nex_animation/backend/tests/test_subtitle_modes.py

210 lines
12 KiB
Python
Raw Permalink Blame History

This file contains ambiguous Unicode characters!

This file contains ambiguous Unicode characters that may be confused with others in your current locale. If your use case is intentional and legitimate, you can safely ignore this warning. Use the Escape button to highlight these characters.

"""字幕交付能力:能力感知选择 ffmpeg + 字幕方式(内嵌软字幕轨 / 烧录 / 外挂)折算与落地。
覆盖三件事:
1) 二进制解析:Homebrew 精简 bottle 缺 libass 时,必须自动改用带 libass 的完整版构建;
2) 字幕方式折算:请求 > 时间线 > 老版本 burn_subtitle,且能力不足时逐级降级不静默;
3) 真实成片:默认内嵌 mov_text 软字幕轨(语言标记 chi),both 模式同时烧进画面。
"""
from __future__ import annotations
from pathlib import Path
import pytest
from app.services import compose
from app.services import ffmpeg as ff
from tests.conftest import absolute
from tests.test_delivery import _prepare
# ---------------------------------------------------------------- 二进制与能力
def test_ffmpeg_binary_resolution_prefers_libass():
"""PATH 上的构建缺 libass 时,必须挑到候选目录里带 libass 的那个(或明确降级可用)。"""
assert ff.available(), "本机应至少有一个可用的 ffmpeg/ffprobe"
assert ff.FFMPEG and Path(ff.FFMPEG).is_file()
caps = ff.capabilities()
for key in ("subtitle_burn", "subtitle_embed", "subtitle_codec", "libass", "ffmpeg_path",
"ffmpeg_version", "drawtext"):
assert key in caps, f"capabilities 缺少字段 {key}"
assert caps["ffmpeg_path"] == ff.FFMPEG
# 内嵌字幕只依赖 mov_text(几乎所有构建都有),必须可用
assert ff.can_embed_subtitles(".mp4") is True and ff.subtitle_codec(".mp4") == "mov_text"
assert ff.can_embed_subtitles(".webm") == ("webvtt" in ff.list_encoders())
if ff.can_embed_subtitles(".mkv"):
assert "mov_text" in {ff.subtitle_codec(x) for x in (".mp4", ".mov", ".mkv")}
# 选中的二进制自身能力必须与 capabilities 口径一致
assert caps["subtitle_burn"] == bool(ff.subtitle_filter())
assert (ff.subtitle_filter() in ff.list_filters()) or not ff.subtitle_filter()
def test_drawtext_available_on_resolved_binary():
"""中文字幕样式(drawtext)在解析出的二进制上必须可用,否则 ASS 烧录会缺字体渲染。"""
if not ff.can_draw_text():
pytest.skip("本机所有 ffmpeg 构建都没有 drawtext")
assert ff.find_font(), "需要系统中文字体才能烧录"
# ---------------------------------------------------------------- 烧录字体
_ASS_TMPL = """[Script Info]
ScriptType: v4.00+
PlayResX: 640
PlayResY: 200
ScaledBorderAndShadow: yes
[V4+ Styles]
Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding
Style: Default,{family},54,&H00FFFFFF,&H00FFFFFF,&H00101010,&H96000000,-1,0,0,0,100,100,0,0,1,3,1,2,20,20,40,1
[Events]
Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
Dialogue: 0,0:00:00.00,0:00:05.00,Default,,0,0,0,,深夜面馆,灯给你留着
"""
def _render_text(sample: str, family: str) -> str:
"""把一句话用指定字体烧到纯色画面上,返回原始像素(用于逐像素比对)。"""
ass = Path(sample)
ass.write_text(_ASS_TMPL.format(family=family), encoding="utf-8")
out = ass.with_suffix(".rgb")
ff.run([ff.FFMPEG, "-v", "error", "-y", "-f", "lavfi", "-i",
"color=c=0x22303a:s=640x200:d=1", "-vf", f"ass={ff.escape_filter_path(ass)}",
"-frames:v", "1", "-f", "rawvideo", "-pix_fmt", "rgb24", str(out)], timeout=120)
return out.read_bytes()
def test_ass_fontname_is_family_not_path():
"""ASS 的 Fontname 必须是字体**族名**:写成文件路径时 libass 找不到字体、
静默回退成无中文字形的默认字体,成片里就是一排豆腐块(本轮真实踩过)。"""
font = ff.subtitle_font()
assert font["family"], "本机没有可用于中文烧录的字体"
assert not Path(font["family"]).suffix, f"Fontname 不能是文件路径:{font['family']}"
catalog = {f for f, _ in ff.cjk_font_catalog()}
assert not catalog or font["family"] in catalog, "选中的字体不在 fontconfig 的中文字形清单里"
# 老配置里存的是字体文件路径,也必须折算成族名
if font["path"]:
again = ff.subtitle_font(font["path"])
assert again["family"] == font["family"] or not Path(again["family"]).suffix
def test_write_ass_header_uses_family(tmp_path):
cues = [{"start_ms": 0, "end_ms": 1000, "text": "深夜面馆,灯给你留着", "speaker": "林秀云"}]
out = ff.write_ass(cues, tmp_path / "s.ass", width=1280, height=720)
style = next(line for line in out.read_text(encoding="utf-8").splitlines()
if line.startswith("Style: Default"))
family = style.split(",")[1]
assert family and not family.lower().endswith((".ttf", ".ttc", ".otf"))
def test_fontconfig_charset_query_finds_cjk(tmp_path):
""":charset=4e2d 是我们判定「这个字体能不能画中文」的唯一依据,必须真有结果。"""
rows = ff.cjk_font_catalog()
assert rows, "fontconfig 查不到含中文字形的字体,烧录一定出豆腐块"
assert all(Path(p).exists() for _, p in rows[:5])
assert all(not f.startswith(".") for f, _ in rows), "内部名(.xxx Interface)画不出字"
def test_burn_render_draws_real_glyphs(tmp_path):
"""端到端证明「画出来的是汉字」:同一句中文分别用选中的中文字体、
一个不含中文的拉丁字体、一个不存在的字体名渲染。
若中文字体的像素和后两者之一完全相同,说明 libass 根本没用到中文字形。"""
if not ff.can_burn_subtitles() or not ff.subtitle_font()["family"]:
pytest.skip("本机不支持字幕烧录或缺少中文字体")
family = ff.subtitle_font()["family"]
cjk = _render_text(tmp_path / "a.ass", family)
latin = _render_text(tmp_path / "b.ass", "Verdana")
bogus = _render_text(tmp_path / "c.ass", "NoSuchFont 99")
assert cjk != latin, f"{family} 渲染结果与拉丁字体一致 → 中文没被真正渲染"
assert cjk != bogus, f"{family} 渲染结果与未知字体一致 → 中文没被真正渲染"
# ---------------------------------------------------------------- 字幕方式折算
def test_subtitle_mode_precedence():
conf_embed = {"subtitle_mode": "embed", "burn_subtitle": True}
assert compose.resolve_subtitle_mode(conf_embed)["mode"] == "embed"
# 请求显式 subtitle_mode 覆盖时间线
assert compose.resolve_subtitle_mode(conf_embed, subtitle_mode="burn")["mode"] == "burn"
# 老版本只有 burn_subtitle 布尔时保持原语义(向后兼容)
assert compose.resolve_subtitle_mode({"burn_subtitle": True})["mode"] == "burn"
assert compose.resolve_subtitle_mode({"burn_subtitle": False})["mode"] == "sidecar"
# 请求里的 burn_subtitle 优先于时间线 subtitle_mode
assert compose.resolve_subtitle_mode(conf_embed, burn_subtitle=True)["mode"] == "burn"
# 非法值回落到时间线老字段,坏输入不会让渲染报错
assert compose.resolve_subtitle_mode({"subtitle_mode": "????", "burn_subtitle": False})["mode"] == "sidecar"
def test_subtitle_mode_downgrades_by_capability(monkeypatch):
"""缺 libass 时 burn 自动降级为内嵌;两者都缺才降到外挂(并标记可报错)。"""
monkeypatch.setattr(ff, "can_burn_subtitles", lambda: False)
monkeypatch.setattr(ff, "can_embed_subtitles", lambda container=".mp4": True)
down = compose.resolve_subtitle_mode({}, burn_subtitle=True)
assert down["mode"] == "embed" and down["downgraded"] and down["explicit"]
assert down["burn"] is False and down["embed"] is True
monkeypatch.setattr(ff, "can_embed_subtitles", lambda container=".mp4": False)
last = compose.resolve_subtitle_mode({}, burn_subtitle=True)
assert last["mode"] == "sidecar" and last["downgraded"] and last["explicit"]
assert last["burn_available"] is False and last["embed_available"] is False
monkeypatch.setattr(ff, "can_burn_subtitles", lambda: True)
monkeypatch.setattr(ff, "can_embed_subtitles", lambda container=".mp4": False)
assert compose.resolve_subtitle_mode({"subtitle_mode": "both"})["mode"] == "burn"
monkeypatch.setattr(ff, "can_burn_subtitles", lambda: False)
monkeypatch.setattr(ff, "can_embed_subtitles", lambda container=".mp4": True)
assert compose.resolve_subtitle_mode({"subtitle_mode": "both"})["mode"] == "embed"
# 没有字幕句时不产生「降级」噪音
assert compose.resolve_subtitle_mode({}, has_subtitles=False)["downgraded"] is False
# ---------------------------------------------------------------- 真实成片
def test_compose_embeds_soft_subtitle_track_by_default(api, studio, script_text, make_image_png):
"""默认字幕方式 = 内嵌软字幕轨:成片必须带一条 mov_text 字幕轨,语言标记 chi,画面不糊字。"""
pid, _ = _prepare(api, studio, script_text, make_image_png)
api.post(f"/api/projects/{pid}/timeline", json={})
report = api.get(f"/api/projects/{pid}/timeline/report")["report"]
assert report["subtitle_mode"] == "embed" and report["subtitles"] > 0
result = api.post(f"/api/projects/{pid}/compose", json={"preset": "draft_720p"})["result"]
export = result["exports"][0]
assert export["status"] == "succeeded", export["error_message"]
# 内嵌字幕轨绝不能把成片截短(-shortest 与 SRT 流长度冲突的回归点)
total_ms = report["total_duration_ms"]
assert abs(export["duration_ms"] - total_ms) / total_ms < 0.06, export["error_message"]
assert "偏差" not in (export["error_message"] or "")
assert export["subtitle_burned"] is False # 内嵌模式不烧进画面
probed = ff.probe(absolute(export["path"]))
assert [t["codec_name"] for t in probed.subtitle_tracks] == ["mov_text"]
assert probed.subtitle_tracks[0]["language"] == "chi"
sidecar = Path(absolute(export["path"])).with_suffix(".srt")
assert sidecar.exists() and "-->" in sidecar.read_text(encoding="utf-8")
body = api.get(f"/api/exports/{export['id']}")["export"]
assert body["status"] == "succeeded"
def test_compose_both_mode_burns_and_embeds(api, studio, script_text, make_image_png):
"""both:字幕既烧进画面(硬字幕)又带一条可开关的软字幕轨。"""
if not ff.can_burn_subtitles():
pytest.skip("本机 ffmpeg 缺 libass,无法验证烧录")
pid, _ = _prepare(api, studio, script_text, make_image_png)
api.post(f"/api/projects/{pid}/timeline", json={"settings": {"subtitle_mode": "both"}})
result = api.post(f"/api/projects/{pid}/compose", json={"preset": "draft_720p"})["result"]
export = result["exports"][0]
assert export["status"] == "succeeded", export["error_message"]
assert export["subtitle_burned"] is True
probed = ff.probe(absolute(export["path"]))
assert len(probed.subtitle_tracks) == 1 and probed.subtitle_tracks[0]["codec_name"] == "mov_text"
actions = [e["action"] for e in api.get(f"/api/projects/{pid}/events")["events"]]
assert "compose.subtitle_skipped" not in actions
def test_compose_off_mode_emits_no_subtitle_files(api, studio, script_text, make_image_png):
"""off:成片无字幕轨,也不再产出外挂 SRT(交付纯净版)。"""
pid, _ = _prepare(api, studio, script_text, make_image_png)
api.post(f"/api/projects/{pid}/timeline", json={"settings": {"subtitle_mode": "off"}})
result = api.post(f"/api/projects/{pid}/compose", json={"preset": "draft_720p"})["result"]
export = result["exports"][0]
assert export["status"] == "succeeded", export["error_message"]
assert export["subtitle_burned"] is False
out = Path(absolute(export["path"]))
assert ff.probe(out).subtitle_tracks == []
assert not out.with_suffix(".srt").exists()