Samsung Keyboard Link Boundary Node Splitting and Escaping

Phenomenon

When using Chrome on Android with Samsung Keyboard Text Prediction ON, typing Korean at the boundary of a link (<a>) element causes the following structural issues:

1. Forced Text Node Splitting: What should be a single continuous text node is physically split into two.
2. Caret/Input Escape and Duplication (Leakage): The input value jumps out of the link, and the last character of the link is often duplicated into the new node (e.g., typing ‘Ma’ after ‘com’ results in ‘com’ + ‘mMa’).
3. Loss of Composition Reference: The IME loses context between the link text and the new input, leading to the duplication and broken composition described above.

Reproduction Steps

1. On a Samsung Galaxy device, open Chrome and enable ‘Predictive text’ in Samsung Keyboard settings.
2. Create a link (<a>Link</a>) within a contenteditable region.
3. Place the caret at the very end of the link text.
4. Type a Korean character (e.g., ‘안녕’).

Observed Behavior (Based on Zero Trace Log)

• Step 72 (beforeinput): The target ranges correctly point inside the A tag during the insertCompositionText event.
• Step 73 (input): The moment the input is reflected in the DOM, the character (ㅇ) is inserted into the P tag, outside the A tag.
• As a result, the DOM structure becomes <p><a>Link</a>ㅇ</p>, unintentionally breaking the link formatting.

Expected Behavior

• The typed character should remain within the text node at the end of the <a> tag.
• The text node should not split; new content should be merged into the existing node.

Circular Sync Normalization Points

1. Normalization Cycle: Immediately after the input event, if a node split is detected, the algorithm reconciles the state back to the Model’s truth (ID-based structure) by “rotating” the stray text back into the link.
2. Shadow Composition Tracking: Even if the browser splits the DOM, the Model treats it as a single node, ensuring a merged output in the next rendering cycle.