- SpellCheck
- Autofill
- Sync
- Safe Browsing
- Translate
Instead, these features are implemented in chrome, while content only
provides generic extension points that allow these features to subscribe to the
events they require. Some features will require adding new extension points: for
more information, see How to Add New Features (without bloating
RenderView/RenderViewHost/WebContents).
Finally, there are a number of browser features that require interaction with
online services supplied by the vendor, e.g. from the above list, Safe Browsing,
Translate, Sync, and Autofill all require various network services to function.
The chrome layer is the natural place to encapsulate that vendor-specific
integration behavior. For the rare cases where a web platform feature
implemented in content has a dependency on a network service (e.g. the network
location service used by Geolocation), content should provide a way for the
embedder to inject an endpoint (e.g. chrome might provide the service URL to
use). The content module itself must remain generic, with no hardcoded
vendor-specific logic.
Architectural Diagram
See an older diagram at: https://www.chromium.org/developers/content-module.
The diagram illustrates the layering of the different modules. A module can include code directly from lower modules. However, a module can not include code from a module that is higher than it. This is enforced through DEPS rules. Modules can implement embedder APIs so that modules lower than them can call them. Examples of these APIs are the WebKit API and the Content API.
Content API
The Content API is how code in content can indirectly call
Chrome. Where possible, Chrome features try to hook in by filtering IPCs and
listening to events per How to Add New Features (without bloating
RenderView/RenderViewHost/WebContents).
When there isn't enough context (i.e. callback from WebKit) or when the
callback is a one-off, we have a ContentClient interface that the embedder
(Chrome) implements. ContentClient is available in all processes. Some
processes also have their own callback API as well, i.e.
ContentBrowserClient/ContentRendererClient/ContentPluginClient.
Status and Roadmap
The current status is content doesn't depend on chrome at all (see the meta
bug and all bugs
it depends on). We now have a basic browser built on top of content
("content_shell") that renders pages using content on all platforms. This
allow developers working on the web platform and core code to only have to
build/test content, instead of all of chrome.
We have a separate target for content's unit tests in content_unittests, and
integration tests in content_browsertests.
content is build at a separate dll to speed up the build.
We've created an API around content, similar to our WebKit API. This isolates
embedders from content's inner workings, and makes it clear to people working on
content which methods are used by embedders.