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

 private char[] data = " DOCTYPE abc".ToCharArray();

[Benchmark(Baseline = true)]
public bool Before() => StrEqual(data, 1, 7, "DOCTYPE");

[Benchmark]
public bool After() => StrEqual2(data, 1, 7, "DOCTYPE");

internal static bool StrEqual(char[] chars, int strPos1, int strLen1, string str2)
{
    if (strLen1 != str2.Length)
    {
        return false;
    }

    Debug.Assert(chars != null);

    int i = 0;
    while (i < strLen1 && chars[strPos1 + i] == str2[i])
    {
        i++;
    }

    return i == strLen1;
}

internal static bool StrEqual2(char[] chars, int strPos1, int strLen1, string str2)
{
    if (strLen1 != str2.Length)
    {
        return false;
    }

    Debug.Assert(chars != null);

    return chars.AsSpan(strPos1, strLen1).SequenceEqual(str2.AsSpan());
}

I think changes are ready for review. Failing tests are not related with changes.

It looks good. Do we have tests that appropriately exercise these code paths? e.g. If you do a code coverage run, are all of these hit?

All that I know is that on first commit several tests failed.
When I try VS "Analyze Code Coverage" menu from tests - cannot see code coverage of System.Xml.dll

If somebody tell me how to receive results, I can check.

@stephentoub If possible to ask something not related to this PR?
I saw that there is NameTable class which is behaving as standard HashSet, but can search by string or char[].
My attempt was to replace it with HashSet<Node> and play with Node's equal methods.
It works, but when I benchmark - it says regression with ~ 30%.
The question is : Do you know any other trick that can squeeze few more nanoseconds :)

Here it is how my table looks like :

public class ImprovedNameTable
	{
		private readonly struct Node : IEquatable<Node>
		{
			public Node(string data)
			{
				Data = data;
				_dataAsArray = null;
				_start = 0;
				_len = 0;
			}

			//Search ctor only
			public Node(char[] dataAsArray, int start, int len)
			{
				Data = null;
				_dataAsArray = dataAsArray;
				_start = start;
				_len = len;
			}

			public string Data { get; }

			private readonly char[] _dataAsArray;
			private readonly int _start;
			private readonly int _len;

			public bool Equals(Node other)
			{
				return other.Data?.Equals(Data) ?? other._dataAsArray.AsSpan(other._start, other._len).SequenceEqual(Data.AsSpan());
			}

			public override bool Equals(object obj)
			{
				return obj is Node n && Equals(n);
			}

			public override int GetHashCode()
			{
				return string.GetHashCode(Data != null ? Data.AsSpan() : _dataAsArray.AsSpan(_start, _len));
			}
		}

		private readonly HashSet<Node> _items = new HashSet<Node>();

		public string Add(string key)
		{
			ArgumentNullException.ThrowIfNull(key);

			int len = key.Length;
			if (len == 0)
			{
				return string.Empty;
			}

			var checkItem = new Node(key);
			if (_items.TryGetValue(checkItem, out var actualValue))
				return actualValue.Data;

			_items.Add(checkItem);

			return checkItem.Data;
		}

		public string Add(char[] key, int start, int len)
		{
			if (len == 0)
			{
				return string.Empty;
			}

			// Compatibility check to ensure same exception as previous versions
			// independently of any exceptions throw by the hashing function.
			// note that NullReferenceException is the first one if key is null.
			if (start >= key.Length || start < 0 || (long)start + len > (long)key.Length)
			{
				throw new IndexOutOfRangeException();
			}

			// Compatibility check for len < 0, just throw the same exception as new string(key, start, len)
			if (len < 0)
			{
				throw new ArgumentOutOfRangeException(nameof(len));
			}

			var checkItem = new Node(key, start, len);
			if (_items.TryGetValue(checkItem, out var actualValue))
				return actualValue.Data;

			checkItem = new Node(new string(key, start, len));
			_items.Add(checkItem);

			return checkItem.Data;
		}
	}

When I try VS "Analyze Code Coverage" menu from tests - cannot see code coverage of System.Xml.dll

dotnet build /t:test /p:Coverage=true

inside of the relevant test directory. Unfortunately the System.Private.Xml tests are a bit of a mess and are spread out over many test projects (I'd really like to see us consolidate them all into a single project, but that's not your problem 😄). @krwq, which is the right test project for this functionality?

The question is : Do you know any other trick that can squeeze few more nanoseconds :)

I'd need to profile it to see where the time is being spent :) But for starters in the case of it needing to add to the table, it's doing two lookups, one in TryGetValue and one in Add; I'm not sure what your test is actually exercising, but if it's the add path, I'd start there and find a way to reduce that to a single lookup rather than two.

which is the right test project for this functionality?

I've check code coverage with System.Private.Xml\tests\XmlSerializer and System.Private.Xml\tests\XmlDocument and was able to find most of code paths covered, but for example changes in XmlTextReaderImpl.IncrementalRead are not executed there.

but for example changes in XmlTextReaderImpl.IncrementalRead are not executed there

Thanks for checking. Is it possible to add tests that ensures they are covered?

dotnet build /t:test /p:Coverage=true

I've tried to throw exception in

and then to run all tests in System.Private.Xml, but none of them failed. Looks like this is not covered or maybe some other module is using it. Unfortunately I am not so deep in the code to understand use case of it. If possible somebody to give me a hint of sample xml , and I will create tests for it. I suspect that this is something connected with binary data inside xml, or DataContract serializer is using it.

@stephentoub First of all - thanks for consolidating xml tests.
I've run them and verified that IncrementalRead() is not covered.

I will be glad to add tests, just need help from somebody to give me hint - when it is expected this method to be used.

I've tried all benchmarks with Xml in it's name from performance repo and here it is the result:

Statistics

Total: 69
Same: 71.01 %
Slower: 1.45 %
Faster: 18.84 %
Noise: 8.70 %
Unknown: 0.00 %

Statistics per Architecture

Statistics per Operating System

Statistics per Namespace

System.Xml.Linq.Perf_XElement.GetValue

MicroBenchmarks.Serializers.Xml_FromStream.DataContractSerializer_

MicroBenchmarks.Serializers.Xml_FromStream.XmlSerializer_

MicroBenchmarks.Serializers.Xml_FromStream.DataContractSerializer_

MicroBenchmarks.Serializers.Xml_ToStream.DataContractSerializer_

MicroBenchmarks.Serializers.Xml_FromStream.XmlSerializer_

MicroBenchmarks.Serializers.Xml_FromStream.XmlSerializer_

MicroBenchmarks.Serializers.Xml_ToStream.DataContractSerializer_

Microsoft.Extensions.Configuration.Xml.XmlConfigurationProviderBenchmarks.Load(FileName: "names.xml")

MicroBenchmarks.Serializers.Xml_FromStream.XmlSerializer_

MicroBenchmarks.Serializers.Xml_ToStream.DataContractSerializer_

System.Xml.Linq.Perf_XElementList.Enumerator

MicroBenchmarks.Serializers.Xml_ToStream.DataContractSerializer_

MicroBenchmarks.Serializers.Xml_FromStream.XmlSerializer_

MicroBenchmarks.Serializers.Xml_FromStream.XmlSerializer_

Microsoft.Extensions.Configuration.Xml.XmlConfigurationProviderBenchmarks.Load(FileName: "repeated.xml")

MicroBenchmarks.Serializers.Xml_FromStream.XmlSerializer_

Thanks, I think this looks good. @stephentoub any objection to merging without added coverage?

any objection to merging without added coverage?

How do we know it didn't break something?

just need help from somebody to give me hint - when it is expected this method to be used

If you start from https://source.dot.net/#q=incrementalread and click on IncrementalRead in the left list, that will bring you to the source for that method. Then click on the method name in the code pane on the right, and the pane on the left will show you all the call sites. You can then click on one of those to navigate the right pane to that usage, and continue that process to look for all the ways to get to this function.

@eiriktsarpalis, @stephentoub
XmlTextReader.IncrementalRead method was used by XmlTextReader.ReadChars.
I've added one test with sample xml found in the solution which contains a lot of different cases and validating output.
Now all changes are covered + we have better coverage.
Before(based on consolidating tests commit) :
+--------------------+--------+--------+--------+
| Module | Line | Branch | Method |
+--------------------+--------+--------+--------+
| System.Private.Xml | 52.68% | 43.27% | 60.92% |
+--------------------+--------+--------+--------+

Now:
+--------------------+--------+--------+--------+
| Module | Line | Branch | Method |
+--------------------+--------+--------+--------+
| System.Private.Xml | 52.84% | 43.44% | 61.01% |
+--------------------+--------+--------+--------+

Thanks @TrayanZapryanov!