A front-end data model for the Industry Foundation Classes (IFC), the open BIM exchange standard maintained by buildingSMART International. COMPAS IFC sits between IFC's full schema and the people who need to work with it, exposing a small, intentional API where the schema offers thousands of classes and dozens of relationship patterns.
The toolkit is the open-source artefact described in chapter 4 of Future Data Models for AEC: From Simplicity for Humans to Interoperability by AI (Li Chen, ETH Zürich, 2026). It targets researchers and developers who need to read, analyse, modify, and write IFC data without becoming experts in the schema itself.
- Opens any IFC file (IFC2X3, IFC4, or IFC4X3) and exposes its content through one container class and one element class.
- Materialises the spatial hierarchy as an explicit
tree
with direct parent/child pointers, so traversal does not require chasing
IfcRelContainedInSpatialStructureinstances. - Materialises non-hierarchical relationships (structural connections, system flows, material associations, geometric dependencies) as a separate graph.
- Couples geometric representations to executable kernels — the COMPAS core library for primitives and meshes, OpenCascade (via compas_occ) for B-Rep and NURBS, and CGAL (via compas_cgal) for boolean operations and predicates — so volumes, surface areas, and bounding boxes can be computed without leaving Python.
- Treats every product subclass uniformly through a single
GenericElementtype, falling back toIfcBuildingElementProxy(with the original type name preserved asObjectType) for custom or domain-specific elements. - Validates element metadata against Pydantic schemas and rejects non-conforming data at the point of insertion.
┌───────────────────────────────────────────────────────────────────┐
│ Front-end API │
│ BuildingInformationModel GenericElement │
│ SpatialTree (model.tree) InteractionGraph (model.graph) │
└───────────────────────────────────────────────────────────────────┘
▲
┌───────────────────────────────────────────────────────────────────┐
│ Bidirectional mapping layer │
│ relationship resolution · placement-chain rectification · │
│ representation conversion · type normalisation · │
│ property serialisation │
└───────────────────────────────────────────────────────────────────┘
▲
┌───────────────────────────────────────────────────────────────────┐
│ IFC schema interface — IfcOpenShell (IFC2X3 · IFC4 · IFC4X3) │
└───────────────────────────────────────────────────────────────────┘
pip install compas_ifcFor development:
git clone https://github.com/compas-dev/compas_ifc.git
cd compas_ifc
pip install -e ".[dev]"Optional dependencies:
- compas_occ — high-fidelity B-Rep and NURBS via OpenCascade.
- compas_viewer —
model.show()andmodel.show_collisions().
from compas_ifc.bim import BuildingInformationModel
# Load an existing model
model = BuildingInformationModel("data/Duplex_A_20110907.ifc")
# Walk the spatial hierarchy
for storey in model.storeys:
print(storey.name)
for child in storey.children:
print(f" - [{child.ifc_type}] {child.name}")
# Query by IFC type or GlobalId
walls = model.get_elements_by_type("IfcWall")
wall = model.get_element_by_global_id("3cUkl32yn9qRSPvBJZ3dB2")
# Computed geometric properties
print(wall.volume, wall.surface_area)
# Save back to disk
model.save("modified.ifc")Creating a new model from scratch:
from compas.geometry import Box, Frame, Point, Vector
from compas_ifc.bim import BuildingInformationModel
model = BuildingInformationModel.template(schema="IFC4", unit="m")
storey = model.storeys[0]
model.create_wall(
name="South wall",
parent=storey,
geometry=Box(8.0, 0.2, 3.0),
frame=Frame(Point(0, 0, 0), Vector.Xaxis(), Vector.Yaxis()),
)
model.save("from_scratch.ifc")Custom elements with declarative validation:
from pydantic import BaseModel, Field
from compas_ifc.bim import BuildingInformationModel
from compas_ifc.validation import Specification
class ConcreteRecipe(BaseModel):
strength_class: str = Field(pattern=r"^C\d{2}/\d{2}$")
cement_type: str
cover_mm: float = Field(gt=0)
model = BuildingInformationModel.template(schema="IFC4", unit="m")
model.specifications = [
Specification(
name="Concrete recipe",
ifc_types=["IfcSlab"],
required_psets={"ConcreteRecipe": ConcreteRecipe},
)
]
# Slabs added without a valid ConcreteRecipe pset are rejected at insertion.The complete evaluation suite from appendix A of the thesis lives in
thesis/appendix/A/. Run them all and produce a consolidated report with:
conda run -n compas-ifc python thesis/appendix/A/run_all.pyThe report is written to thesis/appendix/A/outputs/A-summary.txt.
compas_ifc/
├── src/compas_ifc/
│ ├── bim.py # BuildingInformationModel — entry point
│ ├── element.py # GenericElement — unified element abstraction
│ ├── factory.py # ElementFactoryMixin — typed creators + template
│ ├── tree.py # TreeMixin — IFC import/export, placement rectify
│ ├── interactions.py # InteractionMixin — graph + connection/collision
│ ├── validation.py # Specification + Pydantic-based enforcement
│ ├── representations/ # Parametric geometry (Extrusion, Pipe, …)
│ ├── brep/ # TessellatedBrep + viewer plugins
│ ├── algorithms/ # Vectorised contact and collision detection
│ ├── conversions/ # IFC ↔ COMPAS converters
│ ├── entities/ # IfcOpenShell wrappers (back-end implementation)
│ └── file.py # IFCFile — ifcopenshell adapter
├── tests/ # pytest suite
├── thesis/appendix/A/ # Reproducible evaluation suite (chapter 4)
├── data/ # Reference IFC files
├── docs/ # Sphinx documentation
└── scripts/ # Worked examples
MIT. See LICENSE.
If you use COMPAS IFC in academic work, please cite the thesis:
Chen, L. Future Data Models for AEC: From Simplicity for Humans to Interoperability by AI. Doctoral dissertation, ETH Zürich, 2026.
Issues and feature requests: https://github.com/compas-dev/compas_ifc/issues. Questions: li.chen@arch.ethz.ch.