[Bhavy-Ranka]

Motivation

IndexSchema.to_dict() serializes enum values to lowercase via model_to_dict. Because CompressionType has mixed-case values (LVQ4, LVQ4x4, LVQ4x8, LVQ8, LeanVec4x8, LeanVec8x8) and lacked case-insensitive validation, schemas exported via to_dict() or to_yaml() failed with pydantic.ValidationError when reloaded via from_dict() or from_yaml(). This also caused MigrationPlanner.merge_patch() and AsyncMigrationPlanner.create_plan_from_patch() to fail for any index containing a compressed SVS-VAMANA vector field (#808).

Changes

Case-insensitive CompressionType enum lookup

Added a _missing_ classmethod to CompressionType in redisvl/schema/fields.py. It matches string inputs case-insensitively against member values, allowing lowercase values like "lvq8" or "leanvec4x8" emitted by to_dict() to resolve to their canonical members (e.g. CompressionType.LVQ8) while falling back to None for invalid values so ValueError / ValidationError is raised as expected.

Migration wizard cleanup

Replaced the manual uppercase normalization mapping dictionary in redisvl/migration/wizard.py with direct CompressionType(compression_input).value lookup and imported CompressionType.

Regression and unit tests

• tests/unit/test_schema.py: Added test_svs_vamana_compression_yaml_round_trip parametrized across all members of CompressionType to ensure clean YAML round-tripping.
• tests/unit/test_migration_planner.py: Added test_merge_patch_preserves_svs_vamana_compression verifying merge_patch() does not fail and preserves compression attributes.
• tests/unit/test_async_migration_planner.py: Added test_async_planner_merge_patch_preserves_svs_vamana_compression verifying async schema patch planning preserves compression.
• tests/unit/test_fields.py: Added tests covering exact-case input, case-insensitive lowercase input, validation errors on unknown values, and canonical uppercase argument generation in as_redis_field() / redis_args().

Release Notes

Allow case-insensitive vector compression type values in SVS-VAMANA index schemas, ensuring schemas exported to dictionaries or YAML round-trip cleanly without validation errors.

Reviewer Brief

Start here: redisvl/schema/fields.py, CompressionType._missing_. It resolves string inputs case-insensitively, enabling lower-cased enum strings emitted by model_to_dict() to map back to canonical enum members.

Where it could fail

• redisvl/schema/fields.py:105: Ensure non-matching strings fall through to return None so standard Enum ValueError (and downstream Pydantic ValidationError) is raised for invalid options.
• redisvl/migration/wizard.py:832: Ensure empty inputs bypass CompressionType parsing and leave compression unconfigured.

Not verified: Live Redis instances or Docker testcontainers (tests/unit run with --noconftest).

Beyond CI: Confirmed regression tests 1, 2, 3, 5, and 7 fail on main before CompressionType._missing_ is added, and pass with this change.

_{Reviewed 0302739 · Antigravity (Gemini 3.8 Flash) · lenses: correctness, tests, design}

(AI assistance has been taken in solving this)

---

Note

Low Risk
Narrow enum parsing and migration-wizard input change; invalid compression still fails validation, and Redis index behavior is unchanged for already-valid schemas.

Overview
Fixes schema reload and migration failures when SVS-VAMANA compression values are not exact-case (e.g. after to_dict() / YAML export).

CompressionType now resolves string inputs case-insensitively via _missing_, mapping values like lvq8 to canonical members while still rejecting unknown strings with validation errors. Migration wizard compression prompts use that enum instead of a hand-maintained uppercase map, with a user-facing message on invalid input.

Regression tests cover YAML round-trip for every compression member, canonical FT.CREATE args from lowercase input, and sync/async merge patch flows that keep compression when adding unrelated fields.

^{Reviewed by Cursor Bugbot for commit 0302739. Bugbot is set up for automated code reviews on this repo. Configure here.}

[Bhavy-Ranka]

@claude @joshrotenberg @bsbodden @abrookins