<filesystem> support junctions in read_symlink

[strega-nil-ms]

• is_symlink(junction) is still false
• read_symlink(junction) works and does the obvious thing
• copy_symlink(junction) doesn't work - creating a directory symlink is bad, and creating a junction is not really supported by the windows API (and also a junction is not a symlink)
• copy(junction) works and does the obvious thing

Fixes #2818

[StephanTLavavej]

P0218R1_filesystem is failing with:

test_status @ 2885: check failed: is_other(file_status{file_type::junction})
Assertion failed: pass, file C:\a\1\s\tests\std\tests\P0218R1_filesystem\test.cpp, line 4044
abort() has been called

STL/tests/std/tests/P0218R1_filesystem/test.cpp

Line 2885 in ef62d3f

EXPECT(is_other(file_status{file_type::junction}));

Ideally, it's best to run a full test pass for x64 before submitting a PR. (For maintainers, if your local machine is slow, you can create a VM internally to run tests.) At a minimum, strongly related tests should be run, which will take just a few minutes. For example:

C:\GitHub\STL\tests\std\tests>dir /b *filesystem*
Dev11_1066931_filesystem_rename_noop
Dev11_1180290_filesystem_error_code
GH_001010_filesystem_error_encoding
P0218R1_filesystem
VSO_0121275_filesystem_canonical_should_handle_many_double_dots

[cpplearner]

I'm all for treating junctions as symlinks, but I'm not sure about some details.

1. The standard says ([fs.op.is.symlink]):

   bool filesystem::is_symlink(file_status s) noexcept;
   Returns: s.type() == file_­type​::​symlink.

   But this PR makes is_symlink return something different.

2. IIUC this PR makes copy_symlink create a symlink (as opposed to a junction) if the source is a junction. Since copy calls copy_symlink if is_symlink returns true, this means copy (still) will not make a faithful copy of a junction.

[strega-nil-ms]

@cpplearner you're absolutely correct, thanks for pointing that out.

[strega-nil-ms]

Waiting on LWG issue to link.

[strega-nil-ms]

We now have a LWG issue - LWG-3744!

[StephanTLavavej]

Thanks for fixing this bug! I pushed minor stylistic changes, nothing should be controversial but please meow if you have concerns. 😺

[strega-nil-ms]

LGTM!

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[StephanTLavavej]

Thanks for yet another <filesystem> improvement for Windows! 💾 📀 📂

[LewisPringle]

I'm confused about the reasoning for is_symlink () returning false for juntions.

How is one to know to call read_symlink?

https://en.cppreference.com/w/cpp/filesystem/read_symlink says its an error to call read_symlink for soemthing that isn't a symlink.

My code:
if (filesystem::is_symlink (p)) {
p = filesystem_read_symlink(p);
}

doesn't work with our chosen fix. Not sure why the above should not work?