Solidify Even Thickness

A runnable example that solidifies an open, folded strip with the Solidify modifier and proves, fold by fold, what use_even_offset does to the shell's thickness. The strip is a zigzag profile extruded along Y, folded at bend angles of 60°, 90° and 120°. The evaluated shells are read through the depsgraph lifetime contract from depsgraph-and-evaluated-data (evaluated_get → to_mesh → to_mesh_clear), and the strip is built with bmesh in a try/finally, as mesh-editing-and-bmesh requires.

What it witnesses: in the default Simple mode (solidify_mode = 'EXTRUDE') with offset = -1, the original surface stays put and every vertex gets a copy pushed thickness along its vertex normal. At a fold the vertex normal is the bisector of the two face normals, which meet at the bend angle φ. So:

• without use_even_offset the copy moves t along the bisector, and the shell's perpendicular thickness at the fold is t · cos(φ/2), which is t · sin(θ/2) for the interior angle θ = 180° − φ. With t = 0.16 that is 0.1386 at 60°, 0.1131 at 90° and 0.0800 at 120°: half the thickness you asked for;
• with use_even_offset the copy moves t / cos(φ/2) along the bisector, and the thickness is exactly t at every fold.

The check builds both closed forms from the profile itself, taking the face planes from the segment directions and never from Blender's normals. It measures each shell's perpendicular thickness at every fold and free edge and requires agreement to 1e-5 (measured: 3.7e-8). It also asserts the topology the measurement relies on: 2N evaluated vertices, the first N being the untouched original surface, and copy i at vertex N + i. Finally it checks that the thinning is real, so the witness cannot pass vacuously.

The trap it exposes: use_even_offset is off by default. A script that solidifies a bent panel, a folded bracket or a box with thickness = t gets a wall that is noticeably thinner at every corner, down to half at a 120° bend, and nothing errors. --no-even leaves the flag off on the shell that the check expects to be even, and check 4 fails with the measured thickness.

The still shows both shells in teal glaze side by side on a walnut plinth, viewed nearly end-on. The cut section, Solidify's rim, is drawn in selection-orange enamel through material_offset_rim, so the band's width is the shell's thickness. On the left, with use_even_offset = False, the band pinches at every fold, and to half at the sharp V. A red line on that cut face traces where a t-thick shell's inner face would run: it is the even shell's own evaluated copy positions, placed through the plain shell's transform, so the gap between the orange band and the red line is the thickness lost at each fold, drawn at true size. On the right, with use_even_offset = True, the band stays one width all the way round. Engraved brass plaques name each setting, and brass dowels hold the raised folds off the plinth.

Run#

# Cheap correctness check (no render) — the CI check:
blender --background --python solidify_even_thickness.py --

# Falsifier: leave use_even_offset off on the shell checked as even. Must exit 4.
blender --background --python solidify_even_thickness.py -- --no-even

# Also render a still (EEVEE on a GPU host; use --engine cycles on GPU-less hosts):
blender --background --python solidify_even_thickness.py -- --output solidify.png --engine cycles

Version notes#

The Solidify modifier properties used here (solidify_mode, thickness, offset, use_even_offset, use_rim, material_offset_rim) and the output vertex order (originals first, then copies) are the same on 4.5 LTS, 5.1 and 5.2 LTS. The defaults are also the same: 'EXTRUDE', offset −1, even off, rim on, quality normals off. All values were measured identically on 4.5.11, 5.1.2 and 5.2.1.

Exit codes#

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 falsifier is --no-even (expects exit 4).