Re: requirements for anonymous dimensions (and question for John)

Ed Hartnett wrote:


How would you like the ids of private dimensions to work?

For example, if I define two public dimensions (0, 1), then a private
dimension, does it get id 2?

What problem is this intended to solve? What I mean is, how is this
feature intended to be used in the real universe?


ideally, only define a private dimension in the context of its variable.
the order of the dimensions in the variable is obviously very important, but from my POV the ids are not, so anything that works for you is ok with me. might be interesting to allow promotion of private dimension to shared when modifying a file.

problems to solve:

1. shared dimesions are sometimes not useful, and just clutter things up, so i think its useful to allow private dimensions. string length dimesnions for example. once you allow Strings, though this argument is less valid, and i could see an argument that advocates forcing the use of shared dimensions in netcdf-4.

2. hdf5 and dods only have private dimensions, so i guess they find them useful. OTOH, shared dimesnions are essential for the coordinate system extentions we have been talking about.

3. both hdf5 and dods have agreed to allow shared dimesnions, but will still have private dimesnions, so we need them to retrofit existing hdf5 and dods datsets to new netcdf data model.

it might help me to understand how hdf5 proposes to add shared dimensions, and the status of coordinates aka dimension scales?