v2.18

• Support for much larger range of sample rates. Outside of the range 32767 to 1/32767 integer sample rates are approximated whenever possible.
• Rewrite of all decoding and encoding routines to replace those used from qlib2, now the entire project is LGPL licensed.
• Remove limitation on sample rate before calling ms_genfactmult() in the normal path of packing records. Previously generating the factor and multiplier was not attempted for rates higher than 32,767.
• ms_genfactmult() now support a much larger range of integer sample rates and periods.
• ms_genfactmult() now sets the factor and multiplier using the SECONDS/SAMPLE notation for sample rates less than 1.0 to retain precision for low rates.
• ms_genfactmult() now assumes the specified rate is a sample period in seconds if the samprate value is negative.
• Add ms_rsqrt64() as a general use reciprocal sqrt for doubles.
• Use memcpy() instead of assignment when unpacking float32 and float64 samples to avoid problems with NaN's. Thanks Lion Krischer.
• Add test for reading records without Blockette 1000.
• Reformat all source code using included clang-format profile.
• A more elegant sanity check for output length in packing by mbyt.
• Improvements for test suite, more consistency.
• Remove msr_decode_steim? from libmseed.def, they are internal.
• Add sanity to length check before memset calls in packing functions.
• Check for environment variables ENCODE_DEBUG and DECODE_DEBUG and set debugging output, at this point it is Steim frame details and differences being encoded/decoded.
• Fix padding in steim[12] encoding routines.
• Remove unneeded output buffer checks in steim[12] decoding routines.
• Replace data sample packing and unpacking routines from qlib2 with new routines developed from scratch. All code is now LGPL licensed.
• Add test suite with tests for encoding, decoding, parsing, etc.
• Update licensing to GNU-LGPL version 3 and include (L)GPL licenses in LICENSE.txt.
• Define needed C99 int types for MSVC 2012 or earlier. Previously this was only done for versions earlier than MSVC 2010.

v2.17

• Round Fixed Section Data Header start time values to the nearest tenth of millisecond and restrict the microsecond offset value to a range between -50 and +49 as recommended in SEED. Previously start times were truncated at tenths of millisecond resolution and the microsecond offset value was between 0 and +99. This also addresses a bug where microsecond offsets were off by 100ms for times before Jan 1 1970. Thanks to Lion Krischer for reporting.

v2.16

• Cleanup of lmplatform.h removing unneeded headers and using C99 standard headers except for a few platform specific cases.
• Convert all printf() and scanf() usage of %lld for 64-bit integers to use the C99 PRId64 and SCNd64 macros for portability (MingGW).
• Change detection of Linux/Cygwin to set global define LMP_LINUX instead of LMP_GLIB2 (now marked as deprecated).
• Change detection of Windows to set global define LMP_WIN instead of LMP_WIN32 (now marked as deprecated).
• Add detection of MINGW64 define.
• Tested building on Win7 with: Open Watcom 1.9, MinGW gcc 4.8.1 and Cygwin 1.7.35 (gcc 2.9.2).
• Define NTP-Posix time epoch conversion constant specifically as a long long integer to avoid warnings on some compilers.

v2.15

• Fix infinite loop if blockette chain is corrupt. Patch submitted by Elliott Sales de Andrade.

v2.14

• Fix memory leak when msr_pack() returns after an error. Patch contributed by Larry Baker and Eric Thomas.
• Change casting of values passed to isdigit() to int instead of unsigned character. Apparently this was confusing on ARM systems.
• Add ms_readleapseconds() and ms_readleapsecondfile() routines to read a leap seconds file into an internal list.
• Modify msr_endtime() to check for an internal leap second list, check for overlap with the record and adjust the end time as needed when leap seconds are present. When a leap second list is present any indication of positive leap seconds in the fixed section data header are ignored.
• Add casting to size_t and int to avoid build warnings on certain build systems (e.g. older MS Visual Studio). The effective maximum sample buffer and record buffer size is ~2GB.

v2.13

• Clean up Makefile and example/Makefile, remove all GCC-specific and debug targets. Makefiles are compatible with both GNU and BSD make. As before, the shared target works only for GCC-compatible compilers. Thanks to Elliott Sales de Andrade for pointing out that shared library targets did not work with parallel builds, prompting this clean up.
• Support 128-byte record length by changing MINRECLEN to 128 (was 256). The current SEED specification is a minimum record length of 256-bytes, but there are cases (e.g. low latency data flow) using 128-byte records.
• Add declarations, casting and string truncation to clear warnings uncovered in ObsPy testing (thanks to Elliott Sales de Andrade).
• Add CYGWIN defined test to Linux section of lmplatform.h.

v2.16m

• Add defines for needed integer types and macros from inttypes.h missing in older MSVC versions. MSVC 2010 and later appear to have enough C99 support.
• Add define tests for _WIN32 and _WIN64 to cover all WINs.
• ms_fread(): Add cast to quiet warning for conversion from size_t to int. In this case the read will always be <= MAXRECLEN, much smaller than int, making the conversion safe.