Skip to content

Latest commit

 

History

History
80 lines (62 loc) · 3.05 KB

README.md

File metadata and controls

80 lines (62 loc) · 3.05 KB
Raw

Reproducing the Experiments and Analysis of the Paper

Before running the analysis notebooks:

  1. unzip the three zip files within this folder (callgraph_seq.zip, pycg_output.zip, DynaPyt_callgraphs.zip).
  2. create a Python virtual environement (Use Python 3.10 or above) 
    python3.10 -m venv .venv
source .venv/bin/activate
  3. Install requirments 
    pip install -r requirements.txt

1. DyPyBench Overview (Sec3)

You can reproduce the plots of the overview of the benchmark by running the notebook: dypybench_overview_plots.ipynb

2. Call Graphs (Sec4.1)

In this part we generate Dynamic Call Graphs using DynaPyt and we also generate Static Call Graphs using PyCG, then we compare them.

Step 1. Generating Dynamic Call Graphs

To generate a dynamic call graph for a project (e.g, project 1), the following commands should be executed (note that it takes on average around 4 Hours to complete the generation per project, so run carefully)

# Run test (necessary to setup the project)
python3 dypybench.py --test 1

# Run instrumentation
python3 dypybench.py --dynapyt_instrument 1 --dynapyt_file ./text/includes.txt --dynapyt_analysis CallGraph

# Run the analysis
python3 dypybench.py --dynapyt_run 1 --dynapyt_analysis CallGraph

Once the the commands finished executed, a json file would appear under the project folder (~/temp/project1).

Step 2. Generating Static Call Graphs

To generate a static call graph for a given project, run the following commands (For documentation, check: https://github.com/vitsalis/PyCG):

# Run test (necessary to setup the project)
python3 dypybench.py --test 1

# call PyCG
pycg --package project1 $(find project1 -type f -name "*.py") -o project1.json

Step 3. Parsing the generated files and analysing the call graphs

The analysis of callgraphs was done in the notebook: callgraphs.ipynb

LExecutor (Sec4.2)

Step 1. Generate LExecutor traces

# Run LExecutor instrumentation
python3 dypybench.py --lex_instrument 1 --lex_file ./text/includes.txt

# Generate Traces
python3 dypybench.py --lex_test 1

A ".h5" file would be generate within the project folder.

Step 2. Retraining LExecutor

To retrain LExector, follow the instruction of the LExecutor repository.

The plot presented in the paper can reproduced by calling the Python script: generate_lexecutor_acc.py

Mining Specs (Sec4.3)

Step 1: Generate Sequences of API calls

Example on project 1.

# Run test (necessary to setup the project)
python3 dypybench.py --test 1

# Run instrumentation
python3 dypybench.py --dynapyt_instrument 1 --dynapyt_file ./text/includes.txt --dynapyt_analysis CallGraphSeq

# Run the analysis
python3 dypybench.py --dynapyt_run 1 --dynapyt_analysis CallGraphSeq

Step 2: Patterns Mining

The specification mining is done in the notebooks: spec_mine.ipynb and plots_sec4.3.ipynb