Why does Path.Combine not properly concatenate filenames that start with Path.DirectorySeparatorChar?

This is kind of a philosophical question (which perhaps only Microsoft can truly answer), since it's doing exactly what the documentation says.

System.IO.Path.Combine

"If path2 contains an absolute path, this method returns path2."

Here's the actual Combine method from the .NET source. You can see that it calls CombineNoChecks, which then calls IsPathRooted on path2 and returns that path if so:

public static String Combine(String path1, String path2) {
    if (path1==null || path2==null)
        throw new ArgumentNullException((path1==null) ? "path1" : "path2");
    Contract.EndContractBlock();
    CheckInvalidPathChars(path1);
    CheckInvalidPathChars(path2);

    return CombineNoChecks(path1, path2);
}

internal static string CombineNoChecks(string path1, string path2)
{
    if (path2.Length == 0)
        return path1;

    if (path1.Length == 0)
        return path2;

    if (IsPathRooted(path2))
        return path2;

    char ch = path1[path1.Length - 1];
    if (ch != DirectorySeparatorChar && ch != AltDirectorySeparatorChar &&
            ch != VolumeSeparatorChar) 
        return path1 + DirectorySeparatorCharAsString + path2;
    return path1 + path2;
}

I don't know what the rationale is. I guess the solution is to strip off (or Trim) DirectorySeparatorChar from the beginning of the second path; maybe write your own Combine method that does that and then calls Path.Combine().

This is the disassembled code from .NET Reflector for Path.Combine method. Check IsPathRooted function. If the second path is rooted (starts with a DirectorySeparatorChar), return second path as it is.

public static string Combine(string path1, string path2)
{
    if ((path1 == null) || (path2 == null))
    {
        throw new ArgumentNullException((path1 == null) ? "path1" : "path2");
    }
    CheckInvalidPathChars(path1);
    CheckInvalidPathChars(path2);
    if (path2.Length == 0)
    {
        return path1;
    }
    if (path1.Length == 0)
    {
        return path2;
    }
    if (IsPathRooted(path2))
    {
        return path2;
    }
    char ch = path1[path1.Length - 1];
    if (((ch != DirectorySeparatorChar) &&
         (ch != AltDirectorySeparatorChar)) &&
         (ch != VolumeSeparatorChar))
    {
        return (path1 + DirectorySeparatorChar + path2);
    }
    return (path1 + path2);
}


public static bool IsPathRooted(string path)
{
    if (path != null)
    {
        CheckInvalidPathChars(path);
        int length = path.Length;
        if (
              (
                  (length >= 1) &&
                  (
                      (path[0] == DirectorySeparatorChar) ||
                      (path[0] == AltDirectorySeparatorChar)
                  )
              )

              ||

              ((length >= 2) &&
              (path[1] == VolumeSeparatorChar))
           )
        {
            return true;
        }
    }
    return false;
}

I wanted to solve this problem:

string sample1 = "configuration/config.xml";
string sample2 = "/configuration/config.xml";
string sample3 = "\\configuration/config.xml";

string dir1 = "c:\\temp";
string dir2 = "c:\\temp\\";
string dir3 = "c:\\temp/";

string path1 = PathCombine(dir1, sample1);
string path2 = PathCombine(dir1, sample2);
string path3 = PathCombine(dir1, sample3);

string path4 = PathCombine(dir2, sample1);
string path5 = PathCombine(dir2, sample2);
string path6 = PathCombine(dir2, sample3);

string path7 = PathCombine(dir3, sample1);
string path8 = PathCombine(dir3, sample2);
string path9 = PathCombine(dir3, sample3);

Of course, all paths 1-9 should contain an equivalent string in the end. Here is the PathCombine method I came up with:

private string PathCombine(string path1, string path2)
{
    if (Path.IsPathRooted(path2))
    {
        path2 = path2.TrimStart(Path.DirectorySeparatorChar);
        path2 = path2.TrimStart(Path.AltDirectorySeparatorChar);
    }

    return Path.Combine(path1, path2);
}

I also think that it is quite annoying that this string handling has to be done manually, and I'd be interested in the reason behind this.