sdf (originally sdfx)

A rewrite of the original CAD package sdfx for generating 2D and 3D geometry using Go.

• Objects are modelled with 2d and 3d signed distance functions (SDFs).
• Objects are defined with Go code.
• Objects are rendered to an STL file to be viewed and/or 3d printed.

Highlights

• 3d and 2d objects modelled with signed distance functions (SDFs)
• Minimal and idiomatic API
• End-to-end testing using image comparison
• must and form packages provide panicking and normal error handling basic shape generation APIs for different scenarios
• Dead-simple, single method Renderer interface

Examples

For real-world examples with images see examples directory README.

See images of rendered shapes in render/testdata.

Here is a rendered bolt from one of the unit tests under form3_test.go

Roadmap

• Add a 2D renderer and it's respective Renderer2 interface.
• Make 3D renderer multicore

Comparison

deadsy/sdfx

Advantages of deadsy/sdfx:

• Widely used
• More helper functions
• Working 2D renderer

Advantages of soypat/sdf:

• Very fast rendering
  • deadsy/sdfx is over 2 times slower and has ~5 times more allocations.
• Minimal and idiomatic API
• Renderer interface is dead-simple, idiomatic Go and not limited to SDFs
  • deadsy/sdfx Renderer3 interface has filled render package with technical debt. See Questionable API design.
• Has SDFUnion and SDFDiff interfaces for blending shapes easily
• No nil valued SDFs
  • deadsy/sdfx internally makes use of nil SDFs as "empty" objects. This can later cause panics during rendering well after the point of failure causing hard to debug issues.
• Well defined package organization.
  • deadsy/sdfx dumps helper and utility functions in sdf
• End-to-end tested.
  • Ensures functioning renderer and SDF functions using image comparison preventing accidental changes.
• Error-free API under must3 and must2 packages for makers.
  • For simple projects these packages allow for streamlined error handling process using panic instead of returned errors.
  • deadsy/sdfx only allows for Go-style error handling like the form3 and form2 packages.
• Sound use of math package for best precision and overflow prevention.
  • math.Hypot used for all length calculations. deadsy/sdfx does not use math.Hypot.
• Uses gonum's spatial package
  • sdfx has own vector types with methods which hurt code legibility

Contributing

See CONTRIBUTING.

Why was sdfx rewritten?

The original sdfx package is amazing. I thank deadsy for putting all that great work into making an amazing tool I use daily. That said, there are some things that were not compatible with my needs:

Performance

sdfx is needlessly slow. Here is a benchmark rendering a threaded bolt:

$ go test -benchmem -run=^$ -bench ^(BenchmarkSDFXBolt|BenchmarkBolt)$ ./render
goos: linux
goarch: amd64
pkg: github.com/soypat/sdf/render
cpu: AMD Ryzen 5 3400G with Radeon Vega Graphics    
BenchmarkSDFXBolt-8   	       6	 196941244 ns/op	14700786 B/op	   98261 allocs/op
BenchmarkBolt-8       	      13	  87547265 ns/op	18136785 B/op	   20754 allocs/op
PASS
ok  	github.com/soypat/sdf/render	4.390s

BenchmarkBolt-8 is this implementation of Octree. BenchmarkSDFXBolt-8 is the sdfx implementation of said algorithm.

Questionable API design

• deadsy/sdfx#48 Vector API redesign
• deadsy/sdfx#35 Better STL save functions.
• deadsy/sdfx#50 Removing returned errors from shape generation functions

The vector math functions are methods which yield hard to follow operations. i.e:

return bb.Min.Add(bb.Size().Mul(i.ToV3().DivScalar(float64(node.meshSize)).
    Div(node.cellCounts.ToV3().DivScalar(float64(node.meshSize))))) // actual code from original sdfx.

A more pressing issue was the Renderer3 interface definition method, Render

type Renderer3 interface {
    // ...
    Render(s sdf.SDF3, meshCells int, output chan<- *Triangle3)
}

This presented a few problems:

1. Raises many questions about usage of the function Render- who closes the channel? Does this function block? Do I have to call it as a goroutine?

2. To implement a renderer one needs to bake in concurrency which is a hard thing to get right from the start. This also means all rendering code besides having the responsibility of computing geometry, it also has to handle concurrency features of the language. This leads to rendering functions with dual responsibility- compute geometry and also handle the multi-core aspect of the computation making code harder to maintain in the long run

3. Using a channel to send individual triangles is probably a bottleneck.

4. I would liken meshCells to an implementation detail of the renderer used. This can be passed as an argument when instantiating the renderer used.

5. Who's to say we have to limit ourselves to signed distance functions? With the new proposed Renderer interface this is no longer the case.

sdf and sdfx consolidation

None planned.

My understanding is the sdfx author has a very different design goal to what I envision. See the bullet-list of issues at the start of Questionable API design.