STL#

.stl read + write lazy: binary only

Summary of the specification#

STL describes a solid as an unordered set of triangles — a triangle soup with no vertex sharing and no topology. Each facet carries its own normal and its three corner points. The ASCII flavour spells this out with solid / facet normal / outer loop / vertex keywords; the binary flavour is an 80-byte free-text header, a uint32 triangle count, then a fixed 50-byte record per facet (twelve float32 values plus a two-byte attribute field).

Specification at a glance#

ascii form	solid <name> … facet normal / outer loop / vertex x3 / endloop / endfacet
binary form	80-byte header, uint32 count, then 50 bytes per facet
binary record	3 float32 normal + 9 float32 vertices + uint16 attribute byte count
byte order	little-endian
topology	none — vertices are repeated per triangle

Read the full STL specification ↗

Reading#

import polyxios as px

mesh = px.read("model.stl")
mesh.vertices          # (n, 3)
mesh.element_types     # element groups found in the file

Binary bodies can be memory-mapped instead of loaded:

mesh = px.read("big.stl", lazy=True)

Writing#

px.write(mesh, "out.stl")

Format-specific options:

Option	Default	Effect
`binary`	`False`	Write the 50-byte-per-facet binary form instead of ASCII.

Quirks worth knowing#

Vertices are deduplicated on read so the mesh has shared topology — except in binary lazy mode, which returns them as-is, three per triangle, to avoid a second pass over the data.
Facet normals are read but not trusted for orientation; they are kept as element attributes.
The declared triangle count is validated against the real file size before memory is allocated.