Object-oriented Fortran 2008 HDF5 interface

Simple, robust, thin HDF5 polymorphic read/write interface. Reading or writing {real64,real32,int64,int32} from scalar to 7d is as simple as

use h5fortran

call h5write('foo.h5', '/x', x)

call h5read('bar.h5', '/y', y)

• For NetCDF4 see nc4fortran.
• Designed for easy use as a Meson "subproject" or CMake "ExternalProject / FetchContent" using static or shared linking.

Uses Fortran 2008 submodule for clean template structure. This easy-to-use, thin object-oriented modern Fortran library abstracts away the messy parts of HDF5 so that you can read/write various types/ranks of data with a single command. In distinction from other high-level HDF5 interfaces, h5fortran works to deduplicate code, using polymorphism wherever feasible and extensive test suite.

Polymorphic API with read/write for types int32, int64, real32, real64 with rank:

• scalar (0-D)
• 1-D .. 7-D

as well as character (string) variables and attributes. Mismatched datatypes are coerced as per standard Fortran rules. For example, reading a float HDF5 variable into an integer Fortran variable: 42.3 => 42

Tested on systems with HDF5 1.8, 1.10 and 1.12 including:

• MacOS (homebrew)
• Ubuntu 16.04 / 18.04 (gfortran 6 or newer)
• Windows Subsystem for Linux
• Windows MSYS2

Currently, Cygwin does not have Fortran HDF5 libraries.

Build

Requirements:

• modern Fortran compiler (this project uses submodule and error stop). For example, Gfortran ≥ 6.
• HDF5 Fortran library (>= 1.8.7, including 1.10.x and 1.12.x)
  • Mac / Homebrew: brew install gcc hdf5
  • Linux: apt install gfortran libhdf5-dev
  • Windows Subsystem for Linux: apt install gfortran libhdf5-dev
  • Windows MSYS2: pacman -S mingw-w64-x86_64-hdf5

Note that some precompiled HDF5 libraries include C / C++ without Fortran. Platforms that currently do not have Fortran HDF5 libraries, and thus will not work with h5fortran unless you compile HDF5 library for Fortran include:

• Cygwin
• Conda

Build this HDF5 OO Fortran interface with Meson or CMake. The library libh5fortran is built, link it into your program as usual.

Meson

To build h5fortran as a standalone project

meson build

meson test -C build

Meson ≥ 0.53.0 has enhanced HDF5 dependency finding and is recommended. To include h5fortran as a Meson subproject, in the master project meson.build (that uses h5fortran) have like:

hdf5_proj = subproject('h5fortran')
hdf5_interface = hdf5_proj.get_variable('hdf5_interface')

my_exe = executable('myexe', 'main.f90', dependencies: hdf5_interface)

and have a file in the master project subprojects/h5fortran.wrap containing:

[wrap-git]
directory = h5fortran
url = https://github.com/scivision/h5fortran.git
revision = head

CMake

cmake -B build

cmake --build build --parallel

Optionally run self-tests:

cd build

ctest -V

To specify a particular HDF5 library, use

cmake -DHDF5_ROOT=/path/to/hdf5lib -B build

or set environment variable HDF5_ROOT=/path/to/hdf5lib

To use CMake target h5fortran via CMake FetchContent:

include(FetchContent)

FetchContent_Declare(h5fortran_proj
  GIT_REPOSITORY https://github.com/scivision/h5fortran.git
  GIT_TAG master  # whatever desired version is
)

FetchContent_MakeAvailable(h5fortran_proj)

# ------------------------------------------------------
# whatever your program is
add_executable(myProj main.f90)
target_link_libraries(myProj h5fortran::h5fortran)

Usage

All examples assume:

use h5fortran, only: hdf5_file
type(hdf5_file) :: h5f

• gzip compression may be applied for rank ≥ 2 arrays by setting comp_lvl to a value between 1 and 9. Shuffle filter is automatically applied for better compression
• string attributes may be applied to any variable at time of writing or later.
• chunk_size and comp_lvl options must be set to enable compression

integer, intent(out) :: ierr is a mandatory parameter. It will be non-zero if error detected. This value should be checked, particularly for write operations to avoid missing error conditions. The design choice to keep error stop out of h5fortran was in line with the HDF5 library itself. Major Fortran libraries like MPI also make this design choice, perhaps since Fortran doesn't currently have exception handling.

Create new HDF5 file, with variable "value1"

call h5f%initialize('test.h5', ierr, status='new',action='w')

call h5f%write('/value1', 123., ierr)

call h5f%finalize(ierr)

Add/append variable "value1" to existing HDF5 file "test.h5"

• if file test.h5 exists, add a variable to it
• if file test.h5 does not exist, create it and add a variable to it.

call h5f%initialize('test.h5', ierr, status='unknown',action='rw')

call h5f%write('/value1', 123., ierr)

call h5f%finalize(ierr)

Add gzip compressed 3-D array "value2" to existing HDF5 file "test.h5"

real :: val2(1000,1000,3) = 0.

call h5f%initialize('test.h5', ierr, comp_lvl=1)

call h5f%write('/value2', val2, ierr)

call h5f%finalize(ierr)

chunk_size may optionally be set in the %write() method.

compression and chunking are disabled if:

• any element of chunk_size is less than 1
• chunk_size is not given in the initialize() call AND not specified in %write()

check if a variable exists

the logical method %exists() checks if a dataset (variable) exists in the initialized HDF5 file.

exists = h5f%exists("/foo", ierr)

Read scalar, 3-D array of unknown size

call h5f%initialize('test.h5', ierr, status='old',action='r')

integer(hsize_t), allocatable :: dims(:)
real, allocatable :: A(:,:,:)

call h5f%shape('/foo',dims, ierr)
allocate(A(dims(1), dims(2), dims(3)))
call h5f%read('/foo', A)

call h5f%finalize(ierr)

is dataset contiguous or chunked?

Assumed file handle h5f was already initialized, the logical status is inspected:

is_contig = h5f%is_contig('/foo', ierr)

is_chunked = h5f%is_chunked('/foo', ierr)

Create group "scope"

real :: val2(1000,1000,3) = 0.

call h5f%initialize('test.h5', ierr)

call h5f%write('/scope/', ierr)

call h5f%finalize(ierr)

Permissive syntax

We make the hdf5%open(..., status=...) like Fortran open()

• overwrite (truncate) existing file: open with status='new' or status='replace'
• append to existing file or create file: status='old' or status='unknown'

Note the trailing / on /scope/, that tells the API you are creating a group instead of a variable.

Notes

• The first character of the filename should be a character, NOT whitespace to avoid file open/creation errors.
• Using compilers like PGI or Flang may require first compiling the HDF5 library yourself.
• Intel compiler HDF5 compile notes
• Polymorphic array rank is implemented by explicit code internally. We could have used pointers, but the code is simple enough to avoid the risk associated with explicit array pointers. Also, select rank support requires Gfortran-10 or Intel Fortran 2020, so we didn't want to make too-new compiler restriction.

Missing datatypes

• arrays of rank > 7: this has been stubbed in reader_nd.f90, writer_nd.f90. Only the latest compilers support Fortran 2008 arrays up to rank 15.

The datatypes below are more complex to handle and may see little use due to their downsides.

• complex64/complex128: this is not natively handled in HDF5. There are performance impacts for compound datatypes, thus many choose to just write two datasets, one each for real and imaginary like foo_r and foo_i
• non-default character kind