This directory predates C++23's std::float16_t. It exists for users on toolchains that don't yet ship <stdfloat>.

The modern path

For C++23 toolchains (libstdc++ 13+, libc++ 18+), use std::float16_t directly. h5cpp already ships an h5::dt_t<std::float16_t> spec in h5cpp/H5Tall.hpp that emits a 16-bit IEEE binary16 layout. See:

examples/datatypes/ — section 5 demonstrates the round-trip
h5cpp/H5Tall.hpp:181 — the spec itself

#include <stdfloat>
#include <h5cpp/all>
 
std::vector<std::float16_t> vec = { /* ... */ };
h5::write(fd, "/half/float16", vec);
auto back = h5::read<std::vector<std::float16_t>>(fd, "/half/float16");

The third-party variants

The two subdirectories here show the same idea using vendored half-float libraries — useful when your toolchain doesn't have <stdfloat> yet:

Variant	Backing type	Source
`christian-rau/`	`half_float::half`	http://half.sourceforge.net/ (header-only)
`open-exr/`	`Imath::half`	https://github.com/AcademySoftwareFoundation/Imath

Both use exactly the same h5cpp specialization pattern as the C++23 native path — specialize h5::impl::detail::hid_t<T, H5Tclose, true, true, hdf5::type> and have the ctor emit a 16-bit IEEE binary16 type id. The wrappers in each subdirectory's utils.hpp are the reference for hooking any third-party half-float library that does not ship its own h5cpp glue.

When to use which

You have…	Use…
GCC 13+, Clang 18+, MSVC with `/std:c++latest`	`examples/datatypes/` section 5 (native `std::float16_t`)
An older toolchain + Christian Rau's `half.hpp`	`examples/half-float/christian-rau/`
OpenEXR / Imath already in your project	`examples/half-float/open-exr/`

For the general "how does h5cpp let me add a custom datatype?" question, the focused tutorial is in examples/datatypes/ — these subdirectories are deliberately narrow.