fix: reframe and relight the two newest heroes for card-scale readability #84

Workflow file for this run

.github/workflows/blender-smoke.yml at 3c8d0e0

	name: Blender Smoke Test

	# Executes the snippets' and skills' headline examples inside REAL Blender, headless,
	# on the current stable (5.1.x) and the active LTS (4.5.x), and fails on any error or
	# empty-output assertion. py_compile (in validate.yml) cannot catch API-level regressions
	# like the EEVEE-id inversion, the slotted-actions boundary, the driver TypeError, or the
	# dead SDF link -- this gate runs the code so those surface in CI, not in users' files.

	on:
	workflow_dispatch: {}
	schedule:
	- cron: "0 7 * * 1" # weekly, Monday 07:00 UTC
	pull_request:
	branches: [main]

	permissions:
	contents: read

	concurrency:
	group: blender-smoke-${{ github.ref }}
	cancel-in-progress: true

	jobs:
	smoke:
	name: Blender ${{ matrix.series }} smoke
	runs-on: ubuntu-latest
	timeout-minutes: 30
	strategy:
	fail-fast: false
	matrix:
	include:
	- series: "5.1" # current stable
	- series: "4.5" # active LTS
	steps:
	- uses: actions/checkout@v7

	- name: Install Blender runtime libraries
	run: \|
	set -euo pipefail
	sudo apt-get update
	# Blender needs these shared libs even in --background (GPU/GL module init);
	# xvfb provides a virtual display so GL/EGL init does not abort the process.
	sudo apt-get install -y --no-install-recommends \
	xvfb libgl1 libegl1 libxrender1 libxxf86vm1 libxfixes3 libxi6 \
	libxkbcommon0 libsm6 libice6

	- name: Resolve and download Blender ${{ matrix.series }}
	run: \|
	set -euo pipefail
	series="${{ matrix.series }}"
	base="https://download.blender.org/release/Blender${series}/"
	echo "Listing $base"
	# pick the highest point release for this series (linux x64 portable)
	file=$(curl -fsSL "$base" \
	\| grep -oE "blender-${series}\.[0-9]+-linux-x64\.tar\.xz" \
	\| sort -V \| uniq \| tail -1)
	if [ -z "$file" ]; then
	echo "::error::Could not resolve a linux-x64 build for Blender ${series} at $base"
	exit 1
	fi
	url="${base}${file}"
	echo "Downloading $url"
	mkdir -p "$RUNNER_TEMP/bl"
	curl -fSL --retry 3 -o "$RUNNER_TEMP/bl.tar.xz" "$url"
	tar -xf "$RUNNER_TEMP/bl.tar.xz" -C "$RUNNER_TEMP/bl"
	bl=$(find "$RUNNER_TEMP/bl" -maxdepth 2 -type f -name blender \| head -1)
	if [ -z "$bl" ]; then
	echo "::error::blender binary not found after extraction"
	exit 1
	fi
	echo "BLENDER=$bl" >> "$GITHUB_ENV"

	- name: Print Blender version
	run: \|
	set -euo pipefail
	"$BLENDER" --version \| head -1
	# series guard: confirm we actually got the matrix series
	"$BLENDER" --version \| head -1 \| grep -q "Blender ${{ matrix.series }}\." \
	\|\| { echo "::error::version does not match series ${{ matrix.series }}"; exit 1; }

	- name: Run in-Blender smoke driver
	run: \|
	set -euo pipefail
	mkdir -p "$RUNNER_TEMP/out"
	xvfb-run -a "$BLENDER" --background --python tests/smoke/run_smoke.py -- "$RUNNER_TEMP/out"

	- name: Build template input scene
	run: \|
	set -euo pipefail
	xvfb-run -a "$BLENDER" --background --python tests/smoke/make_input.py -- "$RUNNER_TEMP/out/input.blend"

	- name: Headless glTF template runs (exit 0, .glb produced)
	run: \|
	set -euo pipefail
	xvfb-run -a "$BLENDER" --background "$RUNNER_TEMP/out/input.blend" \
	--python tests/smoke/tmpl_gltf.py -- \
	--output "$RUNNER_TEMP/out/out.glb" --apply-modifier SUBSURF
	test -s "$RUNNER_TEMP/out/out.glb" \|\| { echo "::error::glTF output missing/empty"; exit 1; }
	head -c4 "$RUNNER_TEMP/out/out.glb" \| grep -q "glTF" \|\| { echo "::error::not a glTF binary"; exit 1; }

	- name: Headless template no-mesh path returns exit 2
	run: \|
	set +e
	xvfb-run -a "$BLENDER" --background "$RUNNER_TEMP/out/empty.blend" \
	--python tests/smoke/tmpl_gltf.py -- --output "$RUNNER_TEMP/out/none.glb"
	code=$?
	set -e
	[ "$code" -eq 2 ] \|\| { echo "::error::expected exit 2 for no-mesh input, got $code"; exit 1; }
	echo "no-mesh exit code = $code (correct)"

	- name: Headless render template runs (exit 0, PNG produced)
	run: \|
	set -euo pipefail
	# Cycles (CPU) so this is reliable on GPU-less runners; the EEVEE-id regression
	# itself is gated in run_smoke.py via engine assignment.
	xvfb-run -a "$BLENDER" --background "$RUNNER_TEMP/out/input.blend" \
	--python tests/smoke/tmpl_render.py -- \
	--output "$RUNNER_TEMP/out/render.png" --engine CYCLES
	test -s "$RUNNER_TEMP/out/render.png" \|\| { echo "::error::render PNG missing/empty"; exit 1; }

	- name: Shipped example - swatch grid renders and self-verifies
	run: \|
	set -euo pipefail
	# Run the SHIPPED example file (not a copy). It asserts non-black AND
	# distinct-region-count == material-count internally and exits non-zero on
	# failure. Cycles (CPU) for pixels on the GPU-less runner; the example still
	# asserts the version-correct EEVEE engine id before rendering. Small/low-sample
	# for speed -- region detection holds at this size.
	xvfb-run -a "$BLENDER" --background \
	--python examples/swatch-grid/swatch_grid.py -- \
	--output "$RUNNER_TEMP/out/swatch.png" --engine cycles --samples 8 --width 640
	test -s "$RUNNER_TEMP/out/swatch.png" \|\| { echo "::error::swatch grid output missing/empty"; exit 1; }

	- name: Shipped example - turntable correctness (slotted actions)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): asserts the rotation keys drive
	# playback via the cross-version channelbag path; exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/turntable/turntable.py --

	- name: Shipped example - GN SDF remesh correctness (GridToMesh)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): asserts the SDF->GridToMesh remesh
	# yields geometry (eval vcount > 0 and != base); exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/gn-sdf-remesh/gn_sdf_remesh.py --

	- name: Shipped example - depsgraph-evaluated export
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): asserts wm.obj_export ships the
	# depsgraph-evaluated (modifier-applied) geometry -- exported vcount == evaluated
	# > base. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/depsgraph-export/depsgraph_export.py --

	- name: Shipped example - wave displace (foreach bulk IO)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): 9409 verts displaced via one
	# foreach_get + one foreach_set; asserts the Z span matches the amplitude
	# and every vertex matches the closed-form wave. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/wave-displace/wave_displace.py --

	- name: Shipped example - driver wave (driver_namespace + depsgraph)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): sixteen SCRIPTED drivers call a
	# driver_namespace function; asserts the driven values on the evaluated copy
	# AND the flushed-back original both match the closed form. Exits non-zero
	# on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/driver-wave/driver_wave.py --

	- name: Shipped example - bmesh gear (ownership + watertight topology)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): parametric gear via bmesh with
	# bm.free() in try/finally; asserts closed-form vert/edge/face counts and
	# that every edge borders exactly two faces. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/bmesh-gear/bmesh_gear.py --

	- name: Shipped example - shader node group (interface sockets + sharing)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): TintedGloss group declared via
	# tree.interface.new_socket, instanced by two materials; asserts the sockets
	# exist, the group datablock is shared (users == 2), and the per-instance
	# Tint values differ. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/shader-node-group/shader_node_group.py --

	- name: Shipped example - temp-override join (context override)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): three-step staircase joined under
	# bpy.context.temp_override; asserts one mesh remains, sources are gone,
	# topology verts=24 faces=18, and local Z spans all steps. Exits non-zero
	# on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/temp-override-join/temp_override_join.py --

	- name: Shipped example - GN instance grid (Instance on Points)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): generative GN tree instances a
	# cube on a 3x3 grid and realizes; asserts eval verts=72 faces=54, Set
	# Material carries Lime, and corner center matches the grid. Exits non-zero
	# on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/gn-instance-grid/gn_instance_grid.py --

	- name: Shipped example - shape-key blend (data API + evaluated mesh)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): relative Tall key lifts+flares
	# at value=0.5; asserts undeformed mesh stays at Basis and every evaluated
	# vert matches basis + value*(key-basis). Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/shape-key-blend/shape_key_blend.py --

	- name: Shipped example - curve bevel arc (Bezier + bevel_depth)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): beveled Bezier semicircle via
	# curve data API; asserts 8 points, bevel_depth=0.15, fill caps, eval
	# verts=1044 faces=1028, tube rests on floor. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/curve-bevel-arc/curve_bevel_arc.py --

	- name: Shipped example - compositor glare (Fog Glow + 5.x node group)
	run: \|
	set -euo pipefail
	# Two tiny 1-sample Cycles renders (compositor on vs off, ~seconds each):
	# asserts the cross-version compositor plumbing (compositing_node_group
	# on 5.x, scene.node_tree on 4.x), the Glare Fog Glow configuration, a
	# strictly-falling halo with the compositor on, and exactly-zero halo
	# pixels with it off. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/compositor-glare/compositor_glare.py --

	- name: Shipped example - damped-track aim (DAMPED_TRACK data API)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): twelve needles aim at an
	# emissive core via Object.constraints.new('DAMPED_TRACK') on TRACK_Z;
	# asserts one unmuted constraint per needle and evaluated local +Z
	# aligns toward the core (dot ≥ 0.998). Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/damped-track-aim/damped_track_aim.py --

	- name: Shipped example - color attribute wheel (color_attributes CORNER domain)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): an HSV wheel disc colored via
	# mesh.color_attributes.new() on the CORNER domain; asserts the attribute
	# is sized to loop count (not vertex count), is active_color, and probe
	# loops match the closed-form HSV for their vertex. Exits non-zero on
	# failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/color-attribute-wheel/color_attribute_wheel.py --

	- name: Shipped example - parent inverse orrery (matrix_parent_inverse)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): a data-API orrery whose arms,
	# planets, and moon are parented with the keep-world idiom; asserts bare
	# `.parent =` teleports a probe, matrix_parent_inverse restores it exactly,
	# matrix_world is stale until view_layer.update(), and every orbit lands
	# on its closed form. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/parent-inverse-orrery/parent_inverse_orrery.py --

	- name: Shipped example - grease pencil rosette (GPv3 attribute API)
	run: \|
	set -euo pipefail
	# Frame-independent check only (no render): five nested rose curves drawn
	# as GPv3 strokes; asserts the version-gated datablock address (4.5:
	# grease_pencils_v3 with grease_pencils still legacy; 5.x: grease_pencils
	# is GPv3, legacy names gone), lazy attribute-layer materialization from
	# point writes, and a closed-form round-trip of every position through
	# the raw POINT attribute buffer. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/grease-pencil-rosette/grease_pencil_rosette.py --

	- name: Shipped example - armature bend (edit_bones + LBS closed form)
	run: \|
	set -euo pipefail
	# Check only (no render): a four-bone chain skins a tapered tube via
	# name-bound vertex groups; asserts edit_bones is empty outside edit
	# mode and holds the closed-form chain inside it, then compares every
	# depsgraph-evaluated vertex against re-implemented linear blend
	# skinning (root ring pinned, tip deflected — a straight tube fails).
	# Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/armature-bend/armature_bend.py --

	- name: Shipped example - text version stamp (TextCurve solids + version string)
	run: \|
	set -euo pipefail
	# Check only (no render): a TextCurve whose body is the live
	# bpy.app.version_string; asserts the Curve subclass and built-in
	# font contracts, that flat text is filled but planar, the closed
	# form z-extent = 2*(extrude+bevel_depth), body edits regenerating
	# geometry, that version_string starts with the dotted version tuple
	# ("4.5.11 LTS" is not bare semver), and that a Mesh reference dies
	# at to_mesh_clear(). Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/text-version-stamp/text_version_stamp.py --

	- name: Shipped example - image pixels testcard (flat RGBA buffer + save lifecycle)
	run: \|
	set -euo pipefail
	# Check only (no render): a procedural broadcast test card written with
	# one pixels.foreach_set; asserts the buffer is always flat RGBA
	# (channels == 4 even with alpha=False, RGB-stride writes raise), the
	# byte round-trip error is <= 0.5/255 and strictly > 0 (storage really
	# is 8-bit) while float_buffer=True round-trips at float32 precision,
	# scale() reallocates so stale-size bulk reads raise, and save() flips
	# source to FILE and drops the buffer (pixels silently re-load from
	# disk — proven with an imposter file) while save_render() preserves
	# it. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/image-pixels-testcard/image_pixels_testcard.py --

	- name: Shipped example - uv layer grid (calc_uvs silent no-op)
	run: \|
	set -euo pipefail
	# Check only (no render): create_grid(..., calc_uvs=True) without a
	# pre-existing UV layer leaves 0 layers (silent no-op); pre-create +
	# calc_uvs fills loops to the closed-form grid UVs
	# ((x/size+1)/2, (y/size+1)/2) within 1e-6 including mesh persistence;
	# explicit loop assignment is the calc_uvs-free fallback. Exits
	# non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/uv-layer-grid/uv_layer_grid.py --

	- name: Shipped example - png vs exr alpha (float PNG false unpremul)
	run: \|
	set -euo pipefail
	# Check only (no render): float_buffer=True Image.save() to PNG writes
	# RGBA16 and unpremultiplies as if associated-alpha, so straight-authored
	# dark values at low alpha clamp to white (closed-form err 0.98 at
	# RGB 0.02 / a=1/255); OpenEXR round-trips within 1e-5; byte→PNG is
	# straight RGBA8; EXR color_mode=RGB drops alpha to opaque. Exits
	# non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/png-exr-alpha/png_exr_alpha.py --

	- name: Shipped example - vse cut list (sequencer 4.5 vs 5.x rename)
	run: \|
	set -euo pipefail
	# Check only (no render): asserts the accessor rename (.sequences
	# gone on 5.x, bridged on 4.5), the new_effect end-kwarg rename
	# (frame_end on 4.5, length on 5.x — the wrong kwarg TypeErrors),
	# closed-form spans on each version's canonical accessors
	# (frame_final_* vs left_handle/right_handle/duration with the
	# deprecation bridge), GC wiring and clamping, scene-strip span,
	# and a save/reload round-trip. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/vse-cut-list/vse_cut_list.py --
	# Compositing witness on a tiny Cycles render: mosaic cells carry
	# their strip colors, the cross cell is a true mid blend, and a
	# consumed cross source never composites independently.
	xvfb-run -a "$BLENDER" --background \
	--python examples/vse-cut-list/vse_cut_list.py -- \
	--check-pixels --engine cycles

	- name: Shipped example - glTF export round-trip (+Y-up, applied modifiers, UVs)
	run: \|
	set -euo pipefail
	# Check only (no render): a 35-part supply crate exported GLTF_SEPARATE
	# and re-imported; asserts the exporter/importer kwarg set still exists
	# (RNA guard), the on-disk +Y-up conversion ((x,y,z)->(x,z,-y) baked
	# into POSITION bounds, no node rotation), disk vertex count ==
	# evaluated loop count (export_apply), V-flipped TEXCOORD_0, and the
	# full round-trip of positions/normals/UVs/per-triangle material
	# bindings within float tolerances. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/gltf-export-roundtrip/gltf_export_roundtrip.py --

	- name: Shipped example - LOD decimate chain (depsgraph + ratio bounds)
	run: \|
	set -euo pipefail
	# Check only (no render): a lathed retro rocket evaluated at LOD0/1/2
	# through the Decimate modifier; asserts closed-form base topology and
	# bbox, evaluated-vs-original non-destructiveness (obj.data unchanged),
	# evaluated triangle counts within 5% of ratio*base, and silhouette
	# (bbox) preservation within 1e-3. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/lod-decimate-chain/lod_decimate_chain.py --

	- name: Shipped example - vertex weight limit (4-influence cap + LBS)
	run: \|
	set -euo pipefail
	# Check only (no render): a five-bone mech arm with rich 5-influence
	# boots, pruned to the 4-influence engine cap via the data API;
	# asserts pre/post-limit influence counts, unit weight sums, pose
	# preservation (pre vs post pruning), exact LBS from the mesh's own
	# deform layer, and a pinned Root mount. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/vertex-weight-limit/vertex_weight_limit.py --

	- name: Shipped example - triangulate + tangents (mikktspace contract)
	run: \|
	set -euo pipefail
	# Check only (no render): a lathed buckler with polar/strip/planar
	# UV charts; asserts closed-form loop-triangle count, unit and
	# orthogonal tangent frames, bitangent == sign*(n x t), tangents
	# matching the independent edge/UV-delta formula on smooth-field
	# triangles, zero flips off seams, and the re-fetched UV layer
	# surviving calc_tangents. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/triangulate-tangents/triangulate_tangents.py --

	- name: Shipped example - glTF skin round-trip (skins, weights, deform)
	run: \|
	set -euo pipefail
	# Check only (no render): a 7-bone mech scorpion exported with
	# export_skins and re-imported; asserts skins[0] joints, JOINTS_0/
	# WEIGHTS_0 unit sums, bone/parent/rest-matrix round-trip, bit-exact
	# weights via position-key lookup, and identical deformation of the
	# re-imported rig. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/gltf-skin-roundtrip/gltf_skin_roundtrip.py --

	- name: Shipped example - VSE gamma cross (blend-curve closed form)
	run: \|
	set -euo pipefail
	# Check only: renders tiny frames across a GAMMA_CROSS and asserts
	# every sample against the gamma-0.5 closed form ((1-t)*sqrt(A) +
	# t*sqrt(B))^2 with t = (frame-start)/duration, plus the material
	# lerp deviation at mid. Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/vse-gamma-cross/vse_gamma_cross.py --

	- name: Shipped example - light link studio (receiver collections in pixels)
	run: \|
	set -euo pipefail
	# Check only (two tiny Cycles renders): a key linked to the hero
	# collection must light only the hero (luminance ratio >= 3x at
	# projected centers), and unlinking must raise the decoy >= 50%
	# with hero drift <= 5%. Also guards the Object-level API path
	# (ld.light_linking is an AttributeError). Exits non-zero on failure.
	xvfb-run -a "$BLENDER" --background \
	--python examples/light-link-studio/light_link_studio.py --

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: reframe and relight the two newest heroes for card-scale readability #84

Workflow file

fix: reframe and relight the two newest heroes for card-scale readability #84

Uh oh!

Workflow file for this run