Teralizer - Replication Package

This (https://doi.org/10.5281/zenodo.17950380) is a replication package for the paper:

Teralizer: Semantics-Based Test Generalization from Conventional Unit Tests to Property-Based Tests

Our work proposes a semantics-based test generalization approach that automatically transforms conventional unit tests into property-based tests by extracting specifications from implementations via single-path symbolic analysis. We demonstrate this approach through Teralizer, a prototype tool for Java that transforms JUnit tests into property-based jqwik tests.

---

Contents

• Links
• Package Contents
• Quick Start
• Verification Workflows
• Complete Reproduction
• Project Structure
• Citation
• License

---

Links

Resource	Location
Zenodo Archive	10.5281/zenodo.17950380
Paper (arXiv)	arXiv:2512.14475
Artifact Repository	glockyco/Teralizer
Paper Repository	glockyco/Teralizer-Paper

---

Package Contents

Archive	Size	Contents
teralizer-results	~1 MB	Tables, figures, HTML notebooks
teralizer-core	~250 MB	Code, database dumps, reference outputs
teralizer-projects-primary	~45 MB	EqBench + commons-utils source code
teralizer-projects-extended-sample	~170 MB	100 sampled RepoReapers projects
teralizer-projects-extended	~1.7 GB	All 1161 RepoReapers projects
teralizer-data-primary	~1.1 GB	Logs, tool reports, generalized tests
teralizer-data-extended	~260 MB	Logs, tool reports, generalized tests

What to download:

• Browse results only: teralizer-results
• Verify analysis: teralizer-core
• Verify pipeline: teralizer-core + teralizer-projects-extended-sample
• Full reproduction: teralizer-core + teralizer-projects-primary + teralizer-projects-extended

---

Quick Start

See REQUIREMENTS.md for system requirements and INSTALL.md for detailed setup instructions.

cd replication
./quick-start.sh

This starts PostgreSQL, imports the database dumps, and launches Jupyter Lab.

Access points (open in browser after setup completes):

• Jupyter Lab: http://localhost:8888
• Database UI (Adminer): http://localhost:18080
  • System: PostgreSQL
  • Server: postgres
  • Username: teralizer
  • Password: teralizer
  • Database: postgres_dev (or postgres_test)

Stopping services:

docker compose down      # Stop containers (preserves data)
docker compose down -v   # Stop and remove all data

---

Verification Workflows

Three workflows verify the artifact at increasing levels of depth:

Workflow	What it does	Archives needed	Output
1. Inspect	Browse pre-computed results	teralizer-results	—
2. Verify analysis	Re-run notebooks on existing data	teralizer-core	verify/
3. Verify pipeline	Re-run data collection	teralizer-core + teralizer-projects-*	replicate/

Workflow 2 should produce outputs identical to original/. Workflow 3 outputs may differ due to resource limits and external factors (see Complete Reproduction).

Workflow 1: Inspect Pre-computed Results (~5 min)

Browse the pre-computed results without re-running anything.

1. Run quick-start:

   cd replication && ./quick-start.sh

2. Verify database import:

   ./scripts/verify-results.sh

3. Explore:

   • Jupyter: http://localhost:8888 — browse notebooks
   • Adminer: http://localhost:18080 — query databases
   • Files: analysis/output/original/ — pre-computed tables and figures

Workflow 2: Verify Analysis (~10 min)

Confirm the analysis code produces identical results on the same data.

1. Re-run all notebooks:

   ./scripts/run-notebooks.sh verify

2. Compare outputs:

   ./scripts/verify-outputs.sh original verify

Expected: All outputs match exactly. The analysis is deterministic.

Workflow 3: Verify Pipeline (~15 min)

Confirm the data collection pipeline executes successfully.

1. Run pipeline on a subset of projects:

   ./scripts/run.sh --dataset extended --count 5

2. Run analysis on the new data:

   ./scripts/run-notebooks.sh replicate

3. Compare outputs (differences expected due to non-determinism):

   ./scripts/verify-outputs.sh original replicate

For full reproduction of all projects, see Complete Reproduction.

Analysis Notebooks

All evaluation figures and tables in the paper are generated by notebooks in analysis/notebooks/. Outputs are saved to analysis/output/ as LaTeX tables, PDF figures, and CSV data.

Notebook	Paper Section	Description
dataset-characteristics.ipynb	Evaluation Setup	Dataset statistics and characteristics
rq1-mutation-detection.ipynb	RQ1, RQ2	Mutation scores, constraint complexity
rq2-test-suite-effects.ipynb	RQ3	Test suite size and runtime effects
rq3-runtime-requirements.ipynb	RQ4	Teralizer efficiency analysis
rq4-limitations.ipynb	RQ5, RQ6	Exclusion causes (primary + extended)

---

Complete Reproduction

Full reproduction requires significant compute time and may produce non-identical results due to:

• Machine-dependent resource limits (timeouts, memory)
• Evaluated projects with unavailable dependencies (artifacts removed from repositories)
• Evaluated projects with unpinned dependency versions (breaking changes in newer versions)

Extended Dataset (~15 hours)

./scripts/run.sh --dataset extended

Processes all 1161 RepoReapers projects. Runtime is relatively short because most projects fail to complete the full processing pipeline.

Primary Dataset (~100+ hours)

The primary dataset requires a two-phase workflow:

1. Generate tests (EvoSuite):

   ./scripts/run.sh --dataset primary --phase generation

2. Generalize tests:

   ./scripts/run.sh --dataset primary --phase generalization

Analyzing Reproduced Data

./scripts/run-notebooks.sh replicate
./scripts/verify-outputs.sh original replicate

---

Project Structure

teralizer/
├── README.md                   # This file
├── INSTALL.md                  # Installation instructions
├── REQUIREMENTS.md             # System requirements
├── LICENSE-MIT                 # MIT license (code)
├── LICENSE-CC-BY-4.0           # CC BY 4.0 license (data, docs)
├── src/                        # Teralizer Java source code
├── analysis/
│   ├── notebooks/              # Jupyter analysis notebooks
│   ├── src/                    # Python analysis modules
│   └── output/                 # Generated tables, figures, data
├── replication/
│   ├── docker-compose.yml      # Docker services configuration
│   ├── quick-start.sh          # One-command setup script
│   ├── datasets/               # Database dumps
│   └── scripts/                # Automation scripts
├── project-configs/            # Pipeline configuration files
└── docs/                       # Architecture documentation

---

Citation

@misc{glock_2025_teralizer,
  title={Teralizer: Semantics-Based Test Generalization from Conventional Unit Tests to Property-Based Tests},
  author={Johann Glock and Clemens Bauer and Martin Pinzger},
  year={2025},
  eprint={2512.14475},
  archivePrefix={arXiv},
  primaryClass={cs.SE},
  url={https://arxiv.org/abs/2512.14475},
}

---

License

This artifact uses dual licensing:

Component	License
Source code (Java, Python, scripts)	MIT
Data, documentation	CC BY 4.0

Analyzed projects retain their original licenses.