exLong Dataset

This dataset is used to train and evaluate the exLong models on generating exceptional-behavior tests.

It has two subsets:

• 'with-EBT-name': provides the target test name in the prompt
• 'no-EBT-name': does not provide the target test name in the prompt

NOTE: the data format is customized for Code Llama models.

Language

This is a Java dataset

Dataset Structure

The input for the model contains the following context:

1. Method under test (MUT)
2. Relevant non-exceptional behavior test (nEBT)
3. Guard expression: the logical formula representing the constraints on the symbolic variables that must be true to follow that particular trace
4. Stack trace: the sequence of method invocations that start from the MUT and lead to the target throw statement
5. (Optional) Exceptional-behavior test name

Data fields

1. id: the identifier for each data example
2. instruction: prompt containing all the context
3. output: target exceptional-behavior test