/** \page GLS Glossary \section GLS_A A
A variable that can store (logically) dense, rectilinear, multidimensional arrays of elements of a given HDF5 datatype.
The combination of array rank (dimensionality) and extent is called an array variable's shape. This includes the degenerate array shapes of a singleton (scalar) and the empty array (null).
The array element datatype is sometimes referred to as the array variable's type, which is not entirely accurate because the array variable's type is 'array of element type' rather than 'element type'.
In HDF5, there are two kinds of array variables, attributes and datasets, and the distinction is functional (i.e., how they can be used) rather than conceptual. Attributes are commonly used for descriptive "light-weight" HDF5 object metadata while datasets are HDF5 objects used to store "heavy-weight" problem-sized data.
A named array variable that is associated with an HDF5 object, its owner or attributee, and used to represent application domain-specific metadata of the object. Intuitively, the set of an object's attributes can be thought of as its key-value pair collection. Attribute names (keys) can be arbitrary Unicode strings, but must be unique per object, i.e., an object can have at most one attribute with a given name.
A scalar attribute is an attribute backed by a singleton array variable. A null attribute is attribute backed by an empty array variable.
A dataset storage layout where the dataset elements are partitioned into fixed-size multidimensional chunks or tiles. Chunked layout is mandatory for datasets with one or more dimensions of indefinite (infinite) extent or where compression or other filters are applied to the dataset elements.
Chunked layout may improve I/O performance for certain access patterns.
A dataset storage layout where the dataset elements are stored in the object header of the dataset. This layout is suitable for very small datasets that can easily fit in the object header.
Compact layout can improve storage and access performance for files that have many very small datasets.
A family of HDF5 datatypes whose elements are records with named fields of other HDF5 datatypes. Currently, on ASCII field names are supported.
Similar to a struct
in C or a COMMON
block in
Fortran.
A kind of HDF5 object, a linked array variable. which can be located in an HDF5 file through a path name. Datasets are commonly used to store "heavy-weight" problem-sized data.
The HDF5 library offers a lot of features aimed at optimized dataset access and storage, including compression and partial I/O.
An HDF5 datatype consists of an abstract data type (a set of elements) and a bit-level representation of these elements in storage such as an HDF5 file or memory.
The HDF5 library comes with a large set of predefined datatypes and offers mechanisms for creating user-defined datatypes.
The ten major families or classes of HDF5 datatypes are:
A kind of HDF5 object that stores a collection of HDF5 links. Each HDF5 file contains at least one group, it's root group.
Among the destinations of an HDF5 group's links may be other HDF5 groups (including the group itself!). This ability is sometimes referred to as the closure property of groups. It is the basis for creating hierarchical or more general graph-like structures.
A regular multidimensional pattern described by four vectors whose length equals the rank of the pattern.
start
- the offset where the first block of the hyperslab beginsstride
- the offset between pattern blockscount
- the number of blocksblock
- the extent of an individual pattern blockFor example, the black squares on a (two-dimensional) chessboard with
origin at (0,0)
can be represented as the union of two
hyperslabs representing the even (0,2,4,6)
and
odd (1,3,5,7)
rows, respectively.
The hyperslab parameters for the even rows are: start (0,0)
,
stride (2,2)
, count (4,4)
, block
(1,1)
. Likewise the parameters for the odd rows are: start
(1,1)
, stride (2,2)
, count
(4,4)
, block (1,1)
.
A named, uni-directional association between a source and a destination. In HDF5, the source is always the HDF5 group that hosts the link in its link collection.
There are several ways to specify a link's destination:
A link name can be any Unicode string that does not contain slashes
("/"
) or consists of a single dot character
("."
). A link name must be unique in a group's link
collection.
A link destination is sometimes referred to as a member of the link's source (group). This way of speaking invites confusion: A destination (e.g., object) can be the destination of multiple links in the same (!) or different groups. It would then be a "member" of a given group with multiplicity greater than one and be a member of multiple groups.
It is the link that is a member of the group's link collection and not the link destination.
A Unicode string that depending on the item it names might be subject to certain character restrictions, such as ASCII-encoded only. In HDF5, the user might encounter the following names:
'/'
). In HDF5, path names are used to locate and refer
to HDF5 objects.An HDF5 API construct, a means of customizing the behavior of the HDF5 library when creating, accessing or modifying HDF5 items.
While the default property settings are sufficient in many cases, certain HDF5 features, such as compression, can be reasonably controlled only by the user who has to provide the desired settings via property lists.
An HDF5 group that is present in all HDF5 files and that acts as the entry or base point for all other data stored in an HDF5 file.
The root group is "the mother of all objects" in an HDF5 file in the sense that all objects (and their attributes) can be discovered, beginning at the root group, by combinations of the following operations:
This discovery is portable and robust with respect to file-internal storage reorganization.
A variable-length (1-4 bytes per code point) encoding of the Unicode set of code points. This is the encoding supported by HDF5 to represent Unicode strings.
The ASCII encoding is a proper subset of UTF-8.