Hack around a getting precise time in Windows

[da-x]

This takes the approach from rust-lang/rust#67266 by @dbergman. This may be necessary for the v0.1 branch of time because the widely-used chrono crate still depends on it. When writing logs on Windows, crates such as simplelog which depend on chrono would benefit from a better timestamp. @quodlibetor @Drakulix

[jhpratt]

A few things:

• CI is failing due to the use of a static atomic
• I don't have knowledge of Windows APIs. Please comment the code so I can follow what it's doing and why.
• This code does not appear to be compatible with Windows < 8, per the rust-lang issue.
• What benefit does this provide, exactly? Time v0.1 is no longer supported, so only major issues are being fixed.

[da-x]

• Seems to be failing due to Rust 1.24 compatibility. I'll look into resolving it.
• Sure, I'll add the comments.
• It should be compatible with older versions of Windows, as it implements a fallback to the older function if the newer function is not exported from kernel32. The pending PR I've opened for std does something similar but in a cleaner approaching using existing std machinery.
• The major benefit is the accuracy of a timestamp going down significantly from 10ms to the 1usec area. This is useful for applications that write log files.

[jhpratt]

I'm still not quite sure this is worthy of another release. As I mentioned, 0.1 isn't supported any more. The current limitation of ~10ms is how it's been for a very long time. If this is released in a 0.1.43, there are a number of other things that I have previously closed that people would want to revisit.

Is there anywhere this has caused major issues that can't be worked around?

[da-x]

I guess that a possible workaround is to bake this functionality straight into the chrono crate, and have it not depend on old time, maybe only for Windows. Maybe @quodlibetor can comment?

[jhpratt]

As I decided in #240, I'll be creating one final release for 0.1 at some point in the future. As such, I'll review this PR along with others when I get the chance.

[pfmooney]

It should be compatible with older versions of Windows, as it implements a fallback to the older function if the newer function is not exported from kernel32. The pending PR I've opened for std does something similar but in a cleaner approaching using existing std machinery.

It would be nice to have a test plan for these older versions of Windows, both with stable rust and 1.21 (if possible).

[github-actions]

This pull request has not had any activity recently. It will be closed without further activity.