[hyoban]

Summary

API unit and controller test jobs have gotten much slower since early October (merge-queue medians: unit shard ~4m → ~8m, controllers ~3m → ~10m50). This PR fixes the two largest causes that aren't just "more tests".

1. IM provider SDKs imported by create_app() (from #44864)

create_app() reaches core/human_input_v2/im_integration/adapters/factory.py, which imported every IM provider adapter and its vendor SDK at module load. lark_oapi alone costs ~3.5s, so each xdist worker and every subprocess-based test paid it, and existing tests that happened to run first in a worker jumped from ~0.05s to 3–6s.

Each adapter import now lives in its match branch, matching the pattern already used by _card_decoder in services/human_input_v2/im_webhook_runtime.py.

• create_app() no longer loads lark_oapi, slack_sdk, botbuilder, botframework, alibabacloud_dingtalk, or wechatpy.
• python -c "from app_factory import create_app; create_app()": ~11.6s → ~7.9s.

2. Error contract check copies every surface model into each test API (from #44001)

The autouse error_response_contract fixture builds an OpenAPI document for each ExternalApi that serves an error response. Test-built APIs borrowed models by copying all ~2200 surface models into api.models, so a one-path document carried every schema and took ~90ms in finalize_openapi_payload. About 1200 test APIs per run made that ~107s of CPU, about 18% of the controller suite locally.

Models are now borrowed through a fallback mapping. Name lookups (the only way Swagger resolves models) fall back to the surfaces, but iteration (the way document finalization merges models) yields only the API's own models. The swap is scoped to the build, so api.models itself is no longer mutated.

• For all 1200 test APIs in the controller suite, the new documents have identical paths, every $ref resolves, and every referenced component is identical to the old document.
• Document building time: 108s → 13s of CPU.
• Controller suite locally (-n 8, two runs each): 74s / 98s → 62s / 79s.
• Added test_documents_only_the_surface_models_a_test_api_references; it fails on the old helper.

Checklist

• I've verified the change and added or updated tests where meaningful regression risk justifies coverage.
• I ran make lint && make type-check (backend) and vp staged (frontend) to appease the lint gods

From Claude Code

[github-actions]

Pyrefly Type Coverage

Metric	Base	PR	Delta
Type coverage	83.01%	83.01%	-0.00%
Strict coverage	82.66%	82.66%	-0.00%
Typed symbols	85,604	85,605	+1
Untyped symbols	17,597	17,598	+1
Modules	4373	4373	0

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.80%. Comparing base (ac0f594) to head (c7c404d).
⚠️ Report is 8 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff             @@
##             main   #45381       +/-   ##
===========================================
+ Coverage   68.81%   89.80%   +20.98%     
===========================================
  Files        6103     6239      +136     
  Lines      368940   390949    +22009     
  Branches    77825    79291     +1466     
===========================================
+ Hits       253897   351075    +97178     
+ Misses     112233    33774    -78459     
- Partials     2810     6100     +3290     

Flag	Coverage Δ
api	89.64% <100.00%> (+37.30%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.