[pmeems]

Phase 3 of the polygon triangulation plan: verify the complete rendering pipeline, measure triangulation cost and mesh storage, and bring the documentation in line with the implementation.

Depends on #66 (Core triangulation) and #68 (OpenGL mesh rendering); stacked on the #68 branch.

What changed

• Lifecycle verification (Core): MapViewTest.ComplexSceneMeshesSurviveViewportChanges proves the triangulated fill meshes of a concave, holed multipart scene survive viewport resize, zoom in/out, anchored zoom, and pan without the scene being rebuilt or re-triangulated, and that layer add/remove rebuilds the scene with fresh meshes while keeping the original rings.
• GDI fallback verification (ActiveX): VectorLayerRenderer.GdiFallbackRendersConcaveHoledPolygonFromRings proves a polygon without a fill mesh is rendered by GDI from its original rings — concave indentation and interior ring stay unfilled — and that the coordinator recovers to OpenGL for a later mesh-complete frame.
• Performance baseline: RenderingBaseline.MeasuresComplexPolygonCosts (opt-in, same env vars as the existing baseline) generates 50k concave L-shapes, 50k holed two-component multipolygons, and 12.5k 32-vertex star polygons across three layers, and reports scene-preparation time, mesh vertex/index storage, memory growth, and GDI/OpenGL frame costs.
• Documentation: docs/ARCHITECTURE.md, docs/PORTING.md, and docs/TESTING.md no longer claim OpenGL supports only convex single-ring polygons; the measured baseline numbers (including the ~15 µs per polygon triangulation cost that now dominates scene preparation) are recorded in docs/TESTING.md.

Measured (Release, x64, one workstation)

• Complex mix: scene preparation 2,733.7 ms for 162,500 triangulated meshes; mesh storage 32.1 MiB; OpenGL median frame 42.2 ms; GDI median frame 168.6 ms.
• Triangulation runs once per polygon during scene preparation and is reused on redraw; per-frame times are unaffected.

Validation

• msbuild MapWinGIS6.slnx -restore -p:Configuration=Debug -p:Platform=x64 -m — clean (warnings are errors in this configuration).
• MapWinGIS.Core.Tests.exe: 312 passed, 1 skipped (network).
• MapWinGIS.ActiveX.Tests.exe: 67 passed, 3 skipped (interactive host + two opt-in baselines); both baselines pass in Release when opted in.

No memory changes were needed: the measured numbers live in docs/TESTING.md, and no new reusable pitfalls or tooling knowledge surfaced.

[Copilot]

🟡 Changes recommended

The tests can miss a permanent OpenGL fallback regression, and the rendering baseline uses invalid overlapping MultiPolygons, requiring fixture corrections and refreshed measurements.

2 open issues

• Move satellite outside body interiors and validate multipart geometry · New
• Verify OpenGL availability before accepting the GDI fallback · New

What changed in this PR

This PR adds verification and performance documentation for the complex-polygon rendering pipeline introduced by Core triangulation in #66 and OpenGL mesh rendering in #68.

Changes:

• Adds tests for mesh retention across viewport changes and ring-based GDI fallback.
• Adds an opt-in complex-polygon benchmark covering scene preparation, mesh storage, memory growth, and backend frame costs.
• Updates rendering documentation to describe mesh-based OpenGL support and record performance measurements.

Builds and tests were not run during this review.

File	Description
tests/​MapWinGIS.Core.Tests/​MapViewTests.cpp	Adds complex-scene mesh lifecycle coverage.
tests/​MapWinGIS.ActiveX.Tests/​VectorLayerRendererTests.cpp	Adds concave and holed polygon fallback coverage.
tests/​MapWinGIS.ActiveX.Tests/​RenderingBaselineTests.cpp	Adds complex geometry fixtures and performance measurements.
docs/​TESTING.md	Documents benchmark workloads, usage, and recorded results.
docs/​PORTING.md	Updates supported polygon rendering and fallback behavior.
docs/​ARCHITECTURE.md	Describes retained meshes, backend responsibilities, and rendering costs.

_{Review effort: Balanced}

[Julian-MW]

I reviewed the latest changes and did not find anything new to report.