Dumping the grid versus making dumping optional

[egull]

Andrey: I would like to strongly advocate to dump points in the grids. The pros :

1. It will reduce the dependency on domain specific parameters. Real world example : TRIQS-1.0 had 'half-bin' grids in and TRIQS-1.2 does not. The old data therefore is not protected to be loaded/saved automatically. So this will guarantee the compatibility of future versions of the format with the very first one (5 years down the road we won't need to look at how the specific parameter was defined if our goal is just to compare with the data).
2. It is more human-readable and guarantees hardcoded mapping between physical points and data in the file and not in the code (without converters from codes that may suffer from human error). The data is directly plottable - important for benchmarks.

Therefore the grid might look like::

     grid
          \---(points)
          \---kind 
          \---(label)
          \---<domain-specific property1> : statistics, beta, ...
          \---<domain-specific property2> : statistics, beta, ...

For defined kind - the domain specific attributes should primarily be used for reading and writing. Main cons : redundancy for some grids, for example the Matsubara one.