fix: enforce artwork-safe music ingestion

This commit is contained in:
2026-07-20 01:56:13 +08:00
parent 4c68e415ee
commit 484c9c0d24
6 changed files with 1223 additions and 78 deletions
+16
View File
@@ -40,6 +40,22 @@ cd frontend && npm ci && npm run dev
cd docker && docker compose up -d --build
```
## 一次性清理历史曲库
正常导入不会扫描或删除 `Library` 中的历史文件。升级后如需清理旧数据,先运行默认的预览模式:
```bash
./scripts/cleanup-library.sh --library /path/to/Library --dry-run
```
确认输出后再显式执行:
```bash
./scripts/cleanup-library.sh --library /path/to/Library --execute
```
脚本会优先把音频内嵌封面提取为专辑目录的 `cover.jpg/png`。只有目录和音频都没有封面时,才删除音频及同 basename 的 `.lrc``.cue``.json``.txt` 和图片 sidecar;不会删除公共 `cover.jpg/png` 或其他曲目文件。
## 技术栈
- 后端:Spring Boot 2.7 + Java 8 + Maven
@@ -153,6 +153,42 @@ public class AudioValidationService {
}
}
// ========== 嵌套内嵌封面结果类 ==========
/**
* 内嵌 attached-picture(封面)探测结果。
* <p>{@link #isPresent()} 为 true 表示源媒体含有 disposition.attached_pic 的图像流;
* {@link #getStreamIndex()} 为该流索引,{@link #getCodecName()} 为其编码名(如 mjpeg/png)。
* 仅识别内嵌 attached-picture,绝不从文件名推断或联网获取。</p>
*/
public static class EmbeddedArtwork {
private final boolean present;
private final int streamIndex;
private final String codecName;
EmbeddedArtwork(boolean present, int streamIndex, String codecName) {
this.present = present;
this.streamIndex = streamIndex;
this.codecName = codecName != null ? codecName : "";
}
public boolean isPresent() {
return present;
}
public int getStreamIndex() {
return streamIndex;
}
public String getCodecName() {
return codecName;
}
static EmbeddedArtwork absent() {
return new EmbeddedArtwork(false, -1, "");
}
}
// ========== 嵌套进程运行结果 ==========
/**
@@ -335,6 +371,39 @@ public class AudioValidationService {
}
}
/**
* 探测文件第一个音频流的编码名(如 {@code aac}/{@code mp3}/{@code flac}/{@code vorbis}/
* {@code opus}/{@code alac}/{@code wmav2})。用于后备 remux 时选择编码兼容的容器,
* 避免把 AAC 之类误封进 ogg 等不兼容容器。无法识别时返回空字符串。
*/
public String probeAudioCodec(Path media) {
int timeout = resolveProbeTimeout();
ProcessResult result;
try {
result = runProcess(getMetadataCommandArgs(media), timeout, MAX_METADATA_JSON_LENGTH);
} catch (IOException e) {
log.warn("FFprobe 编码探测执行失败: {} - {}", media, e.getMessage());
return "";
}
if (result.timedOut || result.exitCode != 0) {
return "";
}
try {
JsonNode root = JSON_MAPPER.readTree(result.output);
JsonNode streams = root.path("streams");
if (streams.isArray()) {
for (JsonNode stream : streams) {
if ("audio".equals(stream.path("codec_type").asText())) {
return stream.path("codec_name").asText("").toLowerCase();
}
}
}
} catch (IOException e) {
log.warn("FFprobe 编码探测 JSON 解析失败: {} - {}", media, e.getMessage());
}
return "";
}
/**
* 判断 FFprobe JSON 中是否存在音频流。
*/
@@ -368,6 +437,136 @@ public class AudioValidationService {
}
}
// ========== 内嵌封面探测与提取(FFprobe / FFmpeg ==========
/** 可直接以 {@code -c copy} 原样导出的图像编码 → 目标扩展名 */
private static final Map<String, String> COPYABLE_IMAGE_CODECS;
static {
Map<String, String> m = new LinkedHashMap<>();
m.put("mjpeg", "jpg");
m.put("jpeg", "jpg");
m.put("png", "png");
COPYABLE_IMAGE_CODECS = m;
}
/**
* 探测源媒体是否含有内嵌 attached-picture(封面)图像流。
* <p>基于 FFprobe 结构化 JSON 的 {@code disposition.attached_pic == 1} 判定,
* 仅识别真正的内嵌封面,普通视频流不计入。</p>
*/
public EmbeddedArtwork probeAttachedPicture(Path media) {
int timeout = resolveProbeTimeout();
ProcessResult result;
try {
result = runProcess(getMetadataCommandArgs(media), timeout, MAX_METADATA_JSON_LENGTH);
} catch (IOException e) {
log.warn("FFprobe 封面探测执行失败: {} - {}", media, e.getMessage());
return EmbeddedArtwork.absent();
}
if (result.timedOut || result.exitCode != 0) {
return EmbeddedArtwork.absent();
}
try {
JsonNode root = JSON_MAPPER.readTree(result.output);
JsonNode streams = root.path("streams");
if (streams.isArray()) {
for (JsonNode stream : streams) {
JsonNode disp = stream.path("disposition").path("attached_pic");
if (disp.asInt(0) == 1) {
int idx = stream.path("index").asInt(-1);
String codec = stream.path("codec_name").asText("");
return new EmbeddedArtwork(true, idx, codec);
}
}
}
} catch (IOException e) {
log.warn("FFprobe 封面 JSON 解析失败: {} - {}", media, e.getMessage());
}
return EmbeddedArtwork.absent();
}
/**
* 返回从内嵌封面流导出到 {@code output} 的 FFmpeg 命令参数列表(用于测试观察)。
* <p>可直接复制的编码(mjpeg/png)使用 {@code -c copy} 保留原始质量;
* 其他可解码的图像格式则重编码为 JPEG(由 {@code output} 扩展名决定)。</p>
*/
List<String> getCoverExtractCommandArgs(Path media, int streamIndex, boolean copyStream, Path output) {
List<String> cmd = new ArrayList<>();
cmd.add(getFfmpegCommand());
cmd.add("-y");
cmd.add("-i");
cmd.add(media.toAbsolutePath().toString());
cmd.add("-map");
cmd.add("0:" + streamIndex);
if (copyStream) {
cmd.add("-c");
cmd.add("copy");
}
cmd.add("-frames:v");
cmd.add("1");
cmd.add(output.toAbsolutePath().toString());
return cmd;
}
/**
* 将内嵌 attached-picture 封面提取到指定目录,返回写入的封面文件路径。
* <p>JPEG/PNG 原样保留({@code -c copy});其他 FFmpeg 可解码的图像格式规范化为
* {@code cover.jpg}。绝不联网或从文件名推断。</p>
*
* @param media 源媒体
* @param art {@link #probeAttachedPicture} 的探测结果(必须 present
* @param coverTarget 目标封面文件路径(扩展名应与保留/规范化策略一致,调用方决定去重命名)
* @return true 表示成功写入非空封面文件
*/
public boolean extractCoverTo(Path media, EmbeddedArtwork art, Path coverTarget) {
if (art == null || !art.isPresent() || art.getStreamIndex() < 0) {
return false;
}
String codec = art.getCodecName().toLowerCase();
boolean copyable = COPYABLE_IMAGE_CODECS.containsKey(codec);
int timeout = resolveDecodeTimeout();
List<String> cmd = getCoverExtractCommandArgs(media, art.getStreamIndex(), copyable, coverTarget);
try {
ProcessResult result = runProcess(cmd, timeout);
if (result.timedOut || result.exitCode != 0) {
String diag = capDiagnostic(result.output);
log.warn("封面提取失败: {} (exit={}) diag={}", media, result.exitCode, diag);
deleteQuietly(coverTarget);
return false;
}
} catch (IOException e) {
log.warn("封面提取执行失败: {} - {}", media, e.getMessage());
deleteQuietly(coverTarget);
return false;
}
try {
if (!java.nio.file.Files.exists(coverTarget) || java.nio.file.Files.size(coverTarget) == 0) {
deleteQuietly(coverTarget);
return false;
}
} catch (IOException e) {
return false;
}
return true;
}
/**
* 依据封面图像编码返回目标扩展名:mjpeg/jpeg→jpgpng→png,其余规范化为 jpg。
*/
public String coverExtensionFor(EmbeddedArtwork art) {
if (art == null) return "jpg";
String codec = art.getCodecName().toLowerCase();
return COPYABLE_IMAGE_CODECS.getOrDefault(codec, "jpg");
}
private static void deleteQuietly(Path p) {
if (p == null) return;
try {
java.nio.file.Files.deleteIfExists(p);
} catch (IOException ignored) {
}
}
// ========== 解码(FFmpeg,严格模式) ==========
/**
@@ -192,6 +192,10 @@ public class IngestService {
}
}
// 历史 Library 清理由 scripts/cleanup-library.sh 显式执行一次。
// 正常入库不执行破坏性全量扫描,避免每批导入反复删除历史数据。
int cleanupRemoved = 0;
// 扫描 Library 中已有的文件作为重复检测参考(t2s 归一化)
Set<IdentityKey> libraryIdentities = scanLibraryIdentities(libraryPath);
log.info("Library 中已有 {} 个曲目用于重复检测", libraryIdentities.size());
@@ -206,6 +210,7 @@ public class IngestService {
AtomicInteger unreadable = new AtomicInteger(0);
AtomicInteger convFailed = new AtomicInteger(0);
AtomicInteger otherRejected = new AtomicInteger(0);
AtomicInteger missingCover = new AtomicInteger(0);
AtomicInteger processed = new AtomicInteger(0);
// 批内重复检测集合
@@ -223,7 +228,7 @@ public class IngestService {
outcome = processSingleFile(srcFile, libraryPath, rejectedPath,
libraryIdentities, batchIdentities,
ingested, duplicates, missingMeta, unreadable,
convFailed, otherRejected);
convFailed, otherRejected, missingCover);
} catch (Exception e) {
otherRejected.incrementAndGet();
moveToRejected(srcFile, rejectedPath, "Other", fileName);
@@ -242,15 +247,16 @@ public class IngestService {
// 写入结构化报告
writeReport(taskId, fileOutcomes, rejectedPath, ingested.get(), duplicates.get(),
missingMeta.get(), unreadable.get(), convFailed.get(), otherRejected.get());
missingMeta.get(), unreadable.get(), convFailed.get(), otherRejected.get(),
missingCover.get(), cleanupRemoved);
// 完成
sendProgress(taskId, total, processed.get(), ingested.get(), duplicates.get(),
missingMeta.get(), unreadable.get(), convFailed.get(),
otherRejected.get(), null,
String.format("导入完成!成功: %d, 重复: %d, 缺元数据: %d, 不可读: %d, 转码失败: %d, 其他: %d",
ingested.get(), duplicates.get(), missingMeta.get(),
unreadable.get(), convFailed.get(), otherRejected.get()),
String.format("导入完成!成功: %d, 重复: %d, 缺元数据: %d, 缺封面: %d, 不可读: %d, 转码失败: %d, 其他: %d, 清理: %d",
ingested.get(), duplicates.get(), missingMeta.get(), missingCover.get(),
unreadable.get(), convFailed.get(), otherRejected.get(), cleanupRemoved),
true);
} catch (Exception e) {
@@ -306,7 +312,8 @@ public class IngestService {
Set<IdentityKey> batchIdentities,
AtomicInteger ingested, AtomicInteger duplicates,
AtomicInteger missingMeta, AtomicInteger unreadable,
AtomicInteger convFailed, AtomicInteger otherRejected) throws Exception {
AtomicInteger convFailed, AtomicInteger otherRejected,
AtomicInteger missingCover) throws Exception {
String fileName = srcFile.getFileName().toString();
String baseName = getBaseName(fileName);
@@ -380,6 +387,9 @@ public class IngestService {
// 3. 繁简转换文本标签并写回源文件(仅 jaudiotagger 路径;
// FFprobe 后备路径稍后用 FFmpeg 无损 remux 写入简体标签)
// 若 commit 失败但已读到完整元数据,则切换到 FFmpeg remux 恢复路径(对受支持容器),
// 避免把合法音频误判为 Other。
boolean commitFailedNeedsRemux = false;
if (jaudioOk) {
boolean tagsModified = false;
for (FieldKey key : TEXT_FIELDS) {
@@ -396,10 +406,17 @@ public class IngestService {
try {
audioFile.commit();
} catch (Exception e) {
otherRejected.incrementAndGet();
moveToRejected(srcFile, rejectedPath, "Other", fileName);
log.warn("标签写入失败且无法恢复: {} - {}", fileName, e.getMessage());
return "rejected:tag-write-failed";
// 已读到完整元数据;对可 remux 的容器改走 FFmpeg 无损恢复路径。
if (audioValidationService != null && isFallbackEligible(srcFile)) {
commitFailedNeedsRemux = true;
tag = null; // 不再依赖 jaudiotagger 标签对象,改用 remux 后 FFprobe 重读
log.info("标签 commit 失败,改用 FFmpeg remux 恢复: {} - {}", fileName, e.getMessage());
} else {
otherRejected.incrementAndGet();
moveToRejected(srcFile, rejectedPath, "Other", fileName);
log.warn("标签写入失败且无法恢复: {} - {}", fileName, e.getMessage());
return "rejected:tag-write-failed";
}
}
}
}
@@ -430,6 +447,7 @@ public class IngestService {
// 5. 格式转换 / 后备 remux(如需要)
Path effectiveFile = srcFile;
boolean needsConversion = isLosslessFormat(srcFile);
String remuxExt = null; // 后备 remux 时的输出容器扩展名(决定入库文件扩展名)
if (needsConversion) {
try {
Path flacFile = convertToFlac(srcFile, srcFile.getParent());
@@ -440,18 +458,21 @@ public class IngestService {
log.warn("转码失败: {} - {}", fileName, e.getMessage());
return "rejected:conversion-failed";
}
} else if (usedFallback) {
// jaudiotagger 无法写回此容器的标签:用 FFmpeg 无损 remux-c copy
} else if (usedFallback || commitFailedNeedsRemux) {
// jaudiotagger 无法解析或无法写回此容器的标签:用 FFmpeg 无损 remux-c copy
// 生成保留音频质量与原有元数据的 Navidrome 兼容输出,并写入简体 Title/Artist/Album。
// 依据探测到的音频编码选择兼容容器,避免把 AAC 等误封进不兼容容器。
String simpTitle = traditionalFilterService.toSimplified(title);
String simpArtist = traditionalFilterService.toSimplified(artist);
String simpAlbum = traditionalFilterService.toSimplified(album);
String simpAlbumArtist = albumArtist.isEmpty()
? "" : traditionalFilterService.toSimplified(albumArtist);
remuxExt = remuxContainerExt(srcFile);
try {
Path remuxed = remuxWithSimplifiedTags(srcFile, simpTitle, simpArtist,
simpAlbum, simpAlbumArtist);
simpAlbum, simpAlbumArtist, remuxExt);
effectiveFile = remuxed;
usedFallback = true; // 后续按 FFprobe 重读元数据 + 从源提取封面处理
} catch (Exception e) {
convFailed.incrementAndGet();
moveToRejected(srcFile, rejectedPath, "ConversionFailed", fileName);
@@ -544,12 +565,12 @@ public class IngestService {
}
}
// 确定最终格式与文件名(后备 remux 统一输出 m4a 容器
// 确定最终格式与文件名(后备 remux 使用编码兼容的容器扩展名
String ext;
if (needsConversion) {
ext = "flac";
} else if (usedFallback) {
ext = "m4a";
} else if (remuxExt != null) {
ext = remuxExt;
} else {
ext = getExtension(fileName);
if (ext == null) ext = "flac";
@@ -573,7 +594,26 @@ public class IngestService {
String destFileName = trackStr + " - " + safeTitle + "." + ext;
Path targetFile = resolveUniqueFile(targetDir, destFileName);
// 8. 移动/复制到目标位置
// 封面是入库硬性不变量:album 目录已有 cover.jpg/png,或能从源媒体提取内嵌封面,
// 二者皆无则拒绝为 MissingCover。必须在源文件删除前从源提取
// (后备 remux 使用 -map 0:a:0 会丢弃 attached picture,故须从原始 srcFile 提取)。
boolean albumAlreadyCovered = hasExistingCover(targetDir);
Path writtenCover = null;
if (!albumAlreadyCovered) {
writtenCover = acquireCover(srcFile, tag, targetDir);
if (writtenCover == null) {
// 无既有封面且无法提取内嵌封面 → MissingCover;清理派生临时文件,隔离源文件
if (derivedFile) {
deleteIfExists(effectiveFile);
}
missingCover.incrementAndGet();
moveToRejected(srcFile, rejectedPath, "MissingCover", fileName);
log.info("缺少封面(无既有 cover 且无内嵌封面): {}", fileName);
return "rejected:missing-cover";
}
}
// 移动/复制到目标位置
try {
FileTransferUtils.moveWithFallback(effectiveFile, targetFile);
} catch (IOException e) {
@@ -582,34 +622,128 @@ public class IngestService {
if (derivedFile) {
deleteIfExists(effectiveFile);
}
// 本曲目刚写入的封面(原本无既有封面)也需清理,避免遗留孤立 cover
if (writtenCover != null) {
deleteIfExists(writtenCover);
}
moveToRejected(srcFile, rejectedPath, "Other", fileName);
log.warn("移动文件到 Library 失败: {} - {}", fileName, e.getMessage());
return "rejected:move-failed";
}
// 9. 如果生成了派生文件(转换或后备 remux)且成功,删除原文件
// 如果生成了派生文件(转换或后备 remux)且成功,删除原文件
if (derivedFile) {
deleteIfExists(srcFile);
}
// 10. 更新身份标识集合
// 更新身份标识集合
libraryIdentities.add(identity);
batchIdentities.add(identity);
ingested.incrementAndGet();
// 11. 处理关联 LRC 文件(可选)
// 处理关联 LRC 文件(可选)
handleAssociatedLrc(srcFile, targetDir, baseName, destFileName);
// 12. 提取嵌入式歌词与封面(可选,仅 jaudiotagger 路径有 tag 对象
// 后备 remux 已保留原始嵌入式封面/歌词于输出容器中)
// 提取嵌入式歌词(可选,仅 jaudiotagger 路径有 tag 对象
if (tag != null) {
extractEmbeddedLyrics(tag, targetDir, trackStr, safeTitle, title, effectiveArtist);
extractCover(tag, targetDir);
}
return "ingested";
}
// ========== 封面处理 ==========
/**
* album 目录是否已存在有效封面文件(cover.jpg/png,大小写不敏感)。
*/
private boolean hasExistingCover(Path albumDir) {
return findExistingCover(albumDir) != null;
}
/**
* 返回 album 目录中已存在的封面文件(cover.jpg/jpeg/png,大小写不敏感),无则 null。
*/
private Path findExistingCover(Path albumDir) {
if (albumDir == null || !Files.isDirectory(albumDir)) return null;
try (DirectoryStream<Path> ds = Files.newDirectoryStream(albumDir)) {
for (Path p : ds) {
if (!Files.isRegularFile(p)) continue;
String n = p.getFileName().toString().toLowerCase();
if (n.equals("cover.jpg") || n.equals("cover.jpeg") || n.equals("cover.png")) {
try {
if (Files.size(p) > 0) return p;
} catch (IOException ignored) {
}
}
}
} catch (IOException e) {
log.debug("扫描既有封面失败: {} - {}", albumDir, e.getMessage());
}
return null;
}
/**
* 从源媒体提取内嵌封面写入 album 目录,返回写入的封面路径;无内嵌封面或提取失败返回 null。
* <p>优先使用 jaudiotagger 已解析的 artwork 字节(保留 JPEG/PNG 原样);不可用或为
* 不支持的图像格式时,改用 FFmpeg 从源的 attached-picture 流提取/规范化为 cover.jpg。
* 绝不联网或从文件名推断。</p>
*
* @param srcFile 原始源媒体(须在删除前调用,remux 输出不含封面)
* @param tag jaudiotagger 标签对象(可为 null
* @param albumDir 目标 album 目录
*/
private Path acquireCover(Path srcFile, Tag tag, Path albumDir) {
// 1. jaudiotagger 已解析的内嵌封面字节(JPEG/PNG 直接保留)
if (tag != null) {
try {
List<Artwork> artworks = tag.getArtworkList();
if (artworks != null && !artworks.isEmpty()) {
Artwork artwork = artworks.get(0);
byte[] data = artwork.getBinaryData();
if (data != null && data.length > 0) {
String mime = artwork.getMimeType();
String coverExt = null;
if (mime != null) {
String m = mime.toLowerCase();
if (m.contains("png")) coverExt = "png";
else if (m.contains("jpeg") || m.contains("jpg")) coverExt = "jpg";
}
if (coverExt != null) {
Path coverFile = resolveUniqueFile(albumDir, "cover." + coverExt);
Files.write(coverFile, data);
log.info("已从标签提取封面到: {}", coverFile);
return coverFile;
}
// 不支持的图像格式:交由 FFmpeg 规范化(下面走 ffprobe/ffmpeg 路径)
}
}
} catch (Exception e) {
log.debug("jaudiotagger 封面提取失败,尝试 FFmpeg: {} - {}", srcFile, e.getMessage());
}
}
// 2. FFmpeg:探测源的 attached-picture 流并提取(保留 JPEG/PNG,其他规范化为 jpg
if (audioValidationService != null) {
AudioValidationService.EmbeddedArtwork art =
audioValidationService.probeAttachedPicture(srcFile);
if (art.isPresent()) {
String coverExt = audioValidationService.coverExtensionFor(art);
try {
Path coverFile = resolveUniqueFile(albumDir, "cover." + coverExt);
if (audioValidationService.extractCoverTo(srcFile, art, coverFile)) {
log.info("已从源媒体提取内嵌封面到: {}", coverFile);
return coverFile;
}
deleteIfExists(coverFile);
} catch (IOException e) {
log.debug("FFmpeg 封面提取失败: {} - {}", srcFile, e.getMessage());
}
}
}
return null;
}
// ========== 关联文件处理 ==========
/**
@@ -658,33 +792,6 @@ public class IngestService {
}
}
/**
* 从标签中提取封面(可选),写入 cover.jpg
*/
private void extractCover(Tag tag, Path targetDir) {
try {
List<Artwork> artworks = tag.getArtworkList();
if (artworks == null || artworks.isEmpty()) return;
Artwork artwork = artworks.get(0);
byte[] data = artwork.getBinaryData();
if (data == null || data.length == 0) return;
String mime = artwork.getMimeType();
String coverExt = "jpg";
if (mime != null) {
if (mime.contains("png")) coverExt = "png";
else if (mime.contains("gif")) coverExt = "gif";
}
Path coverFile = resolveUniqueFile(targetDir, "cover." + coverExt);
Files.write(coverFile, data);
log.info("已提取封面到: {}", coverFile);
} catch (Exception e) {
log.debug("提取封面失败(可选,忽略): {}", e.getMessage());
}
}
// ========== 结构化报告 ==========
/**
@@ -693,7 +800,8 @@ public class IngestService {
private void writeReport(String taskId, LinkedHashMap<String, String> fileOutcomes,
Path rejectedPath, int ingestedCount, int duplicateCount,
int missingMetaCount, int unreadableCount,
int convFailedCount, int otherRejectedCount) {
int convFailedCount, int otherRejectedCount,
int missingCoverCount, int cleanupRemovedCount) {
try {
Path reportsDir = rejectedPath.resolve(REPORT_SUBDIR);
Files.createDirectories(reportsDir);
@@ -713,7 +821,10 @@ public class IngestService {
json.append(" \"missingMetadata\": ").append(missingMetaCount).append(",\n");
json.append(" \"unreadable\": ").append(unreadableCount).append(",\n");
json.append(" \"conversionFailed\": ").append(convFailedCount).append(",\n");
json.append(" \"otherRejected\": ").append(otherRejectedCount).append("\n");
json.append(" \"otherRejected\": ").append(otherRejectedCount).append(",\n");
// 新增字段:追加在已有字段之后,保持既有消费者向后兼容
json.append(" \"missingCover\": ").append(missingCoverCount).append(",\n");
json.append(" \"libraryCleanupRemoved\": ").append(cleanupRemovedCount).append("\n");
json.append(" },\n");
json.append(" \"files\": [\n");
@@ -833,18 +944,41 @@ public class IngestService {
}
}
/** FFprobe 后备读取仅对这些容器格式启用(jaudiotagger 常拒绝的合法 AAC/M4A/MP4 */
private static final Set<String> FALLBACK_ELIGIBLE_EXTENSIONS = new HashSet<>(Arrays.asList(
"m4a", "mp4", "aac", "m4b", "m4p"
));
/**
* 判断文件是否适用 FFprobe 元数据后备读取路径。
* 仅限受支持的 M4A/MP4/AAC 容器,避免对其他格式做无意义的后备。
* 判断文件是否适用 FFprobe 元数据后备/恢复路径。
* <p>凡是受支持的音频类型(已通过 {@link #isAudioFile} 扫描)且启用了校验服务,
* 均允许在 jaudiotagger 无法解析/写回时改用 FFprobe 元数据分类并 remux 恢复,
* 使支持的可读文件由 FFprobe 元数据决定归类,而非一律 Unreadable。</p>
*/
private boolean isFallbackEligible(Path file) {
String ext = getExtension(file.getFileName().toString());
return ext != null && FALLBACK_ELIGIBLE_EXTENSIONS.contains(ext);
return audioValidationService != null && isAudioFile(file);
}
/** 音频编码 → 后备 remux 输出容器扩展名(选择编码兼容的容器,避免不兼容封装) */
private static final Map<String, String> CODEC_TO_CONTAINER_EXT;
static {
Map<String, String> m = new HashMap<>();
m.put("aac", "m4a");
m.put("alac", "m4a");
m.put("mp3", "mp3");
m.put("flac", "flac");
m.put("vorbis", "ogg");
m.put("opus", "opus");
m.put("wmav1", "wma");
m.put("wmav2", "wma");
m.put("wmapro", "wma");
CODEC_TO_CONTAINER_EXT = m;
}
/**
* 依据探测到的音频编码选择后备 remux 的输出容器扩展名。
* 未知编码回退到 {@code m4a}(AAC 系最常见),保证仍产出可用的音乐资产。
*/
private String remuxContainerExt(Path srcFile) {
String codec = audioValidationService != null
? audioValidationService.probeAudioCodec(srcFile) : "";
String ext = CODEC_TO_CONTAINER_EXT.get(codec);
return ext != null ? ext : "m4a";
}
/**
@@ -863,7 +997,7 @@ public class IngestService {
cmd.add("-y");
cmd.add("-i");
cmd.add(input.toAbsolutePath().toString());
// 仅映射预期的第一个音频流,避免把普通视频/字幕/数据等流带入生成的 M4A
// 仅映射预期的第一个音频流,避免把普通视频/字幕/数据等流带入生成的输出
// 确保输出是干净的 Navidrome 兼容音乐资产。
cmd.add("-map");
cmd.add("0:a:0");
@@ -872,6 +1006,12 @@ public class IngestService {
// 保留原有 format 级元数据(不含未映射的流),随后覆盖简体标签
cmd.add("-map_metadata");
cmd.add("0");
// MP3 输出使用 ID3v2.3 以获得最广泛的兼容性
String outExt = getExtension(output.getFileName().toString());
if ("mp3".equals(outExt)) {
cmd.add("-id3v2_version");
cmd.add("3");
}
// 覆盖简体标签(不伪造缺失字段:调用方已确保三项必需字段非空)
cmd.add("-metadata");
cmd.add("title=" + title);
@@ -888,10 +1028,10 @@ public class IngestService {
}
private Path remuxWithSimplifiedTags(Path input, String title, String artist,
String album, String albumArtist)
String album, String albumArtist, String containerExt)
throws IOException, InterruptedException {
String baseName = getBaseName(input.getFileName().toString());
Path output = resolveUniqueFile(input.getParent(), baseName + ".fallback.m4a");
Path output = resolveUniqueFile(input.getParent(), baseName + ".fallback." + containerExt);
List<String> cmd = buildRemuxCommand(input, output, title, artist, album, albumArtist);
runFfmpegDrained(cmd, FFMPEG_CONVERT_TIMEOUT_SECONDS, output, "remux");
@@ -986,6 +1126,88 @@ public class IngestService {
}
}
/**
* 既有 Library 封面清理:扫描各 album 目录,删除既无 album 封面文件(cover.jpg/png)、
* 且自身也不含内嵌 attached-picture 封面的音频文件。
* <p>安全边界:仅删除 Library 内的音频文件;不删除封面、非音频侧车、任何 Rejected 文件;
* album 目录中一旦存在有效 cover.jpg/png,则该目录内音频全部保留。
* 操作确定性且幂等(相同输入多次运行结果一致)。</p>
*
* @return 被删除的音频文件数量
*/
int cleanupLibraryWithoutCover(Path libraryPath) {
if (libraryPath == null || !Files.isDirectory(libraryPath)) return 0;
// 收集所有含音频的目录(album 目录以其直接父目录音频为单位判定封面)
List<Path> audioFiles = new ArrayList<>();
try {
Files.walkFileTree(libraryPath, new SimpleFileVisitor<Path>() {
@Override
public FileVisitResult visitFile(Path file, BasicFileAttributes attrs) {
if (isAudioFile(file)) {
audioFiles.add(file);
}
return FileVisitResult.CONTINUE;
}
});
} catch (IOException e) {
log.warn("封面清理扫描失败: {}", e.getMessage());
return 0;
}
// 确定性:按路径排序;按 album 目录缓存“是否已有封面文件”避免重复扫描
audioFiles.sort(Comparator.comparing(Path::toString));
Map<Path, Boolean> dirHasCover = new HashMap<>();
int removed = 0;
for (Path audio : audioFiles) {
Path albumDir = audio.getParent();
if (albumDir == null) continue;
Boolean covered = dirHasCover.get(albumDir);
if (covered == null) {
covered = hasExistingCover(albumDir);
dirHasCover.put(albumDir, covered);
}
if (covered) {
continue; // 目录已有封面 → 保留其内所有音频
}
// 目录无封面文件:若该音频含内嵌封面,则提取为 album 目录的 cover.jpg/png
// 使 Navidrome 可发现封面;随后该目录标记为 covered,其余音频一并保留。
Path extracted = extractCoverFromAudio(audio, albumDir);
if (extracted != null) {
dirHasCover.put(albumDir, Boolean.TRUE);
log.info("封面清理:从内嵌封面为 album 目录生成 {}", extracted.getFileName());
continue;
}
// 既无目录封面、也无内嵌封面 → 删除该音频文件(仅音频,不动侧车/封面)
try {
if (Files.deleteIfExists(audio)) {
removed++;
log.info("封面清理:移除无封面音频 {}", audio.getFileName());
}
} catch (IOException e) {
log.warn("封面清理删除失败: {} - {}", audio.getFileName(), e.getMessage());
}
}
return removed;
}
/**
* 若音频含内嵌 attached-picture 封面,则提取到 album 目录并返回写入的 cover 路径;
* 无内嵌封面或提取失败返回 null。jaudiotagger 已解析的字节优先(保留 JPEG/PNG),
* 否则用 FFmpeg 从源提取/规范化。绝不联网或从文件名推断。
*/
private Path extractCoverFromAudio(Path audio, Path albumDir) {
Tag tag = null;
try {
AudioFile af = AudioFileIO.read(audio.toFile());
tag = af.getTag();
} catch (Exception ignored) {
// jaudiotagger 无法解析 → acquireCover 内部会走 FFmpeg 路径
}
return acquireCover(audio, tag, albumDir);
}
/**
* 扫描 Library 目录中已有的音频文件,构建身份标识集合
* (所有文本字段经过 t2s + lowercase 归一化,与 incoming 文件一致)
@@ -519,6 +519,106 @@ class AudioValidationServiceTest {
assertFalse(meta.isAvailable(), "ffprobe 失败应返回 unavailable");
}
// ========== 内嵌封面探测与提取 ==========
@Test
void probeAttachedPicture_detectsEmbeddedCover() throws Exception {
assumeFfmpegFfprobe();
Path tmpDir = Files.createTempDirectory("probe-cover-");
try {
Path m4a = createM4aWithCover(tmpDir, "withcover.m4a");
AudioValidationService.EmbeddedArtwork art = service.probeAttachedPicture(m4a);
assertTrue(art.isPresent(), "应检测到内嵌 attached-picture");
assertTrue(art.getStreamIndex() >= 0, "封面流索引应有效");
assertFalse(art.getCodecName().isEmpty(), "应报告封面编码名");
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void probeAttachedPicture_realVideoNotTreatedAsCover() throws Exception {
assumeFfmpegFfprobe();
Path tmpDir = Files.createTempDirectory("probe-cover-");
try {
// 普通视频流(非 attached_pic)不应被识别为封面
Path m4a = tmpDir.resolve("realvideo.m4a");
ProcessBuilder pb = new ProcessBuilder(
"ffmpeg", "-y",
"-f", "lavfi", "-i", "anullsrc=r=44100:cl=mono",
"-f", "lavfi", "-i", "testsrc=s=64x64:d=0.3",
"-map", "0:a", "-map", "1:v",
"-c:a", "aac", "-c:v", "libx264", "-shortest", "-t", "0.3",
"-f", "mp4", m4a.toAbsolutePath().toString());
pb.redirectErrorStream(true);
Process p = pb.start();
readAllBytes(p.getInputStream());
p.waitFor(15, TimeUnit.SECONDS);
AudioValidationService.EmbeddedArtwork art = service.probeAttachedPicture(m4a);
assertFalse(art.isPresent(), "普通视频流不应被当作封面");
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void extractCoverTo_writesCoverFile() throws Exception {
assumeFfmpegFfprobe();
Path tmpDir = Files.createTempDirectory("probe-cover-");
try {
Path m4a = createM4aWithCover(tmpDir, "withcover.m4a");
AudioValidationService.EmbeddedArtwork art = service.probeAttachedPicture(m4a);
assertTrue(art.isPresent());
String ext = service.coverExtensionFor(art);
Path cover = tmpDir.resolve("cover." + ext);
boolean ok = service.extractCoverTo(m4a, art, cover);
assertTrue(ok, "封面提取应成功");
assertTrue(Files.exists(cover) && Files.size(cover) > 0, "应写入非空封面文件");
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void getCoverExtractCommandArgs_copiesForJpegPng() {
Path media = Paths.get("/tmp/a.m4a");
Path out = Paths.get("/tmp/cover.png");
List<String> copyArgs = service.getCoverExtractCommandArgs(media, 1, true, out);
assertTrue(copyArgs.contains("copy"), "可复制编码应使用 -c copy");
assertTrue(copyArgs.contains("0:1"), "应按流索引映射");
List<String> reencodeArgs = service.getCoverExtractCommandArgs(media, 1, false, out);
assertFalse(reencodeArgs.contains("copy"), "重编码路径不应带 -c copy");
}
private static Path createM4aWithCover(Path dir, String name) throws Exception {
Path cover = dir.resolve(".cover_src.png");
runProcessLocal(new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "color=c=blue:s=48x48:d=0.1",
"-frames:v", "1", cover.toAbsolutePath().toString()));
Path out = dir.resolve(name);
runProcessLocal(new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "anullsrc=r=44100:cl=mono",
"-i", cover.toAbsolutePath().toString(),
"-map", "0:a", "-map", "1:v", "-disposition:v:0", "attached_pic",
"-c:a", "aac", "-c:v", "png", "-t", "0.3", "-f", "mp4",
"-metadata", "title=T", "-metadata", "artist=A", "-metadata", "album=Al",
out.toAbsolutePath().toString()));
Files.deleteIfExists(cover);
return out;
}
private static void runProcessLocal(ProcessBuilder pb) throws Exception {
pb.redirectErrorStream(true);
Process p = pb.start();
readAllBytes(p.getInputStream());
boolean ok = p.waitFor(15, TimeUnit.SECONDS);
if (!ok) { p.destroyForcibly(); throw new RuntimeException("ffmpeg 超时"); }
if (p.exitValue() != 0) throw new RuntimeException("ffmpeg 失败 exit=" + p.exitValue());
}
/**
* 使用 FFmpeg 创建一个带标签的有效短 M4A(AAC)文件。
*/
@@ -505,8 +505,16 @@ class IngestServiceE2ETest {
String simpAlbum = "\u98CE\u9A6C"; // 风马
Path src = inputDir.resolve("trad.mp3");
// 内嵌 PNG 封面,使 jaudiotagger artwork 路径可提取并满足入库不变量
Path coverSrc = inputDir.resolve(".trad.cover.png");
runProcess(new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "color=c=red:s=48x48:d=0.1",
"-frames:v", "1", coverSrc.toAbsolutePath().toString()));
ProcessBuilder pb = new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "anullsrc=r=44100:cl=mono",
"-i", coverSrc.toAbsolutePath().toString(),
"-map", "0:a", "-map", "1:v", "-disposition:v:0", "attached_pic",
"-c:v", "png", "-id3v2_version", "3",
"-metadata", "title=" + tradTitle,
"-metadata", "artist=" + tradArtist,
"-metadata", "album=" + tradAlbum,
@@ -514,6 +522,7 @@ class IngestServiceE2ETest {
src.toAbsolutePath().toString()
);
runProcess(pb);
Files.deleteIfExists(coverSrc);
assertTrue(Files.exists(src), "FFmpeg 应创建带标签的 MP3");
assertTrue(Files.size(src) > 2000, "MP3 文件应有足够的音频帧");
@@ -876,7 +885,7 @@ class IngestServiceE2ETest {
Files.createDirectories(rejDir);
Path src = createTaggedMp3(inputDir, "corrupt.mp3",
"Valid Title", "Valid Artist", "Valid Album");
"Valid Title", "Valid Artist", "Valid Album", false);
// 截断文件到极小尺寸以破坏音频数据,但保留文件头部的 ID3v2 标签
// 使 jaudiotagger 仍可读取标签,但 FFprobe/FFmpeg 无法处理
long truncatedSize = Math.min(Files.size(src), 512);
@@ -987,8 +996,9 @@ class IngestServiceE2ETest {
assertEquals("ingested", result1);
// 第二个相同内容的文件应为 duplicate,不应触发解码验证
// (无封面以保证 512 字节截断后 ID3 标签仍可读,从而进入去重判定)
Path src2 = createTaggedMp3(inputDir, "second.mp3",
"Dup Title", "Dup Artist", "Dup Album");
"Dup Title", "Dup Artist", "Dup Album", false);
// 截断第二个文件使其无法通过验证,但 duplicate 检测应提前阻止验证
long truncatedSize2 = Math.min(Files.size(src2), 512);
try (java.io.RandomAccessFile raf = new java.io.RandomAccessFile(src2.toFile(), "rw")) {
@@ -1089,6 +1099,11 @@ class IngestServiceE2ETest {
Path src = createTaggedLossless(inputDir, "song.wav",
"Flac Title", "Flac Artist", "Flac Album");
// WAV 容器无法内嵌封面:预置 album 目录的 cover.jpg,验证“既有封面”即满足不变量。
Path expectedDir = libDir.resolve("Flac Artist").resolve("Flac Album");
Files.createDirectories(expectedDir);
Files.write(expectedDir.resolve("cover.jpg"), new byte[]{(byte) 0xFF, (byte) 0xD8, (byte) 0xFF});
IngestService service = buildServiceWithValidation();
AtomicInteger ingested = new AtomicInteger();
AtomicInteger duplicates = new AtomicInteger();
@@ -1108,7 +1123,6 @@ class IngestServiceE2ETest {
assertEquals(0, convFailed.get());
// 文件应作为 FLAC 存入 Library/Flac Artist/Flac Album/01 - Flac Title.flac
Path expectedDir = libDir.resolve("Flac Artist").resolve("Flac Album");
assertTrue(Files.isDirectory(expectedDir), "Library 应包含 Artist/Album 目录");
boolean foundFlac = Files.list(expectedDir).anyMatch(p ->
@@ -1264,6 +1278,86 @@ class IngestServiceE2ETest {
}
}
// ========== Finding metadata-fallback-coverage:非 m4a/aac 扩展名也应走 FFprobe 分类 ==========
@Test
void mislabeledSupportedFile_classifiedByFfprobe_ingestedWithCompatibleContainer() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-fallback-");
try {
Path inputDir = tmpDir.resolve("Input");
Path libDir = tmpDir.resolve("Library");
Path rejDir = tmpDir.resolve("Rejected");
Files.createDirectories(inputDir);
Files.createDirectories(libDir);
Files.createDirectories(rejDir);
// MP4/AAC 内容但命名为 .ogg(受支持但此前不在 fallback-eligible 集合):
// 旧逻辑会直接判为 Unreadable;修复后应由 FFprobe 元数据分类并 remux 入库。
Path src = createMp4ContainerAsAac(inputDir, "mislabel.ogg",
"Mis Title", "Mis Artist", "Mis Album", true);
assertJaudiotaggerCannotRead(src);
IngestService service = buildServiceWithValidation();
AtomicInteger ingested = new AtomicInteger();
AtomicInteger unreadable = new AtomicInteger();
AtomicInteger missingMeta = new AtomicInteger();
String result = invokeProcessSingleFile(service, src, libDir, rejDir,
new HashSet<>(), new HashSet<>(),
ingested, new AtomicInteger(),
missingMeta, unreadable,
new AtomicInteger(), new AtomicInteger());
assertEquals("ingested", result,
"受支持但被 jaudiotagger 拒绝的文件应由 FFprobe 分类并入库,而非 Unreadable");
assertEquals(1, ingested.get());
assertEquals(0, unreadable.get());
// AAC 内容应封入编码兼容的 m4a 容器
Path albumDir = libDir.resolve("Mis Artist").resolve("Mis Album");
assertTrue(Files.list(albumDir).anyMatch(p ->
p.getFileName().toString().equals("01 - Mis Title.m4a")),
"AAC 内容应 remux 为编码兼容的 .m4a 容器");
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void mislabeledSupportedFile_missingTag_classifiedMissingMetadataNotUnreadable() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-fallback-");
try {
Path inputDir = tmpDir.resolve("Input");
Path libDir = tmpDir.resolve("Library");
Path rejDir = tmpDir.resolve("Rejected");
Files.createDirectories(inputDir);
Files.createDirectories(libDir);
Files.createDirectories(rejDir);
// MP4/AAC 内容命名为 .ogg,缺 album → FFprobe 可读但元数据不全,应为 MissingMetadata
Path src = createMp4ContainerAsAac(inputDir, "mislabel2.ogg",
"Only Title", "Only Artist", null, true);
assertJaudiotaggerCannotRead(src);
IngestService service = buildServiceWithValidation();
AtomicInteger unreadable = new AtomicInteger();
AtomicInteger missingMeta = new AtomicInteger();
String result = invokeProcessSingleFile(service, src, libDir, rejDir,
new HashSet<>(), new HashSet<>(),
new AtomicInteger(), new AtomicInteger(),
missingMeta, unreadable,
new AtomicInteger(), new AtomicInteger());
assertEquals("rejected:missing-metadata", result,
"FFprobe 可读但缺必填标签应为 MissingMetadata,而非 Unreadable");
assertEquals(1, missingMeta.get());
assertEquals(0, unreadable.get());
assertTrue(Files.exists(rejDir.resolve("MissingMetadata").resolve("mislabel2.ogg")));
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void jaudiotaggerUnreadableM4a_missingAlbumRejectedMissingMetadata() throws Exception {
assumeFfmpeg();
@@ -1577,6 +1671,344 @@ class IngestServiceE2ETest {
}
}
// ========== 6. 封面不变量:提取 / 缺失拒绝 / 清理 ==========
@Test
void embeddedArtworkExtractedAsCoverOnIngest() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-cover-");
try {
Path inputDir = tmpDir.resolve("Input");
Path libDir = tmpDir.resolve("Library");
Path rejDir = tmpDir.resolve("Rejected");
Files.createDirectories(inputDir);
Files.createDirectories(libDir);
Files.createDirectories(rejDir);
// 带内嵌 PNG 封面的 MP3
Path src = createTaggedMp3(inputDir, "song.mp3",
"Cov Title", "Cov Artist", "Cov Album", true);
IngestService service = buildServiceWithValidation();
String result = invokeProcessSingleFile(service, src, libDir, rejDir,
new HashSet<>(), new HashSet<>(),
new AtomicInteger(), new AtomicInteger(),
new AtomicInteger(), new AtomicInteger(),
new AtomicInteger(), new AtomicInteger());
assertEquals("ingested", result);
Path albumDir = libDir.resolve("Cov Artist").resolve("Cov Album");
boolean coverExists = Files.exists(albumDir.resolve("cover.jpg"))
|| Files.exists(albumDir.resolve("cover.png"));
assertTrue(coverExists, "内嵌封面应被提取为 album 目录下的 cover.jpg/png");
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void completeMetadataButNoCover_rejectedAsMissingCover() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-cover-");
try {
Path inputDir = tmpDir.resolve("Input");
Path libDir = tmpDir.resolve("Library");
Path rejDir = tmpDir.resolve("Rejected");
Files.createDirectories(inputDir);
Files.createDirectories(libDir);
Files.createDirectories(rejDir);
// 完整元数据但无内嵌封面,且 album 目录也无既有 cover
Path src = createTaggedMp3(inputDir, "nocover.mp3",
"NC Title", "NC Artist", "NC Album", false);
IngestService service = buildServiceWithValidation();
AtomicInteger missingCover = new AtomicInteger();
AtomicInteger ingested = new AtomicInteger();
String result = invokeProcessSingleFile(service, src, libDir, rejDir,
new HashSet<>(), new HashSet<>(),
ingested, new AtomicInteger(),
new AtomicInteger(), new AtomicInteger(),
new AtomicInteger(), new AtomicInteger(), missingCover);
assertEquals("rejected:missing-cover", result);
assertEquals(1, missingCover.get());
assertEquals(0, ingested.get());
assertTrue(Files.exists(rejDir.resolve("MissingCover").resolve("nocover.mp3")),
"无封面文件应进入 Rejected/MissingCover");
// Library 中不应有任何音频文件或遗留 cover(可能存在空目录,属正常)
try (Stream<Path> walk = Files.walk(libDir)) {
boolean hasFile = walk.filter(Files::isRegularFile).findAny().isPresent();
assertFalse(hasFile, "Library 中不应有任何音频或遗留封面文件");
}
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void noEmbeddedArtworkButAlbumHasCover_accepted() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-cover-");
try {
Path inputDir = tmpDir.resolve("Input");
Path libDir = tmpDir.resolve("Library");
Path rejDir = tmpDir.resolve("Rejected");
Files.createDirectories(inputDir);
Files.createDirectories(libDir);
Files.createDirectories(rejDir);
Path src = createTaggedMp3(inputDir, "song.mp3",
"AC Title", "AC Artist", "AC Album", false);
// 预置 album 封面
Path albumDir = libDir.resolve("AC Artist").resolve("AC Album");
Files.createDirectories(albumDir);
Files.write(albumDir.resolve("cover.png"), new byte[]{(byte) 0x89, 'P', 'N', 'G'});
IngestService service = buildServiceWithValidation();
AtomicInteger ingested = new AtomicInteger();
String result = invokeProcessSingleFile(service, src, libDir, rejDir,
new HashSet<>(), new HashSet<>(),
ingested, new AtomicInteger(),
new AtomicInteger(), new AtomicInteger(),
new AtomicInteger(), new AtomicInteger());
assertEquals("ingested", result, "album 已有封面时无内嵌封面也应入库");
assertEquals(1, ingested.get());
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void fallbackRemux_extractsEmbeddedCoverBeforeSourceDeletion() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-cover-");
try {
Path inputDir = tmpDir.resolve("Input");
Path libDir = tmpDir.resolve("Library");
Path rejDir = tmpDir.resolve("Rejected");
Files.createDirectories(inputDir);
Files.createDirectories(libDir);
Files.createDirectories(rejDir);
// .aac 扩展名的 M4A 容器(jaudiotagger 拒绝)+ 内嵌封面 → 走 remux 后备
Path src = createMp4ContainerAsAac(inputDir, "fb.aac",
"FB Title", "FB Artist", "FB Album", true);
assertJaudiotaggerCannotRead(src);
IngestService service = buildServiceWithValidation();
AtomicInteger ingested = new AtomicInteger();
String result = invokeProcessSingleFile(service, src, libDir, rejDir,
new HashSet<>(), new HashSet<>(),
ingested, new AtomicInteger(),
new AtomicInteger(), new AtomicInteger(),
new AtomicInteger(), new AtomicInteger());
assertEquals("ingested", result, "后备 remux 路径应在删除源前提取封面并入库");
assertEquals(1, ingested.get());
Path albumDir = libDir.resolve("FB Artist").resolve("FB Album");
boolean coverExists = Files.exists(albumDir.resolve("cover.jpg"))
|| Files.exists(albumDir.resolve("cover.png"));
assertTrue(coverExists, "封面应在源删除前从源提取到 album 目录");
assertFalse(Files.exists(src), "源文件应已删除");
} finally {
deleteDirectory(tmpDir);
}
}
// ========== 7. 既有 Library 封面清理:安全性与幂等 ==========
@Test
void cleanup_removesUncoveredAudio_preservesCoveredAndSidecars() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-cleanup-");
try {
Path libDir = tmpDir.resolve("Library");
// Album A:有 cover.jpg → 其内音频与侧车全部保留
Path albumA = libDir.resolve("Artist").resolve("Album A");
Files.createDirectories(albumA);
Files.write(albumA.resolve("cover.jpg"), new byte[]{(byte) 0xFF, (byte) 0xD8, (byte) 0xFF});
Path aAudio = createTaggedMp3(albumA, "01 - a.mp3", "a", "Artist", "Album A", false);
Path aLrc = albumA.resolve("01 - a.lrc");
Files.write(aLrc, "[00:00]x".getBytes(StandardCharsets.UTF_8));
// Album B:无 cover 文件,音频也无内嵌封面 → 应被删除
Path albumB = libDir.resolve("Artist").resolve("Album B");
Files.createDirectories(albumB);
Path bAudio = createTaggedMp3(albumB, "01 - b.mp3", "b", "Artist", "Album B", false);
Path bLrc = albumB.resolve("01 - b.lrc");
Files.write(bLrc, "[00:00]y".getBytes(StandardCharsets.UTF_8));
// Album C:无 cover 文件,但音频有内嵌封面 → 应保留并从内嵌封面提取出 cover 文件
Path albumC = libDir.resolve("Artist").resolve("Album C");
Files.createDirectories(albumC);
Path cAudio = createTaggedMp3(albumC, "01 - c.mp3", "c", "Artist", "Album C", true);
IngestService service = buildServiceWithValidation();
int removed = service.cleanupLibraryWithoutCover(libDir);
assertEquals(1, removed, "只应删除 Album B 中的无封面音频");
assertTrue(Files.exists(aAudio), "有封面 album 的音频应保留");
assertTrue(Files.exists(aLrc), "有封面 album 的侧车应保留");
assertFalse(Files.exists(bAudio), "无封面且无内嵌封面的音频应被删除");
assertTrue(Files.exists(bLrc), "侧车 .lrc 不应被删除(仅删音频)");
assertTrue(Files.exists(cAudio), "有内嵌封面的音频应保留");
// Finding library-cover-extraction:内嵌封面应被提取为 Navidrome 可发现的 cover 文件
boolean cCover = Files.exists(albumC.resolve("cover.jpg"))
|| Files.exists(albumC.resolve("cover.png"));
assertTrue(cCover, "内嵌封面应被提取为 album 目录的 cover.jpg/png");
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void cleanup_isIdempotent() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-cleanup-");
try {
Path libDir = tmpDir.resolve("Library");
Path albumB = libDir.resolve("Artist").resolve("Album B");
Files.createDirectories(albumB);
createTaggedMp3(albumB, "01 - b.mp3", "b", "Artist", "Album B", false);
createTaggedMp3(albumB, "02 - b2.mp3", "b2", "Artist", "Album B", false);
IngestService service = buildServiceWithValidation();
int first = service.cleanupLibraryWithoutCover(libDir);
int second = service.cleanupLibraryWithoutCover(libDir);
assertEquals(2, first, "首次应删除两首无封面音频");
assertEquals(0, second, "再次运行应无可删除项(幂等)");
} finally {
deleteDirectory(tmpDir);
}
}
@Test
void cleanup_extractsCoverFromEmbeddedArtwork_thenIdempotent() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-cleanup-");
try {
Path libDir = tmpDir.resolve("Library");
// 无 cover 文件,但两首音频均含内嵌封面 → 首轮应提取出 cover,且保留所有音频
Path album = libDir.resolve("Artist").resolve("Album E");
Files.createDirectories(album);
Path a1 = createTaggedMp3(album, "01 - e1.mp3", "e1", "Artist", "Album E", true);
Path a2 = createTaggedMp3(album, "02 - e2.mp3", "e2", "Artist", "Album E", true);
IngestService service = buildServiceWithValidation();
int removed1 = service.cleanupLibraryWithoutCover(libDir);
assertEquals(0, removed1, "含内嵌封面的音频不应被删除");
boolean cover = Files.exists(album.resolve("cover.jpg"))
|| Files.exists(album.resolve("cover.png"));
assertTrue(cover, "应从内嵌封面提取出 cover 文件");
assertTrue(Files.exists(a1) && Files.exists(a2), "两首音频应保留");
// 幂等:已存在 cover,再次运行不删除、不重复
long coverCountBefore = countCovers(album);
int removed2 = service.cleanupLibraryWithoutCover(libDir);
assertEquals(0, removed2, "再次运行应无删除(幂等)");
assertEquals(coverCountBefore, countCovers(album), "不应重复生成 cover 文件");
} finally {
deleteDirectory(tmpDir);
}
}
private static long countCovers(Path dir) throws Exception {
try (Stream<Path> s = Files.list(dir)) {
return s.filter(p -> {
String n = p.getFileName().toString().toLowerCase();
return n.equals("cover.jpg") || n.equals("cover.jpeg") || n.equals("cover.png");
}).count();
}
}
@Test
void cleanup_doesNotTouchRejectedOrCoverFiles() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-cleanup-");
try {
Path libDir = tmpDir.resolve("Library");
Path albumA = libDir.resolve("Artist").resolve("Album A");
Files.createDirectories(albumA);
Path cover = albumA.resolve("cover.png");
Files.write(cover, new byte[]{(byte) 0x89, 'P', 'N', 'G'});
// 无音频,只有 cover → cover 不应被删除
IngestService service = buildServiceWithValidation();
int removed = service.cleanupLibraryWithoutCover(libDir);
assertEquals(0, removed);
assertTrue(Files.exists(cover), "cover 文件不应被清理删除");
} finally {
deleteDirectory(tmpDir);
}
}
// ========== 8. M4A 标签 commit 失败经 remux 恢复 ==========
@Test
void m4aCommitFailure_recoveredViaRemux_notRejectedAsOther() throws Exception {
assumeFfmpeg();
Path tmpDir = Files.createTempDirectory("ingest-commit-");
try {
Path inputDir = tmpDir.resolve("Input");
Path libDir = tmpDir.resolve("Library");
Path rejDir = tmpDir.resolve("Rejected");
Files.createDirectories(inputDir);
Files.createDirectories(libDir);
Files.createDirectories(rejDir);
// 合法 .m4a(含繁体标签 + 内嵌封面),jaudiotagger 可读取但设为只读使 commit() 失败
Path src = inputDir.resolve("commit.m4a");
Path coverSrc = inputDir.resolve(".c.png");
runProcessChecked(new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "color=c=blue:s=48x48:d=0.1",
"-frames:v", "1", coverSrc.toAbsolutePath().toString()), "ffmpeg");
// 繁体标签迫使 ingest 尝试写回(触发 commit
runProcessChecked(new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "anullsrc=r=44100:cl=mono",
"-i", coverSrc.toAbsolutePath().toString(),
"-map", "0:a", "-map", "1:v", "-disposition:v:0", "attached_pic", "-c:v", "png",
"-t", "0.3", "-c:a", "aac", "-f", "mp4",
"-metadata", "title=體", // 體
"-metadata", "artist=門", // 門
"-metadata", "album=風", // 風
src.toAbsolutePath().toString()), "ffmpeg");
Files.deleteIfExists(coverSrc);
// 前置:jaudiotagger 能读取该 m4a
AudioFile pre = AudioFileIO.read(src.toFile());
assumeTrueLocal(pre.getTag() != null, "前置:jaudiotagger 应能读取该 m4a");
// 设为只读,使 audioFile.commit() 抛异常,触发 remux 恢复
src.toFile().setWritable(false);
IngestService service = buildServiceWithValidation();
AtomicInteger ingested = new AtomicInteger();
AtomicInteger otherRejected = new AtomicInteger();
AtomicInteger convFailed = new AtomicInteger();
String result = invokeProcessSingleFile(service, src, libDir, rejDir,
new HashSet<>(), new HashSet<>(),
ingested, new AtomicInteger(),
new AtomicInteger(), new AtomicInteger(),
convFailed, otherRejected);
src.toFile().setWritable(true);
assertEquals("ingested", result,
"commit 失败的合法 m4a 应经 remux 恢复入库,而非 Other,实际: " + result);
assertEquals(0, otherRejected.get(), "不应归类为 Other");
assertEquals(1, ingested.get());
} finally {
deleteDirectory(tmpDir);
}
}
/** 本类内联 assume(避免与 assumeFfmpeg 混淆) */
private static void assumeTrueLocal(boolean cond, String msg) {
org.junit.jupiter.api.Assumptions.assumeTrue(cond, msg);
}
// ========== 工具方法 ==========
/**
@@ -1586,15 +2018,36 @@ class IngestServiceE2ETest {
*/
private static Path createMp4ContainerAsAac(Path dir, String name,
String title, String artist, String album) throws Exception {
// 默认内嵌封面,使后备路径可从源提取封面并满足入库不变量。
return createMp4ContainerAsAac(dir, name, title, artist, album, true);
}
private static Path createMp4ContainerAsAac(Path dir, String name,
String title, String artist, String album,
boolean withCover) throws Exception {
Path file = dir.resolve(name);
java.util.List<String> cmd = new java.util.ArrayList<>(Arrays.asList(
"ffmpeg", "-y", "-f", "lavfi", "-i", "anullsrc=r=44100:cl=mono",
"-t", "0.3", "-c:a", "aac", "-f", "mp4"));
"ffmpeg", "-y", "-f", "lavfi", "-i", "anullsrc=r=44100:cl=mono"));
Path cover = null;
if (withCover) {
cover = dir.resolve("." + name + ".cover.png");
runProcessChecked(new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "color=c=blue:s=48x48:d=0.1",
"-frames:v", "1", cover.toAbsolutePath().toString()), "ffmpeg");
cmd.add("-i");
cmd.add(cover.toAbsolutePath().toString());
}
cmd.addAll(Arrays.asList("-map", "0:a"));
if (withCover) {
cmd.addAll(Arrays.asList("-map", "1:v", "-disposition:v:0", "attached_pic", "-c:v", "png"));
}
cmd.addAll(Arrays.asList("-t", "0.3", "-c:a", "aac", "-f", "mp4"));
if (title != null) { cmd.add("-metadata"); cmd.add("title=" + title); }
if (artist != null) { cmd.add("-metadata"); cmd.add("artist=" + artist); }
if (album != null) { cmd.add("-metadata"); cmd.add("album=" + album); }
cmd.add(file.toAbsolutePath().toString());
runProcessChecked(new ProcessBuilder(cmd), "ffmpeg");
if (cover != null) Files.deleteIfExists(cover);
return file;
}
@@ -1605,18 +2058,27 @@ class IngestServiceE2ETest {
private static Path createMp4WithVideoAndAudio(Path dir, String name,
String title, String artist, String album) throws Exception {
Path file = dir.resolve(name);
// 内嵌一张 attached-picture 封面 + 一个普通视频流:验证 remux 丢弃普通视频,
// 而封面由 acquireCover 从源提取满足入库不变量。
Path cover = dir.resolve("." + name + ".cover.png");
runProcessChecked(new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "color=c=green:s=48x48:d=0.1",
"-frames:v", "1", cover.toAbsolutePath().toString()), "ffmpeg");
java.util.List<String> cmd = new java.util.ArrayList<>(Arrays.asList(
"ffmpeg", "-y",
"-f", "lavfi", "-i", "anullsrc=r=44100:cl=mono",
"-f", "lavfi", "-i", "testsrc=s=64x64:d=0.3",
"-map", "0:a", "-map", "1:v",
"-c:a", "aac", "-c:v", "libx264",
"-i", cover.toAbsolutePath().toString(),
"-map", "0:a", "-map", "1:v", "-map", "2:v",
"-disposition:v:1", "attached_pic",
"-c:a", "aac", "-c:v:0", "libx264", "-c:v:1", "png",
"-shortest", "-t", "0.3", "-f", "mp4"));
if (title != null) { cmd.add("-metadata"); cmd.add("title=" + title); }
if (artist != null) { cmd.add("-metadata"); cmd.add("artist=" + artist); }
if (album != null) { cmd.add("-metadata"); cmd.add("album=" + album); }
cmd.add(file.toAbsolutePath().toString());
runProcessChecked(new ProcessBuilder(cmd), "ffmpeg");
Files.deleteIfExists(cover);
return file;
}
@@ -1674,24 +2136,39 @@ class IngestServiceE2ETest {
);
}
/** 现有调用方无需感知 missingCover 计数器;内部分配一个丢弃即可。 */
private String invokeProcessSingleFile(
IngestService service, Path srcFile, Path libraryPath, Path rejectedPath,
Set<Object> libraryIdentities, Set<Object> batchIdentities,
AtomicInteger ingested, AtomicInteger duplicates,
AtomicInteger missingMeta, AtomicInteger unreadable,
AtomicInteger convFailed, AtomicInteger otherRejected) throws Exception {
return invokeProcessSingleFile(service, srcFile, libraryPath, rejectedPath,
libraryIdentities, batchIdentities,
ingested, duplicates, missingMeta, unreadable,
convFailed, otherRejected, new AtomicInteger());
}
private String invokeProcessSingleFile(
IngestService service, Path srcFile, Path libraryPath, Path rejectedPath,
Set<Object> libraryIdentities, Set<Object> batchIdentities,
AtomicInteger ingested, AtomicInteger duplicates,
AtomicInteger missingMeta, AtomicInteger unreadable,
AtomicInteger convFailed, AtomicInteger otherRejected,
AtomicInteger missingCover) throws Exception {
Method m = IngestService.class.getDeclaredMethod("processSingleFile",
Path.class, Path.class, Path.class,
Set.class, Set.class,
AtomicInteger.class, AtomicInteger.class,
AtomicInteger.class, AtomicInteger.class,
AtomicInteger.class, AtomicInteger.class);
AtomicInteger.class, AtomicInteger.class,
AtomicInteger.class);
m.setAccessible(true);
return (String) m.invoke(service, srcFile, libraryPath, rejectedPath,
libraryIdentities, batchIdentities,
ingested, duplicates, missingMeta, unreadable,
convFailed, otherRejected);
convFailed, otherRejected, missingCover);
}
private Class<?> getIdentityKeyClass() throws Exception {
@@ -1703,12 +2180,39 @@ class IngestServiceE2ETest {
private static Path createTaggedMp3(Path dir, String name,
String title, String artist, String album) throws Exception {
// 默认带内嵌封面:封面是入库硬性不变量,绝大多数成功路径测试需要它。
return createTaggedMp3(dir, name, title, artist, album, true);
}
/**
* 创建带标签的 MP3{@code withCover=true} 时内嵌一张 PNG 封面。
*/
private static Path createTaggedMp3(Path dir, String name,
String title, String artist, String album,
boolean withCover) throws Exception {
Path file = dir.resolve(name);
ProcessBuilder pb = new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "anullsrc",
"-t", "0.1", file.toAbsolutePath().toString()
);
runProcessChecked(pb, "ffmpeg");
if (withCover) {
Path cover = dir.resolve(".__cover_src.png");
ProcessBuilder mkCover = new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "color=c=blue:s=48x48:d=0.1",
"-frames:v", "1", cover.toAbsolutePath().toString());
runProcessChecked(mkCover, "ffmpeg");
ProcessBuilder pb = new ProcessBuilder(
"ffmpeg", "-y",
"-f", "lavfi", "-i", "sine=frequency=440:duration=0.2",
"-i", cover.toAbsolutePath().toString(),
"-map", "0:a", "-map", "1:v", "-disposition:v:0", "attached_pic",
"-c:a", "libmp3lame", "-c:v", "png", "-id3v2_version", "3",
file.toAbsolutePath().toString());
runProcessChecked(pb, "ffmpeg");
Files.deleteIfExists(cover);
} else {
ProcessBuilder pb = new ProcessBuilder(
"ffmpeg", "-y", "-f", "lavfi", "-i", "anullsrc",
"-t", "0.1", file.toAbsolutePath().toString()
);
runProcessChecked(pb, "ffmpeg");
}
writeTags(file, title, artist, album);
return file;
}
+104
View File
@@ -0,0 +1,104 @@
#!/usr/bin/env bash
set -euo pipefail
usage() {
cat <<'EOF'
Usage: cleanup-library.sh --library PATH [--dry-run|--execute]
Default mode is --dry-run. --execute is required to delete files.
The script only handles audio files below PATH and same-basename sidecars.
EOF
}
LIBRARY=""
MODE="dry-run"
while [[ $# -gt 0 ]]; do
case "$1" in
--library) [[ $# -ge 2 ]] || { usage >&2; exit 2; }; LIBRARY="$2"; shift 2 ;;
--dry-run) MODE="dry-run"; shift ;;
--execute) MODE="execute"; shift ;;
-h|--help) usage; exit 0 ;;
*) echo "Unknown option: $1" >&2; usage >&2; exit 2 ;;
esac
done
[[ -n "$LIBRARY" && -d "$LIBRARY" ]] || { echo "Library directory is required" >&2; exit 2; }
command -v ffprobe >/dev/null || { echo "ffprobe is required" >&2; exit 2; }
command -v ffmpeg >/dev/null || { echo "ffmpeg is required" >&2; exit 2; }
is_audio() {
local ext="${1##*.}"
ext="${ext,,}"
case "$ext" in
mp3|m4a|mp4|aac|flac|ogg|oga|opus|wma|wav|ape|aiff|aif|wv|tta) return 0 ;;
*) return 1 ;;
esac
}
has_cover_file() {
[[ -s "$1/cover.jpg" || -s "$1/cover.jpeg" || -s "$1/cover.png" ]]
}
embedded_stream() {
ffprobe -v error -select_streams v \
-show_entries stream=index:stream_disposition=attached_pic \
-of csv=p=0 "$1" 2>/dev/null | awk -F, '$2 == 1 { print $1; exit }'
}
extract_cover() {
local audio="$1" album_dir="$2" stream codec ext target
stream="$(embedded_stream "$audio")"
[[ -n "$stream" ]] || return 1
codec="$(ffprobe -v error -show_entries stream=index,codec_name -of csv=p=0 "$audio" 2>/dev/null | awk -F, -v idx="$stream" '$1 == idx { print $2; exit }')"
case "$codec" in
mjpeg|jpeg) ext="jpg"; target="$album_dir/cover.jpg" ;;
png) ext="png"; target="$album_dir/cover.png" ;;
*) ext="jpg"; target="$album_dir/cover.jpg" ;;
esac
[[ -e "$target" ]] && return 0
if [[ "$MODE" == "dry-run" ]]; then
echo "WOULD EXTRACT $target FROM $audio"
return 0
fi
if [[ "$ext" == "jpg" && "$codec" =~ ^(mjpeg|jpeg)$ || "$ext" == "png" ]]; then
ffmpeg -v error -y -i "$audio" -map "0:$stream" -frames:v 1 -c copy "$target" 2>/dev/null
else
ffmpeg -v error -y -i "$audio" -map "0:$stream" -frames:v 1 "$target" 2>/dev/null
fi
[[ -s "$target" ]]
}
delete_sidecars() {
local audio="$1" dir base sidecar ext
dir="${audio%/*}"; base="${audio##*/}"; base="${base%.*}"
for sidecar in "$dir/$base".*; do
[[ -f "$sidecar" && "$sidecar" != "$audio" ]] || continue
ext="${sidecar##*.}"
case "${ext,,}" in
lrc|cue|json|txt|jpg|jpeg|png|webp)
[[ "${sidecar##*/}" == cover.jpg || "${sidecar##*/}" == cover.jpeg || "${sidecar##*/}" == cover.png ]] && continue
if [[ "$MODE" == "dry-run" ]]; then echo "WOULD DELETE SIDECAR $sidecar"; else rm -f -- "$sidecar"; fi ;;
esac
done
}
removed=0
sidecars=0
extracted=0
while IFS= read -r -d '' audio; do
album_dir="${audio%/*}"
if has_cover_file "$album_dir"; then continue; fi
if extract_cover "$audio" "$album_dir"; then
((extracted += 1)) || true
continue
fi
if [[ "$MODE" == "dry-run" ]]; then
echo "WOULD DELETE AUDIO $audio"
else
rm -f -- "$audio"
fi
delete_sidecars "$audio"
((removed += 1)) || true
done < <(find "$LIBRARY" -type f -print0 | while IFS= read -r -d '' f; do is_audio "$f" && printf '%s\0' "$f"; done)
echo "mode=$MODE extracted=$extracted audio_removed=$removed"