Meshing

After geometry is synchronized, generate a mesh up to a target dimension.

gmsh.model.mesh.generate(3); // 1D -> 2D -> 3D

generate(dim) runs all lower dimensions first, so generate(3) also produces the 1D and 2D meshes.

Reading the mesh

Nodes

const { nodeTags, coord } = gmsh.model.mesh.getNodes();
// nodeTags: number[]      — N node ids
// coord:    number[]      — 3N values, [x0,y0,z0, x1,y1,z1, ...]
for (let i = 0; i < nodeTags.length; i++) {
  const [x, y, z] = coord.slice(i * 3, i * 3 + 3);
}

Elements

const { elementTypes, elementTags, nodeTags } = gmsh.model.mesh.getElements();
// elementTypes: number[]    — one entry per element type present
// elementTags:  number[][]  — element ids, grouped by type
// nodeTags:     number[][]  — connectivity, grouped by type (flattened per element)

Common element type ids: 1 = 2-node line, 2 = 3-node triangle, 3 = 4-node quad, 4 = 4-node tetrahedron, 5 = 8-node hexahedron.

const triIdx = elementTypes.indexOf(2);
const numTriangles = triIdx >= 0 ? elementTags[triIdx].length : 0;

Algorithms and options

Set options before generate() with gmsh.option.setNumber.

Option	Meaning	Notable values
Mesh.Algorithm	2D algorithm	1 MeshAdapt, 5 Delaunay, 6 Frontal-Delaunay (default), 8 Frontal-Quad
Mesh.Algorithm3D	3D algorithm	1 Delaunay (default), 4 Frontal, 10 HXT
Mesh.MeshSizeMin / Mesh.MeshSizeMax	global size bounds
Mesh.MeshSizeFactor	scale all sizes
Mesh.Optimize	optimise tets	0/1
Mesh.ElementOrder	element order	1 linear, 2 quadratic, ...
General.Verbosity	log level	0 silent … 5 debug

gmsh.option.setNumber('Mesh.MeshSizeMax', 0.1);
gmsh.model.mesh.generate(3);

Hex-dominant meshing (experimental)

Gmsh's hex-tet hybrid recombiner is only reachable through the experimental RTree 3D algorithm (Mesh.Algorithm3D = 9) — Mesh.Recombine3DAll is a no-op under Delaunay/Frontal/HXT, and combining Mesh.RecombineAll with Frontal throws Cannot use frontal 3D algorithm with quadrangles on boundary (an ordering bug in upstream Gmsh, not this build). This is how Gmsh itself gates the feature, not a gap in the WASM build:

gmsh.option.setNumber('Mesh.Algorithm3D', 9);      // RTree
gmsh.option.setNumber('Mesh.Recombine3DAll', 1);
gmsh.model.mesh.generate(3);
// -> a mix of element type 4 (tet) and 5 (hex), stitched by type-140
//    "trihedron" connector elements at the tet/hex interface.

Treat this path as experimental (upstream describes it as such); prefer the default Delaunay or HXT algorithms for production meshes.

Higher-order and optimisation

gmsh.option.setNumber('Mesh.ElementOrder', 2);     // quadratic elements
gmsh.model.mesh.generate(3);
gmsh.model.mesh.optimize('Netgen');                // optimise

Size fields

For non-uniform sizing, use the field submodule:

const f = gmsh.model.mesh.field.add('Distance');
gmsh.model.mesh.field.setNumbers(f, 'PointsList', [p1, p2]);

const t = gmsh.model.mesh.field.add('Threshold');
gmsh.model.mesh.field.setNumber(t, 'InField', f);
gmsh.model.mesh.field.setNumber(t, 'SizeMin', 0.01);
gmsh.model.mesh.field.setNumber(t, 'SizeMax', 0.2);
gmsh.model.mesh.field.setNumber(t, 'DistMin', 0.1);
gmsh.model.mesh.field.setNumber(t, 'DistMax', 0.5);

gmsh.model.mesh.field.setAsBackgroundMesh(t);
gmsh.model.mesh.generate(2);

Size callback

For sizing that can't be expressed as a field, setSizeCallback installs a JS function Gmsh calls once per mesh vertex:

gmsh.option.setNumber('General.NumThreads', 1); // required, see below
gmsh.model.mesh.setSizeCallback((dim, tag, x, y, z, lc) => {
  return x < 0.5 ? 0.02 : lc; // refine near x=0, leave the rest as-is
});
gmsh.model.mesh.generate(2);
gmsh.model.mesh.removeSizeCallback();

Main-thread only: the callback is a native function pointer installed via Emscripten's addFunction, which is per-thread — it is not reachable from OpenMP worker pthreads. Keep General.NumThreads at 1 (Gmsh's default) while a callback is set; the runtime warns once if it detects otherwise.

Threads

This build is OpenMP-enabled (pthreads). Gmsh still defaults to 1 thread; opt in per session:

gmsh.option.setNumber('General.NumThreads', 0); // 0 = all cores, or an explicit count

Per-stage limits are available via Mesh.MaxNumThreads1D, Mesh.MaxNumThreads2D and Mesh.MaxNumThreads3D. The OMP_NUM_THREADS environment variable does not exist in the browser (and is not read by the WASM module) — use the options.

In the browser, threads require a cross-origin-isolated page (COOP/COEP headers) — see Browser usage. Avoid setting the thread count above navigator.hardwareConcurrency: the pthread worker pool is sized to the core count, and oversubscribing it can deadlock the main thread.

The OpenMP-parallel 3D mesher is HXT (Mesh.Algorithm3D = 10); it works on CAD re-imported through STEP/IGES the same as the default Delaunay algorithm. setSizeCallback (see API overview) is main-thread-only and incompatible with General.NumThreads > 1.