User:T. Canens/date issues

• operator<<(duration):

• Wording is unclear that we may need to copy the result of to_w?string if there's a traits mismatch

• to_stream(duration):

• Unclear how exactly formatting is to be done for a pure duration - e.g. "12-hour"/"24-hour"?

• file_clock's to/from functions are specified with vigorous handwaving
• clock_time_conversion

• We should say that user specializations need to be DefaultConstructible; otherwise clock_cast can't use them.
• Can specializations be an empty class as well?
• "takes a parameter of type time_point<SourceClock, Duration>" - mean "argument"

• Calendar classes

• operator<< specification is all kinds of unclear in terms of interaction with stream formatting flags.

• e.g., operator<<(day): if inserting format(...) as specified, would apply padding etc. But reference impl does nothing of that sort and writes into the stream directly.
• These need to be respecified to avoid depending on format per the new direction that to_stream set failbit if !ok()
• Also the specification does not handle traits correctly - attempts to insert a string with mismatched traits if not the default.

• literal interaction with overflow is interesting - e.g., "4294967297d" on a typical implementation is defined to be 1d because overflow is checked post-static_cast.

• Howard: this is meant to be unspecified

• year::ok has quasi-editorial bug comparing y_ with min()/max() which are years
• year_month + months is insufficiently specified (need z.ok()) and needs wording to avoid UB by omission
• year_month_day_last::day()'s handling of !ok() case is unclear (and UB in the reference impl) - this has implications on a bunch of noexcepts if we keep it UB
• weekday_indexed allows [1,5] as index range, but year_month_weekday took special care to handle 0. Suggested allowing [0, 7].

• [time.format]

• What happens if an invalid specifier is encountered?

• Howard: They are treated as ordinary characters.

• "ordinary multibyte characters" wchar_t?
• Definition of "information the conversion specifier refers to" is unclear with the ISO week-based year thingys

• Howard: this is meant to be based on the general case, i.e., %G requires day-precision.

• Handling of stream flags unclear
• "unspecified value" or failbit not pinned down

• Howard: These should all set failbit, never unspecified value.

• format throws an exception in that case. Need to revisit << that are specified with format

• [time.parse]

• Same problem with ISO week-based year.
• duplicated flags?
• Check %g's interpretation if no century info?
• parse needs to use addressof for abbrev - let's check the entire subclause
• No overload taking const charT*?!
• Result/type of extraction not specified.

• time_of_day

• "Each specialization of time_­of_­day is a trivially copyable and standard-layout class type." Not possible for arbitrary Rep unless you ban it
• What exactly happens if constructed with a duration > 24h?

• tzdb_list

• I don't think "thread-safe" is a defined term. Double check and replace with the usual formulation

• Yep, use "does not introduce a data race".

• tzdb

• Can default construct? Is aggregate?
• Public exposition-only member that's not used anywhere