SourceText.GetTextChanges produces invalid edits after successive changes

[ajaybhargavb]

When working on formatting for .razor files, I noticed something odd. Under certain conditions, GetTextChanges would return invalid edits. I'm not sure exactly under what conditions. I was able to come up with a (somewhat) minimal repro.

Version Used:
Microsoft.CodeAnalysis 3.4.0
dotnet 3.1.101

Steps to Reproduce:

1. dotnet new console -o SourceTextBug
2. cd SourceTextBug
3. Replace Program.cs content with the following,

Program.cs

using System;
using Microsoft.CodeAnalysis.Text;

namespace SourceTextBug
{
    class Program
    {
        static void Main(string[] args)
        {
            var content = @"@functions{
    public class Foo
    {
void Method()
{
    
}
    }
}";
            Console.WriteLine("Initial:");
            Console.WriteLine(content);
            Console.WriteLine();

            var text = SourceText.From(content);
            var edits1 = new TextChange[]
            {
                new TextChange(new TextSpan(39, 0), "    "),
                new TextChange(new TextSpan(42, 0), "            "),
                new TextChange(new TextSpan(57, 0), "            "),
                new TextChange(new TextSpan(58, 0), "\r\n"),
                new TextChange(new TextSpan(64, 2), "        "),
                new TextChange(new TextSpan(69, 0), "    "),
            };
            var changedText = text.WithChanges(edits1);

            Console.WriteLine("After first set of edits:");
            Console.WriteLine(changedText.ToString());
            Console.WriteLine();

            var edits2 = new TextChange[]
            {
                new TextChange(new TextSpan(35, 4), string.Empty),
                new TextChange(new TextSpan(46, 4), string.Empty),
                new TextChange(new TextSpan(73, 4), string.Empty),
                new TextChange(new TextSpan(88, 0), "    "),
                new TextChange(new TextSpan(90, 4), string.Empty),
                new TextChange(new TextSpan(105, 4), string.Empty),
            };
            var changedText2 = changedText.WithChanges(edits2);

            Console.WriteLine("After second set of edits:");
            Console.WriteLine(changedText2.ToString());
            Console.WriteLine();

            Console.WriteLine("Diff of final text vs original:");
            var changes = changedText2.GetTextChanges(text);
            foreach (var change in changes)
            {
                Console.WriteLine(change);
            }
        }
    }
}

4. Add the following in the csproj,

<PackageReference Include="Microsoft.CodeAnalysis" Version="3.4.0" />`

5. dotnet run

Expected Behavior:
The set of TextChanges produced should be valid and not overlap.

Actual Behavior:
Overlapping changes are produced.

Output

Initial:
@functions{
    public class Foo
    {
void Method()
{

}
    }
}

After first set of edits:
@functions{
    public class Foo
        {
            void Method()
            {

            }
        }
}

After second set of edits:
@functions{
    public class Foo
    {
        void Method()
        {

        }
    }
}

Diff of final text vs original:
TextChange: { [35..39), "    " }
TextChange: { [42..42), "        " }
TextChange: { [57..57), "        " }
TextChange: { [58..58), "
" }
TextChange: { [64..66), "        }
 " }
TextChange: { [62..66), "" }
TextChange: { [69..69), "    " }
TextChange: { [73..77), "" }

Notice the two overlapping edits,
[64..66) and [62..66)

[jasonmalinowski]

FYI to @jinujoseph and @vatsalyaagrawal that this is potentially impacting ASP.NET's team to take on some ownership of the ASP.NET formatting code paths. I'm not sure if we really have an owner of this code or not. It was most recently touched in #39258 but it seems there's still some issues in it.

[ajaybhargavb]

Looks like this also happens with GetChangeRanges(). Calling this method after successive changes, sometimes provides bad ranges.