sandbox-code-guidelines

name	sandbox-code-guidelines
description	Guidelines for ClrProfiler experiments and manual demonstrations using `sandbox/ConsoleApp` and `sandbox/CustomConsoleApp`. Use for exploratory CLR event generation, timer sampling, lifecycle checks, custom callback handlers, console or Datadog metric inspection, and small performance probes that are not yet production tests.

Sandbox Code Guidelines

Use the existing sample applications for manual or exploratory work:

• sandbox/ConsoleApp: exercise the packaged Datadog adapter and local DogStatsD-style flow.
• sandbox/CustomConsoleApp: exercise IClrTrackerCallbackHandler and custom metric handling.

Do not use dotnet-script. Prefer a focused TUnit test when the behavior can be verified deterministically; see test-first-development for the conventions.

Workflow

1. Choose the smallest existing sample that exposes the behavior.
2. Keep the experiment isolated from library API changes until it proves the approach.
3. Run the sample in Release mode from the repository root:

dotnet run -c Release --project sandbox/ConsoleApp/ConsoleApp.csproj
dotnet run -c Release --project sandbox/CustomConsoleApp/CustomConsoleApp.csproj

4. Convert confirmed behavior into the appropriate test project before changing production code.
5. Remove task-specific probes, delays, and forced allocations unless they improve the maintained sample itself.

Guardrails

• Treat sample output as manual evidence, not a regression test. CLR scheduling, GC timing, process counters, and UDP delivery vary by machine.
• Use bounded waits and cancellation. Do not add infinite loops that cannot terminate through CancellationToken, disposal, or a clear user action.
• Do not add secrets, production Datadog endpoints, private tags, host identities, or user payloads.
• Keep network tests local. Use loopback or a fake callback/handler; do not require a running cloud agent for automated verification.
• Do not infer channel losslessness or event ordering from a console run. Verify those properties with synthetic listener inputs in tests/ClrProfiler.UnitTest.
• Do not use a sample stopwatch as a benchmark conclusion. Follow performance-requirements for repeatable measurements.
• Preserve the purpose of each sample. Add a new sample project only when the scenario cannot fit either existing application without obscuring it.
• Keep sample target frameworks and package versions centralized and consistent with the repository project files.