Split (partition into multiple meshes)

meshioplusplus.split(mesh, by=...) partitions one mesh into several submeshes — by cell type, by connected component, by region (a named cell_sets group, or an integer cell_data tag), or by named Cell regions (by="regions", plural) — each pruned to its own points with connectivity and data remapped. It is a mesh operation (like crop and merge), not a file format, and uses only standard C++/numpy, so it runs under every mesh backend.

import meshioplusplus

mesh = meshioplusplus.read("mixed.vtu")

pieces = meshioplusplus.split(mesh, by="type")        # {"triangle": ..., "quad": ...}
pieces = meshioplusplus.split(mesh, by="component")   # {"0": ..., "1": ...}
pieces = meshioplusplus.split(mesh, by="region")      # by cell_sets / tag
pieces = meshioplusplus.split(mesh, by="regions")     # one piece per named Cell region

for key, piece in pieces.items():
    meshioplusplus.write(f"part_{key}.vtu", piece)

split returns an ordered dict {key: Mesh}.

Criteria

by	partitions by	keys
"type"	cell type (one piece per type)	the meshio type name ("triangle", …)
"component"	connected components (flood-fill over node-sharing adjacency)	"0", "1", …
"region"	named cell_sets if present, else an integer cell_data tag	set name / tag value
"regions"	named Cell regions (cross-binding — runs in the C++ core, not just the Python shim)	region name

For by="region", if the mesh carries cell_sets they define the pieces (one per set, keyed by name); otherwise the split falls back to an integer cell_data tag — tag="name" selects it, or the first integer cell_data is auto-detected. Connected components use a union-find over cells sharing a node (never O(N²)).

by="regions" (plural) is the odd one out: it is not a partition. Each named Cell region is processed independently, so a cell belonging to two regions lands in both output pieces, and a cell in no region lands in none. Point/Side regions produce no piece at all — there is no sound default for turning "these facets" or "these points" alone into a whole submesh without first deciding which surrounding cells to pull in, so split does not guess. Unlike by="region", this criterion runs in the C++ core and so reaches every binding uniformly (see Other languages), not just Python; --tag is unused by it. Use meshioplusplus regions first to see what a mesh's regions are before splitting by them.

What changes

Each piece is pruned to only its own points, with connectivity remapped and point_data / cell_data subset to it (field_data carried through). point_sets / cell_sets are remapped into each piece (done in the Python layer). The numpy fallback handles rectangular cell blocks; ragged/polyhedron blocks are handled by the C++ core only.

CLI

meshioplusplus split in.vtu 'out_{key}.vtu' --by type
meshioplusplus split in.vtu 'out_{key}.vtu' --by component
meshioplusplus split in.vtu 'out_{key}.vtu' --by region --tag gmsh:physical
meshioplusplus split in.vtu 'out_{key}.vtu' --by regions

The output pattern must contain {key}, which is replaced by each piece's key. The number of pieces and their sizes are printed. See the CLI reference.

Other languages

The C++ core returns the pieces (the file-writing pattern is a caller concern). by="region" there means the integer cell_data tag (cell_sets are Python-only); by="regions" (plural) means named Cell regions, and reaches every binding below identically, since split's by string is resolved once by the shared split_by_from_name:

• C API — mio_split(mesh, by, tag_name) returns a mio_split_result (mio_split_result_count, ..._key, ..._mesh / ..._take_mesh, ..._free). See the C API reference.
• Fortran — mesh%split(by, tag_name=..., keys=...) returns an array of meshes. See the Fortran reference.
• WebAssembly / JavaScript — split(mesh, by, tagName) returns an array of {key, mesh}. See the WebAssembly reference.