Implement optimised MySqlDataReader.GetX methods

[bgrainger]

trafficstars

MySqlDataReader.GetInt32(int ordinal) is essentially implemented as return (int) GetValue(ordinal);

This boxes the int as an object then unboxes it. It would be more efficient to return the int value directly and would eliminate allocations. (Coupled with 00199a0248bf436dd77e3f517d12a8fde2678d63 this would also optimise GetFieldValue<int>.)

For the text protocol, it might be possible to implement this without significant code duplication if each GetX method just needs to check for an integral column type, parse the text, and check for overflow.

A quick-and-dirty implementation yielded the following benchmark results:

.NET Core 3.0:

Method	Mean	Error	StdDev	Median	Gen 0	Gen 1	Gen 2	Allocated
GetInt32	21.21 ns	0.4360 ns	0.6112 ns	21.06 ns	-	-	-	-
GetValue	30.96 ns	0.7450 ns	1.4878 ns	30.56 ns	0.0051	-	-	24 B
GetFieldValue	26.04 ns	0.5446 ns	1.0228 ns	25.57 ns	-	-	-	-

Before, GetInt32 and GetFieldValue<int> were 31-37ns and also allocating.

.NET 4.8

Method	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated
GetInt32	54.19 ns	1.0760 ns	1.1513 ns	-	-	-	-
GetValue	57.13 ns	1.1467 ns	1.6076 ns	0.0050	0.0001	-	24 B

The improvement is less pronounced here, with the method taking ~59ns before the change.

[bgrainger]

Optimising GetInt32 specifically would be useful for the TechEmpower Framework Benchmarks: https://github.com/TechEmpower/FrameworkBenchmarks/blob/a94705893dd6a0e7e8312c79d0e4dff7ba8eb990/frameworks/CSharp/aspnetcore/PlatformBenchmarks/Data/RawDb.cs#L49-L50

[bgrainger]

For the text protocol, it might be possible to implement this without significant code duplication if each GetX method just needs to check for an integral column type, parse the text, and check for overflow.

It looks like naively extending the current code might result in quite a bit of duplication and potential inconsistency between different access methods. A different approach may be needed.

[ugavnholt]

Another related optimization, could be the ability to access the buffer of the reader, to be able to use JsonSerializer.DeserializeAsync<JsonDocument> directly on the buffer of the reader.

[bgrainger]

access the buffer of the reader

If I understand correctly, that sounds like https://github.com/dotnet/runtime/issues/24899. The current best workaround is to call GetStream, which will return a MemoryStream that reads directly from the buffer (https://github.com/mysql-net/MySqlConnector/issues/592).