HDF5 documents and links 
Introduction to HDF5 
HDF5 Reference Manual 
HDF5 User's Guide for Release 1.6 
And in this document, the HDF5 User's Guide from Release 1.4.5:    
Files   Datasets   Datatypes   Dataspaces   Groups  
References   Attributes   Property Lists   Error Handling  
Filters   Caching   Chunking   Mounting Files  
Performance   Debugging   Environment   DDL  

The Datatype Interface (H5T)

1. Introduction

The datatype interface provides a mechanism to describe the storage format of individual data points of a data set and is hopefully designed in such a way as to allow new features to be easily added without disrupting applications that use the datatype interface. A dataset (the H5D interface) is composed of a collection or raw data points of homogeneous type organized according to the data space (the H5S interface).

A datatype is a collection of datatype properties, all of which can be stored on disk, and which when taken as a whole, provide complete information for data conversion to or from that datatype. The interface provides functions to set and query properties of a datatype.

A data point is an instance of a datatype, which is an instance of a type class. We have defined a set of type classes and properties which can be extended at a later time. The atomic type classes are those which describe types which cannot be decomposed at the datatype interface level; all other classes are compound.

2. General Datatype Operations

The functions defined in this section operate on datatypes as a whole. New datatypes can be created from scratch or copied from existing datatypes. When a datatype is no longer needed its resources should be released by calling H5Tclose().

Datatypes come in two flavors: named datatypes and transient datatypes. A named datatype is stored in a file while the transient flavor is independent of any file. Named datatypes are always read-only, but transient types come in three varieties: modifiable, read-only, and immutable. The difference between read-only and immutable types is that immutable types cannot be closed except when the entire library is closed (the predefined types like H5T_NATIVE_INT are immutable transient types).

hid_t H5Tcreate (H5T_class_t class, size_t size)
Datatypes can be created by calling this function, where class is a datatype class identifier. However, the only class currently allowed is H5T_COMPOUND to create a new empty compound datatype where size is the total size in bytes of an instance of this datatype. Other datatypes are created with H5Tcopy(). All functions that return datatype identifiers return a negative value for failure.

hid_t H5Topen (hid_t location, const char *name)
A named datatype can be opened by calling this function, which returns a datatype identifier. The identifier should eventually be released by calling H5Tclose() to release resources. The named datatype returned by this function is read-only or a negative value is returned for failure. The location is either a file or group identifier.

herr_t H5Tcommit (hid_t location, const char *name, hid_t type)
A transient datatype (not immutable) can be committed to a file and turned into a named datatype by calling this function. The location is either a file or group identifier and when combined with name refers to a new named datatype.

htri_t H5Tcommitted (hid_t type)
A type can be queried to determine if it is a named type or a transient type. If this function returns a positive value then the type is named (that is, it has been committed perhaps by some other application). Datasets which return committed datatypes with H5Dget_type() are able to share the datatype with other datasets in the same file.

hid_t H5Tcopy (hid_t type)
This function returns a modifiable transient datatype which is a copy of type or a negative value for failure. If type is a dataset identifier then the type returned is a modifiable transient copy of the datatype of the specified dataset.

herr_t H5Tclose (hid_t type)
Releases resources associated with a datatype. The datatype identifier should not be subsequently used since the results would be unpredictable. It is illegal to close an immutable transient datatype.

htri_t H5Tequal (hid_t type1, hid_t type2)
Determines if two types are equal. If type1 and type2 are the same then this function returns TRUE, otherwise it returns FALSE (an error results in a negative return value).

herr_t H5Tlock (hid_t type)
A transient datatype can be locked, making it immutable (read-only and not closable). The library does this to all predefined types to prevent the application from inadvertently modifying or deleting (closing) them, but the application is also allowed to do this for its own datatypes. Immutable datatypes are closed when the library closes (either by H5close() or by normal program termination).

3. Properties of Atomic Types

An atomic type is a type which cannot be decomposed into smaller units at the API level. All atomic types have a common set of properties which are augmented by properties specific to a particular type class. Some of these properties also apply to compound datatypes, but we discuss them only as they apply to atomic datatypes here. The properties and the functions that query and set their values are:

H5T_class_t H5Tget_class (hid_t type)
This property holds one of the class names: H5T_INTEGER, H5T_FLOAT, H5T_TIME, H5T_STRING, or H5T_BITFIELD. This property is read-only and is set when the datatype is created or copied (see H5Tcreate(), H5Tcopy()). If this function fails it returns H5T_NO_CLASS which has a negative value (all other class constants are non-negative).

size_t H5Tget_size (hid_t type)
herr_t H5Tset_size (hid_t type, size_t size)
This property is total size of the datum in bytes, including padding which may appear on either side of the actual value. If this property is reset to a smaller value which would cause the significant part of the data to extend beyond the edge of the datatype then the offset property is decremented a bit at a time. If the offset reaches zero and the significant part of the data still extends beyond the edge of the datatype then the precision property is decremented a bit at a time. Decreasing the size of a datatype may fail if the H5T_FLOAT bit fields would extend beyond the significant part of the type. Adjusting the size of an H5T_STRING automatically adjusts the precision as well. On error, H5Tget_size() returns zero which is never a valid size.

H5T_order_t H5Tget_order (hid_t type)
herr_t H5Tset_order (hid_t type, H5T_order_t order)
All atomic datatypes have a byte order which describes how the bytes of the datatype are layed out in memory. If the lowest memory address contains the least significant byte of the datum then it is said to be little-endian or H5T_ORDER_LE. If the bytes are in the oposite order then they are said to be big-endian or H5T_ORDER_BE. Some datatypes have the same byte order on all machines and are H5T_ORDER_NONE (like character strings). If H5Tget_order() fails then it returns H5T_ORDER_ERROR which is a negative value (all successful return values are non-negative).

size_t H5Tget_precision (hid_t type)
herr_t H5Tset_precision (hid_t type, size_t precision)
Some datatypes occupy more bytes than what is needed to store the value. For instance, a short on a Cray is 32 significant bits in an eight-byte field. The precision property identifies the number of significant bits of a datatype and the offset property (defined below) identifies its location. The size property defined above represents the entire size (in bytes) of the datatype. If the precision is decreased then padding bits are inserted on the MSB side of the significant bits (this will fail for H5T_FLOAT types if it results in the sign, mantissa, or exponent bit field extending beyond the edge of the significant bit field). On the other hand, if the precision is increased so that it "hangs over" the edge of the total size then the offset property is decremented a bit at a time. If the offset reaches zero and the significant bits still hang over the edge, then the total size is increased a byte at a time. The precision of an H5T_STRING is read-only and is always eight times the value returned by H5Tget_size(). H5Tget_precision() returns zero on failure since zero is never a valid precision.

size_t H5Tget_offset (hid_t type)
herr_t H5Tset_offset (hid_t type, size_t offset)
While the precision property defines the number of significant bits, the offset property defines the location of those bits within the entire datum. The bits of the entire data are numbered beginning at zero at the least significant bit of the least significant byte (the byte at the lowest memory address for a little-endian type or the byte at the highest address for a big-endian type). The offset property defines the bit location of the least signficant bit of a bit field whose length is precision. If the offset is increased so the significant bits "hang over" the edge of the datum, then the size property is automatically incremented. The offset is a read-only property of an H5T_STRING and is always zero. H5Tget_offset() returns zero on failure which is also a valid offset, but is guaranteed to succeed if a call to H5Tget_precision() succeeds with the same arguments.

herr_t H5Tget_pad (hid_t type, H5T_pad_t *lsb, H5T_pad_t *msb)
herr_t H5Tset_pad (hid_t type, H5T_pad_t lsb, H5T_pad_t msb)
The bits of a datum which are not significant as defined by the precision and offset properties are called padding. Padding falls into two categories: padding in the low-numbered bits is lsb padding and padding in the high-numbered bits is msb padding (bits are numbered according to the description for the offset property). Padding bits can always be set to zero (H5T_PAD_ZERO) or always set to one (H5T_PAD_ONE). The current pad types are returned through arguments of H5Tget_pad() either of which may be null pointers.

3.1. Properties of Integer Atomic Types

Integer atomic types (class=H5T_INTEGER) describe integer number formats. Such types include the following information which describes the type completely and allows conversion between various integer atomic types.

H5T_sign_t H5Tget_sign (hid_t type)
herr_t H5Tset_sign (hid_t type, H5T_sign_t sign)
Integer data can be signed two's complement (H5T_SGN_2) or unsigned (H5T_SGN_NONE). Whether data is signed or not becomes important when converting between two integer datatypes of differing sizes as it determines how values are truncated and sign extended.

3.2. Properties of Floating-point Atomic Types

The library supports floating-point atomic types (class=H5T_FLOAT) as long as the bits of the exponent are contiguous and stored as a biased positive number, the bits of the mantissa are contiguous and stored as a positive magnitude, and a sign bit exists which is set for negative values. Properties specific to floating-point types are:

herr_t H5Tget_fields (hid_t type, size_t *spos, size_t *epos, size_t *esize, size_t *mpos, size_t *msize)
herr_t H5Tset_fields (hid_t type, size_t spos, size_t epos, size_t esize, size_t mpos, size_t msize)
A floating-point datum has bit fields which are the exponent and mantissa as well as a mantissa sign bit. These properties define the location (bit position of least significant bit of the field) and size (in bits) of each field. The bit positions are numbered beginning at zero at the beginning of the significant part of the datum (see the descriptions of the precision and offset properties). The sign bit is always of length one and none of the fields are allowed to overlap. When expanding a floating-point type one should set the precision first; when decreasing the size one should set the field positions and sizes first.

size_t H5Tget_ebias (hid_t type)
herr_t H5Tset_ebias (hid_t type, size_t ebias)
The exponent is stored as a non-negative value which is ebias larger than the true exponent. H5Tget_ebias() returns zero on failure which is also a valid exponent bias, but the function is guaranteed to succeed if H5Tget_precision() succeeds when called with the same arguments.

H5T_norm_t H5Tget_norm (hid_t type)
herr_t H5Tset_norm (hid_t type, H5T_norm_t norm)
This property determines the normalization method of the mantissa.

H5T_pad_t H5Tget_inpad (hid_t type)
herr_t H5Tset_inpad (hid_t type, H5T_pad_t inpad)
If any internal bits (that is, bits between the sign bit, the mantissa field, and the exponent field but within the precision field) are unused, then they will be filled according to the value of this property. The inpad argument can be H5T_PAD_ZERO if the internal padding should always be set to zero, or H5T_PAD_ONE if it should always be set to one. H5Tget_inpad() returns H5T_PAD_ERROR on failure which is a negative value (successful return is always non-negative).

3.3. Properties of Date and Time Atomic Types

Dates and times (class=H5T_TIME) are stored as character strings in one of the ISO-8601 formats like "1997-12-05 16:25:30"; as character strings using the Unix asctime(3) format like "Thu Dec 05 16:25:30 1997"; as an integer value by juxtaposition of the year, month, and day-of-month, hour, minute and second in decimal like 19971205162530; as an integer value in Unix time(2) format; or other variations.

3.4. Properties of Character String Atomic Types

Fixed-length character string types are used to store textual information. The offset property of a string is always zero and the precision property is eight times as large as the value returned by H5Tget_size() (since precision is measured in bits while size is measured in bytes). Both properties are read-only.

H5T_cset_t H5Tget_cset (hid_t type)
herr_t H5Tset_cset (hid_t type, H5T_cset_t cset)
HDF5 is able to distinguish between character sets of different nationalities and to convert between them to the extent possible. The only character set currently supported is H5T_CSET_ASCII.

H5T_str_t H5Tget_strpad (hid_t type)
herr_t H5Tset_strpad (hid_t type, H5T_str_t strpad)
The method used to store character strings differs with the programming language: C usually null terminates strings while Fortran left-justifies and space-pads strings. This property defines the storage mechanism and can be

A C-style string which is guaranteed to be null terminated. When converting from a longer string the value will be truncated and then a null character appended.

A C-style string which is padded with null characters but not necessarily null terminated. Conversion from a long string to a shorter H5T_STR_NULLPAD string will truncate but not null terminate. Conversion from a short value to a longer value will append null characters as with H5T_STR_NULLTERM.

A Fortran-style string which is padded with space characters. This is the same as H5T_STR_NULLPAD except the padding character is a space instead of a null.

H5Tget_strpad() returns H5T_STR_ERROR on failure, a negative value (all successful return values are non-negative).

3.5. Properties of Bit Field Atomic Types

Converting a bit field (class=H5T_BITFIELD) from one type to another simply copies the significant bits. If the destination is smaller than the source then bits are truncated. Otherwise new bits are filled according to the msb padding type.

3.6. Character and String Datatype Issues

The H5T_NATIVE_CHAR and H5T_NATIVE_UCHAR datatypes are actually numeric data (1-byte integers). If the application wishes to store character data, then an HDF5 string datatype should be derived from H5T_C_S1 instead.


HDF5 defines at least three classes of datatypes: integer data, floating point data, and character data. However, the C language defines only integer and floating point datatypes; character data in C is overloaded on the 8- or 16-bit integer types and character strings are overloaded on arrays of those integer types which, by convention, are terminated with a zero element. In C, the variable unsigned char s[256] is either an array of numeric data, a single character string with at most 255 characters, or an array of 256 characters, depending entirely on usage. For uniformity with the other H5T_NATIVE_ types, HDF5 uses the numeric interpretation of H5T_NATIVE_CHAR and H5T_NATIVE_UCHAR.


To store unsigned char s[256] data as an array of integer values, use the HDF5 datatype H5T_NATIVE_UCHAR and a data space that describes the 256-element array. Some other application that reads the data will then be able to read, say, a 256-element array of 2-byte integers and HDF5 will perform the numeric translation. To store unsigned char s[256] data as a character string, derive a fixed length string datatype from H5T_C_S1 by increasing its size to 256 characters. Some other application that reads the data will be able to read, say, a space padded string of 16-bit characters and HDF5 will perform the character and padding translations.
                hid_t s256 = H5Tcopy(H5T_C_S1);
                             H5Tset_size(s256, 256);
To store unsigned char s[256] data as an array of 256 ASCII characters, use an HDF5 data space to describe the array and derive a one-character string type from H5T_C_S1. Some other application will be able to read a subset of the array as 16-bit characters and HDF5 will perform the character translations. The H5T_STR_NULLPAD is necessary because if H5T_STR_NULLTERM were used (the default) then the single character of storage would be for the null terminator and no useful data would actually be stored (unless the length were incremented to more than one character).
                hid_t s1 = H5Tcopy(H5T_C_S1);
                           H5Tset_strpad(s1, H5T_STR_NULLPAD);


The C language uses the term char to represent one-byte numeric data and does not make character strings a first-class datatype. HDF5 makes a distinction between integer and character data and maps the C signed char (H5T_NATIVE_CHAR) and unsigned char (H5T_NATIVE_UCHAR) datatypes to the HDF5 integer type class.

4. Properties of Opaque Types

Opaque types (class=H5T_OPAQUE) provide the application with a mechanism for describing data which cannot be otherwise described by HDF5. The only properties associated with opaque types are a size in bytes and an ASCII tag which is manipulated with H5Tset_tag() and H5Tget_tag() functions. The library contains no predefined conversion functions but the application is free to register conversions between any two opaque types or between an opaque type and some other type.

5. Properties of Compound Types

A compound datatype is similar to a struct in C or a common block in Fortran: it is a collection of one or more atomic types or small arrays of such types. Each member of a compound type has a name which is unique within that type, and a byte offset that determines the first byte (smallest byte address) of that member in a compound datum. A compound datatype has the following properties:

H5T_class_t H5Tget_class (hid_t type)
All compound datatypes belong to the type class H5T_COMPOUND. This property is read-only and is defined when a datatype is created or copied (see H5Tcreate() or H5Tcopy()).

size_t H5Tget_size (hid_t type)
Compound datatypes have a total size in bytes which is returned by this function. All members of a compound datatype must exist within this size. A value of zero is returned for failure; all successful return values are positive.

int H5Tget_nmembers (hid_t type)
A compound datatype consists of zero or more members (defined in any order) with unique names and which occupy non-overlapping regions within the datum. In the functions that follow, individual members are referenced by an index number between zero and N-1, inclusive, where N is the value returned by this function. H5Tget_nmembers() returns -1 on failure.

char *H5Tget_member_name (hid_t type, int membno)
Each member has a name which is unique among its siblings in a compound datatype. This function returns a pointer to a null-terminated copy of the name allocated with malloc() or the null pointer on failure. The caller is responsible for freeing the memory returned by this function.

size_t H5Tget_member_offset (hid_t type, int membno)
The byte offset of member number membno with respect to the beginning of the containing compound datum is returned by this function. A zero is returned on failure which is also a valid offset, but this function is guaranteed to succeed if a call to H5Tget_member_class() succeeds when called with the same type and membno arguments.

hid_t H5Tget_member_type (hid_t type, int membno)
Each member has its own datatype, a copy of which is returned by this function. The returned datatype identifier should be released by eventually calling H5Tclose() on that type.

Properties of members of a compound datatype are defined when the member is added to the compound type (see H5Tinsert()) and cannot be subsequently modified. This makes it imposible to define recursive data structures.

6. Predefined Atomic Datatypes

The library predefines a modest number of datatypes having names like H5T_arch_base where arch is an architecture name and base is a programming type name. New types can be derived from the predifined types by copying the predefined type (see H5Tcopy()) and then modifying the result.

Architecture Name Description
IEEE This architecture defines standard floating point types in various byte orders.
STD This is an architecture that contains semi-standard datatypes like signed two's complement integers, unsigned integers, and bitfields in various byte orders.
UNIX Types which are specific to Unix operating systems are defined in this architecture. The only type currently defined is the Unix date and time types (time_t).
Types which are specific to the C or Fortran programming languages are defined in these architectures. For instance, H5T_C_STRING defines a base string type with null termination which can be used to derive string types of other lengths.
NATIVE This architecture contains C-like datatypes for the machine on which the library was compiled. The types were actually defined by running the H5detect program when the library was compiled. In order to be portable, applications should almost always use this architecture to describe things in memory.
CRAY Cray architectures. These are word-addressable, big-endian systems with non-IEEE floating point.
INTEL All Intel and compatible CPU's including 80286, 80386, 80486, Pentium, Pentium-Pro, and Pentium-II. These are little-endian systems with IEEE floating-point.
MIPS All MIPS CPU's commonly used in SGI systems. These are big-endian systems with IEEE floating-point.
ALPHA All DEC Alpha CPU's, little-endian systems with IEEE floating-point.

The base name of most types consists of a letter, a precision in bits, and an indication of the byte order. The letters are:

B Bitfield
D Date and time
F Floating point
I Signed integer
R References
S Character string
U Unsigned integer

The byte order is a two-letter sequence:

BE Big endian
LE Little endian
VX Vax order


H5T_IEEE_F64LE Eight-byte, little-endian, IEEE floating-point
H5T_IEEE_F32BE Four-byte, big-endian, IEEE floating point
H5T_STD_I32LE Four-byte, little-endian, signed two's complement integer
H5T_STD_U16BE Two-byte, big-endian, unsigned integer
H5T_UNIX_D32LE Four-byte, little-endian, time_t
H5T_C_S1 One-byte, null-terminated string of eight-bit characters
H5T_INTEL_B64 Eight-byte bit field on an Intel CPU
H5T_CRAY_F64 Eight-byte Cray floating point
H5T_STD_ROBJ Reference to an entire object in a file

The NATIVE architecture has base names which don't follow the same rules as the others. Instead, native type names are similar to the C type names. Here are some examples:


Corresponding C Type
H5T_NATIVE_SCHAR signed char
H5T_NATIVE_UCHAR unsigned char
H5T_NATIVE_USHORT unsigned short
H5T_NATIVE_UINT unsigned
H5T_NATIVE_ULONG unsigned long
H5T_NATIVE_LLONG long long
H5T_NATIVE_ULLONG unsigned long long
H5T_NATIVE_LDOUBLE long double

Example: A 128-bit integer

To create a 128-bit, little-endian signed integer type one could use the following (increasing the precision of a type automatically increases the total size):

hid_t new_type = H5Tcopy (H5T_NATIVE_INT);
H5Tset_precision (new_type, 128);
H5Tset_order (new_type, H5T_ORDER_LE);

Example: An 80-character string

To create an 80-byte null terminated string type one might do this (the offset of a character string is always zero and the precision is adjusted automatically to match the size):

hid_t str80 = H5Tcopy (H5T_C_S1);
H5Tset_size (str80, 80);

A complete list of the datatypes predefined in HDF5 can be found in HDF5 Predefined Datatypes in the HDF5 Reference Manual.

7. Defining Compound Datatypes

Unlike atomic datatypes which are derived from other atomic datatypes, compound datatypes are created from scratch. First, one creates an empty compound datatype and specifies it's total size. Then members are added to the compound datatype in any order.

Usually a C struct will be defined to hold a data point in memory, and the offsets of the members in memory will be the offsets of the struct members from the beginning of an instance of the struct.

This macro computes the offset of member m within a struct s.
This macro defined in stddef.h does exactly the same thing as the HOFFSET() macro.

Each member must have a descriptive name which is the key used to uniquely identify the member within the compound datatype. A member name in an HDF5 datatype does not necessarily have to be the same as the name of the member in the C struct, although this is often the case. Nor does one need to define all members of the C struct in the HDF5 compound datatype (or vice versa).

Example: A simple struct

An HDF5 datatype is created to describe complex numbers whose type is defined by the complex_t struct.

typedef struct {
   double re;   /*real part*/
   double im;   /*imaginary part*/
} complex_t;

hid_t complex_id = H5Tcreate (H5T_COMPOUND, sizeof tmp);
H5Tinsert (complex_id, "real", HOFFSET(complex_t,re),
H5Tinsert (complex_id, "imaginary", HOFFSET(complex_t,im),

Member alignment is handled by the HOFFSET macro. However, data stored on disk does not require alignment, so unaligned versions of compound data structures can be created to improve space efficiency on disk. These unaligned compound datatypes can be created by computing offsets by hand to eliminate inter-member padding, or the members can be packed by calling H5Tpack() (which modifies a datatype directly, so it is usually preceded by a call to H5Tcopy()):

Example: A packed struct

This example shows how to create a disk version of a compound datatype in order to store data on disk in as compact a form as possible. Packed compound datatypes should generally not be used to describe memory as they may violate alignment constraints for the architecture being used. Note also that using a packed datatype for disk storage may involve a higher data conversion cost.

hid_t complex_disk_id = H5Tcopy (complex_id);
H5Tpack (complex_disk_id);

Example: A flattened struct

Compound datatypes that have a compound datatype member can be handled two ways. This example shows that the compound datatype can be flattened, resulting in a compound type with only atomic members.

typedef struct {
   complex_t x;
   complex_t y;
} surf_t;

hid_t surf_id = H5Tcreate (H5T_COMPOUND, sizeof tmp);
H5Tinsert (surf_id, "x-re", HOFFSET(surf_t,x.re),
H5Tinsert (surf_id, "x-im", HOFFSET(surf_t,x.im),
H5Tinsert (surf_id, "y-re", HOFFSET(surf_t,y.re),
H5Tinsert (surf_id, "y-im", HOFFSET(surf_t,y.im),

Example: A nested struct

However, when the complex_t is used often it becomes inconvenient to list its members over and over again. So the alternative approach to flattening is to define a compound datatype and then use it as the type of the compound members, as is done here (the typedefs are defined in the previous examples).

hid_t complex_id, surf_id; /*hdf5 datatypes*/

complex_id = H5Tcreate (H5T_COMPOUND, sizeof c);
H5Tinsert (complex_id, "re", HOFFSET(complex_t,re),
H5Tinsert (complex_id, "im", HOFFSET(complex_t,im),

surf_id = H5Tcreate (H5T_COMPOUND, sizeof s);
H5Tinsert (surf_id, "x", HOFFSET(surf_t,x), complex_id);
H5Tinsert (surf_id, "y", HOFFSET(surf_t,y), complex_id);

8. Enumeration Datatypes

8.1. Introduction

An HDF enumeration datatype is a 1:1 mapping between a set of symbols and a set of integer values, and an order is imposed on the symbols by their integer values. The symbols are passed between the application and library as character strings and all the values for a particular enumeration type are of the same integer type, which is not necessarily a native type.

8.2. Creation

Creation of an enumeration datatype resembles creation of a compound datatype: first an empty enumeration type is created, then members are added to the type, then the type is optionally locked.

hid_t H5Tcreate(H5T_class_t type_class, size_t size)
This function creates a new empty enumeration datatype based on a native signed integer type. The first argument is the constant H5T_ENUM and the second argument is the size in bytes of the native integer on which the enumeration type is based. If the architecture does not support a native signed integer of the specified size then an error is returned.
/* Based on a native signed short */
hid_t hdf_en_colors = H5Tcreate(H5T_ENUM, sizeof(short));
hid_t H5Tenum_create(hid_t base)
This function creates a new empty enumeration datatype based on some integer datatype base and is a generalization of the H5Tcreate() function. This function is useful when creating an enumeration type based on some non-native integer datatype, but it can be used for native types as well.
/* Based on a native unsigned short */
hid_t hdf_en_colors_1 = H5Tenum_create(H5T_NATIVE_USHORT);

/* Based on a MIPS 16-bit unsigned integer */
hid_t hdf_en_colors_2 = H5Tenum_create(H5T_MIPS_UINT16);

/* Based on a big-endian 16-bit unsigned integer */
hid_t hdf_en_colors_3 = H5Tenum_create(H5T_STD_U16BE);
herr_t H5Tenum_insert(hid_t etype, const char *symbol, void *value)
Members are inserted into the enumeration datatype etype with this function. Each member has a symbolic name symbol and some integer representation value. The value argument must point to a value of the same datatype as specified when the enumeration type was created. The order of member insertion is not important but all symbol names and values must be unique within a particular enumeration type.
short val;
H5Tenum_insert(hdf_en_colors, "RED",   (val=0,&val));
H5Tenum_insert(hdf_en_colors, "GREEN", (val=1,&val));
H5Tenum_insert(hdf_en_colors, "BLUE",  (val=2,&val));
H5Tenum_insert(hdf_en_colors, "WHITE", (val=3,&val));
H5Tenum_insert(hdf_en_colors, "BLACK", (val=4,&val));
herr_t H5Tlock(hid_t etype)
This function locks a datatype so it cannot be modified or freed unless the entire HDF5 library is closed. Its use is completely optional but using it on an application datatype makes that datatype act like a predefined datatype.

8.3. Integer Operations

Because an enumeration datatype is derived from an integer datatype, any operation which can be performed on integer datatypes can also be performed on enumeration datatypes. This includes:

H5Topen() H5Tcreate() H5Tcopy() H5Tclose()
H5Tequal() H5Tlock() H5Tcommit() H5Tcommitted()
H5Tget_class() H5Tget_size() H5Tget_order() H5Tget_pad()
H5Tget_precision() H5Tget_offset() H5Tget_sign() H5Tset_size()
H5Tset_order() H5Tset_precision() H5Tset_offset() H5Tset_pad()

In addition, the new function H5Tget_super() will be defined for all datatypes that are derived from existing types (currently just enumeration types).

hid_t H5Tget_super(hid_t type)
Return the datatype from which type is derived. When type is an enumeration datatype then the returned value will be an integer datatype but not necessarily a native type. One use of this function would be to create a new enumeration type based on the same underlying integer type and values but with possibly different symbols.
hid_t itype = H5Tget_super(hdf_en_colors);
hid_t hdf_fr_colors = H5Tenum_create(itype);

short val;
H5Tenum_insert(hdf_fr_colors, "ouge",  (val=0,&val));
H5Tenum_insert(hdf_fr_colors, "vert",  (val=1,&val));
H5Tenum_insert(hdf_fr_colors, "bleu",  (val=2,&val));
H5Tenum_insert(hdf_fr_colors, "blanc", (val=3,&val));
H5Tenum_insert(hdf_fr_colors, "noir",  (val=4,&val));

8.4. Type Functions

A small set of functions is available for querying properties of an enumeration type. These functions are likely to be used by browsers to display datatype information.

int H5Tget_nmembers(hid_t etype)
When given an enumeration datatype etype this function returns the number of members defined for that type. This function is already implemented for compound datatypes.

char *H5Tget_member_name(hid_t etype, int membno)
Given an enumeration datatype etype this function returns the symbol name for the member indexed by membno. Members are numbered from zero to N-1 where N is the return value from H5Tget_nmembers(). The members are stored in no particular order. This function is already implemented for compound datatypes. If an error occurs then the null pointer is returned. The return value should be freed by calling free().

herr_t H5Tget_member_value(hid_t etype, int membno, void *value/*out*/)
Given an enumeration datatype etype this function returns the value associated with the member indexed by membno (as described for H5Tget_member_name()). The value returned is in the domain of the underlying integer datatype which is often a native integer type. The application should ensure that the memory pointed to by value is large enough to contain the result (the size can be obtained by calling H5Tget_size() on either the enumeration type or the underlying integer type when the type is not known by the C compiler.
int i, n = H5Tget_nmembers(hdf_en_colors);
for (i=0; i<n; i++) {
    char *symbol = H5Tget_member_name(hdf_en_colors, i);
    short val;
    H5Tget_member_value(hdf_en_colors, i, &val);
    printf("#%d %20s = %d\n", i, symbol, val);


#0                BLACK = 4
#1                 BLUE = 2
#2                GREEN = 1
#3                  RED = 0
#4                WHITE = 3

8.5. Data Functions

In addition to querying about the enumeration type properties, an application may want to make queries about enumerated data. These functions perform efficient mappings between symbol names and values.

herr_t H5Tenum_valueof(hid_t etype, const char *symbol, void *value/*out*/)
Given an enumeration datatype etype this function returns through value the bit pattern associated with the symbol name symbol. The value argument should point to memory which is large enough to hold the result, which is returned as the underlying integer datatype specified when the enumeration type was created, often a native integer type.

herr_t H5Tenum_nameof(hid_t etype, void *value, char *symbol, size_t size)
This function translates a bit pattern pointed to by value to a symbol name according to the mapping defined in the enumeration datatype etype and stores at most size characters of that name (counting the null terminator) to the symbol buffer. If the name is longer than the result buffer then the result is not null terminated and the function returns failure. If value points to a bit pattern which is not in the domain of the enumeration type then the first byte of the symbol buffer is set to zero and the function fails.
short data[1000] = {4, 2, 0, 0, 5, 1, ...};
int i;
char symbol[32];

for (i=0; i<1000; i++) {
    if (H5Tenum_nameof(hdf_en_colors, data+i, symbol,
                       sizeof symbol))<0) {
        if (symbol[0]) {
            strcpy(symbol+sizeof(symbol)-4, "...");
        } else {
            strcpy(symbol, "UNKNOWN");
    printf("%d %s\n", data[i], symbol);



8.6. Conversion

Enumerated data can be converted from one type to another provided the destination enumeration type contains all the symbols of the source enumeration type. The conversion operates by matching up the symbol names of the source and destination enumeration types to build a mapping from source value to destination value. For instance, if we are translating from an enumeration type that defines a sequence of integers as the values for the colors to a type that defines a different bit for each color then the mapping might look like this:

Enumeration Mapping

That is, a source value of 2 which corresponds to BLUE would be mapped to 0x0004. The following code snippet builds the second datatype, then converts a raw data array from one datatype to another, and then prints the result.

/* Create a new enumeration type */
short val;
hid_t bits = H5Tcreate(H5T_ENUM, sizeof val);
H5Tenum_insert(bits, "RED",   (val=0x0001,&val));
H5Tenum_insert(bits, "GREEN", (val=0x0002,&val));
H5Tenum_insert(bits, "BLUE",  (val=0x0004,&val));
H5Tenum_insert(bits, "WHITE", (val=0x0008,&val));
H5Tenum_insert(bits, "BLACK", (val=0x0010,&val));

/* The data */
short data[6] = {1, 4, 2, 0, 3, 5};

/* Convert the data from one type to another */
H5Tconvert(hdf_en_colors, bits, 5, data, NULL, plist_id);

/* Print the data */
for (i=0; i<6; i++) {
    printf("0x%04x\n", (unsigned)(data[i]));



If the source data stream contains values which are not in the domain of the conversion map then an overflow exception is raised within the library, causing the application defined overflow handler to be invoked (see H5Tset_overflow()). If no overflow handler is defined then all bits of the destination value will be set.

The HDF library will not provide conversions between enumerated data and integers although the application is free to do so (this is a policy we apply to all classes of HDF datatypes). However, since enumeration types are derived from integer types it is permissible to treat enumerated data as integers and perform integer conversions in that context.

8.7. Symbol Order

Symbol order is determined by the integer values associated with each symbol. When the integer datatype is a native type, testing the relative order of two symbols is an easy process: simply compare the values of the symbols. If only the symbol names are available then the values must first be determined by calling H5Tenum_valueof().

short val1, val2;
H5Tenum_valueof(hdf_en_colors, "WHITE", &val1);
H5Tenum_valueof(hdf_en_colors, "BLACK", &val2);
if (val1 < val2) ...

When the underlying integer datatype is not a native type then the easiest way to compare symbols is to first create a similar enumeration type that contains all the same symbols but has a native integer type (HDF type conversion features can be used to convert the non-native values to native values). Once we have a native type we can compare symbol order as just described. If foreign is some non-native enumeration type then a native type can be created as follows:

int n = H5Tget_nmembers(foreign);
hid_t itype = H5Tget_super(foreign);
void *val = malloc(n * MAX(H5Tget_size(itype), sizeof(int)));
char *name = malloc(n * sizeof(char*));
int i;

/* Get foreign type information */
for (i=0; i<n; i++) {
    name[i] = H5Tget_member_name(foreign, i);
    H5Tget_member_value(foreign, i,

/* Convert integer values to new type */
H5Tconvert(itype, H5T_NATIVE_INT, n, val, NULL, plist_id);

/* Build a native type */
hid_t native = H5Tenum_create(H5T_NATIVE_INT);
for (i=0; i<n; i++) {
    H5Tenum_insert(native, name[i], ((int*)val)[i]);

It is also possible to convert enumerated data to a new type that has a different order defined for the symbols. For instance, we can define a new type, reverse that defines the same five colors but in the reverse order.

short val;
int i;
char sym[8];
short data[5] = {0, 1, 2, 3, 4};

hid_t reverse = H5Tenum_create(H5T_NATIVE_SHORT);
H5Tenum_insert(reverse, "BLACK", (val=0,&val));
H5Tenum_insert(reverse, "WHITE", (val=1,&val));
H5Tenum_insert(reverse, "BLUE",  (val=2,&val));
H5Tenum_insert(reverse, "GREEN", (val=3,&val));
H5Tenum_insert(reverse, "RED",   (val=4,&val));

/* Print data */
for (i=0; i<5; i++) {
    H5Tenum_nameof(hdf_en_colors, data+i, sym, sizeof sym);
    printf ("%d %s\n", data[i], sym);

H5Tconvert(hdf_en_colors, reverse, 5, data, NULL, plist_id);

/* Print data */
for (i=0; i<5; i++) {
    H5Tenum_nameof(reverse, data+i, sym, sizeof sym);
    printf ("%d %s\n", data[i], sym);



8.8. Equality

The order that members are inserted into an enumeration type is unimportant; the important part is the associations between the symbol names and the values. Thus, two enumeration datatypes will be considered equal if and only if both types have the same symbol/value associations and both have equal underlying integer datatypes. Type equality is tested with the H5Tequal() function.

8.9. Interacting with C's enum Type

Although HDF enumeration datatypes are similar to C enum datatypes, there are some important differences:

Difference Motivation/Implications
Symbols are unquoted in C but quoted in HDF. This allows the application to manipulate symbol names in ways that are not possible with C.
The C compiler automatically replaces all symbols with their integer values but HDF requires explicit calls to do the same. C resolves symbols at compile time while HDF resolves symbols at run time.
The mapping from symbols to integers is N:1 in C but 1:1 in HDF. HDF can translate from value to name uniquely and large switch statements are not necessary to print values in human-readable format.
A symbol must appear in only one C enum type but may appear in multiple HDF enumeration types. The translation from symbol to value in HDF requires the datatype to be specified while in C the datatype is not necessary because it can be inferred from the symbol.
The underlying integer value is always a native integer in C but can be a foreign integer type in HDF. This allows HDF to describe data that might reside on a foreign architecture, such as data stored in a file.
The sign and size of the underlying integer datatype is chosen automatically by the C compiler but must be fully specified with HDF. Since HDF doesn't require finalization of a datatype, complete specification of the type must be supplied before the type is used. Requiring that information at the time of type creation was a design decision to simplify the library.

The examples below use the following C datatypes:

/* English color names */
typedef enum {
} c_en_colors;

/* Spanish color names, reverse order */
typedef enum {
} c_sp_colors;

/* No enum definition for French names */

Creating HDF Types from C Types

An HDF enumeration datatype can be created from a C enum type simply by passing pointers to the C enum values to H5Tenum_insert(). For instance, to create HDF types for the c_en_colors type shown above:

c_en_colors val;
hid_t hdf_en_colors = H5Tcreate(H5T_ENUM, sizeof(c_en_colors));
H5Tenum_insert(hdf_en_colors, "RED",   (val=RED,  &val));
H5Tenum_insert(hdf_en_colors, "GREEN", (val=GREEN,&val));
H5Tenum_insert(hdf_en_colors, "BLUE",  (val=BLUE, &val));
H5Tenum_insert(hdf_en_colors, "WHITE", (val=WHITE,&val));
H5Tenum_insert(hdf_en_colors, "BLACK", (val=BLACK,&val));

Name Changes between Applications

Occassionally two applicatons wish to exchange data but they use different names for the constants they exchange. For instance, an English and a Spanish program may want to communicate color names although they use different symbols in the C enum definitions. The communication is still possible although the applications must agree on common terms for the colors. The following example shows the Spanish code to read the values assuming that the applications have agreed that the color information will be exchanged using Enlish color names:

c_sp_colors val, data[1000];
hid_t hdf_sp_colors = H5Tcreate(H5T_ENUM, sizeof(c_sp_colors));
H5Tenum_insert(hdf_sp_colors, "RED",   (val=ROJO,   &val));
H5Tenum_insert(hdf_sp_colors, "GREEN", (val=VERDE,  &val));
H5Tenum_insert(hdf_sp_colors, "BLUE",  (val=AZUL,   &val));
H5Tenum_insert(hdf_sp_colors, "WHITE", (val=BLANCO, &val));
H5Tenum_insert(hdf_sp_colors, "BLACK", (val=NEGRO,  &val));

H5Dread(dataset, hdf_sp_colors, H5S_ALL, H5S_ALL, H5P_DEFAULT, data);

Symbol Ordering across Applications

Since symbol ordering is completely determined by the integer values assigned to each symbol in the enum definition, ordering of enum symbols cannot be preserved across files like with HDF enumeration types. HDF can convert from one application's integer values to the other's so a symbol in one application's C enum gets mapped to the same symbol in the other application's C enum, but the relative order of the symbols is not preserved.

For example, an application may be defined to use the definition of c_en_colors defined above where WHITE is less than BLACK, but some other application might define the colors in some other order. If each application defines an HDF enumeration type based on that application's C enum type then HDF will modify the integer values as data is communicated from one application to the other so that a RED value in the first application is also a RED value in the other application.

A case of this reordering of symbol names was also shown in the previous code snippet (as well as a change of language), where HDF changed the integer values so 0 (RED) in the input file became 4 (ROJO) in the data array. In the input file, WHITE was less than BLACK; in the application the opposite is true.

In fact, the ability to change the order of symbols is often convenient when the enumeration type is used only to group related symbols that don't have any well defined order relationship.


The HDF enumeration type conversion features can also be used to provide internationalization of debugging output. A program written with the c_en_colors datatype could define a separate HDF datatype for languages such as English, Spanish, and French and cast the enumerated value to one of these HDF types to print the result.

c_en_colors val, *data=...;

hid_t hdf_sp_colors = H5Tcreate(H5T_ENUM, sizeof val);
H5Tenum_insert(hdf_sp_colors, "ROJO",   (val=RED,   &val));
H5Tenum_insert(hdf_sp_colors, "VERDE",  (val=GREEN, &val));
H5Tenum_insert(hdf_sp_colors, "AZUL",   (val=BLUE,  &val));
H5Tenum_insert(hdf_sp_colors, "BLANCO", (val=WHITE, &val));
H5Tenum_insert(hdf_sp_colors, "NEGRO",  (val=BLACK, &val));

hid_t hdf_fr_colors = H5Tcreate(H5T_ENUM, sizeof val);
H5Tenum_insert(hdf_fr_colors, "OUGE",  (val=RED,   &val));
H5Tenum_insert(hdf_fr_colors, "VERT",  (val=GREEN, &val));
H5Tenum_insert(hdf_fr_colors, "BLEU",  (val=BLUE,  &val));
H5Tenum_insert(hdf_fr_colors, "BLANC", (val=WHITE, &val));
H5Tenum_insert(hdf_fr_colors, "NOIR",  (val=BLACK, &val));

nameof(lang_t language, c_en_colors val, char *name, size_t size)
    switch (language) {
    case ENGLISH:
        H5Tenum_nameof(hdf_en_colors, &val, name, size);
    case SPANISH:
        H5Tenum_nameof(hdf_sp_colors, &val, name, size);
    case FRENCH:
        H5Tenum_nameof(hdf_fr_colors, &val, name, size);

8.10. Goals That Have Been Met

The main goal of enumeration types is to provide communication of enumerated data using symbolic equivalence. That is, a symbol written to a dataset by one application should be read as the same symbol by some other application.

Architecture Independence Two applications shall be able to exchange enumerated data even when the underlying integer values have different storage formats. HDF accomplishes this for enumeration types by building them upon integer types.
Preservation of Order Relationship The relative order of symbols shall be preserved between two applications that use equivalent enumeration datatypes. Unlike numeric values that have an implicit ordering, enumerated data has an explicit order defined by the enumeration datatype and HDF records this order in the file.
Order Independence An application shall be able to change the relative ordering of the symbols in an enumeration datatype. This is accomplished by defining a new type with different integer values and converting data from one type to the other.
Subsets An application shall be able to read enumerated data from an archived dataset even after the application has defined additional members for the enumeration type. An application shall be able to write to a dataset when the dataset contains a superset of the members defined by the application. Similar rules apply for in-core conversions between enumerated datatypes.
Targetable An application shall be able to target a particular architecture or application when storing enumerated data. This is accomplished by allowing non-native underlying integer types and converting the native data to non-native data.
Efficient Data Transfer An application that defines a file dataset that corresponds to some native C enumerated data array shall be able to read and write to that dataset directly using only Posix read and write functions. HDF already optimizes this case for integers, so the same optimization will apply to enumerated data.
Efficient Storage Enumerated data shall be stored in a manner which is space efficient. HDF stores the enumerated data as integers and allows the application to chose the size and format of those integers.

9. Variable-length Datatypes

9.1. Overview And Justification

Variable-length (VL) datatypes are sequences of an existing datatype (atomic, VL, or compound) which are not fixed in length from one dataset location to another. In essence, they are similar to C character strings -- a sequence of a type which is pointed to by a particular type of pointer -- although they are implemented more closely to FORTRAN strings by including an explicit length in the pointer instead of using a particular value to terminate the sequence.

VL datatypes are useful to the scientific community in many different ways, some of which are listed below:

9.2. Variable-length Datatype Memory Management

With each element possibly being of different sequence lengths for a dataset with a VL datatype, the memory for the VL datatype must be dynamically allocated. Currently there are two methods of managing the memory for VL datatypes: the standard C malloc/free memory allocation routines or a method of calling user-defined memory management routines to allocate or free memory. Since the memory allocated when reading (or writing) may be complicated to release, an HDF5 routine is provided to traverse a memory buffer and free the VL datatype information without leaking memory.

Variable-length datatypes cannot be divided

VL datatypes are designed so that they cannot be subdivided by the library with selections, etc. This design was chosen due to the complexities in specifying selections on each VL element of a dataset through a selection API that is easy to understand. Also, the selection APIs work on dataspaces, not on datatypes. At some point in time, we may want to create a way for dataspaces to have VL components to them and we would need to allow selections of those VL regions, but that is beyond the scope of this document.

What happens if the library runs out of memory while reading?

It is possible for a call to H5Dread to fail while reading in VL datatype information if the memory required exceeds that which is available. In this case, the H5Dread call will fail gracefully and any VL data which has been allocated prior to the memory shortage will be returned to the system via the memory management routines detailed below. It may be possible to design a partial read API function at a later date, if demand for such a function warrants.

Strings as variable-length datatypes

Since character strings are a special case of VL data that is implemented in many different ways on different machines and in different programming languages, they are handled somewhat differently from other VL datatypes in HDF5.

HDF5 has native VL strings for each language API, which are stored the same way on disk, but are exported through each language API in a natural way for that language. When retrieving VL strings from a dataset, users may choose to have them stored in memory as a native VL string or in HDF5's hvl_t struct for VL datatypes.

VL strings may be created in one of two ways: by creating a VL datatype with a base type of H5T_NATIVE_ASCII, H5T_NATIVE_UNICODE, etc., or by creating a string datatype and setting its length to H5T_VARIABLE. The second method is used to access native VL strings in memory. The library will convert between the two types, but they are stored on disk using different datatypes and have different memory representations.

Multi-byte character representations, such as UNICODE or wide characters in C/C++, will need the appropriate character and string datatypes created so that they can be described properly through the datatype API. Additional conversions between these types and the current ASCII characters will also be required.

Variable-width character strings (which might be compressed data or some other encoding) are not currently handled by this design. We will evaluate how to implement them based on user feedback.

9.3. Variable-length Datatype API


VL datatypes are created with the H5Tvlen_create() function as follows:
type_id = H5Tvlen_create(hid_t base_type_id);

The base datatype will be the datatype that the sequence is composed of, characters for character strings, vertex coordinates for polygon lists, etc. The base datatype specified for the VL datatype can be of any HDF5 datatype, including another VL datatype, a compound datatype, or an atomic datatype.

Query base datatype of VL datatype

It may be necessary to know the base datatype of a VL datatype before memory is allocated, etc. The base datatype is queried with the H5Tget_super() function, described in the H5T documentation.

Query minimum memory required for VL information

It order to predict the memory usage that H5Dread may need to allocate to store VL data while reading the data, the H5Dget_vlen_size() function is provided:
herr_t H5Dget_vlen_buf_size(hid_t dataset_id, hid_t type_id, hid_t space_id, hsize_t *size)
(This function is not implemented in Release 1.2.)

This routine checks the number of bytes required to store the VL data from the dataset, using the space_id for the selection in the dataset on disk and the type_id for the memory representation of the VL data in memory. The *size value is modified according to how many bytes are required to store the VL data in memory.

Specifying how to manage memory for the VL datatype

The memory management method is determined by dataset transfer properties passed into the H5Dread and H5Dwrite functions with the dataset transfer property list.

Default memory management is set by using H5P_DEFAULT for the dataset transfer property list identifier. If H5P_DEFAULT is used with H5Dread, the system malloc and free calls will be used for allocating and freeing memory. In such a case, H5P_DEFAULT should also be passed as the property list identifier to H5Dvlen_reclaim.

The rest of this subsection is relevant only to those who choose not to use default memory management.

The user can choose whether to use the system malloc and free calls or user-defined, or custom, memory management functions. If user-defined memory management functions are to be used, the memory allocation and free routines must be defined via H5Pset_vlen_mem_manager(), as follows:

herr_t H5Pset_vlen_mem_manager(hid_t plist_id, H5MM_allocate_t alloc, void *alloc_info, H5MM_free_t free, void *free_info)

The alloc and free parameters identify the memory management routines to be used. If the user has defined custom memory management routines, alloc and/or free should be set to make those routine calls (i.e., the name of the routine is used as the value of the parameter); if the user prefers to use the system's malloc and/or free, the alloc and free parameters, respectively, should be set to NULL

The prototypes for the user-defined functions would appear as follows:

typedef void *(*H5MM_allocate_t)(size_t size, void *info) ;
typedef void (*H5MM_free_t)(void *mem, void *free_info) ;

The alloc_info and free_info parameters can be used to pass along any required information to the user's memory management routines.

In summary, if the user has defined custom memory management routines, the name(s) of the routines are passed in the alloc and free parameters and the custom routines' parameters are passed in the alloc_info and free_info parameters. If the user wishes to use the system malloc and free functions, the alloc and/or free parameters are set to NULL and the alloc_info and free_info parameters are ignored.

Recovering memory from VL buffers read in

The complex memory buffers created for a VL datatype may be reclaimed with the H5Dvlen_reclaim() function call, as follows:
herr_t H5Dvlen_reclaim(hid_t type_id, hid_t space_id, hid_t plist_id, void *buf);

The type_id must be the datatype stored in the buffer, space_id describes the selection for the memory buffer to free the VL datatypes within, plist_id is the dataset transfer property list which was used for the I/O transfer to create the buffer, and buf is the pointer to the buffer to free the VL memory within. The VL structures (hvl_t) in the user's buffer are modified to zero out the VL information after it has been freed.

If nested VL datatypes were used to create the buffer, this routine frees them from the bottom up, releasing all the memory without creating memory leaks.

9.4. Code Examples

The following example creates the following one-dimensional array of size 4 of variable-length datatype.
          0 10 20 30
            11 21 31
               22 32
Each element of the VL datatype is of H5T_NATIVE_UINT type.

The array is stored in the dataset and then read back into memory. Default memory management routines are used for writing the VL data. Custom memory management routines are used for reading the VL data and reclaiming memory space.

Example: Variable-length Datatypes

#include <hdf5.h>

#define FILE   "vltypes.h5"
#define MAX(X,Y)        ((X)>(Y)?(X):(Y))

/* 1-D dataset with fixed dimensions */
#define SPACE_NAME  "Space"
#define SPACE_RANK	1
#define SPACE_DIM	4

void *vltypes_alloc_custom(size_t size, void *info);
void vltypes_free_custom(void *mem, void *info);

**  vltypes_alloc_custom():  VL datatype custom memory
**      allocation routine.  This routine just uses malloc to
**      allocate the memory and increments the amount of memory
**      allocated.
void *vltypes_alloc_custom(size_t size, void *info)

    void *ret_value=NULL;       /* Pointer to return */
    int *mem_used=(int *)info;  /* Get the pointer to the memory used */
    size_t extra;               /* Extra space needed */

     *  This weird contortion is required on the DEC Alpha to keep the
     *  alignment correct.
    extra=MAX(sizeof(void *),sizeof(int));

    if((ret_value=(void *)malloc(extra+size))!=NULL) {
        *(int *)ret_value=size;
    } /* end if */
    ret_value=((unsigned char *)ret_value)+extra;
**  vltypes_free_custom(): VL datatype custom memory
**      allocation routine.  This routine just uses free to
**      release the memory and decrements the amount of memory
**      allocated.
** ****************************************************************/
void vltypes_free_custom(void *_mem, void *info)

    unsigned char *mem;
    int *mem_used=(int *)info;  /* Get the pointer to the memory used */
    size_t extra;               /* Extra space needed */    
     *  This weird contortion is required on the DEC Alpha to keep the
     *  alignment correct.      
    extra=MAX(sizeof(void *),sizeof(int));
    if(_mem!=NULL) {        
        mem=((unsigned char *)_mem)-extra;
       *mem_used-=*(int *)mem; 
    } /* end if */

int main(void)

    hvl_t wdata[SPACE_DIM];   /* Information to write */
    hvl_t rdata[SPACE_DIM];   /* Information read in */
    hid_t		fid;	   /* HDF5 File IDs */  
    hid_t		dataset;   /* Dataset ID */
    hid_t		sid;       /* Dataspace ID */
    hid_t		tid;       /* Datatype ID	   	 */
    hid_t       xfer_pid;   /* Dataset transfer property list ID */
    hsize_t		dims[] = {SPACE_DIM};
    uint       i,j;        /* counting variables */
    int         mem_used=0; /* Memory used during allocation */
    herr_t		ret;		/* Generic return value	 */

     * Allocate and initialize VL data to write 
    for(i=0; i<SPACE_DIM; i++) {

        wdata[i].p= (unsigned int *)malloc((i+1)*sizeof(unsigned int));
        for(j=0; j<(i+1); j++)
            ((unsigned int *)wdata[i].p)[j]=i*10+j;
    } /* end for */

     * Create file. 

     * Create dataspace for datasets. 
    sid = H5Screate_simple(SPACE_RANK, dims, NULL);

     * Create a datatype to refer to. 
    tid = H5Tvlen_create (H5T_NATIVE_UINT);

     * Create a dataset. 
    dataset=H5Dcreate(fid, "Dataset", tid, sid, H5P_DEFAULT);

     * Write dataset to disk. 
    ret=H5Dwrite(dataset, tid, H5S_ALL, H5S_ALL, H5P_DEFAULT, wdata);

     * Change to the custom memory allocation routines for reading 
     * VL data 

    ret=H5Pset_vlen_mem_manager(xfer_pid, vltypes_alloc_custom,
                                &mem_used, vltypes_free_custom, 

     * Read dataset from disk. vltypes_alloc_custom and
     * will be used to manage memory.
    ret=H5Dread(dataset, tid, H5S_ALL, H5S_ALL, xfer_pid, rdata);   

     * Display data read in 
    for(i=0; i<SPACE_DIM; i++) {
        printf("%d-th element length is %d \n", i, 
                                   (unsigned) rdata[i].len);
        for(j=0; j<rdata[i].len; j++) {
            printf(" %d ",((unsigned int *)rdata[i].p)[j] );   
    } /* end for */

     * Reclaim the read VL data. vltypes_free_custom will be used 
     * to reclaim the space. 
    ret=H5Dvlen_reclaim(tid, sid, xfer_pid, rdata);

     * Reclaim the write VL data.  C language free function will be 
     * used to reclaim space. 
    ret=H5Dvlen_reclaim(tid, sid, H5P_DEFAULT, wdata);

     * Close Dataset 
    ret = H5Dclose(dataset);

     * Close datatype 
    ret = H5Tclose(tid);

     * Close disk dataspace 
    ret = H5Sclose(sid);
     * Close dataset transfer property list 
    ret = H5Pclose(xfer_pid);
     * Close file 
    ret = H5Fclose(fid);

And the output from this sample code would be as follows:

Example: Variable-length Datatypes, Sample Output

0-th element length is 1 
1-th element length is 2 
10  11 
2-th element length is 3 
20  21  22 
3-th element length is 4 
30  31  32  33 

For further samples of VL datatype code, see the tests in test/tvltypes.c in the HDF5 distribution.

10. Array Datatypes

The array class of datatypes, H5T_ARRAY, allows the construction of true, homogeneous, multi-dimensional arrays. Since these are homogeneous arrays, each element of the array will be of the same datatype, designated at the time the array is created.

Arrays can be nested. Not only is an array datatype used as an element of an HDF5 dataset, but the elements of an array datatype may be of any datatype, including another array datatype.

Array datatypes cannot be subdivided for I/O; the entire array must be transferred from one dataset to another.

Within the limitations outlined in the next paragraph, array datatypes may be N-dimensional and of any dimension size. Unlimited dimensions, however, are not supported. Functionality similar to unlimited dimension arrays is available through the use of variable-length datatypes.

The maximum number of dimensions, i.e., the maximum rank, of an array datatype is specified by the HDF5 library constant H5S_MAX_RANK. The minimum rank is 1 (one). All dimension sizes must be greater than 0 (zero).

One array dataype may only be converted to another array datatype if the number of dimensions and the sizes of the dimensions are equal and the datatype of the first array's elements can be converted to the datatype of the second array's elements.

10.1 Array Datatype APIs

The functions for creating and manipulating array datadypes are as follows:
H5Tarray_create    Creates an array datatype.
hid_t H5Tarray_create( hid_t base, int rank, const hsize_t dims[/*rank*/], const int perm[/*rank*/] )
H5Tget_array_ndims    Retrieves the rank of the array datatype.
int H5Tget_array_ndims( hid_t adtype_id )
H5Tget_array_dims    Retrieves the dimension sizes of the array datatype.
int H5Tget_array_dims( hid_t adtype_id, hsize_t *dims[], int *perm[] )

10.2 Transition Issues in Adapting Existing Software
       (Transition to HDF5 Release 1.4 Only)

The array datatype class is new with Release 1.4; prior releases included an array element for compound datatypes.

The use of the array datatype class will not interfere with the use of existing compound datatypes. Applications may continue to read and write the older field arrays, but they will no longer be able to create array fields in newly-defined compound datatypes.

Existing array fields will be transparently mapped to array datatypes when they are read in.

10.3 Code Example

The following example creates an array datatype and a dataset containing elements of the array datatype in an HDF5 file. It then writes the dataset to the file.

Example: Array Datatype

#include <hdf5.h>

#define FILE        "SDS_array_type.h5"
#define DATASETNAME "IntArray" 
#define ARRAY_DIM1     5                      /* array dimensions and rank */
#define ARRAY_DIM2     4 
#define ARRAY_RANK     2 
#define SPACE_DIM     10                      /* dataset dimensions and rank */ 
#define RANK  1 

main (void)
    hid_t       file, dataset;         /* file and dataset handles */
    hid_t       datatype, dataspace;   /* handles */
    hsize_t     sdims[] = {SPACE_DIM};              /* dataset dimensions */
    hsize_t     adims[] = {ARRAY_DIM1, ARRAY_DIM2}; /* array dimensions */
    hsize_t     adims_out[2]; 
    herr_t      status;                             
    int         data[SPACE_DIM][ARRAY_DIM1][ARRAY_DIM2];   /* data to write */
    int         k, i, j;
    int         array_rank_out; 

     * Data  and output buffer initialization. 
    for (k = 0; k < SPACE_DIM; k++) {
      for (j = 0; j < ARRAY_DIM1; j++) {
	for (i = 0; i < ARRAY_DIM2; i++)
               data[k][j][i] = k;
     * Create a new file using H5F_ACC_TRUNC access,
     * default file creation properties, and default file
     * access properties.
    file = H5Fcreate(FILE, H5F_ACC_TRUNC, H5P_DEFAULT, H5P_DEFAULT);

     * Describe the size of the array and create the data space for fixed
     * size dataset. 
    dataspace = H5Screate_simple(RANK, sdims, NULL); 

     * Define array datatype for the data in the file.
    datatype = H5Tarray_create(H5T_NATIVE_INT, ARRAY_RANK, adims, NULL);

     * Create a new dataset within the file using defined dataspace and
     * datatype and default dataset creation properties.
    dataset = H5Dcreate(file, DATASETNAME, datatype, dataspace,

     * Write the data to the dataset using default transfer properties.
    status = H5Dwrite(dataset, datatype, H5S_ALL, H5S_ALL,
		      H5P_DEFAULT, data);

     * Close/release resources.
     * Reopen dataset, and return information about its datatype.
    dataset = H5Dopen(file, DATASETNAME);
    datatype = H5Dget_type(dataset);
    array_rank_out = H5Tget_array_ndims(datatype);
    status = H5Tget_array_dims(datatype, adims_out, NULL); 
    printf(" Array datatype rank is %d \n", array_rank_out);
    printf(" Array dimensions are %d x %d \n", (int)adims_out[0],   

    return 0;

11. Sharing Datatypes among Datasets

If a file has lots of datasets which have a common datatype, then the file could be made smaller by having all the datasets share a single datatype. Instead of storing a copy of the datatype in each dataset object header, a single datatype is stored and the object headers point to it. The space savings is probably only significant for datasets with a compound datatype, since the atomic datatypes can be described with just a few bytes anyway.

To create a bunch of datasets that share a single datatype just create the datasets with a committed (named) datatype.

Example: Shared Datatypes

To create two datasets that share a common datatype one just commits the datatype, giving it a name, and then uses that datatype to create the datasets.

hid_t t1 = ...some transient type...;
H5Tcommit (file, "shared_type", t1);
hid_t dset1 = H5Dcreate (file, "dset1", t1, space, H5P_DEFAULT);
hid_t dset2 = H5Dcreate (file, "dset2", t1, space, H5P_DEFAULT);

And to create two additional datasets later which share the same type as the first two datasets:

hid_t dset1 = H5Dopen (file, "dset1");
hid_t t2 = H5Dget_type (dset1);
hid_t dset3 = H5Dcreate (file, "dset3", t2, space, H5P_DEFAULT);
hid_t dset4 = H5Dcreate (file, "dset4", t2, space, H5P_DEFAULT);

12. Data Conversion

The library is capable of converting data from one type to another and does so automatically when reading or writing the raw data of a dataset, attribute data, or fill values. The application can also change the type of data stored in an array.

In order to insure that data conversion exceeds disk I/O rates, common data conversion paths can be hand-tuned and optimized for performance. The library contains very efficient code for conversions between most native datatypes and a few non-native datatypes, but if a hand-tuned conversion function is not available, then the library falls back to a slower but more general conversion function. The application programmer can define additional conversion functions when the libraries repertoire is insufficient. In fact, if an application does define a conversion function which would be of general interest, we request that the function be submitted to the HDF5 development team for inclusion in the library.

Note: The HDF5 library contains a deliberately limited set of conversion routines. It can convert from one integer format to another, from one floating point format to another, and from one struct to another. It can also perform byte swapping when the source and destination types are otherwise the same. The library does not contain any functions for converting data between integer and floating point formats. It is anticipated that some users will find it necessary to develop float to integer or integer to float conversion functions at the application level; users are invited to submit those functions to be considered for inclusion in future versions of the library.

A conversion path contains a source and destination datatype and each path contains a hard conversion function and/or a soft conversion function. The only difference between hard and soft functions is the way in which the library chooses which function applies: A hard function applies to a specific conversion path while a soft function may apply to multiple paths. When both hard and soft functions apply to a conversion path, then the hard function is favored and when multiple soft functions apply, the one defined last is favored.

A data conversion function is of type H5T_conv_t, which is defined as follows:

typedef herr_t (*H5T_conv_t) (hid_t src_id, 
                              hid_t dst_id, 
                              H5T_cdata_t *cdata,
                              hsize_t nelmts, 
                              size_t buf_stride, 
                              size_t bkg_stride, 
                              void *buffer, 
                              void *bkg_buffer,
                              hid_t dset_xfer_plist);

The conversion function is called with the source and destination datatypes (src_id and dst_id), the path-constant data struct (cdata), the number of instances of the datatype to convert (nelmts), a conversion buffer (buffer) which initially contains an array of data having the source type and on return will contain an array of data having the destination type, a temporary or background buffer (bkg_buffer, see description of H5T_BKG_YES below), conversion and background buffer strides (buf_stride and bkg_stride) that indicate what data is to be converted, and a dataset transfer properties list (dset_xfer_plist).

buf_stride and bkg_stride are in bytes and are related to the size of the datatype. If every data element is to be converted, the parameter's value is equal to the size of the datatype; if every other data element is to be converted, the parameter's value is equal to twice the size of the datatype; etc.

dset_xfer_plist may contain properties that are passed to the read and write calls. This parameter is currently used only with variable-length data.

bkg_buffer and bkg_stride are used only with compound datatypes.

The path-constant data struct, H5T_cdata_t, is declared as follows:

typedef struct *H5T_cdata_t (H5T_cmd_t command, 
                             H5T_bkg_t need_bkg, 
                             hbool_t *recalc,
                             void *priv)

The command field of the cdata argument determines what happens within the conversion function. It's values can be:

This command is to hard conversion functions when they're registered or soft conversion functions when the library is determining if a conversion can be used for a particular path. The src_type and dst_type are the end-points of the path being queried and cdata is all zero. The library should examine the source and destination types and return zero if the conversion is possible and negative otherwise (hard conversions need not do this since they've presumably been registered only on paths they support). If the conversion is possible the library may allocate and initialize private data and assign the pointer to the priv field of cdata (or private data can be initialized later). It should also initialize the need_bkg field described below. The buf and background pointers will be null pointers.

This command indicates that data points should be converted. The conversion function should initialize the priv field of cdata if it wasn't initialize during the H5T_CONV_INIT command and then convert nelmts instances of the src_type to the dst_type. The buffer serves as both input and output. The background buffer is supplied according to the value of the need_bkg field of cdata (the values are described below).

The conversion function is about to be removed from some path and the private data (the cdata->priv pointer) should be freed and set to null. All other pointer arguments are null, the src_type and dst_type are invalid (negative), and the nelmts argument is zero.

Other commands might be implemented later and conversion functions that don't support those commands should return a negative value.

Whether a background buffer is supplied to a conversion function, and whether the background buffer is initialized depends on the value of cdata->need_bkg which the conversion function should have initialized during the H5T_CONV_INIT command. It can have one of these values:

No background buffer will be supplied to the conversion function. This is the default.

A background buffer will be supplied but it will not be initialized. This is useful for those functions requiring some extra buffer space as the buffer can probably be allocated more efficiently by the library (the application can supply the buffer as part of the dataset transfer property list).

An initialized background buffer is passed to the conversion function. The buffer is initialized with the current values of the destination for the data which is passed in through the buffer argument. It can be used to "fill in between the cracks". For instance, if the destination type is a compound datatype and we are initializing only part of the compound datatype from the source type then the background buffer can be used to initialize the other part of the destination.

The recalc field of cdata is set when the conversion path table changes. It can be used by conversion function that cache other conversion paths so they know when their cache needs to be recomputed.

Once a conversion function is written it can be registered and unregistered with these functions:

herr_t H5Tregister(H5T_pers_t pers, const char *name, hid_t src_type, hid_t dest_type, H5T_conv_t func)
Once a conversion function is written, the library must be notified so it can be used. The function can be registered as a hard (H5T_PERS_HARD) or soft (H5T_PERS_SOFT) conversion depending on the value of pers, displacing any previous conversions for all applicable paths. The name is used only for debugging but must be supplied. If pers is H5T_PERS_SOFT then only the type classes of the src_type and dst_type are used. For instance, to register a general soft conversion function that can be applied to any integer to integer conversion one could say: H5Tregister(H5T_PERS_SOFT, "i2i", H5T_NATIVE_INT, H5T_NATIVE_INT, convert_i2i). One special conversion path called the "no-op" conversion path is always defined by the library and used as the conversion function when no data transformation is necessary. The application can redefine this path by specifying a new hard conversion function with a negative value for both the source and destination datatypes, but the library might not call the function under certain circumstances.

herr_t H5Tunregister (H5T_pers_t pers, const char *name, hid_t src_type, hid_t dest_type, H5T_conv_t func)
Any conversion path or function that matches the critera specified by a call to this function is removed from the type conversion table. All fields have the same interpretation as for H5Tregister() with the added feature that any (or all) may be wild cards. The H5T_PERS_DONTCARE constant should be used to indicate a wild card for the pers argument. The wild card name is the null pointer or empty string, the wild card for the src_type and dest_type arguments is any negative value, and the wild card for the func argument is the null pointer. The special no-op conversion path is never removed by this function.

Example: A conversion function

Here's an example application-level function that converts Cray unsigned short to any other 16-bit unsigned big-endian integer. A cray short is a big-endian value which has 32 bits of precision in the high-order bits of a 64-bit word.

 1 typedef struct {
 2     size_t dst_size;
 3     int direction;
 4 } cray_ushort2be_t;
 6 herr_t
 7 cray_ushort2be (hid_t src, hid_t dst,
 8                 H5T_cdata_t *cdata, hsize_t nelmts, 
 9                 size_t buf_str, size_t bkg_str, void *buf, 
10                 const void *background, hid_t plist)
11 {
12     unsigned char *src = (unsigned char *)buf;
13     unsigned char *dst = src;
14     cray_ushort2be_t *priv = NULL;
16     switch (cdata->command) {
17     case H5T_CONV_INIT:
18         /*
19          * We are being queried to see if we handle this
20          * conversion.  We can handle conversion from
21          * Cray unsigned short to any other big-endian
22          * unsigned integer that doesn't have padding.
23          */
24         if (!H5Tequal (src, H5T_CRAY_USHORT) ||
25             H5T_ORDER_BE != H5Tget_order (dst) ||
26             H5T_SGN_NONE != H5Tget_signed (dst) ||
27             8*H5Tget_size (dst) != H5Tget_precision (dst)) {
28             return -1;
29         }
31         /*
32          * Initialize private data.  If the destination size
33          * is larger than the source size, then we must
34          * process the elements from right to left.
35          */
36         cdata->priv = priv = malloc (sizeof(cray_ushort2be_t));
37         priv->dst_size = H5Tget_size (dst);
38         if (priv->dst_size>8) {
39             priv->direction = -1;
40         } else {
41             priv->direction = 1;
42         }
43         break;
45     case H5T_CONV_FREE:
46         /*
47          * Free private data.
48          */
49         free (cdata->priv);
50         cdata->priv = NULL;
51         break;
53     case H5T_CONV_CONV:
54         /*
55          * Convert each element, watch out for overlap src
56          * with dst on the left-most element of the buffer.
57          */
58         priv = (cray_ushort2be_t *)(cdata->priv);
59         if (priv->direction<0) {
60             src += (nelmts - 1) * 8;
61             dst += (nelmts - 1) * dst_size;
62         }
63         for (i=0; i<n; i++) {
64             if (src==dst && dst_size<4) {
65                 for (j=0; j<dst_size; j++) {
66                     dst[j] = src[j+4-dst_size];
67                 }
68             } else {
69                 for (j=0; j<4 && j<dst_size; j++) {
70                     dst[dst_size-(j+1)] = src[3-j];
71                 }
72                 for (j=4; j<dst_size; j++) {
73                     dst[dst_size-(j+1)] = 0;
74                 }
75             }
76             src += 8 * direction;
77             dst += dst_size * direction;
78         }
79         break;
81     default:
82         /*
83          * Unknown command.
84          */
85         return -1;
86     }
87     return 0;
88 }

The background argument is ignored since it's generally not applicable to atomic datatypes.

Example: Soft Registration

The convesion function described in the previous example applies to more than one conversion path. Instead of enumerating all possible paths, we register it as a soft function and allow it to decide which paths it can handle.

H5Tregister(H5T_PERS_SOFT, "cus2be",

This causes it to be consulted for any conversion from an integer type to another integer type. The first argument is just a short identifier which will be printed with the datatype conversion statistics.

NOTE: The idea of a master soft list and being able to query conversion functions for their abilities tries to overcome problems we saw with AIO. Namely, that there was a dichotomy between generic conversions and specific conversions that made it very difficult to write a conversion function that operated on, say, integers of any size and order as long as they don't have zero padding. The AIO mechanism required such a function to be explicitly registered (like H5Tregister_hard()) for each an every possible conversion path whether that conversion path was actually used or not.

HDF5 documents and links 
Introduction to HDF5 
HDF5 Reference Manual 
HDF5 User's Guide for Release 1.6 
And in this document, the HDF5 User's Guide from Release 1.4.5:    
Files   Datasets   Datatypes   Dataspaces   Groups  
References   Attributes   Property Lists   Error Handling  
Filters   Caching   Chunking   Mounting Files  
Performance   Debugging   Environment   DDL  

HDF Help Desk
Describes HDF5 Release 1.4.5, February 2003
Last modified: 2 August 2001