cleanup(phase7): backend destructive cleanup — B1-B6 + partial §3.2 audit migration#1665
Merged
Conversation
…allers Delete the Phase 3/4/5 pure re-export shims under ``aperag/service/``: api_key_service, audit_service, bot_service, chat_collection_service, chat_document_service, chat_service, chat_title_service, collection_service, collection_summary_service, default_model_service, document_service, graph_service, llm_available_model_service, llm_provider_service, marketplace_collection_service, marketplace_service, turn_feedback_service. All consumer imports rewritten to the canonical ``aperag.domains.<d>.service.<name>`` paths (app.py wire-up, views/main.py, views/collections.py, utils/audit_decorator.py, graph_curation/service.py, the two remaining ACTIVE_LEGACY services chat_completion_service.py / question_set_service.py, config/celery_tasks.py, and 9 test modules including the monkeypatch target in tests/unit_test/tasks/test_reconciler.py). PERMANENT (§1.1): quota_service, prompt_template_service, search_pipeline_service remain untouched. Gate: tests/unit_test/test_modularization_boundaries.py 20 passed. Ghost-check: none — batch B1; no PERMANENT §1.1 files touched; G17/G18-alt wirings in app.py lines 101–145 unchanged. Per docs/modularization/cleanup-inventory.md §B1. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…it__ Delete the Phase 3/4/5 pure router re-export shims under ``aperag/views/``: agent_runtime, api_key, audit, bots_v2, chat, collections_v2, documents_v2, evaluation_v2, llm, marketplace, marketplace_collections, providers_v2, plus the empty ``__init__.py`` (no consumers do ``from aperag.views import …``). All consumer imports rewritten to the canonical domain router paths across 6 OpenAPI-contract test modules and the llm rerank helper test. Do not touch §3.2 ACTIVE_LEGACY views (auth.py, main.py, prompts.py, settings.py, config.py, export.py, quota.py, openai.py, collections.py, test.py, utils.py, chat_documents.py). app.py mount list already uses canonical domain routers (lines 227–250) — no wiring change. Gate: tests/unit_test/test_modularization_boundaries.py 20 passed. Ghost-check: none — batch B2; no PERMANENT §1.1 files touched; G17/G18-alt wirings in app.py lines 101–145 unchanged. Per docs/modularization/cleanup-inventory.md §B2. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…kage Delete the Phase 5 step 5-S5b shim package ``aperag/agent_runtime/`` (__init__.py + runtime.py + schemas.py + services.py + storage.py), all of which were pure re-exports of ``aperag.domains.agent_runtime.*``. Migrated the 2 remaining live callers: - ``aperag/service/chat_completion_service.py`` (ACTIVE_LEGACY) now imports runtime/schemas/services from the domain path. - ``tests/unit_test/agent_runtime/test_agent_runtime_v3.py`` now imports schemas/services from the domain path. Gate: tests/unit_test/test_modularization_boundaries.py 20 passed. G17 / G18-alt CRITICAL_WIRINGS still pass — app.py already uses ``aperag.domains.agent_runtime.*`` for wire-up. Ghost-check: none — batch B3; no PERMANENT §1.1 files touched. Per docs/modularization/cleanup-inventory.md §B3. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…kage Delete the Phase 5 step 5-S6 shim package ``aperag/evaluation_v2/`` (__init__.py + constants.py + judges.py + schemas.py + services.py + tasks.py + worker.py). All were pure re-exports of ``aperag.domains.evaluation.*``. The orphaned ``EVALUATION_V2_SCHEMA_VERSION`` constant at the shim ``__init__`` had zero consumers. Celery task registry: Celery task NAME ``aperag.evaluation_v2.tasks.run_evaluation_run`` stays unchanged (preserved at the domain ``tasks.py`` ``@app.task(name=...)``), so in-flight messages keep routing. The ``include=`` list in ``config/celery.py`` is updated to ``aperag.domains.evaluation.tasks`` so Celery loads the module that registers the task. Migrated call sites: - ``aperag/domains/evaluation/tasks.py`` lazy import of ``execute_evaluation_run`` now from domain path - ``aperag/domains/evaluation/services.py`` ``launch_run`` lazy import of ``run_evaluation_run`` now from domain path - ``tests/unit_test/test_evaluation_v2_worker.py`` imports + monkeypatch module aliases now point at domain path Gate: tests/unit_test/test_modularization_boundaries.py 20 passed. Ghost-check: none — batch B4; no PERMANENT §1.1 files touched. Per docs/modularization/cleanup-inventory.md §B4. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…block Drop the ~70 LOC per-domain DB re-export block at ``aperag/db/models.py`` lines 395–462. Physical owners have been at ``aperag/domains/<d>/db/models.py`` since Phase 3/4/5 — this removes the back-compat aggregate. Kept unchanged (per inventory §1.2 + §1.4 + B5 local-models list): - Top-level ``from aperag.domains.identity.db.models import OAuthAccount, Role, User`` (Invitation class body binds Role at load time — G15-exempt). - Local models: ConfigModel, UserQuota, ModelServiceProvider, ModelServiceProviderStatus, QuestionType, EvaluationStatus, EvaluationItemStatus, QuestionSet, Question, Evaluation, EvaluationItem, Setting, ExportTaskStatus, ExportTask, PromptTemplate, Invitation. - ``Base = declarative_base()`` anchor and the ``from aperag.db.base`` re-export. Migrated 43 import blocks across 30 files to route moved symbols through ``aperag.domains.<d>.db.models``: app.py, views (auth/main/ collections/quota/test/export/prompts/openai — ACTIVE_LEGACY imports updated in place), service (chat_completion/quota/question_set/export/ evaluation — ACTIVE_LEGACY imports updated in place), all ``aperag/db/repositories/*``, domain lazy imports, llm helpers, tasks, graph_curation, config/celery_tasks, config/export_tasks, and the corresponding test modules. ``User`` / ``Role`` / ``OAuthAccount`` imports remain on ``aperag.db.models`` path per §1.2. Alembic metadata: ``aperag/migration/env.py`` now explicitly imports every domain's ``db.models`` module so all mappers register against ``Base.metadata`` before autogen runs (previously they rode on the aggregate re-export block's side effects). Ruff-I001 import-sort: ``ruff check --fix`` applied uniformly. Gates: - tests/unit_test/test_modularization_boundaries.py: 20 passed. - Spot-check of 15 B1-B5 affected test modules: 117 passed. - ruff check aperag/ tests/ config/: clean. Ghost-check: none — batch B5; §1.1 PERMANENT files unchanged; §1.2 Invitation/Role binding preserved; G17/G18-alt wirings unchanged. Per docs/modularization/cleanup-inventory.md §B5. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Phase 7 B5 stripped the cross-domain re-export block from ``aperag.db.models``; the old G11 audit at ``tests/unit_test/test_phase3_reexport_audit.py`` asserted symbols remained importable from that legacy aggregate. Rewrite the audit to assert every Phase 3 symbol resolves from its canonical ``aperag.domains.<d>.db.models`` path AND every Phase 3 table is still registered on ``Base.metadata`` (so Alembic autogen keeps seeing them). G13 single-definition audit is unchanged. Gate: tests/unit_test/test_phase3_reexport_audit.py 2 passed. Full unit_test suite: 578 passed, 29 skipped, 0 failed. Per docs/modularization/cleanup-inventory.md §B5 (delete gate). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…allers
Strips the 7 dual-hook `try:`/`except ImportError:` re-export blocks at
the bottom of `aperag/schema/view_models.py` (KB / identity / governance
/ marketplace / model_platform / conversation / agent_runtime) and the
symmetric `_bind_view_models_reexports()` write-back hooks inside each
of the 7 domain `schemas.py` modules.
All callers are rewritten to import directly from the canonical
`aperag.domains.<d>.schemas` path. `CollectionConfig` / `ModelSpec`
callers migrate to `aperag.schema.common` (their actual home since
Phase 3 Step 4b).
`view_models.py` retains its local schemas only (FailResponse, Settings,
ParserHealth*, Prompt*, Quota*, Question*, Evaluation*, ExportTask*,
LLMConfig, TargetEntityData*, NodeMerge*, ChatSuccessResponse,
EvaluationChatWith*) plus the Phase 2 retrieval / knowledge_graph
re-export block (not a dual-hook).
The 4 v2-covered v1 view files that depended on dual-hook-resolved
symbols are removed here as a unit (they cannot survive dual-hook strip
and have no migration destination per §3.2):
* `aperag/views/collections.py` (v1 upload flow; v2 provides
`/api/v2/collections/{id}/documents/{upload,confirm,fetch-url,staged}`)
* `aperag/views/main.py` (v1 llm_providers + default_models +
available_models; v2 providers_v2_routes covers equivalent under
`/api/v2/providers/*` + `/api/v2/default-models`)
* `aperag/views/chat_documents.py` (empty stub, no mount)
* `aperag/views/test.py` (dev-only test-admin-register helper)
`aperag/app.py` drops the 4 imports + 3 include_router calls + the
now-unused `import os` / DEPLOYMENT_MODE dev guard.
Per hard-cut口径: `X is view_models.X is domain.schemas.X` identity
invariant is intentionally broken; every caller has migrated to the
canonical per-domain path so the invariant is no longer load-bearing.
Ghost-check: none — §1 permanent whitelist untouched (quota_service,
prompt_template_service, search_pipeline_service, db/models.py line 36,
G17/G18-alt CRITICAL_WIRINGS in app.py:~101-145, shared-infra dirs);
G1-G19 boundary gate green (20/20); broad unit test sweep 577 passed /
29 skipped / 1 pre-existing FE-only test deselected
(test_phase1_fe_complete_identity_auth_admin_audit_adapter_boundary —
flags missing `listUserQuotas` in features/admin/client-api.ts, same
failure on `origin/bryce/phase7-backend-cleanup` HEAD, not introduced
by B6); ruff clean.
See `docs/modularization/cleanup-inventory.md §3.1 F` (dual-hook strip)
+ §3.2 (v2-covered view direct-rm entries).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…n domain Moves `aperag/service/chat_completion_service.py` (468 LOC) to its canonical location `aperag/domains/conversation/service/chat_completion_service.py`. The OpenAI-compat chat completions stream belongs to the conversation domain (chats + agent runtime integration). Rewires the two callers: * `aperag/views/openai.py` — import path update * `tests/unit_test/chat/test_chat_completion_service.py` — module path update Also drops the `from aperag.schema import view_models` + sole `view_models.ModelSpec` use-site in favor of a direct `from aperag.schema.common import ModelSpec` import (canonical home). Ghost-check: none — §1 permanent whitelist untouched; G1-G19 gate green (20/20 + targeted completion tests). See `docs/modularization/cleanup-inventory.md §3.2` row 1. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Moves `aperag/service/setting_service.py` (86 LOC) to `aperag/domains/governance/service/setting_service.py` — global settings (MinerU token, parser toggles) belong with governance / quota / audit. Only imports `aperag.db.ops`, no view_models or legacy db.models dependency, so the move is purely a path rename. Rewires two callers: * `aperag/views/settings.py` — import path * `aperag/tasks/utils.py` — local import inside indexing task Ghost-check: none — §1 permanent whitelist untouched; G1-G19 gate green (20/20). See `docs/modularization/cleanup-inventory.md §3.2` row 5. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Moves `aperag/service/test_mcp_agent.py` (332 LOC) to `tests/fixtures/mcp_agent.py` — it is a standalone multi-agent RAG smoke script, not runtime service code, and has never been imported by the rest of the backend (zero call sites under aperag/ or tests/). Per §3.2 row 6 its canonical home is the test-fixtures tree. Ghost-check: none — §1 permanent whitelist untouched; G1-G19 gate green (20/20). See `docs/modularization/cleanup-inventory.md §3.2` row 6. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Moves `aperag/views/config.py` (55 LOC) to `aperag/domains/identity/api/config_routes.py`. The `/api/v1/config` mount in `aperag/app.py` is updated to point at the new module; URL contract is byte-preserved (same prefix, same handler, same response shape). `aperag/views/utils.py::get_available_login_methods` stays at its current path for now — it is shared by auth.py (still legacy) and several other consumers; moving it to identity requires resolving the `views/utils.py` co-owner question (mask_api_key / validate_source_connect_config are used by governance / KB / model_platform domains). Deferred along with auth.py in the same follow-up. Ghost-check: none — §1 permanent whitelist untouched; G1-G19 gate green (20/20). See `docs/modularization/cleanup-inventory.md §3.2` row 10. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Moves `aperag/views/openai.py` (83 LOC) to `aperag/domains/conversation/api/openai_routes.py`. OpenAI-compat `/v1/chat/completions` belongs with the conversation domain (it routes into `chat_completion_service` which landed in the same domain two commits ago). Replaces the `from aperag.db.models import User` dependency with a narrow `AuthenticatedUser` `Protocol` (lesson 9a-ter / G1 boundary pattern used by `aperag/domains/knowledge_graph/api/routes.py`). The handler only reads `user.id` so the minimal ``id: object`` protocol is sufficient; the fastapi-users ``User`` row structurally satisfies it. `aperag/app.py` mount for `/v1` prefix updates to the new module; URL contract preserved. Ghost-check: none — §1 permanent whitelist untouched; G1-G19 gate green (20/20). See `docs/modularization/cleanup-inventory.md §3.2` row 13. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
`ruff check --fix` whipped three files into shape after the preceding six migration commits; no logical change, just import-order re-sorting that ruff's stable-sort requires and ``pre-commit`` didn't auto-run in my worktree. Ghost-check: none — §1 permanent whitelist untouched; G1-G19 gate green (20/20); ruff clean. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
… conversation schemas
Weston blocker-level CR (msg=ffe45a23) found 2 mechanical `git diff --check`
blockers from the B6 `_bind_view_models_reexports()` hook removal — both
`aperag/domains/{agent_runtime,conversation}/schemas.py` ended with three
trailing newlines after the closing bracket. Normalize to single newline.
No logic change; diff is whitespace-only.
Ghost-check: none — §1 Layer A/B/C/D/E untouched; whitespace-only EOF fix.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Phase 7 Backend destructive cleanup — task #31
Per
docs/modularization/cleanup-inventory.mdv1 (hard-cut口径) and PM msg=b55ca803 / msg=b2f621b6.Scope completed
aperag/service/*shim delete + caller migrate919a6ceaaperag/views/*shim + empty__init__delete5df80898aperag/agent_runtime/5-file re-export package delete7e4a1e59aperag/evaluation_v2/7-file re-export package delete (Celery task 字符串aperag.tasks.evaluation.*preserved)adfcacf7aperag/db/models.pycross-domain re-export block strip (InvitationRoletop-level import preserved per §1 Layer C)3225e23f25bbb25daperag/schema/view_models.py7 dual-hook blocks strip +_bind_view_models_reexports()hooks removed + callers migrated toaperag.domains.<d>.schemas；also removed v2-covered v1 views (collections.py/main.py/chat_documents.py/test.py)61c071d127c756ef07e790bf00deafc1880edb811c5c86b8chat/history/is ACTIVE code,misc/stopwords.txtis data,graph/already absent)§3.2 ACTIVE_LEGACY migrations done
aperag/service/chat_completion_service.py→aperag/domains/conversation/service/chat_completion_service.pyaperag/service/setting_service.py→aperag/domains/governance/service/setting_service.pyaperag/service/test_mcp_agent.py→tests/fixtures/mcp_agent.pyaperag/views/config.py→aperag/domains/identity/api/config_routes.py(mount/api/v1/configpreserved)aperag/views/openai.py→aperag/domains/conversation/api/openai_routes.py(mount/v1/chat/completionspreserved)aperag/views/{collections,main,chat_documents,test}.py→ rm (v2 已覆盖 / dev-only / empty stub)§3.2 deferred with canonical reasons (need 主架构师 SSoT decision)
以下 migrations 触达 v1 legacy ORM 的 canonical-home 决策或 Layer A PERMANENT Protocol 边界，不属于单个 coding lane scope。
aperag/service/evaluation_service.pyaperag/db/models.py；legacy v1 tables 未 migrate 到aperag/domains/evaluation/db/models.pyaperag/service/export_service.pyaperag/service/question_set_service.pyaperag/views/auth.py+aperag/views/utils.pyaperag/views/export.pyaperag.service.export_serviceaperag/views/prompts.pyprompt_template_service；G1 forbids domain→aperag/service/*aperag/views/quota.pyquota_serviceaperag/views/settings.pySettingsPydantic LOCAL toview_models.py；needs canonical home decision (governance vs KB)aperag/views/graph.py§1 PERMANENT 白名单 self-check (all preserved)
quota_service.py+prompt_template_service.py未动search_pipeline_service.py未动aperag/db/models.pytop-level Role import 保留app.pywiring 字节保留Gates (per inventory §3.4)
uv run pytest tests/unit_test/test_modularization_boundaries.py -x -q→ 20 pass / 0 fail (G1-G19 all green)uv run ruff check aperag/ tests/ config/→ cleanuv run pytest tests/unit_test/ -x --ignore tests/unit_test/objectstore -q→ 577 pass / 29 skip / 0 fail (1 pre-existing FE adapter-boundary test deselected —listUserQuotastrimmed by chore(fe-cleanup): Phase 7 destructive cleanup — F3/F4/F5/F6 (#32) #1664，not introduced by this PR)OpenAPI byte-check 未跑 — per v1 §0 hard-cut 口径 accept byte changes.
Net diff
~4 300 deletions / ~550 insertions / net ~3 750 LOC deleted
Ghost-check
Ghost-check: none — §1 Layer A/B/C/D/E whitelist untouched; G1-G19 20/20 green; deferred §3.2 items documented above with canonical rationale requiring 主架构师 SSoT decision.
🤖 Generated with Claude Code