How hieunguyen2711/parser-fuzzer works
Parser Fuzzer is a grammar-guided, LLM-assisted fuzzer for the mxml XML parser. Its principal workflow generates a Hypothesis strategy, validates it cheaply, runs a campaign through a sanitizer-enabled C harness, then summarizes results and refines the strategy. A separate grammar-check capability compares the standard and mxml-adjusted grammars with mxml’s actual acceptance. The graph preserves the README-documented grammar-check branch; unsampled implementation details are represented only where the README supports them.
Main entry points: loop.py Refinement loop · seed_run.py Seed run · triage.py Crash triage · compare_grammars.py Grammar comparison
Components 22
Strategy generation
- Seed run
fuzzer/seed_run.py - Refinement loop
fuzzer/loop.py - Prompt construction
fuzzer/prompts.py - Strategy loading
fuzzer/strategy_io.py - Hypothesis strategies
fuzzer/strategies
Fuzzing campaign
- Strategy validation
fuzzer/validate.py - Example generation
fuzzer/examples.py - Campaign execution
fuzzer/campaign.py - Harness runner
fuzzer/runner.py - Campaign scoring
fuzzer/score.py - Crash triage
fuzzer/triage.py - Run summary
fuzzer/summary.py - Crash findings
crashes
Parser harness
- C harness
harness/harness.c - mxml adapter
harness/fuzz_target.c - mxml parser
target
Grammar verification
- XML grammar
grammar/XMLParser.g4 - mxml grammar
grammar/XMLmxmlParser.g4 - Grammar comparison
grammar/compare_grammars.py - Acceptance probe
grammar/probe_mxml.c
Other components
- Engineer
- Vertex AI model
Connections 33
- Engineer → Seed run: starts seed
README.md - Engineer → Refinement loop: starts refinement
README.md - Seed run → Prompt construction: builds prompt
fuzzer/seed_run.py - Seed run → Vertex AI model: requests strategy
fuzzer/seed_run.py - Refinement loop → Prompt construction: builds prompts
fuzzer/loop.py - Refinement loop → Vertex AI model: requests revisions
fuzzer/loop.py - Vertex AI model → Seed run: returns strategy
README.md - Vertex AI model → Refinement loop: returns revisions
README.md - Seed run → Strategy loading: extracts and loads
fuzzer/seed_run.py - Refinement loop → Strategy loading: extracts and loads
fuzzer/loop.py - Strategy loading → Hypothesis strategies: loads module
fuzzer/strategy_io.py - Seed run → Strategy validation: checks strategy
fuzzer/seed_run.py - Refinement loop → Strategy validation: checks strategy
fuzzer/loop.py - Strategy validation → Example generation: draws samples
fuzzer/validate.py - Strategy validation → Harness runner: runs samples
fuzzer/validate.py - Refinement loop → Campaign execution: runs campaign
fuzzer/loop.py - Campaign execution → Harness runner: runs inputs
fuzzer/campaign.py - Campaign execution → Crash triage: signs crashes
fuzzer/campaign.py - Refinement loop → Campaign scoring: scores results
fuzzer/loop.py - Refinement loop → Run summary: summarizes results
fuzzer/loop.py - Refinement loop → Crash triage: saves findings
fuzzer/loop.py - Run summary → Campaign execution: reads statistics
fuzzer/summary.py - Run summary → Crash triage: reads findings
fuzzer/summary.py - Crash triage → Harness runner: uses run results
fuzzer/triage.py - Crash triage → Crash findings: writes reports
- Harness runner → C harness: executes process
README.md - C harness → mxml adapter: dispatches input
README.md - mxml adapter → mxml parser: parses document
harness/fuzz_target.c - Engineer → Grammar comparison: runs check
README.md - Grammar comparison → XML grammar: compares grammar
README.md - Grammar comparison → mxml grammar: compares grammar
README.md - Grammar comparison → Acceptance probe: compiles probe
README.md - Acceptance probe → mxml parser: tests acceptance
README.md