0.8 breaks positioning at negative locations (on Windows?)

[oscargus]

0.8 breaks positioning at negative locations (on Windows?)

[oscargus]

Turned out that passing save_steps affected the results, so I started to look into why. It seems like plt.draw wasn't enough in my case for some reason (I use adjustText in a library), so I swapped the order and applied the draw on the canvas instead. (Seems to work for me even when not supplying an Axes, but easy to do.)

[Phlya]

Can you try the version from here? #139

Otherwise I have no clue actually, don't have any Windows machines... Sorry.

[Phlya]

(Note that available arguments are a little bit different, but overall the algorithm is very similar, and way faster)

[oscargus]

I'll try it out later, thanks!

Can just update that the same thing happens on my CentOS 7 office machine... Quite confusing indeed. Will try to debug it a bit more careful as well. I mean, there should be a specific call in adjustText(?) that all of a sudden sets it to zero. The only thing I can think of is the unit handling in Matplotlib (as in that is the only thing from the changelog that seems remotely related).

[Phlya]

Yeah but you don't really have any units here, it's just a regular plot... Really strange. It could be some platform-dependent bug in mpl, if it's related to that.

Are you using the same backends on all machines?