perf: add reproducible font startup profiler

.github/workflows/Benchmark.yml

@@ -14,6 +14,13 @@ jobs:
       with:
         dotnet-version: '10.x'
     - run: dotnet run -p CSharpMath.Rendering.Benchmarks
+    - name: Font profile smoke
+      run: dotnet run -c Release --project CSharpMath.Rendering.Benchmarks/CSharpMath.Rendering.Benchmarks.csproj -- --font-profile --font-profile-smoke --font-profile-output .fontprofile/font-profile-smoke.json
+    - name: Store font profile smoke
+      uses: actions/upload-artifact@v4
+      with:
+        name: font-profile-smoke
+        path: .fontprofile/font-profile-smoke.json
     - name: Store benchmark result
       uses: benchmark-action/github-action-benchmark@v1
       with:
@@ -25,4 +32,4 @@ jobs:
         alert-threshold: 155% # Should not be lower than 155% to take account of fluctuations of CI load, see https://github.com/verybadcat/CSharpMath/pull/156#commitcomment-42067078
         comment-on-alert: true # Show alert with commit comment on detecting possible performance regression
         fail-on-alert: true
-        alert-comment-cc-users: '@Happypig375'
+        alert-comment-cc-users: '@Happypig375'

CSharpMath.Rendering.Benchmarks/CSharpMath.Rendering.Benchmarks.csproj

@@ -4,11 +4,12 @@
     <TargetFramework>net10.0</TargetFramework>
   </PropertyGroup>
   <ItemGroup>
+    <None Include="..\CSharpMath.Rendering.Tests\ComicNeue_Bold.otf" Link="ComicNeue_Bold.otf" CopyToOutputDirectory="PreserveNewest" />
     <Compile Include="..\CSharpMath.Rendering.Tests\TestRenderingMathData.cs" Link="TestRenderingMathData.cs" />
     <Compile Include="..\CSharpMath.Rendering.Tests\TestRenderingSharedData.cs" Link="TestRenderingSharedData.cs" />
   </ItemGroup>
   <ItemGroup>
     <PackageReference Include="BenchmarkDotNet" Version="0.15.8" />
     <ProjectReference Include="../CSharpMath.SkiaSharp/CSharpMath.SkiaSharp.csproj" />
   </ItemGroup>
-</Project>
+</Project>

CSharpMath.Rendering.Benchmarks/FontProfiling.md

@@ -0,0 +1,27 @@
+# Font profiling artifact
+
+Run `dotnet run -c Release -p CSharpMath.Rendering.Benchmarks -- --font-profile`. Use
+`--font-profile-samples`, `--font-profile-iterations`, and `--font-profile-batches` to control work,
+or `CSHARP_MATH_FONT_PROFILE_OUTPUT` / `--font-profile-output` for the output path. Invoke
+`--font-profile-smoke` with or without `--font-profile`; it is a cheap deterministic correctness
+check for CI.
+
+The artifact reports raw samples for direct per-resource SFNT parsing, separate first-math and
+first-mixed cold workers, global rendering, and alternating formula/font batches. It records bytes
+allocated by the managed runtime and bytes retained after a forced GC; process working set is not
+used as managed retention. The three embedded faces are loaded by identity and hashed from their
+actual bytes. The tracked Comic Neue asset is a real runtime `LocalTypeface`. Table tags are parsed
+from each SFNT directory; CFF/glyf and hint/instruction, MATH, GSUB, GPOS, and GDEF presence can be
+read directly from `tables`. The `packaging` object reports the measured bundled-font, custom-asset,
+rendering-assembly, and generated-font-data fields for the current build. `exercisedPaths`,
+`switchingSequence`, and `globalFaces` identify operations actually performed; each `globalFaces`
+entry includes the cmap-validated corpus, LocalTypeface usage, and successful draw evidence.
+Parse-only table inventory is not presented as outline/cache execution evidence.
+
+Compare distributions and batch retained-memory trends, not thresholds: OS, runtime, Skia, and
+process startup noise affect measurements. The artifact helps investigate font startup and cache
+growth discussed by issues #107 and #191; it does not prove causality or impose a performance gate.
+
+Captured smoke result (Release, .NET 10 on the development Windows host): 10 raw samples (three
+parse-only faces, three rendered bundled-face scenarios, first math, first mixed text, global
+alternation, and custom Comic Neue), four font inventories, and zero worker or schema errors.