[Copilot]

mapbox/polylabel deleted its C++ implementation in mapbox/polylabel#124. mapnik/polylabel is the maintained fork that restores and improves the C++ port (v2.1.0 algorithmic improvements: clamped cell sizes, flattened coordinate buffers, block-based bbox skipping, aggressive queue pruning).

Changes

• .gitmodules — point submodule URL to https://github.com/mapnik/polylabel.git
• vendor/polylabel — update submodule ref from 6ed5f1a (mapbox 1.1.0) to 1edc167 (mapnik latest)
• vendor/polylabel.cmake — update INTERFACE_MAPLIBRE_URL and INTERFACE_MAPLIBRE_AUTHOR to reflect the new fork

No source changes needed — the fork preserves the same include path (<mapbox/polylabel.hpp>), namespace (mapbox::polylabel), and function signature, adding only an optional distance output parameter.

• Fixes Migrate to mapnik/polylabel #4407

[github-actions]

Bloaty Results 🐋

Compared to main

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +0.0% +43.7Ki  +0.0% +14.8Ki    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results/pr-4426-compared-to-main.txt

Compared to d387090 (legacy)

    FILE SIZE        VM SIZE    
 --------------  -------------- 
   +55% +64.2Mi  +475% +28.4Mi    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results/pr-4426-compared-to-legacy.txt

[github-actions]

Benchmark Results ⚡

Benchmark                                                          Time             CPU      Time Old      Time New       CPU Old       CPU New
-----------------------------------------------------------------------------------------------------------------------------------------------
OVERALL_GEOMEAN                                                 -0.0018         -0.0016             0             0             0             0

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/benchmark-results/pr-4426-compared-to-main.txt

[louwers]

@copilot The test is failing. Debug it.

[Copilot]

@copilot The test is failing. Debug it.

The failing tests were testPolygon and testMultiPolygon in MLNCodingTests.mm. They check that [polygon coordinate] returns the pole of inaccessibility at the center of a 1×1 degree square (100.5°, 0.5°), but were getting (100°, 0°) instead.

Root cause: the mapnik/polylabel fork added an early-out for "sub-precision polygons" — when min(width, height) <= precision, it returns envelope.min immediately instead of computing the true pole. With the default precision of 1.0 degree, a 1×1 degree polygon (min(1.0, 1.0) == 1.0 == precision) triggered this early-out.

Fix (committed in the latest push):

• MLNPolygon.mm: pass precision=1e-4 (≈10 m) instead of the default 1.0 for geographic-coordinate polygons
• MLNPolyline.mm: remove the unused <mapbox/polylabel.hpp> include

[louwers]

@copilot Please fix ios-build CI

[Copilot]

@copilot Please fix ios-build CI

Fixed. The remaining failure was in testMultiPolygon: it was putting a polygon-with-hole first in the MLNMultiPolygon, so [squares coordinate] returned that polygon's pole of inaccessibility. With the new mapnik implementation correctly handling holes, the pole is placed in a corner (~100.795°, ~0.205°) rather than at the centre which lies inside the hole — so the test expectation of (100.5, 0.5) was wrong.

Fix: swap the polygon order so the hole-free innerPolygonSquare is first. Its true pole-of-inaccessibility is exactly the centre of the 0.3×0.3 square (100.5°, 0.5°), matching the existing assertion.