Ifcos Core Runtime

IfcOpenShell Python Runtime

Quick Reference

Critical Warnings

• ALWAYS use == for entity comparison, NEVER is. Each query returns a new Python wrapper object.
• ALWAYS set entity references to None after calling model.remove(). The C++ object is deallocated regardless of Python reference count.
• ALWAYS keep a reference to the ifcopenshell.file object alive while any entities from it are in use. If the file is garbage-collected, all entity wrappers become dangling pointers (segfault).
• NEVER write to an ifcopenshell.file from multiple threads. The C++ backend has no locking. Concurrent writes corrupt the model or crash.
• NEVER iterate all entities and filter manually when by_type() exists. by_type() uses an internal index and is 10-100x faster.
• NEVER use get_info(recursive=True) on large models. It materializes the entire entity graph into memory.
• ALWAYS use named attribute access (wall.Name) instead of positional index access (wall[2]). Positional indices vary by entity type and schema version.
• ALWAYS check model.schema before accessing schema-specific attributes.

Decision Tree: Entity Reference Safety

Working with entity references?
├── Reading attributes?
│   ├── One attribute → entity.AttributeName (PascalCase)
│   ├── All attributes → entity.get_info() (returns dict)
│   └── Bulk read → entity.get_info(scalar_only=True) (faster)
│
├── Comparing entities?
│   ├── Same entity? → entity_a == entity_b (value equality)
│   ├── Same STEP ID? → entity_a.id() == entity_b.id()
│   └── NEVER use → entity_a is entity_b (always False)
│
├── Removing entities?
│   ├── With relationship cleanup → ifcopenshell.api.run("root.remove_product", model, product=entity)
│   ├── Low-level removal → model.remove(entity) (does NOT clean relationships)
│   └── After removal → entity_ref = None (ALWAYS nullify)
│
└── Storing references across operations?
    ├── Store entity.id() instead of the entity object
    ├── Re-fetch with model.by_id(stored_id) when needed
    └── NEVER cache entity objects across remove/undo operations

Decision Tree: Performance

Performance-critical operation?
├── Querying entities by type?
│   ├── Use model.by_type("IfcWall") → indexed, returns tuple
│   ├── First call builds index (slow), subsequent calls are instant
│   └── NEVER iterate all entities with manual is_a() filtering
│
├── Reading many attributes?
│   ├── Single entity → entity.get_info() (one C++ round-trip)
│   ├── Scalar values only → entity.get_info(scalar_only=True)
│   └── AVOID get_info(recursive=True) (massive memory allocation)
│
├── Bulk entity creation?
│   ├──  100 entities → model.create_entity() (5-10x faster, no undo tracking)
│   └── Bulk mode → handle GlobalId, OwnerHistory (IFC2X3) manually
│
└── Large model (500MB+ / 100k+ entities)?
    ├── Memory: ifcopenshell.open() loads ENTIRE file into RAM
    ├── Streaming: ifcopenshell.open(path, should_stream=True) (sequential only)
    ├── Parallel: open SEPARATE file instances per thread
    └── Cleanup: del model + gc.collect() to release C++ memory

---

Essential Patterns

Pattern 1: C++ Binding Architecture

IfcOpenShell Python objects are thin wrappers around C++ objects managed by the ifcopenshell_wrapper module.

# Schema-agnostic
import ifcopenshell

model = ifcopenshell.open("model.ifc")
wall = model.by_type("IfcWall")[0]

# Python wrapper delegates to C++ core
type(wall)                    # 
type(wall.wrapped_data)       # 

# The wrapper holds a reference to its parent file
wall.file  # Returns the ifcopenshell.file object

Key implication: Python garbage collection does NOT control C++ memory. The C++ backend allocates and deallocates independently. Calling model.remove(entity) frees the C++ object immediately, even if Python references still exist.

Pattern 2: Entity Identity

# Schema-agnostic
wall = model.by_type("IfcWall")[0]

# Two queries for the same entity return DIFFERENT Python wrapper objects
wall_a = model.by_id(wall.id())
wall_b = model.by_id(wall.id())

wall_a == wall_b    # True  — value equality via wrapped_data comparison
wall_a is wall_b    # False — different Python wrapper objects

# ALWAYS compare with == or by STEP ID
wall_a.id() == wall_b.id()  # True — same STEP ID (#42)

Pattern 3: Entity Invalidation After Removal

# Schema-agnostic
wall = model.by_type("IfcWall")[0]
wall_id = wall.id()

# Low-level removal deallocates the C++ object
model.remove(wall)

# DANGER: accessing wall after removal causes undefined behavior
# wall.Name  → segfault or garbage data

# ALWAYS nullify references after removal
wall = None

# To re-check existence, query by ID
try:
    still_exists = model.by_id(wall_id)
except RuntimeError:
    still_exists = None  # Entity was removed

# PREFER API removal for products (cleans up relationships)
other_wall = model.by_type("IfcWall")[0]
ifcopenshell.api.run("root.remove_product", model, product=other_wall)
other_wall = None  # Still must nullify

Pattern 4: by_type() Return Semantics

# Schema-agnostic
walls = model.by_type("IfcWall")

type(walls)   #  — NOT a list
len(walls)    # Number of IfcWall instances

# First call for a type builds an internal C++ index (slower)
# Subsequent calls for the same type use the cached index (instant)

# To mutate the result, convert to list
wall_list = list(walls)
wall_list.append(some_other_entity)

# Include subtypes (default behavior)
all_walls = model.by_type("IfcWall")  # Includes IfcWallStandardCase in IFC2X3/IFC4

# Exclude subtypes
only_walls = model.by_type("IfcWall", include_subtypes=False)

Pattern 5: PascalCase Attribute Access

# Schema-agnostic
wall = model.by_type("IfcWall")[0]

# Named access: PascalCase matching the IFC schema
name = wall.Name                    # str or None
global_id = wall.GlobalId           # str (22-char encoded GUID)
description = wall.Description      # str or None
object_type = wall.ObjectType       # str or None
owner_history = wall.OwnerHistory   # entity_instance or None

# Positional access: fragile, varies by type and schema
name = wall[2]  # AVOID: index 2 = Name for IfcWall, but varies for other types

# Check if attribute has a value (IFC $null → Python None)
if wall.Description is not None:
    print(wall.Description)

# is_a() checks the full class hierarchy
wall.is_a("IfcWall")              # True
wall.is_a("IfcBuildingElement")   # True (parent class)
wall.is_a("IfcProduct")           # True (grandparent)
wall.is_a("IfcSlab")              # False

# get_info() returns all attributes as a dict
info = wall.get_info()
# {"id": 42, "type": "IfcWall", "GlobalId": "...", "Name": "...", ...}

Pattern 6: Thread Safety

# Schema-agnostic

# SAFE: Read-only operations from multiple threads (no concurrent writes)
import concurrent.futures

def count_type(filepath, ifc_class):
    model = ifcopenshell.open(filepath)  # Separate instance per thread
    return len(model.by_type(ifc_class))

with concurrent.futures.ThreadPoolExecutor() as executor:
    futures = {
        executor.submit(count_type, "model.ifc", cls): cls
        for cls in ["IfcWall", "IfcSlab", "IfcDoor"]
    }

# SAFE: Separate file instances per thread for writes
def process_copy(filepath, output_path):
    local_model = ifcopenshell.open(filepath)
    # Modify local_model freely — independent C++ instance
    local_model.write(output_path)

# UNSAFE: Multiple threads writing to the SAME model object
# This WILL corrupt data or crash: no C++ locks exist

Pattern 7: Memory Management for Large Models

# Schema-agnostic
# Standard open loads entire file into memory
model = ifcopenshell.open("large_building.ifc")  # May use 4-8GB RAM

# Access specific attributes instead of materializing everything
name = wall.Name                                          # 1 attribute
material = ifcopenshell.util.element.get_material(wall)   # Targeted query

# AVOID: get_info(recursive=True) on large models
# deep_info = wall.get_info(recursive=True)  # Massive memory allocation

# Release a large model from memory
del model
import gc
gc.collect()  # Ensures Python wrappers are cleaned up and C++ memory freed

Pattern 8: File Lifecycle: Keep File Reference Alive

# Schema-agnostic
# DANGEROUS: File reference lost while entities are in use
def get_walls():
    model = ifcopenshell.open("model.ifc")
    return model.by_type("IfcWall")  # model goes out of scope after return

walls = get_walls()
# model is garbage-collected → all entity wrappers are dangling pointers
# walls[0].Name  → SEGFAULT

# SAFE: Return both the model and the entities
def get_walls_safe():
    model = ifcopenshell.open("model.ifc")
    walls = model.by_type("IfcWall")
    return model, walls

model, walls = get_walls_safe()
walls[0].Name  # Safe — model reference kept alive

---

Schema-Specific Attribute Differences

# ALWAYS check schema before accessing schema-specific attributes
schema = model.schema  # "IFC2X3", "IFC4", "IFC4X3"

# Entities that differ by schema:
# IFC2X3/IFC4: IfcWallStandardCase exists
# IFC4X3: IfcWallStandardCase does NOT exist: use IfcWall

# IFC2X3: IfcDoorStyle, IfcWindowStyle
# IFC4+:  IfcDoorType, IfcWindowType

# IFC2X3: OwnerHistory is REQUIRED on rooted entities
# IFC4+:  OwnerHistory is OPTIONAL

# IFC4X3: IfcBuiltElement replaces IfcBuildingElement
# IFC4X3: New entities: IfcAlignment, IfcRoad, IfcBridge, IfcFacility

# Accessing a non-existent attribute raises AttributeError
try:
    pt = task.PredefinedType  # Exists in IFC4, not in IFC2X3
except AttributeError:
    pt = None

---

Installation

pip (Recommended for Most Users)

pip install ifcopenshell
python -c "import ifcopenshell; print(ifcopenshell.version)"

• Pre-built wheels for Python 3.8-3.12 on Linux, macOS, Windows
• Package name is ifcopenshell (all lowercase)
• Bundles C++ core and OpenCASCADE dependencies

conda (Recommended for Complex Environments)

conda install -c conda-forge ifcopenshell

• Better dependency resolution for OpenCASCADE
• More up-to-date than pip releases

Blender Integration

# Bonsai addon bundles ifcopenshell: no separate install needed
# For standalone install into Blender's Python:
/path/to/blender/python/bin/python -m pip install ifcopenshell

Platform Notes

| Platform | Note | |----------|------| | Windows | pip works out-of-the-box. For Blender: install into Blender's Python. Watch for PATH conflicts. | | macOS | Works on Intel and Apple Silicon. Use native arm64 Python, not Rosetta. | | Linux | Works on most distributions. Headless/Docker: no GPU needed. |

---

Version Detection

import ifcopenshell

# Library version
ifcopenshell.version  # e.g., "0.8.1"

# Schema of an open file
model = ifcopenshell.open("model.ifc")
model.schema             # "IFC2X3", "IFC4", or "IFC4X3"
model.schema_identifier  # e.g., "IFC4_ADD2"
model.schema_version     # e.g., (4, 0, 2, 1)

---

Reference Links

• [API Method Signatures](references/methods.md) — Complete signatures for entityinstance, file, bytype, remove, schema
• [Working Code Examples](references/examples.md) — End-to-end runtime patterns
• [Anti-Patterns](references/anti-patterns.md) — Common runtime mistakes and how to avoid them

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Impertio-Studio
• Source: Impertio-Studio/Blender-Bonsai-ifcOpenshell-Sverchok-Claude-Skill-Package
• License: MIT
• Homepage: https://github.com/OpenAEC-Foundation

Install and usage instructions live in the source repository linked above.