lepton-1.2.1+git.20201230-bp155.2.4

- Remove unneeded BuildRequires on python 2.

- Stop building with march=native (boo#1100677)

- Update to version 1.2.1+git.20201230:
  * fix #122 by setting compiler flags to safer default without march=native optimizations
  * fix #144

- Update to version 1.2.1+git.20190820:
  * Fix #125

- Update to version 1.2.1+git.20190517:
  * Correctly check if CXXFLAGS is set
  * Include <functional> to fix build failure on Windows
  * fix warnings, fix #36
  * fix #111
  * fix #114 by checking for header payloads greater than max file size
  * some additional checks to prevent an infinite loop when presented with certain corrupt files. Also fixing CI.
  * make automake not to install `test_invariants'

- Update to version 1.2.1+git.20180824:
  * only assert similarity of last_dc if segment_size > 1 since the overhang byte will clobber any incorrect value
  * only exit on encode for items that could hazard a divide by zero on decode
  * relax strictness in singlethreaded decompression mode, otherwise we are checking items that multithread mode cannot validate
  * check hpos before reading from hdrs-size array
  * initialize hdr memory to zero when allocating
  * Fix #110 by providing a more explicit error message
  * fix permissive mode on windows

- Update to version 1.2.1+git.20180410:
  * support one version future, so we can roll out smoothly
  * do not trust 16 bit length value to stay below the header length

- Update to version 1.2.1+git.20180111:
  * include version.hh from generic compressor
  * process md5 of entire input, even if subprocess fails. cap out max generic lepton file at 32 megs: users can concatenate files to make larger generic files
  * turn on seccomp for generic lepton compression
  * Relax criteria for Huffman table existence check
  * Fix #101 by allowing a permissive flag to accept any type of nonstandard jpeg and create a valid lepton file to represent a bitexact copy of the nonstandard jpeg
  * integrate concatenation test suite into the make check system
  * test suite for new mechanisms
  * fix concatenation of permissive lepton files
  * better header allowing for multithreaded decoding of the jpeg
  * more permissive with small files and no random SIGPIPE or other hurried subprocess error codes from larger ones
  * prototyping a new version of lepton that can compress arbitrary images or files by having a fallback path to encode data in a passthrough manner
  * unless EOF reached, read next header in addition to the current buffer--this lets us consume all data even if we never have a desire to read past the end
  * put ANS behind an experimental feature flag
  * Have all inter-thread communication use the same length buffers so we can distinguish between extra data after threads have closed and new types of work
  * tweaks to observation and update paremeters to make them slightly faster
  * initialize last pointer for packet reader
  * this worked a little too well.. call me skeptical
  * instantiate some ans functions
  * tests pass by keeping the boolwriters in scope until threads have been joined upon: do so by refactoring function so duplicate code would be in thhreaded_encode_inner and ANSBoolWriter can also lean on that templatized function
  * Add BoolDecoder and BoolEncoder as template arguments rather than as toplevel defines. This required a significant amount of threading of the BoolDecoder through each of the state classes
  * add unit test for ans bool reader and writer
  * move to 512 resolution probability space (considering reducing resolution on 256 instead--but lets get correctness first
  * simplify some of the flushing behavior
  * link to ANS encoder
  * avoid required size from zlib0 plugin

- Update to version 1.2.1+git.20171018:
  * Fix #91 by adding asserts that the garbage data is never more or equal to the full file size
  * Fix #90 by checking for nonzero width/height
  * deal with malformed lepton files that have incorrect ThreadHandoff sizes
  * Fix #95 by adding a workaround for GCC bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81015 that breaks builtin_clz. Also made every run of lepton error out if bitlen(1024) != 11
  * pin local brotli dep
  * first cut at brotli support
  * add missing files
  * fix brotli compression: add test for brotli header compression
  * quality 9 uses too much memory--move to quality 10--which seems to be more acceptible for memory consumption
  * can now produce a .lep file that represents a file with a single embedded jpeg within
  * fix verification to properly understand embedded files (by blanking out the -embedding command line flag on decode) on OSX and linux. Add a test for embedding
  * add an EOF marker at the end of any stream, so that we don't intrude upon concatenated lepton files
  * process files one-by-one if they are concatenated together. Current failure upon threads reading second work-items
  * resetting works for concatenation of 2x the same file
  * fix single-threaded decode of two copies of the same image concatenated
  * clear grbgdata and prefix data between runs
  * proper deallocation of grbgdata
  * account for some variations between the first and second file by calling reset_buffers
  * clear min_noise_threshold if it would be <= 0, so that a second runthrough doesn't reuse previously set values
  * allow number of threads to taper off over time
  * max and min encode threads so you can force a particular thread count for a file....will help with multi-file concatenations where you want a single thread count everywhere
  * get rid of printf
  * do not specify a thread count outside [minthread,maxthread] range
  * fix some memory leaks between decodes of concatenated lepton files
  * teach lepton to understand concatenated huffman tables, JPEG headers, prefix or postfix junk etc...
  * lepcat mode that can concatenate two .lep images together in a more efficient way by putting all header material in the same place
  * checks to make sure files are compatible before concatenation
  * Move concatenation code into its own header
  * fix llvm build
  * allow more than 2 files to be concatenated
  * fix undefined behavior in brotli bit reader, fix allocator type, and have a better strategy for picking a max size for brotlic ompression
  * fix for threads having no work and more even work split by row count rather than by bit density
  * revert thread splitting change for version 1 lepton
  * default to old thread splitting strategy
  * don't allocate extra memory if zlib0 output is selected

- Update to version 1.2.1+git.20170405:
  * explicitly make padbit signed. Try to fix some  quirks in cmake for powerpc
  * make autotools use sane default compiler flags if none are provided in the command line
  * on ppc when -std=c++11 is defined, the __linux macro goes away
  * added documentation to seeming typo
  * fix #87 : always check that threads_required set up the appropriate number of threads---fire off nop functions on unused threads for consistency
  * check divide by zero, fixing #86
  * only build -scalar in ppc case

- Update to version 1.2.1+git.20170328:
  * Fix #85 32 bit platform support

- Update to version 1.2.1+git.20170311:
  * Cleaned up cmake config. CMake Config was messy, it is less so now Fixed inconsistent case usage Fixed indentation
  * Get lepton building with emscripten + cmake
  * Add a lepton-scalar target
  * Reduce memory usage by disabling custom malloc under emscripten, enabled -O2 optimizations (only work for lepton-scalar)
  * Added a javascript wrapper for the emscripten compiled lepton-scalar
  * added scalar checks to ./configure and to travis.yml
  * added config options to force standard memory allocators and adjust seccomp rules accordingly
  * be explicit about if a dev_assert happens versus a always_assert... demote any assert() that's costly to dev_assert
  * add billing as an option independent of NDEBUG
  * added a PacketReader interface to allow for streaming of data into the vpx_reader
  * change rand to simply use modular arithmetic for the test reader class
  * make the streaming work for single threaded case
  * intrusive list nodes
  * use correct thread_id for setting up the packet readers
  * plumb the thread communication through the whole stack. Ugly init semantics for SendToVirtualThread state)
  * getting close: initialize the bool decoder later, when the streams are available
  * make single thread work again for progressive files
  * added baseline version of the progressive reading of data
  * remove some debug printouts:
  * remove preload of the mux_reader, since that is allocated bit-by-bit
  * receive data queued up in batches from the main thread, if available
  * trying to make a unified model where every thread is treated equally, and thus can finish at a more similar timestamp
  * fix singlethread progressive case
  * make unified framework operate with progressive files...both single threaded and multi
  * give the first worker its own full thread to fully decouple the I/O with the processing so that we don't starve the other workers
  * set the thread mapping before kicking off the first thread
  * setup memmgr with right number of threads
  * added test of an image encoded with more threads than available by default
  * fix multithreaded issue when more virtual threads exist than workers
  * remove the legacy nonunified thread model code, coalesce single-threaded code in the recode.cc case
  * simplified the decode_chunk progressive inner loop
  * startup all threads before waiting for them to start
  * wipe pthread
  * fewer assert statements... simpler reasoning?
  * Include a copy of the Apache license
  * added a test for failing round trip
  * prewarm get_nprocs for ubuntu12.04 compatibility with using posix_memalign under seccomp filters
  * this should make perf significantly worse; but it gets better... unclear why
  * Added a benchmark mode for lepton take filename as arg instead of including it... the reasonable include in smalljpg.hh is sufficient benchmark to produce meaningful stats
  * use std::mutex instead of std::atomic for thread synchronization when STANDARD_MEMORY_ALLOCATORS are on...this helps race detection tools and is simpler to reason about.
  * no longer define __APPLE__ to isolate threads
  * added some extra error handling and fixed a loop for decoding of progressive zero runs
  * add some initialization to constructors to make it easier to reason about. Also set EOBrun to better detect runs
  * remove buffer leak from benchmark
  * allow benchmark to finish early if you want to test less system load
  * windows support
  * between logical threads, never assume previous thread overran its bound
  * prewarm get_nprocs for ubuntu12.04 compatibility with using posix_memalign under seccomp filters
  * Fixed benchmark to work with custom allocators. Also added better help explanations to the help output display
  * minor changes to make unused items only defined when they are used
  * better help message on windows

- Update to version 1.2.1+git.20161213:
  * Ensure proc.terminate() is called after unit test
  * ioutil: poll.h -> sys/select.h
  * close opened file descriptors properly
  * Revert "close opened file descriptors properly"
  * resource leak: fp
  * Fix posix_memalign call
  * Added missing 'defined'.
  * fix uninitialized variable

- Update to version 1.2.1+git.20160921:
  * make sockettester and forktester python3 compatible
  * Ensure proc.terminate() is called after unit test
  * ioutil: poll.h -> sys/select.h
  * close opened file descriptors properly
  * Revert "close opened file descriptors properly"
  * resource leak: fp
  * Fix posix_memalign call
  * Added missing 'defined'.

- Update to version 1.2.1+git.20160813:
  * ioutil: poll.h -> sys/select.h
  * close opened file descriptors properly

- Update to version 1.2.1+git.20160802:
  * Added wrapper libraries section to README
  * formatting
  * rename it to LICENSE.md
  * removing the extension
  * fixes for #47 : i386 compat
  * Add php bindings notice to README.md
  * make sockettester and forktester python3 compatible
  * Ensure proc.terminate() is called after unit test

- Update to version 1.2.1+git.20160718:
  * /build for .gitignore
  * 64-bit-related fixes
  * Win32 build improvements
  * Fixed types to be 32-bit in all cases, fixed comments
  * Unified types
  * remove unused components of zlib
  * it builds now, but will it run; the way things are going I don't know...

- Initial package for version 1.0