Context Isolation transition to contextBridge for secure API exposure

Direct answer

With contextIsolation: true (default since Electron 12), the preload script and renderer run in separate V8 contexts. Every argument and return value crossing contextBridge is structured-cloned, so high-frequency APIs pay a per-call cost that grows with payload size and complexity. Unlike shared state with isolation disabled, there is no zero-copy reference sharing. Functions cannot be cloned, so each exposed method becomes an async trampoline that allocates a promise and a message per call. Per-property access on a bridged object also crosses the boundary, meaning deep object graphs or getter-heavy proxies multiply crossings dramatically.

Confirmed facts versus likely explanation

Item	Status	Notes
Values cross via structured clone	Confirmed	Applies to all arguments and return values through contextBridge.
Class instances lose prototypes	Confirmed	They arrive as plain objects; Error stacks are also lost.
Functions are not cloneable	Confirmed	Exposed methods are async wrappers, not direct references.
Exact per-call millisecond cost	Uncertain	Varies by Electron version, V8 version, and payload shape; measure on your target version.
ArrayBuffer copy versus transfer	Uncertain	Behavior is version-sensitive; do not assume zero-copy without benchmarking.

Steps to reduce latency without weakening isolation

1. Expose coarse-grained methods that return complete result objects instead of fine-grained property proxies.
2. Batch many small operations into a single bridge call so one clone replaces many.
3. For streaming or telemetry, push events from the main process and coalesce or throttle updates to display cadence; render from a local renderer buffer.
4. Keep heavy computation in the main process or a worker, and send only summarized results across the bridge.
5. For large binary payloads, prefer ArrayBuffer or TypedArray over deeply nested plain-object trees, and avoid re-wrapping data on every update.
6. Do not set contextIsolation: false or nodeIntegration: true to remove the boundary.

Verification

Build a minimal app with contextIsolation: true and time N round-trips using performance.now(), comparing a small object against a 10,000-element array. Record a Chrome DevTools performance trace on the high-frequency path to check whether serialization blocks the renderer main thread. Compare a fine-grained bridged proxy against one batched method returning a single object to quantify per-crossing overhead.

Missing diagnostic detail: What Electron version are you targeting? ArrayBuffer transfer semantics and preload sandbox behavior differ across versions, and that determines whether binary transfer is a safe optimization for your case.