[libc++] Granularizes vector.

[Mordante]

Summary

https://llvm.org/PR61314 reports a missing include in vector.
Unfortunately adding that include causes a circular reference.
__format/buffer.h uses vector<charT>, which is included by vector
to fix the bug.

The intention is to only let __format/buffer.h use the granularized
header, this will not use vector<bool> so it's clear which
specialization is needed. For other headers this might be possible too,
but it's not planned.

[Mordante]

CI fixes.

[Mordante]

CI fixes.

[Mordante]

CI fixes.

[Mordante]

CI fixes.

[Mordante]

CI fixes.

[Mordante]

CI fixes

[Mordante]

CI fixes.

[Mordante]

CI fixes.

[EricWF]

I think this is a really bad idea.
Splitting the specialization out so that only some files can depend on it, while others cannot, allowing some bits of the code to only see a partial implementation is going to be bug prone.

I'm also not sure I agree with the bug. The standard only requires the declaration of the type to be available without the format include. It says nothing about requiring the type to be complete IMHO.
Like that hash<std::vector<bool>> is declared in <vector> but technically you need to include <utility> to get its definition.

This dance is done specifically so that anytime a type T with a hash specialization is seen by the user, hash<T> has been properly declared, so the compiler knows to not instantiate the default implementation.

I would like to see other alternative solutions explored. I don't think this one should proceed.

libcxx/include/__vector/vector.h

339

Where did this go?

3351

Where did this go?

[EricWF]

Rereading the bug on github. I think this might deserve a standard defect report.

The spec should be a lot more clear about what exactly is provided in the library headers which provide declare specializations.

[Mordante]

In D145800#4207309, @EricWF wrote:

Rereading the bug on github. I think this might deserve a standard defect report.

The spec should be a lot more clear about what exactly is provided in the library headers which provide declare specializations.

I'm not sure this warrants an LWG-issue. Based on http://eel.is/c++draft/format#formatter.spec-4

If the library provides an explicit or partial specialization of formatter<T, charT>, that specialization is enabled and meets the Formatter requirements except as noted otherwise.

I think it's clear the bug report is valid.

The problem is caused by libc++ including <vector> from format. This is not something the Standard mandates.
Do you still feel there is a reason to file an LWG-issue?

I will look whether I can fix this in a different way.

[Mordante]

@EricWF I've discussed this further with @ldionne. He's also not thrilled by this approach and we decided on a different approach. So I will no longer pursue this approach.