TODO

    - test tzoffset
    - make more tests (see Devel::Cover)
    - clean up xs
    - test strptime with '%z' and '%Z'
        - timegm(@tm_parts) to get ephoch, then CORE::localtime($epoch) to return right parts
            - or return 'needs convert' flag, get epoch from GMT parts, then get parts from localtime()
    - allow cdate() to print as different formats
    - Document format flags that don't work everywhere (see comments in *.t files)
    - Every "return 0" in the strptime case statement in Piece.xs needs to be preceded by a useful error statement
    - Check for int and size_t mismatches (like from strlen())
    - _C_time_locale should only get populated when really needed
    - use macros or something to get all the strlen stuff out of xs hash fetches
    - make sure changing class vars effects all instances 
    - add better %V support, newer versions of libc seem to have it: 
            https://opensource.apple.com/source/Libc/Libc-1158.20.4/stdtime/FreeBSD/strptime.c.auto.html

for population of c_epoch (https://rt.cpan.org/Ticket/Display.html?id=118927)
was done because of ephoch call for strftime change:

-    return _strftime($format, (@$time)[c_sec..c_isdst]);
+
+    return _strftime($format, $time->epoch, $time->[c_islocal]);

causes state change and test suites that called is_deeply would see the change and fail.
so figure out how timelocal(@time_parts) gets ephoch and just move that into c perhaps.
    (well that is not going to be so easy... so perhaps just change it back to old behavior and see what breaks on cpan)
        (( perhaps check first if c_epoch is defined, if not just call timelocle and not set c_epoch))

Document Time::Piece limits:
    dates before 1970
    strptime doesn't really handle %z %Z or %V
    add_months calls _mini_maketm which only uses gmtime, probably wrong
    strftime doesn't really handle %s right since it calls system mktime with wrong parameters if given a tm struct built from gmtime (fixed?)
    uses epoch everywhere so don't try to calculate stuff before 1970
    mktime docs even state it is only for localtime (gets used for stuff from gmtime sometimes)
    TP is guaranteed to parse stuff that is returned from strftime, not so much any other random time string