Edge ticks sometimes disappear

[EdwinChan]

I am using

• matplotlib 1.5.1 installed with pip for Python 3.5.1+ on Ubuntu; and
• matplotlib 2.0.0b1+1708.g5de9c44 built from source for Python 3.5.1 on Scientific Linux 5.8.

Consider the example

import matplotlib
from matplotlib import pyplot

pyplot.xlim(0, 1.2)
pyplot.ylim(0, 1.2)
pyplot.annotate(matplotlib.__version__, (0, 0), (0.1, 1.1), 'data', 'data',
  ha='left', va='top')
pyplot.savefig('lim.png')

With the GitHub version of matplotlib, the last tick (1.2) is missing from both axes. This is a regression because matplotlib 1.5.1 can show both ticks without problem.

[anntzer]

Bisected back to #5956. I wonder if the check for interval_contains is really necessary? Can't we just rely on the ticker to only return ticks that are within the interval?

[efiring]

It looks like that check dates back to 2005--tickers have changed a lot since then, so yes, it seems reasonable to leave the responsibility for tick selection entirely to the tickers. It would be a good idea to check all of them to see whether they are handling floating point slop in a reasonable and consistent way.

[anntzer]

The log-ticker seems to use a strict inequality check (e.g. there's no tick at 10 for xlim(1, nextafter(10, 0))) but that's not really a problem (or at most a separate one), right?
Also I am now not even certain that the whole dance with Base to handle floating point errors is necessary?

[efiring]

The Base machinery was put in explicitly for that purpose, because we were having problems, I believe; is there another mechanism that replaces it? (I didn't write Base, and haven't looked at it carefully.)

[efiring]

#5956 has an obvious effect only on tick labels, not on whether a tick is drawn, so I don't see how it could be the cause of the change in behavior.

[efiring]

Closed by #7042.