Program translation is a growing demand in software engineering. Manual program translation requires programming expertise in source and target language. One way to automate this process is to make use of the big data of programs. However, existing code retrieval techniques lack the design to cover cross-language code retrieval. Other data-driven approaches require human efforts in constructing cross-language parallel datasets to train translation models. We built a code translation retrieval system. We use a lightweight but informative program representation, which can be generalized to all imperative PLs. Furthermore, we implement our customized index structure and hierarchical filtering mechanism for efficient code retrieval from a big data.
- Python3
- MongoDB
- pymongo
- ANTLR (Java target)
In the offline phase, system constructs feature reapresentation for each program in the database and save it in a feature database for use.
sh create_feature_db.sh path_to_your_database
System constructs path-type-bucket-index for the featrue representations.
python3 create_index.py
User can specify maximum bucket size in bucket_size.json.
sh translate.sh path_to_input_code target_language
In this repository, we add 4 programming languages: Python, Java, C++, JavaScript.
If you want more languages support, just simply run ANTLR parser in your desired language and enrich the lang_collection list in each file.
We compare the results of effectiveness and efficiency of our system with the following state-of-the-art baselines:
- 1pSMT: phrase-based SMT on sequential programs.
- mppSMT: multi-phase phrase-based SMT.
- Tree2tree: tree-to-tree neural networks.
- TransCoder: weakly-supervise translation model.
