Split the Alloy runtime into bootstrap and style components. Support
creation of Alloy style browsers and windows with the Chrome runtime.
Chrome runtime (`--enable-chrome-runtime`) + Alloy style
(`--use-alloy-style`) supports Views (`--use-views`), native parent
(`--use-native`) and windowless rendering
(`--off-screen-rendering-enabled`).
Print preview is supported in all cases except with windowless rendering
on all platforms and native parent on MacOS. It is disabled by default
with Alloy style for legacy compatibility. Where supported it can be
enabled or disabled globally using `--[enable|disable]-print-preview` or
configured on a per-RequestContext basis using the
`printing.print_preview_disabled` preference. It also behaves as
expected when triggered via the PDF viewer print button.
Chrome runtime + Alloy style behavior differs from Alloy runtime in the
following significant ways:
- Supports Chrome error pages by default.
- DevTools popups are Chrome style only (cannot be windowless).
- The Alloy extension API will not supported.
Chrome runtime + Alloy style passes all expected Alloy ceftests except
the following:
- `DisplayTest.AutoResize` (Alloy extension API not supported)
- `DownloadTest.*` (Download API not yet supported)
- `ExtensionTest.*` (Alloy extension API not supported)
This change also adds Chrome runtime support for
CefContextMenuHandler::RunContextMenu (see #3293).
This change also explicitly blocks (and doesn't retry) FrameAttached
requests from PDF viewer and print preview excluded frames (see #3664).
Known issues specific to Chrome runtime + Alloy style:
- DevTools popup with windowless rendering doesn't load successfully.
Use windowed rendering or remote debugging as a workaround.
- Chrome style Window with Alloy style BrowserView (`--use-alloy-style
--use-chrome-style-window`) does not show Chrome theme changes.
To test:
- Run `ceftests --enable-chrome-runtime --use-alloy-style
[--use-chrome-style-window] [--use-views|--use-native]
--gtest_filter=...`
- Run `cefclient --enable-chrome-runtime --use-alloy-style
[--use-chrome-style-window]
[--use-views|--use-native|--off-screen-rendering-enabled]`
- Run `cefsimple --enable-chrome-runtime --use-alloy-style [--use-views]`
Frame identifiers have changed from int64_t to string type. This is due
to https://crbug.com/1502660 which removes access to frame routing IDs
in the renderer process. New cross-process frame identifiers are 160-bit
values (32-bit child process ID + 128-bit local frame token) and most
easily represented as strings. All other frame-related expectations and
behaviors remain the same.
Write access to the shared memory region is required because JavaScript lacks
the capability to create read-only ArrayBuffers. When a user attempts to modify
an ArrayBuffer that utilizes a ReadOnlySharedMemoryRegion as its BackingStore
it triggers an access violation.
Note that this pull request may be reverted in the future if JavaScript adds
read-only ArrayBuffer support.
The CefFrame::CreateURLRequest method is no longer supported in the renderer
process. Usage of this method was already limited to same-origin requests due
to renderer process CORS restrictions, and the underlying Blink API has now
been removed in https://crbug.com/1413912 (M112+).
Existing alternatives include CefURLRequest usage in the browser process, or
JavaScript XMLHttpRequest/fetch API usage in the renderer process.
Send a FrameDetached message from CefFrameHostImpl::Detach before closing
the RenderFrame connection to avoid an immediate reconnect attempt by the
renderer.
When BFCache is disabled the intentionally detached frame will never be
reconnected. When BFCache is enabled the intentionally detached frame will
be reconnected via CefFrameImpl::OnWasShown if/when it exits the cache.
Run with "--vmodule=frame_*=1 --enable-logging=stderr --no-sandbox" on the
command-line to see these log messages from both the browser and renderer
processes.
CefSharedProcessMessageBuilder supports creation of a CefProcessMessage
backed by a CefSharedMemoryRegion.
Performance tests comparing the existing ArgumentList approach and the new
SharedMemoryRegion approach have been added to cefclient at
http://tests/ipc_performance.
CefMessageRouter has been updated to use SharedMemoryRegion as transport
for larger message payloads. The threshold is configurable via
|CefMessageRouterConfig.message_size_threshold|.
To test:
run `ceftests --gtest_filter=SendSharedProcessMessageTest.*:SharedProcessMessageTest.*:MessageRouterTest.Threshold*`
When bfcache is disabled use the RenderFrameObserver::DidFinishLoad notification
instead of the DidStopLoading notification that originates from the browser
process. During fast navigation the DidStopLoading notification may arrive in
the renderer after a new navigation has commenced and cause the wrong regions to
be reported.
With same-site BFCache enabled every navigation can now potentially be served
via the BFCache. To support this internally a new top-level RenderFrame object
may be created for each new navigation. As a result, OnBrowserCreated may now
be called multiple times with the same browser ID in a given renderer process
(a behavior previously only seen with cross-site navigations and different
renderer processes).
BFCache navigations do not trigger the same Chromium notifications as a normal
load. To avoid breaking CEF API usage expectations we now synthetically
generate the load-related callbacks that would otherwise be missing
(OnLoadingStateChange with isLoading=true, OnLoadStart, OnLoadEnd). The
|httpStatusCode| argument to OnLoadEnd will be 0 in this case.
To test:
- Run `FrameHandlerTest.*:MessageRouterTest.*:NavigationTest.*`
- Run `NavigationTest.LoadSameOriginLoadURL` for OnBrowserCreated behavior.
- Run `NavigationTest.History` for load-related callback behavior.
The mojo channel used for frame communication may disconnect for a variety of
reasons including frame navigation, frame destruction, or insertion into the
BackForwardCache (when the browser-side frame representation is destroyed and
closes the connection). When disconnect occurs we now evaluate the situation
and reconnect if appropriate.
Connections are now initiated solely from the renderer process and the
RenderFrame is passed as an argument to FrameAttached() instead of being
retrieved independently. Messages are queued while the frame is disconnected
and sent only after FrameAttachedAck() is received from the browser process.
The renderer process will be crashed intentionally with a "connection retry
failure" message if the reconnect fails 3 times in a row.
When BackForwardCache is enabled and the user navigates the main frame
back/forward a new RFH may be created for an existing main frame GlobalId value
and CefFrameHostImpl (e.g. an object that was previously Detach()ed after main
frame navigation called SetMainFrame, but for which RenderFrameDeleted was not
subsequently called due to insertion in the BackForwardCache). In this case we
can re-attach the new RFH to the existing main frame CefFrameHostImpl in
RenderFrameHostStateChanged and resume processing of messages.
Swapping back/forward to an existing (already loaded) renderer does not trigger
new notifications for draggable regions (e.g. RenderFrameObserver::
DraggableRegionsChanged is not called by default). We therefore explicitly
request an update of draggable regions by sending the DidStopLoading message to
the renderer.
A new |reattached| parameter is added to CefFrameHandler::OnFrameAttached to
support identification of BackForwardCache usage by the client.
To test with unit tests:
Run `ceftests --gtest_filter=DraggableRegionsTest.DraggableRegionsCrossOrigin
--enable-features=BackForwardCache`
To test manually:
1. Run `cefclient --enable-features=BackForwardCache --use-views
--url=http://tests/draggable`, note that draggable regions work.
2. Load https://www.google.com via the address bar, note that draggable regions
are removed.
3. Go back to http://tests/draggable, note that draggable regions work.
4. Go forward to https://www.google.com, note that draggable regions are
removed.
See the new cef_frame_handler.h for complete usage documentation.
This change includes the following related enhancements:
- The newly added CefBrowser::IsValid method will return false (in the browser
process) after CefLifeSpanHandler::OnBeforeClose is called.
- CefBrowser::GetMainFrame will return a valid object (in the browser process)
until after CefLifeSpanHandler::OnBeforeClose is called.
- The main frame object will change during cross-origin navigation or
re-navigation after renderer process termination. During that time,
GetMainFrame will return the new/pending frame (in the browser process) and
any messages that arrive for the new/pending frame will be correctly
attributed in OnProcessMessageReceived.
- Commands to be executed in the renderer process that may fail during early
frame initialization (ExecuteJavaScript, LoadRequest, etc.) will now be
queued until after the JavaScript context for the frame has been created.
- Logging has been added for any commands that are dropped because they arrived
after frame detachment.
A reference to a received CefProcessMessage object and/or associated argument
list can now be kept outside of the OnProcessMessageReceived callback. The
argument list is no longer explicitly owned by the CefProcessMessage object
and can be individually assigned to other CefValue types as needed (e.g. by
passing to SetList, etc). Depending on client usage this could reduce the
potential for unnecessary copies of the list contents.
Received messages can also be sent back using SendProcessMessage (after which
the CefProcessMessage would become invalid as discussed in issue #3123). This
is not new behavior but we have now added explicit unit test coverage for it.
This also no longer requires a copy of the argument list contents.
Note that a received argument list is initially read-only for logical
consistency. Assignment to another CefValue object could potentially remove
the read-only status because it is not an intrinsic property of the underlying
Chromium data type. This is fine because, at that point, ownership has been
transfered to the new CefValue object and the original logical context (as
part of the CefProcessMessage) no longer applies.
This change introduces a few minor CEF API behavior changes:
- A CefProcessMessage object cannot be reused after being passed to
SendProcessMessage.
- The |extra_info| argument to CefRenderProcessHandler::OnBrowserCreated may
now be NULL.
Where appropriate, we now utilize the default UTF string encoding format and
shared memory to reduce copies and conversions for the cross-process
transfer of arbitrary-length strings. For example, CefFrame::GetSource/GetText
now involves zero UTF conversions and zero copies in the browser process for
the CefString delivered to CefStringVisitor::Visit().
This is the first pass in removing direct dependencies on the Alloy
runtime from code that can potentially be shared between runtimes.
CefBrowserHost and CefRequestContext APIs (including CefCookieManager,
CefURLRequest, etc.) are not yet implemented for the Chrome runtime.
Assert early if these API methods are called while the Chrome runtime
is enabled.
As part of introducing the Chrome runtime we now need to distinguish
between the classes that implement the current CEF runtime and the
classes the implement the shared CEF library/runtime structure and
public API. We choose the name Alloy for the current CEF runtime
because it describes a combination of Chrome and other elements.
Shared CEF library/runtime classes will continue to use the Cef
prefix. Classes that implement the Alloy or Chrome runtime will use
the Alloy or Chrome prefixes respectively. Classes that extend an
existing Chrome-prefixed class will add the Cef or Alloy suffix,
thereby following the existing naming pattern of Chrome-derived
classes.
This change applies the new naming pattern to an initial set of
runtime-related classes. Additional classes/files will be renamed
and moved as the Chrome runtime implementation progresses.
The PDF loading documentation in extension_system.cc has be updated to
describe the new code paths.
To support delivery of input events to the mime handler renderer process it is
now necessary to route events via the correct RWHV interface. For Aura-based
platforms (Windows/Linux) this means RWHVAura::On*Event and for macOS this
means RWHVMac::RouteOrProcess*Event. Since Aura uses UI event types these have
become the source of truth on Aura-based platforms with conversion to Web event
types when needed (primarily for OSR).
This change also adds a timeout for CefProcessHostMsg_GetNewBrowserInfo to
avoid a hung renderer process if the guest WebContents route is not
registered via CefMimeHandlerViewGuestDelegate::OnGuestDetached as expected
prior to CefBrowserInfoManager::OnGetNewBrowserInfo being called. This
timeout can be disabled for testing purposes by passing the
`--disable-new-browser-info-timeout` command-line flag.
The `--disable-features=MimeHandlerViewInCrossProcessFrame` command-line
flag can be used for a limited time to restore the previous implementation
based on BrowserPlugin. That implementation will be deleted starting with
the 3897 branch update.
Known issues:
- ExecuteJavaScript calls on the frame hosting the PDF extension will not
be routed to the mime handler renderer process.
- The PDF extension will not load successfully if blocked by
ChromePluginPlaceholder and then manually continued via the "Run this
plugin" context menu option (see https://crbug.com/533069#c41).
This change moves the SendProcessMessage method from CefBrowser to CefFrame and
adds CefBrowser parameters to OnProcessMessageReceived and
OnDraggableRegionsChanged.
The internal implementation has changed as follows:
- Frame IDs are now a 64-bit combination of the 32-bit render_process_id and
render_routing_id values that uniquely identify a RenderFrameHost (RFH).
- CefFrameHostImpl objects are now managed by CefBrowserInfo with life span tied
to RFH expectations. Specifically, a CefFrameHostImpl object representing a
sub-frame will be created when a RenderFrame is created in the renderer
process and detached when the associated RenderFrame is deleted or the
renderer process in which it runs has died.
- The CefFrameHostImpl object representing the main frame will always be valid
but the underlying RFH (and associated frame ID) may change over time as a
result of cross-origin navigations. Despite these changes calling LoadURL on
the main frame object in the browser process will always navigate as expected.
- Speculative RFHs, which may be created as a result of a cross-origin
navigation and discarded if that navigation is not committed, are now handled
correctly (e.g. ignored in most cases until they're committed).
- It is less likely, but still possible, to receive a CefFrame object with an
invalid frame ID (ID < 0). This can happen in cases where a RFH has not yet
been created for a sub-frame. For example, when OnBeforeBrowse is called
before initiating navigation in a previously nonexisting sub-frame.
To test: All tests pass with NetworkService enabled and disabled.
Requests created using CefURLRequest::Create are not associated with a
browser/frame. When originating from the render process these requests cannot be
intercepted and consequently only http(s) and blob requests are supported. To
work around this limitation a new CefFrame::CreateURLRequest method has been
added that allows the request to be associated with that browser/frame for
interception purposes.
This change also fixes an issue with the NetworkService implementation where
redirected requests could result in two parallel requests being sent to the
target server.
To test: URLRequestTest.* tests pass with NetworkService enabled.
Marshall Greenblatt
Nik Pavlov
Alexander Guettler
Daniel Nitsche
Petra Öhlin