Modernize the build: BenchmarkDotNet, MinVer, one workflow, a net10.0 leg

[jbtule]

Steps 1 and 2 of #63, plus the net10.0 leg. Four commits, each self-contained.

The suite is green for the first time: 190 tests on net8.0 and on net10.0, 0 failed, 0 skipped — where master is 200 passed / 17 failed / 5 skipped. A CI=true dotnet build --no-incremental of the solution has no warnings and no errors.

1. Benchmarks, and the wall-clock tests are gone (1510c13)

Tests/UnitTestImpromptuInterface/SpeedTest.cs was 28 [Test] methods timing 500,000 iterations and asserting Assert.Less(impromptuTime, reflectionTime): red on a loaded machine, vacuous on a fast one, and "handled" in CI by a TestCategory!=Performance filter that did not match them.

All 28 measured Dynamitey — Dynamic.InvokeSet, CacheableInvocation, FastDynamicInvoke — and not one touched ActLike. They were left over from the 2013 split, so they are deleted rather than migrated; they belong to that repo if anywhere.

In their place Benchmarks/, measuring this library: creating a proxy, and reading, writing and calling through one, each against the same operation on a type that declares the interface, so a ratio reads as what the proxy adds.

	vs. a declared type
ActLike<IPoco>(), type already built	~130× (≈560 ns)
read a property through the proxy	~6× (≈5 ns)
write a property through the proxy	~1.7× (≈4 ns)
call a method through the proxy	~1.7× (≈13 ns)

MethodInfo.Invoke is ~4.4× and PropertyInfo.GetValue ~10× in the same run, so a proxy call is cheaper than reflection and creating one is not. The baseline accessors are MethodImpl(NoInlining) on purpose: inlined, the JIT elides them, the baseline measures 0 ns and every ratio is meaningless — that happened on the first run.

2. MinVer, one workflow, NuGet.config (932045c)

Version.props carried the prefix by hand and built the suffix from GITHUB_RUN_NUMBER, so a local build and a CI build of the same commit disagreed. MinVer reads the tags; MinVerMinimumMajorMinor is 8.1, so untagged builds read 8.1.0-alpha.<height>.

dotnet.yml + dotnet48.yml become one build.yml across ubuntu/macOS/windows, with the net47 leg only where the framework exists and a publish job for master. Warnings are errors when CI=true only.

Two bugs fixed in passing: the CopyPackage target resolved $(SolutionDir) to / and failed whenever the library was built on its own rather than through the solution; and UnitTestSupportLibrary targeted netstandard1.0 (NETSDK1215).

3. A net10.0 leg (46beee2)

net40;netstandard2.0;net10.0. Both existing legs stay — NuGet hands a .NET Framework consumer the net40 asset however new their framework is, so dropping it breaks real consumers. net10.0 is purely additive: a .NET 10 consumer gets it instead of falling back to netstandard2.0. Package verified to carry all three lib/ folders with the right dependency group each (net10.0 gets only Dynamitey; Microsoft.CSharp and System.Reflection.Emit are inbox there).

The new leg surfaced that binary serialization is obsolete from .NET 8 (SYSLIB0050/0051). ISerializable is on the public ActLikeProxy base that every emitted proxy derives from, so it is suppressed on that leg rather than compiled out — the public surface should not differ per target framework.

4. Linq tests, IronPython 3.4, and a critical advisory (f2b4cf9)

Same pattern as SpeedTest.cs: four of the nine Linq tests exercised Dynamic.Linq with no proxy involved. Deleted. The four kept are ActLike over a Dynamitey Linq proxy, two driven from IronPython — a proxy called through a foreign DLR binder, which nothing else here covers.

That unblocked IronPython 3.4, which drops DynamicLanguageRuntime 1.3.1 and with it System.Drawing.Common 4.7.0 — a critical advisory (GHSA-rxg9-xrhp-64gj) that only became visible once warnings are errors. The one test that still failed on 3.4 needed System.Func[System.Int32, System.Boolean] rather than System.Func[int, bool]: on IronPython 3 the Python int is arbitrary precision and no longer means Int32.

Not in this PR

Step 3 of #63 (the AnyUnit port) is proven on a local branch but waits on AnyUnit 1.3.0 reaching nuget.org. Step 4 (Tests.Wasm) needs that harness — though the capability question behind it is now answered: proxy emit works on browser-wasm under .NET 10, interpreted and AOT-published alike, which closed #43 and narrowed #48 to trimming.

🤖 Generated with Claude Code

https://claude.ai/code/session_016mwfq4oeZW8SiD4HjTHYdg

[Copilot]

Copilot review overview

🟡 Changes recommended

Unresolved critical CI and publishing failures, along with additional workflow and dependency issues, block approval.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 2 · 2 · 1

Open (5)

• Unconditional build fails on non-Windows .NET Framework targets · New
• Ubuntu packing fails for net40 target · New
• Missing pinned global.json causes non-reproducible SDK selection · New
• Deleting TimeIt breaks Clay test project · New
• Solution retains deleted Version.props entry · New

What changed in this PR

This PR modernizes versioning, benchmarking, CI, dependency management, and adds .NET 10 support.

Changes:

• Replaces flaky timing tests with BenchmarkDotNet benchmarks.
• Introduces MinVer and updates target frameworks and test dependencies.
• Consolidates CI, testing, and publishing into one workflow.

File	Summary
Version.props	Removed for MinVer migration. Nit (1 vote): remove the dangling solution-item entry.
Tests/​UnitTestSupportLibrary/​UnitTestSupportLibrary.csproj	Retargets the support library to netstandard2.0.
Tests/​UnitTestImpromptuInterface/​UnitTestImpromptuInterface.csproj	Adds .NET 10 and IronPython 3.4. Moderate (1 vote): the Clay project still references vulnerable IronPython 2.7. Nit (1 vote): clarify the dependency explanation.
Tests/​UnitTestImpromptuInterface/​Support/​SupportDefinitions.cs	Removes timing helpers. Moderate (2 votes): Clay tests still depend on TimeIt.
Tests/​UnitTestImpromptuInterface/​SpeedTest.cs	Removes obsolete wall-clock performance tests.
Tests/​UnitTestImpromptuInterface/​Linq.cs	Removes unrelated Dynamitey tests and updates retained proxy/LINQ coverage.
NuGet.config	Cleans package sources and cache settings.
ImpromptuInterface/​ImpromptuInterface.csproj	Adds net10.0 packaging and serialization warning handling.
Directory.Build.props	Adds MinVer and CI warning policy. Nit (2 votes): remove the stale solution-item entry.
Benchmarks/​Readme.md	Documents benchmark execution and results.
Benchmarks/​ProxyBenchmarks.cs	Defines proxy performance benchmarks.
Benchmarks/​Program.cs	Adds the BenchmarkDotNet runner.
Benchmarks/​Fixtures.cs	Adds benchmark fixtures and baseline types.
Benchmarks/​Benchmarks.csproj	Adds the BenchmarkDotNet project.
BenchmarkDotNet.Artifacts/​results/​Benchmarks.ProxyBenchmarks-report.html	Adds generated benchmark output. Nit (1 vote): ignore generated artifacts or retain only a curated snapshot.
BenchmarkDotNet.Artifacts/​results/​Benchmarks.ProxyBenchmarks-report.csv	Adds generated benchmark CSV output.
BenchmarkDotNet.Artifacts/​results/​Benchmarks.ProxyBenchmarks-report-github.md	Adds generated benchmark Markdown output.
.github/​workflows/​dotnet48.yml	Removes the superseded workflow.
.github/​workflows/​dotnet.yml	Removes the superseded workflow.
.github/​workflows/​build.yml	Adds unified cross-platform build, test, and publish jobs. Critical (2 votes): non-Windows solution builds target net40/net47; critical (2 votes): Ubuntu packing also requires .NET Framework assets. Moderate (2 votes): SDK selection is not pinned. Moderate (1 vote): publishing lacks packages: write. Moderate (1 vote): the quoted package glob will not expand. Nit (1 vote): update README badges for the renamed workflow.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.