Right now, `<webview>` is the only way to embed additional content in a
`BrowserWindow`. Unfortunately `<webview>` suffers from a [number of
problems](https://github.com/electron/electron/issues?utf8=%E2%9C%93&q=is%3Aissue%20is%3Aopen%20label%3Awebview%20).
To make matters worse, many of these are upstream Chromium bugs instead
of Electron-specific bugs.
For us at [Figma](https://www.figma.com), the main issue is very slow
performance.
Despite the upstream improvements to `<webview>` through the OOPIF work, it is
probable that there will continue to be `<webview>`-specific bugs in the
future.
Therefore, this introduces a `<webview>` alternative to called `BrowserView`,
which...
- is a thin wrapper around `api::WebContents` (so bugs in `BrowserView` will
likely also be bugs in `BrowserWindow` web contents)
- is instantiated in the main process like `BrowserWindow` (and unlike
`<webview>`, which lives in the DOM of a `BrowserWindow` web contents)
- needs to be added to a `BrowserWindow` to display something on the screen
This implements the most basic API. The API is expected to evolve and change in
the near future and has consequently been marked as experimental. Please do not
use this API in production unless you are prepared to deal with breaking
changes.
In the future, we will want to change the API to support multiple
`BrowserView`s per window. We will also want to consider z-ordering
auto-resizing, and possibly even nested views.
Embedding arbitrary web content is problematic when it comes to keyboard
shortcuts because:
* Web content can steal app shortcuts (see e.g. brave/browser-laptop#4408)
* Blocked web content (e.g. a focused <webview> performing expensive
computation) will also prevent app shortcuts from firing immediately
The new before-input-event event can be used to overcome these issues by
always handle certain keyboard events in the main process.
Note that this requires electron/brightray#261 to compile.
The sandbox option allows multiple webContents in one renderer process, so using
the only the renderer id to identify WebContents instances is no longer an
option.
WebContents::GetID now returns a 64-bit integer, which is composed of both the
process id(high 32), and the RenderViewHost routing id(low 32). Also add a
`GetProcessID` that retrieves the renderer process id, a requirement in some of
our javascript code.
- Allow `api::Window` instances to be created from existing `api::WebContents`.
- Override `WebContentsCreated` and `AddNewContents` to wrap renderer-created
`content::WebContents` into `api::WebContents`.
- For `content::WebContents` that should be displayed in new windows, pass the
wrapped `api::WebContents` object to window manager.
- Add an overload to `WebContents::CreateFrom` that accepts a type parameter. If
type is `REMOTE`, initialization is the same as before(a thin wrapper). If
not, the `api::WebContents` will be fully initialized, as if it was created by
`api::WebContents::Create`.
- Move common initialization code to `InitWithSessionAndOptions`.
