{
  "added_at": "2026-10-19",
  "arxiv_id": "2406.04568",
  "arxiv_version_reviewed": "v1",
  "authors": [
    "Weike Fang",
    "Zhejian Zhou",
    "Junzhou He",
    "Weihang Wang"
  ],
  "bibtex_url": "https://awesomerea.org/papers/citations/paper-stacksight.bib",
  "code_url": null,
  "date_basis": "PMLR states that Volume 235 was published on 08 July 2024, and the paper page gives the same citation date. The July 21–27 conference dates and the June 2024 arXiv submission are not used as the publication date.",
  "date_basis_zh": "PMLR 标明第 235 卷于 2024 年 7 月 8 日出版，论文页面给出的引用日期与此相同。本条目不以 7 月 21—27 日的会议日期或 2024 年 6 月的 arXiv 提交日期作为出版日期。",
  "date_precision": "day",
  "description_en": "StackSight decompiles WebAssembly text-format functions into C++ with LLM prompting. A static analysis annotates each WAT instruction with the virtual-stack state, using symbolic notation where branches diverge; a one-shot chain-of-thought prompt then asks the model to recover variable meanings, summarize the function and write C++. With GPT-4 the share of outputs passing all benchmark tests rises from 19.7% to 36.5% on HumanEval-X and from 42.4% to 59.6% on MBXP relative to one-shot prompting; GPT-3.5 and Code Llama-7B-Instruct are also evaluated.",
  "description_zh": "StackSight 借助大语言模型提示，将 WebAssembly 文本格式（WAT）函数反编译为 C++。它先通过静态分析为每条 WAT 指令标注虚拟栈状态，在分支分叉处使用符号记法；再用单样例思维链提示，要求模型恢复变量含义、总结函数功能并写出 C++。使用 GPT-4 时，与单样例提示相比，通过全部基准测试的输出比例在 HumanEval-X 上从 19.7% 升至 36.5%，在 MBXP 上从 42.4% 升至 59.6%；论文还评估了 GPT-3.5 与 Code Llama-7B-Instruct。",
  "doi": null,
  "entry_url": "https://awesomerea.org/papers/paper-stacksight/",
  "entry_url_zh": "https://awesomerea.org/zh/papers/paper-stacksight/",
  "evidence_urls": [
    "https://proceedings.mlr.press/v235/fang24e.html",
    "https://proceedings.mlr.press/v235/",
    "https://raw.githubusercontent.com/mlresearch/v235/main/assets/fang24e/fang24e.pdf",
    "https://arxiv.org/abs/2406.04568v1"
  ],
  "first_arxiv_date": "2024-06-07",
  "id": "paper-stacksight",
  "json_url": "https://awesomerea.org/papers/paper-stacksight/record.json",
  "kind": "research",
  "markdown_url": "https://awesomerea.org/papers/paper-stacksight/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/papers/paper-stacksight/index.md",
  "metadata_notes": "PMLR lists Proceedings of the 41st International Conference on Machine Learning, PMLR 235, pages 13010–13028; no DOI is given. arXiv has a single version, v1 of 2024-06-07, whose comment cites the ICML 2024 proceedings. The models are gpt-3.5-turbo-1106, gpt-4-0125-preview and Code Llama-7b-Instruct; GPT-4 was compared only with one-shot prompting, while the GPT-3.5 and Code Llama baselines also use 3-, 5- and 10-shot prompts. The appendix gives the emcc and wasm2wat commands and the prompts. The paper links third-party tools such as WABT and Emscripten but no author code, prompts repository or evaluation harness.",
  "metadata_notes_zh": "PMLR 列出 Proceedings of the 41st International Conference on Machine Learning、PMLR 第 235 卷第 13010–13028 页，未给出 DOI。arXiv 仅有 2024-06-07 的 v1，其备注引用 ICML 2024 论文集。所用模型为 gpt-3.5-turbo-1106、gpt-4-0125-preview 与 Code Llama-7b-Instruct；GPT-4 仅与单样例提示比较，GPT-3.5 与 Code Llama 的基线另含 3、5、10 样例提示。附录给出 emcc 与 wasm2wat 命令及提示词。论文链接了 WABT、Emscripten 等第三方工具，但未提供作者代码、提示词仓库或评测框架。",
  "pdf_url": "https://raw.githubusercontent.com/mlresearch/v235/main/assets/fang24e/fang24e.pdf",
  "project_url": null,
  "publication_date": "2024-07-08",
  "publication_status": "conference-paper",
  "record_type": "paper",
  "runtime_tested": false,
  "scope_limitations_en": "Inputs are short exercise functions from the C++ splits of HumanEval-X (161 of 164 compiled) and MBXP (773 of 974), built as Emscripten 3.1.46 side modules at -Oz with function names kept by --profiling-funcs and converted with wasm2wat 1.0.33; minified, obfuscated or whole-application WebAssembly is not evaluated. Correctness is the share of outputs that compile and pass the benchmarks' test cases after a test oracle renames functions, inserts tests and adds namespace directives, with manual pass-by-value edits. Passing these tests is not proof of equivalence with the original module. Summaries are scored by BERTScore and SBERT; the user study had 15 participants on 20 snippets per dataset. wasm2c and an earlier neural decompiler were excluded for producing no compilable output, and results depend on specific hosted model versions.",
  "scope_limitations_zh": "输入是 HumanEval-X（164 个中 161 个可编译）与 MBXP（974 个中 773 个）C++ 子集中的简短练习函数，以 Emscripten 3.1.46 按 -Oz 编译为侧模块，通过 --profiling-funcs 保留函数名，再用 wasm2wat 1.0.33 转换；未评估经过压缩、混淆或完整应用规模的 WebAssembly。正确性指输出能够编译并通过基准测试用例的比例；测试判定器会重命名函数、插入测试并补充命名空间指令，按值传参则由人工修改。通过这些测试并不证明与原始模块等价。函数摘要以 BERTScore 和 SBERT 评分；用户研究有 15 名参与者，每个数据集 20 个代码片段。wasm2c 与一种早期神经反编译器因无法生成可编译输出而未纳入比较，结果也依赖特定版本的托管模型。",
  "targets": [
    "binary",
    "website"
  ],
  "title": "StackSight: Unveiling WebAssembly through Large Language Models and Neurosymbolic Chain-of-Thought Decompilation",
  "title_zh": "StackSight：借助大语言模型与神经符号思维链反编译揭示 WebAssembly",
  "url": "https://proceedings.mlr.press/v235/fang24e.html",
  "venue": "ICML 2024",
  "verification": "source-reviewed",
  "verified_at": "2026-10-18",
  "why_fits_en": "Adds LLM-based WebAssembly decompilation guided by static stack analysis, covering the bytecode format shipped in web applications alongside the native-code decompilation papers in this section.",
  "why_fits_zh": "补充由静态栈分析引导的大语言模型 WebAssembly 反编译方法，覆盖网页应用所交付的字节码格式，与本节的原生代码反编译论文并列。",
  "year": 2024
}