Reducing Allocation & GC Pressure When Executing Cached AST in Jint

[koobooteam]

We’ve been using Jint extensively and are very satisfied with its flexibility and design.

Our current execution pipeline is:

• Parse JavaScript source into an Esprima.Ast.Script (once)
• Cache the parsed AST
• Execute via engine.Evaluate(ast)

This avoids repeated parsing costs, but during runtime we’re still observing:

Noticeable memory growth under load
Frequent GC cycles

We see many objects in memory like Jint.Native.JsNumber.

We’d like to better understand the intended performance model of Jint and whether we’re missing a more optimal pattern. Specifically:

• Is AST reuse alone sufficient, or are there additional layers (e.g., compiled/bytecode caching) we should leverage?

• Does Engine reuse vs recreation significantly impact allocation behavior?

We’re aiming to run a high-throughput scenario (many script executions per second), so minimizing allocation rate is critical.