S3M

This repository contains the code of our MSR 2021 paper "S3M: Siamese Stack (Trace) Similarity Measure".

Data

NetBeans

You can find NetBeans dataset from paper to reproduce experiments on Figshare.

Format

You can use other datasets that match the following format. The data presented as a list of report in the following JSON:

[
    {
        "bug_id": 1234,
        "dup_id": 12,
        "creation_ts": 1234567891234,
        "stacktrace": {
            "exception": ["java.lang.Exception"], 
            "frames": [
                {"function": "java.util.ArrayList.get"}, 
                {"function": "com.company.Class1.method1"}, 
                {"function": "com.company.Class2.method2"}, 
                {"function": "com.company.Class1.method2"}
            ]
        }
    },
    {
        "bug_id": 1235,
        "dup_id": null,
        "creation_ts": 1234567898765,
        "stacktrace": {
            "exception": ["com.company.MyException"], 
            "frames": [  
                {"function": "com.company.Class1.method2"}, 
                {"function": "com.company.Class1.main"}
            ]
        }
    }
]

where

• bug_id – identifier of report
• dup_id – identifier of first report in group. If this report is the first in this group, this field should be null
• creation_ts – timestamp of report creation

stacktrace field contains information about report content such as exception classes and stack frames.

Usage

Install

git clone [email protected]:akhvorov/S3M.git
cd S3M
pip install -r requirements.txt

Run

cd src
python main.py --data_path path_to_reports.json

The script will produce some state files in event_state directory in one level with src. This precomputed states will speed up data reading in the next runs.