re(RE-14.1): HostObjAttr (0x006b) signal is bimodal — text vs real re… #299

Workflow file for this run

	name: CI

	on:
	push:
	branches: [main]
	# Docs-only changes don't need to spin up the Windows wheel
	# matrix. Keep the fast Ubuntu jobs (fmt / clippy / doc / pii)
	# for any commit that actually touches code, including tests
	# or workflow files. `paths-ignore` is a whole-run filter so
	# this skips ALL jobs when only ignored paths changed.
	paths-ignore:
	- "**.md"
	- "docs/**"
	- "CHANGELOG.md"
	- "LICENSE*"
	- ".gitignore"
	- ".github/ISSUE_TEMPLATE/**"
	- "CITATION.cff"
	pull_request:
	branches: [main]

	# Supersede any in-flight run for the same ref when a new push
	# arrives. Prevents #52a → #52b → #52c back-to-back pushes from
	# burning 3× runner-minutes on stale Windows matrices nobody will
	# look at. `cancel-in-progress: true` is the budget saver; the
	# group key is per-ref so concurrent PR + push on different refs
	# don't cancel each other.
	concurrency:
	group: ci-${{ github.ref }}
	cancel-in-progress: true

	# Least-privilege at the workflow level. No CI job here modifies the
	# repo, opens issues, comments on PRs, or otherwise needs write
	# access; `contents: read` is enough. Individual jobs can request
	# additional scopes with a per-job `permissions:` block if that
	# ever changes.
	permissions:
	contents: read

	env:
	CARGO_TERM_COLOR: always
	RUSTFLAGS: "-D warnings"

	jobs:
	fmt:
	name: cargo fmt
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- uses: dtolnay/rust-toolchain@stable
	with:
	components: rustfmt
	- run: cargo fmt --all -- --check

	clippy:
	name: cargo clippy
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- uses: dtolnay/rust-toolchain@stable
	with:
	components: clippy
	- uses: Swatinem/rust-cache@42dc69e1aa15d09112580998cf2ef0119e2e91ae # v2
	- run: cargo clippy --all-targets --all-features -- -D warnings

	test:
	name: cargo test / ${{ matrix.os }} / ${{ matrix.rust }}
	runs-on: ${{ matrix.os }}
	strategy:
	fail-fast: false
	matrix:
	os: [ubuntu-latest, macos-latest, windows-latest]
	rust: [stable]
	include:
	- os: ubuntu-latest
	rust: "1.85" # MSRV (Rust 2024 edition minimum)
	steps:
	# Windows' git default `core.autocrlf=true` mangles binary files
	# that aren't explicitly marked `binary` in `.gitattributes`.
	# The phi-ag/rvt corpus stores some `.rfa` files outside LFS
	# without binary attribution — Windows checkout therefore
	# silently CRLF-converts them and CFB parsing fails on affected
	# streams. Disable autocrlf globally before any checkout runs.
	- name: Disable autocrlf on Windows
	if: runner.os == 'Windows'
	run: git config --global core.autocrlf false
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	# Fetch the Autodesk-owned `rac_basic_sample_family` corpus from
	# phi-ag/rvt. rvt-rs intentionally does not redistribute these
	# files (see SECURITY.md §Scope); they're pulled at build time.
	- name: Fetch phi-ag/rvt sample corpus
	uses: actions/checkout@v4
	with:
	repository: phi-ag/rvt
	path: _corpus
	lfs: true
	- uses: dtolnay/rust-toolchain@master
	with:
	toolchain: ${{ matrix.rust }}
	- uses: Swatinem/rust-cache@42dc69e1aa15d09112580998cf2ef0119e2e91ae # v2
	- name: Build
	run: cargo build --release --all-targets
	- name: Unit tests
	run: cargo test --release --lib --bins
	- name: Doc tests
	run: cargo test --release --doc
	# Regression-proof the Q5.2 100%-classification claim. Fails if
	# any field in any release decodes to FieldType::Unknown, or if
	# any expected corpus file is missing (guarantees non-vacuous
	# coverage).
	- name: Coverage gate — 100% field-type classification
	env:
	RVT_SAMPLES_DIR: ${{ github.workspace }}/_corpus/examples/Autodesk
	RVT_REQUIRE_CORPUS: "1"
	run: cargo test --release --test field_type_coverage -- --nocapture
	# Broader integration suite (round-trips, part-atom parsing,
	# history extraction, stream inventory, etc.) runs against the
	# corpus on all three OSes. Previously failed on Windows due to
	# a path-separator bug (CFB stream names came back with `\`
	# instead of `/` on Windows; fixed in reader::stream_names).
	- name: Integration tests (corpus-dependent)
	env:
	RVT_SAMPLES_DIR: ${{ github.workspace }}/_corpus/examples/Autodesk
	run: cargo test --release --test samples
	- name: IFC export integration tests
	env:
	RVT_SAMPLES_DIR: ${{ github.workspace }}/_corpus/examples/Autodesk
	run: cargo test --release --test ifc_roundtrip

	ifc-smoke:
	# IFC-42: run the synthetic-project integration test and assert
	# the emitted STEP file has the expected entity counts. Catches
	# regressions where the writer drops elements, the bridge layer
	# mis-counts openings, or STEP serialisation silently truncates.
	# Uses the committed fixture at `tests/fixtures/synthetic-project.ifc`
	# as the reference — any writer change that alters counts must
	# update this job AND the fixture together.
	name: IFC synthetic-project smoke
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@42dc69e1aa15d09112580998cf2ef0119e2e91ae # v2
	- name: Build + run synthetic-project integration test
	run: cargo test --release --test ifc_synthetic_project
	- name: Assert entity counts in committed fixture
	run: \|
	set -euo pipefail
	f=tests/fixtures/synthetic-project.ifc
	expect() {
	local pat="$1" want="$2"
	local got
	got=$(grep -c "$pat" "$f" \|\| true)
	if [ "$got" != "$want" ]; then
	echo "::error::IFC count drift: $pat got $got, want $want"
	exit 1
	fi
	echo "ok: $pat = $got"
	}
	expect IFCPROJECT 1
	expect IFCBUILDINGSTOREY 3
	expect IFCWALL 4
	expect IFCSLAB 1
	expect IFCDOOR 1
	expect IFCWINDOW 2
	expect IFCMATERIAL 2
	expect IFCPROPERTYSET 2
	expect IFCOPENINGELEMENT 1
	expect IFCRELVOIDSELEMENT 1
	expect IFCRELFILLSELEMENT 1
	expect IFCEXTRUDEDAREASOLID 7

	ifcopenshell-validate:
	# IFC-41: load the committed synthetic-project.ifc fixture in
	# IfcOpenShell and assert it opens without errors, carries the
	# right spatial hierarchy, and resolves every IfcRel* link. This
	# is the independent-validator gate — the Rust-side ifc-smoke
	# job above counts substrings, but IfcOpenShell actually parses
	# the STEP file through the full IFC4 schema, so errors the
	# substring check can't see (dangling refs, wrong attribute
	# arity, enum values outside the spec vocabulary) fail here.
	name: IfcOpenShell validates synthetic project
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- uses: actions/setup-python@v5
	with:
	python-version: "3.12"
	- name: Install IfcOpenShell (prebuilt wheel)
	run: \|
	python -m pip install --upgrade pip
	# ifcopenshell ships prebuilt wheels on PyPI as
	# `ifcopenshell-python` for Linux/macOS/Windows x64. Pin
	# to a stable 0.8.x series — the 0.7 API has semantic
	# differences in entity-by-type iteration.
	python -m pip install 'ifcopenshell>=0.8.0,<0.9.0'
	- name: Validate synthetic-project fixture
	run: \|
	python - <<'PY'
	import ifcopenshell
	path = "tests/fixtures/synthetic-project.ifc"
	f = ifcopenshell.open(path)
	# Schema identifier — we only ship IFC4.
	assert f.schema == "IFC4", f"expected IFC4, got {f.schema}"
	# Spatial hierarchy: 1 project, 1 site, 1 building, 3 storeys.
	assert len(f.by_type("IfcProject")) == 1
	assert len(f.by_type("IfcSite")) == 1
	assert len(f.by_type("IfcBuilding")) == 1
	assert len(f.by_type("IfcBuildingStorey")) == 3, (
	f"expected 3 storeys, got {len(f.by_type('IfcBuildingStorey'))}"
	)
	# Per-element entities.
	assert len(f.by_type("IfcWall")) == 4
	assert len(f.by_type("IfcSlab")) == 1
	assert len(f.by_type("IfcDoor")) == 1
	assert len(f.by_type("IfcWindow")) == 2
	# Openings + rels must have resolved references (no
	# dangling pointers — IfcOpenShell would raise during
	# open() if they did; double-check here).
	openings = f.by_type("IfcOpeningElement")
	assert len(openings) == 1
	for op in openings:
	voids = [r for r in f.by_type("IfcRelVoidsElement")
	if r.RelatedOpeningElement.id() == op.id()]
	fills = [r for r in f.by_type("IfcRelFillsElement")
	if r.RelatingOpeningElement.id() == op.id()]
	assert len(voids) == 1, f"opening {op} has {len(voids)} voids"
	assert len(fills) == 1, f"opening {op} has {len(fills)} fills"
	print("IfcOpenShell validation passed:")
	print(f" schema: {f.schema}")
	print(f" entities: {len(list(f))}")
	print(f" walls/slabs/doors/windows: 4/1/1/2")
	print(f" storeys: 3")
	PY

	# IFC-44: second fixture — structural framing + shared
	# RepresentationMap + faceted-brep furniture. Validates IFC-17 /
	# IFC-20 / IFC-21 / IFC-24 / IFC-30 paths that the project
	# fixture above doesn't exercise.
	- name: Validate synthetic-structural fixture
	run: \|
	python - <<'PY'
	import ifcopenshell
	path = "tests/fixtures/synthetic-structural.ifc"
	f = ifcopenshell.open(path)
	assert f.schema == "IFC4"
	assert len(f.by_type("IfcColumn")) == 2, "expected 2 columns"
	assert len(f.by_type("IfcBeam")) == 2, "expected 2 beams"
	assert len(f.by_type("IfcDoor")) == 3, "expected 3 doors"
	assert len(f.by_type("IfcFurniture")) == 1, "expected 1 furniture"
	# IFC-21: shared representation map, one entity, three
	# mapped items referencing it.
	reps = f.by_type("IfcRepresentationMap")
	assert len(reps) == 1, f"expected 1 IfcRepresentationMap, got {len(reps)}"
	mapped_items = f.by_type("IfcMappedItem")
	assert len(mapped_items) == 3, (
	f"expected 3 IfcMappedItem instances, got {len(mapped_items)}"
	)
	# All three mapped items must reference the same map.
	for mi in mapped_items:
	assert mi.MappingSource.id() == reps[0].id(), (
	"mapped item points at a different representation map"
	)
	# IFC-20: one faceted brep for the furniture.
	assert len(f.by_type("IfcFacetedBrep")) == 1
	# IFC-30: MaterialProfileSet usage on the 2 beams.
	assert len(f.by_type("IfcMaterialProfileSetUsage")) >= 1
	# IFC-24: I-shape profiles (3 total) + 1 circle profile.
	assert len(f.by_type("IfcIShapeProfileDef")) == 3
	assert len(f.by_type("IfcCircleProfileDef")) == 1
	print("IFC-44 structural fixture validation passed:")
	print(f" columns/beams/doors/furniture: 2/2/3/1")
	print(f" IfcRepresentationMap/IfcMappedItem: 1/3 (3:1 sharing ratio)")
	print(f" IfcFacetedBrep: 1 (tetrahedron)")
	PY

	# L5B-11.8: real-file walker → IFC end-to-end gate. Emit a
	# fresh IFC4 STEP from one of the family samples via `rvt-ifc`,
	# then load it in IfcOpenShell and assert: schema is IFC4, the
	# IfcProject exists, and every element entity that the walker
	# threaded through resolves without dangling references. This
	# gate catches regressions that the synthetic-fixture checks
	# can't see — e.g. a walker change that silently emits zero
	# elements on real files, or breaks step_writer's handling of
	# BuildingElement entities appended after the Project.
	- name: Build rvt-ifc + walker->IFC real-file gate
	run: \|
	cargo build --release --bin rvt-ifc
	# The family corpus resolves via RVT_SAMPLES_DIR (used by
	# `tests/common::samples_dir`) — if phi-ag/rvt was checked
	# out, point at it; otherwise skip (family samples aren't
	# redistributed by this repo).
	if [ -d "${RVT_SAMPLES_DIR:-./_corpus/examples/Autodesk}" ]; then
	SAMPLE=$(ls "${RVT_SAMPLES_DIR:-./_corpus/examples/Autodesk}"/*.rfa 2>/dev/null \| head -1)
	fi
	if [ -z "$SAMPLE" ]; then
	echo "skipping walker->IFC real-file gate: no family sample available"
	exit 0
	fi
	./target/release/rvt-ifc "$SAMPLE" -o /tmp/walker-real-file.ifc
	python - <<'PY'
	import ifcopenshell
	f = ifcopenshell.open("/tmp/walker-real-file.ifc")
	assert f.schema == "IFC4", f"schema: {f.schema}"
	# Project must exist (metadata-only baseline).
	assert len(f.by_type("IfcProject")) == 1, "lost IfcProject"
	# Walker may emit zero elements on family files — that's
	# acceptable. But if any BuildingElementProxy entities were
	# emitted, each must resolve without dangling references
	# (IfcOpenShell would have raised during open() otherwise).
	proxies = f.by_type("IfcBuildingElementProxy")
	print(f"L5B-11.8 walker->IFC gate passed:")
	print(f" schema: {f.schema}")
	print(f" IfcProject: 1")
	print(f" IfcBuildingElementProxy (walker-recovered): {len(proxies)}")
	PY

	doc:
	name: cargo doc
	runs-on: ubuntu-latest
	env:
	RUSTDOCFLAGS: "-D warnings"
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@42dc69e1aa15d09112580998cf2ef0119e2e91ae # v2
	- run: cargo doc --no-deps --lib

	# Q-09: benchmark-build regression gate.
	#
	# Runs `cargo bench --no-run` to compile every benchmark target
	# against the current library API. Without this job, a PR that
	# renames a public item or changes a function's signature can
	# break the benchmarks silently — `cargo test` doesn't build the
	# `[[bench]]` targets, and nobody runs benches on every push.
	#
	# A full runtime regression gate (compare wallclock numbers to a
	# baseline, fail on >X% slowdown) would need a dedicated runner
	# class with stable performance characteristics — GitHub's shared
	# runners have variable load, so wallclock comparisons need wide
	# tolerances to avoid false positives. Compile-gate is the
	# high-value, low-noise version we can enforce on every push.
	bench-build:
	name: cargo bench --no-run
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@42dc69e1aa15d09112580998cf2ef0119e2e91ae # v2
	with:
	key: bench-build
	# `--no-run` compiles the benches but doesn't execute them.
	# This catches API drift (renamed public items, changed
	# function signatures, missing traits on types used in
	# benches) that `cargo test` doesn't, since benches live in
	# their own `[[bench]]` targets.
	- run: cargo bench --no-run

	# Q-12: reproduce the docs.rs build locally in CI.
	# docs.rs runs `cargo doc` on nightly with `--all-features` and the
	# `--cfg docsrs` flag set; broken rustdoc (invalid links, missing
	# items, syntax errors in code examples) breaks the docs.rs page —
	# catching it in CI means every PR validates that rvt-rs will render
	# properly on docs.rs before the release cut.
	docs-rs-preview:
	name: docs.rs build preview
	runs-on: ubuntu-latest
	env:
	# Same flags docs.rs passes, plus -D warnings so this job fails
	# instead of silently publishing broken docs.
	RUSTDOCFLAGS: "--cfg docsrs -D warnings"
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	# docs.rs builds with nightly — any `#[cfg_attr(docsrs, ...)]`
	# attributes + `#![feature(...)]` gates must be available.
	- uses: dtolnay/rust-toolchain@nightly
	- uses: Swatinem/rust-cache@42dc69e1aa15d09112580998cf2ef0119e2e91ae # v2
	# Mirror the [package.metadata.docs.rs] settings from Cargo.toml
	# (`all-features = true`) — if these drift, update both.
	- run: cargo +nightly doc --no-deps --all-features

	python-wheel:
	name: Python wheel / ${{ matrix.os }}
	runs-on: ${{ matrix.os }}
	strategy:
	fail-fast: false
	matrix:
	os: [ubuntu-latest, macos-latest, windows-latest]
	steps:
	# Windows CFB stream names need the same autocrlf=false guard
	# we use for the Rust Windows integration job, or LFS-bound
	# corpus bytes get CRLF-mangled. Keeping this idempotent.
	- name: Disable autocrlf on Windows
	if: runner.os == 'Windows'
	run: git config --global core.autocrlf false
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- name: Fetch phi-ag/rvt sample corpus
	uses: actions/checkout@v4
	with:
	repository: phi-ag/rvt
	path: _corpus
	lfs: true
	- uses: actions/setup-python@v5
	with:
	python-version: "3.12"
	- uses: dtolnay/rust-toolchain@stable
	- uses: Swatinem/rust-cache@42dc69e1aa15d09112580998cf2ef0119e2e91ae # v2
	with:
	key: python-wheel-${{ matrix.os }}
	# Build the abi3 wheel. SEC-12/13: wheel sources live in the
	# `rvt-py/` workspace member crate; maturin follows the
	# `manifest-path` in pyproject.toml. `--release` produces an
	# optimised binary matching what users get via `pip install rvt`.
	- name: Build wheel via maturin
	uses: PyO3/maturin-action@v1
	with:
	command: build
	args: --release --out dist
	# Install into this runner's Python, then run the pytest suite
	# we ship in tests/python/. This is the real regression gate —
	# if the wheel doesn't actually import or the bindings return
	# wrong values, CI fails here.
	- name: Install the built wheel
	shell: bash
	run: \|
	python3 -m pip install --upgrade pip pytest
	python3 -m pip install --force-reinstall dist/*.whl
	- name: Run pytest integration tests
	shell: bash
	env:
	RVT_SAMPLES_DIR: ${{ github.workspace }}/_corpus/examples/Autodesk
	run: python3 -m pytest tests/python -v
	- name: Upload wheel artifact
	uses: actions/upload-artifact@v4
	with:
	name: rvt-wheel-${{ matrix.os }}
	path: dist/*.whl
	if-no-files-found: error

	# Supply-chain hardening. `cargo-deny check` enforces our deny.toml:
	# permissive-license allowlist, RustSec advisory deny, crates.io-only
	# source, no wildcards. Keeps the dependency tree auditable as the
	# project grows.
	deny:
	name: cargo deny
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- uses: EmbarkStudios/cargo-deny-action@91bf2b620e09e18d6eb78b92e7861937469acedb # v2.0.17
	with:
	command: check

	# Fails CI on any RustSec advisory published against the current
	# dependency set. Complements cargo-deny's coverage — deny has the
	# per-advisory `ignore` list for triage flexibility, audit catches
	# fresh advisories the moment they land.
	audit:
	name: cargo audit
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- uses: rustsec/audit-check@69366f33c96575abad1ee0dba8212993eecbe998 # v2.0.0
	with:
	token: ${{ secrets.GITHUB_TOKEN }}

	pii-guard:
	name: PII guard
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
	- name: Scan for PII-shaped leak patterns
	run: \|
	set -e
	echo "Scanning repo for patterns that must never be committed..."
	# Each regex catches a class of PII without spelling out
	# any specific sensitive value. The patterns are intentionally
	# broad — they're belt-and-suspenders on top of src/redact.rs.
	PATTERNS=(
	# Absolute home paths (macOS, Linux, Windows).
	'/Users/[a-z][a-z0-9._-]+/'
	'/home/[a-z][a-z0-9._-]+/'
	'C:\\Users\\[a-z][a-z0-9._-]+\\'
	# Autodesk employee-owned OneDrive authoring paths.
	'OneDrive - Autodesk\\[^\\]*\\Revit - [0-9]{3,}'
	# Autodesk build-server paths baked into DLLs.
	'F:\\Ship\\[0-9]{4}_'
	)
	failed=0
	for p in "${PATTERNS[@]}"; do
	if git grep -I -n -P "$p" -- ':!.github/workflows/' ':!src/redact.rs' 2>/dev/null; then
	echo "::error::PII-shaped pattern matched: $p"
	failed=1
	fi
	done
	if [ "$failed" -ne 0 ]; then
	exit 1
	fi
	echo "PII guard: clean."

Navigation Menu

Uh oh!

re(RE-14.1): HostObjAttr (0x006b) signal is bimodal — text vs real re… #299

Workflow file

re(RE-14.1): HostObjAttr (0x006b) signal is bimodal — text vs real re… #299

Uh oh!

Workflow file for this run