Data

Bug-Fixing Commits

Bug-Fixing commits metadata extracted during the mining. The CSV file contains the following fields:

• ID : Commit HASH ID

• Repo_URL : GitHub URL of the repository

• Commit_URL : GitHub URL of the bug-fixing commit

• Message : Commit message of the bug-fixing commit

Data

Download CSV file (900 MB)

Code from Bug-Fixes

Raw source code extracted from the bug-fixing commits.

Each bug-fixing commit is represented by a folder named as the commit hash ID. In each folder there are two sub-folders:

• P_DIR: Java source code files before the bug-fixing commit

• F_DIR: Java source code files after the bug-fixing commit

Data

Download data (15 GB)

Extracted Bug-Fix Pairs (BFP)

Method pairs extracted from the bug-fixing commits.

Each bug-fix is represented by a folder with the corresponding commit hash ID. In each bug-fix folder there is a first level of folders representing the files, then a second level of folders representing the methods. In each method folders there are the following files:

• before.java : Method's source code before the fix

• after.java : Method's source code after the fix

• operations.txt : AST operations performed on the method as extracted by GumTreeDiff

• signature.txt : Fully qualified signatures of the method before/after the fix

Data

Download data (7 GB)

Dataset of Bug-Fix Pairs for small and medium methods. Each dataset is formed by training (80%), validation (10%), and test (10%) sets, as well as the combined files. Each set contains the buggy and the corresponding fixed code, with line-by-line alignment (i-th line in buggy corresponds to the i-th line in fixed). The original source code of the methods is also available. The mapping file allows to map the abstract code with the original source code.

• Small Methods (50 tokens)

  • Dataset Split & Combined

  • Methods' Source Code

  • Mapping

• Medium Methods (50-100 tokens)

  • Dataset Split & Combined

  • Methods' Source Code

  • Mapping

Download data

Predictions of the models for small and medium methods. For each model, there is a folder for each beam size containing files of the predictions performed on the test set of the corresponding dataset (above):

• Predictions - Small Methods (50 tokens)

• Predictions - Medium Methods (50-100 tokens)

Each beam size folder contains the following files:

• prediction.beam.mul.txt : the k predictions performed by the model separated by the <SEP> token. The i-th prediction refer to the i-th buggy code in the test set;

• prediction.beam.vis.txt : same predictions but displayed one by line;

• perfect.beam.mul.txt : the perfect fixes generated by the model. The file shows first the buggy code, then the fixed code, then the predicted code (equal to fixed);

• pred.operations.txt : AST fine grained operations emulated by the model when performing the perfect fixes.