Use built-in "CreateReadOnlySpanFromNullTerminated" on NET6-specific path

[stevenaw]

Adds a .NET6-specific implementation of PtrToStringUTF8 to use the built-in MemoryMarshal.CreateReadOnlySpanFromNullTerminated() function to satisfy the below existing inline comment in the source:

// TODO: Is there a built in / vectorized / better way to implement this?

This API was added in .NET6 for this purpose (when an interop call only returns a pointer and the length of the underlying null-terminated string is unknown) - https://github.com/dotnet/runtime/issues/40202. The internal source uses strlen() which internally delegates to SpanHelpers.IndexOfNullByte which is heavily optimized with vectorization or other special cases.

Performance is comparable. I have run the unit tests locally to validate the change but am unable to run the integration tests.

Benchmarks (using .NET8 runtime)

```pre | Method | Mean | Error | StdDev | Ratio | Gen0 | Gen1 | Allocated | Alloc Ratio | |--------- |---------:|---------:|---------:|------:|-------:|-------:|----------:|------------:| | Original | 12.05 ns | 0.110 ns | 0.103 ns | 1.00 | 0.0064 | 0.0000 | 40 B | 1.00 | | Updated | 11.70 ns | 0.042 ns | 0.040 ns | 0.97 | 0.0064 | - | 40 B | 1.00 | ``` ```csharp [MemoryDiagnoser] public class UnsafeIntPtrToStr { GCHandle handle; IntPtr strPtr; [GlobalSetup] public void Setup() { var strBytes = Encoding.UTF8.GetBytes("TestData"); byte[] strBytesNulTerminated = new byte[strBytes.Length + 1]; // initialized to all 0's. Array.Copy(strBytes, strBytesNulTerminated, strBytes.Length); handle = GCHandle.Alloc(strBytesNulTerminated, GCHandleType.Pinned); strPtr = handle.AddrOfPinnedObject(); } [Benchmark(Baseline = true)] public unsafe string Original() { #if NET6_0_OR_GREATER var bytes = MemoryMarshal.CreateReadOnlySpanFromNullTerminated((byte*)strPtr.ToPointer()); return Encoding.UTF8.GetString(bytes); #else // TODO: Is there a built in / vectorized / better way to implement this? byte* pTraverse = (byte*)strPtr; while (*pTraverse != 0) { pTraverse += 1; } var length = (int)(pTraverse - (byte*)strPtr); return Encoding.UTF8.GetString((byte*)strPtr.ToPointer(), length); #endif } [Benchmark] public unsafe string Updated() { #if NET6_0_OR_GREATER var bytes = MemoryMarshal.CreateReadOnlySpanFromNullTerminated((byte*)strPtr.ToPointer()); return Encoding.UTF8.GetString(bytes); #else // TODO: Is there a built in / vectorized / better way to implement this? byte* pTraverse = (byte*)strPtr; while (*pTraverse != 0) { pTraverse += 1; } var length = (int)(pTraverse - (byte*)strPtr); return Encoding.UTF8.GetString((byte*)strPtr.ToPointer(), length); #endif } [GlobalCleanup] public void Cleanup() { handle.Free(); strPtr = IntPtr.Zero; } } ``` ```

[cla-assistant[bot]]

All committers have signed the CLA.

[cla-assistant[bot]]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}