README.md

metadata

license: mit
language:
  - code
tags:
  - rust
  - go
  - julia
  - code
  - test-generation
  - low-resource-languages
  - unit-tests
  - benchmark
pretty_name: TestEval-LR
size_categories:
  - 1K<n<10K
dataset_summary: >
  TestEval-LR is a benchmark dataset for evaluating unit test generation
  capabilities of models on low-resource programming languages (Rust, Go, and
  Julia). Each sample provides a focal function's source code and metadata for
  generating idiomatic test code.

📚 TestEval-LR: Unit Test Generation Benchmark for Low-Resource Languages

TestEval-LR is a validation benchmark designed to measure how well models can generate unit tests for Low-Resource Programming Languages (LRPLs) — specifically Rust, Go, and Julia.

📌 Purpose

Evaluate how well a model can generate test code, given a focal function's source code and context.

📂 Dataset Structure

Each example contains:

• function_name: Name of the focal function.
• focal_code: Raw source code of the function (used for context).
• function_component: Detail information about the function like function signature,arguments definition,line range,...
• file_content: Content of file have the focal function.
• file_path: Relative path to the file in the repository.

Dataset Size

The dataset contains ~372–412 samples per language, depending on the source repository.

📝 Prompt Example for Test Generation

Suffix syntax for each language: suffix syntax is to put in the end of the prompt ( after wrap in the chat template if need ) to generate the expected test function and avoid hallucination

    "Rust":  "#[test]\nfn test_{function_name}() {"

    "Julia": "@testset \"{function_name} Tests\" begin"

    "Go":    "func Test{function_name_with_uppercase_first_letter}("

This dataset uses a structured prompt format for three programming languages: Rust, Julia, and Go.
Each language has two variants:

• instruct → Full instruction prompt including explicit guidance to generate a unit test for a given function.
• base → Minimal version that only contains the function code and a short comment reminding to check correctness.

Structure

prompts = {
    "Rust": {
        "instruct": (
            "{function_code}\n"
            "Generate Rust unittest for {function_name} function in module {file_path}:\n"
        ),
        "base": (
            "{function_code}\n"
            "// Check the correctness for {function_name} function in Rust\n"
        ),
    },
    "Julia": {
        "instruct": (
            "{function_code}\n"
            "Generate Julia unittest for {function_name} function in module {file_path}:\n"
        ),
        "base": (
            "{function_code}\n"
            "# Check the correctness for {function_name} function in Julia\n"
        ),
    },
    "Go": {
        "instruct": (
            "{function_code}\n"
            "Generate Go unittest for {function_name} function in module {file_path}:\n"
        ),
        "base": (
            "{function_code}\n"
            "// Check the correctness for {function_name} function in Go\n"
        ),
    },
}