version 0.006; 2007-10-01

  * bugfix: change behaviour of nextafter() when both arguments are
    numerically zero to that specified by IEEE 754: the second argument
    is returned rather than the first

  * bugfix: in hex_float(), cleanly reject non-ASCII digits in exponent

  * bugfix: in float_hex(), cleanly reject non-ASCII digits in an
    "EXPONENT="-type "zero_strategy" setting

  * add IEEE 754r functions nextup() and nextdown()

  * add IEEE 754r function totalorder()

  * add constant max_number

  * hex_float(): accept IEEE 754r special input strings "infinity" and
    "snan", in addition to the existing "inf" and "nan"

  * hex_float(): make the "0x" prefix in hexadecimal input optional

  * float_hex(): new option "hex_prefix_string" to control the "0x" prefix

  * test classification functions, examination functions, string
    conversion functions, and manipulation functions (all the functions
    that were not being tested)

  * test all functions for lack of side effects on zero arguments and
    purity of zero results

  * in documentation, note new standard-conforming behaviour of copysign()
    with a NaN second argument

  * in documentation, note that hex_float() accepts the special "0.0"
    form of input without a sign, as well as with one

  * in documentation, where the IEEE standard nature of functions is
    noted, add discussion of non-conforming behaviour on NaNs

  * in documentation, change some variable names in the synopsis for
    clarity

  * test POD syntax and coverage, and rename some internal functions to
    satisfy the coverage test

  * tweak tests on constants to avoid infinite loops if importing constant
    functions fails

  * build with Module::Build instead of ExtUtils::MakeMaker

  * complete dependency list

  * include signature in distribution

  * in documentation, separate "license" section from "copyright" section

version 0.005; 2007-01-25

  * bugfix: change behaviour of copysign() with a NaN as the second
    argument to that specified by IEEE 754: it is not an error but
    (in the context of Perl's opaque NaNs) results in copying an
    unpredictable sign

  * add hex_float() function to input floating point values in hexadecimal

  * add IEEE 754 function signbit()

  * float_id_cmp(): tighten specification of return values to match
    Perl's <=> operator (actual behaviour always matched the tighter spec,
    it just wasn't documented and tested)

  * in documentation, note that the string values of float arguments
    are ignored

  * in documentation, clarify note about implicit conversion of integer
    to float

  * in documentation, note that both arguments to nextafter() must be
    floating point values

  * in documentation, note standard nature of signbit(), copysign(),
    and nextafter()

  * in documentation, note float_id_cmp()'s relation to the IEEE 754r
    function totalorder()

  * in documentation, note that Perl does not distinguish between
    different NaNs

  * in documentation, give a second definition of max_integer (equivalent
    to the first)

  * add test for consistency of constants

  * in documentation, reference Scalar::Number

  * remove now-useless test t/use.t

version 0.004; 2007-01-12

  * bugfix: correct value for max_integer to 2^(significand_bits+1),
    and correct its description in the documentation

  * make pos_zero and neg_zero constants into non-constant functions
    that return fresh floating point zeroes, if running on a Perl where
    floating point zeroes can transmogrify into integer zeroes

  * add float_id_cmp() function for total ordering of floats

  * in documentation, more details of behaviour of zeroes

  * in documentation, discussion of interaction with native integer values

  * in documentation, note truncating behaviour of the % operator

  * in documentation, note the slightly misleading nature of the names
    "significand_bits" and "max_integer"

  * reference Data::Integer and perlnumber(1) in documentation

version 0.003; 2006-08-08

  * float_hex(): add OPTIONS parameter to control details of output
    formatting

  * when looking for NaNs, see whether the string "nan" qualifies

  * slight clarification to documentation of significand_step

version 0.002; 2006-08-03

  * bugfix: in mult_pow2(), copysign(), and nextafter(), take care
    to return a pristine signed zero when returning zero: they were
    returning zeroes that got broken (due to the Perl bug noted in the
    documentation) by internal arithmetic

version 0.001; 2006-08-01

  * add IEEE 754 functions copysign() and nextafter()

  * test sign of zero using sprintf() to avoid being confused by dualvars

  * slight expansion of initial paragraph of documentation

version 0.000; 2006-07-30

  * initial released version