Natural Sort Order in C#

Solution 1:

The easiest thing to do is just P/Invoke the built-in function in Windows, and use it as the comparison function in your IComparer:

[DllImport("shlwapi.dll", CharSet = CharSet.Unicode)]
private static extern int StrCmpLogicalW(string psz1, string psz2);

Michael Kaplan has some examples of how this function works here, and the changes that were made for Vista to make it work more intuitively. The plus side of this function is that it will have the same behaviour as the version of Windows it runs on, however this does mean that it differs between versions of Windows so you need to consider whether this is a problem for you.

So a complete implementation would be something like:

[SuppressUnmanagedCodeSecurity]
internal static class SafeNativeMethods
{
    [DllImport("shlwapi.dll", CharSet = CharSet.Unicode)]
    public static extern int StrCmpLogicalW(string psz1, string psz2);
}

public sealed class NaturalStringComparer : IComparer<string>
{
    public int Compare(string a, string b)
    {
        return SafeNativeMethods.StrCmpLogicalW(a, b);
    }
}

public sealed class NaturalFileInfoNameComparer : IComparer<FileInfo>
{
    public int Compare(FileInfo a, FileInfo b)
    {
        return SafeNativeMethods.StrCmpLogicalW(a.Name, b.Name);
    }
}

Solution 2:

Just thought I'd add to this (with the most concise solution I could find):

public static IOrderedEnumerable<T> OrderByAlphaNumeric<T>(this IEnumerable<T> source, Func<T, string> selector)
{
    int max = source
        .SelectMany(i => Regex.Matches(selector(i), @"\d+").Cast<Match>().Select(m => (int?)m.Value.Length))
        .Max() ?? 0;

    return source.OrderBy(i => Regex.Replace(selector(i), @"\d+", m => m.Value.PadLeft(max, '0')));
}

The above pads any numbers in the string to the max length of all numbers in all strings and uses the resulting string to sort.

The cast to (int?) is to allow for collections of strings without any numbers (.Max() on an empty enumerable throws an InvalidOperationException).

Solution 3:

None of the existing implementations looked great so I wrote my own. The results are almost identical to the sorting used by modern versions of Windows Explorer (Windows 7/8). The only differences I've seen are 1) although Windows used to (e.g. XP) handle numbers of any length, it's now limited to 19 digits - mine is unlimited, 2) Windows gives inconsistent results with certain sets of Unicode digits - mine works fine (although it doesn't numerically compare digits from surrogate pairs; nor does Windows), and 3) mine can't distinguish different types of non-primary sort weights if they occur in different sections (e.g. "e-1é" vs "é1e-" - the sections before and after the number have diacritic and punctuation weight differences).

public static int CompareNatural(string strA, string strB) {
    return CompareNatural(strA, strB, CultureInfo.CurrentCulture, CompareOptions.IgnoreCase);
}

public static int CompareNatural(string strA, string strB, CultureInfo culture, CompareOptions options) {
    CompareInfo cmp = culture.CompareInfo;
    int iA = 0;
    int iB = 0;
    int softResult = 0;
    int softResultWeight = 0;
    while (iA < strA.Length && iB < strB.Length) {
        bool isDigitA = Char.IsDigit(strA[iA]);
        bool isDigitB = Char.IsDigit(strB[iB]);
        if (isDigitA != isDigitB) {
            return cmp.Compare(strA, iA, strB, iB, options);
        }
        else if (!isDigitA && !isDigitB) {
            int jA = iA + 1;
            int jB = iB + 1;
            while (jA < strA.Length && !Char.IsDigit(strA[jA])) jA++;
            while (jB < strB.Length && !Char.IsDigit(strB[jB])) jB++;
            int cmpResult = cmp.Compare(strA, iA, jA - iA, strB, iB, jB - iB, options);
            if (cmpResult != 0) {
                // Certain strings may be considered different due to "soft" differences that are
                // ignored if more significant differences follow, e.g. a hyphen only affects the
                // comparison if no other differences follow
                string sectionA = strA.Substring(iA, jA - iA);
                string sectionB = strB.Substring(iB, jB - iB);
                if (cmp.Compare(sectionA + "1", sectionB + "2", options) ==
                    cmp.Compare(sectionA + "2", sectionB + "1", options))
                {
                    return cmp.Compare(strA, iA, strB, iB, options);
                }
                else if (softResultWeight < 1) {
                    softResult = cmpResult;
                    softResultWeight = 1;
                }
            }
            iA = jA;
            iB = jB;
        }
        else {
            char zeroA = (char)(strA[iA] - (int)Char.GetNumericValue(strA[iA]));
            char zeroB = (char)(strB[iB] - (int)Char.GetNumericValue(strB[iB]));
            int jA = iA;
            int jB = iB;
            while (jA < strA.Length && strA[jA] == zeroA) jA++;
            while (jB < strB.Length && strB[jB] == zeroB) jB++;
            int resultIfSameLength = 0;
            do {
                isDigitA = jA < strA.Length && Char.IsDigit(strA[jA]);
                isDigitB = jB < strB.Length && Char.IsDigit(strB[jB]);
                int numA = isDigitA ? (int)Char.GetNumericValue(strA[jA]) : 0;
                int numB = isDigitB ? (int)Char.GetNumericValue(strB[jB]) : 0;
                if (isDigitA && (char)(strA[jA] - numA) != zeroA) isDigitA = false;
                if (isDigitB && (char)(strB[jB] - numB) != zeroB) isDigitB = false;
                if (isDigitA && isDigitB) {
                    if (numA != numB && resultIfSameLength == 0) {
                        resultIfSameLength = numA < numB ? -1 : 1;
                    }
                    jA++;
                    jB++;
                }
            }
            while (isDigitA && isDigitB);
            if (isDigitA != isDigitB) {
                // One number has more digits than the other (ignoring leading zeros) - the longer
                // number must be larger
                return isDigitA ? 1 : -1;
            }
            else if (resultIfSameLength != 0) {
                // Both numbers are the same length (ignoring leading zeros) and at least one of
                // the digits differed - the first difference determines the result
                return resultIfSameLength;
            }
            int lA = jA - iA;
            int lB = jB - iB;
            if (lA != lB) {
                // Both numbers are equivalent but one has more leading zeros
                return lA > lB ? -1 : 1;
            }
            else if (zeroA != zeroB && softResultWeight < 2) {
                softResult = cmp.Compare(strA, iA, 1, strB, iB, 1, options);
                softResultWeight = 2;
            }
            iA = jA;
            iB = jB;
        }
    }
    if (iA < strA.Length || iB < strB.Length) {
        return iA < strA.Length ? 1 : -1;
    }
    else if (softResult != 0) {
        return softResult;
    }
    return 0;
}

The signature matches the Comparison<string> delegate:

string[] files = Directory.GetFiles(@"C:\");
Array.Sort(files, CompareNatural);

Here's a wrapper class for use as IComparer<string>:

public class CustomComparer<T> : IComparer<T> {
    private Comparison<T> _comparison;

    public CustomComparer(Comparison<T> comparison) {
        _comparison = comparison;
    }

    public int Compare(T x, T y) {
        return _comparison(x, y);
    }
}

Example:

string[] files = Directory.EnumerateFiles(@"C:\")
    .OrderBy(f => f, new CustomComparer<string>(CompareNatural))
    .ToArray();

Here's a good set of filenames I use for testing:

Func<string, string> expand = (s) => { int o; while ((o = s.IndexOf('\\')) != -1) { int p = o + 1;
    int z = 1; while (s[p] == '0') { z++; p++; } int c = Int32.Parse(s.Substring(p, z));
    s = s.Substring(0, o) + new string(s[o - 1], c) + s.Substring(p + z); } return s; };
string encodedFileNames =
    "KDEqLW4xMiotbjEzKjAwMDFcMDY2KjAwMlwwMTcqMDA5XDAxNyowMlwwMTcqMDlcMDE3KjEhKjEtISox" +
    "LWEqMS4yNT8xLjI1KjEuNT8xLjUqMSoxXDAxNyoxXDAxOCoxXDAxOSoxXDA2NioxXDA2NyoxYSoyXDAx" +
    "NyoyXDAxOCo5XDAxNyo5XDAxOCo5XDA2Nio9MSphMDAxdGVzdDAxKmEwMDF0ZXN0aW5nYTBcMzEqYTAw" +
    "Mj9hMDAyIGE/YTAwMiBhKmEwMDIqYTAwMmE/YTAwMmEqYTAxdGVzdGluZ2EwMDEqYTAxdnNmcyphMSph" +
    "MWEqYTF6KmEyKmIwMDAzcTYqYjAwM3E0KmIwM3E1KmMtZSpjZCpjZipmIDEqZipnP2cgMT9oLW4qaG8t" +
    "bipJKmljZS1jcmVhbT9pY2VjcmVhbT9pY2VjcmVhbS0/ajBcNDE/ajAwMWE/ajAxP2shKmsnKmstKmsx" +
    "KmthKmxpc3QqbTAwMDNhMDA1YSptMDAzYTAwMDVhKm0wMDNhMDA1Km0wMDNhMDA1YSpuMTIqbjEzKm8t" +
    "bjAxMypvLW4xMipvLW40P28tbjQhP28tbjR6P28tbjlhLWI1Km8tbjlhYjUqb24wMTMqb24xMipvbjQ/" +
    "b240IT9vbjR6P29uOWEtYjUqb245YWI1Km/CrW4wMTMqb8KtbjEyKnAwMCpwMDEqcDAxwr0hKnAwMcK9" +
    "KnAwMcK9YSpwMDHCvcK+KnAwMipwMMK9KnEtbjAxMypxLW4xMipxbjAxMypxbjEyKnItMDAhKnItMDAh" +
    "NSpyLTAwIe+8lSpyLTAwYSpyLe+8kFwxIS01KnIt77yQXDEhLe+8lSpyLe+8kFwxISpyLe+8kFwxITUq" +
    "ci3vvJBcMSHvvJUqci3vvJBcMWEqci3vvJBcMyE1KnIwMCEqcjAwLTUqcjAwLjUqcjAwNSpyMDBhKnIw" +
    "NSpyMDYqcjQqcjUqctmg2aYqctmkKnLZpSpy27Dbtipy27Qqctu1KnLfgN+GKnLfhCpy34UqcuClpuCl" +
    "rCpy4KWqKnLgpasqcuCnpuCnrCpy4KeqKnLgp6sqcuCppuCprCpy4KmqKnLgqasqcuCrpuCrrCpy4Kuq" +
    "KnLgq6sqcuCtpuCtrCpy4K2qKnLgrasqcuCvpuCvrCpy4K+qKnLgr6sqcuCxpuCxrCpy4LGqKnLgsasq" +
    "cuCzpuCzrCpy4LOqKnLgs6sqcuC1puC1rCpy4LWqKnLgtasqcuC5kOC5lipy4LmUKnLguZUqcuC7kOC7" +
    "lipy4LuUKnLgu5UqcuC8oOC8pipy4LykKnLgvKUqcuGBgOGBhipy4YGEKnLhgYUqcuGCkOGClipy4YKU" +
    "KnLhgpUqcuGfoOGfpipy4Z+kKnLhn6UqcuGgkOGglipy4aCUKnLhoJUqcuGlhuGljCpy4aWKKnLhpYsq" +
    "cuGnkOGnlipy4aeUKnLhp5UqcuGtkOGtlipy4a2UKnLhrZUqcuGusOGutipy4a60KnLhrrUqcuGxgOGx" +
    "hipy4bGEKnLhsYUqcuGxkOGxlipy4bGUKnLhsZUqcuqYoFwx6pilKnLqmKDqmKUqcuqYoOqYpipy6pik" +
    "KnLqmKUqcuqjkOqjlipy6qOUKnLqo5UqcuqkgOqkhipy6qSEKnLqpIUqcuqpkOqplipy6qmUKnLqqZUq" +
    "cvCQkqAqcvCQkqUqcvCdn5gqcvCdn50qcu+8kFwxISpy77yQXDEt77yVKnLvvJBcMS7vvJUqcu+8kFwx" +
    "YSpy77yQXDHqmKUqcu+8kFwx77yO77yVKnLvvJBcMe+8lSpy77yQ77yVKnLvvJDvvJYqcu+8lCpy77yV" +
    "KnNpKnPEsSp0ZXN02aIqdGVzdNmi2aAqdGVzdNmjKnVBZS0qdWFlKnViZS0qdUJlKnVjZS0xw6kqdWNl" +
    "McOpLSp1Y2Uxw6kqdWPDqS0xZSp1Y8OpMWUtKnVjw6kxZSp3ZWlhMSp3ZWlhMip3ZWlzczEqd2Vpc3My" +
    "KndlaXoxKndlaXoyKndlacOfMSp3ZWnDnzIqeSBhMyp5IGE0KnknYTMqeSdhNCp5K2EzKnkrYTQqeS1h" +
    "Myp5LWE0KnlhMyp5YTQqej96IDA1MD96IDIxP3ohMjE/ejIwP3oyMj96YTIxP3rCqTIxP1sxKl8xKsKt" +
    "bjEyKsKtbjEzKsSwKg==";
string[] fileNames = Encoding.UTF8.GetString(Convert.FromBase64String(encodedFileNames))
    .Replace("*", ".txt?").Split(new[] { "?" }, StringSplitOptions.RemoveEmptyEntries)
    .Select(n => expand(n)).ToArray();

Solution 4:

Pure C# solution for linq orderby:

http://zootfroot.blogspot.com/2009/09/natural-sort-compare-with-linq-orderby.html

public class NaturalSortComparer<T> : IComparer<string>, IDisposable
{
    private bool isAscending;

    public NaturalSortComparer(bool inAscendingOrder = true)
    {
        this.isAscending = inAscendingOrder;
    }

    #region IComparer<string> Members

    public int Compare(string x, string y)
    {
        throw new NotImplementedException();
    }

    #endregion

    #region IComparer<string> Members

    int IComparer<string>.Compare(string x, string y)
    {
        if (x == y)
            return 0;

        string[] x1, y1;

        if (!table.TryGetValue(x, out x1))
        {
            x1 = Regex.Split(x.Replace(" ", ""), "([0-9]+)");
            table.Add(x, x1);
        }

        if (!table.TryGetValue(y, out y1))
        {
            y1 = Regex.Split(y.Replace(" ", ""), "([0-9]+)");
            table.Add(y, y1);
        }

        int returnVal;

        for (int i = 0; i < x1.Length && i < y1.Length; i++)
        {
            if (x1[i] != y1[i])
            {
                returnVal = PartCompare(x1[i], y1[i]);
                return isAscending ? returnVal : -returnVal;
            }
        }

        if (y1.Length > x1.Length)
        {
            returnVal = 1;
        }
        else if (x1.Length > y1.Length)
        { 
            returnVal = -1; 
        }
        else
        {
            returnVal = 0;
        }

        return isAscending ? returnVal : -returnVal;
    }

    private static int PartCompare(string left, string right)
    {
        int x, y;
        if (!int.TryParse(left, out x))
            return left.CompareTo(right);

        if (!int.TryParse(right, out y))
            return left.CompareTo(right);

        return x.CompareTo(y);
    }

    #endregion

    private Dictionary<string, string[]> table = new Dictionary<string, string[]>();

    public void Dispose()
    {
        table.Clear();
        table = null;
    }
}

Solution 5:

Matthews Horsleys answer is the fastest method which doesn't change behaviour depending on which version of windows your program is running on. However, it can be even faster by creating the regex once, and using RegexOptions.Compiled. I also added the option of inserting a string comparer so you can ignore case if needed, and improved readability a bit.

    public static IEnumerable<T> OrderByNatural<T>(this IEnumerable<T> items, Func<T, string> selector, StringComparer stringComparer = null)
    {
        var regex = new Regex(@"\d+", RegexOptions.Compiled);

        int maxDigits = items
                      .SelectMany(i => regex.Matches(selector(i)).Cast<Match>().Select(digitChunk => (int?)digitChunk.Value.Length))
                      .Max() ?? 0;

        return items.OrderBy(i => regex.Replace(selector(i), match => match.Value.PadLeft(maxDigits, '0')), stringComparer ?? StringComparer.CurrentCulture);
    }

Use by

var sortedEmployees = employees.OrderByNatural(emp => emp.Name);

This takes 450ms to sort 100,000 strings compared to 300ms for the default .net string comparison - pretty fast!