Circle draw tool must produce circle

[pllim]

User was confused on why they get back elliptical region when they used the circular draw tool on their image. You can basically reproduce this behavior using spacetelescope/jdaviz#697 even though I think @camipacifici reported a different workflow.

With this patch, even though the circle tool obviously produced something that does not look circular in Lab, it still gives a circle.

I don't see the affected code tested anywhere in this repo, so I am unable to add new tests.

🐱

[pllim]

I don't think the pydantic.errors.PydanticImportError: BaseSettings has been moved to the pydantic-settings package. error in CircleCI is related.

[pllim]

p.s. I tried having both tools subclass from a common base class but that broke the click-and-drag feature for some reason.

[astrofrog]

Thanks! This approach doesn't quite work because if I make a simple scatter plot in glue-jupyter where the two axes might be quantities with different scales (e.g. x is velocity between -1 and 1 and y is position between -100 and 100) the current 'circle' tool will produce weird results - e.g.

In this scenario users would expect the circular tool to draw a circle.

I think a cleaner solution would be to keep the current circular tool as is but add a class attribute e.g. _strict_circle that defaults to False and check that attribute where we currently construct a CircularROI. You can then make a 'strict circle' subclass that simply sets _strict_circle = True, and use bqplot:strictcircle in Imviz.

[dhomeier]

Suggested change

	r = (rx + ry) * 0.5
	r = np.sqrt((rx**2 + ry**2) * 0.5)

adds a little computational overhead, but I think using arithmetic mean here could create a circle passing somewhat besides the actual pointer position

[pllim]

@camipacific's workflow on standalone Jdaviz (read: she ran this stuff on voila)

• create circular subset on object of interest
• create annulus subset around that circular subset for background
• go in the subset tool plugin to properly center the circular subset on the object of interest after recentering, notice that the circular subset is actually an ellipse (i could see it in the image when compared to the annulus subset and i noticed that the subset had two radii)

at this point i was confused and thought i did something wrong so I deleted the subsets and started again. by redoing it, i noticed that the circular subset turned automatically in an ellipse, while the annulus stayed a circular annulus. this can be confusing for a user.

[pllim]

As per offline discussions, I will wait for the alternative solution that @astrofrog is going to propose.

[pllim]

This has been superseded by #376