Rework blending method in Variant animation for Int/Array/String

[TokageItLab]

Follow up #80813

Supersedes / Closes #66771
Closes godotengine/godot-proposals#5501

• Fixes AnimationTree animation string values are superseded by RESET animation values when using Continuous update mode #69591
• Fixes Animation::blend_variant() can't handle Array type #73342
• Fixes Continuous animation of Polygon2D points no longer works. (Animation Player) #83841
• Fixes AnimationPlayer negative values of tracks are rounded inconsistently for both update modes #84106
• Fixes Update issues with AnimationPlayer's string-tracks #84612

I wanted to include godotengine/godot-proposals#8085, but it would be a rather complicated implementation, so I will postpone that.

This PR allows for interpolating arrays. In most cases, this is intended for use with polygon animation, so if the arrays are of different lengths, interpolation is performed with the last element of the shorter array.

array_blending.mp4

Also, it allows different strings to be interpolated by treating the characters as numbers. This should provide a behavior similar to interpolating text keyframes in AfterEffects.

For about interpolating discrete Variables (e.g. Int, Rect2i, PackedInt64Array and etc.)

During blending, these Variants must continuous as float.

cast_to_blendwise() / cast_from_blendwise(), ensure that casts are performed before and after the blend calculation.

Interpolations that do not blend, such as Tween and Animation::seek(), are still temporarily transformed, but are refactored to use the above functions.

For about interpolating Strings which have partially same text with different length

interpolate_variant()

string_anim.mp4

interpolate_variant() (used by Tween and Animation::seek()) can only change the length between strings with partially same text. This is because the order is fixed and the priority string is determined.

---

blend_variant()

srting_blend.mp4

blend_variant() interpolates characters (same as with arrays, it will be interpolated with the last element of the shorter string) at the same time as length. This is because the order is not fixed and it is difficult to apply a priority order.

If the implementation of godotengine/godot-proposals#8085 is successful in the future, the priority order may be able to determine, but for now, I submit the current version as an experimental algorithm; when string blending occur, it show warning once.

---

BTW, there is a bug with reset on save for arrays, but it exists before this PR, so I will look into it separately.

Demo project (array_blending_r.zip)

[TokageItLab]

Working on improving the String Tween (add/sub_varaint for Tween delta with String is currently problematic).

[TokageItLab]

Alright, I fixed tween. I believe now it is ready for review.

text_bool_tween.mp4

array_blending_r.zip

[KoBeWi]

I did some tests with Tweens and it works correctly. I did not test animation blending nor any of the linked issues.

Animating emoji is interesting 🤔

godot.windows.editor.dev.x86_64_S9Sh2O1OxP.mp4

[YuriSizov]

I read through changes and it all looks sensible to me. But I can't say if this is a correct or a complete solution to the problem.

[TokageItLab]

At last, I just added exception handling for PackedByteArray to eliminated the possibility of interpolation as PACKED_INT_ARRAY in interpolate_variant(), although I think that's a very rare case where the check would be needed.

[akien-mga]

Tested the MRPs from the linked issues.

• AnimationTree animation string values are superseded by RESET animation values when using Continuous update mode #69591: Confirmed fixed.
• Animation::blend_variant() can't handle Array type #73342: With the test project from Animation::blend_variant() can't handle Array type #73342 (comment), it doesn't seem to work for me. I don't see any change when tweaking the blend value between the 3 animations/polygons. On the other hand I do see polygon blending working on the test project from this PR, so maybe it's a me issue or specific to that MRP.
• AnimationPlayer negative values of tracks are rounded inconsistently for both update modes #84106: Confirmed fixed.
  • When changing the z_index blending to Continuous on the ArmRPlayer walk_left animation, I get this warning: WARNING: Value Track: '.:z_index' has different update modes between some animations may be blended. Blending prioritizes UpdateMode.UPDATE_CONTINUOUS, so the process treat UpdateMode.UPDATE_DISCRETE as UpdateMode.UPDATE_CONTINUOUS with InterpolationType.INTERPOLATION_NEAREST.
    It's not clear to me what this means exactly, there's no AnimationTree/blending in that test, just one property using Continuous while other unrelated properties use Discrete. Is that really a situation that needs a warning?
    The warning is a bit ambiguous, it's not clear what the user needs to do.
• Update issues with AnimationPlayer's string-tracks #84612: Confirmed fixed. The fact that Discrete String animation gives different results when going forward and backward may be a bug, but that's the case in 4.1.3 too so not a regression.

simplescreenrecorder-2023-11-16_12.19.07.mp4

I'm not sure how useful the String blending shown in this PR's demo properly really is in practice, but it warns users that it's experimental so I think we can try it out and see if users are satisfied with it.

I don't know if this really solves everything properly but it seems to be an improvement and fixes a number of bug reports.

[TokageItLab]

#73342: With the test project from #73342 (comment), it doesn't seem to work for me. I don't see any change when tweaking the blend value between the 3 animations/polygons. On the other hand I do see polygon blending working on the test project from this PR, so maybe it's a me issue or specific to that MRP.

This is because that MRP is using UpdateMode::Discrete. If you set UpdateMode::Continuous, the blend will work. Also If you loop the animation or connect a node with a playback function such as NodeOneShot/Transition, the key update of UpdateMode::Discrete will also work.

When changing the z_index blending to Continuous on the ArmRPlayer walk_left animation, I get this warning: WARNING: Value Track: '.:z_index' has different update modes between some animations may be blended. Blending prioritizes UpdateMode.UPDATE_CONTINUOUS, so the process treat UpdateMode.UPDATE_DISCRETE as UpdateMode.UPDATE_CONTINUOUS with InterpolationType.INTERPOLATION_NEAREST.
It's not clear to me what this means exactly, there's no AnimationTree/blending in that test, just one property using Continuous while other unrelated properties use Discrete. Is that really a situation that needs a warning?
The warning is a bit ambiguous, it's not clear what the user needs to do.

This was implemented at the same time as the AnimationMixer implementation. When everything is in DiscreteMode, the AnimationTree only updates the value when it crosses a key. But that MRP has Continuous so AnimationMixer enforce Continuous for that track, meaning that the value is updated even though it does not cross the key.

Also, AnimationMixer does not know when a blend will occur, so it iterates through the entire animation list once and generates a track cache. Changes to blend types are made at the time of cache generation, and warnings are output at the same time, which is why warnings are generated even if the AnimationTree or AnimationPlayer does not have them assigned if that tracks exist in any animation of animation list.

[akien-mga]

Thanks for the quick turnaround fixing these issues!