fix: Bugfix-Sweep – Worker-Lebensdauer, HTML-Export, RAG-Verdrahtung, Extraktion

CHANGELOG.md

@@ -3,6 +3,20 @@
 Alle wesentlichen Änderungen an diesem Projekt werden hier dokumentiert.
 Format basiert auf [Keep a Changelog](https://keepachangelog.com/de/1.1.0/).
 
+## [Unreleased — 2026-10-06]
+
+### Fixed
+- GUI workers (`QThread`) are now kept alive until `finished` (`src/gui/worker_utils.py`); a running extraction is never replaced, queued files are no longer dropped, and `closeEvent` stops and waits for all workers before saving.
+- HTML export escapes all Markdown text and the document title and only renders `http`/`https`/`mailto`/relative links; the GUI HTML export uses the same `ReportExporter` implementation. TXT export strips Markdown syntax only (keeps `C#`, `#12`, `file_name`).
+- Text extraction no longer triggers synchronous embedding on the GUI thread; documents are queued and indexed by the `IndexWorker`. Re-indexing embeds before deleting old chunks; failed indexing resets `is_indexed`; stale `is_indexed` flags are reconciled with the vector index when a project is activated.
+- RAG search uses relevance scores (higher = better) instead of raw distances for confidence and thresholds.
+- Chat always receives the current project's document manager (no unfiltered queries over the shared collection) and a disabled RAG engine is propagated as `None`.
+- New projects inherit provider/model/URL from the saved app config; the LLM client is initialised at startup; projects are opened by ID, the current project is saved first, and project directories get a unique suffix.
+- Supported file types come from one shared set (adds `.pptx`, `.html`, `.htm`; drops unsupported `.odt`, `.ods`); HTML is converted to plain text.
+- Extractor: Excel `0`/`False` cells kept, RTF `\uN` fallback skipping and multi-byte code pages fixed, BOM/cp1252 detection for plain text, PPTX slides sorted numerically, PDF/Excel/MSG handles closed reliably.
+- Ollama availability is re-checked lazily instead of being cached forever after a failed start-up check.
+- YAML front matter quotes title/author, GUI PDF export has a timeout, error message and temp-file cleanup, chat messages render as plain text, translator hint matching uses whole words, and the Web Companion stores review notes per project ID (new optional `workspace.id` in the export).
+
 ## [Unreleased — 2026-09-22]
 
 ### Added

EXPORTFORMAT.md

@@ -47,6 +47,7 @@ Review-Notizen als Markdown exportiert.
     "exported_at": "2026-05-26T00:00:00Z"
   },
   "workspace": {
+    "id": "8f0c6a2e-…",
     "title": "Projektname",
     "question": "Zentrale Fragestellung",
     "workflow_type": "analysis",

README.md

@@ -6,8 +6,8 @@
 [![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/)
 [![Platform: Windows](https://img.shields.io/badge/platform-Windows-lightgrey.svg)]()
 [![Offline-first](https://img.shields.io/badge/offline--first-yes-green.svg)]()
-[![Pytest](https://img.shields.io/badge/Pytest-103%20passed-brightgreen.svg)]()
-[![Web Companion Tests](https://img.shields.io/badge/Web%20Companion-58%20passed-brightgreen.svg)]()
+[![Pytest](https://img.shields.io/badge/Pytest-147%20passed-brightgreen.svg)]()
+[![Web Companion Tests](https://img.shields.io/badge/Web%20Companion-60%20passed-brightgreen.svg)]()
 [![Notice: Invariants](https://img.shields.io/badge/Notice-INV--LOCAL--01..10-blue.svg)](NOTICE)
 [![SBOM: Level 1](https://img.shields.io/badge/SBOM-Level%201-blue.svg)](THIRD_PARTY_LICENSES.md)
 [![Ecosystem: file-bricks](https://img.shields.io/badge/Ecosystem-file--bricks-blue.svg)](https://github.com/file-bricks)

README_de.md

@@ -6,8 +6,8 @@
 [![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/)
 [![Plattform: Windows](https://img.shields.io/badge/Plattform-Windows-lightgrey.svg)]()
 [![Offline-first](https://img.shields.io/badge/offline--first-ja-green.svg)]()
-[![Pytest](https://img.shields.io/badge/Pytest-103%20bestanden-brightgreen.svg)]()
-[![Web Companion Tests](https://img.shields.io/badge/Web%20Companion-58%20bestanden-brightgreen.svg)]()
+[![Pytest](https://img.shields.io/badge/Pytest-147%20bestanden-brightgreen.svg)]()
+[![Web Companion Tests](https://img.shields.io/badge/Web%20Companion-60%20bestanden-brightgreen.svg)]()
 [![Notice: Invarianten](https://img.shields.io/badge/Notice-INV--LOCAL--01..10-blue.svg)](NOTICE)
 [![SBOM: Level 1](https://img.shields.io/badge/SBOM-Level%201-blue.svg)](THIRD_PARTY_LICENSES.md)
 [![Ökosystem: file-bricks](https://img.shields.io/badge/%C3%96kosystem-file--bricks-blue.svg)](https://github.com/file-bricks)

llms.txt

@@ -2,7 +2,7 @@
 
 ## Last-checked: 2026-09-22
 
-> Local NotebookLM alternative for private document analysis, RAG-assisted research, and multi-format report generation. Verified with 161 passing unit tests (103 Python pytest + 58 Web Companion Node.js tests).
+> Local NotebookLM alternative for private document analysis, RAG-assisted research, and multi-format report generation. Verified with 207 passing unit tests (147 Python pytest + 60 Web Companion Node.js tests).
 
 ## Description
 

src/core/document_manager.py

@@ -25,6 +25,8 @@
 from typing import Dict, List, Optional, Set, Callable, TYPE_CHECKING
 import uuid
 
+from .text_extractor import SUPPORTED_EXTENSIONS as _EXTRACTOR_SUPPORTED_EXTENSIONS
+
 if TYPE_CHECKING:
     from ..rag.engine import RAGEngine
 
@@ -147,21 +149,8 @@ class DocumentManager:
     - Manage sub-query associations
     """
 
-    # Supported file types
-    SUPPORTED_EXTENSIONS = {
-        # Text
-        ".txt", ".md", ".rst", ".log",
-        # Documents
-        ".pdf", ".docx", ".doc", ".odt", ".rtf",
-        # Data
-        ".json", ".xml", ".yaml", ".yml", ".csv",
-        # Code (for analysis)
-        ".py", ".js", ".ts", ".java", ".cpp", ".c", ".h",
-        # Spreadsheets
-        ".xlsx", ".xls", ".ods",
-        # Email
-        ".eml", ".msg"
-    }
+    # Supported file types -- abgeleitet aus dem TextExtractor (eine Quelle der Wahrheit)
+    SUPPORTED_EXTENSIONS = _EXTRACTOR_SUPPORTED_EXTENSIONS
 
     def __init__(self, project_path: Optional[Path] = None, rag_engine: Optional['RAGEngine'] = None):
         """
@@ -179,6 +168,7 @@ def __init__(self, project_path: Optional[Path] = None, rag_engine: Optional['RA
         self._auto_extract: bool = True  # Automatisch Text extrahieren bei add_file
         self._text_extractor = None  # Lazy-loaded
         self._pending_extractions: List[str] = []  # doc_ids waiting for async extraction
+        self._pending_index: List[str] = []  # doc_ids waiting for async (re-)indexing
 
         if project_path:
             self._cache_dir = project_path / ".cache"
@@ -492,15 +482,41 @@ def _try_auto_extract(self, doc: DocumentItem) -> None:
             self._notify_change("update", doc)
             logger.error(f"Auto-Extraktion Fehler: {doc.name}: {e}")
 
-    def set_rag_engine(self, rag_engine: 'RAGEngine') -> None:
+    def set_rag_engine(self, rag_engine: Optional['RAGEngine']) -> None:
         """
         Setzt die RAG Engine für semantische Suche.
 
         Args:
-            rag_engine: Die RAG Engine Instanz
+            rag_engine: Die RAG Engine Instanz oder None zum Trennen
         """
         self._rag_engine = rag_engine
-        logger.info("RAG Engine verbunden")
+        if rag_engine is None:
+            self._pending_index.clear()
+            logger.info("RAG Engine getrennt")
+        else:
+            logger.info("RAG Engine verbunden")
+
+    def sync_index_flags(self) -> None:
+        """Gleicht is_indexed mit dem tatsächlichen Inhalt des Vektor-Index ab.
+
+        Nötig, weil der Index projektübergreifend geleert werden kann und
+        documents.json dann veraltete is_indexed-Flags enthält.
+        """
+        if not self._rag_engine:
+            return
+        flagged = [d for d in self._documents.values() if d.is_indexed and not d.is_directory]
+        if not flagged:
+            return
+        try:
+            present = self._rag_engine.get_indexed_document_ids([d.id for d in flagged])
+        except Exception as e:
+            logger.warning("Index-Abgleich fehlgeschlagen: %s", e)
+            return
+        for doc in flagged:
+            if doc.id not in present:
+                doc.is_indexed = False
+                doc.chunk_count = 0
+                self._notify_change("deindexed", doc)
 
     def pop_pending_extractions(self) -> List[tuple]:
         """Gibt ausstehende Extraktionen zurück und leert die Queue.
@@ -516,6 +532,38 @@ def pop_pending_extractions(self) -> List[tuple]:
         self._pending_extractions.clear()
         return result
 
+    def has_pending_extractions(self) -> bool:
+        """True, wenn Dokumente auf die asynchrone Extraktion warten."""
+        return any(
+            doc_id in self._documents and not self._documents[doc_id].is_directory
+            for doc_id in self._pending_extractions
+        )
+
+    def discard_pending_extractions(self, doc_ids) -> None:
+        """Entfernt Dokumente aus der Extraktions-Queue (z.B. weil sie bereits extrahiert werden)."""
+        ids = set(doc_ids)
+        self._pending_extractions = [d for d in self._pending_extractions if d not in ids]
+
+    def pop_pending_index(self) -> List[tuple]:
+        """Gibt Dokumente zurück, die (neu) indexiert werden sollen, und leert die Queue.
+
+        Returns:
+            Liste von (doc_id, doc_name) Tupeln
+        """
+        result = []
+        seen = set()
+        for doc_id in self._pending_index:
+            doc = self._documents.get(doc_id)
+            if doc and not doc.is_directory and doc.extracted_text and doc_id not in seen:
+                seen.add(doc_id)
+                result.append((doc.id, doc.name))
+        self._pending_index.clear()
+        return result
+
+    def index_pending_documents(self) -> Dict[str, bool]:
+        """Indexiert alle wartenden Dokumente synchron (nur außerhalb des GUI-Threads nutzen)."""
+        return {doc_id: self.index_document(doc_id) for doc_id, _ in self.pop_pending_index()}
+
     def set_auto_index(self, enabled: bool) -> None:
         """Aktiviert/Deaktiviert automatische Indexierung."""
         self._auto_index = enabled
@@ -562,12 +610,21 @@ def index_document(self, doc_id: str) -> bool:
                 return True
             else:
                 logger.error(f"Indexierung fehlgeschlagen: {result.error}")
+                self._mark_not_indexed(doc)
                 return False
 
         except Exception as e:
             logger.error(f"Fehler bei Indexierung von {doc_id}: {e}")
+            self._mark_not_indexed(doc)
             return False
 
+    def _mark_not_indexed(self, doc: DocumentItem) -> None:
+        """Setzt den Index-Status nach einem Fehlschlag zurück."""
+        if doc.is_indexed or doc.chunk_count:
+            doc.is_indexed = False
+            doc.chunk_count = 0
+            self._notify_change("deindexed", doc)
+
     def index_all_documents(self) -> Dict[str, bool]:
         """
         Indexiert alle Dokumente mit extrahiertem Text.
@@ -741,17 +798,30 @@ def get_rag_statistics(self) -> Dict:
 
         return stats
 
-    # Override update_content to auto-index
     def update_content(self, doc_id: str, text: str) -> None:
-        """Update extracted text for a document."""
+        """Update extracted text for a document.
+
+        Indexiert NICHT synchron (Embedding-Aufrufe gehen über HTTP und würden
+        den GUI-Thread blockieren). Bei aktivem Auto-Index wird das Dokument in
+        eine Queue gestellt, die die GUI per IndexWorker abarbeitet
+        (pop_pending_index) bzw. Nicht-GUI-Aufrufer per index_pending_documents().
+        """
         doc = self._documents.get(doc_id)
         if doc:
+            new_hash = hashlib.md5(text.encode()).hexdigest()
+            content_changed = new_hash != doc.content_hash
             doc.extracted_text = text
             doc.text_length = len(text)
-            doc.content_hash = hashlib.md5(text.encode()).hexdigest()
+            doc.content_hash = new_hash
             doc.status = DocumentStatus.READY
+            if content_changed and doc.is_indexed:
+                # Vorhandene Chunks passen nicht mehr zum Text
+                doc.is_indexed = False
+                doc.chunk_count = 0
             self._notify_change("update", doc)
 
-            # Auto-indexieren wenn aktiviert
-            if self._auto_index and self._rag_engine and text:
-                self.index_document(doc_id)
+            # Auto-Indexierung vormerken (asynchron)
+            needs_index = content_changed or not doc.is_indexed
+            if (self._auto_index and self._rag_engine and text and needs_index
+                    and doc_id not in self._pending_index):
+                self._pending_index.append(doc_id)

src/core/project.py

@@ -60,6 +60,11 @@ def to_dict(self) -> dict:
             "language": self.language
         }
 
+    @classmethod
+    def from_app_config(cls) -> "ProjectSettings":
+        """Neue Projekt-Settings mit den gespeicherten LLM-Einstellungen der App."""
+        return cls.from_dict({})
+
     @classmethod
     def from_dict(cls, data: dict) -> "ProjectSettings":
         from .app_config import get_app_config
@@ -126,7 +131,9 @@ def create(cls, name: str, main_question: str = "", report_type: str = "analysis
             id=str(uuid.uuid4()),
             name=name,
             main_question=main_question,
-            report_type=report_type
+            report_type=report_type,
+            # LLM-Provider/Modell/URL aus der gespeicherten App-Konfiguration
+            settings=ProjectSettings.from_app_config(),
         )
 
     @property
@@ -359,23 +366,35 @@ def open_project(self, project_id_or_name: str) -> Optional[Project]:
         Returns:
             The project if found
         """
-        # Search by ID or name
-        for item in self.projects_dir.iterdir():
+        # ID hat Vorrang vor dem (nicht eindeutigen) Namen
+        by_name = []
+        for item in sorted(self.projects_dir.iterdir()):
             if item.is_dir():
                 project_file = item / "project.json"
                 if project_file.exists():
                     try:
                         data = json.loads(project_file.read_text(encoding="utf-8"))
-                        if data["id"] == project_id_or_name or data["name"] == project_id_or_name:
-                            project = Project.load(item)
-                            if project:
-                                self._current_project = project
-                                return project
-                    except Exception:
-                        pass
+                    except (OSError, ValueError):
+                        continue  # beschädigte project.json überspringen
+                    if data.get("id") == project_id_or_name:
+                        return self._load_as_current(item)
+                    if data.get("name") == project_id_or_name:
+                        by_name.append((data.get("modified_at", data.get("created_at", "")), item))
+
+        if by_name:
+            # Bei Namensdubletten das zuletzt geänderte Projekt öffnen
+            by_name.sort(key=lambda entry: entry[0], reverse=True)
+            return self._load_as_current(by_name[0][1])
 
         return None
 
+    def _load_as_current(self, directory: Path) -> Optional[Project]:
+        """Load a project directory and make it the current project."""
+        project = Project.load(directory)
+        if project:
+            self._current_project = project
+        return project
+
     def save_current(self) -> bool:
         """Save the current project."""
         if not self._current_project:
@@ -438,9 +457,9 @@ def _safe_dirname(self, name: str) -> str:
         safe = "".join(c for c in name if c.isalnum() or c in " -_").strip()
         safe = safe.replace(" ", "_")
 
-        # Add timestamp for uniqueness
+        # Timestamp + Kurz-UUID: zwei Projekte in derselben Sekunde kollidieren nicht
         timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
-        return f"{safe}_{timestamp}"
+        return f"{safe or 'Projekt'}_{timestamp}_{uuid.uuid4().hex[:8]}"
 
     # Output profiles
     def get_output_profiles(self) -> List[OutputProfile]: