📰 Custom Issue

@bsherratt has provided an interesting case. A Name III .txt file describing:

• 200 x 200 grid
• ~200 phenomena

The file takes ~20s to load. If Iris is used to convert the file to NetCDF, that NetCDF file takes ~105s to load. Both loaders are unreasonably slow in this example, since the grid is shared between all phenomena, yet the Coord instances are created on a Cube-by-Cube basis (see #3172 for discussion on sharing). But why is NetCDF even slower?

File available on request.