pypi / h5py

h5py

What changed in h5py 3, from its official release notes, upgrade guides and advisories. Sources checked 2026-10-30.

21 breaking12 new APIs and options3 performance3 behavior and defaults7 fixes and other

See which of these apply to your code, ranked, in your agent:

npx @unstale/cli init

@unstale/cli 0.2.9 on npm

h5py 346 changes

Breaking changes 21

The deprecated property was removed; use ds[()] to read all data from any dataset.
3.0.0upgrade guide
The broken inherited method was removed because the dimensions interface behaves as a sequence, not a mapping.
3.0.0upgrade guide
The broken inherited method was removed because the dimensions interface behaves as a sequence, not a mapping.
3.0.0upgrade guide
The default mode for opening files is now read-only; specify another mode if you need to write.
3.0.0upgrade guide
The deprecated attribute was removed; use File.id instead.
3.0.0upgrade guide
The deprecated function was removed; see the newer APIs for special types.
3.0.0upgrade guide
The deprecated function was removed; see the newer APIs for special types.
3.0.0upgrade guide
Creating a dataset from a list of bytes objects now uses a variable-length string datatype; use h5py.string_dtype() for fixed-length strings.
3.0.0upgrade guide
The external argument accepts any iterable as its top-level container, bytes or os.PathLike file names, and NumPy integer offsets and sizes; some previous forms are no longer accepted.
3.0.0upgrade guide
Its third argument is now an optional high-level Dataset object rather than a DatasetID.
3.0.0upgrade guide
The deprecated dictionary was removed.
3.0.0upgrade guide
The deprecated module was removed; the high-level API is available directly in the h5py module.
3.0.0upgrade guide
MPI mode no longer supports mpi4py 1.x.
3.0.0upgrade guide
The deprecated function was removed; see the newer APIs for special types.
3.0.0upgrade guide
The deprecated function was removed; see the newer APIs for special types.
3.0.0upgrade guide
h5py now requires Python 3.6 or above and is no longer compatible with Python 2.7.
3.0.0upgrade guide
The command was removed; build configuration can be specified with environment variables instead.
3.0.0upgrade guide
The interface and rules for storing and reading strings have changed and may require changes to existing h5py code.
3.0.0upgrade guide
Variable-length UTF-8 strings in datasets are now read as bytes by default instead of str.
3.0.0upgrade guide
Relative paths are now used for virtual data sources when the source dataset is in the same file as the virtual dataset.
3.0.0upgrade guide
A dtype must now be specified when making a virtual dataset; there is no longer a default dtype.
3.0.0upgrade guide

New APIs and options 12

Reads only selected fields from a dataset with a compound datatype.
3.0.0upgrade guide
Iterates over chunks within the given selection.
3.0.0upgrade guide
Provides the size of a dataset’s data in bytes, unlike size, which gives the number of elements.
3.0.0upgrade guide
File creation now supports setting the file space strategy, including an option to persist empty space tracking between sessions.
3.0.0upgrade guide
The low-level chunk query API provides information about dataset chunks in an HDF5 file.
3.0.0upgrade guide
The low-level chunk query API provides information about dataset chunks in an HDF5 file.
3.0.0upgrade guide
The low-level chunk query API provides information about dataset chunks in an HDF5 file.
3.0.0upgrade guide
Adds allow_unknown_filter for data that will be compressed before writing with the low-level write_direct_chunk() method.
3.0.0upgrade guide
Use a suitable dtype from this API to store fixed-length strings.
3.0.0upgrade guide
Can now be stored and read as HDF5 opaque data, though other tools will not understand them.
3.0.0upgrade guide
Methods accepting a shape tuple, such as dataset creation, now also accept an integer for a 1D shape.
3.0.0upgrade guide
h5py now supports the split file driver.
3.0.0upgrade guide

Performance 3

Selection is implemented in Cython, reducing reading overhead; many small reads from the same dataset can be as much as 10 times faster.
3.0.0upgrade guide
Reading and writing release the GIL so another Python thread can continue during HDF5 access; HDF5’s global lock still prevents faster parallel data access with multithreading.
3.0.0upgrade guide
Writing is more efficient when assigning a scalar to a chunked dataset if the number of elements written is no more than one chunk.
3.0.0upgrade guide

Behavior and default changes 3

Casting data to a specified type on reading can now be used without a with statement.
3.0.0upgrade guide
Now works for any file driver that supports it, not only the sec2 driver.
3.0.0upgrade guide
Virtual datasets can now grow as their source data is resized.
3.0.0upgrade guide

Fixes and other changes 7

obj.attrs can now work on HDF5 stored datatypes.
3.0.0upgrade guide
The property now handles SWMR access modes in addition to plain RDONLY and RDWR modes.
3.0.0upgrade guide
The property is now always available.
3.0.0upgrade guide
Fixed a bug when the source is a high-level object and the destination is a Group.
3.0.0upgrade guide
Now reports storage size of zero bytes without raising an exception.
3.0.0upgrade guide
h5py is now compatible with HDF5 1.12.
3.0.0upgrade guide
When built against HDF5 1.10.6, h5py uses UTF-8 for file names on Windows, allowing all filenames.
3.0.0upgrade guide