System.Xml.XmlTextReaderImpl could possibly benefit from Span

[johnthcall]

Description

In the Azure SDK ecosystem errors are serialized by the server as XML and parsed into XDocument, In the stack trace below the string message is wrapped into a TextReader which is then passed down to the XmlTextReaderImpl where a buffer is allocated. Even in the case when InitTextReaderInput is called with a string source instead of a TextReader the entire string is copied to a new char[] with a (char)0 ending.

It seems that if span can be leveraged that significant performance could be gained. If no one sees a clear path for this, no worries just want to make sure there isn't any low hanging fruit here.

Configuration

• Which version of .NET is the code running on? 5.0.9
• What OS version, and what distro if applicable? Windows Server 2019
• What is the architecture (x64, x86, ARM, ARM64)? x64

Data

clr.dll!JIT_NewArr1
System.Xml.ni.dll!System.Xml.XmlTextReaderImpl.InitTextReaderInput(System.String, System.Uri, System.IO.TextReader)$##6000B97
System.Xml.ni.dll!System.Xml.XmlTextReaderImpl.FinishInitTextReader()$##6000B10
System.Xml.ni.dll!System.Xml.XmlReaderSettings.CreateReader(System.IO.TextReader, System.String, System.Xml.XmlParserContext)$##6000A31
System.Xml.Linq.ni.dll!System.Xml.Linq.XDocument.Parse(System.String, System.Xml.Linq.LoadOptions)$##60001A0
Azure.Core.Pipeline.StorageClientDiagnostics!ExtractFailureContent

Analysis

I initially tried altering the following by verifying that 0 ending array is required and making it a direct copy of the string does cause two unit tests to break.

runtime/src/libraries/System.Private.Xml/src/System/Xml/Core/XmlTextReaderImpl.cs

Lines 3058 to 3061 in 918aaec

	_ps.chars = new char[len + 1];
	str.CopyTo(0, _ps.chars, 0, str.Length);
	_ps.charsUsed = len;
	_ps.chars[len] = (char)0;

If the 0 ending array is strictly required then a copy of the full array will always be required.

I'm happy to contribute changes here but don't know if it is feasible or the best place to start.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-xml
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Description

In the Azure SDK ecosystem errors are serialized by the server as XML and parsed into XDocument, In the stack trace below the string message is wrapped into a TextReader which is then passed down to the XmlTextReaderImpl where a buffer is allocated. Even in the case when InitTextReaderInput is called with a string source instead of a TextReader the entire string is copied to a new char[] with a (char)0 ending.

It seems that if span can be leveraged that significant performance could be gained. If no one sees a clear path for this, no worries just want to make sure there isn't any low hanging fruit here.

Configuration

• Which version of .NET is the code running on? 5.0.9
• What OS version, and what distro if applicable? Windows Server 2019
• What is the architecture (x64, x86, ARM, ARM64)? x64

Data

clr.dll!JIT_NewArr1
System.Xml.ni.dll!System.Xml.XmlTextReaderImpl.InitTextReaderInput(System.String, System.Uri, System.IO.TextReader)$##6000B97
System.Xml.ni.dll!System.Xml.XmlTextReaderImpl.FinishInitTextReader()$##6000B10
System.Xml.ni.dll!System.Xml.XmlReaderSettings.CreateReader(System.IO.TextReader, System.String, System.Xml.XmlParserContext)$##6000A31
System.Xml.Linq.ni.dll!System.Xml.Linq.XDocument.Parse(System.String, System.Xml.Linq.LoadOptions)$##60001A0
Azure.Core.Pipeline.StorageClientDiagnostics!ExtractFailureContent

Analysis

I initially tried altering the following by verifying that 0 ending array is required and making it a direct copy of the string does cause two unit tests to break.

runtime/src/libraries/System.Private.Xml/src/System/Xml/Core/XmlTextReaderImpl.cs

Lines 3058 to 3061 in 918aaec

	_ps.chars = new char[len + 1];
	str.CopyTo(0, _ps.chars, 0, str.Length);
	_ps.charsUsed = len;
	_ps.chars[len] = (char)0;

If the 0 ending array is strictly required then a copy of the full array will always be required.

Author:	johnthcall
Assignees:	-
Labels:	area-System.Xml, tenet-performance, untriaged
Milestone:	-

[eiriktsarpalis]

Hi @johnthcall, we'd be happy to consider a PR that adds spanification improvements to the XmlTextReader implementation. Would it be possible to also share benchmarks that demonstrate the perf improvements?