Skip to content

Adding a new operation¤

A new Brep operation touches four places. fillet is used below as a worked example — read Architecture first if the pieces don't look familiar.

1. Declare the pluggable¤

Add a signature-only function to operations.py. No logic — it exists so both backends have something to implement.

@pluggable(category="brep-operations")
def brep_fillet(brep: Brep, radius: float, edges: list[int] | None = None) -> Brep:
    raise NotImplementedError

2. Register it in each backend¤

In backend/occ/plugins.py and backend/rhino/plugins.py, register a @plugin with the same name, gated by what that backend requires:

# backend/occ/plugins.py
@plugin(category="brep-operations", requires=["OCP"])
def brep_fillet(brep, radius, edges=None):
    from .operations import occ_fillet

    return occ_fillet(brep, radius, edges)

# backend/rhino/plugins.py
@plugin(category="brep-operations", requires=["Rhino"])
def brep_fillet(brep, radius, edges=None):
    from .operations import rhino_fillet

    return rhino_fillet(brep, radius, edges)

3. Implement it, once per backend¤

The real logic goes in backend/occ/operations.py and backend/rhino/operations.py (or factories.py / queries.py, depending on what the operation does), written directly against the native kernel:

# backend/occ/operations.py
def occ_fillet(brep: Brep, radius: float, edges: list[int] | None = None) -> Brep:
    """Fillet edges of a Brep. If edges is None, fillet all edges."""
    shape = brep_to_occ(brep)
    fillet = BRepFilletAPI_MakeFillet(shape)
    ...

Both implementations must produce an equivalent result for the same input — that's the whole point of having one Brep API. They don't need to share code; OCC and Rhino calls look nothing alike.

4. Expose it on Brep¤

In brep.py, add the public method. It does nothing but call the pluggable:

def filleted(self, radius: float, edges: list[int] | None = None) -> Brep:
    """Return a filleted copy of this Brep."""
    return brep_fillet(self, radius, edges)

If the operation should also have an in-place form, follow the existing x() / xed() pattern (e.g. fillet() calls filleted() and replaces self with the result).

5. Test it against both backends¤

Write one test per backend, marked accordingly:

@pytest.mark.occ
def test_fillet_occ():
    ...

@pytest.mark.rhino
def test_fillet_rhino():
    ...

occ tests run on CI. rhino tests are skipped by default (-m 'not rhino' in pyproject.toml) and only run locally, on a machine with a Rhino license. Run the OCC suite yourself before committing — see Installation.

If the operation changes what gets serialized (a new surface or curve type, for instance), it also needs a round-trip test through the exchange format — both backends must be able to read what either one writes.