ReCAP

Implementation of ReCAP ("Efficient Path Query Processing in Relational Database Systems"): a compiler that takes a path query -- a label regex plus a selective aggregate over edge/vertex properties -- and a property graph, and produces optimized recursive SQL that early-filters non-viable paths during exploration instead of generating every path first and filtering afterward. Runs on DuckDB.

Quick start

cd compiler
pip install -e '.[dev]'
python3 -m pytest tests/ -v          # 162 tests, all on the bundled sample dataset
python3 demo_pipeline.py             # see it run end to end, no browser needed

For the browser workbench instead of the script:

cd compiler
pip install -e '.[ui]'
streamlit run webapp/app.py

See compiler/README.md for what each of these actually does, and compiler/CHECKLIST.md for build status (one row per pipeline stage) and implementation notes for anyone extending this.

Where things are

• compiler/ -- the compiler itself, fully self-contained (bundles its own small sample dataset in compiler/sample_data/, no external data needed). Start here.
• new_compiler_requirements/compiler_reqs.md -- the functional requirements spec the compiler is built against.
• experiments/alternative_explorations/ -- standalone experiments exploring compatibility with segmented/non-forward path evaluation strategies; design in navigation_style_experiment.md, runnable implementation in navigation_experiment/ (see its own README).
• experiments/ -- the full real-data evaluation campaign (multiple datasets/engines, up to ~100M edges). Has its own per-experiment READMEs; out of scope for this top-level README since it depends on large external datasets not bundled here.