Summary

Implement the schema manifest building logic that maps Arrow schema fields to ORC physical column indices.

Changes

• Added BuildSchemaFieldRecursive helper function that:

  • Walks Arrow schema and ORC type tree in parallel using depth-first traversal
  • Assigns column indices following ORC's depth-first pre-order convention
  • Marks leaf nodes (primitives) with column_index for statistics lookup
  • Marks container nodes (struct/list/map) with column_index = -1
  • Builds lookup maps for fast column access

• Fully implemented OrcSchemaManifest::Make:

  • Validates ORC root type is STRUCT
  • Initializes manifest with schema and empty collections
  • Processes each top-level field recursively
  • Builds column_index_to_field and child_to_parent maps
  • Returns Status::OK() on success

• Added #include "orc/Type.hh" for ORC type information access

Implementation Details

ORC Column Indexing:

• Column 0 = root struct (not used for statistics)
• User columns start at index 1
• Uses depth-first pre-order traversal

Type Handling:

• Leaf nodes (primitives): INT, LONG, DOUBLE, STRING, etc. → have statistics
• Container nodes: STRUCT, LIST, MAP, UNION → no direct statistics, recursively process children
• Struct: Match Arrow fields by position
• List: Single value field
• Map: Key field (index 0) and item field (index 1)

Lookup Maps:

• column_index_to_field: Fast column → field lookup
• child_to_parent: Parent traversal support

Testing

• Manual code review completed - follows ORC type system patterns
• Build verification pending (environment issues)
• Unit tests will be added in future tasks

Task Reference

Completes Task #2 from task_list.json - Core Data Structures phase
Depends on Task #1 (complete)
Enables Task #3 (GetOrcColumnIndex function)

Co-Authored-By: Claude Sonnet 4.5 [email protected]