DisconnectSpace and TimeSeriesScalar

[rangsjo]

Describe the bug

rerun_example_disconnect_space does not work if TimeSeriesScalar is logged in python.

To Reproduce
Steps to reproduce the behavior:

1. Run:

        # Slightly modified example code for disconnect space
        rr.init("rerun_example_disconnect_space", spawn=True)
   
        # These two points can be projected into the same space..
        rr.log("world/room1/point", rr.TimeSeriesScalar(0))
        rr.log("world/room2/point", rr.TimeSeriesScalar(1))
   
        # ..but this one lives in a completely separate space!
        rr.log("world/wormhole", rr.DisconnectedSpace())
        rr.log("world/wormhole/point", rr.TimeSeriesScalar(3))
   

2. Look in the viewer for disconnected spaces.

Expected behavior

Two different spaces should be shown, but as stated, all signals are shown in one single space.

Screenshots

Desktop:

• OS: ubuntu 22.04

Rerun version

rerun_py 0.10.1 [rustc 1.72.1 (d5c2e9c34 2023-09-13), LLVM 16.0.5] x86_64-unknown-linux-gnu release-0.10.1 e596e4e, built 2023-11-02T16:05:55Z

[Wumpf]

Thanks for filing the issue! This is indeed a bug the way DisconnectedSpace is defined. That said, itself is a workaround for not being able to specify what goes into which space view from code which is luckily about to change! (we've been saying that for a while now, but we're finally almost there 😅 )
We're also meanwhile want to make the way space view heuristic spawning works a bit more scalable and predictable. As part of that we concluded that DisconnectedSpace will likely be renamed to UndefinedTransform and only be applicable to 2D & 3D space views. I.e. it is regarded as a special spatial transform.

.. but as things are this is a bug as described.

[Wumpf]

Related to:

• Re-engineer the space-view heuristics engine #4388

[emilk]

The plan is that setting up spaceviews should be done via the new blueprint API when this issue is done:

• V2 Blueprint APIs for Python #4167