HDF5 version 1.10.6-snap1 currently under development ================================================================================ INTRODUCTION This document describes the differences between this release and the previous HDF5 release. It contains information on the platforms tested and known problems in this release. For more details check the HISTORY*.txt files in the HDF5 source. Note that documentation in the links below will be updated at the time of each final release. Links to HDF5 documentation can be found on The HDF5 web page: https://portal.hdfgroup.org/display/HDF5/HDF5 The official HDF5 releases can be obtained from: https://www.hdfgroup.org/downloads/hdf5/ Changes from Release to Release and New Features in the HDF5-1.10.x release series can be found at: https://portal.hdfgroup.org/display/HDF5/HDF5+Application+Developer%27s+Guide If you have any questions or comments, please send them to the HDF Help Desk: help@hdfgroup.org CONTENTS - New Features - Support for new platforms and languages - Bug Fixes since HDF5-1.10.4 - Supported Platforms - Tested Configuration Features Summary - More Tested Platforms - Known Problems - CMake vs. Autotools installations New Features ============ Configuration: ------------- - Rework CMake command files to fix MPI testing. Added setup fixture to remove any test generated files and added DEPENDS to test properties to execute tests in order expected. (ADB - 2019/02/14, TRILABS-111) - Disable SZIP or ZLIB options if TGZ files are not available. Changed the TGZ option for SZip and ZLib to disable the options if the source tar.gz files are not found. (ADB - 2019/02/05, HDFFV-10697) - Added a new option to enable/disable using pread/pwrite instead of read/write in the sec2, log, and core VFDs. This option is enabled by default when pread/pwrite are detected. Autotools: --enable-preadwrite CMake: HDF5_ENABLE_PREADWRITE (DER - 2019/02/03, HDFFV-10696) - Rework CMake versioning for OSX platforms. Changed the current_version and compatibility_version flags from optional with HDF5_BUILD_WITH_INSTALL_NAME to always setting the flags for OSX. (ADB - 2019/01/22, HDFFV-10685) - Rework CMake command files to eliminate developer CMP005 warning Use variables without quotes in if () statements. (ADB - 2019/01/18, TILABS-105) - Rework CMake configure files to eliminate developer CMP0075 warning Renamed varname to HDF5_REQUIRED_LIBRARIES as the contents were not required for configuration. Also moved check includes calls to top of files. (ADB - 2019/01/03, HDFFV-10546) - Keep stderr and stdout separate in tests Changed test handling of output capture. Tests now keep the stderr output separate from the stdout output. It is up to the test to decide which output to check against a reference. Also added the option to grep for a string in either output. (ADB - 2018/12/12, HDFFV-10632) - Incorrectly installed private header files were removed from CMake installs. The CMake build files incorrectly flagged the following header files as public and installed them. They are private and will no longer be installed. HDF5 library private package files (H5Xpkg.h) H5Edefin.h H5Einit.h H5Eterm.h H5LTparse.h h5diff.h h5tools_dump.h h5tools.h h5tools_ref.h h5tools_str.h h5tools_utils.h h5trav.h (DER - 2018/10/26, HDFFV-10614, 10609) - Autotools installs now install H5FDwindows.h This is simply to align the installed header files between the autotools and CMake. H5FDwindows.h has no functionality on non-Windows systems. (DER - 2018/10/26, HDFFV-10614) Library: -------- - The sec2, log, and core VFDs can now use pread/pwrite instead of read/write. pread and pwrite do not change the file offset, a feature that was requested by a user working with a multi-threaded application. The option to configure this feature is described above. (DER - 2019/02/03, HDFFV-10696) - Add ability to minimze dataset object headers. Creation of many, very small datasets resulted in extensive file bloat due to extra space in the dataset object headers -- this space is allocated by default to allow for the insertion of a small number of attributes within the object header and not require a continuation block, an unnecessary provision in the target use case. Inform the library to expect no attributes on created datasets, and to allocate the least space possible for the object headers. NOTE: A continuation block is created if attributes are added to a 'minimized' dataset, which can reduce performance. NOTE: Some extra space is allocated for attributes essential to the correct behavior of the object header (store creation times, e.g.). This does not violate the design principle, as the space is calculated and allocated as needed at the time of dataset object header creation -- unused space is not generated. New API calls: H5Fget_dset_no_attrs_hint H5Fset_dset_no_attrs_hint H5Pget_dset_no_attrs_hint H5Pset_dset_no_attrs_hint (JOS - 2019/01/04, TRILAB-45) - Added new chunk query functions The following public functions were added to discover information about the chunks in an HDF5 file. herr_t H5Dget_num_chunks(dset_id, fspace_id, *nchunks) herr_t H5Dget_chunk_info_by_coord(dset_id, *coord, *filter_mask, *addr, *size) herr_t H5Dget_chunk_info(dset_id, fspace_id, index, *coord, *filter_mask, *addr, *size) (BMR - 2018/11/07, HDFFV-10615) - Several empty public header files where removed from the distribution The following files were empty placeholders. They are for internal packages that are unlikely to ever have public functionality and have thus been removed. H5Bpublic.h H5B2public.h H5FSpublic.h H5HFpublic.h H5HGpublic.h H5HLpublic.h They were only installed in CMake builds. (DER - 2018/10/26, HDFFV-10614) Parallel Library: ----------------- - Changed the default behavior in parallel when reading the same dataset in its entirety (i.e. H5S_ALL dataset selection) which is being read by all the processes collectively. The dataset must be contiguous, less than 2GB, and of an atomic datatype. The new behavior is the HDF5 library will use an MPI_Bcast to pass the data read from the disk by the root process to the remain processes in the MPI communicator associated with the HDF5 file. (MSB - 2019/01/02, HDFFV-10652) - All MPI-1 API calls have been replaced with MPI-2 equivalents. This was done to better support OpenMPI, as default builds no longer include MPI-1 support (as of OpenMPI 4.0). (DER - 2018/12/30, HDFFV-10566) Fortran Library: ---------------- - Added wrappers for dataset object header minimization calls. (see the note for TRILAB-45, above) New API calls: h5fget_dset_no_attrs_hint_f h5fset_dset_no_attrs_hint_f h5pget_dset_no_attrs_hint_f h5pset_dset_no_attrs_hint_f (DER - 2019/01/09, TRILAB-45) - Added new Fortran derived type, c_h5o_info_t, which is interoperable with C's h5o_info_t. This is needed for callback functions which pass C's h5o_info_t data type definition. (MSB, 2019/01/08, HDFFV-10443) - Added new Fortran API, H5gmtime, which converts (C) 'time_t' structure to Fortran DATE AND TIME storage format. (MSB, 2019/01/08, HDFFV-10443) - Added new Fortran 'fields' optional parameter to: h5ovisit_f, h5oget_info_by_name_f, h5oget_info, h5oget_info_by_idx and h5ovisit_by_name_f. (MSB, 2019/01/08, HDFFV-10443) C++ Library: ------------ - Added new function to the C++ interface Added wrapper for H5Ovisit2: H5Object::visit() (BMR - 2019/02/14, HDFFV-10532) Java Library: ---------------- - Rewrote the JNI error handling to be much cleaner (JTH - 2019/02/12) - Add new functions to java interface Added wrappers for: H5Fset_libver_bounds H5Fget_dset_no_attrs_hint/H5Fset_dset_no_attrs_hint H5Pget_dset_no_attrs_hint/H5Pset_dset_no_attrs_hint (ADB - 2019/01/07, HDFFV-10664) - Fix java unit tests when Time is a natural number Time substitution in java/test/junit.sh.in doesn't handle the case when Time is a natural number. Fixed the regular expression. (ADB - 2019/01/07, HDFFV-10674) - Duplicate the data read/write functions of Datasets for Attributes. Region references could not be displayed for attributes as they could for datasets. Datasets had overloaded read and write functions for different datatypes that were not available for attributes. After adding similar functions, attribute region references work normally. (ADB - 2018/12/12, HDFVIEW-4) Tools: ------ - The h5repart -family-to-sec2 argument was changed to -family-to-single In order to better support other single-file VFDs which could work with h5repart, the -family-to-sec2 argument was renamed to -family-to-single. This is just a name change and the functionality of the argument has not changed. The -family-to-sec2 argument has been kept for backwards-compatibility. This argument should be considered deprecated. (DER - 2018/11/14, HDFFV-10633) High-Level APIs: --------------- - C Packet Table API ------------------ - Internal header file -------------------- - Documentation ------------- - Support for new platforms, languages and compilers. ======================================= - Bug Fixes since HDF5-1.10.4 release ================================== Library ------- - Fixed a bug that would cause an error or cause fill values to be incorrectly read from a chunked dataset using the "single chunk" index if the data was held in cache and there was no data on disk. (NAF - 2019/03/06) - Fixed a bug that could cause an error or cause fill values to be incorrectly read from a dataset that was written to using H5Dwrite_chunk if the dataset was not closed after writing. (NAF - 2019/03/06, HDFFV-10716) - Fixed memory leak in scale offset filter In a special case where the MinBits is the same as the number of bits in the datatype's precision, the filter's data buffer was not freed, causing the memory usage to grow. In general the buffer was freed correctly. The Minbits are the minimal number of bits to store the data values. Please see the reference manual for H5Pset_scaleoffset for the detail. (RL - 2019/3/4, HDFFV-10705) - Fix hangs with collective metadata reads during chunked dataset I/O In the parallel library, it was discovered that when a particular sequence of operations following a pattern of: "write to chunked dataset" -> "flush file" -> "read from dataset" occurred with collective metadata reads enabled, hangs could be observed due to certain MPI ranks not participating in the collective metadata reads. To fix the issue, collective metadata reads are now disabled during chunked dataset raw data I/O. (JTH - 2019/02/11, HDFFV-10563, HDFFV-10688) - Performance issue when closing an object The slow down is due to the search of the "tag_list" to find out the "corked" status of an object and "uncork" it if so. Improve performance by skipping the search of the "tag_list" if there are no "corked" objects when closing an object. (VC - 2019/02/06) - Uninitialized bytes from a type conversion buffer could be written to disk in H5Dwrite calls where type conversion takes place and the type conversion buffer was created by the HDF5 library. When H5Dwrite is called and datatype conversion must be performed, the library will create a temporary buffer for type conversion if one is not provided by the user via H5Pset_buffer. This internal buffer is allocated via malloc and contains uninitialized data. In some datatype conversions (float to long double, possibly others), some of this uninitialized data could be written to disk. This was flagged by valgrind in the dtransform test and does not appear to be a common occurrence (it is flagged in one test out of the entire HDF5 test suite). Switching to calloc fixed the problem. (DER - 2019/02/03, HDFFV-10694) - There was missing protection against division by zero reported to The HDF Group as issue #CVE-2018-17434. Protection against division by zero was added to address the issue #CVE-2018-17434. (BMR - 2019/01/29, HDFFV-10586) - The issue CVE-2018-17437 was reported to The HDF Group Although CVE-2018-17437 reported a memory leak, the actual issue was invalid read. It was found that the attribute name length in an attribute message was corrupted, which caused the buffer pointer to be advanced too far and later caused an invalid read. A check was added to detect when the attribute name or its length was corrupted and report the potential of data corruption. (BMR - 2019/01/29, HDFFV-10588) - H5Ewalk did not stop when it was supposed to H5Ewalk was supposed to stop when the callback function stopped even though the errors in the stack were not all visited, but it did not. This problem is now fixed. (BMR - 2019/01/29, HDFFV-10684) - Revert H5Oget_info* and H5Ovisit* functions In 1.10.3 new H5Oget_info*2 and H5Ovisit*2 functions were added for performance. Inadvertently, the original functions; H5Oget_info, H5Oget_info_by_name, H5Oget_info_by_idx, H5Ovisit, H5Ovisit_by_name were versioned to H5Oget_info*1 and H5Ovisit*1. This broke the API compatibility for a maintenance release. The original functions have been restored. (ADB - 2019/01/24, HDFFV-10686) - Fixed a potential invalid memory access and failure that could occur when decoding an unknown object header message (from a future version of the library). (NAF - 2019/01/07) - Deleting attributes in dense storage The library aborts with "infinite loop closing library" after attributes in dense storage are created and then deleted. When deleting the attribute nodes from the name index v2 B-tree, if an attribute is found in the intermediate B-tree nodes, which may be merged/redistributed in the process, we need to free the dynamically allocated spaces for the intermediate decoded attribute. (VC - 2018/12/26, HDFFV-10659) - There was missing protection against division by zero reported to The HDF Group as issue #CVE-2018-17233. Protection against division by zero was added to address the issue #CVE-2018-17233. In addition, several similar occurrences in the same file were fixed as well. (BMR - 2018/12/23, HDFFV-10577) - Fixed an issue where the parallel filters tests would fail if zlib was not available on the system. Until support can be added in the tests for filters beyond gzip/zlib, the tests will be skipped if zlib is not available. (JTH - 2018/12/05) - A bug was discovered in the parallel library where an application would eventually consume all of the available MPI communicators when continually writing to a compressed dataset in parallel. This was due to internal copies of an HDF5 File Access Property List, which each contained a copy of the MPI communicator, not being closed at the end of each write operation. This problem was exacerbated by larger numbers of processors. (JTH - 2018/12/05, HDFFV-10629) Java Library: ---------------- - Fortran -------- - Fixed issue with Fortran not returning h5o_info_t field values meta_size%attr%index_size and meta_size%attr%heap_size. (MSB, 2019/01/08, HDFFV-10443) - Added symbolic links libhdf5_hl_fortran.so to libhdf5hl_fortran.so and libhdf5_hl_fortran.a to libhdf5hl_fortran.a in hdf5/lib directory for autotools installs. These were added to match the name of the files installed by cmake and the general pattern of hl lib files. We will change the names of the installed lib files to the matching name in the next major release. (LRK - 2019/01/04, HDFFV-10596) - Made Fortran specific subroutines PRIVATE in generic procedures. Affected generic procedures were functions in H5A, H5D, H5P, H5R and H5T. (MSB, 2018/12/04, HDFFV-10511) Tools ----- - High-Level APIs: ------ - Fortran High-Level APIs: ------ - Documentation ------------- - F90 APIs -------- - C++ APIs -------- - Testing ------- - Fixed a test failure in testpar/t_dset.c caused by the test trying to use the parallel filters feature on MPI-2 implementations. (JTH, 2019/2/7) Supported Platforms =================== Linux 2.6.32-696.16.1.el6.ppc64 gcc (GCC) 4.4.7 20120313 (Red Hat 4.4.7-18) #1 SMP ppc64 GNU/Linux g++ (GCC) 4.4.7 20120313 (Red Hat 4.4.7-18) (ostrich) GNU Fortran (GCC) 4.4.7 20120313 (Red Hat 4.4.7-18) IBM XL C/C++ V13.1 IBM XL Fortran V15.1 Linux 3.10.0-327.10.1.el7 GNU C (gcc), Fortran (gfortran), C++ (g++) #1 SMP x86_64 GNU/Linux compilers: (kituo/moohan) Version 4.8.5 20150623 (Red Hat 4.8.5-4) Version 4.9.3, Version 5.2.0 Intel(R) C (icc), C++ (icpc), Fortran (icc) compilers: Version 17.0.0.098 Build 20160721 MPICH 3.1.4 compiled with GCC 4.9.3 SunOS 5.11 32- and 64-bit Sun C 5.12 SunOS_sparc (emu) Sun Fortran 95 8.6 SunOS_sparc Sun C++ 5.12 SunOS_sparc Windows 7 Visual Studio 2015 w/ Intel Fortran 16 (cmake) Windows 7 x64 Visual Studio 2013 Visual Studio 2015 w/ Intel Fortran 16 (cmake) Visual Studio 2015 w/ Intel C, Fortran 2018 (cmake) Visual Studio 2015 w/ MSMPI 8 (cmake) Windows 10 Visual Studio 2015 w/ Intel Fortran 18 (cmake) Windows 10 x64 Visual Studio 2015 w/ Intel Fortran 18 (cmake) Visual Studio 2017 w/ Intel Fortran 18 (cmake) Mac OS X Yosemite 10.10.5 Apple clang/clang++ version 6.1 from Xcode 7.0 64-bit gfortran GNU Fortran (GCC) 4.9.2 (osx1010dev/osx1010test) Intel icc/icpc/ifort version 15.0.3 Mac OS X El Capitan 10.11.6 Apple clang/clang++ version 7.3.0 from Xcode 7.3 64-bit gfortran GNU Fortran (GCC) 5.2.0 (osx1011dev/osx1011test) Intel icc/icpc/ifort version 16.0.2 Mac OS Sierra 10.12.6 Apple LLVM version 8.1.0 (clang/clang++-802.0.42) 64-bit gfortran GNU Fortran (GCC) 7.1.0 (swallow/kite) Intel icc/icpc/ifort version 17.0.2 Tested Configuration Features Summary ===================================== In the tables below y = tested n = not tested in this release C = Cluster W = Workstation x = not working in this release dna = does not apply ( ) = footnote appears below second table = testing incomplete on this feature or platform Platform C F90/ F90 C++ zlib SZIP parallel F2003 parallel Solaris2.11 32-bit n y/y n y y y Solaris2.11 64-bit n y/n n y y y Windows 7 y y/y n y y y Windows 7 x64 y y/y y y y y Windows 7 Cygwin n y/n n y y y Windows 7 x64 Cygwin n y/n n y y y Windows 10 y y/y n y y y Windows 10 x64 y y/y n y y y Mac OS X Mavericks 10.9.5 64-bit n y/y n y y y Mac OS X Yosemite 10.10.5 64-bit n y/y n y y y Mac OS X El Capitan 10.11.6 64-bit n y/y n y y y Mac OS Sierra 10.12.6 64-bit n y/y n y y y CentOS 7.2 Linux 2.6.32 x86_64 PGI n y/y n y y y CentOS 7.2 Linux 2.6.32 x86_64 GNU y y/y y y y y CentOS 7.2 Linux 2.6.32 x86_64 Intel n y/y n y y y Linux 2.6.32-573.18.1.el6.ppc64 n y/y n y y y Platform Shared Shared Shared Thread- C libs F90 libs C++ libs safe Solaris2.11 32-bit y y y y Solaris2.11 64-bit y y y y Windows 7 y y y y Windows 7 x64 y y y y Windows 7 Cygwin n n n y Windows 7 x64 Cygwin n n n y Windows 10 y y y y Windows 10 x64 y y y y Mac OS X Mavericks 10.9.5 64-bit y n y y Mac OS X Yosemite 10.10.5 64-bit y n y y Mac OS X El Capitan 10.11.6 64-bit y n y y Mac OS Sierra 10.12.6 64-bit y n y y CentOS 7.2 Linux 2.6.32 x86_64 PGI y y y n CentOS 7.2 Linux 2.6.32 x86_64 GNU y y y y CentOS 7.2 Linux 2.6.32 x86_64 Intel y y y n Linux 2.6.32-573.18.1.el6.ppc64 y y y n Compiler versions for each platform are listed in the preceding "Supported Platforms" table. More Tested Platforms ===================== The following platforms are not supported but have been tested for this release. Linux 2.6.32-573.22.1.el6 GNU C (gcc), Fortran (gfortran), C++ (g++) #1 SMP x86_64 GNU/Linux compilers: (mayll/platypus) Version 4.4.7 20120313 Version 4.9.3, 5.3.0, 6.2.0 PGI C, Fortran, C++ for 64-bit target on x86-64; Version 17.10-0 Intel(R) C (icc), C++ (icpc), Fortran (icc) compilers: Version 17.0.4.196 Build 20170411 MPICH 3.1.4 compiled with GCC 4.9.3 Linux 3.10.0-327.18.2.el7 GNU C (gcc) and C++ (g++) compilers #1 SMP x86_64 GNU/Linux Version 4.8.5 20150623 (Red Hat 4.8.5-4) (jelly) with NAG Fortran Compiler Release 6.1(Tozai) GCC Version 7.1.0 OpenMPI 3.0.0-GCC-7.2.0-2.29, 3.1.0-GCC-7.2.0-2.29 Intel(R) C (icc) and C++ (icpc) compilers Version 17.0.0.098 Build 20160721 with NAG Fortran Compiler Release 6.1(Tozai) Linux 3.10.0-327.10.1.el7 MPICH 3.2 compiled with GCC 5.3.0 #1 SMP x86_64 GNU/Linux (moohan) Linux 2.6.32-573.18.1.el6.ppc64 MPICH mpich 3.1.4 compiled with #1 SMP ppc64 GNU/Linux IBM XL C/C++ for Linux, V13.1 (ostrich) and IBM XL Fortran for Linux, V15.1 Debian 8.4 3.16.0-4-amd64 #1 SMP Debian 3.16.36-1 x86_64 GNU/Linux gcc, g++ (Debian 4.9.2-10) 4.9.2 GNU Fortran (Debian 4.9.2-10) 4.9.2 (cmake and autotools) Fedora 24 4.7.2-201.fc24.x86_64 #1 SMP x86_64 x86_64 x86_64 GNU/Linux gcc, g++ (GCC) 6.1.1 20160621 (Red Hat 6.1.1-3) GNU Fortran (GCC) 6.1.1 20160621 (Red Hat 6.1.1-3) (cmake and autotools) Ubuntu 16.04.1 4.4.0-38-generic #57-Ubuntu SMP x86_64 GNU/Linux gcc, g++ (Ubuntu 5.4.0-6ubuntu1~16.04.2) 5.4.0 20160609 GNU Fortran (Ubuntu 5.4.0-6ubuntu1~16.04.2) 5.4.0 20160609 (cmake and autotools) Known Problems ============== 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. Three tests fail with OpenMPI 3.0.0/GCC-7.2.0-2.29: testphdf5 (ecdsetw, selnone, cchunk1, cchunk3, cchunk4, and actualio) t_shapesame (sscontig2) t_pflush1/fails on exit The first two tests fail attempting collective writes. CPP ptable test fails on VS2017 with Intel compiler, JIRA issue: HDFFV-10628. This test will pass with VS2015 with Intel compiler. Older MPI libraries such as OpenMPI 2.0.1 and MPICH 2.1.5 were tested while attempting to resolve the Jira issue: HDFFV-10540. The known problems of reading or writing > 2GBs when using MPI-2 was partially resolved with the MPICH library. The proposed support recognizes IO operations > 2GB and if the datatype is not a derived type, the library breaks the IO into chunks which can be input or output with the existing MPI 2 limitations, i.e. size reporting and function API size/count arguments are restricted to be 32 bit integers. For derived types larger than 2GB, MPICH 2.1.5 fails while attempting to read or write data. OpenMPI in contrast, implements MPI-3 APIs even in the older releases and thus does not suffer from the 32 bit size limitation described here. OpenMPI releases prior to v3.1.3 appear to have other datatype issues however, e.g. within a single parallel test (testphdf5) the subtests (cdsetr, eidsetr) report data verfication errors before eventually aborting. The most recent versions of OpenMPI (v3.1.3 or newer) have evidently resolved these isses and parallel HDF5 testing does not currently report errors though occasional hangs have been observed. 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. CMake vs. Autotools installations ================================= While both build systems produce similar results, there are differences. Each system produces the same set of folders on linux (only CMake works on standard Windows); bin, include, lib and share. Autotools places the COPYING and RELEASE.txt file in the root folder, CMake places them in the share folder. The bin folder contains the tools and the build scripts. Additionally, CMake creates dynamic versions of the tools with the suffix "-shared". Autotools installs one set of tools depending on the "--enable-shared" configuration option. build scripts ------------- Autotools: h5c++, h5cc, h5fc CMake: h5c++, h5cc, h5hlc++, h5hlcc The include folder holds the header files and the fortran mod files. CMake places the fortran mod files into separate shared and static subfolders, while Autotools places one set of mod files into the include folder. Because CMake produces a tools library, the header files for tools will appear in the include folder. The lib folder contains the library files, and CMake adds the pkgconfig subfolder with the hdf5*.pc files used by the bin/build scripts created by the CMake build. CMake separates the C interface code from the fortran code by creating C-stub libraries for each Fortran library. In addition, only CMake installs the tools library. The names of the szip libraries are different between the build systems. The share folder will have the most differences because CMake builds include a number of CMake specific files for support of CMake's find_package and support for the HDF5 Examples CMake project.