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.
Readarr/NzbDrone.Core/Parser.cs

183 lines
6.2 KiB

using System;
using System.Collections.Generic;
using System.IO;
using System.Linq;
using System.Text;
using System.Text.RegularExpressions;
using NLog;
using NzbDrone.Core.Model;
using NzbDrone.Core.Providers;
using NzbDrone.Core.Repository.Quality;
using Rss;
namespace NzbDrone.Core
{
internal static class Parser
{
private static readonly Logger Logger = LogManager.GetCurrentClassLogger();
private static readonly Regex[] ReportTitleRegex = new[]
{
new Regex(@"(?<title>.+?)?\W?(?<year>\d+?)?\WS?(?<season>\d+)\w(?<episode>\d+)\W(?!\\)", RegexOptions.IgnoreCase | RegexOptions.Compiled)
};
private static readonly Regex NormalizeRegex = new Regex(@"((\s|^)the(\s|$))|((\s|^)and(\s|$))|[^a-z]", RegexOptions.IgnoreCase | RegexOptions.Compiled);
/// <summary>
/// Parses a post title into list of episodes it contains
/// </summary>
/// <param name="title">Title of the report</param>
/// <returns>List of episodes contained to the post</returns>
internal static List<EpisodeParseResult> ParseEpisodeInfo(string title)
{
Logger.Trace("Parsing string '{0}'", title);
var result = new List<EpisodeParseResult>();
foreach (var regex in ReportTitleRegex)
{
var match = regex.Matches(title);
if (match.Count != 0)
{
var seriesName = NormalizeTitle(match[0].Groups["title"].Value);
var year = 0;
Int32.TryParse(match[0].Groups["year"].Value, out year);
if (year < 1900 || year > DateTime.Now.Year + 1)
{
year = 0;
}
foreach (Match matchGroup in match)
{
var parsedEpisode = new EpisodeParseResult
{
SeriesTitle = seriesName,
SeasonNumber = Convert.ToInt32(matchGroup.Groups["season"].Value),
EpisodeNumber = Convert.ToInt32(matchGroup.Groups["episode"].Value),
Year = year
};
result.Add(parsedEpisode);
Logger.Trace("Episode Parsed. {0}", parsedEpisode);
}
}
}
return result;
}
/// <summary>
/// Parses proper status out of a report title
/// </summary>
/// <param name="title">Title of the report</param>
/// <returns></returns>
internal static bool ParseProper(string title)
{
return title.ToLower().Contains("proper");
}
internal static QualityTypes ParseQuality(string name)
{
Logger.Trace("Trying to parse quality for {0}", name);
var result = QualityTypes.Unknown;
name = name.ToLowerInvariant();
if (name.Contains("dvd"))
return QualityTypes.DVD;
if (name.Contains("bdrip") || name.Contains("brrip"))
{
return QualityTypes.BDRip;
}
if (name.Contains("xvid") || name.Contains("divx"))
{
if (name.Contains("bluray"))
{
return QualityTypes.BDRip;
}
return QualityTypes.TV;
}
if (name.Contains("bluray"))
{
return QualityTypes.Bluray;
}
if (name.Contains("web-dl"))
return QualityTypes.WEBDL;
if (name.Contains("x264") || name.Contains("h264") || name.Contains("720p"))
return QualityTypes.HDTV;
//Based on extension
if (result == QualityTypes.Unknown)
{
switch (new FileInfo(name).Extension.ToLower())
{
case ".avi":
case ".xvid":
case ".wmv":
{
result = QualityTypes.TV;
break;
}
case ".mkv":
{
result = QualityTypes.HDTV;
break;
}
}
}
Logger.Trace("Quality Parsed:{0} Title:", result, name);
return result;
}
/// <summary>
/// Normalizes the title. removing all non-word characters as well as common tokens
/// such as 'the' and 'and'
/// </summary>
/// <param name="title">title</param>
/// <returns></returns>
internal static string NormalizeTitle(string title)
{
return NormalizeRegex.Replace(title, String.Empty).ToLower();
}
//Note: changing case on path is a problem for running on mono/*nix
public static string NormalizePath(string path)
{
if (String.IsNullOrEmpty(path))
throw new ArgumentException("Path can not be null or empty");
var info = new FileInfo(path);
if (info.FullName.StartsWith(@"\\")) //UNC
{
return info.FullName.ToLower().TrimEnd('/', '\\', ' ');
}
return info.FullName.ToLower().Trim('/', '\\', ' ');
}
public static NzbInfoModel ParseNzbInfo(FeedInfoModel feed, RssItem item)
{
NzbSiteModel site = NzbSiteModel.Parse(feed.Url.ToLower());
return new NzbInfoModel
{
Id = site.ParseId(item.Link.ToString()),
Title = item.Title,
Site = site,
Link = item.Link,
Description = item.Description
};
}
}
}