* There has been a great deal of anguish over the encoding of
weekdays: whether [0, 6] maps to [Sunday, Saturday] or
[1, 7] maps to [Monday, Sunday]. This commit attempts
to address that issue, but will break a small amount of
code at compile-time. See below on how to fix that.
* The weekday constructor used to accept [0, 6] to represent
[Sunday, Saturday]. It now accepts [0, 7] to represent
[Sunday, Saturday] with both 0 and 7 mapping to Sunday.
* The conversion from weekday to unsigned has been removed.
* To convert a weekday to unsigned replace:
auto u = unsigned{wd};
with:
auto u = (wd - Sunday).count();
This maps [Sunday, Saturday] to [0, 6], which is the
C/POSIX mapping. If you prefer the ISO mapping
([Monday, Sunday] -> [1, 7]), then do:
auto u = (wd - Monday).count() + 1;
* Added new overloads for operator+/- between year_month and duration
that is convertible to years, so it is better candidate for operands
that are convertible to both years and months. To preserve
functionality, this operator is conditionally noexcept.
* Reworked year_month_day, year_month_day_last, year_month_weekday,
and year_month_weekday_last.
* Added tests for this new functionality.
* For format, all a type must do is implement to_stream.
* For parse, each type X must specialize parse_manip<X,CharT,Traits>.
Each specialization must include a public typedef to itself named
type.
* Each parse_manip specialization must have a stream extraction
operator.
* This commit depends on expression-SFINAE. If this commit breaks
your build, it is likely that your compiler/version does not
support expression-SFINAE. To fix this NO_EXPRESSION_SFINAE
needs to be defined in the configuration area of date.h for
that compiler/version.
* The unit tests should reflect basic definitions.
* Derivative relationships do not need to be tested.
* Derivative tests needlessly increase test times and noise.
* After much consideration I consider this not a good change.
* It introduced a completely arbitrary mathematical relationship
among the units.
* This commit restores a precise relationship among the units which
is based on the Gregorian calendar.
* However this experiment has been preserved in the git history.
* The Gregorian average of months and years is an
integral number of seconds.
* This patch slightly tweaks the average to make it
not an integral number of seconds, and thus these
units no longer implicitly convert to seconds.
* This will flag naive mixing of years, months and seconds
as a compile time error.
