Verion 1.5.3
* Integer rounding error in m5slice
* Add ctypes-based Python bindings 
* Add Python examples (m5spec.py, m5stat.py, m5subband.py, m5time.py, m5tone.py, m5selfcorr.py)
* m5subband: extract a single subband by FFT-based filtering of a wideband input file
* m5pcal: now supports tones at non-integer MHz, better supports wideband recordings, 
  adds FFT length (--chunksize) option, and is updated to behave more DiFX-like
* m5d: adds complex decode (--double) and custom printout format (e.g., --format=%f)
* test5b: will now report skipped frames
* zerocorr: write results with %e to avoid scaling causing underflow results
* more const correctness
* functions added that mask bad values in channel-dependent manner from decoded VDIF streams if EDV4 is used.  Per-channel accounting of zero values is provided.
* Fix some data rate and frame rate calculations in mark5bfile.h
* New function: void snprintmark5bfilesummary(char *str, int maxLength, const struct mark5b_file_summary *sum)
* Change RPM name to vlbi-mark5access.*.rpm
* New option in test5b: can supply frames per second and then the true number of skipped frames can be evaluated.
* fixmark5b: don't seg fault if can't open files
* Many utilities: change from use of signal() to sigaction(), which is supposed to be more reliable.
* Version for DiFX-2.5, May 16, 2017

Version 1.5.2
* Post DiFX-2.4
* Detect Legacy VDIF

Version 1.5.1
* Improve behavior of fixmark5b() in the case that fill pattern is seen at the 1 second transition.
* m5bsum calculates frame rates now rather than just assuming 25600 frames per second
* Improve blanking of mark5b in fix case (make use of the TVG bit to blank frame by frame)
* Added support for "d2k" mode, which is Mark5B with sign and magnitude bits swapped
* m5bstate: complex sampling is now supported
* examples/makeheader: tool to make a SIGPROC pulsar filterbank compatible header.
* m5bfb: Additions to polarisation support.
* Version for DiFX-2.4, Nov 4, 2014

Version 1.5.0
* Add KVN5B format: like Mark5B but different data bit patterns (Richard Dodson)
* Add 3, 5, 6, and 7 channel VDIF decoders
* Add functionality to repair Mark5B data streams (regrid frames, remove stray bytes, fill missing frames with fill pattern)
* Add Mark5B file summary program, m5bsum, akin to vsum (in vdifio)
* Reorder format checks to look for Mark5B earlier than Mark4 or VLBA as that format is becoming much more common
* Utility m5fb (mark5 filterbank) to generate dynamic spectrum from baseband data
* Version for DiFX-2.3, Dec 20, 2013

Version 1.4.4
* Slightly faster and more relaxed frame finding for VLBA and Mark4 formats
* Numerous small improvements based on improved cppcheck exposed by Martin
Ettl
* Add genheaders() for Mark4 and Mark5B formats
* m5test: minor improvements
* c++11 compliance
* Much improved capabilities for reading from stdin
* m5bstate: recognize filled data and don't count
* Remove findvdif program.  new program vsum in the vdifio package is much
better.
* m5timeseries utility added
* 32 channel 2 bit VDIF support (Nicolas Pradel)
* Version for DiFX-2.2, May 31, 2013

Version 1.4.3
* Post 2.0.1 tagging
* Adopt more unix-standard return values in utility programs
* mk5spec: work on complex data (Chris Phillips)
* VDIF format determination by data peeking added
* Version for DiFX-2.1, May 4, 2012

Version 1.4.2
* Post 2.0.0 tagging
* Start adding state counting for use in power detection.  2-bit sampling supported only at this time.
* Add m5tsys
* Can now make RedHat RPM
* Fix some memory leaks identified by valgrind.
* Introduce new_mark5_stream_absorb.  like new_mark5_stream, but it cleanly and immediately deallocates space for the temporary mark5_stream_generic and mark5_format_generic structures
* Version for DiFX-2.0.1, June 10, 2011

Version 1.4.1
* revert to standard "c" types (long long, unsigned char, ...) over int64_t, uint8_t
* disable all K5 code until that is at least somewhat useful
* Some memory leak fixes from Alexander Neidhardt and Martin Ettl (from Wettzell)
* Add m5fold utility
* Add some 4 and 8 bit unpackers for VDIF format
* m5d and m5spec : behave better if -h is given
* mark5_format_mark5b_init: if no nanoseconds in headers, then assume 2048-16-2 (this is to support Mark5B mode via Mark5C)
* Put pcal utility in this package
* Add zerocorr -- a zero baseline cross correlator
* Add sigint catching to several utils
* Improve string (mostly filename) handling safety with snprintf
* Allow read from stdin (at least in some circumstances) via filename = "-"
* Version for DiFX-2.0.0

Version 1.4.0
* Add VDIF format -- but only for formats with 1 thread
* Add VLBN format == VLBA without data modulation
* Fix blank range checking for a small subset of Mark4 modes
	(Thanks to Sergei Pogrebenko and Jan Wagner for the bug report)
* Version for DiFX-1.5.2

Version 1.3.2
* Fix sample granularity calculation for Mark5B
* Version for DiFX-1.5.1

Version 1.3.1
* Version for DiFX-1.5

Version 1.3.0
* Post NRAO-DiFX 1.1
* Fix file based stream offset problem for all data alignments
* Start of Mark5C development

Version 1.2.3
* fix minor bug in blanking code -- end of header not blanked on first frame
  of mark4 format
* add m5spec -- a spectrometer test program.  Needs fftw3 to compile.
* allow offsets > 2 GiB
* add m5test -- look for problems in data files
* add consecutivefails.  
* fix to file stream type -- sometimes 1 frame would be skipped
* fix to mark4 decoding -- sample offset was wrong.  stupid data replacement!

Version 1.2.2
* rename oversampling to decimation

Version 1.2.1
* add oversampling (factors 1, 2, 4, 8+) support to mark4 vlba mark5 formats

Version 1.2.0
* fix some Mark5B issues.  This required changing ns to be a double throughout.

Version 1.1.3
* fix datestamp on mark4 data -- subtract 1 day (Jan 0, not Jan 1 is the zeroth day of year)

Version 1.1.2
* Support < 8 tracks on vlba and mark4 formats
* new unpack function allowing offset into stream

Version 1.1.1
* Fix unpack/blanker issue
* Fix mark5_stream_open()
* New functions:
    const char *mark5_stream_list_formats();
    struct mark5_format *new_mark5_format_from_name(const char *formatname);

Version 1.1.0
* Start of K5 support -- not yet usable
* Implement blanking: look for fill pattern and decode to all zeros.

Version 1.0.1
* Fix new_mark5_format_from_stream()

Version 1.0
* Support of VLBA, Mark4 & Mark5B formats
* Support File and Memory streams
* Support "pure" unpacking -- no frame validation
* Big and Little Endian support