♻️ [PANA-5123] Assign node ids in preorder when serializing #4002
Conversation
| const expectedHost = expectNewNode({ type: NodeType.Element, tagName: 'div' }) | ||
| const shadowRootNode = expectNewNode({ type: NodeType.DocumentFragment, isShadowRoot: true }) | ||
| const child = expectNewNode({ type: NodeType.Element, tagName: 'span' }) |
There was a problem hiding this comment.
The helper function expectNewNode() assigns ids to each node in order, so we have to reverse the order of these calls to get ids that match the new approach.
|
|
||
| expect(serializeDocumentNode(document, NodePrivacyLevel.ALLOW, transaction)).toEqual({ | ||
| type: NodeType.Document, | ||
| id: 0, |
There was a problem hiding this comment.
It was necessary to add this because serializeDocumentNode() used to return a node with no id assigned, but now it returns a node with an id. Conveniently, the id of the document node is always zero with pre-order assignment!
| export * from '../types' | ||
|
|
||
| export { serializeNodeWithId } from '../domain/record' | ||
| export { serializeNode, serializeNode as serializeNodeWithId } from '../domain/record' |
There was a problem hiding this comment.
There are some known external callers of this function in Datadog-internal code. I've exported the function under both the old and new names for compatibility, but I plan to update those callers once this change ships, and then we can remove the old name.
Motivation
The current DOM serialization algorithm serializes a node's children before assigning it an id. This results in a post-order assignment, like this:
This is a fine solution for the current data format; node ids are recorded explicitly, so we can assign them using whatever scheme we wish. However, the new format assigns node ids implicitly in the order that the nodes appear, which means that parent nodes will naturally be assigned a node id before their descendants. The new format uses a pre-order assignment, which looks like this:
This difference in node id numbering is the biggest source of complexity when comparing the output of the current DOM serialization algorithm with the output of the new one. It'd be ideal if both algorithms used the same ids; this would make translation between their representations straightforward.
The good news is that, because the current data format records node ids explicitly, there's no backwards compatibility issue with simply changing how node ids are assigned. So, we can switch the existing algorithm to use a pre-order assignment, and eliminate this impedance mismatch between the two algorithms.
Changes
This PR:
serializeNodeWithId()toSerializationTransaction#assignId().SerializationTransaction#assignId()to do the work.serializeNodeWithId(), which now does nothing, and replaces calls to it with direct calls toserializeNode().1to0, for better alignment between the old and new algorithms.Test instructions
You can test the changes using the "Live Replay" tab of the browser SDK extension.
Checklist