BUG: annotating a sphere then setting font size in a plot results in a crash about yt HOT 3 CLOSED

I did some searching and found one more callback with a similar issue: calling annotate_arrow() then another function that runs the callback changes the arrow size.

from yt.

Looks like this is because the callback overwrites the radius attribute with a scaled version, which changes the units from length to either dimensionless value or 1/length. So when it runs back through the callback (cause calling set_font_size invalidates the plot), the radius attribute will be the already-scaled value with units that can't convert to dimensions of length.

https://github.com/yt-project/yt/blob/14c49c4fa832fb7f04485a16134d9adc2916508d/yt/visualization/plot_modifications.py#L2033C1-L2050C59

        if is_sequence(self.radius):
            self.radius = plot.data.ds.quan(self.radius[0], self.radius[1])
            self.radius = np.float64(self.radius.in_units(plot.xlim[0].units))
        if isinstance(self.radius, YTQuantity):
            if isinstance(self.center, YTArray):
                units = self.center.units
            else:
                units = "code_length"
            self.radius = self.radius.to(units)


        # This assures the radius has the appropriate size in
        # the different coordinate systems, since one cannot simply
        # apply a different transform for a length in the same way
        # you can for a coordinate.
        if self.coord_system == "data" or self.coord_system == "plot":
            self.radius = self.radius * self._pixel_scale(plot)[0]
        else:
            self.radius /= (plot.xlim[1] - plot.xlim[0]).v

simplest fix would be to not overwrite self.radius. but could also add a self._scaled_radius attribute to avoid re-calculating if it's already been through once? I'll keep playing around and push something up shortly.

from yt.

Thanks @yut23 ! I was curious if there'd be any others that failed when invalidated and run again, looks like the arrow callback fails for the same reason (storing calculations in class attributes). I'll add the fix to my current PR

from yt.