Skip to content

code-analysis.md

Latest commit

 

History

History
35 lines (28 loc) · 2.14 KB

code-analysis.md

File metadata and controls

35 lines (28 loc) · 2.14 KB
Raw

Code analysis & analyzers

Rask builds with warnings-as-errors and analyzers on (Directory.Build.props): TreatWarningsAsErrors, EnableNETAnalyzers, EnforceCodeStyleInBuild. The SDK's .NET analyzers (CAxxxx) and code-style (IDExxxx, severities from .editorconfig) run on every build. AnalysisLevel is left at the SDK default because the repo builds clean there; raising it is a deliberate, per-PR cleanup (see below).

Recommended additional analyzers

Adopt these for a published, perf-sensitive framework with source generators. Add via central package management (Directory.Packages.props) + a PackageReference … PrivateAssets="all" in Directory.Build.props.

Analyzer Why
Roslynator.Analyzers Broad, high-signal C# rules + refactorings. Best single add.
Meziantou.Analyzer Correctness/perf rules the SDK misses (culture, async, allocations).
SonarAnalyzer.CSharp Bug & code-smell detection, cognitive-complexity.
Microsoft.CodeAnalysis.PublicApiAnalyzers Track the public API surface of the NuGets (PublicAPI.Shipped/Unshipped.txt) so breaking changes are caught — pairs with SemVer/MinVer.
Microsoft.CodeAnalysis.BannedApiAnalyzers Enforce "use standard libs / safe defaults": ban DateTime.Now, culture-less ToString, etc. (directly addresses the CA1305 class of findings).
Microsoft.CodeAnalysis.Analyzers (RS rules) Author-correctness for the Rask.Generators projects specifically.

Adoption procedure (important)

Because the build is warnings-as-errors, adding an analyzer turns its findings into build errors immediately. So adopt one analyzer per PR:

  1. Add the package; build; triage the findings.
  2. Fix the real ones; suppress intentional ones in .editorconfig with a # justification comment (e.g. CA1051 on RenderFrame's deliberately-public hot-path fields).
  3. Run the rask-ship gate (incl. benchmarks if a fix touches the render hot path).

A ready starter set (already-clean default mode) can be promoted later via <AnalysisLevel>latest-recommended</AnalysisLevel> — that surfaces ~33 CA findings today, so it gets its own PR.