Part of #39418. See that PR for a full description.
Moves:
- `read_json_from_file`
- `write_json_to_file`
- `IpcSendResult`
- `IpcSend`
Renames:
- `CoreResourceThreadPool` to `ThreadPool` (shorter and more
descriptive, as we use it for more than the core resource thread now)
Signed-off-by: Ashwin Naren <arihant2math@gmail.com>
Additionally also improve the warning message if the routed receiver
disconnects and exit the thread.
If the routed receiver disconnects, we can't receive any canvas messages
anymore, and any control messages can't remedy that, so we might as well
exit.
Testing: Channel changes are covered by existing tests. Exiting the
canvas thread if the routed thread disconnects is not tested, and needs
reviewer attention.
Part of https://github.com/servo/servo/issues/38912
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Refactor the background hang monitor channels to use GenericChannel.
Deserialization errors of `BackgroundHangMonitorControlMsg` are now
logged and ignored instead of causing a panic.
Testing: No major functional changes. Covered by BHM tests.
GenericChannel is also already widely used in servo.
Part of #38912
---------
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Previously, each ScriptThread was creating a new image cache with a
separate thread pool. These changes add an image cache to the
constellation and create new image caches by calling the
`create_new_image_cache` method. It reuses the original image cache's
thread pool and reduces the number of spawned threads in the
single-process mode.
Testing: Tested manually, using `ps -M` to see the number of spawned
threads with multiple tabs open in servoshell before and after these
changes.
Fixes: #37770
Signed-off-by: Rodion Borovyk <rodion.borovyk@gmail.com>
This should be the final PR for the Hash Function series that is
trivial.
Of note: I decided to transform `HashMapTracedValues<Atom,..>` to use
FxBuildHasher. This is likely not going to improve performance as Atom's
already have a unique u32 that is used as the Hash but it safes a few
bytes for the RandomState that is normally in the HashMap.
Signed-off-by: Narfinger <Narfinger@users.noreply.github.com>
Testing: Hash function changes should not change functionality, we
slightly decrease the size and unit tests still work.
Signed-off-by: Narfinger <Narfinger@users.noreply.github.com>
There are many important changes here:
- Generalize the presentation buffer into standalone staging buffers
that hold their own state. This allow them to be used by getImage.
- Move all clear handling to the ScriptThread and send the configuration
on each request present/getimage, thus avoiding any recreate/clearing
messages. This means that we prepare staging buffers lazily, on the
first request.
Try run for this change:
https://github.com/sagudev/servo/actions/runs/17341982368
Testing: This is covered by existing WebGPU CTS tests. There are some
bad expectations updates, but they are also on main (presumably from
last update the rendering work) although I think CTS is actually wrong
(see https://github.com/gpuweb/cts/issues/4440).
Fixes: #36820Fixes: #37705Fixes: #33368 (we now keep reference alive in hashmap)
---------
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
FNV is faster for hashing less than 16 bytes of data and the
cryptographic properties of the default HashMap are not needed for the
various ids.
Testing: This does not change functionality.
Signed-off-by: Narfinger <Narfinger@users.noreply.github.com>
There were still some accesses to the inner BrowsingContextId from the
WebViewId. This changes it to completely rely on the From trait for
these methods. This also means we can make the field private.
For testing we add a way to create arbitrary WebViewIds.
Testing: Does not change functionality.
Signed-off-by: Narfinger <Narfinger@users.noreply.github.com>
This PR **removes** `ScriptToConstellationMessage::ForwardToEmbedder`,
and replaces it with an explicit `ScriptToEmbedderChannel`. This new
channel is based on `GenericCallback` and in single-process mode will
directly send the message the to the embedder and wake it. In
multi-process mode, the message is routed via the ROUTER, since waking
is only possible from the same process currently. This means in
multi-process mode there are likely no direct perf benefits, since we
still need to hop the message over the ROUTER (instead of over the
constellation).
In single-process mode we can directly send the message to the embedder,
which should provide a noticable latency improvement in all cases where
script is blocked waiting on the embedder to reply.
This does not change the way the embedder receives messages - the
receiving end is unchanged.
## How was sending messages to the embedder working before?
1. Script wraps it's message to the embedder in
`ScriptToConstellationMessage::ForwardToEmbedder` and sends it to
constellation.
2. The [constellation event loop] receives the message in
[handle_request]
3. If deserialization fails, [an error is logged and the message is
ignored]
4. Since our message came from script, it is handle in
[handle_request_from_script]
5. The message is logged with trace log level
6. If the pipeline is closed, [a warning is logged and the message
ignored]
7. The wrapped `EmbedderMsg` [is forwarded to the embedder]. Sending the
message also invokes `wake()` on the embedder eventloop waker.
[constellation event loop]:
2e1b2e7260/components/constellation/constellation.rs (L755)
[handle request]:
2e1b2e7260/components/constellation/constellation.rs (L1182)
[an error is logged and the message is ignored]:
2e1b2e7260/components/constellation/constellation.rs (L1252)
[handle_request_from_script]:
https://github.com/servo/servo/blob/main/components/constellation/constellation.rs#L1590
[a warning is logged and the message ignored]:
2e1b2e7260/components/constellation/constellation.rs (L1599)
[is forwarded to the embedder]:
2e1b2e7260/components/constellation/constellation.rs (L1701)
Testing: Communication between Script and Embedder is extensive, so this
should be covered by existing tests.
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Migrate `ServiceWorkerMsg` and `SWManagerMsg` to GenericChannel
Testing: Covered by service worker wpt tests
Part of #38912
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
These changes add a custom servo:preferences URL that allows modifying
selected preferences at runtime. The goal of this work is to make it
easy to test pages while toggling experimental web platform features,
and support quickly changing the User-Agent header.
Testing: Manually verified that spacex.com loads correctly after
changing the user agent, and that https://polygon.io/ displays grid
elements correctly and no console errors with the experimental prefs
enabled.
Fixes: #35862
<img width="1136" height="880" alt="Screenshot 2025-07-18 at 1 06 23 AM"
src="https://github.com/user-attachments/assets/2d27c321-6ca0-43c3-a347-7bc4b55272df"
/>
---------
Signed-off-by: Josh Matthews <josh@joshmatthews.net>
This change ports all `EmbedderMsg` reply channels that don't use the
`ROUTER` to GenericChannel.
The remaining reply channels that use the router are blocked until
#38973 is merged.
This is a breaking change in the API between libservo and embedders.
Future work: A lot of the reply channels in this PR look like they
conceptually should be oneshot ipc channels. It might make sense to
provide a `OneshotGenericChannel` abstraction that encodes this.
Testing: No functional changes - covered by existing tests. None of the
channels changed here uses the Router
Part of #38912
---------
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Adds epoch to each WR image op command that is sent to compositor. The
renderer now has a `FrameDelayer` data structure that is responsible for
tracking when a frame is ready to be displayed. When asking canvases to
update their rendering, they are given an optional `Epoch` which denotes
the `Document`'s canvas epoch. When all image updates for that `Epoch`
are seen in the renderer, the frame can be displayed.
Testing: Existing WPT tests
Fixes: #35733
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
Co-authored-by: Martin Robinson <mrobinson@igalia.com>
This change was previously part of
fb1c0a4c48, which got reverted due to an
issue
with the compositor channel.
Split this change out into a separate PR, as it probably should have
been in the first place. Presumably it was one change before, since
serialization of crossbeam generic channels in single-process mode was
not implemented yet at the time.
Testing: Covered by existing tests. No custom callbacks involved.
Part of #38912
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Port the reply / back channels of StorageThreadMsg to GenericChannel.
Testing: No functional changes
Part of #38912
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Instead of doing font selection and text shaping in `canvas`, move this
to `script`. This allows canvas to use the shared `Document`
`FontContext`, which has access to web fonts. In addition, ensure that
there is a font style accessible for `OffscreenCanvas` in workers.
Testing: This causes a number of WPT tests to start to pass as web fonts
are
supported on canvas again. In addition, some start to fail as they
expose other
issues:
- The lack of support for the `Context2D.fontStretch` property
- Issues with zerosize gradient interpolation.
- Differences between quoted and unquoted font family names. This seems
like
a timing issue with the way we are handling web fonts. The test seems to
be
expecting Local fonts to be available immediately (without waiting for
them
to load). This isn't how Servo works ATM. Seems like an issue with the
test.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
In general, `raqote` is essentially umaintained and has issues with
quality (for instance text rendering has lots of issues) and removing it
finally lets us remove our dependency on `font-kit`. Although,
`vello_cpu` performance is not yet equal to raqote, rendering quality is
a lot better. It's expected that `vello` and `vello_cpu` performance
will keep improving.
Testing: This is covered by existing WPT tests.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
This change includes the following additions to GenericChannel:
- Add a GenericSend trait which is meant to replace the `IpcSend` trait
over time, as channels are migrated. For the time being this means, that
we often need to use `GenericSend::send()` to disambiguate from the
`IpcSend::send` function, until all usages of `IpcSend` have been
replaced.
- Add an OpaqueSender impl for GenericSender
- Add a profiled version of GenericChannel. The profiling is 1:1 the
same as for the existing profiled IPC channel, namely that only the
blocked time during `recv` is measured.
Testing: No functional changes, covered by existing tests
Part of #38912
---------
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
This change makes it so that every `GlobalScope` can contain an optional
`FontContext`. This will be necessary for every `GlobalScope` that
accesses canvas. Currently, `FontContext` is created and accessed via
the canvas worker thread, but this means that web fonts are not
available to canvas. This change will eventually make it possible for
canvas to share web fonts with the `Document` that owns them.
Testing: This should not change behavior and is thus covered by existing
tests.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
To help ensure our internal threads have shut-down before we deinit
Servo, the last thing the constellation should do, is sending the
`EmbedderMsg::ShutdownComplete`.
Testing: Manual testing by starting Servo and then closing the window.
Fixes: Part of https://github.com/servo/servo/issues/30849
---------
Signed-off-by: gterzian <2792687+gterzian@users.noreply.github.com>
Signed-off-by: Gregory Terzian <2792687+gterzian@users.noreply.github.com>
Co-authored-by: Martin Robinson <mrobinson@igalia.com>
This reverts commit fb1c0a4c48.
Previously in `create_compositor_channel`, the [routing callback][1] was
setup so that a message received on the Compositor's IPC receiver will
be
forwarded to the local receiver using the `CompositorProxy` which also
takes care of waking up the event loop. In #38782, this was changed so
that the routing callbacks simply forwards the message directly without
going via the `CompositorProxy`. This breaks behaviours that rely on the
event loop being woken up on message sending, e.g. updating image frames
for animated gifs.
Since the GenericChannel API doesn't allow custom routing callbacks,
revert this change until we figure out a better solution.
[1]:
d2ccce6052/components/servo/lib.rs (L1114)
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
To ensure a clean-shutdown of Servo, we need to shutdown our internal
threading before deinit. This shuts down the style thread pool either
using the constellation(in single-process mode), or when a content
process exits.
Testing: Manual testing by opening a window, opening and closing
multiple tabs, and closing the window, with a sleep added before deinit
and a sample taken to ensure the loose threads previously noticed are
gone.
Fixes: part of https://github.com/servo/servo/issues/30849
---------
Signed-off-by: gterzian <2792687+gterzian@users.noreply.github.com>
Ports the channel for WebDriverLoadStatus to GenericChannel.
Testing: No functional changes - Covered by existing webdriver tests
Part of #38912
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Migrates the namespace sender and receiver to use GenericChannel
Testing: Covered by existing tests
Part of #38912
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Besides migrating the channel to GenericChannel, this PR adds
`routed_channel_with_local_sender()` to `generic_channel`. This is for
existing use-cases, where we want to provide both an IPC capable
GenericSender, as well as a crossbeam Sender, for efficient sending if
the sender is in the same process.
Testing: All of our channels should send / receive at least some
messages during WPT tests.
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
This changes makes a variety of changes to ensure that the cursor is set
back to the default cursor when it leaves the `WebView`:
1. Display list updates can come after a mouse leaves the `WebView`, so
when refreshing the cursor after the update, base the updated cursor
on the last hovered location in the `DocumentEventHandler`, rather
than the compositor. This allows us to catch when the last hovered
position is `None` (ie the cursor has left the `WebView`).
2. When handling `MouseLeftViewport` events for the cursor leaving the
entire WebView, properly set the
MouseLeftViewport::focus_moving_to_another_iframe` on the input event
passed to the script thread.
3. When moving out of the `WebView` entirely, explicitly ask the
embedder to set the cursor back to the default.
Testing: This change adds a unit test verifying this behavior.
Fixes: #38710.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
Previously we immediately passed the KeyboardEvent to embedder. Now we
make element send keys go through the dispatch action which required by
spec. CompositionEvent still immediately passed through embedder
Testing: Should make
`./tests/wpt/tests/webdriver/tests/classic/element_send_keys/` more
stable.
Fixes: https://github.com/servo/servo/issues/38354
Fixes: https://github.com/servo/servo/issues/38442
---------
Signed-off-by: PotatoCP <Kenzie.Raditya.Tirtarahardja@huawei.com>
Co-authored-by: Euclid Ye <euclid.ye@huawei.com>
Building on the preference observer work from #38649, we now
automatically install an observer when multiprocess mode is enabled.
This observer notifies the constellation of updated preferences, which
in turn notifies each content process so the changes will be reflected
into script/layout as expected. There's a unit test that verifies this
works correctly by checking a preference-gated WebIDL property before
and after the preference is toggled.
Testing: New unit test added.
Fixes: #35966
Depends on #38649.
---------
Signed-off-by: Josh Matthews <josh@joshmatthews.net>
Motivation:
Using our GenericChannel abstraction allows us to optimize IPC in
single-process mode to just use cross-beam channel.
To keep the diff low, and get early feedback, this PR only tackles a
single channel, but the intention is to port all ipc channels to the
generic channel, which allows us to skip serializing and deserializing
messages in single process mode.
Based on:
- https://github.com/servo/servo/pull/38638
- https://github.com/servo/servo/pull/38636
Testing: Covered by existing tests
---------
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Implment get window handles according to
[spec](https://w3c.github.io/webdriver/#dfn-window-handles).
- Window handles are supposed to identify `browsing context`. However,
based on `get window handle command` and `get window handles command`,
we only need to care about top level browsing context.
- Back then, we use a random generated uuid for eacch webview id, it is
not correct but still work because all commands depend on `webview id`
and `browsing context id`. The only case we need window handle is is
when webdriver gets window object with js script. Since the object is
converted to the id of window's document node, `get window handle`
should return the same thing.
Action run (with updated expectation):
https://github.com/longvatrong111/servo/actions/runs/16957610535https://github.com/longvatrong111/servo/actions/runs/16957612027
Some tests may sporadically timeout due to unstable hit test.
cc: @xiaochengh
---------
Signed-off-by: batu_hoang <hoang.binh.trong@huawei.com>
In single process mode, there is a race condition on the initialization
of the global fetch thread: once initialized the global fetch thread
will always use a given core resource thread, and this will be
determined by the component who first initializes it. For example, if
the canvas paint thread first does an async fetch, then this will set
the public core resource as used for all future fetches, including those
coming from a pipeline in private mode.
In multi-process mode, there is a race condition per window event-loop:
the first pipeline to use the fetch will set the core resource thread
for all others.
To ensure the fetch thread uses the correct core resource thread(private
vs public), we need to
pass the core resource thread to each fetch thread operation for which
is it needed.
Testing: It should not break existing fetch WPT tests. The race
condition is not something that can be tested reliably, but it seems to
be based on solid logic.
Fixes: follow-up from
https://github.com/servo/servo/pull/38421/files#r2248950924
---------
Signed-off-by: gterzian <2792687+gterzian@users.noreply.github.com>
Add some tracing/instrumenting for canvas messages processing.
Testing: We have not tests for tracing code.
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
Properly send `mouseleave` events when the cursor moves between
`<iframe>`s. This allows a better handling of cursor changes and status
text updates. Specifically, we do not need to continuously update the
cursor and the value can be cached in the `Document`. In addition,
status updates can now be sent properly when moving focus between
`<iframe>`s.
Note that style updates for `:hover` values are still broken, but less
so than before. Now the hover state on the `Node` is updated, but for
some
reason the restyle isn't taking place properly. This maintains the
status quo as far as behavior goes when hover moves between `<iframe>`s.
This change also adds a helper data structure to `Document` which will
eventually be responsible for event handling.
Testing: Cursor and status change are currently very hard to test as
the API test harness makes this difficult at the moment.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
Co-authored-by: Oriol Brufau <obrufau@igalia.com>
This change replaces our custom `panic` / `unwrap` lint with the one
from clippy. This rule as not properly applied in servoshell, so this
change fixes some clippy errors raised by the new configuration.
Testing: This change removes the tidy tests for the custom lints, but
otherwise the behavior is tested as part of clippy itself.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
Instead of using WebRender hit testing to update the cursor, base it on
layout hit tests. This allows removing the majority of WebRender hit
test items and finally opens up the possibility of adding support for
custom cursors. In addition, this change fixes an issue where cursors
were not set properly on areas of the viewport that extended past the
page content.
Testing: This is difficult to test as verifying that the cursor changed
properly is beyond the capabilities of Servo's test harnesses.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
Co-authored-by: Oriol Brufau <obrufau@igalia.com>
The fetch thread is currently not shut down, meaning it is one of the
threads counted in the "threads are still running after shutdown (bad)"
counter shown when Servo has shut-down.
This PR introduces a mechanism to start, and shut-down, one fetch thread
per process that requires one.
Testing: WPT tests and unit tests(for font context). Also manually
tested loading and closing "about:blank": this change indeed brings down
the count of threads still running after shutdown by one.
Fixes: https://github.com/servo/servo/issues/34887
---------
Signed-off-by: gterzian <2792687+gterzian@users.noreply.github.com>
We currently send exit signals to script-threads, and we also join on
the BHM worker. This ensure the constellation shuts-down only after
script has dropped it's sender to the BHM worker. By joining on the
script-threads, we have a guarantee that they have exited(which is
stronger than having dropped their senders) by the time the
constellation exits.
Testing: Manually opened many tabs and closed the window, both in
single- and multi-process modes.
Fixes: Part of - https://github.com/servo/servo/issues/30849
---------
Signed-off-by: gterzian <2792687+gterzian@users.noreply.github.com>
The previous use of a static variable for the runtime prevented it from
shutting down cleanly, because shutdown requires dropping or taking
ownership of it. This PR switches the static variable to a handle only,
and introduces a new trait to pass a handle to the async runtime to the
constellation, where it can be shut-down along with other components and
help reduce our count of still running threads after shutdown.
Testing: manual testing, and covered by unit-test in net, and wpt tests.
Fixes: part of - https://github.com/servo/servo/issues/30849
---------
Signed-off-by: gterzian <2792687+gterzian@users.noreply.github.com>
["When applicable, unstable sorting is preferred because it is generally
faster than stable sorting and it doesn’t allocate auxiliary
memory."](https://doc.rust-lang.org/std/vec/struct.Vec.html#method.sort)
Binary also reduced by 1KB in Release.
Testing: No behaviour change as semantically all current usage does not
have any pair with `std::cmp::Ordering::Equal`.
Signed-off-by: Euclid Ye <euclid.ye@huawei.com>
With some adjustment for `NamedKey`. The two crates need to be bumped
together to avoid duplicate of `keyboard-types` action.
---------
Signed-off-by: PotatoCP <Kenzie.Raditya.Tirtarahardja@huawei.com>
`webview.focus` return `Result<(),()>`, hence it gives warning if the
result is not handled.
Signed-off-by: PotatoCP <Kenzie.Raditya.Tirtarahardja@huawei.com>
Shut-down of the background hang monitor(bhm) is currently problematic:
- it does not always run until the monitored script-thread does(see
"BackgroundHangMonitor has gone away" mentioned in
https://github.com/servo/servo/issues/34158).
- it shuts-down before the constellation(good, so actually
https://github.com/servo/servo/issues/24850 was "fixed" but in a way
that introduced a new problem), but using a mechanism that allows it to
shutdown before script(the problem above).
- there are various mechanism(see the doc comments removed by this PR)
in place which are meant to ensure a clean shutdown despite the above
problems; those are complicated, and become unnecessary once those
problems are fixed.
All of the above is fixed by the changes in this PR, which ensure the
bhm does not shut-down before script, and also maintains the invariant
that it must shut-down before the constellation(in single-process mode)
or before the main thread(in multi-process mode), but using a mechanism
which allows it to keep running until script shuts-down.
An unnecessary option around the exit signal is also removed.
As a positive side-effect, it also ensures that any script-thread is
shut-down before the constellation(because for the bhm worker to exit,
the monitored script must have exited first), so this should also fix a
host of other problems noted in
https://github.com/servo/servo/issues/30849, but each should be
confirmed independently(and various other improvements seem possible in
their specific contexts, such as joining on script threads, and removing
the `ScriptThreadMessage::ExitScriptThread`).
Fixes: https://github.com/servo/servo/issues/24850 and part of
https://github.com/servo/servo/issues/34158
Testing: Unit tests in `component/background_hang_monitor/tests`. Also
manually tested loading "about-blank" in single- and multi-process mode.
---------
Signed-off-by: gterzian <2792687+gterzian@users.noreply.github.com>
For current implementation, when a command may trigger a navigation,
webdriver only waits for document readiness state.
However, not all navigations make change in document.
This PR handles more cases for waiting for a navigation, and apply to
`element_click`.
- Before sending a command which may trigger a navigation, `webdriver`
sets `load status send` to `embedder`, `constelltation` and `script
thread` to listen to `navigation events`.
- Webdriver check if there is a navigation with `script thread`.
- If the navigation is loading a new url, webdriver checks if the
request is approved with `constellation`, then waits for document
readiness state.
- If the navigation is a hashchange, webdriver waits untill all new
generated dom events have been processed.
Testing:
`tests/wpt/tests/webdriver/tests/classic/element_click/navigate.py`
`tests/wpt/tests/webdriver/tests/classic/element_click/user_prompts.py`
https://github.com/longvatrong111/servo/actions/runs/16488690749
cc: @xiaochengh
---------
Signed-off-by: batu_hoang <hoang.binh.trong@huawei.com>
We want to eventually remove raqote backend, but for now we can gate it
behind a feature (still enabled by default in servoshell) like the rest
of backends. `dom_canvas_backend=auto` will select first available
backend. Builds on top of https://github.com/servo/servo/pull/38310 to
support cases where no backend is available.
Testing: It compiles with or without feature
---------
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
