When an iframe is removed before its post-connection session history
steps finish processing, the queued steps still run and call
navigate() on the now-destroyed navigable. This calls
set_delaying_load_events(true), creating a DocumentLoadEventDelayer
on the parent document. Since the navigable is destroyed, no
finalize step ever runs to clear the delayer, permanently blocking
the parent document's load event.
Fix this by checking has_been_destroyed() at the start of
begin_navigation() and bailing out early.
When the visual viewport scrolls (e.g. panning while zoomed in), the
transform matrix baked into the AccumulatedVisualContext tree becomes
stale. Previously only set_needs_display(InvalidateDisplayList::No) was
called, which re-rendered with the old transform — making hit testing
and painting incorrect after visual viewport panning.
I noticed that when performing a smooth scroll up or down, we would
repeatedly enter `Navigable::perform_a_scroll_of_the_viewport()` when
the top or bottom (respectively) of the page was reached. Optimize this
case by returning early when we've reached the intended target scroll
position and we're not performing any vertical or horizontal scrolling.
In preparation for handling input events on the rendering thread, move
backpressure management to RenderingThread. The rendering thread needs
to manage this independently without querying the main thread.
Previously, the main thread would block when the UI process hadn't yet
released the backing surface. Now, the main thread can continue
producing display lists while the UI process is busy, allowing more work
to happen in parallel. When rasterization is slow and display lists are
produced faster than they can be consumed, presentation requests are
naturally coalesced using a flag-based approach -
multiple present_frame() calls simply update the pending state,
resulting in a single rasterization with the latest display list.
See: 976d272
This fixes a timeout in the included test. In order for the
test to run properly in our test harness it is modified slightly
(as we are loading the HTML from a file:// URL).
This is preparation for future work where the rendering thread will
initiate rasterization independently and notify the UI process without
requiring coordination with the main thread.
This change prepares for a future where the rendering thread handles
input events directly, allowing it to trigger repainting without
waiting for the main thread. To support this, the compositor needs to
own the display list, scroll state, and backing stores rather than
receiving them per-frame from the main thread.
Previously, when loading a document, we would try to sniff the MIME
type by reading from the response body's source. However, for streaming
HTTP responses, the body source is Empty (the data comes through the
stream instead), so we had no bytes to sniff.
This caused pages like hypr.land (which sends no Content-Type header)
to be misidentified as plain text instead of HTML, since the MIME
sniffing algorithm would receive zero bytes and fall back to the
default type.
The fix captures the first bytes of the response body during fetch,
storing them on the Body object. These bytes are the "resource header"
defined by the MIME Sniffing spec - up to 1445 bytes, which is enough
to identify any MIME type the spec can detect.
Since bytes may arrive asynchronously during streaming, we use a
callback mechanism: if bytes aren't ready yet when load_document()
needs them, it registers a callback that fires once enough bytes have
been captured (or the stream ends).
The flow is:
1. FetchedDataReceiver receives network bytes, buffers them
2. When Body is created, buffered bytes are flushed to Body's sniff
buffer, and subsequent bytes are appended as they arrive
3. Before calling load_document(), Navigable waits for sniff bytes
4. load_document() passes the bytes to MimeSniff::Resource::sniff()
When scrolling with a visual viewport offset (from pinch-to-zoom),
scroll_viewport_by_delta() was passing m_viewport_scroll_offset + delta
to perform_a_scroll_of_the_viewport(). However, that function calculates
the scroll delta as `position - page_top()`, where page_top() includes
the visual viewport offset. This caused the effective scroll delta to be
reduced by the visual offset amount.
Fix by using the current page position (which includes the visual
offset) as the base for the delta calculation.
Regression from 0a57e1e8ac.
Cache the scroll state snapshot in ViewportPaintable when
refresh_scroll_state() is called.
The upcoming AccumulatedVisualContext integration requires access to
the scroll state snapshot during hit testing to transform screen
coordinates through scroll frames. Without caching, each hit test would
allocate a new snapshot (a Vector<Entry>), causing many temporary
allocations during mouse movement. Caching the snapshot eliminates this
overhead.
A version of this was added in a610639119
and reverted in 70671b4c11. The bugs
there (confusing scroll-to-position and scroll-by-delta, and not having
an execution context in some cases) have been fixed in this version.
A lot of our scrolling code is quite old, and doesn't match the spec,
but does use some similar names. This is quite confusing. In particular
`perform_scroll_of_viewport()` is not the same as the spec algorithm.
That algorithm is actually almost implemented in
`scroll_viewport_by_delta()`.
To clarify things, this commit makes a few changes:
- Rename perform_scroll_of_viewport() to
perform_scroll_of_viewport_scrolling_box(). This is a better match
for how we use this method, even if it's not actually a match for the
algorithm. (:yakbait:)
- Move `scroll_viewport_by_delta()`'s code into a new
`perform_a_scroll_of_the_viewport()` method, and make it take a
position like it should. `scroll_viewport_by_delta()` now calls it
with a calculated position.
I've avoided reusing the original `perform_scroll_of_viewport()` name to
avoid accidents.
This function essentially performs a BFS traversal over document states.
With this change, we let `doc_states` grow instead of removing traversed
states, avoiding shifting elements on every iteration.
This reduces `./test-web -j 1` from ~7m to ~5m on my machine.
Fixes crashing introduced in a610639 when `scroll_viewport_by_delta()`
is called from `EventHandler::handle_mousewheel()` and there's no
running execution context to grab current realm from to allocate a
promise.
Stops this WPT test from crashing:
navigation-api/ordering-and-transition/transition-to.html
...it now times out instead, so more work is needed before importing.
The end goal here is for LibHTTP to be the home of our RFC 9111 (HTTP
caching) implementation. We currently have one implementation in LibWeb
for our in-memory cache and another in RequestServer for our disk cache.
The implementations both largely revolve around interacting with HTTP
headers. But in LibWeb, we are using Fetch's header infra, and in RS we
are using are home-grown header infra from LibHTTP.
So to give these a common denominator, this patch replaces the LibHTTP
implementation with Fetch's infra. Our existing LibHTTP implementation
was not particularly compliant with any spec, so this at least gives us
a standards-based common implementation.
This migration also required moving a handful of other Fetch AOs over
to LibHTTP. (It turns out these AOs were all from the Fetch/Infra/HTTP
folder, so perhaps it makes sense for LibHTTP to be the implementation
of that entire set of facilities.)
The spec declares these as a byte sequence, which we then implemented as
a ByteBuffer. This has become pretty awkward to deal with, as evidenced
by the plethora of `MUST(ByteBuffer::copy(...))` and `.bytes()` calls
everywhere inside Fetch. We would then treat the bytes as a string
anyways by wrapping them in StringView everywhere.
We now store these as a ByteString. This is more comfortable to deal
with, and we no longer need to continually copy underlying storage (as
ByteString is ref-counted).
This work is largely preparatory for an upcoming HTTP header refactor.
If multiple cross-document navigations are queued on
SessionHistoryTraversalQueue, running the next entry before the current
document load is finished may result in a deadlock. If the new document
has a navigable element of its own, it will append steps to SHTQ and
hang in nested spin_until.
This change uses promises to ensure that the current document loads
before the next entry is executed.
Fixes timeouts in the imported tests.
Co-authored-by: Sam Atkins <sam@ladybird.org>
NavigationObserver register itself in Navigable from constructor and
unregister itself from `finalize()`. The problem is that `finalize()`
won't be invoked for as long as NavigationObserver is visited by
Navigable, leading to GC leaks.
Instead of painting DecodedImageData by first asking it for a bitmap
and then painting that, this commit adds two new APIs:
- frame_rect(frame_index):
Gets the size of the animation frame at the given index.
- paint(context, ...):
Paints the DecodedImageData into a DisplayListRecordingContext.
The main powerful thing here is that this allows SVGDecodedImageData
to render itself using the GPU when available.
Previously, we enqueued a task on the main thread's event loop to
execute the callback. This meant that even though the rendering thread
had finished producing the next frame, there was still a delay before
the main thread notified the UI process.
This change makes the rendering thread execute the callback directly.
This should be safe, as the only pointer captured by the callback is the
traversable `PageClient`, which is expected to remain alive for as long
as the rendering thread exists. The callback then invokes either
`page_did_paint()` or `page_did_take_screenshot()`, both of which
enqueue an IPC message, which is safe to do since `SendQueue` is
protected by a mutex.
Implements spec algorithm for viewport scrolling that first checks if
it's possible to use delta to move the visual viewport before falling
back to scrolling the layout viewport. This is a part of pinch-to-zoom
support.
This also fixes a bug in the view transitions code that was
required to get the imported test to pass. The code for setting
the initial containing block size just did not set the right thing,
since doing so would trigger an error later on.
That later error resulted from walking up the tree, without
considering that the document element has a parent that is not
itself an element. (and then doing element things to it)
Partly corresponds to 80ebad5fbf
This is mostly to handle null source_documents, which is something that
needs more work elsewhere. The spec change above is about the deferred
fetch quota.
