Reimplementing constraint solver

[DeepSOIC]

Implemented:
* ParameterStore
* ParameterRef
(with equality-constraint support by redirection)
* ParameterSet
* ValueSet

Next up: base classes for constraints.

[Kunda1]

Implemented:
....
Next up: base classes for constraints.

Woohoo!

[Zolko]

Next up: base classes for constraints.

9. use in assembly workbench

How do you intend to represent the constraints ?
What sort of FreeCAD/Python object do you plan to use ?
Do you think that App::FeaturePython objects could suit your needs ?

[DeepSOIC]

How do you intend to represent the constraints ?

An object that provides an error function with derivative information.

What sort of FreeCAD/Python object do you plan to use ?

Base class. It will be up to the "user" (i.e., sketcher/assembly developer) to formulate DocumentObject-based variant.

I plan introducing a python-extendable variant of a constraint, something like ConstraintPython, to work similarly to how App::FeaturePython works in terms of how it is extended by python (i.e., by means of a Proxy object)

[triplus]

Finally! Been waiting for something like that, since that now legendary assembly video. Especially as it looks like the effort, a traditional assembly solver, was put on a halt, half a decade back:

https://forum.freecadweb.org/viewtopic. ... 02#p344274

Much of what you wrote makes sense. Things like being able to introduce an addition relation in another workbench. Running multiple (Sketcher) solvers on different cores, sounds fun, not sure if practical. But this was mentioned in the past a few times already. Therefore if you have interest in making it happen, for sure, why not.

P.S. Likely this is a multi-year effort, though.

[triplus]

How far away do you believe you are, demonstrating (pre-alpha version) one assembly relation? I feel that if you could demosntrate that in sensible amount of time and in code, that should attract developers.

[DeepSOIC]

How far away do you believe you are, demonstrating (pre-alpha version) one assembly relation? I feel that if you could demosntrate that in sensible amount of time and in code, that should attract developers.

Halfway there, probably. But I'm slowing down, unfortunately.

[triplus]

OK, i will read that as we are in trouble.

What if you would, for now, remove goals, such as to improve current Sketcher solver, to run multiple Sketcher solvers at the same time, 3D Sketcher ... And the only goal remaining would be to demonstrate one assembly relation (pre-alpha version). Would that be doable in reasonable amount of time and before you reach an on halt state?

[DeepSOIC]

What if you would, for now, remove goals, such as to improve current Sketcher solver

That's pointless, isn't it? The whole point of me doing it started as a "improve the solver" initiative. Making good py interface is taking a lot of effort, possibly a good 70% of it

3D Sketcher

3D sketcher goal isn't getting in the way, it's mostly an extensive growth.

to run multiple Sketcher solvers at the same time,

There is quite a lot of stuff done in the direction. Revising this point is going to further slow me down at this point, I think.

[triplus]

The way i see it, you have a basic idea sorted out, on how you will try to tackle each of this areas. All of them are in my opinion noble goals and would benefit FreeCAD greatly. What i guess i am asking is, would it be possible to demonstrate in code, one assembly relation (pre-alpha version). It doesn't have to have Python interface for now. That would in my opinion be the breakthrough we need.