pykentron

Python bindings for kentron, with a CadQuery-compatible API, so that CadQuery scripts run on kentron unchanged.

Installing

pip install pykentron              # the kernel and pykentron's own CadQuery-like API
pip install "pykentron[cadquery]"  # also what CadQuery's own layers need, to run CadQuery scripts

Python 3.10 or later. The wheels on PyPI are built for any x86-64 Linux machine.

Using it

import pykentron as cq
box = cq.Workplane("XY").box(10, 20, 30).faces(">Z").workplane().hole(5)
print(box.val().Volume())

To run CadQuery scripts unchanged, import pykentron.shim first: import cadquery then gives CadQuery's own pure-Python layers (Workplane, Sketch, selectors, cqgi, free functions, assemblies: copied unchanged from CadQuery v2.8.0, Apache-2.0, see pykentron/cadquery_shim/cadquery/VENDORED.txt) on pykentron's shapes. Or run a script as it is with pykentron-run script.py. Nothing named cadquery is installed, so a real CadQuery beside pykentron is unaffected unless the shim is activated (the two cannot share one process).

Names pykentron does not have yet, and the OCP names scripts use directly (beyond Bnd_Box, BRepBndLib.AddOptimal_s, BRepTools.Clean_s), import but raise NotImplementedError when used.

Building from source

pip install ./crates/pykentron                 # with maturin, as the PyPI wheels are built
python3 crates/pykentron/build.py              # in place, optimized (LTO, AVX2/FMA), into python/pykentron/_kentron.abi3.so
python3 crates/pykentron/build.py --portable   # in place, for any x86-64 CPU
python3 crates/pykentron/build.py --quick      # in place, the tests' release profile

The in-place build installs the extension by renaming, so interpreters that have it loaded keep running; put crates/pykentron/python on sys.path to use it. The CadQuery layers need pyparsing, multimethod, runtype, typing_extensions and numpy (and nlopt, casadi for constraints), as CadQuery does: the cadquery extra.

Memory and speed

Shapes no Python object refers to any more are collected from time to time. How much is freed, and which allocator kentron's memory comes from, is a profile, chosen by PYKENTRON_PROFILE before pykentron is imported:

pykentron.profile() gives the profile in use and pykentron.set_profile(name) switches between low-memory and balanced at any time; the allocator is chosen once, at startup, so fast is only had through the variable. Building without the mimalloc feature (cargo build -p pykentron --no-default-features) drops mimalloc and the fast profile.

What is there

• Workplane: sketching (polylines, lines, arcs, rectangles, circles, polygons, points, workplanes and transforms), extrude/cutBlind/revolve/box/cylinder, booleans, fillet, selection (faces, edges, ... with string selectors |, #, +, -, <, >, %TYPE), placement.
• Shapes (Vertex, Edge, Wire, Face, Shell, Solid, Compound): construction (boxes, cylinders, cones, spheres, tori, prisms, revolutions, lofts, offset2D, shell offsets and thick solids), booleans, clean, transforms and mirrors, queries (volume, area, length, centres, bounding boxes, validity, normals, geometry types, distance, isInside, tessellate). As in CadQuery, a shape's bounding box is tight until it is tessellated, then bounded by its meshes, widened by their deviation.
• Files: STEP import and export (shapes, or an Assembly with names, colours and placements), BRep import and export, STL export.
• Assembly, Location, Plane, Vector, Color, BoundBox.

Not yet: DXF import, SVG and other exporters, cadquery.vis.show (warns).

Checking it

• tests/test_against_cadquery.py compares pykentron with the real CadQuery case by case (run it with a Python that has CadQuery: PYTHON crates/pykentron/tests/test_against_cadquery.py).
• scripts/cadquery_build.py --project DIR MODULE runs a CadQuery project's build() on pykentron and reports each part's volume, faces and validity; --oracle runs it on CadQuery instead.
• scripts/cadquery_sweep.py --project DIR --python PYTHON builds every module of a project both ways and checks that each part is valid and within 0.1% of CadQuery's volume (--oracle-cache FILE keeps CadQuery's results, which do not change).
• scripts/cadquery_tests.py runs CadQuery's own test suite (release v2.8.0, fetched from GitHub into target/cadquery-tests with a virtual environment of its own, made with uv) against pykentron, unchanged: python3 crates/pykentron/scripts/cadquery_tests.py [pytest arguments]. It fails on any test failing that is not in tests/cadquery_baseline.txt and reports those of the baseline that pass; --update-baseline records the current failures. The suite reaches CadQuery's internals (cadquery.occ_impl.*, cadquery.func, OCP): the shim resolves those names to pykentron's where it has them, and makes the rest placeholders that raise NotImplementedError when used.