Dependency Update Metrics

A Python tool for analyzing time-to-update (TTU) and time-to-remediate (TTR) metrics for package dependencies across different ecosystems (npm, PyPI, Cargo/crates.io).

Features

• Multi-ecosystem support: Analyze packages from npm, PyPI, and Cargo/crates.io
• Time-to-Update (TTU): Measure how long dependencies stay behind the latest version
• Time-to-Remediate (TTR): Measure how long known vulnerabilities remain unpatched
• Flexible weighting: Support for linear, exponential, inverse, or no weighting
• OSV integration: Uses the Open Source Vulnerabilities database for security analysis
• Export capabilities: Export detailed worksheets and OSV data

Installation

From PyPI

pip install dependency-update-metrics

From source

git clone https://github.com/imranur-rahman/dependency-update-metrics.git
cd dependency-update-metrics
pip install -e .

Usage

As a command-line tool

# Basic usage
dependency-metrics --ecosystem npm --package express

# With date range
dependency-metrics --ecosystem npm --package express \
  --start-date 2020-01-01 \
  --end-date 2023-12-31

# With exponential weighting
dependency-metrics --ecosystem npm --package express \
  --weighting-type exponential \
  --half-life 180

# Build OSV database first (one-time operation)
dependency-metrics --ecosystem npm --package express --build-osv

# Export detailed worksheets
dependency-metrics --ecosystem npm --package express --get-worksheets

# Get OSV vulnerability data
dependency-metrics --ecosystem npm --package express --get-osv

# Test a PyPI package
dependency-metrics --ecosystem pypi --package requests --start-date 2020-01-01 --end-date 2021-01-01 --get-worksheets

# Test a Cargo/crates.io package
dependency-metrics --ecosystem cargo --package serde --start-date 2020-01-01 --end-date 2024-01-01

# Bulk CSV input (parallel by default)
dependency-metrics --input-csv ./input.csv --workers 8 --output-dir ./output

# Resume an interrupted bulk run
dependency-metrics --input-csv ./input.csv --output-dir ./output --resume

# Per-release analysis: MTTU/MTTR at every release of the parent package
dependency-metrics --ecosystem pypi --package requests \
  --start-date 2020-01-01 --end-date 2021-01-01 \
  --per-release --get-worksheets

# Per-release in bulk CSV mode
dependency-metrics --input-csv ./input.csv --per-release --workers 4

Input CSV columns: ecosystem, package_name, optional start_date, optional end_date. Accepted ecosystem values are npm, pypi, cargo, and crates.io; crates.io is normalized to cargo in outputs. If end_date is omitted, the CLI --end-date value is used; when --end-date is also omitted, it defaults to today. Extra columns are allowed; duplicates are removed by ecosystem, package_name, end_date. Bulk mode computes metrics per package using the latest dependency set.

As a Python module

from datetime import datetime
from pathlib import Path
from dependency_metrics.analyzer import DependencyAnalyzer

# Create analyzer
analyzer = DependencyAnalyzer(
    ecosystem="npm",
    package="express",
    start_date=datetime(2020, 1, 1),
    end_date=datetime(2023, 12, 31),
    weighting_type="exponential",
    half_life=180,
    output_dir=Path("./output")
)

# Run analysis
results = analyzer.analyze()

print(f"Average TTU: {results['ttu']:.2f} days")
print(f"Average TTR: {results['ttr']:.2f} days")

Command-line Arguments

• --ecosystem: Ecosystem to analyze (npm, pypi, cargo, or crates.io) [Required unless --input-csv]
• --package: Package name to analyze [Required unless --input-csv]
• --input-csv: CSV file with columns ecosystem, package_name, optional start_date, and optional end_date. If end_date is omitted, --end-date is used.
• --start-date: Start date for analysis (YYYY-MM-DD) [Default: 1900-01-01]
• --end-date: End date for analysis (YYYY-MM-DD) [Default: today]
• --weighting-type: Weighting method (linear, exponential, inverse, disable) [Default: disable]
• --half-life: Half-life in days (required for exponential weighting)
• --build-osv: Build or rebuild the OSV vulnerability database. It downloads the npm, PyPI, and crates.io OSV zips (about 260 MB in total). Without --package, it only rebuilds the database.
• --osv-index-date: Print the build date of the OSV database as JSON (null when there is no database), then exit
• --cache-stats: Show the rows and the size of each namespace in the resolver cache (<output-dir>/cache/cache.db), then exit
• --cache-prune-gb N: Remove the least recently used cache rows until the cache holds less than N GB of data, then exit. The prune does not run VACUUM, so the file does not shrink; SQLite reuses the freed pages
• --cache-clear: Remove all cache rows, then exit
• --cache-namespace X: With --cache-clear, remove only the rows of namespace X
• --get-osv: Export OSV data for the package's dependencies
• --get-worksheets: Export detailed analysis worksheets to Excel
• --output-dir: Output directory for results [Default: ./output]
• --workers: Number of parallel workers for bulk CSV mode [Default: min(8, CPU count)]
• --package-timeout: Per-package analysis timeout in seconds (bulk CSV mode only). Packages that exceed this time receive an error row and are skipped [Default: disabled]
• --resume: Resume bulk CSV runs by skipping completed rows and retrying errors
• --per-release: Compute MTTU/MTTR at every release of the parent package within the window (one output row per release)
• --json: Print the result as one JSON object on stdout, in the result shape of the Host API (needs --package; see USAGE.md)

Weighting Methods

Disable (default)

No weighting applied. All time periods are weighted equally.

Linear

Weight decreases linearly with age:

weight = 1 - (age / max_age)

Exponential

Weight decreases exponentially based on half-life:

weight = exp(-λ * age)
where λ = ln(2) / half_life

Inverse

Weight is inversely proportional to age:

weight = 1 / (1 + age)

Output

The tool generates several outputs in the specified output directory:

1. JSON results file: Contains TTU, TTR, and metadata
2. OSV data (with --get-osv): CSV file with vulnerability information
3. Excel worksheets (with --get-worksheets): Detailed analysis for each dependency
4. Bulk CSV results (with --input-csv): Summary CSV, plus an optional dependency details CSV with --write-dependency-details
5. Per-release results (with --per-release): One row per package release with package_version, package_release_date, window_start, window_end, mttu, mttr

Example JSON output

{
  "package": "express",
  "ecosystem": "npm",
  "version": "4.18.2",
  "start_date": "2020-01-01T00:00:00",
  "end_date": "2023-12-31T00:00:00",
  "weighting_type": "exponential",
  "half_life": 180,
  "ttu": 45.32,
  "ttr": 12.15,
  "num_dependencies": 30
}

How It Works

1. Package Analysis: Fetches package metadata from the registry
2. Dependency Extraction: Extracts dependencies from the package version closest to the end date
3. Timeline Construction: Creates a timeline of version releases for each dependency
4. Version Resolution: Resolves dependency versions at each time interval via ecosystem resolvers
5. Remediation Check: Uses the OSV service to determine vulnerability remediation
6. Metric Calculation:
   • TTU: Measures time when dependency is not at the highest available version
   • TTR: Measures time when dependency has known vulnerabilities
7. Weighting: Applies optional time-based weighting to prioritize recent periods
8. Aggregation: Averages metrics across all dependencies

Architecture Notes

• Ecosystem-specific resolution is encapsulated in dependency_metrics/resolvers.py.
• OSV remediation checks are handled by dependency_metrics/osv_service.py.
• Reporting/export helpers live in dependency_metrics/reporting.py.
• Bulk CSV mode groups rows by (ecosystem, package_name) and processes each group sequentially while running groups in parallel to maximize cache reuse.
• For bulk runs, a single package analysis computes metrics across multiple end dates using the latest dependency set.
• --per-release mode uses analyze_at_release_points() which fetches per-version dependency sets and builds a shared dep-cache covering the full window; it slices that cache for each release sub-window [start_date, release_date].
• Cache invalidation: ResolverCache (dependency_metrics/resolvers.py) persists all HTTP responses to a SQLite file at <output-dir>/cache/cache.db, keyed only by (namespace, key) — there is no TTL. This is fine for per-version data (a specific version's dependencies/metadata never changes), but package-level version-list lookups (DepsDevClient.get_package()'s depsdev_package namespace, and each resolver's fetch_package_metadata()/_get_npm_time_data() under the metadata/npm_time namespaces) cache the entire, mutable version list. Those lookups compare the requested end_date against the newest publish date already in the cached listing and transparently re-fetch whenever the cache can't yet rule out a newer release — so re-running against the same --output-dir won't silently miss versions published since the last run. The tradeoff: analyzing "as of today" re-fetches each package's version list on every run (since "today" is always newer than the last known release), which is the correct behavior — only truly historical, fixed end_date values benefit from the cache staying warm across runs.

Requirements

• Python 3.10+
• git (for auto-installing the patched pip on first PyPI analysis)
• npm CLI (for npm ecosystem analysis) sudo apt install nodejs npm
• Internet connection (for fetching package metadata and OSV data)

Note: PyPI analysis uses a patched pip fork that adds date-bounded version resolution. The tool clones it automatically to ~/.dependency_metrics/vendor/pip on first use — no manual setup required.

Development

Setup development environment

git clone https://github.com/imranur-rahman/dependency-update-metrics.git
cd dependency-update-metrics
pip install -e ".[dev]"

Run tests

pytest

Code formatting

black dependency_metrics/

Publishing to PyPI

Prepare for publishing

# Install build tools
pip install build twine

# Build distribution
python -m build

# Check distribution
twine check dist/*

Upload to PyPI

# Upload to Test PyPI first
twine upload --repository testpypi dist/*

# If all looks good, upload to PyPI
twine upload dist/*

License

MIT License - see LICENSE file for details

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Citation

If you use this tool in your research, please cite:

@software{dependency_metrics,
  author = {Rahman, Imranur},
  title = {Dependency Update Metrics},
  year = {2025},
  url = {https://github.com/imranur-rahman/dependency-update-metrics}
}

Acknowledgments

• Uses data from Open Source Vulnerabilities (OSV)
• Package metadata from npm Registry and PyPI