chore: harden startup environment and simplify worker paths (#41)

* test: cover startup display environment

* fix(worker): kaputten API-Key-Pre-Check im TranscribeWorker entfernen

Der Pre-Check referenzierte self.config, das im Worker nicht existiert.
Bei fehlendem API-Key sah der Nutzer dadurch einen AttributeError statt
des Env-Var-Hinweises. LLMService.rewrite() meldet den fehlenden Key
bereits selbst korrekt; der doppelte Check entfällt ersatzlos.
Regressionstest deckt den Fehlerpfad ab.

* refactor: doppeltes Thread-Wiring und Completion-Call zusammenführen

ComposeWindow: _start_worker und _start_worker_raw teilten 13 Zeilen
identisches QThread-Wiring; jetzt gemeinsamer _launch_worker-Helper.
LLMService: dampf_ablassen, text_improver, emoji_text und rewrite_raw
wiederholten denselben Completion-Aufruf inkl. Leerantwort-Check;
jetzt gemeinsamer _chat_completion-Helper. Verhalten unverändert.

Author: TimInTech

ROADMAP.md

@@ -1,7 +1,6 @@
 """AudioRecorder for BlitztextLinux.
 
-Startet und stoppt Audioaufnahmen via parec (PulseAudio/PipeWire).
-Extrahiert aus whisper-dictation scripts/dictate_toggle.py v0.2.19.
+Starts and stops audio recordings via parec (PulseAudio/PipeWire).
 
 Design:
 - Synchron (kein Thread): der aufrufende Thread blockiert nicht,
@@ -23,7 +22,7 @@
 
 logger = logging.getLogger("blitztext.audio_recorder")
 
-# Aufnahme-Parameter (identisch zu whisper-dictation)
+# Recording parameters for Whisper-compatible 16 kHz mono WAV input.
 _RATE = 16000
 _CHANNELS = 1
 _LATENCY_MS = 100
@@ -102,7 +101,7 @@ def start(self, device: str = "@DEFAULT_SOURCE@") -> Path:
                 "parec nicht gefunden. Bitte installieren: sudo apt install pulseaudio-utils"
             )
 
-        # Stale-PID-Schutz (aus whisper-dictation v0.2.19)
+        # Stale PID guard for interrupted recording processes.
         self._cleanup_stale_pid()
 
         if self._pid_file.is_file():

app/audio_recorder.py

@@ -31,7 +31,7 @@
     sys.path.insert(0, PROJECT_DIR)
 
 from app.config import Config, DEFAULTS, VALID_HOTKEY_KEYS
-from app.llm_service import LLMService, WorkflowType, LLM_WORKFLOWS, LLMServiceError
+from app.llm_service import LLMService, WorkflowType, LLM_WORKFLOWS
 from app.writing_presets import WRITING_PRESET_KEYS, get_preset, preset_index
 from app.hotkey_service import HotkeyWorker
 from app.audio_recorder import AudioRecorder, AudioRecorderError
@@ -587,13 +587,10 @@ def run(self) -> None:
             if not transcript or not transcript.strip():
                 raise TranscribeError("Keine Sprache im Audio erkannt.")
 
-            # LLM rewrite if it is an LLM workflow
+            # LLM rewrite if it is an LLM workflow. rewrite() meldet einen
+            # fehlenden API-Key selbst als LLMServiceError mit Env-Var-Hinweis.
             if self.workflow in LLM_WORKFLOWS:
                 self._emit("status_changed", "rewriting")
-                if not self.llm_service.is_available():
-                    raise LLMServiceError(
-                        f"OpenAI API-Key nicht gesetzt. Bitte {self.config.openai_api_key_env} in ~/.config/blitztext-linux/secrets.env setzen."
-                    )
                 result_text = self.llm_service.rewrite(self.workflow, transcript)
             else:
                 result_text = transcript

app/blitztext_linux.py

@@ -655,7 +655,6 @@ def _start_worker(self, text: str) -> None:
         else:
             writing_preset = self._selected_preset()
 
-        thread = QThread(self)
         worker = _ComposeWorker(
             self._llm_service,
             workflow,
@@ -664,6 +663,11 @@ def _start_worker(self, text: str) -> None:
             tone=tone,
             custom_prompt=custom_prompt,
         )
+        self._launch_worker(worker)
+
+    def _launch_worker(self, worker: _ComposeWorker) -> None:
+        """Gemeinsames Thread-Wiring für Standard- und Raw-Prompt-Worker."""
+        thread = QThread(self)
         worker.moveToThread(thread)
         thread.started.connect(worker.run)
         worker.finished.connect(self._on_worker_result)
@@ -739,7 +743,6 @@ def _on_show_prompt_clicked(self) -> None:
             self._start_worker_raw(dialog.get_system_prompt(), dialog.get_user_message())
 
     def _start_worker_raw(self, system_prompt: str, user_message: str) -> None:
-        thread = QThread(self)
         worker = _ComposeWorker(
             self._llm_service,
             self._selected_workflow(),
@@ -748,19 +751,7 @@ def _start_worker_raw(self, system_prompt: str, user_message: str) -> None:
             raw_system_prompt=system_prompt,
             raw_user_message=user_message,
         )
-        worker.moveToThread(thread)
-        thread.started.connect(worker.run)
-        worker.finished.connect(self._on_worker_result)
-        worker.error.connect(self._on_worker_error)
-        worker.finished.connect(thread.quit)
-        worker.error.connect(thread.quit)
-        thread.finished.connect(worker.deleteLater)
-        thread.finished.connect(thread.deleteLater)
-        thread.finished.connect(self._on_worker_thread_finished)
-        self._worker_thread = thread
-        self._worker = worker
-        self._set_busy(True)
-        thread.start()
+        self._launch_worker(worker)
 
     @pyqtSlot(str)
     def _on_worker_result(self, result_text: str) -> None:

app/compose_window.py

@@ -1,7 +1,4 @@
-"""Verlauf-/Diktat-Panel fuer BlitztextLinux.
-
-Portiert und an die monolithische Blitztext-Architektur angepasst aus
-whisper-dictation app/gui/history_panel.py.
+"""History window for Blitztext Linux.
 
 GUI-freie Logik (Notiz speichern, Diktat zusammenfuehren) liegt in
 Modulfunktionen, damit sie ohne Qt/Display testbar ist.

app/history_panel.py

@@ -176,18 +176,13 @@ def _rewrite_for_workflow(
                 raise
             raise LLMServiceError(f"OpenAI API-Fehler: {exc}") from exc
 
-    def dampf_ablassen(self, transcript: str, custom_system_prompt: str = "") -> str:
-        self._check_openai()
-        if not transcript or not transcript.strip():
-            raise ValueError("transcript must not be empty")
-
-        system = (custom_system_prompt.strip() or self.dampf_system_prompt.strip() or _DAMPF_SYSTEM) + self._custom_terms_instruction()
-
+    def _chat_completion(self, system: str, user: str) -> str:
+        """Führt den Chat-Completion-Aufruf aus; gemeinsamer Pfad aller Workflows."""
         response = self.client.chat.completions.create(
             model=self.model,
             messages=[
                 {"role": "system", "content": system},
-                {"role": "user", "content": transcript.strip()},
+                {"role": "user", "content": user.strip()},
             ],
             temperature=0.7,
         )
@@ -196,6 +191,14 @@ def dampf_ablassen(self, transcript: str, custom_system_prompt: str = "") -> str
             raise LLMServiceError("OpenAI hat eine leere Antwort zurückgegeben.")
         return content.strip()
 
+    def dampf_ablassen(self, transcript: str, custom_system_prompt: str = "") -> str:
+        self._check_openai()
+        if not transcript or not transcript.strip():
+            raise ValueError("transcript must not be empty")
+
+        system = (custom_system_prompt.strip() or self.dampf_system_prompt.strip() or _DAMPF_SYSTEM) + self._custom_terms_instruction()
+        return self._chat_completion(system, transcript)
+
     def text_improver(self, transcript: str, tone: str = "neutral", custom_prompt: str = "") -> str:
         self._check_openai()
         if not transcript or not transcript.strip():
@@ -204,19 +207,7 @@ def text_improver(self, transcript: str, tone: str = "neutral", custom_prompt: s
             raise ValueError(f"invalid tone: {tone}")
 
         system = (custom_prompt.strip() or _TEXT_IMPROVER_SYSTEM_TEMPLATE.format(tone=tone)) + self._custom_terms_instruction()
-
-        response = self.client.chat.completions.create(
-            model=self.model,
-            messages=[
-                {"role": "system", "content": system},
-                {"role": "user", "content": transcript.strip()},
-            ],
-            temperature=0.7,
-        )
-        content = response.choices[0].message.content
-        if content is None:
-            raise LLMServiceError("OpenAI hat eine leere Antwort zurückgegeben.")
-        return content.strip()
+        return self._chat_completion(system, transcript)
 
     def emoji_text(self, transcript: str, density: str = "mittel") -> str:
         self._check_openai()
@@ -226,19 +217,7 @@ def emoji_text(self, transcript: str, density: str = "mittel") -> str:
             raise ValueError(f"invalid density: {density}")
 
         system = _EMOJI_SYSTEM_TEMPLATE.format(density=density) + self._custom_terms_instruction()
-
-        response = self.client.chat.completions.create(
-            model=self.model,
-            messages=[
-                {"role": "system", "content": system},
-                {"role": "user", "content": transcript.strip()},
-            ],
-            temperature=0.7,
-        )
-        content = response.choices[0].message.content
-        if content is None:
-            raise LLMServiceError("OpenAI hat eine leere Antwort zurückgegeben.")
-        return content.strip()
+        return self._chat_completion(system, transcript)
 
     def rewrite(self, workflow: WorkflowType, transcript: str) -> str:
         """Send transcript to OpenAI and return the rewritten text.
@@ -306,18 +285,7 @@ def rewrite_raw(self, system_prompt: str, user_message: str) -> str:
         if not user_message or not user_message.strip():
             raise ValueError("user_message must not be empty")
         try:
-            response = self.client.chat.completions.create(
-                model=self.model,
-                messages=[
-                    {"role": "system", "content": system_prompt},
-                    {"role": "user", "content": user_message.strip()},
-                ],
-                temperature=0.7,
-            )
-            content = response.choices[0].message.content
-            if content is None:
-                raise LLMServiceError("OpenAI hat eine leere Antwort zurückgegeben.")
-            return content.strip()
+            return self._chat_completion(system_prompt, user_message)
         except Exception as exc:
             if isinstance(exc, LLMServiceError):
                 raise

app/llm_service.py

@@ -1,7 +1,7 @@
-"""Desktop-Benachrichtigungen via notify-send.
+"""Desktop notifications for BlitztextLinux.
 
-Portiert aus whisper-dictation app/gui/notify.py. Schlaegt nie hart fehl --
-fehlt notify-send, wird die Meldung still uebersprungen.
+Notifications never fail hard. If notify-send is unavailable, the message is
+silently skipped.
 """
 from __future__ import annotations
 

app/notify.py

@@ -1,7 +1,5 @@
 """PasteService for BlitztextLinux.
 
-Kopiert/extrahiert aus whisper-dictation scripts/dictate_toggle.py v0.2.19.
-
 Zwei Schritte:
   1. wl-copy/xclip  -- Text in Clipboard schreiben
   2. ydotool       -- Ctrl+V simulieren (nur wenn autopaste=True)
@@ -20,9 +18,9 @@
 
 logger = logging.getLogger("blitztext.paste_service")
 
-# Verzoegerung zwischen wl-copy und ydotool key (identisch zu whisper-dictation)
+# Delay between clipboard write and simulated paste key.
 _PASTE_DELAY = 0.15
-# ydotool key-delay in ms (identisch zu whisper-dictation)
+# ydotool key-delay in ms.
 _KEY_DELAY_MS = 80
 # Strg+V als rohe Keycodes (`<keycode>:<pressed>`). ydotool >=1.0
 # interpretiert KEINE Tastennamen mehr wie "ctrl+v" -- solche Werte werden

app/paste_service.py

@@ -1,13 +1,6 @@
 #!/usr/bin/env python3
 """Transcribe a WAV file with Whisper.
 
-Kopiert aus whisper-dictation app/transcribe.py v0.2.19.
-Aenderungen gegenueber Original:
-  - Logger-Name: blitztext.transcribe
-  - Neue oeffentliche Funktion transcribe() fuer direkten Python-Import
-    (BlitztextLinux ruft nicht per subprocess auf)
-  - main() und CLI-Interface bleiben unveraendert erhalten
-
 Supports both openai-whisper and faster-whisper backends.
 """
 from __future__ import annotations
@@ -33,7 +26,7 @@ class TranscribeError(Exception):
 
 
 # ---------------------------------------------------------------------------
-# Public API (neu gegenueber whisper-dictation)
+# Public API
 # ---------------------------------------------------------------------------
 
 def transcribe(
@@ -87,7 +80,7 @@ def transcribe(
 
 
 # ---------------------------------------------------------------------------
-# Backend-Implementierungen (unveraendert aus whisper-dictation)
+# Backend implementations
 # ---------------------------------------------------------------------------
 
 def _normalize_language(language: str) -> Optional[str]:
@@ -203,7 +196,7 @@ def _transcribe_faster(
 
 
 # ---------------------------------------------------------------------------
-# CLI (unveraendert aus whisper-dictation)
+# CLI
 # ---------------------------------------------------------------------------
 
 def main() -> int:

app/transcribe.py

@@ -37,6 +37,22 @@ systemctl --user disable blitztext-linux
 ```
 </details>
 
+## Desktop session notes
+
+BlitztextLinux is developed for KDE Plasma on Wayland, with X11 fallbacks where the
+underlying tools support them.
+
+- GUI startup needs a real desktop session: either a usable `WAYLAND_DISPLAY` socket
+  or `DISPLAY` must be available. In headless shells, `scripts/verify.sh` can report
+  a warning even when the installed dependencies are otherwise correct.
+- Qt prefers Wayland when `WAYLAND_DISPLAY` points to an existing socket. If that
+  variable is stale but `DISPLAY` is set, the launcher falls back to X11.
+- Clipboard support uses `wl-copy`/`wl-paste` on Wayland and `xclip` on X11.
+- Auto-paste uses `ydotool`; terminal windows may need `Ctrl+Shift+V` instead of
+  `Ctrl+V`, so the app detects known terminal window classes when possible.
+- Global hotkeys still use `evdev`/the `input` group. A future desktop-native XDG
+  GlobalShortcuts integration would be a larger replacement, not a small setup fix.
+
 ## Manual install
 
 If you want to debug the Linux setup path step by step: