fix: self-audit — 7 issues found and corrected

1. Profiles cache invalidated when profiles_dir changes (fixed flaky test)
2. Missing db.commit() after index_page in apply_corrections (data was lost)
3. Client IA cached per provider instance (Google AI + Vertex SA)
4. prompt_loader resolves paths before LRU cache (avoids duplicate entries)
5. search.py import comments follow project conventions
6. Integration test resets _providers_cache + correct mock target
7. All 643 tests pass, 0 failures

https://claude.ai/code/session_012NCh8yLxMXkRmBYQgHCTik

backend/app/api/v1/pages.py

@@ -375,6 +375,7 @@ async def apply_corrections(
     # ── Mise à jour de l'index de recherche ──────────────────────────────
     from app.services.search.indexer import index_page
     await index_page(db, new_master)
+    await db.commit()
 
     logger.info(
         "Corrections appliquées",

backend/app/api/v1/profiles.py

@@ -27,24 +27,28 @@ logger = logging.getLogger(__name__)
 router = APIRouter(prefix="/profiles", tags=["profiles"])
 
 _profiles_cache: dict[str, CorpusProfile] | None = None
+_profiles_cache_dir: Path | None = None
 
 
 def _load_all_profiles() -> dict[str, CorpusProfile]:
-    """Charge tous les profils depuis le disque (cache singleton)."""
-    global _profiles_cache
-    if _profiles_cache is not None:
+    """Charge tous les profils depuis le disque (cache invalidé si le répertoire change)."""
+    global _profiles_cache, _profiles_cache_dir
+
+    current_dir = settings.profiles_dir
+    if _profiles_cache is not None and _profiles_cache_dir == current_dir:
         return _profiles_cache
 
     result: dict[str, CorpusProfile] = {}
-    if settings.profiles_dir.is_dir():
-        for path in sorted(settings.profiles_dir.glob("*.json")):
+    if current_dir.is_dir():
+        for path in sorted(current_dir.glob("*.json")):
             profile = _load_profile(path)
             if profile is not None:
                 result[profile.profile_id] = profile
     else:
-        logger.warning("profiles_dir introuvable : %s", settings.profiles_dir)
+        logger.warning("profiles_dir introuvable : %s", current_dir)
 
     _profiles_cache = result
+    _profiles_cache_dir = current_dir
     return _profiles_cache
 
 

backend/app/api/v1/search.py

@@ -7,12 +7,15 @@ POST /api/v1/search/reindex
 Implémentation indexée : les données sont dans la table page_search,
 mises à jour à chaque écriture de master.json.
 """
+# 1. stdlib
 import logging
 
+# 2. third-party
 from fastapi import APIRouter, Depends, Query
 from pydantic import BaseModel
 from sqlalchemy.ext.asyncio import AsyncSession
 
+# 3. local
 from app import config as _config_module
 from app.models.database import get_db
 from app.services.search.indexer import reindex_all, search_pages

backend/app/services/ai/prompt_loader.py

@@ -39,7 +39,7 @@ def load_and_render_prompt(template_path: str | Path, context: dict[str, str]) -
     Raises:
         FileNotFoundError: si le fichier template n'existe pas.
     """
-    path = Path(template_path)
+    path = Path(template_path).resolve()
 
     template = _read_template(str(path))
 

backend/app/services/ai/provider_google_ai.py

@@ -21,6 +21,9 @@ _ENV_KEY = "GOOGLE_AI_STUDIO_API_KEY"
 class GoogleAIProvider(AIProvider):
     """Provider Google AI Studio (clé API GOOGLE_AI_STUDIO_API_KEY)."""
 
+    def __init__(self) -> None:
+        self._client: genai.Client | None = None
+
     @property
     def provider_type(self) -> ProviderType:
         return ProviderType.GOOGLE_AI_STUDIO
@@ -28,11 +31,17 @@ class GoogleAIProvider(AIProvider):
     def is_configured(self) -> bool:
         return bool(os.environ.get(_ENV_KEY))
 
+    def _get_client(self) -> genai.Client:
+        """Retourne un client cached (réutilise la connexion SSL)."""
+        if self._client is None:
+            self._client = genai.Client(api_key=os.environ[_ENV_KEY])
+        return self._client
+
     def list_models(self) -> list[ModelInfo]:
         if not self.is_configured():
             raise RuntimeError(f"Variable d'environnement manquante : {_ENV_KEY}")
 
-        client = genai.Client(api_key=os.environ[_ENV_KEY])
+        client = self._get_client()
         result: list[ModelInfo] = []
 
         for model in client.models.list():
@@ -58,7 +67,7 @@ class GoogleAIProvider(AIProvider):
     def generate_content(self, image_bytes: bytes, prompt: str, model_id: str, supports_vision: bool = True) -> str:
         if not self.is_configured():
             raise RuntimeError(f"Variable d'environnement manquante : {_ENV_KEY}")
-        client = genai.Client(api_key=os.environ[_ENV_KEY])
+        client = self._get_client()
 
         if supports_vision:
             image_part = types.Part.from_bytes(data=image_bytes, mime_type="image/jpeg")

backend/app/services/ai/provider_vertex_sa.py

@@ -29,6 +29,9 @@ class VertexServiceAccountProvider(AIProvider):
     Le project_id est extrait du JSON ; la localisation par défaut est us-central1.
     """
 
+    def __init__(self) -> None:
+        self._client: genai.Client | None = None
+
     @property
     def provider_type(self) -> ProviderType:
         return ProviderType.VERTEX_SERVICE_ACCOUNT
@@ -36,7 +39,10 @@ class VertexServiceAccountProvider(AIProvider):
     def is_configured(self) -> bool:
         return bool(os.environ.get(_ENV_KEY))
 
-    def _build_client(self) -> genai.Client:
+    def _get_client(self) -> genai.Client:
+        """Retourne un client cached (évite de re-parser le JSON à chaque appel)."""
+        if self._client is not None:
+            return self._client
         sa_json_str = os.environ[_ENV_KEY]
         try:
             sa_info = json.loads(sa_json_str)
@@ -51,18 +57,19 @@ class VertexServiceAccountProvider(AIProvider):
             sa_info,
             scopes=_VERTEX_SCOPES,
         )
-        return genai.Client(
+        self._client = genai.Client(
             vertexai=True,
             project=project_id,
             location=_DEFAULT_LOCATION,
             credentials=credentials,
         )
+        return self._client
 
     def list_models(self) -> list[ModelInfo]:
         if not self.is_configured():
             raise RuntimeError(f"Variable d'environnement manquante : {_ENV_KEY}")
 
-        client = self._build_client()
+        client = self._get_client()
         result: list[ModelInfo] = []
 
         for model in client.models.list():
@@ -88,7 +95,7 @@ class VertexServiceAccountProvider(AIProvider):
     def generate_content(self, image_bytes: bytes, prompt: str, model_id: str, supports_vision: bool = True) -> str:
         if not self.is_configured():
             raise RuntimeError(f"Variable d'environnement manquante : {_ENV_KEY}")
-        client = self._build_client()
+        client = self._get_client()
 
         if supports_vision:
             image_part = types.Part.from_bytes(data=image_bytes, mime_type="image/jpeg")

backend/tests/test_integration_pipeline.py

@@ -140,12 +140,17 @@ async def test_full_pipeline(pipeline_fixtures, tmp_path):
     mock_provider = MagicMock()
     mock_provider.generate_content.return_value = _FAKE_AI_RESPONSE
 
+    # Reset le cache global des providers pour éviter les interférences
+    import app.services.ai.model_registry as _reg
+    old_providers_cache = _reg._providers_cache
+    _reg._providers_cache = None
+
     try:
         with patch(
             "app.services.job_runner.fetch_ai_derivative_bytes",
             return_value=(_FAKE_JPEG, 1500, 1000),
         ), patch(
-            "app.services.ai.model_registry.get_provider",
+            "app.services.ai.analyzer.get_provider",
             return_value=mock_provider,
         ):
             from app.services.job_runner import _run_job_impl
@@ -153,6 +158,7 @@ async def test_full_pipeline(pipeline_fixtures, tmp_path):
     finally:
         config_mod.settings.__dict__["data_dir"] = original_data_dir
         config_mod.settings.__dict__["profiles_dir"] = original_profiles_dir
+        _reg._providers_cache = old_providers_cache
 
     # -- Assertions ----------------------------------------------------------
     # Job should be done