BUG: xlim on plots with shared axes (GH2960, GH3490)

[rosnfeld]

Fixes #2960
Fixes #3490.

Summary of changes:

• for irregular timeseries plots, considers all lines already plotted when calculating xlim
• for plots with secondary_y axis, considers lines on both left and right axes when calculating xlim. There are tests for non-timeseries plots (always worked, but now a test to confirm this), irregular timeseries plots, and regular timeseries plots.

This does not fix #6608, which I now think is a fairly specific/rare case that could always be manually avoided by x_compat=True. I will still try and fix that, but it looks a bit uglier and I didn't want to hold back these more important changes.

[jreback]

@TomAugspurger ?

[rosnfeld]

I see from another PR that @TomAugspurger is travelling... is there anyone else who could take a look at this? Seems like a nice patch to have in 0.14.1.

(admittedly I am travelling also)

[jreback]

@cpcloud can you take a look?

@rosnfeld pls rebase this

[rosnfeld]

Rebased. Thanks for reviving this. Looks like @sinhrks is working on similar material in #7459 and #7515.

[jreback]

cc @sinhrks can you have a look?

[sinhrks]

Should this also try to search on both right and left axes?

[rosnfeld]

Good catch. I had created test_secondary_y_irregular_ts_xlim to test a secondary_y case but as written it only makes sure that a "primary" (non-secondary) axis is taken into consideration when you plot on the secondary axis, but not the other way around. I've strengthened the test and updated the code.

[jreback]

@rosnfeld pls rebase

[rosnfeld]

Rebased.

[jreback]

one more rebase pls (i just merged bunch of plot stuff)

[rosnfeld]

Ok, rebased again.

[jreback]

ok, ping when green

[rosnfeld]

Travis is green on my fork but will take a while on the main pydata one I think. I am in Asia at the moment and unfortunately need to sleep. :)

[jreback]

hahha...np...thanks

[rosnfeld]

Alright, it's green now.

[jreback]

thanks!