Changes for version v0.012 - 2016-07-14
- test coverage: Devel::Cover showed two conditionals which can only ever trigger one of the two (impossible to hit the opposite); remove the unnecessary conditional
- bugfix <https://rt.cpan.org/Public/Bug/Display.html?id=116005>: get SNAN, expect QNAN
- original code always expected QNAN, because all of the developer's machines automatically quiet SNAN to QNAN
- v0.011_001: Test suite for :floatingpoint now converts SNAN to expect /[SQ]NAN/, because it doesn't really matter that much whether Perl silences a signaling NAN
- bugfix <https://rt.cpan.org/Public/Bug/Display.html?id=116006>
- getting NAN instead of expected value on nextup(NEGATIVE NORMAL) or nextdown(POSITIVE NORMAL)
- v0.011_001: when LSB underflows (goes to -1), the hexification is too many F's; misinterpreted when the two were being recombined, so masked the MSB and LSB before merging
- bugfix <https://rt.cpan.org/Public/Bug/Display.html?id=116004>
- test suite getting -0.160000000000000003 when expecting -0.16
- caused by systems with $Config{nvsize} > 8: code expected 64bit NV (some CPAN Testers have 128bit NV)
- v0.011_002: change test suite to only compare enough significant figures for a 64bit double, so that even 128bit floats (quad?) will pass
- v0.011_004: added documentation to explain that the module expects 64bit NV, rounds off more-precise NV floats to 64bit, and will fail tests on 32bit NV (but might coerce to 64bit floats and might still "work" if the installation is forced)
Modules
Various tools for understanding and manipulating the underlying IEEE-754 representation of floating point values