Single Merge for load_cube

[rcomer]

See #3334. Just removed the first merge to see if anything breaks!

[bjlittle]

@rcomer Thanks for this.

Yup, so it appears there is a double merge being performed here within iris.load_cube 😱

Once by the underlying iris.cube._CubeFilter.merged for the cubes associated with the matching constraint. Then once afterwards within iris.load_cube itself, when it calls merge_cube on the iris.cube.CubeList returned from the filter.

This current functionality all works just fine, but is really a tad inefficient. Plus, it results in a not so informative traceback from merge_cube due to the prior (and unnecessary) merge.

With regards to the example given in #3334, we have a traceback that goes from:

iris.exceptions.ConstraintMismatchError: failed to merge into a single cube.
  cube.shape differs: (3, 4) != (2, 4)

to:

ConstraintMismatchError: failed to merge into a single cube.
  Coordinates in cube.aux_coords (non-scalar) differ: baz.

For me, the second traceback wins the day. From a users perspective, they have a fighting chance to understand what the heck is wrong.

LGTM 👍

[bjlittle]

@rcomer I'm more than happy to merge this, but first I need to fix the master travis-ci tests, which are failing at the moment 😢

So, let me do that first, then we can merge this with confidence. Does that sound like a reasonable plan?

[rcomer]

@bjlittle sounds good to me 👍

[bjlittle]

@rcomer Fancy doing a re-base?

Just to make sure all is well before merging...

Reference #3349

[rcomer]

Rebased 🤞

[bjlittle]

It's been a long day, clumsily clicked the thumbs-down emoji by mistake 😊

[bjlittle]

@rcomer Awesome, thanks 👍