Add event-loop-friendly async serialization and deserialization

[kyken]

Motivation

Processing large values synchronously can block the event loop, delaying timers, I/O, and other application work.

This change adds asynchronous APIs for transport-boundary processing without changing the existing synchronous APIs.

Summary

• Add serializeAsync and deserializeAsync as instance, static, and named exports.
• Yield to the event loop while traversing and copying large values.
• Add a configurable yieldRate with a default of 1024.
• Preserve support for special values, circular references, referential equality, custom transformers, and in-place deserialization.
• Keep the synchronous serialize and deserialize APIs unchanged.

Measured effect

Benchmark setup

Item	Value
Host	Mac15,13 (Apple M3, arm64)
Memory	16 GiB
OS	macOS 26.6.2
Runtime	Node.js v20.15.0
Package	Compiled output imported from dist/index.js
Input	Array of 500,000 numeric primitives
Payload	3,388,900 bytes as UTF-8 JSON

The synchronous serialized payload was created once and reused for all deserialization cases.

Each case was run once in the same Node.js process. The benchmark was started with --expose-gc; global.gc() was called between cases, followed by a 20 ms settling period.

Immediately before each operation, a 1 ms interval timer was started. The benchmark recorded elapsed time, timer callbacks until completion, and the maximum timer gap. The maximum gap includes time during which the operation blocked timer callbacks.

Operation	Elapsed	Timer callbacks	Maximum timer gap
serialize	219.47 ms	0	219.69 ms
serializeAsync, yieldRate=1	6430.25 ms	6311	63.15 ms
serializeAsync, yieldRate=1024	243.69 ms	165	70.02 ms
serializeAsync, yieldRate=8192	237.12 ms	60	89.49 ms
serializeAsync, yieldRate=65536	234.15 ms	6	121.99 ms
deserialize	2.48 ms	0	2.52 ms
deserializeAsync, yieldRate=1	6180.00 ms	6160	20.40 ms
deserializeAsync, yieldRate=1024	29.79 ms	30	1.51 ms
deserializeAsync, yieldRate=8192	26.07 ms	25	2.24 ms
deserializeAsync, yieldRate=65536	24.88 ms	6	10.59 ms

These are single-run directional measurements and may vary with CPU load, garbage collection, and operating-system scheduling.

yieldRate=1 causes substantial scheduling overhead. In this measurement, the default yieldRate=1024 keeps serialization close to the synchronous baseline while allowing timer callbacks to run. Higher values reduce overhead further but allow longer uninterrupted processing periods.

Impact and overhead

The overhead applies only to the asynchronous APIs. Existing synchronous consumers are unaffected.

yieldRate controls the trade-off between responsiveness and throughput: lower values yield more frequently but add overhead, while higher values reduce overhead but allow longer uninterrupted processing. The default is 1024 and can be adjusted for the workload.

[greptile-apps]

PR author is not in the allowed authors list.