avm2: Implement flash.geom.PerspectiveProjection

[cookie-s]

doc: https://docs.ruffle.rs/en_US/FlashPlatform/reference/actionscript/3/flash/geom/PerspectiveProjection.html

This PR implements flash.geom.PerspectiveProjection features. However, assignment to flash.display.DisplayObject through transform.perspectiveProjection is not yet supported. No effect to visual, either.

[cookie-s]

I don't know where this value 500 comes from. It fits the output from the test in FP.

[cookie-s]

From TestTransform() result, fieldOfView should be saved and focalLength should be computed from it.
PP(FOV:55, FL:480.25) assignment to Sprite.transform.perspectiveProjection produces PP(FOV:55, FL:288.15) (when SWF width is 300).

[emergence75]

this might be useful, take a look
https://github.com/emergence75/ruffle/blob/flash.geom-alternate/core/src/avm2/globals/flash/geom/PerspectiveProjection.as

[adrian17]

I think it all looks fine, but if you don’t mind, I wanna investigate a bit into the remaining unknowns; but I’ll only be able to do it on wednesday.

[adrian17]

To be sure, this is a complete matrix impl? Asking due to the removed stub.

[cookie-s]

Yes, I believe so. For other methods, especially setters are related to unimplemented rendering/visual, but I believe this toMatrix3D just returns the Matrix3D object.

Ah, but only if the small but not very small floating point number differences are allowed. Please see precision in TestToMatrix3D test function. There is unexplainable numerical errors which is now rounded to approximation in the test.

Except for that, I believe all features in toMatrix3D are implemented.

[adrian17]

As for emergence75’s code, note that we did find issues with some of the changes there (see: #18495 (comment) and the subsequent comment). I know that since then it was replaced by a newer PR (which I’m sorry for not commenting on), but as far as I’m aware, the listed concerns about the changes still apply :( With that, I’d be careful with treating it as a source.

[cookie-s]

this might be useful, take a look

Oh thank you. My PR might be better to implement the input validation for fieldOfView setter.

I wanna investigate a bit into the remaining unknowns

Sure thank you!

[adrian17]

If this has the same logic as the one below, any reason for it to be separate?

[cookie-s]

Good point. No! Thank you, merged to the line below at f6f2aba

[adrian17]

Does this actually need the live reference to the DO? What if instead, only the width value was stored?

[cookie-s]

True, storing only width passes the current test, but another observation by for example this code:

        private function TestTransformUpdate(): void {
            var s: Sprite = new Sprite();

            trace("// Set non-null to transform.perspectiveProjection");
            s.transform.perspectiveProjection = new PerspectiveProjection();
            printProps(s.transform.perspectiveProjection);

            trace("// Set null to transform.perspectiveProjection");
            s.transform.perspectiveProjection.fieldOfView = 100;
            printProps(s.transform.perspectiveProjection);
        }

shows some setter in DO.transform.perspectiveProjection : PerspectiveProjection should update the values in DO.

It's the similar behavior to transform.matrix3D rather than transform.matrix in terms of what we discussed before: #18810 (comment)
(although that setter behavior is not implemented to matrix3D and transform.perspectiveProjection setter is almost imcomplete as of now.)

Given that, we will anyway need reference to DO at some point I believe.

[adrian17]

Oh, I see, yeah.

            var a = dobj.transform.perspectiveProjection;
            var b = dobj.transform.perspectiveProjection;
            trace(a === b); // false
            a.fieldOfView = 123;
            trace(b.fieldOfView); // 123

[adrian17]

Can you please add some TODO comments explaining the above, and maybe a known-failure test? But just a comment is enough, I think I'm gonna approve the PR in a moment.

although that setter behavior is not implemented to matrix3D and transform.perspectiveProjection setter is almost imcomplete as of now

Sure, I think this is an okay compromise.

[cookie-s]

Thank you! I added a few TODO/FIXME comments.

[adrian17]

Thanks!

(as for the z-order PR, I'm trying to find time to take a look at it, sorry for a delay 😅 )

[cookie-s]

Thank you for merging!

(No problem at all, I know it's not a tiny PR. I have just pushed my last update before its review.)