time::Instant is a different size of different platforms

[Thomasdezeeuw]

time::Instant on Linux is 16 bytes, while it is 8 bytes on MacOs. This makes it hard to assert the size of any structs using it.

Playground link

use std::mem;
use std::time::Instant;

fn assert_size<T>(want: usize) {
    assert_eq!(mem::size_of::<T>(), want);
}

fn main() {
    // This is 16 on Linux, but 8 on MacOS (and Windows?).
    assert_size::<Instant>(16);

    // This is 24 on Linux, but 16 on MacOS (and Windows?).
    assert_size::<Option<Instant>>(24);
}

[sfackler]

Why do you want to make assertions about the size of a struct with an Instant in it?

[Thomasdezeeuw]

I’m using it as a deadline. And I want to make sure the struct doesn’t grow in the future, it’s big enough already. Met vriendelijke groet, Thomas de Zeeuw https://thomasdezeeuw.nl <https://thomasdezeeuw.nl/> [email protected] <mailto:[email protected]>

…

On 19 Jul 2017, at 17:40, Steven Fackler ***@***.***> wrote: Why do you want to make assertions about the size of a struct with an Instant in it? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#43332 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADA0GOswRaZTctqyIqV6pG0UspWNHvikks5sPiNXgaJpZM4Ocr8R>.

[Mark-Simulacrum]

Perhaps you could use #[cfg(unix)]? I don't think it's likely that we'll change anything to increase or decrease the size of Instant based on platform... On OS X today, an Instant is defined by u64 whereas on other unix platforms it's a libc::timespec... I would presume that this is correct and won't change.

[Thomasdezeeuw]

That is what I’m currently using. It just seemed odd so I filed this bug for it. But if no change is possible or desired it might be worth documenting it? Met vriendelijke groet, Thomas de Zeeuw https://thomasdezeeuw.nl <https://thomasdezeeuw.nl/> [email protected] <mailto:[email protected]>

…

On 19 Jul 2017, at 18:05, Mark Simulacrum ***@***.***> wrote: Perhaps you could use #[cfg(unix)]? I don't think it's likely that we'll change anything to increase or decrease the size of Instant based on platform... On OS X today, an Instant is defined by u64 whereas on other unix platforms it's a libc::timespec... I would presume that this is correct and won't change. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#43332 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADA0GDGVlzPDcnC88ct0mePxUu5_ReTwks5sPik3gaJpZM4Ocr8R>.

[Mark-Simulacrum]

Hm, well, I'll nominate for libs team, but I personally don't think documenting differences in size really makes sense today.

[durka]

@Mark-Simulacrum stuff like this could be a good candidate for the "portability lint" discussed here and here.

[alexcrichton]

The libs team decided that we're not going to make any guarantees about sizes of structs, and we've opened #43601 to track documenting that the sizes may be different.

[Thomasdezeeuw]

Sounds good to me.