lyrics aggregate block in digest.md

sort_ym/digest.py

@@ -9,11 +9,14 @@ from . import classify, genres, report
 
 DIGEST_FILE = "digest.md"
 DIGEST_SOURCE_FILE = "digest_source.md"
+LYRICS_DIGEST_FILE = "lyrics_digest.md"
 
 DECADE_UNKNOWN = "unknown"
 DECADE_MIN_YEAR = 1900
 DECADE_MAX_YEAR = 2100
 DECADE_TOP_GENRES = 3  # константа вёрстки (сколько жанров показывать на десятилетие), не config-knob
+LYRICS_TOP_MOODS = 6
+LYRICS_TOP_THEMES = 8
 
 
 @dataclass(frozen=True)
@@ -257,6 +260,60 @@ def _pct(n: int, total: int) -> int:
     return round(100 * n / total) if total else 0
 
 
+def analyzed_entries(analysis: dict[str, dict]) -> list[dict]:
+    """Только успешные записи - таймауты и битые ответы хранятся в том же кэше с полем error."""
+    return [e for e in analysis.values() if "error" not in e]
+
+
+def _counted(counter: Counter, key: str, limit: int | None = None) -> list[dict]:
+    items = sorted(counter.items(), key=lambda kv: (-kv[1], kv[0]))
+    if limit is not None:
+        items = items[:limit]
+    return [{key: name, "tracks": n} for name, n in items]
+
+
+def mood_stats(analysis: dict[str, dict]) -> list[dict]:
+    counter: Counter[str] = Counter()
+    for entry in analyzed_entries(analysis):
+        counter.update(entry.get("mood", []))
+    return _counted(counter, "mood")
+
+
+def theme_stats(analysis: dict[str, dict], limit: int) -> list[dict]:
+    counter: Counter[str] = Counter()
+    for entry in analyzed_entries(analysis):
+        counter.update(entry.get("themes", []))
+    return _counted(counter, "theme", limit)
+
+
+def lyrics_block(analysis: dict[str, dict]) -> list[str]:
+    """Компактный агрегат разбора лирики для digest.md.
+
+    Сам разбор (266 нарративных блоков) живёт в out/lyrics_digest.md - digest.md существует
+    ради фиксированного потолка ~150 строк, и класть их сюда нельзя.
+    """
+    entries = analyzed_entries(analysis)
+    if not entries:
+        return []
+
+    moods = ", ".join(f"{m['mood']} {m['tracks']}" for m in mood_stats(analysis)[:LYRICS_TOP_MOODS])
+    themes = ", ".join(f"{t['theme']} {t['tracks']}" for t in theme_stats(analysis, LYRICS_TOP_THEMES))
+    stances = Counter(e["stance"] for e in entries if "stance" in e)
+    povs = Counter(e["pov"] for e in entries if "pov" in e)
+    stance_part = " / ".join(f"{name} {n}" for name, n in sorted(stances.items(), key=lambda kv: (-kv[1], kv[0])))
+    pov_part = " / ".join(f"{name} {n}" for name, n in sorted(povs.items(), key=lambda kv: (-kv[1], kv[0])))
+
+    return [
+        "## Лирика (RU-подмножество)",
+        f"Разбор текстов {len(entries)} RU-треков локальной моделью; подробнее — out/{LYRICS_DIGEST_FILE}.",
+        f"- Настроения: {moods}",
+        f"- Темы: {themes}",
+        f"- Тон: {stance_part}",
+        f"- Ракурс: {pov_part}",
+        "",
+    ]
+
+
 def render_digest(
     rows: list[dict],
     tracks_cache: dict[str, dict],
@@ -264,6 +321,7 @@ def render_digest(
     top_artists: int,
     top_albums: int,
     wording: Wording = DEFAULT_WORDING,
+    analysis: dict[str, dict] | None = None,
 ) -> str:
     total = len(rows)
     unique_artists = {name for t in tracks_cache.values() for name in t["artists"]}
@@ -320,6 +378,8 @@ def render_digest(
         lines.append(f"- {label}: {n} ({_pct(n, total)}%)")
     lines.append("")
 
+    lines.extend(lyrics_block(analysis or {}))
+
     lines.append("## Десятилетия")
     lines.append("В скобках — доля от библиотеки; далее — три самые частые крупные корзины десятилетия.")
     for d in decade_rows:

tests/test_digest.py

@@ -275,3 +275,58 @@ def test_digest_line_count_stays_bounded_on_large_library():
     text = digest.render_digest(rows, tracks_cache, {}, top_artists=40, top_albums=15)
 
     assert len(text.splitlines()) < 250
+
+
+ANALYSIS = {
+    "100": {"mood": ["melancholy", "longing"], "themes": ["city", "memory"], "stance": "sincere", "pov": "confessional"},
+    "200": {"mood": ["melancholy"], "themes": ["city"], "stance": "ironic", "pov": "narrative"},
+    "300": {"error": "TimeoutError: модель зависла"},
+}
+
+
+def test_mood_stats_counts_every_mood_of_a_track():
+    assert digest.mood_stats(ANALYSIS) == [
+        {"mood": "melancholy", "tracks": 2},
+        {"mood": "longing", "tracks": 1},
+    ]
+
+
+def test_theme_stats_limits_and_orders():
+    assert digest.theme_stats(ANALYSIS, limit=1) == [{"theme": "city", "tracks": 2}]
+
+
+def test_analyzed_entries_excludes_error_markers():
+    assert len(digest.analyzed_entries(ANALYSIS)) == 2
+
+
+def test_lyrics_block_is_compact():
+    lines = digest.lyrics_block(ANALYSIS)
+
+    assert lines[0] == "## Лирика (RU-подмножество)"
+    assert len(lines) <= 7, "агрегат в digest.md должен оставаться компактным"
+    assert any("melancholy 2" in line for line in lines)
+    assert any("city 2" in line for line in lines)
+
+
+def test_lyrics_block_empty_without_analysis():
+    assert digest.lyrics_block({}) == []
+    assert digest.lyrics_block({"300": {"error": "TimeoutError: ..."}}) == []
+
+
+def test_render_digest_without_analysis_has_no_lyrics_section():
+    # Обратная совместимость: без analysis дайджест прежний, как в test_default_wording_matches_likes_output.
+    tracks_cache = {"100:1000": _track(100, "Тоска", ["Артист"], "rock")}
+    rows = _rows(tracks_cache)
+
+    text = digest.render_digest(rows, tracks_cache, {}, top_artists=10, top_albums=10)
+
+    assert "## Лирика" not in text
+
+
+def test_render_digest_with_analysis_includes_lyrics_section():
+    tracks_cache = {"100:1000": _track(100, "Тоска", ["Артист"], "rock")}
+    rows = _rows(tracks_cache)
+
+    text = digest.render_digest(rows, tracks_cache, {}, top_artists=10, top_albums=10, analysis=ANALYSIS)
+
+    assert "## Лирика (RU-подмножество)" in text