We now replace the current history entry if the page-load has been
caused because of a redirect. This makes it able to traverse the
history if one of the entries redirects you, which previously
caused an infinite history traversion loop.
Depends on https://github.com/SerenityOS/serenity/pull/16004
Similar to https://github.com/SerenityOS/serenity/commit/9782660. Unlike
Serenity's browser, this doesn't affect reloading the page, as Ladybird
refers to the History object for reloading (which is updated already on
page load). However, this URL is used for e.g. crash reporting, so let's
update it here as well.
WebContent now needs to interact with these dialogs asynchronously. This
updates WebContentView to hold a pointer to whatever dialog is open, and
implements the methods to interact with that dialog.
This adds a WebDriver binary for Ladybird to make use of Serenity's
WebDriver implementation. This has to use the same IPC socket handling
that was used to make WebContent work out-of-process. Besides that, we
are able to reuse almost everything from Serenity.
The WebDriver will pass the --webdriver-fd-passing-socket command line
option when it launches Ladybird. Forward this flag onto the WebContent
process, where it will create the WebDriverConnection for IPC.
Rather than needing to set another environment variable for WebDriver's
passing socket, let's forward these FDs by command line. This also moves
the creation of the WebContent connection to a helper function so that
the WebDriver connection can re-use it.
Previously, reloading went back to the first page loaded by
WebView::load() or WebView::load_html(), as they are the only methods
that modify m_url, which is what the reload loaded. Now we handle
reloads in Tab.cpp by simply loading the last entry in the m_history.
The hackish initial loading of about:blank was previously added to the
history, so you could go back to it (which wasn't very ergonomic). Now
we set the m_is_history_navigation flag before loading it so it doesn't
get added to the history.
Steps to reproduce:
1. Open the Cookie Clicker game in a tab.
2. Open another website in another tab and make that the current tab.
3. Observe how the window's title mentions Cookie Clicker.
Previously we were always pushing to history on the on_load_start
callback. Now we only do that if we are NOT navigating through the
history navigation (loading pages by going back/forward). This is what
the SerenityOS browser does:^)
We now emit a new signal for backward mouse button's mouseup and forward
mouse button's mouseup which is handled by going back and forward in the
history respectively:))
There's no point in busy-waiting for the condition to come true.
By passing the `WaitForMoreEvents` flag to `processEvents()`, we allow
Qt to block until it has something for us to react to.
This was extremely noticeable when waiting for large resources to
finish loading.
The slowdown is sometimes 5x, possibly more.
This is trivially confirmed by adding a large JS file to a page and
comparing the load time with a simple wget.
- Add Qt::Core, Qt::Gui, LibGfx, LibIPC, and LibJS to the ladybird
target, remove LibGL, LibSoftGPU, and LibWebSocket
- Add LibJS to the WebContent target, remove LibWebView
- Order them properly :^)
Regressed in https://github.com/SerenityOS/serenity/pull/15746.
Fixes#108.
There are no custom changes for Ladybird in the current copies of those
files, so we just need to ensure to keep Ladybird up to date for any
changes made upstream.
There are no custom changes for Ladybird in the current copies of those
files, so we just need to ensure to keep Ladybird up to date for any
changes made upstream.
This fixes a build issue introduced by https://github.com/SerenityOS/serenity/pull/15736.
Qt can wrap any number of cookies into a single Set-Cookie header in the
network responses it gives us. We now use the QNetworkReply::header()
API to get a "cooked" list of the cookies, and then rewrap them in a
format suitable for LibWeb.
Sites that send multiple Set-Cookie headers in one response now work
a lot better. :^)
These didn't work, for two reasons:
1. Qt swallows all Tab key presses by default. We have to override
the event() function in order to receive them.
2. Qt transforms Shift+Tab into a fake "Backtab" key. We have to
undo this transformation and send Shift+Tab to WebContent.
When spawning a WebContent process, we have to close the file
descriptors belonging to the "other side" in both processes, or they
will not get naturally "cleaned up" when one of the processes exits.
Fixes#93
Always call platform_init after there's a QApplication, because in the
installed configuration that's how we find the resources.
Try QCoreApplication::applicationDirPath() after looking in ./WebContent
for the WebContent process. In an installed configuration, ladybird and
WebContent will both be in $PREFIX/bin.
Add install rules for WebContent and its linked libraries, for if they
ever differ from ladybird's.
This causes CMake to output a WebContent build, without this it would
not build WebContent and Ladybird would be unusable since it couldn't
find the WebContent executable.
While this adds a fair bit of widget code, we're also increasing code
sharing by using the same bits in WebContentClient for interacting with
the JS console.
That said, we should look for more ways to share code here.
This patch brings over the WebContent process over from SerenityOS
to Ladybird, along with a new WebContentView widget that renders
web content in a separate process.
There's a lot of jank and FIXME material here, notably I had to re-add
manually pumped Core::EventLoop instances on both sides, in order to get
the IPC protocol running. This introduces a lot of latency and we should
work towards replacing those loops with improved abstractions.
The WebContent process is built separately here (not part of Lagom) and
we provide our own main.cpp for it. Like everything, this can be better
architected, it's just a starting point. :^)
This prevents memory leaks detected by both Valgrind and ASAN/LSAN.
Valgrind is still suspicious of the leaked JS::VM from
Web::Bindings::main_thread_vm() but there's other issues with leak
checking all the GC'd objects.
Co-Authored-By: Diego Iastrubni <diegoiast@gmail.com>
For the first cut, the file path is not configurable and the content
filter cannot be toggled on or off. If we fail to apply the content
filters for any reason (e.g. the filter file doesn't exist), we simply
just stop loading the content filters to allow using Ladybird without
content filters.
We use a ModelTranslator to expose a DOMTreeModel from LibWebView :^)
It allows you to select the currently inspected node, which causes
the engine to render a little box model overlay above the web content.
This will allow us to share code with LibWebView from SerenityOS.
(This would otherwise not work, since its "WebView" namespace collides
with our "WebView" class.)
Also, we should eventually move towards a more sophisticated
multi-process WebView like OOPWV.
There are a lot of unsupported mouse click events that the engine
cannot handle. It such event (0) reaches the web engine - it will
assert.
Don't even propagate them - this is the safe way. As of today!
I also added back/forward buttons to the translation.
Should fix#27
This commit changes how we set the back and forward button key bindings
to use platform-specific standard key sequences.
For example, in Mac OS X, the back action will be now triggered via
Cmd+← and Cmd+[, whereas previously the action was mapped to Alt+←,
which is not standard in Mac OS X.
This setup should allow the package maintainers who are looking to
distribute ladybird on their distributions to use CMake to install
ladybird using cmake install rules rather than having to write their own
Reorganize the logic for initializing s_serenity_resource_root.
Now, we initialize it in platform_init(), and move platform_init to the
top of initialize_web_engine.
Add a branch at the end of the function to check
``QApplication::applicationDirPath()`` for the location of the
executable, and base the location of resources on that.
In an installed configuration, this will be /some/path/bin, with the
resource root set to /some/path/share/, looking for files in
/some/path/share/res/resource-type.
This matches up with some upcoming CMake changes to install resources in
CMAKE_INSTALL_DATADIR.
In the future, ladybird should probably use a QOpenGLWidget or similar
platform plugin to use the native GL implementation instead of the one
in serenity.
This allows requests like dumping the DOM tree to be forwarded to the
correct tab. Otherwise, the event would be forwarded to the most
recently created tab.
This patch removes the dual-event-loop setup, leaving only the Qt event
loop. We teach LibWeb how to drive Qt by installing an EventLoopPlugin.
This removes the 50ms latency on all UI interactions (and network
requests, etc.)
This handles most (?) of keyboard input. For some reason, "Ctrl+A"
and Enter are not working on google.com.
It can handle plain ASCII, I tested also Hebrew input,
and https://playbiolab.com/ (which is playable now!)
Build an Android APK file that, when configured properly in Qt Creator,
can be used to deploy the browser to an Android device.
The current build requires NDK 24, targets no less than Android API 30,
and Qt Creator 6.4.0.
- Silences the -Wuser-defined-literals warning which is triggered by our
use of the `sv` suffix for StringView
- Removes an unused captured `this` pointer [-Wunused-lambda-capture]
- Changes a JSONArray.h include to JSONObject.h to get the definition
for `JSONValue::serialize`. This is needed because template functions
are not exported for dylibs on macOS. This is a hack; the JSON headers
should be refactored so that each one includes the definition of
the template functions it sees. -- Maybe we should build with
-fvisibility-inlines-hidden on Linux to catch issues like this?
This gives the actual webcontent more space to work with,
it also emulates how other browsers does it.
In the future we'd like to do something else since only ToolTip
can be visible at the same time.
This patch takes the browser history code from the Serenity browser and
wires it up to the QT interface. This is tied in with a few extra
toolbar buttons associated with each tab.
This patch removes the browser WebView from the window and places it
inside a Tab object, all wrapped up in a QT tab control. So far you can
create tabs, but can't close them.
Until we can get our own RequestServer infrastructure up and running,
running the TLS and HTTP code in-process was causing lots of crashes
due to unexpected reentrancy via nested event loops.
This patch adds a simple backend for HTTP and HTTPS requests that simply
funnels them through QNetworkAccessManager.
This patch adds an event handler to the main window which allows it to
respond to a user closing the window. This event is then passed on to
the LibCore event loop, which allows the application quit itself.
Previously the application would hang, only running in the background,
until killed by an external force.
On high-dpi systems QT will automatically scale up painting operations.
This results in an ugly blurry browser window surrounded by a nicely
scaled window frame. This patch applies an inverse scale to the WebView
widget, ensuring the painting and mouse events are lined up and draw
with a 1:1 pixel ratio with respect to the display's device pixels.
