Thanks for testing it out! I doubt any of the fixes will get into 2.92 as that's releasing any day now.

I never knew you can Dnd from the outliner. Thanks Looch!

design a new baking system. With some class-based architecture, with a simple (maybe even abstract) base class that's made so that subclasses can implement/override specifics, could simplify things a lot.

Can you give examples of what you have in mind in terms of subclasses and overrides? I'm having trouble understanding what you meant. Adding further abstractions seems like it would add unnecessary layering. I agree that the signaling is confusing and I just tried to split the patch (Smart baking separate from the signaling changes) but then ran into the issue that the new signaling adds support for subframe baking. Without it, smart baking will only bake whole frame keys.

upload isolated diff without the dependencies

• ghost fcurve drawing uses Show Extrapolation setting

• Overlooked remap_domain writing, now writes remap_domain on output even when remapping succeeds. This didn't have any effect for the user as the input blended_snapshot and output upper_snapshot were always the same.

• rebase and minor cleanup

I'll see if I can finish a patch for the simple approach . The upgrades appear non-trivial as I'm not sure how to properly determine the fcurve extents when multiple fmodifiers exists and how they interact with eachother (cycles modifier, limits modifier, restricted ranges, etc).

• Animator-Level Changes

Yeah, that's exactly how it works right now.

• rename:

Auto Frame Range -> Frame Range
None -> Custom
Scene Render Range -> Scene Range
Scene Preview Range -> Preview Range

• missed typing hint

Note to self: Todo/Check/Question:
D10383: Fix: NLA Apply Scale Sets Wrong Action Range (referenced for the change in anim_deps.c)

Added inline notes for source of bugs.

uploaded a diff with the isolated change. It only depends on D10101: Cleanup: Nla: Refactor nla.c functions for the function BKE_nlatrack_add_strip().

Shouldn't it just always work

Seeing the complete result will always work. But keyframe remapping won't always work. If your layers look like

[Full Replace LocXYZ]
[(Tweaking) LocXYZ]

Then keyframe remapping will fail when you try to insert LocXYZ keyframes on the bottom layer. This is an expected case where it fails ofc.

• revert to defaults for hotkeys
• apply review notes (better bool name, removed trailing space, simplified comment)

• rename sampled to blend_domain for clarity

• applied review notes:
• assertions on else() cases for future proofing
• renamed invertible to remap_domain for clarity, thanks
• should now work when blended_snapshot and output r_upper_snapshot are different pointers. Currently they're the same so nothing broke. This means that we now explicitly copy over the disabled remap_domain bits from blended_snapshot to r_upper_snapshot. They're of the form:

language=C
if (!BLI_BITMAP_TEST_BOOL(blended_necs->remap_domain.ptr, j)) {
   BLI_BITMAP_DISABLE(result_necs->remap_domain.ptr, j);
   continue;
}

being able to drag and drop them from the outliner into the NLA as clips would be handy

I don't think anywhere else in Blender does outliner Dnd to another editor so the support probably doesn't already exist, and thus too out of scope.

• ensure transitions evaluate properly, accounting for new sampled property

Abandon reason: Got the OK the commit directly.

Fixed. I'm not 100% on the guidelines on committing without review, so I just submitted a patch for now.

I'll take a look.

rebase and minor cleanup

rebase

proper rebase since dependency patch accepted.