GSOC 26: render multi-material models per part by Nixxx19 · Pull Request #8955 · processing/p5.js

Nixxx19 · 2026-06-24T10:19:40Z

part of my google summer of code 2026 project, "full texture support for .mtl files in p5.js". works toward #6924.

this is phase 3: the renderer side, building on the data + parser foundation in #8879. multi-material models now actually draw, one part per material, each with its own look. public api is unchanged, you still just call loadModel() and model().

project phases for context:

phase 1: design + pipeline study
phase 2: data model + obj/mtl parser (GSOC 26: multi-material model parts (obj/mtl) #8879)
phase 3: per-part rendering (this PR)

what's in here:

the renderer loops over a geometry's parts and draws each one
each part applies its own material (fill, specular, ambient, shininess, texture) around its draw, restored after so nothing leaks between parts
a single-material geometry is its own part, so that path renders exactly as before, byte for byte
p5.GeometryPart gains what the renderer needs to treat it as a drawable (hasFillTransparency, userVertexProperties)

tests:

a multi-material model (octa-color, 8 materials) draws one fill per part and passes the instance count through
the existing obj+mtl diffuse-colour visual test still matches its reference screenshot
full unit suite green, including the webgpu and visual (pixel-compare) tests

stacked on #8879, so it should go in after that one merges.

p5-bot · 2026-06-24T10:20:54Z

Continuous Release

CDN link

https://raw.esm.sh/pr/p5@24319e8/lib/p5.min.js

Published Packages

https://pkg.pr.new/p5@24319e8

Commit hash: 59b2cce

Previous deployments

`aba7591`

`c8f0ec2`

`e303901`

`c7db707`

`c9c0cc4`

`ef33f8e`

This is an automated message.

davepagurek

Looking good! I left a few comments. One thing that would be good to clarify is what the valid states of a geometry are (should they always have parts?) so that that can be an invariant in the code and simplifies implementation.

davepagurek · 2026-07-04T13:00:47Z

+      // material around the draw.
+      const parts = geometry.parts && geometry.parts.length
+        ? geometry.parts
+        : [geometry];


What's a scenario where we'd end up with no parts?

there isn't one. the constructor always sets parts = [this] and loaders replace it, so a geometry always has at least one part. i'll make that an invariant and just loop geometry.parts directly, dropping the fallback

davepagurek · 2026-07-04T13:01:40Z

+    if (!partState) return;
+    if (partState.fill) {
+      const c = partState.fill;
+      this.states.setValue('curFillColor', [c[0], c[1], c[2], 1]);


Is there a reason why it's not possible for a part to have alpha in the fill? Although the obj/mtl spec puts transparency into its own command, in p5 fills can have alpha (e.g. geometry that comes from buildGeometry would be able to have fills with alpha if a user calls something like fill(255, 100) in the middle of building it.) Maybe we can store the transparency from a mtl in the alpha field of its fill color?

yeah that makes sense. fills carry alpha in p5, so i'll store the mtl d/Tr transparency in the fill colour's alpha (4th component) instead of forcing 1. that also covers the buildGeometry fill(255,100) case. nice, this brings transparency back in the right place

davepagurek · 2026-07-04T13:06:20Z

+  suite('multi-material rendering', function() {
+    test('draws each part and passes the instance count through', async function() {
+      const renderer = myp5.createCanvas(50, 50, myp5.WEBGL);
+      const model = await new Promise(resolve =>


I think loadModel(url) should already returns a promise, without needing to wrap it further

right, will simplify to const model = await myp5.loadModel('/test/unit/assets/octa-color.obj') and drop the wrapper

davepagurek · 2026-07-04T13:07:14Z

+
+      const spy = vi.spyOn(renderer, '_drawFills');
+      myp5.background(255);
+      myp5.model(model, 4);


The second parameter here is for WebGL instanced rendering, is that intended to be set in this test?

the 4 was to confirm the instance count reaches each part's draw, but you're right it muddies this test, i'll split it, assert per-part drawing on its own and keep the instancing check separate

davepagurek · 2026-07-04T13:08:43Z

+}
+
+function geometryPart(p5, fn) {
+  p5.GeometryPart = GeometryPart;


I think this maybe could be done in the existing geometry addon, since there's not a way to use geometry without geometryPart currently. So this file could just export the class, and p5.Geometry can import it and add it to the p5 object

agreed, no reason for a separate addon. i'll have p5.GeometryPart.js just export the class and let p5.Geometry import it and add it to the p5 object, dropping the extra registerAddon

davepagurek · 2026-07-04T13:10:14Z

+      // octa-color has several materials, so several parts.
+      expect(model.parts.length).toBeGreaterThan(1);
+
+      const spy = vi.spyOn(renderer, '_drawFills');


We should have a similar spy and assertion for _drawStrokes, which I believe we still need to implement.

good catch, strokes aren't per-part yet, only fills. i'll add per-part stroke drawing and a _drawStrokes spy + assertion alongside the fills one

Nixxx19 · 2026-07-09T06:20:07Z

Looking good! I left a few comments. One thing that would be good to clarify is what the valid states of a geometry are (should they always have parts?) so that that can be an invariant in the code and simplifies implementation.

thanks for the review! on the invariant, yeah a geometry should always have at least one part, the constructor guarantees it (parts = [this]) and loaders replace it. i'll lean into that so the renderer just loops parts with no fallback, like you said, addressing the rest now

…ort, stroke test

Nixxx19 · 2026-07-09T07:05:08Z

heads up, the failing check is just flaky ci, not anything in this pr.

Nixxx19 and others added 8 commits June 7, 2026 15:07

add geometrypart class

bfec2d4

wrap geometry in a single part

e812dae

split obj models into per-material parts

962112d

drop opacity from part state, make parts public

b086d81

Merge branch 'dev-2.0' into geometry-part-refactor

1234024

deprecate direct vertices/faces access for 3.0

0101168

Merge branch 'main' into geometry-part-refactor

40d520a

render multi-material models per part

ef33f8e

Nixxx19 changed the title ~~render multi-material models per part~~ GSOC 26: render multi-material models per part Jun 24, 2026

Nixxx19 added 2 commits June 24, 2026 16:06

Merge branch 'main' into renderer-per-part

c9c0cc4

Merge branch 'main' into renderer-per-part

c7db707

davepagurek reviewed Jul 4, 2026

View reviewed changes

Nixxx19 added 4 commits July 9, 2026 11:53

Merge branch 'main' into renderer-per-part

e303901

address review: drop parts fallback, mtl fill alpha, geometrypart exp…

c8f0ec2

…ort, stroke test

retrigger ci

aba7591

retrigger ci

59b2cce

Uh oh!

Uh oh!

Conversation

Nixxx19 commented Jun 24, 2026

Uh oh!

p5-bot Bot commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Continuous Release

CDN link

Published Packages

Uh oh!

davepagurek left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Nixxx19 commented Jul 9, 2026

Uh oh!

Nixxx19 commented Jul 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

p5-bot Bot commented Jun 24, 2026 •

edited

Loading