Don't allocate in BeginChunkBytes

[benaadams]

Use the BufferWriter's Span instead

Think allocations are up by factor of 1024 in the benchmark (i.e. x IterationCount for the pre)

                Method | DataLength |      Mean |          Op/s |  Gen 0 | Allocated |
 --------------------- |----------- |----------:|--------------:|-------:|----------:|
- WriteBeginChunkBytes |          0 |  33.30 ns |  30,030,302.5 | 0.7324 |     40 KB |
+ WriteBeginChunkBytes |          0 |   8.09 ns | 123,639,689.1 | 0.7324 |       0 B |
- WriteBeginChunkBytes |          1 |  34.48 ns |  28,998,176.5 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |          1 |   8.40 ns | 119,093,216.2 | 0.6714 |       0 B |
- WriteBeginChunkBytes |         16 |  34.07 ns |  29,348,584.0 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |         16 |   9.01 ns | 110,966,446.7 | 0.6714 |       0 B |
- WriteBeginChunkBytes |        256 |  45.12 ns |  22,164,899.8 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |        256 |  10.81 ns |  92,489,663.4 | 0.6714 |       0 B |
- WriteBeginChunkBytes |       4096 |  33.05 ns |  30,257,577.0 | 0.7324 |     40 KB |
+ WriteBeginChunkBytes |       4096 |  11.54 ns |  86,654,067.4 | 0.7324 |       0 B |
- WriteBeginChunkBytes |      65536 |  45.85 ns |  21,809,105.6 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |      65536 |  12.69 ns |  78,783,888.6 | 0.6714 |       0 B |
- WriteBeginChunkBytes |    1048576 |  33.94 ns |  29,459,649.1 | 0.7324 |     40 KB |
+ WriteBeginChunkBytes |    1048576 |  13.92 ns |  71,856,593.8 | 0.7324 |       0 B |
- WriteBeginChunkBytes |   16777216 |  33.30 ns |  30,028,526.6 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |   16777216 |  15.35 ns |  65,169,537.2 | 0.6714 |       0 B |

/cc @pakrym @davidfowl

[davidfowl]

@benaadams perf tests?

[benaadams]

perf tests?

Don't think BDN is dividing allocations by iterations, howevers:

                Method | DataLength |      Mean |          Op/s |  Gen 0 | Allocated |
 --------------------- |----------- |----------:|--------------:|-------:|----------:|
- WriteBeginChunkBytes |          0 |  33.30 ns |  30,030,302.5 | 0.7324 |     40 KB |
+ WriteBeginChunkBytes |          0 |   8.09 ns | 123,639,689.1 | 0.7324 |       0 B |
- WriteBeginChunkBytes |          1 |  34.48 ns |  28,998,176.5 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |          1 |   8.40 ns | 119,093,216.2 | 0.6714 |       0 B |
- WriteBeginChunkBytes |         16 |  34.07 ns |  29,348,584.0 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |         16 |   9.01 ns | 110,966,446.7 | 0.6714 |       0 B |
- WriteBeginChunkBytes |        256 |  45.12 ns |  22,164,899.8 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |        256 |  10.81 ns |  92,489,663.4 | 0.6714 |       0 B |
- WriteBeginChunkBytes |       4096 |  33.05 ns |  30,257,577.0 | 0.7324 |     40 KB |
+ WriteBeginChunkBytes |       4096 |  11.54 ns |  86,654,067.4 | 0.7324 |       0 B |
- WriteBeginChunkBytes |      65536 |  45.85 ns |  21,809,105.6 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |      65536 |  12.69 ns |  78,783,888.6 | 0.6714 |       0 B |
- WriteBeginChunkBytes |    1048576 |  33.94 ns |  29,459,649.1 | 0.7324 |     40 KB |
+ WriteBeginChunkBytes |    1048576 |  13.92 ns |  71,856,593.8 | 0.7324 |       0 B |
- WriteBeginChunkBytes |   16777216 |  33.30 ns |  30,028,526.6 | 0.6714 |     40 KB |
+ WriteBeginChunkBytes |   16777216 |  15.35 ns |  65,169,537.2 | 0.6714 |       0 B |

[pentp]

This could be changed to a ReadOnlySpan<byte> returning property (or inlined in the method) - it would be compiled as just static data and startHex will be a constant then. This would get rid of the static .cctor also.

var hex = new ReadOnlySpan<byte>(new byte[] {(byte)'0', (byte)'1', (byte)'2',/*...*/ (byte)'e', (byte)'f' });
ref var startHex = ref Unsafe.AsRef(in hex[0]);

Sharplab example

[benaadams]

2 unrelated errors
#6135

Microsoft.AspNetCore.FunctionalTests.Microsoft.AspNetCore.Tests.WebHostFunctionalTests.RunsInIISExpressInProcess
Assert.Equal() Failure
          ↓ (pos 0)
Expected: CreateDefaultBuilderApp
Actual:   <!DOCTYPE html>\r\n<html lang="en-US" ns···
          ↑ (pos 0)
   at Microsoft.AspNetCore.Tests.WebHostFunctionalTests.RunsInIISExpressInProcess()
   in /_/src/DefaultBuilder/test/Microsoft.AspNetCore.FunctionalTests/WebHostFunctionalTests.cs:line 179
--- End of stack trace from previous location where exception was thrown ---

#6136

ServerComparison.FunctionalTests.ResponseTests.ResponseFormats_Chunked(variant: 
 Server: IISExpress, TFM: netcoreapp3.0, Type: Portable, Arch: x64, ANCM: V2, Host: InProcess)
/chunked, chunked?
Expected: True
Actual:   (null)
   at ServerComparison.FunctionalTests.ResponseTests.CheckChunkedAsync(HttpClient client, ILogger logger)
    in /_/src/Servers/test/FunctionalTests/ResponseTests.cs:line 211
   at ServerComparison.FunctionalTests.ResponseTests.ResponseFormats(TestVariant variant, Func`3 scenario, String testName)
    in /_/src/Servers/test/FunctionalTests/ResponseTests.cs:line 125
--- End of stack trace from previous location where exception was thrown ---

[benaadams]

Opened issues for the errors #6135 and #6136

[pakrym]

While RunsInIISExpressInProcess constantly fails, I've never seen ResponseFormats_Chunked failing in other RP. Doesn't seem to be connected at all but I'll restart it again.

[pakrym]

I lied, https://github.com/aspnet/AspNetCore/pull/6119/checks?check_run_id=44328480 has it too.