13

For a long path aware process on Windows 10, I'm trying to understand what the argument restrictions are when using the windows shell method PathRelativePathTo.

In my example below, I'm using C# via pinvoke to call the method.
I've given multiple examples below and their output. Note:

  • All of the examples give directory paths for "from" and file paths for "to" (none of these paths actually exist on disk)
  • My observations are that
    • Paths under the "short" MAX_PATH length (260) return success with the expected result.
    • Some paths over the "short" MAX_PATH return success with the correct result.
    • Some paths over the "short" MAX_PATH return success with the wrong answer (yikes!)
    • Some much longer paths return an Error. However, it is not at some fixed max length.

Source:

    class Program
    {
        static class Native
        {
            // https://www.pinvoke.net/default.aspx/shlwapi.pathrelativepathto
            // https://learn.microsoft.com/en-us/windows/win32/api/shlwapi/nf-shlwapi-pathrelativepathtoa
            [DllImport("shlwapi.dll", SetLastError = true, CharSet = CharSet.Auto)]
            [return: MarshalAs(UnmanagedType.Bool)]
            internal static extern bool PathRelativePathTo([Out] StringBuilder pszPath, [In] string pszFrom, [In] int dwAttrFrom, [In] string pszTo, [In] int dwAttrTo);
        }

        static void Main(string[] args)
        {
            string pszFrom, pszTo;
            int i = 0;

            // #1 At "short" max path (259)
            // Succeeds with right answer
            pszFrom = @"c:\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCD123456789";
            pszTo = @"c:\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCD123456789\abcdefghijklmnop.txt";
            TestPathRelativePathTo(++i, pszFrom, pszTo);

            // #2 One over "short" max path
            // Succeeds with right answer
            pszFrom = @"c:\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCD1234567890";
            pszTo = @"c:\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCD1234567890\abcdefghijklmnop.txt";
            TestPathRelativePathTo(++i, pszFrom, pszTo);

            // #3 Shortest path (by experiment) that returned the wrong answer
            pszFrom = @"c:\ABCDEFGHIJKLMNOPQRS\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCD1234567890";
            pszTo = @"c:\ABCDEFGHIJKLMNOPQRS\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCD1234567890\b.txt";
            TestPathRelativePathTo(++i, pszFrom, pszTo);

            // #4: Long path that errors out
            // Errors out
            pszFrom = @"c:\ABCDEFGHIJKLMNOPQRSTUVWXYZABCDEFGH\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890";
            pszTo = @"c:\ABCDEFGHIJKLMNOPQRSTUVWXYZABCDEFGH\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\b.txt";
            TestPathRelativePathTo(++i, pszFrom, pszTo);

            // #5: Same as previous except one character removed from beginning of first folder
            // Succeeds, but wrong return result
            pszFrom = @"c:\BCDEFGHIJKLMNOPQRSTUVWXYZABCDEFGH\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890";
            pszTo = @"c:\BCDEFGHIJKLMNOPQRSTUVWXYZABCDEFGH\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\b.txt";
            TestPathRelativePathTo(++i, pszFrom, pszTo);

            // #6: Same as previous except 3 characters added to filename. 
            // Succeeds, but wrong return result
            pszFrom = @"c:\BCDEFGHIJKLMNOPQRSTUVWXYZABCDEFGH\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890";
            pszTo = @"c:\BCDEFGHIJKLMNOPQRSTUVWXYZABCDEFGH\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\b123.txt";
            TestPathRelativePathTo(++i, pszFrom, pszTo);
        }

        static void TestPathRelativePathTo(int i, string pszFromDir, string pszToFile)
        {
            int maxResult = 10000;
            StringBuilder result = new StringBuilder(maxResult);
            Console.WriteLine($"#{i}: Calling PathRelativePathTo(...): pszFrom.Length: {pszFromDir.Length}; pszTo.Length {pszToFile.Length} ");
            bool bRet = Native.PathRelativePathTo(result, pszFromDir, (int)FileAttributes.Directory, pszToFile, (int)FileAttributes.Normal);
            if (!bRet)
            {
                // *Edit*: As pointed out in the comments, PathRelativePathTo does not set last error, so this part of the code is incorrect, it should really just print out that the method returned false.
                // https://blogs.msdn.microsoft.com/shawnfa/2004/09/10/formatmessage-shortcut-for-win32-error-codes/
                int currentError = Marshal.GetLastWin32Error();
                var errorMessage = new Win32Exception(currentError).Message;
                Console.WriteLine($"  Error: {errorMessage}");
            }
            else
            {
                Console.WriteLine($"  Result: {result}");
            }
        }
    }

Output:

#1: Calling PathRelativePathTo(...): pszFrom.Length: 238; pszTo.Length 259
  Result: .\abcdefghijklmnop.txt
#2: Calling PathRelativePathTo(...): pszFrom.Length: 239; pszTo.Length 260
  Result: .\abcdefghijklmnop.txt
#3: Calling PathRelativePathTo(...): pszFrom.Length: 259; pszTo.Length 265
  Result: ..\ABCD1234567890\b.txt
#4: Calling PathRelativePathTo(...): pszFrom.Length: 481; pszTo.Length 487
  Error: The system cannot find the file specified
#5: Calling PathRelativePathTo(...): pszFrom.Length: 480; pszTo.Length 486
  Result: .\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\b.txt
#6: Calling PathRelativePathTo(...): pszFrom.Length: 480; pszTo.Length 489
  Result: .\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890\b123.txt

Questions:

  • What is the expected behavior of PathRelativePathTo with respect to the above?
  • Is it only expected to work properly with paths under the "short" MAX_PATH limit (and the rest of the behavior is undefined)?
  • Is there something else in the .net framework I can use instead (Note: I see that .NET Core has Path.GetRelativePath, but I can't (yet) use that)?
Agnel Amodia
  • 765
  • 8
  • 18
Matt Smith
  • 17,026
  • 7
  • 53
  • 103
  • Comments are not for extended discussion; this conversation has been [moved to chat](https://chat.stackoverflow.com/rooms/202237/discussion-on-question-by-matt-smith-restrictions-on-arguments-to-pathrelativepa). – Samuel Liew Nov 13 '19 at 01:03
  • Forget `PathRelativePathTo`, it's not meant for long paths. It's actually not safe to use it, since you can't state the size of the destination buffer, the documentation only says it "must be at least MAX_PATH characters in size." – acelent Nov 13 '19 at 09:46
  • official doc is pretty clear on the MAX_PATH limits. As for a replacement, it's easy to get it wrong, you could re-use .NET core source or use it as a starting point: https://github.com/dotnet/corefx/blob/70150e6750c35c9a67803e9cddfda7b1db2ec487/src/Common/src/CoreLib/System/IO/Path.cs#L805 – Simon Mourier Nov 15 '19 at 10:09
  • What are you using so far? Classic .NET or .NET Core, which version? – Pavel Anikhouski Nov 15 '19 at 21:30
  • .net framework. Once I am able to move to .net core 3.0, I will be all set since they have the built in method I mentioned. – Matt Smith Nov 16 '19 at 22:34

4 Answers4

4

From the looks of it, seems like the PathRelativePathTo API is safe only for paths upto MAX_LENGTH. Atleast from Wine documentation, we see that the API has been problematic in the Win32 implementation.

The Win32 version of this function contains a bug where the lpszTo string may be referenced 1 byte beyond the end of the string. As a result random garbage may be written to the output path, depending on what lies beyond the last byte of the string. This bug occurs because of the behaviour of PathCommonPrefix() (see notes for that function), and no workaround seems possible with Win32. This bug has been fixed here, so for example the relative path from "\" to "\" is correctly determined as "." in this implementation.

And from the PathCommonPrefix documentation,

A common prefix of 2 is always returned as 3. It is thus possible for the length returned to be invalid (i.e. Longer than one or both of the strings given as parameters). This Win32 behaviour has been implemented here, and cannot be changed (fixed?) without breaking other SHLWAPI calls. To work around this when using this function, always check that the byte at [common_prefix_len-1] is not a NUL. If it is, deduct 1 from the prefix.

This information and assuming the shlwapi implementation works with buffers of MAX_SIZE length and is similar to what it is in Wine or ReactOS (https://doxygen.reactos.org/de/dff/dll_2win32_2shlwapi_2path_8c_source.html) seems to kind of explain the undefined behavior that you are seeing in testing.

As for a .NET solution, the easiest way (may not be the best) I can think of is to use System.Uri

Uri path1 = new Uri(@"c:\lvl1\lvl2\");
Uri path2 = new Uri(@"c:\lvl1\lvl3\file1.txt");
Uri diff = path1.MakeRelativeUri(path2);
// Uri will switch to forward slashes, so to fix that...
string relPath = 
Uri.UnescapeDataString(diff.OriginalString).Replace("/",@"\");

Or ofcourse you can implement something based on the .NET Core source of Path.GetRelativePath

jester
  • 3,491
  • 19
  • 30
3

.NET 4.6.2 solution

Use the \\?\C:\Verrrrrrrrrrrry long path syntax as described here.

There is also a great blog post about this

In general the biggest issue I have is with Shared folders over the web. The rest is fine.

Older .NET versions

If you are using an older version of .NET you can check out this Win32 API function, you will need P/Invoke for this.

The Windows API has many functions that also have Unicode versions to permit an extended-length path for a maximum total path length of 32,767 characters

Also you can check out this SO question, which is very similar to yours.
How to deal with files with a name longer than 259 characters?

Madara's Ghost
  • 172,118
  • 50
  • 264
  • 308
Gilad
  • 6,437
  • 14
  • 61
  • 119
  • 4
    but all this unrelated to `PathRelativePathTo` – RbMm Nov 17 '19 at 17:00
  • 2
    How is this answering the question? – jester Nov 17 '19 at 20:41
  • This is exactly the same idea behind all of Path functionality. – Gilad Nov 17 '19 at 21:10
  • 1
    no any *all of Path functionality* concrete `PathRelativePathTo` not affected by any prefix. this is pure lexical parsing api, hardcoded to 260 chars limit. also even \\ vs / different - break it – RbMm Nov 18 '19 at 14:59
  • There's even a comment stating it doesn't work: https://chat.stackoverflow.com/transcript/message/47826723#47826723 – user1781290 Nov 18 '19 at 16:27
3

I decided to use a port of the dotnet/corefx Path.GetRelativePath method.

The following code was adapted from the following sources. Read the comments in the code where I list any adjustments or workarounds that I used:

My goal in adapting the code was to

  • make as few modifications as possible (Noted in code comments any modifications made)
  • Keep class structure the same as in the original source
  • Only include methods/properties that were needed to implement the method GetRelativePath

Code

using System;
using System.Diagnostics;
using System.Runtime.CompilerServices;
using System.Text;
using static System.IO.Path;

static class PathExtension
{
    // Port of .net 3.0 Path.GetRelativePath (Windows version)
    // https://learn.microsoft.com/en-us/dotnet/api/system.io.path.getrelativepath?view=netcore-3.0
    // 
    // Adapted from:
    // https://github.com/dotnet/corefx/blob/b123ba4b9107c73cbc02010dc1ee78eb8ffccb93/src/Common/src/CoreLib/System/IO/Path.cs
    // https://github.com/dotnet/corefx/blob/4a7075f188b5777ccb519f2af9b8a284f4383357/src/Common/src/CoreLib/System/IO/Path.Windows.cs
    //
    // Notes:
    // * I didn't have access to ReadOnlySpan<T> nor .AsSpan(), so I removed them.  I just used regular string instead.
    // * I hard coded some resource strings (from exceptions)
    // * Replaced ValueStringBuild with StringBuilder

    /// <summary>
    /// Create a relative path from one path to another. Paths will be resolved before calculating the difference.
    /// Default path comparison for the active platform will be used (OrdinalIgnoreCase for Windows or Mac, Ordinal for Unix).
    /// </summary>
    /// <param name="relativeTo">The source path the output should be relative to. This path is always considered to be a directory.</param>
    /// <param name="path">The destination path.</param>
    /// <returns>The relative path or <paramref name="path"/> if the paths don't share the same root.</returns>
    /// <exception cref="ArgumentNullException">Thrown if <paramref name="relativeTo"/> or <paramref name="path"/> is <c>null</c> or an empty string.</exception>
    public static string GetRelativePath(string relativeTo, string path)
    {
        return GetRelativePath(relativeTo, path, StringComparison);
    }

    private static string GetRelativePath(string relativeTo, string path, StringComparison comparisonType)
    {
        if (relativeTo == null)
            throw new ArgumentNullException(nameof(relativeTo));

        if (PathInternal.IsEffectivelyEmpty(relativeTo.AsSpan()))
            throw new ArgumentException(SR.Arg_PathEmpty, nameof(relativeTo));

        if (path == null)
            throw new ArgumentNullException(nameof(path));

        if (PathInternal.IsEffectivelyEmpty(path.AsSpan()))
            throw new ArgumentException(SR.Arg_PathEmpty, nameof(path));

        Debug.Assert(comparisonType == StringComparison.Ordinal || comparisonType == StringComparison.OrdinalIgnoreCase);

        relativeTo = GetFullPath(relativeTo);
        path = GetFullPath(path);

        // Need to check if the roots are different- if they are we need to return the "to" path.
        if (!PathInternal.AreRootsEqual(relativeTo, path, comparisonType))
            return path;

        int commonLength = PathInternal.GetCommonPathLength(relativeTo, path, ignoreCase: comparisonType == StringComparison.OrdinalIgnoreCase);

        // If there is nothing in common they can't share the same root, return the "to" path as is.
        if (commonLength == 0)
            return path;

        // Trailing separators aren't significant for comparison
        int relativeToLength = relativeTo.Length;
        if (EndsInDirectorySeparator(relativeTo.AsSpan()))
            relativeToLength--;

        bool pathEndsInSeparator = EndsInDirectorySeparator(path.AsSpan());
        int pathLength = path.Length;
        if (pathEndsInSeparator)
            pathLength--;

        // If we have effectively the same path, return "."
        if (relativeToLength == pathLength && commonLength >= relativeToLength) return ".";

        // We have the same root, we need to calculate the difference now using the
        // common Length and Segment count past the length.
        //
        // Some examples:
        //
        //  C:\Foo C:\Bar L3, S1 -> ..\Bar
        //  C:\Foo C:\Foo\Bar L6, S0 -> Bar
        //  C:\Foo\Bar C:\Bar\Bar L3, S2 -> ..\..\Bar\Bar
        //  C:\Foo\Foo C:\Foo\Bar L7, S1 -> ..\Bar

        // Original: var sb = new ValueStringBuilder(stackalloc char[260]);
        var sb = new StringBuilder(260);
        sb.EnsureCapacity(Math.Max(relativeTo.Length, path.Length));

        // Add parent segments for segments past the common on the "from" path
        if (commonLength < relativeToLength)
        {
            sb.Append("..");

            for (int i = commonLength + 1; i < relativeToLength; i++)
            {
                if (PathInternal.IsDirectorySeparator(relativeTo[i]))
                {
                    sb.Append(DirectorySeparatorChar);
                    sb.Append("..");
                }
            }
        }
        else if (PathInternal.IsDirectorySeparator(path[commonLength]))
        {
            // No parent segments and we need to eat the initial separator
            //  (C:\Foo C:\Foo\Bar case)
            commonLength++;
        }

        // Now add the rest of the "to" path, adding back the trailing separator
        int differenceLength = pathLength - commonLength;
        if (pathEndsInSeparator)
            differenceLength++;

        if (differenceLength > 0)
        {
            if (sb.Length > 0)
            {
                sb.Append(DirectorySeparatorChar);
            }

            sb.Append(path.AsSpan(commonLength, differenceLength));
        }

        return sb.ToString();
    }

    /// <summary>Returns a comparison that can be used to compare file and directory names for equality.</summary>
    internal static StringComparison StringComparison =>
        IsCaseSensitive ?
            StringComparison.Ordinal :
            StringComparison.OrdinalIgnoreCase;

    /// <summary>
    /// Returns true if the path ends in a directory separator.
    /// </summary>
    public static bool EndsInDirectorySeparator(string path) // Originally was public static bool EndsInDirectorySeparator(ReadOnlySpan<char> path)
        => path.Length > 0 && PathInternal.IsDirectorySeparator(path[path.Length - 1]);

    #region Resources
    // From https://github.com/dotnet/corefx/blob/c390ce7df50252e11f5d322276e9d19e046d1332/src/Microsoft.IO.Redist/src/Resources/Strings.resx

    static class SR
    {
        public static string Arg_PathEmpty => "The path is empty.";
    }
    #endregion Resources

    #region Path.Windows 
    // Code from 
    // https://github.com/dotnet/corefx/blob/4a7075f188b5777ccb519f2af9b8a284f4383357/src/Common/src/CoreLib/System/IO/Path.Windows.cs

    // https://github.com/dotnet/corefx/blob/4a7075f188b5777ccb519f2af9b8a284f4383357/src/Common/src/CoreLib/System/IO/Path.Windows.cs#L235
    /// <summary>Gets whether the system is case-sensitive.</summary>
    internal static bool IsCaseSensitive => false;

    #endregion Path.Windows

    #region Workarounds

    // Note, this is here just to cause all .AsSpan() calls to return a string since I don't have access to ReadOnlySpan<char>
    // https://learn.microsoft.com/en-us/dotnet/api/microsoft.extensions.primitives.stringsegment.asspan?view=dotnet-plat-ext-3.0
    static string AsSpan(this string s)
    {
        return s;
    }

    // Note, this is here just to cause all .AsSpan() calls to return a string since I don't have access to ReadOnlySpan<char>
    // https://learn.microsoft.com/en-us/dotnet/api/system.memoryextensions.asspan?view=netcore-3.0#System_MemoryExtensions_AsSpan_System_String_System_Int32_System_Int32_
    static string AsSpan(this string s, int startIndex, int length)
    {
        return s.Substring(startIndex, length);
    }


    #endregion Workarounds

    // Code from 
    // https://github.com/dotnet/corefx/blob/b123ba4b9107c73cbc02010dc1ee78eb8ffccb93/src/Common/src/CoreLib/System/IO/PathInternal.cs
    // https://github.com/dotnet/corefx/blob/b123ba4b9107c73cbc02010dc1ee78eb8ffccb93/src/Common/src/CoreLib/System/IO/PathInternal.Windows.cs
    static class PathInternal
    {
        /// <summary>
        /// Returns true if the two paths have the same root
        /// </summary>
        internal static bool AreRootsEqual(string first, string second, StringComparison comparisonType)
        {
            int firstRootLength = GetRootLength(first.AsSpan());
            int secondRootLength = GetRootLength(second.AsSpan());

            return firstRootLength == secondRootLength
                && string.Compare(
                    strA: first,
                    indexA: 0,
                    strB: second,
                    indexB: 0,
                    length: firstRootLength,
                    comparisonType: comparisonType) == 0;
        }

        #region PathInternal.Windows
        // Code from https://github.com/dotnet/corefx/blob/b123ba4b9107c73cbc02010dc1ee78eb8ffccb93/src/Common/src/CoreLib/System/IO/PathInternal.Windows.cs

        // \\?\, \\.\, \??\
        internal const int DevicePrefixLength = 4;

        // \\
        internal const int UncPrefixLength = 2;

        // \\?\UNC\, \\.\UNC\
        internal const int UncExtendedPrefixLength = 8;

        /// <summary>
        /// Returns true if the given character is a valid drive letter
        /// </summary>
        internal static bool IsValidDriveChar(char value)
        {
            return (value >= 'A' && value <= 'Z') || (value >= 'a' && value <= 'z');
        }

        /// <summary>
        /// True if the given character is a directory separator.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        internal static bool IsDirectorySeparator(char c)
        {
            return c == DirectorySeparatorChar || c == AltDirectorySeparatorChar;
        }

        /// <summary>
        /// Returns true if the path uses the canonical form of extended syntax ("\\?\" or "\??\"). If the
        /// path matches exactly (cannot use alternate directory separators) Windows will skip normalization
        /// and path length checks.
        /// </summary>
        internal static bool IsExtended(string path) // Original was internal static bool IsExtended(ReadOnlySpan<char> path)
        {
            // While paths like "//?/C:/" will work, they're treated the same as "\\.\" paths.
            // Skipping of normalization will *only* occur if back slashes ('\') are used.
            return path.Length >= DevicePrefixLength
                && path[0] == '\\'
                && (path[1] == '\\' || path[1] == '?')
                && path[2] == '?'
                && path[3] == '\\';
        }

        /// <summary>
        /// Returns true if the path uses any of the DOS device path syntaxes. ("\\.\", "\\?\", or "\??\")
        /// </summary>
        internal static bool IsDevice(string path) // Original was: internal static bool IsDevice(ReadOnlySpan<char> path)
        {
            // If the path begins with any two separators is will be recognized and normalized and prepped with
            // "\??\" for internal usage correctly. "\??\" is recognized and handled, "/??/" is not.
            return IsExtended(path)
                ||
                (
                    path.Length >= DevicePrefixLength
                    && IsDirectorySeparator(path[0])
                    && IsDirectorySeparator(path[1])
                    && (path[2] == '.' || path[2] == '?')
                    && IsDirectorySeparator(path[3])
                );
        }

        /// <summary>
        /// Returns true if the path is a device UNC (\\?\UNC\, \\.\UNC\)
        /// </summary>
        internal static bool IsDeviceUNC(string path) // Original was: internal static bool IsDeviceUNC(ReadOnlySpan<char> path) 
        {
            return path.Length >= UncExtendedPrefixLength
                && IsDevice(path)
                && IsDirectorySeparator(path[7])
                && path[4] == 'U'
                && path[5] == 'N'
                && path[6] == 'C';
        }

        /// <summary>
        /// Gets the length of the root of the path (drive, share, etc.).
        /// </summary>
        internal static int GetRootLength(string path) // Note: original was internal static int GetRootLength(ReadOnlySpan<char> path)

        {
            int pathLength = path.Length;
            int i = 0;

            bool deviceSyntax = IsDevice(path);
            bool deviceUnc = deviceSyntax && IsDeviceUNC(path);

            if ((!deviceSyntax || deviceUnc) && pathLength > 0 && IsDirectorySeparator(path[0]))
            {
                // UNC or simple rooted path (e.g. "\foo", NOT "\\?\C:\foo")
                if (deviceUnc || (pathLength > 1 && IsDirectorySeparator(path[1])))
                {
                    // UNC (\\?\UNC\ or \\), scan past server\share

                    // Start past the prefix ("\\" or "\\?\UNC\")
                    i = deviceUnc ? UncExtendedPrefixLength : UncPrefixLength;

                    // Skip two separators at most
                    int n = 2;
                    while (i < pathLength && (!IsDirectorySeparator(path[i]) || --n > 0))
                        i++;
                }
                else
                {
                    // Current drive rooted (e.g. "\foo")
                    i = 1;
                }
            }
            else if (deviceSyntax)
            {
                // Device path (e.g. "\\?\.", "\\.\")
                // Skip any characters following the prefix that aren't a separator
                i = DevicePrefixLength;
                while (i < pathLength && !IsDirectorySeparator(path[i]))
                    i++;

                // If there is another separator take it, as long as we have had at least one
                // non-separator after the prefix (e.g. don't take "\\?\\", but take "\\?\a\")
                if (i < pathLength && i > DevicePrefixLength && IsDirectorySeparator(path[i]))
                    i++;
            }
            else if (pathLength >= 2
                && path[1] == VolumeSeparatorChar
                && IsValidDriveChar(path[0]))
            {
                // Valid drive specified path ("C:", "D:", etc.)
                i = 2;

                // If the colon is followed by a directory separator, move past it (e.g "C:\")
                if (pathLength > 2 && IsDirectorySeparator(path[2]))
                    i++;
            }

            return i;
        }

        /// <summary>
        /// Gets the count of common characters from the left optionally ignoring case
        /// </summary>
        internal static unsafe int EqualStartingCharacterCount(string first, string second, bool ignoreCase)
        {
            if (string.IsNullOrEmpty(first) || string.IsNullOrEmpty(second)) return 0;

            int commonChars = 0;

            fixed (char* f = first)
            fixed (char* s = second)
            {
                char* l = f;
                char* r = s;
                char* leftEnd = l + first.Length;
                char* rightEnd = r + second.Length;

                while (l != leftEnd && r != rightEnd
                    && (*l == *r || (ignoreCase && char.ToUpperInvariant(*l) == char.ToUpperInvariant(*r))))
                {
                    commonChars++;
                    l++;
                    r++;
                }
            }

            return commonChars;
        }

        /// <summary>
        /// Get the common path length from the start of the string.
        /// </summary>
        internal static int GetCommonPathLength(string first, string second, bool ignoreCase)
        {
            int commonChars = EqualStartingCharacterCount(first, second, ignoreCase: ignoreCase);

            // If nothing matches
            if (commonChars == 0)
                return commonChars;

            // Or we're a full string and equal length or match to a separator
            if (commonChars == first.Length
                && (commonChars == second.Length || IsDirectorySeparator(second[commonChars])))
                return commonChars;

            if (commonChars == second.Length && IsDirectorySeparator(first[commonChars]))
                return commonChars;

            // It's possible we matched somewhere in the middle of a segment e.g. C:\Foodie and C:\Foobar.
            while (commonChars > 0 && !IsDirectorySeparator(first[commonChars - 1]))
                commonChars--;

            return commonChars;
        }

        /// <summary>
        /// Returns true if the path is effectively empty for the current OS.
        /// For unix, this is empty or null. For Windows, this is empty, null, or
        /// just spaces ((char)32).
        /// </summary>
        /// 
        internal static bool IsEffectivelyEmpty(string path)
        {
            // Note, see the original version below
            return string.IsNullOrWhiteSpace(path);
        }

        // Note: here's the original version.  I've replaced it with the version above that just uses string
        // 
        //internal static bool IsEffectivelyEmpty(ReadOnlySpan<char> path)
        //{
        //    if (path.IsEmpty)
        //        return true;

        //    foreach (char c in path)
        //    {
        //        if (c != ' ')
        //            return false;
        //    }
        //    return true;
        //}

        #endregion PathInternal.Windows
    }
}
Matt Smith
  • 17,026
  • 7
  • 53
  • 103
2

at How can one get an absolute or normalized file path in .NET? I see

public static string NormalizePath(string path)
{
    return Path.GetFullPath(new Uri(path).LocalPath)
           .TrimEnd(Path.DirectorySeparatorChar, Path.AltDirectorySeparatorChar)
           .ToUpperInvariant();
}

so I'd start with that to normalize the two paths (also see https://blogs.msdn.microsoft.com/jeremykuhne/2016/04/21/path-normalization/ in case that covers more cases)

then I'd split them into arrays/lists of subpaths (say with one of the methods from How does one extract each folder name from a path?)

from there I'd find the max N first parts that are common.

then I'd subtract N from first path's count of parts C, aka C-N to get how many ..\ I need to add to the first path in order to get back to the common path.

finally I'd add the rest of the toPath after having removed the first N items from it and return the resulting path

Guess you could also do that (to avoid extra storage) with string parsing (without splitting in lists) once you've found the normalized paths. The idea would be that you'd find the common string prefix and then trim the last part of it if the common part didn't end up with the path separator (since that would be a coincidental extra common part, e.g. c:\a\test1 and c:\a\test2 have common path c:\a\ and not c:\a\test as you'd get with a simple common prefix string extraction).

Alternatively you could use an algorithm that returns character indexes for each \ working up the two normalized paths at the same time in a loop (one step on each) so that you don't need to store something extra. The logic would be similar to the one described above.

George Birbilis
  • 2,782
  • 2
  • 33
  • 35