bugfix: decode prefermance issue fix

[nuclearg]

What this PR does:
decode.go 里面的 Decode() 每次在解析map或list时都会触发一次对 d.refHolders 的轮询 notify()，在解析复杂对象时（我的场景需要处理约10M的数据），这个 Decode() 会被疯狂递归调用，无数次对这个 d.refHolders() 进行无用的 notify()，产生严重的性能问题

我的机器是 MacBook Pro M3Pro 36G，在我本机上，处理10M的数据时，rt将达到5秒。此性能不可接受。

本次改动对 Decode() 的递归次数进行计数，只在最外层的 Decode() 时进行 notify() 操作。

可以通过MR中的 decode_largedata_test.go 验证。此测试用例随机生成一个巨大的map/list相互嵌套的结构并执行序列化/反序列化操作。未引入我的修改时，反序列化的rt远高于序列化（秒级）。引入此修改后序列化与反序列化rt持平（毫秒级）

Which issue(s) this PR fixes:

Fixes #

Special notes for your reviewer:

Does this PR introduce a user-facing change?:

NONE

[Copilot]

Pull Request Overview

This PR addresses a performance bug in the Decode() function where repeated unnecessary calls to notify() on d.refHolders are causing significant slowdowns when processing large, complex data structures. The changes introduce a recursive depth counter in Decode() so that d.refHolders.notify() is only called at the outermost level.

• Introduced a decodeRecursiveDepth counter in the Decoder struct
• Updated Decode() to increment/decrement this counter and conditionally call notify()
• Added a large data test case (decode_largedata_test.go) to validate the performance improvement

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
decode_largedata_test.go	Added a test to generate and verify performance with large data
decode.go	Updated Decode() with a recursive depth counter to control notify() calls

Comments suppressed due to low confidence (2)

decode.go:243

• [nitpick] The variable name 'decodeRecursiveDepth' is descriptive, but consider a shorter variant like 'recursiveDepth' if it would improve consistency and readability within the codebase.

d.decodeRecursiveDepth++

decode.go:242

• If the Decoder type might be used concurrently, please ensure that proper thread-safety measures are in place for 'decodeRecursiveDepth' to avoid potential race conditions.

func (d *Decoder) Decode() (interface{}, error) {

[tiltwind]

@nuclearg 感谢

1. 有一个错误要解决

Error: decode_benchmark_test.go:54:2: ineffectual assignment to `now` (ineffassign)
	now = time.Now()

2. 测试和压测可以考虑改一下：

func TestMultipleLevelRecursiveDep(t *testing.T) {

	// TODO : 测试 encode 和 decode 后的对象和源对象结构是一致的，值是相等的

}

func BenchmarkMultipleLevelRecursiveDep(b *testing.B) {
	// TODO: 构造一个多层循环依赖的对象, 无需大对象， 压测情况大批量请求可以看出是否有性能改进
	for i := 0; i < b.N; i++ {
		// TODO : 循环 encode 和 decode
	}
}

[nuclearg]

@tiltwind 基准测试用例已提交

[wongoo]

LGTM

[AlexStocks]

@nuclearg please fix the ci failure

[AlexStocks]

@nuclearg please fix the ci failure.

[wongoo]

use t.Log instead

[nuclearg]

NO. This is a perf fix, run slow is a failure.

[tiltwind]

the CI failed

=== RUN   TestMultipleLevelRecursiveDep2
hessian2 serialize 3.948985828s 10660KB
hessian2 deserialize 1.218173434s
    decode_benchmark_test.go:75: deserialize too slow
--- FAIL: TestMultipleLevelRecursiveDep2 (11.67s)

[AlexStocks]

@nuclearg any feedback?

[AlexStocks]

please fix the ci failure