Decoding Base64 in C# sometimes gives incorrect result with one extra padding character

Question

I stumbled upon a strange behavior of Convert.FromBase64String in .NET 4.7.2. Normally it would throw an exception when the padding is not correct. But I found a case where adding another padding character produces an incorrect result instead of an exception.

var correct = Convert.FromBase64String("YWE=");

In this case correct is [97, 97] or "aa" in a string form. But when I add another =:

var incorrect = Convert.FromBase64String("YWE==");

instead of getting an exception I get one byte less and incorrect is [88] or "X" is a string form.

Weird. Is this a bug and it should be reported? Or is it a known/documented behavior? I couldn't find any references to this.

Compare to Ruby. This evaluates to "aa":

Base64.strict_decode64 "YWE="

And this raises an exception:

Base64.strict_decode64 "YWE=="

ArgumentError: invalid base64
from /usr/local/Cellar/ruby/2.6.1/lib/ruby/2.6.0/base64.rb:74:in `unpack1'

Answer 1

Looks like this was a bug: https://github.com/dotnet/corefx/issues/30793 It's fixed in .NET Core but still present in .NET Framework up to and including 4.8.

This code should abort with exception instead of printing 1:

using System;

public class Program
{
    public static void Main()
    {
        Console.WriteLine(Convert.FromBase64String("YWE==").Length);
    }
}

Here's a test: https://dotnetfiddle.net/x2X9CT