Runtime: UTF32Encoding and XmlTransformation when ByteOrderMark is set to false still contains the mark on .NET Core 2

[adamhathcock]

I've been updating Cake to be .NET Standard 2.0 and run on netcoreapp2.0 here: https://github.com/cake-build/cake/pull/1812

There was a battery of XML tests that were full framework only that I enabled for netstandard2.0 because they now exist on that footprint. There's one problem test that works still on net46/net461 but not on netcoreapp2.0

            [Fact]
            public void Should_Transform_Xml_String_And_Xsl_String_To_Result_String_With_Utf32Xml_Declaration()
            {
                // Given
                var xml = Resources.XmlTransformation_Xml;
                var xsl = Resources.XmlTransformation_Xsl;
                var settings = new XmlTransformationSettings
                {
                    Encoding = new UTF32Encoding(false, false, true)
                };

                // When
                var result = string.Concat(XmlTransformation.Transform(xsl, xml, settings).Take(39));

                // Then
                Assert.Equal("<?xml version=\"1.0\" encoding=\"utf-32\"?>", result);
            }

I can't tell if it's just an issue with UTF32Encoding (I don't believe so) or something else in the XML Transform.

[karelz]

Can you please isolate the test case from your infra? Make it a standalone app which compiles (incl. Resources). What is the output you get on .NET Core 2.0?

[adamhathcock]

Sorry yes. It was more tied to Cake than I realized:

https://github.com/adamhathcock/utf32bug

The error I get:

utf32bug.UnitTest1.Should_Transform_Xml_String_And_Xsl_String_To_Result_String_With_Utf32Xml_Declaration

Assert.Equal() Failure
          ↓ (pos 0)
Expected: <?xml version="1.0" encoding="utf-32"?>
Actual:   <?xml version="1.0" encoding="utf-32"?
          ↑ (pos 0)
   at utf32bug.UnitTest1.Should_Transform_Xml_String_And_Xsl_String_To_Result_String_With_Utf32Xml_Declaration() in /Users/adam/git/personal/utf32bug/UnitTest1.cs:line 48

Actual seems to have the BOM (the error message lines up better in Rider)

[krwq]

Thank you for repro @adamhathcock

[clarkis117]

@adamhathcock You're right, something in the XmlWriter.Create(result, settings); call is writing the BOM to the MemoryStream, in your repro on netcoreapp2.0.

[clarkis117]

@karelz I would like to pick up this issue. I have identified the bug and have a initial fix for it. However, I'm not sure it's the most elegant fix. I'm also not sure where the test(s) should go since the bug is in CoreLib.

[karelz]

Great, thanks for help! Assigned to you.
I would follow pattern of other tests for sibling methods. In general, most of the CoreFX tests are in CoreFX repo, even if the code lives in CoreCLR. @krwq @pjanotti will be able to say for sure.

[krwq]

@clarkis117, do not worry if it is not super elegant - we can give some recommendations during the code review - correctness is more important. Thank you for investigating this! (please cc us on the code review)