EncodedStringText.Decode does not handle memory stream with non-zero offset correctly

[nguerrera]

Version Used: 2.0 (Synced to master recently)

There's an optimization for memory stream in EncodedStringText.Decode, which retrieves the backing byte array if it can. This optimization does not account for a MemoryStream that was constructed with a non-zero offset. The fix should use MemoryStream.TryGetBuffer instead of MemoryStream.GetBuffer in order to get an array segment with the appropriate offset.

Steps to Reproduce:
Compile and run the following program.

(I'm not sure if the root issue in EncodedStringText.Decode can affect anything other than scripting. This was the first code path I found where the user could pass an arbitrary stream.)

using Microsoft.CodeAnalysis;
using Microsoft.CodeAnalysis.CSharp.Scripting;
using Microsoft.CodeAnalysis.Scripting;
using System.Diagnostics;
using System.IO;
using System.Linq;
using System.Text;

class Resolver : SourceReferenceResolver
{
    public override bool Equals(object other) => ReferenceEquals(this, other);
    public override int GetHashCode() => 42;
    public override string ResolveReference(string path, string baseFilePath) => path;
    public override string NormalizePath(string path, string baseFilePath) => path;

    public override Stream OpenRead(string resolvedPath)
    {
        // Make an ASCII text buffer with Hello World script preceded by padding Qs
        const int padding = 42;
        string text = @"System.Console.WriteLine(""Hello World!"");";
        byte[] bytes = Enumerable.Repeat((byte)'Q', text.Length + padding).ToArray();
        Encoding.ASCII.GetBytes(text, 0, text.Length, bytes, padding);

        // Make a stream over the program portion, skipping the Qs.
        var stream = new MemoryStream(
            bytes,
            padding,
            text.Length,
            writable: false,
            publiclyVisible: true);

        // sanity check that reading entire stream gives us back our text.
        using (var streamReader = new StreamReader(
            stream,
            Encoding.ASCII,
            detectEncodingFromByteOrderMarks: false,
            bufferSize: bytes.Length,
            leaveOpen: true))
        {
            var textFromStream = streamReader.ReadToEnd();
            Debug.Assert(textFromStream == text);
        }

        stream.Position = 0;
        return stream;
    }
}

class Program
{
    static void Main(string[] args)
    {
        var options = ScriptOptions.Default.WithSourceResolver(new Resolver());
        var script = CSharpScript.Create(@"#load ""foo.cs""", options);
        var result = script.RunAsync().Result;
    }
}

Expected Behavior:
Script runs and prints hello world.

Actual Behavior:
Unhandled Exception: Microsoft.CodeAnalysis.Scripting.CompilationErrorException: foo.cs(1,1): error CS0103: The name 'QQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQ' does not exist in the current context

[nguerrera]

#12353 includes a fix for this.

[svick]

@nguerrera It does? I don't see anything similar to #12408 in #12353 and running the above code when referencing code from #12353 still throws the exception for me.

[nguerrera]

@svick You're right. Sorry, I commented on the wrong bug. Thanks for pointing it out.

[TyOverby]

This bug was fixed in 418868bc

I can confirm that your program compiles and runs correctly now.

I'll add a regression test.