Each redraw issued one unbuffered stderr write per row plus cursor
moves, which can tear visually mid-frame. Build the frame in a stack
buffer and emit it with a single locked write, like Spinner's
renderLocked.
Previously tick() tore the connection down without a WebSocket close
frame, so clients saw an abnormal closure (1006). Safe to send from
tick(): the worker thread is the sole writer of the socket.
Review pattern from #2944: an injected `frame` parameter is the frame of
the calling realm, not the frame owning the element. The wheel/touch
cancelability check consulted the caller's event manager and window; for
an element living in another frame (e.g. inside an iframe) its listeners
are registered in its own frame's event manager and its propagation path
ends at that frame's window. Resolve the frame through Element.ownerFrame
like the other cross-realm-safe paths.
Co-Authored-By: Karl Seguin <karlseguin@users.noreply.github.com>
Fixes the 21 failing subtests of WPT
/dom/ranges/Range-mutations-replaceChild.html (39/60 -> 60/60): live
range boundary points anchored on the parent were off by one because we
inserted the new child before removing the old one. Per the DOM
"replace" algorithm, the reference points are captured first (child's
next sibling, or node's own next sibling when they're adjacent), the
old child is removed, and only then is the new node inserted before the
reference - each step running the live-range update rules in that
order.
Self-replacement (replaceChild(c, c)) now also performs the actual
remove-and-reinsert instead of only queueing records, so ranges
anchored inside the node move to its old position, as the remove step
requires.
Node-replaceChild.html and MutationObserver-childList.html stay fully
green.
Coverage: /dom/ranges/Range-mutations-replaceChild.html 39/60 -> 60/60
(fully green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes WPT /dom/events/focus-event-document-move.html: testdriver's
Actions().send() promise resolved immediately (the vendor glue returned
Promise.resolve()) while the action sequence runs on the next scheduler
tick, so tests asserting right after `await ...send()` observed the
pre-action state.
actionSequence now returns a promise that a persisted resolver settles
once the input sources have been performed; the testdriver vendor glue
returns it. The persisted resolver handle is page-managed (persist
tracks it on the context), so the task does not reset it itself - doing
so double-freed the v8 global at page teardown.
Coverage: /dom/events/focus-event-document-move.html 0/1 -> 1/1 (fully
green); /dom/events regression clean (201 files).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes WPT /dom/events/Event-timestamp-high-resolution.https.html
("window[eventType] is not a constructor"): the test constructs a
GamepadEvent and checks its timeStamp against performance.now().
GamepadEvent is a plain Event subclass with a gamepad member that is
always null - there are no gamepads in a headless browser - following
the DeviceMotionEvent pattern.
Coverage: /dom/events/Event-timestamp-high-resolution.https.html
0/1 -> 1/1 (fully green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes the last failing subtest of WPT /dom/events/Event-dispatch-click.html
("pick the first with activation behavior <a href>", 32/33 -> 33/33):
following a link whose href is a javascript: URL must evaluate the URL
body as script in the link's frame. We explicitly ignored such hrefs,
so the test's completion callback never ran.
Anchor activation now schedules the script on the frame's JS scheduler
(navigation is a queued task) and compiles/runs it in the frame's
context, ignoring the completion value (a string result would replace
the document, which nothing relies on here). The URL body should be
percent-decoded per spec; markup hrefs rarely are, so that's left as a
TODO.
Coverage: /dom/events/Event-dispatch-click.html 32/33 -> 33/33 (fully
green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes all 29 failing files under /dom/events/non-cancelable-when-passive/
(13/42 -> 42/42 files passing):
- WebDriver wheel actions only worked with an element origin;
testdriver's Actions().scroll() uses the viewport origin, so nothing
was dispatched. Viewport origins now resolve their target through the
faux layout (a new vertical-axis-only elementFromPoint variant, since
the faux layout gives elements no useful horizontal extent), falling
back to the document element. The faux dimensions also learn
vh/vw units (resolved against the page viewport), which the tests'
200vh scroll containers rely on.
- Wheel dispatch also fires the legacy mousewheel event, like Blink.
- Touch pointer sources (Actions().addPointer("touch")) now dispatch
touchstart/touchmove/touchend (with pointer events, without mouse
events) instead of being treated as a mouse.
- Per the cancelability rules for scroll-blocking events, wheel,
mousewheel and touch events are dispatched cancelable only when some
listener on the propagation path (target chain plus window) is
non-passive: the UA knows preventDefault can't be called otherwise.
Coverage: /dom/events/non-cancelable-when-passive/: all 42 files pass
(29 newly passing, 1 subtest each); no regressions across /dom/events/.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes all remaining failures of WPT /dom/ranges/Range-cloneContents.html
(171/187 -> 187/187) and Range-extractContents.html (153/187 ->
187/187):
- cloneContents only handled same-container and sibling-boundary
ranges. It now implements the DOM "clone the contents" algorithm on
explicit boundary points: partially contained CharacterData
boundaries are cloned with the substring, partially contained
elements are cloned shallow with the subrange recursed into the
clone, contained nodes are cloned deep, and a contained doctype
throws HierarchyRequestError.
- extractContents was cloneContents + deleteContents, which recreated
the extracted nodes ("you created or detached nodes when you weren't
supposed to"). It now implements the "extract" algorithm: contained
nodes MOVE into the fragment preserving identity; only the boundary
CharacterData substrings and the partially contained element shells
are cloned, with the originals truncated via replaceData. Children
are classified before anything moves, since moving a contained child
shifts the indices the boundary comparisons rely on. The range then
collapses to the same new node/new offset rule as deleteContents.
Coverage:
- /dom/ranges/Range-cloneContents.html 171/187 -> 187/187 (fully green)
- /dom/ranges/Range-extractContents.html 153/187 -> 187/187 (fully green)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Review pattern: the JsApi struct is for JS/v8 bridging only; getters
with actual logic belong on the type. getCompatMode moves next to
isQuirksMode and the bridge accessor just references it.
Co-Authored-By: Karl Seguin <karlseguin@users.noreply.github.com>
Fixes the 20 failing subtests of WPT /dom/ranges/Range-deleteContents.html
(105/125 -> 125/125) and 12 of Range-extractContents.html (which
delegates its deletion half): deleteContents only handled same-container
ranges and the "boundary containers are siblings" case, and always
collapsed to the start point.
It now follows the DOM algorithm:
- same-CharacterData ranges replace the data in place (replaceData, so
live ranges update);
- the top-most contained nodes (contained per boundary-point comparison,
with no contained parent) are collected before any mutation and
removed wherever they live in the tree, not just among siblings;
- partially contained CharacterData boundaries are truncated with
replaceData;
- the range collapses to the spec's "new node/new offset": the start
point when the start node is an inclusive ancestor of the end node,
otherwise just after the highest partially-contained ancestor of the
start node.
Coverage:
- /dom/ranges/Range-deleteContents.html 105/125 -> 125/125 (fully green)
- /dom/ranges/Range-extractContents.html 141/187 -> 153/187
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes the 44 failing subtests of WPT /dom/ranges/Range-collapse.html
(142/186 -> 186/186): per the DOM IDL, collapse(optional boolean
toStart = false) collapses the range to its END point when the argument
is omitted; we defaulted to the start.
Internal callers (deleteContents) already pass an explicit true and are
unaffected.
Coverage: /dom/ranges/Range-collapse.html 142/186 -> 186/186 (fully
green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes the 22 failing subtests of WPT /dom/ranges/Range-cloneRange.html
(40/62 -> 62/62): per the DOM spec, document.createRange() returns a
range whose start and end are (document, 0) - the document the method
was called on. The range factory hardcoded the frame's main document,
so foreignDoc.createRange() produced a range rooted in the main
document.
Factory.abstractRangeIn takes the initial container; createRange passes
its own document, and Range.init keeps the main-document default for
internal callers.
Coverage: /dom/ranges/Range-cloneRange.html 40/62 -> 62/62 (fully
green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes WPT /dom/nodes/getElementsByClassName-14.htm: in quirks mode
getElementsByClassName must match class names ASCII
case-insensitively (class="a A" matches "a" twice), while Unicode case
stays significant.
The document had no notion of quirks mode at all (compatMode was
hardcoded to "CSS1Compat"). Document.isQuirksMode approximates the
HTML parser's mode: an HTML document without a doctype child is in
quirks mode (legacy doctypes that also trigger quirks are not
detected). compatMode now reports BackCompat accordingly, and
getElementsByClassName bakes the mode into its live-collection filter
(ClassNameFilter), with classAttributeContainsCase doing the
ASCII-case-insensitive token scan.
Coverage: /dom/nodes/getElementsByClassName-14.htm 1/2 -> 2/2 (fully
green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- replaceChild with a DocumentFragment still queues the fragment's own
removal record (the spec's insert algorithm queues it regardless of
suppressObservers); only the parent-side addition record is suppressed
in favor of the combined record.
- The combined record's previousSibling/nextSibling are captured before
new_child is removed from its old position, per the replace algorithm's
step order, so replacing a child with its immediate previous sibling
reports that sibling.
- The fragment-into-document validity check treats CDATASection as a
Text node, matching the direct-node check.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Goal is to better support attachToBrowserTarget by giving the browser context
its own session_id. Without this calls to attachToBrowserTarget clobber the
bc.session_id which is the page/target session_id.
Fixes "parser insertion mutations" in WPT
/dom/nodes/MutationObserver-document.html (1/4 -> 2/4): a script running
during the initial document parse can observe the document with
{subtree, childList} and must receive records for the nodes the parser
inserts after it. We suppressed all notifications for main-document
parser insertions.
Parser insertions now notify per inserted node. Fragment parses
(innerHTML et al.) stay silent, since Node.setHTML queues a single
combined "replace all" record. There is no overhead in the common case:
notifyChildInserted is gated on any observer existing, which is never
true during a parse unless an earlier script registered one.
The remaining two subtests need synchronous execution of scripts
inserted by other scripts during parsing (record batching across the
two) and parser insertions into a parent removed mid-parse; both are
script-scheduling semantics, not observer plumbing.
Coverage: /dom/nodes/MutationObserver-document.html 1/4 -> 2/4.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes the remaining 8 failing subtests of WPT
/dom/nodes/MutationObserver-childList.html (30/38 -> 38/38, fully
green):
- Inserting a DocumentFragment emitted one record per child for both
the removals from the fragment and the insertions into the parent.
Per the DOM insert algorithm observers get exactly two records: one
on the fragment with all removedNodes and one on the parent with all
addedNodes. appendAllChildren/insertAllChildrenBefore now share
moveAllChildren, which suppresses per-node notification and queues
the two combined records (or none, in the silent mode replaceChild
uses).
- replaceChild emitted separate insertion and removal records; the
"replace" algorithm queues one combined record {addedNodes,
removedNodes}. Removing the new child from its previous position
still notifies separately (internal replacement), and replacing a
node with itself reports a removal record followed by an addition
record with no tree change, matching browsers.
- Range.insertNode into a Text container rebuilt the text as new
before/after nodes (replaceChild + two insertBefores, three records).
It now uses splitText and inserts before the second half: two
records, and live ranges are updated by the split as a bonus
(Range-mutations-replaceChild 37/60 -> 39/60).
The unit test mutation_observer/childlist.html asserted the old
two-record replaceChild; updated to the spec's single combined record.
Coverage:
- /dom/nodes/MutationObserver-childList.html 30/38 -> 38/38 (fully green)
- /dom/ranges/Range-mutations-replaceChild.html 37/60 -> 39/60
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes the failing subtest of WPT /dom/nodes/Node-mutation-adoptNode.html
("Adopting an element ... owner docs of it's attributes"): after
adopting an element into another document, its attribute nodes'
ownerDocument must report the new document.
Attribute nodes are parent-less, so ownerDocument fell through to the
detached-node fallback (the per-frame owner map / main document), which
never changes on adoption. An attribute node with an owning element now
delegates to that element's ownerDocument; detached Attr nodes keep the
old resolution.
Coverage:
- /dom/nodes/Node-mutation-adoptNode.html 1/2 -> 2/2 (fully green)
- /dom/nodes/attributes-namednodemap-cross-document.window.html 0/2 -> 1/2
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes the remaining failures of WPT /dom/nodes/Node-insertBefore.html
(16/40), Node-replaceChild.html (15/29) and Node-appendChild.html
(9/11) - all three are now fully green.
The insertion validity checks were incomplete and ran in the wrong
order:
- The reference-child NotFoundError check ran before the parent-type
and cycle checks, but per "ensure pre-insert validity" the
HierarchyRequestError checks for the parent kind and for node being
an inclusive ancestor of parent come first.
- Inserting a Document node was not rejected.
- DocumentFragment insertion into a document skipped validation
entirely: fragments with multiple elements, with a Text child, or
whose element joins an existing document element now throw.
- The doctype rules were missing: a second doctype, a doctype inserted
after an element, and an element inserted before a doctype all throw,
with replaceChild's variants (the replaced child doesn't count).
- replaceChild threw HierarchyRequestError for a child with the wrong
parent; the spec requires NotFoundError (unit test updated
accordingly).
- insertBefore's reference-child argument is required-but-nullable per
WebIDL: omitting it is now a TypeError (js.Nullable), while passing
null still appends.
appendChild/insertBefore/replaceChild now share ensurePreInsertValidity
(insert and replace modes); replaceChildren keeps its replace-all
validation.
Coverage:
- /dom/nodes/Node-insertBefore.html 16/40 -> 40/40 (fully green)
- /dom/nodes/Node-replaceChild.html 15/29 -> 29/29 (fully green)
- /dom/nodes/Node-appendChild.html 9/11 -> 11/11 (fully green)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes WPT /dom/traversal/NodeIterator-removal.html (0/23 -> 23/23) and
/dom/nodes/moveBefore/moveBefore-nodeiterator.html: removing a node
that is an inclusive ancestor of a live NodeIterator's reference node
must move the reference per the DOM pre-removing steps. We never
adjusted iterators, so the reference kept pointing into the detached
subtree.
The frame keeps an intrusive list of live NodeIterators (mirroring
_live_ranges; iterators are slab-allocated for the frame lifetime, so
they are never unlinked) and Frame.removeNode runs the steps while the
tree is still intact:
- removing the root or an ancestor of the root leaves the iterator
untouched (matching browsers and the WPT model);
- with the pointer before the reference, the reference moves to the
first node following the removed subtree, if any;
- otherwise (or when there is no such node) it moves to the node
immediately preceding the removed node in tree order, clearing the
before-pointer in the fallthrough case.
Coverage:
- /dom/traversal/NodeIterator-removal.html 0/23 -> 23/23 (fully green)
- /dom/nodes/moveBefore/moveBefore-nodeiterator.html 0/1 -> 1/1
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes the 5 failing subtests of WPT
/dom/traversal/TreeWalker-acceptNode-filter.html (7/12 -> 12/12):
- A filter object was converted eagerly at createTreeWalker time, so an
object without a callable acceptNode threw "invalid argument" at
creation. Per WebIDL any object converts to the NodeFilter callback
interface; the TypeError belongs at invocation time.
- The acceptNode member was cached at conversion. Per "call a user
object's operation" it must be looked up with a fresh Get on every
traversal, rethrowing errors from a throwing getter.
- The callback was invoked with the default this; the spec requires the
filter object itself as the this value.
NodeFilter now stores the raw function or object (js.Object.Global) and
performs the per-invocation lookup, callability check and this-binding
in acceptNode.
Coverage: /dom/traversal/TreeWalker-acceptNode-filter.html 7/12 ->
12/12 (fully green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes the last failing subtest of WPT /dom/traversal/TreeWalker.html
("Recursive filters need to throw"): per the DOM traversal "filter"
algorithm, a NodeFilter that re-enters the walker (calling parentNode()
etc. from inside the callback) must get an InvalidStateError.
DOMTreeWalker tracks an active flag around the filter invocation and
throws InvalidStateError when a traversal starts while it is set.
Coverage: /dom/traversal/TreeWalker.html 760/761 -> 761/761 (fully
green).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes WPT /dom/traversal/TreeWalker-previousSiblingLastChildSkip.html
and 59 subtests of TreeWalker.html (701/761 -> 760/761):
TreeWalker.previousSibling()/nextSibling() only scanned the current
node's direct siblings, but the spec's "traverse siblings" algorithm
also:
- descends into a skipped (FILTER_SKIP) sibling's children - only a
rejected sibling excludes its whole subtree - so from B2 with B1
skipped, previousSibling() must return B1's last child;
- climbs to the parent when the siblings are exhausted and continues
from the parent's siblings, stopping at the root or at an accepted
parent.
Both directions now share the spec's traverseSiblings implementation.
Coverage:
- /dom/traversal/TreeWalker-previousSiblingLastChildSkip.html 0/1 -> 1/1
(fully green)
- /dom/traversal/TreeWalker.html 701/761 -> 760/761
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
std.json serializes []u8 that isn't valid UTF-8 as a JSON array of
numbers, so headers like "expires: mié, 15 jul 2026 ..." (Latin-1 0xE9)
appeared as byte arrays in Network.responseReceived. Stream such values
through the JSON writer with Latin-1 -> UTF-8 transcoding, matching
Chrome's behavior for DevTools.
