[libc++][chrono] Implements formatter day.

[Mordante]

Summary

This implements the enabled specializaton
template<class charT> struct formatter<chrono::day, charT>;

and
template<class charT, class traits>

basic_ostream<charT, traits>&
  operator<<(basic_ostream<charT, traits>& os, const day& d);

Implements:

• LWG 3241 chrono-spec grammar ambiguity in §[time.format]

Partially implements:

• P1361 Integration of chrono with text formatting

[Mordante]

libcxx/include/__chrono/formatter.h

61–62

@vitaut suggested something similar. I like this approach better, modulo the trailing comma.

[Mordante]

Addresses review comments and attempts to fix the CI.

[Mordante]

Attempt to fix CI.

[ldionne]

libcxx/include/__chrono/formatter.h

52

I think it might be important to use = tm() to get proper zero initialization, otherwise some members might be left uninitialized.

[ldionne]

libcxx/docs/Status/Cxx20Issues.csv

178

libcxx/include/__chrono/formatter.h

51

I don't think this should be in formatter.h or in the __formatter namespace. This seems generally useful, don't you think? I might suggest something like __chrono/convert_to_tm.h to mirror convert_to_timespec.h.

libcxx/include/__chrono/statically_widen.h

14

36

Equivalent since _LIBCPP_STATICALLY_WIDEN should only be used in our namespace, but still more idiomatic since this is a macro.

46

Ditto.

libcxx/include/module.modulemap.in

445

This shouldn't be here.

459

Not needed.

463

Not needed.

libcxx/test/libcxx/utilities/format/format.formatter/format.formatter.spec/formattable.compile.pass.cpp

146

I would prefer instead:

// The chrono formatters require localization support because {XXXXXXXX}
#ifndef TEST_HAS_NO_LOCALIZATION
  assert_is_formattable<std::chrono::day, CharT>();
#endif

We don't need to test that it's not formattable when localization is disabled.

libcxx/test/std/time/time.cal/time.cal.day/time.cal.day.nonmembers/ostream.pass.cpp

35–37

Leftover?

90–91

Leftover?

libcxx/test/std/time/time.syn/formatter_tests.h

8

Missing header guard.

61–67

89–93

I would suggest putting that above so it doesn't clutter the body of the function so much. You could use something like using Context = ..... The same comment applies above too.

104–105

libcxx/test/std/utilities/format/format.formatter/format.formatter.spec/types.compile.pass.cpp

119

Same comment as above for TEST_HAS_NO_LOCALIZATION.

[Mordante]

• Rebased
• Addresses most review comments
• Qualify several calls
• Comment out some transitive includes that caused include cycles

[Mordante]

Adds work-around to test CI.

[Mordante]

Attempts to fix Windows build.
Adds convert_to_tm private header.

[Mordante]

Fixes CI.

[Mordante]

Attempts to fix CI.

[Mordante]

Tests failing CI.

[Mordante]

Attempt to fix the CI.

[Mordante]

Remove more parts of the experimental library.

[Mordante]

Another attempt to fix the CI.

[Mordante]

CI test

[Mordante]

CI test

[ldionne]

Notes from live review:

1. Let's move transitive includes to the end of each file to break cycles
2. Let's guard transitive includes on standard versions

Also, please update this to its final form and we can take a look at the Windows DLL errors.

[Mordante]

CI

[Mordante]

CI

[Mordante]

Restore CI and try to fix circular dependencies.

[Mordante]

Disable a test and test CI

[Mordante]

libcxx/include/chrono

733

As tested in D132284.

737

The header can be unconditional again.

[Mordante]

Rebased and integrated all upstream changes.

[Mordante]

Attempts to fix CI.

[Mordante]

Rebased and some final polishing.

[ldionne]

libcxx/include/chrono

732

Is this comment still required? I think it is no longer needed since we inlined the destructor of format_error to workaround Windows DLL issues.

[Mordante]

libcxx/include/chrono

732

No it's not, removed it.