I'm making a physics extension/Help with 3D physics extension?

[Brackets-Coder]

video.mp4

I'm making a physics extension. It's not near done yet; this is a proof of concept, but it still kinda works.

it uses cannon.js for physics calculations but I couldn't use ammo.js because it relies on wasm and isn't native javascript...

So even though cannon is kinda old it still works.

Goals for this extension:

• Work well with any 3D renderer
• Great performance
• Support Spheres, planes, boxes, cylinders, convex hulls, and triangle meshes (just spheres and boxes as of now)
• Discrete collision detection (basically <is [body1] touching [body2]?> and <get all bodies touching [body]>)
• Constraint support
  • Point to point
  • Distance
  • Hinge
  • Lock
  • ConeTwist
• Enable/disable collision response for bodies (still detect collision but don't react to forces)

basically it's supposed to support everything cannon.js supports.

Don't know why the video is starting at the middle instead of the beginning

[CubesterYT]

This is amazing! The things you do are awesome, it's great having you a part of this community!

(If you have Discord, you should definitely join our community there!)

[Brackets-Coder]

Thank you. My favorite form of development by far is making websites and Turbowarp extensions, so I try do what I can to learn more and help more.

P.S. I don't have Discord and haven't planned on getting it just because of people on the internet being people on the internet (if you know what I mean)
My main communities will continue to be Scratch and GitHub if that's fine with you.

[CubesterYT]

Honestly, our community is a great place compared to most places (a lot more mature and trustworthy), but I respect your stance!

[Brackets-Coder]

@CubesterYT I don't think this is how physics is supposed to work...

Do you have any idea why this is happening? My guess is it has something to do with quaternion to XYZ degree Euler conversion or something related to gimbal lock or something like that but I thought I fixed that.
It's probably not the library because other renderer implementations work.

It might be a better question for @Xeltalliv because of Simple3D integration and things

video.mp4

[CubesterYT]

This is definitely a question for @Xeltalliv

[Xeltalliv]

Wrong rotation order probably? Possibly wrong signs on rotations as well.

Also, the physics engine probably has the model space to world space matrices already somewhere inside, so you should be able to feed those directly into Simple3D using "start with saved in [list v] at (1)" block, instead of dealing with conversions between different rotation formats.

[Xeltalliv]

Also, from what it seems, in your extension you can only have 1 collider per rigidbody, which is a very big limitation. For example, my custom physics engine allows for multiple colliders per rigidbody, which is how tables and chairs in this scene are possible:

[Brackets-Coder]

@Xeltalliv It's definitely due to some inconsistency of the transformations between your renderer/my engine/my implementation of physics. I considered the possibility of wrong rotation order/sign, but there's a script that turns the quaternions into XYZ degree Eulers. I think that's the main issue, there could also be some gimbal lock things going on because I've seen evidence for that in other examples. Cannon.js actually has quaternion to euler functions, but attempting to pass in the "XYZ" argument to re-order the Euler returns an error that "the order 'XYZ' isn't supported yet", so I had to write some custom conversion scripts that are probably the main issue. Cannon.js is an old(er) library but it's the only native JS solution that's still somewhat prevalent. Ammo requires WebAssembly so it can't be a TW extension

Also Cannon does support multiple shapes/colliders per body but I haven't implemented blocks that add new shapes to existing bodies yet

[Brackets-Coder]

This gets even weirder when I use an infinite plane instead of a finite, thin box

video.mp4

[Brackets-Coder]

I think I'm just going to re-write this whole extension from scratch again. Of course, I'll save a backup of the current version but Cannon.js just has too many limitations and the main thing holding me back from using Ammo.js was the fact that I thought it required WebAssembly but I think I figured out how to minify it and use the native JS version so I'll try that and see if it's any better

[Brackets-Coder]

@CubesterYT I'm having some issues trying to get Ammo.js to work. It works like this:

But not like this:

Of course, I know I'm supposed to put the proper ESLint comments around any minified libraries and they have to be self-contained within the extension, but something in Ammo just won't let it happen unless it's in the global scope. Any idea how I can fix this? I tried assigning it to a script tag and injecting it to the <head> but that didn't work and assigning based on window.Ammo = didn't work either.

[CubesterYT]

Try swapping var for const

[Brackets-Coder]

still doesn't work, it returns Uncaught TypeError: Cannot set properties of undefined (setting 'Ammo') at line 27 when trying to initialize Ammo() unless the entire library is outside the extension's scope for some reason

[Brackets-Coder]

@CubesterYT in what circumstances would it be acceptable to use any variables outside the TW extension IIFE? Could I use bracketsCoder_Ammo = _insertLibraryHere_? I know It's absolutely not best practices because of extension conflicts, async issues, etc. etc., but I'm having a lot of trouble getting Ammo to register before the extension is fully loaded.

[CST1229]

i assume never, but i don't think putting it ouside the IIFE would affect anything.

[Brackets-Coder]

It's using module.exports, which according to ChatGPT isn't supported in the Turbowarp Extension environment, which makes sense because A-frame #1732 and Cannon.js seem to work fine and they don't use module.exports. (so cannon.js works, but it's just outdated and discontinued and technically not even fully featured and doesn't have full support for transformation conversion).

I tried changing the library to re-assign some of those things but I think it's everywhere and the asynchronous nature of the loading mechanism creates a bunch of issues when putting it in an IIFE... I've also tried creating custom promises and things but it doesn't seem to work because Ammo() (the initialization function that returns the proper object with all of the library's properties and methods) doesn't work until it has fully loaded and returns an undefined error.

Anyone have any suggestions?

[CST1229]

Also, it probably fails because you might have put your library code (e.g the definition for Ammo()) after code that uses it (the Ammo() call)?

[Brackets-Coder]

It's using module.exports, which according to ChatGPT isn't supported in the Turbowarp Extension environment, which makes sense because A-frame #1732 and Cannon.js seem to work fine and they don't use module.exports. (so cannon.js works, but it's just outdated and discontinued and technically not even fully featured and doesn't have full support for transformation conversion).

Before the Zip extension was changed to use scratch-vm's JSZip instance, I simply assiged const module = {exports: null} and that seemed to work.

Noted, I'll keep that in mind and try it when I can

Also, it probably fails because you might have put your library code (e.g the definition for Ammo()) after code that uses it (the Ammo() call)?

Definitely not, see images above. I always try to check execution order to make sure it's not a dumb oversight on my part

[Brackets-Coder]

It's using module.exports, which according to ChatGPT isn't supported in the Turbowarp Extension environment, which makes sense because A-frame #1732 and Cannon.js seem to work fine and they don't use module.exports. (so cannon.js works, but it's just outdated and discontinued and technically not even fully featured and doesn't have full support for transformation conversion).

Before the Zip extension was changed to use scratch-vm's JSZip instance, I simply assiged const module = {exports: null} and that seemed to work.

May I ask how you used const module = {exports: null}?

After the library? At the end of the library's code? Before the library? etc, etc.

[CST1229]

After the library? At the end of the library's code? Before the library? etc, etc.

before the library, so that when the library tries setting module.exports, it will set it on my variable instead (which I can then access)

[Brackets-Coder]

After the library? At the end of the library's code? Before the library? etc, etc.

before the library, so that when the library tries setting module.exports, it will set it on my variable instead (which I can then access)

I saw that in the closed PR. However, I tried it and logging Ammo still returns the library function before initialization and attempting to execute the initialization function returns an undefined error, so it's likely due to async issues. I've tried a bunch of things and I didn't really want to do a set timeout as that's too impractical and there's not exactly a promise that can be used so it's difficult. The main problem is the whole IIFE thing mixed with the fact that I'm assuming Ammo is immediately available, which it isn't.

[Brackets-Coder]

I suppose I should just start developing most of the functionality with Ammo.js outside the IIFE until I can figure out the asynchronous issues

[Brackets-Coder]

@CubesterYT @CST1229 I've figured out the issue. So it turns out that Ammo.JS will return an undefined function error every time I try to initialize it when the minified library was pasted inside the extension IIFE. Even if I create a initialize block and execute it several minutes after the page has loaded (which should be unnecessary considering my computer is quite fast), it still returns the error, but it doesn't when the library is declared outside the IIFE, which of course is not best practices at all. At this point it's a scoping issue, not a async issue.

How do you recommend I continue? I could just rename the library's variable to something like masterMath_bracketsCoder_AmmoPhysics_Ammo just to make sure nothing in the global scope can use it and cause conflicts, but if I do that it probably won't get merged. If I do I could then re-assign it to a more usable name inside the IIFE but I can't really think of anything else other than modifying the library itself, which would be a pain in the neck.

[CST1229]

one possible way would be to wrap it around a block, like

{
// ammojs code
(function(Scratch) {...})(Scratch);
}

or alternatively find out the exact cause of the issue (maybe the strict mode caused by "use strict";)

[Brackets-Coder]

one possible way would be to wrap it around a block, like

{
// ammojs code
(function(Scratch) {...})(Scratch);
}

or alternatively find out the exact cause of the issue (maybe the strict mode caused by "use strict";)

@CST1229 Removing "use strict"; makes it work inside the IIFE. I know unsandboxed extensions are supposed to have that included, but would it be acceptable/get merged without it assuming it passes lint? Or does lint only pass if "use strict"; is included?

[CST1229]

not sure

[Brackets-Coder]

I figured it out and I might have another discussion or draft PR open in a few months

[Brackets-Coder]

@CST1229 and @CubesterYT:

I wanted the opinion of some moderators because this is a grey area not explicitly described in the contributing guidelines. This works:

It's a bit unconventional (especially the ESLint handling) but is it acceptable enough to be merged or do I have to do it differently?

Also note that the extension class and registration of the extension happens inside the Ammo().then function because it will only work if the promise resolves and I need to be able to access the Ammo object inside the extension's block functions.

[EverythingMaker]

nice

[Brackets-Coder]

#1974