This is a branch cleaning up old `#if` clauses, `.csproj` bits, Linux build compatibility, and a few other tid-bits (broken out by commit).
In general: we're on 3.0 tooling and all previous cruft has been removed.
Hand coded version did not operate under the same circumstances as others.
- Add CommandBehavior.SingleResult and CommandBehavior.SingleRow. All other framworks does this under the hood.
- Run SqlCommand.Prepare() (perhaps it should be benchmarked as "Compiled"?)
- Use GetNullableValue because it's easier to read and slightly faster
- _table.Rows was incrementally added, which made the benchmark run slower the longer it executed. Added _table.Rows.Clear() to prevent the growth.
- Changed select to use * like other benchmarks
This is prepping a run we can drop into the README easily. And more info on allocations and such along with it. It's finishing off the benchmark port getting it into a good output state (something we can copy/paste). It removes Simple.Data because the package was debug and causes a lot of issues.
This doesn't output 100% like we want yet (fastest to slowest). BenchmarkDotNet joins summary results after ordering...we need to post-order that. I'm asking the maintainers how to do so.
Also note Belgrade was not parameterizing before, giving it a huge unfair advantage. This has been fixed to be apples:apples.
MySql.Data is a travesty at this point - moving to something more stable. Really we weren't testing Dapper with previous test failures, we were testing what mood MySql.Data and NuGet extraction are in and which phase of the moon is about.
Also speeds up the builds with 1 restore and 1 build. We can simplify here because everything is relevant (no wasted builds).
This uses the new non-parallel-at-the-end I added to xUnit in https://github.com/xunit/xunit/pull/1411 This way we can run any global test that are other-test-unfriendly much more easily.
OperationsPerInvoke is an informational, not instructional property.
What we actually want is the UnrollFactor. Moved into the job definition
to simplify things.
This adds a netstandard2.0 build to Dapper (and StrongName until v2),
adds a netstandard2.0 test lineup, restores most functionality (except
UDTs for SQL, e.g. .UdtTypeName isn't in netstandard2), and also breaks
things not actually implemented or implemented correctly in
netstandard2.0. Pushing this up so we can work through these with the
.NET teams.
Several things are still broken here (and fail tests):
- Parameter decimal values (not filed yet)
- DataTables as parameters
(https://github.com/dotnet/corefx/issues/19708)
- .GetSchemaTable() (not yet filed, implementation is completely missing
from SqlDataReader in CoreFX)
While the compile is clean, the above are runtime issues.
`SqlParameter.UdtTypeName` is a separate issue, that one just wasn't in
`netstandard2.0` at all...so we can't support UDTs there until it is.
See https://github.com/dotnet/corefx/issues/17126 for details.
This is based on BenchmarkDotNet (while preserving the legacy format
with minor improvements as well - legacy runs much faster). See #666 for
details. Not an ominus number at all.
Note: this code will get a bit simpler with BenchmarkDotNet categories,
see https://github.com/dotnet/BenchmarkDotNet/issues/248 for details.
Due to the way VS test works (by injecting an executable entry point), the performance tests needed to be factored out. I also organized all of our existing tests better along the way.
Actual code changes to Dapper itself are very minor, only formatting and documentation fixes (which we need many more of).
The build.ps1 script is upated to work, but note that <frameworkAssemblies> is not working in .nuspec inside the packages in the new .csproj system. I consider this to be a blocker. Issue is here: https://github.com/NuGet/Home/issues/4853
Nick Craver
