It looks like the test failures are pep8/pycodestlye releated.

We use and automated tool to check style consistency across the library. It seems silly, but for a library this big with this many contributors the consistent code style really does help.

You have two lines that are too long.

Could you also add a test, probably in test_axes. Something like

@cleanup
def test_twinx_veiw_interval():
    fig, ax = plt.subplots()
    ax2 = ax.twinx()
    # do the thing that exposed the bug
    # assert that it is behaving correctly now

to make sure that this bug does not creep back in.

I feel that self.axis.set_view_interval should also under the control of Axes._autoscaleXon like autoscale_view, then once we called xlim then the axes limit will be fixed, though later xticks may excess the range. Does this make sense?

A stupid question, which place is better for the discussion, in PR or in original issue?

Here is probably better (as it is attached to the code).

I think it comes down to how we should interpret user intent. If the user calls set_xticks then the user is saying "I want my ticks to be exactly at these places" (under the hood it switches to using a FixedLocator) and so the user probably means that those values are interesting and should be include so the limits should be extended to include them. That intent should take priority over the earlier instruction to 'show this region'.

Expanding the view internal on set_xticks has been like that for a long time and many people maybe relying on the view-expanding behavior. Possibly breaking user code (with the exception of breaking work-arounds to bugs) makes changes have a much higher bar to be accepted.

Reasonable, I agree with you.

It looks like there are a couple of other places where viewLim is being directly set:

22:21 $ git grep -n '\.viewLim.interval.*='
lib/matplotlib/axes/_base.py:2876:        self.viewLim.intervalx = (left, right)
lib/matplotlib/axes/_base.py:3150:        self.viewLim.intervaly = (bottom, top)
lib/matplotlib/axis.py:2005:                self.axes.viewLim.intervalx = xmin, xmax
lib/matplotlib/axis.py:2345:                self.axes.viewLim.intervaly = ymin, ymax
lib/matplotlib/projections/polar.py:587:        self.viewLim.intervalx = (0.0, np.pi * 2.0)
lib/matplotlib/ticker.py:226:        self.viewLim.intervalx = vmin, vmax

While you are working on this could you clean those up too?

For ticker.py, it's in class _DummyAxis(object). This class is not a subclass of Axis and seems to be a placeholder or something similar for internal use in ticker.Formatter. I didn't see the relation with real axis of axes but not very sure.

The rest I'll check them later. It occurs more than I thought. And we shall take care of every '\.viewLim.*=' not only '\.viewLim.interval.*='...

Sorry, still have no time for this PR. I'm busy working on my thesis.
Many thanks, if anyone feels interested in this problem and wants to take over.

Closing as not active. Feel free to re-open if someone feels the need...