Use Qt's platform shortcut bindings for back and forward navigation
instead of only installing the primary binding. On macOS this keeps
Command-[ and Command-] while also accepting Command-Left and
Command-Right.
Disable Qt quit-on-last-window behavior on macOS and install an
application-level menu so useful commands remain available after the
last browser window closes.
Reuse shared application actions and menus for Settings, Bookmarks,
History, Help, and global appearance controls. Centralize window-aware
commands such as New Tab, New Window, Open File, Open Location, and
Reopen Recently Closed so the browser window and windowless app menu
use the same behavior. Keep the location editor focused for new blank
windows after the native window has been activated.
Disable AppKit's native resize handling for client-side decorated macOS
windows so AppKit does not show a resize cursor for points it will not
deliver to Ladybird. Reapply this after Qt creates the native surface,
since Qt may restore the resizable style mask during Cocoa window setup.
Handle the resize gesture in BrowserWindow instead, with thin side edges
and a slightly larger corner target that avoids the rounded corner
cutouts.
Move the active macOS web selection color into LibWebView so AppKit
and Qt use the same value when building the WebContent system palette.
Qt now also asks AppKit for inactive selection colors on macOS, matching
the AppKit frontend while leaving other Qt platforms on QPalette.
macOS can be configured to drag windows with Cmd+Ctrl+click via
NSWindowShouldDragOnGesture. AppKit does not see its usual titlebar
drag path for our frameless Qt windows, so handle the native mouse
down in the AppKit event filter when that default is enabled.
Forward the event to performWindowDragWithEvent before Qt turns it into
normal widget input. This restores the system gesture without changing
behavior for users who have not enabled the default.
After web content has focus on macOS, its native QRhiWidget can
remain the AppKit first responder even after Qt focuses the location
edit.
Make the top-level Qt content view first responder before focusing the
location edit so keyboard input returns to chrome.
Allow highlighted rows that are not valid inline completions to preview
their full suggestion text in the location editor. This covers history
entries matched by title or URL substring, while ordinary typing stays
on the inline-only path.
Keep automatic inline autocomplete text when Escape closes the popup.
Escape still cancels hovered or keyboard-selected preview text.
Remember which query produced the visible autocomplete popup rows, and
only activate selected rows when the visible popup still belongs to the
current query.
When Enter is pressed while fresh results are pending, rerun
autocomplete for the current query and activate once matching results
arrive. Ignore late completions after focus leaves. Avoid canceling
autocomplete while its completion callback activates the pending result.
Open the autocomplete popup when the location editor is focused
through the browser action, such as Cmd+L. Keep the selected URL text
intact by skipping inline completion for that explicit popup request.
Position the native popup again after its first show so Qt has realized
the tooltip window before the final placement, avoiding the brief macOS
Y-coordinate jump.
Problem: UBSan crash when computing layout for an element with a giant
negative inset.
Cause: CSSPixels::operator-() returned from_raw(-raw_value()), and
negating the i32 minimum overflows int.
Fix: Negate with saturating_sub(0, raw_value()) — matching the
saturating arithmetic already used by the other CSSPixels operators.
Fixes https://github.com/LadybirdBrowser/ladybird/issues/9997
Problem: Crash when loading an ordered list whose numbering reaches the
i32 limit; e.g. <ol start="2147483647"> with two or more items.
Cause: Element::ordinal_value() kept its numbering in a Checked<i32>
and stepped it once per list item. When the numbering already sits at
the i32 maximum (or minimum, for a reversed list whose value attribute
pins it there), the increment overflowed the Checked value.
Fix: Keep the numbering in a plain i32 instead, and step it with
AK::saturating_add and AK::saturating_sub — so it clamps at the i32
bounds, rather than overflowing.
Fixes https://github.com/LadybirdBrowser/ladybird/issues/10003
Problem: Crash when generating an RSA key — or serializing one —
whose publicExponent is a typed array whose backing ArrayBuffer has
been detached; for example, by calling transfer() on it.
Cause: Two places with big_integer_from_api_big_integer() reading the
bytes of the backing ArrayBuffer directly. But reading the bytes of a
detached buffer aborts.
Fix: Read the bytes with WebIDL get_buffer_source_copy() — which yields
an empty copy for a detached, or OOB resizable, buffer. The empty array
is already mapped to zero — so generation rejects the zero exponent with
an error, rather than crashing.
Fixes https://github.com/LadybirdBrowser/ladybird/issues/9991
Previously we assumed that if the calculation had been simplified to a
single `NumericCalculationNode` that we could resolve it - this isn't
the case if we have a percentage that should be resolved to a different
type.
Fixes#10209
Previously, font selection ignored the Unicode emoji presentation of a
code point. Emoji-capable code points were always resolved through
pre-baked color emoji and symbol fonts. Text-default code points which
lacked the `Emoji_Presentation` property, were therefore rendered as
color emoji rather than text. We now classify each code point's default
presentation from its `Emoji_Presentation` property and any trailing
variation selector.
Problem: TestWebDriverSessionHistory flakes on slow (Sanitizer) CI
runners. The subtest checking that cancellation of a script-initiated
cross-site “back” by a beforeunload handler intermittently failed the
assertion “Expected beforeunload to cancel script-initiated cross-site
history.back(), got [url_b, 0]”. scriptBeforeUnloadCount was 0, not 1.
Cause: history.back() appends its traversal to the traversable’s session
history queue and returns immediately. So the beforeunload prompt to
unload runs async. The test read scriptBeforeUnloadCount in a separate
execute_script right after — with no wait. So on a slow runner, the read
beats the dispatch, and sees 0. Every other script-initiated navigation
in the test waits for its target document to load — and the “refresh”
beforeunload check already polls. But the canceled “back” loads no
target document — and this one waited for nothing.
Fix: Poll with wait_for_script_result until the page is still on /b and
window.scriptBeforeUnloadCount has incremented, before asserting the
final state — mirroring the “refresh” beforeunload check.
Move CSS animation values into a mutable overlay on computed properties
and make base computed style data immutable after construction. Base
style mutation now goes through a builder that is consumed on publish,
so installed styles no longer expose mutation APIs.
Build new base style data for inherited style updates instead of cloning
and mutating installed computed properties. Element-specific computed
style adjustments now run before publication, while animation and
transition updates continue to mutate only the animated overlay.
Problem: Converting an object with a pathologically-deep prototype chain
to a primitive was segfaulting.
Cause: Object::internal_get implements [[Get]] by recursing into the
prototype’s [[Get]] (parent->internal_get) when the property isn’t an
own property. For a sufficiently deep prototype chain, that C++
recursion exhausts the native stack, and segfaults. The bytecode
interpreter’s call-stack limit doesn’t cover this native recursion.
Fix: Before recursing into the prototype in Object::internal_get, check
VM::did_reach_stack_space_limit(), and throw a CallStackSizeExceeded
InternalError — the same way the interpreter and other recursive runtime
operations guard the native stack. The deep-chain get now throws a
catchable call-stack-size-exceeded error, rather than crashing.
Fixes https://github.com/LadybirdBrowser/ladybird/issues/3584
Replace the generated C++ legacy codec implementations with a
small Rust wrapper around encoding_rs.
This keeps the existing LibTextCodec API while moving label lookup,
legacy decode/encode, validation, and streaming decoder state to Rust.
The generated index data and generator are no longer needed.
It also fixes several TextDecoder EOF cases due to a more correct
implementation. encoding_rs finalizes decoders according to the
Encoding Standard, so incomplete UTF-8/Big5 tails and malformed
UTF-16 surrogate tails produce the required single replacement at
end-of-queue instead of being dropped, buffered, or double-counted
by our old hand-written decoders.
Replace the Rust URL form-encoding callback bridge into C++ TextCodec
with a direct encoding_rs encoder.
This keeps percent-encode-after-encoding entirely in Rust and removes
liburl_rust's dependency on LibTextCodec.
It also happens to fix ISO-2022-JP URL encoding of literal U+FFFD.
The LibTextCodec reverse lookup treats generated 0xFFFD table holes as
real JIS0208 mappings, so literal U+FFFD skipped the encoder-error
path. encoding_rs treats U+FFFD as unmappable, so URL encoding emits
the required numeric character reference.
Pass an encoding label to StreamingDecoder instead of requiring
callers to pre-resolve a Decoder reference.
This removes redundant decoder_for() plumbing from callers and lets
StreamingDecoder hide how chunked decoder state is represented.
Use String::from_utf8_with_replacement_character() directly for Infra's
parse JSON bytes algorithm instead of constructing a more generic
TextCodec UTF-8 decoder.
PDFDocEncoding has no remaining users left. Remove the decoder and
label plumbing. Retarget the lazy source-code decoding regression
test to Windows-1252 so it still covers non-UTF source decoding.
Include directory read access in Landlock read-write directory rules.
Without it, paths such as /dev/dri could be writable but not enumerable
after sandboxing, which made Mesa fail while probing GPU devices.
Allow the Compositor to open /dev/udmabuf as well when the device is
present, matching the other GPU device nodes exposed to the process.
Let the Linux Compositor sandbox permit writable executable mappings
used lazily by GPU drivers after WebGL context creation. Keep this
allowance scoped to the Compositor process so WebContent retains the
existing memory mapping restrictions.
UnresolvedStyleValue::create() serialized the component values into a
comparison string unconditionally, but the result is only used when the
caller supplied separate original source text. On the common path that
serialization (including float formatting of every numeric token) was
computed and immediately discarded.
Move the serialization into the branch that actually uses it. This is
hot during var() substitution, where many intermediate
UnresolvedStyleValues are created and never compared.
TransportMachPort wrote to its event-loop notification pipe for every
received message. Large bursts could spend significant time in the IO
thread just waking the main loop, even when a previous read notification
was still pending and would already drain the queued messages.
Track whether a read notification is pending while holding the incoming
queue mutex, and only write a new pipe byte when the queue needs a wake.
EOF still schedules a notification, and synchronous waiters continue to
use the condition variable for every arrival.
Nested compositor contexts used the same vsync scheduling path as
top-level contexts. That made iframe presents depend on a context's own
timer instead of the display refresh rate, and screenshots needed a
separate descendant flush to avoid stale child surfaces.
Keep top-level contexts on vsync, but make nested contexts dirty their
containing context instead of scheduling themselves. When a composited
child surface is resolved, synchronously rasterize the child if it has
a pending frame, using a separate display-list player so recursive
replay does not reuse the active player state. This lets screenshots and
normal presentation share the same child-surface update path.
Allow having separate GC heaps and implement coordinated marking between
them; this is useful for keeping wasm and js GC heaps separated with a
clear boundary.
Previously, these were stored in a vector that was linearly scanned.
For large sites this vector could contain hundreds of entries, so
a HashMap gives a significant speedup.
Build a per-anchor Bloom-style filter for :has() argument matching
after an anchor sees a second check for the same traversal scope. The
filter stores salted tag, id, class, and attribute-name hashes from the
child or descendant scope and rejects arguments whose required
identifiers are absent.
This avoids repeatedly walking the same subtree for unrelated :has()
arguments while preserving the single-check case. More complex
direct-child arguments use the descendant scope so hashes from later
descendant compounds cannot cause false rejections.
Keep the filter conservative for quirks-mode class selectors and for
sibling-combinator relative selectors during invalidation metadata
collection. Text tests cover cache-primed misses for both cases.
Firefox exposes removal operations for IndexedDB databases, object
stores, and records from the Storage panel. Route those actor requests
to LibWeb so DevTools can delete the selected data and receive the
update or clear event needed to refresh the panel.
The Storage panel expects storesUpdate messages after watched storage
changes. Summarize committed IndexedDB mutation logs into DevTools paths
and forward them through WebContent so Firefox can refresh database,
object-store, and record rows without polling.
Firefox's behaviour is less than ideal here. A lot of things don't
update automatically even inspecting a page in Firefox. Some
things (like new databases) won't show up until you fully refresh the
page. So that makes it a bit hard to know that we're doing things
correctly. As far as I can tell, we are at least behaving as well as
Firefox requires.
We do have one workaround: Firefox doesn't display record updates
without a manual refresh, and in fact any change messages for them show
up as rows in the host's database table. So for now, we filter them out
to avoid visual weirdness in the inspector.
Firefox asks the storage watcher for an indexed-db resource before it
shows IndexedDB entries in the Storage panel. Add an IndexedDB actor and
serialize the live LibWeb database registry on demand, so WebContent can
return the host tree and table rows without duplicating database state.
Use the LibWeb inspection helpers to read IndexedDB internals, and keep
the Firefox protocol shape in LibDevTools. WebContent only forwards the
serialized response over the existing DevTools IPC path.
