feat: route imdb_search advanced title searches through Kometa IMDb Service

[chazlarson]

What type of PR is this?

• Bug Fix (non-breaking change which fixes an issue)
• Feature/Tweak (non-breaking change which adds new functionality or enhances existing functionality)
• Breaking Change (fix or feature that would break any existing functionality for users)
• Documentation Update
• Chore (maintenance, dependency bumps, housekeeping - no functional change)
• Other

Description

Routes imdb_search builder advanced title searches through the Kometa IMDb Service (POST /search/advanced) instead of hitting IMDb GraphQL directly.

What changed:

• Added _service_post() - POST counterpart to _service_request, with the same error handling (exception wrapping, 404/400+ checks, JSON parse guard)
• Rewrote the search branch of _pagination(): calls _graphql_json to build constraint variables (preserving all existing translation tables and constraint logic), splits into constraints/sort/limit, and sends a single POST to /search/advanced
• The service handles cursor pagination and caching server-side (1-day TTL by constraint hash); Kometa receives the full ordered result list in one response, eliminating client-side cursor pagination overhead
• List and watchlist paths remain unchanged on direct GraphQL

Tested: 46 existing IMDb tests pass. The change is purely a routing swap - all constraint-building logic and translation tables are preserved as-is.

Have you updated the CHANGELOG.md?

• Yes
• No

[chazlarson]

Review

All CI checks pass and the PR is mergeable. Core change is clean. Notes below.

✅ Addressed

• Wasted search_hash fetch (fixed in 1e5b9da): the search path previously called _graphql_json, which triggered _json_operation/search_hash — a dead GitHub-raw HTTP GET on every imdb_search. Now split into _graphql_variables (builds constraints) and a thin _graphql_json wrapper; the search branch calls _graphql_variables directly and no longer fetches the hash. List/watchlist paths unchanged.

🟡 Still open: no test coverage for the new search path

_service_post() and the search-via-service branch of _pagination are untested. Worth a few small tests covering:

• the constraints/sort/limit split (excluding locale/first/sortBy/sortOrder),
• limit omitted when data["limit"] == 0,
• results parsing + the empty-results Failed path,
• _service_post error handling (≥400, invalid JSON) mirroring the existing _service_request tests.

🟡 PR scope

Branch is stacked on imdb-service-ratings, so the diff into nightly carries all prior service work (ratings/genres/episode-ratings/parental/charts/keywords, gitpython bump, spellcheck tooling). Only ~60 lines are the actual search change. Should land after — or together with — the ratings branch.

✅ Looks good

• _service_post() correctly mirrors _service_request() error handling (exception wrap, 404/not_found_ok, ≥400, JSON-parse guard).
• constraints/sort/limit split matches the service contract.
• Omitting limit when 0 correctly maps Kometa's "no limit" convention onto the service default.
• List/watchlist correctly left on direct GraphQL; item_count = 100 simplification is sound.
• No-fallback-for-search is reasonable (no TSV equivalent exists).

Minor nits

• Search block uses data.get("limit", 0) while the list path uses data["limit"] directly — validate_imdb always sets limit, so the defensive .get is inconsistent (harmless).
• self._service_post(...) or {} turns a None (404) into "No IMDb IDs Found" rather than a distinct not-found message — fine, but less precise than passing not_found_ok=True.

[chazlarson]

Review (updated)

Both follow-up items are now resolved. CI green, mergeable.

✅ Addressed

• Wasted search_hash fetch (1e5b9da): split _graphql_json into _graphql_variables (builds constraints) and a thin _graphql_json wrapper. The search branch calls _graphql_variables directly, so it no longer triggers _json_operation/search_hash — the dead GitHub-raw HTTP GET per imdb_search is gone. List/watchlist paths unchanged.
• Test coverage for the search path (043fdb6): added 11 tests (57 total, up from 46):
  • constraints/sort/limit split (asserts locale/first/sortBy/sortOrder excluded from constraints; sort split correctly),
  • limit omitted when data['limit'] == 0,
  • results parsing, empty-results and None-response Failed paths,
  • service-failure propagation (no fallback dataset for search),
  • _service_post success / 404 not_found_ok / ≥400 / invalid JSON / connection-error handling, mirroring the _service_request tests.
  • Autouse fixture patches the module logger to a no-op (it's None in tests) so the search path's logger.ghost/exorcise calls don't crash. tests/ is excluded from pyright, so the MagicMock-assignment warnings there are non-issues.

🟡 Scope note (unchanged)

Branch is stacked on imdb-service-ratings, so the diff into nightly carries all prior service work (ratings/genres/episode-ratings/parental/charts/keywords, gitpython bump, spellcheck tooling). Only the search change + its tests are new here. Should land after — or together with — the ratings branch.

✅ Looks good

• _service_post() mirrors _service_request() error handling and is now test-covered.
• constraints/sort/limit split matches the service contract.
• Omitting limit when 0 correctly maps Kometa's "no limit" convention onto the service default.
• List/watchlist correctly left on direct GraphQL.
• No-fallback-for-search is reasonable (no TSV equivalent exists) and is now asserted by a test.

Minor nits (non-blocking)

• Search block uses data.get("limit", 0) while the list path uses data["limit"] directly — harmless inconsistency since validate_imdb always sets limit.
• self._service_post(...) or {} turns a None (404) into "No IMDb IDs Found" rather than a distinct not-found message — fine, but less precise than passing not_found_ok=True.

[badja-dev]

the new search branch in _pagination calls logger.ghost(...) / logger.exorcise() unguarded, while every other new method in this PR (_service_request, _service_post, get_rating, etc.) wraps logger calls in if logger:

---

Once rebased onto nightly and the above is fix, looks good to me

[chazlarson]

@badja-dev can you approve if happy?