Implements the ifcgit/Bonsai diff algorithm for the preview service:
diff.py: step_ids_from_diff() parses git diff text (regex on +#NNN= / -#NNN=
lines) to classify step IDs as added/modified/removed. expand_step_ids()
walks IfcShapeRepresentation, IfcObjectPlacement, IfcPropertySet, and
IfcTypeProduct relationships to promote changed sub-entities to the parent
IfcProduct that is visually affected.
render.py: render_diff() does two passes over the same camera. Pass 1 renders
the head model with green=added, blue=modified, ghost=unchanged. Pass 2
renders only the removed elements from the base model in red, using the camera
auto-fitted to pass 1. PIL composites the passes: wherever pass 2 is not
white background it stamps onto pass 1. Moved elements (modified) appear once
in blue with no doubling artefact.
git.py: diff_text() runs git diff against the cached bare repo clone and
returns the raw text for diff.py to parse.
service.py: GET+POST /render_diff endpoint. Fetches both commits, gets
diff text, runs the full pipeline sandboxed. Caches on disk when both refs
are immutable. Pillow added to render/service extras.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
