✨ split RumEvent into RendererRumEvent and MainRumEvent

[bcaudan]

Motivation

This allows the Electron SDK to properly type events coming from the main process versus the renderer process, and to support future execution-context events that will be handled by the main process. Main-process code could previously construct RUM events of any type in the full RumEvent union, including action/long_task, even though those never actually originate from the main process.

This builds on rum-events-format#437 (merged), which introduces schemas/rum-events-electron-schema.json: a RumEvent union schema mirroring the browser schema but limited to the event types the main process actually emit.

Changes

• Generate two separate RUM event unions: RendererRumEvent (browser schema, unchanged) and MainRumEvent (new electron schema), and make ServerRumEvent a discriminated union on source.
• Move RUM event/type definitions under src/domain/rum/types/.
• Remove the now-impossible action branch from ViewCollection's main-only counter path.

Test instructions

1. Run the app in Electron, trigger a click action in the renderer, and confirm the view document's action.count still increments (renderer-origin action events are unaffected by the split).
2. Trigger a main-process error (e.g. a crash) and confirm it still reaches RUM intake with error.source_type: nodejs.
3. Try making main-process code construct an action-typed ServerRumEvent and confirm yarn typecheck now rejects it.

Checklist

• Tested locally (playground)
• Added unit tests for this change.
• Added e2e/integration tests for this change.
• Updated related documentation.
• Agentic code review findings addressed or explicitly dismissed.

[bcaudan]

@codex review

[cursor]

Stale comment

PR Review — Score: 4.7 / 5

This is a well-motivated type-boundary refactor that correctly splits browser vs electron RUM schemas, threads the distinction through ServerRumEvent via source, and removes the now-dead main-process action counter path. Tests and assembly call sites are updated consistently, and the change is compile-time safe without altering runtime intake behavior. I would approve.

Why 4.7: Clear motivation tied to rum-events-format#437, disciplined file moves under src/domain/rum/types/, discriminated ServerRumEvent in event.types.ts, targeted ViewCollection cleanup, and unit tests that now assert the main-process subset (error/resource only).

Why not 5: Shared FormatHooks still types RUM assembly with the broad RumEvent union, which forces a cast in MainAssembly and leaves hook callbacks process-agnostic. docs/DEVELOPMENT.md still documents the old src/rumEvent.types.ts path.

---

Findings

• [Minor] Stale schema docs — docs/DEVELOPMENT.md still references the removed src/rumEvent.types.ts path instead of the new src/domain/rum/types/ layout.
• [Minor] Hook pipeline still process-agnostic — FormatHooks continues to use RecursivePartial<RumEvent>, so main-process assembly needs an explicit cast rather than getting compile-time guarantees end-to-end.
• [Nit] Renderer pipeline typing — RendererPipeline still casts bridge payloads to the broad RumEvent union; RendererRumEvent would complete the split on the renderer side.

---

Architectural flow

sequenceDiagram
    participant Raw as RawRumData
    participant MainAsm as MainAssembly
    participant Hooks as FormatHooks
    participant EM as EventManager
    participant Pipe as RendererPipeline
    participant VC as ViewCollection

    Raw->>MainAsm: main-process raw RUM
    MainAsm->>Hooks: triggerRum MAIN
    Hooks-->>MainAsm: hook overrides
    MainAsm->>EM: ServerMainRumEvent

    Pipe->>Hooks: triggerRum RENDERER
    Pipe->>EM: ServerRendererRumEvent

    EM->>VC: ServerRumEvent by source
    Note over VC: main only counts error and resource

Loading

Before: A single RumEvent union backed by the browser schema was used for both processes, so main-process code could type-check impossible event kinds (action, long_task). ServerRumEvent.data was untyped with respect to origin.

After: Browser and electron schemas generate RendererRumEvent and MainRumEvent; ServerRumEvent is a discriminated union on source, and main-only collection paths (e.g. view counters) narrow to the electron subset.

  

_{Sent by Cursor Automation: electron-sdk reviews}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2409b5708f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[cursor]

PR Review — Score: 4.9 / 5

This is a well-executed type-boundary refactor that splits browser vs electron RUM schemas, threads the distinction through ServerRumEvent via source, and removes the dead main-process action counter path. Follow-up commits since the first review addressed the documentation, static-analysis ignore list, renderer pipeline typing, and barrel-export rationale. Runtime intake behavior is unchanged and the compile-time guarantees are materially stronger. I would approve.

Why 4.9: Clear motivation tied to rum-events-format#437, disciplined relocation under src/domain/rum/types/, discriminated ServerRumEvent in event.types.ts, targeted ViewCollection cleanup, renderer pipeline now typed with RendererRumEvent, and prior Codex/automation feedback incorporated in the latest head.

Why not 5: FormatHooks.triggerRum still returns the broad RumEvent union (you noted this can wait for a follow-up), so MainAssembly keeps a narrow cast at the hook boundary. That is an intentional tradeoff, not a functional gap.

---

Architectural flow

sequenceDiagram
    participant Raw as RawRumData
    participant MainAsm as MainAssembly
    participant Hooks as FormatHooks
    participant EM as EventManager
    participant Pipe as RendererPipeline
    participant VC as ViewCollection

    Raw->>MainAsm: main-process raw RUM
    MainAsm->>Hooks: triggerRum MAIN
    Hooks-->>MainAsm: hook overrides
    MainAsm->>EM: ServerMainRumEvent

    Pipe->>Hooks: triggerRum RENDERER
    Pipe->>EM: ServerRendererRumEvent

    EM->>VC: ServerRumEvent by source
    Note over VC: main counts error and resource only

Loading

Before: A single browser-schema RumEvent union backed both processes, so main-process code could type-check impossible event kinds (action, long_task). ServerRumEvent.data was not tied to source.

After: Browser and electron schemas generate RendererRumEvent and MainRumEvent. ServerRumEvent is a discriminated union on source, and main-only collection paths narrow to the electron subset.

  

_{Sent by Cursor Automation: electron-sdk reviews}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f881df5c25

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".