fix: update Java Comparator to read from test_results table

[mashraf-222]

Summary

• Fixed schema mismatch by updating Comparator to read from test_results table
• Aligns with cross-language schema consistency requirement
• Enables Java behavior verification to work end-to-end

Problem

The Java Comparator was reading from an invocations table with schema:

• call_id, method_id, args_json, result_json, error_json

But Java instrumentation writes to a test_results table with schema:

• test_module_path, test_class_name, test_function_name, function_getting_tested, loop_index, iteration_id, runtime, return_value, verification_type

This prevented behavior verification from working because the Comparator couldn't find any instrumented test data.

Solution

Updated Comparator.java to read from the test_results table that instrumentation actually writes to. This aligns with the requirement for consistent schema across all languages.

Changes

codeflash-java-runtime/src/main/java/com/codeflash/Comparator.java

getInvocations() method:

• Changed SQL query from SELECT ... FROM invocations to SELECT ... FROM test_results
• Updated column mapping:
  • loop_index + iteration_id → call_id (unique identifier)
  • return_value → resultJson (JSON result to compare)
  • test_class_name + "." + function_getting_tested → methodId
  • args_json → null (not captured in test_results)
  • error_json → null (not captured in test_results)

New helper method:

• parseIterationId(): Extracts numeric iteration from string format "iter_testIteration" (e.g., "1_0" → 1)
• Generates unique call_id as (loop_index * 10000) + iteration_number

Testing

✅ All 336 Java tests pass (7 skipped)
✅ All 18 comparator tests pass
✅ Integration tests verify correct schema mapping

Impact

This fix is critical for Java optimization:

• ✅ Behavior verification can now read instrumented test data
• ✅ Comparator can detect correctness differences between original and optimized code
• ✅ Aligns with cross-language schema consistency
• ✅ Required for Java optimization workflow to function end-to-end

Notes

The test_results schema doesn't capture args_json or error_json, so those fields are set to null in the Invocation objects. This is acceptable because:

• Return value comparison is the primary correctness check
• Test pass/fail status provides error detection
• Future enhancement could add these fields to the schema if needed

🤖 Generated with Claude Code