Tracking Issue for `feature(file_set_times)`: `FileTimes` and `File::set_times`

[joshtriplett]

Feature gate: #![feature(file_set_times)]

This is a tracking issue for setting file accesses/modification timestamps via the File::set_times function and associated FileTimes structure.

Public API

// std::fs

#[derive(Copy, Clone, Debug, Default)]
pub struct FileTimes { /* ... */ };

impl FileTimes {
    pub fn new() -> Self;
    pub fn set_accessed(self, t: SystemTime) -> Self;
    pub fn set_modified(self, t: SystemTime) -> Self;
}

impl File {
    pub fn set_times(&self, times: FileTimes) -> std::io::Result<()>;
    pub fn set_modified(&self, t: SystemTime) -> std::io::Result<()>;
}

// std::os::ios::fs
// std::os::macos::fs
// std::os::watchos::fs
// std::os::windows::fs

pub trait FileTimesExt: Sealed {
    fn set_created(self, t: SystemTime) -> Self;
}

impl FileTimesExt for std::fs::FileTimes;

Steps / History

• Implementation: Support setting file accessed/modified timestamps #98246
• Add creation time support to FileTimes on apple and windows #109773
• Final comment period (FCP)
• Stabilization PR

[m-ou-se]

I wonder if it would be good to have a file.set_last_modified(time) short-hand for file.set_times(FileTimes::new().set_modified(time)). FileTimes nicely allows platforms to have a FileTimesExt trait to add more times, but it's a bit verbose for simple use cases.

[joshtriplett]

@m-ou-se No fundamental objection, though I don't think we should provide an alias for accessed (less common to set in isolation, and if we had both people might be tempted to call both rather than making one call to set_times).

Also, I'd probably call it set_modified for symmetry with the various uses of modified.

[joshtriplett]

@m-ou-se Done.

[sunfishcode]

On Unix-family platforms, there's a mechanism for setting a timestamp to the current time (UTIME_NOW), which doesn't appear to be covered in this API. One can get the same effect by using a timestamp of SystemTime::now(); is that fast enough? I expect it is on Linux with the vDSO, but I don't know about other platforms.

[sargun]

One thing I noticed is that there's no way to set the AT_SYMLINK_NOFOLLOW flag for this call. We might want to expose that.

[silence-coding]

Is there a planned stabilization time for this feature?

[Zerowalker]

is it possible to at created times here as well?
I know it exists in Windows, not sure about Unix but i would expect it's a thing there as well.

[rbtcollins]

statx is needed for birth time support, and its getting more and more support, but it still pretty new as these things go. https://www.phoronix.com/news/FAT-File-Creation-Birth-5.19 for instance adding support for FAT32 support.

@sargun This seems to be operating on the FD - does NOFOLLOW make any sense when not operating with paths?