Working with attributes

Every node can be augumented with attributes. Attributes behave a bit like datasets but lack some of the features the latter provides

• there is no partial IO on attributes

• attribute data cannot be compressed.

• you cannot use virtual datasets or SWMR on attributes.

Attributes can be accessed via the hdf5::attribute::Manager interface exposed by the attributes member of every hdf5::node::Node instance. hdf5::attribute::Manager provides an STL compliant container interface to attributes which is somehow similar to that of hdf5::node::LinkView or hdf5::node::NodeView.

An attribute itself is represented by an instance of hdf5::attribute::Attribute.

Basic attribute management

Creating attributes

Attribute creation is done via the create() member function of the attribute manager interface. To create a simple scalar string attribute you could use

using hdf5::attribute::Attribute;
hdf5::node::Dataset dataset = ...;

Attribute attribute = dataset.attributes.create<std::string>("author");

A multidimensional attribute can be created using

Attribute attribute = dataset.attributes.create<float>("orientation_matrix",{3,3});

which would lead to a 3x3 matrix attribute of type float. There exists also a non template version of create() which exposes all the flexibility HDF5 provides. So we could write the last example also with

hdf5::dataspace::Simple space{{3,3}};
auto type = hdf5::datatype::create<float>();
Attribute attr = dataset.attributes.create("orientation_matrix",type,space);

Removing attributes

Attributes can be removed using the remove() member functions of hdf5::attribute::Manager. This function takes either the index of the name of an attribute as its key

hdf5::node::Dataset dataset = ... ;

dataset.attributes.remove(1); //remove attribute with index 1
dataset.attributes.remove("temperature"); //remove attribute "temperature"

Both methods throw std::runtime_error in the case of a failure.

Attribute inquery

If we would like to know how many attributes are attached to a node we could use the size() member function of the manager interface.

hdf5::node::Group group = ...;
std::cout<<"Number of attributes: "<<group.attributes.size()<<std::endl;

the exists() method checks for the existence of a particular attribute with a given name

hdf5::node::Group group = ...;

if(group.attributes.exists("NX_class"))
{
 ....
}

Accessing attributes

Element wise access

Like nodes or links, attributes can be either accessed via their name or by their index.

hdf5::node::Dataset dataset = ...;

std::cout<<dataset.attributes["temperature"].name()<<std::endl;
std::cout<<dataset.attributes[1].name()<<std::endl;

Iteartor access

Iterating over attributes

hdf5::node::Group group = ....;

for(auto attribute: group.attributes)
{
    std::cout<<attribute.name()<<std::endl;
}

or alternatively

hdf5::node::Group group = ....;

std::for_each(group.attribtues.begin(),
              group.attributes.end(),
              [](const hdf5::attribute::Attribute &attr)
              { std::cout<<attr.name()<<std::endl; });

Reading and writing data

As proposed in design-read-write attribute_t provides read() and write() methods to read and write data.

attribute_t attr = group.attributes["NX_class"];

attr.write("NXdetector");

string nx_class;
attr.read(nx_class);