split internal dates methods into separate functions

[aplavin]

another attempt at #53692

[LilithHafner]

I still like the changes.

Need to adjust rata2datetime and datetime2rata to use _yearmonthday, too.

[aplavin]

Tests still fail, but I cannot see any specific failure in the logs.

[LilithHafner]

Yeah, it's confusing to me too. I'm trying to figure out how to rerun the build in case it's a buildkite issue. I've never seen a failure like this.

[aplavin]

bump... not sure what to do here

[LilithHafner]

Thanks for the bump. Next steps are to fix test failures in CI and then run pkgeval. Here's a real test failure from CI:

Error During Test at /Users/julia/.julia/scratchspaces/a66863c6-20e8-4ff4-8a62-49f30b1f605e/agent-cache/default-grannysmith-C07ZQ07RJYVY.0/build/default-grannysmith-C07ZQ07RJYVY-0/julialang/julia-master/julia-8d470900e4/share/julia/stdlib/v1.12/Dates/test/accessors.jl:59
--
  | Test threw exception
  | Expression: (y, m, d) == Dates.yearmonthday(days)
  | MethodError: no method matching yearmonthday(::Int64)
  | The function `yearmonthday` exists, but no method is defined for this combination of argument types.

[aplavin]

Hmmm, some of these methods are tested directly in

julia/stdlib/Dates/test/accessors.jl

Lines 8 to 46 in c6c449c

	@testset "yearmonthday/yearmonth/monthday" begin
	# yearmonthday is the opposite of totaldays
	# taking Rata Die Day # and returning proleptic Gregorian date
	@test Dates.yearmonthday(-306) == (0, 2, 29)
	@test Dates.yearmonth(-306) == (0, 2)
	@test Dates.monthday(-306) == (2, 29)
	@test Dates.yearmonthday(-305) == (0, 3, 1)
	@test Dates.yearmonth(-305) == (0, 3)
	@test Dates.monthday(-305) == (3, 1)
	@test Dates.yearmonthday(-2) == (0, 12, 29)
	@test Dates.yearmonth(-2) == (0, 12)
	@test Dates.monthday(-2) == (12, 29)
	@test Dates.yearmonthday(-1) == (0, 12, 30)
	@test Dates.yearmonth(-1) == (0, 12)
	@test Dates.monthday(-1) == (12, 30)
	@test Dates.yearmonthday(0) == (0, 12, 31)
	@test Dates.yearmonth(-0) == (0, 12)
	@test Dates.monthday(-0) == (12, 31)
	@test Dates.yearmonthday(1) == (1, 1, 1)
	@test Dates.yearmonth(1) == (1, 1)
	@test Dates.monthday(1) == (1, 1)
	@test Dates.yearmonthday(730120) == (2000, 1, 1)
	end
	@testset "year/month/day" begin
	# year, month, and day return the individual components
	# of yearmonthday, avoiding additional calculations when possible
	@test Dates.year(-1) == 0
	@test Dates.month(-1) == 12
	@test Dates.day(-1) == 30
	@test Dates.year(0) == 0
	@test Dates.month(0) == 12
	@test Dates.day(0) == 31
	@test Dates.year(1) == 1
	@test Dates.month(1) == 1
	@test Dates.day(1) == 1
	@test Dates.year(730120) == 2000
	@test Dates.month(730120) == 1
	@test Dates.day(730120) == 1
	end

Any idea whether it was intended as a public api (yearmonthday(x) and others taking any number) or not?

[LilithHafner]

I don't see that usage documented, so no: not public. Those tests should be switched to testing _yearmonthday and friends. Nanosoldier can tell us if packages depend on those undocumented internals.

@nanosoldier runtests()

[nanosoldier]

The package evaluation job you requested has completed - possible new issues were detected.
The full report is available.

[LilithHafner]

The failures in Kedzi, NanoDates, and Dates look real. If this were breaking something that was ever public, I would be concerned. As is, we should investigate Kexzi and NanoDates's failures and it would be polite to make PRs that fix them but I don't view those nanosoldier failures as blocking.