add option to change endianess on variables about netcdf

[magnusuMET]

Implemented in #35

[lwandrebeck]

Would something like this do (partially) the trick ?
lwandrebeck@4ce77a9

[magnusuMET]

Could you create an enum for endianess, which you convert to the corresponding value? It would also be nice to have both a setter and a getter for the endianess. The signature of the setter should be &mut self.

[lwandrebeck]

What about lwandrebeck@e0afced ?

[magnusuMET]

Starting to look good now! Feel free to open a PR so I can comment directly on it, and let travis have a go at it.

endian_value should return error::Result<Endianness> and enum Endianness should #[derive(Debug, Copy, Clone)].

There should also be a test that uses these methods to ensures they are visible to the user. You don't have to test whether the underlying netcdf library actually sets this.

[magnusuMET]

Endianness could also derive PartialEq, but then the equality native <-> little/big has to be decided.

[lwandrebeck]

Ok, I’m on test part. It looks like netcdf does not use Native, but converts silently to Little or Big depending on the arch the file is created on. I need to dig a bit deeper about that.

[magnusuMET]

Maybe one can put Native, but you will never get Native in return?

[lwandrebeck]

It seems even if you give Big on create, it is written as Little on a little-endian arch.
« and the library takes care of the conversions on reading or writing, so that programs using the library alsways deal with the data in native form. That's one of the purposes of the netCDF library, to insulate programmers from having to know anything about characterisitics of the computer on which the data was written. » from https://www.unidata.ucar.edu/support/help/MailArchives/netcdf/msg11257.html

[lwandrebeck]

I left test out of #35 as we need to dive a bit more into netcdf lib and decide what and how to test endianness.

[magnusuMET]

Might be that the endianess is not set before it is synced. You could try putting a value then reading it back out. The only purpose of setting the endianess ourself would be to avoid endianess conversion on the target, otherwise it should be hidden.

[magnusuMET]

We could test the true endianess once #33 is implemented.

[lwandrebeck]

Ok, I’ve fixed up a stupid bug introduced by myself in test.
Native is automatically converted by netcdf into either Little or Big. Adding test to PR.