Added tolerances to Coord.is_contiguous()

[esc24]

The iris.coords.Coord.is_contiguous() method tests for equality between adjacent upper and lower bounds to determine whether bounds are contiguous. This runs into issues with floating point numbers. For example:

import numpy as np
import iris

delta = np.float64(0.00001)
lower = -1.0 + delta
upper = 3.0 - delta
points, step = np.linspace(lower, upper, 2,
                           endpoint=False, retstep=True)
points += step * 0.5 
coord = iris.coords.DimCoord(points)
coord.guess_bounds()
coord.is_contiguous()

returns False.

This PR modifies is_contiguous() to use np.allclose() adding the tolerance keyword args to is_contiguous().

[bblay]

Do we need to reproduce numpy's defaults here?

[esc24]

I cannot come up with a neat way not to. Feel free to suggest one, but I'm happy to have clearly stated, documented defaults and simple code.

[esc24]

@bblay - do you have any outstanding objections to this being merged?

[bblay]

Apologies @esc24 but I could not finish the review.
My comment about defaults, above, was just wondering if we could just pass through None instead,
and refer to the np func for defaults. This would allow np to change them and we'd stay up-to-date.

[pelson]

👍 - I'd be happy to merge this.

[bblay]

Might be worth adding a comment in here as this is an important test, perhaps something like,
bounds[0,1] differs from bounds[1,0] around the 16th decimal place.
Alternatively, it might make the test clearer if you were to simply hard-code the values?

[bblay]

👍 There are a few trivial comments you might choose to act on, but I'd be happy for this to be merged now.
I'd be happier still with a response to the 'defaults' question:
"can't we just pass None through to np, so it can decide (and change) appropriate defaults?".
Perhaps we can't, I'm sure I've seen functions that can't be used like that.
Perhaps we don't want to rely on np incase it breaks our tests in the future.

[pelson]

but I'd be happy for this to be merged now.

I'm guessing you missed the fact that it had already been merged 😄. Its up to you if you'd like Ed to follow up on those comments in a follow-up PR.

I'd be happier still with a response to the 'defaults' question

No, you can't do this. None is a legitimate argument and almost all functions (unless None means "use a default value" and is typically only done for mutable default values).

[bblay]

Doh! 😊 They removed the red box saying merged.
Those comments are not important enough to require more work on this.
Thanks Phil