Path subgroup access

[AntoineRenaud91]

Summary

This pull request introduces new methods for the File, MutableFile, Group, and MutableGroup structs. These methods enable users to fetch a specific subgroup, variable, or attribute within a netCDF structure using a relative path formatted like a Linux file system ("group1/group2/.../var").

Key Features

• Enables direct querying of nested structures in netCDF files.
• Eliminates the need to keep intermediate subgroups alive when only interested in a specific, deeply-nested subgroup, variable, or attribute.

Example Unit Test

Here's a unit test to demonstrate how the new feature works (can be found in netcdf/tests/file.rs):

#[test]
fn fetch_from_path() {
    // Create a temporary directory and netCDF file
    let d = tempfile::tempdir().unwrap();
    let path = d.path().join("cdf5.nc");
    {
        let mut file = netcdf::create(path.clone()).unwrap();
        let mut group = file.add_group("grp").unwrap();
        let mut subgroup = group.add_group("subgrp").unwrap();
        subgroup.add_dimension("dim", 1).unwrap();
        subgroup.add_variable::<f64>("var", &["dim"]).unwrap();
        subgroup.add_attribute("attr", "test").unwrap();
    }

    // Open the netCDF file and test the new methods
    let file = netcdf::open(path).unwrap();
    assert_eq!(
        file.group_from_path("grp/subgrp")
            .unwrap()
            .unwrap()
            .variable("var")
            .unwrap()
            .name(),
        file.variable_from_path("grp/subgrp/var").unwrap().name(),
    );

    match file
        .attribute_from_path("grp/subgrp/attr")
        .unwrap()
        .value()
        .unwrap()
    {
        netcdf::AttrValue::Str(string) => assert_eq!(string, "test"),
        _ => panic!(),
    }
}

This feature enhances user experience by providing a more intuitive and efficient way to navigate complex netCDF files.

[AntoineRenaud91]

I am not sure how to solve the last failed test. It seems that the test is not finding the libnetcdf library, which I don't believe is likely due to my changes.

[magnusuMET]

This feature looks great! In fact, why don't we make this the default so we could do

let var = file.variable("my/group/here/variable")?;

This works like the official python library and makes groups much more pleasant compared to the lifetime struggle with the old approach.

You can ignore the failing test and I will try to figure out what is going wrong there. It is not related to your changes

[AntoineRenaud91]

I'm glad you like this feature. Should I make a commit to set it as the default behavior and subsequently remove all the '_from_path' methods? It seems very unlikely that existing projects using this crate would have group or variable names with slashes.

[magnusuMET]

Would be great if you made that commit. Slashes are not allowed in object names in hdf5, so I would think they need to be disallowed in netCDF too.

[AntoineRenaud91]

Done ! :)

[AntoineRenaud91]

Just avoiding an unnecessary allocation.

I know the scope has increased a bit from the original idea, but this is a great ergonomic boost. Some other cases I have though up:

* Dimensions should also allow a group name

* What should we do if we try to add a variable `in/a/group`? Automatically create the group?

* What should we return if we request a variable in a group which does not exist? Return `None` or an error?

I think the API would be safer if we allowed recursive group creation only through the add_groupmethod. I would expect all other "add" methods to return an error if the subgroup structure doesn't exist.

As for requesting, I think it should return an error if the parent groups do not exist.

Maybe we should add new error variant ?

[magnusuMET]

We should probably add a new error variant, but that would be a breaking change which we could avoid by returning an Error::Str (and a TODO on Error for when we have to break the API)

[AntoineRenaud91]

I just commited these changes. I created a function called 'crate' to transform an 'ncid' and a 'path' string into the 'ncid' of the deepest subgroup and the stem name. It returns an error if any of the subgroups do not exist. This way, I didn't have to make significant modifications to all the 'add' methods. I simply used this function to update the 'ncid' and the stem name.

[magnusuMET]

Thanks a lot for your work on this @AntoineRenaud91, this is an awesome feature!

[AntoineRenaud91]

Happy to provide a useful contribution :)