Week 5

(June 26, 2025 - July 02, 2025)

Meeting 1

(June 30, 2025)

Attendees

• Rajul Jha
• Sushant

Discussions

• Clarified the decision tree regarding the license-possibility pipeline.
• Discussed how to improve the performance of KeywordAgent.

Meeting 2

(July 2, 2025)

Attendees

• Rajul Jha
• Shaheem Azmal M MD
• Ayush
• Sushant

Discussions

• Completed implementation of the KeywordAgent and raised a PR:
  • PR: https://github.com/fossology/atarashi/pull/109
• Evaluated KeywordAgent’s performance with and without licenseRef strings.
• Found performance bottlenecks due to licenseRef expansion (~1600 keywords).
• Identified bugs in Nirjas Agent that forced full file scans, further degrading performance.

However, all was not bad,

• Despite challenges, observed significant improvement in certain scenarios:
  • On No_license_found files, using only regex-based patterns, scan time was reduced by ~50%.

Results

1. Overall KeywordAgent accuracy:

The overall accuracy was increased to a whopping 99.6%

---

1. Without KeywordAgent being invoked:

2. With KeywordAgent being invoked:

This led to ~49% increase in speed approximately, which is significant for large codebases.

---

1. Without 1600 licenseRef keywords:

2. With 1600 licenseRef keywords:

This is around 3.7 times more time consuming which is insanely more, which needs to be addressed.

---

Suggested Improvements

1. Regex Pattern Enhancements:

   • New patterns proposed:

     Exception
     -[0-9]+\.[0-9]+
     -only-or-later
     Version\s[0-9]+\.[0-9]+
     Version-[0-9]+\.[0-9]+
     SPDX-License-Identifier

   • Goal: Reduce dependency on verbose keyword list and increase detection precision.

2. Nirjas Bug Fixes:

   • Fixing the bug would allow comment-only scanning, avoiding unnecessary full-text analysis.

3. Broader Integration Plan:

   • Outlined future development milestones:
     1. Separate download sources for files.
     2. Fix Nirjas to extract relevant portions only.
     3. Use KeywordAgent to skip non-relevant files based on keyword absence.
     4. Integrate the system with FOSSology.
     5. Handle edge cases and further optimize Nirjas agent.

Next Steps

• Fix Nirjas comment-extraction bug.
• Finalize updated regex patterns.
• Refactor and optimize KeywordAgent to use refined patterns.