[FEA] convert_datetime.cu should use proper chrono types for timestamp/duration handling

[jrhemstad]

Is your feature request related to a problem? Please describe.

The functionality in convert_datetime.cu parses timestamps using plain integers instead of chrono types. This should be avoided as it leads to problems like this: #9793

Describe the solution you'd like

Use appropriate chrono types and functions

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[github-actions]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[davidwendt]

It appears the chrono classes do not handle negative timestamps

cudf/cpp/src/strings/convert/convert_datetime.cu

Lines 720 to 724 in dad51a5

	// Note: Tried using: cuda::std::chrono::hh_mm_ss(T::duration(timestamp));
	// and retrieving the hour, minute, second, and subsecond values from it
	// but it did not scale/modulo the components for negative timestamps
	// correctly -- it simply did an abs(timestamp) as documented here:
	// https://en.cppreference.com/w/cpp/chrono/hh_mm_ss/hh_mm_ss

The chrono classes are being used wherever possible:

cudf/cpp/src/strings/convert/convert_datetime.cu

Lines 315 to 320 in dad51a5

	auto const ymd = // convenient chrono class handles the leap year calculations for us
	cuda::std::chrono::year_month_day(
	cuda::std::chrono::year{timeparts.year},
	cuda::std::chrono::month{static_cast<uint32_t>(timeparts.month)},
	cuda::std::chrono::day{static_cast<uint32_t>(timeparts.day)});
	auto const days = cuda::std::chrono::sys_days(ymd).time_since_epoch().count();

cudf/cpp/src/strings/convert/convert_datetime.cu

Lines 598 to 601 in dad51a5

	return cuda::std::chrono::year_month_day(cuda::std::chrono::year{year},
	cuda::std::chrono::month{month},
	cuda::std::chrono::day{day})
	.ok();

and have been helpful in handling tricky calculations like:

• day-of-week:

  cudf/cpp/src/strings/convert/convert_datetime.cu

  Lines 781 to 782 in dad51a5

  	auto const day_of_week =
  	cuda::std::chrono::year_month_weekday(days.value()).weekday().c_encoding();

• week-of-year:

  cudf/cpp/src/strings/convert/convert_datetime.cu

  Lines 888 to 896 in dad51a5

  	[[nodiscard]] __device__ int8_t get_week_of_year(cuda::std::chrono::sys_days const days,
  	cuda::std::chrono::sys_days const start) const
  	{
  	return days < start
  	? 0
  	: (cuda::std::chrono::duration_cast<cuda::std::chrono::weeks>(days - start) +
  	cuda::std::chrono::weeks{1})
  	.count();
  	}

• especially in handling ISO weeks and ISO years:

  cudf/cpp/src/strings/convert/convert_datetime.cu

  Lines 858 to 859 in dad51a5

  	[[nodiscard]] __device__ thrust::pair<int32_t, int32_t> get_iso_week_year(
  	cuda::std::chrono::year_month_day const& ymd) const

I don't believe there is anything left in this file that can benefit from the chrono classes.
Therefore, I'd like to close this issue.