Hi @james-willis ! I addressed most of your comments. For some I left my answers. It looks like we are close to a final version. Could be nice if you can take another look, so I can work on bindings and docs.

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 70.30303% with 49 lines in your changes missing coverage. Please review.
✅ Project coverage is 80.80%. Comparing base (f1db6f4) to head (88001b9).
⚠️ Report is 7 commits behind head on main.

@@            Coverage Diff             @@
##             main     #792      +/-   ##
==========================================
- Coverage   84.94%   80.80%   -4.15%     
==========================================
  Files          68       78      +10     
  Lines        3507     4428     +921     
  Branches      453      527      +74     
==========================================
+ Hits         2979     3578     +599     
- Misses        528      850     +322     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

I will work on tests, bindings and docs.

@james-willis this one is ready for review

Looking at this PR, here are my recommendations:

Code Quality & Performance

1. Memory Management Concerns (AggregateNeighbors.scala:267-268):
   The hardcoded warning for maxHops > 10 should be configurable or include memory estimation logic rather than an arbitrary threshold.

2. Potential Optimization (AggregateNeighbors.scala:303):
   The repartition by SRC could be expensive for smaller datasets. Consider making this optional or using coalesce when appropriate.

API Design

3. Required Parameter Validation:
   The API requires either stoppingCondition or targetCondition but this isn't enforced until run(). Consider validation during builder setup for better UX.

4. Default Accumulator:
   Most use cases will want basic path tracking. Consider adding a convenience method for common path accumulation patterns.

Testing & Documentation

5. Edge Case Testing: Add tests for:

   • Very large graphs (performance characteristics)
   • Disconnected components
   • Graphs with cycles under different stopping conditions

6. Python API Consistency:
   Ensure Python parameter names follow conventions consistently.

Implementation Robustness

7. Convergence Logic (AggregateNeighbors.scala:400):
   Currently only checks if frontier is empty, but doesn't handle cases where states exist but no new edges can be traversed.

Overall, this is a solid implementation that follows GraphFrames patterns well. The main areas for improvement are around performance optimization configuration and edge case handling. The comprehensive test suite and documentation are excellent.

Comment by Claude (AI Assistant)

@james-willis

1. It is a warning, I don't understand how the warning can be configurable.
2. That is false. The join mean repartition by src anyway, so even on a single step it is better to keep the repartition
3. It is impossible to achieve, because we do not know the order of setter-calls
4. I do not understand, what does it mean? If it is about adding implementations, I'm going to do it in follow-up PRs. I'm going to add at least allPaths to find all paths between two nodes
5. I'm not going to add huge graphs (performance testing) to unit tests. All tests are 10+ minutes already, do we want to have 30 minutes tests?
6. All the Python API follows the snake_case way
7. In this case the DataFrame will be empty, so I don't fully understand the comment...

If it is about adding implementations, I'm going to do it in follow-up PRs. I'm going to add at least allPaths to find all paths between two nodes

yeah that was it. sgtm.

All other answer sgtm on that Claude review.