- Revert to previous version of _is_zero() XSub. (More reliable.)
  - Alter Overload_abs() to work correctly with -0 when the compiler/libc
    insists that -0 * -1.0 is still -0.
  - In new(), correct occurrence of "NaNLD(1)" to "NaNLD()" - and add test to new.t.
  - Replace SvUV/SvIV/SvNV with SvUVX/SvIVX/SvNVX where appropriate. (In typemap, too.)
  - Ensure that, in LDtoNV(), we don't cast an extended precision long double "inf" to
    a __float128. Instead cast a double "inf" to __float128. (This is done only when
    nvtype is __float128 && longdblkind is either 3 or 4.)

  - tweak output of try2.in.
  - alter test 12 in t/overload_cmp_reversal.t to cater for dragonfly (in addition
    to freebsd).
  - alter test 2 in t/special.t to cater for dragonfly's reduced erfl and erfcl

  - Additional checks to Makefile.PL, trying to eliminate multiple FAIL reports
    that emanate from Slaven's and Bingos' freebsd and dragonfly machines.

  - Improve the workaround for the __MINGW64__ modfl bug.
  - Align default output precision with mpfr library. (ie Rewrite
    calculation of _DIGITS in LongDouble.xs.)
  - Extend the modf_LD mingw64 workaround to include mingw-runtime version 5.0.
  - Fix cmp overloading.

  - Another rewrite of try.in aimed at eliminating the continuing
    FAIL reports from Bingos' NetBSD machines.
  - Rewrite try2.in to avoid constants.

  - 0.13 failed to pick when Bingos' BSD machines were going to choke on erfl() - so
    I've rewritten try.in so that it actually outputs something.
  - Change try3.in and try4.in to actually output something.

  - add implementation of sincosl(), for use when that function is not available.
  - runtime version 4 modfl() segfaults, so we introduce our own rendition of modfl()
    if that runtime is in place. See http://sourceforge.net/p/mingw-w64/bugs/478/
  - if test 6 of t/int.t reports that overloaded int(-0.6257) returns 0 instead
    of -0, we no longer record that as "not ok". We do however emit a message
    stating that we got '0' instead of the expected '-0'.
  - add some more diagnostics to tests 2, 3 and 5 of t/prec.t in an attempt to
    find out why one (or more ?) of chorny's smokers fails those tests.
  - Add nnumflag(), set_nnum() and clear_nnum() functions. Also check for
    non-numeric strings handed to strtold().

  - approx()imate test 3 result of pow.t if it fails initially.
  - run try.exe to check for undefined references that don't show up until
  - add more diagnostics to nan_pow.t for failing nan**0 tests.
  - tweaks to cmp.t to try to fix or better understand some strange failures on FreeBSD.

  - Add extra checks, workarounds, tests wrt (the possibly unimplemented)
    isnanl(), nanl() and signbitl().

  - Fix checking for nan**0 bug.
  - Modify t/nan_pow.t and t/overload.t to accommodate nan**0 bug.
  - Check for (and workaround) absence of isnanl(). (Assume that if isnanl
    is missing then so is nanl - and work around that, too.)

  - Work around bug on some compilers that has nan**0 is NaN.
  - Allow less stringent checking on cbrt() calculations that should return
    an integer value. (This accommodates a minor, but nonetheless annoying,
    bug with some libc versions.)

  - Provide more float.h and math.h constants.

  - Add math library functions.
  - Add ld_bytes function.
  - Allow overloading of UV, IV, NV and PV types.
  - Remove support for *signed* NaN.

  - Add -DOLDPERL capability to Makefile.PL

  - Provide LD_DBL_DIG LD_LDBL_DIG ld_min_inter_prec ld_max_orig_len
    ld_min_inter_base and ld_max_orig_base functions.
  - Add metadata to Makefile.PL.

  - Use float.h's LDBL_DIG to set the initial value of _DIGITS in LongDouble.xs.
  - Add _LDBL_DIG XSub so that we can see the value of float.h's LDBL_DIG.
  - Amend t/prec.t to work with powerpc64 (where long double has 106 bits of

  - Fix LDtoLD and STRtoLD documentation.
  - Alter the way that strEQ gets hold of HvNAME(SvSTASH(SvRV(sv))) - so that -Wall
    doesn't emit "null argument" warnings. (Thanks Daniel Kahn Gillmor.)
  - Add t/stringification.t.
  - Fix typos in documentation.
  - Fix bug in new() - it wasn't handling infinity/nan args correctly.

  - Replace '_precision' (not exported) with 'ld_get_prec' (exported).
  - Provide 'ld_set_prec'.
  - Provide LdtoSTRP.
  - set default precision of stringified values to 18 decimal digits.