feat: Glossar-Entwurf aus Rohtext und Korrekturlauf (glossar_vorschlag.py)

- Das Modell sucht abschnittsweise vermutete Hörfehler (Namen, Orte,
  Fremdwörter, Komposita); nur Funde, die wörtlich im Rohtext stehen
- Sichere Funde als Einträge, unsichere auskommentiert („# ?“)
- -k: Änderungen eines Korrekturlaufs, immer auskommentiert, da oft
  zusammenhangsabhängig
- -g: vorhandene Glossare werden vorher angewendet und nicht erneut vorgeschlagen
- Modellantworten im Zwischenspeicher <Entwurf>.parts/

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

tests/test_glossar_vorschlag.py

@@ -0,0 +1,129 @@
+"""Tests for glossar_vorschlag.py; the Ollama model is always replaced by fakes."""
+import io
+import json
+import sys
+
+import pytest
+
+import glossar_vorschlag as gv
+import rede_korrigieren as rk
+
+RAW = (
+    "Der Kremmel und Egon Bar trafen Host Telchik. Die Rechtsstaatbahn war gut. "
+    "Herr Grotzki und Ghard Kocher sprachen über das gedeiliche Verhältnis."
+)
+
+
+def fake_model(entries):
+    """urlopen stand-in answering every chunk with the same entries."""
+    calls = []
+
+    def urlopen(request, timeout=None):
+        calls.append(json.loads(request.data))
+        content = json.dumps({"eintraege": entries}, ensure_ascii=False)
+        return io.BytesIO(json.dumps({"message": {"content": content}}).encode("utf-8"))
+
+    urlopen.calls = calls
+    return urlopen
+
+
+ENTRIES = [
+    {"falsch": "Grotzki", "richtig": "Grotzky", "sicher": True},
+    {"falsch": "Ghard Kocher", "richtig": "Gerhard Kocher", "sicher": False},
+    {"falsch": "Kremmel", "richtig": "Kreml", "sicher": True},       # in glossary
+    {"falsch": "Nirgendwo", "richtig": "Irgendwo", "sicher": True},  # not in text
+    {"falsch": "gut", "richtig": "gut", "sicher": True},             # no change
+    {"falsch": "", "richtig": "x", "sicher": True},
+    "kein Objekt",
+]
+
+
+@pytest.fixture
+def raw_file(tmp_path):
+    path = tmp_path / "rede.txt"
+    path.write_text(RAW + "\n", encoding="utf-8")
+    return path
+
+
+@pytest.fixture
+def glossary(tmp_path):
+    path = tmp_path / "g.txt"
+    path.write_text("Kremmel => Kreml\nEgon Bar => Egon Bahr\n", encoding="utf-8")
+    return path
+
+
+def no_model_calls(*args, **kwargs):
+    raise AssertionError("Modell wurde unerwartet aufgerufen")
+
+
+def run_main(monkeypatch, *args):
+    monkeypatch.setattr(sys, "argv", ["glossar_vorschlag.py", *map(str, args)])
+    gv.main()
+
+
+def test_main_filters_and_formats(raw_file, glossary, monkeypatch, capsys):
+    model = fake_model(ENTRIES)
+    monkeypatch.setattr(gv.urllib.request, "urlopen", model)
+    run_main(monkeypatch, "-i", raw_file, "-g", glossary, "-m", "test:1b")
+
+    draft = raw_file.with_name("rede.glossar.entwurf.txt").read_text(encoding="utf-8")
+    lines = draft.splitlines()
+    assert "Grotzki => Grotzky" in lines
+    assert "# ? Ghard Kocher => Gerhard Kocher" in lines
+    for dropped in ("Kremmel", "Nirgendwo", "gut =>", " => x"):
+        assert dropped not in draft
+    assert "Modell test:1b" in lines[0]
+    # The model sees the text with the known replacements already applied
+    sent = model.calls[0]["messages"][1]["content"]
+    assert "Kreml" in sent and "Kremmel" not in sent
+    assert model.calls[0]["model"] == "test:1b"
+
+    out = capsys.readouterr().out
+    assert "Vermutete Hörfehler: 2 (1 sicher, 1 unsicher)" in out
+
+    # Second run uses the cache
+    monkeypatch.setattr(gv.urllib.request, "urlopen", no_model_calls)
+    run_main(monkeypatch, "-i", raw_file, "-g", glossary, "-m", "test:1b")
+
+
+def test_correction_changes_are_commented_out(raw_file, glossary, tmp_path, monkeypatch):
+    monkeypatch.setattr(gv.urllib.request, "urlopen", fake_model(ENTRIES[:2]))
+    corrected = tmp_path / "rede.korrigiert.txt"
+    corrected.write_text(
+        "Der Kreml und Egon Bahr trafen Host Telchik. Die Rechtsstaat Bahn war gut. "
+        "Herr Grotzky und Ghard Kocher sprachen über das gedeihliche Verhältnis.\n",
+        encoding="utf-8",
+    )
+    out = tmp_path / "entwurf.txt"
+    run_main(monkeypatch, "-i", raw_file, "-k", corrected, "-g", glossary, "-o", out)
+
+    lines = out.read_text(encoding="utf-8").splitlines()
+    assert "# ? gedeiliche => gedeihliche   (1×)" in lines
+    assert "# ? Rechtsstaatbahn => Rechtsstaat Bahn   (1×)" in lines
+    # Already a model suggestion, already in the glossary: not listed twice
+    assert not any("Grotzki => Grotzky   (" in line for line in lines)
+    assert not any("Egon Bar =>" in line for line in lines)
+
+
+def test_drafted_entries_are_valid_glossary(raw_file, monkeypatch, tmp_path):
+    monkeypatch.setattr(gv.urllib.request, "urlopen", fake_model(ENTRIES[:2]))
+    out = tmp_path / "entwurf.txt"
+    run_main(monkeypatch, "-i", raw_file, "-o", out)
+    terms, replacements = rk.read_glossary(out)
+    assert terms == []
+    assert replacements == [("Grotzki", "Grotzky")]
+
+
+def test_unreadable_model_answer(raw_file, monkeypatch, tmp_path, capsys):
+    def urlopen(request, timeout=None):
+        return io.BytesIO(json.dumps({"message": {"content": "kein JSON"}}).encode())
+
+    monkeypatch.setattr(gv.urllib.request, "urlopen", urlopen)
+    run_main(monkeypatch, "-i", raw_file, "-o", tmp_path / "e.txt")
+    assert "unlesbar" in capsys.readouterr().out
+
+
+def test_cli_missing_file(monkeypatch, capsys):
+    with pytest.raises(SystemExit):
+        run_main(monkeypatch, "-i", "fehlt.txt")
+    assert "Datei nicht gefunden" in capsys.readouterr().err