Skip identifier walk through empty FunctionEnvironments

[lahma]

Summary

Closures with zero parameters and zero local declarations — a tight inner method like sw.Start() in the stopwatch benchmark — end up with a FunctionEnvironment whose _slots and _dictionary are both null. Their own bindings are nothing. Identifier reads inside the body always resolve outward through the closure chain, but the env-walk in JintEnvironment.TryGetIdentifierEnvironmentWithBinding{,Value} still pays a virtual TryGetBinding / HasBinding per layer.

This PR skips those empty FunctionEnvironment layers outright in the chain walk:

if (record is FunctionEnvironment fenv
    && fenv._slots is null
    && fenv._dictionary is null
    && record._outerEnv is not null)
{
    record = record._outerEnv;
    continue;
}

Only FunctionEnvironment is targeted — ModuleEnvironment may carry indirect-import bindings even when slots/dict are null, and DeclarativeEnvironment block scopes effectively always have slots when they exist.

The hot-path benefit is small in isolation — the slot-binding cache from #2416 already short-circuits this on cache hits — but it speeds up first-call resolve and cache-miss fallback paths.

Benchmark (stopwatch.js)

Ryzen 9 5950X / .NET 10.0.7. Baseline + PR taken back-to-back. Run-to-run StdDev was high on this session (4-17ms), so individual numbers should be read with caution; the directional pattern is consistent though.

Method	Modern	main	PR	Δ time
Execute	var	249.5 ms	203.5 ms	-18%
Execute_ParsedScript	var	250.2 ms	257.1 ms	+3% (noise)
Execute	let	300.0 ms	290.9 ms	-3%
Execute_ParsedScript	let	253.1 ms	242.8 ms	-4%
Allocated	all	~38.5 MB	~38.5 MB	flat

Most of the gain is on the parse-each-iteration Execute (var) path where the slot cache from #2416 starts cold each iteration; here the empty-env skip removes a virtual call per identifier read in the warmup phase. Allocations are unchanged — this is purely a control-flow tweak.

Validation

• Jint.Tests (net10.0): 2861 passed, 0 failed.
• Jint.Tests.Test262 (net10.0): 98567 passed, 506 skipped, 0 failed (clean run; transient flakes under concurrent test load are unrelated).

The skip is gated on is FunctionEnvironment plus _slots is null && _dictionary is null, so any env that could ever resolve a binding is preserved. record._outerEnv is not null ensures we still hit the global lookup at the chain end.

Stacked work

Fifth of six related performance changes. The perf/drop-interlocked branch (originally #2418) was abandoned after CI revealed that pool fields on JintFunctionDefinition.State are genuinely shared across parallel test fixtures via the static Test262 harness script cache, so the atomic exchanges are required.

Last in the queue is perf/block-env-reset on lahma/jint — small block-env reset inlining. Will be PR'd after this one lands.

🤖 Generated with Claude Code