Use mem::MaybeUninit instead of mem::uninitialized() since Rust 1.36

[rhysd]

I noticed that CI fails since this crate is using deprecated std::mem::uninitialized().

https://ci.appveyor.com/project/Stebalien/term/builds/26853323/job/880r4cnu5lhfaccv

This PR uses MaybeUninit which replaces std::mem::uninitialized() using a compiler version switch provided by rustversion crate.

[rhysd]

I have addressed the review comment at 8edc047. Instead of uninitialized(), it wraps GetConsoleScreenBufferInfo() as get_console_screen_buffer_info(). After getting uninitialized memory, it immediately initializes the buffer with GetConsoleScreenBufferInfo(). So the returned buffer is guaranteed to be initialized and there is no undefined behavior.

(I apologize many force pushes. I don't have Windows machine so I needed to check Appveyor result instead)

[Stebalien]

Thanks but I believe this is still unsafe. See: file:///home/steb/.local/lib/rustup/toolchains/nightly-x86_64-unknown-linux-gnu/share/doc/rust/html/std/mem/union.MaybeUninit.html#examples-1

[Stebalien]

This needs to call assume_init after filling the buffer. That is, it should call GetConsoleScreenBufferInfo(handle, buffer_info.as_mut_ptr() then buffer_info.assume_init() (if there was no error).

We can't take a reference to uninitialized memory (which is why this type was introduced).

[rhysd]

Ahh, ok, I misunderstood the semantics of MaybeUninit. Thank you for the point.

[rhysd]

I fixed my patch again at 3e9624f and added a new small test for get_console_screen_buffer_info() at 65801a0.

[Stebalien]

Thanks! LGTM.

And yeah, uninitialized values are tricky business...

[rhysd]

Thank you for correcting me and helping to understand correctly.