There are 4 types of named data objects represented in the H5s graph: the group, the datatype, the dataset, and the attribute.
All the structural components—datatypes, dataspaces, storage classes, and dimensional shapes—that describe an HDF5 named data object are readable in a single line in the H5s hierarchical display.
These "one-liners" are key: grok your datasets effortlessly.
Group
The group is the hierarchical unit of the HDF5 datastore. It contains any of the following: other groups, datatypes, and datasets.
Datatype
The datatype is the storage format of a single data element. An atomic type, either predefined or custom, represents the smallest addressable unit of storage in HDF5. Datatypes are combined with any other datatype to form compound structured datatypes. It is the datatype specification that allows complete portablility of datastores across computing platforms.
Dataset
The elements specified by the datatype are organized by a dataspace, and this entity becomes the HDF5 dataset. Access to the dataset is defined by the storage layout given to the dataset. H5s renders all the principle defining aspects of a dataset in single line as concisely as possible.
Attribute
An attribute is a dataset with a specific storage layout that can be attached to any of the other named data objects. Attributes are part of the "data describing data" capability of HDF5. Attributes are given special placement in the graph using a specific attribute expander:
Attributes are not first-class citizens of the HDF5 graph, so H5s renders them to reflect this. When one or more attributes are attached to a named data object, the attribute expander is rendered as a suffix to the object name. This allows you to independently expand and collapse the attributes, and to see at a glance which named data objects are attributed. Additionally, this design also makes way for advantages in scrolling the graph as well. You can find more on this in the screenplay.
One other consideration in rendering named data objects.
Links
Links to named data objects can be inserted in the HDF5 graph. HDF5 links are either hard–a physical datastore reference to another named data object, or soft–storage of an HDF5 path to a named data object (which does not have to exist).
