Support stand-alone Path objects in Canvas

[corranwebster]

This is experimental right now - backend only, and checking that I haven't broken anything.

Some tests should fail with coverage errors.

PR Checklist:

• All new features have been tested
• All new features have been documented
• I have read the CONTRIBUTING.md file
• I will abide by the code of conduct

[corranwebster]

Hm, that's a good question. I don't think either implementation is wrong, seeing as how this is outside the HTML-matching API and into our (potentially-useful-but-ever-confusing) nonlinear editing API, so ultimately we get to make the call...

My off-the-cuff thought is that in the conceptual framework we've been using, it should be as if we're "going back in time" and altering what the radius was, as if the circle had been created that way from the start — in which case, it should update in both paths. This would imply that path-copying should be shallow.

Thinking about it a bit, I could be convinced either way, but I'm currently leaning towards shallow copy being more "pythonic"; and deepcopy is always there if you need it.

[corranwebster]

I think we're approaching the point where this needs to wait for #4159 since there's overlap between the two at the public API level, and I want this to match the outcome from that.

One question is what to do about things like this once #4159 is merged:

path = Path2D()
path.arc(30, 20, 10)
with canvas.root_state.fill(path=path) as fill:
    fill.rect(10, 20, 30, 40)

Currently this will ignore the fill.rect and draw the path, but having a path in the context-manager version of fill and stroke is awkward and probably shouldn't be allowed: I think if we enter a context manager and path is not None, it should raise an exception.

[HalfWhitt]

I think if we enter a context manager and path is not None, it should raise an exception.

I agree, that makes sense.

[HalfWhitt]

Thinking about the path argument here and the x / y / potential transform parameters, it occurs to me that there are essentially four different kinds of potential overlap between "standalone" and context-manager parameters:

• They work (and do essentially the same thing) in both. This is mostly everything so far.
• They work, but do different things in each (as might be the case with transform). I think we should be really careful about adding any of these.
• They work as a standalone command, but are meaningless in a context manager. If we add any of these (like path), it's easy enough to raise an exception if and when the drawing action is entered.
• They work in the context manager, but are meaningless in the standalone command, like x and y. These are problematic because we don't have any way of detecting if they've been supplied when they won't be used. That's not the end of the world, but could be confusing for users. I think we should be hesitant to add any more of these. I doubt I'd add x/y if they weren't pre-existing; it might even be a good idea to deprecate them, but I'm not yet convinced either way.

[corranwebster]

Looks like a segfault on macOS Intel machines. I can replicate on an old Intel Mac on macOS Sequoia. It looks like it is an issue with CGPathAddArc in Path2d.ellipse when counterclockwise is True. Works fine on Arm machines.