fix(gfql): handle schema-changing operations in chains - #762

Fixes #761

Summary

Fix "Column 'index' not found in edges" error when UMAP or hypergraph operations are mixed with other GFQL operations in chains.

Implemented recursive dispatch that automatically splits chains at schema-changer boundaries, executing them as: before → schema_changer → rest. This ensures schema-changing operations (which create entirely new graph structures) are properly isolated from regular chain machinery.

Changes

Added

• UMAP in chains: Can now use call('umap', {...}) mixed with filters and other operations
• Hypergraph in chains: Removed mixing restriction - now allows [n(...), call('hypergraph', {...})]

Implementation

• graphistry/compute/chain.py (lines 372-406): Recursive dispatch logic
  • Detects schema-changers (['umap', 'hypergraph']) in operation list
  • Handles singleton, multi-op, and nested cases
  • Falls through to existing chain logic when no schema-changers present
• graphistry/tests/compute/test_gfql_schema_changers.py: New comprehensive test suite (8 scenarios)
• graphistry/tests/compute/test_gfql_hypergraph.py: Updated to reflect new behavior

Test Coverage

Unit Tests

✅ Unit tests (compute): 396 passed, 49 skipped
✅ GFQL tests: 17/17 pass
✅ Chain tests: 25 passed, 3 skipped
✅ Lint (flake8): Pass
✅ Typecheck (mypy): Pass

Standalone Validation (10 scenarios in plans/fix-umap-chain-tracking/)

✅ Singleton operations: UMAP alone, hypergraph alone
✅ Multiple schema-changers: UMAP→UMAP, UMAP→hypergraph, n→umap→e→hypergraph→n
✅ Schema-changers at different positions: First, middle, last
✅ Mixed with filters: Filter→UMAP, UMAP→filter
✅ Edge cases: Empty before/after segments, consecutive schema-changers
✅ Let bindings: Schema-changers nested in let() constructs work correctly

All 10 standalone tests pass ✅

Example Usage

# Before: This would fail with "Column 'index' not found"
result = g.gfql([
    n({'type': 'person'}),
    call('umap', {'n_neighbors': 15}),
    e()
])

# Now: Works! Chain is split and executed as:
# g.chain([n(...)]) → g.chain([umap(...)]) → g.chain([e()])

Testing Notes

Schema-changers work correctly in:

1. Direct chains: [n(), umap(), e()]
2. Let bindings: let({'embedded': call('umap', ...)})
3. Multiple schema-changers: [umap(), hypergraph()]
4. Any position: Beginning, middle, or end of chain