[RUM-10500] [V3] Attributes encoding and validation

[marco-saia-datadog]

What does this PR do?

Starting from v3, we are changing the way attributes are handled to improve safety, consistency and compatibility with native SDKs, and to support more attribute types.

What’s New

• Unsupported values (functions, symbols, non-finite numbers, etc.) are automatically dropped, with warnings, to prevent crashes in native SDKs.
• Common types such as Date, Error, and Map are properly serialized
• Nested objects are automatically flattened into dot syntax (a.b.c = value), matching the format expected by native SDKs and avoiding ambiguous outcomes.
• Custom encoders can be registered by users to support domain-specific types.
• Primitives and arrays at the root are wrapped under a context key to maintain schema consistency.

Motivation

In React Native, it’s common for developers to pass complex objects as attributes (e.g. partial or full Redux/Zustand state, raw Error objects). These may contain unserializable values that would break serialization. While ideally users should only pass clean dictionaries of encodable values, in practice they expect the SDK to handle arbitrary objects to some extent.

This change ensures:

• Everything passed from JS to native is safe and JSON-serializable
• Flattening happens in JS, so users don’t need to handle dot syntax manually (which is not idiomatic in React Native), and we don't have to rely on the native SDKs either
• Ambiguities between nested objects and dot-syntax keys are avoided

Example Transformations

• "foo" → { context: "foo" }

• { user: { profile: { name: "Alice" } } } → { "user.profile.name": "Alice" }

• new Error("Boom") → { "context.name": "Error", "context.message": "Boom", "context.stack": "...", ... }

• new Map([["k1", 1], ["k2", "v2"]]) →

  {
    "context": [
      { "key": "k1", "keyType": "string", "value": 1 },
      { "key": "k2", "keyType": "string", "value": "v2" }
    ]
  }

• Unsupported values like Symbol("x") or () => {} → {} (dropped with warning)

Benchmarks

• Flat/nested objects: ~1M ops/sec
• Deeply nested objects (20+ levels): ~233k ops/sec
• Maps with mixed keys: ~405k ops/sec
• Arrays of objects: ~28k–33k ops/sec (expectedly slower due to recursion)

[sbarrio]

Left some minor comments. Good work!

[marco-saia-datadog]

Thanks @sbarrio 🙌 I have addressed your comments in a new commit. I still have to do some further manual testing before merging this. After I am done, I will squash the commits and merge this in v3 :)

[marco-saia-datadog]

@sbarrio I made some changes to introduce the 128 attributes limit + additional testing for this new logic and for making sure encoding does not modify the original objects

[sbarrio]

LGTM!

Just a heads up, due to the rename from DdSDK -> NativeDdSdk you will need to adapt all calls to the attributes API. These are the methods that will need adapting:

Attributes API

addAttribute
removeAttribute
addAttributes
removeAttributes

Please let me know if you want me to handle the merge or if you have any questions with it 🙂

[datadog-datadog-prod-us1]

✅ Tests

🎉 All green!

❄️ No new flaky tests detected
🧪 All tests passed

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: c1145b2 | Docs | Datadog PR Page | Was this helpful? Give us feedback!}