[BugFix][SSE] Resolve ReferenceError for Event in EventSource (d0c402d) @DragonDragonDDD

The EventSource implementation was causing a ReferenceError because the standard Event object is not available in its execution environment.

This change resolves the issue by creating a compatible Event object for the onopen and onerror handlers.

• An Event type alias is created from the existing BaseEventOrig interface to ensure type safety without defining a new class.
• The _dispatchEvent method is updated to construct plain event objects that conform to the BaseEventOrig interface, replacing the new Event() calls.
• The onmessage handler continues to receive the raw EventSourceEvent for direct data access, while onerror and onopen receive the new, compliant Event object.

[BugFix][devtool] Fixed an issue where BaseDevTool would cause a crash by continuing to process messages after it had been destroyed. (0565c21) @popoaichuiniu

  Make native_handler thread-safe with a lock and prevent message passing to BaseDevTool when the handler is invalid.

[BugFix][iOS] Simplify logic executor initialization and add async template fetch callbacks (3452820) @ykai55

This refactor removes direct dependency on passing template bundles and runtime options into LynxDefaultLogicExecutor. The executor now derives its configuration from LynxViewGroup, reducing setup surface and coupling.

Key changes:

• Replace initWithTemplateBundle/initWithRuntimeOptions with initWithLynxViewGroup:debuggable:
• Executor now fetches LynxTemplateBundle and runtime options via LynxViewGroup+Internal
• Add LynxViewGroup fetchTemplate: API to support async callbacks once template load completes or fails
• Introduce internal callback queue, error propagation, and single-shot result handling for template fetch
• Rename _fetch_task to _fetchTask and encapsulate fetch logic in fetchTemplateInternal/setFetchResult
• Update LynxLogicExecutor protocol to use setLynxViewGroup: instead of setTemplateBundle:
• Provide getTemplateBundleNonBlocking accessor for immediate reads

This streamlines initialization, ensures safe async coordination, and improves error handling during template fetching.

[BugFix][EmbeddedMode] Fix timing not work in EmbeddedMode for Android. (a92bce3) @DragonDragonDDD

Add embedded-mode timing collector and wire PerformanceController behavior

Introduce EmbeddedTimingCollector to provide minimal timing tracking for embedded mode (loadBundleStart and paintEnd). PerformanceController now supports embedded mode via setEmbeddedMode, routing timing to the embedded collector and disabling heavy reporting/memory monitoring when embedded.

Key changes:

• Add EmbeddedTimingCollector with minimal pipeline event emission
• Add PerformanceController.setEmbeddedMode/isEmbeddedMode and initialize embedded collector lazily
• In embedded mode, skip native actor interactions and memory monitoring; markTiming/paintEnd handled by EmbeddedTimingCollector
• Replace enableController flag with embedded-mode checks
• LynxTemplateRender sets embedded mode on PerformanceController, passes null to native when embedded, and marks LOAD_BUNDLE_START at render entry points
• Add PAINT_END constant to TimingConstants and align paint end handling

Landed From: !58844

Author: @kechenglong

[BugFix][Event] Modify the way Events are held and passed. (07eda8e) @rAY-ooo

description: On the one hand, the front-end framework needs to implement interfaces such as CreateEvent, DispatchEvent, and StopPropagation through the C++ Event Ref; on the other hand, EventDispatcher needs to hold the Event Ref to avoid double free.

[BugFix][Fiber]Prevent inline conversion for component children in FiberElement (f655c0b) @DragonDragonDDD

Ensure children are only converted to inline elements when the parent is inline and not a component. Previously, component parents that were inline could force their children into inline mode, causing unexpected layout behavior. The added is_component() guard avoids converting child nodes in component contexts.

[BugFix][Event] Fix a crash issue caused by EventDispatcher resetting failure. (37a8ff8) @rAY-ooo

description: Because the EventDispatcher is destroyed before the root node, it might cause UserData to fail to reset, leading to a dangling pointer crash when an event is triggered. This is modified to reset UserData when the root node is destroyed to ensure the effectiveness of the EventDispatcher.

NoLanding: develop, release/3.5

[BugFix] Considering LynxThreadStrategy configged by LynxViewBuilder. (d3e1ee8) @nhsprite

It's reasonable that LynxThreadStrategy set by LynxViewBuilder is with higher priority to LynxViewGroup, fix it in this patch.

1. add a _hasThreadStragySet flag that identifies the threadStrategy has been set manually.
2. when calling getThreadStrategyForRendering, if _hasThreadStragySet is true, we should return the strategy set by lynxViewBuilder else return with the strategy set by lynxViewGroup

[BugFix] support updateViewport with needLayout flag. (35ad348) @nhsprite

support updateViewport with needLayout flag, so that users may manually control whether needLayout needs to be triggered at all.

[BugFix][Binding] Fix remote event precision loss (47c95b6) @ci_lynx

The default setting of std::stringstream loses precision, and we use pointers as identifiers in touch events, which requires a quite high precision.

[BugFix][Android] Add unified security verification for TemplateBundle (0a2fc83) @nhsprite

Introduce ByteBuffer parsing and consolidate TASM security verification for both byte[] and ByteBuffer inputs in TemplateBundle. Key updates:

• Add verifyTasm helper to centralize ILynxSecurityService checks
• Extend internalBuildTemplate to accept ByteBuffer and choose native parsing path accordingly
• Enable security verification for ByteBuffer templates (previously TODO)
• Improve null-safety and early returns; support optional option.getUrl()
• Preserve tracing and error propagation; unify options buffer handling

This change enables DirectByteBuffer inputs with consistent security checks and reduces duplicated code paths.

[BugFix][Event] Fix the issue where MTS's stopPropagation was not working under the new event logic. (7c30932) @rAY-ooo

description: The new event logic is compatible with stopPropagation in MTS events.

[BugFix] Do not use putIfAbsent in lynx (6014afe) @nhsprite

Do not use putIfAbsent in lynx as it may cause crash on lower android versions

[BugFix][Event] The new event logic is adapted to RadonDiff. (501d0ac) @rAY-ooo

description: In the RadonDiff architecture, the framework registers events with the AttributeHolder through RadonNode. At this point, the RadonNode is not bound to an Element, so it cannot add the corresponding EventListener to the Element. Therefore, we add the corresponding EventListener to the corresponding Element during the Diff process.

[BugFix][Event] The new event logic is adapted to SSR. (5fb9496) @rAY-ooo

description: Events bound to the SSR homepage can also be responded to; therefore, the corresponding EventListener needs to be added when restoring events or diffing.

[BugFix][fetch] Persist ReadableStream class across BaseApp singleton transfer (24ceab7) @DragonDragonDDD

Ensure fetch-related ReadableStream factory and class are copied when transferring singleton data between BaseApp and StandaloneApp. Adds createReadableStreamClass and ReadableStreamClass to BaseAppSingletonData, and wires them in transferSingletonData and snapshotSingletonData. This maintains consistent fetch behavior in standalone environments after app rebinds or reloads.

[BugFix] harmony longpress not initial with bindLongPress Event fail (15e863e) @ci_lynx

done.

[BugFix][Android] Fix the issue of not setting font features when using cached fonts. (c1f1a77) @Randycn

Previously, font-feature-settings were part of the FontSettingsKey used for caching Typeface objects. This was incorrect because the cached Typeface itself does not retain these settings. As a result, when a font was retrieved from the cache, the feature settings were lost.

This commit decouples font feature settings from the cache key. The fontFeatureSettings are now applied directly to the TextPaint object after the Typeface is retrieved, regardless of whether it came from the cache or was newly created.

This ensures that font features are consistently applied in all scenarios, fixing the rendering bug for cached fonts.

[BugFix] fix iOS sendGlobalEvent in Embedded. (0e54fca) @nhsprite

fix iOS sendGlobalEvent in Embedded

[BugFix] use id instead of function for timer api (0bd1adf) @lybvinci

Use id instead of directly storing specific JS objects to avoid occasional crashes during JSC garbage collection.

[BugFix][Canvas] correct truncation logic in TruncateToStringView (b3c7ac5) @GhostFlying

The original implementation used std::max which would extend the string view beyond the input length when max_length was larger. Using std::min ensures proper truncation behavior.

[BugFix][DevTool]Fix that reload method parameter missing (0ef0b47) @RovicnowYoyi

Add back reload method without param url.

[BugFix][Renderer] Fix style resolving process when removing style objects with intersecting properties (c95eae2) @zhongyr

This patch fixes an issue in the Style resolving process when multiple StyleObjects define different values on the same property, and one of them is removed. The previous implementation could leave stale property values when StyleObjects were removed.

The fix introduces a new property-based simple style mode that:

1. Tracks parsed styles in a map for proper cleanup
2. Uses incremental updates based on properties rather than StyleObjects
3. Properly resets properties that no longer exist in the new style configuration
4. Adds a configuration flag 'enablePropertyBasedSimpleStyle' to toggle between the old StyleObject-based approach and the new property-based approach

Key changes:

• Add EnablePropertyBasedSimpleStyle() method to ElementManager
• Implement UpdateSimpleStyles() overload for move semantics
• Add ResolveStyleObjectsBasedOnExsitingMap() in StyleResolver
• Update SimpleStyleNode interface to support move semantics
• Add enablePropertyBasedSimpleStyle config to LynxConfig

This ensures proper style cleanup and prevents stale property values when StyleObjects are dynamically added or removed.

[BugFix][EmbeddedMode] Fix timing not work in EmbeddedMode when reuse engine. (ac3c83a) @DragonDragonDDD

Ensure PerformanceController registers mClientV2 as the performance observer when starting a page. Previously, the observer was not set in this code path, causing missing performance callbacks during initial render.

Changes:

• Add null-checked setPerformanceObserver(mClientV2) before dispatchOnPageStart in both branches of the initialization flow.

[BugFix][Renderer] Fix style resolving process when removing style objects with intersecting properties (117f073) @zhongyr

This patch fixes an issue in the Style resolving process when multiple StyleObjects define different values on the same property, and one of them is removed. The previous implementation could leave stale property values when StyleObjects were removed.

The fix introduces a new property-based simple style mode that:

1. Tracks parsed styles in a map for proper cleanup
2. Uses incremental updates based on properties rather than StyleObjects
3. Properly resets properties that no longer exist in the new style configuration
4. Adds a configuration flag 'enablePropertyBasedSimpleStyle' to toggle between the old StyleObject-based approach and the new property-based approach

Key changes:

• Add EnablePropertyBasedSimpleStyle() method to ElementManager
• Implement UpdateSimpleStyles() overload for move semantics
• Add ResolveStyleObjectsBasedOnExsitingMap() in StyleResolver
• Update SimpleStyleNode interface to support move semantics
• Add enablePropertyBasedSimpleStyle config to LynxConfig

This ensures proper style cleanup and prevents stale property values when StyleObjects are dynamically added or removed.

[BugFix][List] Ensure the list-item's view insertion and onNodeReady callback in the same ui operation queue. (bc8316d) @DwwWxx

Currently when using MOST_ON_TASM/MULTI_THREAD thread strategy in list, the list-item's view is added to view tree before onNodeReady callback, but many properties are applied in onNodeReady callback. So in reusing list-item scenario, non-udpated list-item's view may be rendered in some draw frames until onNodeReady callback is invoked.

In this commit, we let list-item's view insertion operation and onNodeReady callback in the same ui operation queue to make sure that the list-item is updated before drawing.

NoLanding: release/3.5,release/3.6,develop

[BugFix] Call LynxDevToolUtils only if lynxDebugEnabled is YES (ce510b1) @talisk

summary: In LynxEnv, initLynxBase will call LynxDevToolUtils. It will make LynxDevToolEnv initialized.

On Android platform, LynxEnv.java, there's a judge for isLynxDebugEnabled before LynxDevToolUtils calling. So make iOS and Android the same way.

[BugFix][iOS] Fix iOS source build error (5d0cd0e) @ci_lynx

[BugFix][iOS] Use LynxThreadSafeDictionary for component class registries (190bc8a) @usczz

Replace NSMutableDictionary with LynxThreadSafeDictionary for uiClasses and shadowNodeClasses in LynxComponentScopeRegistry. This ensures thread-safe access to component and shadow node class registrations, reducing potential race conditions during concurrent initialization and lookup.

[BugFix][Harmony] Add NapiHandleScope to ensure valid N-API handle lifetimes (68554da) @rel-q

Introduce NapiHandleScope in Harmony paths to properly scope N-API handles and prevent leaks/crashes during JS interop.

[BugFix][Devtool]Replace "List.of" into "Arrays.asList" to support JDK version below 9 (743af93) @ci_lynx

[BugFix][Logging] Add info log in UpdateGlobalProps for data length and this pointer (9baa4a0) @nhsprite

Introduce a LOGI statement in TemplateAssembler::UpdateGlobalProps to log the incoming data length and the current object pointer. This aids runtime diagnostics and traceability when updating global properties without altering functional behavior.

[BugFix][iOS] correct memory usage calculation for scaled images. (5617bb6) @Tamerlx

This MR is to correct memory usage calculation for scaled images.

The previous calculation didn't account for scale factor on both dimensions, potentially underestimating memory usage for scaled images. This fix properly includes the scale factor for both height and width in the calculation.

[BugFix] dealloc function trigger autorelease pool in concurrent_message_loop (094dd6e) @ci_lynx

dealloc function trigger autorelease pool in concurrent_message_loop

[BugFix][Lepus] Ensure response listeners invoked and record promise result (7dc8cc3) @nhsprite

Previously, response callbacks could be dropped or executed off the expected thread. This change routes response listener execution through TASM delegate’s InvokeResponsePromiseCallback and stores the resolved value in ResponsePromise to preserve result state.

Key changes:

• ResponseHandlerProxy: simplify listener callback to directly invoke the provided closure with the BundleResourceInfo.
• ResponsePromise: persist resolved result in result_ when setting value, before notifying callbacks.
• Lepus binding: accept Callable instead of Closure and schedule callback via TASM delegate’s InvokeResponsePromiseCallback to ensure proper execution context.
• Testing: implement InvokeResponsePromiseCallback in mock delegate and add TestFetchBundle2 to verify promise resolution and listener path.

fields and values via then-listener.

[BugFix] Fix iOS compile errors (b3f60d7) @ci_lynx

• Use an independent to control compile unit for hotreload
• Move enableNewGesture props to public

[BugFix][RadonFiber] Preserve explicit attribute override on Radon architecture in SetNativeProps (5111616) @DragonDragonDDD

When setting native props, attributes passed via style used the style path, but non-style attributes always called SetAttribute(key_str, value) with default override behavior. On Radon architecture, this caused the explicit override flag to be lost, leading to incorrect attribute precedence.

This change routes non-style attributes through SetAttribute(key_str, value, false) when running on Radon, preserving the explicit override semantics expected by Radon while keeping existing behavior for other architectures.

[BugFix][iOS] copy uiHolder before enumeration to prevent mutation. (3f1fe97) @Tamerlx

This MR is to copy uiHolder before enumeration to prevent mutation.

Prevent potential concurrent modification issues by creating an immutable copy of the uiHolder dictionary before enumeration.

[BugFix] Separate SimpleStyle and CSS resolution to prevent duplicate processing (0061cf9) @zhongyr

Summary of change: This change ensures that SimpleStyle and CSS resolution happen only once and orthogonally by extracting SimpleStyle resolution into a dedicated method and making CSS and SimpleStyle resolution mutually exclusive. The bug was causing duplicate style processing when both systems were active.

The fix adds a new ResolveSimpleStyles() method and modifies PrepareForCreateOrUpdate() to call either ResolveSimpleStyles() or ResolveCSSStyles() based on whether SimpleStyle is enabled, preventing duplicate processing.

[BugFix][DevTool]Fix DevToolGlobalHandler init logic (187e23a) @RovicnowYoyi

Using isLynxDebugEnabled() to ensure the initialization logic is correct.

[BugFix][DevTool][Android] Move shouldLoadDevToolJsBridge to LynxDevtoolEnv. (8beab37) @Dango-2021

Since shouldLoadDevToolJsBridge is only used in LynxDevtoolEnv, move it there and make it private.

[BugFix][Android] Fix class cast exception when copying ellipsized text. (10b9c71) @Randycn

• guard Selection ops to run only when text instanceof Spannable
• copy via internal indices (mSelectStart/mSelectEnd), not Selection.getSelectionStart/End
• draw highlight when mOverflow != 0

[BugFix][canvas] fix clearbuffer[ifui]v performance issue (bc4d99f) @ci_lynx

• Fix OOM issue caused by excessive buffer size in clearBuffer value parameter.
• Adjust the logic for updating Piper header files.

[BugFix][Layout] Correctly handle position:fixed when under display:none nodes (68a4377) @givemefive9

summary: Fix an issue where position:fixed still participated in measurement and painting in certain scenarios (under display:none nodes). Such nodes and their subtrees are now fully excluded from layout, painting.

In addition, add a layout switch to avoid break change, it will display the repair effect when quirksMode >= "3.5" or engineVersion >= "3.5".

[BugFix] fix callback's id reused error (ddde49b) @lybvinci

callback's id should release after callback invoked.

[BugFix][Event] Fix an issue with abnormal exposure when Lynx pages were laid out horizontally off-screen. (30a9132) @rAY-ooo

description: In addition to checking the viewport intersection between the exposure node and its parent node, it is also necessary to check the viewport intersection between the Lynx page root node and the screen.

[BugFix][Android] Should update DisplayMetrics when lynxEnv setup. (81d5a7e) @DragonDragonDDD

Should update DisplayMetrics when lynxEnv setup.

[BugFix][RadonFiber] Validate raw inline style updates when property exists (3d2662d) @DragonDragonDDD

Introduce an overload of SetRawInlineStyle that accepts CSSParserConfigs and applies UnitHandler::Process only when the style property already exists in raw_inline_styles_. This preserves previously valid values if a new value is invalid, while avoiding validation cost for first-time additions. Update SetStaticInlineStyle to use the new overload when fiber-based diffing is enabled.

[BugFix] getTemplateData crash caused by Concurrency. (11a9017) @nhsprite

• fix_detail: fix embeddedmodule crash by deep cloning data returned by getTemplateData in LynxTemplateRender

[BugFix][Event] Fix the registration of __RemoveEventListener. (0bcfa67) @rAY-ooo

description: In LepusNG, the registration of RendererFunctions::FiberRemoveEventListener is corrected to __AddEventListener.

[BugFix] Remove callback's id reused logic (a4ba851) @lybvinci

id is enough, so we don't need reused.

[BugFix][Event] Correct the order in which pointer-events traverse sibling nodes. (040c4e5) @rAY-ooo

description: When the node that hits the touch test has point-events:none set, the previously saved sibling nodes that hit the touch test should be traversed in ascending order.

[BugFix][Harmony][Overlay] Stabilize overlay show/dismiss lifecycle and gesture target restoration (ab75ad6) @AdrianLCA

This change fixes overlay visibility and gesture handling issues on Harmony:

• Guard dialog show/dismiss with is_show_ and native_dialog_ to avoid duplicate operations and null derefs; add informative logs when skipped.
• Track "visible" prop separately (is_visible_props_) and auto-show when parent is set after a visible prop update.
• Restore gesture targets on dismiss and during destruction by updating UIOwner’s root target to the next visible overlay or root.
• Unregister NODE_TOUCH_EVENT and set dialog content to nullptr before closing to prevent event leaks and dangling references.
• Introduce RemoveNode override for proper child removal from overlay stack.
• Minor safety/refactor: lock first_active_target_ once in EventDispatcher::EventThrough to avoid redundant locking.

These changes make overlay lifecycle idempotent, prevent stale event routing, and ensure consistent gesture interception behavior.

[BugFix][Harmony] include lynx_actor_specialization.h where lynx_actor.h is dependent. (a93ecd6) @ChrisChan0668

When including lynx_actor, we didn't include shell/lynx_actor_specialization.h, as a result, harmony has lost LynxActorMixin specialization codes, so fix it.

[BugFix] add loadSuccess field to performance entry. (5b91c4f) @Tamerlx

This MR is to add loadSuccess field to performance entry.

Track whether lazy bundle loading was successful by adding a boolean field to the performance entry. This helps in monitoring and debugging bundle loading issues.

[BugFix][Event] Correct the filtering logic of EventListener. (dae4a5b) @rAY-ooo

description: For events that can be captured or bubbled, the EventListener of the EventTarget needs to be filtered at different stages to ensure that the event callback is triggered correctly.

[BugFix][Event] shouldHitTest adapts to the case where parent is JSUIBase. (bcf525f) @rAY-ooo

description: When the parent node is JSUIBase, in the case of Custom Layout, the parent of FrameNode is nullptr, which will cause the child node to fail to respond to HitTest. Therefore, this needs to be avoided.

[BugFix][LayoutInElement] fix layout tree not dirty issue (1aa3d09) @linxs0211

when insert/remove node,we should mark layout tree dirty!

[BugFix] add handle for top function (5ce08e0) @zys777

[BugFix] Perfer PendingJs By ViewBuilder than LynxViewGroup (80d0376) @nhsprite

LynxViewBuilder is more specific than LynxViewGroup. It's reasonable that PendingJs configged by LynxViewBuilder is perfered to LynxViewGroup

isseu: f-972365

summary: Longer description of change addressing as appropriate: why the change is made, context if it is part of many changes, description of previous behavior and newly introduced differences, etc.

Long lines should be wrapped to 72 columns for easier log message viewing in terminals.

[BugFix][LayoutInElement] fix layoutinelement bug (6bfd20b) @linxs0211

fix the issue when inline element is wrapper element

[BugFix][Android][List] Clamp scrollTo offsets to valid range (fa8c72a) @DwwWxx

Ensure scroll offsets are validated before calling super.scrollTo() to avoid overscroll when updateContentSizeAndOffset callbacks are delayed under MOST_ON_TASM or MULTI_THREAD strategies.

[BugFix][Reland] It's not allowed for platform_data to be modified. (49e8202) @nhsprite

1. there may be case that value_converted_from_platform_data_ is already exist before post to js, make a shallowCopy to make it safe.
2. when GenerateTemplateDataPostedToJs to generate a templateData to be consumed by js thread, we just need to copy the updateActions but now the raw lepus::Value is taken to js thread either, it may causing the lepus::Value freed on the js thread

this relands mr: !58616

[BugFix][Android][List] Fix error scroll state when invoking stop autoScroll (e899962) @DwwWxx

Move setScrollState(SCROLL_STATE_SCROLL_ANIMATION) to UIListAutoScroller#onAutoScrollStart callback to avoid error scroll sta te when developer invoke stop autoScroll method.

[BugFix] fix response handler callback lifetimes (8837bac) @nhsprite

Ensure the response handler listener will be triggered correctly even if there is no reference to the handler object in javascript. Making it work the same way as Promise object.

Introduce unit tests for response handler behavior in both JSI and Lepus runtimes. Tests verify that listener callbacks fire even after the handler object is released, and that delegate invocation occurs as expected.

Key changes:

• Add BUILD targets and test groups for JS and Lepus response handler tests.
• Implement response_handler_in_js_unittest.cc and response_handler_in_lepus_unittest.cc.
• Refactor JS handler to avoid weak_from_this usage:
  • Capture native_app weak_ptr directly and guard against destruction.
  • Use ApiCallBack and delegate-driven invocation for JS callbacks.
  • Make ConvertBundleInfoToPiperValue static and require app pointer.
• Refactor Lepus handler closure capture to use delegate pointer safely, relying on LazyBundleLoader guarantees.
• Fix renderer_functions to copy closure arg to ensure safe asynchronous use.

[BugFix] Abort renderer only when DevTool is enabled in debug without LogBox (8683aa6) @ShouruiSong

RenderFatal previously aborted the renderer whenever debug was enabled and LogBox was disabled. This caused unintended aborts in environments without DevTool. The condition is updated to require DevTool to be enabled in addition to debug mode, ensuring aborts only occur in the intended developer setup.

Change:

• Add IsDevToolEnabled() to should_abort condition alongside IsLynxDebugEnabled() and !IsLogBoxEnabled().

[Optimize] Use RunNowOrPostTask to ensure UI thread execution for callbacks (4be30bf) @nhsprite

Replace direct PostTask usage with fml::TaskRunner::RunNowOrPostTask when dispatching UI-thread work in LynxResourceLoaderHarmony. This ensures tasks execute immediately if already on the UI thread, reducing latency and avoiding unnecessary queuing. Previous behavior always posted asynchronously; new behavior runs inline when possible, otherwise posts to the runner.