Our existing coalescing mechanism for input events didn't prevent
multiple mousemove/mousewheel events from being processed between paint
cycles. Since handling these events can trigger style & layout updates
solely for hit-testing purposes, we might end up doing work that won't
be observable by a user and could be avoided by shceduling input events
processing to happen right before painting the next frame.
Before this change, tasks associated with a destroyed document would get
stuck in the task queue forever, since document-associated tasks are not
allowed to run when their document isn't fully active (and destroyed
documents never become fully active again). This caused everything
captured by task callbacks to leak.
We now treat tasks for destroyed documents as runnable immediately,
which gets them out of the queue.
This fixes another massive GC leak on Speedometer.
Before this change, `m_needs_repaint` was reset in
`Document::record_display_list()` only when the cached display list was
absent. This meant that if the last triggered repaint used the cached
display list, we would keep repainting indefinitely until the display
list was invalidated (We schedule a task that checks if repainting is
required 60/s).
This change also moves `m_needs_repaint` from Document to
TraversableNavigable as we only ever need to repaint a document that
belongs to traversable.
Previously, we were generating timestamps relative to the current time
of the monotonic clock. We now generate timestamps relative to the
event loop's last render opportunity time, per the spec.
For example, running `alert(1)` will pause the event loop, during which
time no JavaScript should execute. This patch extends this disruption to
microtasks. This avoids a crash inside the microtask executor, which
asserts the JS execution context stack is empty.
This makes us behave the same as Firefox in the following page:
<script>
queueMicrotask(() => {
console.log("inside microtask");
});
alert("hi");
</script>
Before the aforementioned assertion was added, we would execute that
microtask before showing the alert. Firefox does not do this, and now
we don't either.
This makes it more convenient to use the 'relvant agent' concept,
instead of the awkward dynamic casts we needed to do for every call
site.
mutation_observers is also changed to hold a GC::Root instead of raw
GC::Ptr. Somehow this was not causing problems before, but trips up CI
after these changes.
Instead of collecting all documents in a big vector and then filtering
the vector (twice!) with remove_all_matching(), we now pass a filter
callback to documents_in_this_event_loop_matching() and avoid all the
extra shuffling work.
Saw this stuff hogging ~20% of CPU time when profiling a WebContent
process in the middle of a WPT run.
The WebSocket spec tells us to queue tasks instead of firing events
synchronously at WebSockets, so this commit does exactly that.
The way we've implemented web sockets means that the work is spread
across multiple libraries and even processes, which is why it doesn't
look like the spec verbatim.
Resulting in a massive rename across almost everywhere! Alongside the
namespace change, we now have the following names:
* JS::NonnullGCPtr -> GC::Ref
* JS::GCPtr -> GC::Ptr
* JS::HeapFunction -> GC::Function
* JS::CellImpl -> GC::Cell
* JS::Handle -> GC::Root
Now that the heap has no knowledge about a JavaScript realm and is
purely for managing the memory of the heap, it does not make sense
to name this function to say that it is a non-realm variant.
