New SemanticFieldInferenceFieldsIT test

[kkharbas]

1. New SemanticFieldInferenceFieldsIT test for inference field inclusion//exclusion with semantic field.
2. Extracts common logic from SemanticTextnferenceFieldsIT into abstract base class AbstractInferenceFieldsIT

Partly addresses #146592

[elasticsearchmachine]

Pinging @elastic/es-search-relevance (Team:Search Relevance)

[Mikep86]

I don't think this is how we should generalize this test case. It adds a lot of duplication that is confusing and hard to maintain. Instead, I think we can run cases for both field types through excludeInferenceFieldsFromSourceTestCase. We just need to parameterize that across a few dimensions:

• The inference endpoint(s) created
• The field(s) created (1:1 with the inference endpoints)
• The field values used during indexing
  • We can use SemanticTextFieldTests.randomSemanticInput to generate a random mix of values appropriate for the field type
• The field(s) queried and asserted on

Most of the logic can live in AbstractInferenceFieldsIT, and the concrete implementations can be thin layers that define the abstract functions that implement these parameters.

[kkharbas]

I don't think this is how we should generalize this test case. It adds a lot of duplication that is confusing and hard to maintain. Instead, I think we can run cases for both field types through excludeInferenceFieldsFromSourceTestCase. We just need to parameterize that across a few dimensions:

• The inference endpoint(s) created

• The field(s) created (1:1 with the inference endpoints)

• The field values used during indexing

  • We can use SemanticTextFieldTests.randomSemanticInput to generate a random mix of values appropriate for the field type

• The field(s) queried and asserted on

Most of the logic can live in AbstractInferenceFieldsIT, and the concrete implementations can be thin layers that define the abstract functions that implement these parameters.

Updated. Its in a better shape now.

[Mikep86]

This isn't quite what I meant. We can generalize the logic further and provide more test coverage with less code duplication. I took a shot at it myself: #155004