empty scatter messes up the limits

[akhmerov]

Tested with 1.1.1 (didn't manage 1.2.0 to show anything yet).

To reproduce:

from matplotlib import pyplot as plt
plt.scatter([0,1],[0,1])
plt.scatter([], []) # The lower limit shifts appreciably, compared to not executing this line.
plt.show()

[WeatherGod]

Confirmed. Furthermore, it should be noted that each subsequent call to empty scatter will grow the limits even further. I suspect it is some sort of issue with autoscale_view() when one has an empty collection. A temporary fix would be to only call autoscale_view() if the length of the collection is greater than zero, but I would rather fix the issue with autoscale_view().

[akhmerov]

@WeatherGod I found the problem, and the issue is not with autoscale_view(). What happens instead is that an empty collection with a set path has non-empty datalimits, so when it is added with add_collection, the datalimits aren't updated correctly (that is they do actually become modified and don't stay the same). The collection datalimits are obtained through path.get_path_collection_extents. This function, in turn calls _path.get_path_collection_extents. That function, when provided with offsets=[] treats the offsets as if they were [0, 0] (this behavior is used in path.get_paths_extents).

The proper fix, as far as I can tell, would be to correct _path.get_path_collection_extents, such that it returns None's if offsets=[], and to call it with offsets=[0, 0] from path.get_paths_extents. I am not sure whether None's would be a legitimate value of datalimits, but the behavior that I described would at least be logical.

Another fix, which doesn't prevent this issue from popping up somewhere else would be to fix axes.add_collection adding an extra check for offsets. I will submit a pull request with this fix shortly, and if it's accepted I recommend to open an extra issue for path.get_path_collection_extents.

[efiring]

Closed by #1497