auto_ticks raises ValueError for some inputs

[mdickinson]

We're running into the following problem with auto_ticks on a downstream project, leading to a blank plot window.

Python 3.6.13 |Enthought, Inc. (x86_64)| (default, Apr 13 2021, 19:22:11) 
[GCC 4.2.1 Compatible Apple LLVM 8.0.0 (clang-800.0.42.1)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from chaco.ticks import auto_ticks
>>> auto_ticks(657022.2328451854, 577935.6054180114, 657022.2328451854, 577935.6054180114, "auto", use_endpoints=False)
/Users/mdickinson/Enthought/ETS/chaco/chaco/ticks.py:392: RuntimeWarning: invalid value encountered in log10
  magnitudes = 10.0 ** floor(log10(candidate_intervals))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/mdickinson/Enthought/ETS/chaco/chaco/ticks.py", line 308, in auto_ticks
    ticks = arange(start, end + (tick_interval / 2.0), tick_interval)
ValueError: arange: cannot compute length

Expected behaviour: a list of tick locations is returned (possibly the empty list).

[mdickinson]

Note that we have data_low > data_high for the inputs to auto_ticks above. While this may be indicative of an error in the calling code, I think it would make sense to also make auto_ticks robust for this kind of input.

[corranwebster]

Not a fix, but a work-around: if you can create a replacement without the error, it is straightforward to create a new tick generator around the replacement and use that in the downstream app.

[corranwebster]

Also, separately, this feels like it is a symptom, not the root cause: the axis bounds being inverted smells like there a problem with the values being passed to the ticking system. In particular, are we in a situation where the width (or height, depending on orientation) of the axis component is negative? If so, then all sorts of things may be messed up beyond ticking.

[aaronayres35]

FWIW -
This same traceback was seen a while back:
#529
and that issue was closed by #636
but 636 was just checking for NaN inputs, not non-NaN inputs that could lead to NaNs inside the method

In this case, tick_interval on this line ends up as NaN

chaco/chaco/ticks.py

Line 272 in 0907d1d

tick_interval = auto_interval(lower, upper)

and that is because log10 is getting negative inputs here

chaco/chaco/ticks.py

Line 392 in 0907d1d

magnitudes = 10.0 ** floor(log10(candidate_intervals))

[corranwebster]

Tentatively closing as this should be resolved (at least in the sense of treating the symptoms) by #848