StudioInfinity/dyna3
2
0
Fork 1
Code Issues 56 Pull requests 1 Projects 3 Releases 4 Wiki Activity Actions

Curvature regulators #80

Merged
glen merged 21 commits from Vectornaut/dyna3:curvature-regulators into main 2025-04-21 23:40:43 +00:00
Conversation 22 Commits 21 Files changed 6 +14 -10
1 changed files with 14 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files
Showing only changes of commit 8f8e806d12 - Show all commits

24
app-proto/src/assembly.rs
View file

@ -323,11 +323,15 @@ impl Assembly {
);
}
fn insert_regulator(&self, regulator: Rc<dyn Regulator>) {
let subjects = regulator.subjects();
fn insert_regulator<T: Regulator + 'static>(&self, regulator: T) {
// add the regulator to the assembly's regulator list
let regulator_rc = Rc::new(regulator);
let key = self.regulators.update(
|regs| regs.insert(regulator.clone())
|regs| regs.insert(regulator_rc.clone())
);
// add the regulator to each subject's regulator list
let subjects = regulator_rc.subjects();
let subject_regulators: Vec<_> = self.elements.with_untracked(
|elts| subjects.into_iter().map(
|subj| elts[subj].regulators
@ -342,10 +346,10 @@ impl Assembly {
let self_for_effect = self.clone();
create_effect(move || {
console::log_1(&JsValue::from(
format!("Updated regulator with subjects {:?}", regulator.subjects())
format!("Updated regulator with subjects {:?}", regulator_rc.subjects())
));
if regulator.set_point().with(|set_pt| set_pt.is_present()) {
regulator.activate(&self_for_effect);
if regulator_rc.set_point().with(|set_pt| set_pt.is_present()) {
regulator_rc.activate(&self_for_effect);
self_for_effect.realize();
}
glen marked this conversation as resolved Outdated
glen commented 2025-04-16 21:08:20 +00:00
Outdated
Copy link

I figured out what's bothering me about this code. It's connected with my feeling that these two methods (insert_new_product_regulator and insert_new_half_curvature_regulator) should either be the same or be more highly factored than they are now. It's that there's a de-localization of concerns at the moment. It appears that the definition of the measurement of a product regulator or a half-curvature regulator currently "lives" in these insertion methods. But such code should be part of the code for that flavor of regulator, not in an "outside" method like this insertion gadget. The current code structure suggests that it would be perfectly legitimate to insert a half-curvature regulator that measures the mean of the first three coordinates. It would not be.

In other words, we don't have ideal encapsulation of regulators at the moment. Please refactor; and hopefully, after refactoring, these two functions will be identical or very thin veneer over something identical. Thanks!

I figured out what's bothering me about this code. It's connected with my feeling that these two methods (insert_new_product_regulator and insert_new_half_curvature_regulator) should either be the same or be more highly factored than they are now. It's that there's a de-localization of concerns at the moment. It appears that the definition of the measurement of a product regulator or a half-curvature regulator currently "lives" in these insertion methods. But such code should be part of the code for that flavor of regulator, not in an "outside" method like this insertion gadget. The current code structure suggests that it would be perfectly legitimate to insert a half-curvature regulator that measures the mean of the first three coordinates. It would not be. In other words, we don't have ideal encapsulation of regulators at the moment. Please refactor; and hopefully, after refactoring, these two functions will be identical or very thin veneer over something identical. Thanks!
Vectornaut commented 2025-04-16 22:59:47 +00:00
Outdated
Copy link

But such code should be part of the code for that flavor of regulator, not in an "outside" method like this insertion gadget.

I think that's a great idea. I propose moving the insertion code into a method of the Regulator trait, analogous to how the code that writes constraint problem data is kept in the pose method of the ProblemPoser trait. If you'd like, we could try to make a supertrait for anything that can be inserted into an assembly, which would be even more analogous to the ProblemPoser trait, but I don't think that's called for yet.

> But such code should be part of the code for that flavor of regulator, not in an "outside" method like this insertion gadget. I think that's a great idea. I propose moving the insertion code into a method of the `Regulator` trait, analogous to how the code that writes constraint problem data is kept in the `pose` method of the `ProblemPoser` trait. If you'd like, we could try to make a supertrait for anything that can be inserted into an assembly, which would be even more analogous to the `ProblemPoser` trait, but I don't think that's called for yet.
glen commented 2025-04-16 23:14:34 +00:00
Outdated
Copy link

yes a method of regulator trait sounds good, no not necessary to generalize into an assembly-inserting supertrait, at least not until we have something else that would also use that supertrait, i.e. no need for premature abstraction. Thanks!

yes a method of regulator trait sounds good, no not necessary to generalize into an assembly-inserting supertrait, at least not until we have something else that would also use that supertrait, i.e. no need for premature abstraction. Thanks!
Vectornaut commented 2025-04-17 21:27:31 +00:00
Outdated
Copy link

Done (commits ee8a01b – 52d9975). The code that creates and inserts regulators looks much better-organized now!

Done (commits ee8a01b – 52d9975). The code that creates and inserts regulators looks much better-organized now!
});
@ -386,11 +390,11 @@ impl Assembly {
}
);
let set_point = create_signal(SpecifiedValue::from_empty_spec());
self.insert_regulator(Rc::new(ProductRegulator {
self.insert_regulator(ProductRegulator {
subjects: subjects,
measurement: measurement,
set_point: set_point
glen marked this conversation as resolved Outdated
glen commented 2025-04-10 23:31:53 +00:00
Outdated
Copy link

This function evokes a few questions:

A) Seems like there is some duplication at least of structure/behavior here with the insert_new_product_regulator; is there anything that can be profitable factored out (some common insert_regulator common functionality that both of these can use)?

B) Why do you have to do so much work updating the guess (I think that's what's going on) when you start regulating curvature, but it seems like you don't do much of anything when you start regulating an inversive distance?

C) This very much has the look of engine code that has leaked into the assembly: nitty gritty dealing with the internal coordinates of an element. Any chance it could be implemented in the engine and just called here, or perhaps even just deferred until a pre-processing step when the realization is called? Either way, it also might make the code between insert curvature regulator and insert product regulator more similar and easier to share between the two.

This function evokes a few questions: A) Seems like there is some duplication at least of structure/behavior here with the insert_new_product_regulator; is there anything that can be profitable factored out (some common insert_regulator common functionality that both of these can use)? B) Why do you have to do so much work updating the guess (I think that's what's going on) when you start regulating curvature, but it seems like you don't do much of anything when you start regulating an inversive distance? C) This very much has the look of engine code that has leaked into the assembly: nitty gritty dealing with the internal coordinates of an element. Any chance it could be implemented in the engine and just called here, or perhaps even just deferred until a pre-processing step when the realization is called? Either way, it also might make the code between insert curvature regulator and insert product regulator more similar and easier to share between the two.
Vectornaut commented 2025-04-14 19:25:02 +00:00
Outdated
Copy link

A) Seems like there is some duplication at least of structure/behavior here with the insert_new_product_regulator; is there anything that can be profitable factored out (some common insert_regulator common functionality that both of these can use)?

I agree that the code is organized similarly, but I haven't found anything we can usefully factor out. For example, I don't see a way to put the shared organization in a shared constructor, because ProductRegulator and HalfCurvatureRegulator are different structures. The only thing that seems really straightforwardly shared is initializing the set point to

create_signal(SpecifiedValue::from_empty_spec()),

which seems like too little code to factor out profitably.

> A) Seems like there is some duplication at least of structure/behavior here with the insert_new_product_regulator; is there anything that can be profitable factored out (some common insert_regulator common functionality that both of these can use)? I agree that the code is organized similarly, but I haven't found anything we can usefully factor out. For example, I don't see a way to put the shared organization in a shared constructor, because `ProductRegulator` and `HalfCurvatureRegulator` are different structures. The only thing that seems really straightforwardly shared is initializing the set point to ```create_signal(SpecifiedValue::from_empty_spec())```, which seems like too little code to factor out profitably.
Vectornaut commented 2025-04-14 20:07:38 +00:00
Outdated
Copy link

B) Why do you have to do so much work updating the guess (I think that's what's going on) when you start regulating curvature, but it seems like you don't do much of anything when you start regulating an inversive distance?

We don't have to be so careful about updating the guess. My first prototype of the curvature regulator didn't do anything special when it started regulating; the curvature component of the representation vector would just get overwritten with the desired value at the beginning of the engine's realization routine. That seemed to work fine in the examples I played with.

I felt compelled to be careful about updating the guess for three related reasons:

  1. Just overwriting the curvature component of the representation vector messes up the vector's normalization. I worried that starting from a non-normalized guess could make the engine more likely to stall, and could make the assembly change especially erratically when the curvature regulator is turned on.
  2. I could imagine using a curvature regulator to make extreme curvature changes. For example, someone who's doing conformal geometry on the boundary of hyperbolic 3-space might switch between the Poincaré model and the upper half-space model by switching the curvature of the boundary sphere between 1 and 0. I worried that if we just took whatever realization the engine happened to find, extreme curvature changes might throw the sphere far outside the field of view.
  3. For a moderately curved sphere with nothing constrained except its curvature, one might expect that changing the curvature shouldn't affect the center much. I wanted to demonstrate that we could meet this expectation, in case you thought it was important.

I'd be open to also choosing the initial guess more carefully when we start regulating an inversive distance. That seems tricker to me, though, and I haven't felt a need for it.

> B) Why do you have to do so much work updating the guess (I think that's what's going on) when you start regulating curvature, but it seems like you don't do much of anything when you start regulating an inversive distance? We don't *have* to be so careful about updating the guess. My first prototype of the curvature regulator didn't do anything special when it started regulating; the curvature component of the representation vector would just get overwritten with the desired value at the beginning of the engine's realization routine. That seemed to work fine in the examples I played with. I felt compelled to be careful about updating the guess for three related reasons: 1. Just overwriting the curvature component of the representation vector messes up the vector's normalization. I worried that starting from a non-normalized guess could make the engine more likely to stall, and could make the assembly change especially erratically when the curvature regulator is turned on. 2. I could imagine using a curvature regulator to make extreme curvature changes. For example, someone who's doing conformal geometry on the boundary of hyperbolic 3-space might switch between the Poincaré model and the upper half-space model by switching the curvature of the boundary sphere between 1 and 0. I worried that if we just took whatever realization the engine happened to find, extreme curvature changes might throw the sphere far outside the field of view. 3. For a moderately curved sphere with nothing constrained except its curvature, one might expect that changing the curvature shouldn't affect the center much. I wanted to demonstrate that we could meet this expectation, in case you thought it was important. I'd be open to also choosing the initial guess more carefully when we start regulating an inversive distance. That seems tricker to me, though, and I haven't felt a need for it.
Vectornaut commented 2025-04-14 20:26:58 +00:00
Outdated
Copy link

C) This very much has the look of engine code that has leaked into the assembly: nitty gritty dealing with the internal coordinates of an element.

Yes, I think this code is on the border between the assembly module's responsibilities and the engine's responsibilities. I decided to put it in the assembly module because its core purpose is to decide how an otherwise unconstrained sphere should behave when you change its curvature. That seems more like user interaction than like constraint solving to me.

On the other hand, the engine does provide representation-specific implementations of other user-facing, mostly-representation-agnostic tasks, like creating a sphere from a center and a radius (sphere) or a direction, an offset, and a curvature (sphere_with_offset). If we can find a mostly-representation-agnostic way to describe how we update the curvature, maybe it would feel similar in spirit to those tasks.

> C) This very much has the look of engine code that has leaked into the assembly: nitty gritty dealing with the internal coordinates of an element. Yes, I think this code is on the border between the assembly module's responsibilities and the engine's responsibilities. I decided to put it in the assembly module because its core purpose is to decide how an otherwise unconstrained sphere should behave when you change its curvature. That seems more like user interaction than like constraint solving to me. On the other hand, the engine does provide representation-specific implementations of other user-facing, mostly-representation-agnostic tasks, like creating a sphere from a center and a radius (`sphere`) or a direction, an offset, and a curvature (`sphere_with_offset`). If we can find a mostly-representation-agnostic way to describe how we update the curvature, maybe it would feel similar in spirit to those tasks.
Vectornaut commented 2025-04-16 06:59:41 +00:00
Outdated
Copy link

On the other hand, the engine does provide representation-specific implementations of other user-facing, mostly-representation-agnostic tasks […]. If we can find a mostly-representation-agnostic way to describe how we update the curvature, maybe it would feel similar in spirit to those tasks.

I've adopted this approach in commit 4654bf0 by moving the half-curvature change routine into the engine module. I haven't come up with a representation-agnostic description of how the sphere is supposed to change, so I'm just calling the routine change_half_curvature for now.

> On the other hand, the engine does provide representation-specific implementations of other user-facing, mostly-representation-agnostic tasks […]. If we can find a mostly-representation-agnostic way to describe how we update the curvature, maybe it would feel similar in spirit to those tasks. I've adopted this approach in commit 4654bf0 by moving the half-curvature change routine into the engine module. I haven't come up with a representation-agnostic description of how the sphere is supposed to change, so I'm just calling the routine `change_half_curvature` for now.
}));
});
}
pub fn insert_new_half_curvature_regulator(&self, subject: ElementKey) {
@ -399,11 +403,11 @@ impl Assembly {
move |elts| elts[subject].representation.with(|rep| rep[3])
);
let set_point = create_signal(SpecifiedValue::from_empty_spec());
self.insert_regulator(Rc::new(HalfCurvatureRegulator {
self.insert_regulator(HalfCurvatureRegulator {
subject: subject,
measurement: measurement,
set_point: set_point
}));
});
}
// --- realization ---