Rotation

[ajyoon]

This is a tracking issue for support gaps relating to object rotation. Most of this is likely out of scope for the current release plan, but they are outlined here for tracking purposes.

• Basic individual rotation of Path, Text, and RichText objects
• Rotation is treated in isolation. Objects with rotated parents do not inherit their parents' rotation. This is needed to support things like rotating object groups, like entire staves.
• Explicitly setting rotation axis position. Rotation currently only works with the rotation axis at the object's Qt "logical coordinate system" origin. (done by @Xavman42)
• Text bounding rects are inaccurate for rotated objects. (Though it is unclear if this is desired behavior or not)
• Rotated text and paths which break across flowable lines do not clip correctly. The results aren't terrible - the clip points are generally quite close to correct, but the clipping occurs at an angle. Resolution would probably require modifying QClippingPath to account for rotation. Care would be needed to ensure rotation respects the rotational axis even with clipping painter offsets.
• Rotated paths with path elements anchored to other objects display incorrectly. This occurs because path element resolution (Path._resolve_path_elements) does not respect rotation

[Xavman42]

I would like to start tackling this issue. My hope is to get to the point where I can rotate staves about an axis and their children will also rotate about the same axis. I'm not worried about flowables or clipping issues - I can work around those for now.

Any suggestions on where how I can start?

[ajyoon]

That would be great! I'll need to take some time to sketch out a dev plan for this, but something we definitely need to support is low level rotation-about-an-axis. If you want to go ahead and add a transform_origin: Point field to the various interface classes which have a rotation field already, then integrate those fields into their underlying Qt objects, that would be a great start. You should be able to test this new field similarly to how rotation is tested, for example see TestTextInterface.test_rotation().

It looks like the corresponding Qt field for this is https://doc.qt.io/qt-5/qgraphicsitem.html#transformOriginPoint, which seems to set the scale origin point too. I'm fine with this change, but the docstrings around scaling should be updated to indicate this. (For now doc updates at the interface layer alone are enough, since this functionality isn't exposed to core yet.)

I've also just opened a new rotation development branch; please make your PRs against this branch then we'll merge it all together into main when it's ready.

[ajyoon]

Ok @Xavman42 the next step is to add the transform_origin field to the core layer and similarly plug it into the created interface objects and add unit tests for the plumbing.

After that, in a separate PR, will be the difficult step of figuring out how to actually compound translation, scaling, and rotation transformations. I have a feeling this might involve doing proper affine transforms instead of my bad homemade mapping methods (see PositionedObject.map_to), though maybe there's a way to hack it in the meantime..

[ajyoon]

another idea would be to change the interface layer so interface objects actually have parents, so we can use Qt's transform inheritance for free. off the top of my head I can't think of a reason this wouldn't work, and it's probably much simpler and faster than building our own parallel transform system.

[ajyoon]

yeah the more I think about it the more I like the idea of using the Qt scene graph to do the work for us. I may just do this myself since I think it would be fairly quick once your work is done, but I may ask for help depending on the time I have available.

I think the main complication is inside flowables, where we can't directly use the scene graph. maybe it's not so unreasonable to use the current system for flowable contexts, then use Qt's outside of them and say that rotation and scaling are not well supported inside flowables (which is already the case)

[ajyoon]

Ok transform origins seem to be working now! I'm going to take a crack at building out inherited transforms this week, will update on progress when I can

[ajyoon]

Still a ways to go, but today I made some progress - this is made with the following simple script:

prev = Text((ZERO, Mm(150)), None, "root")

for i in range(1, 11):
    prev = Text((Mm(20), ZERO), prev, f"desc_{i}", scale=1.1, rotation=5)

(ignore the dark background - it's a bug caused by this stuff)

[ajyoon]

I think I got it! Just need to clean up and update docs for this. Hoping to cut a release today or tomorrow.

[Xavman42]

Wow! That looks perfect - thank you so much!

[ajyoon]

Support for inherited rotation and scaling has now been release in 0.1.7. See the release announcement here for more information. It required a few breaking changes so I recommend reading over it and the changelog.