USD Export Evaluation Mode

A SUBSURF cube exported through wm.usd_export that witnesses evaluation_mode='RENDER' against 'VIEWPORT' — the contract usd-export-evaluation-mode.py names but does not prove.

Catmull-Clark on a cube is closed form: verts = 2 + 6 × 4^n (n=1 → 26, n=2 → 98). TESSELLATE + VIEWPORT writes 26 points / 24 quads; TESSELLATE + RENDER writes 98 / 96. Default export_subdivision='BEST_MATCH' writes the 8-vert cage plus subdivisionScheme = catmullClark, so evaluation_mode is silent — both files are the cage. TESSELLATE is what makes the mode observable.

The still shows the exported files themselves. The --output path builds a turned goblet on an 8-sided lathe cage with SUBSURF levels=0 / render_levels=3, exports it twice through the same wm.usd_export call the check proves (TESSELLATE + VIEWPORT, TESSELLATE + RENDER), re-imports both USDA files with wm.usd_import, and stages the two imported meshes. Left is the VIEWPORT file (the raw 8-sided cage, 130 points), right the RENDER file (8450 points). Both are flat-shaded with the same material, so geometry is the only difference, and the point count on each placard is read from the re-imported mesh at render time. The goblet widens the cube's L1/L2 split to L0/L3 so the difference reads at thumbnail size. The render path also asserts that the re-imported RENDER goblet is denser than the VIEWPORT one (exit 6). Placards use the bundled DejaVu Sans Mono, whose 1 cannot be misread as I.

What failure each check would catch:

• exit 3 — VIEWPORT TESSELLATE did not write the L1 closed form
• exit 4 — RENDER TESSELLATE wrote viewport quality or the BEST_MATCH cage (--evaluation-mode VIEWPORT measured 26/24; --subdivision BEST_MATCH measured 8/6 catmullClark)
• exit 5 — BEST_MATCH stopped writing the cage
• exit 6 — RENDER and VIEWPORT files are identical

Staging: Catmull-Clark pulls the surface inside the cage, and each level shrinks it by a different amount, so each re-imported goblet is seated on its plinth from its own vertices rather than from the cage's z.

Probed on the CI Linux portables Blender 5.2.1 LTS (9e2066aef7ef) and Blender 4.5.13 LTS (daeeeca98fb0): wm.usd_export exists, poll() is true in --background, evaluation_mode is {RENDER, VIEWPORT}. The --output render path measures framing via examples/gallery_framing.py (exit 10 on violation).

Run#

blender --background --python usd_export_evaluation_mode.py --
blender --background --python usd_export_evaluation_mode.py -- --output u.png

Exit codes#

Per-script sequential checks. 9 is a valid check code; there is no rule against it. 10 is the shared framing helper.

The blender-smoke workflow runs the check on Blender 5.2 LTS and 4.5 LTS (5.1 on the weekly cron, the needs-5.1 PR label, or manual dispatch). Smoke does not pass --output. Its catalog falsifiers are --subdivision BEST_MATCH (expects exit 4) and --evaluation-mode VIEWPORT (expects exit 4).