Diffinite — Source Code Comparison

Forensic source-code comparison tool for IP litigation and code audit, now available as a VSCode extension.

Compare two directories of source code with Winnowing fingerprints (Schleimer et al., 2003 — the algorithm behind Stanford MOSS) and generate professional PDF/HTML/Markdown reports — all from within VSCode.

Design Principle: Diffinite reports how similar and where similar. It does not classify the type of copying — that is the expert witness's job.

Features

• 1:1 File Matching — Pairs files across two directories using fuzzy name matching, then computes line-by-line or word-by-word diffs with syntax highlighting.
• N:M Cross-Matching (Deep Mode) — Winnowing fingerprint-based Jaccard similarity across all file pairs. Detects code reuse even across renamed, split, or merged files.
• Comment Stripping — 5-state FSM parser supporting 30+ file extensions (.py, .js, .ts, .java, .c, .cpp, .go, .rs, .rb, .sql, .html, .css, and more).
• Multiple Report Formats — Export to PDF, HTML, or Markdown.
• Forensic Annotations — Page numbers, file numbers, Bates stamps, filenames on every page.
• GUI Options Panel — Configure all analysis parameters visually without touching the CLI.
• Bundled Binary Support — Ships with standalone binaries when available; falls back to Python if needed.

Usage

1. Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P)
2. Run "Diffinite: Compare Directories"
3. Select the original directory (A) and comparison directory (B)
4. Configure options in the GUI panel (mode, thresholds, comment stripping, etc.)
5. View results in the built-in diff viewer or export a report

How It Works

Diffinite runs a two-stage pipeline:

Stage 1: 1:1 File Matching (simple mode)

1. Fuzzy name matching — Pairs files across directories using string similarity (configurable threshold).
2. Comment stripping — Optionally removes comments using a 5-state finite state machine parser.
3. Side-by-side diff — Computes line-by-line (or word-by-word) diffs using difflib.SequenceMatcher.
4. Report generation — Renders syntax-highlighted HTML diffs via Pygments, converts to PDF with xhtml2pdf.

Stage 2: N:M Cross-Matching (deep mode, default)

5. Winnowing fingerprint extraction — Extracts position-independent code fingerprints (K-gram → rolling hash → window selection).
6. Inverted index construction — Builds a hash-to-file mapping for all B-directory fingerprints.
7. Jaccard similarity computation — For each A-file, computes |A∩B| / |A∪B| against all B-files sharing fingerprints.
8. Cross-match reporting — Appends an N:M similarity matrix showing which files from A are similar to which files in B.

Output Report

Cover Page

Summary table for each matched file pair:

Diff Pages

Side-by-side diff for each matched pair:

• 🟢 Green — Lines present only in File B (additions)
• 🔴 Red — Lines present only in File A (deletions)
• No highlight — Identical lines (with configurable context folding)

Deep Compare Section

N:M cross-matching table (deep mode):

Extension Settings

GUI Options Panel

All options are configurable through the built-in GUI panel:

Requirements

This extension requires either:

• Bundled binary (included for Windows / Linux / macOS when available), or
• Python ≥ 3.10 with diffinite installed:

  pip install diffinite
  

Comment Stripping Support

The Strip Comments option removes comments using a 5-state FSM parser:

📊 Benchmark results and dataset rationale → see GitHub README

Limitations

• General-purpose tokenizer — Uses a single regex tokenizer, not language-specific parsers.
• Position-independent — Reordered functions may produce higher similarity than expected.
• No corpus-wide weighting — Pairwise comparison only; no TF-IDF to down-weight common idioms.
• Not a legal opinion — Similarity scores are mathematical measurements, not legal conclusions.

License

Apache License 2.0

See NOTICE for attribution.