New v4.0.0 alpha release

[jacomyal]

Sigma.js: New v4.0.0 alpha release

Hey everyone,

I've been working on a new version of sigma for quite a few months, and I'm really happy to announce the first alpha release of sigma v4. This post is a quick tour of what's in it and how we got there.

Why v4

The need for a v4 started from a few issues that turned out to be fundamentally incompatible with v3's architecture. There wasn't a clean way to fix them without breaking things, which is what motivated v4 in the first place.

• Nodes, edges and labels render in separate layers, so all nodes are always over all edges (#1233)
• Labels were rendered in canvas, so they always appeared hover the WebGL scene, and sigma cannot support label events (#1319)
• Composed programs don't render "all at once", causing weird interleaving artifacts (#1427)
• Sigma consumed too many WebGL contexts per instance (#1321)

Additionally, along the way, some other pain points surfaced:

• Nodes and edges were painful to customize (v3's nodeProgramClasses and edgeProgramClasses)
• Most other graph visualization tools expose a style-like API, which makes customizing item appearance much more approachable
• Node shapes had a leaky abstraction: rendering and interaction supported arbitrary shapes, but labels and edge attachments still assumed circles under the hood
• Similarly, edge paths had to be both implemented in GLSL for the GPU (for rendering and picking), and in JavaScript and canvas for the CPU (for labels rendering)

In late 2025, at @ouestware, we had some time to dedicate to this task, and we've been lucky again to have @arthur-bigeard-gdotv from G.V() sponsor us.

OIT, and what actually worked

The original plan was to focus on Order-independent Transparency. That's still what the milestone for v4 mentions. But after several attempts, OIT wasn't the solution: I couldn't get it to combine fully opaque items with smooth opacity-based antialiasing.

Depth ended up being handled differently:

• Migrate all rendering to WebGL (no more canvas)
• Move all items to the same single WebGL context
• In addition to the z-index, implement a layers management system, that allows rendering all types of items at different depths

Then, about the node shapes issues, when looking into where to anchor edges or labels to any node shape, I found that Signed Distance Fields are the consensus answer. SDFs gave clean math for label and edge attachment. And, even better, they come with an awesome comprehensive GLSL catalog, thanks to Inigo Quilez!

That made the rest of v4 possible: styles and primitives. Sigma no longer needs a custom program per node or edge renderer. Instead, it now compiles one unified node program and one unified edge program, and everything is configured declaratively.

What's in v4

Besides the architectural rework, v4 fills a lot of gaps from earlier versions:

• Declarative styles and primitives APIs for nodes and edges
• Built-in node dragging
• Built-in cursor management
• Parallel edges and self-loops handled natively
• WebGL-rendered labels, which also enables label picking and events
• Edge labels through the same SDF pipeline as node labels
• Node backdrops (shape-aware shadows and borders behind nodes and labels, previously NodeHoverDrawingFunction)
• Label placement (directional anchors for node labels, along-path placement for edge labels)
• Composable edge layers (plain, dashed, ...)

And there's more in the codebase. The full documentation is at v4.sigmajs.org (with various new examples), and the changelog is available here.

Try it

I'd love for people to try upgrading real projects against the alpha and report back, especially on migration pain points, performance, and anything the new styles API makes harder rather than easier. New issues on the repo are all welcome, please make sure to use the v4 label.

Thanks!

[rahst12]

I'm on the v3 line of Sigma.. How mature would you say the alpha release is? If I'm starting or converting a project to use Sigma, is v4 at a point where I should be starting here?

Thanks!

[jacomyal]

Hi @rahst12,

Thanks to @arthur-bigeard-gdotv who's been migrating early so that we can have proper feedback on the v4, there've been a lot of bug fixes since the first alpha release, and the APIs are quite more stable as well. It's now more of a beta version, I just forgot to change the flag...

Basically, here's the current state:

• I consider the current state as full-featured, and the APIs should not move any longer
• The v4 website documentation is up to date, and quite exhaustive
• I am satisfied with the current test coverage, and it's now easy for me to add "regression e2e tests", so I should be quite reactive on new v4 bugs
• There are surely more bugs to find, mostly in interactions between new features that haven't been tested together yet

So, I'd say you can probably start migrating, you should just expect some eventual bugs - in which case, I'll tackle them as quick as I can. And also, the more people start migrating, the faster we will have a final release 🙏

Best!