Move Constraint solver into a separate crate and put it onto a solid mathematical foundation

[TobiasJacob]

Move Constraint solver into a separate crate and put it onto a solid mathematical foundation

[MattFerraro]

I think it would be best to do a completely clean break here rather than keeping it inside the CADmium repo, so I've created https://github.com/CADmium-Co/ISOtope

I picked "ISOtope" because it's an Iterative SOlver and because it sorta fits the naming scheme of elemental things. Also "tope" like topography, and how it's all based on gradient descent? IDK, we'll workshop it.

Don't worry about matching the existing API too much--it was whipped up super fast as a proof of concept and it may well be bad. Build the right API for a standalone 2D Sketch Solver and CADmium will adapt to meet it.

[TobiasJacob]

I was able to make some progress in the repo https://github.com/CADmium-Co/ISOtope.

There is still room for improvement, but I think we can start working on integrating it into the software before the codebases diverge too far.

[TobiasJacob]

Yeah, you can open up an issue for this in the new isotope repo

[MattFerraro]

Thank you Tobias, what a great writeup!

I have no attachment to the second-order nature of the current approach--it was just easy to think about during prototyping. I think the first order formulation is likely to be much better!

I strongly support the idea of breaking out the 2D constraint solver into its own standalone crate and giving it as few dependencies as possible. I also love the idea of trying out a few different approaches to the gradient descent step so we can weigh the pros and cons.

If you feel confident to lead this effort, I will support you however I can!

[TobiasJacob]

Hi @MattFerraro ,

Thanks, great to hear. I think I can start working on refactoring the code to use gradient descent. I can also document the maths behind it in a readme. This shouldnt be too much of a change.

Refactoring the code into a separate crate will probably be a big pullrequest leading to merge conflicts with other braches, so I'm not sure how to deal with this.

You wrote on Discord that you are working on settigung up a Github Organization anyways, which would make it easier to move the code around.

I guess suddenly coordinating 800+ people puts a bit of unexpected stress on you 😅 so yeah, just let me know what you think a good way forward

[MattFerraro]

This is fantastic work, Tobias! Unbelievable job

I pulled it locally and ran cargo test. On my m1 mac I'm seeing:

point_a: Point2 { data: [[0.5700834586643445, -0.06745530512562747]], gradient: [[-25.709630952671287, -3.5250742668599533]] }
point_b: Point2 { data: [[-1.3913236621093832, -0.33256804099923265]], gradient: [[-0.03432845891903502, -0.13300160678809736]] }
point_c: Point2 { data: [[-1.7858823110455389, 2.616758693265621]], gradient: [[0.04124606881310794, -0.04022156009781125]] }
point_d: Point2 { data: [[0.17054457073731163, 2.8818523972647134]], gradient: [[0.02043760840499649, -0.022759707622775023]] }
point_reference: Point2 { data: [[0.5592929825521086, -0.06193753782611579]], gradient: [[25.811662378731278, 3.5916709580860164]] }
thread 'examples::test_rectangle_rotated::tests::test_rectangle_rotated' panicked at src/examples/test_rectangle_rotated.rs:113:9:
assertion failed: (point_a.as_ref().borrow().data() - Vector2::new(0.0, 0.0)).norm() < 0.001


failures:
    examples::test_rectangle_rotated::tests::test_rectangle_rotated

test result: FAILED. 17 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 6.30s

Do you get the same?

[TobiasJacob]

Yes, one test case is not yet working, its written down in the Todos of the Readme.

[TobiasJacob]

PS: Some of the constraints could be improved still, but I think the overall structure shouldnt change too much anymore.
Also having UI would help a lot for debugging, thats why integrating it into the code base would be helpful.

[hrgdavor]

have you considered a wasm build aswell ? for folks that would like to use this directly ?

[MattFerraro]

Excellent work @TobiasJacob! I'm marking this done and we can have other tickets for integrating into the UI more fully