Treebuilder.inspecificscope optimizations

[JH-A-Kim]

Performance Optimization: Algorithmic Improvement for HtmlTreeBuilder.inSpecificScope

Executive Summary

This PR addresses a significant performance bottleneck in the HTML parser's scope-checking logic. Through profiling with Java Flight Recorder (JFR), we identified that HtmlTreeBuilder.inSpecificScope and its dependency on Arrays.binarySearch were responsible for a dominant portion of CPU cycles.

By replacing repeated binary search lookups with a pre-calculated bit flag on the Tag object, we have shifted the algorithmic complexity of stack traversal from $O(n \log m)$ to a strictly linear $O(n)$.

Key Performance Gains

In synthetic workloads testing deep nesting (depths up to 120), the following improvements were observed:

CPU Profiling (JFR Samples)

Method	Before (% / Samples)	After (% / Samples)	Improvement
Arrays.binarySearch	33.8% / 52	19.2% / 25	~52%
HtmlTreeBuilder.inSpecificScope	46 samples	25 samples	~46%
TreeBuilder.stepParser	30 samples	< 5 samples	~83%+
Tokeniser.read	29 samples	< 5 samples	~83%+

Call Stack Hot Paths

Method	Before (Samples)	After (Samples)	Delta
StringUtil.inSorted	52	25	-27
HtmlTreeBuilder.inSpecificScope	46	25	-21
HtmlTreeBuilderState.process	34	5	-29

Algorithmic Improvement

• Original Approach: As the parser walks the open-element stack, it performs membership checks using StringUtil.inSorted, which triggers Arrays.binarySearch for every step. Total complexity: $O(n \log m)$ where $n$ is stack depth and $m$ is the count of boundary tags ($m \approx 9$).
• Optimized Approach: Canonical boundary tags (e.g., applet, caption, html, table, etc.) are now pre-marked with a Tag.InScope bit flag during tag setup. The traversal now uses a constant-time bitwise check (el.tag().is(Tag.InScope)), resulting in a linear $O(n)$ walk.

Memory & Stability

Memory behavior remains neutral. The gains are strictly algorithmic and do not increase heap pressure or Garbage Collection frequency.

Metric	Before	After	Status
Sampled Allocation Weight	1320.1 MB	1319.3 MB	Neutral
Total GC Pause	16.9 ms	17.3 ms	Neutral
Max Heap after GC	6.88 MB	7.01 MB	Neutral

Justification

This optimization fundamentally improves the scalability of the parser for modern, deeply nested web documents. By reducing the per-step overhead of scope checking to a single bitwise instruction, we ensure Jsoup remains performant and deterministic in high-throughput or resource-constrained environments.