I do enjoy using iris, but I must also say that sometimes it tests my patience.
I typically work on projection data from different modelling institutes, which all aim at following the CMIP and CORDEX data specs to the letter. But now and then they miss out on some detail that slip through the CMOR and other quality control tools. But these mistakes invariably show up as problems in iris when merging cube lists and similar operations. But alright, this is something one have to live with, and as long as the mistakes are in parts covered by the CF Conventions or the CMIP/CORDEX data specs it could actually be a good thing.
But I find it more difficult to like is that iris also refuses to merge cubes where non-CF attributes do not agree perfectly. Here is an example from a recent attempt to merge three cubes:

time:units                    days since 1949-12-01 00:0...  days since 1949-12-01T00:0...  days since 1949-12-01 00:0...
latitude:long_name                               'latitude'          'latitude coordinate'                     'latitude'
longitude:long_name                             'longitude'         'longitude coordinate'                    'longitude'
grid_longitude:long_name      'longitude in rotated pole...  'longitude in rotated pole...  'longitude in rotated-pole...

gives the following error:

MergeError: failed to merge into a single cube.
  Coordinates in cube.aux_coords (scalar) differ: time.
  Coordinates in cube.dim_coords differ: grid_latitude, grid_longitude.
  Coordinates in cube.aux_coords (non-scalar) differ: latitude, longitude.

It would be helpful if iris could issue a warning (specifyng what it finds different) rather than just throw an error (that is only moderately helpful) when differences are found in parts not covered by the CF Conventions.