The function iris.analysis.coord_comparison returns groups of coordinates from a set of cubes categorised by various measures. Notably, it is used by the maths operations to determine whether two cubes have matching metadata and are allowed to be combined.

There is a bug in the implementation which means that if exactly the same coordinate object is used by two or more of the source cubes it will be incorrectly categorised as 'ungroupable' (meaning not all source cubes provided an equivalent coordinate). I haven't delved too deeply but my guess is that it's due to this id check