[Desktop] Uri(Uri, String) throws NullReferenceException for certain relativeUri values

[walro]

Short reproduction from C# interactive in Visual Studio 2017 15.6.4 on Windows 10 1709 (I'd assume this runs under .NET 4.7.1):

var baseUri = new Uri("http://example.com");
var relativeUri = "ftp:%c3%b6";
new Uri(baseUri, relativeUri);

Object reference not set to an instance of an object.
  + System.Uri.CreateUriInfo(System.Uri.Flags)
  + System.Uri.EnsureUriInfo()
  + System.Uri.ParseRemaining()
  + System.Uri.EnsureParseRemaining()
  + System.Uri.CreateThis(string, bool, System.UriKind)
  + System.Uri.CreateUri(System.Uri, string, bool)

The exception type is unexpected, I would expect a UriFormatException to be thrown.

• Changing : to %3a works fine
• Changing %c3%b6 to a yields a better exception: System.UriFormatException: Invalid URI: The Authority/Host could not be parsed.

So there seems to be something going on with the UTF-8 decoding combined with special case when a protocol is at start of the string.

(The problem is the same under mono 5.4.1.6 on macOS)

[EDIT] Update code formatting by @karelz

[stephentoub]

With .NET Core, this results in:

Unhandled Exception: System.UriFormatException: Invalid URI: The Authority/Host could not be parsed.
   at System.Uri.CreateThis(String uri, Boolean dontEscape, UriKind uriKind)
   at System.Uri.CreateUri(Uri baseUri, String relativeUri, Boolean dontEscape)
   at System.Uri..ctor(Uri baseUri, String relativeUri)

as expected.

[davidsh]

cc: @rmkerr

[rmkerr]

This will also throw the expected exception on .NET Framework 4.7.2

System.UriFormatException occurred
  Message=Invalid URI: The Authority/Host could not be parsed.
  StackTrace:
   at System.Uri.CreateThis(String uri, Boolean dontEscape, UriKind uriKind)
   at System.Uri.CreateUri(Uri baseUri, String relativeUri, Boolean dontEscape)
   at System.Uri..ctor(Uri baseUri, String relativeUri)

This situation is actually quite interesting. For compatibility reasons, we usually allow creation of relative URIs when there is a scheme but no host. That's based on section 5.4.2 of the RFC:

   Some parsers allow the scheme name to be present in a relative
   reference if it is the same as the base URI scheme.  This is
   considered to be a loophole in prior specifications of partial URI
   [RFC1630].  Its use should be avoided but is allowed for backward
   compatibility.

      "http:g"        =  "http:g"         ; for strict parsers
                      /  "http://a/b/c/g" ; for backward compatibility

It looks like we're running into a situation where we're disallowing that behavior. I think it's worth looking into why that is. I think we should keep this issue open to track that investigation, but leave it in future for now.

[rmkerr]

In .NET Core Preview 2 and .NET Framework 4.7.2 this throws the expected UriFormatException. Closing as this appears to be fixed.