Skip to content

Latest commit

 

History

History
920 lines (576 loc) · 34.7 KB

RELEASE_NOTES.md

File metadata and controls

920 lines (576 loc) · 34.7 KB
Raw

Release notes for C-Blosc2

Changes from 2.15.1 to 2.15.2

#XXX version-specific blurb XXX#

Changes from 2.15.0 to 2.15.1

  • Do not pass -m flags when compiling shuffle.c. This prevents the compiler from incidentally optimizing the code called independently of the runtime CPU check to these instruction sets, effectively causing SIGILL on other CPUs. Fixes #621. Thanks to @t20100 and @mgorny.

  • Internal LZ4 sources bumped to 1.10.0.

  • Allow direct loading of plugins by name, without relying on the presence of python. Thanks to @boxerab.

  • Add b2nd_nans method (PR #624). Thanks to @waynegm.

Changes from 2.14.4 to 2.15.0

  • Removed some duplicated functions. See #503.

  • Added a new io mode to memory map files. This forced to change the io_cb read API. See https://github.com/Blosc/c-blosc2/blob/main/tests/test_mmap.c to see an example on how to use it.

  • Updated the SOVERSION to 4 due to the API change in io_cb read.

  • Added functions to get cparams, dparams, storage and io defaults respectively.

  • Internal zstd sources updated to 1.5.6.

  • Fixed a bug when setting a slice using prefilters.

Changes from 2.14.3 to 2.14.4

  • Bumped SONAME due to recent API changes. See #581.

Changes from 2.14.2 to 2.14.3

  • More fixes for internal fuzzer.

Changes from 2.14.1 to 2.14.2

  • Fixes for CVE-2024-3203 and CVE-2024-3204.

Changes from 2.14.0 to 2.14.1

  • When loading plugins, first try with python and then python3. This is because many linux distros do not have python as a symlink to python3 anymore.

Changes from 2.13.2 to 2.14.0

  • Fixed a bug preventing buffers to be appended to empty (0-sized) b2nd arrays.

  • New acceleration path for b2nd_append(). This new path is much faster (up to 4x) than the previous one, specially for large arrays. See bench/bench_stack_append.c for the bench of use.

  • New examples for using the b2nd_set_slice_cbuffer() and b2nd_append() functions for adding data into existing b2nd arrays. See examples/example_stack_images.c.

  • Now, python3 is used for finding plugins instead of python. This is because many linux distros do not have python as a symlink to python3 anymore.

  • New round of fixing warnings. Now, C-Blosc2 should be relatively free of them.

  • Small performance tweak for clevel 1 in BloscLZ codec.

  • Fixed a leak in frame code. Closes #591. Thanks to @LuMingYinDetect.

  • Disable shuffle repeat in filters pipeline. This was broken since the initial implemented, and it was never documented. Also, compression ratios do not seem to be improved in our experiments, so this capability has been removed completely.

  • Support for new Intel compilers (2023.0.1 and on). Fixes #533. Thanks to Nick Papior.

Changes from 2.13.1 to 2.13.2

Changes from 2.13.0 to 2.13.1

  • Removed private include in b2nd.h. This fixes issue #579.

Changes from 2.12.0 to 2.13.0

  • Added a new BLOSC_FILTER_INT_TRUNC filter for truncating integers to a given number of bits. This is useful for compressing integers that are not using all the bits of the type. See PR #577.

  • Optimized zstd, specially when using dicts. See PR #578.

  • Initialize grok library when loading the plugin. This is needed for other plugins to be able to use it without the need of importing the package.

Changes from 2.11.3 to 2.12.0

  • New blosc2_get_slice_nchunks function for getting the unidimensional chunk indexes of a Blosc2 container slice.

  • Globally registered new codec grok. This will be loaded dynamically.

Changes from 2.11.2 to 2.11.3

  • Frames accept now typesizes that are larger than 255 (and up to 2**31). See PyTables/PyTables#1086. Thanks to @chris-allan for the report.

  • AVX512 runtime dispatching has been fixed (basically disabled) for GCC versions <= 10.

  • Use typedef for blosc_timestamp_t. Thanks to Magnus Ulimoen.

Changes from 2.11.1 to 2.11.2

  • Added support for ARMv7l platforms (Raspberry Pi). The NEON version of the bitshuffle filter was not compiling there, and besides it offered no performance advantage over the generic bitshuffle version (it is 2x to 3x slower actually). So bitshuffle-neon.c has been disabled by default in all ARM platforms.

  • Also, unaligned access has been disabled in all ARM non-64bits platforms. It turned out that, at least the armv7l CPU in Raspberry Pi 4, had issues because __ARM_FEATURE_UNALIGNED C macro was asserted in the compiler (both gcc and clang), but it actually made binaries to raise a "Bus error".

  • Thanks to Ben Nuttall for providing a Raspberry Pi for tracking down these issues.

Changes from 2.11.0 to 2.11.1

  • Fix ALTIVEC header. Only affects to IBM POWER builds. Thanks to Michael Kuhn for providing a patch.

Changes from 2.10.5 to 2.11.0

  • New AVX512 support for the bitshuffle filter. This is a backport of the upstream bitshuffle project (https://github.com/kiyo-masui/bitshuffle). Expect up to 20% better compression speed on AMD Zen4 architecture (7950X3D CPU).

  • Add c-blosc2 package definition for Guix. Thanks to Ivan Vilata.

  • Properly check calls to strtol. Fixes #558.

  • Export the b2nd_copy_buffer function. This may be useful for other projects dealing with multidimensional arrays in memory. Thanks to Ivan Vilata.

  • Better check that nthreads must be >= 1 and <= INT16_MAX. Fixes #559.

  • Fix compile arguments for armv7l. Thanks to Ben Greiner.

Changes from 2.10.4 to 2.10.5

  • Fix a variable name in a test that was causing a segfault in some platforms.

  • Change tuner's functions signature to return always an error code. This allows for better error checking when using pluggable tuners in Blosc2.

  • Do checks when creating contexts.

Changes from 2.10.3 to 2.10.4

  • Remove duplicated tune initialization since it is already done in blosc2_create_cctx. Thanks to Marta Iborra

  • Typos fixed. Thanks to Dimitri Papadopoulos.

Changes from 2.10.2 to 2.10.3

  • Globally registered new codec openhtj2k. This will be loaded dynamically. See PR #557.

  • Added a BLOSC_INFO macro for details on compression params.

  • Added get_blocksize.c example on automatic blocksizes.

  • Warning fixes.

  • Fixes for mingw.

Changes from 2.10.1 to 2.10.2

  • Several fixes for the CMake system. Thanks to Axel Huebl. See PR #541 and #542.

  • Several fixes for mingw platform. Thanks to Biswapriyo Nath. See PR #540 and #543.

Changes from 2.10.0 to 2.10.1

  • blosc2_remove_urlpath(const char *urlpath) does not return an error when path does not exist.

  • Changes in CMake installer to conserve targets and properties on install, so CMake users do not need to write FindBlosc2.cmake files anymore. This also helps to preserve transitive dependencies on CMake targets, especially useful for fully static builds, e.g., for Python wheels. Thanks to @ax3l (Axel Huebl). See PR #537.

  • Fix new typos. Thanks to @DimitriPapadopoulos. See PR #538.

Changes from 2.9.3 to 2.10.0

  • bytedelta filter has been fixed. For backward compatibility, the old bytedelta filter is still available as BLOSC_FILTER_BYTEDELTA_BUGGY symbol, with the same ID (34) than before. The new, fixed bytedelta filter has received a new ID (35) and it can be used via the usual BLOSC_FILTER_BYTEDELTA symbol. That means that old data written with the buggy bytedelta filter should be decompressed without issues. Thanks to @foody (Tom Birch) for the fix. See #531, #532 for more info.

  • Filter buffers are correctly cycled now. Now it is possible to use e.g. shuffle and bitshuffle filters in the pipeline. Thanks to @foody (Tom Birch) for the fix. See #528 and PR #530.

  • Assorted fixes for allowing better inclusion in external projects. Thanks to @ax3l (Axel Huebel). See #525, #527 and #529.

  • Minor fixes in the documentation. Thanks to @ivilata (Ivan Vilata). See #523.

Changes from 2.9.2 to 2.9.3

  • Thanks to Dimitri Papadopoulos for an extensive set of improvements in documentation and code.

  • load_lib is now a private function. Before was public, but never meant to be.

  • Several fixes for bugs discovered by the fuzzer.

Changes from 2.9.1 to 2.9.2

  • Now is possible to register the same plugin (as long as they have the same ID and name) without errors. This is useful for registering the same plugin without worrying on whether it has been registered already.

  • Improved detection of dynamic plugin locations. Now they must implement plugin_module.print_libpath() as the canonical way to find the path for the dynamic library plugin.

  • The blosc2_static has gained the cmake POSITION_INDEPENDENT_CODE property. This should allow to use the static library in more situations.

  • BLOSC_STUNE is defined in blosc2.h now. Fixes #481. Thanks to @DimitriPapadopoulos.

  • Fixed an issue when having incompressible data in combination with lazy_chunks.

  • Fix linking with static -DBUILD_STATIC=0. Fixes #480.

  • Visual Studio 2010 (version 10.0) has been deprecated. Now, users will need to use Visual Studio 2012 (version 11.0) or later.

  • Many small fixes and code improvements. Thanks to @DimitriPapadopoulos, @bnavigator.

Changes from 2.9.0 to 2.9.1

  • Allow the use of BTUNE by detecting the BTUNE_BALANCE environment variable.

Changes from 2.8.0 to 2.9.0

  • Dynamic plugins as Python wheels are supported now! This new feature allows for creating plugins in C, distribute them as wheels, and load them dynamically in runtime. Small example at https://github.com/Blosc/blosc2_plugin_example

  • BloscLZ can achieve more speed in clevel 1 now.

  • Internal Zstd sources updated to latest 1.5.5 version.

  • Copyright notice updated. Thanks to @DimitriPapadopoulos.

Changes from 2.7.1 to 2.8.0

  • New bytedelta filter added. SIMD support for Intel and ARM platforms is there. We have blogged about this: https://www.blosc.org/posts/bytedelta-enhance-compression-toolset.rst Thanks to Aras Pranckevičius for inspiration and initial implementation.

  • Minor improvements in BloscLZ, leading to better compression ratios in general. BLoscLZ version bumped to 2.5.2.

  • Updated internal zlib-ng to 2.0.7.

  • Used const qualifier where possible in b2nd. Thanks to @cf-natali.

Changes from 2.6.1 to 2.7.1

  • Caterva has been merged and carefully integrated in C-Blosc2 in the new b2nd interface. For more info on the new interface, see https://www.blosc.org/c-blosc2/reference/b2nd.html. Thanks to Marta Iborra, Oscar Guiñón, J. David Ibáñez and Francesc Alted. Also thanks to Aleix Alcacer for his great work in the Caterva project.

    We have a blog about this: https://www.blosc.org/posts/blosc2-ndim-intro

  • Updated internal zstd sources to 1.5.4. Thanks to Dimitri Papadopoulos.

  • blosc2_schunk_avoid_cframe_free and blosc2_schunk_append_file are exported as public functions now. Thanks to @bnavigator.

  • BloscLZ codec is now treated exactly the same as LZ4. Before BloscLZ was considered less capable of reaching decent compression ratios, but this has changed quite a bit lately, so there is no point in treating both differently.

  • Fixed some leaks, mainly on the test suite.

  • Fixed quite a bit of compiler warnings.

Changes from 2.6.0 to 2.6.1

  • Add support for macos universal2 binaries (arm64+x86_64 build). Thanks to Thomas Vincent.

Changes from 2.5.0 to 2.6.0

  • [API] Now it is possible to pass filter ID to a User Defined Filter.

  • Unified convention for BLOSC_SPLITMODE environment variable in Blosc and Blosc2. The list of valid values is "ALWAYS", "NEVER", "AUTO" and "FORWARD_COMPAT" now.

  • Unified convention for BLOSC_SPLITMODE enum in Blosc and Blosc2 headers.

Changes from 2.4.3 to 2.5.0

  • Fixed a nasty bug that prevented retrieving data correctly with large super-chunks (> 2^31 elements).

  • Fixed an issue in blosc2_schunk_get_slice_buffer() in the interpretation of the stop param. Now stop is not part of the selected slice (as advertised).

  • Now blosc2_create_cctx() supports the same environment variables than blosc2_compress().

  • Now blosc2_create_dctx() supports the same environment variables than blosc2_decompress().

  • Added support for the split mode to be serialized in cframes/sframes.

  • A new splitmode field has been added to the blosc2_schunk structure.

  • Changed some fields in blosc2_preparams and blosc2_postparams structs:

    • in -> input
    • out -> output
    • out_size -> output_size
    • out_typesize -> output_typesize
    • out_offset -> output_offset This was needed to allow Cython to map the fields (in is a reserved word in Python).

  • Disabled maskout reads in blosc2_schunk_get_slice_buffer() as they are not faster than getitem there.

  • Add an intermediate block size in its automatic calculation based on clevel.

Changes from 2.4.2 to 2.4.3

  • Disable automatic split of blocks when not using shuffle. Experiments are showing that cratio is suffering too much, specially when using BloscLZ.

  • Changed xxhash.h and xxhash.c to the newest version. Thanks to Dimitri Papadopoulos.

Changes from 2.4.1 to 2.4.2

  • Fixed BLOSC1_COMPAT mode. If the symbol BLOSC1_COMPAT is defined, one should actually get a correct map for the BLOSC1 API now. Thanks to @MehdiChinoune for pointing this out.

  • Optimizations for blosc2_schunk_get_slice_buffer(). Now more caution has been put in avoiding memcpy's as much as possible.

  • Now blosc2_set_nthreads() also modifies the schunk compression and decompression defaults.

  • Several other fixes, specially when converting schunks to cframe format.

Changes from 2.4.0 to 2.4.1

  • New blosc2_schunk_avoid_cframe_free() for avoiding the free of a cframe when destroying a super-chunk. Mainly useful for situations where you build a super-chunk out of an existing cframe, so you don't want it to be freed automatically.

Changes from 2.3.1 to 2.4.0

  • New blosc2_schunk_get_slice_buffer() and blosc2_schunk_set_slice_buffer() functions for getting and setting slices from/to a super-chunk.

Changes from 2.3.0 to 2.3.1

  • Support for negative values for BLOSC_TRUNC_PREC filter. Negatives values mean reduce mantissa precision bits, whereas positive values mean keep precision bits.

  • Re-add the check for small buffers and a new test (see Blosc/python-blosc2#46).

  • Make static a couple of funcs to avoid collisions. This can be useful in case someone tries to compile both C-Blosc and C-Blosc2 in the same name space (however, don't do that unless you know what you are doing; better use dynamic libraries which allow for a much better name space separation).

Changes from 2.2.0 to 2.3.0

  • [API change] In order to allow to compile with both C-Blosc and C-Blosc2 libraries, a new API has been created for the symbols and function names that had collisions. Here are the changed symbols and functions:

    • Blosc2 symbols that take different values than in Blosc1:

      • BLOSC_VERSION_MAJOR -> BLOSC2_VERSION_MAJOR
      • BLOSC_VERSION_MINOR -> BLOSC2_VERSION_MINOR
      • BLOSC_VERSION_RELEASE -> BLOSC2_VERSION_RELEASE
      • BLOSC_VERSION_STRING -> BLOSC2_VERSION_STRING
      • BLOSC_VERSION_DATE -> BLOSC2_VERSION_DATE
      • BLOSC_MAX_OVERHEAD -> BLOSC2_MAX_OVERHEAD
      • BLOSC_MAX_BUFFERSIZE -> BLOSC2_MAX_BUFFERSIZE

    • Original Blosc1 API that takes the blosc1_ prefix:

      • blosc_compress -> blosc1_compress
      • blosc_decompress -> blosc1_decompress
      • blosc_getitem -> blosc1_getitem
      • blosc_get_compressor -> blosc1_get_compressor
      • blosc_set_compressor -> blosc_set_compressor
      • blosc_cbuffer_sizes -> blosc1_cbuffer_sizes
      • blosc_cbuffer_validate -> blosc1_cbuffer_validate
      • blosc_cbuffer_metainfo -> blosc1_cbuffer_metainfo
      • blosc_get_blocksize -> blosc1_get_blocksize
      • blosc_set_blocksize -> blosc1_set_blocksize
      • blosc_set_splitmode -> blosc1_set_splitmode

    • API that has been migrated to blosc2_ prefix

      • blosc_init -> blosc2_init
      • blosc_destroy -> blosc2_destroy
      • blosc_free_resources -> blosc2_free_resources
      • blosc_get_nthreads -> blosc2_get_nthreads
      • blosc_set_nthreads -> blosc2_set_nthreads
      • blosc_compcode_to_compname -> blosc2_compcode_to_compname
      • blosc_compname_to_compcode -> blosc2_compname_to_compcode
      • blosc_list_compressors -> blosc2_list_compressors
      • blosc_get_version_string -> blosc2_get_version_string
      • blosc_get_complib_info -> blosc2_get_complib_info
      • blosc_cbuffer_versions -> blosc2_cbuffer_versions
      • blosc_cbuffer_complib -> blosc2_cbuffer_complib

    It is recommended to migrate to the new API as soon as possible. In the meanwhile, you can still compile with the previous API (corresponding to C-Blosc2 pre-2.3.0), by defining the BLOSC1_COMPAT symbol in your C-Blosc2 app (before including the 'blosc2.h' header).

  • Fixed some issues in converting from super-chunks to frames and back. Now it is possible to do a roundtrip without (known) problems.

  • LZ4 codec has been bumped to 1.9.4.

Changes from 2.1.1 to 2.2.0

  • Added new blosc2_schunk_open_offset() and blosc2_schunk_append_file() functions for being able to open a super-chunk inside of file, and append a super-chunk to the end of an existing file. See #409.

  • Protect the update of a field in compression/decompression context. Fixes issues when compressing/decompressing super-chunks in multi-thread mode.

  • Fix issue when inserting a chunk in a super-chunk. See #408.

  • Fix issue when appending a special chunk in a super-chunk. See #407.

  • Optimized the blockshape calculation when using the ZFP plugin in BLOSC_CODEC_ZFP_FIXED_RATE mode. See #406.

  • New blosc2_unidim_to_multidim and blosc2_multidim_to_unidim which are useful for codecs that are meant to deal with multidimensional arrays (like ZFP).

  • Another round of squashing warnings has been carried out. Thanks to Marta Iborra.

  • Added locks in situations where different threads were trying to update the same variable. Thanks to Marta Iborra (and helgrind!).

  • Use proper types for native zlib-ng interface. This allows for linking with zlib-ng native API. Thanks to Klaus Zimmermann.

Changes from 2.1.0 to 2.1.1

  • Fixed a nasty bug introduced in 2.1.0 that affects to the blosclz codec in ARM arch. See #392.

  • Added support for external zlib-ng (see #393). Thanks to Mark Harfouche.

  • Fixed more compiler warnings (#394). Now C-Blosc2 should be relatively free of them.

  • Fixed more fuzzer issues.

Changes from 2.0.4 to 2.1.0

  • Support of the ZFP lossy codec as a plugin. This is mainly meant for compressing multidimensional arrays of floating point data, so it is better used in combination with Caterva. More info at our blog: https://www.blosc.org/posts/support-lossy-zfp/. Thanks to Oscar Guiñon and Francesc Alted.

  • Massive cleanup of warnings. Thanks to Marta Iborra.

  • Fixed some bugs in test updates and insertions in 64-bit super chunks. Thanks to Francesc Alted.

  • Fixed the build on FreeBSD/powerpc64le. Thanks to @pkubaj.

  • Updated internal zstd sources to 1.5.2.

  • Updated internal zlib-ng to 2.0.6.

Changes from 2.0.3 to 2.0.4

  • Restored support for type size that is not a divisor of a buffer size for more info. See #356.

  • Implemented a blosc2_rename_urlpath, a portable function to rename a file or a full directory. See #355.

  • Several improvements for packaging. See #354, #357, #359.

  • Fixed a bug in BloscLZ codec (bumped to 2.5.1).

Changes from 2.0.2 to 2.0.3

  • Improved BloscLZ codec (bumped to 2.5.0) for achieving better compression ratios on data with high entropy. Although in general LZ4 still does a better job in this scenario, BloscLZ can sometimes achieve better compression ratios.

  • Added blosc2_vlmeta_delete() for removing vlmeta data. Thanks to Marta Iborra.

  • Update pkg-config file to support blosc2. Fixes #236. Thanks to Håvard Flaget Aasen.

  • Build system: Change hard coded library path with CMAKE_INSTALL_LIBDIR CMake variable. Thanks to Håvard Flaget Aasen.

Changes from 2.0.1 to 2.0.2

  • Fixed data chunk memory leaks in frame (see #335).

  • Fixed blosc2_stdio_open never returns NULL if it cannot open file.

  • Standardized places for headers in blosc/ and include/ dirs.

  • nthreads is int16_t everywhere in the API. Fixes #331.

  • Add blosc2_remove_urlpath function (see #330).

  • Fixed a bug when a lazy_chunk was created from a small, memcpyed chunk. (see #329).

  • Fixed many issues in documentation (see #333).

Changes from 2.0.0 to 2.0.1

  • The blosc2_schunk_fill_special function was not exported, so not in the shared library. This has been fixed; see #328. Thanks to Mark Kittisopikul.

Changes from 2.0.0.rc2 to 2.0.0 (final)

  • Now Blosc is always compiled with LZ4. See #324.

  • Implemented a system to register plugins (PR #314). See our blog at: https://www.blosc.org/posts/registering-plugins.

  • Added Blosc Lite version. Just activate BUILD_LITE cmake option with: -DBUILD_LITE. See #316.

  • You can deactivate the plugins by setting cmake option BUILD_PLUGINS to OFF.

  • Created include folder. See #310.

  • Moved codecs-registry.h and filters-registry.h to include/blosc2. See #325.

  • Fix error in endian-handler function affecting frames metadata. See #320.

  • Improved tolerance to Win64 workflows failure. See #319.

  • zlib-ng updated to 2.0.5.

  • New COMPILING_WITH_WHEELS.rst doc added.

Changes from 2.0.0.rc.1 to 2.0.0.rc2

  • New compatibility with MinGW32/64. See #302.

  • Improved support for AArch64 (aka ARM64), ARMv7l, ARMv6l and powerpc64le. See #306. Thanks to Alexandr Romanenko, Mark Kittisopikul and Mosè Giordano from the Julia packaging team for their help.

  • BloscLZ updated to 2.4.0. Aligned access in ARM has been enabled, as well as other performance improvements. Expect much better performance, specially on ARM platforms (like Apple Silicon).

  • Zstd sources updated to 1.5.0.

  • zlib-ng sources updated to 2.0.3.

Changes from 2.0.0-beta.5 to 2.0.0.rc.1

  • [API change] blosc2_decompress_ctx() gets a new srcsize parameter to ensure that it does not read past the end of the provided buffer. See #144. Thanks to Nathan Moinvaziri (@nmoinvaz).

  • [BREAKING CHANGE] The format for frames has changed and BLOSC2_VERSION_FRAME_FORMAT is now set to 2. There is no attempt to support previous formats, but there will probably be backward compatibility support starting from version 2 on.

  • New functionality for updating, inserting and deleting chunks in a super-chunk.

  • Support for special values. Large sequences of repeated values can be represented with an efficient, simple and fast run-length representation, without the need to use regular codecs.

  • Internal Zstd sources updated to 1.4.9.

  • Internal LZ4 sources updated to 1.9.3.

  • Internal zlib support is provided now by new zlib-ng 2.0.2 (replacing miniz).

  • The support for Snappy codec has been completely removed. Snappy is a C++ library, which is not good for a library that aims to be fully pure C. Snappy was removed from sources in C-Blosc(1) some years ago, so there should not be a lot of data compressed with Blosc/Snappy out there (and for the existing ones, a transcoding is always possible using C-Blosc(1)).

  • The Lizard codec has been removed. Lizard is a pretty good one, but it looks like it is not timely maintained. Zstd/Zlib can cover its place pretty nicely.

  • The split of blocks only happens for BLOSCLZ and LZ4 codecs. All the rest are not split at all.

  • Public APIs for frames have been removed. Frames should be considered an storage detail, so having them accessible publicly should only bring unnecessary cognitive load. Care have been carried out so as to ensure the same functionality via the super-chunk (schunk) API.

  • [FORMAT] New sparse frame format for on-disk I/O. This allows for storing data chunks in super-chunks in separate files inside a directory. The way to generate sparse frames is via storage.contiguous=false and storage.urlpath=dirname. See README_SFRAME_FORMAT.rst for details.

Changes from 2.0.0-beta.4 to 2.0.0.beta5

  • The prefilter functionality has been introduced and declared stable. With that, you can callback arbitrary functions previous to any filter. This can be useful for performing (parallel) computations on chunks. For an example of use, see tests/test_prefilter.c.

  • New blosc2_set_maskout() function to avoid decompressing blocks. This can be handy when it is not needed to decompress all the blocks in a chunk. This should be always called before blosc2_decompress_ctx() and its effect is reset to the default (decompress all blocks) after that.

  • New ALTIVEC optimizations for Power architecture. These include support for both shuffle and bitshuffle filters. For details, see #98. Thanks to Jerome Kieffer and ESRF for making this happen.

  • New frame_from_bframe() function for getting a blosc2_frame out of an in-memory serialized frame.

  • Zstd codec updated to 1.4.4.

Changes from 2.0.0-beta.3 to 2.0.0-beta.4

  • New pluggable threading backend. Instead of having Blosc use its own thread pool, you can instead call blosc2_set_threads_callback(threads_callback, callback_data) to install your own threading backend. This gives Blosc the possibility to use the same threading mechanism as one you are using in the rest of your program (e.g. OpenMP or Intel TBB), sharing the same threads, rather than starting its own threads that compete with yours for the CPU cores. See #PR 81. Thanks to Steven G. Johnson.

  • The endianness of the platform that is writing the data in chunks is stored now in the headers of the chunks. This info is not used yet, but this should allow a good hint for implementing format compatibility among platforms with different endianness in other layers. See PR #84.

  • Fixed a nasty bug that prevented frames to go more than 2 GB in size.

  • Added a cache for on-disk offsets. This accelerates the reading of slices from disk quite a lot (up to 50% with my benchmarks).

  • Zstd codec upgraded from 1.4.0 to 1.4.3.

Changes from 2.0.0-beta.2 to 2.0.0-beta.3

  • Quick release to fix that beta.2 was tagged on top of a branch, not master.

  • The msgpack trailer now properly starts with 0x90 + 4 value, not plain wrong 0x09 + 4.

  • Trailer version bumped to 1.

Changes from 2.0.0-beta.1 to 2.0.0-beta.2

  • A new vlmetalayers chunk in schunk allows to store arbitrary meta-information that is up to the user. If the schunk has an attached frame, the later will be updated accordingly too. For more info, see PR #74 and docstrings of new blosc2_update_vlmetalayer() and blosc2_get_vlmetalayer() functions.

  • Metalayers must now be attached to super-chunks, not frames. The reason is that frames are increasingly treated as a storage specifier (in-memory or disk now, but can be other means in the future), whereas the actual API for I/O (including metainfo) goes into super-chunks. See PR #75.

  • New frame format documented in README_CFRAME_FORMAT.rst. Remember that the frame format is not written in stone yet, so some changes may be introduced before getting out of beta.

  • BREAKING CHANGE: the format for frames has changed and BLOSC2_VERSION_FRAME_FORMAT is now set to 1. There is no attempt to support previous formats, but there will probably be backward compatibility support starting from version 1 on.

  • BREAKING CHANGE: the next APIs have been renamed:

    • blosc2_frame_has_metalayer -> blosc2_has_metalayer
    • blosc2_frame_add_metalayer -> blosc2_add_metalayer
    • blosc2_frame_update_metalayer -> blosc2_update_metalayer
    • blosc2_frame_metalayer -> blosc2_get_metalayer

    Although the API was declared stable in beta.1, the fact that metalayers are attached now to super-chunks directly, made this change completely necessary.

  • BREAKING CHANGE: the next symbols have been renamed:

    • BLOSC_CPARAMS_DEFAULTS -> BLOSC2_CPARAMS_DEFAULTS
    • BLOSC_DPARAMS_DEFAULTS -> BLOSC2_DPARAMS_DEFAULTS

Changes from 2.0.0a5 to 2.0.0-beta.1

  • The library is called now blosc2 and not blosc anymore. This is necessary so as to prevent collisions with existing blosc deployments.

  • The make install now install all the necessary requirements out-of-the-box.

  • Use Intel IPP's LZ4Safe when compressing/decompressing: this provides better compression ratios and speed (in some cases). It is activated automatically if Intel IPP is found in the system, but you can always disable it with: cmake -DDEACTIVATE_IPP=ON

  • BREAKING CHANGE: the next API have been made private:

    • blosc2_frame_append_chunk -> frame_append_chunk
    • blosc2_frame_get_chunk -> frame_get_chunk
    • blosc2_frame_decompress_chunk -> frame_decompress_chunk

    Now the appending and retrieval of data in frames should always be made via the frame-backed super-chunk API. The idea is to deduplicate the I/O primitives as much as possible, and the super-chunks are the logical way for doing this.

  • BREAKING CHANGE: the next APIs have been renamed:

    • blosc2_get_cparams - > blosc2_schunk_get_cparams
    • blosc2_get_dparams - > blosc2_schunk_get_dparams

  • Internal BloscLZ sources bumped to 2.0.0.

  • Internal LZ4 sources bumped to 1.9.1.

  • Internal Zstd sources bumped to 1.4.0.

Changes from 2.0.0a4 to 2.0.0a5

  • Delta filter now implemented as a XOR instead of a NEG for better numerical stability.

  • Zstd dictionaries are supported. This allows for (much) better compression ratios when using small blocksizes, while improving compression speed.

  • Performance upgrades in the BloscLZ codec, specially for decompression in ARM architectures.

  • Preliminary support for Neon instruction set in ARM CPUs. Shuffle seems to work well, but still some issues with bitshuffle.

  • Internal LZ4 sources bumped to 1.8.2.

  • Internal Zstd sources bumped to 1.3.4.

Changes from 2.0.0a3 to 2.0.0a4

  • New filter pipeline designed to work inside a chunk. The pipeline has a current capacity of 5 slots, and is designed to apply different filters sequentially over the same chunk. For this, a new extended header for the chunk has been put in place (see README_HEADER.rst).

  • New delta filter meant to work inside a chunk. Previously delta was working inside a super-chunk, but the new implementation is both faster and simpler and gets better compression ratios (at least for the tested datasets).

  • New re-parametrization of BloscLZ for improved compression ratio and also more speed (at least on modern Intel/AMD CPUs). Version for internal BloscLZ codec bumped to 1.0.6.

  • Internal zstd sources bumped to 1.3.0.

  • The BLOSC2_MAX_OVERHEAD symbol is always 32 bytes, not 16 as in Blosc1. This is needed in order to allow buffers larger than 32 bits (31 actually :).

Changes from 2.0.0a2 to 2.0.0a3

  • New blosc2_create_context() and blosc2_free_context() for creating contexts so that Blosc can be used from threaded applications without the global lock. For now contexts can be only used from blosc2_compress_ctx() and blosc2_decompress_ctx() (see below).

  • The blosc_compress_ctx() and blosc_decompress_ctx() have been replaced by blosc2_compress_ctx() and blosc2_decompress_ctx() that do accept actual contexts.

  • Added support for new Zstd codec (https://github.com/Cyan4973/zstd). This is a new compressor by Yann Collet, the author of LZ4 and LZ4HC. For details on Zstd, see this nice intro: https://fastcompression.blogspot.com/2015/01/zstd-stronger-compression-algorithm.html.

  • The blosc2_append_chunk() has been removed. This is this because an existing chunk may not fulfill the sequence of filters in super header. It is best that the user will use blosc2_schunk_append_buffer() and compress it internally.

  • The split of blocks only happens for BLOSCLZ and SNAPPY codecs. All the rest are not split at all. This allows for faster operation for the rest of codecs (most specially zstd).

  • The internal zlib 1.2.8 sources have been replaced by the miniz library, which is meant to be fully compatible with Zlib, but much smaller and besides tends to be a bit faster. Also, miniz is preferred to an external Zlib.

  • The internal Snappy sources have been removed. If the Snappy library is found, the support for it is still there.

  • Internal LZ4 sources upgraded to 1.7.1.

Changes from 2.0.0a1 to 2.0.0a2

  • The delta filter runs inside of the compression pipeline.

  • Many fixes for the delta filter in super-chunks.

  • Added tests for delta filter in combination with super-chunks.