His2Trans

His2Trans is a knowledge-guided agentic framework for project-level C-to-Rust migration. It builds a Rust scaffold for the target C project, translates function bodies with retrieved historical migration knowledge, and then refines the assembled crate with compilation, semantic, and unsafe-code feedback.

This repository contains the framework implementation and the minimal paper-aligned artifacts needed to reproduce the current reported metrics.

Repository Layout

• framework/: the C-to-Rust migration pipeline. The main entry is framework/batch_test_staged.sh.
• data/ohos/: compact OpenHarmony headers and five small source-project bundles used by the lightweight framework examples.
• data/test_module_rust_tests/: Rust test harnesses for the open-source RQ2 project set.
• data/paper_artifacts/: minimal final translated Rust crates, metric scripts, and verified result files for the current paper.
• data/paper_metric_exports/: paper-aligned reference tables and generated metric summaries.
• scripts/export_current_plot_metrics.py: regenerates the paper metric CSV/JSON/Markdown summaries from the shipped artifacts.
• tests/: lightweight tests for the paper metric helper code.

Paper-Aligned Results

The current paper uses ten OpenHarmony modules and eight open-source C projects.

Dataset	Incremental compilation	Test pass rate	Unsafe ratio	Required unsafe ratio
OpenHarmony 10 modules	100.00%	94.92%	16.35%	8.48%
OSS8 projects	100.00%	100.00%	8.59%	2.94%

The OpenHarmony ablation table is:

Setting	Incremental compilation	Test pass rate	Unsafe ratio
Knowledge-guided function translation	95.82%	39.29%	15.95%
With compiler-driven per-function repair	100.00%	94.92%	22.52%
With project-level agentic refinement	100.00%	94.92%	16.35%

Regenerate the repository-local metric exports with:

python3 scripts/export_current_plot_metrics.py

This rewrites:

• data/paper_metric_exports/current_plot_metrics_alignment.json
• data/paper_metric_exports/current_plot_metrics_alignment.md
• data/paper_metric_exports/reference_tables/*.csv
• data/paper_metric_exports/generated_structured_json/rq1_ohos10_method_comparison.json
• data/paper_metric_exports/generated_structured_json/rq2_oss8_method_comparison.json
• data/paper_metric_exports/generated_structured_json/rq3_ohos10_ablation.json
• data/paper_metric_exports/generated_structured_json/rq4_case_evidence.json

The export script checks the key numbers above and fails if they drift.

Reproducing Archived Metrics

OpenHarmony 10-module archive

The OHOS10 archive is:

data/paper_artifacts/deepseek_v4_pro_ohos10_0613_harness_fixed_v2/

Fast metric replay:

python3 data/paper_artifacts/deepseek_v4_pro_ohos10_0613_harness_fixed_v2/scripts/run_archived_ohos_metrics.py \
  --output-dir /tmp/his2trans_ohos10_metrics

This recomputes incremental-compilation and unsafe metrics from the archived Rust crates, and reuses the verified gtest result files shipped in the archive. Re-running the full OHOS C gtest bridge requires the original full OpenHarmony/self-contained source tree, which is too large for this minimal open-source artifact.

OSS8 archive

The OSS8 archive is:

data/paper_artifacts/deepseek_v4_pro_oss8_0613_rq2_100pct_minimal/

Full metric replay:

bash data/paper_artifacts/deepseek_v4_pro_oss8_0613_rq2_100pct_minimal/reproduce.sh \
  /tmp/his2trans_oss8_metrics

This reruns the archived OSS8 Rust tests, incremental compilation, unsafe analysis, and warning counting from the shipped final crates.

Environment

The paper experiments were run on Ubuntu Linux with Clang 14, Rust nightly, and Python 3.13. The framework environment used Python 3.11 through the c2r_frame conda environment.

Required system tools:

• Python 3.10+ for framework scripts; Python 3.13 was used for the final paper metric scripts.
• Rust nightly with cargo and clippy.
• Clang and libclang.
• Conda, if you want to use the provided framework environment setup.

Set up the framework conda environment:

bash framework/setup_conda_env.sh
export CONDA_OVERRIDE_CUDA=
export CONDA_NO_PLUGINS=true
conda activate c2r_frame
export NLTK_DATA="$(pwd)/framework/data/nltk_data"

For external-API translation runs:

export RUSTUP_TOOLCHAIN=nightly
export USE_VLLM=false
export EXTERNAL_API_BASE_URL="https://api.deepseek.com/beta"
export EXTERNAL_API_MODEL="deepseek-coder"
export EXTERNAL_API_KEY="YOUR_KEY"
export USE_PREPROCESSING=false
export HF_HOME="$(pwd)/framework/data/my-huggingface"
export TRANSFORMERS_CACHE="$HF_HOME"
export HF_HUB_CACHE="$HF_HOME/hub"

Framework Smoke Run

Run a single shipped OpenHarmony example without RAG or learned-KB retrieval:

cd framework
bash batch_test_staged.sh \
  --layered --incremental --max-repair 1 \
  --max-parallel 1 --max-parallel-workers 1 \
  --run-rag false --skip-learned-kb --use-libclang \
  --suite ohos \
  --only osal__0bc4f21396ad \
  --run-dir smoke_api

Outputs are written to framework/translation_outputs/<run-dir>/.

Development Checks

Run the lightweight metric tests:

python3 -m pytest -q tests

Compile-check the Python sources:

python3 -m py_compile \
  scripts/export_current_plot_metrics.py \
  framework/*.py \
  framework/generate/*.py \
  framework/scripts/*.py \
  framework/scripts/agentic_repair/*.py