Remove Dimensions::getDims

[Flamefire]

This function (https://github.com/ComputationalRadiationPhysics/libSplash/blob/master/src/include/splash/Dimensions.hpp#L252-L263) is wrong, as it does neither honor zero sizes (all dims=0) nor does it work for offsets. As it can't be fixed properly in that design, it should be removed.

[ax3l]

careful, zero-components have still "size==1", e.g., a 2D data set has (X, Y, 1) size and (OX, OY, 0) offset

[Flamefire]

So a 2D zero size is (0, 0, 1)? This looks rather strange... Is this a HDF5/Splash related requirement?

[ax3l]

no there is no zero size in libSplash, the smallest element one can write is of size (1, 1, 1) while its dimensionality (1D, 2D, 3D) is set via dim.

[Flamefire]

Really? What about e.g. writing particles when there are no particles for the current process? Or field slices where the current process has no data because it is outside the slice?

[ax3l]

The limitation only applies to "global" sizes of data sets.
Please check https://github.com/ComputationalRadiationPhysics/picongpu/tree/dev/src/picongpu/include/plugins/hdf5 for details.

[ax3l]

zero-sized would be (0, 1, 1) ... don't judge. it's all from having a fixed 3D Dimensions object in splash instead of using a class that can do ND...

[Flamefire]

Well I did not know, that you can't write a zero-sized record and just did it. It actually writes a single value, but doesn't run into any problems besides that.

[ax3l]

It actually writes a single value, ...

yes, sounds like it.