Skip to content

Update EarCut_jll to v3 - #287

Merged
SimonDanisch merged 1 commit into
JuliaGeometry:masterfrom
thchr-agent:agent/earcut-jll-v3
Nov 17, 2026
Merged

SimonDanisch merged 1 commit into
JuliaGeometry:masterfrom
thchr-agent:agent/earcut-jll-v3

Conversation

@thchr-agent

Copy link
Copy Markdown
Contributor

🤖 Moves to EarCut_jll v3 (3.2.3, available since JuliaPackaging/Yggdrasil#14434). The current v2 JLL was built with an old BinaryBuilder and pulls in Pkg as a dependency; v3.2.3 does not, so downstream packages no longer load Pkg through GeometryBasics.

  • No source changes: the JLL's C wrapper and symbols are unchanged.
  • The triangulations come back in a different order, so the expected faces in test/polygons.jl are updated. In all but one test these are exactly the same triangles (same winding), only with faces listed in a different order or vertices rotated within a face. The exception is the square-with-square-hole test, which still gives 8 valid triangles but with different diagonals, due to earcut 3.2.3's new Delaunay refine pass.
  • Bumps the version to 0.5.14 (patch). It would be great to tag a release once this is merged, so the Pkg dependency drops out for downstream users.

EarCut_jll v2 was built with an old BinaryBuilder and lists `Pkg` as a
dependency; v3.2.3 does not. The C wrapper and its symbols are unchanged
(the major bump tracks earcut.hpp catching up with earcut.js v3), but the
triangulations differ: earcut 3.2.3 adds a Delaunay refine pass, and
triangle and vertex order changed. The expected faces in the tests are
updated accordingly; all are still valid triangulations with the same
winding.

Bump version to 0.5.14.

Assisted-by: Claude Opus 5.5 <noreply@anthropic.com>
@thchr

thchr commented Nov 17, 2026

Copy link
Copy Markdown
Contributor

Gentle bump.

@SimonDanisch
SimonDanisch merged commit c1ec0a1 into JuliaGeometry:master Nov 17, 2026
13 checks passed
@SimonDanisch

Copy link
Copy Markdown
Member

Thanks... I guess this needs to be tagged as breaking though!

@thchr

thchr commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Why does it need a breaking release? It changes outputs, but not in a way that breaks any API contracts, right?

And thanks for merge!

@thchr

thchr commented Nov 17, 2026

Copy link
Copy Markdown
Contributor

(since there's 112 direct dependencies, a breaking release seems really scary?)

@SimonDanisch

Copy link
Copy Markdown
Member

Well you changed tests, in my book that's the best hint that something is breaking existing tested code.

(since there's 112 direct dependencies, a breaking release seems really scary?)

Well a non breaking release which breaks 112 packages seems scarier to me ;)

@thchr

thchr commented Nov 17, 2026

Copy link
Copy Markdown
Contributor

But the changed tests here are not really of a breaking sort though: the same exact triangles are returned, only differently sorted (either in face or vertex sorting). For every vertex-reordering, the resulting face has the same normal also (cyclic permutation).
As Claude noted, the only exception to this is the square-with-square-hole test, which gives a slightly different set of triangles, due to an changed Delaunay pass. But that ought to very comfortably within the scope of minor changes for a meshing tool?

@SimonDanisch

SimonDanisch commented Nov 17, 2026 •

Copy link
Copy Markdown
Member

Earcut things the release is breaking 🤷 Well i'm happy to tag it non breaking, once the Makie CI run through with this:
MakieOrg/Makie.jl#5915

@thchr

thchr commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Earcut things the release is breaking 🤷

I think those breaking changes are all about the JS packaging; none of it appears to touch the C++ code which the jll uses afaiu. See Earcut's 3.0.0 changelog.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants