You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
jellyfin/Emby.Common.Implementations/TextEncoding/TextEncoding.cs

253 lines
7.9 KiB

using System;
using System.Text;
using MediaBrowser.Model.IO;
using MediaBrowser.Model.Text;
using System.IO;
using System.Threading;
using System.Threading.Tasks;
using MediaBrowser.Model.MediaInfo;
using MediaBrowser.Model.Logging;
using UniversalDetector;
using NLangDetect.Core;
using MediaBrowser.Model.Serialization;
namespace Emby.Common.Implementations.TextEncoding
{
public class TextEncoding : ITextEncoding
{
private readonly IFileSystem _fileSystem;
private readonly ILogger _logger;
private IJsonSerializer _json;
public TextEncoding(IFileSystem fileSystem, ILogger logger, IJsonSerializer json)
{
_fileSystem = fileSystem;
_logger = logger;
_json = json;
}
public Encoding GetASCIIEncoding()
{
return Encoding.ASCII;
}
private Encoding GetInitialEncoding(byte[] buffer)
{
if (buffer[0] == 0xef && buffer[1] == 0xbb && buffer[2] == 0xbf)
return Encoding.UTF8;
if (buffer[0] == 0xfe && buffer[1] == 0xff)
return Encoding.Unicode;
if (buffer[0] == 0 && buffer[1] == 0 && buffer[2] == 0xfe && buffer[3] == 0xff)
return Encoding.UTF32;
if (buffer[0] == 0x2b && buffer[1] == 0x2f && buffer[2] == 0x76)
return Encoding.UTF7;
var result = new TextEncodingDetect().DetectEncoding(buffer, buffer.Length);
switch (result)
{
case TextEncodingDetect.CharacterEncoding.Ansi:
return Encoding.ASCII;
case TextEncodingDetect.CharacterEncoding.Ascii:
return Encoding.ASCII;
case TextEncodingDetect.CharacterEncoding.Utf16BeBom:
return Encoding.UTF32;
case TextEncodingDetect.CharacterEncoding.Utf16BeNoBom:
return Encoding.UTF32;
case TextEncodingDetect.CharacterEncoding.Utf16LeBom:
return Encoding.UTF32;
case TextEncodingDetect.CharacterEncoding.Utf16LeNoBom:
return Encoding.UTF32;
case TextEncodingDetect.CharacterEncoding.Utf8Bom:
return Encoding.UTF8;
case TextEncodingDetect.CharacterEncoding.Utf8Nobom:
return Encoding.UTF8;
default:
return null;
}
}
private bool _langDetectInitialized;
public string GetDetectedEncodingName(byte[] bytes, string language, bool enableLanguageDetection)
{
var encoding = GetInitialEncoding(bytes);
if (encoding != null && encoding.Equals(Encoding.UTF8))
{
return "utf-8";
}
if (string.IsNullOrWhiteSpace(language) && enableLanguageDetection)
{
if (!_langDetectInitialized)
{
_langDetectInitialized = true;
LanguageDetector.Initialize(_json);
}
language = DetectLanguage(bytes);
if (!string.IsNullOrWhiteSpace(language))
{
_logger.Debug("Text language detected as {0}", language);
}
}
var charset = DetectCharset(bytes, language);
if (!string.IsNullOrWhiteSpace(charset))
{
if (string.Equals(charset, "utf-8", StringComparison.OrdinalIgnoreCase))
{
return "utf-8";
}
if (!string.Equals(charset, "windows-1252", StringComparison.OrdinalIgnoreCase))
{
return charset;
}
}
if (!string.IsNullOrWhiteSpace(language))
{
return GetFileCharacterSetFromLanguage(language);
}
return null;
}
private string DetectLanguage(byte[] bytes)
{
try
{
return LanguageDetector.DetectLanguage(Encoding.UTF8.GetString(bytes));
}
catch (NLangDetectException ex)
{
}
try
{
return LanguageDetector.DetectLanguage(Encoding.ASCII.GetString(bytes));
}
catch (NLangDetectException ex)
{
}
try
{
return LanguageDetector.DetectLanguage(Encoding.Unicode.GetString(bytes));
}
catch (NLangDetectException ex)
{
}
return null;
}
public Encoding GetEncodingFromCharset(string charset)
{
if (string.IsNullOrWhiteSpace(charset))
{
throw new ArgumentNullException("charset");
}
_logger.Debug("Getting encoding object for character set: {0}", charset);
try
{
return Encoding.GetEncoding(charset);
}
catch (ArgumentException)
{
charset = charset.Replace("-", string.Empty);
_logger.Debug("Getting encoding object for character set: {0}", charset);
return Encoding.GetEncoding(charset);
}
}
public Encoding GetDetectedEncoding(byte[] bytes, string language, bool enableLanguageDetection)
{
var charset = GetDetectedEncodingName(bytes, language, enableLanguageDetection);
return GetEncodingFromCharset(charset);
}
private string GetFileCharacterSetFromLanguage(string language)
{
// https://developer.xamarin.com/api/type/System.Text.Encoding/
switch (language.ToLower())
{
case "hun":
return "windows-1252";
case "pol":
case "cze":
case "ces":
case "slo":
case "srp":
case "hrv":
case "rum":
case "ron":
case "rup":
return "windows-1250";
// albanian
case "alb":
case "sqi":
return "windows-1250";
// slovak
case "slk":
case "slv":
return "windows-1250";
case "ara":
return "windows-1256";
case "heb":
return "windows-1255";
case "grc":
return "windows-1253";
// greek
case "gre":
case "ell":
return "windows-1253";
case "crh":
case "ota":
case "tur":
return "windows-1254";
// bulgarian
case "bul":
case "bgr":
return "windows-1251";
case "rus":
return "windows-1251";
case "vie":
return "windows-1258";
case "kor":
return "cp949";
default:
return "windows-1252";
}
}
private string DetectCharset(byte[] bytes, string language)
{
var detector = new CharsetDetector();
detector.Feed(bytes, 0, bytes.Length);
detector.DataEnd();
var charset = detector.Charset;
// This is often incorrectly indetected. If this happens, try to use other techniques instead
if (string.Equals("x-mac-cyrillic", charset, StringComparison.OrdinalIgnoreCase))
{
if (!string.IsNullOrWhiteSpace(language))
{
return null;
}
}
return charset;
}
}
}