While technically correct, I believe we should move the calculation of data lims down into the artists.

Current logic

Extract all data coordinatesxyz from the artist and feed them intoauto_scale_xyz().

This updates thexy_dataLim,zz_dataLim BBoxes via.Bbox.update_from_data_xy/x/y(), which in turn creates aPath and calls.Bbox.update_from_path().

This is quite a lot data pushing (and maybe copying) just to update dataLim bboxes.

Proposed logic

Separation of concerns: The Artists should have a function to get their data lims. These limits should be used to update the Bbox. Basically someBbox.update_from_box, a kind of an in-placeBbox.union, which we don't seem to have.

This whole block should read something like

if autolim:    self.auto_scale_lim(col.get_data_lims())

Note: "data lim" is intentionally vague as we currently don't have a good structure to describe it (we use two 2D Bboxes,xy_dataLim andzz_dataLim.

The minimal approach - if we don't want to redesign for proper 3d boxes, would be to letget_data_lims() return two BBoxes as well. In that case, I would keep this interface private_get_data_lims() so that we can still change to a single 3D instance later.

Edit: This is basically a generalization ofCollection.get_datalim.

Rough outline: We may start minimal with rolling a small

class _Bbox3d:    def __init__(self, points):        ((self.xmin, self.xmax),         (self.ymin, self.ymax),         (self.zmin, self.zmax)) = points    def to_bbox_xy(self):        return Bbox(((self.xmin, self.xmax), (self.ymin, self.ymax)))    def to_bbox_zz(self):        # first component contains z, second is unused        return Bbox(((self.zmin, self.zmax), (0, 0)))

and implement_get_datalim3d() -> _Bbox3d on the 3d collections.

Then

def add_collection(col, autolim=True):    ...    if autolim:        self.auto_scale_lim(col.get_datalim3d())

and

def auto_scale_lim(bbox3d):     ...     self.dataLim_xy.update_from_box(bbox3d.to_bbox_xy())     self.dataLim_zz.update_from_box(bbox3d.to_bbox_zz())

I like this structure, but could we make that refactor its own issue?