Reading sections from a 1D dataset

[MillieFD]

I am working with a hdf5 file that contains several one-dimensional datasets and zero higher-dimensionality datasets. I have been reading the shorter datasets using read_raw which returns the data points contained in a Vec. This is fine for smaller datasets, but for larger datasets it would be preferable to read only the required section into a Vec and leave the rest of the dataset alone.

Current Solution

I found an example using read_slice_1d(ndarray::s![start..stop]); however, this returns a ndarray::Array1 which must then be converted into a Vec. It also requires yet another dependency (ndarray).

Question

Does anyone know of a "built-in" solution to read a slice from a 1D hdf5 dataset into a Vec without needing additional dependencies?

[mulimoen]

There are no convenience methods from reading into a buffer, and all reader functions returns an ndarray. You can use into_raw_vec_and_offset on an ndarray to get a Vec.

An alternative is using the netcdf crate which has get_values and get_values_into for when you want a vec or to put values into a user supplied buffer

[MillieFD]

Thanks for the feedback! I've implemented f.dataset("name")?.read_slice_1d(ndarray::s![start..end])?.to_vec() where f: hdf5::File for the time being, although this approach still requires adding ndarray as a dependency just to use the ndarray::s! macro. It functions fine, but feels a bit wasteful as ndarray isn't needed anywhere else in the project.

pub fn read_slice_1d<T, S>(&self, selection: S) -> Result<Array1<T>>
where
    T: H5Type,
    S: TryInto<Selection>,
    Error: From<S::Error>,

The signature for read_slice_1d (shown above for convenience) requires our input to implement TryInto<Selection>. Is there a way to construct S without needing the ndarray::s! macro?

I'm not familiar with the netcdf crate but it sounds like a promising alternative. Any chance you could provide a little example of how netcdf should be used in this instance?

[mulimoen]

You should be able to use a slice index directly as we have this impl: https://docs.rs/hdf5-metno/latest/hdf5_metno/enum.Selection.html#impl-From%3CRangeInclusive%3Cusize%3E%3E-for-Selection

[mulimoen]

For netcdf it would be something like

let var = todo!();
let mut buf = vec![0; 10];
var.get_values_into(buf, (start..start+10))?;

https://docs.rs/netcdf/0.11.0/netcdf/struct.Variable.html#method.get_values_into