PREM: Program Repair using Error Messages

This repo contains the source code and evaluation artifact of the paper Compilation Error Repairing by Examples.

System Requirements

Before building PREM, the following platforms/tools are required as external dependencies. All libraries that PREM depends on are specified in the project files and will be proceed automatically by dotnet.

.NET

.NET is a cross-platform development environment. PREM is built as a netcoreapp2.0. To build and run PREM, you need to install a .NET Core SDK with version 2.x. The recommended version is v2.1.26. After installation, use dotnet --version to check the version number.

UTrans Parser

PREM relies on UTrans, an external parser (for C# and Java) developed with ANTLR in Java. The standalone jar file is provided in the lib/ folder. Please make sure that you put this jar somewhere on your file system, and let the environment variable UTRANS_JAR point to the full path of this jar, e.g. by adding export UTRANS_JAR=<full path to UTrans-1.0.jar> in your shell profile: bashrc, zshrc, or whatever.

C# Compiler

To allow PREM invoke C# compiler from command line to check if a repaired program compiles, you should have csc on your system PATH. You may use the one supported by .NET (if any) or Mono instead.

Building & Running

In the project root directory:

cd src/Prem/
dotnet build

Type dotnet run to see command line option helps.

Please remember that every dotnet command must be executed under the folder src/Prem/, otherwise it cannot resolve the correct project file.

Dataset Folder Structure

Datasets passed to options --learn and --bench (see data/CSharp/ for an example) must have the following folder structure:

<root>/
     example_group_1/
         1/
             [E]* erroneous version
             [C]* corrected version
             [P]* error info
         2/
             ...
     example_group_2/
         ...
     ...

PREM recognizes a file with prefix [E] as the erroneous version, [C] as the corrected version, and [P] as the error info (first line error position, second line error message).

To make filters --learn-with and --bench-with work normally, you must have your example folder (itself is a subfolder of some example group folder) named as a digit, e.g. 2 rather than example-2.

Datasets passed to option --predict (see data/Mutation/ for an example) must have the following folder structure:

<root>/
    case_1/
        [E]* testcase
        [P]* error info
    case_2/
    ...

Reproducing

See Reproducing Guide.