mirrors/hdf5
2
0
Fork 0
mirror of https://github.com/HDFGroup/hdf5.git synced 2026-09-25 04:09:44 +03:00
Code Issues Packages Projects Releases Wiki Activity
Files
05676d1abe6014326abd342d54dab5fca632b10c
hdf5/release_docs/HISTORY-2.X.md
T
Larry Knox 8f2ee48e8f Add 2.1.0 CHANGELOG to HISTORY-2.X.md and set release default to draft. (#6291) 
* Add 2.1.0 CHANGELOG to HISTORY-2.X.md.
Set release default to draft.

* Updated CHANGELOG.md Executive Summary "Important" section similar to
that in HDF5 2.1.0 release.
2026-03-17 09:27:10 -05:00

85 KiB
Raw Blame History

HDF5 History

============

This file contains development history of the HDF5 2.X.X releases

  1.  Release Information for hdf5-2.1.0

  2.  Release Information for hdf5-2.0.0

[Search on the string '%%%%' for section breaks of each release.]

%%%%2.1.0%%%%

HDF5 version 2.1.0 released on 2026-03-02

🔺 HDF5 Changelog

All notable changes to this project will be documented in this file. This document describes the differences between this release and the previous HDF5 release, platforms tested, and known problems in this release.

For releases prior to version 2.0.0, please see the release.txt file and for more details check the HISTORY*.txt files in the HDF5 source.

🔗 Quick Links

📖 Contents

🔆 Executive Summary: HDF5 Version 2.1.0

Important

  • The format of the GitHub tag for HDF5 releases has been changed to Major.Minor.Patch, consistent with the versioning policy change to follow the Semantic Versioning Specification described in this Wiki page. The previous tag format hdf5_Major_Minor_Patch that was created in addition for HDF5 2.0.0 and this release will not be continued.

Performance Enhancements:

Significant Advancements:

Enhanced Features:

Java Enhancements:

Acknowledgements:

We would like to thank the many HDF5 community members who contributed to this release of HDF5.

⚠️ Breaking Changes

🚀 New Features & Improvements

Configuration

Added a CMake module to locate libaec for SZIP support

A new Findlibaec.cmake CMake module has been added. This module is intended to locate libaec on the system for SZIP support in HDF5 when libaec was built with Autotools instead of CMake. When SZIP support is enabled in HDF5 with the HDF5_ENABLE_SZIP_SUPPORT option, this module will first check for an existing CMake-built libaec and use that if it's available. Otherwise, the module will heuristically search for libaec on the system. If necessary, the module can be hinted toward a particular libaec installation by setting the CMake variable libaec_ROOT to point to a directory. If it is known that a CMake-built libaec installation exists on the system in a non-standard location, the CMake variable libaec_DIR can instead be set to a directory containing a libaec-config.cmake file to cause the module to prefer that libaec installation.

Refactored CMake library export mechanism

The CMake build system now uses separate export targets for different library types: static libraries use ${HDF5_EXPORTED_TARGETS}_static, Java libraries use ${HDF5_EXPORTED_TARGETS}_java, and shared libraries continue using the main ${HDF5_EXPORTED_TARGETS}. The generated hdf5-config.cmake file conditionally includes the appropriate target files (hdf5-targets_static.cmake and hdf5-targets_java.cmake), allowing downstream projects to selectively link against specific library variants. This refactoring improves build system maintainability and applies consistently across all HDF5 components (core, C++, Fortran, high-level APIs, tools, and Java JNI).

Library

Added predefined datatypes for FP6 data

Predefined datatypes have been added for FP6 data in E2M3 and E3M2 formats.

The following new macros have been added:

  • H5T_FLOAT_F6E2M3
  • H5T_FLOAT_F6E3M2

These macros map to IDs of HDF5 datatypes representing a 6-bit floating-point datatype with 1 sign bit and either 2 exponent bits and 3 mantissa bits (E2M3 format) or 3 exponent bits and 2 mantissa bits (E3M2 format).

Note that support for a native FP6 datatype has not been added yet. This means that any datatype conversions to/from the new FP6 datatypes will be emulated in software rather than potentially using specialized hardware instructions. Until support for a native FP6 type is added, an application can avoid datatype conversion performance issues if it is sure that the datatype used for in-memory data buffers matches one of the above floating-point formats. In this case, the application can specify one of the above macros for both the file datatype when creating a dataset or attribute and the memory datatype when performing I/O on the dataset or attribute.

Also note that HDF5 currently has incomplete support for datatype conversions involving non-IEEE floating-point format datatypes. Refer to the 'Known Problems' section for information about datatype conversions with these new datatypes.

Added predefined datatype for FP4 data

A predefined datatype has been added for FP4 data in E2M1 format.

The following new macro has been added:

  • H5T_FLOAT_F4E2M1

This macro maps to the ID of an HDF5 datatype representing a 4-bit floating-point datatype with 1 sign bit, 2 exponent bits and 1 mantissa bit.

Note that support for a native FP4 datatype has not been added yet. This means that any datatype conversions to/from the new FP4 datatype will be emulated in software rather than potentially using specialized hardware instructions. Until support for a native FP4 type is added, an application can avoid datatype conversion performance issues if it is sure that the datatype used for in-memory data buffers matches the above floating-point format. In this case, the application can specify the above macro for both the file datatype when creating a dataset or attribute and the memory datatype when performing I/O on the dataset or attribute.

Also note that HDF5 currently has incomplete support for datatype conversions involving non-IEEE floating-point format datatypes. Refer to the 'Known Problems' section for information about datatype conversions with these new datatypes.

Parallel Library

Fortran Library

C++ Library

Java Library

Tools

High-Level APIs

C Packet Table API

Internal header file

Documentation

🪲 Bug Fixes

Library

Fixed a problem with using filters with variable length datatypes

When using data filters with a dataset with a variable length datatype, the library would not invoke the filters' can_apply and set_local callbacks. This has been addressed and the library will now make those callbacks no matter what the datatype is.

Fixed a potential out of bound read

When a file is corrupted such that an array datatype's size, the number of elements, and the element size are not in agreement, it can trigger an out of bounds read. A check has been added to detect such situation.

Fixed security issue CVE-2025-44904

For unfiltered dataset chunks, the size on disk should be constant for all chunks in a dataset. In some cases the size of each chunk is stored even in this case where it can be inferred from the chunk dimensions and datatype. The code previously assumed this stored size was equal to the inferred size, leading to a mismatch in the expected and actual buffer size. Modified the library to throw an error if the size does not match the expected size.

Fixed security issue CVE-2025-2309

Fixed a critical vulnerability in H5T__bit_copy of the component Type Conversion Logic which could lead to heap-based buffer overflow.

Fixed security issue CVE-2025-2308

Fixed a critical vulnerability in the function H5Z__scaleoffset_decompress_one_byte of the component Scale-Offset Filter which could lead to heap-based buffer overflow.

Fixed a double-free bug in H5D__chunk_copy

Fixed a double-free bug in the internal H5D__chunk_copy() function which occurred when a buffer was re-allocated without updating the original pointer freed later on.

Fixes GitHub issues #6123 #6124 #6125 #6126 #6133

Fixes potential security issues

The get_name API functions allow passing NULL when querying the object name length. However, passing a non-NULL buffer with size == 0 will result in security vulnerability of invalid write. That was because the library wrote a null terminator to the buffer regardless of what the size of the buffer was as long as the buffer was non-NULL. These functions are now fixed to treat (buffer != NULL, size == 0) as a length-only query to eliminate Valgrind error of invalid write.

Fixed a performance issue with chunked dataset I/O

When dataset chunks are unable to be placed in the dataset chunk cache (for example, if a chunk is too large), the library falls back to an alternative approach for I/O on dataset chunks. An issue with the logic in this approach prevented chunked dataset I/O from making use of the library's data sieve buffer I/O optimization functionality. For chunk shapes that are non-contiguous with the memory layout of a buffer, this could result in severely degraded I/O performance, with the worst-case behavior causing I/O to be performed on a single data element at a time.

The data sieve buffer functionality has been extended to cover the case of uncached chunks and will be used as long as the underlying Virtual File Driver supports data sieving.

Java Library

Configuration

Removed force-setting of ZLIB_USE_EXTERNAL and SZIP_USE_EXTERNAL CMake variables to ON

When the CMake variable HDF5_ALLOW_EXTERNAL_SUPPORT is set to GIT or TGZ, the library's build process previously force-set the ZLIB_USE_EXTERNAL and SZIP_USE_EXTERNAL variables to ON. This prevented the ability to independently choose whether zlib and szip are built from system libraries or from external sources. These variables are no longer forced to ON in this case and can be set individually.

Tools

Performance

Fortran API

Added Fortran wrappers for SWMR functionality

Added four new Fortran wrappers that provide direct access to SWMR (Single Writer Multiple Reader) C APIs:

  • h5fstart_swmr_write_f - Enables SWMR writing mode for a file
  • h5dflush_f - Flushes dataset buffers to disk
  • h5pset_append_flush_f - Sets append flush property values including optional callback function
  • h5pget_append_flush_f - Retrieves append flush property values including callback function

High-Level Library

Fortran High-Level APIs

Documentation

F90 APIs

C++ APIs

Testing

✨ Support for new platforms and languages

☑️ Platforms Tested

A table of platforms tested can be seen on the wiki. Current test results are available here.

⛔ Known Problems

  • When performing implicit datatype conversion on specific non-IEEE floating-point format data, HDF5 may improperly convert some data values:

    When performing I/O operations using a non-IEEE floating-point format datatype, HDF5 may improperly convert some data values due to incomplete handling of non-IEEE types. Such types include the following pre-defined datatypes:

    H5T_FLOAT_F8E4M3
    H5T_FLOAT_F8E5M2
    H5T_FLOAT_F6E2M3
    H5T_FLOAT_F6E3M2
    H5T_FLOAT_F4E2M1

    If possible, an application should perform I/O with these datatypes using an in-memory type that matches the specific floating-point format and perform explicit data conversion outside of HDF5, if necessary. Otherwise, read/written values should be verified to be correct.

  • When the library detects and builds in support for the _Float16 datatype, an issue has been observed on at least one MacOS 14 system where the library fails to initialize due to not being able to detect the byte order of the _Float16 type #4310:

    #5: H5Tinit_float.c line 308 in H5T__fix_order(): failed to detect byte order major: Datatype minor: Unable to initialize object

    If this issue is encountered, support for the _Float16 type can be disabled with a configuration option:

    CMake: HDF5_ENABLE_NONSTANDARD_FEATURE_FLOAT16=OFF

  • When HDF5 is compiled with NVHPC versions 23.5 - 23.9 (additional versions may also be applicable) and with -O2 (or higher) and -DNDEBUG, test failures occur in the following tests:

    • H5PLUGIN-filter_plugin
    • H5TEST-flush2
    • H5TEST-testhdf5-base
    • MPI_TEST_t_filters_parallel

    Sporadic failures (even with lower -O levels):

    • Java JUnit-TestH5Pfapl
    • Java JUnit-TestH5D

    Also, NVHPC will fail to compile the test/tselect.c test file with a compiler error of use of undefined value when the optimization level is -O2 or higher.

    This is confirmed to be a bug in the nvc compiler that has been fixed as of 23.11. If you are using an affected version of the NVidia compiler, the work-around is to set the optimization level to -O1.

  • CMake files do not behave correctly with paths containing spaces

    Do not use spaces in paths because the required escaping for handling spaces results in very complex and fragile build files.

  • At present, metadata cache images may not be generated by parallel applications. Parallel applications can read files with metadata cache images, but since this is a collective operation, a deadlock is possible if one or more processes do not participate.

  • The subsetting option in ph5diff currently will fail and should be avoided

    The subsetting option works correctly in serial h5diff.

  • Flang Fortran compilation will fail (last check version 17) due to not yet implemented: (1) derived type argument passed by value (H5VLff.F90), and (2) support for REAL with KIND = 2 in intrinsic SPACING used in testing.

  • Fortran tests HDF5_1_8.F90 and HDF5_F03.F90 will fail with Cray compilers greater than version 16.0 due to a compiler bug. The latest version verified as failing was version 17.0.

  • Several tests currently fail on certain platforms: MPI_TEST-t_bigio fails with spectrum-mpi on ppc64le platforms.

    MPI_TEST-t_subfiling_vfd and MPI_TEST_EXAMPLES-ph5_subfiling fail with cray-mpich on theta and with XL compilers on ppc64le platforms.

  • File space may not be released when overwriting or deleting certain nested variable length or reference types.

Known problems in previous releases can be found in the HISTORY*.txt files in the HDF5 source. Please report any new problems found to help@hdfgroup.org.

%%%%2.0.0%%%%

HDF5 version 2.0.0 released on 2025-11-11

🔺 HDF5 Changelog

All notable changes to this project will be documented in this file. This document describes the differences between this release and the previous HDF5 release, platforms tested, and known problems in this release.

For releases prior to version 2.0.0, please see the release.txt file and for more details check the HISTORY*.txt files in the HDF5 source.

🔗 Quick Links

📖 Contents

🔆 Executive Summary: HDF5 Version 2.0.0

Performance Enhancements:

Significant Advancements:

Updated Foundation:

Important

  • Transitioned to CMake-only builds, and Autotools is no longer in use.
  • Renamed library state variables, notably HDF5_ENABLE_PARALLEL is now HDF5_PROVIDES_PARALLEL, see PR #5716 for more details.
  • The default setting for H5Fset_libver_bounds has been updated to set the lower bound to the HDF5 library version 1.8. This change ensures that users can take advantage of the library's optimal performance and the latest features by default. If users need their files to be compatible with older versions of the HDF5 library, they will need to adjust this lower bound manually.

Enhanced Features:

  • Improved ROS3 VFD capabilities using the aws-c-s3 library.

Java Enhancements:

  • Java FFM bindings generated for the C library if Java 25+ available and JNI option is FALSE.
  • Enhanced Maven artifact deployment with comprehensive multi-platform support (Linux, Windows, macOS x86_64, macOS aarch64).
  • Complete Java examples Maven integration (org.hdfgroup:hdf5-java-examples) with cross-platform CI/CD testing.

Acknowledgements:

We would like to thank the many HDF5 community members who contributed to HDF5 2.0.

⚠️ Breaking Changes

Updated default file format to 1.8

By default, HDF5 will now use the 1.8 file format (H5F_LIBVER_V18). This provides improved performance and space efficiency, particularly with groups and links. However, HDF5 library versions 1.6 and earlier will not be able to read files created with the default settings. The previous behavior can be restored using H5Pset_libver_bounds(fapl_id, H5F_LIBVER_EARLIEST, H5F_LIBVER_LATEST).

Renamed the option: HDF5_ENABLE_Z_LIB_SUPPORT

The option has been renamed to HDF5_ENABLE_ZLIB_SUPPORT to be consistent with the naming of other options. Also, the option defaults to OFF. This requires the user to explicitly enable zlib support when configuring the library.

Autotools support was removed from HDF5

CMake is now the build system available in HDF5 code. Version 3.26 or later is required. See the AutotoolsToCMakeOptions.md file for highlights of the CMake HDF5 install layout and CMake options to use in place of former Autotools options.

Fixed problems with family driver and user block

When using a user block with the family driver, the driver would inappropriately subtract the user block size for each member file when calculating member EOAs. This could cause a failure when an address overflowed the calculated eoa. The driver would also add the user block size when returning the EOF. Modified the family driver to not consider the user block, as it is handled by the H5FD layer. The user block now spans the first X bytes of the family array, for example a 4 KiB user block with 3 KiB member size will take up the entire first member and the first 1 KiB of the second. This may cause compatibility issues with preexisting family files with user blocks, though the way it worked before was inconsistent if it worked at all.

🚀 New Features & Improvements

Configuration

Improved the cross-compile support in the build system

The CMake build system has been improved to better support cross-compiling. This includes the following changes:

  • The CMake option, CMAKE_CROSS_COMPILING, indicates that the library is being cross-compiled. This option, usually in a toolchain file, can be set to ON when cross-compiling.
  • Removed the CMAKE_CROSSCOMPILING_EMULATOR when cross-compiling, CMake will automatically insert it in the command.
  • Added an option HDF5_USE_PREGEN to supply pre-generated files located in HDF5_USE_PREGEN_DIR directory, to bypass running feature detection programs when cross-compiling.
  • Added a variable CROSSCOMPILING_PATH to specify a path to search for programs when cross-compiling. This is useful when the build system needs to run programs that were built for the host system.

Refactored HDF5_BUILD/ENABLE_{feature} variable in hdf5-config.cmake file

The variables used in hdf5-config.cmake to indicate what options were used to build the installed library have been renamed. All HDF5_BUILD/ENABLE_{feature} variables are now HDF5_PROVIDES_{feature}. This more clearly indicates that these variables reflect whether the feature is supported by the installed library, instead of whether the feature is an option that can be changed when building an application with the library.

Created macro EXTERNAL_HDF5_STATUS to convert between the old and new names. The macro is in the config/examples/HDF5SubdirMacros.cmake file and can be copied into a project's CMakeLists.txt file to provide backward compatibility.

CMake minimum version is now 3.26

The minimum version of CMake is now 3.26; this will remove workarounds to handle versions between 3.18 (the previous minimum) and 3.26.

Removed HDF5_ENABLE_THREADS option

The HDF5_ENABLE_THREADS option has been removed, as it no longer functions as a proper build option. The library will always check for thread support and set the internal status variable, HDF5_THREADS_ENABLED. The HDF5_ENABLE_THREADSAFE option is still available to build with thread-safe API calls.

Enhanced Maven repository deployment support

Added comprehensive Maven integration with optimized workflows for Java artifact deployment:

  • New CMake options: HDF5_ENABLE_MAVEN_DEPLOY and HDF5_MAVEN_SNAPSHOT for Maven repository deployment
  • Minimal build presets: Added ci-MinShar-*-Maven* presets for efficient Java-only artifact generation
  • Multi-platform support: Automated generation of platform-specific JARs with classifiers (linux-x86_64, windows-x86_64, macos-x86_64, macos-aarch64)
  • CI/CD integration: Enhanced GitHub Actions workflows (maven-staging.yml, maven-deploy.yml) with cross-platform build matrix
  • Artifact validation: Comprehensive validation framework for Maven artifacts before deployment
  • Deployment targets: Support for GitHub Packages and Maven Central staging repositories
  • Java Examples Maven Integration: Added complete Maven artifact for Java examples (org.hdfgroup:hdf5-java-examples) with cross-platform compatibility
  • Multi-platform testing: Comprehensive CI/CD testing of Java examples across all supported platforms (Linux, Windows, macOS x86_64, macOS aarch64)
  • Native library error handling: Enhanced validation logic for Maven-only environments to properly handle expected native library loading errors
  • Dynamic repository support: Enhanced workflows to use github.repository variable for seamless testing on forks before canonical deployment
  • Fork-based testing: Complete testing framework allowing validation on repository forks (e.g., fork-name/hdf5) before merging to HDFGroup/hdf5
  • Multi-artifact deployment: Enhanced deployment workflow to handle both hdf5-java (platform-specific) and hdf5-java-examples (platform-independent) artifacts
  • Production deployment validation: Successfully resolved HTTP 409 version conflicts through snapshot versioning strategy
  • Deployment status: ✅ Fully validated and production-ready with comprehensive error resolution and testing documentation

Reorganized the files in the config/cmake folder into the config folder structure

The config folder CMake files have been reorganized to make it easier to maintain and add new features. This includes the following changes:

  • The files in the config folder are the macros and templates for the build process.
  • The files in the config/cmake folder are primarily for optional features.
  • The files in the config/install folder are for installation support.
  • The files in the config/flags folder are for compiler support.
  • The remaining folders remain unchanged.

Added CMake configuration options to override compilers in h5cc:
HDF5_H5CC_C_COMPILER for the C compiler
HDF5_H5CC_CXX_COMPILER for the C++ compiler
HDF5_H5CC_Fortran_COMPILER for the Fortran compiler

These default to the currently used compiler, preserving the current behavior. However, they can be overridden by users who need to use a different compiler at runtime, for example, when they build via cache.

Aligned the CMake compiler wrappers with the old Autotools versions

The versions of h5cc, h5fc, h5c++, etc. generated by CMake were missing several options and features from the Autotools counterparts. Some of these options and features have now been implemented in the CMake versions, while some of them have not:

  • The missing --help/-h, -c, -echo, and -shlib/-noshlib options have been implemented.
  • The -prefix option was not implemented, as it didn't appear to function in the Autotools wrappers and is generally covered by pkg-config in the CMake wrappers.
  • A new -nohl option has been added to avoid building and linking against the high-level HDF5 libraries if desired.
  • Similar to the Autotools wrappers, the CMake wrappers now add the HDF5 installation library directory to the rpath of the resulting executable/library by default when linking against shared HDF5 libraries. This behavior can be avoided by specifying the new -norpath option.
  • Parsing of the HDF5_USE_SHLIB environment variable has been added to determine whether to link against shared or static HDF5 libraries. Precedence is still given to the -shlib/-noshlib options.
  • Parsing of the HDF5_PKG_CONFIG_ARGS environment variable has been added to separate pkg-config-specific options from compiler-specific options and prevent conflicts between them.

Several issues were also fixed in the pkg-config files that are generated by CMake.

Added CMAKE_INSTALL_PREFIX to the default plugin path

To help users find their plugins, the default plugin path has been changed to include the CMAKE_INSTALL_PREFIX. Adding the install prefix allows users to skip setting the HDF5_PLUGIN_PATH environment variable when using plugins with the default lib/plugin location.

Converted documentation in the source folder, doc, to doxygen files.

Added configuration option for API concurrency support:

CMake: HDF5_ENABLE_CONCURRENCY (ON/OFF) (Default: OFF)

This option enables support for concurrent multithreaded operation of supported API routines. This option also provides thread-safe execution of all other, non-concurrent operations. The 'concurrency' option thus is a superset of the existing 'threadsafe' option. Both options are currently available, although mutually exclusive. As the 'concurrency' code becomes more stable over time, the 'threadsafe' option may be deprecated in favor of the new 'concurrency' option.

The following API routines support concurrent multithreaded operation:

Added support for MinGW + MSYS2 when building with CMake

We added support for this to the appropriate configure-time checks in CMake. CMake + MinGW + MSYS2 is now tested with the following environments:

  • mingw32
  • mingw64
  • ucrt64
  • clang64

Added CMake build mode flags to the libhdf5.settings file

Flags from the CMake build mode (e.g., optimization) are not a part of CMAKE_<language>_FLAGS and were not exported to the libhdf5.settings file. This has been fixed, and the C, Fortran, and C++ build mode flags are now exported to the file.

This also affects the text output of H5check_version() and the libhdf5.settings string stored in the library (for those who use strings(1), etc. to get build info from the binary).

CMake: Split compiler-specific flags into separate files

The compiler-specific flags have been split into separate files to make it easier to maintain and add new compiler flags. The flags for NVHPC, Intel, GNU and Clang compilers are now in separate files included from the current compiler flags files; HDFCompiler<language>Flags.cmake.

Added support for native zlib-ng compression

Changed the zlib-ng CMake logic to prefer the native zlib-ng library. Added #ifdef around the compression function calls. Added including the correct header file with the same #ifdef.

Renamed HDF5Examples build options from H5EXAMPLE_<option> to H5EXAMPLE_<option>

Changed the prefix to better distinguish the examples build options from the library options when building the examples along with the library.

Renamed remaining HDF5 library CMake options except for CMake BUILD* variables

Old New
DEFAULT_API_VERSION HDF5_DEFAULT_API_VERSION
DISABLE_PDB_FILES HDF5_DISABLE_PDB_FILES
ONLY_SHARED_LIBS HDF5_ONLY_SHARED_LIBS
ALLOW_UNSUPPORTED HDF5_ALLOW_UNSUPPORTED
TEST_SHELL_SCRIPTS HDF5_TEST_SHELL_SCRIPTS

All other HDF5 library CMake options are prefixed with HDF5_

bin/cmakehdf5 has been removed

This was an unsupported build script that made building HDF5 via CMake work like building HDF5 via the Autotools. It has been unmaintained for a long time, has been marked deprecated, and is being removed.

Generated files in src are now checked into version control

These files are infrequently updated, and generating them adds a dependency on Perl. The listed files are now checked in and do not need to be recreated when checking out development branches.

  • H5Edefin.h
  • H5Einit.h
  • H5Emajdef.h
  • H5Emindef.h
  • H5Epubgen.h
  • H5Eterm.h
  • H5overflow.h
  • H5version.h

Dropped some old Solaris Studio work-arounds

Solaris Studio no longer seems to be maintained and the last version (12.4, circa 2015) doesn't seem to fully support C11. We've removed some hacks that work around things like __attribute__() support.

Dropped support for the traditional MSVC preprocessor

Visual Studio has recently started using a standards-compliant preprocessor (In VS2019+) and this is the default in C11. Because of this, we've dropped support for the traditional MSVC preprocessor.

The standard for building the library is now C11. We have updated the build files to set the C standard to C11, though some platforms use gnu11 to get some GNU things to work.

Library

Added support for large chunks

The library now supports chunks larger than 4 GiB using 64 bit addressing. Creating chunks with size >= 4 GiB will upgrade the file format and prevent the dataset from being opened with earlier versions of the library. 32 bit systems will not be able to use these chunks in all circumstances, such as with data filters or a fill value.

Changed default chunk cache hash table size to 8191

In order to reduce hash collisions and take advantage of modern memory capacity, the default hash table size for the chunk cache has been increased from 521 to 8191. This means the hash table will consume approximately 64 KiB per open dataset. This value can be changed with H5Pset_cache() or H5Pset_chunk_cache(). This value was chosen because it is a prime number close to 8K.

Updated default file format to 1.8

By default, HDF5 will now use the 1.8 file format (H5F_LIBVER_V18). This provides improved performance and space efficiency, particularly with groups and links. This behavior can be overridden with H5Pset_libver_bounds().

Added predefined datatypes for bfloat16 data

Predefined datatypes have been added for little- and big-endian bfloat16 data.

The following new macros have been added:

- `H5T_FLOAT_BFLOAT16LE` / `H5T_FLOAT_BFLOAT16BE`

These macros map to IDs of HDF5 datatypes representing a little- or big-endian 16-bit floating-point datatype with 1 sign bit, 8 exponent bits and 7 fraction bits.

Note that support for a native bfloat16 datatype has not been added yet. This means that any datatype conversions to/from the new bfloat16 datatypes will be emulated in software rather than potentially using specialized hardware instructions. Until support for a native bfloat16 type is added, an application can avoid datatype conversion performance issues if it is sure that the datatype used for in-memory data buffers matches the above floating-point format (such as the __bf16 type). In this case, the application can specify one of the above macros for both the file datatype when creating a dataset or attribute and the memory datatype when performing I/O on the dataset or attribute.

Added predefined datatypes for FP8 data

Predefined datatypes have been added for FP8 data in E4M3 and E5M2 formats (https://arxiv.org/abs/2209.05433).

The following new macros have been added:

- H5T_FLOAT_F8E4M3
- H5T_FLOAT_F8E5M2

These macros map to IDs of HDF5 datatypes representing an 8-bit floating-point datatype with 1 sign bit and either 4 exponent bits and 3 mantissa bits (E4M3 format) or 5 exponent bits and 2 mantissa bits (E5M2 format).

Note that support for a native FP8 datatype has not been added yet. This means that any datatype conversions to/from the new FP8 datatypes will be emulated in software rather than potentially using specialized hardware instructions. Until support for a native FP8 type is added, an application can avoid datatype conversion performance issues if it is sure that the datatype used for in-memory data buffers matches one of the above floating-point formats. In this case, the application can specify one of the above macros for both the file datatype when creating a dataset or attribute and the memory datatype when performing I/O on the dataset or attribute.

Also note that HDF5 currently has incomplete support for datatype conversions involving non-IEEE floating-point format datatypes. Refer to the 'Known Problems' section for information about datatype conversions with these new datatypes.

Removed hbool_t from public API calls

The hbool_t type was introduced before the library supported C99's Boolean type. Originally typedef'd to an integer, it has been typedef'd to C99's bool for many years.

It had been previously purged from the bulk of the library code and only remained in public API signatures. In HDF5 2.0, it has also been removed from public API signatures.

The hbool_t typedef remains in H5public.h so existing code does not need to be updated.

H5public.h no longer includes features.h

features.h is supposed to be included by glibc headers and not used in application code. It is unnecessary given our use of feature test macros like _POSIX_C_SOURCE and has been removed.

Improved performance of opening a virtual dataset with many mappings

When opening a virtual dataset, the library would previously decode the mappings in the object header package, then copy them to the dataset struct, then copy them to the internal dataset creation property list. Copying the VDS mappings could be very expensive if there were many mappings. Changed this to delay decoding the mappings until the dataset code, and delay copying the layout to the DCPL until it is needed. This results in only the decoding and no copies in most use cases, as opposed to the decoding and two copies with the previous code.

Changed the default page buffer size for the ROS3 driver

Calling H5Pset_fapl_ros3() now has the side effect of setting the page buffer size in the FAPL to 64 MiB if it was not previously set. This will only have an effect if the file uses paged allocation. Also added the H5F_PAGE_BUFFER_SIZE_DEFAULT to allow the user to unset the page buffer size in an FAPL so it can be similarly overridden.

Default dataset chunk cache size increased

The default dataset chunk cache size was increased to 8 MiB (8,388,608 bytes).

The file format has been updated to 4.0

The Virtual Dataset Global Heap Block format has been updated to version 1 to support shared string storage for source filenames and dataset names, reducing file size when multiple mappings reference the same sources. This new format is only used when the HDF5 library version bounds lower bound is set to 2.0 or later.

Use of the shared strings option for Virtual Datasets reduces memory overhead and optimizes dataset close operations.

The chunked dataset file format has been updated to always use 64 bits to encode the size of filtered chunks. This will allow data filters that expand the chunks by a large amount to still work. Chunk sizes are still limited to 2^32 - 1. This new format is only used when the HDF5 library version bounds lower bound is set to 2.0 or later.

The H5Dread_chunk() signature has changed

A new parameter, nalloc, has been added to H5Dread_chunk(). This parameter contains a pointer to a variable that holds the size of the buffer buf. If *nalloc is not large enough to hold the entire chunk being read, no data is read. On exit, the value of this variable is set to the buffer size needed to read the chunk.

The old signature has been renamed to H5Dread_chunk1() and is considered deprecated:

   herr_t H5Dread_chunk1(hid_t dset_id, hid_t dxpl_id,
                         const hsize_t *offset, uint32_t *filters,
                         void *buf);

The new signature is H5Dread_chunk2(). All code should be updated to use this version:

   herr_t H5Dread_chunk2(hid_t dset_id, hid_t dxpl_id,
                         const hsize_t *offset, uint32_t *filters,
                         void *buf, size_t *nalloc);

H5Dread_chunk() will map to the new signature unless the library is explicitly configured to use an older version of the API.

Replaced the ROS3 VFD's S3 backend based on libcurl with a new backend based on the aws-c-s3 library

The ROS3 VFD now requires the aws-c-s3 library in order to be built. This library offers several useful features for the VFD, including the following:

  • Automatic retries of non-fatal failed requests (where the libcurl backend would simply return an error),
  • Built-in sourcing of credentials and other configuration information from standard AWS configuration files and environment variables,
  • Automatic splitting and parallelization of large S3 requests,
  • Built-in handling of the formation of HTTP request authentication headers (where the libcurl backend had to manage this manually), and more.

Additional benefits with the rewrite of this backend include:

  • Support for s3:// object URIs,
  • The ability to enable debugging information printouts at runtime by setting a new environment variable HDF5_ROS3_VFD_DEBUG (refer to H5FDros3.h),
  • The ability to capture logging information from the aws-c-s3 library by setting a new environment variable HDF5_ROS3_VFD_LOG_LEVEL (refer to H5FDros3.h).

With these changes, the behavior of the authenticate field of the ROS3 VFD's FAPL structure has changed slightly. If authenticate is true, the ROS3 VFD will only use credentials specified in the FAPL structure and will not attempt to load credentials from other places. In this case, the secret_id and secret_key fields must still both be non-empty strings, or an error will be returned when opening a file. If a session token is to be used, it must be specified with H5Pset_fapl_ros3_token().

If authenticate is false, the ROS3 VFD will instead attempt to load credentials from several different places, in this order:

  • From the environment, by checking AWS environment variables such as AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_SESSION_TOKEN and AWS_ACCOUNT_ID,
  • From the AWS profile files, by reading from ~/.aws/config and ~/.aws/credentials, by default. The specific files read from can be overridden with the AWS_CONFIG_FILE and AWS_SHARED_CREDENTIALS_FILE environment variables,
  • From STS, by using AssumeRoleWithWebIdentity,
  • From EC2 instance metadata.

If the ROS3 VFD cannot source credentials from any of these locations, it will fall back to using anonymous credentials.

This functionality effectively deprecates the --s3-cred option for the h5dump, h5ls, and h5stat tools. However, this option has been kept for compatibility reasons and can still be used to force the specified credentials to take precedence over any credentials that the VFD would otherwise try to source.

Note that with these changes, the AWS region to be used must always be specified. However, the region can now be specified in several ways other than just in the FAPL structure. The VFD will search for a specified AWS region in the following order:

  • The FAPL, if aws_region is not an empty string
  • The AWS_REGION environment variable
  • The AWS_DEFAULT_REGION environment variable
  • The AWS configuration file (~/.aws/config by default)
  • The default profile from this file is used, unless a different profile is specified with the AWS_PROFILE environment variable

If the ROS3 VFD cannot determine an AWS region from one of these locations, an error will be returned when opening a file.

New API functions H5Pset_fapl_ros3_endpoint() and H5Pget_fapl_ros3_endpoint() have been added for use with the ROS3 VFD. These functions set/get an alternate endpoint URL to use when opening files with the ROS3 VFD. This is useful in cases where the application needs to access files that are in a location other than the standard s3.<region-code>.amazonaws.com, which is what the ROS3 VFD uses when an alternate endpoint URL isn't specified. The ROS3 VFD also checks the AWS_ENDPOINT_URL_S3 and AWS_ENDPOINT_URL environment variables for an alternate endpoint URL if one isn't specified with H5Pset_fapl_ros3_endpoint().

Instructions for building the ROS3 VFD with the aws-c-s3 library are in the INSTALL_S3.txt file. The ROS3 VFD and information about the usage of the driver are described in the HDF5 user's guide.

Renamed some API decorations

Some API decorations (used to hide __declspec on Windows, among other things) have been renamed:

Old New
H5_DLLCPP(VAR) H5CPP_DLL(VAR)
H5_HLDLL(VAR) H5HL_DLL(VAR)
H5_HLCPPDLL(VAR) H5CPP_HL_DLL(VAR)
H5_FCDLL(VAR) H5FC_DLL(VAR)
H5_FCTESTDLL(VAR) H5FC_TEST_DLL(VAR)
HDF5_HL_F90CSTUBDLL(VAR) H5FC_HL_DLL(VAR)

The H5Iregister_type() signature has changed

The hash_size parameter has not been used since early versions of HDF5 1.8, so it has been removed, and the API call has been versioned.

The old signature has been renamed to H5Iregister_type1() and is considered deprecated:

H5I_type_t H5Iregister_type1(size_t hash_size, unsigned reserved, H5I_free_t free_func);

The new signature is H5Iregister_type2(). New code should use this version:

H5I_type_t H5Iregister_type2(unsigned reserved, H5I_free_t free_func);

H5Iregister_type() will map to the new signature unless the library is explicitly configured to use an older version of the API.

The H5Tdecode() signature has changed

When provided malformed or too-small buffers, H5Tdecode() would crash. The new buffer size parameter allows this to be reliably avoided.

The old signature has been renamed to H5Tdecode1() and is considered deprecated:

  hid_t H5Tdecode1(const void *buf);

The new signature is H5Tdecode2(). New code should use this version:

  hid_t H5Tdecode2(const void *buf, size_t buf_size);

H5Tdecode() will map to the new signature unless the library is explicitly configured to use an older version of the API.

H5F_LIBVER_LATEST is now an enum value

This was previously #defined to the latest H5F_libver_t API version, but is now an enum value with an integer value equal to the latest H5F_libver_t API version's value. e.g.:

      <snip>
      H5F_LIBVER_V200 = 5,
      H5F_LIBVER_LATEST = 5,
      </snip>

Added support for complex number datatypes

Support for the C99 float _Complex, double _Complex and long double _Complex (with MSVC, _Fcomplex, _Dcomplex and _Lcomplex) types has been added for platforms/compilers that support them. These types have been implemented with a new datatype class, H5T_COMPLEX. Note that any datatypes of class H5T_COMPLEX will not be readable with previous versions of HDF5. If a file is accessed with a library version bounds "high" setting less than H5F_LIBVER_V200, an error will occur if the application tries to create an object with a complex number datatype. If compatibility with previous versions of HDF5 is desired, applications should instead consider adopting one of the existing conventions.

The following new macros have been added:

Macro Description
H5_HAVE_COMPLEX_NUMBERS This macro is defined in H5pubconf.h and will have the value 1 if native support for complex numbers is available. It will not be defined otherwise.
H5_HAVE_C99_COMPLEX_NUMBERS This macro is defined in H5pubconf.h and will have the value 1 if native support for C99 complex numbers is available. It will not be defined otherwise. If this macro is not defined but H5_HAVE_COMPLEX_NUMBERS is defined, the complex number types supported are the MSVC types.
H5_SIZEOF_FLOAT_COMPLEX This macro is defined in H5pubconf.h and will have a value corresponding to the size of the native float complex datatype, as computed by sizeof(). If C99 complex number support is available, this will be the size of the "float _Complex" type. Otherwise, it will be the size of the _Fcomplex type. It will have the value 0 if support for a native float complex datatype is not available.
H5_SIZEOF_DOUBLE_COMPLEX This macro is defined in H5pubconf.h and will have a value corresponding to the size of the native double complex datatype, as computed by sizeof(). If C99 complex number support is available, this will be the size of the double _Complex type. Otherwise, it will be the size of the _Dcomplex type. It will have the value 0 if support for a native double complex datatype is not available.
H5_SIZEOF_LONG_DOUBLE_COMPLEX This macro is defined in H5pubconf.h and will have a value corresponding to the size of the native long double complex datatype, as computed by sizeof(). If C99 complex number support is available, this will be the size of the long double _Complex type. Otherwise, it will be the size of the _Lcomplex type. It will have the value 0 if support for a native long double complex datatype is not available.
H5T_NATIVE_FLOAT_COMPLEX This macro maps to the ID of an HDF5 datatype representing the native C float complex datatype (either "float _Complex" or _Fcomplex) for the platform. If support for a native float complex datatype is not available (H5_HAVE_COMPLEX_NUMBERS is not defined), the macro will map to H5I_INVALID_HID and should not be used.
H5T_NATIVE_DOUBLE_COMPLEX This macro maps to the ID of an HDF5 datatype representing the native C double complex datatype (either double _Complex or _Dcomplex) for the platform. If support for a native double complex datatype is not available (H5_HAVE_COMPLEX_NUMBERS is not defined), the macro will map to H5I_INVALID_HID and should not be used.
H5T_NATIVE_LDOUBLE_COMPLEX This macro maps to the ID of an HDF5 datatype representing the native C long double complex datatype (either long double _Complex or _Lcomplex) for the platform. If support for a native long double complex datatype is not available (H5_HAVE_COMPLEX_NUMBERS is not defined), the macro will map to H5I_INVALID_HID and should not be used.
H5T_COMPLEX_IEEE_F16LE / H5T_COMPLEX_IEEE_F16BE These macros map to IDs of HDF5 datatypes representing a complex number of two parts, each of which is an IEEE 754 16-bit floating-point datatype in little- or big-endian order. These datatypes are available regardless of whether complex number support is available or not.
H5T_COMPLEX_IEEE_F32LE / H5T_COMPLEX_IEEE_F32BE These macros map to IDs of HDF5 datatypes representing a complex number of two parts, each of which is an IEEE 754 32-bit floating-point datatype in little- or big-endian order. These datatypes are available regardless of whether complex number support is available or not.
H5T_COMPLEX_IEEE_F64LE / H5T_COMPLEX_IEEE_F64BE These macros map to IDs of HDF5 datatypes representing a complex number of two parts, each of which is an IEEE 754 64-bit floating-point datatype in little- or big-endian order. These datatypes are available regardless of whether complex number support is available or not.

The following new API function has been added:

Function Description
hid_t H5Tcomplex_create(hid_t base_type_id) Creates a new complex number datatype from the base datatype specified by the given HDF5 ID base_type_id. The base datatype must be a floating-point datatype.

The following new hard datatype conversion paths have been added, but will only be used when complex number support is available:
H5T_NATIVE_SCHAR <-> H5T_NATIVE_FLOAT_COMPLEX H5T_NATIVE_UCHAR <-> H5T_NATIVE_FLOAT_COMPLEX
H5T_NATIVE_SHORT <-> H5T_NATIVE_FLOAT_COMPLEX H5T_NATIVE_USHORT <-> H5T_NATIVE_FLOAT_COMPLEX
H5T_NATIVE_INT <-> H5T_NATIVE_FLOAT_COMPLEX H5T_NATIVE_UINT<-> H5T_NATIVE_FLOAT_COMPLEX
H5T_NATIVE_LONG<-> H5T_NATIVE_FLOAT_COMPLEX H5T_NATIVE_ULONG <-> H5T_NATIVE_FLOAT_COMPLEX
H5T_NATIVE_LLONG <-> H5T_NATIVE_FLOAT_COMPLEX H5T_NATIVE_ULLONG <-> H5T_NATIVE_FLOAT_COMPLEX
H5T_NATIVE_FLOAT16 <-> H5T_NATIVE_FLOAT_COMPLEX H5T_NATIVE_FLOAT <-> H5T_NATIVE_FLOAT_COMPLEX
H5T_NATIVE_DOUBLE <-> H5T_NATIVE_FLOAT_COMPLEX H5T_NATIVE_LDOUBLE <-> H5T_NATIVE_FLOAT_COMPLEX
H5T_NATIVE_SCHAR <-> H5T_NATIVE_DOUBLE_COMPLEX H5T_NATIVE_UCHAR <-> H5T_NATIVE_DOUBLE_COMPLEX
H5T_NATIVE_SHORT <-> H5T_NATIVE_DOUBLE_COMPLEX H5T_NATIVE_USHORT <-> H5T_NATIVE_DOUBLE_COMPLEX
H5T_NATIVE_INT <-> H5T_NATIVE_DOUBLE_COMPLEX H5T_NATIVE_UINT<-> H5T_NATIVE_DOUBLE_COMPLEX
H5T_NATIVE_LONG<-> H5T_NATIVE_DOUBLE_COMPLEX H5T_NATIVE_ULONG <-> H5T_NATIVE_DOUBLE_COMPLEX
H5T_NATIVE_LLONG <-> H5T_NATIVE_DOUBLE_COMPLEX H5T_NATIVE_ULLONG <-> H5T_NATIVE_DOUBLE_COMPLEX
H5T_NATIVE_FLOAT16 <-> H5T_NATIVE_DOUBLE_COMPLEX H5T_NATIVE_FLOAT <-> H5T_NATIVE_DOUBLE_COMPLEX
H5T_NATIVE_DOUBLE <-> H5T_NATIVE_DOUBLE_COMPLEX H5T_NATIVE_LDOUBLE <-> H5T_NATIVE_DOUBLE_COMPLEX
H5T_NATIVE_SCHAR <-> H5T_NATIVE_LDOUBLE_COMPLEX H5T_NATIVE_UCHAR <-> H5T_NATIVE_LDOUBLE_COMPLEX
H5T_NATIVE_SHORT <-> H5T_NATIVE_LDOUBLE_COMPLEX H5T_NATIVE_USHORT <-> H5T_NATIVE_LDOUBLE_COMPLEX
H5T_NATIVE_INT <-> H5T_NATIVE_LDOUBLE_COMPLEX H5T_NATIVE_UINT<-> H5T_NATIVE_LDOUBLE_COMPLEX
H5T_NATIVE_LONG <-> H5T_NATIVE_LDOUBLE_COMPLEX H5T_NATIVE_ULONG <-> H5T_NATIVE_LDOUBLE_COMPLEX
H5T_NATIVE_LLONG <-> H5T_NATIVE_LDOUBLE_COMPLEX H5T_NATIVE_ULLONG <-> H5T_NATIVE_LDOUBLE_COMPLEX
H5T_NATIVE_FLOAT16 <-> H5T_NATIVE_LDOUBLE_COMPLEX H5T_NATIVE_FLOAT <-> H5T_NATIVE_LDOUBLE_COMPLEX
H5T_NATIVE_DOUBLE <-> H5T_NATIVE_LDOUBLE_COMPLEX H5T_NATIVE_LDOUBLE <-> H5T_NATIVE_LDOUBLE_COMPLEX
H5T_NATIVE_FLOAT_COMPLEX <-> H5T_NATIVE_DOUBLE_COMPLEX
H5T_NATIVE_FLOAT_COMPLEX <-> H5T_NATIVE_LDOUBLE_COMPLEX
H5T_NATIVE_DOUBLE_COMPLEX <-> H5T_NATIVE_LDOUBLE_COMPLEX

Alternative software implementation conversion paths have been added for all of the above for use when native complex number support is not available. All of these conversion paths follow the behavior outlined in the C standard for conversions of complex number values.

Additionally, a special datatype conversion path has been added between complex number datatypes and array or compound datatypes, where the in-memory layout of data is the same between the datatypes and data can be directly converted. This conversion path is subject to the following rules:

  • An array datatype must consist of exactly two elements, where each element is of the same floating-point datatype as the complex number datatype's base floating-point datatype.
  • A compound datatype must consist of exactly two fields, where each field is of the same floating-point datatype as the complex number datatype's base floating-point datatype. The compound datatype must not have any leading or trailing structure padding or any padding between its two fields. The fields must also have compatible names, must have compatible offsets within the datatype and must be in the order of "real" part -> "imaginary" part, such that the compound datatype matches the following representation:
H5T_COMPOUND {
<float_type> "r(e)(a)(l)";  OFFSET 0
<float_type> "i(m)(a)(g)(i)(n)(a)(r)(y)"; OFFSET SIZEOF("r(e)(a)(l)")
}

where r(e)(a)(l) means the field may be named any substring of "real", such as "r", or "re", and i(m)(a)(g)(i)(n)(a)(r)(y) means the field may be named any substring of "imaginary", such as "im" or "imag".

Support for complex numbers has been added to the h5dump, h5ls, and h5diff/ph5diff tools. The h5dump command-line option '-m' can be used to change the floating-point printing format for the float complex and double complex datatypes, as well as the long double complex datatype if it has the same size as a double complex datatype.

Support for the predefined complex number datatypes and the H5Tcomplex_create function has been added to the Java wrappers. However, Java does not have official types for complex numbers, so an application must be sure that data is in an appropriate format in-memory when using these datatypes.

Support for the Fortran wrappers has not yet been added.

Support for the predefined complex number datatypes and the H5Tcomplex_create function has been added to the high-level library, allowing them to work with the H5LTtext_to_dtype and H5LTdtype_to_text functions.

Simple example programs showing how to use complex number datatypes have been added in the following files:

  • HDF5Examples/C/H5T/200/h5ex_t_complex.c (Uses C99 complex number types)
  • HDF5Examples/C/H5T/200/h5ex_t_complex_msvc.c (Uses MSVC complex number types)
  • HDF5Examples/C/H5T/200/h5ex_t_complex_custom.c (Uses H5Tcomplex_create() to create a custom complex number type)

FOR VOL DEVELOPERS: Renamed H5VLstart_lib_state and H5VLfinish_lib_state

The APIs H5VLstart_lib_state and H5VLfinish_lib_state have been renamed to H5VLopen_lib_context and H5VLclose_lib_context, respectively, with the addition of a "context" argument.

Removed H5FDperform_init API routine.

Virtual File Driver (VFD) developers who wish to provide an ID for their driver should create a routine specific to their individual implementation.

H5Pset_external() now uses HDoff_t, which is always a 64-bit type

The H5Pset_external() call took an off_t parameter in HDF5 1.14.x and earlier. On POSIX systems, off_t is specified as a 64-bit type via POSIX large-file support (LFS). On Windows, however, off_t is defined as a 32-bit type, even on 64-bit Windows.

HDoff_t has been added to H5public.h and is defined to be int64_t on Windows and the library has been updated to use HDoff_t in place of off_t throughout. The H5Pset_external() offset parameter has also been updated to be HDoff_t.

There is no API compatibility wrapper for this change.

Fixes GitHub issue #3506

H5Pset* routines now fail when used on default property lists

Modifying default property lists was never fully supported and could produce inconsistent and unexpected behavior.

H5Pset_vol() now fails when used on a non-file-access property list

Similar to the above. Setting the connector on a non-FAPL had no effect on library behavior, and the connector ID and information could not be read back from that plist later.

Optimized Virtual Dataset opens by delaying layout copy

On dataset open, the dataset performed an internal copy of the layout in order to populate its internal DCPL. For virtual datasets, this added a significant amount of overhead to the open operation.

This layout copy is now delayed until either a user requests the DCPL, or until the start of an operation that needs to read the layout from the DCPL.

Virtual datasets now use a spatial tree to optimize searches

Virtual dataset operations with many (>1,000) mappings were much slower than corresponding operations on normal datasets. This was due to the need to iterate through every source dataset's dataspace and check for an intersection with the user-selected region for a read/write in the virtual dataset.

Virtual datasets with many mappings now use an r-tree (defined in H5RT.c) to perform a spatial search. This allows the dataspaces that intersect the user-selection to be computed with, in most cases, much fewer intersection checks, improving the speed of VDS read/write operations.

Virtual datasets will use the r-tree by default, since the majority of use cases, should see improvements from use of the tree. However, because some workflows may find that the overhead of the tree outweighs the time saved on searches, there is a new Dataset Access Property List (DAPL) property to control use of the spatial tree.

This property can be set or queried with the new API functions H5Pset_virtual_spatial_tree()/H5Pget_virtual_spatial_tree().

Parallel Library

Added H5FDsubfiling_get_file_mapping() API function for subfiling VFD

Added H5FDsubfiling_get_file_mapping() API function to retrieve the names of all physical subfiles that collectively make up a logical HDF5 file when using the subfiling Virtual File Driver.

Fortran Library

Added Fortran wrapper h5fdsubfiling_get_file_mapping_f() for subfiling VFD

Added Fortran wrapper h5fdsubfiling_get_file_mapping_f() for the subfiling file mapping functionality, ensuring complete language binding support.

C++ Library

Java Library

Tools

Added AWS endpoint command option to allow specifying an alternate endpoint URL when using the ROS3 VFD

The new option is --endpoint-url, which allows the user to set an alternate endpoint URL other than the standard "protocol://service-code.region-code.amazonaws.com". If --endpoint-url is not specified, the ROS3 VFD will first check the AWS_ENDPOINT_URL_S3 and AWS_ENDPOINT_URL environment variables for an alternate endpoint URL before using a default one, with the region-code being supplied by the FAPL or standard AWS locations/environment variables.

This option is supported by the following tools: h5dump, h5ls, h5stat

Specifying ROS3 VFD on the command line is not required when using S3 URI

If using an S3 URI to reference an HDF5 file in S3 (example: s3://mybucket/myfile.h5), then ROS3 VFD will be automatically selected unless the command-line option for the virtual file driver is used.

This feature applies to the following tools: h5dump, h5ls, h5stat.

Deprecated h5dump XML option

The h5dump XML option is deprecated and will be removed in a future release. The XML output format has not been maintained and is not up-to-date with the latest features of HDF5.

Added h5dump command option to set the floating point format for long double

The new option is --lformat, which allows the user to set the floating point format for long double. The default format is %Lg. There is already an option --format to set the floating point format for double and float. The default format is %g.

Removed the high-level GIF tools

The high-level GIF tools, h52gif and gif2h5, have unfixed CVE issues (with no proof-of-concept files). They are not critical tools, are not well-maintained, and are an odd fit for building with the library. Because of this, they have been removed. We may move them to a separate repository in the future.

This also removed the following configure options: CMake: HDF5_BUILD_HL_GIF_TOOLS

High-Level APIs

C Packet Table API

Internal header file

Documentation

The COPYING file has been renamed to LICENSE

This is where most people will expect to find license information. The COPYING_LBNL_HDF5 file has also been renamed to LICENSE_LBNL_HDF5. The licenses are unchanged.

🪲 Bug Fixes

Library

Fixed problems with the family driver and user block

When using a user block with the family driver, the driver would inappropriately subtract the user block size for each member file when calculating member EOAs. This could cause a failure when an address overflowed the calculated eoa. The driver would also add the user block size when returning the EOF. Modified the family driver to not consider the user block, as it is handled by the H5FD layer. The user block now spans the first X bytes of the family array, for example a 4 KiB user block with 3 KiB member size will take up the entire first member and the first 1 KiB of the second. This may cause compatibility issues with preexisting family files with user blocks, though the way it worked before was inconsistent if it worked at all.

Fixed security issue CVE-2025-7067

Fixed a heap buffer overflow in H5FS__sinfo_serialize_node_cb() by discarding file free space sections from the file free space manager when they are found to be invalid. Specifically crafted HDF5 files can result in an attempt to insert duplicate or overlapping file free space sections into a file free space manager, later resulting in a buffer overflow when the same free space section is serialized to the file multiple times.

Fixes GitHub issue #5577

Fixed security issue CVE-2025-2915 and OSV-2024-381

Fixed a heap-based buffer overflow in H5F__accum_free caused by an integer overflow when calculating new_accum_size. Added validation in H5O__mdci_decode to detect and reject invalid values early, preventing the overflow condition.

Fixes GitHub issue #5380

Fixed security issue CVE-2025-7068

Failures during the discard process on a metadata cache entry could cause the library to skip calling the callback to free the cache entry. This could result in resource leaks and issues with flushing and closing the metadata cache during file close. This has been fixed by noting errors during the discard process, but attempting to fully free a cache entry before signalling that an error has occurred.

Fixes GitHub issue #5578

Fix bugs in object header operations

In some rare circumstances, such as deleting hard links that point to their own parent group in a file using the new file format, memory corruption could occur due to recursive operations changing data structures being operated on by multiple levels of recursion. Made changes to delay changing the data structure in a dangerous way until recursion is complete.

Fixes GitHub issue #5854

Fixed security issues CVE-2025-6816, CVE-2025-6818, CVE-2025-6856 and CVE-2025-2923

A specially constructed HDF5 file could contain a corrupted object header with a continuation message that points back to itself. This could result in an internal buffer being allocated with too small of a size, leading to a heap buffer overflow. This has been fixed by checking the expected number of object header chunks against the actual value as chunks are being deserialized.

Fixes GitHub issues #5571, #5573, #5574 and #5381

Fixed security issue CVE-2025-6750

A heap buffer overflow occurred because an mtime message was not properly decoded, resulting in a buffer of size 0 being passed into the encoder. This has been fixed by decoding old and new mtime messages which will allow invalid message size to be detected.

Fixes GitHub issue #5549

Fixed CVE-2025-7069

An invalid page size in file space info was decoded, which caused a heap buffer overflow when reading certain compromised data files. The invalid value has been detected for a graceful failure.

Fixes GitHub issues #5550 and OSS issues OSV-2024-379, OSV-2024-575, and OSV-2024-772

Fixed CVE-2025-6269 and CVE-2025-6516

There were several security vulnerabilities found in the function H5C__reconstruct_cache_entry(), including buffer overflows and memory leaks. The function has been hardened with bounds checks, input validation, and safe cleanup.

Fixes GitHub issues #5579 and #5581

Fixed a problem with the scale-offset filter

A security fix added to 1.14.6 introduced a regression where certain data values could trigger a library error (not a crash or segfault).

Fixes GitHub issue #5861

Fixed security issue CVE-2025-2153

The message flags field could be modified such that a message that is not sharable according to the share_flags field in H5O_msg_class_t can be treated as sharable. An assert has been added in H5O__msg_write_real to make sure messages that are not sharable can't be modified to shared. Additionally, the check in H5O__chunk_deserialize that catches unsharable messages being marked as sharable has been improved.

Fixes GitHub issue #5329 and OSV-2023-76

Fixed security issue CVE-2025-2925

Actual_len + H5C_IMAGE_EXTRA_SPACE, which was used by H5MM_realloc as the size input, could equal 0 due to bad inputs. When H5MM_realloc was called, it freed image, but then could get sent to done before new_image could be assigned to image. Because the pointer for image wasn't null, it was freed again in done, causing a double-free vulnerability. H5C__load_entry() now checks for an image buffer length of 0 before calling H5MM_realloc.

Fixes GitHub issue #5383

Fixed security issue CVE-2025-6857

An HDF5 file had a corrupted v1 B-tree that would result in a stack overflow when performing a lookup on it. This has been fixed with additional integrity checks.

Fixes GitHub issue #5575

Check for overflow in decoded heap block addresses

Currently, we do not check for overflow when decoding addresses from the heap, which can cause overflow problems. We've added a check in H5HL__fl_deserialize to ensure no overflow can occur.

Fixes GitHub issue #5382

Fixed a segfault with a corrupted file based on netCDF-4

When the internal function H5G__dense_iterate failed early, allocated structures were never populated, causing cleanup code to read uninitialized values and attempt to free invalid pointers. The structures are now initialized with proper values and NULL pointers immediately after allocation.

Fixes GitHub issue #5375

Fixed security issues CVE-2025-2913, CVE-2025-2926, CVE-2025-6817, and CVE-2025-6858

The size of a continuation message was decoded as 0, causing multiple vulnerabilities. An error check was added to return failure to prevent further processing of invalid data.

Fixes GitHub issue #5376, #5384, #5572, and #5576

Revised handling of Unicode filenames on Windows

In the HDF5 1.14.4 release, a change was made to address some issues with the library's handling of code pages and file paths on Windows. This change introduced other issues with the handling of UTF-8 file names that caused breakage for software using the 1.14.4 and 1.14.5 releases of HDF5. That change was reverted for the 1.14.6 release and the behavior has been slightly modified for this release.

On Windows, the library once again assumes that filename strings will be UTF-8 encoded strings and will attempt to convert them to UTF-16 before passing them to Windows API functions. However, if the library fails to convert a filename string to UTF-16, it will now fallback to the equivalent Windows "ANSI" API functions which will interpret the string according to the active Windows code page.

Support for a new environment variable, HDF5_PREFER_WINDOWS_CODE_PAGE, was added in order to instruct HDF5 to prefer interpreting filenames according to the active Windows code page rather than assuming UTF-8 encoding. If this environment variable is set to "1" or "TRUE" (case-insensitive), the active code page will be preferred. If it is unset or set to "0" or "FALSE" (case-insensitive), UTF-8 will be preferred.

Fixed an issue with caching in the ROS3 VFD

The ROS3 VFD uses a very simple caching mechanism that caches the first 16MiB of a file during file open and serves later reads from that cache if the offset + length falls within the cached range of bytes. Combinations of offset + length that extended exactly to the end of the cached range of bytes (for example, offset=0 and len=16777216) would end up not being served from the cache due to an incorrect range check. This has now been fixed.

Fixed an error with H5Fget_file_image() with the latest file format

When using H5Fget_file_image() on a file created with the latest file format (or any format newer than the earliest), the library failed to recalculate the superblock checksum after changing the access flags in the superblock, causing any subsequent attempt to open the returned file image to fail due to the checksum failing to verify. Fixed H5Fget_file_image() to recalculate the checksum.

Fixed GitHub issue #1915

Fixed an assertion failure in H5S__hyper_make_spans()

Calling H5Sselect_hyperslab() on dataspaces with invalid extents could result in an assertion failure in debug builds of the library if the dataspace has an extent with a rank value of 0. This has been fixed by converting the assertion failure into a normal error check.

Fixed an assertion failure in H5S__hyper_new_span_info()

Calling H5Scopy() on hyperslab selection dataspaces with invalid extents could result in an assertion failure in debug builds of the library if the dataspace has an extent with a rank value of 0. This has been fixed by converting the assertion failure into a normal error check.

Fixed a segfault in H5S__get_select_hyper_blocklist()

When attempting to retrieve the list of hyperslab blocks selected within a dataspace, a segfault or bus error could occur when the dataspace has an extent with a rank value of 0. This would cause indexing into an array variable on the stack using a negative value. An error check was added to return failure from the function for such dataspaces.

Fixed an error in H5Ddebug

H5Ddebug would fail for any chunked dataset with a chunk index, due to its failure to tag the dataset before performing metadata operations. This caused h5ls -va to silently fail to print chunk addresses. This has been fixed.

Fixed a bug in the H5Oexists and H5Oexists_by_name API routines that would cause those routines to return FAIL instead of FALSE when checking the existence of a non-existent object with a file ID instead of a group ID.

Fixed a segfault in h5dump when a B-tree node level is corrupted

h5dump produced a segfault on a malformed file because a B-tree node level was corrupted.

An internal function was modified to help detect when a decoded B-tree node level has an unexpected value, and an error will be produced.

Fixed GitHub issue [#4432](https://github.com/HDFGroup/hdf5/issues/4432)

Fixed H5Ovisit2 to recursively visit all objects

`H5Ovisit2` visited only the root group and not all the nested groups.

This behavior occurred when the fields are not `H5O_INFO_BASIC` or `H5O_INFO_ALL` because an internal function did not obtain the basic information needed by its caller. This problem is now fixed.

Fixed GitHub issue #4941

Only clear FE_INVALID when that symbol is present on the system

When we initialize the floating-point types at library startup, it's possible to raise floating-point exceptions when we check which things are supported. Normally, we clear these floating-point exceptions via feclearexcept(FE_INVALID), but FE_INVALID may not be present on all systems. Specifically, this was reported as being a problem when using Emscripten 3.1.68 to compile HDF5 1.14.5 to WebAssembly.

We've added an #ifdef FE_INVALID block around the exception clearing code to correct this.

Fixed GitHub issue #4952

Fixed security issue CVE-2025-2310

A malformed HDF5 file could have an attribute with a recorded name length of zero.This would lead to an overflow and an invalid memory access. An integrity check has been added to detect this case and safely stop file decoding.

Java Library

Renamed the Callbacks.java file to H5Callbacks.java

The Callbacks.java file was renamed to H5Callbacks.java to match the file pattern used by doxygen. This change only affects the Java filenames and does not change the classname or the package name.

Configuration

Reorganized CMake HDF5 configuration options

The CMake configuration options have been reorganized to identify the primary options that are relevant to the build. These options are now in a separate file, CMakeBuildOptions.cmake, which is included by the root CMakeLists.txt file. In addition, some options have been converted to cmake_dependent_option() calls, which allows the options to be hidden from the CMake GUI when they are not relevant to the build.

Remove default setting of CMAKE_DEBUG_POSTFIX

Move the default setting of CMAKE_DEBUG_POSTFIX to the cacheinit.cmake file usually used by testing. If CMAKE_DEBUG_POSTFIX is not set with a -D option then CMAKE_DEBUG_POSTFIX will be the default provided by CMake itself.

The relative rpaths ($ORIGIN / @loader_path) are appended to the CMAKE_INSTALL_RPATH

The RPATH settings were removed by a pull-request #5271, but the settings are needed under certain conditions. These settings have been restored by appending the necessary paths and will not override/overwrite any existing settings.

When using a system-installed zlib library, the shared library is expected to be found in the system library path.

Setting the HDF5_MODULE_MODE_ZLIB option to OFF will force find_package to use config mode first. An installed zlib, or an alternate installed zlib library, is expected to have a correct zlib-config.cmake file for config mode. Current zlib installs usually do not have a zlib-config.cmake file, so the option is set to ON by default.

Use pre-installed libaec compression library

The CMake logic for finding the libaec compression library has been modified for a system-installed version of the library. Two options must be set:

HDF5_ALLOW_EXTERNAL_SUPPORT:STRING=NO
<LIB_PKG_NAME>_USE_EXTERNAL:BOOL=OFF

where <LIB_PKG_NAME> is one of ZLIB, ZLIBNG, SZIP, PLUGIN.

Note that HDF5_ALLOW_EXTERNAL_SUPPORT:STRING=NO disables building all plugins and external libraries in-line with the HDF5 library.

In addition, the <LIB_PKG_NAME>_ROOT environment variables must be set, where <LIB_PKG_NAME> is one of ZLIB, ZLIBNG, SZIP, libaec, PLUGIN. Note that libaec is the expected name for using the libaec library in place of original szip.

See INSTALL_CMake.txt for more detailed information.

Changed the zlib/szip compression find message to FATAL ERROR

The message was changed to indicate that zlib/szip compression was requested and that it was not found. If an option is requested, not finding it should always be an error.

Removed the module search find_package for szip library

There is not an szip module file to use, so the find_package only uses find_package in config mode. The choice then is to either build szip, with libaec, inline, or find a system installed szip library, built with CMake.

Tools

h5repack did not properly parse User Defined filters

The h5repack tool did not properly parse user-defined filter command-line arguments when the number of elements value was 0 (zero). Also, using a colon without a preceding object was enforced to behave the same as not using a colon.

Fixed GitHub issue #5132

Changed the default value for number of cd_values in filters.

The tools used an arbitrary value 0f 20 for the number of cd_values used in a filter. Created a new define DEFAULT_CDELEMTS in H5tools.h for the default value, which currently matches the library restriction of 256.

Fixed GitHub issue #5414

Performance

Fortran API

Added missing parameters H5F_ACC_SWMR_READ_F and H5F_ACC_SWMR_WRITE_F Fixed GitHub issue #5959

High-Level Library

Fixed an issue with H5TB functions

The H5TB functions were not correctly creating the FILL_INFO attribute for tables. This has been fixed by using the field offsets array from the call to H5TBAget_fill instead of using the compound type member offsets from the H5Tget_member_offset call for each compound type.

Fortran High-Level APIs

Documentation

F90 APIs

C++ APIs

Testing

Fixed an allocation in the t_bigio parallel test on 32-bit systems

A test in t_bigio.c attempts to allocate more than 4 GiB of memory, which will overflow (and wrap) the size_t type on 32-bit systems, creating a very small allocation instead of a very large allocation. The test then segfaults when it accesses memory outside of the small buffer.

The test has been fixed by limiting the buffer to 2 GiB on 32-bit systems.

Fixed GitHub #2510

Added skipping of a few parallel tests for OpenMPI 5.0.5

An issue in OpenMPI 5.0.5 causes a few parallel HDF5 tests (mpiodup, props, fapl_preserve) to fail. These tests are now skipped for that release of OpenMPI. The issue has been fixed in the 5.0.6 release of OpenMPI.

✨ Support for new platforms and languages

☑️ Platforms Tested

A table of platforms tested can be seen on the wiki. Current test results are available here.

⛔ Known Problems

  • When performing implicit datatype conversion on specific non-IEEE floating-point format data, HDF5 may improperly convert some data values:

    When performing I/O operations using a non-IEEE floating-point format datatype, HDF5 may improperly convert some data values due to incomplete handling of non-IEEE types. Such types include the following pre-defined datatypes:

    H5T_FLOAT_F8E4M3 H5T_FLOAT_F8E5M2

    If possible, an application should perform I/O with these datatypes using an in-memory type that matches the specific floating-point format and perform explicit data conversion outside of HDF5, if necessary. Otherwise, read/written values should be verified to be correct.

  • When the library detects and builds in support for the _Float16 datatype, an issue has been observed on at least one MacOS 14 system where the library fails to initialize due to not being able to detect the byte order of the _Float16 type #4310:

    #5: H5Tinit_float.c line 308 in H5T__fix_order(): failed to detect byte order major: Datatype minor: Unable to initialize object

    If this issue is encountered, support for the _Float16 type can be disabled with a configuration option:

    CMake: HDF5_ENABLE_NONSTANDARD_FEATURE_FLOAT16=OFF

  • When HDF5 is compiled with NVHPC versions 23.5 - 23.9 (additional versions may also be applicable) and with -O2 (or higher) and -DNDEBUG, test failures occur in the following tests:

    • H5PLUGIN-filter_plugin
    • H5TEST-flush2
    • H5TEST-testhdf5-base
    • MPI_TEST_t_filters_parallel

    Sporadic failures (even with lower -O levels):

    • Java JUnit-TestH5Pfapl
    • Java JUnit-TestH5D

    Also, NVHPC will fail to compile the test/tselect.c test file with a compiler error of 'use of undefined value' when the optimization level is -O2 or higher.

    This is confirmed to be a bug in the nvc compiler that has been fixed as of 23.11. If you are using an affected version of the NVidia compiler, the work-around is to set the optimization level to -O1.

  • CMake files do not behave correctly with paths containing spaces

    Do not use spaces in paths because the required escaping for handling spaces results in very complex and fragile build files.

  • At present, metadata cache images may not be generated by parallel applications. Parallel applications can read files with metadata cache images, but since this is a collective operation, a deadlock is possible if one or more processes do not participate.

  • The subsetting option in ph5diff currently will fail and should be avoided

    The subsetting option works correctly in serial h5diff.

  • Flang Fortran compilation will fail (last check version 17) due to not yet implemented: (1) derived type argument passed by value (H5VLff.F90), and (2) support for REAL with KIND = 2 in intrinsic SPACING used in testing.

  • Fortran tests HDF5_1_8.F90 and HDF5_F03.F90 will fail with Cray compilers greater than version 16.0 due to a compiler bug. The latest version verified as failing was version 17.0.

  • Several tests currently fail on certain platforms: MPI_TEST-t_bigio fails with spectrum-mpi on ppc64le platforms.

    MPI_TEST-t_subfiling_vfd and MPI_TEST_EXAMPLES-ph5_subfiling fail with cray-mpich on theta and with XL compilers on ppc64le platforms.

  • File space may not be released when overwriting or deleting certain nested variable length or reference types.

Known problems in previous releases can be found in the HISTORY*.txt files in the HDF5 source. Please report any new problems found to help@hdfgroup.org.

Reference in New Issue View Git Blame Copy Permalink